1: \begin{abstract}

2: 	Numerical methods for stochastic partial differential equations

3: 	typically

4: 	estimate moments of the solution

5: 	from sampled paths.

6: 	Instead,

7: 	we shall directly target

8: 	the deterministic equations

9: 	satisfied by the first and second moments, as well as the covariance.

10: 

11: 	In the first part, we focus on stochastic ordinary differential equations.

12: 	For the canonical examples

13: 	with additive noise (Ornstein--Uhlenbeck process)

14: 	or

15: 	multiplicative noise (geometric Brownian motion)

16: 	we derive these deterministic equations

17: 	in variational form

18: 	and

19: 	discuss their well-posedness in detail.

20: 	Notably,

21:     the second moment equation

22: 	in the multiplicative case

23:     is naturally posed on

24: 	projective--injective tensor product spaces

25: 	as trial--test spaces.

26: 	We construct Petrov--Galerkin discretizations

27: 	based on tensor product piecewise polynomials

28: 	and analyze their stability and convergence

29: 	in these natural norms.

30: 

31: 	In the second part,

32: 	we proceed with

33: 	parabolic stochastic partial differential equations

34: 	with affine multiplicative noise.

35: 	We prove well-posedness of the deterministic variational problem

36: 	for the second moment,

37: 	improving an earlier result.

38: 	We then propose conforming space-time Petrov--Galerkin discretizations,

39:     which we show to be stable and quasi-optimal.

40: 

41:     In both parts, the outcomes are illustrated by numerical examples.

42: \end{abstract}

43: