Scale-invariant multilevel Monte Carlo method and application to linear elasticity

We propose a novel scale-invariant version of the mean and variance multi-level Monte Carlo estimate. The computation cost across grid levels is optimised using a normalized error based on t-statistics. By doing so, the algorithm achieves convergence independent of the physical scale at which the estimate is computed. The effectiveness of this algorithm is demonstrated through testing on a linear elastic example, where material uncertainty incorporating both heterogeneity and anisotropy is considered in the constitutive law.