Continuous time integration for changing type systems

Sebastian Franz

REVIEW 3 major objections 5 minor 11 references

Continuous time integration for changing type systems

T0 review · 3 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read This paper proves that the continuous Galerkin–Petrov time-stepping method converges with optimal order r+1 in time and k in space, in a cGP-type norm, for evolutionary systems whose type changes between hyperbolic, parabolic, and…

desk verdict First cGP analysis for changing-type systems, but the key convergence proof rests on an invalid extension of the temporal interpolation identity to the full space-time error. read the letter →

arxiv 1908.00728 v3 pith:64VKFGTF submitted 2019-08-02 math.NA cs.NA

Sebastian Franz This is my paper · ORCID

classification math.NAcs.NA MSC 65J0865J1065M1265M60

keywords evolutionaryequationschangingtypesystemcontinuousGalerkinPetrovspace-timeapproacherrorestimatesmixedfiniteelementsvariationaltimeintegrationweightednorms

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

This paper establishes an optimal-order error estimate for a continuous Galerkin–Petrov (cGP) time discretization applied to evolutionary systems whose type changes over the spatial domain. These systems are written in the compact first-order form $(\partial_t M_0 + M_1 + A)U = F$, where $M_0$ and $M_1$ are bounded selfadjoint operators, $A$ is skew-selfadjoint, and the type change between hyperbolic, parabolic, and elliptic regions is encoded by characteristic functions in $M_0$ and $M_1$, so no transmission conditions have to be imposed at the interfaces. For solutions with the weighted time-regularity stated in Theorem 4.2, the fully discrete scheme converges with order $r+1$ in time and order $k$ in space in a cGP-type norm. A further result covers the purely hyperbolic case $M_0>0$ in the weighted $L^2$ norm. The numerical experiments show the predicted high orders for smooth solutions, while for nonsmooth data the observed rates drop to between one and two.

What carries the argument

The argument rests on three pieces. The first is the error decomposition $U-U^\tau_h=\eta+\xi$ with $\eta=U-P_r IU$ and $\xi=P_r IU-U^\tau_h$, where $P_r$ is the temporal interpolation operator fixing both endpoints of each time interval and orthogonal to lower-degree polynomials, and $I$ is the spatial interpolation acting separately on the $H^1$ and $H(\mathrm{div})$ components. The second is the stability estimate for the discrete scheme in the cGP norm, obtained by integration by parts and the structural inequality $\rho M_0+M_1\ge\gamma$. The third is the error equation derived from Galerkin orthogonality: the identity $\langle\partial_t M_0(v-P_r v),w\rangle=2\rho\langle M_0(v-P_r v),w\rangle$ converts the residual of the interpolation error into the algebraic term $(2\rho M_0+M_1+A)\eta$, whose norms are then bounded by the time and space interpolation estimates. This chain reduces the proof of Theorem 4.2 to bounding interpolation errors in the weighted spaces.

What would settle it

On a single time interval, take a smooth $U$ with a nontrivial spatial interpolation part $\eta_2=P_r(U-IU)$ and compute the full expression $\langle(\partial_t M_0+M_1+A)\eta,V_h^\tau\rangle$ after integration by parts, keeping the boundary and $\langle\eta_2,\partial_t M_0 V_h^\tau\rangle$ terms; if the result is not $-\langle(2\rho M_0+M_1+A)\eta,V_h^\tau\rangle$, then the error equation (4.5) and the stability bound built on it are not established. More directly, running the scheme on a smooth known solution with the regularity of Theorem 4.2 should show the rate $r+1$ in the cGP norm; a persistent lower rate would contradict the theorem.

Watch

Extended reading notes

Core claim

The central claim is Theorem 4.2: under the regularity assumptions $U\in H^1_\rho([0,T];H^k)\cap H^{r+1}_\rho([0,T];H)$ and $AU\in H_\rho([0,T];H^k)\cap H^{r+1}_\rho([0,T];H)$, the fully discrete solution $U^\tau_h$ of the cGP method satisfies $$\|\|\|U-U^\tau_h\|\|\|_\rho \le C\big(\$tau^{{r+1}}$(\|\$partial_t^{{r+1}}$U\|_\rho+\|\$partial_t^{{r+1}}$AU\|_\rho)+h^k(\|U\|_{H^k,\rho}+\|AU\|_{H^k,\rho}+\|U(T)\|_{H^k}$e^{{-\rho T}}$+\|NU_0\|_{H^k})\big).$$ The norm here is the cGP norm $\|\|\|v\|\|\|_\rho^2=\tfrac12\|M_0^{1/2}v(T)\|_H^2e^{-2\rho T}+\|Nv(0)\|_H^2+\gamma\|\Pi^\tau_h v\|_\rho^2$, which controls the projected test-space part, the initial nullspace component, and the terminal $M_0$-weighted component. The proof combines Galerkin orthogonality with a stability estimate for the discrete scheme and interpolation estimates in time and space, and it requires no interface or transmission conditions across the subdomains of different type. For the case $M_0>0$ the paper additionally proves a weighted $L^2$ error bound of the same order in time and space.

Load-bearing premise

The proof assumes that the integration-by-parts simplification used for the time-interpolation error also applies to the spatially interpolated part of the error; if that does not hold, the stability estimate and the convergence theorem are not established by the given argument.

Editorial extensions

If this is right

For smooth enough solutions, the cGP method achieves full order $r+1$ in time and $k$ in space in the cGP norm, making it a provably converging continuous time integrator for changing-type systems without interface conditions.
The same analysis applies to any evolutionary problem satisfying $\rho M_0+M_1\ge\gamma$; only the spatial interpolation must be changed when the operator $A$ is different.
In the purely hyperbolic case $M_0>0$, the additional weighted $L^2$ bound shows the scheme controls the solution values themselves, not only the projected test-space norm.
The numerical experiments indicate that the benefit of raising polynomial degrees is limited by the solution's regularity; for the tested problems with coefficient jumps the observed rates remained between 1 and 2, with the discontinuous Galerkin comparison giving smaller errors.

Reading between the lines

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

If the missing terms in Lemma 4.1 turn out to be nonnegligible, the theorem could still be rescued by choosing a spatial interpolation $I$ with additional orthogonality, or by adding correction terms; testing this on a single element would settle the proof.
One testable extension is local mesh refinement near the type-change interfaces: if the observed rate loss is caused by low spatial regularity at the interfaces, refining there should recover higher convergence rates in the cGP norm.
Because the cGP norm weights the terminal value only through $M_0$, the method's control of components in the elliptic region is indirect; a sharper norm or additional regularity would be needed to make pointwise-in-time statements about those components.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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

Dear colleague,

Two things to know. This is the first continuous Galerkin-Petrov (cGP) time-integrator analysis for evolutionary systems of changing type, and the method itself is a natural and honest attempt. But the proof of the main convergence theorem has a specific, load-bearing gap: Lemma 4.1 applies the identity for temporal interpolation errors to the full space-time error eta = U - Pr(IU), which also contains the term Pr(U-IU). That term is not of the form v-Pr(v), so the boundary and derivative contributions do not vanish. The stress-test example (H = R^2, rho = 0, r = 1, with a non-orthogonal spatial interpolant) cleanly exposes the failure of equation (4.5). I think the stress-test note is correct.

What is good: the problem is real, the cGP formulation is a credible alternative to the discontinuous Galerkin method of [6], and the stability theory in Lemma 3.2 looks sound. The interpolation estimates are standard, and if the error equation were valid the stated orders would follow. The numerical examples are honest about the limited regularity of the test problems, and the author says smooth test cases give the theoretical orders, though no data for those smooth cases is shown.

The soft spots, in proportion. The Lemma 4.1 gap is the main issue: without a valid error equation, Theorem 4.2 is not established by the given argument. It may be repairable by a more careful splitting of the error (e.g., handling the spatial interpolation error separately or using a projection that commutes with the time interpolant), but that is not in the paper. Second, the claimed benefit of continuity over the discontinuous method is not quantified. Third, the numerical verification of the optimal rate depends on an unshown smooth-solution experiment.

Who is this for? Researchers working on space-time discretizations of evolutionary problems, especially with degenerate or varying type, will find the formulation worth attention even if the theorem is currently unproven. I would not cite the convergence result, but I would bring the paper to a reading group to discuss the gap and possible repairs.

Recommendation: send it to peer review, with instructions to scrutinize the derivation of Lemma 4.1. The idea and the stability analysis deserve referee time; the flaw is central but looks fixable.

Best, [You]

Referee Report

3 major / 5 minor

Summary. The paper analyzes a continuous Galerkin-Petrov (cGP) time discretization combined with conforming spatial finite elements for evolutionary systems of changing type, written as (∂tM0+M1+A)U=F with bounded selfadjoint M0, M1 and skew-selfadjoint A. The main theoretical contributions are a stability-based existence proof for the discrete solution and an optimal-order error estimate in a cGP-type norm under spatial and temporal regularity assumptions on the exact solution. The paper also reports numerical experiments on two changing-type model problems and compares the method with a discontinuous Galerkin approach. The central error analysis, however, relies on an identity for the full space-time interpolation error that the proof does not justify; the same is true for a key equality in the stability lemma. As a result, the main theorem is not established by the arguments presented.

Significance. If the error estimate were valid, it would be a useful theoretical foundation for a fully discrete cGP method for changing-type systems, complementing the earlier dG analysis in [6] and providing a non-dissipative alternative for such problems. The paper is clearly written and honestly reports that the numerical rates are below the theoretical predictions for nonsmooth solutions. The abstract operator framework connects naturally to the solution theory of [7], and the numerical implementation is reproducible in principle. However, the central proof has load-bearing gaps, so the main claim is currently unsubstantiated and the paper requires substantial revision.

major comments (3)

[Section 4, Lemma 4.1] The proof of Lemma 4.1 applies the identity ⟨∂tM0(v−Prv),w⟩=2ρ⟨M0(v−Prv),w⟩, which is valid for the temporal interpolation error v−Prv, to the full error η=η1+η2 with η2=Pr(U−IU). Since η2 is not of the form v−Prv, the endpoint terms and the term ⟨η2,∂tM0V⟩ do not vanish, so Eq. (4.5) does not follow from the stated argument. The failure is not merely a missing estimate: in an algebraic model problem with H=R², V=span{e1}, I(x,y)=(x+y)e1, M0=I, M1=0, A=0, ρ=0, one interval, r=1, and U(t)=t e2, the discrete solution Uτh=0 gives ξ=t e1 and η=t(e2−e1), and Eq. (4.5) with test function e1 yields left-hand side τ and right-hand side 0. Consequently the stability bound (4.6) and Theorem 4.2 are not established by the proof as written.
[Section 3, Lemma 3.2] The displayed equality Bm(Uτh,ΠτhUτh)=⟨∂tM0Uτh,Uτh⟩+⟨M1ΠτhUτh,ΠτhUτh⟩ is not justified by the projection property (3.1). The argument requires, for instance, that ∂tM0Uτh is orthogonal to Uτh−ΠτhUτh, which would follow if ∂tM0Uτh belonged to the test space; but M0 is a multiplication by characteristic functions of subdomains and does not map the conforming spaces V1⊗V2 into themselves. The same obstruction affects the treatment of the M1 and A terms, since A does not map the trial space into the test space either. Therefore the lower bound (3.2) and the stability estimate (3.3), which support the existence proof, are not proven.
[Theorem 4.2, proof of the initial-value term] The proof asserts ‖N(U−IU)(0)‖H≤Chk‖NU0‖Hk without justification. In the changing-type setting, N is multiplication by a characteristic function of a subdomain and does not commute with the Scott–Zhang or Raviart–Thomas interpolants; moreover, NU0 need not belong to Hk under the stated regularity assumptions because the multiplication introduces jumps at subdomain interfaces. The term should at least be replaced by ‖U0‖Hk, or an additional assumption such as NU0∈Hk must be stated explicitly, together with a proof of the interpolation estimate for NU0.

minor comments (5)

[Section 4, interpolation estimates] In Eq. (4.1), the first inequality should read ‖v−I1v‖0≤Chs‖v‖s, not ‖v‖r.
[Section 4, proof of Theorem 4.2] The equalities ‖Πτhη1‖ρ=‖η1‖ρ and ‖Πτhη2‖ρ=‖η2‖ρ should be inequalities of the form ≤, since Πτh is a projection onto the test space; the subsequent bounds remain valid with ≤.
[Section 2, function spaces] The notation H1ρ([0,T];Hk) in Theorem 4.2 is ambiguous: it should be clarified whether the time derivative is taken with values in H and the spatial regularity is pointwise in time, or whether the Bochner norm uses Hk in space throughout.
[Section 5, tables] The relation between M and N in Tables 1 and 2 is stated as M=2N, but the tables list M values only; the corresponding N values should be given for reproducibility.
[Section 1, introduction] There is a typo in the phrase "changing type problems problems" in the introduction.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation; the error analysis is self-contained and numerical comparison cites prior work only for context.

full rationale

The central claim, Theorem 4.2's optimal-order error estimate for the continuous Galerkin-Petrov time discretization, is obtained from Galerkin orthogonality (4.3), standard interpolation estimates (Scott-Zhang [9], Brezzi-Fortin [5]), and the evolutionary-equation solution theory of Picard [7]. No parameter is fitted to data and later reported as a prediction; the a priori error bound contains only norms of U, AU, and initial data with constants that are not tuned to the numerical examples. The cGP method itself is credited to external sources [1-4,8,11]. The only self-citation is [6], the author's earlier discontinuous-Galerkin paper, which is used for motivation and numerical comparison (Sections 5.1-5.2) and is not invoked in the proofs of Lemma 4.1 or Theorem 4.2, so it is not load-bearing. A possible proof gap in Lemma 4.1 regarding the extension of the temporal interpolation identity to the full interpolation error eta would be a correctness concern, not a circularity, since the claimed error estimate does not reduce to its assumptions by construction.

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

The central bound rests on well-posedness and regularity assumptions from the solution theory, plus an unjustified interpolation identity in the error proof.

assumptions (5)

domain assumption Assumption (1.2): existence of rho0, gamma with ⟨(rho M0 + M1)x, x⟩ >= gamma ||x||^2 for all rho >= rho0.
Imported from Picard's solution theory [7]; guarantees well-posedness and is used in Lemma 3.2 and Theorem 4.2.
domain assumption Compatibility conditions (1.3): F continuous with F=0 for t<0, U0 in D(A), (M1+A)U0=F(0+).
Needed for the continuous solution to have U(0+)=U0; restricts initial data (Corollary 1.1).
domain assumption Regularity assumptions in Theorem 4.2 for U and AU.
The optimal-order bound is conditional on this smoothness; the numerical examples note their solutions likely lack it.
ad hoc to paper The extension of the temporal interpolation identity to the full space-time interpolation error in Lemma 4.1.
Unproved and generally not valid; the main gap in the proof.
domain assumption Mesh resolves Omega_ell, Omega_par, Omega_hyp so that M0 and M1 map V1⊗V2 into itself.
Stated in Section 2; used to justify ⟨M1U,ΠU⟩=⟨M1ΠU,ΠU⟩ in Lemma 3.2.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Continuous time integration for changing type systems." pith.science (2026). https://pith.science/paper/64VKFGTF

@misc{pith2026190800728,
  author       = {Pith},
  title        = {Pith review of: Continuous time integration for changing type systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/64VKFGTF}},
  note         = {Machine review of arXiv:1908.00728}
}

read the original abstract

We consider variational time integration using continuous Galerkin Petrov methods applied to evolutionary systems of changing type. We prove optimal-order convergence of the error in a cGP-like norm and conclude the paper with some numerical examples and conclusions.

Figures

Figures reproduced from arXiv: 1908.00728 by the authors.

**Figure 2.** First component of U at times t = 5`/16 for ` ∈ {1, . . . , 6} (top left to bottom right) of problem (5.2) from [6] that uses globally discontinuous piecewise polynomials of degree r in time and the same approximation in space as the method described in this paper. The errors given in the remaining columns show a similar behaviour with convergence rates between 1 and 2. Nevertheless, the errors are smaller for the d… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

11 extracted references · 10 canonical work pages

[6]

Franz, S

S. Franz, S. Trostorﬀ, and M. Waurick. Numerical methods for changing type systems. IMAJNA, 39(2):1009–1038, 2019

work page 2019
[7]

R. Picard. A structural observation for linear material laws in classical mathematical physics. Math. Methods Appl. Sci. , 32(14):1768–1803, 2009

work page 2009
[1]

Ahmed and G

N. Ahmed and G. Matthies. Higher order continuous Galerkin-Petrov time stepping schemes for transient convection-diﬀusion-reaction equations. ESAIM Math. Model. Nu- mer. Anal., 49(5):1429–1450, 2015

work page 2015
[2]

Akrivis and Ch

G. Akrivis and Ch. Makridakis. Galerkin time-stepping methods for nonlinear parabolic equations. M2AN Math. Model. Numer. Anal. , 38(2):261–289, 2004

work page 2004
[3]

Akrivis, Ch

G. Akrivis, Ch. Makridakis, and R.H. Nochetto. Galerkin and Runge-Kutta methods: uniﬁed formulation, a posteriori error estimates and nodal superconvergence. Numer. Math., 118(3):429–456, 2011. 13 CTS_cGP_ETNA_v5_arxiv January 6, 2021

work page 2011
[4]

A. K. Aziz and P. Monk. Continuous ﬁnite elements in space and time for the heat equation. Math. Comp, 52(186):255–274, 1989

work page 1989
[5]

Brezzi and M

F. Brezzi and M. Fortin. Mixed and hybrid ﬁnite element methods , volume 15 of Springer Series in Computational Mathematics . Springer-Verlag, New York, 1991

1991
[8]

Schieweck

F. Schieweck. A-stable discontinuous Galerkin-Petrov time discretization of higher order. J. Numer. Math. , 18(1):25–57, 2010

work page 2010

Show all 11 references

[9]

Scott and S

L.R. Scott and S. Zhang. Finite element interpolation of nonsmooth functions satisfying boundary conditions. Math. Comp., (54):483–493, 1990

1990
[10]

Trostorﬀ and M

S. Trostorﬀ and M. Waurick. On higher index diﬀerential-algebraic equations in inﬁnite dimensions. In A. B¨ ottcher, D. Potts, P. Stollmann, and D. Wenzel, editors,The Diversity and Beauty of Applied Operator Theory , pages 477–486, Cham, 2018. Springer Interna- tional Publishing

2018
[11]

R. Winther. A stable ﬁnite element method for initial-boundary value problems for ﬁrst- order hyperbolic systems. Math. Comp., 36(153):65–86, 1981. 14

1981

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [6]

Franz, S

S. Franz, S. Trostorﬀ, and M. Waurick. Numerical methods for changing type systems. IMAJNA, 39(2):1009–1038, 2019

work page 2019

[2] [7]

R. Picard. A structural observation for linear material laws in classical mathematical physics. Math. Methods Appl. Sci. , 32(14):1768–1803, 2009

work page 2009

[3] [1]

Ahmed and G

N. Ahmed and G. Matthies. Higher order continuous Galerkin-Petrov time stepping schemes for transient convection-diﬀusion-reaction equations. ESAIM Math. Model. Nu- mer. Anal., 49(5):1429–1450, 2015

work page 2015

[4] [2]

Akrivis and Ch

G. Akrivis and Ch. Makridakis. Galerkin time-stepping methods for nonlinear parabolic equations. M2AN Math. Model. Numer. Anal. , 38(2):261–289, 2004

work page 2004

[5] [3]

Akrivis, Ch

G. Akrivis, Ch. Makridakis, and R.H. Nochetto. Galerkin and Runge-Kutta methods: uniﬁed formulation, a posteriori error estimates and nodal superconvergence. Numer. Math., 118(3):429–456, 2011. 13 CTS_cGP_ETNA_v5_arxiv January 6, 2021

work page 2011

[6] [4]

A. K. Aziz and P. Monk. Continuous ﬁnite elements in space and time for the heat equation. Math. Comp, 52(186):255–274, 1989

work page 1989

[7] [5]

Brezzi and M

F. Brezzi and M. Fortin. Mixed and hybrid ﬁnite element methods , volume 15 of Springer Series in Computational Mathematics . Springer-Verlag, New York, 1991

1991

[8] [8]

Schieweck

F. Schieweck. A-stable discontinuous Galerkin-Petrov time discretization of higher order. J. Numer. Math. , 18(1):25–57, 2010

work page 2010