1: \begin{abstract}

2: Randomness is ubiquitous in modern engineering. The uncertainty is

3: often modeled as random coefficients in the differential equations

4: that describe the underlying physics.  In this work, we describe a

5: two-step framework for numerically solving semilinear elliptic partial

6: differential equations with random coefficients: 1) reformulate

7: the problem as a functional minimization problem based on the

8: direct method of calculus of variation; 2) solve the minimization

9: problem using the stochastic gradient descent method.  We

10: provide the convergence criterion for the resulted stochastic gradient

11: descent algorithm and discuss some useful technique to overcome the issues of ill-conditioning and large variance. The accuracy and efficiency of the algorithm are demonstrated by numerical experiments.

12: \end{abstract}

13: