Discrete Applied Mathematics 156 (2008) 2793–2803 www.elsevier.com/locate/dam Note Identities via Bell matrix and Fibonacci matrix Weiping Wanga,∗, Tianming Wanga,b a Department of Applied Mathematics, Dalian University of Technology, Dalian 116024, PR China b Department of Mathematics, Hainan Normal University, Haikou 571158, PR China Received 31 July 2006; received in revised form 3 September 2007; accepted 13 October 2007 Available online 21 February 2008 Abstract In this paper, we study the relations between the Bell matrix and the Fibonacci matrix, which provide a unified approach to some lower triangular matrices, such as the Stirling matrices of both kinds, the Lah matrix, and the generalized Pascal matrix. To make the results more general, the discussion is also extended to the generalized Fibonacci numbers and the corresponding matrix. Moreover, based on the matrix representations, various identities are derived. c 2007 Elsevier B.V. All rights reserved. Keywords: Combinatorial identities; Fibonacci numbers; Generalized Fibonacci numbers; Bell polynomials; Iteration matrix 1. Introduction Recently, the lower triangular matrices have catalyzed many investigations. The Pascal matrix and several generalized Pascal matrices first received wide concern (see, e.g., [2,3,17,18]), and some other lower triangular matrices were also studied systematically, for example, the Lah matrix [14], the Stirling matrices of the first kind and of the second kind [5,6]. In this paper, we will study the Fibonacci matrix and the Bell matrix. Let us first consider a special n × n lower triangular matrix Sn which is defined by = 1, i j, (Sn)i, j = −1, i − 2 ≤ j ≤ i − 1, for i, j = 1, 2,..., n. (1.1) 0, else, Thus, we have 1 0 ········· 0 −1 1 0 ······ 0 −1 −1 1 0 ··· 0 S = . n − − .. . 0 1 1 1 . . .. .. .. .. 0 ··· 0 −1 −1 1 ∗ Corresponding author. E-mail addresses: [email protected] (W. Wang), [email protected] (T. Wang). 0166-218X/$ - see front matter c 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.dam.2007.10.025 2794 W. Wang, T. Wang / Discrete Applied Mathematics 156 (2008) 2793–2803 By some computation, it is not difficult to find that the inverse of Sn is 1 0 ········· 0 1 1 0 ······ 0 2 1 1 0 ··· 0 S−1 = . n .. . 3 2 1 1 . .. .. .. .. ··· 3 2 1 1 −1 Now, one can observe that the inverse matrix Sn is related to the famous Fibonacci numbers, which are defined by −1 F0 = 0, F1 = 1 and Fn = Fn−1 + Fn−2 for n ≥ 2. Actually, if we denote Sn by Fn, then Fi− j+1, i − j + 1 > 0, (Fn)i j = fi j = for i, j = 1, 2,..., n. (1.2) , , 0, i − j + 1 ≤ 0, For this reason, Fn is called the n × n Fibonacci matrix. The Fibonacci matrix and several generalizations have already been studied in many works. In [11], the authors gave the Cholesky factorization of the Fibonacci matrix Fn and discussed the eigenvalues of the symmetric Fibonacci T matrix FnFn . In [8,9], the authors studied the generalized Fibonacci matrix as well as the k-Fibonacci matrix. Moreover, the relations between the Fibonacci matrix and some other lower triangular matrices, such as the Pascal matrix, the Stirling matrices of the first kind and of the second kind as well as the Bernoulli matrix, were investigated in [10,19], respectively. It is well known that many combinatorial sequences, for instance, the Stirling numbers and the Lah numbers, are special cases of the Bell polynomials (see [1], [4, Chapter 11] and [7, Chapter 3]). Therefore, by means of the study of the matrix related to the Bell polynomials, we will have a unified approach to various lower triangular matrices. In [15], we have studied the factorizations of the Bell matrix, and in the present paper, we will do some further researches on the relations between the Bell matrix and the Fibonacci matrix. Additionally, from the matrix representations, we will also give some combinatorial identities. This article is organized as follows. In Section 2, we consider the relations between the matrix related to the exponential partial Bell polynomials and the Fibonacci matrix. Section 3 is devoted to the identities concerning the Fibonacci numbers and Section 4 is devoted to the Bell polynomials with respect to Ω as well as the iteration matrix. Finally, in Section 5, we extend the discussion to the generalized Fibonacci numbers and their corresponding matrix. 2. Relations between Bell matrix and Fibonacci matrix The exponential partial Bell polynomials are defined as follows [7, pp. 133 and 134]: Definition 2.1. The exponential partial Bell polynomials are the polynomials Bn,k = Bn,k(x1, x2,..., xn−k+1) in an infinite number of variables x1, x2,... , defined by the series expansion !k 1 X tm X tn x = B , k = 0, 1, 2,.... k! m m! n,k n! m≥1 n≥k Their exact expression is n! = X c1 c2 ··· Bn,k(x1, x2,..., xn−k+1) c c x1 x2 , c1!c2!··· (1!) 1 (2!) 2 ··· where the summation takes place over all integers c1, c2, c3,... ≥ 0, such that c1 + 2c2 + 3c3 + · · · = n and c1 + c2 + c3 + · · · = k. By the definition, we can readily obtain some special values of the Bell polynomials. Particularly, we have B0,0 = 1, Bn,0 = 0, Bn,1 = xn for n ≥ 1 and Bn,k = 0 for n < k. In addition to these, the following lemma holds [7, p. 135]. W. Wang, T. Wang / Discrete Applied Mathematics 156 (2008) 2793–2803 2795 Lemma 2.2. For positive integers n and k, we have Bn,k(1, 1, 1, . .) = S(n, k), (Stirling number of the second kind), n − 1 n! Bn k(1!, 2!, 3!, . .) = = L(n, k), (Lah number), , k − 1 k! Bn,k(0!, 1!, 2!, . .) = s(n, k), (unsigned Stirling number of the first kind), n B (1, 2, 3, . .) = kn−k,(idempotent number). n,k k The readers are referred to [1] and [4, Chapter 11] for some other sequences which can be obtained from the Bell polynomials. × = = −1 = 0 Now, define the n n Bell matrix Bn by (Bn)i, j Bi, j and denote (Sn)i, j (Fn )i, j fi, j , where i, j = 1, 2,..., n. In the next lemma, we will consider the matrix multiplications Sn Bn and Bn Sn. Lemma 2.3. We have Sn Bn = Nn and Bn Sn = Mn, where the n × n matrices Nn and Mn are defined by (Nn)i, j = qi, j = Bi, j − Bi−1, j − Bi−2, j , (Mn)i, j = pi, j = Bi, j − Bi, j+1 − Bi, j+2, respectively, for i, j = 1, 2,..., n. Proof. From Definition 2.1 and the remark after it, we can determine the elements of the matrix Nn. Especially, q1,1 = B1,1, q1, j = 0 for j ≥ 2, q2,1 = B2,1 − B1,1, q2,2 = B2,2, and q2, j = 0 for j ≥ 3. The elements of Mn can also be determined in a similar way. Let us first verify the equation Sn Bn = Nn. 0 = = ≥ Pn 0 = 0 = = Pn 0 = 0 = Since f1, j B1, j 0 for j 2, then k=1 f1,k Bk,1 f1,1 B1,1 B1,1 q1,1, k=1 f1,k Bk, j f1,1 B1, j = ≥ 0 = − 0 = 0 = ≥ Pn 0 = 0 + 0 = 0 q1, j for j 2. Since f2,1 1, f2,2 1 and f2, j 0 for j 3, then k=1 f2,k Bk,1 f2,1 B1,1 f2,2 B2,1 − = Pn 0 = 0 + 0 = = Pn 0 = 0 + 0 = B2,1 B1,1 q2,1, k=1 f2,k Bk,2 f2,1 B1,2 f2,2 B2,2 B2,2 q2,2, and k=1 f2,k Bk, j f2,1 B1, j f2,2 B2, j 0 = q2, j for j ≥ 3. ≥ Pn 0 = 0 + 0 + 0 = − − Next, let i 3. In view of (1.1), k=1 fi,k Bk, j fi,i Bi, j fi,i−1 Bi−1, j fi,i−2 Bi−2, j Bi, j Bi−1, j Bi−2, j = qi, j . Therefore, we have Sn Bn = Nn. Similar to the preceding process, we can also verify the equation Bn Sn = Mn. Since the Fibonacci matrix Fn is the inverse of Sn, the following theorem holds. Theorem 2.4. The Bell matrix Bn can be factorized as Bn = FnNn = MnFn. (2.1) From the factorizations, we have for 1 ≤ k ≤ n that n X Bn,k = Fn−l+1(Bl,k − Bl−1,k − Bl−2,k) (2.2) l=k n X = (Bn,l − Bn,l+1 − Bn,l+2)Fl−k+1. (2.3) l=k T Let En = (1, 1,..., 1) . Since Bn En = MnFn En and [10, Corollary 2.2] F1 + F2 + · · · + Fn−2 = Fn − 1, (2.4) then (2.3) implies that n n X X Bn,k = (Bn,k − Bn,k+1 − Bn,k+2)(Fk+2 − 1). (2.5) k=1 k=1 2796 W. Wang, T. Wang / Discrete Applied Mathematics 156 (2008) 2793–2803 Making use of the general identities (2.2), (2.3) and (2.5), we can obtain the corresponding ones for some special combinatorial sequences, which are given by the corollaries below. It should be noticed that, from the generating functions (see [7, pp. 156, 206 and 213]), S(i, j), s(i, j) and L(i, j) will vanish for 0 ≤ i < j. Additionally, we also follow the convention that S(−1, 1) = s(−1, 1) = L(−1, 1) = 0.