The outline of the paper is as follows. Section 2, for the sake of completeness, is an overview on Riemannian polyhedra, energy of maps, harmonic maps and morphisms, Brownian motion on Riemannian manifolds, martingales etc. In section 3, we prove that Brownian motions is stochastically continuous Markov processes and consequently it has a unique infinitesimal generator on some Banach space. The forth section is devoted to the behavior of Brownian motion under harmonic functions (in the sens of Korevaar-Schoen [18] ). The key of this characterization is to establish in some sense that the Brownian motion in Riemannian polyhedra has as an infinitesimal generator the ”Laplacian” (Theorem  4.1 ).

In the last section we show that harmonic maps, with target smooth Riemannian manifolds, in the sense of [11] , are exactly those which maps Brownian motion in Riemannian polyhedron into a martingale (see Theorem  5.1 ), while harmonic morphisms are exactly the maps which are Brownian preserving paths (see Theorem  5.2 ).

Acknowledgments. The authors would like to thank Professor A. Verjovsky to encourage them to investigate these problems and the ICTP Trieste for hospitality during this work.

2 Preliminaries.

This section is devoted to some basic notions and known results which will be used in the next sections.

2.1 Riemannian polyhedra.

2.1.1 Riemannian admissible complexes ([2] , [5] , [6] , [9] , [23] ).

Let $C$  be a locally finite simplicial complex, endowed with a piecewise smooth Riemannian metric $g$  ( i.e. $g$  is a family of smooth Riemannian metrics $g_S$  on simplexes $S$  of $C$  , such that the restriction $(g_S{)}_{|S^{\prime }}=g_{S^{\prime }}$  , for any simplexes $S^{\prime }$  and $S$  with $S^{\prime }\subset S$  ).

Let $C$  be a finite dimensional simplicial complex which is connected locally finite. A map $f$  from $[a,b]$  to $C$  is called a broken geodesic if there is a subdivision $a=t_0<t_1<...<t_{p+1}=b$  , such that $f\left(\right[t_i,t_{i+1}\left]\right)$  is contained in some cell and the restriction of $f$  to $[t_i,t_{i+1}]$  is a geodesic inside that cell. Then define the length of the broken geodesic map $f$  to be:

$$\mathcal{ℒ}\left(f\right)={\sum }_{i=0}^{p}d\left(f\right(t_i),f(t_{i+1}\left)\right).$$  The length inside each cell is measured with respect to its metric.

For every two points $x,y$  in $C$  , define $\tilde{d}(x,y)$  to be the lower bound of the lengths of broken geodesics from $x$  to $y$  . $\tilde{d}$  is a pseudo-distance.

If $C$  is connected and locally finite, then $(C,\tilde{d})$  is a length space and hence a geodesic space (i.e. a metric space where every two points are connected by a curve with length equal to the distance between them ) if complete.

We say that the complex $C$  is admissible, if it is dimensionally homogeneous and for every connected open subset $U$  of $C$  , the open set $U\backslash \{U\cap \{(n-2)-\text{skeleton}\left\}\right\}$  is connected, where $n$  is the dimension of $C$  (i.e. $C$  is $(n-1)$  -chainable).

We call an admissible connected locally finite simplicial complex, endowed with a piecewise smooth Riemannian metric, an admissible Riemannian complex.

2.1.2 Riemannian polyhedron [11] .

We mean by polyhedron a connected locally compact separable Hausdorff space $K$  for which there exists a simplicial complex $C$  and homeomorphism $\theta :C\to K$  . Any such pair $(C,\theta )$  is called a triangulation of $K$  . The complex $C$  is necessarily countable and locally finite (cf. [22] page 120) and the space $K$  is path connected and locally contractible.

The dimension of $K$  is by definition the dimension of $C$  and it is independent of the triangulation.

If $K$  is a polyhedron with specified triangulation $(C,\theta )$  , we shall speak of vertices, simplexes, $i$  -skeletons (the set of simplexes of dimensions lower or equal to $i$  ) of $K$  as the image under $\theta$  of vertices, simplexes, $i$  -skeletons of $C$  . Thus our simplexes become compact subsets of $K$  .

If for given triangulation $(C,\theta )$  of the polyhedron $K$  , the homeomorphism $\theta$  is locally bi-lipschitz then $K$  is said to be Lip polyhedron and $\theta$  Lip homeomorphism.

A null set in a Lip polyhedron $K$  is a set $Z\subset K$  such that $Z$  meets every maximal simplex $S$  , relative to a triangulation $(C,\theta )$  (hence any) in a set whose pre-image under $\theta$  has $n$  -dimensional Lebesgue measure $0$  , $n=dimS$  . Note that 'almost everywhere' (a.e.) means everywhere except in some null set.

A Riemannian polyhedron $K=(K,g)$  is defined as a Lip polyhedron $K$  with a specified triangulation $(C,\theta )$  such that $C$  is a simplicial complex endowed with a covariant bounded measurable Riemannian metric tensor $g$  , satisfying the ellipticity condition below.

In fact, suppose that $K$  has homogeneous dimension $n$  and choose a measurable Riemannian metric $g_S$  on the open euclidean $n$  -simplex ${\theta }^{-1}\left(S^o\right)$  of $C$  . In terms of euclidean coordinates $\{x_1,...,x_n\}$  of points $x={\theta }^{-1}\left(p\right)$  , $g_S$  thus assigns to almost every point $p\in S^o$  (or $x$  ), an $n\times n$  symmetric positive definite matrix $g_S=\left(g_{ij}^S\right(x){)}_{i,j=1,...,n}$  with measurable real entries and there is a constant ${\Lambda }_S>0$  such that (ellipticity condition):

$${\Lambda }_S^{-2}{\sum }_{i=0}^n({\xi }^i{)}^2\le {\sum }_{i,j}{g}_{ij}^S(x){\xi }^i{\xi }^j\le {\Lambda }_S^2{\sum }_{i=0}^n({\xi }^i{)}^2$$  for $a.e.$  $x\in {\theta }^{-1}\left(S^o\right)$  and every $\xi =({\xi }^1,...,{\xi }^n)\in {\mathbf{R}}^n$  . This condition amounts to the components of $g_S$  being bounded and it is independent not only of the choice of the euclidean frame on ${\theta }^{-1}\left(S^o\right)$  but also of the chosen triangulation.

For simplicity of statements we shall sometimes require that, relative to a fixed triangulation $(C,\theta )$  of Riemannian polyhedron $K$  (uniform ellipticity condition), $$\Lambda :=\text{sup}\{{\Lambda }_S:S\text{is simplex of}K\}<\infty .$$  A Riemannian polyhedron $K$  is said to be admissible if for a fixed triangulation $(C,\theta )$  (hence any) the Riemannian simplicial complex $C$  is admissible.

We underline that (for simplicity) the given definition of a Riemannian polyhedron $(K,g)$  contains already the fact (because of the definition above of the Riemannian admissible complex) that the metric $g$  is continuous relative to some (hence any) triangulation (i.e. for every maximal simplex $S$  the metric $g_S$  is continuous up to the boundary). This fact is sometimes omitted in the literature. The polyhedron is said to be simplexwise smooth if relative to some triangulation $(C,\theta )$  (and hence any), the complex $C$  is simplexwise smooth. Both continuity and simplexwise smoothness are preserved under subdivision.

2.2 Energy of maps.

The concept of energy in the case of a map of Riemannian domain into an arbitrary metric space $Y$  was defined and investigated by Korevaar and Schoen [18] . Later this concept was extended by Eells and Fuglede [11] to the case of a map from an admissible Riemannian polyhedron $K$  with simplexwise smooth Riemannian metric. Thus, the energy $E\left(\phi \right)$  of a map $\phi$  from $K$  to the space $Y$  is defined as the limit of suitable approximate energy expressed in terms of the distance function $d_Y$  of $Y$  .

It is shown in [11] that the maps $\phi :K\to Y$  of finite energy are precisely those quasicontinuous (i.e. has a continuous restriction to closed sets), whose complements have arbitrarily small capacity, (cf. [11] page 153) whose restriction to each top dimensional simplex of $K$  has finite energy in the sense of Korevaar-Schoen, and $E\left(\phi \right)$  is the sum of the energies of these restrictions.

Now, let $(K,g)$  be an admissible $m$  -dimensional Riemannian polyhedron with simplexwise smooth Riemannian metric. It is not required that $g$  is continuous across lower dimensional simplexes. The target $(Y,d_Y)$  is an arbitrary metric space.

Denote $L_{loc}^2(K,Y)$  the space of all ${\mu }_g$  -measurable ( ${\mu }_g$  the volume measure of $g$  ) maps $\phi :K\to Y$  having separable essential range and for which the map $d_Y\left(\phi \right(.),q)\in L_{loc}^2(K,{\mu }_g)$  (i.e. locally ${\mu }_g$  -squared integrable) for some point $q$  (hence by triangle inequality for any point). For $\phi ,\psi \in L_{loc}^2(K,Y)$  define their distance $\mathcal{D}(\phi ,\psi )$  by:

$${\mathcal{D}}^2(\phi ,\psi )={\int }_K{d}_Y^2\left(\phi \right(x),\psi (y\left)\right)d{\mu }_g\left(x\right).$$  Two maps $\phi ,\psi \in L_{loc}^2(K,Y)$  are said to be equivalent if $\mathcal{D}(\phi ,\psi )=0$  ,(i.e. $\phi \left(x\right)=\psi \left(x\right)$  ${\mu }_g$  -a.e.). If the space $K$  is compact then $\mathcal{D}(\phi ,\psi )<\infty$  and $\mathcal{D}$  is a metric on $L_{loc}^2(K,Y)=L^2(K,Y)$  which is complete if the space $Y$  is complete [18] .

The approximate energy density of the map $\phi \in L_{loc}^2(K,Y)$  is defined for $\epsilon >0$  by:

$$e_{\epsilon }\left(\phi \right)\left(x\right)={\int }_{B_K(x,\epsilon )}\frac{d_Y^2\left(\phi \right(x),\phi (x^{\prime }\left)\right)}{{\epsilon }^{m+2}}d{\mu }_g\left(x^{\prime }\right).$$  The function $e_{\epsilon }\left(\phi \right)\ge 0$  is locally ${\mu }_g$  -integrable.

The energy $E\left(\phi \right)$  of a map $\phi$  of class $L_{loc}^2(K,Y)$  is:

$$E\left(\phi \right)={sup}_{f\in {\mathcal{C}}_c(K,[0,1\left]\right)}\left({limsup}_{\epsilon \to 0}{\int }_{K}f{e}_{\epsilon }\right(\phi \left)d{\mu }_g\right),$$  where ${\mathcal{C}}_c(K,[0,1\left]\right)$  denotes the space of continuous functions from $K$  to the interval $[0,1]$  with compact support.

A map $\phi :K\to Y$  is said to be locally of finite energy, and we write $\phi \in W_{loc}^{1,2}(K,Y)$  , if $E\left({\phi }_{|U}\right)<\infty$  for every relatively compact domain $U\subset K$  , or equivalently if $K$  can be covered by domains $U\subset K$  such that $E\left({\phi }_{|U}\right)<\infty$  . For example (cf. [11] Lemma 4.4), every Lip continuous map $\phi :K\to Y$  is of class $W_{loc}^{1,2}(K,Y)$  . In the case when $K$  is compact $W_{loc}^{1,2}(K,Y)$  is denoted $W^{1,2}(K,Y)$  the space of all maps of finite energy.

$W_c^{1,2}(K,Y)$  denotes the linear subspace of $W^{1,2}(K,Y)$  consisting of all maps of finite energy of compact support in $K$  .

2.3 Harmonic maps and harmonic morphisms on Riemannian polyhedra [11] .

Let $(K,g)$  be an arbitrary admissible Riemannian polyhedron ( $g$  just bounded measurable with local elliptic bounds), $dimK=m$  and $(Y,d_Y)$  a metric space .

A continuous map $\phi :K\to Y$  of class $W_{loc}^{1,2}(K,Y)$  is said to be harmonic if it is bi-locally E-minimizing, i.e. $K$  can be covered by relatively compact subdomains $U$  for each of which there is an open set $V\supset \phi \left(U\right)$  in $Y$  such that $$E\left({\phi }_{|U}\right)\le E\left({\psi }_{|U}\right)$$  for every continuous map $\psi \in W_{loc}^{1,2}(K,Y)$  , with $\psi \left(U\right)\subset V$  and $\psi =\phi$  in $K\backslash U$  .

Let $(N,h)$  denote a smooth Riemannian manifold without boundary, $di{m}_{\mathbf{R}}N=n$  and ${\Gamma }_{\alpha \beta }^k$  the Christoffel symbols on $N$  . By a weakly harmonic map $\phi :K\to N$  we mean a quasicontinuous map (a map which is continuous on the complement of open sets of arbitrarily small capacity; in the case of the Riemannian polyhedron $K$  it is just the complement of open subsets of the $(m-2)$  -skeleton of $K$  ) of class $W_{loc}^{1,2}(K,N)$  with the following property:

For any chart $\eta :V\to {\mathbf{R}}^n$  on N and any quasiopen set $U\subset {\phi }^{-1}\left(V\right)$  of compact closure in $K$  , the equation $${\int }_U\langle \nabla \lambda ,\nabla {\phi }^k\rangle d{\mu }_g={\int }_U\lambda ({\Gamma }_{\alpha \beta }^k\circ \phi )\langle \nabla {\phi }^{\alpha },\nabla {\phi }^{\beta }\rangle d{\mu }_g,$$  holds for every $k=1,...,n$  and every bounded function $\lambda \in W_0^{1,2}\left(U\right)$  .

Denote now, by $K$  and $Y$  two Riemannian polyhedra (or any harmonic spaces in the sense of Brelot; see Chapter 2, [11] ). A continuous map $\phi :K\to Y$  is a harmonic morphism if, for every open set $V\subset Y$  and for every harmonic function $v$  on $V$  , $v\circ \phi$  is harmonic on ${\phi }^{-1}\left(V\right)$  .

2.4 Brownian motion in Riemannian manifolds [8] , [12] , [24] .

Consider $(\Omega ,\mathcal{A},P)$  a probability space, $(E,\varepsilon )$  a measurable space, and $I$  an ordered set. By a stochastic process on $(\Omega ,\mathcal{A},P)$  with values on $(E,\varepsilon )$  and $I$  as time interval, we mean a map (see [12] , or [24] , or [8] ):

$$\begin{array}{ccccc}X:& I\times \Omega & \to & E& \end{array}$$

$$\begin{array}{ccccc}& (t,\omega )& \mapsto & X(t,\omega ),& \text{such that fot each}t\in I,\end{array}$$

$X_t:\omega \in \Omega \mapsto X(t,\omega )\in E\text{is measurable from}(\Omega ,\mathcal{A})\text{to}(E,\varepsilon ).$  A family $\mathcal{ℱ}=({\mathcal{ℱ}}_t{)}_{t\in I}$  of $\sigma$  -subalgebras of $\mathcal{A}$  , such that ${\mathcal{ℱ}}_s\subset {\mathcal{ℱ}}_t$  , for all $s$  , $t$  with $s<t$  , is called a filtration on $(\Omega ,\mathcal{A},P)$  with $I$  time interval.

Given a filtration $\mathcal{ℱ}=({\mathcal{ℱ}}_t{)}_{t\in I}$  , a process $X$  , admitting as time interval a part $J$  of $I$  , is said to be adapted to $\mathcal{ℱ}$  , if for every $t\in J$  , $X_t$  is ${\mathcal{ℱ}}_t$  -measurable.

A real valued process $X$  is said to be a submartingale, with respect to a filtration ${\mathcal{ℱ}}_t$  fixed on $(\Omega ,\mathcal{A},P)$  , if it has the following properties : a) $X$  is adapted; b) each random variable $X_t$  is integrable; c) for each pair of real numbers $s$  , $t$  , $s<t$  , and every $A\in {\mathcal{ℱ}}_s$  we have:

$${\int }_A{X}_{s}dP\le {\int }_A{X}_{t}dP.$$  When the equality holds we say that $X$  is a martingale.

A real-valued process $X$  is said to be a continuous local martingale if and only if it is a continuous ( with respect to the time variable ) adapted process $X$  such that each $X_{t\wedge T_n}{\chi }_{\{T_n>0\}}$  is a martingale, where $\chi$  is the characteristic function and $T_n$  is the stopping time: $inf\{t:|X_t|\ge n\}$  .

A semimartingale is the sum of a continuous local martingale and a process with finite variation. If the process of the finite variation is an increasing one, the semimartingale is called a local submartingale.

Let $M$  be a manifold with a connection ${}^M\nabla$  , and $X$  a $M$  -valued process. Following Schwartz characterization (see [21] ), a ${}^M\nabla$  -martingale tester, $(U_1,U_2,U_3,f)$  will consist of:

• ∙ open sets $U_1$  , $U_2$  , $U_3$  in $M$  with $\overline{U_1}\subset U_2\subset \overline{U_2}\subset U_3,$ 
• ∙ a convex function $f:U_3\to \mathbf{R}.$ 

The process $X=(X_t,{\mathcal{ℱ}}_t)$  is said to be a ${}^M\nabla$  -martingale, if it is a continuous semimartingale on $M$  (i.e. $\forall f\in {\mathcal{C}}^2\left(M\right)$  , $f\circ X$  is a real valued semimartingale), and for all ${}^M\nabla$  -martingale tester $(U_1,U_2,U_3,f)$  , the process $Y=(Y_t,{\mathcal{ℱ}}_t)$  , $Y_t={\int }_0^t{1}_F\left(s\right)d(f\circ X_s),$  is a local submartingale. $F$  denote the previsible set ${\cup }_{i=1}^{\infty }({\sigma }_i,{\tau }_i]$  where ${\sigma }_i,{\tau }_i,i\ge 0$  is the collection of stopping-times, associated to the process $X$  and any ${}^M\nabla$  -martingale tester, defined by:

• ∙ ${\sigma }_0=0,{\tau }_0=0$ 
• ∙ ${\sigma }_i=inf\{t>{\tau }_{i-1}:X_t\in U_1\};{\tau }_i=inf\{t>{\sigma }_i:X_t/\in \overline{U_2}\},i\ge 1.$ 

Suppose $M$  is a Riemannian manifold with Levi-Civita connection ${}^M\nabla$  . A Brownian motion is characterized as a diffusion $B=(B_t,{\mathcal{ℱ}}_t)$  with generator $\frac{1}{2}\Delta$  ; in other words, for all $f:M\to \mathbf{R}$  , the process $C^f$  , where $C_t^f=f\left(B_t\right)-f\left(B_0\right)-\frac{1}{2}{\int }_0^t{1}_F\left(s\right)\Delta \left(f\right)\left(B_s\right)ds$  , is a local martingale.

3 Brownian motions in admissible Riemannian polyhedra.

The Brownian motion in piecewise smooth Riemannian complexes, was obtained in [4] , as a weak limit of isotropic processes. This construction holds obviously also for the piecewise smooth Riemannian polyhedra.

Let us recall some essential facts about this construction. In [4] , the second author define a process: $Y^{\eta }=(\Omega ,{\mathcal{ℱ}}_t^0,Y_t^{\eta },{\theta }_t,P)$  , for $\eta \in (0,1]$  , in the following way:

$$Y_t^{\eta }\left(\omega \right)=\{\begin{array}{ccc}{\Upsilon }_{\eta Z_i\left(\omega \right)}(\frac{t}{{\eta }^2}-{\tau }_i(\omega \left)\right)& \text{if}& {\tau }_i\left(\omega \right)\le \frac{t}{{\eta }^2}\le {\tau }_{i+1}\left(\omega \right)\\ D& \text{if}& \xi \left(\omega \right)\le \frac{t}{{\eta }^2},\\ \end{array}$$  where ${\tau }_i$  are the stopping times such that, for all $i\in \mathbf{N}$  , the real random variable $({\tau }_i-{\tau }_{i+1})$  is exponentially distributed and ${\tau }_0=0$  ; ${\Upsilon }_{\eta }$  is the generalized geodesic flow (see [2] ); $D$  is the one point compactification of $K$  (because $K$  is semicompact) and $\xi$  is the life time of $Y_t^{\eta }$  ; $Z$  is a unit tangent vector randomly chosen in the link of the point ${\Upsilon }_{\eta Z_{i-1}\left(\omega \right)}\left({\tau }_i\right(\omega \left)\right)$  with respect to the volume measure (link is viewed as a spherical Riemannian polyhedron), where $Z_0\left(\omega \right)$  is also a unit tangent vector randomly chosen in the link of the starting point.

In [4] it is also proved that $Y_t^{\eta }$  (for $\eta \in (0,1]$  ) is a continuous Markov process, for each $\eta >0$  , $Y^{\eta }$  generate a measure ${\mu }_{\eta }$  on the space $C({\mathbf{R}}^{+},K):=\{f:{\mathbf{R}}^{+}\to K,fcontinuous\}$  and ${\mu }_{\eta }$  has a subsequence which converges to a measure $W$  on $C({\mathbf{R}}^{+},K)$  , called Wiener measure. This Wiener measure generate Brownian motions in the Riemannian polyhedra, such that the transition functions of the generated Brownian motions are just the projections of the Wiener measure on $K$  (see for details [4] ).

Proposition 3.1 The transition function $W_t$  associate to the K-valued Brownian motion, is stochastically continuous.

Proof: It is enough to show that for $r>0$  :

$${\text{lim}}_{t\to 0}{W}_t(p,B(p,r\left)\right)=1.$$  By construction: $W_t(p,B(p,r\left)\right)={\text{lim}}_{\eta \to 0}P\{Y_t^{\eta }\in B(p,r)/Y_0^{\eta }=p\}.$  Following [4] (Section 3), we have:

$$\begin{array}{c}P\{Y_t^{\eta }\in CB(p,r)/Y_0^{\eta }=p\}\le e^{-\frac{t}{{\eta }^2}},\end{array}$$

(1)

where $CB(p,r)$  denote the complementary of the ball $B(p,r)$  in $K$  .

On the other hand, $W_t(p,B(p,r\left)\right)=1-W_t(p,CB(p,r\left)\right)$  .

But, $${\text{lim}}_{t\to 0}{W}_t(p,B(p,r\left)\right)=1-{\text{lim}}_{t\to 0}{W}_t(p,CB(p,r\left)\right),$$  which is equal to:

$$1-{\text{lim}}_{t\to 0}{\text{lim}}_{\eta \to 0}P\{Y_t^{\eta }\in CB(p,r)/Y_0^{\eta }=p\}.$$  Using ( 1 ), the proof is achieved. $\square$ 

Proposition 3.2 The Brownian motion $(B_t{)}_{t\ge 0}$  on Riemannian polyhedra has an infinitesimal generator $L$  defined on a Banach subspace ${\mathbf{D}}_L$  which is dense in the space $C_c\left(K\right)$  (i.e.

for every $f\in {\mathbf{D}}_L$  , $Lf:={lim}_{t\to 0}\frac{E\left[f\right(B_t\left)\right]-f\left(B_0\right)}{t}$  uniformly) .

Proof: By Proposition  3.1 and Dynkin's result (cf. [10] Theorem 2.3), the existence of $L$  is completely insured. $\square$ 

Remark 3.3 The space $D_L$  and the operator $L$  will be discussed in more details in the next section.

4 Brownian motion and Harmonic functions.

Let $(K,g)$  be an admissible Riemannian polyhedron, of dimension $n$  , endowed with continuous simplexwise smooth metrics.

For $p\in K$  , let $B=(\Omega ,{\mathcal{ℱ}}_t^0,B_t^p,{\theta }_t,P)$  be the $K$  -valued Brownian motion starting at the point $p$  [4] .

Theorem 4.1 Let $f\in W_{loc}^{1,2}\left(K\right)$  , i.e. $K$  is covered by relatively compact subdomains $U$  such that $f$  is of finite energy on each $U$  . Moreover, $f$  is considered at least of class ${\mathcal{C}}^2$  on the interior of each $(n-simplex)$  and the interior of each $\left(\right(n-1)-simplex)$  .

Denote by ${\tau }_u:=inf\{t>0/B_t/\in U\}$  the first exit time of the $B_t$  from $U$  . Then we have:

$\frac{1}{2}\Delta f=\frac{\partial }{\partial t}E\left[f\right(B_{t\wedge {\tau }_u}\left)\right],\text{On the}U\backslash \left(\right(n-2)-skeleton)\text{.}$  where $E\left[f\right(B_{t\wedge {\tau }_u}\left)\right]$  is the expectation with respect to $B_{t\wedge {\tau }_u}$  .

Proof: There are two cases to investigate:

• Case 1:
  If $p$  is in the topological interior of some $n$  -dimensional simplex, using [12] or [14] , then ( 4.1 ) clearly holds in this case.
• Case 2:
  Let $p$  be in the $\left(\right(n-1)-skeleton)\backslash \left(\right(n-2)-skeleton)$  . Suppose that $p$  is in the topological interior of the $\left(\right(n-1)-simplex)$  , $S_{n-1}$  . Let $S_n^1,S_n^2,...,S_n^k$  , $k\ge 2$  , denote the $(n-simplexes)$  adjacent to $S_{n-1}$  .

  Consider ${}^0{S}_{n-1}$  as a submanifold of each $S_n^l,l=1,...,k$  ( ${}^{0}S$  denote the interior of $S$  ).

  Denote by ${\perp }_p^l{(}^0{S}_{n-1})$  the orthogonal complement of the tangent space of ${}^0{S}_{n-1}$  at $p$  , considered as a subspace of the tangent space ${\top }_p\left(S_n^l\right)$  (the space of all tangent vectors at $p$  pointing to $S_n^l$  ). Then the normal bundle of ${}^0{S}_{n-1}$  in $S_n^l$  is $${\perp }^l{(}^0{S}_{n-1})=\{(p,v)/p{\in }^0{S}_{n-1},v\in {\perp }_p^l{(}^0{S}_{n-1}\left)\right\}.$$  The exponential map of the normal bundle ${\perp }^l{(}^0{S}_{n-1})$  is defined:

  $$\begin{array}{cccc}ex{p}_{{\perp }^l{(}^0{S}_{n-1}}):& {\perp }^l{(}^0{S}_{n-1})& \to & S_n^l\end{array}$$

  $$\begin{array}{cccc}& (p,v)& \mapsto & ex{p}_p\left(v\right).\end{array}$$

  By O'Neill's or Gray's result (see [20] , p.199-200, or [15] , p.16), there exist a $\eta >0$  such that the normal neighborhood $${V_{\eta }}^0{S}_{n-1}:=\left\{\right(p,v)\in {\perp }^l{(}^0{S}_{n-1})/\left|\right|v\left|\right|<\eta \}$$  of ${}^0{S}_{n-1}\subset {\perp }^l{(}^0{S}_{n-1})$  is mapped diffeomorphically into a tubular neighborhood $U_{\eta }^l=ex{p}_{{\perp }^l{(}^0{S}_{n-1}}\left)\right({V_{\eta }}^0{S}_{n-1})$  of ${}^0{S}_{n-1}\subset S_n^l$  .

  For $p{\in }^0{S}_{n-1}$  let $B(p,\varepsilon )$  be a open ball centered in $p$  and ray $\varepsilon$  .

  Using Greene-Wu's result (cf. [16] ), in the rest of the paper, we choose a coordinate system $(x^1,x^2,...,x^{n-1})$  in $B(p,\varepsilon ){\cap }^0{S}_{n-1}$  , such that each $x^i,i=1,...,n-1$  , is a smooth harmonic function (because $K$  is also simplexwise smooth).

  Now, we take $\eta$  small enough, such that in the open tubular neighborhood $B(p,\varepsilon )\cap U_{\eta }^l$  , $l=1,...,k$  , of $p$  in $S_n^l$  , we associate to each point $q\in B(p,\varepsilon )\cap U_{\eta }^l$  , $l=1,...,k$  it's Fermi coordinates with respect to the submanifold ${}^0{S}_{n-1}$  . In other words, each point $q\in B(p,\varepsilon )\cap U_{\eta }^l$  , $l=1,...,k$  , is characterized by a pair $\left(\pi \right(q),r_l)$  , where $\pi \left(q\right)$  is the unique orthogonal projection of $q$  on $B(p,\varepsilon ){\cap }^0{S}_{n-1}$  and $r_l=d(q,\pi (q\left)\right)$  viewed as a real coordinate on the unique orthogonal geodesic to ${}^0{S}_{n-1}$  , joining $q$  and $\pi \left(q\right)$  .

  By this choice of coordinates, the system of coordinates on $B(p,\varepsilon )\cap U_{\eta }^l$  , $l=1,...,k$  , becomes harmonic.

  For any $l=1,...,k$  , in the neighborhood $B(p,\varepsilon )\cap U_{\eta }^l$  the Laplace -Beltrami operator has the form:

  $$\begin{array}{c}\begin{array}{ccc}{\Delta }_l& =& {(}^l{g}^{r_l{r}_l}){\partial }_{r_l}{\partial }_{r_l}f+{\sum }_{i,j=1}^{n-1}{(}^l{g}^{ij}){\partial }_i{\partial }_{j}f+\\ & & {\sum }_{i=1}^{n-1}{(}^l{g}^{i{r}_l}){\partial }_i{\partial }_{r_l}f+{\sum }_{j=1}^{n-1}{(}^l{g}^{r_{l}j}){\partial }_{r_l}{\partial }_{j}f+\\ & & \left({\Delta }_l{r}_l\right){\partial }_{r_l}f+{\sum }_{i=1}^{n-1}\left({\Delta }_l{x}_i\right){\partial }_{i}f\\ \end{array}\end{array}$$

  (2)

  But our system coordinates is a Fermi one, $g^{i{r}_l}$  , $g^{r_{l}j}$  are zero and moreover the coordinates are also harmonic the terms ${\Delta }_l{r}_l$  and ${\Delta }_l{x}_i$  are zero. So the Laplace-Beltrami operator reduce to :

  $$\begin{array}{c}{\Delta }_l={\partial }_{r_l}{\partial }_{r_l}f+{\sum }_{i,j=1}^{n-1}{(}^l{g}^{ij}){\partial }_i{\partial }_{j}f.\end{array}$$

  (3)

  The family ${(}^{l}g{)}_l$  of the metrics is supposed to be continuous, thus the restrictions of each ${}^{l}g,l=1,2,...,k$  to $S_{n-1}$  are equal. Consequently the term ${\sum }_{i,j=1}^{n-1}{(}^l{g}^{ij}){\partial }_i{\partial }_j$  is independent of the choice of $l$  and the global Laplacian in $K$  , at a point $p$  in the chosen map can be written:

  $$\begin{array}{c}\Delta =\frac{1}{k}{\sum }_{l=1}^k{\Delta }_l,\end{array}$$

  (4)

  which is equal to:

  $$\begin{array}{c}\Delta =\frac{1}{k}{\sum }_{l=1}^k{\partial }_{r_l^2}^2+{\sum }_{i,j=1}^{n-1}{(}^1{g}^{ij}\left){\partial }_i{\partial }_j\right).\end{array}$$

  (5)

  Now for a point $p_1\in B(p,\varepsilon )\cap U_{\eta }^l,$  the second order Taylors development of $f$  , at the point $p$  in the fixed map (see above), has the form:

  $$\begin{array}{c}\begin{array}{ccc}f\left(p_1\right)& =& f\left(p\right)+{\partial }_{r_l}f\left(p\right){\xi }_{r_l}\left(p_1\right)+\frac{1}{2}{\partial }_{r_l}{\partial }_{r_l}f\left(p\right){\xi }_{r_l}^2\left(p_1\right)+\\ & & \frac{1}{2}{\sum }_{j=1}^{n-1}{\partial }_{r_l}{\partial }_{j}f\left(p\right){\xi }_{r_l}\left(p_1\right){\xi }_j\left(p_1\right)+\frac{1}{2}{\sum }_{i=1}^{n-1}{\partial }_i{\partial }_{r_l}f\left(p\right){\xi }_i\left(p_1\right){\xi }_{r_l}\left(p_1\right)+\\ & & {\sum }_{i=1}^{n-1}{\partial }_{i}f\left(p\right){\xi }_i\left(p_1\right)+{\sum }_{i,j=1}^{n-1}{\partial }_i{\partial }_{j}f\left(p\right){\xi }_i\left(p_1\right){\xi }_j\left(p_1\right)+o\left(\varepsilon \right).\\ \end{array}\end{array}$$

  (6)

  If we denote $\pi \left(p_1\right)$  the geodesic orthogonal projection of $p_1$  on ${}^0{S}_{n-1}$  , then the $\xi$  's which appear in ( 6 ) are completely defined as follows:

  $${\xi }_{r_l}=d(p_1,\pi (p_1\left)\right)$$  and $({\xi }_1,...,{\xi }_{n-1})$  is the tangent vector to the minimal curve joining $p$  and $\pi \left(p_1\right)$  in ${}^0{S}_{n-1}$  with respect to the metric $g_{{|}^0{S}_{n-1}}$  .

  Actually, for $t\ll \eta \wedge \varepsilon$  (then $B_{t\wedge {\tau }_u}\in B(p,\varepsilon )\cap S_n^l$  ), the relation ( 6 ) becomes:

  $$\begin{array}{c}\begin{array}{cc}f\left(B_{t\wedge {\tau }_u}\right)=& f\left(p\right)+{\partial }_{r_l}f\left(p\right)d(B_{t\wedge {\tau }_u},\pi (B_{t\wedge {\tau }_u}\left)\right)+\\ & \frac{1}{2}{\partial }_{r_l}{\partial }_{r_l}f\left(p\right)d^2(B_{t\wedge {\tau }_u},\pi (B_{t\wedge {\tau }_u}\left)\right)+\\ & {\sum }_{i=1}^{n-1}{\partial }_{i}f\left(p\right){\xi }_i\left(B_{t\wedge {\tau }_u}\right)+\\ & \frac{1}{2}{\sum }_{i,j=1}^{n-1}{\partial }_i{\partial }_{j}f\left(p\right){\xi }_i\left(B_{t\wedge {\tau }_u}\right){\xi }_j\left(B_{t\wedge {\tau }_u}\right)+o\left(t\right).\\ \end{array}\end{array}$$

  (7)

  Then,

  $$\begin{array}{c}\begin{array}{cc}{E}^p\left[f\right(B_{t\wedge {\tau }_u}\left)\right]=& f\left(p\right)+\frac{1}{k}{\sum }_{l=1}^k\left\{{\partial }_{r_l}f\right(p\left)E^p\right[d(B_{t\wedge {\tau }_u},\pi (B_{t\wedge {\tau }_u}\left)\right)]+\\ & \frac{1}{2}{\partial }_{r_l}{\partial }_{r_l}f\left(p\right)E^p\left[d^2\right(B_{t\wedge {\tau }_u},\pi \left(B_{t\wedge {\tau }_u}\right)\left)\right]\}+\\ & {\sum }_{i=1}^{n-1}{\partial }_{i}f\left(p\right)E^p\left[{\xi }_i\right(B_{t\wedge {\tau }_u}\left)\right]+\\ & \frac{1}{2}{\sum }_{i,j=1}^{n-1}{\partial }_i{\partial }_{j}f\left(p\right)E^p\left[{\xi }_i\right(B_{t\wedge {\tau }_u}\left){\xi }_j\right(B_{t\wedge {\tau }_u}\left)\right]+o\left(t\right).\\ \end{array}\end{array}$$

  (8)

  As proved in ([4] ), the $K$  -valued Brownian motion $(B_t{)}_{t\ge 0}$  , behaves, inside every $n$  -simplex as a standard Brownian motion with values in Riemannian $n$  -dimensional manifolds, hits almost surely the $\left(\right(n-1)-skeleton)\backslash \left(\right(n-2)-skeleton)$  , and it continues it's motion by choosing with equal probabilities a maximal face.

  Now using Baxter-Cachon's results (see [3] ) and the fact that $t\ll \eta \wedge \varepsilon$  , we have:

  $$\frac{1}{k}{\sum }_{l=1}^k{\partial }_{r_l}f\left(p\right)E^p\left[d\right(B_{t\wedge {\tau }_u},\pi \left(B_{t\wedge {\tau }_u}\right)\left)\right]=0,$$  $$\frac{1}{2k}{\sum }_{l=1}^k{\partial }_{r_l}{\partial }_{r_l}f\left(p\right)E^p\left[d^2\right(B_{t\wedge {\tau }_u},\pi \left(B_{t\wedge {\tau }_u}\right)\left)\right]=\frac{t}{2k}{\sum }_{l=1}^k{\Delta }_{r_l}f\left(p\right),$$  where ${\Delta }_{r_l}f={\partial }_{r_l}{\partial }_{r_l}f$  .

  By [14] , (32.5, p.80), we obtain :

  $$E^p\left[{\xi }_i\right(B_{t\wedge {\tau }_u}\left){\xi }_j\right(B_{t\wedge {\tau }_u}\left)\right]={(}^l{g}^{ij})t,$$  and for the last expectation, if $X_0^t$  is a one dimensional real standard Brownian motion with starting point zero, we have:

  $$E^p\left[{\xi }_i\right(B_{t\wedge {\tau }_u}\left)\right]=\sqrt{l{g}^{ii}}{E}^p\left[X_0^t\right]=0.$$  So we deduce that:

  $$\begin{array}{c}\begin{array}{cc}{E}^p\left[f\right(B_{t\wedge {\tau }_u}\left)\right]=& f\left(p\right)+\frac{t}{2k}{\sum }_{l=1}^k{\Delta }_{r_l}f+\frac{t}{2}{{\sum }_{i,j=1}^{n-1}}^1{g}^{ij}{\partial }_i{\partial }_{j}f\left(p\right)=\\ & f\left(p\right)+\frac{t}{2}\Delta f+o\left(t\right).\\ \end{array}\end{array}$$

  (9)

  which ends the proof of the theorem.

$\square$ 

Lemma 4.2 Using the same notation as in the Proposition  3.2 , the space ${\mathbf{D}}_L$  contains $W_c^{1,2}\left(K\right)\cap \{$  function of class $C^2$  in the interior of the $n-$  simplexes and the $(n-1)-$  simplexes $\}$  .

Proof: All the functions considered are supposed to be at least of class $C^2$  in the interior of each $n-$  simplex and each $(n-1)$  simplex.

By Theorem  4.1 , for $f\in W_c^{1,2}\left(K\right)$  , we have, for every $\psi \in W_c^{1,2}\left(K\right)$  , $$\frac{1}{2}{\int }_{K\backslash \left(\right(n-2)-skeleton)}\psi \Delta fd{\mu }_g={\int }_{K\backslash \left(\right(n-2)-skeleton)}\psi Lfd{\mu }_g.$$  Consider now an operator $\tilde{L}$  defined weakly on the space $W_c^{1,2}\left(K\right)$  by:

$$-\frac{1}{2}{\int }_K\langle \nabla \psi ,\nabla f\rangle d{\mu }_g={\int }_K\psi \tilde{L}fd{\mu }_g.$$  It is clear that:

$${\int }_{K\backslash \left(\right(n-2)-skeleton)}\psi Lfd{\mu }_g={\int }_{K\backslash \left(\right(n-2)-skeleton)}\psi \tilde{L}fd{\mu }_g.$$  On the other hand, the Brownian motion almost surly never hits the $(n-2)-$  skeleton, so $\tilde{L}$  is also an infinitesimal generator associate to the transition probability $W_t$  of the Brownian motion.

But by Proposition 3.3, $W_t$  is stochastically continuous and the space $W_c^{1,2}\left(K\right)\cap \{$  function of class $C^2$  in the interior of the $n-$  simplexes and the $(n-1)-$  simplexes $\}$  is dence in the space $C_c\left(K\right)$  so its infinitesimal generator is uniquely determined (cf. [10] Theorem 2.4) we infer that $\tilde{L}$  is equal to $L$  on the space $W_c^{1,2}\left(K\right)$  .

$\square$ 

Theorem 4.3 Let $(K,g)$  be an admissible Riemannian polyhedra endowed with a picewise smooth Riemannian metrics and $f\in W_{loc}^{1,2}\left(K\right).$  Let $(B_t{)}_{t\ge 0}$  be a $K$  -valued Brownian motion and let $U$  an open sets of $K$  taken as in the hypothesis of the Theorem  4.1 . Then, for any $p\in K\backslash \left(\right(n-2)-skeleton)$  the process $$C_t^{f\left(p\right)}=f\left(B_t^p\right)-f\left(p\right)-{\int }_0^{t\wedge {\tau }_U}L\left(f\right(B_s^p\left)\right)ds$$  is a local martingale, where ${\tau }_u:=inf\{t>0/B_t/\in U\}$  the first exit time of $B_t$  from $U$  .

Proof: By construction, the Brownian motion $(B_t{)}_{t\ge 0}$  almost surely never hits the $(n-2)$  -skeleton.

For any $p\in U_1\backslash \left(\right(n-2)-skeleton)$  , consider the process:

$${\tilde{C}}_t^{f\left(p\right)}={\chi }_{\{B_t^p/\in ((n-2)-skeleton\left)\right\}}f\left(B_t^p\right)-f\left(p\right)-{\int }_0^{t\wedge {\tau }_U}{\chi }_{\{B_s^p/\in ((n-2)-skeleton\left)\right\}}L\left(f\right(B_s^p\left)\right)ds,$$  where $\chi$  denote the characteristic function.

By Theorem  4.1 , $\forall p\in U_1\backslash \left(\right(n-2)-skeleton),$  $${\chi }_{\{B_s^p/\in ((n-2)-skeleton\left)\right\}}L\left(f\right(B_s^p\left)\right)={\chi }_{\{B_s^p/\in ((n-2)-skeleton\left)\right\}}\frac{1}{2}\Delta f\left(B_s^p\right).$$  Taking the expectation, we obtain: $$E\left[{\tilde{C}}_t^{f\left(p\right)}\right]=E\left[f\right(B_t^p\left)\right]-E\left[f\right(p\left)\right]-\frac{1}{2}E[{\int }_0^{t\wedge {\tau }_U}{\chi }_{\{B_s^p/\in ((n-2)-skeleton\left)\right\}}\Delta f(B_s^p\left)ds\right].$$  Moreover, $$\frac{\partial }{\partial t}E\left[{\tilde{C}}_t^{f\left(p\right)}\right]=\frac{\partial }{\partial t}E\left[f\right(B_t^p\left)\right]-\frac{1}{2}\Delta f\left(p\right).$$  Then using Theorem  4.1 we obtain, $\frac{\partial }{\partial t}E\left[{\tilde{C}}_t^{f\left(p\right)}\right]=0,\forall p\in U_1\backslash \left(\right(n-2)-skeleton).$  Thus, ${\tilde{C}}_t^{f\left(p\right)}$  is a local martingale, but ${\tilde{C}}_t$  is equal almost surely to the process $C_t$  so the last process is also a local martingale. $\square$ 

Corollary 4.4 Let $(K,g)$  be an admissible Riemannian polyhedra endowed with a picewise smooth Riemannian metrics and $f\in W_{loc}^{1,2}\left(K\right)$  i.e. $K$  is covered by relatively compact subdomains $U$  such that every $f$  is of finite energy on each $U$  . Moreover, $f$  is considered at least of class ${\mathcal{C}}^2$  on the interior of each $n$  -simplex and the interior of each $(n-1)$  -simplex. Denote by ${\tau }_u:=inf\{t>0/B_t/\in U\}$  the first exit time of $B_t$  from $U$  .

Then $f$  is harmonic if and only if, for any $p\in U\backslash \left(\right(n-2)-skeleton)$  , $f\left(B_{t\wedge {\tau }_u}^p\right)$  is a local martingale ( $(B_t^p{)}_t$  is a $K$  -valued Brownian motion).

Proof: By Theorem  4.3 , the processes:

$$C_t^{f\left(p\right)}=f\left(B_t^p\right)-f\left(p\right)-{\int }_0^t{1}_F\left(s\right)L\left(f\right(B_s^p\left)\right)ds$$  and $${\tilde{C}}_t={\chi }_{\{B_t^p/\in ((n-2)-skeleton\left)\right\}}f\left(B_t^p\right)-f\left(p\right)-{\int }_0^t{\chi }_{\{B_s^p/\in ((n-2)-skeleton\left)1_F\right(s\left)L\right(f\left(B_s^p\right)\left)ds\right\}}$$  are both local martingale for every $p\in U\backslash \left(\right(n-2)-skeleton)$  , where $U$  is taken as in the hypothesis of the theorem.

Suppose that $f$  is harmonic, then:

$${\chi }_{\{B_t^p/\in ((n-2)-skeleton\left)\right\}}f\left(B_{t\wedge {\tau }_u}^p\right)={\tilde{C}}_t^p+f\left(p\right),$$  for every $p\in U\backslash \left(\right(n-2)-skeleton)$  .

So the process ${\chi }_{\{B_t^p/\in ((n-2)-skeleton\left)\right\}}f\left(B_{t\wedge {\tau }_u}^p\right)$  is local martingale, for every $p\in U\backslash \left(\right(n-2)-skeleton)$  . But this last process is almost surely equal to $f\left(B_t^p\right)$  so the process $f\left(B_t^p\right)$  is also a locale martingale.

Conversely, suppose that for every $p\in U\backslash \left(\right(n-2)-skeleton)$  , $f\left(B_{t\wedge {\tau }_u}^p\right)$  is a local martingale. Then by classical theory, this implies that $f$  is harmonic on each $U\backslash \left(\right(n-2)-skeleton)$  , so is an $E$  -minimizer on each $U\backslash \left(\right(n-2)-skeleton)$  . Then we have for every $\psi \in W^{1,2}\left(K\right)$  , with $\psi =f$  on $K\backslash U$  , $${\int }_{U}e\left(f\right)={\int }_{U\backslash \left(\right(n-2)-skeleton)}e\left(f\right)\le {\int }_{U\backslash \left(\right(n-2)-skeleton)}e\left(\psi \right)={\int }_{U}e\left(\psi \right).$$  We infer that $f$  is a continuous locally $E$  -minimizer map on $K$  , which means that $f$  is harmonic on $U$  . $\square$ 

5 Brownian motions, Harmonic maps and morphisms.

Theorem 5.1 Let $(K,g)$  be an admissible Riemannian polyhedra endowed with a picewise smooth Riemannian metrics and $(N,h)$  a smooth Riemannian manifold. Let $\phi :K\to N$  a continuous map such that $\phi \in W_{loc}^{1,2}(K,N)$  , i.e. $K$  is covered by relatively compact subdomains $U$  such that every $\phi$  is of finite energy on each $U$  . Moreover, $\phi$  is considered at least of class ${\mathcal{C}}^2$  on the interior of each $n$  -simplex and the interior of each $(n-1)$  -simplex. Then $\phi$  is harmonic if and only if for almost all $p\in U$  , $\phi \left(B_t^p\right)$  is a ${}^N\nabla$  -martingale, where $(B_t^p{)}_{t\ge 0}$  is a $K$  -valued Brownian motion.

Proof: Let $(U_N,V_N,W_N,f)$  a ${}^N\nabla$  -martingale tester on $N$  , such that ${\phi }^{-1}\left(W_N\right)\subset U.$  Suppose that $\phi$  is a harmonic map.

By Theorem  4.3 , for all $p\in U\backslash \left(\right(n-2)-skeleton)$  , the process:

$$C_{t\wedge {\tau }_u}^{f\circ \phi \left(p\right)}:=f\circ \phi \left(B_t^p\right)-f\circ \phi \left(B_0^p\right)-{\int }_0^{t\wedge {\tau }_u}{1}_F\left(s\right)L(f\circ \phi )\left(B_s^p\right)ds$$  and $${\tilde{C}}_{t\wedge {\tau }_u}^{f\circ \phi \left(p\right)}:={\chi }_{\{B_t^p/\in ((n-2)-skeleton\left)\right\}}f\circ \phi \left(B_t^p\right)-f\circ \phi \left(B_0^p\right)-{\int }_0^{t\wedge {\tau }_u}{\chi }_{\{B_s^p/\in ((n-2)-skeleton\left)\right\}}{1}_F\left(s\right)L(f\circ \phi )\left(B_s^p\right)ds$$  are local martingale, where $F={\cup }_{i=1}^{\infty }\left]\right]{\sigma }_i,{\tau }_i\left]\right]$  , with

$$\begin{array}{c}{\sigma }_i=inf\{t>{\tau }_{i-1};B_t^p\in U_N\}\end{array}$$

$$\begin{array}{c}{\tau }_i=inf\{t>{\sigma }_i;B_t^p\notin {\overline{V}}_N\}\end{array}$$

$$\begin{array}{c}{\sigma }_0=0\end{array}$$

$$\begin{array}{c}{\tau }_0=0\end{array}$$

The map $\phi$  is supposed to be harmonic and $f$  is a convex function, so by Eells-Fuglede's result (cf. [11] , Theorem 12.1) $f\circ \phi$  is a subharmonic function on ${\phi }^{-1}\left(W_N\right)$  . But in our case the subharmonicity can be traduced by:

$$L(f\circ \phi )\left(p\right)\ge 0,\forall p\in U\backslash \left(\right(n-2)-skeleton)$$  The process ${\chi }_{\{B_t^p/\in ((n-2)-skeleton\left)\right\}}(f\circ \phi )\left(B_{t\wedge {\tau }_u}^p\right)$  is then the sum of a local martingale and an increasing process, so it is a local submartingale $\forall p\in U\backslash \left(\right(n-2)-skeleton)$  .

But the process ${\chi }_{\{B_t^p/\in ((n-2)-skeleton\left)\right\}}(f\circ \phi )\left(B_{t\wedge {\tau }_u}^p\right)$  is equal almost surely to $(f\circ \phi )\left(B_{t\wedge {\tau }_u}^p\right)$  thus, this last process is also a local submartingale for every $p\in U\backslash \left(\right(n-2)-skeleton)$  . Conversely, suppose that for any $p\in U\backslash \left(\right(n-2)-skeleton)$  , ( $U$  is as in the hypothesis of the theorem ), $\phi \left(B_t^p\right)$  is a ${}^N\nabla$  -martingale.

So, for any ${}^N\nabla$  -tester function $f:W_N\to \mathbf{R}$  , $(f\circ \phi )\left(B_{t\wedge {\tau }_u}^p\right)$  is a local submartingale.

By the Theorem ( 4.3 ), the process:

$$H_t^{f\circ \phi \left(p\right)}:=f\circ \phi \left(B_{t\wedge {\tau }_u}^p\right)-f\circ \phi \left(B_{t\wedge {\tau }_u}^0\right)-{\int }_0^{t\wedge {\tau }_u}{1}_F\left(s\right)L(f\circ \phi )\left(B_s^p\right)ds,$$  is a local martingale, for any $p\in U\backslash \left(\right(n-2)-skeleton),$  where $F={\cup }_{i=1}^{\infty }\left]\right]{\sigma }_i,{\tau }_i\left]\right]$  , with

$$\begin{array}{c}{\sigma }_i=inf\{t>{\tau }_{i-1};\phi (B_t^p)\in U_N\}\end{array}$$

$$\begin{array}{c}{\tau }_i=inf\{t>{\sigma }_i;\phi (B_t^p)\notin {\overline{V}}_N\}\end{array}$$

$$\begin{array}{c}{\sigma }_0=0\end{array}$$

$$\begin{array}{c}{\tau }_0=0\end{array}$$

So, we have for any $p\in U\backslash \left(\right(n-2)-skeleton),$  $$\frac{\partial }{\partial t}E\left[\right(f\circ \phi \left)\right(B_{t\wedge {\tau }_u}^p\left)\right]=\frac{1}{2}\Delta (f\circ \phi )\left(p\right).$$  Since $(f\circ \phi )\left(B_{t\wedge {\tau }_u}^p\right)$  is a local submartingale, then $$\frac{\partial }{\partial t}E\left[\right(f\circ \phi \left)\right(B_{t\wedge {\tau }_u}^p\left)\right]\ge 0,$$  so $\Delta (f\circ \phi )\left(p\right)\ge 0$  , for any $p\in U\backslash \left(\right(n-2)-skeleton).$  Thus, by Eells-Fuglede (cf. [11] Theorem 12.1), we obtain that $\phi$  is a harmonic map on $U\backslash \left(\right(n-2)-skeleton).$  Now, using the same arguments as in the proof of Proposition  4.4 , for harmonic functions, we conclude that $\phi$  is harmonic on each $U$  . $\square$ 

Theorem 5.2 Let $(K,g)$  be an admissible Riemannian polyhedra endowed with a picewise smooth Riemannian metrics and $(N,h)$  a smooth Riemannian manifold without boundary. Let $\phi :K\to N$  a continuous map such that $\phi \in W_{loc}^{1,2}(K,N)$  , i.e. $K$  is covered by relatively compact subdomains $U$  such that every $\phi$  is of finite energy on each $U$  . Moreover, $\phi$  is considered at least of class ${\mathcal{C}}^2$  on the interior of each $n$  -simplex and the interior of each $(n-1)$  -simplex. Then $\phi$  is a harmonic morphism if and only if $\phi$  maps $K$  -valued Brownian motions $(B_t^p{)}_{t\ge 0}$  , for any $p\in K\backslash \left(\right(n-2)-skeleton)$  , to a Brownian motion on $N$  , i.e. If ${(}^N{B}_t^p{)}_{t\ge 0}$  denote the Brownian motion the manifold $N$  then there exist a continuous increasing process $(A_t{)}_{t\ge 0}$  such that: ${}^N{B}_t^p\circ A_t=\phi \circ B_t^p$  .

Remark 5.3 We will suppose the $dimK\ge dimN$  . Otherwise, $\phi$  is constant. Indeed, if $dimK<dimN$  , by smooth theory (see [1] , p.46), $\phi$  is constant on each interior of maximal simplex (of a choosen fine triangulation of $K$  ). On the other hand, $\phi$  is continuous and $K$  is $(n-1)$  -chainable, so $\phi$  is constant on $K$  .

Proof of Theorem ( 5.2 ):

For the proof of the Theorem ( 5.2 ), we will addapt and complete the proof given by Darling (see [8] ) in the smooth case.

$”\Rightarrow ”$  Suppose $\phi :K\to N$  is a harmonic morphism. Then by Eells-Fuglede's result (cf [11] Theorem 13.2), there exists a function $\lambda \in L_{loc}^1\left(K\right)$  called the dilation, such that:

$$\begin{array}{c}-{\int }_K\langle \nabla \psi ,\nabla (v\circ \phi )\rangle =-{\int }_K\psi \lambda \left[\right({\Delta }_{N}v)\circ \phi ],\end{array}$$

(10)

for every $v\in {\mathcal{C}}^2\left(N\right)$  and $\psi \in Li{p}_c\left(K\right)$  .

Let $(B_t^p{)}_{t\ge 0}$  a $K$  -valued Brownian motion, for any $p\in K\backslash \left(\right(n-2)-skeleton)$  .

Take $U$  as in the hypothesis of the theorem and suppose $p\in U$  .

Define a continuous increasing process $(A_t{)}_{t\ge 0}$  by:

$$\begin{array}{c}{A}_t:={\int }_0^t\lambda \left(B_s\right)ds,\end{array}$$

(11)

and it's inverse as: $C_t=inf\{u;A_u>u\}.$  Denote $\phi \circ B$  by $X=(X_t{)}_{t\ge 0}$  on $N$  . For any function $f:N\to \mathbf{R}$  of class ${\mathcal{C}}^2$  and any $p\in K\backslash \left(\right(n-2)-skeleton)$  we have:

$${\int }_0^t{\Delta }_{N}f(X^{\phi \left(p\right)}\circ C_s)ds={\int }_0^{C_t}{\Delta }_{N}f\left(X_u^{\phi \left(p\right)}\right)d{A}_u.$$  By ( 11 ) we obtain:

$$\begin{array}{c}{\int }_0^t{\Delta }_{N}f(X^{\phi \left(p\right)}\circ C_s)ds={\int }_0^{C_t}\lambda \left(B_u^p\right){\Delta }_{N}f\left(X_u^{\phi \left(p\right)}\right)du.\end{array}$$

(12)

Using ( 10 ) , the right hand side of the equality ( 12 ) is equal to: $2{\int }_0^{C_t}L(f\circ \phi )\left(B_u^p\right)du,{\mu }_g\text{a.e.}$  , for every $p\in K\backslash \left(\right(n-2)-skeleton)$  .

On the other hand, by Theorem  4.3 , the process:

$${\tilde{H}}_t^{f\circ \phi \left(p\right)}:={\chi }_{\{B_t^p/\in ((n-2)-skeleton\left)\right\}}(f\circ \phi )\left(B_t^p\right)-(f\circ \phi )\left(p\right)-{\int }_0^t{\chi }_{\{B_s^p/\in ((n-2)-skeleton\left)\right\}}L(f\circ \phi )\left(B_s^p\right)ds$$  is a continuous local martingale for any $p\in K\backslash \left(\right(n-2)-skeleton)$  .

Consider now the process: $${\tilde{H}}^{f\circ \phi \left(p\right)}\circ C_t\stackrel{denote}{=}{\tilde{R}}_t^{f\circ \phi \left(p\right)}.$$  ${\tilde{R}}_t^{f\circ \phi \left(p\right)}$  is obviously a continuous local martingale but it is also almost surely equal (using  10 and  12 ) to the process:

$$R_t^{f\circ \phi \left(p\right)}:=f(X^{\phi \left(p\right)}\circ C_t)-f\left(\phi \right(p\left)\right)-\frac{1}{2}{\int }_0^t{\Delta }_{N}f(X^{\phi \left(p\right)}\circ C_s)ds.$$  So $R_t^{f\circ \phi \left(p\right)}$  is also a continuous local martingale for any $p\in K\backslash \left(\right(n-2)-skeleton)$  , which means, by definition, that $X^{\phi \left(p\right)}\circ C_s$  is a Brownian motion on $N$  .

$”\Leftarrow ”$  Conversely, suppose that for any $p\in K\backslash \left(\right(n-2)-skeleton)$  , $\left(\phi \right(B_t^p){)}_{t\ge 0}$  is a Brownian motion on $N$  up to a change of time.

Let $V$  be an open set of $N$  such that ${\phi }^{-1}\left(V\right)\subset U$  , where $U$  is taken as in the hypothesis of the theorem. Let $f:V\to \mathbf{R}$  be a local harmonic function on $N$  .

Fix $p_0\in U\backslash \left(\right(n-2)-skeleton)$  with $\phi \left(p_0\right)\in V$  and $\tau$  denote the first exit time of $(B_t^{p_0}{)}_{t\ge }$  from ${\phi }^{-1}\left(V\right)$  .

By hypothesis, the process $\left(f\right(\phi \left(B_{t\wedge \tau }^{p_0}\right)){)}_{t\ge 0}=(f\circ \phi \left(B_{t\wedge \tau }^{p_0}\right){)}_{t\ge 0}$  is equal to $f{(}^N{B}^{\phi \left(p\right)}\circ A_{t\wedge \tau })$  .

But this last process is continuous local martingale (because, by definition the process $\left(f{(}^N{B}_{t\wedge \tau }^{\phi \left(p\right)}\right){)}_{t\ge 0}$  is a continuous local martingale and martingale property is stable under change of time).

So we have shown that for every $p_0\in U\backslash \left(\right(n-2)-skeleton)$  and for every harmonic function on $N$  , $(f\circ \phi (B_{t\wedge \tau }^{p_0}){)}_{t\ge 0}$  is a continuous local martingale but by Corollary  4.4 this means that $f\circ \phi$  is harmonic.

In other words,we have shown that $\phi$  pulls-back harmonic function on $N$  to harmonic function on $K\backslash \left(\right(n-2)-skeleton)$  . But we have already proved in the proof of the Theorem ( 4.3 ) that harmonic function on $K\backslash \left(\right(n-2)-skeleton)$  are harmonic on $K$  .

We conclude that $\phi$  is a harmonic morphism. $\square$  References

1. P. Baird, J.C. Wood: Harmonic Morphisms Between Riemannian Manifolds, Oxford University Press (2003).
2. W. Ballmann, M. Brin: Orbihedra of Nonpositive Curvature, Publications IHES, 82, 169-209 (1995).
3. J.R. Baxter, R.V. Chacon: The equivalence of diffusions on networks to Brownian motion, Contemporary Mathematics, Vol.26, 33-48 (1984).
4. T. Bouziane: Brownian motion in riemannian admissible complex to appear in Illinois J.Math. (2005) or arXiv:math.PR/0403080
5. M.R. Bridson: Geodesics and Curvature in Metric Simplicial Complexes, World Scientific, Eds. E. Ghys, A.Haefliger, A. Verjovsky (1990).
6. M.R. Bridson, A. Haefliger: Metric spaces of Non-positive curvature, Springer-Verlag (1999).
7. M. Brin, Y. Kifer: Brownian motion, harmonic functions and hyperbolicity for Euclidean complexes, Math.Z. 237, 421-468 (2001).
8. R.W.R. Darling: Martingales in Manifolds-definitions, examples, and behaviour under maps, Séminaire de Probabilités XVI, 1980/81 Supplément: Géométrie Différentielle Stochastique, Lecture Notes in Mathematics, no. 921.
9. M. Davis, T. Januszkiewicz: Hyperbolization of polyhedra, Journal of Differential Geometry, 34, no. 2, 347-388 (1991).
10. E. B. Dynkin: Markov Process, vol.1, Springer-Verlag (1965).
11. J. Eells, B. Fuglede: Harmonic maps between Riemannian Polyhedra, Cambridge University Press (2001).
12. M. Emery: Stochastic Calculus in Manifolds, Springer-Verlag (1989).
13. B. Fuglede: Harmonic morphisms between Riemannian manifolds, Ann. Inst. Fourier 28, 107-144 (1978).
14. E.P. Hsu: Stochastic Analysis on Manifold, Graduate Studies in Mathematics, vol.38, AMS.
15. A. Gray: Tubes, Progress in Mathematics Vol.221, Birkhäuser Verlag (2004).
16. R. E. Greene, H. Wu: Embedding of open Riemannian manifolds by harmonic functions, Ann. Inst. Fourier 25, 1, 215-235 (1975).
17. T. Ishihara: A mapping of Riemannian manifolds which preserves harmonic functions, J. Math. Kyoto Univ.(JMKYAZ) 19, 215-229 (1979).
18. N. J. Korevaar, R. M. Schoen: Sobolev spaces and harmonic maps for metric space targets, Comm. Anal. Geom. 1, no. 3-4, 561-659 (1993).
19. P. A. Meyer: A differential geometric formalism for the Ito calculus, Springer Lecture Notes in Mathematics, Vol.851, 256-270 (1981)
20. B. O'Neill :Semi-Riemannian Geometry with applications to relativity, Academic Press (1983).
21. L. Schwartz :Semi-martingales sur des variétés, et martingales conformes, Lect. Notes in Math., vol. 780, Springer (1980).
22. E. H. Spanier: Algebraic Topology, McGraw-Hill, New York (1966).
23. J. Tits: Buildings of spherical type and finite BN-pairs, Lect. Notes in Math., vol. 386, Springer-Verlag (1974).
24. M.E. Vázquez-Abal: Harmonicity and Brownian motion, Proceedings Workshop: Recent Topics in Differential Geometry, Puerto de la Cruz (Tenerife, Spain), Secret.Public.Univ.de la Laguna, Serie Informes no.32, 123-132 (1991).