TIME TRAVEL AND OTHER MATHEMATICAL BEWILDERMENTS
T I M E TRAVEL AND OTHER MATHEMATICAL BEWILDERMENTS
MARTIN GARDNER ...
91 downloads
767 Views
15MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
TIME TRAVEL AND OTHER MATHEMATICAL BEWILDERMENTS
T I M E TRAVEL AND OTHER MATHEMATICAL BEWILDERMENTS
MARTIN GARDNER
me W. H. FREEMAN AND COMPANY
N E W
Y O R K
Library of'Congress CataloguinginPublication Data
Gardner, Martin, 1914Time travel and other mathematical bewilderments. 1nclud~:s index. 1. Mathematical recreations. QA95.G325 1987 793.7'4 ISBN 071671924X ISBN 0716719258 @bk.)
I. Title. 8711849
Copyright la1988 by W. H. Freeman and Company No part of this book may be reproduced by any mechanical, photographic, or electronic process, or in the form of a photographic recording, nor may it be stored in a retrieval system, transmitted, or otherwise copied for public or private use, without written permission from the publisher. Printed in the United States of America 3 4 5 6 7 8 9 0 VB 6 5 4 3 2 1 0 8 9
To David A. Klarner for his many splendid contributions to recreational mathematics, for his friendship over the years, and in !gratitude for many other things.
C
O
N
T
E
N
T
S
CHAPTER O N E
Time Travel
1
CHAPTER T W O
Hexes and Stars
.
15
CHAPTER THREE
Tangrams, Part 1
27
CHAPTER FOUR
Tangrams, Part 2
39
CHAPTER FIVE
Nontransitive Paradoxes
55
CHAPTER SIX
71
Combinatorial Card Problems
CHAPTER SEVEN
MelodyMaking Machines
85
CHAPTER EIGHT
.
Anamorphic Art
97
CHAPTER NINE
The Rubber Rope and Other Problems
111
viii
CONTENTS
CHAPTER TEN
125
Six Sensational Discoveries
CHAPTER ELEVEN
The Csaszar Polyhedron
139
CHAPTER TWELVE
153
Dodgem and Other Simple Games
CHAPTER THIRTEEN
163
Tiling with Convex Polygons
CHAPTER FOURTEEN
Tiling with Polyominoes, Polyiamonds, and Polyhexes
CHAPTER FIFTEEN
Curious Maps
189
CHAPTER SIXTEEN
The Sixth Symbol and Other Problems
CHAPTER SEVENTEEN
Magic Squares and Cubes
213
CHAPTER EIGHTEEN
Block Packing
227
CHAPTER NINETEEN
Induction and ~ r o b a b i l i t ~ 24 1
205
177
CONTENTS
CHAPTER TWENTY
Catalan Numbers
253
CHAPTER TWENTYONE
Fun with a Pocket Calculator CHAPTER TWENTYTWO
TreePlant Problems
INDEX OF NAMES
22 7
291
267
ix
Herewith the twelfth collection of my columns from Scientijic American. As usual, they have been corrected, updated, and expanded, mostly on the basis of letters from knowledgeable and alert readers.
TIME TRAVEL AND OTHER MATHEMATICAL BEWILDERMENTS
2 • CHAPTER ONE
Schools Journal under the horrendous title “The Chronic Argonauts.” Wells himself was so ashamed of this clumsily written tale that he broke it off after three installments and later destroyed all the copies he could find. A completely rewritten version, “The Time Traveller’s Story,” was serialized in The New Review beginning in 1894. When it came out as a book in 1895, it brought Wells instant recognition. One of the many remarkable aspects of Wells’s novella is the introduction in which the Time Traveller (his name is not revealed, but in Wells’s first version he is called Dr. Nebogipfel) explains the theory behind his invention. Time is a fourth dimension. An instantaneous cube cannot exist. The cube we see is at each instant a cross section of a “fixed and unalterable” fourdimensional cube having length, breadth, thickness, and duration. “There is no difference between Time and any of the three dimensions of Space,” says the Time Traveller, “except that our consciousness moves along it.” If we could view a person from outside our spacetime (the way human history is viewed by the Eternals in Isaac Asimov’s The End of Eternity or by the Tralfamadorians in Kurt Vonnegut’s SlaughterhouseFive), we would see that person’s past, present, and future all at once, just as in 3space we see all parts of a wavy line that traces on a time chart the onedimensional spatial movements of mercury in a barometer. Reading these remarks today, one might suppose that Wells had been familiar with Hermann Minkowski’s great work of tidying up Einstein’s special theory of relativity. The line along which our consciousness crawls is, of course, our “world line”: the line that traces our movements in 3space on a fourdimensional Minkowski spacetime graph. (My World Line is the title of George Gamow’s autobiography.) But Wells’s story appeared in its final form ten years before Einstein published his first paper on relativity! When Wells wrote his story, he regarded the Time Traveller’s theories as little more than metaphysical hankypanky designed to make his fantasy more plausible. A few decades later physicists were taking such hankypanky with the utmost seriousness. The notion of an absolute cosmic time, with absolute simultaneity between distant events, was swept out of physics by Einstein’s equations. Virtually all physicists now agree that if an astronaut were to travel to a distant star and back, moving at a velocity close to that of light, he could in theory travel thousands of years into the earth’s future. Kurt Gödel constructed a rotating cosmological model in which one can, in principle, travel to any point in the world’s past as well as future, although travel to the past is ruled out as physically impossible. In 1965 Richard P. Feynman received a Nobel prize for his spacetime approach to quantum mechanics in which antiparticles are viewed as particles momentarily moving into the past.
TIME TRAVEL
3
Hundreds of sciencefiction stories have been written about time travel, many of them raising questions about time and causality that are as profound as they are sometimes funny. T o give the most hackneyed example, suppose you traveled back to last month and shot yourself through the head. Not only do you know before making the trip that nothing like this happened but, assuming that somehow you could murder your earlier self, how could you exist next month to make the trip? Fredric Brown's "First Time Machine" opens with Dr. Grainger exhibiting his machine to three friends. One of them uses the device to go back sixty years and kill his hated grandfather when the man was a youth. The story ends sixty years later with Dr. Grainger showing his time machine to two friends. It must not be thought that logical contradictions arise only when people travel in time. The transportation of anything can lead to paradox. There is a hint of this in Wells's story. When the Time Traveller sends a small model of his machine into the past or the future (he does not know which), his guests raise two objections. If the time machine went into the future, why do they not see it now, moving along its world line? If it went into the past, why did they not see it there before the Time Traveller brought it into the room? One of the guests suggests that perhaps the model moves so fast in time it becomes invisible, like the spokes of a rotating wheel. But what if a timetraveling object stops moving? If you have no memory of a cube on the table Monday, how could you send it back to Monday's table on Tuesday? And if on Tuesday you go into the future, put the cube on the table Wednesday, then return to Tuesday, what happens on Wednesday if on Tuesday you destroy the cube? Objects carried back and forth in time are sources of endless confusion in certain sciencefiction tales. Sam Mines once summarized the plot of his own story, "Find the Sculptor," as follows: "A scientist builds a time machine, goes 500 years into the future. He finds a statue of himself commemorating the first time traveler. He brings it back to his own time and it is subsequently set up in his honor. You see the catch here? It had to be set up in his own time so that it would be there waiting for him when he went into the future to find it. He had to go into the future to bring it back so it could be set up in his own time. Somewhere a piece of the cycle is missing. When was the statue made?" A splendid example of how paradox arises, even when nothing more than messages go back in time, is provided by the conjecture that tachyons, particles moving faster than light, might actually exist. Relativity theory leaves no escape from the fact that anything moving faster than light would move backward in time. This is what inspired A. H. Reginald Buller, a Canadian botanist, to write his often quoted limerick:
4
CHAPTER ONE
There was a young lady named Bright W h o traveled much faster than light. She started one day In the relative way, And returned on the previous night.
Tachyons, if they exist, clearly cannot be used for communication. G. A. Benford, D. L. Book, and W . A. Newcomb (of "Newcomb's paradox," the topic of two chapters in my Knotted Doughnuts a d Other Mathematical Entertainments, W . H . Freeman and Company, 1986), have chided physicists who are searching for tachyons for overlooking this. In "The Tachyonic Antitelephone," they point out that certain methods of looking for tachyons are based on interactions that make possible, in theory, communication by tachyons. Suppose physicist Jones on the earth is in communication by tachyonic antitelephones with physicist Alpha in another galaxy. They make the following agreement. W h e n Alpha receives a message from Jones, he will reply immediately. Jones promises to send a message to Alpha at three o'clock earth time, if and only if he has not received a message from Alpha by one o'clock. Do you see the difficulty? Both messages go back in time. IfJones sends his message at three, Alpha's reply could reach him before one. "Then," as the authors put it, "the exchange &.messages will take place if and only if it does no; take place . . . a genuine . . . causal contradiction." Large sums of money have already gone down the drain, the authors believe, in efforts to detect tachyons by methods that imply tachyonic communication and are therefore doomed to failure. Time dilation in relativity theory, time travel in Godel's cosmos, and reversed time in Feynman's way of viewing antiparticles are so carefully hedged by other laws that contradictions cannot arise. In most timetravel stories the paradoxes are skirted by leaving out any incident that would generate a paradox. In some stories, however, logical contradictions explicitly arise. When they do, the author may leave them paradoxical to bend the reader's mind or may try to escape from paradox by making clever assumptions. Before discussing ways of avoiding the paradoxes, brief mention should be made of what might be called pseudotimetravel stories in which there is no possibility of coitradiction. There can be no paradox, for example, if one simply observes the past but does not interact with it. The electronic machine in Eric Temple Bell's "Before the Dawn," which extracts motion pictures of the past from imprints left by light on ancient rocks, is as free of possible paradox as watching a video tape of an old television show. And paradox
TIME TRAVEL • 5
cannot arise if a person travels into the future by going into suspended animation, like Rip van Winkle, or Woody Allen in his motion picture Sleeper, or the sleepers in such novels as Edward Bellamy’s Looking Backward or Wells’s When the Sleeper Wakes. No paradox can arise if one dreams of the past (as in Mark Twain’s A Connecticut Yankee at King Arthur’s Court, or in the 1986 motion picture Peggy Sue Got Married), or goes forward in a reincarnation, or lives for a while in a galaxy where change is so slow in relation to earth time that when he returns, centuries on the earth have gone by. But when someone actually travels to the past or the future, interacts with it and returns, enormous difficulties arise. In certain restricted situations paradox can be avoided by invoking Minkowski’s “block universe,” in which all history is frozen, as it were, by one monstrous spacetime graph on which all world lines are eternal and unalterable. From this deterministic point of view one can allow certain kinds of time travel in either direction, although one must pay a heavy price for it. Hans Reichenbach, in a muddled discussion in The Philosophy of Space and Time (Dover, 1957, pp. 140–142), puts it this way: Is it possible for a person’s world line to “loop” in the sense that it returns him to a spot in spacetime, a spot very close to where he once had been and where some kind of interaction, such as speech, occurs between the two meeting selves? Reichenbach argues that this cannot be ruled out on logical grounds; it can only be ruled out on the ground that we would have to give up two axioms that are strongly confirmed by experience: (1) A person is a unique individual who maintains his identity as he ages, and (2) a person’s world line is linearly ordered so that what he considers “now” is always a unique spot along the line. (Reichenbach does not mention it, but we would also have to abandon any notion of free will.) If we are willing to give up these things, says Reichenbach, we can imagine without paradox certain kinds of loops in a person’s world line. Reichenbach’s example of a consistent loop is as follows. One day you meet a man who looks exactly like you but who is older. He tells you he is your older self who has traveled back in time. You think him insane and walk on. Years later you discover how to go back in time. You visit your younger self. You are compelled to tell him exactly what your older duplicate had told you when you were younger. Of course, he thinks that you are insane. You separate. Each of you leads a normal life until the day comes when your older self makes the trip back in time. Hilary Putnam, in “It Ain’t Necessarily So,” argues in similar fashion that such worldline loops need not be contradictory. He draws a Feynman graph (see Figure 1) on which particle pairproduction and pairannihilation are replaced by person pairproduction and pairannihilation. The zigzag line is
6
CHAPTER ONE
SPACE
Figure 1 Feynman graph for a time traveler to the past
the world line of time traveler Smith. At time t, he goes back to t,, converses with his younger self, then continues to lead a normal life. How would this be observed by someone whose world line is normal? Simply put a ruler at the bottom of the chart, its edge parallel to the space axis, and move it slowly upward. At toyou see young Smith. At t, an older Smith suddenly materializes out of thin air in the same room along with an antiSmith, who is seated in his time machine and living backward. (If he is smoking, you see his cigarette butt lengthen into a whole cigarette, and so on.) Perhaps the two forward Smiths converse. Finally, at t,, young Smith, backward Smith, and the backwardmoving time machine vanish. The older Smith and his older time machine continue on their way. The fact that we can draw a spacetime diagram of these events, Putnam insists, is proof that they are logically consistent. It is true that they are consistent, but note that Putnam's scenario, like Reichenbach's, involves such weak interaction between the Smiths that it evades the deeper contradictions that arise in timetravel fiction. What happens if the older Smith kills the younger Smith? Will Putnam kindly supply a Feynman graph?
TIME TRAVEL
7
There is only one good way out, and sciencefiction scribblers have been using it for more than half a century. According to Sam Moskowitz, the device was first explicitly employed to resolve timetravel paradoxes by David R. Daniels in "Branches of Time," a tale that appeared in Wonder Stories in 1934. The basic idea is as simple as it is fantastic. Persons can travel to any point in the future of their universe, with no complications, but the moment they enter the past, the universe splits into two parallel worlds, each with its own time track. Along one track rolls the world as if no looping had occurred. Along the other track spins the newly created universe, its history permanently altered. When I say "newly created," I speak, of course, from the standpoint of the time traveler's consciousness. For an observer in, say, a fifth dimension the traveler's world line simply switches from one spacetime continuum to another on a graph that depicts all the universes branching like a tree in a metauniverse. Forking time paths appear in many plays, novels, and short stories by nonsciencefiction writers. J. B. Priestley uses it in his popular play Dangerous Corner) as Lord Dunsany had done earlier in his play Mark Twain discusses it in The Mysterious Stranger. Jorge Luis Borges plays with it in his "Garden of Forking Paths." But it was the sciencefiction writers who sharpened and elaborated the concept. Let us see how it works. Suppose you go back to the time of Napoleon in Universe 1 and assassinate him. The world forks. You are now in Universe 2. If you like, you can return to the present of Universe 2, a universe in which ~a~oleo had n been mysteriouslymurdered. How much would this world differ from the old one? Would you find a duplicate of yourself there? Maybe. Maybe not. Some stories assume that the slightest alteration of the past would introduce new causal chains that would have a multiplying effect and produce vast historical changes. Other tales assume that history is dominated by such powerful overall forces that even major alterations of the past would damp out and the future would soon be very much the same. In Ray Bradbury's "A Sound of Thunder," Eckels travels back to an ancient geological epoch under elaborate precautions to prevent any serious alteration of the past. For example, he wears an oxygen mask to prevent his microbes from contaminating animal life. But Eckels violates a prohibition and accidentally steps on a living butterfly. When he returns to the present, he notices subtle changes in the office of the firm that arranged his trip. He is killed for having illegally altered the future. Hundreds of other stories by fantasy and sciencefiction writers have played variations on this theme. One of the saddest is Lord Dunsany's "Lost" (in The Fourth Book ofJorkens) 1948). A man travels to his past, by way of an Oriental
IJ:
8
CHAPTER ONE
charm, to right some old mistakes. Of course, this alters history. When he gets back to the present, he is missing his wife and home. "Lost! Lost!" he cries. "Don't go back down the years trying to alter anything. Don't even wish to . . . . And, mind you, the whole length of the Milky Way is more easily traveled than time, amongst whose terrible ages I am lost." It is easy to see that in such a metacosmos of branching time paths, it is not possible to generate paradox. The future is no problem. If you travel to next week, you merely vanish for a week and reappear in the future a week younger than you would have been. But if you go back and murder yourself in your crib, the universe obligingly splits. Universe 1 goes on as before, with you vanishing from it when you grow up and make the trip back. Perhaps this happens repeatedly, each cycle creating two new worlds. Perhaps it happens only once. W h o knows? In any case, Universe 2 with you and the dead baby in it rolls on. You are not annihilated by your deed, because now you are an alien from Universe 1 living in Universe 2. In such a metacosmos it is easy (as many sciencefiction writers have done) to fabricate duplicates of yourself. You can go back a year in Universe 1, live for a year with yourself in Universe 2, then again go back a year to visit two replicas of yourself in Universe 3. Clearly, by repeating such loops you can create as many replicas of yourself as you please. They are genuine replicas, not pseudoreplicas as in the scenarios by Reichenbach and Putnam. Each has his independent world line. History might become extremely chaotic, but there is one type of event that can never occur: a logically contradictory one. This vision of a metacosmos containing branching worlds may seem crazy, but respectable physicists have taken it quite seriously. In Hugh Everett 111's Ph.D. thesis "'Relative State' Formulation of Quantum Mechanics" (Reviews ofModern Physics 29, July 1957, pp. 454462) he outlines a metatheory in which the universe at every micromicroinstant branches into countless parallel worlds, each a possible combination of microevents that could occur as a result of microlevel uncertainty. The paper is followed by John A. Wheeler's favorable assessment in which he points out that classical physicists were almost as uncomfortable at first with the radical notions of general relativity. "If there are infinite universes," wrote Fredric Brown in What Mad Universe, "then all possible combinations must exist. Then, somewhere, everything must be true . . . . There is a universe in which Huckleberry Finn is a real person, doing the exact things Mark Twain described him as doing. There are, in fact, an infinite number of universes in which a Huckleberry Finn is doing every possible variation of what Mark Twain might have described him as doing . . . . And infinite universes in which the states of existence are such
TIME TRAVEL
9
that we would have no words or thoughts to describe them or to imagine them." What if the universe never forks? Suppose there is only one world, this one, in which all world lines are linearly ordered and objects preserve their identity, come what may. Brown considers this possibility in his story "Experiment." Professor Johnson holds a brass cube in his hand. It is six minutes to three o'clock. At exactly three, he tells his colleagues, he will place the cube on his time machine's platform and send it five minutes into the past. "Therefore," he remarks, "the cube should, at five minutes before three, vanish from my hand and appear on the platform, five minutes before I place it there." "How can you place it there, then?" asked one of his colleagues. "It will, as my hand approaches, vanish from the platform and appear in my hand to be placed there." At five minutes to three the cube vanishes from Professor Johnson's hand and appears on the platform, having been sent back five minutes in time by his future action of placing the cube on the platform at three. "See? Five minutes before I shall place it there, it is there!" "But," says a frowning colleague, "what if, now that it has already appeared five minutes before you place it there, you should change your mind about doing so and not place it there at three o'clock? Wouldn't there be aparadox of some sort involved?" Professor Johnson thinks this is an interesting idea. T o see what happens, he does not put the cube on the platform at three. There is no paradox. The cube remains. But the entire universe, including Professor Johnson, his colleagues, and the time machine, disappears.
ADDENDUM J. A. Lindon, a British writer of comic verse, sent me his sequel to the limerick about Miss Bright: When they questioned her, answered Miss Bright, "I was there when I got home that night; So I slept with myself, Like two shoes on a shelf, Putup relatives shouldn't be tight!" Ned Block wrote to say he had heard the following blue version from a student at M.I.T.:
10
CHAPTER ONE
There was a young couple named Bright W h o could make love much faster than light. They started one day In the relative way, And came on the previous night. Many readers called attention to two difficulties that could arise from time travel in either direction. If travelers stay at the same spot in spacetime, relative to the universe, the earth would no longer be where it was. They might find themselves in empty space, or inside something solid. In the latter case, would the solid body prevent them from arriving? Would one or the other be shoved aside? Would there be an explosion? The second difficulty is thermodynamic. After the time traveler departs, the universe will have lost a bit of massenergy. When he arrives, the universe gains back the same amount. During the interval between leaving and arriving, the universe would seem to be violating the law of massenergy conservation. I mentioned briefly what is now called the "manyworlds interpretation" of Q M (quantum mechanics). The best reference is a 1973 collection of papers on the topic, edited by Bryce DeWitt and Neil1 Graham. Assuming that the universe constantly splits into billions of parallel worlds, the interpretation provides an escape from the indeterminism of the Copenhagen interpretation of QM, as well as from the many paradoxes that plague it. Some physicists who favor the manyworlds interpretation have argued that the countless duplicate selves and parallel worlds produced by the forking paths are not "real," but only artifacts of the theory. In this interpretation of the manyworlds interpretation, the theory collapses into no more than a bizarre way of saying the same things that are said in the Copenhagen interpretation. Everett himself, in his original 1957 thesis, added in proof this famous footnote: In reply to a preprint of this article some correspondents have raised the question of the "transition from possible to actual," arguing that in "realityH there isas our experience testifiesno such splitting of observer states, so that only one branch can ever actually exist. Since this point may occur to other readers the following is offered in explanation. The whole issue of the transition from "possible" to "actual" is taken care of in the theory in a very simple way there is no such transition, nor is such a transition necessary for the theory to be in
TIME TRAVEL
11
accord with our experience. From the viewpoint of the theory all elements of a superposition (all "branches") are "actual," none any more "real" than the rest. It is unnecessary to suppose that all but one are somehow destroyed, since all the separate elements of a superposition individually obey the wave equation with complete indifference to the presence or absence ("actuality" or not) of any other elements. This total lack of effect of one branch on another also implies that no observer will ever be aware of any "splitting" process. Arguments that the world picture presented by this theory is contradicted by experience, because we are unaware of any branching process, are like the criticism of the Copernican theory that the mobility of the earth as a real physical fact is incompatible with the common sense interpretation of nature because we feel no such motion. In both cases the argument fails when it is shown that the theory itself predicts that our experience will be what it in fact is. (In the Copernican case the addition of Newtonian physics was required to be able to show that the earth's inhabitants would be unaware of any motion of the earth.) The manyworlds interpretation has been called a beautiful theory nobody can believe. Nevertheless, a number of top physicists have indeed acceptedsome still do its outrageous multiplicity of logically possible worlds. Here is DeWitt defending it in "Quantum Mechanics and Reality," a 1970 article reprinted in the collection he edited with Graham: The obstacle to taking such a lofty view of things, of course, is that it forces us to believe in the reality of all the simultaneous worlds . . . in each of which the measurement has yielded a different outcome. Nevertheless, this is precisely what [the inventors of the theoryl would have us believe . . . . This universe is con,. stantly splitting into a stupendous number of branches, all resulting from the measurementlike interactions between its myriads of components. Moreover, every quantum transition taking place on every star, in every galaxy, in every remote corner ofthe universe is splitting our local world on earth into myriads of copies of itself. I still recall vividly the shock I experienced on first encountering this multiworld concept. The idea of 1O1O0+ slightly imperfect copies of oneself all constantly splitting into further copies, which
12
CHAPTER ONE
ultimately become unrecognizable, is not easy to reconcile with commonsense. Although John Wheeler originally supported the manyworlds interpretation, he has since abandoned it. I quote from the first chapter of his Frontiers of Time (Center for Theoretical Physics, 1978): Imaginative Everett's thesis is, and instructive, we agree. W e once subscribed to it. In retrospect, however, it looks like the wrong track. First, this formulation of quantum mechanics denigrates the quantum. It denies from the start that the quantum character of nature is any clue to the plan of physics. Take this Hamiltonian for the world, that Hamiltonian, or any other Hamiltonian, this formulation says. I am a principle too lordly to care which, or why there should be any Hamiltonian at all. You give me whatever world you please, and in return I give you back many worlds. Don't look to me for help in understanding this universe. Second, its infinitely many unobservable worlds make a heavy load of metaphysical baggage. They would seem to defy Mendeleev's demand of any proper scientific theory, that it should "expose itself to destruction." Wigner, Weizsacker, and Wheeler have made objections in more detail, but also in quite contrasting terms, to the relativestate or manyworlds interpretation of quantum mechanics. It is hard to name anyone who conceives of it as a way to uphold determinism. In the paper titled "Rotating Cylinders and the Possibility of Global Causality Violation," physicist Frank Tipler raised the theoretical possibility of constructing a machine that would enable one to go forward or backward in time. (Tipler is one of the few remaining enthusiasts for the manyworlds interpretation, and the coauthor of a controversial book, The Anthropic Cosmological Principle, Oxford University Press, 1986). Taking off from Godel's rotating cosmos and from recent work on the spacetime pathologies surrounding black holes, Tipler imagines a massive cylinder, infinitely long, and rotating so rapidly that its surface moves faster than half the speed of light. Spacetime near the cylinder would be so distorted that, according to Tipler's calculations, astronauts could orbit the cylinder, going with or against its spin, and travel into their past or future. Tipler speculated on the possibility that such a machine could be built with a cylinder of finite length and mass, but later concluded that such a device was
TIME TRAVEL
13
impossible to construct with any known forms of matter and force. Such doubts did not inhibit Poul Anderson from using Tipler's cylinder for time travel in his novel The Avatar, nor did it stop Robert Forward from writing "How to Build a Time Machine" (Omni,May 1980). "We already know the theory," O m n i editors commented above Forward's backward article, "All that's needed is some advanced engineering." I close with two pearls of wisdom from the standup comic "Professor" Irwin Corey: "The past is behind us, and the future lies ahead."
B
I
B
L
I
O
G
R
A
P
H
Y
Travelers in Time. Philip Van Doren Stern, ed. Doubleday, 1947. Science Fiction Adventures in Dimension. Groff Conklin, ed. Vanguard, 1953. "It Ain't Necessarily So." Hilary Putnam, in The Journal ofPhilosophy 59, October 25, 1962, pp. 658671; reprinted in Putnam's Philosophical Papers, Vol. 1, Cambridge University Press, 1975. "Is Time Travel Possible?" J. J. C. Smart, in The Journal of Philosophy 60, 1963, pp. 237241. "Time and Fiction in Drama." J. B. Priestley, in Man and Time. Doubleday, 1964. "Measuring Measuring Rods." John C. Graves and James E. Roper, in Philosophy ofscience 32, January 1965, pp. 3956. "On Going Backward in Time." John Earman, in Philosophy 1967, pp. 211 222.
of Science 34, September
"Particles That Go Faster Than Light." Gerald Feinberg, in Scientific American, February 1970, pp. 6977. "The Tachyonic Antitelephone." G. A. Benford, D. L. Bock, and W. A. Newcomb, in Physical Review D 2, July 15, 1970, pp. 263265.
Time and the SpaceTraveler. L. Marder. Allen & Unwin, 1971. "Tachyon Paradoxes." L. S. Schulman, in American Journal ofphysics 39, May 1971, pp. 481484.
The Many Worlds Interpretation ofQuantum Mechanics. Bryce S. DeWitt and Neil1 Graham, eds. Princeton University Press, 1973. "Rotating Cylinders and the Possibility of Global Causality Violation." Frank J. Tipler, in Physical Review D 9, April 15, 1974, pp. 22032206. "The Paradoxes of Time Travel." David Lewis, in American Philosophical Quarterly 13, 1976, pp. 145 152.
14
CHAPTER ONE
"Time and the Nth Dimension" and "Lost and Parallel Worlds." The VisualEncyclopedia of Science Fiction. Brian Ash, ed. Harmony, 1977. "Time Travel," "Time Paradoxes," "Alternate Worlds," and "Parallel Worlds." The Science Fiction Encyclopedia. Peter Nicholls, ed. Doubleday, 1979. "Time Travel and Other Universes." Chapter 5 of The Science in Science Fiction. Peter Nicholls, ed. Knopf, 1983.
TWO
Hexes and Stars
Ancient Greek mathematicians, particularly the Pythagoreans, were entranced by figurate numbers: numbers that could be represented by arranging points in regular patterns on a plane or in space. Among the plane figurate numbers, the most studied were the polygonal numbers. Figure 2 shows how the first four polygonal numberstriangular, square, pentagonal, and hexagonal are built up as the partial sums of simple arithmetic progressions. Triangular numbers are partial sums of the counting numbers 1 2 3 4 . . . . Square numbers are formed by successive addition of the consecutive odd numbers 1 3 5 7 . . . . Pentagonal numbers derive from the progression 1 4 7 10 . . . and hexagonal numbers from the progression 1 5 9 1 3 . . . . The respective differences are 1, 2, 3,. . . . The study of figurate numbers belongs to a branch of number theory called Diophantine analysis, which has to do with finding integral solutions of equations. An enormous effort was made by the great pioneers of number theory in studying the properties of polygonal numbers. Most of this work is crisply summarized in the second volume of Leonard E. Dickson's History of the Theory of Numbers (Carnegie Institute, 1919).
+
+ + +
+ + + + + + + + + + + +
16 • CHAPTER TWO
Figure 2 Construction of order5 polygonal numbers
Let us start with a classic problem that was solved by Leonhard Euler in 1730. How can we find all the numbers that are both square and triangular? The formula for the nth triangular number is 12 (n 2 + n). If this expression is also square, we have the Diophantine equation 12 (n 2 + n) = m 2. Learning the technique of solving this equation is an excellent introduction to Diophantine analysis. The initial step is to manipulate the equation to get a simpler equation that will be a key to the solution. One way to do it is: 1. Express the equation as n2 + n = 2m2. 2. Multiply each side by 4: 4n2 + 4n = 8m2. 3. Add 1 to each side: 4n2 + 4n + 1 = 8m2 + 1. 4. Factor (2n + 1) (2n + 1) = 2(4m2 ) + 1. 5. Let y = (2n + 1) and x = 2m. 6. Substituting these terms in the preceding equation produces y2 = 2x 2+ 1. This is the simplest form of what is called the Pell equation, about which more below. If we can find an integral solution for it, we can easily work backward to find integral values for n and m in the original equation. A standard algorithm for cracking a Pell equation is to express the square root of the coefficient of x (in this case 2) as a continued fraction, then explore its
HEXES AND STARS
17
convergents for values of x and y that satisfy the Pell. The technique is too involved to explain here, but interested readers will find a good introduction to it in Chapter 22 of Albert H. Beiler's Recreations in the Theory ofNumbers or in any good textbook on Diophantine analysis. It turns out that whenever the coefficient is not a square, the Pell has an infinity of solutions. Since 2 is not a square, there is an infinity of square triangles. The sequence begins 1, 36, 1225, 41616, 1413721, . . . . The recursive procedure for continuing this sequence is to multiply the last square triangle by 34, subtract the preceding square triangle, then add 2. A nonrecursive formula for the nth square triangle is
The irrational numbers in the formula might lead one to suppose that rounding up or down is necessary, but this is not the case. The formula is exact. Substitute any positive integer for n, and the irrationals drop out to give an integral value for the expression. It is surprising how often the problem of finding this formula turns up in the problem departments of mathematical journals, even though it has been shown that the formula goes back to Euler. Square triangles have many unusual properties. One of the most surprising is that when a simple algorithm is applied, each square triangle gives the sides of an integral right triangle with one leg exactly one unit longer than the other. Let the sides of the right triangle be x and x 1 and z the hypotenuse. Let v be the square root of a square triangle and u its side when represented as a triangle. The procedure is merely to solve these two simultaneous equations:
+
For example, if we take the second square triangle, 36, then v = 6, u = 8. The above equations give x a value of 20 and x a value of 29. The Pythagorean triplet therefore is 20, 21, 29. Had we used the first square triangle, 1, the algorithm would have provided the familiar 3 , 4 , 5 right triangle. The third square triangle gives the triplet 119, 120, 169. In this way, all Pythagorean triangles with consecutive legs can be obtained from the square triangles, and of course we can go the other way and derive all the square triangles from consecutivelegged Pythagorean triangles. The simple procedure, or one equivalent to it, explains how Beiler was able to construct his table (pp. 328
18
CHAPTER TWO
and 329 in his book) of the first 100 Pythagorean triangles with consecutive legs. The 100th such triangle has legs that are each expressed by a seventysevendigit number. No Pythagorean triangle can have equal legs, but this monstrosity is so nearly isosceles that, as Beiler graphically points out, if its smaller leg is a lightyear long, the other leg would be longer by an amount so infinitesimal that the difference between the two legs would be millions of times less than the diameter of a proton. W e turn now to two ~ l a n a figurate r numbers that are not polygonal in the classic sense. The first has received scant attention in the past. The second, so far as I know, has not previously been recognized as a figurate number. If we arrange points as shown in Figure 3, we have what are known as centered hexagonal numbers to contrast them with the traditional vertexgenerated numbers. Let us call them "hexes" for short. As the illustration makes clear at a glance, the formula for the nth hex is 3n(n  1) 1. It is the sum of three rhombuses, each of sides n and (n  I), plus the single spot in the center. Figure 4 shows that a hex is also the sum of six triangles plus the central spot. The sequence begins 1,7,19,37,61,91,127,169,. . . . The recursive procedure is to multiply by 2, subtract the preceding number, and add 6. Suppose we build a hex pyramid of coins, starting with a hex that has 100 coins on the side. O n top of this, we put a hex of 99 coins on the side, then one of 98, and so on, until finally we cap the pyramid with a single coin on the
+
Figure 3 Formula for nth hex number is [3 X n(n  I)]
+1
HEXES AND STARS
Figure 4
19
Six triangles plus the center make a hex
central stack. The pyramid is 100 layers high. How many coins are in it? T o answer this, we need to know the formula for the sum of the first n hexes. The answer is unexpectedly simple. It is n3. There are therefore 1003= 1,000,000 coins in the hex pyramid. It follows from this formula that every hex is the difference between two consecutive cubes. W e can demonstrate this elegantly by building a cube, say a 5 X 5 X 5, out of unit cubes. Remove one cube (the first hex) from a top corner. It leaves a 1 X 1 X 1 hole. Around that hole are seven cubes (the second hex). Removing these seven leaves a 2 X 2 X 2 cubical hole. Surrounding this hole are nineteen cubes (the third hex). Removing the nineteen cubes leaves a 3 X 3 X 3 cubical hole, and so on (see Figure 5). Apart from a hex of I, the first triangular hex is 91, and the first square hex is 169. Readers who know the Pellian technique may enjoy searching for recursive procedures that will generate each of these infinite sequences of numbers and for their nonrecursive formulas. The Pellian for square hexes is 3x2 1 = yZ, which is solved by finding the convergents of the continued fraction for the square root of 3. The next square hex after 169 is 32761, and the next is 6355441. Are there hexes that are both square and triangular? Is there a cubical hex? Closely related to hexes are numbers that I have not seen recognized before as figurate, although one often sees them as patterns for drainage holes and for
+
20
CHAPTER TWO
Figure 5 Hexagonal numbers as the difference between consecutive cubes
holes on salt and pepper shakers. Game enthusiasts know them as the patterns of boards for playing Chinese checkers. Let us call them "star" numbers. Figure 6 shows the first four stars, and Figure 7 is a "looksee" proof of the formula for the nth star. Clearly, a star consists of six rhombuses, each n by (n  1) plus the central spot, or 6n(n  1) 1.A star is also the sum of twelve triangles plus the central spot, as is shown in Figure 8. The star sequence begins 1,13,37,73,121,181,253,337,433,541,. . . . Adding 12n to the nth star produces the next star. A hex contains six triangles. Adding six more triangles to its six sides produces a star; consequently any hex number becomes a star number if we double it and subtract 1. The first n stars add up to 2n3  n. Is this sum ever a square? Yes, but only when n = 1 or 169. This was established in 1973 by John Harris, on the basis ofresults reported by Louis J. Mordell on page 271 of his Diophantine Equations. The first triangular star after 1is 253. The recursive procedure is to multiply a triangular star by 194, add 60, and subtract the preceding triangular star. The infinite sequence begins 1, 253, 49141, 9533161, 1849384153,. . . . The nonrecursive formula for the nth triangular star is
+
HEXES AND STARS
Figure 6 The first four stars
Figure 7 Formula for nth star is [6 X n(n  I)]
+1
21
22
CHAPTER TWO
Figure 8 Twelve triangles plus the center make a star
The first square star after 1 is 121. It is the number of holes on the standard Chinese checkers board. The recursive procedure is to multiply a square star by 98, subtract the preceding square star, and add 24. The infinite sequence begins 1, 121, 11881, 1164241, 114083761,. . . . For readers who care to work through the solution of the equation for square stars, 6n(n  1) 1 = m2,I shall say only that this reduces to a solution of 2 x 2 1 = 3y2,where x is the square root of a square star. It can be solved by finding alternate convergents of the square root of 4. The extract nonrecursive formula for the nth square star is
+
+
One of the most remarkable properties of square stars is that they provide a simple algorithm for producing every number that can be expressed as the sum of two consecutive squares and also as the sum of three consecutive squares. The smallest such number is 365 (the number of days in the year), which equals 132 142and also equals l o 2 1 l 2 122.The procedure is simply to take any square star greater than 1, triple it, and add 2. The smallest square star greater than 1 is 121. Three times 121, plus 2, is 365. The next square star,
+
+
+
HEXES A N D STARS
+
23
11881, leads to the number 35645, which equals 1332 1342and also equals 10g2 1092 1102. The third case is 3(1164241) 2 = 3492725 = 13212 13222 = 10782 10792 10802. In each case, the middle term of the triplet of consecutive squares is the original square star. It is a pleasant exercise, demanding no special skills in number theory, to show that the algorithm always works. Can the reader find a simple proof? I have been told by Victor Meally that Matila Ghyka, a Romanian mathematician, published some studies of hexes, but I do not know the references. I was unable to find any discussion of stars as such, although their formula turns up in connection with many Diophantine problems. One can raise all kinds of questions about stars that may be easy or difficult to answer. I do not know, for example, if there are stars that are both square and triangular. Since a star's digital root is 1 or 4, and a triangle's digital root must be 1,3,6, or 9, we can say that a square triangular star must have a digital root of 1, but that is not of much help. The general Pell equation, a key to so much of this kind of number analysis, is ax2 1 = y2, where a is a positive integer. It has an infinity of positive integral solutions for x and y, unless a is a square, in which case there are no solutions. As we have seen, when a = 2, we have the key to finding square triangles, and when a = 3, the key to finding square hexes. The equation was mistakenly named for John Pell, a seventeenth century number theorist, because of a false impression on Euler's part. Pell had nothing to do with the equation. It was known to early Greeks and Hindus, but Pierre Fermat was the first to propose advanced work on it, and the general solution was obtained by John Wallis and others. The classic reference is The Pell Equation by E. E. whitford. The tables in this book can save vast amounts of tedious calculations with continued fractions. As J. A. Lindon has put it elegantly in one of his unpublished mathematical Clerihews:
+ + +
+
+
+
+
T o equations simultaneously Pellian My approach is Machiavellian. Anything goes, rather than resort to such actions As covering the walls with continued fractions. I wish to thank Meally for providing the nonrecursive formulas for square stars and triangular stars, as well as other things in this chapter, and also to thank John Harris and John McKay for additional assistance. I also found A Handbook ofInteger Sequences by N. J. A. Sloane to be an invaluable tool. I shall say no more about this marvelous reference, except that every recreational mathematician should buy a copy forthwith.
24
CHAPTER TWO
ANSWERS The problem was to prove that if a star number is multiplied by 3, and 2 is added, the result is a number that can be expressed as the sum of two consecutive squares and also as the sum of three consecutive squares. As explained, square stars are numbers of the form 6n(n  1) 1 = m2, where n and m are positive integers. If the left side, which defines a star, is multiplied by 3, and 2 is added, the result is 18n(n  1) 3 2 = 18n2 18n 5. The expression (3n  2)2. If the is equal to the sum of two consecutive squares: (3n right side of the equation, which defines a square, is multiplied by 3, and 2 is added, the result is 3m2 2. This equals the sum of three consecutive squares m2 (m (m 
+
+ +
+ + +
+
+
+
ADDENDUM I asked if a hex number could be both triangular and square. The answer is: only hex number 1. The proof was given by Charles M. Grinstead, a mathematician at Swarthmore College, in his paper "On a Method of Solving a Class of Diophantine Equations" (Mathematics of Computation 32, July 1978, pp. 936  940). I also asked if a hex could be a cube. David Chess, Sin Hitotumatu, and Wesley Johnson were the first to tell me how easily it can be shown that the answer is no. Every hex, as I explained, is the difference between two consecutive cubes. The question, therefore, is the same as asking if the formula (x I)3  x3 = y3 has an integral solution. When this is written x3 y3 = (x we see at once that it is a case of Fermat's last theorem when the exponents are 3. This was long ago proved to have no integral solution. Harvey J. Hindin, in theJournal $Recreational Mathematics, showed that the problem of determining all numbers that are both hexes and stars is equivalent to the task of determining when one triangular number is twice another, and also the same as finding Pythagorean triples x, y, z, such that y = x 1. A table in his article provides the first ten starhexes, an infinite sequence that begins 1, 37, 1261, 42841,. . . . Hindin has calculated the first 15,000 stars and hexes. The values are available from him at 5 Kinsella Street, Dix Hills, NY 11746. John Harris pointed out in a letter that every star has a digital root (its value modulo 9) of 1 or 4, and that the final pair of digits must be 01,21,41,61,81, 13,33,53,73,93, or 37. This rules out any star number containing each of the digits 1through 9just once (such a number would have a digital root of 9), and also any "repdigit" star consisting entirely of repetitions of the same digit.
+ +
+
+
HEXES AND STARS
25
Harris also showed that every pair of consecutive stars are relatively prime, and every prime divisor of a star is one more or one less than a multiple of 12. Several readers wrote to say that early unofficial flags of the United States had thirteen stars (for the original thirteen states) arranged in star formation. Note also that this pattern is on the green side of a dollar bill, just above the eagle.
B
I
B
L
I
O
G
R
A
P
H
Y
The Pel1 Equation. E. E. Whitford. Columbia University Press, 1912. Recreations in the Theory ofNumbers. Albert H. Beiler. Dover, 1964. Diophantine Equations. Louis J . Mordell. Academic Press, 1969. A Handbook of Integer Sequences. N. J . A. Sloane. Academic Press, 1973. "Diophantine Analysis and Fermat's Last Theorem." Martin Gardner, in Wheels, Life,and Other Mathematical Amusements. W . H. Freeman and Company, 1983. "Stars, Hexes, Triangular Numbers, and Pythagorean Triples." Harvey J. Hindin, in Journal ofRecreationa1 Mathematics 16, 1983 1984, pp. 191 193. "Figurate Numbers: The Other Alternative." Allan Whitcombe, in Mathematics in School (a publication of England's Mathematical Association), September 1986, pp. 40 42. The set of central figurate numbers is the "other alternative" to the vertex figurate numbers.
THREE
Tangrams, Part 1
"The seven Books of Tan . . . illustrate the creation of the world and the origin of species upon a plan which outDanvins Darwin, the progress of the human race being traced through seven stages . . . up to a mysterious spiritual state which is too lunatic for serious consideration."
SAM LOYD,The Eighth Book of Tan
One of the oldest branches of recreational mathematics has to do with dissection puzzles. Plane or solid figures are cut into various pieces, and the problem is to fit the pieces together to make the original figure or some other figure. The outstanding recreation of this type since the Renaissance is the Chinese puzzle game known as tangrams.
28
CHAPTER THREE
Although tangrams and jigsaw puzzles have a superficial resemblance, they are poles apart in the kinds of challenge they offer. As Ronald C. Read points out in his book Tangrams: 330 Puzzles, a typical jigsaw puzzle consists of hundreds of irregularly shaped pieces that fit together injust one way to make a large pattern. Little skill is required, just time and patience. Tangrams have only seven pieces, called tans. They are of the simplest possible shapes and are used to make an infinite variety of tangrams. In creating these figures a heavy demand is made on one's geometrical intuition and artistic ability. The tans are obtained by slicing a square to produce two large triangles, a middlesize triangle, two small triangles, a square, and a rhomboid (see Figure 9). Note that all the corners are multiples of 45 degrees. If a side of the square tan is taken as unity, a side of any tan has one of four lengths: 1,2, JZ,and 2JZ. "At first we are amazed at thk unfitness of the shapes. . . with which we are expected to accomplish so much," wrote ~ o i d the , American puzzle expert. "The number 7 is an obstinate prime which cannot be divided into symmetrical halves, and the geometrical forms . . . with harsh angles, pre
Figure 9 The seven tans
TANGRAMS, PART 1
29
clude the possibility of variety, curves or graceful lines." After working for a while with the tans, however, one begins to appreciate the subtle elegance of the dissection and the richness of its combinatorial possibilities. All kinds of variant dissections, in imitation of tangrams, have been marketed from time to time, but not one has come even close to tangrams in popularity. As with origami, it is the very simplicity of the material and its apparent unfitness for artistic use that lie at the heart of its charm. Tangram play falls roughly into three major categories: I.
2. 3.
Searching for one or more ways to form a given tangram, or for an elegant proof of the impossibility of forming a tangram. Finding ways to depict, with maximum artistry or humor, or both, silhouettes of animals, human figures, and other recognizable objects. Solving a variety of problems in combinatorial geometry that are posed by the seven tans.
Many books, and even a few encyclopedias, declare that tangram play is about 4000 years old. In my Scientijc American column for September 1959, I called tangrams the oldest of dissection games and said that the Chinese had been amusing themselves with it for several thousand years. This is totally wrong. The man responsible for the myth is none other than Sam Loyd. In 1903, when Loyd was sixtyone and at the height of his fame, he published a little book (now extremely rare) called The Eighth Book ofTan. No Western book on tangrams has been more original or influential. In addition to containing hundreds of excellent new figures, Loyd invented a preposterous legend about the pastime's origin. It was the greatest hoax in the history of puzzledom, and the number of intelligent people taken in by it rivals the number of scholars who accepted H. L. Mencken's spurious history of the bathtub. "According to the late Professor Challenor," Loyd wrote, "whose posthumous papers have come into the possession of the writer, seven books of Tangrams, containing one thousand designs each, are known to have been compiled in China over 4000 years ago. These books are so rare that Professor Challenor says that during a forty years' residence in China he only succeeded in seeing perfect editions of the first and seventh volumes, with stray fragments of the second. "In this connection it may be mentioned that portions of one of the books, printed in gold leaf upon parchment, were found in Pekin by an English soldier who sold it for £300 to a collector of Chinese antiquities, who kindly furnished some of the choicest designs presented in this work."
30
CHAPTER THREE
According to Loyd, Tan was a legendary Chinese writer who was worshiped as a deity. He arranged the patterns in his seven books to display seven stages in the evolution of the earth. His tangrams begin with symbolic representations of chaos and the yinyang principle. These are followed by primitive forms of life, then the figures proceed up the evolutionary tree through fish, birds, and animals to the human race. Scattered along the way are tangrams of human artifacts such as tools, furniture, clothing, and architecture. Loyd inserts remarks by Confucius, a philosopher called Choofootze, a commentator named Li Hung Chang, and his mythical Professor Challenor. Chang is quoted as saying that he knew all the figures in the seven books of Tan before he could talk. And there are references to a "wellknown" Chinese saying about "the fool who would write the eighth book of Tan." All of this, of course, was sheer fabrication. When Henry Ernest Dudeney, Loyd's British counterpart, wrote an article on tangrams for The Strand Magazine (November 1908), he soberly repeated Loyd's legendary history. This aroused the curiosity of Sir James Murray, the distinguished lexicographer and an editor of the Oxford English Dictionary, who made inquiries through one of his sons, then teaching at a Chinese university. Oriental scholars had never heard of Tan or even the word tangrams. The game, Murray informed Dudeney, is known in China as ch'i ch'iao t'u, meaning "seveningenious plan" or, less literally, "clever puzzle of seven pieces." Murray could find no record of the word tangram earlier than in an 1864 Webster's dictionary. It had been coined about 1850, Murray guessed, by an American who probably combined tang, a Cantonese word for "Chinese," with the familiar suffix gram, as in anagram or cryptogram. A different theory about the name has recently been advanced by Peter Van Note in his introduction to a Dover reprint of Loyd7sfanciful book. Chinese families who live on riverboats are called tanka, and tan is a Chinese word for prostitute. American sailors, taught the puzzle by tanka girls, may have called it tangrams the puzzle of the prostitutes. When Dudeney reported Murray's opinions, in Amusements in Mathematics (pp. 43  46), he may have deliberately added a hoax of his own. An American correspondent, Dudeney writes, had told him that he owned a Chinese set of motherofpearl tans with an accompanying ricepaper booklet of more than 300 figures. The correspondent was puzzled by a mysterious inscription on the front page that he said he had tried to have translated, but no Chinese to whom he had shown it had been willing or able to read it. Dudeney reproduced the inscription and asked the reader for help. W e do not know what the response was to this request, but Read, who owns a copy of the same booklet, had no difficulty clearing up the mystery. The, inscription is nothing more
TANGRAMS, PART 1
31
than a caption under the tangrams of two men. The caption reads: "Two men facing each other and drinking. This shows the versatility of the sevenpiece puzzle." N o one knows when tangrams originated. The earliest reference known is a book published in China in 1803. Its title, The Collected Volume ofPatterns of the SevenPiece Puzzle, suggests earlier books. Most scholars believe the game originated in China about 1800, became an Oriental craze, and then spread rapidly to the West. The earliest Western books, says Read, were little more than copies of Chinese ricepaper booklets. The Western books even copied errors in the Chinese illustrations. One of the earliest English books on tangrams, originally owned by Charles Lutwidge Dodgson (better known as Lewis Carroll), came into Dudeney's possession. It is called The Fashionable ChinesePuzzle and was first published in New York in 1817. Dudeney quotes from it a passage stating that the game was a favorite of "exEmperor Napoleon, who, being now in a debilitated state, and living very retired, passes many hours a day in thus exercising his patience and ingenuity." This too is an unsupported statement, undoubtedly false. The puzzle is said by Loyd to have been a favorite of John Quincy Adams's and Gustave Dort's, although I know of no basis for either assertion. W e do know that Edgar Allan Poe enjoyed the game, because his imported set of carved ivory tans is owned by the New York Public Library. An anonymous French work, Recueil des PlusJolie Jeux de Socie'tt ( 18 18), may be a translation of Dodgson's English book, or vice versa. I have not seen a copy of either. An 1817 American book bears the title Chinese Philosophical and Mathematical Trangram. "Trangam" was an old English word for a trinket, toy, or puzzle. Samuel Johnson misspelled it as "trangram," and the spelling persisted in later dictionaries. Did the book's anonymous author revive an obsolete word that later evolved to "tangram," or did he misspell "tangram," a word already in use? One mystery novel, The Chinese Nail Murders by the Dutch diplomat and Orientalist Robert Van Gulik, is woven around a set of tangram patterns. Poe's tans are shown in Figure 10. The delicate filigree carving is characteristic of the old Chinese ivory tans. Note that the pieces pack into a square box in two layers. The two layers are squares of equal size, so that putting away the tans is a puzzle in itself. In nineteenthcentury China, where tangrams were popular among adults (it is considered a child's pastime in the Far East today), the pieces were made in many sizes and from many different materials. Dishes, lacquer boxes, and even small tables were given the shapes of the tans. So much for the historical background. Let us turn now to the first of the three categories of tangram play: solving given figures. Figure 11 shows a dozen interesting shapes on which the reader is invited to try his skill. Each
32
CHAPTER THREE
Figure 10 Edgar Allan Poe's carved ivory tangram set
requires all seven pieces. The rhomboid, the only asymmetrical tan, may be placed either side up. One figure in the illustration is not possible. Can the reader identify it and prove its impossibility? The paired tangrams in Figure 12 are samples of delightful paradoxes introduced by Loyd. (The first three pairs were devised by Loyd, the fourth pair was devised by Dudeney.) Although the figure at the right in each case seems to be exactly the same as its mate, except for a missing portion, each is made with all seven tans! The tangrams in Figure 13 are not intended as patterns to be solved, but as illustrations of the second category of play: creating artistic and amusing pictures. (I confess responsibility for the Nixon caricature.) "One remarkable thing about . . . Tangram pictures," wrote Dudeney, "is that they suggest
TANGRAMS, PART 1
Figure 11 Which tangram is impossible?
33
34
CHAPTER THREE
Figure 12 Tangram paradoxes
to the imagination such a lot that is not really there. Who, for example, can look . . . at Lady Belinda . . . without soon feeling the haughty expression . . . ? Then look again at the stork, and see how it is suggested to the mind that the leg is actually much more slender than any one of the pieces employed. It is really an optical illusion. Again, notice in the case of the yacht
TANGRAMS, PART 1
35
how, by leaving that little angular point at the top, a complete mast is suggested. If you place your tangrams together on white paper so that they do not quite touch one another, in some cases the effect is improved by the white lines; in other cases it is almost destroyed." One can mix two or more sets of tans to produce more elaborate figures. Dudeney gives a number of these "double tangrams" in 536 Puzzles and Curious Problems (pp. 221 222), and others will be found in Read's book. I agree with Read, however, when he writes: "With fourteen pieces to play around with, one cannot help but feel that it should be possible to arrive at a reasonable likeness of just about anything. Consequently, the sense of achievement that one gets on producing a recognizable cow, sailing boat, human figure, or what have you, from a mere seven pieces, is quite lacking." Combining two related tangrams, each made with seven tans, is a different matter. Four classic examples, all devised by Loyd, are a woman pushing a baby carriage, a runner being tagged out at home plate, two Indian braves, and a man with a wheelbarrow (see Figure 14). Note that the man and the wheelbarrow are identical tangrams except for orientation. The third category of tangram play, solving combinatorial problems, is the most interesting of all to mathematicians. There have been some remarkable
LADY BELINDA
RABBIT
~ T O R K
RUNNING BOY
Figure 13 Tangram pictures
YACHT
VULTURE
POLAR BEAR
PRESIDENT NIXON
36
CHAPTER THREE
Figure 14 Double tangrams by Sam Loyd
contributions made to this field by Read, a specialist in graph theory at the University of Waterloo, and by E. S. Deutsch, a computer scientist with P. S. Ross and Partners in Toronto. Some of their results will be presented in the next chapter. T o whet the reader's appetite, here are two problems that will be answered in the next chapter. 1. How many different convex polygons can be formed with the seven tans? There must not be any "windows" in the figures. Rotations and reflections are not, as is customary, considered to be different. Because all threesided polygons are convex, and no nonconvex polygon of four sides can be made with all seven tans, answering this question also gives the number of three and foursided polygons. It is easy to see that only one triangle is possible (since corners must be multiples of 45 degrees, the triangle must be a right isosceles one), but finding all the higher convex polygons is a bit tricky. 2. How many different fivesided polygons can be made?
TANGRAMS, PART 1
37
ANSWERS The impossible tangram in Figure 11 is the square with the central square hole. The two large triangles can go only in opposite corners. The square tan must go in a third corner and the rhomboid must touch the fourth corner, but now there is no spot for the middlesize triangle.
B
I
B
L
I
O
G
R
A
P
H
The Bibliography following Chapter 4 (p. 54) is also for this chapter.
Y
FOUR
Tangrams, Part 2
"The formation of designs by means of seven pieces of wood . . . known as tans . . . is one of the oldest amusements in the East. Many hundreds of figures representing men, women, birds, beasts, fish, houses, boats, domestic objects, designs, etc., can be made, but the recreation is not mathematical, and I reluctantly content myself with a bare mention of it."
W. W . ROUSEBALL,Mathematical Recreations and Essays

Not mathematical? Ball wrote without giving the matter much thought. In this chapter we consider some nontrivial combinatorial problems presented by the seven tans. A question that arises at once is: How many sides can a tangram formed with all seven tans have? Although the answer is obvious, it seems to have been first
40
CHAPTER FOUR
Figure 15 Improper tangram (left) and proper tangram (right), each with twentythree sides
stated by Harry Lindgren in his 1968 article "Tangrams." This is how Ronald C. Read, a mathematician at the University of Waterloo, puts it in a lengthy communication to me from which I shall be quoting liberally: "Tangrams have 23 sides between them, so that a tangram like that (at the left in Figure 15) will clearly have this number of sides. O n the other hand, tangrams with pieces joined only at a corner are mathematically uninteresting . . . . " Let us make a rule, Read proposes, that a tangram must have a perimeter topologically equivalent to a circle, that is, it must not selfintersect. Read calls these "proper tangrams." How many sides can a proper tangram have? Again the answer is twentythree. The proof is supplied by the figure of a bowing man shown at the right in Figure 15 and by almost endless other examples. The proper tangrams contain an important subset that Read calls "snug tangrams." In order to understand the meaning of snug, draw lines on all tans (except the two small triangles) to create sixteen identical rightisosceles triangles with unit legs (see Figure 16). A snug tangram is a proper tangram formed so that, where two tans are in contact, the sides of the small right triangles match exactly, either leg to leg or hypotenuse to hypotenuse. All convex tangrams are snug, and so are many of the traditional figures (see Figure 17). Snugness, by the way, is characteristic of technology in the Orient, where the dimensions of houses, furniture, and so on tend to be exact multiples of a basic length. The Japanese building industry, I am told, is one of the most efficient in the world because Japanese lumber is standardized in lengths that are multiples of a basic "mat" length. In addition to snugness of fit, Read adds two more limitations: A snug tangram must be simply connected (all in one piece), and there must be no
TANGRAMS, PART 2
41
Figure 16 T h e seven tans
interior holes, including holes that touch the perimeter at one or more single points. It is convenient to diagram snug tangrams on graph paper so that all integral edges are on the orthogonals of the matrix. All diagonal edges will then be multiples of and therefore irrational. This suggested to Read a very pretty problem: How many snug tangrams have all sides irrational? Such tangrams, if diagrammed on graph paper, would have every side running diagonally. The tans between them have a total of thirty side segments, Read continues, but "whenever we place two pieces together, the two sides that abut are lost to the perimeter, and it can happen that we lose more than two. Furthermore, in order that the resulting tangram shall be connected, there must be at least six lines along which two pieces come together. Hence we cannot avoid losing 12 segments, therefore the total number of segments on the outside cannot be
Figure 17 A snug tangram of a dog with eighteen sides
42
CHAPTER FOUR
more than 18. Each side of a snug tangram consists of at least one segment, hcncc the total number of sides cannot cxcccd 18 either." 'l'hc dog in Figure 17 provesthat a snug tangram can have this maximum total of sides. 'l'he number of proper tangrams obviousljr is infinite. You have only to observe that two tans can abut in an infinite number of positions. If the question is confined to certain lrinds of tangram, however, interesting problems in combinatorial enumeration arise. For example, how many convex tangrams are there? d convex tangram is a polygon in which all corner angles are less than 180 degrees. That there are only thirteen was proved in 1942 by Fu Tsiang Wang and ChuanChih Hsiung in '54 Theorem on the Tangram" in The America~zhlathematicalhlontbdy.. The thirteen are shown in Figure 18. If mirror images count as different convex tangrams, then there are eighteen. The eighteen convex tangrams appear in Chinese tangram books with solutions showing that all can be made without turning over the asymmetric rhomboid tan. (Interior lines are omitted in the illustration, in case some readers might enjoy solving them.) The thirteen convex tangrams include all three and foursided polygons that can be made with the seven pieces. As stated in the previous chapter, nonconvex quadrilaterals are not possible. (Can you prove this? Hint: The four interior angles of such a foursided figure would have to be three angles of 45 degrees and one of 225 degrees, and the figure would have to consist of sixteen isosceles right triangles congruent with the small triangular tan.) Fivesided polygons, made with the tans, can be nonconvex. How many pentagons, both convex and nonconvex, 1,indgren asked, can be made with the scvcn tans? ' i t this spot in my 1974 column, T gave a proof that there are sixteen snug pentagons and two "loose" pentagons, making eighteen in all. Unfortunately, there was a flaw in the proof, which readers quicliljr Qscovered. For months I was flooded with hundreds of letters from readers of all ages who had found pentagons not among the eighteen that I had given in an illustration. I kept tab of the number of Qfferent pentagons sent by readers until it reached a maximum of twentytwo snug and thirtyone loose ones, or fiftythree in all. Only two readers, each working independently and by hand, found all fiftythree. They are hllan L. Sluizer of Northbrook, Illinois, and Ake Lindgren of Uppsala, Sweden. I n 1975 Dr. I. Takeuchi at the Institute for Electrical Engineers, Musashimo, near Tokyo, confirmed the fiftythree figures by a computer program. Tn 1976, unaware of Takeuchi's program, Michael Beeler of Cambridge, Massachusetts, wrote a program that gave the same results. Beeler's drawings of the fiftythree pentagons are shown in Figure 19.
TANGRAMS, PART 2
43
Figure 18 T h e thirteen convex tangrams
One might now wonder how many tangram polygons have six sides, or seven or more, but (as Read points out) the question is easily answered. For n = 6 through 23 there is an infinity of nsided polygons. You have only to glance at pentagon 28 in Figure 19 to see that by sliding the large triangle on the left along the hypotenuse of the other large triangle, you can create an infinity of hexagons. How many snug hexagons are there? Although it is a finite number, as far as I know the number has not yet been determined.
46
CHAPTER FOUR
There are, of course, only a finite number of snug tangrams, but the exact number (Read calls it the snug number) also is far from known. Read has devised an ingenious procedure by which a computer could be programmed to count the number, but he estimates that it is well into the millions, and no such program has yet been written. Unfortunately, the details of Read's procedure are too complex to give here. A simpler problem, using exactly the same procedure, has been solved, however. Read defines a minitangram as one made with the five pieces that remain after leaving out the two large triangles. The problem of finding all snug minitangrams is so much simpler than finding all snug tangrams that Read was able to write a computer program to solve it and ran it on, appropriately enough, a minicomputer at the University of Waterloo. It took only half an hour, and the count was 951. The computer was hooked up to a display terminal, so that it drew pictures of all the minitangrams. Read's programs are designed only for enumeration, not for solving individual tangrams. Is it possible to write a program that will inspect any given tangram and search for at least one solution? Yes, such a program has been developed and published by E. S. Deutsch, a computer scientist. In theory it is possible to write a program that will systematically examine all possible ways the tans fit a given tangram and then print all the solutions, but the complexity of such a program is so great that no one has yet attempted it. Deutsch's program is not of this type. It is heuristic, which means that it goes about solving a tangram in much the same way a person does: by applying a series of tentative tests, examining the feedback, backtracking and trying again when no solution is found, and continuing until it either discovers a solution or gives up. The program seldom fails, taking an average time of about two seconds to solve a tangram. The program starts by examining the tangram's perimeter, noting the edge lengths, and the angles at each corner. It then attempts to separate the tangram into two or more subtangrams. For instance, if two portions of the tangram meet at a point, each portion clearly is a separate tangram. If, say, a rabbit has two ears formed by the two small triangles, each ear meeting the head at a point, the program instantly identifies the two pieces, removes them and goes to work on the subtangram that remains. Ifthe tangram does not have portions meeting at points, the program explores ways of dividing it into subtangrams by extending edges from a corner into the figure. In many cases the internal extension of an edge clearly divides the tangram into subtangrams; in other cases the extension is only a possible line of division. After the program's preliminary exploration, it applies a series of heuristic tests until it finds a way of fitting tans into a subtangram or a possible subtan
TANGRAMS, PART 2
47
gram. When a fit is found, the subtangram is extracted and the program turns to what remains. The tests are ranked in order of their efficacy, so that the strongest tests can be applied first, then the next strongest, and so on. If no solution is obtained, the program backtracks and begins with the second test. It is impossible to describe the program in more detail, but interested readers will find it fully explained, with flow charts and examples, in "A Heuristic Solution to the Tangram Puzzle," by E. S. Deutsch and Kenneth C. Hayes, Jr., in Machine Intelligence 7. A somewhat similar program was developed in 1972 by Ejvind Lynning, a Danish student working with Jacques Cohen, a physicist at Brandeis University. Snug tangrams are, for obvious reasons, usually more difficult to solve (by a person or a computer) than nonsnug figures, and the difficulty tends to increase as the number of sides decreases. One might suppose a pattern with only one solution would be harder to solve than one with many, but that is not the case. A pattern in which the tans touch only at points has only one solution, but it is an immediately obvious one, and there are patterns with a large number of solutions that are among the most difficult. The construction of tangrams with "holes" raises many curious new problems. It is not hard to form a square hole of area 4 or a triangular hole of area 2 that does not touch the border, or two triangular holes of areas 1 and 3that do not touch each other or the border (see Figure 20); ("touch" includes touching at a single point). Can the reader find a way to make exactly two holes, each 1 X 1 square, that do not touch each other or the perimeter? O r two holes, under the same provisos, one a triangle and one a square, each with an area of 17 They are not difficult tasks, but here are two more that I set for myself and found much harder: (1) Form just three holes, two triangular and one square, that do not touch one another or the border. (2) Form just three holes, two rectangular and one triangular, that do not touch one another or the border. Apparently it is not possible for three holes of this type to be all rectangular or all triangular, or for two triangular holes each to have an area of 1. The "farm problem" is another unsolved hole problem. What is the largest hole not touching the border that can be inside a tangram? The solution is a limit that cannot be reached, but one can come as close to it as one wishes. (The best I can obtain is the limit 10.985+ .) How many sides can a single hole that is simply connected and not touching the border have? The maximum surely is thirteen. What is the largest "farm" not touching the border that is square? Rectangular? Triangular? Another unexplored type of tangram problem is finding ways of transforming one tangram into another with the fewest number of moves. A move
48
CHAPTER FOUR
Figure 20 Tangrams with holes
consists in altering the position of a set of one or more tans without disturbing the pattern of the set. For example, the large square tangram can be changed to the large triangle or the large rhomboid in one move, or to the 2 X 4 rectangle in three moves. As Read notes in his book, the square can be changed to a 3 X 3 square with a missing 1 X 1 corner in just two moves. Still another area of tangram play open to exploration is the devising of competitive games that use one or more sets of tans. The only tangram game I have seen in books inEnglish is the party game of giving each guest a set of tans and awarding prizes to those who are the first to make each of a series of displayed patterns. Read's concept of snugness suggests a variety of twoperson games. Here are three that occurred to me. In playing them it is a good plan to mark the middle of the long edges to facilitate placing the tans snugly. 1. Snuggle up. Begin with the tans forming the large triangle, the square, or any foursided ~ o l ~ g oPlayers n. alternate moves. A move consists in changing the position of a single tan to form a new snug tangram that has more sides than the preceding one. The first player who is unable to make a move loses. 2. Snuggle down. Same as above, except that the initial tangram is an eighteensided snug tangram and each move must decrease the number of sides. As in the preceding game, you cannot move a piece that leaves or forms a hole or that divides the figure into parts that touch only at points. Both games end quickly. Because a snug tangram must have at least three sides and no more than eighteen, a game cannot last beyond fifteen moves. Games that go a full fifteen moves are possible. 3. Snuggle up and down. Start with a ten or elevensided snug tangram. One player must increase the number of sides on each play, the other must decrease the number. The same piece may not be moved twice in succession. Each player keeps a record of his increases and decreases, and the first to score 30 wins. If a player cannot move, he loses. If the up
TANGRAMS, PART 2
49
player makes an eighteensided tangram, he wins. If the down player makes a four or threesided tangram, he wins. This game lasts considerably longer than the other two, and it often takes unexpected turns. One player can get far ahead in scoring only to discover, just before he expects to make a winning move, that no move is available. In all three games it is good to keep a running record of the number of sides because it is easy to forget the number, and time is lost by repeated counting. A cribbage board is a convenient device not only for recording the number of sides but also for keeping score in the upanddown game.
ANSWERS Solutions to the four hole problems are shown in Figure 21. At the top left in the illustration is a way of making two unit square holes. At the top right are two holes, one a unit square and the other a triangle of area I. At the left on the bottom of the illustration is a way of making a square hole and two triangular
Figure 21 Solutions to hole problems
50
CHAPTER FOUR
holes. (The fit is extremely close. The top triangle's hypotenuse is longer than the length it must span by only .I21 .) At the right on the bottom are two rectangular holes and one triangular hole. Read has proved that a tangram that is snug except for one or more holes cannot have more than one hole if the holes do not touch one another or the border. The smallest possible hole is equal to a small triangular tan. No matter how two such holes are placed so that they do not touch, at least seventeen triangles are required to isolate them from a tangram border. Since the seven tans are made of sixteen such triangles, completing the required tangram is impossible. For nonsnug tangrams it appears that no more than three holes not touching one another or the border are possible. The square is the only snug tangram with all its sides irrational. This is how Read proves it. As I explained in the previous chapter, an irrational tangram drawn on graph paper with the unit square tan oriented orthogonally would have all its edges making 45degree angles with the matrix lines. All corners clearly must be either 90 degrees or 270 degrees. Since each side is a multiple of & and the total area is 8, it follows that any irrational snug tangram will be a tetromino composed of four squares, each fi on the side. There are five tetrominoes. One of them, the square, we know can be formed. Each of the other four is easily proved impossible by placing the square tan in each of three possible positions (see Figure 22) and then exploring ways of completing the tetromino. The first tetromino is ruled out at once because there is no way to place the two large triangles. In the other three cases, for each position of the square tan, there are at most only four ways to place the two large triangles. In every case, after the square and two large triangles are placed, there is no spot for the rhomboid. Thus the square is the only possible irrational snug tangram. My solution of the farm problem, with a limit of 10.985 , is shown in Figure 23.
+
+
Figure 22 Impossibility proof for the nonsquare tetrominoes
TANGRAMS, PART 2 • 51
Figure 23 A solution to the farm problem
ADDENDUM Scores, perhaps hundreds, of puzzles similar to tangrams, but with different dissections of squares, rectangles, circles, and other shapes, have been described in books and articles and manufactured around the world. Some of the marketed variants are pictured in Professor Hoffmann’s pioneer book on mechanical puzzles, in Creative Puzzles of the World by Pieter Van Delft and Jack Botermans, and in Puzzles Old and New by Jerry Slocum and Botermans. Of special interest, because it predates any known Chinese publication, is a 32page book published by Kyoto Chobo in Japan in 1742. It gives fortytwo patterns to be made with the seven pieces obtained by dissecting a square as shown in Figure 24. The book’s title translates as The Ingenious Pieces of Sei Shonagon. (Sei Shonagon was a court lady of the late tenth and early eleventh centuries, who wrote the famous Pillow Book.) Nothing is known about the author, who uses the pseudonym GanreiKen. It is highly unlikely that Sei Shonagon knew of the puzzle. Shigeo Takagi, a Tokyo magician, was kind enough to send me a photocopy of this rare book. Unlike the Chinese tans, the Shonagon pieces will form a square in two different ways. Can you find the second pattern? The pieces also will make a square with a central square hole in the same orientation. With the Chinese tans it is not possible to put a square hole anywhere inside a large square. Richard Reiss, a professor of English at Southeastern Massachusetts University, sent a good proof that no foursided nonconvex polygon can be made with all
52
CHAPTER FOUR
Figure 24 T h e Sei Shonagon pieces
seven Chinese tans. Peter Van Note proposed the following three tasks based on forming two congruent replicas of a tan:
It is possible to make one large square. Use the seven pieces to make two congruent small squares. 2. A large isosceles right triangle is possible. Use the seven pieces to make two congruent isosceles right triangles. A large rhombus is possible. Van Note could not prove it, but he is 3. convinced that the seven pieces cannot make two congruent rhombuses.
I.
John H. Conway wrote from Cambridge University to pose an interesting unsolved problem. What would be the shapes of an "optimal" set of tansthat is, seven convex polygons that will form the largest number of distinct convex polygons? Figure 25 is taken from Joost Elffers's and Michael Schuyt's marvelous book. The pattern is made with the traditional Chinese tans. Karl Fulves, author of many books on magic, made the following suggestion for an amusing trick. It involves the tangram paradox at the bottom of
TANGRAMS, PART 2
53
Figure 12. You secretly add to a set of tans a third small triangle. Make the man with the feet, using the pattern shown on the right, and the extra triangle for the feet. Now use sleight of hand to "vanish" one of the small triangles (orjust put it in your pocket), then form the man with the foot again, using the pattern on the left. If no one has bothered to count pieces, it seems as if the vanished piece has mysteriously returned. Similar tricks can, of course, be performed with other paradoxical pairs. Several proposals have been made for using two sets of tangrams to play a board game similar to Solomon W. Golomb's pentomino game. Game inventor Sidney Sackson recommends a 6 X 6 checkerboard, its squares the size of the square tan. Each of two players has a set of seven tans. Players take turns placing any tan on the board, wherever they wish, provided the tan's corners fall on the board's lattice points. The person unable to place a tan loses. Many variations in rules are ~ossible,and larger boards can be used for more than two players.
Figure 25
A tangram to solve
54
B
CHAPTER FOUR
I
B
L
I
O
G
R
A
P
H
Y
Puzzles Old and New. "Professor Hoffmann" (pseudonym of Angelo John Lewis). F. Warne, 1893. Amusements in Mathematics. Henry Ernest Dudeney. Nelson, 1917. The Tangram Book. F. G. Hartswick. Simon and Schuster, 1925. "A Theorem on the Tangram." Fu Tsiang Wang and ChuanChih Hsiung, in American Mathematical Monthly 49, November 1942, pp. 596599.
The Chinese Nail Murders. Robert Van Gulik. Harper & Row, 1961. Tangrams: 330 Puzzles. Ronald C. Read. Dover, 1965. Tangram Teasers. R. C. Bell. NewcastleuponTyne, England (privately published), 1965. Tangrams. Peter Van Note. Charles E. Tuttle, 1966. 536 Puzzles and Curious Problems. Henry Ernest Dudeney. Scribner's, 1967
The Eighth Book of Tan. Sam Loyd. Dover, 1968 (reprint of 1903 edition). "Tangrams." Harry Lindgren, in Journal of Recreational Mathematics 1, July 1968, pp. 184 192.
Tangramath. Dale Seymour. Creative Publications, 1971. A Tangram Tale. William Cameron. Brockhampton Press, 1972. "A Heuristic Solution to the Tangram Puzzle." E. S. Deutsch and Kenneth C. Hayes, Jr., in Machine Intelligence 7. Bernard Meltzer and Donald Michie, eds. Wiley, 1972.
The Fun with Tangrams Kit. Susan Johnson. Dover, 1977 Creative Puzzles ofthe World. Pieter Van Delft and Jack Botermans. Abrams, 1978. Tangrams. Joost Elffers and Michael Schuyt. Abrams, 1979. Puzzles Old and New. Jerry Slocum and Jack Botermans. Plenary (Holland), 1986, distributed in the United States by the University of Washington Press, Seattle.
FIVE
Nontransitive Paradoxes
"I have just so much logic, as to be able to see . . . that for me to be too good for you, and for you to be too good for me, cannot be true at once, both ways." ELIZABETHBARRETT, in a letter to Robert Browning.
Whenever a relation R that applies to xRy and yRz also applies to xRz, the relation is said to be transitive. For example, "less than" is transitive among all real numbers. If 2 is less than 71, and the square root of 3 is less than 2, we can be certain that the square root of 3 is less than n. Equality also is transitive: ifa = b and b = c, then a = c. In everyday life such relations as "earlier than," "heavier than," "taller than," "inside of," and hundreds of others are transitive. It is easy to think of relations that are not transitive. IfA is the father of B and B is the father of C, it is never true that A is the father of C. If A loves B and B loves C, it does not follow that A loves C. Familiar games abound in transitive
56
CHAPTER FIVE
rules (if ~ o k ehand r A beats B and B beats C, then A beats C), but some games have nontransitive (or intransitive) rules. Consider the children's game in which, on the count of three, one either makes a fist to symbolize "rock," extends two fingers for "scissors," or all fingers for "paper." ~ o c breaks k scissors, scissors cut paper, and paper covers rock. In this game the winning relation is nontransitive. Occasionally in mathematics, particularly in probability theory and decision theory, one comes on a relation that one expects to be transitive but that actually is not. If the nontransitivity is so counterintuitive as to boggle the mind, we have what is called a nontransitive paradox. The oldest and bestknown paradox of this type is a voting paradox sometimes called the Arrow paradox after Kenneth J. Arrow because of its crucial role in Arrow's "impossibility theorem," for which he shared a Nobel prize in economics in 1972. In Social Choice and ~ndividualValues,Arrow specified five conditions that almost everyone agrees are essential for any democracy in which social decisions are based on individual preferences expressed by voting. Arrow proved that the five conditions are logically inconsistent. It is not possible to devise a voting system that will not, in certain instances, violate at least one of the five essential conditions. In short, a perfect democratic voting system is in principle impossible. As Paul A. Samuelson has put it: "The search of the great minds of recorded t it turns out, is the search for a chimera, for a history for the ~ e r f e cdemocracy, logical selfcontradiction . . . . Now scholars all over the worldin mathematics, politics, philosophy, and economics are trying to salvage what can be salvaged from Arrow's devastating discovery that is to mathematical politics what Kurt Godel's 1931 impossibilityofprovingconsistency theorem is to mathematical logic." Let us approach the voting paradox by first considering a fundamental defect of our present system for electing officials. It frequently puts in office a man who is cordially disliked by a majority of voters but who has an enthusiastic minority following. Suppose 40 percent of the voters are enthusiastic supporters of candidate A. The opposition is split between 30 percent for B and 30 percent for C. A is elected even though 60 percent of the voters dislike him. One popular suggestion for avoiding such consequences of the split vote is to allow voters to rank all candidates in their order of preference. Unfortunately, this too can produce irrational decisions. The matrix in Figure 26 (left) displays the notorious voting paradox in its simplest form. The top row shows that a third of the voters prefer candidates A, B, and C in the order ABC. The middle row shows that another third rank them BCA, and the bottom row shows that the remaining third rank them CAB. Examine the matrix carefully
NONTRANSITIVE PARADOXES
1
RANK ORDER 2
57
3
Figure 26 T h e voting paradox (left) and the tournament paradox based on the magic square (right)
and you will find that when candidates are ranked in pairs, nontransitivity rears its head. Twothirds of the voters prefer A to B, twothirds prefer B to C, and twothirds prefer C to A. If A ran against B, A would win. If B ran against C, B would win. If C ran against A, C would win. Substitute proposals for men and you see how easily a party in power can rig a decision simply by its choice of which paired proposals to put up first for a vote. The paradox was recognized by the Marquis de Condorcet and others in the late eighteenth century, and is known in France as the Condorcet effect. Lewis Carroll, who wrote several pamphlets on voting, rediscovered it. Most of the early advocates of proportional representation were totally unaware of this Achilles' heel; indeed, the paradox was not fully recognized by political theorists until the mid1940s, when Duncan Black, a Welsh economist, rediscovered it in connection with his monumental work on committee decision making. Today the experts are nowhere near agreement on which of Arrow's five conditions should be abandoned in the search for the best voting system. One surprising way out, recommended by many decision theorists, is that when a deadlock arises, a "dictator" is chosen by lot to break it. Something close to this solution actually obtains in certain democracies, England for instance, where a constitutional monarch (selected by chance in the sense that lineage guarantees no special biases) has a carefully limited power to break deadlocks under certain extreme conditions. The voting paradox can arise in any situation in which a decision must be made between two alternatives from a set of three or more. Suppose that A, B, and C are three men who have simultaneously proposed marriage to a girl. The rows of the matrix for the voting paradox can be used to show how she
58
CHAPTER FIVE
ranks them with respect to whatever three traits she considers most important, say intelligence, physical attractiveness, and income. Taken by pairs, the poor girl finds that she prefers A to B, B to C, and C to A. It is easy to see how similar conflicts can arise with respect to one's choice of a job, where to spend a vacation, and so on. Paul R. Halmos once suggested a delightful interpretation of the matrix. Let A, B, and C stand for apple pie, blueberry pie, and cherry pie. A certain restaurant offers only two of them at any given meal. The rows show how a customer ranks the pies with respect to three properties, say taste, freshness, and size of slice. It is perfectly rational, says Halmos, for the customer to prefer apple pie to blueberry, blueberry to cherry, and cherry to apple. In his Adventures of a Mathematician (Scribner, 1976), Stanislaw Ulam speaks of having discovered the nontransitivity of such preferences when he was eight or nine, and of later realizing that it prevented one from ranking great mathematicians in a linear order of relative merit. Experts differ on how often nontransitive orderings such as this one arise in daily life, but some recent studies in psychology and economics indicate that they are commoner than one might suppose. There are even reports of experiments with rats showing that under certain conditions the painvise choices of individual rats are nontransitive. (See Warren S. McCulloch, "A Heterarchy of Values Determined by the Topology of Nervous Nets," Bulletin ofMathematical Biophysics 7, 1945, pp. 89 93.) Similar paradoxes arise in roundrobin tournaments between teams. Assume that nine tennis players are ranked in ability by the numbers 1 through 9, with the best player given the number 9 and the worst the number 1. The matrix in Figure 26 (right) is the familiar order3 magic square. Let rows A, B, and C indicate how the nine players are divided into three teams with each row comprising a team. In roundrobin tournaments between teams, where each member of one team plays once against each member of the others, assume that the stronger player always wins. It turns out that team A defeats B, B defeats C, and C defeats A, in each case by five games to four. It is impossible to say which team is the strongest. The same nontransitivity holds if columns D, E, and F of the matrix are the teams. Many paradoxes of this type were jointly investigated by Leo Moser and J. W. Moon. Some of the MoserMoon paradoxes underlie striking and littleknown sucker bets. For example, let each row (or each column) of an order3 magicsquare matrix be a set of playing cards, say the ace, 6, and 8 of hearts for set A, the 3 , 5 , and 7 of spades for set B, and the 2,4, and 9 of clubs for C (see Figure 27). ~ a c set h is randomized and placed face down on a table. The sucker is allowed to draw a card from any set, then you draw a card from a different
NONTRANSITIVE PARADOXES
59
Figure 27 Nontransitive sucker bet based on magic square: AB+C+A
set. The high card wins. It is easy to prove that no matter what set the sucker draws from, you can pick a set that gives you winning odds of five to four. Set A beats B, B beats C, and C beats A. The victim may even be allowed to decide each time whether the high or the low card wins. If you play low card wins, simply pick the winning pile with respect to a nontransitive circle that goes the other way. A good way to play the game is to use sets of cards from three
60
CHAPTER FIVE
decks with backs of different colors. The packet of nine cards is shuffled each time, then separated by the backs into the three sets. The swindle is, of course, isomorphic with the tennistournament paradox. Nontransitivity prevails in many other simple gambling games. (See Chapter 5 of my Wheels, Life,and Other Mathematical Amusements, for a description of a set of nontransitive dice.) In some cases, such as the top designed by Andrew Lenard (see Figure 28), the nontransitivity is easy to understand. The lower part of the top is fixed but the upper disk rotates. Each of two players chooses a different arrow, the top is spun (in either direction), and the person whose arrow points to the section with the highest number wins. A beats B, B beats C and C beats A, in each case with odds of two to one. In a set of four bingo cards designed by Donald E. Knuth (see Figure 29), the nontransitivity is cleverly concealed. Two players each select a bingo card. Numbers from 1 through 6 are randomly drawn without replacement, as they are in standard bingo. If a called number is on a card, it is marked with a bean. The first player to complete a horizontal row wins. Here, of course, the numbers are just symbols; they can be replaced by any set of six different symbols. I leave it to the reader to work out the probabilities that show card A beats B, B beats C, Cbeats D, and D beats A. The game is transitive with three players, but the winning odds for the four possible triplets are surprising. One of the most incredible of all nontransitive betting situations, discovered (appropriately) by a mathematician named Walter Penney, was given as a problem in the Journal ofRecreational Mathematics (October 1969, p. 241). It is not well known, and most mathematicians simply cannot believe it when they first hear of it. It is certainly one of the finest of all sucker bets. It can be played
ABCA
Figure 28 Nontransitive top
NONTRANSITIVE PARADOXES
61
A+B+C+D+A
Figure 29 Nontransitive bingo cards
with a penny, or as a side bet on the reds and blacks of a roulette wheel, or in any situation in which two alternatives are randomized with equal odds. W e shall assume that a penny is used. If it is flipped three times, there are eight equally probable outcomes: HHH, HHT, HTH, HTT, THH, THT, TTH, and T T T . One player selects one of these triplets, and the other player selects a different one. The penny is then flipped repeatedly until one of the chosen triplets appears as a run and wins the game. For example, if the chosen triplets are HHT and T H T and the flips are THHHT, the last three flips show that HHT has won. In brief, the first triplet to appear as a run wins. One is inclined to assume that one triplet is as likely to appear first as any other, but it takes only a moment to realize that this is not the case even with doublets. Consider the doublets HH, HT, TH, and T T . HHand HTare equally likely to appear first because, after the first H appears, it is just as likely to be followed by an H as by a T. The same reasoning shows that T T and T H are equal. Because of symmetry, HH = T T and H T = TH. TH beats HH with odds of three to one, however, and H T beats T T with the same probability. Consider H T and T T . T T is always preceded by H T except when T T appears on the first two flips. This happens in the long run only once in four times, and so the probability that H T beats T T is 3/4. Figure 30 shows the probability that B, the second player, will win for all pairs of doublets.
62
CHAPTER FIVE
Figure 30 Probabilities of B winning
When we turn to triplets, the situation becomes much more surprising. Since it does not matter which side of a coin is designated heads, we know that HHH = T T T , T T H = HHT, H T H = T H T , and so on. When we examine the probabilities for unequal pairs, however, we discover that the game is not transitive. No matter what triplet the first player takes, the second player can select a better one. Figure 31 gives the probability that B, the second player, defeats A for all possible pairings. T o find B's best response to a triplet chosen by A, find A's triplet at the top, go down the column until you reach a probability (shown in gray), then move left along the row to B's triplet on the left. Note that B's probability of winning is, at the worst, 2/3 (or odds of two to one) and can go as high as 7/8 (or odds of seven to one). The seventoone odds are easy to comprehend. Consider T H H and HHH. If HHH first appears anywhere except at the start, it must be preceded by a T , which means that T H H has appeared earlier. HHH wins, therefore, only when it appears on the first three flips. Clearly this happens only once in eight flips. Barry Wolk of the University of Manitoba has discovered a curious rule for determining the best triplet. Let X be the first triplet chosen. Convert it to a binary number by changing each H to zero and each T to 1.Divide the number by 2, round down the quotient to the nearest integer, multiply by 5, and add 4. Express the result in binary, then convert the last three digits back to H and T. Nontransitivity holds for all higher ntuplets. A chart supplied by Wolk gives the winning probabilities for B in all possible pairings of quadruplets (see Figure 32). Like the preceding two charts and charts for all higher ntuplets, the matrix is symmetric about the center. The upper right quadrant is the
NONTRANSITIVE PARADOXES
63
lower left quadrant upside down, and the same holds for the upper left and lower right quadrants. The probabilities for B's best responses to A are shown in gray. In studying these figures, Wolk discovered another kind of anomaly as surprising as nontransitivity. It has to do with what are called waiting times. The waiting time for an ntuplet is the average number of tosses, in the long run, until the specified ntuplet appears. The longer you wait for a bus, the shorter becomes the expected waiting time. Pennies, however, have no memory, so that the waiting time for an ntuplet is independent of all previous flips. The waiting time for H and T is 2. For doublets the waiting time is 4 for H T and TH, and 6 for HH and T T . For triplets the waiting times are 8 for HHT,
Figure 3 1 Probabilities of B winning in a triplet game
64
CHAPTER FIVE
Figure 32 Probabilities of B winning in a quadruplet game
H T T , THH, and T T H ; 10 for HTH and T H T , and 14 for HHH and T T T . None of this contradicts what we know about which triplet ofa pair is likely to show first. With quadruplets, however, contradictions arise with six pairs. For example, T H T H has a waiting time of 20 and HTHH a waiting time of 18. Yet, T H T H is more likely to turn up before H T H H w i t h a probability of 9/14, or well over onehalf. In other words, an event that is less frequent in the long run is likely to happen before a more frequent event. There is no logical contradiction involved here, but it does show that "average waiting time" has peculiar properties. There are many ways to calculate the probability that one ntuplet will precede another. You can do it by summing infinite series, by drawing tree
NONTRANSITIVE PARADOXES
65
diagrams, by recursive techniques that produce sets of linear equations, and so on. One of the strangest and most efficient techniques was devised by John Horton Conway of the University of Cambridge. I have no idea why it works. It just cranks out the answer as if by magic, like so many of Conway's other algorithms. The key to Conway's procedure is the calculation of four binary numbers that Conway calls leading numbers. Let A stand for the 7tuplet HHTHHHT and B for THHTHHH. W e want to determine the probability of B beating A. T o do this, write A above A, B above B, A above B, and B above A (see Figure 33). Above the top tuplet of each pair a binary number is constructed as follows. Consider the first pair, AA. Look at the first letter of the top tuplet and ask yourself if the seven letters, beginning with this first one, correspond exactly to the first seven letters of the tuplet below it. Obviously they do, and so we put a 1 above the first letter. Next, look at the second letter of the top tuplet and ask if the six letters, starting with this one, correspond to thejrstsix letters of the tuplet below. Clearly they do not, and so we put zero above the second letter. Do the five letters starting with the third letter of the top tuplet correspond to the first five letters of the lower tuplet? No, and so this letter also gets zero. The fourth letter gets another zero. When we check the fifth letter of the top A, we see that HHTdoes correspond to the first three letters of the lower A, and so the fifth letter gets a 1. Letters six and seven each get zero. The " A leading A number," or AA, is 1000100, in which each 1 corresponds to ayes answer, each zero to a no. Translating 1000100 from binary to decimal gives us 68 as the leading number for AA. Figure 33 shows the results of this procedure in calculating leading numbers AA, BB, AB, and BA. Whenever an ntuplet is compared with itself, the first
1000100=68
A = HHTHHHT A = HHTHHHT
1000000=64
B = THHTHHH B = THHTHHH
0000001 =I
A = HH T H H H T B = THH THHH AA  AB:BB  BA 681:6435 67 :29 0 1 0 0 0 1 1 =35
B = THH THHH A =HHTHHHT
Figure 33 John Horton Conway's algorithm for calculating odds of B's ntuplet beating A's ntuplet
66
CHAPTER FIVE
digit of the leading number must, of course, be 1. When compared with a different tuplet, the first digit may or may not be 1. The odds in favor of B beating A are given by the ratio A A  A B :BB BA. In this case 68  1 : 64  35 = 67 :29. As an exercise, the reader can try calculating the odds in favor of THH beating HHH. The four leading numbers will be A A = 7, BB = 4, A B = 0, and B A = 3. Plugging these into the formula, A A  A B :BB  B A gives odds of 7  0 : 4  3, or seven to one, as expected. The algorithm works just as well on tuplets of unequal lengths, provided the smaller tuplet is not contained within the larger one. If, for example, A = HH and B = HHT, A obviously wins with a probability of 1. I conclude with a problem by David L. Silverman, who was the first to introduce the Penney paradox in the problems department that he then edited for theJourna1 ofRecreationa1Mathematics (Vol. 2, October 1969, p. 241). The reader should have little difficulty solving it by Gonway's algorithm. TTHH has a waiting time of 16 and HHH has a waiting time of 14. Which of these tuplets is most likely to appear first and with what probability?
ANSWERS Which pattern of heads and tails, TTHH or HHH, is more likely to appear first as a run when a penny is repeatedly flipped? Applying John Horton Conway's algorithm, we find that TTHH is more likely to precede HHH with a probability of 7/12, or odds of seven to five. Some quadruplets beat some triplets with even greater odds. For example, THHH precedes HHH with a probability of 718, or odds of seven to one. This is easy to see. HHH must be preceded by a T unless it is the first triplet of the series. Ofcourse, the probability of that is 118. The waiting time for TTHH and for THHH is 16, compared with a waiting time of 14 for HHH. Both cases of the quadruplet versus the triplet, therefore, exhibit the paradox of a less likely event occurring before a more likely event with a probability exceeding 112.
ADDENDUM Numerous readers discovered that Barry Wolk's rule for picking the best triplet B to beat triplet A is equivalent to putting in front of A the complement of its nexttolast symbol, then discarding the last symbol. More than half these correspondents found that the method also works for quadruplets except for the two in which H a n d T alternate throughout. In such cases the symbol put in front of A is the same as its next to last one.
NONTRANSITIVE PARADOXES
67
Since October 1974, when this chapter first appeared in ScientificAmerican, many papers have been published that prove Conway's algorithm and give procedures for picking the best ntuplet for all values of n. Two important early articles are cited in the bibliography. The paper by Guibas and Odlyzko gives twentysix references. Readers David Sachs and Bryce Hurst each noted that Conway's "leading number," when an ntuplet is compared with itself, automatically gives that tuplet's waiting time. Simply double the leading number. Ancient Chinese philosophers (I am told) divided matter into five categories that form a nontransitive cycle: wood gives birth to fire, fire to earth, earth to metal, metal to water, and water to wood. Rudy Rucker's sciencefiction story "Spacetime Donuts" (Unearth, Summer 1978) is based on a much more bizarre nontransitive theory. If you move down the scale of size, to several steps below electrons, you get back to the galaxies of the same universe we now occupy. Go up the scale several stages beyond our galactic clusters, and you are back to the elementary particles not larger ones, but the very same particles that make our stars. The word "matter" loses all meaning. The following letter was published in ScientificAmerican (January 1975):
Sirs: Martin Gardner's article on the paradoxical situations that arise from nontransitive relations may have helped me win a bet in Rome on the outcome of the Ali v. Foreman world heavyweight boxing title match in Zalre on October 30. Ali, though slower than in former years, and a 4 1 betting underdog, may have had a psychological and motivational advantage for that particular fight. But in addition, Gardner's mathematics might be relevant. Even though Foreman beat Frazier, who beat Ali, Ali could still beat Foreman because there may be a nontransitive relation between the three. I ranked the three fighters against the criteria of speed, power, and technique (including psychological technique) as reported in the press, and spotted a nontransitive relation worth betting on:
Speed Power Technique
Ali 2
3
Frazier 1 2
1
3
Foreman
3 1 2
68
CHAPTER FIVE
Foreman's power and technique beat Frazier, but Ali's technique and speed beat Foreman. It was worth the bet. The future implications are, however, that Frazier can still beat Ali! PIEL ANTHONY Vaud, Switzerland David Silverman (Journal $Recreational Mathematics 2, October 1969, p. 241) proposed a twoperson game that he called "blind Penneyante." It is based on the nontransitive triplets in a run of fair coin tosses. Each player simultaneously chooses a triplet without knowing his opponent's choice. The triplet that shows up first wins. What is a player's best strategy?This is not an easy problem. A full solution, based on an 8 X 8 game matrix, is given in The College Mathematics Journal as the answer to Problem 299 (January 1987, p. 74 76).
B
I
B
L
I
O
G
R
A
P
H
Y
Nontransitive Voting and Tournament Paradoxes Social Choice and Individual Values. Kenneth J. Arrow. Wiley, 1951. Games and Decision. R. Duncan Luce and Howard Raiffa. Wiley, 1957. The Theory of Committees and Elections. Duncan Black. Cambridge University Press, 1958.
"Voting and the Summation of Preferences: An Interpretive Bibliographical Review of Selected Developments During the Last Decade." William H. Riker, in The American Political Science Review 55, December 1961, pp. 900  91 1. Topics on Tournaments. John W . Moon. Holt, Rinehart and Winston, 1968. Theory of Voting. R. Farquharson. Yale University Press, 1969. Paradoxes in Politics. Steven J. Brams. Free Press, 1976. Arrow's Theorem: The Paradox of Social Choice, Alfred F. MacKay. Yale University Press, 1980.
Nontransitive Dice and Roulette Betting "Nontransitive Dominance." Richard L. Tenney and Caxton C. Foster, in Mathematics Magazine 49, May 1976, pp. 115 120. "A Coin Tossing Game." James C. Frauenthal and Andrew B. Miller, in Mathematics Magazine 53, September 1980, pp. 239243. "Lucifer at Las Vegas." Martin Gardner, Problem 27 in Science Fiction Puzzle Tales. Clarkson Potter, 1981.
NONTRANSITIVE PARADOXES
69
"Nontransitive Dice and Other Probability Problems." Martin Gardner, in Chapter 5 of Wheels, LLife, and Other Mathematical Amusements. W . H . Freeman and Company, 1983. "Sheep Fleecing with Walter Funkenbusch." Ross Honsberger, in Mathematical Gems III. Mathematical Association of America, 1985. Conway's Algorithm "String Overlaps, Pattern Matching, and Nontransitive Games." L. J. Guibas and A. M. Odlyzko, in the Journal ofcombinatorial Theory 30, March 1981, pp. 183208. "Coin Sequence Probabilities and Paradoxes." Stanley Collings, in the Bulletin of the Institute of Mathematics and its Applications 18, November/December 1982, pp. 227  232.
SIX
Combinatorial Card Problems There is no end to the making of mathematical tricks, puzzles, and other recreations that employ playing cards. In this chapter we look at some new card problems and games, with emphasis on how they lead into significant areas of combinatorial theory. Consider the following combinatorial way of dramatizing an important number theorem. Remove all the cards of one suit (say, spades) from a deck and arrange them in serial order from ace to king. (The jack, queen, and king, respectively, represent 11,12, and 13.) Place them face down in a row with the ace at the left. The following turning procedure is now applied, starting at the left at each step and proceeding to the right:
1. Turn over every card. 2. Turn over every second card. (Cards 2 , 4 , 6 , 8 , 1 0 ,and Q are turned face down.) 3. Turn over every third card. 4. Continue in this manner, turning every fourth card, every fifth card, and so on until you turn over only the last card. Inspect the row. Note that all the cards except the ace, the 4, and the 9 are face down as shown in Figure 34. These values happen to be square numbers.
72
CHAPTER SIX
Is this an accident? O r is it an authentic hint of a general rule? A good classroom exercise is to prepare 100 small cards bearing numbers 1 through 100, stand them with their backs out in serial order on a blackboard ledge and apply the turning ~rocedure.Sure enough, at the finish the only visible numbers will be the squares: 1,4,9,16,25,36,49,64,81,and 100. That is too large a sampling to be coincidental. T h e next step is to prove that no matter how large the deck, only squares survive the turning procedure. A simple roof introduces one of the oldest and most fundamental of number theorems: A positive integer has an odd number of divisors (the divisors include 1 and the number itself), if and only if the number is a square. This is easy to see. Most divisors of a number come in pairs. Consider 72. The smallest divisor, 1, goes into the number 72 times, giving the pair 1 and 72. The nextlarger divisor, 2, goes into the number 36 times, giving the pair 2 and 36. Similarly, 72 = 3 X 24 = 4 X 18 = 6 X 12 = 8 X 9. The only divisor of a number that is not paired with a different number is a divisor that is a square root. Consequently, all nonsquares have an even number of divisors, and all squares have an odd number of divisors, How does this apply to the row of cards? Consider the eight of spades in the first cardturning example. Since 8 is not a square, it has an even number of divisors: 1 , 2 , 4 , 8 . It will be turned four times: when you turn each card, each second card, each fourth card, and each eighth card. An even number of turns applied to a facedown card will leave that card face down. Since every nonsquare card will be turned an even number of times, it will be face down at the finish. The only cards that are turned an odd number of times and left face up are those with an odd number of divisors, namely the squares. Is there a better way to etch this basic number theorem in the memory of a highschool student than to have him witness such a demonstration? Let us see how cards can be used for modeling a combinatorial problem that D. H. Lehmer once described as follows. Mr. Smith manages a motel. It consists of n rooms in a straight row. There is no vacancy. Smith is a psychologist who plans to study the effects of rearranging his guests in all possible ways. Every morning he gives them a new permutation. The weather is miserable, raining almost daily. T o minimize his guests' discomfort, each daily rearrangement is made by exchanging the occupants of two adjoining rooms. Is there a simple algorithm that will run through all possible arrangements by switching adjacent occupants at each step? The problem is easily modeled with cards. A row of spades, ace to king, corresponds to a thirteenroom motel. T h e number of permutations of n elements is factorial n. Our problem is to exchange two adjacent cards at each step and run through every possible permutation in just (n!  1) steps. (We
Figure 34
Card algorithm for generating squares
74
CHAPTER SIX
subtract 1 from n! because we begin with one permutation on the table.) Such an algorithm has important applications in computer science. Many problems require a computer in order to run through all permutations of n elements, and if this can be done by exchanging adjacent pairs, there is a significant reduction in computer time. It turns out that there is a simple, beautiful algorithm for doing this; it leads to the fastestknown way for a computer to permute n elements. Hugo Steinhaus, a Polish mathematician, was the first to discover it. It provides a solution for the abacus ~ r o b l e mon page 49 of his One Hundred Problems in Elementary Mathematics, first published in Poland in 1958. In the early 1960's the procedure was independently rediscovered at almost the same time by H. F. Trotter and Selmer M. Johnson, each of whom published it separately. Solving the problem for thirteen cards would require 13!  1 = 6,227,020,799 steps (you can see why a fast computer algorithm is desirable); hence let us start with smaller sets. It is easy to find a solution for three cards, but four cards present difficulties. Besides, we want not just a solution for a specified number of elements but a general method that will apply to any number. With two cards (n = 2) the solution is trivial (see Figure 35A). Simply move the deuce from right to left. For n = 3 we list each of the preceding permutations three times: 12,12,12; 21,21,21. The numeral 3 is now added to the list by a twisting procedure (see Figure 35B). The 3 starts at the right of the first permutation, weaves left through the series, pauses once at the left, then weaves back to end at the right of the final permutation. This generates the series 123, 132, 312, 321, 231, and 213. If YOU start with an ace, deuce, and trey on the table and run through the series, you will see that each permutation is derived from the preceding one by switching two adjacent cards. For n = 4 each permutation of the series for n = 3 is repeated four times. (You always use n repetitions of the series for n  1.) The numeral 4 is then added, weaving it left and right as before (see Figure 35C). The algorithm can be defined by a nonrecursive formula, but the recursive procedure just explained is easier to understand. The trouble with defining a procedure for any n is that when the weaving card pauses at each side, the position of the pair to be switched varies in a curious way. The procedure given here is recursive, of course, because for each n, we must make use of the results obtained for ( n  1). It works for all higher n. For n = 5 we obtain 5! = 120 permutations, beginning with 12345,12354,12534,15234,51234, 51243,15243, . . . , and ending with 21345. Note that a switch of the first two numerals of the final permutation will give the first permutation. This holds for all n. The procedure is cyclic, restoring the original sequence in one more step. Note that after n!/2 steps the cards are in reverse order.
Figure 35
Recursive algorithm for solving motel problem
76
CHAPTER SIX
I am indebted to Donald E. Knuth for this means of displaying the recursive procedure, as well as for the algorithm's history. In the forthcoming fourth volume of his great series on The Art ofComputer Programming, he will discuss the algorithm and show how its "inversion table" is equivalent to what is called a reflected Gray code with mixed bases. The problem is a special case of a more general "motel problem," which in turn is a special case ofwhat Lehmer calls the "travelingburglar problem." A few years ago John Horton Conway of the University of Cambridge invented a series of card problems and games based on the technique of permuting a set of elements by reversing the order of subsets according to various rules. Take, for example, the thirteen spades from a deck, shuffle the packet and hold it face up in your left hand. Note the value of the uppermost card (we shall call it the top card) of the packet. Let us say it is a 9. Call the number out or to yourself, then with your right thumb slide nine cards off the packet one at a time into your right hand. This automatically reverses the order of the nine cards. Now put the ninecard packet on top of the cards in your left hand. A new card is face up on top. Note its value, call it out and repeat the same procedure. In other words, if n is the value of the top card, you always count off n cards, which reverses their order, and replace them face up on top. The game ends if an ace appears on top, because the ace produces a "oneloop" that consists in repeatedly counting off the ace and replacing it. Must the game always end with the call of an ace? Yes, although it may take quite a while. It is impossible to get into a loop before the ace is called. If the game continues long enough without the call of an ace, a king might eventually be called. If this happens, however, the next reversal of cards puts the king on the bottom. Once the king is on the bottom, there is no way it can leave. As the game continues, the queen might eventually be called. If this happens, the queen goes to the twelfth position from the top, just above the king, and stays there. By mathematical induction the same thing must happen to the jack, then the 10 and so on (each card going to a position that corresponds with its value) until eventually, if not sooner, the ace is called and the game terminates. Indeed, each card can be called only once after the latest appearance of all higher cards. The general form of the game involves one or more packets of n cards each. It is called kswops if the kth card from the top is called. The called number gives the number of cards to be counted and replaced on top. It is called kdrops if the same procedure is followed, with the reversed set going to the bottom. When there is only one packet and k equals 1 (the top card) and the counted cards go on top, Conway calls the game topswops. That is the game we have analyzed. The same game with the counted cards going to the bottom is called
COMBINATORIAL CARD PROBLEMS
77
topdrops. In topdrops the top card is called, the cards are counted, and the reversed set is placed on the bottom. Topdrops is less interesting than topswops. You begin at once with a loop that may be long or short and that may or may not contain an ace. When there is one packet and k equals n (the bottom card), the game is botswops or botdrops, depending on whether the counted set goes on top or on the bottom. Botswops is boring. If you play with thirteen spades and the king is not on the bottom, you are immediately in a twocard loop. Suppose the bottom card is a 4. It stays there while you repeatedly reverse the four cards on top. If the king is at the bottom, it goes to the top, and you find yourself in a similar twoloop based on the new bottom card. Botdrops (call the bottom card, count, and put the reversed set on the bottom) is more interesting. If you play it for a while, Conway writes, you might convince yourself that it always loops in a KQKQKQ . . . sequence, but that is not always the case. O n rare occasions other loops are possible. (Can you find one?) In this game, as in all the others, you start, of course, with a shuffled packet. When the game is extended to two or more players, each with a packet, it becomes much harder to analyze. For instance, suppose two players have packets of thirteen cards each. One has spades, the other hearts. They play topswops as follows. Each shuffles his packet. Player A calls his top card, then B counts that number off his packet and replaces the reversed cards on top of his packet. B now calls his top card, A counts and replaces the reversed cards on top of his packet. This continues with players alternating calls. It is a curious fact, reports Conway, that as soon as an ace is called, the calls go into a loop that starts with an ace, then a sequence, then an ace again (either the same ace or the other one), then the same sequence is repeated in reverse. For example, the first called ace might generate the following loop: 1 3  26  4  1 4  6  2  3  1. Note that the sequence between the first two ace calls is the reverse of the sequence between t i e second and third ace calls. It is an unproved conjecture (or was when I last heard from Conway) that in twoplayer topswops an ace is always called. It is not known if the game can conclude in a loop without an ace, although it is known that if a loop includes an ace, it includes it just twice. Let it not be supposed that these Conway card games are trivial. They deal with the theory of set permutations and not only may provide deep theorems but also may have a bearing on practical problems that arise in seemingly unrelated fields. I conclude with three unusual combinatorial card problems. The first is extremely difficult, the second is easy but elegant, and the third is tricky.
78
CHAPTER SIX
1. A Langford problem. In Chapter 5 of my Mathematical Magic Show (Knopf, 1977), I discussed a combinatorial problem, first posed by C. Dudley Langford, that could be worked on with a set of cards containing doublets of values from 1 to n. When the ~ r o b l e mis extended to triplets, the smallest value of n for which a solution is known is 9. Here is the task. Remove from a deck all the cards of three suits that bear values of ace through 9. Try to arrange these twentyseven cards in a single row to meet the following proviso. Between the first two cards of every value k there are exactly k cards, and between the second and third cards of every value k there also are exactly k cards. For instance, between the first and second 7's there must be just seven cards, not counting the two 7's. Similarly, seven cards separate the second and third 7's. The rule applies to each value from 1 through 9. 2. A Silverman problem. David L. Silverman is the inventor of this puzzle. Remove the spades and hearts from a deck. Put the spades face up in a row in serial order with the ace at the left and the king at the right. Place a heart card under each spade so that the sum of the two cards is a square number. Prove that the solution is unique. 3. A Ransom problem. This comes to me through the courtesy of T o m Ransom, a Canadian amateur magician and puzzle collector. Here is how Ransom has been showing his puzzle to magician friends. Five cards are placed in a row as shown in Figure 36. All card backs, Ransom states correctly, are either colored or black. Are all the cards with colored backs jokers? The problem is not to answer the question but to determine the minimum number of cards that must be turned over in order to answer it. In other words, assuming any possible variation of the hidden card sides (eachjoker may have a black or a colored back, the card with the visible colored back may or may not be a joker and so on), how many cards must you turn over before you can answer the question: "Are all the cards with colored backs jokers?" It is a confusing problem and one that calls for careful reasoning. There is a surprise in the solution that is closely related to an old joke about three professors on a train in Scotland. Through the window they see a black sheep.
Figure 36 Are all the cards with colored backs jokers?
COMBINATORIAL CARD PROBLEMS • 79
“How interesting,” says the astronomer. “All sheep in Scotland are black.” “A totally unwarranted inference,” the physicist replies. “We can conclude only that some sheep in Scotland are black.” “Correction,” says the logician. “At least one sheep in Scotland is black on at least one side.”
ANSWERS The three combinatorial problems are answered as follows: 1. Cards 1 through 9, of three suits, are to be arranged in a row so that for every card value k, just k cards are between the first and second cards of value k and between the second and third cards of value k. Not counting reversals, there are three solutions: 1, 8, 1, 9, 1, 5, 2, 6, 7, 2, 8, 5, 2, 9, 6, 4, 7, 5, 3, 8, 4, 6, 3, 9, 7, 4, 3 1, 9, 1, 2, 1, 8, 2, 4, 6, 2, 7, 9, 4, 5, 8, 6, 3, 4, 7, 5, 3, 9, 6, 8, 3, 5, 7 1, 9, 1, 6, 1, 8, 2, 5, 7, 2, 6, 9, 2, 5, 8, 4, 7, 6, 3, 5, 4, 9, 3, 8, 7, 4, 3 The third solution was found without computer aid in 1966 by Eugene, Levine, now a mathematician at Adelphi University. It was published in 1968. Levine proves that a solution for triplets exists only when n, the highest value of a card, has a digital root of 1, 8, or 9 (that is, when n equals –1, 0, or 1, modulo 9), and that 9 is the smallest n that has a solution. Levine found solutions for the next higher cases of n = 10, 17, 18, and 19, and conjectured that there are solutions for all higher values meeting his proviso. D. P. Roselle and T. C. Thomasson, Jr., writing on the problem in 1971, reported computer results that confirm there is no solution for n = 8, and came up with the same solution for n = 9 that Levine found. An exhaustive computer search for n = 9 and n = 10 was made by G. Baron, who reported his results at a conference on combinatorics held in Hungary in 1969. He found the three solutions for n = 9 and five solutions for n = 10. No solution has yet been found for this problem if there are more than three duplicates of each value.
80
CHAPTER SIX
2. The solution to David L. Silverman's problem of pairing each spade with a heart so that the pairsum is square is: 1 8, 2 2, 3 K, 4 Q, 5 J, 6 10,79'81,97,lO6,J5, Q4,K3. AsSilvermanobserves,9,10, andJmust be paired with 7,6, and 5. This establishes six pairings. Since the 6's have been used, 3 pairs only with K. Since the 5's have been used, 4 pairs only with Q. The remaining three gaps are filled in only one way, proving the solution's uniqueness. 3. Tom Ransom's problem asked for the minimum number of cards, in a set of five, that must be turned over to answer the question: Are all coloredback cardsjokers? Letter the cards in Figure 36 from A through E. Obviously, D must be turned to see if it is a joker, and E must be turned to see if it has a colored back. This gives four possibilities:
1. 2. 3. 4.
D is a joker, E has a black back. D is a joker, E has a colored back. D is not a joker, E has a black back. D is not a joker, E has a colored back.
For cases 2,3, and 4 the answer to the question is no. N o more cards need be turned. For case 1 the answer is yes, but it takes more thinking to realize that turning the other three cards cannot contradict this answer. B is irrelevant because it has a black back. Seeing the back of eitherjoker is also irrelevant. If a joker's back is black, it is not involved in the question. If it is colored, the answer is still yes. Most people staring at an actual row of cards have such an overwhelming desire to see the backs of thejokers that they usually answer: A, C, D, E. One might conclude, therefore, that turning D and E is sufficient to answer the question. It is not! Recall the story about the cautious logician who observed a black sheep in Scotland and concluded that at least one sheep in Scotland is black on at least one side? When someone thinks he has solved the problem, Ransom turns over card B to reveal that its other side is a colored back! This, ofcourse, contradicts ayes answer. The correct solution, therefore, is that card B as well as cards D and E must be turned. Ransom has a second "kicker," suggested by his friend P. Howard Lyons. So that a person working on the problem will not forget the exact phrasing of the question ("Are all coloredback cards jokers?"), Ransom writes it on a file card which is placed above the row of cards. This card also must be turned to determine whether its back is colored or black!
COMBINATORIAL CARD PROBLEMS
81
ADDENDUM The combinatorial playingcard problems produced memorable letters from readers who analyzed and extended the problems. Alan Hadsell and Stoddard Vandersteel together used a minicomputer to generalize David L. Silverman's problem. When the highest card value does not exceed 13, solutions exist only for n = 3,5,8,9,lO, 12, and 13, and each solution is unique. From 14 through 31 all values of n have multiple solutions. They report that the number of solutions, beginning with n = 14, are 2,4,3,2,5,15,21,66,37,51,144,263, 601, 1333,2119,2154,2189,3280,. . . . Angeloo Papaikonomou, a bioengineer at the Free University of Amsterdam, after exploring the general problem for squares, turned to the general problem when cubes are substituted for squares. He was surprised to find that for every solvable n the solution is unique. The series of solvable n's begins 7, 19,26,37,44,56,63, . . . . Papaikonomou found a simple recursive procedure that both gives this series and constructs the solutions. Roland Silver of San Cristobal, N.M., exploring botdrops for cycles, discovered why the game usually terminates in a king queen loop. There are no other 2cycles and none of 3 and 4. Although 5cycles exist, the probability of entering one is low. For example, if the faceup packet, reading up from the bottom, starts with 10,J, 2 , 3 and has an ace on top, the ~ a c k eis t in a botdrop 5loop. Herbert S. Wilf of the University of Pennsylvania reported a delightful discovery about topswops that provides a proof of the game's finiteness. A card is in "natural position" if its value is the same as its position in the packet. For example, if the faceup packet, reading down from the top, is
there are five cards (2,5,9,10, Q) in natural position. Ifwe take these values as powers of two, we can create what I shall call the Wilf number: 22 2' Z9 21° 212 = 5668. After any move in topdrops the Wilf number must increase. "The reason that the number increases," Wilf writes, "is that the cards which were in natural position and which were too far down in the deck to be reached by the reversal operation will still be in natural position afterward. The fate of the cards which are involved in the reversal is less clear, except for one thing: the card which was on top before the move will be in natural position after the move, and its power oftwo is large enough to drown out any changes from cards above it. (A power of two is larger than the sum of all
+ +
+ +
82
CHAPTER SIX
earlier powers of two by exactly one unit, a fact which is the basis of binary counting.) "Since the numbers increase steadily but cannot exceed 16,382, it follows that the game must halt after at most that many moves. A slightly more careful study shows, in fact, that for a game with n cards, no more than 2"' moves can take place." This raises an interesting unsolved question: What arrangement of the thirteen cards provides the longest possible game of topswops? A simpler fourcard version of T o m Ransom's puzzle, without Ransom's joke of using what magicians call a doubleback card, has been the object of research by psychologists. It is considered, for example, in The Psychology of Reasoning, by Peter Wason and Philip JohnsonLaird. Reaction to the disclosure that a card could be doublebacked varied enormously. Jay Snyder began a letter: "Wait a minute! Hold it! Point of order! Not so fast! One moment please! Pull over, buddy." Stover sent a set of five tapeddown cards to Wason at University College London for analysis. When Wason was later informed of what he would have found if he had pulled off the cards to check their undersides, he opened his next letter, "Mea culpa!" Several readers found the problem's wording ambiguous as to the meaning of "minimum number of cards" to be turned. The problem is to specify a minimum set of cards that, when they are reversed, will in all possible cases guarantee a correct answer to the question: Are all redbacked cards jokers? (The problem is best presented with cards that may have either red or blue backs.) If we do not care whether or not our answer is correct, obviously we can "answer" the question by turning no cards. In some cases turning one card will guarantee a correct answer. What we want is the smallest set that will provide a correct answer for all cases. James Weinrich has given me permission to reproduce here a letter he sent in December 1974: Dear Mr. Gardner, Here are third and fourth "kickers" regarding Ransom's question, "Are all the coloredback cards jokers?" 3) The paperback American Heritage Dictionary (page 385) defines a joker as "an element in a situation that acts in an unexpected way." Clearly, a blackfronted card (B) that turns out to be redbacked is a joker by this definition. Accordingly: if B is not redbacked, it is irrelevant to the question; if it is redbacked, it is certainly an element acting in an unexpected way, and is thus a joker, but then
COMBINATORIAL CARD PROBLEMS
83
does not affect the answer as you explained for the other jokers. Thus it is not necessary to turn over B to know if all the redbacked cards are jokers.
4) The same dictionary (page 110)defines a card as "an amusing or eccentric person." If Ransom's card B is indeed redbacked, then he has played an amusing trick, and thus himself qualifies as a card. Accordingly, he could be turned over (or at least around) to see if he has acquired a sunburn on his back. One might reason that Ransom is ajoker in any case, and thus it is not necessary to see if he has a red back to answer the question. However, Ransom is not ajoker if the blackfronted card B is a standard one; in that case it is necessary to check Ransom's back. Thus, the correct number of cards to turn is three or four: cards D and E in any case, card B (to see if Ransom is a joker), and Ransom (if B is not a joke card). More or less sincerely yours, JAMES D. WEINRICH
P.S.: Is Ransom Caucasian? If not, it can be assumed he has a colored back, and it would not be necessary to turn him over if B were a joke card. Note that in this case, his back might be both colored and black at the same time . . . . B
I
B
L
I
O
G
R
A
P
H
Y
The Motel Problem "Algorithm 115." H. F. Trotter, in Communications ofthe A C M 5, 1962, pp. 434435. "Generation of Permutations by Adjacent Transposition." Selmer M. Johnson, in Mathematics ofComputation 17, July 1963, pp. 282285. One Hundred Problems in Elementary Mathematics. Hugo Steinhaus. Basic, 1964. (See pp. 4445,4950,17374.)
"Permutation by Adjacent Interchanges." D. H. Lehmer, in The American Mathematical Monthly 72, February 1965, pp. 3646. "Generating Permutations with Nondistinct Items." T. C. Hu and B. N. Tien, in The American Mathematical Monthly 83, October 1976, pp. 629631. Langford's Problem "The Existence of Perfect 3Sequences." Eugene Levine, in The Fibonacci Quarterly 6, November 1968, pp. 108 112.
84
CHAPTER SIX
"Ubcr Vcrallgcmcincrungcn dcs 1,angfordSchcn Problems." G. Haron, in Cbmhinatorial Theory and 1t.r Applications I . Paul Erdos et al., eds. Proceedings of the Conference in Halatonfiircd, 1969. North Holland, 1970, pp. 8192. "On Generalized Langford Sequences." D. P. Roselle and T. C. Thomasson, Jr., in Journal oJ' Cbmhinatorial Theory 11, Scptcmbcr 1971, pp. 196199. "Langford Sequences: a progress report." Takanon Hyasaka and Sadao Saito. in Mathematical Gayette, 63, December 1979, pp. 261262. Ransom's Problem: FourCard ITersions The Psychology of Reasoning. Peter Wason and Philip Johnson1,aird. Harwrd University Press, 1972. "The Psychology of Deceptive Problems." Peter Wason, in New Scientirt, August 15, 1974, pp. 382385. '/'he Mind? ~ V e wScience: A I lirtnry cf the (.7n,nitil/e Keuobtinn, Chapter 13. Howard Gardner. Basic, 1985.
SEVEN
~elod~M king a Machines
"Mathematics and music! The most glaring possible opposites of human thought! Yet connected, mutually sustained!" HERMANN VON HELMHOLTZ,
Popular Scientific Lectures
There is a trivial sense in which any work of art is a combination of a finite number of discrete elements. Not only that, the precise combination of the elements can be expressed by a sequence of digits or, if you will, by one enormous number. Consider a poem. Assign distinct numbers to each letter of the alphabet, to each punctuation symbol, and so on. A certain digit, say zero, can be used to separate the numbers. It is obvious that one long string of digits can express the poem. If the books of a vast library contain every possible combination of words and punctuation marks, as they do in Jorge Luis Borges's famous story
86
CHAPTER SEVEN
"The Library of Babel," then somewhere in the collection is every poem ever written or that can be written. Imagine those poems coded as digital sequences and indexed. If one had enough time, billions on billions of years, one could locate any specified great poem. Are there algorithms by which one could find a great poem not yet written? Consider a painting. Rule the canvas into a matrix of minute cells. The precise color of each cell is easily coded by a number. Scanning the cells yields a chain of numbers that expresses the painting. Since numbers do not decay, a painting can be recreated as long as the number sequence is preserved. Future computers will be able to reproduce a painting more like the original than the original itself, since after a few decades the original will have physically . deteriorated to some extent. If a vast art museum contains every combination of colored cells for matrixes not exceeding a certain size, somewhere in that monstrous museum will hang every picture ever painted or that can be painted. Are there algorithms by which a computer could search a list of the museum's code numbers and identify a sequence for a great painting not yet painted? Consider a symphony. It is a fantastically complex blend of discreteness and continuity; a violin or a slide trombone can move up and down the scale continuously, but a piano cannot produce quartertones. W e know, however, from Fourier analysis that the entire sound of a symphony, from beginning to end, can be represented by a single curve on an oscilloscope. "This curve," wrote Sir James Jeans in Science andMusic (Dover, 1968), "is the symphonyneither more nor less, and the symphony will sound noble or tawdry, musical or harsh, refined or vulgar, according to the quality of this curve." O n a longplaying record a symphony is actually represented by one long space curve. Because curves can be coded to any desired precision by numbers, a symphony, like a painting or a poem, can be quantized and expressed by a number chain. A vast library of computer tapes, recording all combinations of symphonic sounds, would contain every symphony ever written or that could be written. Are there algorithms by which a computer could scan the number sequences of such a library and pick out a great symphony not yet written? Such procedures would, of course, be so stupendously complex that man may never come close to formulating them, but that is not the point. Do they exist in principle? Is it worthwhile to look for bits and pieces of them? Consider one of the humblest of such aesthetic tasks, the search for rules that govern the invention of a simple melody. Is there a procedure by which a person or a computer can compose a pleasing tune, using no more than a set of combinatorial rules?
MELODYMAKING MACHINES
87
If we restrict the tune to a finite length and a finite number of pure tones and rhythms, the number of possible melodies is finite. John Stuart Mill, in his autobiography, recalls that as a young man he was once "seriously tormented by the thought of the exhaustibility of musical compositions." Suppose our tune is made up ofjust ten notes chosen from the set of eight notes in a single octave. The number of melodies is the same as the number of tenletter words that can be formed with eight distinct letters, allowing duplications. It is 81° = 1,073,741,824, and this without even considering varying rhythms which create, in effect, varying melodies. Most of these tunes will be dull (mi, mi, mi, mi, mi, mi, mi, mi, mi, mi for instance), but many will be extremely pleasing. Are there rules by which a computer or a person could pick out the pleasing combinations? Attempts to formulate such rules and embody them in a mechanical device for composing tunes have a curious history that began in 1650 when Athanasius Kircher, a German Jesuit, published in Rome his Musurgia universalis sive ars magna consoni et dissoni (see Figure 37). Kircher was an ardent disciple of Ramon Lull, the Spanish medieval mystic whose Ars magna derived from the crazy notion that significant new knowledge could be obtained in almost every field simply by exploring all combinations of a small number of basic elements. It was natural that Kircher, who later wrote a 500page elaboration of Lull's "great art," would view musical composition as a combinatorial problem. In his music book, he describes a Lullian technique of creating polyphony by sliding columns alongside one another, as with Napier's bones, and reading off rows to obtain various permutations and combinations. Like all of Kircher's huge tomes, the book is a fantastic mix of valuable information and total nonsense, illustrated with elaborate engravings  of vocal cords, bones in the ears of various animals, birds and their songs, musical instruments, mechanical details of music boxes, wateroperated organ pipes with animated figures of animals and people, and hundreds of other curious things. The Lullian device described by Kircher was actually built, circa 1670, for the diarist Samuel Pepys, who owned a copy of ~ i r d h e r ' smusic book and much admired it. The original machine, called "musarithmica mirifica," is in the Pepys Museum at Pepys's alma mater, Magdalene College, Cambridge. During the early eighteenth century many German music scholars became interested in mechanical methods of composition. Lorenz Christoph Mizler wrote a book in 1739 describing a system that produced figured bass for baroque ensemble music. In 1757 Bach's pupil, Johann Philipp Kirnberger, published in Berlin his Everready Composer ofPolonaises and Minuets, using a die for randomizing certain choices. In 1783 another book by Kirnberger extended his methods to symphonies and other forms of music.
Figure 37 Frontispiece of Kircher's Musurgia universalis (1650). The artist was J. Paul Schor. Here is how the picture is described in Athanasius Kircher: A Renaissance Man and the Questfor Lost Knowledge, by Joscelyn Godwin (Thames and Hudson, Ltd., 1979): "The symbol of the Trinity sheds its rays on the nine choirs of angels, who sing a 36part canon (by Romano Micheli), and thence on the earth. The terrestrial sphere is shown encircled by the Zodiac and surmounted by Musica, who holds Apollo's lyre
MELODYMAKING MACHINES
89
Toward the close of the eighteenth century the practice of generating melodies with the aid of tables and randomizers such as dice or teetoturns became a popular pastime. Maximilian Stadler, an Austrian composer, published in 1779 a set of musical bars and tables for producing minuets and trios with the help of dice. At about the same time, a London music publisher, Welcker, issued a "tabular system whereby any person, without the least knowledge of music, may compose ten thousand minuets in the most pleasing and correct manner." Similar anonymous works were falsely attributed to wellknown composers such as C. P. E. Bach (son of Johann Sebastian Bach) and Joseph Haydn. The "Haydn" work, "Gioco filarmonico" ("Philharmonic Joke"), Naples, 1790, was discovered by the Glasgow mathematician Thomas H. O'Beirne to be a plagiarism. Its bars and tables are identical with Stadler's. The most popular work explaining how a pair of dice can be used "to compose without the least knowledge of music" as many German waltzes as one pleases was first published in Amsterdam and in Berlin in 1792, a year after Mozart's death. The work was attributed to Mozart. Most Mozart scholars say it is spurious, although Mozart was fond of mathematical puzzles and did leave handwritten notes showing his interest in musical permutations. (The same pamphlet was issued in Bonn a year later, with a similar work, also attributed to Mozart, for dice composition of country dances. The contredanses pamphlet was reprinted in 1957 by Heuwekemeijer in Amsterdam.) Mozart's Musikalisches Wiirfelspiel, as the waltz pamphlet is usually called, has been reprinted many times in many languages. In 1806 it appeared in London as Mozart's Musical Game, Fitted in an Elegant Box, Showing by Easy System How to Compose an Unlimited Number of Waltzes, Rondos, Hornpipes and Reels. In New York in 1941 the Hungarian composer and concert pianist Alexander Laszlo brought it out under the title The Dice Composer, orchestrating the music so that it could be played by chamber groups and orchestras. The system popped up again in West Germany in 1956 in a score published by B. Schott. Photocopies of Schott's charts and musical bars appear in the instruction booklet for The Melody Dicer, issued early in 1974 by Carousel Publishing
and the panpipes of Marsyas. In the landscape are seen dancing mermaids and satyrs, a shepherd demonstrating an echo, and Pegasus, the winged horse of the Muses. O n the left is Pythagoras, the legendary father of musical theory. He points with one hand to his famous theorem, and with the other to the blacksmiths whose hammers, ringing on the anvil, first led him to discover the relation of tone to weight. O n the right is a muse (Polymnia?) with a birdperched on her headpossibly one of the nine daughters of Pierus, who for their presumption in attempting to rival the Muses were turned into birds. These figures are surrounded respectively by antique and modern instruments."
90
CHAPTER SEVEN
Corporation, Brighton, Mass. This boxed set also includes a pair of dice and blank sheets of music paper. The "Mozart" system consists of a set of short measures numbered 1 through 176. The two dice are thrown sixteen times. With the aid of a chart listing eleven numbers in each of eight columns, the first eight throws determine the first eight bars of the waltz. A second chart is used for the second eight throws that complete the sixteenbar piece. The charts are constructed so that the waltz opens with the tonic or keynote, modulates to the dominant, then finds its way back to the tonic on its final note. Because all bars listed in the eighth column of each chart are alike, the eleven choices (sums 2 through 12 on the dice) are available for only fourteen bars. This allows the system to produce 11l4 waltzes, all with a distinct Mozartean flavor. The number is so large that any waltz you generate with the dice and actually play is almost certainly a waltz never heard before. If you fail to preserve it, it will be a waltz that will probably never be heard again. The first commercial recording of "Mozart" dice pieces was made by O'Beirne. Both the randomizing of the bars and the actual playing of the melodies was done by Solidac, a small and slow experimental computer designed and built between 1959 and 1964 by the Glasgow firm of Barr and Stroud, where O'Beirne was then chief mathematician. It was the first computer built in Scotland. O'Beirne programmed Solidac to play the pieces in clarinetlike tones, and a longplaying recording of selected waltzes and contredanses was issued by Barr and Stroud in 1967. (This recording is no longer available.) O'Beirne is the author of an excellent book on mathematical recreations, Puzzles and Paradoxes (Oxford University Press, 1965). He has been of invaluable help in the preparation of this account. Other methods of producing tunes mechanically were invented in the early 19th century. Antonio Calegari, an Italian composer, used two dice for composing pieces for the pianoforte and harp. His book on the system was published in Venice in 180 1, and later in a French translation. The Melographicon, an anonymous and undated book issued in London about 1805, is subtitled: "A new musical work, by which an interminable number of melodies may be produced, and young people who have a taste for poetry enabled to set their verses to music for the voice and pianoforte, without the necessity of a scientific knowledge of the art." The book has four parts, each providing music for poetry with a certain meter and rhyme scheme. Dice are not used. One simply selects any bar from group A, any from group B, and so on to the last letter of the alphabet for that section. A photograph of a boxed dice game appears in Plate 42 of The Oxford Companion to Music, but without mention of date, inventor, or place of publication. Apparently, it uses thirtytwo dice, their sides marked to indicate
MELODYMAKING MACHINES
91
tones, intervals, chords, modulations, and so on. There also are ivory men whose purpose, the caption reads, is "difficult to fathom." In 1822 a machine called the Kaleidacousticon was advertised in a Boston music magazine, The Euterpiad. By shuffling cards it could compose 214 million waltzes. The Componium, a pipe organ that played its own compositions, was invented by M. Winkel of Amsterdam and created a sensation when it was exhibited in Paris in 1824. Listeners could not believe that the machine actually constructed the melodies it played. Scientists from the French Academy investigated. "When this instrument has received a varied theme," their report stated, "which the inventor has had time to fix by a process of his own, it decomposes the variations of itself, and reproduces their different parts in all the orders of possible permutation . . . . None of the airs which it varies lasts above a minute; could it be supposed that but one of these airs was played without interruption, yet, the principle of variability which it possesses, it might, without ever resuming precisely the same combination, continue to play . . . during so immense a series of ages that, though figures might be brought to express them, common language could not." The report, endorsed by physicist Jean Baptiste Biot, appeared in a British musical journal (The Harmonicon 2, 1824, pp. 40  41). Winkel's machine inspired a Vienna inventor, Baron J. Giuliani, to build a similar device, the construction of which is given in detail on pages 198200 of the same volume. In 1865, a composing system called the Quadrille Melodist, invented by J. Clinton, was advertised in The Euterpiad. By shuffling a set of composing cards, a pianist at a quadrille party could "keep the evening's pleasure going by means of a modest provision of 428,000,000 quadrilles." Joseph Schillinger, a Columbia University mathematician who died in 1943, published his mathematical system of musical composition in a booklet, Kaleidophone, in 1940. George Gershwin is said to have used the system in writing Porgy and Bess. In 1940 Heitor VillaLobos, using the system, translated a silhouette of New York City's skyline into a piano composition (see Figure 38). The Schillinger System ofMusical Composition is a twovolume work by L. Dowling and A. Shaw, published by Carl Fischer in 1941. A footnote on page 673 of Schillinger's eccentric opus The Mathematical Basis of the Arts (Philosophical Library, 1948) says that he left plans for musiccomposing machines, protected by patents, but nothing is said about their construction. In the 1950s, information theory was applied to musical composition by J. R. Pierce and others. In a pioneering article "Information Theory and Melody," chemist Richard C. Pinkerton included a graph which he called the "banal tunemaker." By flipping a coin to determine paths along the network,
92
CHAPTER SEVEN
THE s x n m HAS m OWN MUSICAL PATIERN TRANSLATED FROM SILHOUElTE TO MUSIC NO=
WITH THEHELP OF
THE SCHllllNGER SYSTEM OF MUSICAL COMPOSITION Figure 38 New York skyline translated into music by VillaLobos, who used the Schillinger system
one can compose simple nursery tunes. Most of them are monotonous, but hardly more so, Pinkerton reminds us, than "A Tisket, a Tasket." During the 1960s and early 1970s the proliferation of computers and the development of sophisticated electronic tone synthesizers opened a new era in machine composition of music. It is now possible to write computer programs that go far beyond the crude devices of earlier days. Suppose one wishes to compose a melody in imitation of one by Chopin. A computer analysis is made of all Chopin melodies so that the computer has in its memory a set of "transition probabilities." These give the probability that any set of one, two, three, or more notes in a Chopin melody is followed by any other note. Of course, one must also take into account the type of melody one wishes to compose, the rhythms, the position of each note within the melody, the overall pattern, and other things. In brief, the computer makes random choices within a specified general structure, but these choices are subject to rules and weighted by Chopin's transition preferences. The result is a "Markoff chain" melody, undistinguished but nevertheless sounding curiously like Chopin. The computer can quickly dash off several hundred such pieces, from which the most pleasing may be selected.
MELODYMAKING MACHINES
93
There is now a rapidly growing literature on computer composition, not only of music in traditional styles but also music that takes full advantage of the computer's ability to synthesize weird sounds that resemble none of the sounds made by familiar instruments. Microtones, strange timbres, unbelievably complex rhythms, and harmonics are no problem. The computer is a universal musical instrument. In principle, it can produce any kind of sound the human ear is capable of hearing. Moreover, a computer can be programmed to play one of its own compositions at the same time it is composing it. How can we sum up? Computers certainly can compose mediocre music, frigid and forgettable, even though the music has the flavor of a great composer. N o one, however, has yet found an algorithm for producing even a simple melody that will be as pleasing to most people of a culture as one of their traditional popular songs. W e simply do not know what magic takes dace inside the brain of a comDoser when he creates a su~eriortune. W e do not even know to what extent a tune's merit is bound up with cultural conditioning or even with hereditary traits. About all that can be said is that a good melody is a mixture of predictable patterns and elements of surprise. What the proportions are and how the mixture is achieved, however, still eludes everybody, including composers. O'Beirne has called my attention to how closely some systems of musical r e This is a composition resemble the buzzphrase generator.(ree ~ i ~ u 39). giveaway of Honeywell Incorporated. Pick at random any fourdigit number, such as 8751, then read off phrase 8 of module A, phrase 7 of module B, and so on. The result is a SIMP (Simplified Integrated Modular Prose) sentence. "Add a few more fourdigit numbers," the instructions say, "to make a SIMP paragraph. After you have mastered the basic technique, you can realize the full potential of SIMP by arranging the modules in DACB order, BACD order, or ADCB order. In these advanced configurations, some additional commas may be required." SIMP sounds very much like authentic technical prose, but on closer inspection one discovers that something is lacking. Computermade melodies are perhaps less inane, closer to the random abstract art of a kaleidoscope, but still something essential (nobody knows what) is missing. Indeed, a !good simple tune is much harder to compose than an orchestral piece in the extreme avantgarde manner, so loaded with randomness and dissonance that one hesitates to say, as Mark Twain (or was it Bill Nye?) said of Wagner's music: It is better than it sounds. When a computer generates a melody that becomes as popular as (think of the title of your favorite song), you will know that a colossal breakthrough has been made. Will it ever occur? If so, when? Experts disagree on the answers as I
94
CHAPTER SEVEN
SIMP
table A
In particular, On the other hand, However, Similarly, As a resultant implication, In this regard, Based on integral subsystem considerations, For example, Thus, In respect to specific goals,
SIMP table
B
1. a large portion of the interface coordination communication. 2. a constant flow of effective information 3. the characterization of specific criteria 4. initiation of critical subsystem development 5. the fully integrated test program 6. the product configuration baseline 7. any associated supporting element 8. the incorporation of additional mission constraints 9. the independent functional principle 0. a primary interrelationship between subsystem and/or subsystem technologies
Figure 39
Honeywell's buzzphrase generator for writing Simplified Integrated Modular Prose (SIMP)
much as they do on if and when a computer will write a great poem, paint a great picture, or play grandmaster chess.
ADDENDUM Carousel, the company that brought out Mozart's The Melody Dicer, later issued a similar set called The Scottloplin Melody Dicer. Using the same system of dice and cards, one can compose endless rags of the Scott Joplin variety. In 1977 a curious 284page book titled The Directory of Tunes and Musical Themes,by Denys Parsons, was published in England by the mathematician G. Spencer Brown. Parsons discovered that almost every melody can be identified by a ridiculously simple method. Put down an asterisk for the first note. If the second note is higher, write down U for "up." If lower, write D for "down." If the same, use R for "repeat." Continue with succeeding notes until you have a sequence of up to sixteen letters. This is almost always sufficient to identify the melody. For example *UDDUUUU is enough to key "White Christmas." The book lists alphabetically in two sections, popular and classical, some 15,000 different sequences followed by the title of the work, its composer, and the date.
MELODYMAKING MACHINES
SIMP
table C
1. must utilize and be functionally interwoven with 2. maximizes the probability of project success and minimizes the cost and time required for 3. adds explicit performance limits to 4. necessitates that urgent consideration be applied to 5. requires considerable systems analysis and trade off studies to amve at 6. is further compounded, when taking into account 7. presents extremely interesting lenges to 8. recognizes the importance of other systems and the necessity for 9. effects a significant implementation of 0. adds ovemding performance constraints to
SIMP
95
table D
1. the sophisticated hardware 2. the anticipated fourth generation equipment 3. the subsystem compatibility testing 4. the structural design, based on system engineering concepts 5. the preliminary qualification limit 6. the evolution of specifications over a given time period 7. the philosophy ofcommonality and standardization 8. the greater fightworthinessconcept 9. any discrete configuration mode
The idea behind the buzzphrase generatorrandom selection of words and phrases to create prose or poetry is an old idea. Rational Recreations, a fourvolume work by W. Hooper (the fourth edition was published in London in 1794), has a section in Volume 2 on how to use dice for composing Latin verse. The technique surely is much older. Similar randomizing is used in modern computer programs that generate "poems" and various kinds of imitation prose. Firstgrade teachers, who call it "stringing," use the technique for teaching reading. Children are given a simple pattern sentence with blanks into which they insert words. The New York Times Book Review (June 4, 1978), printed Randolph Hogan's list of buzzwords that enable anyone to write impressive literary criticism. (Someone should do a similar list, perhaps already has, for art critics.) Mad Magazine (October 1974) featured Frank Jacobs's twelve columns of buzzwords and phrases for writing impeachment newspaper stories. Tom Koch (Mad Magazine, March 1982) gave a similar technique for standup comics. Jacobs returned in Mad (September 1982) with another twelve columns of words and phrases for writing the lyrics of countrywestern songs. For Richard Voss's application of Benoit Mandelbrot's concept of fractals to the computer composition of music, see my Scientijc American column for April 1978, and the references cited.
96
B
CHAPTER SEVEN
I
B
L
I
O
G
R
A
P
H
Y
"Information Theory and Melody." Richard C. Pinkerton, in Scientific American, February 1956, pp. 77  86. "The Ars Magna of Ramon Lull." Martin Gardner, in Logic Machines and Diagrams. McGrawHill, 1958. Rev. ed. University of Chicago Press, 1982.
Experimental Music. L. A. Hiller, Jr., and L. M. Isaacson. McGrawHill, 1959. "Computer Music." L. A. Hiller, Jr., in Scientific American, December 1959,pp. 109  120.
Music by Computers. Edited by Heinz Von Foerster and James W . Beauchamp. Wiley, 1969. Cybernetic Serendipity: T h e Computer and the Arts. Jasia Reichardt, ed. Frederick A. Praeger, 1969.
"Composition Systems and Mechanisms." Percy A. Sclioles, in The Oxford Companion to Music, John 0 . Ward, ed. Oxford University Press, 1970. "From Mozart to the Bagpipe with a Small Computer." T. H. O'Beirne, in Bulletin ofthe Institute ofMathematics and its Applications 7, January 1971, pp. 38.
Athanasius Kircher. Joscelyn Godwin. Thomas and Hudson, 1979.
EIGHT
Anarnorphic Art
Anamorphic art is a term unfamiliar to most people; indeed, it is unfamiliar to most artists. From the Greek ana (again) and m o r p h ~(form), it refers to realistic art so monstrously distorted by a projective transformation that it is difficult to recognize. The distortion can be "formed again'' by viewing it on a slant or as a reflection in a suitable mirror. The mirror, called an anamorphoscope, is usually a polished cylinder or cone. The appearance of the undistorted reflection is so magical and surprising that few people seeing it for the first time fail to exclaim in wonder. At this point the reader may want to pause to make the best cylindrical anamorphoscope he can in order to view some of the anamorphic art reproduced here and on the cover. For the best results, the cylinder should have a base that fits the circle in the picture. Acceptable results, however, can be obtained with a cylinder that has a larger or a smaller base. A small juice or soup can with the label removed and the residual glue on the can washed off may do the trick. A transparent cylindrical bottle with a cylinder of black paper inserted into it works fairly well. Chromiumplated tubing, available in some hardware stores for plumbing fixtures, is even better. Aluminum foil is not smooth and stiff enough, but silvered Mylar paper, taped around a cylinder of the right size, makes an excellent anamorphoscope. The reader is urged to get several square feet of this material (it is sold in artsupply and hobby stores),
98
CHAPTER EIGHT
not only for cylindrical viewing, but also because it can be used, as will be explained, to make a conical anamorphoscope. It was in the early Renaissance that European painters, who were just beginning to master perspective, became fascinated by the simplest kind of anamorphic art: stretched pictures that are seen correctly when viewed on a slant. ~ i l first e known examples are in Leonardo da Vinci's notebooks; this is not surprising, because Leonardo was one of the earliest contributors to the geometry of perspective. Surfaces viewed on a slant are, of course, anamorphically distorted even though we are usually not aware of it. A door seen from a certain angle is a trapezoid, but our brain, conditioned by experience, perceives it as a tilted rectangle. When people who are not used to television see a television screen from the side, the images appear to be too skinny. The rest of us have learned to correct for this bias so well that squeezed images on television seem normal. When the Renaissance painters discovered how to transform flat shapes to give a depth illusion to the canvas, they discovered simultaneously how to do it in reverse. A picture stretched according to the same rules of perspective becomes a grotesque form. Hans Holbein's painting The Ambassadors (1533) contains a famous example of anamorphic art (see Figure 40). You can see the stretched shape at the bottom of the painting normally by closing one eye and slanting the page away from you, with the lower left corner of the page pointing toward your open eye and about six inches from it. Another way to see the skull is to place the edge of a flat mirror about three inches from the lower left corner and to look into the mirror with both eyes while tipping it until the skull appears normal. Holbein's painting was probably intended to hang near the top of a stairway so that people going up would be startled by the skull. Another slant picture is an old newspaper puzzle by Sam Loyd (see Figure 41). It has a concealed portrait of the mature George Washington. Can you find it? (A second puzzle consists in dividing the square Washington pie into six square pieces, not necessarily the same size.) Slant pictures of this kind occasionally appear in children's books and on advertising premiums. Sometimes, printing is stretched out so it can be read only by slanting the page. This technique is often used for the word STOP on streets so that the letters appear normal to a driver approaching the intersection. The geometric technique for drawing slant pictures was explained in detail in the first major treatise on anamorphic art: La Perspective Curieuse, by Jean Franqois Niceron (Paris, 1638). The picture is first ruled into square cells. The matrix is stretched to a trapezoid, then the artist copies the picture by filling in the trapezoidal cells, stretching the contents of each cell as accurately as he can to match its corresponding cell on the original picture. The finer the matrix, the more accurate the copy.
ANAMORPHIC ART
99
Figure 40 The Ambassadors, a slant anamorphic painting by Hans Holbein
The exact shape of the trapezoidal matrix depends on the position of the eye when it sees the shape as a normal square. The full threedimensional structure is complex, but it turns out that there is a simple way to construct the trapezoid, given the desired position of the eye. Consider a square of side 8 ruled into 64 squareinch cells. W e want to distort it so that it will appear normal when the eye is 25 units from the midpoint of the picture's top edge and seven units above the plane (see Figure 42). The construction is as follows. XY is the square's width. FB, on the perpendicular bisector of XY, is 25 units. BE, perpendicular to FB, is 7 units. Draw XB, YB, and YE. This determines CD, the trapezoid's bottom edge. Other lines from B andE to the unit marks onXY
100
CHAPTER EIGHT
Figure 4 1 A Sam Loyd puzzle with a hidden anamorphic picture
determine the lines to XY that complete the 64 trapezoidal cells. Neither E nor B indicates the eye's position. The eye is 7 units above the horizontal plane of the paper. A perpendicular dropped from the eye to the plane intersects FB at point G. The construction assumes that G is at least 8 units from F. Another way to draw the picture is to close one eye, view the paper on a bias, and then draw the picture so that it looks normal. This is better than trying to do it with a mirror, because in the mirror your hand moves along one coordinate in a direction opposite to the way it moves on the actual sheet, and your reflexes find that hard to manage. A simple photographic method of making slant pictures is to project the picture (with an enlarger or a slide projector) onto enlarger paper so that the light strikes the paper at the angle intended for viewing. Although there is no evidence that the ancient Greeks played with anamorphic pictures, they sometimes deformed the columns of temples to correct the distortion perceived by someone near the front of the building. For similar reasons, Renaissance painters occasionally deformed murals so that viewers looking up at them would see them with less distortion. Slant pictures were sometimes concealed in paintings or stretched along the side of a long corridor to be viewed from an entrance or an exit. Another popular practice was to put slant pictures inside boxes with a peephole at one end for viewing the picture at the proper angle.
101
ANAMORPHIC ART
Anamorphic paintings for cylindrical and conical mirrors were fashionable toys in both Europe and the Orient during the seventeenth and eighteenth centuries. They were usually done by anonymous artists and were sold with handsomely made anamorphoscopes. Occasionally the pictures carried messages of political protest; at times, they were pornographic or scatological. Several examples of erotic anamorphic pictures appear in ChineseErotic Art, by Michel Beurdeley et al. (Charles E. Tuttle Co., 1969). This early anamorphic art is now a collector's item. In December 1973 a set of ten oil paintines, for both cylinders and cones, from eighteenthcentury France was sold at a Sotheby ParkeBernet auction in New York for $10,800, which is a bargain in view of today's prices. Herbert Tannenbaum, a New York art dealer, had found the paintings in 1939 in a curiosity shop in Amsterdam and had bought them without even knowing what they were. One of these paintings is shown in Figure 43. Figure 44 produces the ten of hearts when viewed in a cylinder. T o make a conical anamorphoscope, cut a circular disk of the proper size from silvered Mylar, cut a radius, overlap the cut edges, and then glue or tape the overlapping edges in place. For the illustration in Figure 45, designed for conical viewing, the radius of the disk that makes a cone of the right propor
I
Figure 42
25
Geometric method of making slant anamorphic pictures
*
102
1
CHAPTER EIGHT
Figure 43 With a Tender Little Song, a cylindrical anamorphic painting from the Tannenbaum Collection
ANAMORPHIC ART
Figure 44
103
Cylindrical anamorphic picture
tions is about 1inch. Adjust the overlap until the base of the cone fits the inner circle on the picture. Place the cone on this circle and view directly from above with one eye. The restored picture is small and circular, completely within the circumference of the cone. If you press on the apex of the cone with a fingertip or a paper clip, it will make the cone more rigid and produce a better picture. For ideal results, you need a solid conical mirror made with great accuracy. As with slant pictures, there are three ways to deform a picture for cylindrical or conical viewing. The geometric procedures, reproduced from the Niceron treatise cited above, are shown in Figures 46 and 47. The methods for an exact construction of the distorted matrix are complicated; interested readers can do no better than to consult the Niceron book for details. Note how the conical reflection literally turns a picture inside out. Point A, at the center of the picture in Figure 47, becomes the circumference of the distorted drawing, and the original circumference becomes the inner circle of the distortion. When Salvador Dali made a set of erotic anamorphic paintings (prints were sold in Switzerland with a cylindrical anamorphoscope), he simply looked into the cylinder while he painted on the surface under it. That is not easy, because in the mirror your hand motion is reversed. You see what you are doing right side up, but your hand must paint the picture upside down. One can make crude anamorphic photographic prints by wrapping a negative halfway around a transparent cylinder and by sending slanting light
104
CHAPTER EIGHT
Figure 45 Newman
Venus and Adonis, a conical anamorphic painting. Photograph O Arnold
through the cylinder from a point source outside and behind the cylinder to project the picture onto enlarging paper under the cylinder. More accurate prints are made by projecting the picture onto the side of an accurately made cylindrical mirror so that it reflects to the enlarging paper (see Figure 48). One should use an enlarger or a 35millimeter slide projector with a diaphragm at the lens, and stop down the lens until the image is sharp on the easel. All light from the enlarger that does not fall directly on the cylinder should be blocked off by black cardboard with a rectangular hole. Conical prints are made in similar fashion. The mirror cone should be large (about 6 inches in diameter).
ANAMORPHIC ART
105
The picture is projected straight down on top of the cone through a circular hole in black cardboard to block off extraneous light. The term anamorphic is used by photographers for any lens that stretches or compresses an image along one coordinate, as well as for the deformed images it produces. In 1953, Twentieth CenturyFox introduced the wide screen with its motion picture TheRobe. Anamorphic lenses squeezed the wide image onto standard 35mm. film, then anamorphic lenses in the projector stretched the image back to fit the wide screen. Most motion pictu;es today are taken and projected with sophisticated anamorphic lens systems. Similar systems adapt widescreen motion pictures to videotape. Psychologists who study perception have experimented with threedimensional anamorphic models of chairs, tables, and other objects. The deformed models appear to be normal when seen from a certain angle. The Ames room is a radically distorted room that seems to be normal when viewed through  a hole in the wall. A person in the room appears to grow or shrink when he or she moves from one part of the room to another (see "Experiments in Perception," by W. H. Ittelson and F. P. Kilpatrick; Scientific American, August
Figure 46 Jean Franqois Niceron's method of drawing cylindrical anamorphic pictures
106
CHAPTER EIGHT
Figure 47
Niceron's method of drawing conical anamorphic pictures
1951). Seventeenthcentury architects did not discover the Ames room, but they did play games with false perspective. The most startling example, which can still be seen in the Palazzo Spada in Rome, is an anamorphic arcade designed by Francesco Borromini about 1638. You seem to be looking down a longcorridor at a large statue beyond the exit. Actually, the deformedcorridor is only 28 feet long, and the statue is 3 feet high. The illusion was created by making the entrance 19 feet high and 10 feet wide and the exit only 8 feet high and 31/3 feet wide (a trick, by the way, long familiar to designers of equipment for stage magicians). There are many other forms of flat anamorphic art: pictures to be reflected in spheres, in properly laced nsided and other polyhedrons, and pictures to be seen through various kinds of distorting lenses. The wavy mirrors in fun houses produce anamorphic images. What is a good caricature if not a complex set of anamorphic distortions that our mind sees as more like the person than the actual person? And there are extreme ways to transform a picture and to restore it again (a hologram, for instance, or the broadcasting of a television image), but the term anamorphic is best confined to coordinate  . transformations, particularly of the three types we have considered. Map makers do not use the term anamorphic, but the many ways in which the earth's surface is projected on the planecylindrically, conically, and othenvise are coordinate transformations closely related to anamorphic art.
ANAMORPHIC ART
107
Botanists apply the word anamorphic to radical changes that certain plants undergo when they are grown in different environments. Zoologists have used the term for the evolutionary modifications of animal forms. D'Arcy Wentworth Thompson's classic work On Growth and Form (Cambridge University Press, 1961) has a chapter filled with diagrams showing animal species that differ from one another by anamorphic distortions so much like the types discussed here that if you view, say, one species of fish on a slant or in a cylindrical mirror, it becomes identical with another species. Similarly, a profile of the human skull, mildly anamorphosed, becomes the skull of a chimpanzee or a baboon. The ability of our visual system to correct anamorphic distortion suggests that vision is concerned more with topologically invariant properties than
Figure 48 Cylindrical anamorphic photograph by Alan Fontaine
108
CHAPTER EIGHT
with Euclidean ones. The visual system not only utilizes the upsidedown images on the two retinas in such a way as to provide a rightsideup impression of a threedimensional world but also corrects for anamorphic distortion by irregular lenses and corneas. A person with marked astigmatism, fitted for the first time with glasses, perceives the world as being deformed because his brain is still correcting for the old distortions. It may take weeks before he sees the world normally again. Experiments have been conducted in which the subject wears special glasses that produce extreme topological transformations. After a few weeks and headaches, the world begins to look normal once again. When the glasses are removed, the world looks distorted, although fortunately only for a short time. Hamlet advised some actors to hold the mirror up to nature. Is the mirror of a great play, novel, painting, or motion picture a distorting mirror or is it a magic anamorphoscope that gives pleasing form to an ugly, shapeless world? Are philosophical systems and religions, even the views of crazy little cults, anamorphic distortions of truth or are they, too, anamorphoscopes designed to give meaning to a meaningless reality? "It was to correct their anamorphosis of the Deity," wrote Thomas ~efferson,"that Jesus preached." T o the outsider, a system of beliefs appears to twist truth like a grotesque anamor~hicpainting. T o the insider, who sees the world in the specially shaped mirror of his perceptual system, everything appears to be normal. Is there a metaphysical system that reflects truth like a flat, untipped mirror? Alas, every true believer is convinced his own anamorphoscope is precisely that.
ANSWERS Although it has nothing to do with anamorphic art, Figure 41 contains a dissection puzzle that I did not answer in my column, but will do so here. It is one of Loyd's jokes. One assumes the square is to be divided along the lattice lines. Here is how Loyd answered the problem in his famous Cyclopedia of 5,000 Puzzles (Morningside Press, 19 14). "The simplest way to cut a square into six squares is to mark it off into nine squares, then the largest one will be made up of four squares, and there will be five more little ones." It is not only the simplest, it is the only solution. For a discussion of the general problem of dissecting squares into smaller squares, not necessarily alike, see the chapter on "Mrs. Perkins' Quilt" in my Mathematical Carnival (Knopf, 1975).
ANAh4ORPHIC ART
B
I
B
L
I
O
G
R
A
P
H
109
Y
The Magic Mirror: A n Antique Optical Toy. McLoughlin Brothers, ca. 1900. Dover reprint, 1979. "Speaking of Pictures: Distorted Paintings Must Be Seen in Special Mirrors to Make Sense." Life 27, September 12 1949, pp. 1819. "The Grand Illusion." Fabrizio Clerici, in Art News Annual 23, 1954, pp. 98 180.
Anamorphoses: O u Magie Artificielle Des Efets Merveilleux. Jurgis BaltruSaitis. Olivier Perrin Editeur, 1969. "Anamorphic Art." Alan Fontaine. Lithopinion 7, Fall 1972, pp. 5059.
Anamorphoses: Games of Perception and Illusion in Art. Joost Elffers, Fred Leeman, and Michael Schuyt. Abrams, 1976. "FunFair Illusions." Robert Hughes, in Time, October 4, 1976, pp. 9293. "Portrait of a Man Standing Before a Balustrade." Reprinting as ajigsaw puzzle of a 1630 Swedish painting, to be viewed in a cylindrical mirror. Abrams, 1977. "Anamorphoscopes: A Visual Aid for Circle Inversion." Philip W . Kuchel, in The Mathematical Gazette 63, June 1979, pp. 82 89. "Anamorphic Pictures." Jearl Walker, "The Amateur Scientist" column, in Scientific American, July 1981, pp. 176 187.
NINE
T h e Rubber Rope and Other Pro blerns 1. THE RUBBER ROPE A worm is at one end of a rubber rope that can be stretched indefinitely (see Figure 49). Initially the rope is one kilometer long. The worm crawls along the rope toward the other end at a constant rate of one centimeter per second. At the end of each second the rope is instantly stretched another kilometer. Thus, after the first second, the worm has traveled one centimeter and the length of the rope has become two kilometers. After the second second, the worm has crawled another centimeter and the rope has become three kilometers long, and so on. The stretching is uniform, like the stretching of a rubber band. Only the rope stretches. Units of length and time remain constant. W e assume an ideal worm, ofpoint size, that never dies, and an ideal rope that can stretch as long as needed. Does the worm ever reach the end of the rope? If it does, estimate how long the trip will take and how long the rope will be. This delightful problem, which has the flavor of a Zeno paradox, was devised by Denys Wilquin of New Caledonia. It first appeared in December 1972 in Pierre Berloquin's lively puzzle column in the French monthly Science et V i e .
112
CHAPTER NINE
Figure 49 A worm crawling on a stretching rubber rope
2. THE SIGIL OF SCOTEIA One of James Branch Cabell's finest novels, The Cream of theJest, involves a series of hypnotic dreams that Felix Kennaston induces by staring at half of the Sigil of Scoteia. At the end of the novel, Kennaston discovers that the Sigil is nothing more than the broken top of a coldcream jar designed by someone who "just made it up out of his head." It "is in no known alphabet. It blends meaningless curlicues and dots and circles with an irresponsible hand." The artist had sketched a crack across it "just to make it look ancient like." A picture of the complete Sigil, which appears in most editions of the book, is reproduced in Figure 50. I remember puzzling over it many years ago and vainly trying to decode it. Designed by Cabell himself, it is not a cipher at all. Can the reader read it? Cabell is still out of favor with most literary critics, but he has a loyal band of admirers who support the James Branch Cabell Society and its official organ, Kalki. It was in the sixth issue of Kalki (1968) that I learned the Sigil's secret.
3. INTEGERCHOICE GAME Two mathematicians are drinking beer. After each has been served a glass, they arrive at who pays for the round by simultaneously writing any positive integer on slips of paper. They compare the slips. Whoever wrote the larger integer has to pay for the round, unless the integer is larger by only 1. In that case the person who wrote the smaller integer pays for that round and the next as well. If both players have chosen the same integer, they play again. T o describe the game positively: The person who writes the smaller number scores a point, unless it is smaller by only 1. In that case the other player scores two points. For example, if the numbers were 12 and 20,12 would win a point. If the numbers were 12 and 13, 1 3 would win two points. The game is fair, but what strategy is best in the sense that no other strategy can beat it in the long run, and if any different strategy is followed, there is a counterstrategy to beat it? The answer is surprising. I shall not have space for a proof, but I shall give the strategy and references to where the proof can be found. The game was
THE RUBBER ROPE AND OTHER PROBLEMS
113
devised by N. S. Mendelsohn and Irving Kaplansky, but it was Paul Halmos who called it to my attention. 4. THREE CIRCLES Draw three nonoverlapping circles of three different sizes anywhere on a piece of paper. For each pair of circles draw their two common tangents. Ifyou have not encountered this beautiful theorem before, you may be surprised to find that the intersections of the three pairs of tangents lie on a straight line (see Figure 51). As one would expect, there are many ways the theorem can be proved by adding construction lines to the figure. Popular Computing reported in its issue
Figure 50 T h e Sigil of Scoteia
114
CHAPTER NINE
Figure 5 1 The threecircle problem
of December 1974, however, that the theorem lends itself to an elegant solution if one leaves the twodimensional plane for a threedimensional extension. Quoting from an earlier book in which they found the problem, the editors of the magazine report that when the theorem was shown to John Edson Sweet, a professor of engineering at Cornell University who died in 1916, he studied the picture for a moment and then said, "Yes, that's perfectly selfevident." What was Professor Sweet's sweet solution?
5. THE MUTILATED SCORE SHEET Figure 52 reproduces the score sheet of a chess game played in a German chess club in 1897. As you can see, a cigar or a cigarette has burned a hole in the
THE RUBBER ROPE AND OTHER PROBLEMS
115
sheet, so that Black's moves, which ended in Black's win on his fourth move, have been obliterated. Can you reconstruct the game? I wish to thank Randolph W. Banner of Newport Beach, Calif., for passing along this amusing problem. He writes that he thinks he saw it in an English periodical published about 1920. 6. SELFNUMBERS
D. R. Kaprekar is a mathematician, diminutive in body but large in brain and heart, who lives in India. For more than forty years he has been doing highly original work in recreational number theory, at times aided by grants from Indian universities. He contributes frequently to Indian mathematics journals, speaks at conferences, and has published some two dozen booklets written in broken English. Kaprekar is best known outside India for his discovery, more than twenty years ago, of "Kaprekar's constant." Start with any fourdigit number in which not all the digits are alike. Arrange the digits in descending order,
Figure 52 What were Black's moves?
116
CHAPTER NINE
reverse them to make a new number, and subtract the new number from the first number. If YOU keep repeating this process with the remainders, you will (in eight steps or fewer) arrive at Kaprekar's constant, 6174, which then generates itself. Zeros must be preserved. Thus, if you start with 21 11 and subtract 1112, you get 0999. Rearranging the digits gives 9990 from which 0999 is taken, and so on. Here we concern ourselves with a remarkable class of numbers called selfnumbers, discovered by Kaprekar in 1949. He has written many pamphlets about them. They are virtually unknown outside India, although last year they turned up briefly (under another name) in an article in The American Mathematical Monthly (April 1974, p. 407). The article contains a proof that there is an infinity of selfnumbers. In explaining selfnumbers, it is best to start with a basic procedure that Kaprekar calls digitadition. Select any positive integer and add to it the sum of its digits. Take 47, for example. The sum of 4 and 7 is 11, and 47 and 11 is 58. The new number, 58, is called a !generated number. The original number, 47, is the generator. The process can be repeated endlessly, forming a digitadition series: 47, 58, 71, 79, 95, . . . . No one has yet found a nonrecursive formula for the partial sum of a digitadition series, given its first and last terms, but there is a simple formula for the sum of all the digits in a digitadition series. Simply subtract the first number from the last and add the sum of the digits in the last number. "Is this not a wonderful new result?" Kaprekar asks in one of his booklets. "The Proof of all this rule is very easy and I have completely written it with me. But as soon as the proof is seen the charm of the whole process is lost, and so I do not wish to give it just now." Can a generated number have more than one generator? Yes, but not until the number exceeds 100. The smallest such number (Kaprekar calls it a junction number) is 101. It has two generators: 91 and 100. The smallest junction number with three generators is 10,000,000,000,001. It is generated by 10,000,000,000,000, 9,999,999,999,901 and 9,999,999,999,892. The smallest number with four generators, discovered by Kaprekar on June 7, 1961, has twentyfive digits. It is 1 followed by twentyone zeros and 102. Since then he has found what he conjectures to be the smallest numbers with five and six generators. A selfnumber is simply a number that has no generator. In Kaprekar's words, "It is self born." There is an infinity of such numbers, but they are much scarcer than generated numbers. Below 100 there are thirteen: 1,3,5,7, 9, 20, 31, 42, 53, 64, 75, 86, and 97. Selfnumbers that are prime are called selfprimes. The familiar cyclic number 142,857 is a selfnumber (multiply it
THE RUBBER ROPE AND OTHER PROBLEMS
117
by the digits 1 through 6 and you always get the same six digits in the same cyclic order). The numbers 11,111,111,111,111,111 and 3,333,333,333 are selfnumbers. Previous years of this century that are selfnumbers are 1906, 1917,1919, 1930,1941, 1952,1963, and 1974. Consider the powers of ten. The number 10 is generated by 5, 100 by 86, 1000 by 977, 10,000 by 9968, and 100,000 by 99,959. W h y is a millionaire such an important man? Because, answers Kaprekar, 1,000,000 is a selfnumber! The next power of ten that is a selfnumber is 1016. N o one has yet discovered a nonrecursive formula that generates all selfnumbers, but Kaprekar has a simple algorithm by which any number can be tested to determine whether it is selfborn or generated. Readers are asked to see if they can discover the procedure. If they can, it should be easy for them to answer this question: What is the next year after 1974 that is a selfnumber?
7. THE COLORED POKER CHIPS What is the smallest number of poker chips that can be placed on a flat surface so that chips of three different colors are required to meet the condition that no two touching chips are the same color? As Figure 53 shows, the answer is obviously three. Our problem is to determine the smallest number of chips, all of the same size, that can be placed flat on the plane so that chips of not three but four colors are necessary for meeting the same proviso.
Figure 53 Mapcoloring problem with poker chips
118
CHAPTER NINE
8. ROLLING CUBES For this beautiful combinatorial puzzle, invented by John Harris of Santa Barbara, Calif., you must obtain eight unit cubes. O n each cube, color one face and make the opposite face black. (Of course, you may distinguish the two faces in any other way you like.) Place the cubes in a shallow 3by3 box (or on a 3by3 matrix) with the middle cell vacant and all cubes black on top (see Figure 54). A move consists in rolling a cube to an empty cell by tipping it over one of its four bottom edges so that it makes a quarter turn. The problem is to invert all eight cubes so their colored sides are up and the center cell is vacant as before. This is to be done in a minimum number of moves. For recording solutions, you can use U, D, L, R for roll up, down, left, and right, and start all solutions with URD.(Any other way of starting is symmetrically the same.)
ANSWERS 1. Regardless of the parameters (the initial length of the rubber rope, the worm's speed, and how much the rope stretches after each unit of time), the worm will reach the end of the rope in a finite time. This is also true if the stretching is a continuous process, at a steady rate, but the problem is easier to analyze if the stretching is done in discrete steps. One is tempted to think one can see that the worm will make it. Since the rope expands uniformly, like a rubber band, the expansion is like looking at the rope through increasingly strong magnifying lenses. Because the worm is
Figure 54 Rollingcubes puzzle
THE RUBBER ROPE AND OTHER PROBLEMS
119
always making progress, must it not eventually reach the end? Not necessarily. One can progress steadily toward a goal forever without ever reaching it. The worm's progress is measured by a series of decreasing fractions of the rope's length. The series could be infinite and yet converge at a point far short of the end of the rope. Indeed, such is the case if the rope stretches by doubling its length after each second. The worm, however, does make it. There are 100,000 centimeters in a kilometer, so that at the end of the first second, the worm has traveled 1/100,00Oth of the rope's length. During the next second, the worm travels (from its previous spot after the stretching) a distance of 1/2OO,OOOth of the rope's length, after the rope has stretched to two kilometers. During the third second it covers 1/3OO,OOOth of the rope (now three kilometers), and so on. The worm's progress, expressed as fractional parts of the entire rope, is
The series inside parentheses is the familiar harmonic one that diverges and therefore can have a sum as large as we please. The partial sum of the harmonic series is never an integer. As soon as it exceeds 100,000, however, the above expression will exceed 1, which means that the worm has reached the end of the rope. The number of terms, n, in this partial harmonic series will be the number of seconds that have elapsed. Since the worm moves one centimeter per second, n is also the final length of the rope in kilometers. This enormous number, correct to within one minute, is
where y is Euler's constant. It gives a length of rope that vastly exceeds the diameter of the known universe, and a time that vastly exceeds present estimates of the age of the universe. (For a derivation of the formula see "Partial Sums of the Harmonic Series," by R. P. Boas, Jr., and J. W . Wrench, Jr., American Mathematical Monthly 78, October 1971, pp. 864 870.) Some notion of the enormous length of the rope when the worm finally reaches the end can be gained from an observation by reader H. E. Rorschach. If the rope starts with a crosssectional area of one square kilometer, it ends up as a single line of atoms, the space between each adjacent pair of atoms being many times the size of the known universe. The time lapse is comparably greater than the age of the universe.
120
CHAPTER NINE
2. As David M. Keller disclosed in his article "The Sigil of Scoteia" (Kalki 2, 1968), one simply turns the Sigil upside down. "Additional difficulties are found in the division of words at the ends of lines," Keller writes, "and in the substitution of odd characters for some of the letters." The Sigil reads: "James Branch Cabell made this book so that he who will may read the story of mans eternally unsatisfied hunger in search of beauty. Ettarre stays inaccessible always and her lovliness is his to look on only in his dreams. All men she must evade at the last and many are the ways of her elusion." 3. It is hard to believe, but the best strategy in the integerchoosing game is to limit one's choices of numbers to 1,2,3,4, and 5. The selection is made at random, with the relative frequencies of 1/16 for numbers 1 and 5,4/16 for number 3, and 5/16 for numbers 2 and 4. One could have in one's lap a spinner designed for picking numbers according to these frequencies. For a roof of the strategy see "A Psychological Game," by N. S. Mendelsohn (American Mathematical Monthly 53, February 1946, pp. 8688) and pages 212215 of I. N. Herstein and I. Kaplansky's Matters Mathematical (Harper & Row, 1974). Walter Stromquist, in a letter, proposed using a pair of dice as follows: "After a few beers, you cannot be expected to distinguish the fives and sixes, so that if either of these numbers appear on either die, you will have to roll again. Also, since you are really using only 8 of each die, it is only natural to multiply the total by 2 (dropping all fractions) before writing it down. For example, the largest number you can roll with two dice (without rolling again) is 8, so that the largest number you would ever choose is 3 of that, or 5. Out of 16 plays, you should expect to choose with these frequencies: 1 once, 2 five times, 3 four times, 4 five times, 5 once." These are precisely the desired frequencies for playing the best strategy. 4. John Edson Sweet's solution to the threecircle theorem is given in the answer to Problem 62 in L. A. Graham's Ingenious Mathematical Problems and Methods (Dover, 1959). Instead of circles, suppose you are looking down on three unequal spheres. The tangent lines for each pair of balls are the edges of three cones into which the two balls fit snugly. T h e cones rest on the plane that supports the balls, and the apexes of the cones therefore lie on the plane. Now imagine that a flat plate is placed on top of all three balls. Its underside is a second plane, tangent to all three balls, and tangent to all three cones. This second plane also will contain the three apexes of the cones. Because the apexes lie on both planes, they must lie on the intersection of the two planes, which of course is a straight line. C. Stanley Ogilvy wrote to say he had included the threecircle problem in his book Excursions in Geometry (Oxford University Press, 1969). He too assumed that the proof by way of the three cones took care of all cases; but one
THE RUBBER ROPE AND OTHER PROBLEMS
121
day, after giving the problem to his class at Hamilton College, a student pointed out that the proof does not apply when a small sphere is between two larger ones. In such cases it is not possible for the two intersecting planes to be mutually tangent to all three spheres. Many readers sent in other ways of proving the theorem. Bernard F. Burke, Richard I. Felver, Clyde E. Holvenstot, David B. Shear, and Radu Vero each proposed turning the drawing so that the line (on which the intersections of the tangent pairs lie) is horizontal and above the circles. The circles can now be viewed as being equal spheres inside three mutually intersecting pipes of identical circular cross section, seen in perspective. The tangent lines become the parallel sides of the three pipes. Since the pipes all must rest on a plane, their parallel sides seen in perspective will all have vanishing points on the horizon line. It is not necessary that the circles be nonintersecting; indeed, the theorem can be stated in a more general way, in terms of "centers of similitude" instead of tangents, to hold for circles that lie entirely within one another. I am indebted to Donald Keeler for explaining this, as well as for pointing out that the theorem is known as "Monge's theorem" after the French mathematician and friend of Napoleon, Gaspard Monge, who gave it in a 1798 treatise. R. C. Archibald, in The American MathematicalMonthly (vol. 22, 1915, p. 65) traced the theorem back to the ancient Greeks (writes Keeler). Daniel Sleator found that the theorem has an analog with four spheres in space. Each of the four triplets will have the apexes of its three cones on a straight line. Because these four lines intersect each other at six points, the four lines must be coplanar. Therefore, the vertexes of the six cones all lie on a plane. The theorem generalizes to all higher Euclidian dimensions. (See "Monge's Theorem in Many Dimensions," by Richard Walker, in The Mathematical Gazette 60, October 1976, pp. 185 188.) Monge's theorem, for three circles on the plane, is mentioned in Herbert Spencer's autobiography. It is, writes Spencer, "a truth which I never contemplate without being struck by its beauty at the same time that it excites feelings of wonder and of awe: the fact that apparently unrelated circles should in every case be held together by this plexus of relations, seeming so utterly incomprehensible." 5. The obliterated chess game is:
1. 2. 3. 4.
PKB3 KB2 KKT.~ KR4
1. 2. 3. 4.
P  K 4 o r K3 QB3 QxP (check) BK2 (mate)
122
CHAPTER NINE
Several readers thought they had found a second solution. Black's first move is P 4 4 . His second move is either P KN3, PKR4, or N KB3. (Black's first two moves may be interchanged.) O n his third move Black checks with Q 4 3 , then presumably mates with Q KB5. Unfortunately, the mate can be thwarted by White's fourth move, P  KN4. 6. D. R. Ka~rekar'smethod of testing a number, N, to see if it is J selfnumber is as follows. Obtain N's digital root by adding its digits, then adding the digits of the result, and so on, until only one digit remains. If the digital root is odd, add 9 to it and divide by 2. If it is even, simply divide by 2. In either case call the result C. Subtract C from N. Check the remainder to see if it generates N.If it does not, subtract 9 from the last result and check again. Continue subtracting 9's, each time checking the result to see if it generates N. If this fails to produce a generator of N in k steps, where k is the number of digits in N, then N is a selfnumber. For example, we want to test the year 1975. Its digital root, 4, is even, so that we divide 4 by 2 to obtain C = 2.1975 minus 2 is 1973, which fails to generate 1975. 1973 minus 9 is 1964. This also fails. But 1964 minus 9 is 1955, and 1955 plus the sum of its digits, 20, is 1975; therefore 1975 is a generated number. Since 1975 has four digits, we would have had only one more step to go to settle the matter. With this simple procedure, it does not take long to determine that the next selfyear after 1974 is 1985. There will be only one more selfyear in this century; 1996. For progress on the problem of finding a nonrecursive formula for the sum of a digitadition series, see "The Sum of a Digitadition Series," by Kenneth B. Stolarsky (Proceedings of the American Mathematical Society 59, August 1976, pp. 15). Among his references on digitadition, the earliest is a 1906 French article. 7. It is easy to prove that the pattern of 11 circles (seeFigure 55) requires at least four colors to ensure that no pair of touching circles are the same color. Assume that it can be done with three colors. Label circles 1, 2, and 3 with colors A, B, and C as shown. This determines the colors of 4, 5, and 6. W e have a choice of two ways to color 7, but either way forces 11to have the same color as 1, which it touches. Three colors are therefore not enough. A number of readers sent proofs that 11 is the minimal number of circles. A proof by Allen J. Schwenk was published in The American Mathematical Monthly (vol. 83, June 1976, pp. 485  486) as a solution to part of question E 2527. In three dimensions the least number of colors needed for any arrangement of identical spheres is not known, although it has been narrowed to 5 , 6 , 7 , 8 ,
THE RUBBER ROPE AND OTHER PROBLEMS
Figure 55
123
Solution to the pokerchip problem
or 9. For this and many other unsolved coloring problems, see "Coloring of Circles," by Brad Jackson and Gerhard Ringel (The American Mathematical Monthly 91, January 1984, pp. 4249). 8. In his article "Single Vacancy Rolling Cube Problems" Journalof Recreational Mathematics 7 , Summer 1974, pp. 220  224), John Harris gave a thirtyeightmove solution to his rolling cube puzzle. About a dozen readers who wrote computer programs for the problem found a unique minimalmove solution in thirtysix moves. (Reversals, rotations, and reflections are not considered different.) The solution is:
U R D LLD RRU LDL U R D RUL DLU U R D RUL DRD LUL DRU Harris ends his article with a difficultproblem that also involves eight cubes on an order3 matrix. Color the cubes so that when they are on the matrix, with the center cell vacant, every exposed face is red and all hidden faces are uncolored. There will be just 24 red sides and 24 uncolored sides. The problem is to roll the cubes until they are back on the same eight cells, with the center cell vacant but with all redsides hidden and all visible sides uncolored.
124 • CHAPTER NINE
Harris reported a solution in eightyfour moves, then later lowered it to seventyfour. In May 1981 I received a letter from Hikoe Enomoto, Kiyoshi Ishihata, and Satoru Kawai, all in the Department of Information Science, University of Tokyo. Their computer program produced the following minimalmove solution in seventy steps: DRUUL LDLDR RULDR RURDL RDLUR LDRUL RULDL
DDRUU ULURD DLULD ULURD DLURU DLURD DRRUL
For two earlier rolling cube puzzles invented by Harris, and a description of a board game based on rolling cubes, see Chapter 9 of my Mathematical Carnival (Knopf, 1975).
TEN
Six Sensational Discoveries
Has the steady rise of public interest in occultism and pseudoscience over the past ten years in the U.S. been something apart from public understanding of scientific knowledge? The two have interacted more strongly than most people realize. Important advances in science have been crowded out of newspapers, magazines, radio, and television to make room for reports on poltergeists, demon possession, psychic healing, prehistoric visits to the earth by astronauts from other worlds, the vanishing of ships and planes in the Bermuda Triangle, the emotional life of plants, the primal scream, and so on ad nauseam. The effect is intensified by an increasing backlog of articles submitted to scientific journals. It is not unusual for several years to elapse between the acceptance of a scientific paper and its publication. In the meantime the author of an unpublished article about an important new discovery may keep his results secret for fear that a rival colleague might steal them and publish first. As a public service, I shall comment briefly on six major discoveries of 1974 that for one reason or another were inadequately reported to both the scientific community and the public at large. The most sensational of last year's discoveries in pure mathematics was surely the finding of a counterexample to the notorious fourcolormap conjecture. That theorem, as all readers of this
126
CHAPTER TEN
book must know, is that four colors are both necessary and sufficient for coloring all planar maps so that no two regons with a common boundary are the same color. It is easy to construct maps that require at least four colors, and topologists long ago proved that five colors are enough to color any map. Closing the gap, however, had eluded the greatest minds in mathematics. Most mathematicians have believed that the fourcolor theorem is true and that eventually it would be established. A few suggested it might be Godelundecidable. H. S. hf. Coxeter, a geometer at the University of Toronto, stood almost alone in believing that the conjecture is false. Coxeter's insight has now been vindcated. In November 1974 William McGrcgor, a graph theorist of Wappingcrs Falls, N.Y., constructed a map of 110 regions that cannot be colored with fewer than five colors (see Figure 56). McGrcgor7s technical report will appear in 1978 in the Journal of Combinatorial Theor_, Series B. In number theory the most exciting Qscovery of the past year is that when the the transcendental number e is raised to the power of z times result is an integer. The Indian mathematician Srinivasa Ramanujan had conjectured that e to the power of is integral in a note in the Quartedj Journal of Pare and Applied &lathematics (vol. 45, 19131 914, p. 350). Working by hand, he found the value to be 262,537,412,640,768,743.999,999,999,999,.. .. The calculations were teQous, and he was unable to verify the next decimal Qgit. Modern computers extended the 9's much farther; indeed, a French program of 1972 went as far as two million 9's. Unfortunately, no one was able to prove that the sequence of 9's continues forever (which, of course, would make the number integral) or whether the number is irrational or an integral fraction. In May 1974 John Brillo of the University of ,Irizona found an ingenious way of applying Euler's constant to the calculation and managed to prove that the number exactly equals 262,537,412,640,768,744. How the prime number 163 manages to convert the expression to an integer is not yet fully understood. Brillo's proof is scheduled to appear in a few years in Mathematics of Computation. There were rumors late in 1974 that ~c would soon be calculated to six million decimal places. This may seen impressive to laymen, but it is a mere computer hiccup compared wit11 the achievement of a specialpurpose chessplaying computer built in 1973 by the Artificial Intelligence Laboratory at the hfassachusetts Institute of Technology. Richard Pinlileaf, who designed the computer with the help of exworldchesschampion AIikhail Botvinnik of the U.S.S.K., calls his machine hIacHic because it so often plays as if it were intoxicated.
J163,
~ f i
SIX SENSATIONAL DISCOVERIES
127
Figure 56 The fourcolormap theorem is exploded
Unlike most chessplaying programs, MacHic is a learning machine that profits from mistakes, keeping a record of all games in its memory and thus steadily improving. Early in 1974 Pinkleaf started MacHic playing against itself, taking both sides and completing a game on an average of every 1.5 seconds. The machine ran steadily for about seven months. At the end of the run, MacHic announced an extraordinary result. It had established, with a high degree of probability, that pawn to king's rook 4 is a win for White. This was quite unexpected because such an opening move has traditionally been regarded as poor. MacHic could not, of course, make an
128 • CHAPTER TWO
exhaustive analysis of all possible replies. In constructing a “game tree” for the opening, however, MacHic extended every branch of the tree to a position that any chess master would unhesitatingly judge to be so hopeless for Black that Black should at once resign. Pinkleaf has been under enormous pressure from world chess leaders to destroy MacHic and suppress all records of its analysis. The Russians are particularly concerned. I am told by one reliable source that a meeting between Kissinger and Brezhnev will take place in June, at which the impact on world chess of MacHic’s discovery will be discussed Bobby Fischer reportedly said that he had developed an impregnable defense against PKR4 at the age of eleven. He has offered to play it against MacHic, provided that arrangements can be made for the computer to play silently and provided that he (Fischer) is guaranteed a winorlose payment of $25 million. The reaction of chess grand masters to MacHic’s discovery was mild compared with the shock waves generated among leading physicists by last year’s discovery that the special theory of relativity contains a logical flaw. The crucial “thought experiment” is easily described. Imagine a meterstick traveling through space like a rocket, on a straight line collinear with the stick. A plate with a circular hole one meter in diameter is parallel to the stick’s path and moving perpendicularly to it (see Figure 57). We idealize the experiment by assuming that both the plate and the meterstick have zero thickness. The
Figure 57 A thought experiment that disproves special relativity
SIX SENSATIONAL DISCOVERIES
129
two objects are on a precise collision course. At the same instant, the center of the meter stick and the center of the hole will coincide. Assume that the plate is the fixed inertial frame of reference and the meterstick is moving so fast that it is Lorentzcontracted by a factor of 10. In this inertial frame the stick has a length of 10 centimeters. As a result it will pass easily through the hole in the rapidly rising plate. (The speed of the rising plate is immaterial.) Now consider the situation from the standpoint of the meterstick's inertial frame. The plate is moving in the opposite horizontal direction, and so now it is the hole that is Lorentzcontracted, along its diameter parallel to the stick, to 10 centimeters. There is no way the 10 centimeter by 1 meter elliptical hole can move up past the meterstick without a collision. The two situations are not equivalent, and thus a fundamental assumption of special relativity is violated. Physicists have long realized that the general theory of relativity is weakly confirmed, but the special theory had been confirmed in so many ways that its sudden collapse came as a great surprise. Humbert Pringle, the British physicist who discovered the fatal Gedankenexperiment, reported it in a short note last summer in Reviews ofModern Physics, but the full impact of all of this has not yet reached the general public. When facsimiles of two lost notebooks of Leonardo da Vinci's were published in 1974 by McGrawHill, they were widely reviewed. The public learned of many hitherto unknown inventions made by Leonardo: a system of ball bearings surrounding a conical pivot (thought to have been first devised by Sperry Gyroscope in the 1920s), a worm screw credited to an eighteenthcentury clockmaker, and dozens of other devices, including a bicycle with a chain drive. In view of the publicity given the McGrawHill volumes, it is hard to understand why the media failed to report in December, 1974 on thk discovery of a drawing that had been missing from the first notebook. This notebook, known as Codex Madrid I (it had been found ten years earlier in the National Library in Madrid), is a systematic treatise of 382 pages on theoretical and applied mechanics (see "Leonardo on Bearings and Gears," by Ladislao Reti; ScientificAmerican, February 1971). There had been much speculation on the nature of the missing page. Augusto Macaroni of the Catholic University of Milan observed that the sketch was in a section on hydraulic devices, and he speculated that it dealt with some type of flushing mechanism. The missing page was found shortly before Christmas by Ramon Paz y Bicuspid, head of the manuscript division of the Madrid Library. It was Bicuspid who had originally found the two lost notebooks. The missing page had been torn from the manuscript and inserted in a fifteenthcentury treatise
130
CHAPTER TEN
on the Renaissance art of perfume making. Figure 58 reproduces a photocopy of the original drawing. As the reader can see at once, Professor Macaroni was on target. The drawing establishes Leonardo as the first inventor of the valve flush toilet. It had long been known that Leonardo had invented a folding toilet seat and had proposed a water closet with continuously running water in channels inside walls, a ventilating shaft to the roof, and suspended weights to make sure the entrance door closed. Until now, however, the first valve flush toilet has always been credited to Sir John Harington, a godson of Queen Elizabeth. Harington described it amusingly in his book The Metamorphosis of Ajax (1596) a cloaca1 satire that got him banished from the court. Although his "Ajax" actually was built at Kelston near Bath, it was not until 200 years later that it came into general use. The first English patent for a valve flush toilet was granted in 1775 to Alexander Cummings, a watchmaker. Modern mechanisms, in which a ball float and automatic cutoff stopper limit the amount of water released with
Figure 58 Lost Leonardo da Vinci drawing [courtesy N.Y. Public Lit
SIX SENSATIONAL DISCOVERIES
131
each flush, date from the early nineteenthcentury patents of Thomas Crapper, a British manufacturer of plumbing fixtures who died in 1910. (See Clean and Decent: T h e Fascinating History of the Bathroom and Water Closet, by Lawrence Wright, Routledge and Kegan Paul, 1960, and Flushed with Pride: T h e Story of Thomas Crapper, by Wallace Reyburn, PrenticeHall, 1971.) Although hundreds of books on parapsychology spewed forth from reputable publishing houses in 1974, not one reported the most sensational psi discovery of the century: a simple motor that runs on psi energy. It was constructed in 1973 by Robert Ripoff, the noted Prague parapsychologist and founder of the International Institute for the Investigation of Mammalian Auras. When Henrietta Birdbrain, an American expert on Kirlian photography, visited Prague early last year, Dr. Ripoff taught her how to make his psychic motor. Ms. Birdbrain demonstrated the device many times in her lectures, but as far as I am aware the only published report on it appeared in the Boston monthly newspaper East WestJournal (May 1974, p. 21). Readers are urged to construct and test a model of the motor. The first step is to cut a threebyseveninch rectangle from a good grade of bond paper. Make a tiny slot in the paper at the spot shown (see Figure 59). The slot must be3 inch long and exactly in the center of the strip, inch from the top edge. Bend the paper into a cylinder, overlapping the ends & inch, and glue the ends together. Cut a second slot in the center of the overlap, directly opposite the preceding one. It must be the same size and the same distance from the top. From a file card or a piece of pasteboard of similar weight, cut a strip 2 inch by three inches. Insert a fine, sharppointed needle twice through the center of the strip, as shown in step 3. The point of the needle should be no more than f inch below the bottom edge ofthe strip. Push the ends ofthe strip through the cylinder's two slots, as shown in step 4, taking care not to bend the strip. The final step is to balance the needle on top of a narrow bottle at least four inches high (step 5). It is essential that the top of the bottle be either glass (preferable) or very hard smooth plastic. Adjust the strip in the slots until the cylinder hangs perfectly straight, its side the same distance from the bottle all around. With scissors snip the ends of the strip so that each end projects inch on each side. Place the little motor on a copy of the Bible or the I Ching, with the book's spine running due north and south. Sit in front of the motor, facing north. Hold either hand, cupped as shown in Figure 60, as close to the cylinder as you can without actually touching it. You must be in a quiet room, where the air is still. Make your mind blanker than usual and focus your mental energy on the motor. Strongly will it to rotate either clockwise or counterclockwise. Be patient. It is normally at least one full minute before the psi energy from your aura takes effect. When it does, the cylinder will start to rotate slowly.
+
STEP 1 INCH y e 3 1 8 INCH
3 INCHES
7 INCHES
STEP 2
STEP 3
3/8 INCH
3 INCHES
5 / 16 INCH OVERLAP
STEP 4
Figure 59 A psychic motor is made
STEP 5
SIX SENSATIONAL DISCOVERIES
133
Figure 60 How to apply psi energy to the psychic motor
Some people, of course, have a stronger psi field than others. A lot depends on your mental state. At times the motor refuses to turn. At other times it begins to turn almost as soon as you start concentrating. Experiments show that for most people it is easier to make the motor rotate counterclockwise with psi energy from your right hand and clockwise with the energy from your left hand. At times negative psi takes over, and the motor turns in the direction opposite to the direction being willed. As Dr. J. B. Rhine has taught us, psi effects are elusive, skittish, and unpredictable. The motor is currently under extensive investigation at numerous parapsychology laboratories around the world. Russian experts are convinced the energy that turns the motor is the same as the psychokinetic energy that enables the Israeli psychic Uri Geller to bend silverware, the Russian "sensitive" Nine1 Kulagina to levitate tabletennis balls, and the Brooklyn psychic Dean Kraft to make pieces of candy leap out of bowls, and pens crawl across rugs. When Kulagina holds both hands near the motor, the cylinder flies straight up in the air for several meters. A book on the Ripoff rotor (as it is called in Prague), with papers by twelve of the world's leading parapsychologists, is being edited by Ms. Birdbrain.
134
CHAPTER TEN
James Randi, the magician, contends that by using trickery, he can make the motor spin rapidly in either direction. Of course, that does not explain why the motor operates so efficiently for thousands of people who know nothing about conjuring.
ADDENDUM The foregoing chapter, when it ran in the April 1975 issue of Scientijic American, was intended as an April Fools'joke. It was so crammed with preposterous ideas and outlandish names that I never dreamed anyone would take it seriously, yet it produced more than a thousand letters from readers who did not recognize the column as a hoax. The map was designed by correspondent William McGregor (his real name), who gave me permission to print it. Hundreds of readers sent me copies of the map, colored with four colors. Some said they had worked on it for days before they found a way to do it. The fourcolor map theorem is no longer a conjecture. It was proved in 1976 by Wolfgang Haken and Kenneth Appel, with the aid of a longrunning computer program. (See "The Solution of the FourColorMap Problem," in Scientijic American, October 1977, pp. 108121 .) Whether a simple, elegant proof not requiring a computer will ever be found, is still an open question. When Norman K. Roth published an article, "Map Coloring," in Mathematics Teacher (December 1975), many readers informed him that Scientijic American had published a map disproving the fourcolor theorem. A letter from Roth in the following May issue pointed out that my column was "an apparently successful April Fools' article." In 1977 the Vancouver Sun reported a British mathematician's claim (it turned out to be invalid) that he had proved the fourcolor map theorem. O n January 17,1977, the newspaper ran a letter from a lady in Port Moody, which said in part: T o set the record straight, I would like to bring to your notice the fact that the theorem has already been disproved by William McGregor, a graph theorist . . . in November 1975. He constructed a map of 110 regions that cannot be colored with less than five colors. . . . The following letter, signed by "Ivan Guffvanoff 111," who claimed to be a mathematician at the University of Wisconsin, was a bit frightening to the staff of Scientijic American, until they realized that it, too, was a joke:
SIX SENSATIONAL DISCOVERIES
135
This is to inform you that my lawyer will soon be contacting you for a damage case of $25 million. In the mathematics section of your April 1975 issue, Martin Gardner wrote that the fourcolor problem had been solved. I have been working on this problem for 25 years. I had prepared a paper to be submitted to the American Mathematical Monthly. The paper was over 300 pages in length. In it I had proved that the answer to the fourcolor problem was no and that it would take five colors instead of four. Upon reading Gardner's article that someone else would publish the solution before I could, I destroyed my paper. Last week I read in Time magazine that Gardner's article was a farce. I did not read Gardner's entire article, only the part on the fourcolor problem, so I was not aware of the farce. Now that I have destroyed my article, it will not be possible to reproduce all 300 pages, since the work has extended over such a long time. I therefore believe that damages are due me. I believe that Gardner's article was the most unprofessional article I have ever seen in your's or any other journal. This kind of activity is below the dignity of what I thought your magazine stood for. I am not only suing you but I am cancelling my membership, and I will ask all my friends to cancel theirs.
In Italy the noted mathematician Beniamino Segre published a serious research note (Rendiconti 59, 1975, pp. 411 412) in which he reproduced McGregor's map, showing how it could be fourcolored. "It is shown the falsehood," his summary reads, "of a presumed counterexample for the fourcolor conjecture." Manifold, a journal published by mathematics students at the University of Warwick, ran the following lines in its Autumn 1975 issue. They are to be sung to the tune of "Oh Mr. Porter, what shall I do?" "Oh Mr. Gardner, What have you done? You've started up a rumour You should never have begun! A fourcolour hoax can't Be undone so quick . . . O h Mr. Gardner, what A bloody silly trick!"
136
CHAPTER TEN
When e is raised to the power of the product of 71 and the square root of 163, the result is the eighteendigit number I gave, minus .000,000,000,000,75. . . .John Brillo, to whom I attributed this hoax, is a play on the name of the distinguished number theorist John Brillhart. The reference to Ramanujan's paper is legitimate. In it the Indian mathematician discusses a family of remarkable nearinteger numbers to which this one belongs, but of course he knew that none were integral. Indeed, as many readers pointed out, it is not hard to prove that they are transcendental. The value of the number to thirtynine significant decimal digits was given by D. H. Lehmer in Mathematical Tables and Aids to Computation (vol. 1, January 1943, pp. 3031). The digit following the run of 9's is 2. See also "What is the Most Amazing Approximate Integer in the Universe?" by I. J. Good, in the Pi M u Epsilon Journal (vol. 5, Fall 1972, pp. 314315). The description of Richard Pinkleaf's chessplaying program, MacHic, is a play on the chess program MacHack, written by Richard Greenblatt of the Massachusetts Institute of Technology. The relativity ~ a r a d o xthat I hung on Humbert Pringle (a play on the name of Herbert Dingle, a British physicist who maintained that relativity theory is disproved by the famous twin paradox) is well known. It appears as a problem on page 99 of the paperback edition of SpacetimePhysics, by Edwin F. Taylor and John A. Wheeler (W. H. Freeman and Company, 1966), and the solution is given on page 25 of the answer section. The paradox is discussed at greater length by George Gamow in Mr. Tompkins in Wonderland (Macmillan, 1947), W . Rindler in American . Journal ofPhysics, (vol. 29, 1961, p. 365 ff.), R. Shaw (ibid., vol. 3 0 , 1 9 6 2 , ~72 ff.), and P. T. Landsberg in The Mathematical Gazette, (vol. 47,1964, p. 197 4. A stationary outside observer will see the meterstick just make it through the hole. If the plate and the stick have thickness, the stick must, of course, be a trifle shorter than the hole to prevent an end from catching. T o an observer on the plate the stick will appear Lorentzcontracted, but it will also appear rotated, so that it seems to approach the hole on a slant. The stick's back end actually seems to go through the hole before its front end, so that it gets through with the same clearance as before. T o an observer on the stick the plate will appear Lorentzcontracted, its hole becoming elliptical, but the plate also appears to be rotated. In this case the hole first goes over thefront end of the slanted stick, again with the same close fit. "Contractions" and "rotations" are ways of speaking in a Euclidean language. In a fourdimensional, nonEuclidean language of spacetime the objects retain their shapes and orientations. Having at one time written a book on relativity, I was abashed to receive more than 100 letters from physicists pointing out the stupid "blunder" I had made.
SIX SENSATIONAL DISCOVERIES
137
Those who enjoyed explaining the paradox may wish to consider how to escape from the following variant. Assume that the meterstick is sliding at high speed along the surface of an enormous flat plate of metal toward a hole slightly larger than the stick. W e idealize the thought experiment by assuming that there is no friction and that the stick and the plate are extremely thin. When the stick is over the hole, gravity (or some other force) pulls it down and through. For an observer on the stick the sheet slides under it, and the hole is Lorentzcontracted enough to prevent the stick from dropping through. In this case the stick and the plate cannot rotate relative to each other. How does the stick get through? (Please, no letters! I know the answer.) The Leonardo da Vinci drawing was done by Anthony Ravielli, a graphic artist well known for his superb illustrations in books on sports, science, and mathematics. It was an earlier version of the sketch that suggested to me the idea of a hoax column. Many years ago a friend of Ravielli's hadjokingly made a bet with a writer that Leonardo had invented the first valve flush toilet. The friend persuaded Ravielli to do a Leonardo drawing in brown ink on faded paper. It was smuggled into the New York Public Library, stamped with a catalogue file number and placed in an official library envelope. Confronted with this evidence, the writer paid off the bet. Augusto Macaroni is a play on Augusto Marinoni, a da Vinci specialist at the Catholic University of Milan, and Ramon Paz y Bicuspid is a play on Ramon Paz y Remolar, the man who actually found the two missing da Vinci notebooks. My data on the history of the water closet are accurate, including the reference to Thomas Crapper. The book by Wallace Reyburn Flushed with Pride: The Story of Thomas Crupper does exist. For many years I assumed that Reyburn's book was the funniest plumbing hoax since H. L. Mencken wrote his fake history of the bathtub. I thought this for two reasons: (1)The book implies that the slang words "crap" and "trapper" derived from Mr. Crapper's name, but "crap" and "crapping case" are both listed in The Slang Dictionary, published in London in 1873. (2) Reyburn wrote a later book titled Bustup: The Uplifting Tale of Otto Titzling and the Development of the Bra. It turns out, though, that both Thomas Crapper and Otto Titzling were real people, and neither of Reyburn's books is entirely a hoax. The Ripoff Rotor is a modification of a psychic motor described in Hugo Gernsback's lurid magazine Science and Invention (November 1923, p. 651). Prizes were awarded in March 1924 to readers who gave the best explanations of why the cylinder turned. The motion can be caused by any of three forces: slight air currents in the room, convection currents produced by heat from the hand, and currents from breathing. The three forces combine in unpredictable
138
CHAPTER TEN
ways. If a person who believes he or she has psychokinetic powers is willing the motor to turn, it may turn in the direction willed, or it may go the other way. Nandor Fodor, in his Encyclopedia ofPsychic Science (Citadel, 1966), under the heading "Fluid Motor," credits the papercylinder device to one Count de Tromelin, but he doesn't say who the Count was or when he invented the motor. I had no individual in mind when I mentioned Ms. Henrietta Birdbrain, but there is an East West newspaper in Boston, and the reader who bothers to check the issue cited, will find a sober report by Stanley Krippner on a psychic motor that was demonstrated to him in Prague by Robert Pavalita. (On Pavalita, see Chapter 28 in Psychic Discoveries behind theIron Curtain (PrenticeHall, 1970), by Sheila Ostrander and Lynn Schroeder.) Many readers, in the spirit of the hoax, sent hilarious explanations for the Ripoff Rotor. Mark J. Hagmann found that the rate of the relative rotation of the rotor and the room was a function of the contents of the liquor bottle he used to support the needle. The rotation increased as the level of the liquid went down. In concluding this addendum, let me pass along some sage advice supplied by my brother Jim. Every responsible science writer should constantly keep in mind the following four words: "Accuracy above all."
ELEVEN
The Csa'sxa'rPolyhedron
Donald W . Crowe, a mathematician at the University of Wisconsin, discovered a surprising correspondence between the skeletons of ndimensional cubes and the solution of a classic puzzle called the Tower of Hanoi. I described his discovery in a column reprinted in The ScientificAmerican Book of MathematicalPuzzles G Diversions (Simon & Schuster, 1959). Professor Crowe has now done it again. In studying the skeletal structure of a bizarre solid known as the Csaszar polyhedron, he has found some remarkable isomorphisms that involve the sevencolor map on a torus, the smallest "finite projective plane," the solution of an old puzzle about triplets of seven girls, the solution of a bridgetournament problem about eight teams, and the construction of a new kind of magic square known as a Room square. The polyhedron that leads into so many apparently unrelated recreations is enormously interesting in itself. It is the only known polyhedron, apart from the tetrahedron, that has no diagonals. (A diagonal is a line joining any two vertexes not connected by an edge.) Consider, for example, the tetrahedron. It has four vertexes, six edges, four faces, and no diagonals. An edge joins every pair of corners. Skeletons of polyhedrons are isomorphic with graphs, that is, with sets of points (vertexes) joined by lines (edges). If an edge connects every pair of
140
CHAPTER ELEVEN
points in a set of n points, it is called a complete graph for n points. Any polyhedron without diagonals clearly must have a skeleton that is a complete graph. Since no polyhedron can have fewer than four corners, the complete graph for four points is the simplest such graph that corresponds to a polyhedral skeleton. W e must now be careful to make precise definitions. A simple polyhedron is one that is topologically equivalent to a sphere and whose faces are all simple polygons: polygons topologically equivalent to a disk. (If you think of a simple polyhedron as having elastic faces, the polyhedron can be inflated like a balloon to form a sphere.) This rules out such nonsimple structures as two polyhedrons joined by an edge or a corner, stellated polyhedrons with intersecting faces, solids with tunnels or interior holes, and so on. Imagine a polyhedron with a surface like that of a cube with a smaller cube on the center of one face. The polyhedron can be inflated to a sphere, but is it simple? It is not, because one face is a ring. The tetrahedron is the only simple polyhedron with no diagonals. An interesting question now arises. Let us define a toroid as a polyhedron whose faces are all simple polygons, but the solid itself is topologically equivalent to a sphere with one tunnel or more going all the way through it. Is it possible to construct a toroid with no diagonals? The answer to the question was not known until the late 1940s, when a Hungarian topologist, ~ k o CsLszir, s succeeded in constructing such a polyhedron. Its skeleton is the complete graph for seven points, shown in symmetrical form at the left of Figure 61. The graph is isomorphic with the skeleton of a sixdimensional simplex, the 6space analogue of the tetrahedron. Since an edge joins every pair of points, the polyhedron has twentyone edges and fourteen triangular faces. It is not difficult to make a paper model of the Csaszar polyhedron. Copy the two patterns in Figure 62 on paper of good quality (or thin cardboard) and cut the copies out. Color the seven shaded triangles on both sides. Crease the paper to make "mountain folds" along each broken line, and "valley folds" along each solid line.
1. With the pattern for the base, fold the two largest triangles to the center and tape the A edges to each other. Turn the paper over. Fold the two smaller triangles to the center and tape the B edges together to obtain a completed base. 2. The sixfaced conical top is formed by taping the Cedges together. Place it on the base as shown in the drawing of the completed model. It will fit
THE CSASZAR POLYHEDRON
141
Figure 61 Skeleton of the Csaszar polyhedron (left) and its dual, the sevencolor torus map (right)
in two ways. Choose the fit that joins white to shaded triangles, then tape each of its six edges to the corresponding six edges of the base. It is not yet known if there is another toroid without diagonals. By applying elementary Diophantine analysis (finding integral solutions of equations), however, Crowe has shown that if there is another, it will have at least twelve vertexes and six tunnels. His proof is as follows: For simple polyhedrons there is a famous formula, discovered by Leonhard Euler, that relates the vertexes (v), edges (e) and faces ( f ) . It is
T h e formula is easily proved, and it takes only a bit of extra work to modify it for toroids. Letting h stand for holes, the formula is
If a toroid is without diagonals, its skeleton is a complete graph, and for any complete graph, edges and vertexes are related by the formula
142
CHAPTER ELEVEN
PATTERN FOR TOP
PATTERN FOR BASE
Figure 62 Patterns for making a model of the Csaszir polyhedron
The faces and edges of a toroid without diagonals (all its faces must be triangles, otherwise a face would have a diagonal on it) are related by the formula
THE CSASZAR POLYHEDRON
Substituting the values for e and f in the formula v  e simplifying, we get
+f
=2
143
 2h and
Factoring the right side gives h a value of
The values of v and h must be integers, and we also know that v must be geater than 3. If v is 4, h has avalue of zero. These values fit the tetrahedron. If v is 5 or 6, h is not integral, proving that no toroid without diagonals can have five or six corners. When v is 7, h is 1. This corresponds to the Csaszar polyhedron. The next solution in integers is v = 12, h = 6. Whether a toroid can be constructed for these values remains unknown. Nor is it known if toroids exist for the next two solutions: v = 15, h = 11, and v = 16, h = 13. From here on, Crowe points out, the number of tunnels exceeds the number of vertexes, so that we can probably rule out all higher toroids. The Csaszar polyhedron has one tunnel, which means that if our model's surface consisted of rubber rather than paper, we could blow it up to the shape of an inner tube. Its twentyone edges would then form a complete graph for seven points on the surface of the torus. None of these edges would intersect one another, proving that the complete graph of seven points has a toroidal crossing number of zero. (On crossing numbers, see Chapter 11 of my Knotted Doughnuts and Other Mathematical Amusements, W . H . Freeman and Company, 1986). Suppose now that on this torus we change the complete graph to its dual. That is done by putting a point inside each of the fourteen triangular faces, then drawing an edge from each point to the points inside its three neighboring faces. Since each new edge crosses one old edge, the number of new edges remains twentyone. The number of faces and the number of vertexes, however, are switched. The new graph is shown in a symmetrical planar form at the right in Figure 61. Crowe has made the fourteen vertexes alternately gray and black spots and has numbered them, as shown, for reasons that will shortly be clear. Its "faces" are hard to see, but you can trace out seven regions, each region surrounded by six edges. This graph can also be drawn on a torus without any intersecting edges. When that is done, the result turns out to be the familiar sevencolor toroidal map (see Figure 63). Note that every two of the seven hexagonal regions share
144
CHAPTER ELEVEN
Figure 63 A sevencolor map o n the torus (regions 3, 4, and 7 wrap around)
a common edge. This means that if the map is colored so that no two contiguous regions have the same color, seven colors are required. O n a plane, no more than four regions can be mutually contiguous, but on a torus the maximum is seven. As an exercise, the reader may like to put a dot inside each region on the torus and see if the map can be converted to its dual by the same procedure described above. Simply draw an edge to connect each pair of dots. Each edge must crossjust one boundary segment, and there must be no intersecting ofthe new edges with each other. Some lines must, of course, wrap around the torus. If you succeed, the new graph will be isomorphic with the complete graph of seven points and also with the skeleton of the Csaszar polyhedron. Here is how Crowe used the dual graph of the skeleton of the Cdszar polyhedron to solve the following puzzle. Seven girls live in a house. Each day a triplet of girls is allowed to leave the house for a visit to town. How can the triplets be chosen so that at the end of seven days every pair of girls will have been in exactly one of the seven triplets? The graph at the right in Figure 61 provides two solutions. For one solution, note each black vertex and write down the (unordered) triplet of numbers on the three gray vertexes that are adjacent (connected by an edge) to it. For the other solution write down the numbers on the black vertexes adjacent to each gray vertex. The two sets of triplets are
THE CSASZAR POLYHEDRON
Black vertex 124 235 346 457 561 672 713
145
Gray vertex 126 237 34 1 452 563 674 715
Each set is called a "Steiner triple system" of order 7, or a "finite projective plane" of order 2. Steiner systems and projective planes are topics of great importance in modern combinatorial theory, but we can (regrettably) mention them only in passing. Because the sevencolor graph is the dual of the Csbzar skeleton, with each of its fourteen vertexes corresponding to one of the fourteen faces of the model of the Csaszar polyhedron (black spots to white faces, gray spots to shaded faces), we can just as easily extract the two solutions from the model. O n the model, no face shares a border with a face of the same color. If the faces are numbered to correspond to the numbers on the vertexes of the graph, the triplet of numbers on the three white faces adjacent to each shaded face gives one solution, and the triplet of numbers on the three shaded faces adjacent to each white face gives the second solution. Another way to find the same two sets of triplets on the model is to number the vertexes of the model, any way you like, from 1through 7. The numbers at the three corners of each shaded face give one set of triplets, and those at the three corners of each white face give the other set. The two solutions will be equivalent to the two sets of triplets listed above, although the numbers may not match. The numbers are no more than arbitrary symbols on a symmetrical graph. T o see the equivalence, it may be necessary to permute them in some way, such as changing all 1's to 5's, all 5's to 3's, and so on. The two solutions obtained by any of these methods are also identical in the sense that one can be changed to the other by permuting the elements; in other words, there is only one basic solution. The next higher Steiner triple system is of order 9. It too has a unique basic solution. Nine girls go out in daily triplets for twelve days, each pair appearing injust one triplet. Can the reader find the solution? The two variants we obtained for the order7 solution are, as Crowe recognized, related in a curious way. N o triplet appears in both sets, and if two pairs of girls in one set appear with the same third girl (for example, in the first set 1,2 and 3,6 each appear with 4), the same pair appears with different girls (5,6)
146
CHAPTER ELEVEN
in the second set. When both of these properties hold, the two solutions are called orthogonal. A Steiner triple system of order n is possible only when n is equal to 1 or 3 (modulo 6). Every orthogonal pair of such systems of order n, Crowe goes on to explain, provides a solution to the following bridge problem for n 1 teams. Suppose there are eight teams of card players and seven tables. Each team must play exactly once with each of the other teams and also exactly once at each table. This is how Crowe tells us to construct the tournament. First draw a square matrix seven cells by seven cells. Consider the pair 1,2. In the first set of triplets, it is associated with 4 and in the second set with 6, so that we put 1,2 in the cell at the intersection of the fourth column and the sixth row (see Figure 64). Consider another pair, 1,3. It is with 7 in the first set and with 4 in the other, so that 1,3 goes in the seventh column and the fourth row. Follow this procedure for all pairs of numbers. The final step is to combine 8 with 1 , 2 , 3 , 4 , 5 , 6 , 7 along a diagonal from the cell at the upper left to the cell at the lower right. Each column indicates a table, and each row indicates a round of simultaneous play at four of the seven tables. All conditions of the desired tournament are now met. The matrix is called a Room square of order 8. Such a square is an arrangement of an even number of objects, n 1, in a square array of side n. Each cell is either empty or holds exactly two different objects. In addition, each object appears exactly once in every row and column, and each (unordered) pair of objects must occur in exactly one cell. The smallest Room square is trivial. It is of order 2 and consists of one cell that contains 1,2. N o Room squares are possible for four or six objects, so that the order8 square is the smallest nontrivial Room square. For years, I assumed that such squares were called Room squares because they concern objects placed in "rooms," but it turns out that they are named for Thomas G. Room, a mathematician who defined them in a brief note in 1955. Combinatorialists have been investigating them ever since. Later it was discovered that Room squares had been in use before 1900 in bridge tournaments, but mathematicians seem not to have been interested until Room wrote his note. There is still more! H. S. M. Coxeter, in editing the twelfth revised edition of W . W . Rouse Ball's classic Mathematical Recreations and Essays (University of Toronto Press, 1974), explains how the Steiner triple system of order 7 can be used for constructing an "anallagmatic pavement" of order 8. Consider the order8 chessboard. If we place any two rows alongside each other, either every cell in one row will match the color of its neighbor in the other, or every
+
+
THE CSASZAR POLYHEDRON
147
TABLES
1
2
3
4
5
6
7
Figure 64 The smallest nontrivial Room square
cell will not match its neighbor's color. W e want to color the sixtyfour cells with two colors so that the following property holds: If any two rows are brought together, half of the paired cells will match and halfwill not, and the same will be true of any two columns. Such squares are also known as Hadamard matrixes, after the French mathematician Jacques Hadamard, who studied them in the 1890s. Apart from the trivial case of order 2, no Hadamard matrix is possible unless the order is a multiple of 4. It is not yet known if matrixes for all such orders exist. Figure 65 shows how our first Steiner triple system provides a Hadamard matrix for the chessboard. The problem involves eight girls and eight days.
148
CHAPTER ELEVEN
1
2
3
GIRLS 4 5
6
7
8
Figure 65 A Hadamard chessboard
Number the rows and columns as shown. The eighth girl is an older girl who chaperones the triplets daily, and on the eighth day all eight girls go into town. For each day (indicated by a row), color the cells that indicate the three girls (plus the chaperone) who walk to town. The result: a Hadamard matrix! There is a simple technique for generating Hadamard matrixes for all orders that are powers of two (see Figure 66). The order2 pattern is placed in three corners of the order4 square, and its negative (colors reversed) goes into the lower right corner. The same procedure with the order4 pattern generates the order 8, and so on for the higher powers. More generally, given two Hadamard matrixes of orders m and n, a matrix of order mn can be created simply by replacing each colored cell of m by the entire pattern of n and each uncolored cell of m by the negative of n. The large matrix is called the tensor product of the two smaller ones. It does not matter whether m is larger, smaller, or equal to n. If, for example, you have a Hadamard matrix of order 12, the tensor product of two such matrixes is a Hadamard matrix of order 144. Hadamard matrixes are more than playthings. They are used for the construction of valuable errorcorrecting binary codes. For descriptions of such applications, the reader is referred to the new edition of Ball's book (so heavily revised as to be almost a new work) and to the references it cites. When the Mariner spacecraft of 1969 sent back pictures of Mars, Coxeter tells us, they were sent in an errorcorrecting code based on the order8 Hadamard matrix. In closing, let us return to the Csaszar polyhedron and pose an intriguing problem. The Csaszar toroid cannot be constructed if all its faces are equilateral triangles. Suppose a onehole toroid is made entirely of congruent trian
THE CSASZAR POLYHEDRON
149
gular faces, all equilateral. What is the minimum number of faces it can have? Bonnie M. Stewart, a mathematician at Michigan State University, considers this problem on page 48 of his book Adventures among the Toroids (1970). Stewart gives the construction of such a torus with fiftyfour faces, twentyseven vertexes, and eightyone edges. One of his students, Kurt Schmucker, found a onehole toroid with fortyeight equilateral triangular faces. Whether that is the minimum, however, is another tantalizing, unanswered toroidal question.
ANSWERS The unique solution for the Steiner triple system of order 9 is obtained from the following matrix: a b c d e f g h i
8
Figure 66 Hadamard matrixes for powers of two
150
CHAPTER ELEVEN
The rows (abc, def; g h i ) provide three triplets. The columns (adg, beh, cf;) provide three more. The main diagonals (aei, ceg) and the broken diagonals (bfg, cdh, afh, bdi) complete the list of twelve triplets. The orders of Steiner triple systems are numbers that have a remainder of 1 or 3 when they are divided by 6. The next higher system, order 13, has two basic solutions. Order 15 is known to have eighty solutions. A model of Kurt Schmucker's onehole toroid, with fortyeight congruent equilateral triangle faces, is easy to make. It consists of a ring of eight regular octahedrons joined by their faces (see Figure 67). Schmucker found that rings could be made by joining eight replicas of each of the Platonic solids except the tetrahedron. N o matter how many tetrahedrons are joined by their faces, no ring is possible even when the solids are allowed to intersect one another. A proof is given by J. H. Mason in his paper "Can Regular Tetrahedra Be Glued Together Face to Face to Form a Ring?" ( T h e Mathematical Gazette 56, October 1972, pp. 194 197).
Figure 67 Ring of eight octahedrolis
THE CSASZAR POLYHEDRON
151
ADDENDUM When I wrote briefly about Hadamard matrixes, I had not realized that their application to telemetry codes was discovered by Golomb and that the use of such codes in several Mariner probes of Mars was the result of Golomb's affiliation with the Jet Propulsion Laboratory of the California Institute of Technology (see Bibliography for Golomb's coauthored 1963 paper). A Hadamard matrix larger than 2 by 2 must have a side that is a multiple of 4, but it is not yet known if matrixes exist for all such orders. Golomb tells me that a 268 Hadamard matrix was reported by Kanue Sawade in Graphs and Combinatorics (vol. 1, 1985, pp. 185 187). The next unknown case, now the smallest, is order 428. Hadamard matrixes have also found extensive application in the processing of pictorial information. The "Hadamard transform" (analogous to the "fast Fourier transform") actually produces a mathematical hologram of the original image. Golomb has been a pioneer in recognizing that classic combinatorial design problems often provide optimum solutions to dataprocessing problems and that it pays to look for engineering problems to which these designs are the solution.
B
I
B
L
I
O
G
R
A
P
H
Y
"A Polyhedron without Diagonals." ~ k o Csaszar, s in Acta Scientiarum Mathematicarum 13, 1949 1950, pp. 140142.
Adventures Among the Toroids. B. M. Stewart. Published by the author, 1952. 2d rev. ed., 1984. "The Search for Hadamard Matrices." Solomon W . Golomb and Leonard D. Baumert, in American Mathematical Monthly 70, January 1963, pp. 12 17. "Euler's Formula for Polyhedra and Related Topics." Donald W . Crowe, in Excursions in Mathematics, Anatole Beck et al., eds. Worth, 1969. "Steiner Triple Systems, Heawood's Torus Coloring, Csaszar's Polyhedron, Room Designs, and Bridge Tournaments." Donald W . Crowe, in Delta 3, Spring 1972, pp. 27  32. Room Squares "A New Type of Magic Square." Thomas G. Room, in The Mathematical Gazette 39,1955, p. 307.
152
CHAPTER ELEVEN
"On Furnishing Room Squares." R. C. Mullin and E. Nemeth, injournal ofCombinatorial Theory 7, November 1969, pp. 266  272.
+
"On Room Squares of Order 6m 2." C. D. O'shaughnessy, ibid, vol. 13, series A. November 1972, pp. 306314. "Solution of the Room Square Existence Problem." W. D. Wallis, ibid, vol. 17, series A Novemer 1974, pp. 379383. "The Existence of Room Squares." R. C. Mullin and W. D. Wallis, in Aequationes Mathematicae 13, 1975, pp. 17.
TWELVE
Dodfem and Other Simple Games Consider a twoperson game with the following characteristics: (1) It is a game of perfect information; that is, both players have complete knowledge of the game's structure after every move. (2) The players move alternately. (3) Decisions are not made by chance. (4) The game ends after a finite number of moves with a win by one player. (No draw is possible.) It is not hard to see that there must be a winning strategy for either the first player or the second. If the first player (henceforth called A) does not have a winning strategy, he must lose. This means that the second player, B, has a winning strategy. Does the argument apply ifwe rescind the requirement that the game end in a finite number of moves? curiously, it all depends on whether or not one accepts the "axiom of choice." This notorious axiom says that from any collection (finite or infinite) of nonempty sets, with no elements in common, you can form a new set by taking one element from each set. In the 1930s, Stefan Banach, Stanislaw Mazur, and Stanislaw Ulam discovered a type of infinite game in which neither A nor B has a winning strategy if the axiom of choice is accepted. Someone argued that this proves the Unitarian dogma that there is "at the most" one God, because if two gods could play such a game, neither could know a winning strategy and therefore neither could be called omniscient!
154
CHAPTER TWELVE
That, however, is by the way. Here we shall examine some new twoperson nonchance games for which the rules are extremely simple and for which a winning strategy is either known or capable of being known. All but one of the games are played with counters on boards that are easily drawn on cardboard. Two differently colored sets of counters, such as go stones or small poker chips, will be useful for any reader who wants to play or to analyze the games. An example of an almost trivial game of the nim type, but one with a strategy that is not immediately apparent, is played on the star pattern shown in Figure 68. Put a counter on each ofthe star's nine points. A and B take turns removing either one counter or any two counters joined by a straight line segment. The player who takes the last counter wins. B can always win at star nim by a strategy based on the board's symmetry. Imagine that the black lines are strings. The pattern can be opened up to a circle that is topologically equivalent to the star. If A takes one counter from this circle, B takes the two counters that are directly opposite. If A takes two counters, B takes one counter that is directly opposite. In each case two sets of three counters are left. Now, whatever A takes from one set, B takes the corresponding counter or counters from the other set. Obviously B will get the last counter. If the reader plays a few games on the circle, translating each move to its equivalent on the star, he or she will soon see how to use the star's symmetry for playing the strategy. In the late 1960s, G. W. Lewthwaite, of Thurso, Scotland, invented a delightful game with an artfully concealed "pairing strategy" that gives the second a sure win. O n a 5by5 square matrix place thirteen black
Figure 68 Star nim (left) and its winning strategy (right)
DODGEM AND OTHER SIMPLE GAMES
155
Figure 69 G. W. Lewthwaite's counter game (left) and a pairing strategy f o ~ Lewthwaite's game (right)
counters and twelve white counters in alternating checkerboard fashion. Any one of the black counters, say the one in the center, is removed (see Figure 69, left). Player A controls the white counters and B the black. They take turns moving one of their counters orthogonally to the vacant square until a player loses by being unable to move. If the board is colored like a checkerboard, it is obvious that on each move, a counter goes to a square of different color and that no counter can be moved twice. The game, therefore, cannot go beyond twelve moves for each player. It may end before then, however, in a win for either player unless B plays rationally. B's strategy is to imagine that the matrix, except for the initially vacant cell, is covered with twelve nonoverlapping dominoes. It does not matter how they are placed. Figure 69 (right) shows a sample pattern. Whenever A moves, B simply moves his counter that is on the domino A has just vacated. Since this ensures that B always has a move to follow a move by A, B is sure to win in twelve or fewer moves. The game can be played not only with counters but also with square tiles or cubes that slide within a matrix surrounded by a rim. Suppose the rules are amended to allow either player at any time to move any number of adjacent counters (I through 4) in a row or column provided that the two end counters are of his or her color. This is a splendid example of how an apparently trivial alteration of a rule can enormously complicate a game's analysis. Lewthwaite was unable to find a winning strategy for either player in this variant of his
156
CHAPTER TWELVE
Games based on the sliding of unit squares within a square matrix offer a e an attractive game plethora of unexplored possibilities. ~ e w t h w a i t proposes that he calls meander. It uses twentyfour identical tiles placed in a 5by5 tray to form the pattern shown in Figure 70 (left). The players take turns sliding a single counter or a straight row or column of two, three, or four counters. The play continues until a player wins by creating a pattern in which at least three tiles form a continuous line or path thatjoins two edges (opposite or adjacent) of the tray. Figure 70 (right) shows a winning pattern, with the winning line indicated by the two arrows. The game is probably too complex for solving without a computer program, and perhaps too complex for solving even with one. In 1972, when Colin Vout was a mathematics student at the University of Cambridge, he invented an intriguing counter game that he calls dodgem because it is so often necessary for a piece to dodge around enemy pieces. It is playable on a checkerboard of any size. Even the game on a 3by3 board is complicated enough to be interesting. Two black counters and two white ones are initially placed as shown in Figure 71 (top). Black sits on the south side of the board and White sits on the west. The players alternately move a counter one cell forward or to their left or right, unless it is blocked by another counter of either color or by an edge of
Figure 70 Meander, with example of pattern on a tile at top (left) and a possible winning pattern in meander (right)
BLACK LEAVES BOARD HERE
WHITE'S DIRECTIONS
t
C
WHITE LEAVES BOARD HERE
BLACK'S DIRECTIONS
Figure 71 Colin Vout's dodgem (top) and a dodgem game won by Black (bottom)
158
CHAPTER TWELVE
the board. Each player's goal is to move all his pieces off the far side of the board. In other words, Black moves orthogonally north, west, or east and attempts to move both of his pieces off the north side of the board. White moves east, north or south and tries to move his pieces off the east side of the board. There are no captures. A player must always leave his opponent a legal move or else forfeit the game. The first to get all his pieces off the board wins. The bottom of Figure 71 shows a typical game won by Black. Vout assures me that the first player has the win on the order3 board, but as far as I know, no games on higherorder boards have yet been solved. O n a board of side n, each player has n  1 pieces placed on the west and south edges, with the southwest corner cell vacant. Played with seven checkers or pawns of one color and seven of another color on the standard order8 checkerboard or chessboard, it is a most enjoyable game. Piet Hein's now classic game of hex (see Chapter 8 of my Scientijic American Book ofMathematica1 Puzzles G Diversions, Simon & Schuster, 1959) remains unsolved, except for small boards. For readers unfamiliar with the game, it is played on an nbyn rhombus of hexagons such as the order4 board shown in Figure 72. White opens by placing a white counter on a cell. Black follows with a black counter. They take turns placing counters on vacant cells (there are no moves or captures) until a player wins by forming a chain of adjacent counters that joins his side of the board to the opposite side, White by joining the north and south edges, Black by joining the east and west edges. It is easy to see that no draw is possible. There is a famous proof by John F. Nash (who independently invented hex) that on a rhombus of any size, the first player has a winning strategy, although the proof gives no hint of what the strategy is. Suppose White allows Black to tell him where he must make his first move. Can White still always win if he plays rationally? This modified version of hex has been called Beck's hex after Anatole Beck, who both proposed and solved it. Writing on hex in Chapter 5 of Excursions into Mathematics, Beck shows that Black can always win if he tells White to open by taking an acute corner cell. In other words, such an opening is a sure loss for White, although Beck's proof does not provide Black's winning strategy. However, as a footnote comments, it "wrecks Beck's hex." What about misire, or reverse, hex, known as rex, in which the first player to join his sides loses? As is so often the case in twoperson games, the reverse game proves to be much harder to crack. No general strategy is known, although Robert 0 .Winder, in unpublished arguments, has shown the existence of a firstplayer winning strategy in rex of even order and a secondplayer winning strategy for all odd orders. More recently Ronald J. Evans has
DODGEM AND OTHER SIMPLE GAMES
159
WHITE
WHITE
Figure 72 Rex, a reverse hex game, with White to play and win
carried Winder's arguments a step further by showing that on evenorder boards there is a winning strategy if White opens in the acute corner. Rex on the order2 rhombus is trivial, and it is not difficult to analyze exhaustively on the order3 rhombus. Play on the order4 rhombus is so complicated, however, that even though it is known that an acutecorner opening initiates a win, the strategy itself remains unformulated. The position shown in Figure 72 is an order4 rex problem composed by Evans. Can the reader determine White's only correct move? Here is an even simpler game for which no general strategy is known. It is played on a 1byn board (a single row of n squares) with counters that are all alike. A and B take turns placing a counter until one player wins by getting three counters adjacent. Could anything be simpler? A can always win when n is odd by first taking the center cell, then playing symmetrically opposite the opponent thereafter. For even n, however, things are not so simple. O n most even rows, A seems to have the win, but not necessarily, and the exceptions follow no known rule. Take n = 6, for example. The reader may enjoy working it out to see who has the win. John Horton Conway has pointed out that this game is equivalent to a game I called 1byn cram in a column reprinted as Chapter 19 of my Knotted
160 • CHAPTER TWELVE
Doughnuts and Other Mathematical Entertainments, (W. H. Freeman and Company, 1986) except that it is played with trominoes instead of dominoes. It is easy to see the isomorphism. In playing the game as described above, it is obviously disastrous to place a counter either next to another counter or one cell from it, since either move gives the opponent an instant win. Hence we might as well prohibit both moves. An easy way to do it is to require that each play consist of a triplet of adjacent counters, which is the same as placing a tromino on the field. (The middle of the triplet corresponds to placing a single counter, and the ends of the triplet enforce the two new rules.) The winner is the player who places a tromino last. (To complete the equivalence, we must allow the placing of a tromino at either end of the field so that it extends one cell beyond the end.) Of course, the game can also be played by forming a row of n counters and by removing them by alternate moves of taking three adjacent counters. This triplet version of cram is considered in a classic paper by Richard K. Guy and Cedric A. B. Smith, “The GValues of Various Games.” Because it is coded as game .007, it has been called the James Bond game. Elwyn Berlekamp has computeranalyzed the game to very high even n without finding any periodicity in the Grundy numbers, which means that no one is even close yet to a general rule. The misère version of 1byn tromino cram, regardless of the parity of n, is also unsolved. Ulam has proposed extending the counter form of tromino cram to a square matrix. The players take turns placing single counters until one player wins by getting three in a row orthogonally or diagonally. As before, oddorder fields are trivial because the first player wins by taking the center, then playing symmetrically until the opponent offers a win. On evenorder boards the order 4 is trivial, but no one yet knows who has the win on orders 6 or 8. Figure 73, supplied by Ulam in a letter, shows a position on the order6 board for which the next player must lose. Here again we can play an equivalent game by alternately placing polyominoes, in this case squares of nine cells, but it is not very convenient because in addition to allowing the pieces to extend into a unit border around the field, we must also allow them to overlap one another by just two cells (corner and side). No one has even begun to find a general strategy for the game in standard or reverse form. ANSWERS The answers to the chapter’s two problems are that the second player has the win on 1by6 tromino cram, and White wins the 4by4 rex (reverse hex)
DODGEM AND OTHER SIMPLE GAMES
161
Figure 73 Stanislaw Ulam's triplet game
game by taking the cell at the intersection of row C and diagonal 1. Although no winning strategy for the first player of order4 rex is known, a winning secondplayer pairing strategy for the order5 game has been found by David L. Silverman. All higher orders remain unsolved. G. W. Lewthwaite's game generalizes in an obvious way to rectangular boards of any size and shape. If the rectangle has an odd number of cells, the second player wins; if it has an even number of cells, the first player wins. (In the latter case the domino covering strategy includes the initially vacant cell.) Karl Fulves has suggested that instead of visualizing a domino pattern, you play the game with counters secretly marked so that you can place them on the board in orientations that group them in pairs. For example, a small pinhole on the rim of a pawn or a checker would enable you to orient the pieces so that the pinholes of each pair are adjacent. You could then play the domino strategy without having to remember a domino pattern. If coins are used, designate a spot on the rim of each side of each coin, say the N in the ONE on a penny, as your mark and orient the coins accordingly.
ADDENDUM David Fremlin and Dennis Rebertus independently wrote computer programs that verified the first player's win in order3 dodgem. White wins only by first moving his piece at the corner. A full analysis of the order3 game is in Winning Ways. Order4 dodgem is still unsolved.
162
CHAPTER TWELVE
Figure 74 John Beidler's results for Stanislaw Ulam's triplet game
John Beidler, who heads the computer science department at the University of Scranton, found by computer that Stanislaw Ulam's triplet game in standard play on a 6by6 field is a win for the first player only if his first move is on one of the four central cells. Beidler generalized the game to rectangular boards and obtained the results shown in Figure 74. The numbers give winning moves by row and column for the first player. The asterisks indicate a win for the second player. If the game is played in reverse form, Beidler found that the second player has the win on 3by3, 3by4,3by5, and 4by4 boards. For tromino cram, played in reverse on a 1byn field, with n less than 19, Beidler proved firstplayer wins for n = 4 , 5 , 7 , 8 , 11, 14,15, 17, and 18, and secondplayer wins for all other values.
B
I
B
L
I
O
G
R
A
P
H
Y
"The GValues of Various Games." R. K. Guy and C. A. B. Smith, in Proceedings ofthe Cambridge Philosophical Society 52, July 1956, pp. 514526. "The Game of Hex." Anatole Beck, in Excursions into Mathematics, Beck et al., eds. Worth, 1969. "A Winning Opening in Reverse Hex." R. J. Evans, inJourna~ofRecreationalMathematics7, Summer 1974, pp. 189 192. "Reverse Cram with Block Sizes Not Exceeding 13." R. J. Evans, in Delta 6, 1976, pp. 5776. "Dodgem." Winning Ways. E. R. Berlekamp, J. H. Conway, and R. K. Guy, pp. 685  688. Academic Press, 1982.
THIRTEEN
Tiling with Convex Polygons
"Many of the brightly coloured, tilecovered walls and floors of the Alhambra in Spain show us that the Moors were masters in the art of filling a plane with similar interlocking figures, bordering each other without gaps. What a pity that their religion forbade them to make images!" M.
C. ESCHER
Imagine that you have an infinite supply ofjigsaw puzzle pieces, all identical. If it is possible to fit them together without gaps or overlaps to cover the entire plane, the piece is said to tile the plane, and the resulting pattern is called a tessellation. From the most ancient times such tessellations have been used throughout the world for floor and wall coverings and as patterns for furniture, rugs, tapestries, quilts, clothing, and other objects. The Dutch artist M.
164
CHAPTER THIRTEEN
C. Escher amused himself by tessellating the plane with intricate shapes that resemble birds, fish, animals, and other living creatures (see Figure 75). A tile that tessellates obviously can have an infinite variety of shapes, but by imposing severe restrictions on the shape, the task of classifying and enumerating tessellations is reduced to something manageable. Geometers have been particularly interested in polygonal tiles, of which even the simplest present formidable problems. In this chapter we are concerned only with the task of finding all convex polygons that tile the plane. It is a task that was not completed until 1967, when Richard Brandon Kershner, assistant director of the Applied Physics Laboratory of Johns Hopkins University, found three pentagonal tilers that had been missed by all predecessors who had worked on the problem. Let us begin by asking how many of the regular polygons tile the plane. As the ancient Greeks knew and proved, there are just three: the equilateral triangle, the square, and the regular hexagon. The hexagonal tiling, so familiar to bees and users of bathrooms, is a fixed pattern (see Figure 76). The patterns formed by equilateral triangles or by squares can be infinitely varied by sliding rows of triangles or squares along lattice lines.
Figure 75 Tessellation by M. C.Esche~
TILING WITH CONVEX POLYGONS
165
Figure 76 T h e three regular polygons that tile the plane
If we remove the restriction that a convex polygon must be regular, the tiling problem grows in interest. It has been proved that no convex polygon of more than six sides can tile the plane. Thus we need to investigate only polygons of three, four, five, and six sides. The triangle is easy. Any triangle tiles the plane. Simply fit two identical triangles together, with the corresponding edges coinciding as shown in Figure 77, and you create a parallelogram. Replicas of any parallelogram obviously will go side by side to make an endless strip with parallel sides, and the strips, in turn, go side by side to fill the plane. The quadrilateral is almost as easy, although much more surprising. Any quadrilateral tiles the plane! As before, take a pair of identical quadrilaterals, one inverted with respect to the other, join the corresponding edges and you create a hexagon (see Figure 78). Each edge of the hexagon is necessarily equal to and parallel to its opposite edge. Such a hexagon, by a simple translation operation (altering its position on the plane without changing its orientation), will form a tiling pattern. The quadrilateral need not be convex. Exactly the same technique creates a tiling pattern for any nonconvex quadrilateral. The case of the hexagon was settled in 1918 by K. Reinhardt in his doctoral thesis at the University of Frankfurt. He showed that any tessellating convex hexagon belongs to one of three classes. Kershner, in a 1969 article, "On Paving the Plane," explains the three types as follows. Label the sides and angles of a hexagon as shown in Figure 79. A convex hexagon will tile the plane if and only if it belongs to one or more of the following classes:
166
CHAPTER THIRTEEN
Figure 77 Any triangle tiles the plane
1. A + B + C = 3 6 0 ° , and a = d. 2. A + B + D = 3 6 0 ° , and a = d, c = e. 3. A = C = E = 120°, and a = b, c = d, e =f: The illustration gives an example of each type of convex hexagon tiler and a portion of its tiling patterns. The gray lines outline a "fundamental region" that tiles the plane by translation. Note that Type 2 requires reflection if the hexagon is asymmetric. In similar fashion the tessellating convex pentagons can be classified in eight ways. Five were found by Reinhardt. Kershner describes them by labeling the pentagon as shown in Figure 80. A convex pentagon paves the plane if it belongs to one or more of the following classes:
1. A + B + C = 3 6 0 ° . 2. A + B D=360°, and a = d. 3. A = C = D = 120°, and a = b, d = c e.
+
+
TILING WITH CONVEX POLYGONS • 167
4. A = C = 90º, and a = b, c = d. 5. A = 60º, C = 120º, and a = b, c = d. Examples of each type and its tiling pattern are reproduced in the illustration with gray lines outlining fundamental regions. Only Type 2 requires reflection. “At this point,” writes Kershner, “either [Reinhardt’s] technique or his fortitude failed him, and he closed the thesis with the statement that in principle it ought to be possible to complete the consideration of pentagons along the lines of his considerations up to that point, but it would be very tedious and there was always the possibility that no further types would emerge Indeed, it is quite clear that Reinhardt and everyone else in the field thought that the Reinhardt pentagon list was probably complete.
Figure 78 Any quadrilateral tiles the plane
168
CHAPTER THIRTEEN
TYPE 1
TYPE 2
TYPE 3
Figure 79 T h e three types of convex hexagon tiler
"For reasons that I would have difficulty explaining I have been intrigued by this problem for some thirtyfive years. Every five or ten years I have made some kind of attempt to solve the problem. Some two years ago I finally discovered a method of classifying the possibilities for pentagons in a more convenient way than Reinhardt's to yield an approach that was humanly possible to carry to completion (thoughjust barely). The result of this investigation was the discovery that there were just three additional types of pentagon . . . that can pave the plane. These pavings are totally surprising. The discovery of their existence is a source of considerable gratification."
TYPE 1
u /
TYPE 2
TYPE 3
d
TYPE 4
h
TYPE 5
b
Figure 80 T h e five types of tiling convex pentagon known in 1918
170
CHAPTER THIRTEEN
The three additional types (see Figure 81) are described by Kershner as follows (Type 7 and Type 8 require reflection):
6. A + B + D = 3 6 0 ° , A = 2 C , and a = b = e, c = d. 7. 2 B + C = 2 D + A = 3 6 O 0 , anda= b=c=d. 8. 2 A + B = 2 D + C = 3 6 0 ° , anda=b=c=d. Kershner's paper does not include a proof that there are no other convex pentagons that tile the plane, "for the excellent reason," reads the editor's introductory note, "that a complete proof would require a rather large book." TYPE 6
C
TYPE 7
TYPE 8
Figure 81 Three new types of tiling convex pentagon discovered in 1967
TILING WITH CONVEX POLYGONS
171
Note that Kershner has deliberately drawn his tessellations with polygons that are as irregular as possible, within the limits of their type, in order to bring out the nature of the tessellation. The most regular hexagonal tessellation is, of course, the familiar beehive mosaic. One can readily see that it belongs to all three hexagon types. If the beehive hexagons are bisected, the result is a pentagonal pattern that belongs to Type 1(see Figure 82A). The pattern formed by six pentagons in a flowerlike arrangement (see Figure 82B) uses a tile that belongs to Type 1, Type 5, and Type 6. The most remarkable of all the pentagonal patterns is a tessellation of equilateral pentagons (see Figure 82C). It belongs to Types 2 and 4. Observe how quadruplets of these pentagons can be grouped into oblong hexagons in two different ways, each set tessellating the plane at right angles to the other. This beautiful tessellation is frequently seen as a street tiling in Cairo and occasionally in the mosaics of Moorish buildings. It underlies many of Escher's tessellations. The equilateral pentagon is readily constructed with a compass and a straightedge (see Figure 83). First draw a side of the pentagon, AB. Construct its perpendicular bisector, CD, then draw lines CE and C F at 45 degrees to AB. With center at A and radius AB, draw a circular arc cutting C E at P. The same construction is repeated on the other side, with center at B and the arc cutting CF at R. Keeping the compass with radius AB, let R be the center and strike an arc that cuts the perpendicular bisector, CD, at Q. The pentagon's corner angles at P and R are right angles. The corner at Q is a little more than 131 degrees, and corners A and B are a trifle more than 114 degrees. The length from Q to B is the product of a side of the pentagon and the square root of two. The pentagon's area (it is easy to prove) is re cis el^ the square of line segment CR. Among the infinite tessellations of the plane that can be made with congruent nonconvex polygons, combinatorial geometers have given special attention in recent years to tiling with polyominoes and their cousins the . are formed by joining unit polyiamonds and ~ o l ~ h e x e s(Polyominoes squares, polyiamonds by joining equilateral triangles, and polyhexes by joining regular hexagons.) Many fascinating problems have been formulated, some solved and some not. That will be the topic of the next chapter.
ADDENDUM A remarkable letter on this chapter after it first appeared as a column in 1975, came from Richard E. James 111, a computer scientist with the Control Data Corporation. He sent a strange tessellation (see Figure 84) along with a note
172
CHAPTER THIRTEEN
Figure 82 Pentagonal tessellations of unusual symmetry
describing the pentagon (in R. B. Kershner's notation) as A = 90 degrees, C D = 270 degrees, 2 0 E = 2C B = 360 degrees, and a = b = c e. "Do you agree that Kershner missed this one?" he asked. Kershner had indeed missed it. This means that the problem of classifying all convex pentagons that tile the plane is not solved, as Kershner had supposed. I must say that Kershner received the blow with grace and good humor. In a letter to him, I mentioned that James's discovery illustrates the pragmatic side of mathematical proof, namely that proofs are not known to be proofs until there is a consensus among experts. Kershner replied as follows: "In connection with your philosophical comments on the nature of a proof you might be interested in an observation by that eminent authority, me. In The Anatomy ojMathematics,Kershner and L. R. Wilcox (Ronald Press, 1950) I wrote:
+
+
+
+
TILING WITH CONVEX POLYGONS
Figure 83 How to construct equilateralpentagon tiler
Figure 84 A remarkable new tessellation with congruent convex pentagons by Richard E. James I11
173
174
CHAPTER THIRTEEN
Now it must be said that there is no simple test that can be applied to determine the validity of a proof, that is, to determine that an alleged proof really is a proof. Mathematical history contains rare instances of arguments that were generally accepted as proofs for hundreds of years, before being successfully challenged by a very ingenious mathematician, who pointed out a possibility that had been overlooked in the alleged proof. And more recently, every year there appear, in the mathematical journals of the world, a certain number of papers which point out that some statement, allegedly proved in a preceding paper, was not only erroneously proved (that is, not proved) but was, in fact, incorrect. These facts are mentioned for the benefit of those who feel that there is some magic formula for a proof which makes it immutable and unarguable henceforth and forevermore. "I must say that when I wrote this paragraph I did not at the time propose eventually to illustrate its validity so gaphically myself." James's tessellation can be varied in ways that have been analyzed in a 1978 paper by Doris Schattschneider of Moravian College, Bethlehem, Pa. It is a basic pattern that could have been discovered by the medieval Moors or even by the ancient Greeks or Romans, but it is probable that the pattern had never before been seen by human eyes until James first put it on paper! The discovery of new types of tiling pentagons did not end with James's finding. Marjorie Rice, a San Diego housewife with no mathematical training beyond the minimum required in high school, began a systematic search for new patterns. In 1976 she discovered a tenth type (see Figure 85), two more types later that year, and still another one the following year, bringing the total number of types to thirteen. A fourteenth type was found in 1985 by Rolf Stein, a mathematics graduate student at the University of Dortmund in West Germany. Its tiling pattern is on the cover of Mathematics Magazine (November 1985), with a note about it on page 308. As far as I know, no more new types have been discovered, although there is as yet no proof that the list is complete. Nor is there a full listing of all nonconvex pentagons that tile the plane. Doris Schattschneider gives a brief account of Mrs. Rice's fantastic achievements in her 1978 paper, and a more detailed account in her contribution to The Mathematical Gardner. The latter paper includes three color plates of beautiful Escherlike tesselations (bees, fish, and flowers) that Mrs. Rice based on her new tiling patterns, and a color plate of a handwoven rug based on the James tesselation. The bee pattern provides the book's jacket.
TILING WITH CONVEX POLYGONS
Figure 85 Marjorie Rice's tenth pentagonal tiling
175
176
CHAPTER THIRTEEN
B
I
B
L
I
O
G
R
A
P
H
Y
"On Paving the Plane." R. B. Kershner, in The American Mathematical Monthly 75, October 1968, pp. 839844. "On Paving the Plane." R. B. Kershner, in A P L Technical Digest 8, JulyAugust 1969, pp. 4 10. "The Laws of Sines and Cosines for Polygons." R. B. Kershner, in Mathematics Magazine 44, May 1971, pp. 150 153. "Tesselations with Pentagons." J. A. Dunn, in The Mathematical Gazette 55, December 1971, pp. 366369; see also 56, December 1972, pp. 332335. "Tesselations of Pentagons." Stanley R. Clemens, in Mathematics Teaching (a British journal for elementary teachers), June 1974, pp. 18 19. See followup by John Parker, ibid. March 1975, p. 34. "Tiling the Plane with Congruent Pentagons." Doris Schattschneider, in Mathematics Magazine 51, January 1978, pp. 2944. Reprinted in Percy Alexander MacMahon: Collected Papers, Vol. 2, George E. Andrews, ed. MIT Press, 1986. "Convex Polygons that Cannot Tile the Plane." Ivan Niven, in The American Mathematical Monthly 85, December 1978, pp. 785789. "A Note on a Result of Niven on Impossible Tesselations." M. S. Klamkin and A. Liu, in The American Mathematical Monthly 87, October 1980, pp. 651 653. "In Praise of Amateurs." Doris Schattschneider, in The Mathematical Gardner, David A. Klarner, ed. Prindle, Weber and Schmidt, 1981. "Equilateral Convex Pentagons Which Tile the Plane." M. D. Hirschhorn and D. C. Hunt, in Journal ofCombinatoria1 Theory 39, May 1985, pp. 1 18. "Tilings by Polygons." Chapter 9 of Tilings and Patterns. Branko Griinbaum and Geoffrey C. Shephard. W. H. Freeman and Company, 1987. This marvelous 700page work is the first comprehensive treatise on all aspects of tiling theory. Tiling with Nonconvex Pentagons "Tesselations with Equilateral Reflex Polygons." Gillian Hatch and David Simonds, in Mathematics Teaching September 1978, pp. 3234. "Tesselations of Nonconvex Pentagons." Michael Palmer, ibid, March 1979, pp. 8  11. "Spiral Tilings and Versatiles." Branko Griinbaum and G. C. Shephard, ibid, September 1979, pp. 5051. "The Incredible Pentagonal Versatile." Doris Schattschneider, designs by Marjorie Rice, ibid, December 1980, pp. 52 53. Solution to Problem 67D, posed by H. Martyn Cundy, based on results by Doris Schattschneider, in The Mathematical Gazette 67, December 1983, pp. 307309.
FOURTEEN
Tiling with Polyom inoes, Polyiarnonds, and Polyhexes "I often wondered at my own mania for making periodic drawings. . . . What can be the reason of my being alone in this field? Why does none of my fellowartists seem to be fascinated as I am by these interlocking shapes?"
The previous chapter was devoted to tessellating the infinite plane with congruent, nonoverlapping convex polygons. In this chapter we extend the topic into the vast domain of nonconvex polygons, with special attention to polyominoes, polyiamonds and polyhexes. Polyominoes are shapes formed by joining unit squares at their edges. They were first introduced to the mathematical world in 1953 by Solomon W. Golomb. His book Polyominoes (Scribner's, 1965) is the standard reference on
178
CHAPTER FOURTEEN
this popular recreation. Some of my previous reports on polyominoes are reprinted in several earlier book collections of my ScientijcAmerican columns. The monomino (single square) and domino obviously tile the plane, and so do the two kinds of tromino. It takes only a few moments to discover that each of the five tetrominoes will tile in a simple ~ e r i o d i pattern c in which all the tiles are identically oriented; that is, no tile needs to be rotated or reflected (flipped over). Each of the twelve fivecelled figures will tile the plane. All but three the T, U, and R pentominoes tile by simple translation (sliding without rotations or reflections). Each of the three that require rotation can be fitted into pairs by turning one upside down to form an order 10 polyomino that tiles by translation (see Figure 86). There are thirtyfive distinct hexominoes, and all of them tile without being reflected. ~ b m etile by simple translation. Those that do not can be joined in pairs by turning one upside down to make an order12 polyomino that tiles by translation. John H. Conway has long been interested in the tiling properties of polyominoes and other polygons. When he examined the 108 heptominoes, it was clear that the task of identifying the tilers would be tedious unless he could find a criterion that would, as he put it, "dispose of most of them rapidly without diagrams all over the graph paper." The criterion discovered by Conway is an efficient one that applies to any polygon. It is based on a hexagonal tiling pattern (see diagrams at left and middle in Figure 87). Note that edges a and d are equal and parallel and that throughout the pattern a joins d with the two hexagons in the same orientation. Note also that each of the other four edges joins its corresponding edge on a tile that has been rotated 180 degrees. With these facts in mind it is easy to understand the basis of Conway's criterion. W e examine a given polygon to see if its perimeter can be divided into six parts, a, b, c, d, e, andJ that meet the following requirements:
u Figure 86 T h e three pentoi .noes that require rotation t:o tile periodically
TILING WITH POLYOMINOES, POLYIAMONDS, AND POLYHEXES
179
Figure 87 Criterion for periodic tiling without reflection
1. Two opposite edges, a, and d, are "parallel" in the sense that they are congruent and in the same orientation. 2. Each of the other four edges, b, c, e, a n d l are centrosymmetric; that is, they are unaltered by a 180degree rotation around a midpoint.
If the polygon meets both requirements, it will tile the plane periodically. N o flipping over of tiles is necessary. The tiles are paired simply by turning one upside down to form a figure that tiles by translation. Conway's example will make this procedure clear (see diagram at right in Figure 87). The two gray lines mark sides a and d, which are "parallel." The two X's distinguish the edges b, c, e, andf. Each of the four edges has central symmetry; therefore the figure will tile the plane by pairing. One figure of the pair is rotated 180 degrees with respect to the other, without reflection, and the double shape tiles by translation. It is important to realize, Conway adds, that any of the six edges may be empty (nonexistent). The criterion is quite general, applying to triangles, quadrilaterals, pentagons, and all higher polygons. Of the 108 heptominoes, 101 meet Conway's criterion. That means each can be paired with an upsidedown mate to form a 14cell shape that tiles by translation. The seven that fail to meet the criterion are shown in Figure 88. The first heptomino obviously cannot tile the plane because there is no way to fill the hole. The fourth one tiles by pairing with a 90degree rotation as is shown in Figure 89, (lower left). The fifth, which Conway considers the most interesting heptomino, tiles in three ways. It tiles by pairing, with 90degree rotation and reflection. Without reflection it tiles in quadruplets of four orientations (see Figure 89, (top right, shows on of two ways). The second heptomino in this group will not tile without reflection. The smallest region, which tiles by translation, contains four replicas in four orientations; two of the replicas are reflected, as is shown in Figure 89, (lower right). This is the lowestorder polyomino, unique among the heptominoes, that requires reflection to tile the plane.
180 • CHAPTER FOURTEEN
1
2
3
4
5
6
7
Figure 88 The seven heptominoes that do not meet Conway’s criterion for tiling periodically
Figure 89 Periodic tiling of heptomino five (top), heptomino four tiles with 90degree rotation (bottom left), and heptomino two tiles ony with reflection (bottom right)
TILING W I T H POLYOMINOES, POLYIAMONDS, AND POLYHEXES
181
The third, sixth, and seventh heptominoes will not tile, making four nontilers in all. Proving impossibility in each case is not difficult, although it is sometimes tedious. First you try all possible ways of fitting two replicas together, eliminating the ways that have a hole or a space that cannot be filled by a third replica. After all such pairs have been found, you test all ways of adding a third piece that allow the placing of a fourth. Eventually you reach a set of n heptominoes such that, no matter how they are combined, there is no way to add another replica. I leave these proofs to interested readers. Although the third heptomino will not tile the plane, it can be combined with 3by3 square tiles to form a striking tessellation. The reader is urged to make a set of cardboard replicas of this heptomino, together with replicas of the order3 square, and see if he can combine the shapes to form a tessellation. As far as I know, only David Bird of North Shields, England, has isolated the nontilers among the 369 octominoes. The six with holes can of course be eliminated immediately. Bird reports twenty other nontilers, making twentysix in all (see Figure 90).
Figure 90 The twentysix nontiling octominoes
182
CHAPTER FOURTEEN
Is there a general algorithm that can be applied to any polyomino to decide if it will tile? Conway's criterion identifies certain polyominoes as tilers, but if a polyomino does not meet his criterion, it may or may not tile in other ways. Conway conjectures that there is no general algorithm, but this conjecture has not yet been proved. A major step toward establishing the undecidability of the algorithm was reported by Golomb in his 1970 paper "Tiling with Sets of Polyominoes." In that paper Golomb considers whether there is a procedure for deciding if any given finite set of different polyominoes (assuming an unlimited supply of each kind) will tile the plane. He shows that the problem is equivalent to a problem of tiling the plane with a finite set of edgecolored squares by matching colors at the joined edges. Since Hao Wang and his colleagues had earlier proved the latter problem to be undecidable (see "Games, Logic and Computers," by Hao Wang; Scientif;~American, November 1965) the corresponding problem for polyominoes is also undecidable. Golomb's paper extended the results published in his 1966 article "Tiling with Polyominoes." In that paper he considered the tiling of such subsets of the plane as the halfplane, the quarterplane, straight strips, bent strips, and rectangles. A chart summarizes results through the hexominoes. If there is no region in a tessellation that tiles by translation, the tessellation with is said to be nonperiodic. One curious way of tiling nonperi~dicall~ congruent polygons is to use a set of them to build a larger replica. Obviously sets of these larger replicas can then be combined in the same way to make a still larger replica, and in that way the entire plane can be covered. A tile with the property of selfreplication was named by Golomb a "reptile" [See his paper "Replicating Figures in the Plane," and the chapter on reptiles in my Unexpected Hanging (Simon & Schuster, 1969).] Less work has been done on polyiamonds (figures formed by joining congruent equilateral triangles) than on polyominoes. (On the polyiamonds see Golomb's book and Chapters 18 and 24 in my Sixth Book ofMathematica1 Games (University of Chicago Press, 1983). It is not hard to establish by Conway's criterion that all polyiamonds through order 6 will tile. Of the twentyfour heptiamonds only the V heptiamond is not a tiler. (A simple impossibility proof is given in Chapter 24 of my Sixth Book). Bird, Gregory J. Bishop, Andrew L. Clarke, John W . Harris, Wade Philpott, and others have found that all octiamonds will tile. As far as I know, the 160 enneiamonds (order 9) have not yet been settled, although two correspondents, Bird and Clarke, agree that the nontilers are the twentyone shown in Figure 91. The polyhexes, formed by joining congruent regular hexagons (see the chapter on them in my Mathematical Magic Show, Knopf, 1971), have received
TILING W I T H POLYOMINOES, POLYIAMONDS, AND POLYHEXES
183
Figure 91 The twentyone enneiamonds believed to be nontilers
even less attention than the polyiamonds. Bird and Bishop have established that all polyhexes through order 5 are planefillers. O f the eightytwo hexahexes Bird believes only five are nontilers (see Figure 92). Once it is determined that a given polygon will tile, one can ask in how many distinct ways the tiling can be accomplished. It can be a very sticky question. A. W. Bell, in his paper "Tessellations of Polyominoes," has made a beginning by attempting to classify all patterns for the L tetromino. He obtained nineteen patterns but made no claim for completeness. Major P. A. MacMahon's little book New Mathematical Pastimes (Cambridge University Press, 1921; why has no publisher reprinted it?) explains with a wealth of illustrations how a simple polygon tessellation is easily transformed into a more complicated one. You merely take a pair of straight edges, which
Figure 92 T h e five hexahexes thought to be nontilers
184
CHAPTER FOURTEEN
always go together in the pattern, and change the straight boundary to a crooked one (curves are allowed), subject to certain symmetry restrictions. It was by just such techniques that Islamic craftsmen created the intricate abstract mosaics in the Alhambra and the Taj Mahal. Here is how Escher described the difficulty of adapting the technique to the creation of tiles that resemble living things: "The borderline between two adjacent shapes having a double function, the act of tracing such a line is a complicated business. O n either side of it, simultaneously, a recognizability takes shape. But the human eye and mind cannot be busy with two things at the same moment, and so there must be a quick and continual jumping from one side to the other." Tessellation theory, quite apart from its usefulness to artists who design patterns for walls, floors, fabrics, and so on, has a practical application in industry. In cutting congruent shapes from thin sheets of metal, plastic, cardboard, leather, and other materials, a tessellation pattern obviously provides the only way of doing it without waste. An unusual art book could be made by collecting pictures of the tessellations used in modern manufacturing, from the simple rectangular patterns of postage stamps, dollar bills, and playing cards to complicated machine parts. There is also a potential application to jigsaw puzzles. In 1958 the British mathematical physicist Roger Penrose amused himself by transforming a parallelogram tessellation into tilings of order1 8 polyiamonds. Some of them make excellent puzzles. Consider a "loaded wheelbarrow" polyiamond (see Figure 93). If the reader will make twelve cardboard replicas of the figure, he will find it a challenging task to fit them together to make a region that tiles the plane by translation.
Figure 93 Roger Penrose's polyiamond puzzle
TILING W I T H POLYOMINOES, POLYIAMONDS, AND POLYHEXES
185
Penrose is Rouse Ball Professor of Mathematics at the Mathematical Institute of the University of Oxford. He is best known among physicists for his contributions to relativity theory and cosmology. He and his father, L. S. Penrose, were the first to discover "impossible objects" such as the famous Penrose staircase that Escher used so effectively in his lithograph Ascending and Descending. For about a decade it has been known that there are sets of polygons that A together will not tile the plane periodically but will do so nonperi~dicall~. few years ago Raphael M. Robinson constructed a set of six tiles that tile only nonperiodically. Penrose later found a set of four and finally a set ofjust two. Is there a single shape, duplicates of which will tile only nonperiodically? This is one of the deepest unsolved problems in tessellation theory. The subject of nonperiodic tiling is, however, another story.
ANSWERS One way of tesselating the plane with replicas of a certain heptomino and a 3by3 square is shown in Figure 94. There are other ways to do it. A tiling pattern for Roger Penrose's wheelbarrow is shown in Figure 95. Replicas of the hexagonal region (consisting of twelve polyiamonds in twelve different orientations) will tile the plane periodically in the manner of the familiar beehive pattern of regular hexagons. Although the tiling pattern is unique, there are many different ways a fundamental region of twelve pieces can be outlined on the infinite plane.
Figure 94 A heptominoandsquare tessellation
186
CHAPTER FOURTEEN
Figure 95 How the Penrose polyiamond tiles the plane
B
I
B
L
I
O
G
R
A
P
H
Y
"Replicating Figures in the Plane." S. W. Golomb, in Mathematical Gazette 48, December 1964, pp. 403412. "Tiling with Polyominoes." S. W. Golomb, in Journal ofCombinatoria1 Theory 1, September 1966, pp. 280 296. "Tesselations of Polyominoes." A. W. Bell, in Mathematical Reflections, edited by members of the Association of Teachers of Mathematics, Cambridge University Press, 1970. "Tiling with Sets of Polyominoes." S. W. Golomb, in Journal ofCombinatoria1 Theory 9, July 1970, pp. 6071.
TILING WITH POLYOMINOES, POLYIAMONDS, AND POLYHEXES
187
"Will it Tile? Try the Conway Criterion!" Doris Schattschneider, in Mathematics Magazine 53, September 1980, pp. 224233. "Tiling with Congruent Tiles." Branko Griinbaum and G. C. Shephard, in Bulletin ofthe American Mathematical Society (new series) 3, November 1980, pp. 951 973. "Polymorphic Polyominoes." Anne Fontaine and George E. Martin, in Mathematics Magazine 57, November 1984, pp. 275283. "Tilings by Polygons." Chapter 9 of Tilings and Patterns. Branko Griinbaum and Geoffrey C. She~hard.W. H. Freeman and Company, 1987.
FIFTEEN
Curious Maps
"I quite realized," said Columbus, "That the Earth was not a rhombus, But I am a little annoyed T o find it an oblate spheroid."
If only the earth were flat, as Martin Luther, John Calvin, and the fathers of the Catholic Church believed, the labors of cartographers would be greatly simplified. Indeed, flatearth maps of the first six centuries of the Christian Era presented no serious geometrical problems. A few learned churchmen agreed with Pythagoras, Plato, Aristotle, and Archimedes that the earth was round, but to most churchmen such a belief was heresy. Early medieval maps of the world were "Scripturepreserving." They were either rectangular to preserve the "four corners" in Isaiah 11:12 and in Revelation 7: l , or they were circular or oval to preserve the "circle of the earth" in Isaiah 40:22. There was, of course, no need for meridians and parallels. Jeru
190
CHAPTER FIFTEEN
salem was exactly in the center, as Ezekiel 5:5 suggests. The top of the map pointed east and included the site of Eden. The land was surrounded by the "great waters" that had once flooded the earth and also by the sources of the "four winds" (Daniel 7:2, Revelation 7:l) that blew so erratically toward the Holy City. After the eighth century the rotundity of the earth gradually became acceptable to the Church, with such eminent Catholics as Thomas Aquinas and Dante Alighieri defending it. It slipped out of favor among the Reformers, but during the Renaissance it quickly won the day. The rapid increase in travel and exploration, particularly the great seavoyages, made it necessary to have better maps, and that naturally revived a troublesome mathematical question: How can a portion of the earth's surface be drawn on a plane so that all distances are accurately represented? The answer is that it cannot. The side of a cylinder or cone will map perfectly onto a plane, but the surface of a sphere will not. You can flatten a cylinder or cone without distorting its surface, but even a small region of a sphere's surface will not press flat without cracking, folding, or stretching. Every flat map of all or part of the earth distorts something. The cartographer's tricky task is to design maps that will show the least distortion or no distortion ofthose properties the map's user deems desirable. At the same time, the distortion of all other properties should be minimal. W e shall take a quick look at some classical methods of mapmaking before turning to methods that result in more bizarre maps. One of the most desirable features of a map is that angles between any two lines on the map be the same as the angles between those same lines on the globe. This feature is enormously useful at sea because it means that observed angles between two landmarks correspond to angles measured on the map with a protractor, and also because small regions on such a map preserve their shape. Maps of that type are called conformal. The simplest way to produce a conformal map is by "stereographic projection." As Figure 96 shows, a sphere is projected by straight lines from point B on the sphere's surface to a plane tangent to the sphere at a point opposite. The projection is called equatorial, polar, or oblique, depending respectively on whether the antipodes are on the equator, the poles, or somewhere else. The price paid for conformality is a distortion of the scale factor that increases with distance from the center of the map. If the projection to the tangent plane is made from the globe's center, it is a gnomonic projection, so called because it is related to the construction of a sundial with a gnomon. Every great circle on the globe becomes a straight line on a gnomonic map. The map is not conformal, but for navigators it has one merit that all other planar projections lack. A straight line between any two
CURIOUS MAPS
191
I
A ORTHOGRAPHIC
PLANE OF PROJECTION
STEREOGRAPHIC
Figure 96 Three azimuthal projections: orthographic (A), stereographic (B), and gnomonic (C)
points on a gnomonic map corresponds to a great circle on the globe and therefore provides the geodesic: the shortest distance between the two points. Since the point of projection can be at any spot inside or outside the globe, there is an endless variety of perspective projections. If the point is at infinity (all projecting lines parallel), the projection is orthographic. Our view of the moon or a view of the earth from the moon is essentially orthographic. Distance distortions are great at the edges of an orthographic map. The map preserves neither area nor angles, but if it is skillfully drawn, it gives a strong illusion of the earth's roundness. Perspective maps with the "eye" above the earth inay be among the least accurate with respect to many properties, but they are the most accurate in matching our visual perceptions of a sphere. Projections need not be made onto a plane. They can be made onto surrounding cylinders or cones that can then be cut and unrolled. Imagine the earth snugly fitted inside a cylinder. The projecting lines are parallel to the plane that cuts the great circle where the globe and the cylinder touch (see Figure 97). The resulting map has the amazing property, highly desirable for
192 • CHAPTER FIFTEEN
Figure 97 Cylindrical projection method for making an equalarea map
many purposes, that areas are preserved: All closed curves have the same areas as their corresponding curves on the globe, and they have them in scale. If the cylinder touches the earth along the Equator, all meridians and parallels on the map become straight lines meeting at right angles. The equalarea cylindrical map is not conformal, and it severely distorts shapes and distances. Indeed, it is not hard to prove that no map can simultaneously be conformal and areapreserving. A great variety of other areapreserving maps have been devised. In modern atlases one of the most popular of equalarea world maps is an elliptical projection worked out by Karl B. Mollweide in 1805. The cylindrical projection suggested to Gerhardus Mercator, a sixteenthcentury Flemish geographer, the famous conformal map that bears his name. Imagine the earth’s surface punctured at the poles, the two holes enlarged until the surface is a cylinder, the cylinder stretched along its length until the map is conformal, then cut along a meridian and unrolled. There is an enormous distortion of scale near the poles. As we learned in grade school, this familiar world map shows Greenland larger than South America when actually it is much smaller. (In order to minimize such scale variations, modern atlases use a modification of the Mercator map called the Miller projection.)
CURIOUS MAPS
193
The Mercator projection has, however, one remarkable property that makes it invaluable to navigators. If you rule a straight line between any two points on the map, the line is a loxodrome, or rhumb line, connecting the two points. A loxodrome is a line that keeps a constant angle with parallels and meridians (see Figure 98). Imagine a point on the globe that starts at the Equator and moves northward in any constant compass direction. The path will be a loxodrome that spirals toward the North Pole, finally strangling the pole after an infinity of turns around it. O n a stereographic map (with its plane tangent to the North Pole) a loxodrome projects as a logarithmic spiral. The loxodrome is not the shortest distance between two points, but for small distances it is reasonably close to a geodesic, and it has the practical value of being a path that does not require constant changes of bearing. For long distances, navigators usually determine the geodesic from a gnomonic projection, then break it into shorter rhumb lines on a Mercator map to minimize changes in compass settings. So much for the classic projections. Let us turn now to more radical distortions. For a twopoint equidistant projection, points A and B are selected. The map is then drawn so that all distances from A and B to any other point on the map are in true scale. Such a map is useful to a person traveling from A to B. N o matter how circuitous one's route is, one can at any time measure with a ruler
Figure 98 A Mercator conformal map with loxodromes, or rhumb lines, from New York
194
CHAPTER FIFTEEN
exactly how far one is from both points. Suppose the two points on a twopoint equidistant map of the world are the two poles. What will the map look like? Another curious specialpurpose map is the "Mecca map," designed to show at once to a Moslem the exact direction he must face when he prays at any spot on the globe. One way to draw such a map is to make a stereographic projection with Mecca at the plane's tangent point. Because the map is conformal, Mecca's bearing angle can be determined by measuring the angle between a straight line to Mecca and a meridian. Unfortunately, the meridians on such a map are curved, making it difficult to measure the angle exactly. One can, however, construct a Mecca map on which all meridians are straight lines, making it possible to measure bearing angles with a protractor. Such a map, with Mecca replaced by another holy place, Wall Street, is given in an internal Bell Laboratories memorandum on map oddities written by Edgar N. Gilbert, a mathematician (see Figure 99). The map's upper boundary is the North Pole.
Figure 99 Edgar N. Gilbert's "Mecca map" for Wall Street
CURIOUS MAPS
195
SOUTH POLE
Figure 100 Johann Werner's cordiform (heartshaped) equalarea map
Gilbert's memorandum contains even stranger maps. One is a cordiform (heartshaped) equalarea map invented by Johann Werner (see Figure 100). It was popular in the sixteenth century but, writes Gilbert, has "now fallen into undeserved obscurity. The highly distorted parts of the map lie away from most major land masses. The curved latitude lines give the map a pleasing illusion of roundness. . . . The latitude lines are arcs of circles, evenly spaced and centered at the North Pole. The longitude lines are drawn to make distances along the latitude lines the same in the map as in the sphere." The bunching of land masses on Werner's map reflects an actual bunching of land on the earth. The Pacific Ocean is so immense that if you look at a globe from a point above the English Channel, you will see more than 80
196
CHAPTER FIFTEEN
percent of the world's land, and the hemisphere opposite will be almost entirely water. Is such lopsidedness surprising? Gilbert came up with an answer by replacing the continents with small nonoverlapping circular caps. caps are randomly distributed around the globe, whaF is Assuming thHt N the probability that the centers of all N circles lie in one hemisphere? In a paper titled "The Probability of Covering a Sphere with N Circular Caps," Gilbert shows that the probability is 2  N ( W  N 2). With N = 7 continents the probability is 11/32, proving that the earth's lopsided land distribution is not at all remarkable. The reader may enjoy testing the formula by determining the probability that the centers of one, two, or three continents all lie in one hemisphere. he strangest of all Gilbert's maps was made by taking a conformal world map based on a conical projection and projecting it back onto a sphere to produce a conformal "twoworld map." Figure 101 shows how the globe looks in perspective from a point about five radii from its center. When people visit Gilbert's office, he likes to ask them what is wrong with his globe. If the visitor cannot see what is wrong, Gilbert gives the globe one slow, complete turn. "Even this hint," he writes, "does not always succeed." Actually every spot on the globe has a duplicate on the other side! Unless you are an experienced !geographer, however, it is not easy to realize that you are seeing much more of the world than can normally be seen on one hemisphere. With the aid of computergenerated graphics it is now possible to write programs that distort a map so that the areas express some desired value such as annual rainfall, retail sales, and so on. Regions on the map still retain recognizable shapes in spite of the distortions. The joke map showing a New Yorker's idea of the United States, although it goes back (in many variants) to precomputer days, is a familiar example of such maps. One of the top experts on such specialpurpose map projections is Waldo R. Tobler, a geographer at the University of Michigan. In a paper titled "A Continuous Transformation Useful in Districting," he explains a computer program that distorts a map to show relative populations of regions by their relative sizes, and he shows how such a technique could be a valuable aid in planning voting districts. In 1973 a geographer with a fondness for floppy bow ties was given a presentation award that consisted of a framed world map distorted to the shape of his tie. Drafting the map was no problem for ~ o b l e r ' scomputer Dissection maps (as I call them) are world maps projected on a pattern of squares, triangles, or polygonal tiles of other shapes. The tiles can be fitted together to make an "interrupted map" (a map with discontinuities) of any portion of the globe. The philosopher and mathematician Charles Sanders Peirce designed such a conformal map. It is a projection on eight isoceles right
+
CURIOUS MAPS
197
NORTH POLE
Figure 101 Gilbert's "twoworld" globe and an oblique view of the globe
triangles that may be regarded as the faces of an octahedron that has been flattened until a space diagonal is zero. The vertexes of the zero diagonal are the North and South poles of Peirce's map. B. J. S. Cahill of Oakland, Calif. patented his butterfly map in 1913, and it enjoyed a considerable vogue in the 1930s. The world is projected onto the eight equilateraltriangular faces of a regular octahedron. Cahill had several versions of his map, based on different projections, but they all consisted of eight triangular tiles that could be fitted together as one pleased (see Figure 102). R. Buckminster Fuller's first Dymaxion map was a projection of the world onto the fourteen faces (six squares and eight equilateral triangles) of a cuboc
198
CHAPTER FIFTEEN
SOUTH POLE
SOUTH POLE
Figure 102 Butterfly map by B. J. S. Cahill
tahedron. Gerard Piel, who was then science editor of Life, was so intrigued by it that he asked the cartographer Richard Edes Harrison to draw an unfolded net of the solid. Staff artists completed a color map of the drawing, and it was published on coverstock paper in the March 1, 1943, issue of Life. It was a great success. All over the country, in homes and laboratories, one would see the little cuboctahedron hanging on a cord and rotating with currents of air. At about the same time, Irving Fisher, a distinguished Yale economist, thought of a similar idea: a gnomonic projection of the world to the twenty triangular faces of an icosahedron. Figure 103 shows an unfolded net of this Platonic solid as it appears inside the jacket of World Maps and Globes, an entertaining introduction to cartography by Fisher and 0. M. Miller. Harrison was the map's cartographer. The reader may enjoy pasting a copy of Fisher's gnomonic projection on heavy paper and folding it into the icosahedral globe or cutting out the triangles and fitting them together in different ways. If the poles are put at opposite vertexes, the Equator becomes a straight line. Fisher marketed several versions of his Likaglobe, as he called it, and wrote an article about its merits in GeographicalReview (October 1943). The globe can, of course, be projected on other regular solids, but distortion is exaggerated on the cube and the tetrahe
CURIOUS MAPS
199
dron. Although the dodecahedron makes a handsome pseudoglobe, its faces cannot be used as tiles because regular pentagons will not tessellate the plane. The icosahedron seems to be the ideal polyhedron for a dissection map, and Fuller himself has now adopted it. In 1954 he copyrighted his Dymaxion Skyocean Projection World Map, drawn by Shoji Sadao. It differs from Fisher's Likaglobe in having the North and South poles on opposite faces at points slightly off center. For a small fee, readers can obtain a punchout version printed in four colors on heavy stock. It folds into a beautiful icosahedron that rests on a cardboard stand that comes with the map. (Orders should go to Dymaxion Maps, 3500 Market Street, Philadelphia, PA, 19104). Also available are several wallmap versions of the unfolded icosahedron and an explanation sheet on which Fuller discusses the philosophy and mechanics of the projection. W e have considered only a small portion of the many curious specialpurpose maps designed by ingenious cartographers. Harrison once drew a world map consisting of nothing but ocean shipping lanes. From a distance, you could see the continents clearly, but from close up you discovered that the map contained not a single shoreline. At the Christian Science Publishing House in Boston, you can step inside a world globe 30 feet in diameter. If such a globe were transparent and viewed from the outside, everything would be mirrorreversed. From the inside the land masses and oceans appear normal.
NORTH POLE
NORTH POLE
SOUTH POLE
Figure 103 Irving Fisher's Likaglobe, which folds into an icosahedron
SOUTH POLE
200
CHAPTER FIFTEEN
The major tenet of a bizarre little sect called Koreshanity it flourished in America late in the nineteenth centurywas that we actually live on the undersurface of such a hollow earth, with the entire cosmos inverted and compressed to fill the interior. The sect was founded by a baptist named Cyrus Reed Teed, of Utica, N.Y. You will find Teed's insideout maps in his great scientific work The Cellular Cosmogony (1870)and in the pages of The Flaming Sword, a periodical that the cult kept going until 1949. How did Teed explain the fact that we cannot point a telescope straight up and see the earth's other side? Well, he took care of it with special optical laws describing how light travels in bent paths. Teed's cosmos has not been sufficiently appreciated by philosophers of science. By performing an inversion operation on the universe, then inventing new laws of physics, one can obtain an insideout cosmology that is not easy to refute except by slashing it with Occam's razor. A large class of eccentric maps that we have not considered are the maps of imaginary regions. "Mighthavebeen" maps show what the world might have looked like if major wars had ended differently. Fantasy maps depict Oz, Hell, Eden, Poictesme, Narnia, Barsoom, MiddleEarth, Atlantis, and other fanciful realms. J. B. Post has collected ninetyeight of them in his beautiful Atlas ofFantasy. And we must not forget the Bellman's ocean chart in Fit 2 of Lewis Carroll's Hunting ofthe Snark: He had bought a large map representing the sea, Without the least vestige of land: And the crew were much pleased when they found it to be A map they could all understand "What's the good of Mercator's North Poles and Equators, Tropics, Zones, and Meridian Lines?" So the Bellman would cry: and the crew would reply, "They are merely conventional signs! "Other maps are such shapes, with their islands and capes! But we've got our brave Captain to thank" (So the crew would protest) "that he's bought us the bestA perfect and absolute blank!"
CURIOUS MAPS
201
Fisher's book on world maps closes with thirtysix excellent questions and answers about world geography. How well can the reader do (without consulting a map) on the following selection?
You are on a ship five miles from an entrance to the Panama Canal and sailing due west toward it. In what body of water is your ship? Flying due south from Detroit, what foreign country do you reach first? Which is nearer Miami, California or Brazil? Which is farther north, Venice or Halifax? Which is farther south, Venice or Vladivostok? Which is larger, Japan or Great Britain? What four states in the U.S. touch at one point? Does a geodesic (great circle) from Tokyo to the Panama Canal pass east or west of San Francisco?
The illustrations for this chapter are by Richard Edes Harrison. He also gave me invaluable assistance on the text.
ANSWERS A twopoint equidistant map of the world, when the two points are any pair of antipodes, is a straight line. The next problem dealt with the distribution of continents around the world. If one, two, or three nonoverlapping circular caps are randomly distributed around the globe, the probability that their centers lie on one hemisphere (any half sphere, not necessarily the Northern or the Southern Hemisphere) is 1. The answers to the geographical questions are:
1. 2. 3. 4. 5. 6. 7. 8.
Pacific. Canada. California. Venice. Vladivostok. Japan. Arizona, Colorado, New Mexico, Utah. East.
202
CHAPTER FIFTEEN
With reference to question 7, I learned in 1984 that the Four Corners Highway passes within a few hundred yards of the spot where the four states meet. An access road leads to the spot, where you find local Indians selling wares to tourists.
ADDENDUM So many strange projections ofworld maps came to my attention after I wrote this chapter that I have space only to comment briefly on a few. I had mentioned the New Yorker's map of the United States on which New York, California, and Florida are huge, all other states tiny. In the late 1970s, the Danish writer and inventor Piet Hein marketed in Denmark what he called the Denmark Globe, on which Denmark is enlarged to symbolize how big it seems to Danes. In 1985 I clipped an advertisement from The New York Review of Books (February 14) for a map of the Americas with South America on top, and the United States and Canada at the bottom. Of course all place names are rightside up. "Run of the mill maps place the U.S. on top," the ad read. "Since 'upper' is equated with 'superior,' this breeds misconceptions and mischief. The Turnabout Map offers a corrective perspective." Professor Tobler, cited in the chapter, wrote to me about his unpublished Mobiusstrip map. The earth is projected onto the strip, like a Mercator projection that has been stretched horizontally, twisted, and the ends joined. The two stretched polar points become the single edge of the onesided surface. The map has the following remarkable property: A pin puncturing it at any point emerges at the spherical antipodal point! W h o but a topologist, Tobler asked, would suspect this similarity of a sphere and a Mobius surface?
B
I
B
L
I
O
G
R
A
P
H
Y
"Quincuncial Projections of the Sphere." Charles S. Peirce, in AmericanJournal afMathematics 2, 1879, pp. 394396.
World Maps and Globes. Irving Fisher and 0. M. Miller. Essential Books, 1944. Elements of Cartography. A. H. Robinson. Wiley, 1960. "The Probability of Covering a Sphere with N Circular Caps." E. N. Gilbert, in Biometrika 52, 1965, p. 323.
CURIOUS MAPS
203
Notes and Comments on the Composition of Terrestrial and Celestial Maps. J. H. Lambert. Translated and introduced by W. R. Tobler. University of Michigan Publication 8, 1972. Atlas ofFantasy. J . B. Post. Mirage Press, 1973. "A Continuous Transformation Useful in Districting." W. R. Tobler, in Annals ofthe N e w York Academy ofsciences 119, 1973, pp. 215220. "Distortion in Maps." E. N. Gilbert, in S I A M Review 16, January 1974, pp. 4762.
Cartographical Curiosities. Gillian Hill. British Museum Publications, 1978.


SIXTEEN
The Sixth Symbol and Other Problems
1. WHAT SYMBOL COMES NEXT?
Can you sketch the sixth figure in the sequence of five symbols shown in Figure 1041
2. WHICH SYMBOL IS DIFFERENT? Every person who ever took an I.Q. test has surely been annoyed by questions that involve a row of symbols and a request to identify the symbol that "does not belong7'to the set. Sometimes there are so many different ways a symbol can be different from the others that brighter students are penalized by having to waste time deciding which symbol is "most obviously different" to the person who designed the test. It was irritation over such ambiguity that prompted Tom Ransom, a puzzle expert of Toronto, Ontario, to devise a delightful parody of such a test. The reader is asked to inspect the five symbols in Figure 105 and pick out the one that is "most different."
206
CHAPTER SIXTEEN
Figure 104 What is the sixth figure in this series?
3. CUTTING A CAKE A cake has been baked in the form of a rectangular parallelepiped with a square base. Assume that the square cake is frosted on the top and four sides and that the frosting's thickness is negligible (zero). W e want to cut the cake into n pieces so that each piece has the same volume and the same area of frosting. The slicing is conventional. Seen from above, the cuts are like spokes radiating from the square's center, and each cutting plane is perpendicular to the cake's base. If n is 2, 4, or 8, the problem is easily solved by slicing the cake into two, four, or eight congruent solids. Suppose, however, that n is 7. How can we locate the required seven points on the perimeter ofthe cake's top? If you solve it for 7, you will be able to generalize to any n. This pretty problem is given by H. S. M. Coxeter on page 38 of his classic Introduction to Geometry (Wiley, 1967). Coxeter does not answer it, but it should give readers little difficulty. The general solution is surprisingly simple.
4. TWO CRYPTARITHMS Here are two elegant cryptarithms by Alan Wayne that have not been published before. The first is in French, the second in German. Each letter represents just one decimal digit, and we adopt the usual convention that zero must not begin a number. Both have unique solutions.
Figure 105 Tom Ransom's I.Q. test: Which symbol is the "most different"?
THE SIXTH SYMBOL AND OTHER PROBLEMS
+
V I N G T C I N Q C I N Q T R E N T E
+
E E E E
I I I I
207
N N N N
V I E R
5. LEWIS CARROLL'S "SONNET" In 1887 Lewis Carroll included in a letter to Maud Standen, a "childfriend," a sixline poem that he called an "anagrammatic" sonnet. (He was using "sonnet" in an older sense, meaning any short piece of verse.) "Each line has four feet," Carroll wrote to Maud, "and each foot is an anagram, i.e., the letters of it can be rearranged so as to make one word." Most of the anagrams, he said, had been devised "for some delicious children" he had met the previous summer at Eastbourne. The words vary in length from four through seven letters, and it is assumed that proper names are not allowed. Here is the "sonnet7':
As to the war, try elm. I tried. The wig cast in, I went to ride 'Ring? Yes.' W e rang. 'Let's rap.' W e don't. '0 shew her wit!' As yet she won't. Saw eel in Rome. Dry one: he's wet. I am dry. O forge! Th' rogue! W h y a net?
In most cases there is little doubt about the correct word. For example, the first foot, "As to," could be "oast" or "stoa," but more likely Carroll meant the commoner word "oats." For several of the feet, however, the intended word is not clear. N o solution by Carroll has survived, and to this day there is contention among Carrollians over the precise set of twentyfour words Carroll had in mind. Readers are invited to make their own list to compare with the conjectures in the answer section.
208
CHAPTER SIXTEEN
6. THIRDMAN THEME Two kings are the sole occupants of a chessboard (see Figure 106). The task is to add a third man, creating a position that meets these provisos: 1. Neither king is in check. 2. The position can be reached in a legal game. 3. It can be proved, by retrograde analysis of previous legal moves, that neither side has a legal play. Note carefully the wording. It asks not for a double stalemate but only for a position in which neither side can move. The solution is unique. This sophisticated little problem appeared in The Problemist (September 1974, p. 471), where it was credited to G. Husserl ofIsrael. Newman Guttman called it to my attention. The original problem asked for the minimum number of men that must be added to the board to meet the conditions, but I have made the problem easier by stating that the minimum is one.
ANSWERS 1. The sixth figure is given in Figure 107, with the vertical lines of symmetry shown in gray. O n the right of each gray line are the numerals 1 through 6; on the left are their mirror reflections. 2. The first symbol of Tom Ransom's I.Q. test differs from the other four in having a gray border. The second symbol differs in not being shaded. The
Figure 106 Where and what is the third man?
THE SIXTH SYMBOL AND OTHER PROBLEMS
209
Figure 107 Solution to the sixthsymbol problem
third differs in not being square, and the fourth in having a gray spot. The fifth is the only symbol that is not unique with respect to some property, and therefore it is the one symbol "most different" from the others. T o put it differently, the first four are each different in the same general way, whereas the fifth is different in a different way, which makes it radically different. One is reminded of the story about the report of a psychiatrist on the most remarkable case in his experience. The patient was a person who showed no trace of any kind of neurosis. In my opinion Ransom has found a simple, elegant model of a semantic paradox that is frequently encountered but seldom explicitly recognized. 3. T o cut a square cake, frosted on the top and four sides, into n pieces of equal volume and equal frosting area, we need only mark the perimeter into n equal parts and cut the cake in the usual manner (see Figure 108). T o understand why this is so, we adopt a stratagem given by Norman N. Nelson and Forest N . Fisch in their article "The Classical Cake Problem" (TheMathematics Teacher 66, November 1973, pp. 659  661). Imagine the cake cut along its diagonals into four congruent triangular prisms, and the prisms arranged in a row as shown at the right in the illustration. The base line is the cake's perimeter. Divide the cake into seven equal portions, A, B, C, D, E, F, G, as shown by the seven gray lines. Those lines correspond to the seven cutting planes in the solution. It is easy to see that the area of icing on each portion is the same. The sums of the one or two rectangles that form the sides of each portion clearly are equal.
Figure 108 Solution to the squarecake problem
210
CHAPTER SIXTEEN
The tops of each portion (composed of one or two triangles) are also equal in area because the sums of the bases of these triangles are equal, and all triangles have the same altitude. Finally, the volumes of the seven portions are the same because their heights are equal and their tops are equal in area. The solution obviously generalizes to n portions. W e cannot, however, apply the procedure to a cake on which the frosting has a finite thickness, because the four corners of such a cake introduce complexities. Stephen I. Warshaw of the Lawrence Livermore Laboratory of the University of California generalized the cakecutting problem. The procedure given for the square cake, he showed, also applies to any cake with a polygonal base each side of which is tangent to the same circle. This includes all triangles and rhombuses, all regular polygons (as well as the limiting case of the circle), and all polygons of unequal sides that meet the proviso. Even more surprising, Warshaw found that this generalization led to the solving of a sticky problem in his work: "How to find a simple way to divide a discrete mesh cell into parts of given proportions in computer simulation studies involving hydrodynamic motions. The solution, once seen, is 'obvious.'" 4 . The French cryptarithm is solved with 94,851 6,483 6,483 = 107,817, and the German one with 821 821 821 821 = 3284. Many readers pointed out that the numeral 1 is commonly called eins in German, not ein as it appeared in the German cryptarithm. William C . Giessen was the first to add that if this change is made, the cryptarithm still has a unique solution: 1329 1329 1329 1329 = 5316. 5. The first published attempt to solve Lewis Carroll's anagram poem was made by Sidney H. Williams and Falconer Madan in their Handbook of the Literature ofthe Rev. C . L. Dodgson (Oxford University Press, 1931). Some of their errors were corrected in a revised edition, the Lewis Carroll Handbook (Oxford University Press, 1962),by Roger L. Green. For further criticism and speculation see Philip S. Benham, "Sonnet Illuminate," in Jabberwocky, the journal of the Lewis Carroll Society (Summer 1974). Drawing on the above sources, and with help from Spencer D. Brown, here is how matters stand on the words Carroll most likely had in mind.
+
+
+
+
+
As to: oats (not oast, stoa). The war: wreath (not thawer). Try elm: myrtle. I tried: tidier. The wig: weight. Cast in: antics (not sciant, actins, or nastic). I went: twine. T o ride: editor or rioted.
+
+
+
THE SIXTH SYMBOL AND OTHER PROBLEMS
211
Ring yes: syringe. W e rang: gnawer. Let's rap: plaster (not stapler, persalt, palters, psalter, or platers). W e don't: wonted. 0 shew: whose. Her wit: writhe (not wither, whiter). As yet: yeast. She won't: snoweth. Saw eel: weasel. In Rome: merino (not moiretz, minore). Dry one: yonder. He's wet: seweth (not thewes, hewest). I am dry: myriad. 0 forge: forego. Th' rogue: tougher (not rougeth). w h y a net: yawneth. Carroll's letter containing the puzzle poem first appeared in Six Letters by Lewis Carroll, privately printed in 1924. It is reprinted as Letter XLV in A Selectiorzfrom the Letters ofLewis Carroll to His Child~riends,edited by Evelyn M. Hatch (Folcroft, 1973). Now for some hot news concerning number 16 above. Can you rearrange the letters of "she won't" to make another common twoword phrase? 6. The only solution to the thirdman chess problem is to place a white bishop on the board as shown in Figure 109. Black obviously cannot move,
Figure 109 Solution to the thirdman theme
212
CHAPTER SIXTEEN
and White cannot move because it is not his move. T o prove that it must be Black's move (therefore Black is stalemated) requires elementary retrograde analysis that I leave to the reader. Many readers could not comprehend why the thirdman chess problem does not have many other solutions, such as a white pawn in place of the white bishop. The reason is that none of these positions rules out the possibility that it is White's move, and if White can move, the position fails to meet the demand that neither side can move. For instance, assume that the black king is on the QB1 cell. It is checked by an advancing white pawn. The black king moves to the corner. Now it is White's move. The given solution is the only one for which it can be shown by retrograde analysis that it must be Black's move, proving that the position is a stalemate for Black.
SEVENTEEN
Magic Squares and Cubes
"In my younger days, having once some leisure (which I still think 1 might have employed more usefully), I had amused myself in making . . . magic squares." BENJAMINFRANKLIN, from a letter
Two major breakthroughs have been made in the study of magic squares and magic cubes: All order5 magic squares have been counted, and the first perfect magic cube has been constructed. I am pleased to be the first to publish both results. So that the magnitude of the two achievements can be fully appreciated, let us take a brief look at the history of magic squares. Although some of the greatest mathematicians have done work on magic squares, and even though such work leads into the theories of groups, lattices, Latin squares, determinants, partitions, matrices, congruence arithmetic, and
2 14
CHAPTER SEVENTEEN
other nontrivial areas of mathematics, the most enthusiastic square constructors have been amateurs. The famous Franklin square, an ingenious 16by16 matrix that Benjamin Franklin called "the most magically magical of any magic square ever made by any magician," is itself the topic of many articles and monographs. The literature on magic squares in general is vast, most of it written by laymen who had become hooked on the elegant symmetries of these interlocking number patterns. A standard magic square, as few readers need to be told, is a square array of positive integers from 1 through N 2 , arranged so that the sum of every row, every column, and each of the two main diagonals is the same. N is the "order" of the square. It is easy to see that the magic constant is the sum of all the numbers divided by N. The formula is
The trivial square of order 1 is simply the number 1, and of course it is unique. It is equally trivial to prove that no order2 square is possible. There are eight ways to arrange the digits 1 through 9 in an order3 array that is magic. It is traditional, however, not to count rotations and reflections. When they are excluded, the order3 square is unique. T o appreciate the gemlike beauty of this most ancient of all combinatorial curiosities, consider all the ways that its constant, 15, can be partitioned into a triplet of distinct positive integers. There are exactly eight:
Now, in the order3 square each of eight lines of three numbers must total 15: the six orthogonals (rows and columns) and the two diagonals. The eight lines exactly match the number of triplets we have available. Since the center number belongs to a row, a column and both diagonals, it clearly must be a digit that appears in four of the eight triplets. The only such digit is 5. W e therefore know that 5 is the central number.
MAGIC SQUARES AND CUBES
215
Consider 9. It belongs to only two triplets. W e cannot place it in a corner since each corner cell belongs to three lines. Consequently it must go in a side cell. Because of the square's symmetry, it does not matter which side cell we choose, so let us put it above the 5. For the top corners, on each side of 9, we have no choices except 2 and 4. Again it does not matter which digit goes where, since one arrangement is merely a mirror reflection of the other. The rest of the square follows automatically. W e have by this simple construction proved its uniqueness. The completed square, in the form shown in Figure 110, is the Lo shu of ancient China. According to legend the pattern was first revealed on the shell of a sacred turtle that crawled out of the Lo River in the twentythird century B.c., but today's Chinese scholars trace references to it back no further than the fourth century B.C. From then until the tenth century the pattern was a mystical Chinese symbol of enormous significance. The even numbers were identified with yin, the female ~rinciple,and the odd numbers with yang, the male principle. The central 5 represented the earth, around which, in evenly balanced yin and yang, were the other four elements: 4 and 9 symbolizing metal, 2 and 7 fire, 1 and 6 water, and 3 and 8 wood. There are 880 magic squares (excluding rotations and reflections) of order 4. They were first given by Bernard Frenicle de Bessy in 1693. There are many ways to classify them. One of the best was devised by Henry Ernest Dudeney, who explains his system in an excellent article on magic squares in early printings of the fourteenth edition of the Encyclopaedia Britannica. The last printing of that edition substitutes for Dudeney's article a superb historical article by Schuyler Cammann. The current (fifteenth) edition has a trivial microarticle on magic squares in the Micropaedia. MFTAI
WOOD
WATER
Figure 110 T h e Lo shu magic square of ancient China
216
CHAPTER SEVENTEEN
How many magic squares are of order 5? The best estimate was given by Albert L. Candy in his Construction, Classijication and Census $Magic Squares of Order Five, privately published in Lincoln, Neb., in 1938. Candy arrived at a total of 13,288,952. The exact number was not known until 1973, when the counting was completed by a computer program developed by Richard Schroeppel, a mathematician and computer programmer at Information International. The program, using a standard backtracking procedure, consists of about 3500 "words" and took about 100 hours of running time on a PDP10. A final report, written by Michael Beeler, was issued in October 1975. Candy's estimate was low by a wide margin. Not counting rotations and reflections, there are 275,305,224 magic squares of order 5. Schroeppel prefers to divide that number by 4 and give the total as 68,826,306. The reason is that in addition to the eight variants obtained by rotation and reflection, there are four other variants generated by the following two transformations, which also preserve magic:
1. Exchange the left and right border columns, then exchange the top and bottom border rows. 2. Exchange rows 1 and 2 and rows 4 and 5. Then exchange columns 1 and 2 and columns 4 and 5. When these two transformations are combined with the two reflections and four rotations, the result is 2 X 4 X 2 X 2 = 32 forms that can be called isomorphic. With this definition of isomorphic the count becomes 68,826,306. That number can be lowered even more by considering another wellknown transformation. If every number in a magic square is subtracted from N2 1 (in this case 26), the result, called the complement, is also magic. When the center of an order5 square is 13, the complement is isomorphic with the original. If it is not 13, a different square results. If we broaden the term isomorphic to include complements, the count of order5 squares drops to about 35 million. The task of classifying order5 squares in meaningful ways is staggering. Dudeney once wrote that certain ways of dividing magic squares into types seemed to him about as useful as dividing people into those who take snuff and those who do not. Nevertheless, certain divisions yield unexpected results. Consider, for example, the total number of order5 squares with centers that are numbers 1 through 13:
+
MAGIC SQUARES AND CUBES
217
Note that the totals steadily increase from 1 through 8 but that the totals for 9 and 10 and for 11 and 12 are reversed. That there are more squares with a center of 11 than squares with a center of 12, and more squares with a center of 9 than squares with a center of 10, came as a surprise. Of course, these same anomalies occur in the counts of squares with centers 14 through 25, since every square with a center that is not 13 has its complement. There are as many squares with 1 in the center as there are with 25, and the same is true for all numbers except 13. Figure 111 shows an order5 square of a type that is more powerfully magic than any other. It is associative, which means that any pair of numbers sym
Figure 111 A pandiagonal, associative magic square of order 5
218
CHAPTER SEVENTEEN
+
metrically opposite the center add up to N2 1. And it is pandiagonal (sometimes called Nasik or diabolic), which means that its broken diagonals add up to 65, the constant. T o put it another way, ifwe tile the plane with this square, we can outline a 5by5 square anywhere on this infinite pattern and it will be magic, although not necessarily associative (see Figure 112). T o be associative too, it must have 13 in the center. The Lo shu is associative but not pandiagonal. An order4 square may be pandiagonal or associative but not both. The order5 square is the smallest one that can have both properties. Excluding rotations and reflections, 3600 order5 squares are pandiagonal, or if we also exclude variants obtained by the cyclic permutation of rows and columns, 144 are pandiagonal. In other words, there are 144 infinite patterns of the type shown here, each containing 25 pandiagonal order5 squares. Of the 144, just 16 contain a square that is also associative. All of this, by the way, was known before Schroeppel developed his computer program. o f the sixteen associative pandiagonal squares of order 5, four have 1 in the first cell, four have 1 in the third cell, four have 1 in the seventh and four have 1 in the eighth. The medieval Moslems were particularly intrigued by pandiagonal squares with 1 in the center. The patterns were not, of course, associa
Figure 112 Cyclic permutations of the order5 square
MAGIC SQUARES AND CUBES
219
tive, but the Moslems thought of the central 1 as being symbolic of the unity of Allah. Indeed, they were so awed by that symbol that they often left blank the central cell on which 1 was supposed to go. It is natural to extend the concept of magic squares to three dimensions and even higher ones. A perfect magic cube is a cubical array of positive integers from 1 to N 3 such that every straight line of N cells adds up to a constant. These lines include the orthogonals (the lines parallel to an edge), the two main diagonals of every orthogonal cross section and the four space diagonals. The constant is
There is, of course, a unique perfect cube of order 1, and it is trivially true that there is none of order 2. Is there one of order 31 Unfortunately, 3 does not quite make it. I do not know who first proved the impossibility, but Richard Lewis Myers, Jr., has a simple way of doing it. Consider any 3by3 cross section. Let A , B, C be the numbers of the first row, D, E, F the numbers ofthe third, and X the central number. Since the diagonals and the center column each must add up to 42, we can write
+ + + + +
From this we subtract A B C D E F = 2(42) to get 3X = 42, and X = 14. Since 14 cannot be the center of every cross section, the cube is impossible. Annoyed by the refusal of such a cube to exist, magiccube buffs have relaxed the requirements to define a species of semiperfect cube that apparently does exist in all orders higher than 2. These are cubes where only the orthogonals and four space diagonals are magic. Let us call them Andrews cubes, since W. S. Andrews devotes two chapters to them in his pioneering Magic Squares and Cubes (1917). The order3 Andrews cube must be associative, with 14 in its center. There are four such cubes, not counting rotations and reflections. All are given by Andrews, although he seems not to have realized that they exhaust all basic types. N o perfect cube of order 4 exists. As far as I know, the first proof was published by Schroeppel in "Artificial Intelligence Memo 239," (MIT, 1972). The first step is to show that on any 4by4 section (orthogonal or diagonal) the four corners must add up to the constant. Let Q be the constant, and label the sixteen cells with other letters (see Figure 113). The gray lines indicate six quadruplets that catch all sixteen cells. Since each corner cell is common to
220
CHAPTER SEVENTEEN
A
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Figure 113 Richard Schroeppel's proof, lemma 1
+ + +
three lines, 3 A 3 0 3M 3P plus each of the other cells taken once must equal 6Q. Ifwe subtract from this the values of the four rows, we are left with 2A 2 0 2M 2P = 2Q, which reduces to A D M P = Q , our first lemma. Now consider the cube's eight corners. W e prove that any two corners connected by an edge must have a sum of Q / 2 . Call the corners A and B. Let C, D and E, F be the corners of any two edges parallel to A , B (see Figure 114). ABDC, EFBA, and EFDC are each the corners ofa 4by4 cross section, so that their total is 3Q. Gather the like terms:
+ +
+
Figure 114 Richard Schroeppel's proof, lemma 2
+ + +
MAGIC SQUARES A N D CUBES
221
Divide each side by 2:
+ + +
+
From this we subtract C D E F = Q to obtain A B = Q/2, our second lemma. Now consider corner B. It is joined to corners A, D, F. Since A B = F B = D B, we can take B from each equality to prove that A = F = D. That is impossible, and so our proof is complete. Is there a perfect magic cube of order 5? N o one knows. Schroeppel has made a beginning by proving (using algebra and combinatorial thinking) that if such a cube exists, its center must be 63. There are perfect magic cubes of order 8. A method of constructing them by the millions was discovered in the spring of 1970 by Myers, when he was sixteen and a student at William Tennant High School in Johnsville, Pa. He sent me a short note about it, saying that he had obtained his first cube "after three months, seven theories, and thirtyone sheets of graph paper." I am embarrassed to confess that I did not appreciate the significance of his claim. He did not send an actual cube, and I replied by suggesting a mathematical journal that could evaluate his work. I next heard about Myers's cubes in December 1972 from John H. Staib, a mathematician at Drexel University in Philadelphia, where Myers had enrolled as a freshman. Staib sent an order8 cube (see Figure 115), and although he extolled its symmetries and provided hints about how Myers had constructed it (by superposing three Latin cubes and applying base8 notation), I still failed to comprehend the cube's importance. I had glanced too quickly at Andrew's book, noting references to magic cubes of order 3 and higher, without realizing that those cubes were only semiperfect. It was not until I started researching this report that I awoke to a full realization ofwhat Myers had done. Every orthogonal and diagonal line of eight numbers on the Myers cube shown in Figure 115, including the four space diagonals, add up to 2052. The cube is associative: Any two numbers symmetrically opposite the center add up to 513. It follows that not only do the eight corner cells of the cube total 2052 but also (as Staib pointed out) the corners of every rectangular solid centered in the cube add up to the same number. Ifthat is not enough, the cube can be sliced into sixtyfour order2 cubes, and the eight numbers in each of these cubes add up to the constant!
+
+
+
222
CHAPTER SEVENTEEN
Figure 115 Cross section of Richard Lewis Myers's magic cube of order 8 [Computer printout courtesy of William Gosper]
MAGIC SQUARES AND CUBES
223
These remarkable symmetries make possible an enormous number of rearrangements of the cube, all in a sense isomorphic, and of course every arrangement can be rotated and reflected fortyeight ways. Imagine this cube with each of its 512 cells replaced by the same cube in any of its thousands of rearrangements or orientations. In cell 1 we put a cube that starts with 1. In cell 2 we put a cube that starts with 83 1 = 513, in cell 3 we put a cube that starts with (2 X 83) 1 = 1025, and so on for the other cells. The result is a perfect magic cube of order 64.Order64 cubes in turn will build perfect magic cubes of order 512, and the same is true for all higher orders that are powers of 8. How many order8 magic cubes are there? By choosing different Latin cubes to superpose, Myers can construct millions different from one another and from the one given here, although not all will be associative. The number of Latin squares of order 8 have been counted (there are billions), but the number of Latin cubes has not, so that the problem ofcounting just the order8 cubes that can be generated by Myers's procedure alone is horrendous. Is 8 the lowest order aperfect magic cube can have? Are there magic cubes of orders not in the 8power series? Both are open questions.
+
+
ADDENDUM When I wrote the foregoing text 1believed that Richard Myers was the first to construct a perfect magic cube of order 8. As I quickly learned (of course, this in no way detracts from Myers's achievement), this is not the case. Victor Meally called my attention to the construction of a perfect magic cube of order 8 in the late 1930s by J. Barkley Rosser and Robert J. Walker. It is pandiagonal in the sense that if any of its three sets of parallel sections are cyclically permuted, the cube remains perfectly magic. The cube was not published, but its construction is given in a manuscript deposited in a library at Cornell University. In the manuscript, the authors show that perfect pandiagonal cubes exist for all orders that are multiples of 8 and all odd orders higher than 8. A brief summary of the construction technique for the order8 cube is given in W. W . Rouse Ball's Mathematical Recreations and Essays (Dover, 1987), revised by H. S. M. Coxeter. Rosser and Walker were not, however, the first to construct a perfect magic cube of order 8. James G. Mauldon, an Amherst College mathematician, wrote to me about a remarkable 1888 paper by F. A. P. Barnard. He credits the first such cube to an unknown person who apparently published it in The Commercial, a Cincinnati newspaper, on March 11, 1875. (Details can be found in the fifth chapter of Magic Cubes by William Benson and Oswald
224
CHAPTER SEVENTEEN
Jacoby.) Mauldon also sent his construction of a perfect associative magic cube of order 7, and perfect associative pandiagonal magic cubes of orders 9 and 11. The first perfect pandiagonal cube of order 8 seems to have been constructed by C. Planck, who reported it in his rare, littleknown Theory o a t h Nasiks (privately published in Rugby, England, 1905). Planck showed that in kspace, where k is 2 or higher, the smallest perfect pandiagonal "cube" is of order 2', and the smallest that is also associative (symmetrically opposite pairs 1. W e saw how that is true of magic of numbers have the same sum) is 2' squares, where k = 2. In 3space the smallest perfect pandiagonal cube is of order 8, and the smallest that is also associative is of order 9. Myers's cube is associative but not pandiagonal. The first publication known to me of a perfect magic cube of order 7 is in Play Mathematics by Harry Langman (Hafner, 1962, pp. 75 76). After this chapter was published as a column (January 1976), scores of readers sent me such order7 cubes, as well as construction procedures for orders 9, 11, and higher. Readers also explored magic cubes in dimensions above 3. For a listing of articles on magic cubes and hypercubes that have appeared in the]ournal of RecreationalMathematics, see the bibliography ofJohn Hendricks's 1985 paper. As far as I know, no one has yet found a perfect magic cube of orders 5,6, or 10, or proved that such cubes cannot exist. The 880 magic squares of order 4 are depicted in the 1976 book on magic squares by Benson and Jacoby, along with considerable original material that appears in print for the first time. The book gives the fantastic trimagic square of order 32 constructed by Captain Benson in 1949. This is a square that retains its magic not only when every number is squared but also when every number is cubed. The simplestknown square of that type was of order 64, before Captain Benson found a way to halve the order.
+
B
I
B
L
I
O
G
R
A
P
H
Y
"Theory of Magic Squares and Cubes." F. A. P. Barnard, in The Memoirs ofthe National Academy of Sciences 4, 1888, pp. 209270.
Magic Squares and Cubes. W . S. Andrews. Open Court, 1917. Dover reprint, 1960. "The Algebraic Theory of Diabolic Magic Squares." J. Barkley Rosser and Robert J. Walker, in Duke Mathematical Journal 5, December 1939, pp. 705 728. "Magic Squares." Martin Gardner, in The Second Scientific American Book $Mathematical Puzzles and Diversions. Simon and Schuster, 1961.
Magic Squares. John L. Fults. Open Court, 1974.
MAGIC SQUARES AND CUBES
225
"Perfect Magic Cubes of Order 7." B. E. Wynne, in Journal $Recreational Mathematics 8, 1975 1976, pp. 285293.
New Recreations with Magic Squares. William H . Benson and Oswald Jacoby. Dover, 1976. "Magic Cubes and Prouhet Sequences." Allan Adler and ShuoYen R. Li, in The American Mathematical Monthly 84, October 1977, pp. 618627.
Magic Cubes. William H. Benson and Oswald Jacoby. Dover, 1981. The Wonders $Magic Squares. Jim Moran. Vintage, 1982. Researches in Magic Squares. Akira Hirayama and Gakuho Abe. Osaka Kyoikutosho Co., Osaka, Japan, 1983. This 316page book is so filled with new material on magic squares, cubes, and related constructions that it deserves an English translation. "Ten Magic Tessaracts of Order Three." John R. Hendricks, in Journal Mathematics 18, 1985 1986, pp. 125 134.
of Recreational
EIGHTEEN
Block Packing
"Pack my box with five dozen liquor jugs." Anonymous
pangram
A pangram is an attempt to pack as many different letters as possible into the shortest intelligible sentence. It is considered not cricket to use names and initials, such as "Schwartz" and "X. Q. Zym," or strange words, such as "pyrzqxgl," which in The Magic ofOz enables you to change instantly into any kind of animal you like if you know how to pronounce it correctly. Many ultimate pangrams of twentysix different letters have been constructed by word players, but they tend to be inelegantly obscure, for example "Vext cwm fly zing jabs kurd qoph," by Dmitri Borgmann. It means that an annoyed fly in a Welsh mountain hollow, humming shrilly, pokes at the nineteenth letter of the Hebrew alphabet drawn by a Kurd. Cryptographers find a perfect pangram amusing because it can be written in a simple lettersubstitution cipher as ABCDE FGHIJ KLMNO PQRST UVWXY Z.
228
CHAPTER EIGHTEEN
The difference between creating pangrams and working on packing problems in combinatorial geometry is not as great as one might suppose. The restraints in the former are the formation rules of English spelling and grammar, and those in the latter are the rules of mathematics. At least two eminent mathematicians, Augustus De Morgan and Claude E. Shannon, are on record as having spent considerable time composing pangrams, and I know of many lesser mathematicians who have tried their hand at it. In mathematics a packing problem in general is one in which a given set of mathematical objects are to be packed as efficiently as possible into a given space according to given rules. Computer scientists, for example, are concerned with finding fast algorithms for packing sets of numbers into "bins," with the sum of the numbers in each bin not tdexceed a specified limit. Such algorithms are needed for the efficient storage  and retrieval of information. Geometrically the task can be viewed as a problem in onedimensional packing: packing rods of varying lengths inside long pipes into which the rods fit snugly. In a complex industrial society all kinds of problems arise involving the packing of threedimensional objects into a specified area: the storing of objects in a warehouse; the packing of supplies into ships, planes, and freight cars; the packing of objects in cartons for distribution to stores, and so on. Perhaps it is the increasing need for packing algorithms that has stimulated some mathematicians in recent years to spend more time on such problems. Here, we consider only the simplest kind of solid packing: the packing of "bricks" (rectangular parallelepipeds) into a "box" (also a rectangular parallelepiped). T o simplify still more, we assume that all three dimensions of both bricks and box are integral, and that the volume of the box exactly equals the total volume of the bricks to be put inside. As David A. Klarner says in his article "Brickpacking Puzzles," many people are surprised to learn that even with these strong simplifications there are problems that are both elegant and challenging. By elegant Klarner means the following. If the bricks will pack the box perfectly, the problem is elegant if finding a way to pack it seems simple but is actually difficult. And if the bricks will not pack the box, the problem is elegant if there is a simple but subtle way to prove impossibility. Klarner, now at the University of Nebraska in Lincoln, is one of the pioneers of brickpacking theory. It is to him that I am indebted for most of what follows. About 1960 the Dutch mathematician Nicolaas G. de Bruijn was struck by the fact that his son, age seven, was unable to fill a 6 X 6 X 6 box with twentyseven bricks, each 1 X 2 X 4. The two volumes are the same and the packing seems easy, but one always ends up with at least one hole that the last
BLOCK PACKING
229
brick will not fill. Studying the matter led de Bruijn to interesting results. They were first published as problems in a Hungarian journal, then later summarized by de Bruijn in his paper "Filling Boxes with Bricks". De Bruijn calls a brick harmonic if its three measurements are integral and can be ordered so that each length is a multiple of the preceding one. In algebraic terms a harmonic brick has the form a X ah X abc, where the letters are positive integers. The 1 X 2 X 4 brick is, of course, harmonic. It is called the canonical brick because it not only is the simplest harmonic brick of three distinct measurements but also approximates the shape of ordinary bricks used in masonry (see Figure 116). De Bruijn was able to prove that a collection of identical harmonic bricks, each a X ab X abc, will perfectly pack a box if and only if the box is a x X aby X abcz. "Perfectly pack" means to fill completely; it is the same as saying that the brick will tile the box. De Bruijn showed that perfect packing is possible only if the box's dimensions are multiples of the brick's dimensions. T o put it another way, if the bricks pack the box at all, there will be a way to do it trivially. That means they will pack when all are identically oriented. (Of course, they may also pack in nontrivial ways.) If the bricks are not harmonic, there are boxes they will fill only in a nontrivial way. For example, five nonharmonic bricks of 1 X 2 X 3 will pack a 1 X 5 X 6 box, but they cannot pack the box if they are all parallel. De Bruijn's results generalize to hyperbricks in all higher Euclidean spaces, and they also hold for 2space "bricks" (rectangles). The canonical plane brick the 1 X 2 domino will pack a rectangle only if one side is even, and then, of course, a trivial packing is possible. Let us return to the task that puzzled de Bruijn's son. Since 6 is not a multiple of 4, we know from de Bruijn's work that the canonical brick will not pack an order6 cubical box. Is there a simple impossibility proof?
Figure 116 The canonical brick
230
CHAPTEK EIGHTEEN
There is, and it is a generalization of the solution to the old brainteaser about an order8 checkerboard that has had two diagonally opposite corner squares removed. Can the board be covered with thirtyone dominoes?The fact that it cannot is evident once you rcalizc that the two missing squares are the same color. The board therefore contains thirtytwo squares of one color and thirty of the other. Since a domino must cover two squares of opposite color, after thirty dominoes are put down there will always be two uncovered squares of the same color that cannot be covered by the last domino. T o apply the same kind of parity check to the cubepacking problem, imagine that the order6 cube is divided into twentyseven order2 cubes and that the order2 cubes are shaded as is shown in Figure 117. No matter how a canonical brick is oriented inside such a cube, it will fill four shaded cells and four white cells. The cube, however, has eight more shaded cells than white
Figure 117 Color scheme for an order6 cubepacking problem
BLOCK PACKING
231
cells. Therefore, after twentysix bricks are placed, eight cells of the same color will remain. Clearly the last brick cannot fill them. Will 125 canonical bricks pack a 10 X 10 X 10 box? They will not, and the same impossibility proof applies. Indeed, the proof applies to any cube with a side that is even and not a multiple of 4. Will 250 bricks of 1 X 1 X 4 pack the order10 cube? As the impossibility proof shows, the answer is no. Sometimes more than two colors are needed for elegant impossibility proofs, but it is surprising how much can be done with just two colors. Here is a delightful tiling problem from Klarner that is easily proved impossible by twocoloring, although not a checkerboard coloring. You have a 25 X 25 square you want to tile (no overlap, no vacancies) with a mixture of 2 X 2 squares and 3 X 3 squares. I shall give his impossibility proof in the answer section. One of Klarner's theorems introduces the concept of cleavability. If a rectangle can be tiled with identical rectangles, there is always a way to tile it so that it can be cut into two smaller rectangles, each ofwhich can also be tiled. Such a rectangle is said to be cleavable. Does this unusual theorem have a 3space analogue? That is, if a box can be fully packed with identical bricks, can it always be cut into two smaller boxes, each packable with the same bricks? The answer, Klarner found, surprisingly is no. The smallest example (discovered by David Singmaster) is the 5 X 5 X 12 box. It can be packed with 1 X 3 X 4 bricks, but not in a way that is cleavable. Is there an infinite number of noncleavable boxes a given brick will pack? Again the answer, surprisingly, is no. Klarner was the first to show that in any Euclidean space an infinite set of boxes packable by a given brick has a finite subset of packable boxes that can be used for packing all the others. In addition, he showed that for every brick, there is a finite set ofpackable but noncleavable boxes. In 1971 two Hungarian mathematicians, G. Katona and D. Szasz, refined Klarner's findings by giving a constructive proof that included specific numerical bounds. If instead of identical bricks, we allow a mixture of different bricks (as in Klarner's tiling problem), many beautiful new problems arise. One of the simplest is a 3 X 3 X 3 box puzzle that, Klarner says, first appeared in a Dutch book in 1970. W e want to pack it with six 1 X 2 X 2 bricks and three unit cubes (see Figure 118). It looks ridiculously easy, yet many find it irritatingly difficult. The reader is urged to construct a set, either by making the pieces out of wood or by gluing cubes together. The packing has a unique solution (not counting rotations and reflections) requiring that the three unit cubes be along a space diagonal (see Figure 119).
232
CHAPTER EIGHTEEN
Figure 118 The 3 X 3 X 3 packing puzzle
T o prove it, first consider any 3 X 3 cross section. If it is checkerboardcolored, five cells are of one color and four are of the other. No matter how a 1 X 2 X 2 brick is placed, it will occupy 0,2, or 4 cells of each cross section, with the colors of the occupied cells evenly divided. As a result, each of the nine sections must have one and only one cell occupied by a unit cube. (A
Figure 119 Key to the 3 X 3 X 3 packing puzzle
BLOCK PACKING
233
section cannot contain all three cubes because that would force some sections to be without cubes.) In addition, in every section the unit cube must be at the center or at one of the corners. The only way to meet these requirements is to place the cubes along a space diagonal. Placing the six bricks then follows automatically. John Horton Conway of the University of Cambridge set himself the task a few years ago of designing a more difficult cubepacking puzzle. Conway's cube, although it appears to be almost as easy as the 3 X 3 X 3 puzzle, is so hard that some people cannot solve it until they are told its whimsical secret. Conway's cube has many variants. The one he likes best requires eighteen harmonic bricks (see Figure 120). The task is to pack them into a 5 X 5 X 5 box or, what is the same thing, to build an order5 cube. It is fiendishly difficult if one tries to solve it by trial and error. Returning to the packing of identical bricks, we can ask the following general question. If a box is not perfectly packable with a set of bricks, what is the maximum number of bricks that will go into it? Even when the bricks are harmonic, it is an extremely difficult problem, although a good start has been made in a paper published in 1974 by Richard A. Brualdi and Thomas H. Foregger. The authors define a "representing set," abbreviated R, as a set of cells in the box such that no matter where a brick is placed, it will occupy at least one cell of R. When R is made as small as possible, it is said to have minimum cardinality. The authors show that the maximum number of identical bricks
Figure 120 The eighteen bricks of John Horton Conway's cubepacking puzzle
234
CHAPTER EIGHTEEN
(not necessarily harmonic) that will go into a box is equal to or less than the minimum cardinality of R. In 2space the maximum number of harmonic bricks (dominoes) invariably equals the minimum cardinality of R, but that does not hold for higher spaces. Consider canonical bricks and cubical boxes. The order4 cube is the smallest that is packable, and ofcourse the packing is trivial. The order6 cube, as we have seen, is not packable, but all bricks but one are easily put inside. What about the order5 cube? It has 125 cells. That is not a multiple of 8, so that it is not fully packable by canonical bricks. Will fifteen such bricks (total volume 120) go inside? T o put it another way, can you build an order5 cube with fifteen canonical bricks and five unit cubes? The minimum cardinality of R is 15, but try as you will it is impossible to get fifteen bricks inside. After fourteen bricks have been placed, the thirteen remaining cells will never accommodate the last brick. Brualdi and Foregger have a complicated coloring proof of the impossibility, but in meditating about it one afternoon, I had a happy inspiration. It led to the following reductio ad absurdurn proof. Assume that fifteen bricks will pack. The total surface area of the order5 cube is 6 X 25 = 150. Since each face has an odd number of squares, one cell on each face must be filled by a unit cube. (No more than one unit cube can occupy a cross section because there are fifteen such sections and only five unit cubes.) This leaves a surface of 150  6 = 144 to be packed by faces of the canonical bricks. Now each brick must have one and only one of its 1 X 2 ends on the surface. This leaves a surface of 144  (15 X 2) = 114 to be packed by 1 X 4 and 2 X 4 rectangles. But 114 is not evenly divisible by 4. Therefore the original assumption is false. A similar proof is obtained by considering the cube's three planar midsections. Each plane is a 5 X 5 matrix, making seventyfive cells in all. One cell in each 5 X 5 square must be intersected by a unit cube. (This could be done with a single unit cube at the center, or two or three unit cubes suitably placed.) Each of the fifteen canonical bricks must intersect two cells, or thirty cells in all. Taking thirtythree cells from seventyfive leaves fortytwo to be intersected in sets of four or eight, and since 42 is not a multiple of 4, impossibility follows. The next cube of interest is the order 7. It is easy to put fortyone canonical bricks inside it, but will it hold fortytwo, leaving seven holes to be filled by unit cubes? Surprisingly the answer is not known. Foregger posed this as an unsolved problem (E 2524) in the March 1975 issue of American Mathematical Monthly. Although every canonical brick must intersect a pair of cells in the planar midsections, after the necessary subtractions are made, the remaining cells are a multiple of 4. Therefore the previous impossibility proof does not apply.
BLOCK PACKING
235
Brualdi and Foregger found many special cases in which the maximum number of harmonic bricks equals the minimal cardinality of R.For example, if the smallest face of the brick packs each face of the box, there is equality. For canonical bricks there is equality if one of the box's dimensions is even. The general problem, however, is far from solved. I can think of no better way to conclude than by quoting the final sentence of Klarner's article: "A word of warning in closing! Engaging in experiments with little wooden blocks is fraught with the danger that friends, family and colleagues will assume you are entering your second childhood, and that you should be put away. A good defense is to have a few copies of Conway's puzzle on hand to divert their attention while you make a getaway."
ANSWERS Readers were asked to prove that it is impossible to tile a square of side 25 with a mixture of 2 X 2 and 3 X 3 squares. David A. Klarner's proof begins by twocoloring the square with stripes (see Figure 121). N o matter how a 2 X 2 square is placed, it will cover two colored cells and two white ones. Since the large square contains an excess of twentyfive colored cells, the excess will remain no matter how many 2 X 2 squares are placed or where they are placed. If the square can be tiled, we must therefore find a way to place a set of 3 X 3 squares so that they cover twentyfive more colored cells than white ones. It is apparent that a 3 X 3 square, however placed, covers either six colored cells and three white ones or six white cells and three colored ones. In each case, the difference is 3. But 25 is not a multiple of 3, and so there is no way the 3 X 3 squares can cover an excess of twentyfive colored cells. Therefore the tiling is impossible. Klarner has shown in general that if a rectangle can be tiled at all by squares of a X a and b X b, it can always be split into two rectangles (one possibly empty), one of which is tilable by a X a squares alone and the other tilable by b X b squares alone. The second task was to build a 5 X 5 X 5 cube with eighteen specified bricks. In working on puzzles of that type one tends to place large pieces first, then try to fit smaller ones into the gaps. In this case it is a fatal strategy. If you worked on the puzzle by trial and error, you probably found that if you left the three small bricks to the last, they never fitted. The puzzle is difficult precisely because those three pieces must be in a unique configuration, and it is unlikely that you would hit on it by chance. Let us approach the problem by way of combinatorial geometry. If we checkerboardcolor a 5 X 5 cross section, we find that it contains thirteen cells
236
CHAPTER EIGHTEEN
25
Figure 121 David A. Klarner's impossibility proof
of one color and twelve of the other (see Figure 122). Consider all the bricks except the three small ones. No matter how one brick is placed, it must fill an even number of cells (0,2,4, or 8) in every section. Half of the cells it fills will be one color and half will be the other. Since there are an odd number of cells in each section, one or three of its cells must be occupied by a 1 X 1 X 3 brick. In addition, the brick must be placed so that if it fills one cell, it will be a cell of the same color as the central cell. If it lies entirely within the section, two of its three cells must match the color of the central cell.
BLOCK PACKING
237
N o matter how a small piece is placed, it will occupy five sections. It follows that the three pieces, if they contribute to all fifteen sections (as they must), will have to be placed so that no two contribute to the same section. There is only one way to do that and at the same time meet the coloring requirements. Once the small pieces are correctly positioned, it is not hard to find one of the more than 500 ways of packing the larger bricks around them. John Horton Conway, who invented this cube puzzle, could have substituted another 1 X 2 X 4 for the 2 X 2 X 2, thus providing the maximum number of canonical bricks that will go into a 5 X 5 X 5 box. Adding the order2 cube, however, provides amusing misdirection. The puzzle can also be given with twentynine bricks of 1 X 2 X 2, together with the three 1 X 1 X 3 bricks. Whether most people would find that easier or more difficult is hard to say. The principle behind Conway's cube generalizes in the sense that three 1 X 1 X (2n  1)bricks have a unique configuration in a cubical box of side 2n 1 that makes it possible to pack the rest of the box with 1 X 2 X 2 bricks.
+
ADDENDUM The chapter posed an unsolved problem about the packing of "canonical bricks" ( I X 2 X 4) into cubes of side n, with n greater than 3. When n is even, the cube can be fully packed (in a trivial way) only if n is a multiple of 4. I gave a simple proof that if n is even and not a multiple of 4, complete packing is impossible. It is necessary to omit one brick.
Figure 122 Key to John Horton Conway's cube puzzle
238
CHAPTER EIGHTEEN
When n is odd, the situation is more interesting. Each n X n "slab" must contain a unit hole; consequently the maximum number of bricks that an oddorder cube can take is (n3  n)/8. For what values of n can this maximum packing be achieved? When I wrote, only the 5cube had been proved impossible. The proof was later generalized by David A. Klarner to cover exactly half of all odd n, namely cubes with sides equal to plus or minus 1 (modulo 8). Many impossibility proofs for the 7cube were found by readers, but they did not generalize. The problem is now solved. Robert Ammann of Lowell, Mass., first proved impossibility of maximal packing for all odd n except n equal to plus or minus 1 (modulo 16). He then found a maximal packing for the 15cube that generalized to provide maximal packings for all odd n equal to plus or minus 1 (modulo 16). Frank Barnes, working independently in England, obtained Ammann's earlier impossibility result and conjectured that maximum packings exist for the 15cube and all others with sides plus or minus 1 (modulo 16). Told of Ammann's work, Barnes verified its soundness. In addition, he found a way to show that all impossible cubes could be packed by omitting just one brick from the maximum number. It is hoped that all these results will eventually be published. Neither Ammann nor Barnes has an academic post. Ammann is a computer programmer working in nonmathematical areas. Although Barnes has lectured on mathematics at the University of Michigan and the University of Reading, he has been earning a living for the past two years by piloting a hotair balloon for advertising purposes. After the above comments ran in my column for October 1976, a proof by Michael Mather, that the order7 box could not be packed with fortytwo canonical bricks, was published (The American MathematicalMonthly 83, November 1976, pp. 741  742) as a solution to Problem E 2524. Mather showed more generally that it is possible to pack a cube of side 2n 1with 2n(2n I) bricks, each of size 1 X 2 X (n I), if and only if n is even or equal to 1. David Klarner found what he calls a "divide and conquer" way to put fortyone canonical bricks into the order7 cube. Cut the cube into three subboxes 2 X 7 X 7, 3 X 5 X 7, and 4 X 5 X 7. "It is easy," he writes in a 1987 letter, "to pack twelve bricks into the 2 X 7 X 7, and even easier to pack seventeen into the 4 X 5 X 7." Putting twelve bricks into the 3 X 5 X 7 is not so easy, but it can be done. There are nine holes. One arrangement has three 1 X 1 X 3 holes. Klarner uses the term "atomic" for a box (not necessarily a cube) that cannot be optimally packed by divide and conquer. In other words, no maximal
+
+
+
BLOCK PACKING • 239
packing will split into smaller boxes that can be optimally packed. The most interesting packing problems involve atomic boxes. We can ask: does the canonical brick have an infinite or finite number of atomic boxes it will pack? The question is open. Some years ago Klarner proved that in the case of tiling a box (packing completely with no holes), for every brick there is a finite set of boxes with integer sides that are atomic. For several years he thought a similar theorem applied to optimal packing with holes. That is, for any given brick there is a finite number of atomic boxes it will optimally pack. “How wrong I was!” he writes. He has since proved that for the 1 × 2 × 2 brick there is an infinite number of atomic cubical boxes it will pack optimally. Much remains unknown about packing canonical bricks into noncubical boxes. Klarner tells me that his student Wade Satterfield has shown that twentyone bricks cannot fit into a 5 × 5 × 7 box. (It is a simple, clever proof, not yet published, based on a parity contradiction.) The same proof also shows that twentyseven canonical bricks will not pack the 5 × 5 × 9. Will the proof generalize to show that 3n such bricks will not pack a 5 × 5 × n box? No, writes Klarner, it breaks down for all odd n greater than 9 “It is an open question,” he continues, “whether there exists any odd number n such that one can pack 3n canonical bricks into a 5 × 5 × n box. Since one can pack six canonical bricks into a 5 × 5 × 2 box, it follows that if one can pack 3k such bricks into a 5 × 5 × k box for some odd number k, then this can be done for all numbers larger than k.” B
I
B
L
I
O
G
R
A
P
H
Y
“Filling Boxes with Bricks.” N. G. de Bruijn, in The American Mathematical Monthly 76, January 1969, pp. 37–40. “Matching Problems.” G. Katona and D. Szász, in Journal of Combinatorial Theory 10, February 1971, pp. 60–92. “BrickPacking Puzzles.” David A. Klarner, in Journal of Recreational Mathematics 6, Spring 1973, pp. 112–117. “Packing Boxes with Harmonic Bricks.” Richard A. Brualdi and Thomas H. Foregger, in Journal of Combinatorial Theory 17, October 1974, pp. 81–114. “A Finite Basis Theorem for Packing Boxes with Bricks.” N. G. de Bruijn and David A. Klarner, in Philips Research Reports 30, 1975, pp. 337–343. “Box Packing Problems.” Ross Honsberger, in Mathematical Gems 2, Chapter 8. Mathematical Association of America, 1976. “Packing Problems and Inequalities.” D. G. Hoffman, in The Mathematical Gardner. David A. Klarner, ed. Prindle, Weber and Schmidt, 1981.
NINETEEN
Induction and Probability
"The universe, so far as known to us, is so constituted, that whatever is true in any one case, is true in all cases of a certain description; the only difficulty is, to find what description." JOHN
STUARTMILL, A System of Logic
Imagine that we are living on an intricately patterned carpet. It may or may not extend to infinity in all directions. Some parts of the pattern appear to be random, like an abstract expressionist painting; other parts are rigidly geometrical. A portion of the carpet may seem totally irregular, but when the same portion is viewed in a larger context, it becomes part of a subtle symmetry. The task of describing the pattern is made difficult by the fact that the carpet is protected by a thick plastic sheet with a translucence that varies from place to place. In certain places we can see through the sheet and perceive the pattern;
242
CHAPTER NINETEEN
in others the sheet is opaque. The plastic sheet also varies in hardness. Here and there we can scrape it down so that the pattern is more clearly visible. In other places the sheet resists all efforts to make it less opaque. Light passing through  the sheet is often refracted in bizarre ways, so that as more ofthe sheet is removed, the pattern is radically transformed. Everywhere there is a mysterious mixing of order and disorder. Faint lattices with beautiful symmetries appear to cover the entire rug, but how far they extend is anyone's guess. N o one knows how thick the plastic sheet is. At no place has anyone scraped deep enough to reach the carpet's surface, if there is one. Already the metaphor has been pushed too far. For one thing, the patterns of the real world, as distinct from this imaginary one, are constantly changing, like a carpet that is rolling up at one end while it is unraveling at the other end. Nevertheless, in a crude way the carpet can introduce some of the difficulties philosophers of science encounter in trying to understand why science works. Induction is the procedure by which carpetologists, after examining parts of the carpet, try to guess what the unexamined parts look like. Suppose the carpet is covered with billions of tiny triangles. Whenever a blue triangle is found, it has a small red dot in one corner. After finding thousands of blue triangles, all with red dots, the ~arpetolo~ists conjecture that all blue triangles have red dots. Each new blue triangle with a red dot is a confirming instance of the law. Provided that no counterexample is found, the more confirming instances there are, the stronger is the carpetologists' belief that the law is true. The leap from "some" blue triangles to "all7' is, of course, a logical fallacy. There is no way to be absolutely certain, as one can be in working inside a deductive system, what any unexamined portion of the carpet looks like. O n the other hand, induction obviously works, and philosophers justify it in other ways. John Stuart Mill did so by positing, in effect, that the carpet's pattern has regularities. He knew this reasoning was circular, since it is only by induction that carpetologists have learned that the carpet is patterned. Mill did not regard the circle as vicious, however, and many contemporary philosophers (R. B. Braithwaite and Max Black, to name two) agree. Bertrand Russell, in his last major work, tried to replace Mill's vague "nature is uniform" with something more precise. He proposed a set of five posits about the structure of the world that he believed were sufficient to justify induction. Hans Reichenbach advanced the most familiar of several pragmatic justifications. If there is any way to guess what unexamined parts of the carpet look like, Reichenbach argued, it has to be by induction. If induction does not work, nothing else will, and so science might as well use the only tool it has. "This answer is not fallacious," wrote Russell, "but I cannot say that I find it very satisfying." 

INDUCTION AND PROBABILITY
243
~ u d o l Carnap f agreed. His opinion was that all these ways of justifying induction are correct but trivial. If "justifym is meant in the sense that a mathematical theorem is justified, then David Hume was right: There is no justification. But if "justify" is taken in any of several weaker senses, then, of course, induction can be defended. A more interesting task, Carnap insisted, is to see whether it is possible to construct an inductive logic. It was Carnap's great hope that such a logic could be constructed. He foresaw a future in which a scientist could express in a formalized language a certain hypothesis together with all the relevant evidence. Then by applying inductive logic, he could assign a probability value, called the degree of confirmation, to the hypothesis. There would be nothing final about that value. It would go up or down or stay the same as new evidence accumulated. Scientists already think in terms of such a logic, Carnap maintained, but only in a vague, informal way. As the tools of science become more powerful, however, and as our knowledge of probability becomes more precise, perhaps eventually we can create a calculus of induction that will be of practical value in the endless search for scientific laws. In Carnap's Logical Foundations ofProbability (University of Chicago Press, 1950) and also in his later writings, he tried to establish a base for such a logic. How successful he was is a matter of dispute. Some philosophers of science share his vision (John G. Kemeny for one) and have taken up the task where Carnap left off. Others, notably Karl Popper and Thomas S. Kuhn, regard the entire project as having been misconceived. Carl G. Hempel, one of Carnap's admirers, has argued sensibly that before we try to assign quantitative values to confirmations, we should first make sure we know in a qualitative way what is meant by "confirming instance." It is here that we run into the worst kinds of difficulty. Consider Hempel's notorious paradox of the raven. Let us approach it by way of 100 playing cards. Some of them have a picture of a raven drawn on the back. The hypothesis is: "All raven cards are black." You shuffle the deck and B. C.
by Jahnnv hart CCULD BE A LILT