Académique Documents
Professionnel Documents
Culture Documents
Golomb Reviewed work(s): Source: Mathematics Magazine, Vol. 77, No. 1 (Feb., 2004), pp. 46-55 Published by: Mathematical Association of America Stable URL: http://www.jstor.org/stable/3219230 . Accessed: 30/10/2011 14:16
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Mathematical Association of America is collaborating with JSTOR to digitize, preserve and extend access to Mathematics Magazine.
http://www.jstor.org
OT
E S
SOLOMON
W. GOLOMB
Figure A tromino 1 We will classify a varietyof nearlyrectangular shapesinto those thatcan be tiled by trominoesand those thatcannot.Fromnow on we will simply say tiled to mean tiled by trominoes. will considershapesthat are integer-dimensioned We rectangles with each dimensionat least 2, and with one or two 1 x 1 squaresremoved.If one square removed, the resultant is call shapea deficientrectangle. the removedsquare If was a cornersquare,call the resultingdeficientrectanglea dog-earedrectangle.The areaof a trominois 3, so, evidentially, only shapeswhose areais a multipleof 3 can be tiled by trominoes. this paperwe will determine In whichdeficientrectangles with areadivisibleby 3 aretileableandwhich arenot;in particular, the dog-eared all ones aretileable.We will also get some partialresultsfor the samequestionfor rectangles with two squares removedandremaining areadivisibleby 3. We especiallyrecommend proof of the DeficientS x S Lemmato the casual the reader. Trominoeswere introducedby Golomb [3], who proved that deficient squares whose side lengthis a powerof two canbe tiled.ChuandJohnsonbaugh extended first Golomb'swork to the generalcases of deficientsquares[1]. They later went on to rectanglesand proveda slightly weakerversion [2] of what we call the Deficient RectangleTheorem.The Proposition the last section answersa questionposedby in ChuandJohnsonbaugh [1]. Before proceeding with the businessat hand,we will mentiona few generalfacts abouttrominoes aboutanother and tilingquestion involvingtiles otherthantrominoes.
46
47
A tromino is a speeial ease of a polyomino, a shape made by eonneeting eertain numbers of 1 x 1 squares, eaeh joined together with at least one other square along an The edge. The polyomino of area 1, a single 1 x 1 tile is ealled a monomino. polyomino of area 2 is the domino. Let P (n) be the number of distinet polyominoes of area n. For example, P(3) = 2, sinee there are aetually two trominoes: a straight tromino, whieh has the shape of a 1 x 3 reetangle, and the objeet shown in FIGURE1, whieh is ealled a right tromino when it needs to be distinguished from the straightone, but whieh will be the only kind of tromino diseussed in this paper.Notiee that in defining P, orientation is ignored. For example, the three objeets ereated by rotating the tromino shown in FIGURE1 by 90, 180, and 270 are not eounted as distinet from the original tromino. For results and open questions about the exponentially growing values of P (n), see [4, Appendix D]. Four trorninoescan be fit together to form a tromino=shaped4-reptile,that is, a set in the plane that ean be tiled by four eongruent sealed down eopies of itself. A tromino has order 2, whieh means that the minimum number of trominoes required to form a reetangle is 2, as in FIGURE2. Finding the order of other polyominoes provides ehallenging problems [4, Chapter 8]. The entire plane ean be tiled in a periodie way by any polyomino of finite order by simply repeating copies of the minimal reetangle. Roger Penrose has given a remarkably simple aperiodic tiling of the entire plane using copies of only two unit-edged rhombi, one with acute angle 36 and the other with acute angle 72 [8]. Tilings are often found in Moorish architecture;some tromiures, a great Arabian style palace noes can be seen in a display case in the RecllesAlccla built during various epochs in Seville, Spain. A comprehensive and interesting book clnd concerning tiling is Tilingksz Patterns [5]. There is lots of information about tiling available on the internet;typing "tromino"into a search engine produced 577 hits. We eppstein/junkyard/polyomino. recommend http://www.ics.uci.edu/ nstarr/. html and http://www.amherst.edu/ Elementary results for rectangles A basic tiling result that we will need identifies precisely which rectangles can be tiled. Let's startwith some simple cases. First of all, a 2 x 3 rectangle can be tiled by trominoes.
Figure 2
TiIing R(2, 3)
Denote a reetangle with i rows and j columns by R(i, j). We will indicate deeompositions into nonoverlapping subreetangles by means of an additive notation. For example, a 3i x 2j rectangle ean be deeomposed into ij 3 x 2 subreetangles and we write this faet as R(3i, 2j) = ,=1 EJ=I R(3, 2) = ijR(3, 2). It follows from this 2 and the tiling in FIGURE that any 3i x 2j or 2i x 3j reetangle ean be tiled. (1)
From now on, any reetangle deeomposed into a eombination of 3i x 2j subreetangles, 2i x 3j subreetangles, and trominoes will be eonsidered as sueeessfully tiled by trominoes. Denote the 1 x 1 square lying in row i and eolumn j as (i, j).
48
Now let's look at some rectangles that cannot be tiled. Suppose that a 3 x 3 square Q has been tiled. Some tromino must cover square (3, 1). Here are the three possible ways that it can do that.
Orientation A is immediately ruled out, since square (1, 1) cannot be tiled. But in cases B and C, the tiling must tile the leftmost 3 x 2 subrectangle of Q, so that the original tiling is also a tiling of the third column of Q, which is an R(3, 1). This is impossible. Similarly, suppose that a 3 x 5 rectangle R has been tiled. This argument shows that the tiling must tile the first two columns of R, and hence also the rightmost three columns of R. This is a contradiction since we have just shown 3 x 3 square to be untileable. Iteratingthis procedure shows no R(3, odd) can be tiled. It turns out that there are no other untileable rectangles with area divisible by 3. The integers m and n will always be greaterthan or equal to 2.
CHU JOHNSONBAUGHTHEOREM [2]. Anm x n rectangle can alwaysbe tiledby trominoesif 3 divides its area mn, exceptwhenone dimensionis 3 and the otheris odd.
The proof of this is not hard. We suggest that the reader give it a try. Here are a few hints. Use fact (1) several times. First do the cases R(3k, even); then do the cases R(6k, odd). This leaves only the cases R(9 + 6k, n), where n > 5 is an odd integer. Reduce such a case to R(9, 5). Finally tile R(9, 5) by trial and error.If you have trouble with the last step, leaf ahead to the top left picture in FIGURE9. Dog-ears An m x n dog-earedrectangleis an m x n rectangle with a l x 1 corner square removed. We will denote the dog-eared rectangle by R(m, n)-, so that R(m, n)- = R(m, n) \ {(1, n)}. Note that the area of R(m, n)- is mn-1. If this rectangle is rotated 180, a similar figure with missing lower left-hand corner is created. If it is reflected about a central vertical (resp., horizontal) axis, a similar figure with missing upper left-hand (resp., lower right-hand) corner is created. The problem of tiling the original figure is clearly equivalent to tiling any one of the other three, even though the original figure cannot be rotated into either of the last two figures.
DOG-EARED RECTANGLETHEOREM. An m x n dog-eared rectangle be tiled can withtrominoes and only if 3 dividesits area. if
To understandwhat this theorem means, note that if mn is congruent to 0 or 2 modulo 3, then the area of R(m, n)- is not congruent to 0 modulo 3 and so that dog-eared rectangle cannot be tiled by trominoes, since the area of any region tiled by trominoes must be an integral multiple of 3. So the only m x n dog-eared rectangles that
49
| m - 1
I
. n
Figure A dog-eared 4 rectangle couldpossiblybe tiledby trominoes thosefor whichmn is congruent 1. In other are to words,the only m x n dog-eared rectangles couldpossiblybe tiled by trominoes that are those for whichm _ n _ 1(mod3) or m _ n-2 (mod3), and,indeed,all those dog-eared rectangles be tiled. can We startwith a family of specialcases of the Dog-earedRectangleTheorem,the dog-eareddyadic squares,R(2t, 2t)-. This is a special case of a well-knownand beautifulexampleof mathematical induction[3], [4, page 4], [6, page 45], [9, problem 2.3.38]. If k = 1, note that R(2, 2)- is itself a tromino.If k = 2, see FIGURE5 for a coveringof R(4, 4)-. In FIGURE5, R(4, 4)- was tiled by dividingit into4 quadrants.The upperrightquadrant an R(2, 2)-while the otherthreequadrants was were all congruent R(2, 2). Thenthe black trominocoveringsquares(2, 2), (2, 3), and to (3, 3) was placedat the center. Thisreduced lowerleft quadrant an R(2, 2)- and the to the remaining quadrants rotations R(2, 2)-. In short,the tiling of R(4, 4)two to of was reducedto the tilingo-ffourcopiesof R(2, 2)-. The readerkshould do k = 3, next by dividing R(8, S) into 4 qlladrants, then coveling tle centralsquclres 4), and (4, (4, 5), and (5, 5) with a tromino.This reducesthe tiling of R(8, S)-- to the tiling of fourcopies of R(4, 4)-. The generalinductive proofshouldnow be clear.
Figure 5
TiIing R(4, 4) -
Proof of the Dog-eured Rectungle Theorem: Let m < n. As mentionedabove,the necessaryconditionthat 3 divide the area of R- = R(m, n)- splits into the cases m _ n--1 (mod3) andm-n _ 2 (mod3). It is not hardto see thatthe m x n dogearedrectangleis congruent the n x m one. So, in the formercase, we have either to R(4,3k+4)withk > O, R(7,6k+7)withk > O, R(7,6k+4)withk > 1, or R(3 j + 4, 3k + 4)- with j > 2 and k > 2. Therecorrespond these four decompositions:
50
R(4, 3k + 4)- = R(4, 3k) + R(4, 4)-, k > O R(7,6k+7)= R(7,6k) +R(7,7)-,k >O
1 and
then by
another the
from
Theorem, and
as in FIGURE
R(7, 7)we
Johnsonbaugh
must
R- = R(3j + 2, 3k + 2)-
O <
<
k. If
j 7& we 1,
is a tromino. Let j = 1. Either k is odd, 3k + 2 = 6E+ 5; or else k is even, 3k + 2 = 6E+ 8. Correspondingly, either R- = R(5, 6E+ 5)- = R(5, 6f) + R(5, 5)- where the first term is tiled with the Chu-Johnsonbaugh Theorem and R(5, 5)- is tiled as in FIGURE 6, or else R- = R(5, 6E+ 8)- = R(5, 6f) + R(5, 8)actually where have again the first term is tiled with the Chu-Johnsonbaugh the first two Theorem terms being and tiled we by also the
Chu-Johnsonbaugh
is a tromino.
L
3x 2 2x 3
Figure 6
is
an
application the
of practical
the
Dog-eared as possible
The-
Consider
question
mx n
value Theoof
where m and n both exceed mn modulo 3. If mn-O, tile the entire rem. If mn _ 1, remove a single corner Theorem turns are very out both to that tile if the rest of and remains to prove the if rectangle. a corner can the
depending
use
Rectangle
mn _ 2, we
and tiled. be two This
squares. to
It it
mn-2
what those used
square
a boundary can We
adjacent by its
removed, to
always other
methods proof as
similar
theorems.
an exercise.
Call
a rectangle
with correspond
one
1 x
1 square
missing rectangles,
R(m, n)
mn deficient
51
removingone squarefrom R(m, n); exaetly 4 of these are eongruentto R(m, n)-. The questionof whethera deficientrectanglecan be tiled with trominoesis clearly equivalent the questionof whether full rectangle to the consistingof the disjointunion of the deficientrectangle the 1 x 1 square be tiled by a set of trominoes a and can and single monomino, with the monomino coveringthe missingsquare. Say thata 1 x 1 squareis good if its removalfroma full m x n rectangle produces a defieientrectanglethatcan be tiled. We will now enumerate some m x n deficient reetangles eannotbe tiled, even though3 dividesmn-1. This enumeration that will be verypreeisein the sensethatforeaehm andn the loeationof thebadsquares be will specified. the veryinteresting whenm = n = 5, the followinglemmaproduees In ease 16 badsquares.
DEFICIENT S X S LEMMA. If thesquare(i, j) is removedfrom S the whereeitheri or j is even,thenthe resulting shape is not tileable.
x S
rectangles
Proobf: Forma kindof checkerboard designby marking eachof the nine squares (l, l), (3, l), (5, l), (l, 3), (l, 5), (3, 3), (3, 5), (5, 3), (5 5)
] 0
J
and assumethatone of the 16 unmarked squareshas been removedfrom R(5, 5) to form R-. Then a proposedtiling of R- mustcontainone trominofor each of the 9 marked squares,so thattiling musthaveareaat least9 3 = 27, whichis absurd since the areaof R- is 24. Thusall 16 of the unmarked squaresarebad. n Next we note that bad squarescan also occur when (nl, s1)= (2, 5 + 3k), k = O,l, 2, .... HeresomebadLsquares thoseof the lorm (x, 3 j), j = 1, 2, ... k + l, ale x = 1 or 2. By symmetrywe may assumethatx = l. To show that R(2, 5 + 3k) \ {(1, 3j)) cannotbe tiled, assumethe oppositeand let T be the trominocoveringthe square(2, 3 j). Thento the left of T + {(1, 3j)} lies eitherthe rectangleR(2, 3j-1 ) or the rectangleR(2, 3j-2), neitherof whichhas areadivisibleby 3. Finally,the square(3, 2) is bad in the 5 x (5 + 3k) case, thatis, R(S, n) \ {(3, 2)) cannotbe tiled. For if (3, 2) were good, some trominoT would have to cover the square(3, 1). If T lay above (3, 2) the square( l, l ) could not be reached,otherwise the square(5, l) could not be reached.Symmetrically, 4 + 3k) is also bad in this (3, case.
?
DEFICIENT RECTANGLE THEOREM (COMPARE [2]). An m x n deficientrectangle, 2 < m < n, 3Imn-1, has a tiling, regardlessof the position of the missing square,if and only if (a) neitherside has length 2 unless both of themdo, and (b) m 7& Furthermore, all the exceptionalcases the only bad squaresare those 5. in enumerated theprecedingdiscussion. in
Proof: Forthisproofonly,we changenotation slightlyandlet R(m, n)- denoteany m x n rectangle deficiency1. The "outlier" of R(2, 2)- is tiledwithone tromino. First assumethatm > 4, m 7& and3 t m. The methodof proofis to proceedinductively 5, aftertreating casesm = 4, 7, 8, 10, and 11 individually. m > 13, thenm-6 > 6 the If so thatwe may slice a full rectangleof height6 off of eitherthe top or the bottomof R(m, n)-, thatis, R(m, n)- = R(m-6, n)- + R(6, n). Sinee the last termis tileable by the Chu-Johnsonbaugh Theorem,this firstreducesthe cases m E [13, 17] to the cases m E [7, 11], thenthe cases m E [19, 23] to the cases m E [13, 17], andso on. If m = 4, write R(4, 3k + 1)- = (k-1)R(4, 3) + R(4, 4)-. Apply the ChuJohnsonbaugh Theorem the firstk-1 terms.For the last term,observethatin [3], to Golombshowedthatall 2k x 2kdeficientsquarescan be tiled. (Its proofis an induc-
52
MATHEMATICS MAGAZINE
tion argument almost identical to the one used above to tile the 2k x 2k dog-eared squares.) If m = 7, we may write R(7, n)- = R(3, n) + R(4, n)- and thus reduce the m = 7 case to the m = 4 case when n is even; while if n is odd, 6 divides n-7 and R(7, n)- = ((n-7)/6)R(7, 6) + R(7, 7)- is tiled using the Chu-Johnsonbaugh Theorem and reference [2]. If m = 10; then R(10, n)- = R(7, n)- + R(3, n) so that the Chu-JohnsonbaughTheorem provides a reduction to the m = 7 case if n is even, while R(10, n)- = R(10, n-3)+ R(10, 3) reduces the odd n case to the even n case. If m = 8, R(8, n)- = R(8, 8 + 3k)- = kR(8, 3) + R(8, 8)- is tiled by the ChuJohnsonbaugh Theorem and reference [3]. Finally if m = 11, n must be congruent to either 8 or 11 modulo 6. If n = 8 + 6k,R(l l, 8 + 6k)- = kR(ll, 6) + R(l l, 8)- with the first terms tiled by the Chu-JohnsonbaughTheorem and the last term tiled by the m = 8 case since R(l l, 8)- = R(8, 11)-, while if n = 11 + 6k, R(l l, 11 + 6k)- = kR(ll, 6) + R(ll, 11)-; the first terms are tiled by the Chu-JohnsonbaughTheorem and the tiling of the last term can be found in reference [1]. In view of the treatment of all the bad cases before the statement of this theorem, it remains only to analyze the exceptional good cases. Since 3 divides mn-1, if m = 2, we must have n = 2 + 3k, k = O, 1, . . ., while if m = 5, we must have n = 5 + 3k, k = O, 1, 2, .... Also notice that the (m,n) = (2, 2) case is notan exception. We'll start with the 5 x 5 good cases. The tiling in FIGURE6 above shows that (1, 5) is good, while these two tilings show (3, 5) and (3, 3) to be good. Symmetry considerations show that the remaining six marked tiles are also good. Thus all nine markedtiles are good.
2x3 3x2
k
2x3 3x2
Azti,:, MYS,,>
2x3
3x2
2x3
...
....
>
><.X,i.,.&.>*S v:*:.-b
53
cases (i, j) = (3, 1), (S, 3), (4, 3), (3, 3), and (3, 4). These are done in ad hoc fashion in FIGURE 8.
3x2
2x3
2x3
X
.:, O,..:.:X,
3x2
3x2
(S,I)
(5,3) Rotate the outlined square 90 v* clockwise to tile the (4,3) ca<3e.> l
1 3x2 1 x2 nx2 1
2x3
2x3
2x
2xx (3,3 )
nx2
x2
(3 4)
Figure8
Deficient 5 x 9 tilings
The cases R(S, 11). By symmetry we may assume i > 3 and j < 6. Since (3, 2) is bad, we have :17cases to show good. If i and j are both odd, (i, j) is a good square of R(S, S), so the decomposition of R(5, 11)- into a left R(S, S)- and a full right R(S, 6) works for these 6 cases. Five of the remaining 11 cases are done in ad hoc fashion in FIGURE A dark outlined 2 x 2 square appears in the tiling for the (4, 1) case that is 9. shown as the top left picture of FIGURE Rotate that square90 clockwise to produce 9. a tiling for the (4, 2) case; then rotate it another 90 to produce a tiling for the (S, 2)
X
(4,1)
e=
(4,3)
X
(4,5)
1X1
(3,4) (3,6)
Figure9
Deficient 5 x 11 tilings
54
producetilings of the (4, 4) and (5, 4) cases fromthe case. Similarpairsof rotations displayedtilingof the (4, 3) case, as well as tilingsof the (4, 6) and (5, 6) cases from X the displayedtilingof the (4, 5) case. Results and questions about 2-deficiency If two squaresareremovedfroma rectdisrectangle.The followingproposition shapea 2-deficient angle, call the resultant definition deficiencyof orderk, k > 2 of allows the possibilityof makingthe natural for andthenfindinga directextensionof the DeficientRectangleTheorem higherdenclencles.
,^ . .
PROPOSITION. No rectanglehas the propertythat no matterwhich two 1 x 1 the shapeof area mn-2 can be tiled. squaresare removed, remaining
For if the squares(1, 2) and (2, 1) are removed,then the square(1, 1) cannotbe coveredby a tromino.(This also shows the propositionstill holds even if "tiling" is extendedto mean "tiling by any collection of polyominoeswhich containsno monomino.") Theorem, of Eventhoughtherewill notbe a directanalogue theDeficientRectangle for workto be done here.Hereis a program what thereis roomfor some interesting A We to do about2-deficiency. extendthe definitionof good to 2-deficiency. pairof squaresis good if theirremovalfrom a m x n rectangleleaves a figurethatcan be tiled. for find all bad pairs of squares PROBLEM. For the general case of 2-deficiency, all m x n rectangleswheremn-2 (mod3). Slightlyless generally,exactlywhencan sucha rectanglebe coveredby one dominoand (mn-2)/3 trominoes? the On the negativeside, as we pointedout in the proof of the Proposition, pair {(2, l), (1, 2)} is bad, that is, if square(2, 1) and square(1, 2) are removed,then no trominocan cover square(1,1). On the positive side, recall thatin the application given afterthe proof of the Dog-earedRectangleTheoremwe pointedout that a tiling is always possible if the two removedsquaresare adjacentand in a corner of the rectangle.In other words, if mn _ 2(mod3), then the pair {(1, n), (2, n)} is good. Now considerthe 5 x 7 case. As in the analysis of the 5 x 5 case for defiby eachof the pattern marking cient rectangles done above,forma checkerboard-like odd 12 squaresthathave both coordinates and assumethattwo of the 23 unmarked squareshave been removedfrom R(5, 7) to form R=. Then a proposedtiling of R= squares,so thattiling musthave mustcontainone trominofor each of the 12 marked areaat least 12 3 = 36, which is absurdsince the areaof R= is 33. This reasoning of = disqualifies(223) 253 pairs.Similarreasoningidentifiesa largenumber badpairs forR(5, 13),...,R(5,7+6k),....
Acknowledgments. MarshallAsh had profitablediscussions with Ayse Sahin and Roger Jones. MartinGardner providedencouragementand some references.ElizabethBreuermotivatedthe applicationgiven after the proof of the Dog-eared Rectangle Theorem.JerroldGriggs and a referee providedhelpful historicalinformation.We thank the referees for style-improvingsuggestions. This researchwas partially supportedby NSF grantDMS 9707011 and a grantfrom the Faculty and DevelopmentProgramof the College of LiberalArts and Sciences, DePaul University.
55
59 Tiling deficientboardswith trominoes,this MAGAZINE (1986), 3440, MR 1. I-P.Chu and R. Johnsonbaugh, 87c:05044. 2. , Tiling boardswith trominoes,J. Rec. Math. 18 (1985-86), 188-193. Math.Monthly,61 (1954), 675-682, MR 16,644a. 3. S. W. Golomb, CheckerBoardsand Polyominoes,Ame75 4. , Polyominoes.Puzzles,patterns,problems,and packings, 2nd ed., PrincetonUniversityPress, Princeton, 1994, MR 95k:00006. 5. B. Grunbaum G. C. Shephard,Tilingsand Patterns,W. H. Freeman,New York, 1987, MR 88k:52018. and 6. R. Johnsonbaugh, Discrete Mathematics,5th ed., PrenticeHall, Upper Saddle River,NJ, 2001. 7. Amer Math.MonthlyProblem 10641, Proposedby J. R. Griggs, 105 (1998), 175; solutionby Michael Woltermann, 107 (2000), 179. 8. R. Penrose, TheEmperor'sNew Mind, OxfordUniversityPress, New York, 1989, MR 91c:00010. 9. P. Zeitz, TheArtand Craftof ProblemSolving, JohnWiley, New York, l999, MR 2000f:00001.
ProofWithoutWords: x z Is a CountableSet