Papers

We present a generalization of the well-known homogeneous self-dual embedding model, which is widely used in conic optimization. The new embedding applies to a problem of minimizing the sum of two proper lower-semicontinuous convex functions and can be represented as a single inequality that uses perspectives of these functions and of their conjugates. A solution to the proposed embedding encodes a primal-dual solution to the original problem when available, or an infeasibility certificate otherwise. We then use the Douglas-Rachford algorithm to find a solution to the embedding and discuss its efficient implementation by exploiting the problem structure. The resulting algorithm recovers an existing method for solving quadratic cone programs as a special case. We demonstrate the generality and effectiveness of the algorithm on a class of convex optimization problems with non-smooth objective function and non-conic constraints.

A data-driven inverse optimization problem (DDIOP) is the problem of estimating the objective-function parameters (weights) that explain observed optimal-solution data, and it arises in many applications, including integer linear programming (ILP). It is known that, by applying gradient-based optimization methods to the suboptimality loss, the inverse optimization of ILPs can be solved exactly within finitely many oracle iterations, and that the required number of iterations is bounded as $T=O(1/\gamma(\ell_{\mathrm{sub}})^2)$ in terms of a problem-dependent geometric constant $\gamma(\ell_{\mathrm{sub}})$. However, no means of bounding $\gamma(\ell_{\mathrm{sub}})$ from below as a function of the problem size has been available, and hence the number of iterations could not be given as an explicit function of the problem size. We therefore give, when the forward problem is an integer linear program (ILP), the number of iterations sufficient for projected subgradient descent applied to the suboptimality loss to achieve exact consistency with the observed data, as a fully explicit function of the number of samples, the dimension of the features, the ranges of the features, and the structure of the constraint coefficient matrix, up to polynomial factors in the basic constants (the diameter of the weight set, the step-size parameter, and the Lipschitz constant of the suboptimality loss).

This paper presents an optimal control framework for achieving finite-time stabilization of linear systems. By introducing a suitably constructed integral cost function, we derive a new class of nonlinear controllers that guarantee finite-time stability through the application of the optimality principle. The relationship between the resulting optimal control law and the associated value function is analyzed, leading to the derivation of a Hamilton-Jacobi-Bellman (HJB) equation and the study of its regularity properties. Numerical simulations validate the theoretical findings and illustrate the effectiveness of the proposed method. Furthermore, a discussion on estimating the convergence time is provided.

The conjugate gradient method (CG) solves a symmetric positive definite (SPD) system $Ax=b$, but does not respect $x\geq0$. We develop non-negative conjugate gradients, a solver for the bound-constrained quadratic $\min_x \tfrac12 x^\top A x - b^\top x$ subject to $x\geq0$, $Bx=c$, $A\succ0$, with $Bx=c$ present only in the equality-augmented variant. The solver wraps CG in a primal-dual active-set loop: each step solves the unconstrained system over the free variables, drops variables that turn negative, re-admits any bound variable with negative reduced gradient, and restarts CG. These toggles are the principal pivots of the linear complementarity problem $(A,-b)$, well defined since $A$ is a $P$-matrix. Termination uses the block-principal-pivoting construction of Júdice and Pires, with a least-index Bland fallback, reaching the unique global minimiser in finitely many outer steps. Our contribution carries this guarantee to the inexact, matrix-free regime the Krylov inner solve requires: once CG's residual is tied to the decision margin, the approximate loop makes the same primal and dual sign decisions as the exact one, and a direct factorisation may replace CG unchanged. Each inner solve is matrix-free: for $A=M^\top M$, the operator $v\mapsto M^\top(Mv)$ needs two products with $M$, no $O(n^2)$ storage, and retains CG's $O(\sqrt\kappa)$ rate, $\kappa=\kappa(A)$. Regularisation or preconditioning that compresses the spectrum lowers the iteration count; a ridge/Tikhonov split $A\mapsto(1-\alpha)A+\alpha R^\top R$ also secures the needed $P$-matrix property. A loop-free projected-gradient reference method, analysed alongside, converges at the slower $O(\kappa)$ rate. On synthetic SPD problems the loop reaches the planted optimum in at most 6 outer steps, and with a direct free-set solve outpaces Lawson–Hanson and an interior-point solver by an order of magnitude on dense instances.

The increasing deployment of separate biowaste collection across Europe raises questions regarding environmentally and economically optimal treatment strategies. From this perspective, the balance between decentralised and centralised infrastructure is a potential lever to consider. The choice of technologies is another. Composting and anaerobic digestion (AD) offer viable circular bioeconomy pathways, but their performance depends on transport distance, facility size, seasonal waste fluctuations, and technological efficiency. Threshold distances and scale-dependent interactions affecting the environmental and economic performance of treatment plants remain poorly characterised. This study develops a unified techno-economic and life cycle assessment framework to quantify how transport constraints and treatment capacities shape biowaste valorisation systems. Three facility scales for composting and AD are evaluated across a range of transport distances and compared with local incineration in terms of environmental and economic performance. Transport burdens are modelled using vehicle life cycle assessment (LCA) and realistic heavy-duty truck data for electric-, gas-, and diesel-powered vehicles. Results reveal non-linear trade-offs between scale and transport, with zero-distance composting yielding the lowest combined impacts. Across all scenarios, 45 configurations outperform local incineration economically and environmentally, while 111 remain environmentally preferable despite higher costs, thus providing quantitative guidance for infrastructure planning.

Upstream suppliers often set wholesale prices and reserve capacity without direct access to the detailed demand information held by downstream retailers. We study how a supplier can learn from a short history of wholesale prices, the retail prices subsequently chosen by a better-informed retailer, and the associated purchase probabilities. A quantile-only method (Q) uses each retail price and purchase probability as a demand observation. Our decision-informed method (DI) also uses the wholesale price under which the retailer chose that price. This additional context rules out demand curves that fit the observed sales outcomes but cannot explain the retailer's past choices. We characterize the worst-case retailer response to a new wholesale price under a broad, nonparametric class of demand curves. When the retailer has a unique best price, the relevant uncertainty is summarized by the lowest demand that remains possible. When several prices are tied, the supplier instead evaluates a finite collection of induced-demand cases. We also distinguish two forms of imperfect behavior. Under condition error (CE), the observed price is retained but its link to the wholesale price may be imperfect. Under price error (PE), the observed price may lie near an unobserved exact optimum. Both extensions remain computationally finite. The resulting downstream-demand summary leads directly to robust wholesale-pricing and stocking decisions. Numerical experiments across six demand environments show that DI provides substantial value with only a few transactions and remains useful under moderate error; with three observations and $\alpha=0$, DI improves the profit-to-oracle ratio by 9.1–11.8 percentage points.

Low-rank adaptation (LoRA) optimizes $J(B,A)=\mathcal L(W_\mathrm{base}+sBA)$ over two adapters $B \in \mathbb{R}^{m \times r}$ and $A \in \mathbb{R}^{r \times n}$ that form a low-rank update to a frozen pretrained weight matrix $W_\mathrm{base} \in \mathbb{R}^{m \times n}$. The prior analysis shows LoRA-GD takes $\exp\{\mathcal{O}(\epsilon^{-2})\}$ oracle calls to find an $\epsilon$-stationary point such that $\|\nabla J(B,A)\|\leq \epsilon$ in the deterministic setting. We sharpen the analysis and show that $\mathcal{O}(\epsilon^{-4})$ full-gradient evaluations suffice for the same first-order criterion. We further study stochastic LoRA under unbiased gradient estimates and finite variance. We propose LoRA-NSGDM, which finds an $\epsilon$-stationary point with $\mathcal{O}(\epsilon^{-8})$ stochastic oracle complexity. Under the additional mean-square smoothness condition, we use variance reduction strategy and propose LoRA-STORM, which improves the stochastic oracle complexity to $\mathcal{O}(\epsilon^{-6})$.

We study destination inference for adversarial unmanned underwater navigation in a spatially varying current field. The red vehicle is modeled as approximately following a Hamilton-Jacobi (HJ) time-optimal path toward an unknown destination, while a blue vehicle observes a noisy realization of that trajectory. We derive a continuous-time likelihood model for this problem and obtain a closed-form local maximum likelihood estimator, a constant-memory multi-period estimator, and an asymptotic Cramér-Rao efficiency result. The Fisher information is governed by a combined score kernel with two additive components, a policy-mean sensitivity and a reference-path sensitivity, and a multiplicative drift sensitivity whose leading geometric contribution is a contraction of the current Hessian with the Jacobi field of the HJ characteristic flow. All three sensitivities are induced by that Jacobi field, yielding a computable link between current-field geometry and destination identifiability. We further extend the framework to a sweep-dependent observation model and formulate an active sweep-design problem for the blue team. Numerical experiments in vortex and channel-shear currents validate the Cramér-Rao prediction and illustrate how current geometry determines which destinations can be reliably inferred.

In this paper we study a Stackelberg game with one leader and multiple followers. Given the leader's control, the followers solve a Nash game with possibly multiple equilibria. We consider a robust leader who considers the worst scenario, namely the followers would select the equilibrium worst for the leader. By using the weak formulation, the problem induces a zero sum game problem with open loop controls, which is time inconsistent. We shall characterize the last problem through an HJB equation on the Wasserstein space of probability measures. The principal-agent problem with one principal and multiple agents can be viewed as a special case of our problem, but with certain constraints.

Bacterial motility is a fundamental trait for territorial expansion and resource acquisition. While existing models of nutrient-dependent motility often do not explicitly account for the metabolic costs associated with motility, these costs become critical in nutrient-limited or closed systems. In this study, we developed a mathematical framework using partial differential equations (PDEs) that explicitly incorporates the energetic trade-offs of motility. By formulating an optimization problem focused on maximizing population yield—defined by the total cell count—we evaluated various motility strategies across different environmental contexts. Our results demonstrate that the optimal motility response is highly sensitive to resource distribution. Specifically, we show that in unpredictable environments, a non-monotonic motility response emerges as the optimal strategy, providing a robust theoretical explanation for dose-response curves observed in experimental microbiology. This framework serves as a powerful, interpretable tool for predicting bacterial behavior in resource-constrained ecosystems and offers new insights into how such adaptive strategies are shaped by environmental pressures.

This study evaluates the potential of electric vehicle (EV) charging flexibility to enhance both energy efficiency and power grid stability. Using real-world data from public charging stations in Prague, we analyze individual and aggregated charging sessions to explore how optimizing charging times can reduce energy waste, minimize grid imbalances, and support the integration of renewable energy. By aligning EV charging with periods of lower grid demand and higher renewable generation, we demonstrate a significant improvement in energy efficiency, reduc ing the need for costly system support and ancillary services. Our findings suggest that cooperation between power distributors and transmission system operators can unlock new opportunities for maintaining grid stability while promoting sustainable energy use in an increasingly uncertain energy landscape.

This paper studies a new class of nonlocal optimal control problems where the constraints involve approximations of minimizers of quasiconvex energies. These problems are parameterized by a fractional parameter $s \in (0, 1)$ and a horizon parameter $\delta > 0$, and the energy density depends on a nonlocal fractional gradient. Here, the constraint consists of finding quasi-minimizers with respect to the quasiconvex energy. Despite the fact that the energies of interest may not have unique minimizers, we may prove the existence of solutions to this class of control problems. The constraint in Cueto-Siktar 2026 was finding global minimizers of the energy, and this problem's main limitation was an inability to prove convergence of solutions for the nonlocal control problems to those of a corresponding local, PDE-constrained optimal control problem. While this issue arises from the lack of uniqueness of minimizers for the constraining energy, we get stronger convergence results with our new choice of constraints. Namely, we obtain convergence of minimizers for nonlocal optimal control problems that have a general cost functional depending on the nonlocal gradient.

Control barrier functions (CBFs) have become an important controller design tool for autonomous systems subject to safety constraints. Despite their popularity, recent works have shown that CBF-based controllers can destabilize the internal dynamics of the system. In this paper, we consider linear systems with affine safety constraints and design linear feedback controllers that satisfy high-order CBF (HOCBF) constraints while rendering the origin globally exponentially stable. We first characterize the exact class of all linear gain matrices that globally satisfy the HOCBF constraints, including necessary and sufficient conditions for when this class is nonempty. Then, by leveraging the recently introduced notion of CBF output dynamics and CBF internal dynamics, we provide the necessary and sufficient conditions for the existence of stabilizing gain matrices within that class. Finally, we show that Linear Quadratic Regulator (LQR) and robust control problems can be solved while being constrained within this class of safe and stabilizing gain matrices, through standard linear control techniques such as algebraic Riccati equations (AREs) and Linear Matrix Inequalities (LMIs). We illustrate our results in a simulation example.

In this paper, we propose and study a robust variant of the smart predict-then-optimize approach that accounts for prediction shifts due to disturbance in the covariate feature space. While traditional integrated-learning-and-optimization models assume that side information is perfectly revealed, empirical data-driven features are frequently corrupted or noisy at the time of decision-making, leading to fragile operational policies. To bridge this gap, we integrate principles of robust optimization directly into the predictive-prescriptive pipeline via a smart predict-then-robustly optimize loss and establish a computationally tractable convex surrogate, designed to hedge against worst-case feature perturbations. On the theoretical front, we formalize the structural validity of this surrogate by proving its approximation error probability decays exponentially according to a sub-Gaussian concentration profile. Furthermore, we establish that under mild assumptions, the surrogate is Fisher consistent with high probability. We also prove necessary conditions under which our framework outperforms standard smart predict-then-optimize and maintain its superiority even when the standard method is equipped with regularized upstream predictions. Numerical experiments validate that our robust framework consistently yields significant performance improvements over standard methods, both in out-of-sample terms and in training stability.

Due to the complexity of neural network loss landscapes, optimization theory is forced to rely on idealized models, and there is generally a tradeoff between how theoretically tractable the model is, and how accurately it describes the true optimization dynamics. In this work, we stress test the simplest possible model of optimization – the quadratic model – and show that it can be surprisingly predictive in an LLM setting with 150M parameters and 3B training tokens. Specifically, we show that Taylor expanding the model and the loss function at intermediate checkpoints through training can accurately predict the optimization dynamics over windows that can last up to 10% of training. Having established this agreement, we then turn to analyzing the structure of these local quadratic optimization problems through two lenses: the Hessian spectrum and local stability. Using Lanczos quadrature with extremely deep probes, we are able to estimate the Hessian spectrum deep into the tail, and we find a surprising amount of structure in both the eigenvalues and eigenvectors, which depends on the batch size, preconditioner, and training time. We also empirically test local linear stability at intermediate checkpoints and compare it to theoretical predictions to demonstrate that optimization in LLMs typically occurs at a stochastic edge of stability, whose nature is also determined by batch size. Our results indicate the quadratic model may be a theoretically tractable proxy for pretraining optimization dynamics.

Digital twins are often used to predict how a system would respond to an intervention. In systems with feedback, a twin must reproduce an equilibrium counterfactual, and a twin developed in one domain may fail after mechanisms change. We study when these predictions can be validated and transported. For equilibrium causal games, we give conditions on the mechanisms, equilibrium selection, and intervention design under which agreement with experimental distributions identifies the counterfactual of interest. We show why agreement of means and covariances is insufficient for distributional queries. We then introduce cyclic selection diagrams and derive criteria for direct reuse and for hybrid models that combine invariant source mechanisms with target information. An impossibility result constructs systems that agree under every experiment in a finite design but disagree on the target counterfactual, showing that validation requires structural assumptions. For linear models, we derive intervention requirements that depend on the mechanisms that changed, the observation model, and graph support. When point identification fails, we characterize the remaining range of query values. We also provide statistical tests for reconstructed means and covariances and illustrate the theory in synthetic feedback systems.

Barzilai–Borwein (BB) method has shown strong practical performance in continuous optimization, yet its convergence dynamics remains poorly understood. In particular, a central unresolved question is whether BB converges superlinearly for almost every strictly convex quadratic problem and initialization. We provide a negative answer to this question. Specifically, for every finite dimension $n\geq4$, we construct a nonempty open, hence positive-Lebesgue-measure, family of strictly convex quadratic problems and initial points for which the long Barzilai–Borwein method (BB1) converges but cannot converge root-superlinearly. More precisely, with the explicit constants $\rho_{\min}=10^{-6},\rho_{\max}=0.61$, every spectral component of the gradient is bounded above and below by the corresponding geometric sequence. Consequently, the gradient norm and the energy norm of the error satisfy two-sided geometric estimates with the same rates, while the objective gap satisfies the corresponding estimates with squared rates. In particular, all three quantities are bounded below by geometric sequences, ruling out superlinear convergence. The construction is highly nontrivial, based on a computer-assisted proof of a nonresonant, attracting seven-cycle of the projectivized BB dynamics in dimension four.

We derive statistical limit theorems for sample-based approximations of infinite-horizon discounted stochastic optimal control problems in discrete time. Our first result is a functional central limit theorem for the sample-based value function under a uniqueness-type condition on population optimal policies. The limiting law is a mean-zero Gaussian process characterized by a linear fixed point equation that resembles a dynamic programming principle. We compare these asymptotics with those obtained from sample-based policy optimization and illustrate that their limiting variances can be different. We also derive a limit theorem for models with nonunique optimal policies, where the limiting law may be non-Gaussian. Applications to inventory control and renewable harvesting illustrate the theory.

Reliable electric vehicle (EV) charging infrastructure is a cornerstone of sustainable, low-carbon cities, yet urban climate stress such as extreme heat, heavy precipitation, and humidity increasingly raises equipment fault risk and undermines the resilience of urban energy and mobility services. Shifting operation from reactive repair to preventive maintenance depends on accurate, forward-looking fault-risk prediction, a task complicated by the heterogeneous time scales of physical, behavioral, contextual, and historical signals and by forecasting over a multi-week horizon. We develop FGDSE, a feature-governed dynamic stacking ensemble that forms an interpretable decision-support system for climate-resilient charging-asset management. It partitions heterogeneous signals into four feature families, assigns each to a domain expert whose inductive bias matches the data, and adds two deep temporal experts for short-term pulses and long-term degradation; a horizon-wise gating mechanism then learns adaptive weights to forecast daily fault risk over 1 to 30 days. SHAP attribution and an X-learner extend the probabilistic output into causal decision support with post-level treatment effects. On 25 months of data from 13 stations, FGDSE surpasses twelve baselines beyond the ten-day horizon, sustains about 85% macro-recall at 30 days with an AUC decay of only 3.2 points, and reveals a shift of dominance from fault history toward climate stress. It identifies extreme heat as the sole exposure whose causal effect amplifies over time, flagging roughly 30% of posts as heat-sensitive and yielding quantitative thresholds for climate-adaptive maintenance that strengthens urban mobility resilience and sustains low-carbon travel.

Demand response (DR) operation of electrolysis devices is gaining traction to capitalize on volatile electricity markets, but their dynamic operation poses challenges to the durability and lifespan of these systems. Our recently developed multi-scale optimization framework for grid-integrated electrolysis systems studied the impacts of DR and effects on device durability. A major hurdle in that work is tractably scaling the model to include participation in high-frequency electricity markets and/or longer time horizons. To address this, in this work, we developed an aggregate-informed Benders decomposition method to tractably solve large instances of this problem structure. To illustrate the efficacy, we solve a case study with market participation in the day-ahead market (DAM) and real-time markets (RTM) for up to 40 years to effectively capture the effects of device lifespan change. We compare our decomposition algorithm to both commercial solvers and traditional Benders decomposition. We find that incorporating aggregate subproblem information accelerates convergence, reducing the final optimality gap by up to ~81% relative to traditional Benders on the 40-year horizon instances that otherwise stall near 85%. We additionally apply the algorithm to find that RTM participation offers economic advantages under the higher price volatility.

We study a free boundary problem of minimising a functional containing a non-local term rewarding depth into the zero phase: for $u\geqslant 0$ on a bounded, open set $\Omega\subset\mathbb R^d$, we minimise $$J(u) = \int_\Omega \left(\frac12|\nabla u|^2 - fu\right) \;-\; \int_{\{u=0\}} F\big(\mathrm{dist}(x,\partial \{u=0\})\big)\;\mathrm{d}x. $$ This kind of functional arises, for example, from a two-membranes problem with an adhesive contact energy. We first address a well-definedness issue caused by the non-local, boundary-sensitive nature of the functional and prove existence of minimisers, establishing along the way a weak lower semicontinuity result for the non-local term. We then derive stationarity conditions for minimisers, including a variational (Euler–Lagrange type) inequality, a PDE on the positivity set, and, under a mild non-degeneracy assumption, a free boundary condition obtained via inner variations and a Danskin-type differentiation of the distance function.

The paper studies a dynamical system in the Wasserstein space, where the evolution is governed by a Fokker-Planck-Kolmogorov equation with coefficients chosen from a prescribed set that depends at each point on that point and on the current measure. Under the assumption that the multivalued mapping determining the constraints is convex-valued, upper semicontinuous and satisfies certain growth conditions, we prove an existence theorem and establish the compactness of the solution set. Moreover, we study an optimal control problem for such dynamical systems.

We introduce a new generalized Hessian, called the Goldstein second-order $\delta$-subdifferential, and an associated notion of $(\epsilon_1,\epsilon_2,\delta)$-second-order stationary point for continuously differentiable functions with locally Lipschitz gradients. We propose a zeroth-order algorithm based on cubic regularization and Gaussian smoothing with homotopy to find such approximate second-order stationary points for Lipschitz differentiable functions, and derive the iteration complexity under a mild coercivity-type assumption on the objective function.

We study a system that is governed by neural dynamics. Neural dynamics are a model for deep neural networks with a large number of layers. For a differential equation where the right-hand side is given by a neural network, we analyze the exact ensemble controllability of the system. This property plays an essential role in Machine Learning. It concerns the ability to learn to perform different tasks simultaneously with a single neural dynamics: Exact ensemble controllability requires that $N$ different initial states are steered to corresponding $N$ target states with a single set of control parameters. We present a constructive solution to the problem. The construction is based on the solution of a neural interpolation problem. We show that if the right-hand side of the differential equation is given by a neural network of depth two, the interpolation problem can be reduced to the solution of a system of linear equations.

This paper develops a unified method for the exponential stabilization of first-order linear hyperbolic balance laws under general actuation, including both underactuated boundary and in-domain control. The proposed framework brings together a wide range of underactuated configurations within a single formulation and substantially extends existing results restricted to particular actuation settings. Using cutting and folding transformations, the proposed approach is further applied to networks of hyperbolic balance laws, including configurations with cycles. The control design is developed under a stabilizability condition and a robustness assumption. It is based on an invertible backstepping transformation, which partially decouples the system, followed by a reformulation of the stabilization problem at the level of an Integral Difference Equation (IDE). The gains of the resulting dynamic feedback law are constructed at the IDE level by combining a stable rank-reduction procedure, which reduces the problem to a single-input design, with the solution of an interpolation equation arising from a Corona problem. Numerical simulations are presented for a relevant cycle network that cannot be addressed by existing methods in the literature.

In this paper, we study the control co-design (CCD) synthesis problem for a class of systems with parabolic partial differential equation (PDE) dynamics. We first derive a sufficient stability condition for the PDE. By spatially discretizing the PDE and using the sufficient stability condition, we propose a computationally tractable approximate CCD problem. We solve the approximate CCD problem using a gradient-based method. Finally, we justify our proposed approach through an example.

AI product launches require service-capacity decisions before sufficient product-specific workload histories are available. Early service observations update beliefs about remaining workload, but workload does not physically deplete capacity as inventory demand does. Preserving rapid-response capability may also require costly capacity-access and operational arrangements. We study how a firm should jointly decide current capacity deployment and future response readiness. We formulate a finite-horizon dynamic program with workload beliefs, serviceable capacity, and an absorbing response-channel state. A preservation-deployment decomposition solves one deployment problem for each continuation choice and compares the resulting optimized values. This structure identifies when response readiness should be preserved and how much capacity should be deployed. We operationalize the framework through the Cold-Start Belief Actionability Policy (CBAP), which estimates side-specific values using workload scenarios drawn from predictive beliefs. We also establish bounds on its decision and lifecycle performance losses. Numerical experiments and a BurstGPT-based evaluation show that CBAP reduces lifecycle cost relative to benchmark policies by avoiding premature deployment and unnecessary readiness spending. The results distinguish forecast informativeness from operational actionability and show that current deployment and future response readiness are separate decision margins. Zero deployment may represent Standby rather than Exit, while positive deployment need not imply preserving the response channel.

A mathematical model of gas flow in a pipeline controlled by the inlet pressure and the outlet mass flow flux is considered in the form of the isothermal Euler equations with an appropriate equation of state. Within the framework of boundary control systems, this model is transformed into a nonlinear abstract differential equation using an appropriate lifting operator. For this abstract equation, a representation in terms of Fourier coefficients is derived analytically. Conditions for the existence of periodic solutions to a broad class of nonlinear control systems with continuously differentiable controls are established in abstract spaces. This framework is applied to a realistic hydrogen transport model to evaluate periodic operating regimes under periodic fluctuations in supply and demand.

We consider the real-time reoptimisation of log-truck routing and scheduling in the Canadian forestry industry following unforeseen disruptions. Road closures, vehicle breakdowns, travel delays, and demand fluctuations invalidate pre-established tactical plans and call for recovery decisions that simultaneously restore feasibility and limit deviation from the original schedule, two objectives inherently in tension under tight time constraints. We formulate recovery as a sequence of neighbourhood-restricted mixed-integer linear programs parameterised by a deviation bound $k$, the $\ell_1$ distance between the recovered and baseline plans. We show that the minimum feasible value $k_{\min}$ can be computed exactly, anchoring the search at its most stable extreme. Rather than targeting a single $k$, we seek a Pareto front of non-dominated recovery plans spanning the stability–cost space, giving dispatchers a structured set of operational options. To navigate this space efficiently, we propose a reinforcement learning policy, trained with the REINFORCE policy-gradient algorithm, that adaptively selects successive values of $k$ from solver feedback, concentrating computational effort in productive regions and terminating exploration when further improvement is unlikely. Evaluated on weekly instances derived from historical data of a Canadian forestry partner, under disruption scenarios covering all event categories considered, the approach recovers richer Pareto fronts with fewer solver calls than fixed-$k$ grid and dichotomic search baselines, and produces feasible recovery plans within operationally acceptable reoptimization times across all configurations and disruption types.

This work considers the design of first-order convex optimization algorithms and convergence proofs. In particular, we consider nonsmooth Lipschitz and smooth problems accessed through a subgradient or gradient oracle, respectively. For the general class of fixed-step first-order methods, prior work on Performance Estimation Problems (PEPs) has shown that structured, tight convergence proofs typically exist. Under mild conditions, we further show that any first-order method guaranteeing a bound on the primal objective gap $f(x_N)-f(x_\star)$ assuming only a bound on $\|x_0-x_\star\|$ actually has a stronger guarantee on an explicit, computable primal-dual gap at the same rate. These implicit optimal dual certificates, which take the form of affine lower bounds, also provide insight into the role of auxiliary sequences in momentum methods.