Stability of Haagerup Property under Graph Product Shubhabrata Das and Partha Sarathi Ghosh Abstract . In this paper, we prove that any graph product of finitely many groups, all satisfying the Haagerup property (or Gromov’s a-T-menability) also satisfies Haagerup property. 1. Introduction Amenable groups were introduced by von Neumann, in connection with the Banach-Tarski paradox, as precisely the groups not admitting a paradoxical decomposition. Since then the notion of amenability was studied extensively and admits a number of equivalent characterizations. One such characterization is in terms of the existence of an approximate identity, consisting of finitely supported positive definite functions. A survey on this can be found in the book [ 14 ] and in the references therein. The reduced group C^{*} -algebra, associated to a group G , denoted by C^{*}_{\lambda}(G) , is the completion of the image of the left regular representation \lambda:\mathbb{C}G\rightarrow\mathbb{B}(\ell^{2}G) , under the operator norm, where \mathbb{C}G is the group ring of finitely supported \mathbb{C} -valued functions on G . A nuclear C^{*} -algebra means that the identity map (on itself) can be approximated by finite rank maps. G is amenable if and only if C^{*}_{\lambda}(G) is nuclear. The most prominent of all non-amenable groups is the non-abelian free group \mathbb{F}_{n} , on n -generators. In [ 11 ] , U. Haagerup proved that the non-nuclear reduced group C^{*} -algebra, C^{*}_{\lambda}(\mathbb{F}_{n}) , admits some approximation property, later termed as the Haagerup approximation property or simply Haagerup property , by Akermann-Walter [ 1 ] and Choda [ 7 ] (see definition 2.2 ). The class of groups with Haagerup property is quite large, containing all compact groups (finite groups in the discrete setting), all amenable groups (hence all abelian groups, solvable groups etc.), non-abelian free groups, fundamental groups of closed orientable surfaces etc. An equivalent definition for the groups with Haagerup property is the existence of a proper conditionally negative definite function on such groups. Groups acting properly on a tree [ 11 ] , or on a CAT(0) -cube complex [ 13 ] , admit such functions, and hence are shown to have Haagerup property. For a good survey on this topic see [ 6 ] , by Cherix et al. Yet another way to characterise Haagerup property in a group, often a very useful one, is by showing such groups admit a proper affine isometric action on a Hilbert space. SO(n,1),~SU(n,1) admit this kind of action on a Hilbert space, and hence have Haagerup property. Bekka, Cherix and Valette showed that amenable groups also admit such actions, [ 3 ] . In this light, Property (T) introduced by D. Kazhdan in [ 12 ] , can be seen as a strong negation of Haagerup property, which is perhaps the reason behind M. Gromov calling Haagerup groups a-T-menable , [ 10 ] . For example, as SL_{n}(\mathbb{Z}) has property (T) for n\geq 3 , they do not have Haagerup property. Other such examples are SL_{n}(\mathbb{Z})\ltimes\mathbb{Z}^{n} , Sp(n,1) for n\geq 2 etc. Direct product of two groups with Haagerup property is again a group with Haagerup property. Also, Haagerup property is stable under taking free products and amalgamated products along finite subgroups [ 6 ] . Graph product of groups is a notion where groups corresponding to vertices in a finite graph are combined, and relators are commutators given by the edges (see definition 3.1 ). This notion in a way interpolates between the free product and direct product of groups. Since Haagerup property is stable under both, it seems a natural question to ask whether the Haagerup property remains preserved under any graph product, i.e. to ask: if we take groups \{G_{v}\}_{v\in V(\Gamma)} , one for each vertex in a finite graph \Gamma and all with Haagerup property, does the graph product satisfy Haagerup property? We answer this question in the positive. Theorem 1.1 . Suppose \Gamma is a finite graph and \{G_{v}\}_{v\in V(\Gamma)} is a collection of Haagerup groups. Then the graph product G(\Gamma) has Haagerup property. This result was first proved by Y. Antolin and D. Dreesen in [ 2 ] , following a very different approach. In some sense our proof is more direct. We show Haagerup property by producing a conditionally negative function on the graph product by gluing such functions on the vertex groups. 2. Haagerup Property Suppose G is a group. From now on G will always be a finitely generated group. Given a function \phi:G\rightarrow\mathbb{C} , it will induce a kernel k_{\phi}:G\times G\rightarrow\mathbb{C} defined by k_{\phi}(g,h):=\phi(h^{-1}g) . By a positive definite function \phi:G\rightarrow\mathbb{C} , we mean that for each n\in\mathbb{N} , given any \{c_{1},c_{2},\ldots,c_{n}\}\subset\mathbb{C} and \{g_{1},g_{2},\ldots,g_{n}\}\subset G we have: \sum_{i,j=1}^{n}c_{i}\overline{c_{j}}k_{\phi}(g_{i},g_{j})=\sum_{i,j=1}^{n}c_{i}\overline{c_{j}}\phi(g_{j}^{-1}g_{i})\geq 0. The set \ell^{2}G is the space of all square summable functions on G , which will form a Hilbert space with respect to the inner product \langle f_{1},f_{2}\rangle:=\sum_{g\in G}f_{1}(g)\overline{f_{2}(g)} . (\ell^{2}G)_{1} is the unit sphere of the Hilbert space. An alternative characterization of \phi being positive definite is ( [ 8 ] ): the existence of an orthogonal representation \pi:G\rightarrow\mathbb{B}(\ell^{2}G) and \xi\in(\ell^{2}G)_{1} such that, \phi(g)=\langle\pi(g)\xi,\xi\rangle Similarly a \phi will be called conditionally negative definite (CND) if for any n\in\mathbb{N} the n\times n matrix \begin{bmatrix}K_{ij}=\phi(g_{j}^{-1}g_{i})\end{bmatrix} is a conditionally negative definite matrix for arbitrary choice of subsets \{g_{1},g_{2},\ldots,g_{n}\}\subset G . An equivalent characterization will be the existence of a Hilbert space \mathcal{H} and a map R:G\rightarrow\mathcal{H} such that, (1) k_{\phi}(g,h)=\phi(h^{-1}g)=||R(g)-R(h)||^{2} Given any kernel k on G , either positive definite or conditionally negative definite, it will induce a function \phi_{k} of same type on G defined by the relation \phi_{k}(g)=k(g,e) , if k is a G -invariant i.e. k(g,h)=k(h^{-1}g,e) for all (g,h)\in G\times G . The interaction of positive definite kernels and conditionally negative definite is due to Schoenberg. Lemma 2.1 . ( [ 16 ] , Theorem D.11 [ 4 ] ) k:G\times G\rightarrow\mathbb{C} is conditionally negative definite if and only if e^{-k}:G\times G\rightarrow\mathbb{C} is positive definite. Definition 2.2 . A group G is called Haagerup group (or a group with Haagerup property) if there is a sequence of functions \phi_{n}:G\rightarrow\mathbb{C} with the following properties: a) \phi_{n}(g)\rightarrow 1 point-wise b) each \phi_{n} ’s vanishes at \infty i.e. \phi_{n}(g)\rightarrow 0 as |g|\rightarrow\infty , c) each \phi_{n} ’s are positive definite. Suppose one has a proper function \phi on G , which is conditionally negative definite, then form lemma 2.1 , the sequence \phi_{n}=e^{-\frac{\phi}{n}} will satisfy all the conditions of definition 2.2 . The other direction i.e. given a sequence \phi_{n} with above properties one can construct a proper conditionally negative definite function \phi on G , see [ 6 ] for the proof of this fact. Therefore we have an equivalent definition: Definition 2.3 . G has Haagerup property if there is a proper, symmetric, conditionally negative definite function on G . Whenever G is Haagerup group, we have a proper CND function \phi , and a map R from G to a Hilbert space \mathcal{H} , satisfying equation ( 1 ). We will call this data, denoted by (G,\phi,\mathcal{H},R) a Haagerup tuple corresponding to G . 3. Graph Product Suppose \Gamma is a finite graph, without loops and bigons. V(\Gamma) is vertex set of \Gamma and E(\Gamma) is the set of edges of \Gamma . An unordered edge between vertices v and w , is denoted by [v,w] . Also let \{G_{v}\}_{v\in V(\Gamma)} be a fixed family of groups. A graph product G(\Gamma) of \{G_{v}\}_{v\in V(\Gamma)} is obtained by combining G_{v} ’s along \Gamma . Definition 3.1 . For a finite graph \Gamma and a family of groups \{G_{v}\}_{v\in V(\Gamma)} , the graph product G(\Gamma) is the quotient *_{v\in V(\Gamma)}G_{v}\Big/\langle\langle\{[G_{v},G_{w}]:[v,w]\in E(\Gamma)\}\rangle\rangle . G_{1} G_{2} G_{3} G_{4} G_{5} Figure 1. \Gamma is disconnected G(\Gamma)=*_{i=1}^{5}G_{i} G_{1} G_{2} G_{3} G_{4} G_{5} Figure 2. \Gamma is complete G(\Gamma)=\oplus_{i=1}^{5}G_{i} The free product and the direct product of groups are the two extreme examples of graph product of groups. One can easily see that if \Gamma is a graph with n -vertices and no edges then G(\Gamma) is the free product of the vertex groups, and if \Gamma is a complete graph with n -vertices, then G(\Gamma) is the direct product (see the above figures). Graph product of groups were studied by Green in her thesis [ 9 ] . For \{G_{v}\}_{v\in V(\Gamma)} , clearly the graph product G(A)\leq G(\Gamma) , where A is a subgraph of \Gamma . Define lk(v):=\{w\in V(\Gamma):[v,w]\in E(\Gamma)\} and st(v):=lk(v)\cup\{v\} , for v\in V(\Gamma) . By ST(v) we denote the maximal subgraph of \Gamma , whose vertex set is st(v) . Then G(st(v)) simply means the graph product G(ST(v)) . Each element g\in G(\Gamma) can be written in the form g=g_{1}g_{2}\ldots g_{m} where each g_{i}\in G_{v_{i}} , for v_{i}\in V(\Gamma) , for some m . If [v_{i},v_{i+1}]\in E(\Gamma) then g=g_{1}g_{2}\ldots g_{m}=g_{1}\ldots g_{i-1}g_{i+1}g_{i}g_{i+2}\ldots g_{m} We call the above operation a ‘shuffle’. One can perform the following three types of operations on the form g=g_{1}g_{2}\ldots g_{m} : a) perform shuffle, b) drop identity element if it appears, c) multiply g_{i} and g_{i+1} if both come from G_{v_{i}} Remark 3.2 . Shuffle does not alter the number of elements. Operations (b) and (c) reduce the number of elements by 1 . A form g=g_{1}g_{2}\ldots g_{m} , is called reduced if the number of elements remains the same after performing the above operations any number of times. This reduced form is the usual normal form in case of free product, which is uniquely determined. The presence of \mathcal{R}_{\Gamma} relations in a general graph product G(\Gamma) prevents the reduced form to be unique. However, the number of elements in any reduced form of a g\in G(\Gamma) , is unique [ 9 , Corollary 3.13] . Therefore it will induce a length function l_{r} , given by the reduced length |\cdot|_{r} on G(\Gamma) i.e. (2) \begin{array}[]{rl}l_{r}:G(\Gamma)&\rightarrow\mathbb{R}\\ g&\mapsto l_{r}(g)=|g|_{r}:=m\end{array} where the reduced length |g|_{r} of a g\in G(\Gamma) in reduced form g=g_{1}g_{2}\ldots g_{m} , is m . Remark 3.3 . Whenever g_{1}g_{2}\ldots g_{m} is a reduced form, and g_{i},g_{j}\in G_{v} , then there is some q\in\mathbb{N} and u\notin st(v) such that i