Given an open set $\Omega$ of compact closure in $\mathbb{R}^m$, the classical Dirichlet problem is to extend a given continuous function $\psi : \partial \Omega \to \mathbb{R}$ to a continuous function $\phi : \overline \Omega \to \mathbb{R}$ such that $\phi$ is harmonic (that is, satisfies the Laplace equation) in $\Omega$. The set $\Omega$ is termed regular if the Dirichlet problem has a (necessarily unique) solution for any continuous boundary function $\psi$. For example, every simply connected planar domain is regular (but may have a 'bad' boundary $\partial\Omega$, for instance, a fractal).

In this article it is shown that, when $\Omega$ is regular (in the above sense), every continuous map $\psi$ from $\partial\Omega$ to a simply connected complete Riemannian manifold $(N, h)$ of sectional curvature at most 0 has a unique continuous extension $\phi : \overline\Omega \to N$ which is harmonic in $\Omega$. This is done with $\mathbb{R}^m$ replaced more generally by an $m$-dimensional Riemannian manifold $(M, g)$.

The proof relies on the unique solvability of the corresponding variational Dirichlet problem (for any open set $\Omega \Subset M$). And for that, the above target manifold $N$ can be replaced more generally by any simply connected complete geodesic space $Y$ of curvature at most 0 in the sense of A. D. Alexandrov. Assuming that $M$ satisfies the Poincaré inequality, we show that, for any map $\psi : M \to Y$ of finite energy in the sense of N. J. Korevaar and R. M. Schoen, there exists a unique map $\phi : M \to Y$ with $\phi = \psi$ on $M \setminus \Omega$ such that $\phi$ minimizes the energy of all maps $M \to Y$ which agree with $\psi$ on $M \setminus \Omega$. If $\Omega$ is regular then $\phi$ is continuous at any point of $\partial\Omega$ at which $\psi$ is continuous. For a Lipschitz (and hence regular) domain $\Omega \Subset M$, existence and uniqueness of the variational solution $\phi$ was obtained by Korevaar and Schoen, and earlier for suitable polyhedral targets $Y$ by Gromov and Schoen.

Instead of the Riemannian manifold $M$ our domain space can still more generally be an admissible Riemannian polyhedron (as studied in the recent Cambridge Tract by J. Eells and the present author); the variational solution $\phi$ is then in general only Hölder continuous in $\Omega$.

The proofs of the stated results of this article rely in part on potential theory relative to the fine topology of H. Cartan.