Linear and Multilinear Algebra Vol. 59, No. 12, December 2011, 1393–1397 Inverse of the distance matrix of a block graph R.B. Bapata and Sivaramakrishnan Sivasubramanianb* aStat-Math Unit, Indian Statistical Institute, Delhi, 7-SJSS Marg, New Delhi 110 016, India; bDepartment of Mathematics, Indian Institute of Technology, Mumbai 400 076, India Communicated by S. Kirkland (Received 28 May 2009; final version received 19 January 2011) A connected graph G, whose 2-connected blocks are all cliques (of possibly varying sizes) is called a block graph. Let D be its distance matrix. By a theorem of Graham, Hoffman and Hosoya, we have det(D) 6¼ 0. We give a formula for both the determinant and the inverse, DÀ1 of D. Keywords: distance matrix; determinant; Laplacian matrix AMS Subject Classifications: 15A09; 15A15; 15A24 1. Introduction Graham et al. [3] proved a very attractive theorem about the determinant of the distance matrix DG of a connected graph G as a function of the distance matrix of its 2-connected blocks. In a connected graph, the distance between two vertices d(u, v)is the length of the shortest path between them. Let A be an n  n matrix. Recall that iþj for 1 i, j n, the cofactor ci,j is defined as (À1) times the determinant of the submatrix obtained by deleting row i and column j of A. For a matrix A, let #(A) ¼ i,jci,j be the sum of its cofactors. Graham et al. [3] showed the following theorem. P THEOREM 1 If G is a connected graph with 2-connected blocks G1, G2, ..., Gr, then # r # r # ðDGÞ¼ i¼1 ðDGi Þ and detðDGÞ¼ i¼1 detðDGi Þ j6¼i ðDGj Þ. Let DQbe the distance matrix ofPa connectedQgraph, all of whose blocks are cliques. Such graphs are called block graphs in [2] and let Gi denote the blocks of G (for 1 i r). See Figure 1 for an example. Further, we give a formula for det(D) for the distance matrix D of a block graph G in terms of its block sizes and n, its number of vertices. From the formula it will be clear that det(D) 6¼ 0. Hence, we are interested in À1 finding D . For the case when all blocks are K2’s (i.e. the graph G is a tree) it is À1 L 1 t known [1,4] that D ¼À2 þ 2ðnÀ1Þ , where L is the Laplacian matrix of G and is the n  1 column vector with v ¼ 2 À degv. Similarly, it is known that when all blocks À1 L 1 t are K3’s [5], we have D ¼À3 þ 3ðnÀ1Þ where L is again the Laplacian of G and *Corresponding author. Email: [email protected] ISSN 0308–1087 print/ISSN 1563–5139 online ß 2011 Taylor & Francis http://dx.doi.org/10.1080/03081087.2011.557374 http://www.tandfonline.com 1394 R.B. Bapat and S. Sivasubramanian 2 6 3 7 1 5 4 9 8 10 11 Figure 1. An example of a block graph. À1 is the column vector with v ¼ 3 À degv. Thus, D is a constant times L plus a multiple of a rank one matrix. We show a similar statement for block graphs. 2. Determinant and inverse of D Let G be a block graph on n vertices with blocks Gi,1 i r, where each Gi is a pi-clique. Denote by G the non-zero constant r pi À 1 G ¼ : ð1Þ i¼1 pi X The following theorem is easily derived from Theorem 1. THEOREM 2 Let G be a block graph on n vertices with blocks Gi,1 i r, where each nÀ1 r Gi is a pi-clique. Let D be its distance matrix. Then, detðDÞ¼ðÀ1Þ G j¼1 pj. Proof As each DGi is the matrix J À I where J is the all ones matrix andQ I is the piÀ1 identity matrix of dimension pi  pi, it is easy to see that detðDGi Þ¼ðÀ1Þ ð pi À 1Þ piÀ1 and #ðDGi Þ¼ðÀ1Þ pi (the #ðDGi Þ calculation is immediate if we use [3, Lemma 1]). r Since i¼1 pi ¼ n þ r À 1, the equality of the theorem follows from Theorem 1. g P For a block graph G, consider the jV(G)j-dimensional column vector defined as follows. Let a vertex v 2 V be in k 1 cliques of sizes p1, p2, ..., pk (where each pi 4 1). Let k 1 v ¼ Àðk À 1Þ: ð2Þ "#i¼1 pi X 163 t For the block graph given in Figure 1, we have G ¼ 60 , and ¼ (À1/4, 1/4, 1/4, 1/4, À3/10, 1/5, 1/5, 1/5, À7/15, 1/3, 1/3). LEMMA 1 Let G be a block graph and let be the vector defined above. Then, v2V(G) v ¼ 1. PProof By induction on b, we have the number of blocks of G, with the case b ¼ 1 being clear. When G has more than one block, let H be any leaf block (i.e. a block whose deletion does not disconnect G) connected through cut-vertex c. Clearly, a leaf block H exists and let F ¼ G À {H À c} be the smaller graph obtained by deleting H À c from G. Let H be a p-clique (i.e. H ¼ Kp ). By induction, for the graph F,we know S ¼ v2V(F) v ¼ 1. It is simple to note that when we move to G from F, the P Linear and Multilinear Algebra 1395 vector is different from that for F only for the vertices of H. The change in for G is easily seen to be (1/p À 1) for c and 1/p for the other p À 1 vertices of H. Thus the sum of the changes is zero, completing the proof. g LEMMA 2 Let D be the distance matrix of a block graph G. Let jVj¼n and be the vector defined by Equation (2). Let 11 be the n-dimensional vector with all components equal to 1. Then D ¼ G11, where G is as given in Equation (1). Proof We again induct on b, the number of blocks of G with the case b ¼ 1 being simple. Delete a leaf block H connected to G through c and let F ¼ G À {H À c}. Let H be a p-clique and let DF be the distance matrix of F. Let 11F be the vector 11 restricted to vertices of F. Let be the column of DF corresponding to the vertex c. The v-th component of is v ¼ dv,c where du,v is the distance between vertices u, v 2 F. It is simple to note that DF þ 11F ÁÁÁ þ 11F t ð þ 11FÞ 0 ÁÁÁ 1 D ¼ 0 1: . .. B . C B C B ð þ 11 Þt 1 ÁÁÁ 0 C B F C @ A If F is the restriction of to F, then by induction we have DF F ¼ F11F. Here F is the vector as in Equation (1) for the graph F. Let t ¼ D and for v 2 F and let Rv(DF) be the v-th row of DF. For vectors a, b with identical dimension, ha, bi denotes the usual (real) inner product of two vectors. For a vertex v 2 F, the v-th component 1 pÀ1 pÀ1 of t is tv ¼hRvðDFÞ, Fiþðp À 1Þ v þð v þ 1ÞÁ p . Hence, tv ¼ F þ p . Thus for all pÀ1 vertices in F, we have G ¼ F þ p . For vertices u 2 H À {c}, we have 1 p À 2 t ¼hð þ 11 Þt, iþ À 1 þ u F F p p p À 1 p À 1 ¼ F þ ð FÞv À 1 þ ¼ F þ , v 2 F p p X where in the first line we have used the fact that c ¼ 0 and in the second line we have pÀ1 pÀ1 used Lemma 1. Thus, for all vertices u 2 V(G), tu ¼ F þ p . Since G ¼ F þ p , the proof is complete. g Let G have vertex set [n] and blocks Gi where 1 i r. Each Gi is also considered as a graph on [n] with perhaps isolated vertices and let its edge set be Ei (i.e. Gi is a clique on say pi vertices, but consider it as a graph on [n]). Let Li be the Laplacian of Gi ¼ ([n], Ei). Let I be the jVjÂjVj identity matrix. Define r ^ 1 L ¼ Li: i¼1 pi X t LEMMA 3 With the above notation, LD^ þ I ¼ 11 . Proof We again induct on b, the number of blocks of G with the base case b ¼ 1 ^ being simple. Let H, F, c be as in the proof of Lemma 1 and let H be a p-clique. Let LF be the combination of the Laplacian as before, but only for the blocks of F and let DF be the distance matrix of F. Similarly, let IF be the identity matrix of order jFjÂjFj. 1396 R.B. Bapat and S. Sivasubramanian Let ec be the jFj-dimensional column vector with a 1 in position c and zero elsewhere and let ¼ DFec. Let 11HÀc be a (jHjÀ1)-dimensional all ones column vector. H is a leaf-block, but considering it as a graph on [n], let its Laplacian be denoted by LH.