1: \begin{abstract}

2: We formulate standard and multilevel Monte Carlo methods for  

3: the $k$th moment $\mathbb{M}^k_\varepsilon[\xi]$ 

4: of a Banach space valued random variable 

5: $\xi\colon\Omega\to E$, 

6: interpreted as an element 

7: of the $k$-fold injective 

8: tensor product space~$\otimes^k_\varepsilon E$. 

9: For the standard Monte Carlo estimator of 

10: $\mathbb{M}^k_\varepsilon[\xi]$, 

11: we prove the $k$-independent convergence rate 

12: $1-\tfrac{1}{p}$ 

13: in the $L_q(\Omega;\otimes^k_\varepsilon E)$-norm,  

14: provided that 

15: (i) $\xi\in L_{kq}(\Omega;E)$  

16: and 

17: (ii) $q\in[p,\infty)$, where $p\in[1,2]$ is  

18: the Rademacher type of $E$. 

19: % 

20: By using the fact that 

21: Rademacher averages 

22: are dominated by Gaussian sums 

23: combined with 

24: a version of Slepian's inequality for Gaussian processes 

25: due to Fernique, 

26: we moreover derive 

27: corresponding results 

28: for multilevel Monte Carlo methods, 

29: including a rigorous error estimate 

30: in the $L_q(\Omega;\otimes^k_\varepsilon E)$-norm 

31: and the optimization of the computational cost  

32: for a given accuracy.  

33: Whenever the type of the 

34: Banach space~$E$ is $p=2$, 

35: our findings coincide with known results 

36: for Hilbert space valued random variables. 

37: 

38: We illustrate the abstract results 

39: by three model problems: 

40: second-order elliptic PDEs with 

41: random forcing or random coefficient, 

42: and stochastic evolution equations.

43: In these cases, the 

44: solution processes naturally take values 

45: in non-Hilbertian Banach spaces. 

46: Further applications, where physical modeling

47: constraints impose a setting in Banach spaces 

48: of type $p<2$, are indicated. 

49: \end{abstract}

50: