A Posteriori Second-Order Guarantees for Bolza Problems via Collocation
Dongzhe Zheng, Wenjie Mei · 2026
Read A Posteriori Second-Order Guarantees for Bolza Problems via Collocation by Dongzhe Zheng, Wenjie Mei (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-05811.txt.
Introduction
Zheng and Mei certify second-order optimality from discrete collocation outputs for Bolza control problems. Rebuilding state, control, and costate paths, they bound the second variation by discrete curvature minus rest terms. A positive bound proves sufficiency and informs growth and trust-region sizing, with mesh refinement support. The test runs straight from solver output.
Catalog record
| CREATOR | Dongzhe Zheng, Wenjie Mei |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 33098 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.05811
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A Posteriori Second-Order Guarantees for Bolza Problems via Collocation Dongzhe Zheng, Wenjie Mei 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. I Introduction Continuous-time second-order sufficient conditions (SSOC) for Bolza optimal control are well understood: strong regularity, coercivity of the second variation, and quadratic growth have been thoroughly characterized in the variational literature [ 18 , 2 ] . However, practical solvers based on direct collocation return only discrete nonlinear programming (NLP) data, as is typical in modern Gaussian-quadrature direct collocation transcriptions [ 16 ] , iterates (x^{N},u^{N},\lambda^{N}) , reduced Hessians, and KKT Jacobians, not continuous trajectories or functional second variations. This paper addresses the resulting gap by developing an a posteriori certification framework that constructs continuous optimality certificates from discrete solver output. On the continuous side, we work in the KKT framework for the endpoint Lagrangian and invoke standard SSOC results for strong local optimality [ 18 ] . On the discrete side, we use the reduced Hessian of the collocation transcription as a computable curvature surrogate [ 12 ] , together with standard polynomial and costate reconstruction techniques [ 1 , 5 ] . Recent work on Gauss and Radau collocation has focused primarily on convergence and stability of discrete solutions to continuous ones [ 7 , 8 , 6 ] ; in contrast, we provide a posteriori certificates for continuous SSOC starting from an already-computed discrete KKT point. Starting from a discrete KKT point, we reconstruct piecewise polynomial trajectories and evaluate residuals of dynamics, boundary, and stationarity. Combining a right-inverse bound for the linearized KKT mapping with smoothness bounds of the Lagrangian second derivatives yields a lower bound for the continuous second variation. The acceptance test reads \hat{\alpha}_{N}\bigl(1-C_{T}E_{N}^{(2)}\bigr)^{2}\;>\;\bigl(\Gamma+C_{\mathrm{quad}}+C_{T}^{\prime}\bigr)\,E_{N}^{(2)}, (1) where \hat{\alpha}_{N} is the minimum eigenvalue of the discrete reduced Hessian, E_{N}^{(2)} collects L^{2} -residual measures, and the constants C_{T} , \Gamma , C_{\mathrm{quad}} , C_{T}^{\prime} are conservatively estimable from reconstructed discrete data. Positivity of the left-hand side minus the right-hand side provides a sufficient certificate for SSOC at the continuous level, together with quantitative bounds relevant to quadratic growth and trust-region sizing. Algorithmically, the framework delivers an explicit acceptance test using only quantities available from the discrete solve: if the discrete curvature exceeds the residual-weighted threshold, the continuous problem is certified; otherwise, the mesh is refined or the polynomial degree is increased. The residual decomposition naturally guides adaptive refinement by identifying dominant error contributors [ 15 , 9 , 10 , 11 ] . We also outline an extension to path inequalities with isolated transversal switches, where modified geometric constants preserve the same acceptance structure; see Section VI for discussion, with details deferred to future work. II Problem Statement We consider the Bolza problem with control u\in L^{2}([0,T];\mathbb{R}^{m}) and no box or path inequality constraints in this section. The stationary point satisfies the interior optimality condition H_{u}(t,x,u,p)=0 . The costate p denotes the Lagrange multiplier for the state dynamics \dot{x}=f(t,x,u) . The second-order analysis is conducted for the endpoint Lagrangian via its second variation and SSOC, following the variational framework in [ 18 ] . Let T>0 . Consider x\in W^{1,2}([0,T];\mathbb{R}^{n}) , u\in L^{2}([0,T];\mathbb{R}^{m}) , we analyze the following optimal control problem: \displaystyle\min_{x,u}\ J(x,u) \displaystyle=K\big(x(0),x(T)\big)+\int_{0}^{T}L\big(t,x(t),u(t)\big)\,\,\mathrm{d}t, \displaystyle\dot{x}(t) \displaystyle=f\big(t,x(t),u(t)\big),\quad t\in[0,T], (2) where the functions f,L,K\in C^{2} (see [ 18 , 2 ] for the standard regularity assumptions in variational analysis). Problem Formulation: The KKT condition defined below corresponds to the problem stated in ( 2 ) with optional boundary equality constraints b(x(0),x(T))=0 . Let n_{b}\geq 0 denote the number of boundary constraints and let \lambda\in\mathbb{R}^{n_{b}} be the associated multiplier. When n_{b}=0 the terms involving b and \lambda vanish. Define the Hamiltonian H and the endpoint Lagrangian \mathcal{L} : \displaystyle H(t,x,\,u,\,p) \displaystyle=L(t,x,u)+p^{\top}f(t,x,u), (3) \displaystyle\mathcal{L}(x,u,p,\lambda) \displaystyle=K(x(0),x(T))+\lambda^{\top}b\big(x(0),x(T)\big) \displaystyle\quad+\int_{0}^{T}\big(L(t,x,u)+p^{\top}(f(t,x,u)-\dot{x})\big)\,\mathrm{d}t. (4) Define the KKT mapping F=(F_{1},\dots,F_{6}) by \displaystyle F_{1} \displaystyle=\dot{x}-f(t,x,u)\;\in L^{2}, (5) \displaystyle F_{2} \displaystyle=-\dot{p}-H_{x}(t,x,u,p)\in L^{2}, (6) \displaystyle F_{3} \displaystyle=H_{u}(t,x,u,p)\in L^{2}, (7) \displaystyle F_{4} \displaystyle=b\big(x(0),x(T)\big)\in\mathbb{R}^{n_{b}}, (8) \displaystyle F_{5} \displaystyle=p(T)-K_{x_{T}}(x_{0},x_{T})-\big(b_{x_{T}}(x_{0},x_{T})\big)^{\top}\lambda\in\mathbb{R}^{n}, (9) \displaystyle F_{6} \displaystyle=-p(0)-K_{x_{0}}(x_{0},x_{T})-\big(b_{x_{0}}(x_{0},x_{T})\big)^{\top}\lambda. (10) We equip the KKT mapping with the mixed norm \|F\|:=\|F_{1}\|_{L^{2}}+\|F_{2}\|_{L^{2}}+\|F_{3}\|_{L^{2}}+|F_{4}|+|F_{5}|+|F_{6}|, (11) where |\cdot| denotes the Euclidean norm on finite-dimensional vectors. In the reconstruction, we additionally report an L^{\infty} -based diagnostic for the dynamics defect (see Section III ). We let T_{\mathrm{cont}} stand for the feasible tangent space of (x,u) -variations that satisfy the linearized dynamics and boundary conditions at a KKT point, and T_{\mathrm{disc}} represents its discrete counterpart defined by the collocation scheme. For v=(\delta x,\delta u)\in T_{\mathrm{cont}} , let Q(x,u)(v) denote the second Gâteaux derivative of the endpoint Lagrangian restricted to the feasible tangent space T_{\mathrm{cont}} at (x,u) . If there exists \alpha>0 such that Q(x^{\star},u^{\star})(v)\geq\alpha\|v\|^{2} for all v\in T_{\mathrm{cont}} , then (x^{\star},u^{\star}) is a strong local minimizer and exhibits quadratic growth in a neighborhood, a standard result for optimization problems whose defining functions are of class C^{2} , i.e. , twice continuously differentiable [ 2 ] . Assumption II.1 (Strengthened Legendre Condition) There exists \rho>0 such that along interior arcs (where no control bound is active), H_{uu}(t,x,u,p)\succeq\rho I_{m\times m} (12) holds almost everywhere. Assumption II.1 ensures coercivity in the control direction and is required for the product-norm estimates in Section IV . For problems with active control bounds or singular arcs, the condition is replaced by second-order sufficiency on the critical cone; see [ 14 ] . III Collocation, reconstruction, and residuals This section defines the reconstruction used in the a posteriori analysis, the associated residuals, and the stability constants that relate the continuous and discrete second-order objects. Throughout, constants such as C_{\mathrm{geo}} , C_{T} , C_{\mathrm{quad}} , C_{T}^{\prime} , and \Lambda denote computable upper bounds obtained from the discrete solution and standard interpolation estimates. These bounds are generally conservative but sufficient for the certification test. Reconstruction and residuals : Let a Gauss, Radau, or Lobatto collocation scheme of degree p on a mesh of size h produce a discrete KKT point (x^{N},u^{N}) (and the associated discrete multipliers). From this discrete solution, we reconstruct piecewise polynomial trajectories (X^{N},u^{N},p^{N}) for state, control, and costate that interpolate the discrete variables and satisfy the discrete adjoint relations, with a terminal condition p^{N}(T)=K_{x_{T}}\bigl(X^{N}(0),X^{N}(T)\bigr). (13) The reconstruction (X^{N},u^{N},p^{N}) is used to evaluate the continuous KKT equations. Then, we distinguish two residual indicators. For theoretical perturbation and projection stability, define the L^{2} -aggregate \displaystyle E_{N}^{(2)}:=\bigl\|\dot{X}^{N}-f(\cdot,X^{N},u^{N})\bigr\|_{L^{2}} \displaystyle+\bigl\|H_{u}(\cdot,X^{N},u^{N},p^{N})\bigr\|_{L^{2}}+\bigl|b(X^{N}(0),X^{N}(T))\bigr|. (14) For diagnostics and visualization, define the