On Error Analysis of Multilevel Picard Iterations for Parabolic Stochastic Partial Differential Equations with Gradient-Independent Nonlinearities
In our previous work [42], we introduced a multilevel Picard iteration method for semilinear parabolic stochastic partial differential equations (SPDEs) and demonstrated its numerical performance. The present paper specifies a conditional least-squares regression formulation of that approach and develops its error analysis in a gradient-independent setting. We explain how the conditional expectations arising in the Picard fixed-point formulation can be computed, for a fixed path of the external noise, by ordinary least-squares regressions over simulated forward paths. We prove pathwise stability estimates and non-asymptotic regression bounds. For a common histogram basis we obtain convergence to the semi-discrete fixed point and explicit finite-grid cost bounds. The expected-cost bound is subexponential, rather than polynomial, in the number of time steps, with strong dimension dependence.
Authors
- He Zhang
Institutions
- Jilin University (CN)
Publication Details
- Journal
- Communications in Mathematical Research
- Published
- 2026-11-18
- DOI
- https://doi.org/10.4208/cmr.2026-0053
- Primary Topic
- Model Reduction and Neural Networks
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- National Natural Science Foundation of China