Claremont Colleges

Scholarship @ Claremont
All HMC Faculty Publications and Research HMC Faculty Scholarship

8-1-2010

Combinatorial Trigonometry with Chebyshev

Polynomials
Arthur T. Benjamin
Harvey Mudd College

Larry Ericksen

Pallavi Jayawant
Bates College

Mark Shattuck
University of Tennessee - Knoxville

Recommended Citation
Benjamin, A.T., Ericksen, L., Jayawant, P., & Shattuck, M. (2010). Combinatorial trigonometry with Chebyshev polynomials. Journal
of Statistical Planning and Inference, 140(8): 2157-2160. DOI: 10.1016/j.jspi.2010.01.011.

This Article is brought to you for free and open access by the HMC Faculty Scholarship at Scholarship @ Claremont. It has been accepted for inclusion
in All HMC Faculty Publications and Research by an authorized administrator of Scholarship @ Claremont. For more information, please contact
[email protected].
This article appeared in a journal published by Elsevier. The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/copyright
 Author's personal copy
ARTICLE IN PRESS
Journal of Statistical Planning and Inference 140 (2010) 2157–2160

Contents lists available at ScienceDirect

Journal of Statistical Planning and Inference

journal homepage: www.elsevier.com/locate/jspi

Combinatorial trigonometry with Chebyshev polynomials

Arthur T. Benjamin a,, Larry Ericksen b, Pallavi Jayawant c, Mark Shattuck d
a
Department of Mathematics, Harvey Mudd College, Claremont, CA 91711, USA
b
P.O. Box 172, Millville, NJ 08332, CA, USA
c
Department of Mathematics, Bates College, Lewiston, ME 04240-6028, USA
d
Department of Mathematics, University of Tennessee, Knoxville, TN 37006-1300, USA

a r t i c l e in fo abstract

Available online 20 January 2010 We provide a combinatorial proof of the trigonometric identity cosðnyÞ ¼ Tn ðcos yÞ,
Keywords: where Tn is the Chebyshev polynomial of the first kind. We also provide combinatorial
Chebyshev polynomials proofs of other trigonometric identities, including those involving Chebyshev poly-
Combinatorial proof nomials of the second kind.
Tiling & 2010 Elsevier B.V. All rights reserved.

1. Introduction

The Chebyshev polynomials have many beautiful properties and countless applications, arising in a variety of
continuous settings. They are a sequence of orthogonal polynomials appearing in approximation theory, numerical
integration, and differential equations. In this paper we approach them instead as discrete objects, counting the sum of
weighted tilings. Using this combinatorial approach, one can prove numerous identities, as is done in Benjamin and Walton
(2009), Benjamin and Walton (2010), Shapiro (1981), and Walton (2007). In this note we provide a combinatorial proof of
perhaps the most fundamental of Chebyshev properties, namely the trigonometric identity

cosðnyÞ ¼ Tn ðcos yÞ;

where Tn is the Chebyshev polynomial of the first kind. We also provide combinatorial proofs of other trigonometric
identities, including those involving Chebyshev polynomials of the second kind.
The Chebyshev polynomials of the first kind are defined by T0 ðxÞ ¼ 1, T1 ðxÞ ¼ x, and for n Z 2,

Tn ðxÞ ¼ 2xT n1 ðxÞTn2 ðxÞ:

The next few polynomials are T2 ðxÞ ¼ 2x2 1, T3 ðxÞ ¼ 4x3 3x, T4 ðxÞ ¼ 8x4 8x2 þ 1, T5 ðxÞ ¼ 16x5 20x3 þ 5x.
The Chebyshev polynomials of the second kind differ only in the initial conditions. They are defined by U0 ðxÞ ¼ 1,
U1 ðxÞ ¼ 2x, and for n Z2,

Un ðxÞ ¼ 2xU n1 ðxÞUn2 ðxÞ:

The next few polynomials are U2 ðxÞ ¼ 4x2 1, U3 ðxÞ ¼ 8x3 4x, U4 ðxÞ ¼ 16x4 12x2 þ 1, U5 ðxÞ ¼ 32x5 32x3 þ6x.

 Corresponding author.
E-mail addresses: [email protected] (A.T. Benjamin), [email protected] (L. Ericksen).

0378-3758/$ - see front matter & 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.jspi.2010.01.011
 Author's personal copy
ARTICLE IN PRESS
2158 A.T. Benjamin et al. / Journal of Statistical Planning and Inference 140 (2010) 2157–2160

2. Chebyshev polynomials of the first kind

Chebyshev polynomials have a simple combinatorial interpretation, which were first explored by Shapiro (1981). We
define an n-tiling to be a sequence of squares (of length one) and dominoes (of length two) with a total length of n. For
example, there are exactly five 4-tilings, namely ssss, ssD, sDs, Dss, and DD, where s denotes a square and D denotes a
domino. For Chebyshev polynomials of the first kind, we assign each of our square tiles, with one possible exception, a
weight of 2x, and each domino is assigned a weight of 1. The exception occurs if a square occurs as the first tile in the
tiling, and here we assign that initial square a weight of x, but all subsequent squares still get a weight of 2x. We define the
weight of the tiling to be the product of the weights of its tiles. For example, the aforementioned 4-tilings, ssss, ssD, sDs, Dss,
and DD, have respective weights 8x4 , 2x2 , 2x2 , 4x2 , and 1. Note that the sum of the weights of all 4-tilings is
8x4 8x2 þ 1, which is equal to T4 ðxÞ.
In general, a simple induction argument on n reveals (or see Benjamin and Quinn, 2003, Chapter 3) the following
theorem.

Theorem 1. For n Z 0, Tn ðxÞ is the sum of the weights of n-tilings where dominoes have weight 1 and squares have weight 2x,
except for a square at the beginning of the tiling, which has weight x.

Our first trigonometric identity concerns the case where x ¼ cos y. Since 2 cos y ¼ eiy þ eiy , we can refine the previous
theorem. Here, we allow squares to come in two colors, white and black, where, except for an initial square, white squares
have weight eiy , and black squares have weight eiy . An initial square has half the weight given above, namely 12 eiy or 12 eiy ,
and all dominoes still have weight 1. For example, if we let D, w, and b, respectively, denote a domino or white square or
black square, then the 13-tiling bbbDwbwwDD would have weight
1
2 eiy eiy eiy ð1Þeiy eiy eiy eiy ð1Þð1Þ ¼ 12eiy :
In general, we have:

Corollary 2. For n Z0, Tn ðcos yÞ is the sum of the weights of n-tilings with dominoes, white squares, and black squares with
weights described above.

We are now ready to establish our first theorem of combinatorial trigonometry.

Identity 1. For n Z 0, Tn ðcos yÞ ¼ cosðnyÞ.

Proof. Observe that the n-tiling consisting of all white squares has weight 12 einy and that the n-tiling with all black squares
has weight 12 einy , and so the sum of their weights is ðeiny þ einy Þ=2 ¼ cosðnyÞ. The identity is proved by showing that the
sum of all the other n-tilings is zero.
Let X be an arbitrary n-tiling, and let j be the first cell of the tiling such that cells j and j þ1 are occupied by a domino or
cells j and j þ 1 are occupied by squares of opposite colors. We consider two cases.
Case I: j 4 1. Suppose that X has squares of opposite color on cells j and j þ 1. Then let X 0 be the tiling obtained by
replacing these two squares with a domino (and leaving the rest of the tiling unchanged). Alternatively, if X has a domino
on cells j and j þ1, then we obtain X 0 by replacing that domino with two squares of opposite color, where the square on cell
j has the same color as the square on cell j1. Note that this rule is reversible: ðX 0 Þ0 ¼ X. Since two squares of opposite color
multiply to a weight of eiy eiy ¼ 1 and dominoes have weight 1, then X and X 0 have weights that are opposite in sign and
so the weights sum to zero.
Case II: j ¼ 1. Here X must be of the form wbY or bwY or DY, where Y is an ðn2Þ- tiling. But since an initial wb contributes
a weight of 12 and an initial bw contributes a weight of 12 and an initial domino contributes a weight of 1, then it is clear
that the weights of wbY, bwY, and DY must sum to zero. &

We note that using the same logic, and replacing 2 cos y with z þ 1=z (as in Ericksen, 2008, 2010), gives us

Identity 2. For n Z 0, and any real or complex number z,

0 1
1 1
zþ zn þ n
B z C z
Tn @ A¼ :
2 2

It should be noted that this identity can be proved directly by a simple induction argument.
Another immediate corollary to Identity 1, is the well-known composition theorem for Chebyshev polynomials.

Identity 3. For m; n Z 0, Tm ðTn ðxÞÞ ¼ Tmn ðxÞ.

Proof. When x ¼ cos y,

Tm ðTn ðcos yÞÞ ¼ Tm ðcos nyÞ ¼ cosðmnyÞ ¼ Tmn ðcos yÞ:
 Author's personal copy
ARTICLE IN PRESS
A.T. Benjamin et al. / Journal of Statistical Planning and Inference 140 (2010) 2157–2160 2159

Since Tm ðTn ðxÞÞ and Tmn ðxÞ are polynomials that agree for an infinite number of values of x, then they must be equal
for all x. &

The form of this theorem suggests that there may also be a bijective proof between mn-tilings and something like
m-tilings of n-tilings. Such a bijection is presented in Benjamin and Walton (2009) and Walton (2007).

3. Chebyshev polynomials of the second kind

Due to its ideal initial conditions U0 ðxÞ ¼ 1 and U1 ðxÞ ¼ 2x, Chebyshev polynomials of the second kind have a simpler
combinatorial interpretation (Benjamin and Quinn, 2003) where all squares are given the same weight, including a square
that occurs as the initial tile.
For example, the five 4-tilings ssss, ssD, sDs, Dss, and DD have respective weights 16x4 , 4x2 , 4x2 , 4x2 , and 1, with a
total weight of 16x4 12x2 þ1 ¼ U4 ðxÞ. And in general,

Theorem 3. For n Z 0, Un ðxÞ is the sum of the weights of n-tilings where dominoes have weight 1 and squares have weight 2x.

Likewise, by giving all white squares a weight of eiy , all black squares a weight of eiy , and all dominoes a weight of 1,
we have

Corollary 4. For n Z 0, Un ðcos yÞ is the sum of the weights of n-tilings with dominoes, white squares, and black squares with
weights described above.

Using ideas from the proof of Identity 1, it is easy to show:

Identity 4. For n Z 0,
sinððn þ 1ÞyÞ
Un ðcos yÞ ¼ :
sin y

Proof. Equivalently, we prove ðsin yÞUn ðcos yÞ ¼ sinððn þ 1ÞyÞ. Here we exploit the fact that sin y ¼ ðeiy eiy Þ=2i. Hence the
left side of our identity is the sum of the weights of all ðn þ1Þ-tilings, where the first tile is either a white square with
weight eiy =2i or a black square with weight eiy =2i, and is followed by an n-tiling where all white squares, black squares,
and dominoes, have respective weights eiy , eiy , and 1.
As in the proof of Identity 1, let X be an ðn þ 1Þ-tiling that does not consist of all white squares or all black squares. We let
j denote the first cell of X such that cells j and j þ 1 contain either two squares of opposite color or a domino. When j 41, we
pair up X with X 0 as before. If j ¼ 1, then X must be of the form X ¼ wbY or X ¼ bwY, where Y is an ðn1Þ-tiling. But since an
initial wb contributes a weight of ðeiy =2iÞeiy ¼ 1=2i and an initial bw contributes a weight of ðeiy =2iÞeiy ¼ 1=2i then the
weights of wbY and bwY must sum to zero.
Hence the only ðn þ1Þ-tilings that are not canceled out are wn þ 1 with weight ðeiy =2iÞeiny ¼ eiðn þ 1Þy =2i and bn þ 1 with
weight ðeiy =2iÞeiny ¼ eiðn þ 1Þy =2i. The combined weight of these two tilings is ðeiðn þ 1Þy eiðn þ 1Þy Þ=2i ¼ sinððn þ 1ÞyÞ, as
desired. &

More generally, as in Ericksen (2008, 2010), the same logic gives us:

Identity 5. For n Z 0, and any real or complex number z,

0 1
  1
1 z þ 1
B zC
z Un @ A ¼ zn þ 1  n þ 1 :
z 2 z

Identity 4 provides us with a composition theorem for Chebyshev polynomials of the second kind.

Identity 6. For m; n Z 0,
Um1 ðTn ðxÞÞUn1 ðxÞ ¼ Umn1 ðxÞ:

Proof. It suffices to prove this when x ¼ cos y. Here we have

sinðnyÞ sinðmnyÞ sinðnyÞ
Um1 ðTn ðcos yÞÞUn1 ðcos yÞ ¼ Um1 ðcosðnyÞÞ ¼ ¼ Umn1 ðcos yÞ: &
sin y sinðnyÞ sin y

4. More trigonometric identities

We conclude with two alternative expressions for cosðnyÞ and sinððn þ1ÞyÞ based on combinatorial arguments. First we
prove a more general identity.
 Author's personal copy
ARTICLE IN PRESS
2160 A.T. Benjamin et al. / Journal of Statistical Planning and Inference 140 (2010) 2157–2160

Identity 7. For n Z 0,
X n 
Tn ðxÞ ¼ ðx2 1Þk xn2k :
kZ0
2k

Proof. For this proof, we use a slightly different combinatorial interpretation of Tn ðxÞ, as given in Benjamin and Walton
(2009), Benjamin and Walton (2010), and Walton (2007). Here we consider n-tilings with white squares, black squares, and
dominoes with respective weights x, x, and 1. But in order to accommodate the initial condition T1 ðxÞ ¼ x, the first cell
must either be a domino or a white square. As usual, the weight of the tiling is the product of the weights of its tiles and
Tn ðxÞ is the sum of the weights of all n-tilings.
Next observe that any such n-tiling can be decomposed in a unique way into ‘‘intervals of darkness’’ and ‘‘intervals of
k
lightness,’’ where an interval of darkness is either of the form Db (a domino followed by k black squares) where k Z 0, or of
k
the form wb (a white square followed by k black squares) where kZ 1. Note that an interval of darkness must have length
at least two. Everything outside an interval of darkness is necessarily a white square, and these comprise the intervals of
lightness. For example, the tiling
wwwwDDDDbbwbwwwbbbDwDbDw
would decompose as
wwww D D D Dbb wb ww wbbb D w Db D w
where the intervals containing only w are intervals of lightness, and the rest are intervals of darkness.
We claim that the summand gives the total weight of all n-tilings with exactly k intervals of darkness. To see this, note
n
that once you choose the endpoints for the k intervals (which can be done in ð2k Þ ways), we can give the first two cells of the
interval a weight of x2 (for a white square followed by a black square) or a weight of 1 (for a domino). There is exactly one
way to assign colors to the rest of the tiling, and those squares will contribute a weight of xn2k . &

Substituting x ¼ cos y then gives the following identity (which has a simpler non-combinatorial proof, given in Chentzov
et al., 1993).

Identity 8. For n Z 0,
X n 
cosðnyÞ ¼ ð1Þk ðsin yÞ2k ðcos yÞn2k :
kZ0
2k

In a similar way, we can prove

Identity 9. For n Z 0,
!
X n þ1
Un ðxÞ ¼ ðx2 1Þk xn2k :
kZ0
2kþ 1

Proof. Here Un ðxÞ has the same interpretation as in the preceding proof, but we no longer have a restriction on the initial
tile. This time, we choose 2k þ1 points from the set f0; 1; . . . ; ng. We begin the tiling with j black squares where j is the
smallest chosen point ðj Z 0Þ. Then we proceed as in the Tn identity with the remaining 2k endpoints. &

This time when we let x ¼ cos y we get another identity from Chentzov et al. (1993).

Identity 10. For n Z 0,

!
X nþ1
sinððn þ 1ÞyÞ ¼ ð1Þk ðsinyÞ2k þ 1 ðcosyÞn2k :
kZ0
2k þ1

References

Benjamin, A.T., Quinn, J.J., 2003. Proofs That Really Count: The Art of Combinatorial Proof. Mathematical Association of America, Washington, DC.
Benjamin, A.T., Walton, D., 2009. Counting on Chebyshev polynomials. Mathematics Magazine 82, 117–126.
Benjamin, A.T., Walton, D., 2010. Combinatorially composing Chebyshev polynomials. Journal of Statistical Planning and Inference 140, 2161–2167 [this
issue].
Chentzov, N., Shklarsky, D., Yaglom, I., 1993. USSR Olympiad Problem Book. Dover Publications, Inc., New York.
Ericksen, L., 2008. Multiple product identities: balanced weights, thetas, finite gold. In: William, W. (Ed.), Applications of Fibonacci Numbers, vol. 11,
Utilitas Mathematica Publishing Co., Winnipeg, to appear.
Ericksen, L., 2010. Lattice path combinatorics for multiple product identities. Journal of Statistical Planning and inference, this issue, doi:10.1016/j.jspi.
2010.01.017.
Shapiro, L.W., 1981. A combinatorial proof of a Chebyshev polynomial identity. Discrete Math. 34, 203–206.
Walton, D., 2007. A Tiling approach to Chebyshev polynomials. Senior Thesis, Harvey Mudd College, Claremont, CA.