HARMONIC MAPS FROM A SIMPLICIAL COMPLEX AND GEOMETRIC RIGIDITY. Georgios Daskalopoulos & Chikako Mese. Abstract

Transcription

1 . differential geometry 78 ( HARMONIC MAPS FROM A SIMPLICIAL COMPLEX AND GEOMETRIC RIGIDITY Georgios Daskalopoulos & Chikako Mese Abstract We study harmonic maps from an admissible flat simplicial complex to a non-positively curved Riemannian manifold. Our main regularity theorem is that these maps are C,β at the interfaces of the top-dimensional simplices in addition to satisfying a balancing condition. If we assume that the domain is a 2-complex, then these maps are C. As an application, we show that the regularity, the balancing condition and a Bochner formula lead to rigidity and vanishing theorems for harmonic maps. Furthermore, we give an explicit relationship between our techniques and those obtained via combinatorial methods.. Introduction Harmonic maps are critical points of the energy functional. The energy of a map ϕ : X N between two spaces X and N is defined to be the integral over the domain space of the energy density function which measures the total stretch of the map at each point of X. In the case when X and N are smooth Riemannian manifolds, the energy density function is the squared norm of the differential of the map. One of the highlights of the harmonic map theory has been its successful applications to study representation of discrete groups. Suppose Γ is a fundamental group of a manifold X acting on a space N by ρ : Γ Isom(N. The idea is to associate the action with an equivariant harmonic map f : X N where X is the universal cover of X. Once the existence is established, one can use the curvature assumptions on the domain and the target spaces to make strong statements about f and hence about the representation ρ. To illustrate this, let X be a compact Riemannian manifold of nonnegative Ricci curvature and with fundamental group Γ, and N be a complete Riemannian manifold of non-positive sectional curvature. Consider a representation ρ : Γ Isom(N and let f : X N be a The first author was partially supported by research grant NSF DMS The second author was partially supported by research grant NSF DMS Received 09/0/

2 270 G. DASKALOPOULOS & C. MESE Γ equivariant harmonic map. Such a map f exists as long as the action ρ does not fix an equivalence class of rays in N. The Eells-Sampson Bochner formula implies ( 2 f 2 d f 2 + d f(ric X (e k, d f(e k f TN R N (d f(e k, d f(e l d f(e l, d f(e k f TN where Ric X and R N are the Ricci and sectional curvatures of X and N respectively. The right hand side of the equation above is non-negative. Furthermore, Stoke s theorem says (2 f 2 0, X from which we conclude that each term on the right hand side of equation ( is also zero. In particular, d f 0, i.e., the map f is totally geodesic. The representation ρ : Γ Isom(N is then said to be rigid. Notice that the two important ingredients that were used here are the Bochner formula and Stoke s Theorem. Further rigidity formulas were discovered by Siu, Corlette, and others; see for further examples [Si], [C], and [MSiY]. In the seminal work of Gromov-Schoen [GS] and subsequently Korevaar-Schoen [KS], [KS2], the situation in which N is only a complete metric space rather than a smooth manifold was considered. This enabled them to prove superrigidity of p-adic representations along the lines of [C]. In a different direction, one can ask if it is possible to allow the domain X to be singular, for example a simplicial complex. This idea goes back to the work of Garland [G] and was subsequenctly elaborated by several groups of authors (cf. [BS], [Gr], [IN], [W], [W2] and [Z] in connection with Kazhdan property (T. For the nonlinear versions, the key idea is to define a combinatorial version of harmonic maps and relate them via a Bochner formula to a combinatorial analogue of curvature on X. This is in essence a refinement of Garland s notion of p-adic curvature. (For a definition of combinatorial harmonic maps, see [J] and [W], [W2]. The actual notion of a harmonic map on a polyhedral domain, rather than its combinatorial counterpart, was first introduced in [Ch] and was further developed in [EF], [DM] and [M]. In the special case when X is a flat n-dimensional simplicial complex, it was shown in [DM] and [M] that the harmonic map is Lipschitz across the edges. In the case of a 2-dimensional flat simplicial complex, the harmonic map has particular growth rate at the vertices depending on the order of the map. (See Section 2, Theorems and 2 for precise statements of the results. The main goal of this paper is to further improve the regularity of harmonic maps defined on polyhedral domains. In particular, we show

3 HARMONIC MAPS FROM A SIMPLICIAL COMPLEX 27 that the harmonic map must be C,β across the strict (n -skeleton and satisfy a natural balancing condition. (See Theorem 4 and Corollary 5. Furthermore, if n 2, then we show that the harmonic map must be C at the interfaces of the -skeleton. The second goal of the paper is to generalize the Bochner technique described earlier in the introduction to the case when X is a simplicial complex. Here, we restrict to the case dimx 2 and introduce weights (see the beginning of Section 2 for a precise definition. This imposes no restriction as far as applications to group theory is concerned. We show that under the assumption that f is bounded, Stoke s formula (2 still holds. By combining this with a simplex-wise Bochner formula (, we obtain that f is totally geodesic (see Theorem 8. It therefore remains to establish the assumption on X for which harmonic maps from X are forced to satisfy the condition f C. It is not hard to see that the latter condition is equivalent to the condition that ord p f for all points p in X, which in turn is equivalent to the condition that the first nonzero eigenvalue of the Laplacian of the link of p, Lk(p, is. Of course the Lk(p is a graph and we can easily relate the spectrum of the Laplacian on Lk(p to the spectrum of the discrete Laplacian (cf. Proposition 3 and Corollary 4. In particular, we show that the condition that the first nonzero eigenvalue being > is equivalent to the first nonzero eigenvalue of the discrete Laplacian being > 2, which is precisely the condition appearing in the combinatorial approach (cf. [BS], [IN], [W], [W2], [Z]. This allows us to deduce the main theorem in [W], that if Σ n is compact simplicial n-complex with admissible weight whose first nonzero eigenvalue of each link of a vertex is > 2 then any isometric action of Γ on a complete simply connected manifold of nonpositive sectional curvature has a fixed point, as a direct consequence of the Eells-Sampson Bochner formula for polyhedral domains. Furthermore, our approach enables us to prove rigidity in the borderline case when the eigenvalue of the discrete laplacian is 2 as conectured by M.-T. Wang. There are important examples of simplicial complexes associated to p-adic groups where the eigenvalues are indeed equal to /2 [W2]. We now turn to the organization of the paper. Throughout the rest of the paper, X will be a compact, admissible and flat simplicial complex and (N, g a complete Riemannian manifold. Section 2 is a review of the results in [DM] about the existence and regularity of harmonic maps. The only difference in this paper is that our maps depend on weights w(f associated to the n-simplices F of X. This slight modification, which from the point of view of the analysis amounts to taking certain weighted Sobolev spaces associated to the complex, is necessary in order to cover all p-adic buildings. We therefore talk about the w-energy of

4 272 G. DASKALOPOULOS & C. MESE a map or w-harmonic maps, but this imposes no real analytical difficulty. Section 3 contains our main regularity result and the balancing condition. More precisely, we show: Theorem (cf. Theorem 4 and Corollary 6. Let f : X (N m, g be a w-harmonic map. Then for any n-dimensional simplex F, the restriction of f to F X (n 2 is a C,β -map. If n 2, then f is C. Theorem (cf. Corollary 5. Let f : X (N m, g be a w-harmonic map. For any point p X (n X (n 2, let F,...,F J be the n- simplices containing p and let E J F. Choose coordinates (x,..., x n on X near p so that E corresponds to the equation x n 0 and coordinates (y,...,y m on N near f(p. Set f α yα f F. Then w(f fα n (x,...,x n, 0 0 where w(f are the weights associated to F. We also prove: Theorem (cf. Theorem 8. Let f : X (N, g be a w-harmonic map where dim X 2, N has nonpositive sectional curvature and f is a bounded function. Then f is totally geodesic on each simplex of X. Furthermore, if N has negative sectional curvature, then f maps each 2-simplex into a geodesic. Finally, in Section 4, we relate the question of regularity of the harmonic map with spectral theory of graphs. In particular, we show: Theorem (cf. Theorem 2. Suppose that X is a 2-complex such that every nonzero eigenvalue of the link of every vertex in X satisfies λ. If f : X (N, g is a w-harmonic map into a complete Riemannian manifold of nonpositive sectional curvature, then f is totally geodesic on each 2-simplex of X. In particular, this implies that if the sectional curvature of N is negative, then f maps each simplex into a geodesic. If the eigenvalues satisfy the stronger condition λ > then f is a constant map. We also establish the equivalence of the eigenvalue condition in Theorem 2 with the one appearing in the combinatorial approach. More precisely, we show: Theorem (cf. Corollary 4. The condition λ (> in the previous theorem is equivalent to the condition that the first nonzero eigenvalue of the discrete Laplacian be (> 2. By taking (Σ n, c an arbitrary compact n-dimensional simplical complex, by reducing the weights c to its 2-skeleton X Σ (2 and by applying the previous two theorems on X, we immediately obtain the main theorem in [W].

5 HARMONIC MAPS FROM A SIMPLICIAL COMPLEX 273 Corollary (cf. Corollary 5. Let (Σ n, c be a compact simplicial complex with admissible weights. Assume that the first nonzero eigenvalue of the link of every vertex is > 2. Then π (Σ Γ has property F; i.e., any isometric action of Γ on a complete, simply connected manifold N of nonpositive sectional curvature has a fixed point on N N N. Acknowledgement. The first author would like to thank Professor M.- T. Wang for communicating his results to him. The second author would like to thank Professors H. Izeki and S. Nayatani for a useful discussion at the Conference on Differential Geometry in Tokyo and Nagoya in the winter of Definitions and known results A simplicial complex of dimension n is referred to as a n-complex. A connected locally finite n-complex is called admissible (cf. [Ch] and [EF] if the following two conditions hold: (i X is dimensionally homogeneous, i.e., every simplex is contained in a n-simplex, and (ii X is locally (n -chainable, i.e., for any (n 2-simplex v, every two n-simplices A and B can be oined by a sequence A F 0, e 0, F, e,...,f k, e k, F k B where F i is a n-simplex containing v and e i is a (n -simplex contained in F i and F i+. The boundary X of X is the union of all simplices of dimension n, which is contained in only one n dimensional simplex. We call a n-complex flat if each k-simplex F is isometric to the convex hull of k + equidistant points of distance in R k and every l-simplex L incident to a k-simplex K (l < k can be seen as a totally geodesic subset of K. In the sequel, all complexes are admissible, flat, compact and without boundary. An isometric action of a group Γ is a homomorphism ρ : Γ Isom(N. Let Γ π (X. A map ϕ : X N is said to be equivariant if ρ(γ ϕ(p ϕ(γp for γ Γ and p X. If Γ acts freely and properly discontinuously on N, then the map ϕ is a lift of the map ϕ : X N/Γ. By identifying X with a fundamental domain of X, we can think of ϕ also being defined on X. In order to include certain important examples appearing in p-adic geometry (e.g., p adic buildings, we will assume that for each n- dimensional simplex F in X, we have an associated weight w(f > 0 and we define the w-measure dµ w by setting dµ w w(fdx

6 274 G. DASKALOPOULOS & C. MESE where dx is the standard Lebesgue measure on F. We define the w- energy E w ( ϕ of a map ϕ : X (N, g as E w ( ϕ ϕ 2 dµ w w(f ϕ 2 dx X F F where F indicates the sum over all n-dimensional simplices F of X and ϕ 2 is defined as usual; i.e., ϕ 2 g k ( ϕ k, ϕ k. Of course, if w(f for all F, then we recover the usual notion of harmonicity defined in [DM]. For the sake of notational simplicity, we will fix weights w(f on F and we will denote dµ dµ w, E E w, etc. A map f : X N is said to be w-harmonic if E( f E( ϕ for all W,2 equivariant maps ϕ : X N (cf. [EF]. The following existence and regularity results for w-harmonic maps from a 2-complex into a non-positively curved Riemannian manifold follows by minor modification of the arguments presented in [DM] and [M]. (In [DM] and [M], we only considered weight function w so that w(f for all 2-simplices F of X. Theorem. Let X be a 2-complex with Γ π (X, N be a complete Riemannian manifold of non-positive curvature and ρ : Γ Isom(N be an isometric action of Γ. Assume that ρ does not fix an equivalent class of rays. Then there exists an equivariant w-harmonic map f : X N. Theorem 2. Let X be a 2-complex, N a complete Riemannian manifold of non-positive curvature and f : X N a w-harmonic map. Then f is Lipschitz continuous away from the 0-simplices of X with the Lipschitz bound dependent only on the total w-energy of f and the distance to the 0-simplices. Let p be a 0-simplex and be the order of f at p. (The definition of order is given in Section 4. Then there exists σ > 0 so that f 2 (q Cr 2 2 for all q B σ (p where C depends on E(f and r d X (p, q. More generally, if X is a n-complex and N and f are as above, then f is Lipschitz continuous away from the (n 2-simplices. 3. Regularity results Let X be a n-complex and N a complete Riemannian manifold as above. For 0 k n, let X (k denote the k-skeleton of X. Given p X (n X (n 2, choose an (n -simplex E containing p and let P U (y,...,y m R m be a local coordinate system of a

7 HARMONIC MAPS FROM A SIMPLICIAL COMPLEX 275 neighborhood U of f(p. Assume g is given by (g αβ in terms of this coordinate system. Choose a neighborhood X of p sufficiently small so that does not intersect any (n 2-simplex and f( is compactly contained in this coordinate patch. If F,...,F J are the n-simplices of X intersecting and E ǫ is a ǫ-neighborhood of E, we will use the coordinate system q (F E (x,...,x n R n for,...,j so that a point in E is given by (x,...,x n, 0 and a point in E ǫ is given by (x,...,x n with 0 x n < ǫ. Let E ǫ be the ǫ-neighborhood of E. Furthermore, for a map f : N, f(x,...,x n (f (x,...,x n,...,f m (x,...,x n in and let f α fα F for α,...,m and,...,j. Theorem 3. Let F,...,F J, E, E ǫ, (x,...,x n, and (f,...,f m as above. If f : N is a harmonic map, then for any Lipschitz function η : R with compact support and any α,...,m, lim w(f η(x,...,x n, ǫ fα F E ǫ n (x,...,x n, ǫdx dx n 0. Proof. Let ϕ (ϕ,...,ϕ m : X R m be a Lipschitz continuous map with compact support. For sufficiently small t, define f t : X U by setting The w-energy of f t in is E(f t ; k α,β f t f + tϕ (f + tϕ,...,f m + tϕ m. ( ( f α f β g αβ (f(x + tϕ(x k + t ϕα k k + t ϕβ k dµ and since f f 0 is w-energy minimizing, 0 d dt E(f t; t0 2 k α,β + 2 k α,β,γ k α,β g αβ (f(x fα k ϕ β k dµ d dx k g αβ,γ (f(x fα k f β k ϕγ dµ (g αβ (f(x fα k ϕβ dµ

8 276 G. DASKALOPOULOS & C. MESE k α,β k α,β,γ k α,β,γ k α,β k α,β k α,β,γ k α,β k α,β,γ k α,β 2 k α,β k α,β,γ g αβ (f(x 2 f α (x k 2ϕβ dµ g αβ,γ (f(x fα k f γ k ϕβ dµ g αβ,γ (f(x fα k f β k ϕγ dµ d (g dx k αβ (f(x fα k ϕβ dµ g αβ (f(x 2 f α (x k 2ϕβ dµ g αβ,γ (f(x fα k f γ k ϕβ dµ g γβ,α (f(x fα k f γ k ϕβ dµ g αβ,γ (f(x fα k f β k ϕγ dµ d (g dx k αβ (f(x fα k ϕβ dµ g αβ (f(x 2 f α (x k 2ϕβ dµ (g αβ,γ + g γβ,α g αγ,β fα k f γ k ϕβ dµ. Let η α m β g αβϕ β. Then ϕ β m α gαβ η α and we see that the last two terms above equal 2 f α η α + α 2 α k α,β,γ,δ f α η α + 2 k α,γ,δ g δβ (g αβ,γ + g γβ,α g αγ,β fα k f γ k η δdµ Γ δ αγ(f(x fα f γ k k η δ dµ 0.

9 HARMONIC MAPS FROM A SIMPLICIAL COMPLEX 277 The last equality is because f is a smooth harmonic map in the interior of each n-simplex F and we have the pointwise equality, f α + k β,γ Γ α f γ βγ (f(x fβ k k 0 in F. Therefore, by the monotone convergence theorem and the fact that f is Lipschitz away from the (n 2-skeleton by Theorem 2, we conclude d 0 (g dx k αβ (f(x fα k ϕβ (x dµ k α,β lim k α,β lim w(f E ǫ k α,β d (g dx k αβ (f(x fα k ϕβ (x dµ F E ǫ Since ϕ β has compact support in, α,β F E ǫ ( d dx k g αβ (f (x fα k ϕβ (x dx. ( d k g αβ (f (x fα k ϕβ (x dx 0 for k,...,n and,...,j. Hence, 0 lim lim w(f w(f α,β α,β F E ǫ ϕ β (x,...,x n, ǫdx. ( d dx n g αβ (f (x fα k ϕβ (x dx F E ǫ g αβ (f (x,...,x n, ǫ fα n Let η : R be a Lipschitz continuous compactly supported function and ϕ β g βα 0 η for α 0, β {,...,m}. Then α,β g αβ f α n ϕβ α,β α η fα 0 n g αβ g βα 0 η fα n δ αα0 η fα n

10 278 G. DASKALOPOULOS & C. MESE and therefore lim w(f η(x,...,x n, ǫ f α 0 E ǫ n (x,...,x n, ǫdx 0. We now prove our main regularity result. q.e.d. Theorem 4. Let f : X (N n, g be a w-harmonic map so that f( maps into a coordinate neighborhood. For any point p X (n X (n 2, let E be the (n -simplex containing p and F,...,F J be the n-simplices containing E in its closure. As in Theorem 3, let f α be the restriction of the coordinate function f α to the n-simplex F and (x,...,x n be a coordinate system for. Then there exist a neighborhood Ω of p so that f α C,β (Ω F, f α W 2,2 (Ω F and fα k W 2,2 (Ω F for k,...,n. Proof. Assume p corresponds to (0,...,0, let w w(f /(w(f + + w(f J and B r R n be a ball of radius r centered at the origin. Fix α 0 {,...,m} and 0 {,...,J} and, for r > 0 sufficiently small, define ψ : B r R by setting { f α 0 ψ(ˆx, x n 0 (ˆx, x n when x n 0 f α 0 0 (ˆx, x n + 2 J J w f α 0 (ˆx, x n when x n < 0 where we simplify notation by using (ˆx, x n (x,...,x (n, x n. By the (Lipschitz continuity of f, f α 0 (ˆx,0 f α 0 0 (ˆx,0 for all,...,j. Therefore, f α 0 0 (ˆx,0 + 2 J w f α 0 (ˆx,0 f α 0 0 (ˆx,0 + 2 J fα 0 0 (ˆx,0 f α 0 0 (ˆx,0. This implies that ψ is Lipschitz continuous in B r. Additionally, for fixed ǫ > 0, ψ fα0 n(ˆx, ǫ 0 n (ˆx, ǫ 2 f α 0 w (ˆx, ǫ. J n Since lim B r {x n ǫ} f α 0 ϕ(ˆx, ǫw (ˆx, ǫ 0 n w

11 HARMONIC MAPS FROM A SIMPLICIAL COMPLEX 279 for any ϕ Cc (B r by Theorem 3, we see that lim ϕ(ˆx, ǫ ψ n(ˆx, ǫ B r {x n ǫ} lim B r {x n ǫ} 2 lim B r {x n ǫ} lim B r {x n ǫ} lim B r {x n ǫ} ϕ(ˆx, ǫ fα 0 0 (ˆx, ǫ n f α 0 ϕ(ˆx, ǫw (ˆx, ǫ n ϕ(ˆx, ǫ fα 0 0 (ˆx, ǫ n ϕ(ˆx, ǫ fα 0 0 (ˆx, ǫ n + lim B (ϕ(ˆx, ǫ ϕ(ˆx, ǫ fα 0 0 (ˆx, ǫ r {x n ǫ} n lim B r {x n ǫ} ϕ(ˆx, ǫ ψ n(ˆx, ǫ + lim B (ϕ(ˆx, ǫ ϕ(ˆx, ǫ fα 0 0 (ˆx, ǫ. r {x n ǫ} n The last term is equal to zero by the dominated convergence theorem since ϕ(ˆx, ǫ ϕ(ˆx, ǫ converges uniformly to 0 as ǫ 0 and fα 0 0 is n bounded. Thus, (3 lim ϕ B ψ r {x n ǫ} n lim ϕ ψ n. B r {x n ǫ} Also, since f is a smooth harmonic map in the interior of a n-simplex, (4 f α + Let l β,γ Γ α βγ (f (x fβ γ f l l 0. Γ α (ˆx, x n l β,γ Γ α βγ (f (ˆx, x n fβ l (ˆx, xn f γ l (ˆx, xn and F : B r R be defined by { Γ α 0 F(ˆx, x n 0 (ˆx, x n when x n 0 Γ α 0 0 (ˆx, x n + 2 J w Γ α 0 (ˆx, xn when x n < 0.

12 280 G. DASKALOPOULOS & C. MESE The equality (4 implies that ψ(ˆx, x n F(ˆx, x n for x n > 0 and x n < 0. Thus, integration by parts implies ϕ ψ ϕ ψ + ϕ ψ B r {x n >ǫ} B r {x n >ǫ} B r {x n ǫ} n ϕf ϕ ψ n and ϕ ψ B r {x n < ǫ} B r {x n >ǫ} B r {x n < ǫ} B r {x n < ǫ} B r {x n ǫ} ϕ ψ ϕf + B r {x n ǫ} B r {x n ǫ} ϕ ψ n ϕ ψ n for any ϕ Cc (B r. Summing the above two equalities, letting ǫ 0 and noting (3, we see that ϕ ψ ϕf. B r B r Since F is bounded, the standard elliptic regularity theorem implies that ψ C,β (B r and ψ W 2,2 (B r for r < r. Since ψ(ˆx, x n f α 0 0 (ˆx, x n for x n 0 and α 0 {,...,m}, 0 {,...,J} are arbitrary choices, this shows f α C,β (B r {x n 0} and f α W 2,2 (B r {x n > 0} for any α,...,n and,...,j. Thus, the fact that 2 f α l k L 2 (B r {x n > 0} combined with fα being bounded implies that l k Γ α f γ βγ (f(x fβ l l l β,γ l β,γ +2 l β,γ Γ α f γ f δ βγ,δ (f(x fβ l l k B r {x n >ǫ} Γ α βγ (f(x 2 f β f γ l k l is L 2 (B r {x n > 0}. This implies F is L 2 (B k r. Now let k,...,n. Then ϕ ψ B r {x n >ǫ} k ϕ ψ B r {x n >ǫ} k + ϕ 2 ψ B r {x n ǫ} k n ϕ F k + ϕ 2 ψ k n B r {x n >ǫ} ϕ F k B r {x n ǫ} B r {x n ǫ} ϕ k ψ n

13 HARMONIC MAPS FROM A SIMPLICIAL COMPLEX 28 and B r {x n < ǫ} ϕ ψ k B r {x n < ǫ} B r {x n < ǫ} B r {x n < ǫ} ϕ ψ k ϕ F k ϕ F k + B r {x n ǫ} B r {x n ǫ} B r {x n ǫ} ϕ 2 ψ k n ϕ 2 ψ k n ϕ ψ k n. Adding the above two equalities, letting ǫ 0 and using (3 with ϕ replaced by ϕ, we obtain k B r ϕ ψ k B r ϕ F k. Since F L 2 (B k r, we conclude ψ W 2,2 (B k r for r < r. Thus, f α is W 2,2 (B k r {x n > 0}. q.e.d. Corollary 5. Let, F,...,F J, E, (x,...,x n, (f,...,f m, f α be as in the paragraph proceeding Theorem 3. If f : N is a harmonic map, then w(f fα n (x,...,x n, 0 0 for all α,...,m and all (x,...,x n, 0 E. Proof. This pointwise equality follows immediately from Theorem 3 and the regularity result of Theorem 4. q.e.d. The following corollary to Theorem 4 will be important in Theorem 7 when we restrict our attention to two-dimensional domains. Corollary 6. Assume n 2. Then for every 2-simplex F, the restriction of f α to the closure of F is C away from the 0-simplices. Proof. We can push the argument for the proof of Theorem 4 further in the case n 2. We let x x and y x 2. Using the Lipschitz continuity to bound the first partial derivatives of f, we see that 2 2 β,γ Γ α βγ (f(x ( f β f γ + fβ C 0 + C D 2 f + C 2 D 2 f 2 + C 3 D 3 f f γ where C 0, C, C 2, C 3 are constants, and D i f is the sum of the absolute values of the ith order partial derivatives of f that involves at most one. By Theorem 4, fα W 2,2 (B r + for some r > 0 where we use the notation B + r B r {y 0}. Thus, 2 f α 2, 2 f α W,2 (B r and hence

14 282 G. DASKALOPOULOS & C. MESE the Sobolev embedding theorem implies that 2 f α, 2 f α 2 Lq (B r for any q <. Therefore, we conclude that 2 F L 2 (B 2 r. Using an analogous argument as in the proof of Theorem 4, we can show that B r ϕ 2 ψ 2 B r ϕ 2 F 2 for any ϕ C c (B r. Thus, 2 ψ 2 W 2,2 (B r2 for some r 2 < r which immediately implies 2 f α 2 W 2,2 (B + r 2. Now we continue inductively. l+ f α l+, l+ f α l Assume l f α W 2,2 (B + l r l. Then W,2 (B r + l and l+ f α, l+ f α L q (B + l+ l r l for any q, 2,.... Then l+ F L 2 (B + l+ r l since l+ ( f l+ Γ α β βγ (f(x β,γ f γ + fβ P( D 2 f,..., D l+ f + C l+2 D l+2 f f γ where C l+ is a constant, and D i f is the sum of absolute values of the ith order partial derivatives of f that involves at most one and P(X 2,...,X l+ is a polynomial involving variables X 2,...,X l+. Using the weak differential equality ϕ l+ ψ B rl l+ ϕ l+ F B rl l+, ϕ C c (B rl we see that l+ f α l+ W 2,2 (B + r l+ for r l+ < r l. We now show that f α is C. By the trace theory, l+ f α W,2 (B l+ rl+ {y 0} and hence f α W l+2,2 (B rl+ {y 0} which in turn implies f α C l+ (B rl+ {y 0}. Since f α is a smooth harmonic map in B rl+ {y > 0}, this implies that f α C l (B r + l+ by the boundary regularity theory of harmonic maps. Since l can be made arbitrarily large, f α is C. q.e.d. For a 2-complex X, the balancing condition along the -simplex stated in Corollary 5 implies that Stoke s theorem can be applied to the integral of f 2 as long as f is a bounded function in X. Theorem 7. Let X be a 2-complex and f : X (N, g be a w- harmonic map. If there exists C so that f C, then f 2 dµ 0. X

15 HARMONIC MAPS FROM A SIMPLICIAL COMPLEX 283 Proof. The Bochner formula gives 2 f 2 (x, y df 2 R N ( f, f f, f where R N (, 0 by hypothesis. In particular, f 2 0 and the monotone convergence theorem implies (5 0 f 2 dµ X lim f σ 0 X 2 dµ. v B σ(v Now applying Stoke s theorem to each 2-simplex F of X, we have (6 0 f 2 dµ X (7 lim w(f σ 0 F F v F B σ(v η f 2 ds + lim w(f σ 0 v B σ(v F η f 2 ds F where F and v indicates the sum over all 2-simplices F and 0- simplices v of X respectively and η is the outward pointing normal to F v F B σ (v. For an edge point p X, let (x, x 2 (x, y, (f,...,f m, F,...,F J and f α be as in the paragraph proceeding Theorem 3. Then for every (x,0 w(f η f 2 (x,0 w(f f 2 (x,0 w(f α,β α,β g αβ (f (x, y ( f α f β + fα f β Now recall that the function φ : F C defined by ( f α f β φ (x, y g αβ (f (x, y fα f β 2i fα (x,y(x,0. f β (x, y

16 284 G. DASKALOPOULOS & C. MESE is a holomorphic function [S]. By Corollary 5, Im w(f φ (x,0 2 α,β g αβ (f (x,0 fα (x,0 w(f fβ (x,0 0. Here, we have used the fact that f is smooth and hence f α (x,0 f α(x,0 and fα (x,0 fα (x,0 for each,...,j. Hence φ(x, y w(f φ (x, y extends across y 0 as a holomorphic function. Riemann equation, ( 0 Imφ (x,0 ( Reφ (x,0 w(f Hence, α,β g αβ (f (x, y w(f f 2 (x,0 2 2 w(f w(f + 4 w(f α,β α,β α,β ( f α f β fα g αβ (f (x, y fα g αβ,γ (f (x,0 fα g αβ (f (x,0 fα f β f β f β By the Cauchy- (x, y (x,y(x,0. (x, y (x,y(x,0 f γ (x,0 2 f β (x,0.

17 HARMONIC MAPS FROM A SIMPLICIAL COMPLEX 285 Again, using the fact that f is smooth, which implies f (x,0 f (x,0 and f (x,0 f (x,0, we have w(f f 2 (x, α,β α,β By Corollary 5, Thus, we conclude (8 g αβ,γ (f (x,0 fα g αβ (f (x,0 fα (x,0 F w(f fγ w(f F v B σ(v for any σ > 0. This implies (9 0 f 2 dµ X v B σ(v F f β (x,0 w(f fγ (x,0 (x,0 0. w(f fβ η f 2 ds 0 w(f v B σ(v F (x,0. η f 2 ds by equation (5. With r denoting the distance from the vertex v, ( r f 2 ds σ d f 2 ds. dσ σ B σ(v F Since η r d dσ, equation (9 implies ( F w(f v B σ(v F f 2 ds σ B σ(v F and σ w(f f 2 ds σ F v B σ(v F is monotone non-increasing. On the other hand, the hypothesis implies that f 2 ds C σ for some constant C. Thus, L lim σ 0 F B σ(v F w(f v 0 f 2 ds σ B σ(v F

18 286 G. DASKALOPOULOS & C. MESE exists. Let 0 for ǫ 0 G(ǫ F w(f B ǫ(v F f 2 ds for ǫ (0, 2 ]. v Since f 2 is C away from the vertices, G is a differentiable function on (0, 2. Furthermore, since f is bounded, G is continuous on [0, 2 ]. Therefore, for σ i 0, we can choose ǫ i (0, σ i by the mean value theorem so that Then d dǫ w(f v F ( w(f F v w(f σ i v F B ǫi (v F w(f d ǫ i dǫ F v ( d w(f dǫ F v w(f ǫ F v i w(f σ F v i w(f ǫ F v i Letting i 0, we obtain 0 f 2 dµ lim X i 0 F w(f v B ǫ(v F B σi F f 2 ds f 2 ds. r f 2 ds ( f 2 ds ǫ B ǫ(v F f 2 B ǫ(v F B ǫi (v F B σi (v F B ǫi (v F f 2 ds f 2 B ǫi (v F We are now ready to prove the following: f 2 ds. ǫǫi ǫǫi ǫǫi r f 2 ds L L 0. q.e.d. Theorem 8. Suppose X is a 2-complex and (N, g a complete Riemannian manifold of nonpositive sectional curvature. If f : X (N, g is a w-harmonic map and f is bounded, then f is totally geodesic on each simplex of X. Furthermore, if the sectional curvature of N is strictly negative, then f maps each 2-simplex into a geodesic.

19 HARMONIC MAPS FROM A SIMPLICIAL COMPLEX 287 Proof. By Theorem 7, X f 2 dµ 0. On the other hand, for any point on the interior of a 2-simplex, the Bochner formula implies (0 2 f 2 df 2 0 by the hypothesis on the flatness of X and the nonpositive curvature of N. Therefore df 2 0 and f is totally geodesic on each simplex. The second statement follows because the second inequality of (0 is a strict inequality if f is nonconstant and the sectional curvature of N is negative. q.e.d. 4. Tangent maps and eigenvalues of the link By Theorem 8, a harmonic map is totally geodesic provided that the energy density function is bounded. This behavior can be guaranteed by certain assumption on the link of a vertex which is defined in terms of a spectral theory on graphs. Let G be a graph. We denote the edges and vertices of G by e,...,e L and v,...,v K respectively. For each k,...,k, let E k be the set of edges incident to vertex v k. We identify each edge e l with the interval [0, π 3 ] (the π 3 corresponding to the fact that all our 2-simplices are equilateral triangles and we assume that each edge e l has an associated weight ŵ l ŵ(e l. In the case G Lk(v where v is a vertex of a 2-dimensional simplex X with weights w(f, we define ŵ(e l w(f l where F l is the oin (i.e., convex hull of v and e l. Returning to the case of a general graph G with weights ŵ l, l,...,l, we let G be the set of functions ϕ : G R so that ϕ l ϕ el is smooth up to the endpoints and ϕ l ( ŵ l η (v k 0 e l E k where η is the outward pointing unit normal at the vertex v k. On each edge e l, define the measure ŵ l dτ, where dτ is the Lebesgue measure on e l [0, π 3 ], and let dν be the measure on G so that dν e l ŵ l dτ. Lemma 9. For any ϕ G, ψϕ dν G G ψ ϕ dν where ψ : G R is a continuous function so that ψ l ψ el is C.

20 288 G. DASKALOPOULOS & C. MESE Proof. By integration by parts, we get L π/3 ψϕ ( dν ŵ l (ψl ϕ l ψ l ϕ l dτ G l 0 ( L ŵ l ψ l (0 dϕ l dη (0 + ψ l(π/3 dϕ l dη (π/3 l L l On the other hand, by equation (. ( ŵ l ψ l (0 dϕ l dη (0 + ψ l(π/3 dϕ l dη (π/3 L l π/3 ( ŵ l ψ l (0 dϕ l dη (0 + ψ l(π/3 dϕ l dη (π/3 K ψ l (v k dϕ l ŵ l dη (v k 0 e l E l k 0 G ψ l ϕ l dτ ψ ϕ dτ. q.e.d. Definition. The eigenfunction and eigenvalue (of the Laplacian on G is a function ϕ G, not identically 0, and λ R so that for ϕ l ϕ el, ϕ l + λϕ l 0. Lemma 0. The eigenvalue λ of G is positive. Proof. By Lemma 9, λ ϕ 2 dν which immediately implies λ > 0. G G ϕϕ dν (ϕ 2 dν, q.e.d. Next, we show that a tangent map of a w-harmonic map at 0-simplex v defines an eigenfunction on the link of v. First, we review the notion of tangent map f of f. For a more detailed discussion for the special case of w so that w(f, see [DM]. The general case considered here follows by a trivial modification of the argument in [DM]. Let p X and d g be the distance function on N induced by its metric g. Define E(σ f 2 dµ, I(σ B σ(p B σ(p µ(σ (I(σσ 2, d 2 g(f, f(pds,

21 HARMONIC MAPS FROM A SIMPLICIAL COMPLEX 289 and lim σ 0 σe(σ I(σ where ds is the measure on B σ (p so that ds F Bσ(p w(fdθ where dθ is the Lebesgue measure on B σ (p. The above limit always exists as the quotient appearing on the right is a monotone non-decreasing function of σ. This limit is called the order of f at p. Let p X. The star St(p is the union of all simplices whose closure contains p. Let σ be sufficiently small so that B σ (p St(p, let B be B σ (p rescaled by a factor of σ so that it has radius, and S : B B σ (p be the natural identification map defined by the scaling. Define the σ-blow up map of f at p as the map σ f : B (N, g σ given by σ f(p f S(p and where (N, g σ is the manifold (N, g rescaled by a factor of µ(σ 2, i.e., g σ µ(σ 2 g. The map σ f converges locally uniformly in the pullback sense (see [KS2] and [DM] to a w-harmonic homogeneous map f (f,...,f m : B R m of order. By using polar coordinates (r, θ on each face F incident to p so that r measures the distance from the vertex p, we can write for α,...,m. f α (r, θ r f α (, θ Lemma. Let p be a vertex of X and f : B R n be a tangent map of f at p. The function ϕ : G R defined by letting G B and ϕ (θ f α (, θ is an eigenfunction of G with eigenvalue 2. Proof. Let v be a vertex in Lk(p. Let F l, l,...,l, be the faces of B which contain v. Without the loss of generality, we arrange the polar coordinates (r, θ on F l so that (, 0 corresponds to v. By Corollary 5, L w(f l fα l(, 0 0 θ l where f l f Fl, which is equivalent to ϕ α l ŵ l η (v 0 e l E where E is the set of edges of Lk(p containing v. Furthermore, f α w-harmonic function on each face and hence 0 f α f α 2 f α r 2 + r r + 2 f α r 2 θ 2 ( r 2 ϕ α + r 2 ϕ α + r 2 (ϕ α ( 2ϕ α + (ϕ α r 2 is a

22 290 G. DASKALOPOULOS & C. MESE which shows (ϕ α + 2ϕ α 0. q.e.d. By combining Lemma with Theorem 8, we obtain: Theorem 2. Suppose that X is a 2-complex such that every nonzero eigenvalue of the link of every vertex in X satisfies λ. (i If f : X (N, g is a w-harmonic map into a complete Riemannian manifold of nonpositive sectional curvature, then f is totally geodesic on each 2-simplex of X. In particular, this implies that if the sectional curvature of N is negative, f maps each 2-simplex into a geodesic. (ii If the eigenvalues satisfy the stronger condition λ > then f is a constant map. Proof. If is the order of f at a vertex p, then by Lemma, 2 is an eigenvalue. Hence by assumption and the first assertion of the theorem follows from Theorems 2 and 8. Let (r, θ be the polar coordinate where r measures the distance from a vertex p. Since f is totally geodesic, we have f(r, θ rf(, θ. If f is not identically constant then f is of order at p. Again, by Lemma, the second assertion of the theorem follows immediately. q.e.d. Finally, we need to characterize the assumption on the complex X for which the eigenvalue assumption of Theorem 2 is satisfied. For this, we need to divert our attention to the notion of the discrete Laplacian on a graph. Again, let G be a weighted graph as in the beginning of this section. Let (G {v,...,v K } denote the vertex set of G and let A(G R K denote the space of functions ϕ : (G R. Given v i, v G, we set { ŵs if v ŵ i i is adacent to v and v i v e s 0 otherwise and d i k ŵ i. Finally, the discrete Laplacian is defined to be the linear operator defined as ( disc ϕ(v i disc : A(G A(G K i, ŵ i d i (ϕ(v i ϕ(v. Notice that disc is self-adoint with respect to the inner product K ϕ, ψ d i ϕ(v i ψ(v i i

23 HARMONIC MAPS FROM A SIMPLICIAL COMPLEX 29 and disc ϕ, ψ K i i, K ŵ i (ϕ(v i ϕ(v (ψ(v i ψ(v. The next proposition relates the spectra of and disc. Proposition 3. A real number ( λ 9k 2 λπ, k, 2,..., is an eigenvalue of if and only if cos is an eigenvalue of disc. Proof. Assume λ > 0 is an eigenvalue of with eigenfunction ϕ, λ 9k 2. Then for each l,...,l, 0 x π 3, 3 ϕ l (x ϕ l(0sin( λ(π/3 x + ϕ l (π/3sin λx sin(. λπ/3 We claim that the balancing condition implies that ϕ (G : (G R is an eigenfunction of disc with eigenvalue cos λ. Indeed, fix v i (G. Then 0 ϕ l w l η (0 e l E i ϕ l (0 λ cos( λ(π/3 x + ϕl (π/3cos λx w l sin( x0 λπ/3 e l E i hence λ λπ sin( 3 λ λπ sin( 3 λ λπ sin( 3 λ λπ sin( 3 i, ( ( λπ ( π w l ϕ l (0cos + ϕ l 3 3 e l E i ( λπ cos w l ϕ(v i + ( π w l ϕ l 3 3 e l E i e l E i ( λπ cos d i ϕ(v i + ŵ i ϕ(v 3 i ( ( λπ cos d i ϕ(v i + ŵ i (ϕ(v ϕ(v i 3, ŵ i d i (ϕ(v i ϕ(v ( ( λπ cos ϕ(v i 3 and the proof of the claim is complete. By reversing our argument the converse is also true. q.e.d. Corollary 4. Any nonzero eigenvalue of is (> if and only if the first nonzero eigenvalue of disc is (> 2.

24 292 G. DASKALOPOULOS & C. MESE Proof. It follows ( immediately from Proposition 3 and the equivalence cos π 3 λ (> 2 if and only if λ (> for λ 0. q.e.d. Now let Σ n be a compact n-dimensional simplicial complex with an admissible weight c (see Definition 2. of [W] for a precise definition. Let w be the induced weights on the 2-skeleton Σ (2 X. By applying Theorem 2 and Corollary 4 to X with weights w, we immediately obtain as a corollary the main theorem of [W] (cf. [W] Theorem. Corollary 5. Let (Σ n, c be a compact simplicial complex with admissible weight. Assume that the first nonzero eigenvalue of the link of vertex is > 2. Then π (Σ Γ has property F; i.e., any isometric action of Γ on a complete, simply connected manifold N of nonpositive sectional curvature has a fixed point on N N N. [BS] [Ch] [C] [DM] [EF] References W. Ballman & J. Swiatkowski, On L 2 -cohomology and property (T for automorphism groups of polyhedral cell complexes, GAFA 7 ( , MR , Zbl J. Chen, On energy minimizing mappings between and into singular spaces, Duke Math. J. 79 ( , MR , Zbl K. Corlette, Archimedean superrigidity and hyperbolic geometry, Ann. of Math. 35 ( , MR 4796, Zbl G. Daskalopoulos & C. Mese. Harmonic maps from a 2-complex, Comm. Anal. Geom. 4 ( , MR J. Eells & B. Fuglede, Harmonic maps between Riemannian polyhedra, Cambridge Tracts in Mathematics, 42, Cambridge University Press, Cambridge, 200, MR , Zbl [G] H. Garland, p-adic curvature and the cohomology of discrete subgroups of p- adic groups, Ann. of Math. 97 ( , MR , Zbl [GT] D. Gilbarg & N. Trudinger. Elliptic Partial Differential Equations of Second Order, Springer-erlag, Berlin, 977, MR , Zbl [Gr] M. Gromov, Random walks in random groups, GAFA 3 ( , MR [GS] [IN] [J] M. Gromov & R. Schoen, Harmonic maps into singular spaces and p-adic superrigidity for lattices in groups of rank one, Publ. Math. IHES 76 ( , MR 25595, Zbl H. Izeki & S. Nayatani, Combinatorial harmonic maps and discretegroup actions on Hadamard spaces, Geom. Dedicata 4 ( , MR J. Jost, Nonpositive curvature: geometric and analytic aspects, Lectures in Mathematics, ETH Zürich, Birkhäuser erlag, 997, MR 45625, Zbl

25 HARMONIC MAPS FROM A SIMPLICIAL COMPLEX 293 [KS] [KS2] [KS3] [M] N. Korevaar & R. Schoen, Global existence theorems for harmonic maps to non-locally compact spaces, Comm. Anal. Geom. ( , MR , Zbl , Global existence theorem for harmonic maps to non-locally compact spaces, Comm. Anal. Geom. 5 ( , MR , Zbl , Global existence theorems for harmonic maps: finite rank spaces and an approach to rigidity for smooth actions, preprint. C. Mese, Regularity of harmonic maps from a flat complex, in ariational Problems in Riemannian Geometry: Bubbles, Scans and Geometric Flows (P. Baird, et al, eds. Birkhäuser, Basel, 2004, MR , Zbl [MSiY] N. Mok, Y.T. Siu, & S.K. Yang, Geometric superrigidity, Invent. Math. 3 ( , MR , Zbl [S] [Si] [W] [W2] [Z] R. Schoen, Analytic aspects of the harmonic map problem, Math. Sci. Res. Inst. Publ., 2, , Springer, Berlin, 984, MR , Zbl Y.T. Siu, The complex analyticity of harmonic maps and the strong rigidity of compact Kähler manifolds, Ann. of Math. 2 (980 73, MR , Zbl M.-T. Wang, A fixed point theorem of discrete group actions on Riemannian manifolds, J. Differential Geom. 50 ( , MR , Zbl , Generalized harmonic maps and representations of discrete groups, Comm. Anal. Geom. 8 ( , MR 77538, Zbl A. Zuk, La propriete (T de Kazhdan pour les groupes agissant sur les polyedres, C.R. Acad. Sci. Paris, 323, Serie I ( , MR , Zbl Brown University address: daskal@math.brown.edu Johns Hopkins University address: cmese@math.hu.edu