pith. sign in

arxiv: 2604.03892 · v1 · submitted 2026-04-04 · 📡 eess.SY · cs.LG· cs.SY· math.OC

Lotka-Sharpe Neural Operators for Control of Population PDEs

Miroslav Krstic , Iasson Karafyllis , Luke Bhan , Carina Veil This is my paper

Pith reviewed 2026-05-13 16:42 UTC · model grok-4.3

classification 📡 eess.SY cs.LGcs.SYmath.OC
keywords Lotka-Sharpe operatorneural operatorsage-structured population PDEsfeedback controlpractical asymptotic stabilitypredator-prey modelsoperator learning
0
0 comments X

The pith

Replacing the exact Lotka-Sharpe operator with a neural-operator approximation preserves semi-global practical asymptotic stability in feedback control of age-structured population PDEs.

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

The paper first proves the Lotka-Sharpe operator is Lipschitz continuous on compact sets of fertility and mortality functions, which guarantees that neural operators can approximate it to any desired accuracy. It then tracks how the approximation error propagates through the other nonlinear maps in the closed-loop system all the way to the control input. The central result is that the resulting approximate feedback law still renders the age-structured predator-prey system semi-globally practically asymptotically stable, with an explicit margin that shrinks as the operator error grows. Numerical experiments show the operator can be trained once and then reused online even when fertility and mortality rates must be estimated.

Core claim

We prove that the Lotka-Sharpe operator is Lipschitz continuous, guaranteeing the existence of arbitrarily accurate neural operator approximations over a compact set of fertility and mortality functions. We then show that the resulting approximate feedback law preserves semi-global practical asymptotic stability under propagation of the operator approximation error through various other nonlinear operators, all the way through to the control input.

What carries the argument

The neural-operator approximation of the Lotka-Sharpe operator, which maps fertility and mortality functions to the scalar ζ defined implicitly by the nonlinear integral condition, and the explicit propagation of its approximation error through the remaining nonlinear operators to the control input.

If this is right

  • The neural Lotka-Sharpe operator can be trained once and reused for control of any other age-structured population interconnection.
  • Online feedback remains effective when fertility and mortality functions are only estimated from measurements.
  • The practical stability margin is a continuous, explicitly computable function of the operator approximation error.
  • Arbitrarily small approximation error is achievable because of the established Lipschitz property on compact sets.

Where Pith is reading between the lines

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

  • The same once-and-for-all operator learning step could be applied to other implicitly defined scalars that appear in population or epidemiological models.
  • Real-time control of high-dimensional age-structured systems becomes feasible without repeatedly solving the nonlinear integral equation at each time step.
  • Pairing the learned operator with online parameter estimators would allow the method to track slowly drifting environmental conditions that affect birth and death rates.

Load-bearing premise

Fertility and mortality functions lie inside a compact set on which the Lotka-Sharpe operator is Lipschitz continuous.

What would settle it

A simulation or experiment in which the closed-loop trajectories escape every practical stability ball once the neural-operator approximation error is made larger than the explicit margin derived from the stability proof.

Figures

Figures reproduced from arXiv: 2604.03892 by Carina Veil, Iasson Karafyllis, Luke Bhan, Miroslav Krstic.

Figure 1
Figure 1. Figure 1: Computational breakdown of the Lotka-Sharpe operator so that k(a)Π(a) is the net maternity function, i.e. fertility at age a weighted by survival up to age a. Then the defining equation becomes Z ∞ 0 k(a)Π(a)e −ζa da = 1, (10) or equivalently L{kΠ}(ζ) = 1. (11) where L is the Laplace transform. Hence, the Lotka-Sharpe operator may be viewed as GLS(k, µ) = (L{kΠ}) −1 (1), (12) namely: first form the net mat… view at source ↗
Figure 2
Figure 2. Figure 2: Graph of the operator dependencies for the control construction in (28). B. Approximate controller introduces a perturbation When the Lotka-Sharpe parameters ζi are known only approximately as ˆζi , ζi is replaced in the controller by ˆζi . The resulting approximation error does not remain localized, but enters every operator that depends on ζi . The approximation of controller (18) is given by u = uˆ(η, e… view at source ↗
Figure 3
Figure 3. Figure 3: Explicit operator mappings in the predator-prey control law. and Z A 0 kmin(a)e − R a 0 µmax(s) ds da > 1. (39) Let ζmin ≤ ζmax be the unique solutions of the following equa￾tions for (kmin,µmax) and (kmax, µmin) respectively. Then, Z A 0 kmin(a)e −ζmina− R a 0 µmax(s) ds da = 1 , (40) Z A 0 kmax(a)e −ζmaxa− R a 0 µmin(s) ds da = 1. (41) Let G > 0 be a sufficiently large constant so that ∥f∥∞ + sup a,s∈[0,… view at source ↗
Figure 4
Figure 4. Figure 4: Example of various (k, µ) used in training and performance of learned operator GbLS. (c) highlights the residuals of all 100 test examples during training in blue and the samples corresponding to (a) and (b) in orange. 0.0 0.5 1.0 Age a 0.0 2.5 5.0 7.5 10.0 Time t 2 4 6 8 10 0.0 0.5 1.0 Age a 0.0 2.5 5.0 7.5 10.0 Time t 2 4 6 0 5 10 Time t 0.9 1.0 1.1 1.2 u(t) u ∗ (a) State x1(a, t) (b) State x2(a, t) (c) … view at source ↗
Figure 5
Figure 5. Figure 5: Simulation profiles of the (a) population x1, (b) population x2(t), and (c) dilution control using (ζˆ1, ζˆ2) = GbLS(k, µ) in the control law (28). k1, k2 and µ1, µ2 correspond with samples eight and nine in [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Adaptive control simulation where k, µ, g are samples eight and nine from [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
read the original abstract

Age-structured predator-prey integro-partial differential equations provide models of interacting populations in ecology, epidemiology, and biotechnology. A key challenge in feedback design for these systems is the scalar $\zeta$, defined implicitly by the Lotka-Sharpe nonlinear integral condition, as a mapping from fertility and mortality rates to $\zeta$. To solve this challenge with operator learning, we first prove that the Lotka-Sharpe operator is Lipschitz continuous, guaranteeing the existence of arbitrarily accurate neural operator approximations over a compact set of fertility and mortality functions. We then show that the resulting approximate feedback law preserves semi-global practical asymptotic stability under propagation of the operator approximation error through various other nonlinear operators, all the way through to the control input. In the numerical results, not only do we learn ``once-and-for-all'' the canonical Lotka-Sharpe (LS) operator, and thus make it available for future uses in control of other age-structured population interconnections, but we demonstrate the online usage of the neural LS operator under estimation of the fertility and mortality functions.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper claims that the Lotka-Sharpe operator mapping fertility and mortality functions to the implicit scalar ζ is Lipschitz continuous on a compact set, permitting arbitrarily accurate neural-operator approximations. It further asserts that substituting the neural approximation into the feedback law for age-structured predator-prey integro-PDEs preserves semi-global practical asymptotic stability, with the practical stability margin explicitly depending on the operator approximation error. Numerical results demonstrate one-time learning of the canonical operator and its online deployment under online estimation of the fertility and mortality functions.

Significance. If the Lipschitz property and error-propagation argument hold, the work supplies a reusable, data-driven surrogate for the Lotka-Sharpe condition that can be applied across multiple age-structured control problems. The explicit dependence of the stability margin on the approximation error and the “once-and-for-all” learning strategy constitute concrete, reusable contributions to the control of nonlinear population PDEs.

major comments (2)
  1. [stability theorem and numerical experiments] The stability theorem (presumably the main result in the analysis section) establishes semi-global practical asymptotic stability only under the standing assumption that the estimated fertility and mortality functions remain inside the compact set on which the Lotka-Sharpe operator is proved Lipschitz. The numerical section demonstrates online estimation but supplies no projection, saturation, or invariant-set argument that would guarantee the estimates stay inside this set; if they exit, both the Lipschitz bound and the propagated error margin cease to apply.
  2. [error propagation analysis] The error-propagation argument through the remaining nonlinear operators (from operator approximation error to control input) is asserted to be controllable, yet the manuscript does not exhibit the explicit chain of inequalities or the dependence of the practical stability radius on the neural-operator error norm; without these quantitative bounds it is impossible to verify that the margin remains positive for any fixed approximation accuracy.
minor comments (2)
  1. [preliminaries] Notation for the compact set of fertility/mortality functions and the associated Lipschitz constant should be introduced once in the preliminaries and used consistently thereafter.
  2. [numerical results] The numerical figures would be clearer if they overlaid the theoretical error bound on the observed approximation error to illustrate the practical stability margin.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive comments on the stability theorem and error-propagation analysis. We address each major comment below and indicate the revisions we will make to strengthen the manuscript.

read point-by-point responses

  1. Referee: [stability theorem and numerical experiments] The stability theorem (presumably the main result in the analysis section) establishes semi-global practical asymptotic stability only under the standing assumption that the estimated fertility and mortality functions remain inside the compact set on which the Lotka-Sharpe operator is proved Lipschitz. The numerical section demonstrates online estimation but supplies no projection, saturation, or invariant-set argument that would guarantee the estimates stay inside this set; if they exit, both the Lipschitz bound and the propagated error margin cease to apply.

    Authors: We agree that the stability result holds only when the online estimates of fertility and mortality remain inside the compact set on which Lipschitz continuity is established. The current numerical experiments illustrate estimation without an explicit safeguard. In the revised manuscript we will augment the online estimation procedure with a projection operator onto the compact set. This modification will be described in the numerical section, and the statement of the stability theorem will be updated to reference the projection, thereby ensuring the assumption is enforced at all times. revision: yes

  2. Referee: [error propagation analysis] The error-propagation argument through the remaining nonlinear operators (from operator approximation error to control input) is asserted to be controllable, yet the manuscript does not exhibit the explicit chain of inequalities or the dependence of the practical stability radius on the neural-operator error norm; without these quantitative bounds it is impossible to verify that the margin remains positive for any fixed approximation accuracy.

    Authors: The referee correctly observes that the manuscript asserts controllability of the error propagation without supplying the full chain of inequalities. We will add a dedicated subsection (or appendix) that derives the explicit bounds step by step, showing how the neural-operator approximation error propagates through the remaining nonlinear operators to the control input and determines the practical stability radius. The new material will include the quantitative dependence of the radius on the operator error norm and will confirm that the margin stays positive for any prescribed approximation accuracy. revision: yes

Circularity Check

0 steps flagged

No circularity: derivation rests on independent Lipschitz proof and explicit error propagation

full rationale

The paper establishes Lipschitz continuity of the Lotka-Sharpe operator over a compact set of fertility/mortality functions as a standalone result, then constructs a neural-operator approximation whose uniform error bound follows directly from that Lipschitz property. Stability of the approximate feedback is obtained by propagating the operator error through the remaining nonlinear maps to the control input, producing an explicit practical-stability margin that depends on the approximation error. None of these steps reduces a claimed prediction to a fitted quantity by construction, nor does the central argument rely on a load-bearing self-citation whose validity is presupposed inside the paper. The numerical examples illustrate online use under parameter estimation but do not substitute for or circularly presuppose the analytic bounds.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the existence of a compact set of fertility and mortality functions on which the Lotka-Sharpe operator is Lipschitz continuous; this compactness is invoked to guarantee neural-operator approximability and to bound the propagated error in the stability proof.

axioms (1)
  • domain assumption The Lotka-Sharpe operator is Lipschitz continuous on a compact set of fertility and mortality functions.
    Stated in the abstract as the key property enabling neural-operator approximation and error propagation.

pith-pipeline@v0.9.0 · 5500 in / 1316 out tokens · 40290 ms · 2026-05-13T16:42:35.761875+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

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

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

Reference graph

Works this paper leans on

33 extracted references · 33 canonical work pages

  1. [1]

    Global stabilization and emergence tracking via aquatic control in an age-structured mosquito model,

    M. Bargo and Y . Simpor ´e, “Global stabilization and emergence tracking via aquatic control in an age-structured mosquito model,” Nonlinear Analysis: Real World Applications, vol. 92, p. 104629, 2026. [Online]. Available: https://www.sciencedirect.com/science/article/pii/ S1468121826000295

    work page 2026

  2. [2]

    Stabilization of nonlinear systems with unknown delays via delay-adaptive neural operator approximate predictors,

    L. Bhan, M. Krsti ´c, and Y . Shi, “Stabilization of nonlinear systems with unknown delays via delay-adaptive neural operator approximate predictors,” 2025. [Online]. Available: https://arxiv.org/abs/2509.26443

    work page arXiv 2025

  3. [3]

    Neural operators for bypassing gain and control computations in PDE backstepping,

    L. Bhan, Y . Shi, and M. Krsti ´c, “Neural operators for bypassing gain and control computations in PDE backstepping,”IEEE Transactions on Automatic Control, vol. 69, no. 8, pp. 5310–5325, 2024

    work page 2024

  4. [4]

    Adaptive control of reaction–diffusion PDEs via neural operator- approximated gain kernels,

    ——, “Adaptive control of reaction–diffusion PDEs via neural operator- approximated gain kernels,”Systems & Control Letters, vol. 195, p. 105968, 2025. [Online]. Available: https://www.sciencedirect.com/ science/article/pii/S0167691124002561

    work page 2025

  5. [5]

    Universal approximation to nonlinear operators by neural networks with arbitrary activation functions and its application to dynamical systems,

    T. Chen and H. Chen, “Universal approximation to nonlinear operators by neural networks with arbitrary activation functions and its application to dynamical systems,”IEEE transactions on neural networks, vol. 6, no. 4, pp. 911–917, 1995

    work page 1995

  6. [6]

    Explaining the optimality of u-shaped age-specific mortality,

    C. Y . C. Chu, H.-K. Chien, and R. D. Lee, “Explaining the optimality of u-shaped age-specific mortality,”Theor. Popul. Biol., vol. 73, no. 2, pp. 171–180, Mar. 2008

    work page 2008

  7. [7]

    Knowledge about the impact of age on fertility: a brief review,

    I. Delbaere, S. Verbiest, and T. Tyd ´en, “Knowledge about the impact of age on fertility: a brief review,”Ups. J. Med. Sci., vol. 125, no. 2, pp. 167–174, May 2020

    work page 2020

  8. [8]

    Dochain,Automatic Control of Bioprocesses

    D. Dochain,Automatic Control of Bioprocesses. John Wiley & Sons, 2013

    work page 2013

  9. [9]

    Stabilization of age-structured chemostat hyperbolic PDE with actuator dynamics,

    P.-E. Haacker, I. Karafyllis, M. Krsti ´c, and M. Diagne, “Stabilization of age-structured chemostat hyperbolic PDE with actuator dynamics,” International Journal of Robust and Nonlinear Control, 2024

    work page 2024

  10. [10]

    The basic approach to age-structured population dynamics,

    M. Iannelli and F. Milner, “The basic approach to age-structured population dynamics,”Lecture Notes on Mathematical Modelling in the Life Sciences. Springer, Dordrecht, 2017

    work page 2017

  11. [11]

    Inaba,Age-Structured Population Dynamics in Demography and Epidemiology

    H. Inaba,Age-Structured Population Dynamics in Demography and Epidemiology. Springer, 2017, vol. 1

    work page 2017

  12. [12]

    Stability of integral delay equations and stabilization of age-structured models,

    I. Karafyllis and M. Krsti ´c, “Stability of integral delay equations and stabilization of age-structured models,”ESAIM: Control, Optimisation and Calculus of Variations, vol. 23, no. 4, pp. 1667–1714, 2017

    work page 2017

  13. [13]

    The age-structured chemostat with substrate dynamics as a control system,

    I. Karafyllis, D. Theodosis, and M. Krsti ´c, “The age-structured chemostat with substrate dynamics as a control system,” 2025. [Online]. Available: https://arxiv.org/abs/2511.09963

    work page arXiv 2025

  14. [14]

    Neural operators for control of age-structured population PDEs,

    M. Krsti ´c, I. Karafyllis, L. Bhan, and C. Veil, “Neural operators for control of age-structured population PDEs,” Honolulu, Hawaii, USA, Dec. 2026, submitted to 2026 IEEE 65th Conference on Decision and Control (CDC)

    work page 2026

  15. [15]

    Neural operators for control of age-structured population PDEs,

    ——, “Neural operators for control of age-structured population PDEs,” https://github.com/lukebhan/NeuralOperatorsLoktaSharpePredatoryPrey, 2026, github repository

    work page 2026

  16. [16]

    Neural operators of backstepping controller and observer gain functions for reaction–diffusion PDEs,

    M. Krsti ´c, L. Bhan, and Y . Shi, “Neural operators of backstepping controller and observer gain functions for reaction–diffusion PDEs,” Automatica, vol. 164, p. 111649, Jun. 2024

    work page 2024

  17. [17]

    Control of age-structured population dynamics with intraspecific competition in context of bioreactors,

    A.-C. Kurth and O. Sawodny, “Control of age-structured population dynamics with intraspecific competition in context of bioreactors,” Automatica, vol. 152, p. 110944, 2023

    work page 2023

  18. [18]

    Tracking-control for age-structured population dynamics with self-competition governed by integro-PDEs,

    A.-C. Kurth, K. Schmidt, and O. Sawodny, “Tracking-control for age-structured population dynamics with self-competition governed by integro-PDEs,”Automatica, vol. 133, p. 109850, 2021

    work page 2021

  19. [19]

    Adaptive neural-operator backstepping control of a benchmark hyperbolic PDE,

    M. Lamarque, L. Bhan, Y . Shi, and M. Krsti ´c, “Adaptive neural-operator backstepping control of a benchmark hyperbolic PDE,”Automatica, vol. 177, p. 112329, 2025

    work page 2025

  20. [20]

    Gain scheduling with a neural operator for a transport PDE with nonlinear recirculation,

    M. Lamarque, L. Bhan, R. Vazquez, and M. Krsti ´c, “Gain scheduling with a neural operator for a transport PDE with nonlinear recirculation,” IEEE Transactions on Automatic Control, 2025

    work page 2025

  21. [21]

    Nonlocality and nonlinearity implies universality in operator learning,

    S. Lanthaler, Z. Li, and A. M. Stuart, “Nonlocality and nonlinearity implies universality in operator learning,”Constructive Approximation, vol. 62, pp. 261–303, 2025. [Online]. Available: https://doi.org/10.1007/ s00365-025-09718-3

    work page 2025

  22. [22]

    Fourier Neural Operator for Parametric Partial Differential Equations,

    Z. Li, N. Kovachki, K. Azizzadenesheli, B. Liu, K. Bhattacharya, A. Stuart, and A. Anandkumar, “Fourier Neural Operator for Parametric Partial Differential Equations,” May 2021

    work page 2021

  23. [23]

    Decoupled weight decay regularization,

    I. Loshchilov and F. Hutter, “Decoupled weight decay regularization,” in International Conference on Learning Representations, 2019. [Online]. Available: https://openreview.net/forum?id=Bkg6RiCqY7 16

    work page 2019

  24. [24]

    Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators,

    L. Lu, P. Jin, G. Pang, Z. Zhang, and G. E. Karniadakis, “Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators,”Nature machine intelligence, vol. 3, no. 3, pp. 218–229, 2021

    work page 2021

  25. [25]

    Neural operators for adaptive control of traffic flow models,

    K. Lyu, J. Wang, Y . Zhang, and H. Yu, “Neural operators for adaptive control of traffic flow models,”IFAC-PapersOnLine, vol. 59, no. 8, pp. 13–18, 2025, 5th IFAC Workshop on Control of Systems Governed by Partial Differential Equations - CPDE 2025. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S2405896325006433

    work page 2025

  26. [26]

    Applications of mathematics to medical problems,

    A. G. McKendrick, “Applications of mathematics to medical problems,” Proceedings of the Edinburgh Mathematical Society, vol. 44, pp. 98– 130, 1925

    work page 1925

  27. [27]

    Yield trajectory tracking for hyperbolic age-structured population systems,

    K. Schmidt, I. Karafyllis, and M. Krsti ´c, “Yield trajectory tracking for hyperbolic age-structured population systems,”Automatica, vol. 90, pp. 138–146, 2018

    work page 2018

  28. [28]

    A problem in age-distribution,

    F. Sharpe and A. Lotka, “A problem in age-distribution,”The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science, vol. 21, no. 124, pp. 435–438, Apr. 1911

    work page 1911

  29. [29]

    Stabilization of age- structured competing populations,

    C. Veil, M. Krsti ´c, P. McNamee, and O. Sawodny, “Stabilization of age- structured competing populations,”arXiv preprint arXiv:2507.23013, 2025

    work page arXiv 2025

  30. [30]

    Stabiliza- tion of predator–prey age-structured hyperbolic PDE when harvesting both species is inevitable,

    C. Veil, M. Krsti ´c, I. Karafyllis, M. Diagne, and O. Sawodny, “Stabiliza- tion of predator–prey age-structured hyperbolic PDE when harvesting both species is inevitable,”IEEE Transactions on Automatic Control, vol. 71, no. 1, pp. 123–138, 2026

    work page 2026

  31. [31]

    Stabilization of age-structured competition (predator-predator) population dynamics,

    C. Veil, P. McNamee, M. Krsti ´c, and O. Sawodny, “Stabilization of age-structured competition (predator-predator) population dynamics,” in 2025 IEEE 64th Conference on Decision and Control (CDC), 2025, pp. 2058–2063

    work page 2025

  32. [32]

    Backstepping Neural Operators for 2x2 Hyperbolic PDEs,

    S. Wang, M. Diagne, and M. Krsti ´c, “Backstepping Neural Operators for 2x2 Hyperbolic PDEs,” Jul. 2024

    work page 2024

  33. [33]

    Neural operators for boundary stabilization of stop-and-go traffic,

    Y . Zhang, R. Zhong, and H. Yu, “Neural operators for boundary stabilization of stop-and-go traffic,” inProceedings of the 6th Annual Learning for Dynamics & Control Conference, ser. Proceedings of Machine Learning Research, A. Abate, M. Cannon, K. Margellos, and A. Papachristodoulou, Eds., vol. 242. PMLR, 15–17 Jul 2024, pp. 554–565. [Online]. Available:...

    work page 2024