Maximum-norm a posteriori error bounds for parabolic equations discretised by the extrapolated Euler method in time and FEM in space

Goran Radojev; Torsten Lin{\ss}

REVIEW 4 major objections 5 minor 22 references

Maximum-norm a posteriori error bounds for parabolic equations discretised by the extrapolated Euler method in time and FEM in space

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper establishes a fully computable maximum-norm a posteriori error bound for the L-step Richardson-extrapolated backward Euler method combined with FEM, valid at the final time for linear parabolic equations.

desk verdict The arbitrary-order reconstruction machinery is a genuine extension, but a factor error in Lemma 1 makes the printed η_t not a guaranteed bound until fixed. read the letter →

arxiv 2411.13617 v1 pith:WOB33WRA submitted 2024-11-20 math.NA cs.NA

Torsten Lin{ss} , Goran Radojev This is my paper · ORCID

classification math.NAcs.NA MSC 65M1565M5065M60

keywords parabolicproblemsmaximum-normaposteriorierrorestimatesbackwardEulerRichardsonextrapolationellipticreconstructionsGreen'sfunctionfiniteelementmethod

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper derives guaranteed, computable upper bounds on the pointwise (maximum-norm) error at the final time for linear parabolic equations solved by Richardson extrapolation of the backward Euler method in time and finite elements in space. The central claim is that, although the extrapolation produces high-order approximations only at temporal mesh points, one can still build a piecewise-polynomial-in-time reconstruction whose residual controls the error, and all ingredients in the bound are evaluated from the computed discrete solution and given data. This matters because maximum-norm a posteriori bounds of this type have previously been available only for low-order or special time discretisations; the paper extends them to extrapolations of arbitrary order. If the bound is correct, an adaptive solver can certify the pointwise accuracy of its final output without knowing the exact solution.

What carries the argument

The machinery is the elliptic reconstruction $R(t)$, defined on each time slice as the solution of $LR(t)=\hat f(t)+\psi(t)$, where $\psi$ measures the deviation of the discrete equation from the continuous one; the finite element solution $u_h(t)$ is then an approximation of $R(t)$, and the elliptic estimator from Assumption 1 bounds $R-u_h$ pointwise. The second key ingredient is the temporal reconstruction: on each interval the numerical solution is represented in a basis built from integrated Legendre polynomials $N_i$, and the free coefficients are chosen so that $\psi+\partial_t u_h$ reduces to a linear combination of the two highest Legendre polynomials on each interval. Lemma 1 then bounds the Green's-function-weighted time integrals of such terms using the assumed decay bounds on $G$ and its derivative.

What would settle it

Take a concrete linear parabolic problem with known exact solution, run the L-step extrapolated Euler-FEM scheme, evaluate each term of $\eta_M$, and compare the true error $\|u(T)-u_M^h\|_{\infty,\Omega}$ with the estimator; because Theorem 1 claims a guaranteed upper bound, any single computed example with true error larger than $\eta_M$ would refute the bound.

Watch

Extended reading notes

Core claim

The paper's central result is Theorem 1: for the L-step Richardson extrapolation method applied to the backward Euler discretisation, combined with any finite-element discretisation that satisfies the paper's assumptions, the error at the final time obeys $$\|u(T)-u_M^h\|_{\infty,\$\Omega$} \le \eta_M := \eta_{\mathrm{init}} + \eta_f + \eta_t + \eta_{\mathrm{ell}},$$ where each term is assembled from the computed approximations, the interpolated source term, elliptic reconstructions, an available elliptic a posteriori estimator, and constants depending only on the Green's function of the parabolic operator. The bound is fully computable from the numerical solution and the data, with no reference to the unknown exact solution. The proof uses the Green's function representation of the error and an elliptic reconstruction of the discrete solution, with the temporal reconstruction chosen so that the residual term $\psi+\partial_t u_h$ collapses to a sum of two Legendre polynomials on each time interval.

Load-bearing premise

The paper's bound depends on the unproved assumption that a computable maximum-norm error estimator exists for the chosen finite-element discretisation of the elliptic problem, including the quadrature and mesh actually used.

Editorial extensions

If this is right

For any extrapolation order L ≥ 2, the final-time solution carries a pointwise error certificate computed from the discrete solution and data alone.
The four additive terms in the estimator separate initial-data, source-interpolation, temporal, and spatial contributions, so an adaptive procedure can target the dominant source of error.
Evaluating the estimate requires only the mesh-point values of the extrapolated solution, auxiliary extrapolation values, the interpolated source, and one elliptic estimator per time level; no additional parabolic solves are needed.
For L = 2 the estimator reduces to the previously known second-order framework, which the present construction generalises systematically.
Any improvement in the available maximum-norm elliptic estimators immediately improves the parabolic bound without changing the temporal analysis.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The paper does not report numerical experiments; a natural check is to compute the effectivity ratio on problems with known solutions to see how far the estimator is from one.
Because the Legendre-bubble construction is the only method-dependent step, the same reconstruction may extend to Richardson extrapolation of other one-step methods, though the paper does not make that claim.
The estimator's per-interval terms suggest a time-adaptive strategy driven by the temporal contribution and local Green's-function weights, an algorithmic step not developed here.
If sharper Green's-function constants became available for specific operators, the bound would tighten automatically; the paper takes the constants as input data.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Two things to know. First, the core idea is real: the paper extends the elliptic reconstruction framework to Richardson extrapolation of arbitrary order L, and the systematic construction of degree L-1 temporal reconstructions from nodal values via integrated Legendre polynomials is a genuine contribution, not in the authors' earlier L=2 paper. Second, as printed, Lemma 1 has a factor error that breaks the guarantee in Theorem 1's η_t, and the L=2 case is mishandled.

What is actually new and good: the recursive definition of the auxiliary unknowns Δ_i ψ and Δ_i u is clever and nontrivial. The proof structure is coherent, the paper honestly lists its external assumptions, and no fitted parameters or manufactured predictions enter. If the algebraic issues are fixed, this gives a useful computable maximum-norm a posteriori bound for a wide class of parabolic FEM discretisations.

Soft spots, in order of severity. (1) In Lemma 1, after integration by parts the proof reaches 1/τ_j ∫ φ_1(T-s)(t_j-s)(s-t_{j-1}) ds. Since that integral is τ_j^3/6, the second constant should be κ_1' τ_j^2/6, not κ_1' τ_j^2/24. Because μ_{j,i} is defined as a minimum, the too-small printed constant can make μ_{j,i} four times too small, so η_t is not a guaranteed upper bound as written. This is load-bearing, not cosmetic. (2) Theorem 1's formula for η_t uses 1/(2L-5), which for L=2 equals -1 and does not match the P0 coefficient in equation (21); a separate L=2 case is needed. Also, the theorem announces η_ψ + η_Ψ while only η_t is defined. (3) Assumption 1 is a real condition: the elliptic estimator is cited from the literature in special settings (quasiuniform meshes, certain operators), but the paper states it for general FEM with quadrature on arbitrary meshes. That makes the main theorem conditional on an assumption that may not hold in the stated generality. Minor point: there are no numerical experiments, so nothing tests sharpness or verifies the constants.

Who this is for: numerical analysts working on adaptive methods for parabolic problems, especially those interested in maximum-norm control. The framework is worth engaging with, but not in its current form. I would send it to a serious referee—the construction is valuable and the errors seem fixable—but the referee should be asked to verify Lemma 1 and the L=2 case carefully. I would not cite it until the factor error and edge case are corrected.

Referee Report

4 major / 5 minor

Summary. The paper develops a framework for guaranteed maximum-norm a posteriori error bounds for linear parabolic initial--boundary value problems discretised by the L-step Richardson extrapolation of the backward Euler method in time and conforming FEM in space. The final-time error is represented via the parabolic Green's function; the spatial error is controlled by elliptic reconstructions together with a generic maximum-norm elliptic estimator (Assumption 1), and the temporal error is controlled by a piecewise polynomial reconstruction of degree L-1 built from integrated Legendre polynomials, chosen so that the residual psi + partial_t u_h collapses to its two highest Legendre modes (Eq. (22)). Lemma 1 bounds the resulting mode contributions, and Theorem 1 assembles the claimed computable bound eta_M = eta_init + eta_f + eta_t + eta_ell. The overall strategy is coherent and the displayed computations in the proof of Lemma 1 are essentially correct, but the printed statements of Lemma 1 and Theorem 1 contain defects that, as printed, invalidate the guaranteed-bound claim and require correction.

Significance. If corrected, the paper gives a useful and reusable extension of the authors' second-order analysis [19] to extrapolation of arbitrary order L >= 2. The Legendre-bubble reconstruction with its recursive coefficient choice (Section 4.4.2) is explicit, elegant, and implementable; the estimator contains no fitted parameters and is derived from the PDE, the scheme, and two clearly stated external assumptions, so the approach is genuinely a posteriori and not circular. The main theorem is, however, conditional on an elliptic estimator whose verified instances in the literature are narrower than the paper's advertised scope, which limits the practical reach but not the value of the framework once the defects below are fixed. The manuscript contains no numerical experiments and would be materially strengthened by one simple test problem with reported effectivity indices.

major comments (4)

[Section 4.4.2, Lemma 1 and Eq. (24)] The constant multiplying kappa_1' in the second bound of Lemma 1 is printed as kappa_1' tau_j^2/24, both in the lemma statement and in the displayed bound (24), but the proof's own computation gives kappa_1' tau_j^2/6: after the substitution ||P_i'||_\infty = i(i+1)/2, the bound contains (kappa_1'/tau_j) \int_{I_j} (t_j-s)(s-t_{j-1}) ds, and since this integral equals tau_j^3/6, the surviving constant is kappa_1' tau_j^2/6. Because mu_{j,i} is defined as the minimum of the two candidate bounds, the too-small printed constant can make mu_{j,i}, and hence the temporal estimator eta_t in Theorem 1, four times smaller than the true guaranteed bound. This error is load-bearing: with the printed constant, Theorem 1 does not assert a valid upper bound. The first term in the same displayed bound should also be made consistent between the lemma statement and the proof, which yields (kappa_1/tau_j) \int ... (the typeset formula may be read as kappa_1 tau_j \int ...).
[Section 4.4.4, Theorem 1] The displayed definition of eta_M reads eta_M := eta_init + eta_f + eta_psi + eta_Psi + eta_ell, but the symbols eta_psi and eta_Psi are never defined; the subsequent list defines only eta_init, eta_f, eta_t, and eta_ell. As printed, the equality defining eta_M is unverifiable. The intended temporal term is evidently eta_t, and the display should say so explicitly.
[Section 4.4.4, Theorem 1 (L = 2 case)] The formula for eta_t contains the coefficient 1/(2L-5) in front of Delta_{L-2} psi_j. For L = 2 this coefficient equals -1, whereas the P0 coefficient of psi + partial_t u_h for L = 2 in Eq. (21) is +Delta_0 psi_j + 2 Delta_1 u_j^h / tau_j; moreover, the reduction to the two-mode form (22) is derived only for L > 2, since the recursion defining Delta_2 psi_j does not exist when L = 2. Because the paper explicitly claims L >= 2, Theorem 1 either needs a separate statement for L = 2 based on (21), or should be restricted to L >= 3 with the L = 2 case referred to the authors' earlier treatment [19].
[Section 4.1, Assumption 1; Section 3] The Introduction states that 'any finite-element method might be used' in space, but every eta_ell term of Theorem 1 rests on Assumption 1, which postulates a maximum-norm elliptic estimator for the discrete problem (4) with, in general, quadrature-perturbed forms a_h(.,.) and (.,.)_h. The works cited in Section 4.1 establish such estimators only in restricted settings (specific operators, quasiuniform or anisotropic meshes, and in several cases without quadrature), and the paper proves no such estimator for the general FEM with quadrature described in Section 3. The advertised computable, guaranteed bound is therefore contingent on a premise that is not verified for the method actually analyzed. The authors should either delimit the theorem's scope to FEMs for which Assumption 1 is known to hold or state concrete verified instances, including quadrature if possible.

minor comments (5)

[Section 4.4.4] In the definition of eta_f, the summation is printed as \sum_{m}^{j=1}; it should read \sum_{j=1}^{M} sigma_j \int_{I_j} ||(f - \hat f)(s)||_{\infty,\Omega} ds.
[Section 3] The formula defining the extrapolation weights alpha_ell through the matrix B is garbled in the printed algorithm; please state the weights explicitly, for example as the solution of the Vandermonde system \sum_{\ell=1}^{L} alpha_ell / \ell^k = \delta_{k0} for k = 0,...,L-1, so that the extrapolation coefficients are unambiguous.
[Numerical examples] The manuscript contains no numerical experiments. In view of the paper's claim that the bound is computable and of the approximations described in Remark 3, at least one academic test problem with reported effectivity indices would substantially support the practical claims; I recommend adding one during revision.
[Section 4.4.2] In expansion (18), the coefficients Delta_i phi_j for i >= 2 are introduced only verbally; a sentence stating that they are the Legendre coefficients of the piecewise polynomial phi (or equivalently the coefficients determined by interpolation at the L-2 interior points) would make the reconstruction fully explicit, especially since these coefficients are later determined through (19).
[Various] Typos such as 'differ to some extend' (Introduction) and 'finite dimentional' (Section 3) should be corrected in the revision.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the parabolic error bound is conditional on external elliptic and Green's-function estimates, not on its own conclusion.

full rationale

The paper's derivation chain is conditional, not circular. The target is the parabolic max-norm bound in Theorem 1, and the proof starts from the Green's function representation (5), a standard identity, then bounds each term using two stated assumptions: Assumption 1 (existence of an elliptic max-norm estimator, imported from external literature including [3,6,11,14,17,18,21]) and Assumption 2 (Green's function decay bounds, discussed in [12,13]). Neither assumption is derived from Theorem 1, and Theorem 1 is not used to justify either assumption; the cited prior works are independent published analyses. The quantities eta_init, eta_f, eta_ell and eta_t are computed from the numerical solution, data, and the elliptic estimator; they are not fitted parameters, and no 'prediction' is obtained by renaming an input. The temporal reconstruction in Section 4.4.2 is an explicit construction: the coefficients Delta_i psi_j and Delta_i u_j^h are defined so that psi + partial_t u_h has a prescribed two-term Legendre form; this is a proof device, not a circular identification of the estimator with the error. The algebraic issue in Lemma 1 (kappa_1' tau_j^2/24 versus tau_j^2/6) and the notational inconsistency eta_psi + eta_Psi versus eta_t are correctness concerns, not circularity. Self-citations [12,13,17,18,19] support technical assumptions but do not carry the target claim by themselves, so per Rule 4 they do not raise the circularity score. If Assumption 1 fails for a chosen finite element method, the theorem does not apply, but that is a limitation of the hypothesis, not a circular derivation.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The paper's result is conditional on two external assumptions (elliptic estimator, Green's function bounds) taken from the literature; no parameters are fitted to data, and no new entities are introduced.

assumptions (5)

domain assumption Assumption 1: existence of a maximum-norm a posteriori error estimator η for the elliptic FEM (4).
Invoked in (11) and (20) to bound elliptic reconstruction errors; not proved in this paper, only cited to the literature.
domain assumption Assumption 2: Green's function bounds (6) with constants κ0, κ1, κ1', γ.
Used in Lemma 1 and the spatial error bound in Section 4.4.3; cited to [13,12].
standard math Green's function representation formula (5) for the parabolic solution.
Standard representation for parabolic operators; cited to [8].
domain assumption The solution u of (1) exists and is continuous on the closed space-time domain under the stated regularity and compatibility conditions.
Invoked implicitly to justify pointwise evaluation; cited to [15].
standard math Properties of Legendre polynomials and integrated Legendre polynomials used in Section 4.4.2, including boundary values and derivative norm.
Standard identities; used in the polynomial reconstruction and in Lemma 1.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Maximum-norm a posteriori error bounds for parabolic equations discretised by the extrapolated Euler method in time and FEM in space." pith.science (2026). https://pith.science/paper/WOB33WRA

@misc{pith2026241113617,
  author       = {Pith},
  title        = {Pith review of: Maximum-norm a posteriori error bounds for parabolic equations discretised by the extrapolated Euler method in time and FEM in space},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WOB33WRA}},
  note         = {Machine review of arXiv:2411.13617}
}

read the original abstract

A class of linear parabolic equations is considered. We derive a framework for the a posteriori error analysis of time discretisations by Richardson extrapolation of arbitrary order combined with finite element discretisations in space. We use the idea of elliptic reconstructions and certain bounds for the Green's function of the parabolic operator. The crucial point in the analysis is the design of suitable polynomial reconstructions in time from approximations that are given only in the mesh points.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

22 extracted references · 22 canonical work pages

[19]

T. Linß, M. Ossadnik, and G. Radojev. A uniﬁed approach t o maximum-norm a posteriori er- ror estimation for second-order time discretizations of pa rabolic equations. IMA J. Numer . Anal. , 44(3):1644–1659, 2024

work page 2024
[1]

Bieterman and I

M. Bieterman and I. Babuˇ ska. The ﬁnite element method fo r parabolic equations. I. A posteriori error estimation. Numer . Math., 40(3):339–371, 1982

work page 1982
[2]

Bieterman and I

M. Bieterman and I. Babuˇ ska. The ﬁnite element method fo r parabolic equations. II. A posteriori error estimation and adaptive approach. Numer . Math., 40(3):373–406, 1982

work page 1982
[3]

Demlow and N

A. Demlow and N. Kopteva. Maximum-norm a posteriori erro r estimates for singularly perturbed elliptic reaction-diﬀusion problems. Numer . Math., 133(4):707–742, 2016

work page 2016
[4]

Demlow, O

A. Demlow, O. Lakkis, and C. Makridakis. A posteriori err or estimates in the maximum norm for parabolic problems. SIAM J. Numer . Anal., 47(3):2157–2176, 2009

work page 2009
[5]

Demlow and Ch

A. Demlow and Ch. Makridakis. Sharply local pointwise a p osteriori error estimates for parabolic problems. Math. Comp., 79(271):1233–1262, 2010. 9

work page 2010
[6]

Maxi mum norm a posteriori error estimates for convection-diﬀusion problems

Alan Demlow, Sebastian Franz, and Natalia Kopteva. Maxi mum norm a posteriori error estimates for convection-diﬀusion problems. IMA J. Numer . Anal., 43(5):2562–2584, 2023

work page 2023
[7]

Eriksson and C

K. Eriksson and C. Johnson. Adaptive ﬁnite element metho ds for parabolic problems. II. Optimal error estimates in L∞L2 and L∞L∞. SIAM J. Numer . Anal., 32(3):706–740, 1995

work page 1995

Show all 22 references

[8]

Friedman

A. Friedman. Partial diﬀerential equations of parabolic type. Prentice-Hall, Englewood Cliﬀs, 1964

1964
[9]

Grossmann, H.-G

Ch. Grossmann, H.-G. Roos, and M. Stynes. Numerical treatment of partial di ﬀerential equations. Universitext. Springer, Berlin, 2007

2007
[10]

Johnson, Yi Y ong Nie, and V

C. Johnson, Yi Y ong Nie, and V . Thom´ ee. An a posteriori error estimate and adaptive timestep control for a backward Euler discretization of a parabolic problem. SIAM J. Numer . Anal., 27(2):277–291, 1990

1990
[11]

N. Kopteva. Maximum-norm a posteriori error estimates for singularly perturbed reaction-di ﬀusion problems on anisotropic meshes. SIAM J. Numer . Anal., 53(6):2519–2544, 2015

2015
[12]

Kopteva and T

N. Kopteva and T. Linß. Maximum norm a posteriori error e stimation for parabolic problems using elliptic reconstructions. SIAM J. Numer . Anal., 51(3):1494–1524, 2013

2013
[13]

Kopteva and T

N. Kopteva and T. Linß. Improved maximum-norm a posteri ori error estimates for linear and semi- linear parabolic equations. Adv. Comput. Math., 43(5):999–1022, 2017

2017
[14]

Pointwise a poster iori error estimates for discontinuous Galerkin methods for singularly perturbed reaction-di ﬀusion equations

Natalia Kopteva and Richard Rankin. Pointwise a poster iori error estimates for discontinuous Galerkin methods for singularly perturbed reaction-di ﬀusion equations. SIAM J. Numer . Anal. , 61(4):1938–1961, 2023

1938
[15]

O. A. Ladyzhenskaya, V . A. Solonnikov, and N. N. Ural’ts eva. Linear and quasi-linear equations of parabolic type. American Mathematical Society, 1968

1968
[16]

M. G. Larson and F. Bengzon. The ﬁnite element method: theory, implementation, and appl ications, volume 10 of T exts in Computational Science and Engineering. Springer, Heidelberg, 2013

2013
[17]

T. Linß. Maximum-norm error analysis of a non-monotone FEM for a singularly perturbed reaction- diﬀusion problem. BIT, 47(2):379–391, 2007

2007
[18]

T. Linß. A posteriori error estimation for arbitrary or der FEM applied to singularly perturbed one- dimensional reaction-diﬀusion problems. Appl. Math., 59(3):241–256, 2014

2014
[20]

Makridakis and R

Ch. Makridakis and R. H. Nochetto. Elliptic reconstruc tion and a posteriori error estimates for parabolic problems. SIAM J. Numer . Anal., 41(4):1585–1594, 2003

2003
[21]

R. H. Nochetto, A. Schmidt, K. G. Siebert, and A. V eeser. Pointwise a posteriori error estimates for monotone semi-linear equations. Numer . Math., 104(4):515–538, 2006

2006
[22]

Thom´ ee.Galerkin ﬁnite element methods for parabolic problems , volume 25 of Springer Series in Computational Mathematics

V . Thom´ ee.Galerkin ﬁnite element methods for parabolic problems , volume 25 of Springer Series in Computational Mathematics. Springer-V erlag, Berlin, second edition, 2006. 10

2006

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.

[1] [19]

T. Linß, M. Ossadnik, and G. Radojev. A uniﬁed approach t o maximum-norm a posteriori er- ror estimation for second-order time discretizations of pa rabolic equations. IMA J. Numer . Anal. , 44(3):1644–1659, 2024

work page 2024

[2] [1]

Bieterman and I

M. Bieterman and I. Babuˇ ska. The ﬁnite element method fo r parabolic equations. I. A posteriori error estimation. Numer . Math., 40(3):339–371, 1982

work page 1982

[3] [2]

Bieterman and I

M. Bieterman and I. Babuˇ ska. The ﬁnite element method fo r parabolic equations. II. A posteriori error estimation and adaptive approach. Numer . Math., 40(3):373–406, 1982

work page 1982

[4] [3]

Demlow and N

A. Demlow and N. Kopteva. Maximum-norm a posteriori erro r estimates for singularly perturbed elliptic reaction-diﬀusion problems. Numer . Math., 133(4):707–742, 2016

work page 2016

[5] [4]

Demlow, O

A. Demlow, O. Lakkis, and C. Makridakis. A posteriori err or estimates in the maximum norm for parabolic problems. SIAM J. Numer . Anal., 47(3):2157–2176, 2009

work page 2009

[6] [5]

Demlow and Ch

A. Demlow and Ch. Makridakis. Sharply local pointwise a p osteriori error estimates for parabolic problems. Math. Comp., 79(271):1233–1262, 2010. 9

work page 2010

[7] [6]

Maxi mum norm a posteriori error estimates for convection-diﬀusion problems

Alan Demlow, Sebastian Franz, and Natalia Kopteva. Maxi mum norm a posteriori error estimates for convection-diﬀusion problems. IMA J. Numer . Anal., 43(5):2562–2584, 2023

work page 2023

[8] [7]

Eriksson and C

K. Eriksson and C. Johnson. Adaptive ﬁnite element metho ds for parabolic problems. II. Optimal error estimates in L∞L2 and L∞L∞. SIAM J. Numer . Anal., 32(3):706–740, 1995

work page 1995