A Posteriori Second-Order Guarantees for Bolza Problems via Collocation
Pith reviewed 2026-05-10 18:54 UTC · model grok-4.3
The pith
From discrete KKT points of a collocated Bolza problem one can compute a lower bound on the continuous second variation that certifies second-order sufficiency when positive.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
Starting from a discrete KKT solution obtained via collocation, piecewise polynomial state, control, and costate trajectories are reconstructed. Residuals of the dynamics, boundary conditions, and stationarity are evaluated. A computable lower bound for the continuous second variation is then derived as the discrete reduced curvature minus explicit residual-dependent correction terms. A positive value of this bound serves as a sufficient certificate for continuous second-order sufficiency, with constants estimable conservatively from the discrete data.
What carries the argument
The a posteriori lower bound on the continuous second variation, formed by subtracting residual-dependent correction terms from the discrete reduced curvature.
If this is right
- The test supplies quantitative information on local growth and suitable trust-region radii directly from solver output.
- Residual decomposition inside the bound supports adaptive mesh refinement that targets regions where curvature is most affected.
- The same construction extends to problems with path inequalities that exhibit isolated transversal switches.
- The entire certificate is operationally verifiable from the primal-dual iterates, reduced Hessians, and Jacobians that standard collocation solvers already expose.
Where Pith is reading between the lines
- The approach could serve as an independent post-processing check that lets existing solvers terminate with a rigorous continuous optimality statement.
- Similar residual-correction ideas might be applied to other transcription methods such as shooting or pseudospectral discretizations.
- In real-time settings the bound could inform online trust-region updates or early termination when the discrete curvature is already sufficiently positive.
Load-bearing premise
The effect of the reconstruction residuals on the second variation can be bounded tightly enough by the derived correction terms to keep the test valid without excessive conservatism.
What would settle it
A concrete numerical example in which the computed bound is positive yet the true continuous second variation possesses a negative direction, or the bound is negative yet the continuous problem is second-order sufficient.
Figures
read the original abstract
Direct collocation for Bolza optimal control yields discrete Karush-Kuhn-Tucker (KKT) points, while practical solvers expose only discrete quantities such as primal-dual iterates, reduced Hessians, and Jacobians. This creates a gap between continuous second-order optimality theory and what can be certified from solver output. We develop an a posteriori certification framework that bridges this gap. Starting from a discrete KKT solution, we reconstruct piecewise polynomial state, control, and costate trajectories, evaluate residuals of the dynamics, boundary, and stationarity conditions, and derive a computable lower bound for the continuous second variation. The bound is expressed as the discrete reduced curvature minus explicit residual-dependent correction terms. A positive bound yields a sufficient certificate for continuous second-order sufficiency and provides quantitative information relevant to local growth and trust-region sizing. The constants entering the certification inequality are conservatively estimable from reconstructed discrete data. The resulting test is operationally verifiable from collocation outputs and naturally supports adaptive mesh refinement through residual decomposition. We also outline an extension to path inequalities with isolated transversal switches.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript develops an a posteriori certification framework for continuous second-order sufficiency in Bolza optimal control problems. From a discrete KKT point obtained via direct collocation, it reconstructs piecewise-polynomial state, control, and costate trajectories, evaluates residuals of the dynamics, boundary, and stationarity conditions, and derives a computable lower bound on the continuous second variation expressed as the discrete reduced curvature minus explicit residual-dependent correction terms. A positive value of this bound is claimed to certify continuous second-order sufficiency and to supply quantitative information for local growth and trust-region sizing; constants in the inequality are conservatively estimated from the reconstructed discrete data. The test is designed to be verifiable from standard collocation solver outputs and to support adaptive mesh refinement via residual decomposition. An extension to problems with path inequalities and isolated transversal switches is outlined.
Significance. If the central inequality is rigorously established and accounts for all relevant terms, the framework would close a longstanding gap between discrete solver outputs and continuous optimality theory in optimal control. It would enable post-hoc verification of second-order conditions directly from primal-dual iterates and reduced Hessians without requiring access to the full continuous problem, while naturally guiding adaptive refinement. The conservative estimation of constants from discrete data and the residual-based decomposition are practical strengths that could make the method immediately usable in existing collocation codes.
major comments (2)
- [main derivation of the certification inequality (abstract and §3–4)] The derivation of the lower bound on the continuous second variation (sketched in the abstract and presumably detailed in the main technical section) must explicitly account for all cross terms induced by the piecewise-polynomial reconstruction. In particular, the expansion should include interpolation errors between collocation nodes, possible costate jumps or discontinuities at mesh points, and boundary-condition residuals; if any such term is absorbed only into a generic constant rather than derived explicitly, the claimed inequality continuous second variation ≥ discrete reduced curvature − corrections can fail to hold even when the continuous problem satisfies the second-order condition.
- [constant estimation procedure (abstract and §5)] The conservative estimation of constants from the same reconstructed discrete data introduces a mild circularity that must be quantified. The manuscript should provide a precise statement of how these constants are computed (e.g., via supremum norms of residuals or curvature bounds) and demonstrate that the resulting test remains sufficient rather than merely necessary; otherwise the positivity certificate may be invalidated by the estimation procedure itself.
minor comments (2)
- [notation section] Notation for the discrete reduced curvature and the various residual operators should be introduced with explicit definitions and distinguished from their continuous counterparts to avoid reader confusion.
- [numerical results] The numerical examples (if present) should report both the value of the computed bound and the actual continuous second variation (or a high-fidelity reference) to illustrate the tightness of the correction terms.
Simulated Author's Rebuttal
We thank the referee for the thorough review and for recognizing the potential significance of the a posteriori certification framework. We address the major comments point by point below, providing clarifications and indicating the revisions we will make to strengthen the manuscript.
read point-by-point responses
-
Referee: [main derivation of the certification inequality (abstract and §3–4)] The derivation of the lower bound on the continuous second variation (sketched in the abstract and presumably detailed in the main technical section) must explicitly account for all cross terms induced by the piecewise-polynomial reconstruction. In particular, the expansion should include interpolation errors between collocation nodes, possible costate jumps or discontinuities at mesh points, and boundary-condition residuals; if any such term is absorbed only into a generic constant rather than derived explicitly, the claimed inequality continuous second variation ≥ discrete reduced curvature − corrections can fail to hold even when the continuous problem satisfies the second-order condition.
Authors: The derivation in Sections 3 and 4 proceeds by substituting the piecewise-polynomial reconstructions into the continuous second-variation functional and integrating by parts where appropriate. All cross terms arising from the interpolation errors (which are bounded by the collocation residuals via standard polynomial approximation theory), the possible discontinuities in the reconstructed costate at mesh points (accounted for through the stationarity residual jumps), and the boundary-condition residuals are collected explicitly as additive correction terms that depend on the residual norms and the mesh size. These terms are not absorbed into a generic constant; the only constants that appear are explicit factors such as the maximum mesh interval length and bounds on the second derivatives of the data functions, which are estimated conservatively but remain explicit in the final inequality. We will insert a detailed term-by-term expansion immediately following the statement of the main theorem to make this accounting completely transparent. revision: yes
-
Referee: [constant estimation procedure (abstract and §5)] The conservative estimation of constants from the same reconstructed discrete data introduces a mild circularity that must be quantified. The manuscript should provide a precise statement of how these constants are computed (e.g., via supremum norms of residuals or curvature bounds) and demonstrate that the resulting test remains sufficient rather than merely necessary; otherwise the positivity certificate may be invalidated by the estimation procedure itself.
Authors: The constants in the certification inequality are upper bounds on the norms of the second derivatives of the problem data and on the Lipschitz constants of the dynamics, evaluated over the reconstructed trajectories. These are computed by taking the maximum of the absolute values of the relevant quantities sampled at the collocation nodes and at a finite number of additional quadrature points within each interval; because the reconstructions are polynomial, this yields rigorous upper bounds. Since we employ upper bounds for the correction coefficients, the resulting lower bound on the second variation is smaller than or equal to the true value. Consequently, if this conservatively computed quantity is positive, the true continuous second variation is necessarily positive, preserving the sufficiency of the test. We will add a dedicated subsection in §5 that states the precise estimation procedure, including pseudocode, and proves that the overestimation does not invalidate sufficiency. revision: yes
Circularity Check
No significant circularity; bound derived independently from discrete residuals
full rationale
The paper starts from discrete KKT points, reconstructs piecewise-polynomial trajectories, computes explicit residuals of dynamics/boundary/stationarity conditions, and derives a lower bound on the continuous second variation as discrete reduced curvature minus residual-dependent correction terms whose constants are conservatively estimated from the same data. This construction is a direct inequality relating computable discrete quantities to the continuous target; it does not define the target in terms of itself, rename a fitted input as a prediction, or rely on a self-citation chain for the core inequality. The abstract and description present the result as an a-posteriori certificate whose validity rests on the explicit bounding of interpolation and residual effects rather than on any tautological reduction. No load-bearing step reduces by construction to its own inputs.
Axiom & Free-Parameter Ledger
free parameters (1)
- conservative constants in certification inequality
axioms (1)
- domain assumption Piecewise polynomial reconstruction from discrete KKT points yields residuals that can be evaluated and bounded to produce a valid lower estimate of the continuous second variation.
Reference graph
Works this paper leans on
- [1]
-
[2]
J. F. Bonnans and A. Shapiro,Perturbation analysis of optimization problems. Springer Science & Business Media, 2013
work page 2013
-
[3]
M. A. Patterson and A. V . Rao, “GPOPS-II: A MATLAB software for solving multiple-phase optimal control problems using hp-adaptive gaus- sian quadrature collocation methods and sparse nonlinear programming,” ACM Transactions on Mathematical Software (TOMS), vol. 41, no. 1, pp. 1–37, 2014
work page 2014
-
[4]
J. Nocedal and S. J. Wright,Numerical Optimization, 2nd ed. New York: Springer, 2006
work page 2006
-
[5]
J. T. Betts,Practical methods for optimal control and estimation using nonlinear programming. SIAM, 2010
work page 2010
-
[6]
C. C. Franc ¸olin, D. A. Benson, W. W. Hager, and A. V . Rao, “Costate approximation in optimal control using integral gaussian quadrature orthogonal collocation methods,”Optimal Control Applications and Methods, vol. 36, no. 4, pp. 381–397, 2015
work page 2015
-
[7]
Convergence rate for a gauss collocation method applied to unconstrained optimal control,
W. W. Hager, H. Hou, and A. V . Rao, “Convergence rate for a gauss collocation method applied to unconstrained optimal control,”Journal of Optimization Theory and Applications, vol. 169, no. 3, pp. 801–824, 2016
work page 2016
-
[8]
Convergence rate for a gauss collocation method applied to constrained optimal control,
W. W. Hager, J. Liu, S. Mohapatra, A. V . Rao, and X.-S. Wang, “Convergence rate for a gauss collocation method applied to constrained optimal control,”SIAM Journal on Control and Optimization, vol. 56, no. 2, pp. 1386–1411, 2018
work page 2018
-
[9]
Con- vergence rate for a radau hp collocation method applied to constrained optimal control,
W. W. Hager, H. Hou, S. Mohapatra, A. V . Rao, and X.-S. Wang, “Con- vergence rate for a radau hp collocation method applied to constrained optimal control,”Computational Optimization and Applications, vol. 74, no. 1, pp. 275–314, 2019
work page 2019
-
[10]
A ph mesh refinement method for optimal control,
M. A. Patterson, W. W. Hager, and A. V . Rao, “A ph mesh refinement method for optimal control,”Optimal Control Applications and Methods, vol. 36, no. 4, pp. 398–421, 2015
work page 2015
-
[11]
F. Liu, W. W. Hager, and A. V . Rao, “Adaptive mesh refinement method for optimal control using nonsmoothness detection and mesh size reduction,”Journal of the Franklin Institute, vol. 352, no. 10, pp. 4081–4106, 2015
work page 2015
-
[12]
——, “Adaptive mesh refinement method for optimal control using decay rates of legendre polynomial coefficients,”IEEE Transactions on Control Systems Technology, vol. 26, no. 4, pp. 1475–1483, 2017
work page 2017
-
[13]
A. T. Miller, W. W. Hager, and A. V . Rao, “Mesh refinement method for solving optimal control problems with nonsmooth solutions using jump function approximations,”Optimal Control Applications and Methods, vol. 42, no. 4, pp. 1119–1140, 2021
work page 2021
-
[14]
N. P. Osmolovskii and H. Maurer,Applications to regular and bang- bang control: second-order necessary and sufficient optimality condi- tions in calculus of variations and optimal control. SIAM, 2012
work page 2012
-
[15]
Strongly regular generalized equations,
S. M. Robinson, “Strongly regular generalized equations,”Mathematics of Operations Research, vol. 5, no. 1, pp. 43–62, 1980
work page 1980
-
[16]
A. L. Dontchev and R. T. Rockafellar,Implicit functions and solution mappings. Springer, 2009, vol. 543
work page 2009
-
[17]
Bounds for integration matrices that arise in gauss and radau collocation,
W. Chen, W. Du, W. W. Hager, and L. Yang, “Bounds for integration matrices that arise in gauss and radau collocation,”Computational Optimization and Applications, vol. 74, no. 1, pp. 259–273, 2019
work page 2019
-
[18]
Finite elements with switch detection for direct optimal control of nonsmooth systems,
A. Nurkanovi ´c, M. Sperl, S. Albrecht, and M. Diehl, “Finite elements with switch detection for direct optimal control of nonsmooth systems,” Numerische Mathematik, vol. 156, no. 3, pp. 1115–1162, 2024
work page 2024
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.