pith. sign in

arxiv: 1907.00806 · v1 · pith:JRTJDQXPnew · submitted 2019-07-01 · 🧮 math.NA · cs.NA· physics.comp-ph· stat.ML

A data-driven approach for multiscale elliptic PDEs with random coefficients based on intrinsic dimension reduction

Sijing Li , Zhiwen Zhang , Hongkai Zhao This is my paper

Pith reviewed 2026-05-25 11:34 UTC · model grok-4.3

classification 🧮 math.NA cs.NAphysics.comp-phstat.ML
keywords multiscale elliptic PDEsrandom coefficientsdata-driven approachdimension reductionGreen's functionslow-dimensional manifoldnumerical methods
0
0 comments X

The pith

The Green's functions of multiscale elliptic PDEs with random coefficients are highly separable, revealing a low-dimensional solution space that supports efficient data-driven approximation.

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

The paper develops a data-driven method for solving multiscale elliptic PDEs with random coefficients by identifying and exploiting an intrinsic low-dimensional structure in the solution space. It establishes this structure through the high separability of the underlying Green's functions and extracts a reduced basis from sampled data in an offline phase. This basis then allows efficient solution of new PDE instances in an online phase, with error bounds depending on sampling error and the truncation threshold for the basis. A reader would care because these problems are computationally expensive due to multiscale features and randomness, and the method separates the cost of learning the structure from repeated solves.

Core claim

The existence of low dimension structure is established by showing the high separability of the underlying Green's functions; the extracted basis can then be used to solve a new multiscale elliptic PDE efficiently with error controlled by sampling error and truncation threshold.

What carries the argument

The reduced basis extracted from data samples of the solution operator, justified by the high separability of the Green's functions that maps random coefficients to a low-dimensional manifold.

If this is right

  • The method allows efficient solution of new problems using the pre-extracted basis without full resolution each time.
  • Error in the approximation is controlled by the quality of the data samples and the chosen truncation level for the basis dimension.
  • Different online methods can adapt to whether the new coefficient field is fully known or only partially observed.
  • The offline-online split improves efficiency for scenarios with multiple queries under similar random coefficient distributions.

Where Pith is reading between the lines

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

  • The separability property may hold for broader classes of random fields or other elliptic operators beyond those tested.
  • Combining this basis extraction with machine learning techniques could further automate the dimension reduction.
  • Error analysis suggests that increasing sample size reduces the approximation error predictably, which could guide practical implementation.
  • Such reduced bases might aid in uncertainty quantification by representing the solution manifold compactly.

Load-bearing premise

The Green's functions exhibit high separability for the class of random fields considered, mapping the solution operator into a low-dimensional manifold.

What would settle it

A counterexample where the numerical rank of the Green's function matrix remains high even for large sample sets, or where the approximation error does not decrease as predicted with more samples or lower truncation.

Figures

Figures reproduced from arXiv: 1907.00806 by Hongkai Zhao, Sijing Li, Zhiwen Zhang.

Figure 1
Figure 1. Figure 1: Green’s function G(x, y) with dependence on x ∈ D1 and y ∈ D2. Proposition 2.3 (Theorem 2.8 of [8]). Let D1, D2 ⊂ D be two subdomains and D1 be convex (see [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The decay properties of the eigenvalues in the local problem of Sec. [PITH_FULL_IMAGE:figures/full_fig_p015_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Plots of data-driven basis φ1 and φ2 and mappings c1(ξ1, ξ2; ξ3, ξ4, ξ5) and c2(ξ1, ξ2; ξ3, ξ4, ξ5) with fixed [ξ3, ξ4, ξ5] T = [0.25, 0.5, 0.75]T . 15 [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Relative L 2 and H1 error with increasing number of basis for the local problem of Sec.4.1. In [PITH_FULL_IMAGE:figures/full_fig_p016_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The relative testing/projection errors in [PITH_FULL_IMAGE:figures/full_fig_p017_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The decay properties of the eigenvalues for the global problem of Sec. [PITH_FULL_IMAGE:figures/full_fig_p017_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The relative errors with increasing number of basis for the global problem of Sec. [PITH_FULL_IMAGE:figures/full_fig_p018_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: The decay properties of the eigenvalues in the problem of Sec. [PITH_FULL_IMAGE:figures/full_fig_p019_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: The relative errors with increasing number of basis in the problem of Sec. [PITH_FULL_IMAGE:figures/full_fig_p019_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Two realizations of the coefficient (43) in the interface problem. We now solve the local problem of (40) with the coefficient (43), where the domain of interest is D1 = [ 1 4 , 3 4 ] × [ 11 16 , 15 16 ]. The force function is f(x, y) = cos(2πx) sin(2πy) · ID2 (x, y), where D2 = [ 1 4 , 3 4 ]×[ 1 16 , 5 16 ]. In Figure 11a and Figure 11b we show the magnitude of dominant eigenvalues and approximate accura… view at source ↗
Figure 11
Figure 11. Figure 11: The decay properties of the eigenvalues in the problem of Sec. [PITH_FULL_IMAGE:figures/full_fig_p021_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: The relative errors with increasing number of basis in the local problem of Sec. [PITH_FULL_IMAGE:figures/full_fig_p022_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: The decay properties of the eigenvalues in the problem of Sec. [PITH_FULL_IMAGE:figures/full_fig_p022_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Structure of neural network, where r1 = 18 and r2 = 2. We use N1 = 5000 samples for network training in the offline stage and use N2 = 200 sam￾ples for testing in the online stage. The sample data pairs for training are {(ξ n (ω), θ n (ω)), c n (ω)} N1 n=1, where ξ n (ω) ∈ [− 1 5 , 1 5 ] 18 , θ n (ω)) ∈ [ 1 4 , 3 4 ] × [ 1 16 , 5 16 ], and c n (ω) ∈ RK. We define the loss func￾tion of network training as … view at source ↗
Figure 15
Figure 15. Figure 15: First column: the value of loss function during training procedure. Second column and third [PITH_FULL_IMAGE:figures/full_fig_p024_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: The relative errors with increasing number of basis in the local problem of Sec. [PITH_FULL_IMAGE:figures/full_fig_p026_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: The relative errors with increasing number of basis in the global problem of Sec. [PITH_FULL_IMAGE:figures/full_fig_p026_17.png] view at source ↗
read the original abstract

We propose a data-driven approach to solve multiscale elliptic PDEs with random coefficients based on the intrinsic low dimension structure of the underlying elliptic differential operators. Our method consists of offline and online stages. At the offline stage, a low dimension space and its basis are extracted from the data to achieve significant dimension reduction in the solution space. At the online stage, the extracted basis will be used to solve a new multiscale elliptic PDE efficiently. The existence of low dimension structure is established by showing the high separability of the underlying Green's functions. Different online construction methods are proposed depending on the problem setup. We provide error analysis based on the sampling error and the truncation threshold in building the data-driven basis. Finally, we present numerical examples to demonstrate the accuracy and 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

0 major / 2 minor

Summary. The manuscript proposes a data-driven approach for multiscale elliptic PDEs with random coefficients. It consists of an offline stage that extracts a low-dimensional basis from sampled data by exploiting high separability of the underlying Green's functions, and an online stage that uses this basis to solve new instances efficiently. Error analysis is provided in terms of sampling error and truncation threshold, with numerical examples demonstrating accuracy and efficiency.

Significance. If the claimed separability of Green's functions holds for the considered random fields, the method supplies a parameter-free dimension reduction for the solution operator together with explicit error control by sampling and truncation. This is a substantive contribution to data-driven solvers for random-coefficient elliptic problems, as the construction directly ties the reduced basis to an intrinsic property of the Green's function rather than to fitted parameters.

minor comments (2)
  1. [Abstract] Abstract: the phrase 'different online construction methods are proposed depending on the problem setup' is not expanded; a one-sentence indication of the two main variants would improve readability.
  2. The error analysis section would benefit from an explicit statement of the norm in which the sampling-plus-truncation bound is derived, to make the comparison with standard finite-element error estimates immediate.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive summary, significance assessment, and recommendation of minor revision. The referee's description accurately reflects the offline/online structure, the role of Green's function separability, and the error analysis in the manuscript.

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper's central claim rests on establishing low-dimensional structure via high separability of Green's functions for the considered random fields, which is presented as an external property of the operators rather than a quantity defined or fitted from the same data used for testing. Error bounds are controlled by sampling error and truncation threshold in basis construction, with no reduction of predictions to fitted inputs by construction. No load-bearing self-citations, uniqueness theorems from prior author work, or ansatzes smuggled via citation are indicated. The derivation chain from separability to controlled approximation remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the unproven (in the abstract) assertion that Green's functions of the random-coefficient operators are highly separable; no free parameters or invented entities are mentioned.

axioms (1)
  • domain assumption Green's functions of the multiscale elliptic operators with random coefficients exhibit high separability, inducing a low-dimensional structure in the solution space.
    Invoked in the abstract to justify both the dimension reduction and the existence of an efficient basis.

pith-pipeline@v0.9.0 · 5674 in / 1418 out tokens · 20585 ms · 2026-05-25T11:34:29.072690+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

41 extracted references · 41 canonical work pages

  1. [1]

    Abdulle, A

    A. Abdulle, A. Barth, and C. Schwab , Multilevel Monte Carlo methods for stochastic elliptic multiscale PDEs , Multiscale Modeling & Simulation, 11 (2013), pp. 1033–1070

    work page 2013

  2. [2]

    Arnst and R

    M. Arnst and R. Ghanem , Probabilistic equivalence and stochastic model reduction in multiscale analysis, Comput. methods Appl. Mech. Engrg, 197(43) (2008), pp. 3584– 3592

    work page 2008

  3. [3]

    B. V. Asokan and N. Zabaras , A stochastic variational multiscale method for dif- fusion in heterogeneous random media , Journal of Computational Physics, 218 (2006), pp. 654–676

    work page 2006

  4. [4]

    Babuska, F

    I. Babuska, F. Nobile, and R. Tempone , A stochastic collocation method for elliptic partial differential equations with random input data , SIAM J. Numer. Anal., 45 (2007), pp. 1005–1034

    work page 2007

  5. [5]

    Babuska, R

    I. Babuska, R. Tempone, and G. Zouraris , Galerkin finite element approxi- mations of stochastic elliptic partial differential equations , SIAM J. Numer. Anal., 42 (2004), pp. 800–825

    work page 2004

  6. [6]

    Barrault, Y

    M. Barrault, Y. Maday, N. C. Nguyen, and A. T. Patera , An empirical interpolation method: application to efficient reduced-basis discretization of partial dif- ferential equations, Comptes Rendus Mathematique, 339(9) (2004), pp. 667–672

    work page 2004

  7. [7]

    Bebendorf and W

    M. Bebendorf and W. Hackbusch , Existence of H-matrix approximants to the inverse FE-matrix of elliptic operators with L infinity-coefficients , Numerische Mathe- matik, 95 (2003), pp. 1–28

    work page 2003

  8. [8]

    , Existence of H-matrix approximants to the inverse FE-matrix of elliptic operators with L infinity coefficients , Numerische Mathematik, 95 (2003), pp. 1–28. 27

    work page 2003

  9. [9]

    Benner, S

    P. Benner, S. Gugercin, and K. Willcox , A survey of projection-based model reduction methods for parametric dynamical systems, SIAM Review, 57 (2015), pp. 483– 531

    work page 2015

  10. [10]

    Berkooz, P

    G. Berkooz, P. Holmes, and J. L. Lumley , The proper orthogonal decomposi- tion in the analysis of turbulent flows , Annual review of fluid mechanics, 25(1) (1993), pp. 539–575

    work page 1993

  11. [11]

    Bryson, H

    J. Bryson, H. Zhao, and Y. Zhong , Intrinsic complexity and scaling laws: from random fields to random vectors, SIAM Journal on Multiscale Modeling and Simulation, 17 (2019), pp. 460–481

    work page 2019

  12. [12]

    H. J. Bungartz and M. Griebel , Sparse grids, Acta Numerica, 13 (2004), pp. 147– 269

    work page 2004

  13. [13]

    Cheng, T

    M. Cheng, T. Y. Hou, M. Yan, and Z. Zhang , A data-driven stochastic method for elliptic PDEs with random coefficients , SIAM J. UQ, 1 (2013), pp. 452–493

    work page 2013

  14. [14]

    Chung, Y

    E. Chung, Y. Efendiev, W. Leung, and Z. Zhang , Cluster-based generalized multiscale finite element method for elliptic PDEs with random coefficients , Journal of Computational Physics, 371 (2018), pp. 606–617

    work page 2018

  15. [15]

    Dolzmann and S

    G. Dolzmann and S. M ¨uller, Estimates for green’s matrices of elliptic systems byl p theory, Manuscripta mathematica, 88 (1995), pp. 261–273

    work page 1995

  16. [16]

    Efendiev, C

    Y. Efendiev, C. Kronsbein, and F. Legoll , Multilevel Monte Carlo approaches for numerical homogenization, Multiscale Modeling & Simulation, 13 (2015), pp. 1107– 1135

    work page 2015

  17. [17]

    Engquist and H

    B. Engquist and H. Zhao , Approximate separability of the Green’s function of the Helmholtz equation in the high frequency limit , Communications on Pure and Applied Mathematics, 71 (2018), pp. 2220–2274

    work page 2018

  18. [18]

    Ghanem and P

    R. Ghanem and P. Spanos , Stochastic finite elements: a spectral approach., Springer- Verlag, New York, 1991

    work page 1991

  19. [19]

    Graham, F

    I. Graham, F. Kuo, D. Nuyens, R. Scheichl, and I. Sloan , Quasi-Monte Carlo methods for elliptic PDEs with random coefficients and applications , Journal of Com- putational Physics, 230 (2011), pp. 3668–3694

    work page 2011

  20. [20]

    I. G. Graham, F. Y. Kuo, J. A. Nichols, R. Scheichl, C. Schwab, and I. H. Sloan, Quasi-Monte Carlo finite element methods for elliptic PDEs with lognormal random coefficients, Numerische Mathematik, 131(2) (2015), pp. 329–368

    work page 2015

  21. [21]

    Gr ¨uter and K

    M. Gr ¨uter and K. Widman , The green function for uniformly elliptic equations , Manuscripta Mathematica, 37 (1982), pp. 303–342. 28

    work page 1982

  22. [22]

    K. He, X. Zhang, S. Ren, and J. Sun , Deep residual learning for image recognition, in Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 770–778

    work page 2016

  23. [23]

    Hou and P

    T. Hou and P. Liu , A heterogeneous stochastic FEM framework for elliptic PDEs , Journal of Computational Physics, 281 (2015), pp. 942–969

    work page 2015

  24. [24]

    T. Hou, P. Liu, and Z. Zhang , A localized data-driven stochastic method for elliptic PDEs with random coefficients , Bull. Inst. Math. Acad. Sin. (N.S.), 1 (2016), pp. 179– 216

    work page 2016

  25. [25]

    T. Hou, D. Ma, and Z. Zhang , A model reduction method for multiscale elliptic PDEs with random coefficients using an optimization approach , Multiscale Modeling & Simulation, 17 (2019), pp. 826–853

    work page 2019

  26. [26]

    T. Y. Hou, W. Luo, B. Rozovskii, and H. M. Zhou , Wiener chaos expansions and numerical solutions of randomly forced equations of fluid mechanics , J. Comput. Phys., 216 (2006), pp. 687–706

    work page 2006

  27. [27]

    I. G. Kevrekidis, C. W. Gear, J. M. Hyman, P. G. Kevrekidid, O. Run- borg, and C. Theodoropoulos , Equation-free, coarse-grained multiscale computa- tion: Enabling mocroscopic simulators to perform system-level analysis , Communica- tions in Mathematical Sciences, 1(4) (2003), pp. 715–762

    work page 2003

  28. [28]

    Manohar, B

    K. Manohar, B. Brunton, J. Kutz, and S. Brunton , Data-driven sparse sensor placement for reconstruction, arXiv:1701.07569, (2017)

    work page arXiv 2017

  29. [29]

    H. G. Matthies and A. Keese , Galerkin methods for linear and nonlinear elliptic stochastic partial differential equations, Comput. Method Appl. Mech. Eng., 194 (2005), pp. 1295–1331

    work page 2005

  30. [30]

    H. N. Najm , Uncertainty quantification and polynomial chaos techniques in computa- tional fluid dynamics , Annual Review of Fluid Mechanics, 41 (2009), pp. 35–52

    work page 2009

  31. [31]

    Nobile, R

    F. Nobile, R. Tempone, and C. Webster , A sparse grid stochastic collocation method for partial differential equations with random input data, SIAM J. Numer. Anal., 46 (2008), pp. 2309–2345

    work page 2008

  32. [32]

    Rozza, D

    G. Rozza, D. B. Huynh, and A. T. Patera , Reduced basis approximation and a posteriori error estimation for affinely parametrized elliptic coercive partial differential equations, Archives of Computational Methods in Engineering, 15(3) (2007), pp. 1–47

    work page 2007

  33. [33]

    Sapsis and P

    T. Sapsis and P. Lermusiaux , Dynamically orthogonal field equations for continuous stochastic dynamical systems, Physica D: Nonlinear Phenomena, 238 (2009), pp. 2347– 2360

    work page 2009

  34. [34]

    Sirovich , Turbulence and the dynamics of coherent structures

    L. Sirovich , Turbulence and the dynamics of coherent structures. I. Coherent struc- tures, Quarterly of applied mathematics, 45(3) (1987), pp. 561–571. 29

    work page 1987

  35. [35]

    W ald and V

    I. W ald and V. Havran , On building fast kd-trees for ray tracing, and on doing that in O (N log N) , in 2006 IEEE Symposium on Interactive Ray Tracing, IEEE, 2006, pp. 61–69

    work page 2006

  36. [36]

    W an and N

    J. W an and N. Zabaras , A probabilistic graphical model approach to stochastic mul- tiscale partial differential equations , Journal of Computational Physics, 250 (2013), pp. 477–510

    work page 2013

  37. [37]

    X. L. W an and G. Karniadakis , Multi-element generalized polynomial chaos for arbitrary probability measures, SIAM J. Sci. Comp., 28 (2006), pp. 901–928

    work page 2006

  38. [38]

    Xiu, Fast numerical methods for stochastic computations: a review , Commun

    D. Xiu, Fast numerical methods for stochastic computations: a review , Commun. Com- put. Phys., 5 (2009), pp. 242–272

    work page 2009

  39. [39]

    Xiu and J

    D. Xiu and J. S. Hesthaven , High-order collocation methods for differential equa- tions with random inputs , SIAM J. Sci. Comp., 27 (2005), pp. 1118–1139

    work page 2005

  40. [40]

    Xiu and G

    D. Xiu and G. Karniadakis , Modeling uncertainty in flow simulations via generalized polynomial chaos, J. Comput. Phys., 187 (2003), pp. 137–167

    work page 2003

  41. [41]

    Zhang, M

    Z. Zhang, M. Ci, and T. Y. Hou , A multiscale data-driven stochastic method for elliptic PDEs with random coefficients , SIAM Multiscale Model. Simul., 13 (2015), pp. 173–204. 30

    work page 2015