Optimal control problems for quasi-linear parabolic equations and their linear approximations

Meisai Wang; Xu Liu; Xu Zhang

REVIEW 3 major objections 5 minor 15 references

Optimal control problems for quasi-linear parabolic equations and their linear approximations

T0 review · 3 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read This paper proves that, for small data, the optimal control problem of a quasi-linear parabolic equation can be replaced by the linear heat-equation LQ problem, with error of second order in the data when the optimal controls are interior p

desk verdict Genuinely new quantitative comparison between quasi-linear and linear parabolic optimal control, with a sharp quadratic interior estimate; two proof-level gaps need referee attention but look patchable. read the letter →

arxiv 2607.26585 v1 pith:XRVISUA2 submitted 2026-07-29 math.OC

Xu Liu , Meisai Wang , Xu Zhang This is my paper

classification math.OC MSC 35K5949J2049K2093C20

keywords quasi-linearparabolicequationslinearoptimalcontroltrajectorylinear-quadraticapproximationsmall-datainteriorvsboundarycontrolssharporder

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 studies whether a controlled quasi-linear parabolic equation—describing, for example, heat diffusion with temperature-dependent conductivity—can be replaced by the classical linear heat equation when solving an optimal control problem. It proves that for sufficiently small initial data and targets, the optimal controls and optimal trajectories of the two problems are close: the gap is of order 1+θ in the data in general, and of order 2 when the optimal controls lie inside the admissible set. The quadratic order is shown to be the best possible in general, because the linear equation is exactly the first-order linearization of the quasi-linear one at zero, so all linear-order behavior matches and the first genuine difference is quadratic. The paper also gives explicit criteria, based on the ratio β/α of control cost to tracking cost, that tell when the optimal control is an interior point rather than on the boundary of the control ball. If correct, this justifies using the simpler linear (LQ) problem as a stand-in for the nonlinear one in small-data thermal control applications.

What carries the argument

The key mechanism is the optimality system for each problem: for the quasi-linear problem, the pair (y,η) satisfies a coupled forward-backward system (4.1) with the control given by u=β^{-1}η, while the linear problem has the analogous system (4.2) with control v=β^{-1}z. The difference (Y,W)=(y−w,η−z) therefore solves a linear system whose source terms F1 and F2 are products of the already-small state and adjoint variables, because aij(y)−aij(0) and f(y), f'(y) vanish like |y| at y=0. A bootstrap through Sobolev embedding raises the integrability from L^2 to the desired L^p, yielding the quadratic bound. The structural identity that makes all this work is that the linear equation is the fir

What would settle it

For the sharpness example (Ω=(0,π), f(s)=s^2, y0=yd=δ sin x, β large), compute the first-order state y1 and check whether it is positive on (π/2,3π/4)×(0,T]; then compute |u−v|_{L^2(Ω×(0,T))} for decreasing δ and verify it is bounded below by a positive constant times δ^2. If the sign or the lower bound fails, the optimality of order two in Remark 1.1 needs additional hypotheses.

Watch

Extended reading notes

Core claim

The central discovery is that the optimal control problem for the quasi-linear parabolic equation (1.1) and the linear heat equation (1.4) share their first-order response to the data. Under the assumption f(0)=f'(0)=0 and small data, the difference between the quasi-linear optimal pair (u,y) and the linear optimal pair (v,w) satisfies |u−v|_{L^p}+|y−w|_{W^{2,1}_p} ≤ C(|y_d|^2 + |y_0|^2) when both controls are interior points of the admissible set Uad (Theorem 1.2), and this quadratic rate cannot be improved in general (Remark 1.1). In the general case, including boundary controls, the gap is of order 1+θ (Theorem 1.1). The proof couples the two optimality systems and shows, by a bootstrap a

Load-bearing premise

The main estimates rest on the assumption that the linear equation is the exact first-order approximation of the quasi-linear one at the zero state (via f(0)=f'(0)=0 and small data); for the sharp lower-bound example, the load-bearing premise is that the first-order solution stays positive in a certain region, a point the paper asserts without a detailed proof.

Editorial extensions

If this is right

For small initial data and targets, one can compute the optimal control for the linear heat equation and use it for the quasi-linear problem; the error in control and trajectory is bounded by a constant times the square of the data size.
In the interior-point case, the quadratic rate is best possible: no general improvement beyond order two is possible unless the nonlinear terms have extra degeneracy or cancellation.
The ratio β/α of control-cost to tracking-cost weight determines whether the optimal control is interior or on the boundary, telling practitioners which approximation order (quadratic or only 1+θ) applies.
The same quadratic approximation holds when the tracking term in the cost uses an L^p norm instead of L^2, and the method is designed to extend to broader quasi-linear parabolic and hyperbolic control problems.
The result gives a rigorous justification for the common engineering practice of replacing a nonlinear thermal-diffusion model by its linear approximation in optimal-control design.

Reading between the lines

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

If the boundary case could be handled with an improved, possibly optimal, rate, the practical range of small-data linearization would widen; the paper leaves this open.
The sharpness example indicates a broader heuristic: whenever a nonlinear model is linearized at an equilibrium with nonzero quadratic nonlinearity, model-in-the-loop optimal controls inherit a second-order gap; the same mechanism should appear in other PDE-constrained optimization problems.
A testable extension: for nonlinearities whose quadratic part also vanishes, e.g., f(s)=s^3, the proof's logic suggests the approximation order may increase (to cubic) unless cross-terms in the diffusion coefficient produce quadratic contributions; computing this case numerically would probe the limit of the result.
The verification criteria in Proposition 1.1 can be read as an a posteriori test: an estimate of the adjoint η can be used in an algorithm to detect whether the optimal control is hitting its bound, which is useful for numerical solvers.

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

First thing to know: the main theorem is probably right, but the proof as written has a real gap in its bootstrap, and the sharpness claim in the abstract rests on a sketch. Neither looks fatal; both need fixing before the paper is citable as is.

What is new: this is the first quantitative statement I know of showing that, for small data, the quasi-linear parabolic optimal control problem can be replaced by its linear LQ counterpart. Theorem 1.1 gives a higher-order (1+θ) error bound when controls may hit the boundary; Theorem 1.2 improves to a quadratic bound in the interior-point case, in the same Lp and W^{2,1}_p spaces as the statement. The quadratic order is the natural one — the linear model is the first-order linearization around zero, so under f(0)=f'(0)=0 the first-order errors cancel. Proposition 1.1's verification criteria for interior versus boundary controls are also useful. The proofs are honest: the estimates flow from the optimality systems (4.1)–(4.2), and the Lp bounds on the nonlinear terms are careful and plausible.

The stress-test concern about Theorem 1.2's proof is valid as written: the iteration asserts |y|_{W^{2,1}_{p1}} ≤ C(|η|_{Lp1} + |y0|_{W^{2,p1}}) with p1 = 2(n+2)/(n−2), but the theorem only assumes y0 ∈ W^{2,p}. When p < p1 (say n=3, p=6, p1=10), the W^{2,p1} norm of y0 need not be finite, so the step is unjustified. I think the theorem survives, though: in the p < p1 case the bootstrap is an inessential detour, since the W^{2,1}_2 bound, Sobolev embedding, and estimates (1.5) and (2.4) close the argument directly. In the p ≥ p1 case the step is fine because W^{2,p} ⊂ W^{2,p1} and Lp ⊂ L^{p1}. The fix is a case split, but a referee should require it.

Second, Remark 1.1's sharpness example is a sketch: it assumes an O(δ³) expansion of the optimality system and asserts y1 > 0 on a subdomain without proof, and 'ρ' is undefined. The contradiction idea is plausible but not a proof. Since the abstract advertises sharpness, this should be completed or downgraded to a heuristic.

Bottom line: a genuine contribution for people working on parabolic control and small-data model reduction. Send it to a careful referee, and require the bootstrap fix and a real lower-bound proof. The main comparison theorem is worth having even if the sharpness claim is softened.

Referee Report

3 major / 5 minor

Summary. The paper studies two optimal control problems: a quasi-linear parabolic equation (P1) and its linearization, the heat equation (P2), with quadratic costs and the same Lp-ball admissible set. It claims (i) verification criteria for interior vs boundary optimal controls (Prop. 1.1), (ii) a general Hölder-type approximation error O(δ^{1+θ}) (Theorem 1.1), and (iii) a quadratic approximation error O(δ²) when the optimal controls are interior (Theorem 1.2), with an example intended to prove sharpness (Remark 1.1), plus an analogous result for higher-order tracking costs (Theorem 1.3). The proofs use optimality systems, Lp estimates and an energy/bootstrap argument.

Significance. The motivation is natural and the main statement is appealing: for small data, a quasi-linear optimal control problem may be replaced by an LQ problem with controlled error. The paper contains several correct and useful ingredients: the optimality systems (4.1)–(4.2), the L2-based estimate in Theorem 1.1, and the Proposition 1.2 state approximation. However, two load-bearing parts are not yet justified: the bootstrap in Theorem 1.2 requires more regularity of the data than assumed, and the sharpness example is only heuristic. If the technical gaps are repaired (e.g., by adding a lower bound on p or an alternative regularity argument), the paper would make a solid contribution.

major comments (3)

[§2, proof of Proposition 1.1(1)] The proof of the interior-point criterion is not valid as stated. In Step 3 the estimate |η|_{W^{2,1}_p} ≤ C1α(|u|_{Lp}+|y0|_{W^{2,p}}+|yd|_{Lp}) ≤ C1αρ1 uses |yd|_{Lp} ≤ ρ1, an assumption absent from Proposition 1.1. Without smallness of yd, the threshold β/α>ρ1* cannot be independent of yd: for a large target the optimal control of the LQ problem saturates the bound |v|_{Lp}≤ρ1 even for large β/α. The statement should either include a smallness condition on yd (and y0) or allow ρ1* to depend on the data.
[§4, proof of Theorem 1.2 (bootstrap after (4.5))] The bootstrap from W^{2,1}_2 to W^{2,1}_p is unjustified for n>2 and p<p1=2(n+2)/(n−2). The proof invokes well-posedness for (4.1) to write |y|_{W^{2,1}_{p1}} ≤ C(|η|_{L^{p1}}+|y0|_{W^{2,p1}}) and |η|_{W^{2,1}_{p1}} ≤ C(|y0|_{W^{2,p1}}+|yd|_{L^{p1}}), but Theorem 1.2 assumes only y0∈W^{2,p} and yd∈L^p with p>n+2. For n=3, p=6, p1=10 neither embedding holds. Hence the final estimate with |y0|_{W^{2,p}}+|yd|_{L^p} does not follow. Either the theorem must assume p≥p1 (or a comparable condition), or the proof must be replaced by an argument that avoids the W^{2,p1} regularity of the data.
[Remark 1.1] The sharpness claim is not established. The argument assumes an asymptotic expansion u=δu1+δ²u2+O(δ³), y=δy1+δ²y2+O(δ³), η=δη1+δ²η2+O(δ³) without justification; the solution map of the optimality system is not proved to be δ-analytic. The positivity assertion y1>0 in (π/2,3π/4)×(0,T] is stated with the undefined parameter ρ ('when ρ is sufficiently small') and no proof; it is essential for the conclusion |y2(·,T)|_{L²}>0. Until these points are supplied, the lower bound |u−v|_{L²}≥C0δ² remains heuristic.

minor comments (5)

[Appendix, proof of Proposition 1.2] In the final displayed estimate, the control norm should be |u|²_{L^p(ω×(0,T))}, not |u|²_{L^p(Q)}.
[Remark 1.1] The symbol ρ in 'when ρ is sufficiently small' is never defined. If it denotes T, δ, or a smallness radius, it should be stated explicitly.
[§4, proof of Theorem 1.2] The elimination of the √ε|W|_{L²(Q)} term in the energy estimate before (4.5) should be justified explicitly; for instance by |W|_{L²(Q)}≤|W|_{W^{2,1}_2(Q)} and then absorbing into the left-hand side.
[Theorem 1.1 and Theorem 1.2] The paper does not prove uniqueness of the optimal control for the quasi-linear problem (P1), but the statements refer to 'the optimal pair'. If uniqueness is not known, the estimates should be formulated for any optimal control or a uniqueness result should be supplied.
[Proposition 1.1(2)] The phrase 'for each sufficiently small ρ1' should state explicitly that ρ2* may depend on ρ1, d0, T, Ω, and the nonlinear data.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the quasi-linear/linear approximation is derived from the coupled optimality systems and Lp estimates, not from its own hypotheses.

full rationale

The paper's central claim is not circular. The linear problem (P2)/(1.4) is introduced independently as the heat equation with the same initial datum, target and cost structure; the approximation estimate is not assumed in the hypotheses. Theorem 1.2 is proved from the interior-point optimality systems (4.1)-(4.2), with the difference (Y,W)=(y-w, η-z) satisfying the forced linear system (4.4). The quadratic source bounds in (4.5) follow from Assumption (H) (f(0)=f'(0)=0 and aij(0)=δij) together with Lp estimates, and the final estimate is obtained by linear parabolic well-posedness, not by any fitted parameter or by assuming the desired conclusion. The sharpness example in Remark 1.1 constructs a lower bound by matching δ- and δ²-terms in the optimality system and proving η2≠0 by contradiction; this is not circular, although it does contain proof gaps (unproved positivity of y1, an undefined ρ, and a bootstrap step in Theorem 1.2 requiring regularity not assumed in the theorem). Those are correctness concerns, not circularity. The only self-citation, reference [11], is not load-bearing; the well-posedness citation [8] is external classical parabolic theory. No self-definitional, fitted-input-as-prediction, uniqueness-import, ansatz-smuggling, or renaming pattern is present. Therefore the appropriate circularity score is 0.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

The main theorems rely on standard parabolic Lp estimates, Sobolev embeddings, and the stated regularity/ellipticity hypotheses. The only ad hoc element is the asymptotic-expansion and positivity assumption in the sharpness remark.

assumptions (6)

standard math Lp-regularity and Schauder fixed point theory for quasi-linear parabolic equations (Ladyzhenskaya et al. [8])
Used to assert local well-posedness and estimates (1.5), the starting point for all control estimates.
domain assumption Uniform ellipticity and regularity aij,f ∈ C², f(0)=0, p>n+2, bounded smooth domain
Needed for strong W^{2,1}_p solutions and C^{1,0} embedding; stated in Section 1.
domain assumption Assumption (H): f(0)=f'(0)=0
Makes the heat equation the exact first-order linearization and is used in every quadratic-order estimate; if f'(0)≠0 the linear model should include f'(0)y.
standard math Sobolev embeddings and interpolation inequalities
Used throughout Theorems 1.1-1.3 to bootstrap from L² to L^p estimates.
standard math Comparison principle for linear parabolic equations
Used in Proposition 1.1 Step 3 to construct sub/super-solutions and locate boundary controls.
ad hoc to paper Smooth asymptotic expansion in δ for the sharpness example
Remark 1.1 assumes u=δu1+δ²u2+O(δ³) etc. and y1>0 without proof; the lower-bound Cδ² depends on this.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Optimal control problems for quasi-linear parabolic equations and their linear approximations." pith.science (2026). https://pith.science/paper/XRVISUA2

@misc{pith2026260726585,
  author       = {Pith},
  title        = {Pith review of: Optimal control problems for quasi-linear parabolic equations and their linear approximations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XRVISUA2}},
  note         = {Machine review of arXiv:2607.26585}
}

read the original abstract

Motivated by thermal diffusion models, a controlled quasi-linear parabolic equation can be formally replaced by its linear approximation, when the initial datum, target and control actions are sufficiently small. This paper makes this approximation precise at the level of optimal control problems. First, we provide verification criteria which distinguish whether an optimal control is an interior point or a boundary point of the admissible control set. We then prove that the differences between the optimal controls and the corresponding optimal trajectories for the quasi-linear and linear parabolic equations are higher-order infinitesimals with respect to the small initial datum and target appearing in the cost functional. Consequently, for small data, the optimal control problem governed by the quasi-linear equation can be approximated by the corresponding linear problem. In the interior-point case, the resulting approximation order is sharp.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

15 extracted references

[1]

Barbu, Convexity and optimization in Banach spaces, F ourth edition, Springer Monogr

V. Barbu, Convexity and optimization in Banach spaces, F ourth edition, Springer Monogr. Math., Springer, 2012

2012
[2]

Bonifacius and I

L. Bonifacius and I. Neitzel, Second order optimality co nditions for optimal control of quasi- linear parabolic equations, Math. Control Relat. Fields, 8 (2018), no. 1, 1–34

2018
[3]

Casas and K

E. Casas and K. Chrysaﬁnos, Analysis and optimal control of some quasilinear parabolic equations, Math. Control Relat. Fields, 8 (2018), no. 3-4, 6 07–623

2018
[4]

Casas and K

E. Casas and K. Kunisch, Inﬁnite horizon optimal control problems for a class of semilinear parabolic equations, SIAM J. Control Optim., 60 (2022), no. 4, 2070–2094

2022
[5]

H. O. Fattorini, Inﬁnite-dimensional optimization and control theory, Cambridge University Press, 1999

1999
[6]

Frankowska and Q

H. Frankowska and Q. L¨ u, Second order necessary conditi ons for optimal control problems of evolution equations involving ﬁnal point equality constra ints, ESAIM Control Optim. Calc. Var., 27 (2021), Paper No. 71, 38 pp

2021
[7]

Hoppe and I

F. Hoppe and I. Neitzel, Optimal control of quasilinear p arabolic PDEs with state- constraints, SIAM J. Control Optim., 60 (2022), no. 1, 330–354

2022
[8]

O. A. Ladyˇ zenskaja, V. A. Solonnikov and N. N. Ural’ceva , Linear and quasilinear equations of parabolic type, Transl. Math. Monogr., Vol. 23, American Mathematical Society, 1968. 23

1968

Show all 15 references

[9]

Li and J

X. Li and J. Yong, Optimal control theory for inﬁnite-dim ensional systems, Birkh¨ auser Boston, Inc., 1995

1995
[10]

Lions, Optimal control of systems governed by par tial diﬀerential equations, Die Grundlehren der mathematischen Wissenschaften, Band 170, Springer-Verlag, 1971

J.-L. Lions, Optimal control of systems governed by par tial diﬀerential equations, Die Grundlehren der mathematischen Wissenschaften, Band 170, Springer-Verlag, 1971

1971
[11]

X. Liu, Q. L¨ u and X. Zhang, Finite codimensional contro llability and optimal control problems with endpoint state constraints, J. Math. Pures Appl., 138 ( 2020), 164–203

2020
[12]

Lou, Optimality conditions for semilinear paraboli c equations with controls in leading term, ESAIM Control Optim

H. Lou, Optimality conditions for semilinear paraboli c equations with controls in leading term, ESAIM Control Optim. Calc. Var., 17 (2011), no. 4, 975–994

2011
[13]

Tr´ elat, C

E. Tr´ elat, C. Zhang and E. Zuazua, Steady-state and per iodic exponential turnpike property for optimal control problems in Hilbert spaces, SIAM J. Cont rol Optim., 56 (2018), no. 2, 1222–1252

2018
[14]

Tr¨ oltzsch, Optimal control of partial diﬀerential e quations, Theory, methods and applica- tions, Grad

F. Tr¨ oltzsch, Optimal control of partial diﬀerential e quations, Theory, methods and applica- tions, Grad. Stud. Math., 112, American Mathematical Socie ty, 2010

2010
[15]

G. Wang, L. Wang, Y. Xu and Y. Zhang, Time optimal control of evolution equations, Progr. Nonlinear Diﬀerential Equations Appl., 92, Subser. Control , Birkh¨ auser/Springer, 2018. 24

2018

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.

[1] [1]

Barbu, Convexity and optimization in Banach spaces, F ourth edition, Springer Monogr

V. Barbu, Convexity and optimization in Banach spaces, F ourth edition, Springer Monogr. Math., Springer, 2012

2012

[2] [2]

Bonifacius and I

L. Bonifacius and I. Neitzel, Second order optimality co nditions for optimal control of quasi- linear parabolic equations, Math. Control Relat. Fields, 8 (2018), no. 1, 1–34

2018

[3] [3]

Casas and K

E. Casas and K. Chrysaﬁnos, Analysis and optimal control of some quasilinear parabolic equations, Math. Control Relat. Fields, 8 (2018), no. 3-4, 6 07–623

2018

[4] [4]

Casas and K

E. Casas and K. Kunisch, Inﬁnite horizon optimal control problems for a class of semilinear parabolic equations, SIAM J. Control Optim., 60 (2022), no. 4, 2070–2094

2022

[5] [5]

H. O. Fattorini, Inﬁnite-dimensional optimization and control theory, Cambridge University Press, 1999

1999

[6] [6]

Frankowska and Q

H. Frankowska and Q. L¨ u, Second order necessary conditi ons for optimal control problems of evolution equations involving ﬁnal point equality constra ints, ESAIM Control Optim. Calc. Var., 27 (2021), Paper No. 71, 38 pp

2021

[7] [7]

Hoppe and I

F. Hoppe and I. Neitzel, Optimal control of quasilinear p arabolic PDEs with state- constraints, SIAM J. Control Optim., 60 (2022), no. 1, 330–354

2022

[8] [8]

O. A. Ladyˇ zenskaja, V. A. Solonnikov and N. N. Ural’ceva , Linear and quasilinear equations of parabolic type, Transl. Math. Monogr., Vol. 23, American Mathematical Society, 1968. 23

1968