Boundary sampled-data feedback stabilization for parabolic equations

Hanbing Liu

REVIEW 2 major objections 4 minor 19 references

Boundary sampled-data feedback stabilization for parabolic equations

T0 review · 2 major / 4 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 Sampled-data boundary feedback stabilizes parabolic equations at any sampling rate.

desk verdict A genuine sampled-data extension of Munteanu's boundary feedback with arbitrary sampling period, but the nonlinear stabilization proof has a real gap for part of the stated hypothesis and the promised numerics are missing. read the letter →

arxiv 1908.03100 v1 pith:RWTLFAQ5 submitted 2019-08-08 math.OC

Hanbing Liu This is my paper · ORCID

classification math.OC MSC 93C2093D1593C5735K58

keywords parabolicequationssampled-datacontrolboundaryfeedbackstabilizationsemilinearfinite-dimensionaleigenfunctionexpansionexponentialstabilityDirichlet

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 designs an explicit finite-dimensional boundary feedback controller of sampled-data form to stabilize equilibrium solutions of semilinear parabolic equations. The controller is built from the eigenfunctions corresponding to the unstable eigenvalues of the linearized operator and works for any sampling period $T$, not necessarily small. The main result says that every initial datum sufficiently close to the equilibrium produces a solution that decays exponentially in $H^{1/2-\varepsilon}$, at any prescribed rate $\mu<\rho$. The same feedback globally stabilizes the linearized equation at rate $\rho$, and it tends to the continuous-time controller of [8,9] as $T\to0$.

What carries the argument

The load-bearing object is the boundary feedback operator $F(w)=\mathbf{1}_{\Gamma_1}\langle \Lambda B Q_N(w), \partial\Phi_N/\partial n\rangle_N$, assembled from a weighted sum $\Lambda=\sum_{k=1}^N\Lambda_{\gamma_k}$, the invertible matrix $B=(B_1+\cdots+B_N)^{-1}$ with $B_k=\Lambda_{\gamma_k}B_0\Lambda_{\gamma_k}$, $B_0=(\langle\partial\varphi_i/\partial n,\partial\varphi_j/\partial n\rangle_0)$, and the elliptic lifting maps $D_{\gamma_k}:L^2(\Gamma_1)\to H^{1/2}(\Omega)$ defined by (2.9). The argument turns on the discrete identity $z_N((i+1)T)=\sum_{k=1}^N e^{-\gamma_k T}B_kB z_N(iT)$ for the controlled low modes, which yields the contraction $|B^{1/2}z_N((i+1)T)|\le e^{-\gamma_1 T}|B^{1/2}z_N(iT)|$; the high-frequency modes decay at rate $\rho$, and the nonlinear terms are controlled by a Gronwall argument over blocks of length $KT$.

What would settle it

Take an explicit multidimensional example, such as the unit square with a known unstable equilibrium, compute the normal-derivative matrix $B_0$ and the weighted matrices $B_k$, and check numerically whether $B_1+\cdots+B_N$ is invertible and whether the spectral inequality $\lambda_{\max}(\sum_k e^{-\gamma_k T}B^{1/2}B_kB^{1/2})\le e^{-\gamma_1 T}$ holds. If either fails, the contraction step collapses. Alternatively, simulate the closed-loop nonlinear system with a large sampling period $T$ and a small initial deviation; the theorem predicts $\|y(t)-y_e\|_{1/2-\varepsilon}\le Ce^{-\mu t}\|y_0-y_e\|_{1/2-\varepsilon}$, so observing growth between sampling instants for arbitrarily small initial data would refute it.

Watch

Extended reading notes

Core claim

The central claim is Theorem 3.1: for any $0<\mu<\rho$, under assumption $(H_f)$, there exist constants $C>0$ and $\delta>0$ such that for every $y_0$ with $\|y_0-y_e\|_{1/2-\varepsilon}\le\delta$, the sampled-data feedback $u(t)=\sum_i\chi_{[iT,(i+1)T)}(t)F(y(iT)-y_e)+y_e$ locally exponentially stabilizes the semilinear parabolic equation, with $\|y(t)-y_e\|_{1/2-\varepsilon}\le C e^{-\mu t}\|y_0-y_e\|_{1/2-\varepsilon}$. The proof first shows that the same feedback globally stabilizes the linearized equation (Theorem 2.2) at rate $\rho$, then treats the nonlinear remainder under the polynomial growth condition $(H_f)$ as a perturbation that can be absorbed by choosing the initial data small and the block length large. The controller is finite-dimensional, using only the $N$ eigenfunctions below the threshold $\rho$, and the controlled low modes contract by a factor $e^{-\gamma_1 T}$ at each sampling step, so no smallness of $T$ is required.

Load-bearing premise

The construction requires the elliptic lifting maps $D_{\gamma_k}$ to be well posed with a coercivity estimate and the matrix $B_1+\cdots+B_N$ to be invertible; the paper states both facts without proofs, and if either fails the feedback operator is not defined.

Editorial extensions

If this is right

The same sampled-data feedback globally stabilizes the linearized parabolic equation at exponential rate $\rho$, for any sampling period $T$.
For the semilinear equation under $(H_f)$, local exponential stability holds at any prescribed rate $\mu<\rho$, with the admissible size $\delta$ depending on $T$ and $\mu$.
The controller is explicit and finite-dimensional: it requires only the $N$ eigenfunctions of the linearized operator with eigenvalues below $\rho$, together with their normal derivatives on the controlled boundary.
As $T\to0$, the sampled-data law recovers the continuous-time boundary feedback of [8,9], so the construction is a genuine sampled-data extension rather than a separate design.
Because the gains $\Lambda_{\gamma_k}$ are recomputed for each $T$, the same formula applies to large sampling periods instead of requiring $T$ to be small.

Reading between the lines

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

Not developed in the paper: the proof displays a quantitative trade-off between the sampling period and the contraction factor $e^{-\gamma_1 T}$, so one could tune $\gamma_1$ to keep a desired decay while increasing $T$; the paper does not optimize this choice.
A natural extension would be an output-feedback sampled-data version that estimates the $N$ low modes from boundary observations, following the same low-mode philosophy; the paper stops at state feedback.
The proof's feedback depends on $T$ through the matrices $\Lambda_{\gamma_k}$, but the numerical size of the stability radius $\delta$ as a function of $T$ is left qualitative; a reader tuning the controller would need to compute it for a concrete case.

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

The paper does something real: it takes Munteanu's finite-dimensional boundary feedback and shows the same construction works as a zero-order-hold sampled-data controller for arbitrary sampling period T, with the feedback tending to the continuous-time controller as T goes to 0. The linear part of the paper is solid. The discrete-time recurrence (2.30), the semipositivity argument, and the decay estimate (2.32) all check out. This is a genuine step beyond the 1-D or small-sampling or globally Lipschitz results in [14,15,16], and the explicit form of the feedback is valuable.

The nonlinear theorem is where the trouble is. The stress-test note is correct: the step leading to (3.8) uses ∫‖z‖^{2p_i} ds ≤ η^{2(p_i−1)}∫‖z‖² ds on the set where ‖z‖≤η, and that inequality only holds for p_i≥1. Hypothesis (Hf) explicitly allows 0

Two smaller issues. First, the feedback F is only defined if the elliptic lifting problem (2.9) is well-posed and if the matrix B1+...+BN is invertible. The paper says these follow by "not difficult" estimates and "following the method in [8]", but gives no proof and no precise citation. This is likely fixable, but as written it is an unproved load-bearing assertion. Second, the abstract and introduction promise numerical examples in Section 4; the paper ends at the references. That is a missing advertised section, not a fatal flaw, but it should be supplied or the promise removed.

Who is this for: people working on sampled-data boundary control of PDEs, especially those using Munteanu's lifting method. It deserves a serious referee—the linear result and the overall strategy are worth engaging—but I would expect major revision before acceptance. I would not cite it in its current form.

Referee Report

2 major / 4 minor

Summary. The paper designs an explicit finite-dimensional boundary feedback controller of sampled-data form for semilinear parabolic equations with Dirichlet boundary conditions on part of the boundary. The controller is built from the unstable eigenfunctions of the linearized operator, uses an arbitrary sampling period T, and is shown to recover the continuous-time controller of [8,9] as T→0. The main results are a global exponential stabilization theorem for the linearized equation (Theorem 2.2) and a local exponential stabilization theorem for the semilinear equation around an equilibrium (Theorem 3.1) under a polynomial growth condition (Hf). The proof proceeds by lifting the sampled boundary data through elliptic problems Dγk, reducing the controlled linear equation to the finite-dimensional recurrence (2.30), and then treating the nonlinearity by an a priori bound and a discrete-time contraction argument.

Significance. If the technical gaps are repaired, the paper makes a useful contribution to constructive PDE control: it gives an explicit finite-dimensional feedback that works for any sampling period rather than only for small T, and it displays the convergence of the sampled-data law to a known continuous-time controller. The linear part of the paper is largely sound: the algebraic derivation of the discrete-time recurrence (2.30) and the spectral radius estimate (2.31) are consistent, and the high-frequency estimate in Theorem 2.2 follows along standard dissipativity lines. The paper does not provide machine-checked proofs or reproducible code, and its main nonlinear closing argument needs essential repair; nevertheless, the explicit construction and the continuous-time limit are valuable features that merit publication after the proof issues are resolved.

major comments (2)

[Sec. 3, Step 2 and Step 3 (Eqs. (3.8), (3.15))] The proof of Theorem 3.1 does not cover all nonlinearities allowed by (Hf). In (3.8) the passage from the sum of integrals of ||z||^{2p_i} to φ(η)∫||z||^2 uses the inequality ∫||z||^{2p_i} ≤ η^{2(p_i−1)}∫||z||^2 on the set {||z||≤η}; this inequality is valid only for p_i≥1. For p_i=1, φ(η)=C5 is independent of η, and for 0<p_i<1, φ(η)→∞ as η→0. Therefore the Step 3 requirement that C3(K)e^{φ(η)KT/2}φ(η)≤(1/2)e^{−μKT} can be achieved by taking η small is not available within the stated range of (Hf). Since (Hf) explicitly allows 0<p_i<1 and p_i=1, Theorem 3.1 is not proved for all admissible nonlinearities; the argument must be repaired or the hypothesis restricts p_i.
[Sec. 2.2, before Eq. (2.9) and after Eq. (2.11)] The feedback F in (2.12) is well defined only if the elliptic lifting problem (2.9) has a unique solution for every boundary datum v∈L2(Γ1) and if the matrix B1+⋯+BN is invertible. Both facts are asserted without proof: the coercivity estimate is introduced with 'It is not difficult to show...' before (2.9), and the invertibility is deferred with 'Following the method in [8]...' after (2.11). These are load-bearing conditions, since the matrices Λγk, B, Bk and the maps Dγk, and hence the feedback itself, do not exist if either assertion fails. The manuscript should either supply the missing proofs or give a precise theorem statement from [8] that covers these facts in the present setting.

minor comments (4)

[Sec. 3, Eqs. (3.5)–(3.6)] The identities involving the lifted functions are dimensionally inconsistent as printed: F_j maps into L2(Γ1), so writing y=z+∑χ∑F_j(z(iT)) in (3.5) and z(0)=y0−∑F_j(y(0)) in (3.6) cannot be correct in L2(Ω); these expressions should involve h_j or Dγ_jF_j.
[Sec. 3, Eq. (3.8)] The first displayed estimate in (3.8) has a term C1||z(0)||_{1/2−ε} on the right-hand side, while the final displayed estimate uses L0||z(0)||^2_{1/2−ε}; the powers should be made consistent.
[Theorem 2.2 statement] The statement says 'Assume that y0∈L2(0,1)', but the domain is a general bounded open set Ω; this should read y0∈L2(Ω).
[Abstract and throughout] There are minor language issues, for example 'not necessary to be small enough' should be 'not necessarily small enough', and 'Deﬁntion' in Definition 2.1 is a typo.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the sampled-data feedback is explicitly constructed and its stability is derived, not fitted or imported from the author's prior claims.

full rationale

The derivation chain is self-contained for the central claims. In Section 2.2 the feedback F in (2.12)-(2.14) is built from spectral data and the elliptic lifting maps D_{gamma_k}; Theorem 2.2 does not assume the decay rate rho but derives the discrete-time recursion (2.29)-(2.30) and then uses the matrix spectral bound (2.31) to obtain exponential decay. The controller depends on T by design, so the 'any sampling rate' statement is a theorem about the constructed controller, not a prediction fitted to target data. The nonlinear proof in Section 3 again proceeds by a priori estimates and a contraction step on sampling instants; no parameter is fitted to the desired rate mu. The citations to [8], [9] and [17] are technique attributions or prior-work comparisons; the invertibility assertion for B1+...+BN is deferred to [8], but this is reliance on an external method rather than a self-citation chain or an input-output equivalence. There are genuine technical gaps (the unproved coercivity/invertibility assertions and the eta-smallness argument in (3.15), which fails for p_i<1), but these are correctness risks, not circularity: none of the displayed reductions equates a derived result to an input by construction.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central claim rests on standard spectral and elliptic assumptions plus two unproved lemmas imported from Munteanu's method (well-posedness of D_{gamma_k} and invertibility of B). The design parameters rho, gamma_k, epsilon, mu are chosen by hand but are not fitted to data. No new physical entities are introduced.

free parameters (4)

rho (spectral threshold)
Chosen to satisfy lambda_N < rho <= lambda_{N+1}; sets the linear decay rate in Theorem 2.2 and the gap for the nonlinear rate mu < rho. Not estimated from data.
gamma_1,...,gamma_N (feedback decay constants)
Constants with rho < gamma_1 < ... < gamma_N; they set the per-period contraction factor e^{-gamma_1 T} in (2.32). Chosen by the designer.
epsilon (regularity exponent)
Arbitrary number in (0,1/2) used for the nonlinear well-posedness and decay norm H^{1/2-epsilon}; chosen by hand.
mu (desired nonlinear decay rate)
Any rate with 0 < mu < rho; the theorem delivers exponential decay with rate mu. Not fitted to data.

assumptions (4)

domain assumption The linearized operator A = -Delta - f_y(.,y_e) has at least one negative eigenvalue and the eigenvalues satisfy (2.1) with lambda_N < rho <= lambda_{N+1}.
The feedback construction targets finitely many unstable modes; if there are none the equilibrium is already stable. This is assumed in Section 2.1.
domain assumption The domain Omega is bounded with smooth boundary split into connected parts Gamma1 and Gamma2, and f, f_y are continuous on the closure of Omega.
Used for elliptic regularity, trace theorems, and compact resolvent; stated in Sections 1 and 2.1.
domain assumption The nonlinearity satisfies the polynomial growth condition (Hf) with exponents p_i bounded by the stated Sobolev-embedding thresholds.
Needed for the estimates in Step 2 of Theorem 3.1 and local well-posedness.
ad hoc to paper The elliptic lifting problem (2.9) is well-posed for every v in L2(Gamma1) and the matrix B1+...+BN is invertible.
Asserted without proof in Section 2.2; the invertibility is deferred to the method of [8]. This is load-bearing for the definition of F in (2.12).

how reviews work

0 comments

Cite this review

Pith. "Pith review of Boundary sampled-data feedback stabilization for parabolic equations." pith.science (2026). https://pith.science/paper/RWTLFAQ5

@misc{pith2026190803100,
  author       = {Pith},
  title        = {Pith review of: Boundary sampled-data feedback stabilization for parabolic equations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RWTLFAQ5}},
  note         = {Machine review of arXiv:1908.03100}
}

read the original abstract

The aim of this work is to design an explicit finite dimensional boundary feedback controller of sampled-data form for locally exponentially stabilizing the equilibrium solutions to semilinear parabolic equations. The feedback controller is expressed in terms of the eigenfunctions corresponding to unstable eigenvalues of the linearized equation. This stabilizing procedure is applicable for any sampling rate, not necessary to be small enough, and it tends to the continuous-times version when the sampling period tends to zero.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

19 extracted references · 19 canonical work pages

[8]

Boundary stabilization of parabolic nonlinear equations

I. Munteanu. Boundary stabilization of Parabolic nonli near equations. arXiv:1501.05737, 2015

work page Pith review arXiv 2015
[1]

Triggiani

R. Triggiani. Boundary feedback stabilization of parab olic equations. Applied Mathematic and Optimization, 6(1980), 210-220

work page 1980
[2]

V . Barbu. Boundary stabilization of equilibrium soluti ons to parabolic equations. IEEE Trans. Automat. Control, 58(9)(2013), 2416–2420

work page 2013
[3]

V . Barbu. Stabilization of Navier-Stokes equations by o blique boundary feedback con- trollers. SIAM J. Control Optim., 50(4)(2012), 2288–2307

work page 2012
[4]

V . Barbu. Stabilization of Navier-Stokes ﬂows. Springe r, London, xii+276, 2011

work page 2011
[5]

Munteanu

I. Munteanu. Boundary stabilization of the phase ﬁeld sy stem by ﬁnite-dimensional feed- back controllers. Journal of Mathematical Analysis and App lications, 412(2)(2014), 964– 975

work page 2014
[6]

Munteanu

I. Munteanu. Stabilization of Semilinear Heat Equation s, with Fading Memory, by bound- ary feedbacks. Journal of Differential Equations, 259(201 5), 454-472

work page
[7]

Munteanu

I. Munteanu. Boundary stabilization of the Navier-Stok es Equation with Fading Memory. International Journal of Control, 88(3)(2015), 531-542

work page 2015

Show all 19 references

[9]

Liu, Hanbing, P

H. Liu, Hanbing, P . Hu, I. Munteanu. Boundary feedback st abilization of Fisher’s equa- tion. Systems Control Lett., 97 (2016), 55C60. 12

2016
[10]

Balogh and M

A. Balogh and M. Krstic, Inﬁnite dimensional backstepp ing-style feedback transforma- tions for the heat equation with arbitrary level of instabil ity. Eur. J. Control, 8(2002), 165-176

2002
[11]

D. M. Boskovic, M. Krstic, and W. Liu, Boundary control o f an unstable heat equa- tion via measurement of domain-averaged temperature. IEEE Trans. Autom. Control, 46(12)(2001), 2028-2038

2001
[12]

W. J. Liu, Boundary feedback stabilization of an unstab le heat equation. SIAM J. Control Optim., 42(3)(2003), 1033-1043

2003
[13]

X. Y u, C. Xu and J. Chu, Local exponentially stabilizati on of Fisher’s equaiton using the backstepping technique. Systems Control Lett., 74(2014), 1-7

2014
[14]

Karafyllis, M

I. Karafyllis, M. Krstic. Sampled-data boundary feedb ack control of 1-D parabolic PDEs. Automatica J. IFAC 87 (2018), 226C237

2018
[15]

Fridman and N

E. Fridman and N. Bar Am. Sampled-data distributed H∞ control of transport reaction systems. SIAM J. Control Optim., 51(2):1500C1527, 2013

2013
[16]

Fridman and A

E. Fridman and A. Blighovsky. Robust sampled-data cont rol of a class of semilinear parabolic systems. Automatica J. IFAC, 48(5):826C836, 201 2
[17]

P . Lin, H. Liu, G. Wang. Output feedback stabilization f or heat equations with sampled- data controls, arXiv:1807.01220v3, 2018

2018 arXiv
[18]

Lasiecka , R

I. Lasiecka , R. Triggiani. Control Theory for Partial D ifferential Equations: Continuous and Approximation Theory. Cambridge University Press, Cam bridge, 2000

2000
[19]

J. H. Wilkinson. The algebraic eigenvalue problem. Oxf ord University Press, New Y ork,

Pith tools

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

[1] [8]

Boundary stabilization of parabolic nonlinear equations

I. Munteanu. Boundary stabilization of Parabolic nonli near equations. arXiv:1501.05737, 2015

work page Pith review arXiv 2015

[2] [1]

Triggiani

R. Triggiani. Boundary feedback stabilization of parab olic equations. Applied Mathematic and Optimization, 6(1980), 210-220

work page 1980

[3] [2]

V . Barbu. Boundary stabilization of equilibrium soluti ons to parabolic equations. IEEE Trans. Automat. Control, 58(9)(2013), 2416–2420

work page 2013

[4] [3]

V . Barbu. Stabilization of Navier-Stokes equations by o blique boundary feedback con- trollers. SIAM J. Control Optim., 50(4)(2012), 2288–2307

work page 2012

[5] [4]

V . Barbu. Stabilization of Navier-Stokes ﬂows. Springe r, London, xii+276, 2011

work page 2011

[6] [5]

Munteanu

I. Munteanu. Boundary stabilization of the phase ﬁeld sy stem by ﬁnite-dimensional feed- back controllers. Journal of Mathematical Analysis and App lications, 412(2)(2014), 964– 975

work page 2014

[7] [6]

Munteanu

I. Munteanu. Stabilization of Semilinear Heat Equation s, with Fading Memory, by bound- ary feedbacks. Journal of Differential Equations, 259(201 5), 454-472

work page

[8] [7]

Munteanu

I. Munteanu. Boundary stabilization of the Navier-Stok es Equation with Fading Memory. International Journal of Control, 88(3)(2015), 531-542

work page 2015