pith. sign in

arxiv: 2604.08135 · v2 · submitted 2026-04-09 · 🧮 math.NA · cs.NA

A Multilevel Monte Carlo Virtual Element Method for Uncertainty Quantification of Elliptic Partial Differential Equations

Paola F. Antonietti , Francesca Bonizzoni , Ilaria Perugia , Marco Verani This is my paper

Pith reviewed 2026-05-10 17:43 UTC · model grok-4.3

classification 🧮 math.NA cs.NA
keywords multilevel Monte Carlovirtual element methoduncertainty quantificationstochastic elliptic PDEsrandom diffusion coefficientsmesh agglomerationnumerical analysis
0
0 comments X

The pith

The multilevel Monte Carlo Virtual Element method yields significant cost reductions compared to standard Monte Carlo for uncertainty quantification of elliptic PDEs while preserving convergence rates.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a Monte Carlo estimator using virtual element discretizations for stochastic elliptic partial differential equations with random diffusion coefficients. It introduces a multilevel variant that constructs coarser spaces through mesh agglomeration to form a practical hierarchy. This matters because standard Monte Carlo requires many independent simulations for statistical accuracy, which becomes expensive on fine meshes for complex geometries. The multilevel approach reduces the number of samples needed at the finest level while maintaining error bounds for the solution and linear quantities of interest. Numerical experiments confirm the predicted cost savings.

Core claim

The authors establish that the Multilevel Monte Carlo Virtual Element method converges at the same rates as standard Monte Carlo methods but with substantially lower computational complexity. This follows from error estimates for the statistical approximation of the solution and quantities of interest, combined with a complexity analysis that accounts for the flexibility of virtual element spaces on general polytopal meshes. The key enabler is the use of mesh agglomeration to generate the multilevel hierarchy without losing approximation properties.

What carries the argument

The multilevel hierarchy of virtual element spaces obtained by successive mesh agglomeration, which supplies the different resolution levels for the Monte Carlo sampling and reduces the work on the finest grid.

If this is right

  • Fewer samples are required on the finest level to achieve a prescribed accuracy.
  • Convergence rates for both the solution and suitable linear quantities of interest remain the same as in standard Monte Carlo.
  • The method applies directly to general polytopal meshes where traditional approaches may be limited.
  • Overall computational complexity is reduced compared to plain Monte Carlo for the same accuracy.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The agglomeration technique for building hierarchies could transfer to other discretization families that handle irregular meshes.
  • Similar cost reductions might appear in uncertainty quantification for time-dependent or nonlinear problems if the virtual element framework extends there.
  • The approach suggests a general template for multilevel sampling when geometric flexibility is more important than strict mesh regularity.

Load-bearing premise

That sequences of coarser virtual element spaces built by mesh agglomeration preserve the approximation and stability properties needed for the multilevel error analysis to hold on complex geometries.

What would settle it

A computation on a sequence of refined polytopal meshes for a fixed target accuracy where the observed total cost fails to decrease as predicted by the complexity analysis, or where the statistical error bound is violated.

Figures

Figures reproduced from arXiv: 2604.08135 by Francesca Bonizzoni, Ilaria Perugia, Marco Verani, Paola F. Antonietti.

Figure 1
Figure 1. Figure 1: Sequence of non-nested Voronoi meshes used in the numerical tests of Section [PITH_FULL_IMAGE:figures/full_fig_p017_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: VE error for the QoI, |Q(uex) − Qh(uh)|, for p = 1, 2, 3. 7.2 Practical selection of MC and MLMC sample sizes We discuss the practical selection of MC and MLMC sample sizes, taking into account the comparison between the MC-VE and MLMC-VE methods. Since the smoothness of the solution u may not be known a priori, we adopt the following choice of the number M of MC samples: M =  p 2ph −2p , for EM[u], p 4ph… view at source ↗
Figure 3
Figure 3. Figure 3: (Left) Number of samples per level {Mℓ} 6 ℓ=1 needed to compute the MLMC-VE estimator EL[u] = PL ℓ=1 EMℓ [wℓ] with L = 6 according to (55) with ε = 1e−10. (Right) Total number of samples PL ℓ=1 Mℓ needed to compute the MLMC estimator EL[u] for increasing L = 2, . . . , 6. Both figures are in semilog scale. For the orders p = 1, 2, we compute the MLMC-VE estimator EL[u] for increasing values of L (L = 1, . … view at source ↗
Figure 4
Figure 4. Figure 4: (right) reports the MLMC-VE errors versus the maximum level L in semilog scale, for L = 1, . . . , 6. The results shows exponential decay in L with approximate slope of 0.8 for p = 1 and 1.5 for p = 2. In the same setting, to assess the relative accuracy of MLMC-VE and MC-VE, we fix several levels of computational complexity and plot in loglog scale the corresponding errors for each method in [PITH_FULL_I… view at source ↗
Figure 5
Figure 5. Figure 5: MLMC-VE and MC-VE errors on E[u] in the H1 (D) norm versus the computational complexity in loglog scale. 20 [PITH_FULL_IMAGE:figures/full_fig_p020_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: (Left) MLMC-VEM errors on E[Q(u)] plotted versus hL in loglog scale. (Right) MLMC-VEM errors on E[Q(u)] versus the computational complexity in loglog scale. Q(u) = ´ D u dx. 7.4 Validation test In this section, we present some numerical results to demonstrate the practical capabilities of the proposed scheme. We consider problem (1) defined over the rectangular domain D = (0, 4) × (0, 1), which is split in… view at source ↗
Figure 7
Figure 7. Figure 7: Subdivision of the rectangular domain into subregions, as considered in Section [PITH_FULL_IMAGE:figures/full_fig_p021_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Sequence of nested meshes used in the numerical tests of Section [PITH_FULL_IMAGE:figures/full_fig_p022_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: (Top): MLMC-VE approximation of E[u] for a(ω, x) = P7 r=1 χDr (x)Yr(ω), where Yr ∼ U([1, 10]) for all r = 1, . . . , 7. (Middle): MC approximation of E[u]. (Bottom): Absolute value of the difference of the two approximations divided by the maximum norm of the MLMC-VE approximation. 22 [PITH_FULL_IMAGE:figures/full_fig_p022_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: (Top): MLMC-VE approximation of E[u] for a(ω, x) = P7 r=1 χDr (x)Yr(ω), where Yr ∼ U([1, 2]) for all r ̸= 7 and Y7 ∼ U([1, 100]). (Middle): MC approximation of E[u]. (Bottom): Absolute value of the difference of the two approximations divided by the maximum norm of the MLMC-VE approximation. smaller interval [1, 2]), whereas the seventh region (the red one in [PITH_FULL_IMAGE:figures/full_fig_p023_10.png] view at source ↗
read the original abstract

We introduce a Monte Carlo Virtual Element estimator based on Virtual Element discretizations for stochastic elliptic partial differential equations with random diffusion coefficients. We prove estimates for the statistical approximation error for both the solution and suitable linear quantities of interest. A Multilevel Monte Carlo Virtual Element method is also developed and analyzed to mitigate the computational cost of the plain Monte Carlo strategy. The proposed approach exploits the flexibility of the Virtual Element method on general polytopal meshes and employs sequences of coarser spaces constructed via mesh agglomeration, providing a practical realization of the multilevel hierarchy even in complex geometries. This strategy substantially reduces the number of samples required on the finest level to achieve a prescribed accuracy. We prove convergence of the multilevel method and analyze its computational complexity, showing that it yields significant cost reductions compared to standard Monte Carlo methods for a prescribed accuracy. Extensive numerical experiments support the theoretical results and demonstrate the efficiency of the proposed method.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces a Monte Carlo Virtual Element estimator for stochastic elliptic PDEs with random diffusion coefficients, proves statistical error estimates for the solution and linear quantities of interest, develops a Multilevel Monte Carlo Virtual Element method using mesh agglomeration to realize the hierarchy on general polytopal meshes, proves convergence and complexity bounds demonstrating significant cost reductions relative to standard Monte Carlo, and validates the approach with numerical experiments.

Significance. If the results hold, the work offers a practical route to efficient uncertainty quantification for elliptic problems on complex geometries by exploiting VEM flexibility together with MLMC variance reduction. Explicit credit is due for the proved statistical error estimates, the convergence and complexity analysis, and the numerical demonstration of efficiency gains on polytopal meshes.

major comments (2)
  1. [Section 3.2] Section 3.2 (construction of the multilevel hierarchy via agglomeration): the argument that agglomerated meshes preserve the VEM stability and approximation properties with constants independent of level is load-bearing for the subsequent MLMC rates. Standard VEM theory requires control on star-shapedness, edge-count bounds, and related regularity parameters; agglomeration on arbitrary input geometries can make these level-dependent, which would invalidate the claimed preservation of single-level rates in the multilevel setting.
  2. [Theorem 5.1] Theorem 5.1 (MLMC complexity analysis): the cost-reduction claim relative to plain Monte Carlo for a prescribed accuracy rests on the variance and bias bounds holding with the same rates as the underlying VEM discretization. If the agglomeration step introduces level-dependent factors into the approximation constants, both the proved convergence and the complexity result cease to apply at the stated rates; an explicit additional assumption or bound on the regularity parameters after agglomeration is needed.
minor comments (2)
  1. [Notation] The notation for the random coefficient, the QoI, and the multilevel spaces would benefit from a consolidated list of symbols to improve readability across the analysis sections.
  2. [Numerical Experiments] Numerical experiments section: the captions of the cost-versus-accuracy plots should explicitly state the sequence of mesh sizes, the number of samples per level, and the specific geometries used so that the reported speed-ups can be directly reproduced.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the positive assessment and the detailed comments on the multilevel hierarchy construction and the complexity analysis. We address each major comment below and propose revisions to strengthen the manuscript.

read point-by-point responses

  1. Referee: Section 3.2 (construction of the multilevel hierarchy via agglomeration): the argument that agglomerated meshes preserve the VEM stability and approximation properties with constants independent of level is load-bearing for the subsequent MLMC rates. Standard VEM theory requires control on star-shapedness, edge-count bounds, and related regularity parameters; agglomeration on arbitrary input geometries can make these level-dependent, which would invalidate the claimed preservation of single-level rates in the multilevel setting.

    Authors: We agree that ensuring the mesh regularity parameters remain bounded independently of the level is essential for the analysis to hold. The manuscript's agglomeration procedure, as described in Section 3.2, is constructed to maintain the star-shapedness and other necessary properties from the fine mesh, with the constants controlled uniformly across levels under the assumptions stated in the paper (e.g., the initial mesh satisfies the standard VEM regularity conditions, and agglomeration preserves them). To address the referee's concern explicitly, we will add a clarifying remark in Section 3.2 stating that the agglomeration algorithm guarantees level-independent bounds on these parameters, possibly with a short justification or reference to relevant literature on mesh agglomeration for VEM. revision: yes

  2. Referee: Theorem 5.1 (MLMC complexity analysis): the cost-reduction claim relative to plain Monte Carlo for a prescribed accuracy rests on the variance and bias bounds holding with the same rates as the underlying VEM discretization. If the agglomeration step introduces level-dependent factors into the approximation constants, both the proved convergence and the complexity result cease to apply at the stated rates; an explicit additional assumption or bound on the regularity parameters after agglomeration is needed.

    Authors: The complexity analysis in Theorem 5.1 indeed depends on the single-level error estimates holding with constants independent of the level. We will revise the manuscript to include an explicit assumption on the uniformity of the VEM constants after agglomeration, as suggested. This assumption will be stated clearly, and the proof of Theorem 5.1 will reference it to confirm that the rates remain as claimed. With this addition, the cost-reduction results hold under the stated conditions. revision: yes

Circularity Check

0 steps flagged

No significant circularity; central claims rest on independent proofs

full rationale

The paper states that it proves statistical error estimates for the Monte Carlo VEM estimator, proves convergence of the multilevel variant, and analyzes computational complexity showing cost reductions. These are presented as new derivations that combine VEM discretization theory with MLMC sampling, using mesh agglomeration only as a practical construction for the hierarchy. No step reduces a claimed prediction or rate to a fitted quantity defined inside the paper, nor does any load-bearing premise collapse to a self-citation whose content is merely renamed or assumed without external verification. The derivation chain therefore remains non-circular.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claims rest on standard assumptions from elliptic PDE theory and virtual element approximation properties; no free parameters or new entities are introduced in the abstract.

axioms (2)
  • domain assumption The random diffusion coefficient satisfies standard measurability and boundedness conditions that guarantee well-posedness of the stochastic elliptic problem.
    Required for the existence of the solution and for the statistical error estimates to hold.
  • domain assumption Virtual element spaces on agglomerated meshes retain the approximation properties needed for the multilevel error analysis.
    Central to constructing the hierarchy without losing convergence rates.

pith-pipeline@v0.9.0 · 5461 in / 1331 out tokens · 59424 ms · 2026-05-10T17:43:19.795953+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

27 extracted references · 27 canonical work pages

  1. [1]

    P. F. Antonietti, L. Beirão da Veiga, and G. Manzini, editors. The Virtual Element Method and its Appli- cations, volume 31 of SEMA SIMAI Springer Series . Springer, Cham, Switzerland, 2022

    work page 2022

  2. [2]

    P. F. Antonietti, M. Caldana, I. Mazzieri, and A. Re Fraschini. MAGNET: an open-source library for mesh agglomeration by Graph Neural Networks. Engineering with Computers , 41(6):4825 – 4850, 2025

    work page 2025

  3. [3]

    P. F. Antonietti, M. Corti, and G. Martinelli. Polytopal mesh agglomeration via geometrical deep learning for three-dimensional heterogeneous domains. Mathematics and Computers in Simulation , 241:335–353, 2026

    work page 2026

  4. [4]

    P. F. Antonietti, N. Farenga, E. Manuzzi, G. Martinelli, and L. Saverio. Agglomeration of polygonal grids using graph neural networks with applications to multigrid solvers. Computers & Mathematics with Applications, 154:45–57, 2024

    work page 2024

  5. [5]

    Barth, C

    A. Barth, C. Schwab, and N. Zollinger. Multi-level Monte Carlo finite element method for elliptic PDEs with stochastic coefficients. Numerische Mathematik , 119(1):123–161, 2011

    work page 2011

  6. [6]

    J. Beck, Y. Liu, E. von Schwerin, and R. Tempone. Goal-oriented adaptive finite element multilevel Monte Carlo with convergence rates. Computer Methods in Applied Mechanics and Engineering , 402:115582, 2022

    work page 2022

  7. [7]

    Beirão da Veiga, A

    L. Beirão da Veiga, A. Chernov, L. Mascotto, and A. Russo. Basic principles of hp virtual elements on quasiuniform meshes. Mathematical Models and Methods in Applied Sciences , 26(8):1567–1598, 2016

    work page 2016

  8. [8]

    Beirão da Veiga, F

    L. Beirão da Veiga, F. Brezzi, A. Cangiani, G. Manzini, L. D. Marini, and A. Russo. Basic principles of Virtual Element methods. Mathematical Models and Methods in Applied Sciences , 23(01):199–214, 2013

    work page 2013

  9. [9]

    Beirão da Veiga, F

    L. Beirão da Veiga, F. Brezzi, and L. D. Marini. Virtual Elements for linear elasticity problems. SIAM Journal on Numerical Analysis , 51(2):794–812, 2013

    work page 2013

  10. [10]

    Beirão da Veiga, F

    L. Beirão da Veiga, F. Brezzi, L. D. Marini, and A. Russo. The Hitchhiker’s guide to the Virtual Element Method. Mathematical Models and Methods in Applied Sciences , 24(08):1541–1573, 2014

    work page 2014

  11. [11]

    Beirão da Veiga, F

    L. Beirão da Veiga, F. Brezzi, L. D. Marini, and A. Russo. The Virtual Element Method. Acta Numerica, 32:123–202, 2023

    work page 2023

  12. [12]

    S. C. Brenner, Q. Guan, and L.-Y. Sung. Some estimates for virtual element methods. Computational Methods in Applied Mathematics , 17(4):553–574, 2017

    work page 2017

  13. [13]

    Cangiani, Z

    A. Cangiani, Z. Dong, E. H. Georgoulis, and P. Houston. hp-Version Discontinuous Galerkin Methods on Polygonal and Polyhedral Meshes . Springer Publishing Company, Incorporated, 1st edition, 2017

    work page 2017

  14. [14]

    Charrier, R

    J. Charrier, R. Scheichl, and A. L. Teckentrup. Finite element error analysis of elliptic pdes with random coefficients and its application to multilevel Monte Carlo methods. SIAM Journal on Numerical Analysis , 51(1):322–352, 2013

    work page 2013

  15. [15]

    Cicuttin, A

    M. Cicuttin, A. Ern, and N. Pignet. Hybrid high-order methods: a primer with applications to solid mechanics. SpingerBriefs in Mathematics. Springer, 2021

    work page 2021

  16. [16]

    K. A. Cliffe, M. B. Giles, R. Scheichl, and A. L. Teckentrup. Multilevel Monte Carlo methods and appli- cations to elliptic PDEs with random coefficients. Computing and Visualization in Science , 14(3), 2011

    work page 2011

  17. [17]

    D. A. Di Pietro and J. Droniou. The Hybrid High-Order Method for Polytopal Meshes: Design, Analysis, and Applications. Springer Nature, 2020

    work page 2020

  18. [18]

    D. A. Di Pietro and A. Ern. Mathematical Aspects of Discontinuous Galerkin Methods , volume 69 of Mathématiques et Applications . Springer, 2012

    work page 2012

  19. [19]

    Feder, A

    M. Feder, A. Cangiani, and L. Heltai. R3MG: R-tree based agglomeration of polytopal grids with applica- tions to multilevel methods. Journal of Computational Physics , 526, 2025. 24

    work page 2025

  20. [20]

    M. Giles. Improved Multilevel Monte Carlo Convergence using the Milstein Scheme. In A. Keller, S. Hein- rich, and H. Niederreiter, editors, Monte Carlo and Quasi-Monte Carlo Methods 2006 , pages 343–358, Berlin, Heidelberg, 2008. Springer Berlin Heidelberg

    work page 2006

  21. [21]

    M. B. Giles. Multilevel Monte Carlo path simulation. Operations Research, 56(3):607–617, 2008

    work page 2008

  22. [22]

    M. B. Giles. Multilevel Monte Carlo methods. Acta Numerica, 24:259–328, 2015

    work page 2015

  23. [23]

    Haji-Ali, F

    A.-L. Haji-Ali, F. Nobile, E. von Schwerin, and R. Tempone. Optimization of mesh hierarchies in multilevel Monte Carlo samplers. Stochastics and Partial Differential Equations Analysis and Computations , 4:76–112, 2016

    work page 2016

  24. [24]

    Schwab and A

    C. Schwab and A. Stein. Multilevel Monte Carlo FEM for elliptic PDEs with Besov random tree priors. Stochastics and Partial Differential Equations: Analysis and Computations , 12:1574–1627, 2024

    work page 2024

  25. [25]

    A. L. Teckentrup, R. Scheichl, M. B. Giles, and E. Ullmann. Further analysis of multilevel Monte Carlo methods for elliptic PDEs with random coefficients. Numerische Mathematik , 125(3):569–600, 2013

    work page 2013

  26. [26]

    Čertík, F

    O. Čertík, F. Gardini, G. Manzini, L. Mascotto, and G. Vacca. The p- and hp-versions of the virtual element method for elliptic eigenvalue problems. Computers & Mathematics with Applications , 79(7):2035– 2056, 2020

    work page 2035

  27. [27]

    Zheng and U

    Z. Zheng and U. Nackenhorst. Stochastic virtual element methods for uncertainty propagation of stochastic linear elasticity. Computational Mechanics , 73(3):667–684, 2024. 25

    work page 2024