Monotone approximation of energy functionals for mappings into metric spaces, I. (Q4339897)

A harmonic map \(f:(M,g) \to(N,h)\) between Riemannian manifolds is the critical point of the energy functional \(E(f)= {1\over 2} \int_M |df(x) |^2 \mu (dx)\) where \(df(x)\) is the differential of \(f\) at \(x\in M\) and \(\mu(dx)\) is the volume measure on \(M\). For any \(L^2\)-function \(f:M\to \mathbb{R}\) the energy \(E(f)\) can be approximated monotonically by \[ E^t(f) ={1\over 2t} \int_M\int_M \bigl|f(x) -f(y)\bigr |^2P_t (x,dy) \mu(dx) \tag{1} \] for \(t \to 0\), where \((P_t)_{t>0}\) denotes the heat semigroup associated with the heat equation on \(M\). More generally, let \((P_t)_{t>0}\) be any symmetric Markov semigroup on some measure space \((M, {\mathcal M}, \mu)\). Then for any real-valued \(f\in L^2 (M, {\mathcal M}, \mu)\) the approximating energy \(E^t(f)\) defined as in (1) is a decreasing function of \(t>0\) and \(E(f): = \lim_{t \to 0} E^t (f)= \sup_{t>0} E^t(f)\) defines a Dirichlet form \((E, {\mathcal D} (E))\) with the domain \({\mathcal D} (E)= \{f \in L^2 (M, {\mathcal M}, \mu): E(f)< \infty\}\). [cf. \textit{M. Fukushima}, \textit{Y. Oshima} and \textit{M. Takeda}, `Dirichlet forms and symmetric Markov processes', de Gruyter, New York (1994; Zbl 0838.31001)]. In order to define and study the energy functional for mappings \(f:M\to N\) with values in some metric space \((N,d)\), J. Jost proposed to use the approximations \[ E^t(f)= {1 \over 2t} \int_M \int_M d^2 \bigl(f(x), f(y)\bigr) P_t(x,dy) \mu (dx) \tag{2} \] for \(t\to 0\). If \((X_t, \mathbb{P}_x)_{t>0, x\in M}\) is a Markov process on \(M\) associated with the semigroup \((P_t)_{t>0}\) in the sense that \(P_t(x,dx) =\mathbb{P}_x (X_t\in dy)\), then (2) can be expressed in probabilistic terms as \[ E^t(f)= {1\over 2t} \mathbb{E}_\mu \biggl[d^2 \bigl(f(X_t), f(X_0) \bigr) \biggr] \] where \(\mathbb{E}_\mu [\cdot]= \int.d \mathbb{P}_x \mu(dx)\). Particularly, for \(k\in \mathbb{N}\), \(E^{2^{-k}} (f)= {1\over 2} \mathbb{E}_\mu [\text{Var}_k(f)]\) with \(\text{Var}_k (f)= \sum^{2^k}_{i=1} d^2 (f(X_{i2^{-2}})\), \(f(X_{(i-1) 2^{-k}}))\) being an approximation for the quadratic variation of the process \(((f \circ X)_t)_{0\leq t\leq 1}\). The main results of this paper are as follows: (i) \(E^t(f)\) as defined in (2) is not always decreasing in \(t\); a counterexample is given. (ii) For any symmetric Markov semigroup \((P_t)_{t>0}\) on any measure \((M,{\mathcal M}, \mu)\) and for any mapping \(f:M\to N\), the approximations \(E^{2^{-k}} (f)\) defined as in (2) are increasing in \(k\in\mathbb{N}\), provided the metric space \((N,d)\) is nonnegatively curved in the sense of some new integral condition. Instead of using the semigroup \((P_t)_{t>0}\) one can also use resolvents \((U_\alpha)_{\alpha>0}\) which leads to \[ E^\alpha (f)= {\alpha^2 \over 2} \int_M \int_M d^2(f(x),\;f(y) U_\alpha (x,dy) \mu (dx) \tag{3} \] for \(\alpha\to \infty\). (iii) \(E^\alpha(f)\) is an increasing function of \(\alpha\) provided the target space \((N,d)\) has nonnegative integral curvature.