Tiling Proofs of Recent Sum Identities Involving Pell Numbers

Arthur T. Benjamin
Department of Mathematics, Harvey Mudd College, Claremont, CA 91711 E-mail: [email protected]

Sean S. Plott
Department of Mathematics, Harvey Mudd College, Claremont, CA 91711 E-mail: [email protected]

and James A. Sellers

July 15, 2006

Abstract

In a recent note, Santana and DiazBarrero proved a number of sum identities involving the wellknown Pell numbers. Their proofs relied heavily on the Binet formula for the Pell numbers. Our goal in this note is to reconsider these identities from a purely combinatorial viewpoint. We provide bijective proofs for each of the results by interpreting the Pell numbers as enumerators of certain types of tilings. In turn, our proofs provide helpful insight for straightforward generalizations of a number of the identities.

Keywords: Pell numbers, combinatorial identities, tilings, NSW numbers

Motivation

In a recent article, Santana and DiazBarrero [3] proved a number of sum identities involving the Pell numbers Pn dened by P0 = 0, P1 = 1, and Pn = 2Pn1 + Pn2 for n 2. (See Sloanes Online Encyclopedia of Integer Sequences [4, A000129] for more details about the Pell numbers.) For example, Santana and DiazBarrero proved the following for all n 0 :
4n+1

Theorem 1.
k=1

Pk is a perfect square.
n

Theorem 2.

r=0

2n + 1 r 2 = P2n + P2n+1 . 2r

Theorem 3.
n

r=0

2n 2n r 2n2r 2 = P2n1 + P2n+1 2n r r

and

r=0

2n + 1 r 2n+12r 2n + 1 2 = P2n + P2n+2 . 2n + 1 r r

Theorem 4. P2n+1 divides

P2k+1
k=0 2n

and P2n divides

P2k1 .
k=1

The proofs of Theorems 14 provided by Santana and DiazBarrero, along with the proofs of three lemmas which also appear in [3], rely heavily on the Binet formula for Pn along with various binomial coecient identities. Our primary goal in this work is to view the above identities combinatorially, providing bijective arguments from the context of tilings as discussed in Benjamin and Quinns recent book Proofs That Really Count [1]. In this way, we 3

provide elegant proofs of the identities of Santana and DiazBarrero, proofs that can clearly be generalized to yield a variety of related sum identities, which are highlighted at the conclusion of this paper. In order to provide combinatorial arguments for the identities above, we will consider the sequence of numbers denoted pn and dened as pn = Pn+1 for n 1. The motivation for considering pn is simple: pn counts the number of tilings of a board of length n using white squares, black squares, and gray dominoes. Thus, for example, p0 = 1 counts the empty tiling, p1 = 2 because a board of length 1 can be covered (exactly) by either one white square or one black square. Similarly, p2 = 5 since a board of length 2, or a 2board, can be covered by two white squares or two black squares or one white square and one black square or one gray domino. See Figure 1.

Figure 1: The tilings of a 2board It is within the context of such tilings that we restate and combinatorially prove Theorems 14.

Tiling Proofs

We will consider Theorems 14 in reverse order, since our proofs of these results will consequently grow in complexity.

2.1

Theorem 4

First, the two divisibility properties stated in this theorem can be combined into one statement. Moreover, this one divisibility property, when couched in the terms of pn rather than Pn , reads as follows: for all n 0,
n

(1)

pn divides
k=0

p2k

We begin our combinatorial proof of (1) by proving the following lemma: Lemma 5. For all n 0,
n

p2n+1 = 2
k=0

p2k .

Proof. We begin by remembering that p2n+1 counts all tilings of a (2n + 1) board (i.e., a board of length 2n + 1 with cells numbered from 1 to 2n + 1). Note that each such tiling must contain at least one square since 2n + 1 is odd. By considering the location of the last square, which must occur in an oddnumbered cell, we see that there are 2p2k tilings whose last square occurs on cell 2k + 1 (0 k n), where the factor of 2 indicates the color choice of this last square. Hence, the righthand side of the equation also counts all tilings of a (2n + 1)board. With Lemma 5 in hand, we can now prove (1), as follows:
n

Theorem 6. For all n 0, pn divides p2n+1 . Moreover, pn divides

p2k .

Proof. We consider whether or not a tiling of length 2n + 1 is breakable at cell n, i.e., if the (2n + 1)-tiling can be split into a tiling of length n followed by a tiling of length n + 1. There are pn pn+1 breakable tilings; the number of unbreakable tilings (with a domino covering cells n and n + 1) is pn1 pn . Thus, p2n+1 = pn pn+1 + pn1 pn = pn (pn+1 + pn1 ) is clearly a multiple of pn . Moreover, since pn+1 = 2pn + pn1 , it follows that pn+1 and pn1 have the same parity, and therefore their sum is an even 5

integer. Consequently, by Lemma 5, it is even true that pn divides 1 p2n+1 = 2 n p2k . k=0

2.2

Theorem 3

Before proving Theorem 3 combinatorially, we rst consider a lemma from [3]. Although this lemma is not necessary for our proof of Theorem 3, we include it here for completeness sake. We will also provide a generalization of it in the nal section of this paper. When rephrased in terms of pn , it reads as follows: Lemma 7. For all n 0, pn =
r0

n r n2r 2 . r

Proof. The lefthand side of this equality counts all tilings of length n. We now consider how many of these tilings contain exactly r dominoes. Such a tiling contains n 2r squares and, therefore, n r tiles altogether. There are nr ways to decide which of these n r tiles are dominoes and then 2n2r r ways to color the squares. Hence there are nr 2n2r tilings with exactly r r dominoes. (Note that this summand is zero whenever r exceeds n .) Summing 2 over all values of r yields the desired result. We now proceed to Theorem 3, which we combine into a single identity: Theorem 8. For all n 2, pn2 + pn =
r0

n n r n2r 2 . nr r

Proof. In order to prove this theorem, we consider special modied tilings we call bracelets. A bracelet of length n is a tiling of an nboard with an added wraparound feature whereby the board is bent into a circle so that cells n and 1, the last and the rst cells on the board, are joined. (See [1, Chapter 2] for additional discussion on bracelets.) For bracelets, a domino 6

is allowed to simultaneously cover cells n and 1 (which we call covering the clasp). When a domino covers the clasp of a bracelet, we say the bracelet is out of phase; otherwise, the bracelet is said to be in phase. We now proceed by showing that both sides of the equality in this theorem count all bracelets of length n that can be created with white squares, black squares, and gray dominoes. First, with an out of phase bracelet, we can remove the domino covering the clasp to produce a (straight) tiling of length n2. Similarly, an in phase bracelet can be undone at the clasp to produce a tiling of length n. Altogether, there are pn2 + pn such tilings. Next, we ask how many such tilings exist with exactly r dominoes. Arguing as in the proof of Lemma 7 above, there are nr 2n2r in phase bracelets r nr1 n2r with r dominoes, and r1 2 out of phase bracelets with r dominoes. nr nr1 n r Since r1 = nr r , we know there are nr nr 2n2r bracelets with r exactly r dominoes. Summing over r counts all bracelets of length n.

2.3

Theorem 2

We now wish to prove our reformulation of Theorem 2. Before proving this theorem, it should be noted that the values of pn1 + pn are wellknown [4, A001333]. They are the numerators of the continued fraction convergents of 2. When n is even, these are known as the NSW numbers studied by Newman, Shanks, and Williams [2] approximately twentyve years ago in connection with the order of certain simple groups. The interested reader should consult Sloane [4, A002315] for more information on these numbers. Theorem 9. For all n 0, pn1 + pn =
r0

n+1 r 2. 2r

Proof. We prove this result by counting (n + 1)-boards with white squares, black squares, and gray dominoes, with the restriction that the tilings do not end in a black square. First, by considering the last tile used in such a tiling (either a domino or a white square), the answer to the question is pn1 + pn . 7

Next, we note that such a tiling can be constructed by rst choosing X, a subset of {1, . . . , n + 1} of even cardinality, say X = {x1 , . . . , x2r } with x1 < x2 < < x2r . From X we create r intervals [x2j1 , x2j ] and can then create 2r dierent tilings as follows. Any cell on the (n + 1)board that does not belong to an interval is covered by a white square. An interval of k 2 cells can be tiled in one of two ways, either as k 1 black squares followed by a white square, or as k 2 black squares followed by a domino. (See Figure 2 for one of the four possible colored 11-tilings generated by the intervals [2,5] and [8,10])

Figure 2: An example of a tiling via intervals Notice that no interval will end with a black square, and thus the tiling of the (n + 1)board, created as the concatenation of all of these smaller tilings, will not end in a black square. Conversely, one can easily show by induction on n that every tiling of a (n + 1)board that does not end with a black square has a unique construction in this manner. Summing over r yields the desired result.

2.4

Theorem 1

We close this section by considering the following reformulation of Theorem 1. Note that the righthand side of the equality is the square of an NSW number. 8

Theorem 10. For all n 0,

pk = (p2n1 + p2n )2 .
k=0

Proof. The lefthand side counts all tilings of boards of length at most 4n using white squares, black squares, and gray dominoes. Thus we simply need to show that the righthand side also counts the same tilings. Note rst that the righthand side clearly counts tiling pairs (X, Y ) where X and Y are tilings of (2n+1)boards that end in a domino or a white square (but not a black square). We now create a bijection from this set to the set of all tilings of length at most 4n by considering the last tile of X and Y. Four cases arise: Case 1: X and Y both end in a white square. That is, X = Aw and Y = Bw where A and B are tilings of length 2n. Here, (X, Y ) is mapped to the tiling T = AB which has length 4n and is breakable at its midpoint. Case 2: X ends in a domino, Y ends in a white square. That is, X = Ad and Y = Bw where A has length 2n 1 and B has length 2n. Here (X, Y ) is mapped to the tiling AB, which has length 4n 1, and is breakable at its midpoint. (Note that we dene the midpoint here to be 4n1 or 2n 1.) 2 Case 3: X ends in a white square, Y ends in a domino. That is X = Aw and Y = Bd where A has length 2n and B has length 2n 1. (To avoid confusion, we remove the white square at the end of X and the domino at the end of Y. Thus A covers cells 1 through 2n and B covers cells 1 through 2n 1.) From a visual perspective, we set A above B and look for the last fault line that passes through A and B. (A fault line occurs at cell i if both A and B are breakable at cell i.)

Figure 3: The last fault line occurs at cell k. This fault line occurs at some cell k where 0 k 2n 1. (When k = 0, the only fault line is at cell 0, to the left of cell 1.) To the right of this last fault line, A and B consist entirely of dominoes in staggered formation, except for a single square s on cell k + 1 in A or B (but not both). In this case, (X, Y ) will be mapped to a tiling of even length as follows: Let C denote the subtiling of A covering cells 1 through k and let D denote the subtiling of B covering cells 1 through k. If the square s (sitting on cell k + 1) is white, then (X, Y ) is mapped to the tiling CD, a tiling of length 2k that is breakable at its midpoint. If the square s is black, then (X, Y ) is mapped to the tiling CdD, a tiling of length 2k + 2, that is not breakable at its midpoint since a domino starts on cell k + 1. Notice that Case 3 covers all of the even length tilings, except for the tilings of length 4n that were covered in Case 1. It remains to map onto the tilings of odd length at most 4n 1, excluding the tilings generated by Case 2. Case 4: X and Y both end in a domino. That is X = Ad and Y = Bd where A and B have length 2n 1. First we oset the tilings by shifting X to the right by one cell. (See Figure 4.) X still has length 2n + 1 but it covers cells 2 through 2n + 2. Again, 10

Figure 4: A Case 4 Example we nd the last fault line, which will occur at some cell k where 1 k 2n 1. (Note that a fault must exist since X and Y have odd length and must therefore contain at least one square. Also note that k cannot equal 2n or 2n + 1 since X and Y both end in dominoes.) As before, to the right of the fault line at cell k, we have dominoes in staggered formation, and just a single square s on cell k + 1 in A or B (but not both). Here (X, Y ) will be mapped to a tiling of odd length as follows: Let C denote the subtiling of X covering cells 2 through k, and let D denote the subtiling of Y covering cells 1 through k. If the square s (sitting on cell k + 1) is white, then (X, Y ) is mapped to the tiling CD, a tiling of length 2k 1, that is breakable at its midpoint. If the square s is black, then (X, Y ) is mapped to the tiling CdD, a tiling of length 2k + 1, that is not breakable at its midpoint, since a domino starts on cell k. The mapping is easily reversed: Given any tiling of length j (0 j 4n) we can nd its preimage by noting the length and parity of j and whether j the tiling is breakable at 2 .

11

Generalizations

We close by highlighting a number of generalizations of the results above which are obtained almost eortlessly thanks to these tiling proofs. We do not attempt to mention all possible generalizations herein but highlight some of the more straightforward ones that are readily available. First, we consider a oneparameter generalization of Theorem 9. Let pb n b b b b b be dened by p0 = 1, p1 = 2, and pn = 2pn1 + bpn2 for n 2 where b is a natural number. By denition, p1 = pn and it is easy to see that pb counts n n the number of tilings of an nboard with white squares, black squares, and b dierent colors of dominoes. Then Theorem 9 can be generalized as follows: Theorem 11. For all n 0, bpb + pb = n1 n
r0

n+1 (b + 1)r . 2r

The proof of this generalized theorem closely follows the proof of Theorem 9 given earlier. The only dierence is that b choices of colors for the dominoes are available throughout. To close this section, we consider a twoparameter generalization of pn , denoted pa,b and dened by pa,b = 1, pa,b = a, and pa,b = apa,b + bpa,b for 0 1 n1 n2 n n 2,1 n 2 where a and b are natural numbers. Thus pn = pn and pa,b counts n the number of tilings of an nboard where a dierent colors of squares and b dierent colors of dominoes are available. We can then extend our proofs above in the context of these values pa,b to prove the following generalizations n of Lemma 5, the rst part of Theorem 6, Lemma 7, and Theorem 8 above as follows: Lemma 12. For all n 0,
n

pa,b 2n+1

=a
k=0

bnk pa,b . 2k

Theorem 13. For all n 0, pa,b divides pa,b . 2n+1 n 12

Lemma 14. For all n 0, pa,b = n

n r n2r r a b. r

Theorem 15. For all n 0, pa,b + bpa,b = n2 n

n n r n2r r a b. nr r

References
[1] A. T. Benjamin and J. J. Quinn, Proofs That Really Count: The Art of Combinatorial Proof, The Dolciani Mathematical Expositions, 27, Mathematical Association of America, Washington, DC, 2003 [2] M. Newman, D. Shanks, and H. C. Williams, Simple groups of square order and an interesting sequence of primes, Acta Arithmetica 38, no. 2 (1980), 129140 [3] S. F. Santana and J. L. DiazBarrero, Some properties of sums involving Pell numbers, Missouri Journal of Mathematical Sciences 18, no. 1 (2006), 3340 [4] N. J. A. Sloane, The On-Line Encyclopedia of Integer Sequences, published electronically at http://www.research.att.com/njas/ sequences/, 2006

13