A coordinated effort to formalize the mathematics of operations research in Lean 4.
Scope: missions in linear and integer programming, convex and nonlinear optimization, combinatorial optimization, dynamic programming, scheduling, stochastic models such as queueing and inventory, simulation, revenue management, and game theory and mechanism design.
Online Resource Allocation under Partially Predictable Demand: The Non-Adaptive Threshold Algorithm Is p + (1−p)/(2−a) − O(√(log n / b)/(a(1−p)p)) CompetitiveResearch Paper
A firm sells b identical units over n periods. In each period at most one customer arrives, pays 1 (type 1) or a∈(0,1) (type 2), and must be accepted or rejected on arrival. This two-fare problem is the basic model of revenue management and of online resource allocation (advertising budgets, hotel rooms, cloud capacity). Two arrival models dominate its…
Delay-Adaptive Learning in Generalized Linear Contextual Bandits: Under Independent Delays with a Tail Envelope, DUCB-GLCB Satisfies the Regret Bound (16) with Probability at Least 1 − 5δResearch Paper
In many sequential decision problems the outcome of a decision arrives late. A recommendation's conversion, a clinical response or a sale is observed hours or weeks after the action, and in the meantime the decision maker has to keep acting. Contextual bandits model the decision side: at each round a learner sees side information, chooses one of K actions…
Rate-Optimal Bayesian Simple Regret in Best Arm Identification: Two-Stage Exploration Has Bayesian Simple Regret at Most Copt/T′+o(1/T)Research Paper
A clinical trial, an A/B test or a simulation study often ends with one decision: which treatment to adopt. Multi-armed bandit theory calls this best arm identification. When the arm means are fixed, the frequentist simple regret after T pulls decays exponentially at each instance, with a rate that degenerates as the gap closes. A practitioner who has a…
The Exact Modulus of the Generalized Concave Kurdyka-Łojasiewicz Property: The Exact Modulus Is the Smallest Concave Desingularizing FunctionResearch Paper
The Kurdyka–Łojasiewicz (KL) inequality is the regularity condition behind most convergence proofs for first-order methods on nonconvex, nonsmooth problems: proximal gradient, PALM, proximal ADMM and their variants. Near a point xˉ, it asks that
On Polyhedral Approximations of the Positive Semidefinite Cone: Every Polytope Between (1−ε)(D − A) and D − A Has Extension Complexity at Least exp(c√n)Research Paper
Linear programs are the workhorse of optimization, and a recurring question is which convex sets can be described, or approximated, by a linear program of small size. The relevant measure is the extension complexityxc(P) of a polytope P: the smallest number of facets of a polytope that projects linearly onto P. If xc(P)=N,…
Hölderian Error Bounds and Kurdyka-Łojasiewicz Inequality for the Trust Region Subproblem: The KL Inequality Holds at Every Optimal Solution with Exponent 3/4 in the TRS-Ill Case and 1/2 OtherwiseResearch Paper
The trust region subproblem (TRS) asks for the minimum of a quadratic function over the Euclidean unit ball. It is the inner step of trust region methods for nonlinear optimization, and it also arises in robust optimization and regularized least squares. When the quadratic form is indefinite the problem is nonconvex, yet it has hidden convexity: strong…
Innovative Business Models in Ocean-Bound Plastic Recycling: additionality raises a for-profit offset seller's social and environmental impacts when the offset market is small, lowers them when largeResearch Paper
About 30 million tons of plastic waste reach the oceans every year (Baron et al.), and over 80% of it is estimated to enter from low- and middle-income coastal countries (Jambeck et al. 2015). Firms such as Plastic Bank and rePurpose Global try to prevent this by working with local recycling supply chains and financing the work through plastic offsets: a…
Adaptive Discretization in Online Reinforcement Learning: AdaMB and AdaQL Have Regret Scaling with the Zooming DimensionResearch Paper
Reinforcement learning in operations problems (inventory control, ambulance routing, pricing, queue admission) often has continuous state and action spaces. A fixed grid of mesh ϵ wastes samples on regions where nothing is learned, and its regret grows with the ambient dimension of S×A. Adaptive discretization refines a…
Approximation Schemes for Capacity-Constrained Assortment Optimization Under the Nested Logit Model: A Dynamic Program Earns 1 − ε of Optimal in Time Polynomial in Input Size, 1/ε and γ⁺/γ⁻Research Paper
The Nested Logit model is one of the most widely used discrete choice models in revenue management. It captures substitution that the Multinomial Logit model cannot: products are grouped into nests, a customer first chooses a nest and then a product in it (Davis, Gallego, Topaloglu, 2014; Gallego, Topaloglu, 2014). A retailer that decides which products to…
An SDE Perspective on Stochastic Convex Optimization: The Gradient SDE Has a Unique Solution and, Under Square-Integrable Noise, Converges Almost Surely to a MinimizerResearch Paper
First-order methods for convex minimization rarely see an exact gradient: in stochastic gradient descent and its relatives the gradient is perturbed by sampling or measurement noise. A standard way to study the long-time behaviour of such methods is to pass to continuous time, where the deterministic gradient flow x˙=−∇f(x) is replaced by a…
Quantitative Convergence for Displacement Monotone Mean Field Games with Controlled Volatility: Open-Loop Nash Equilibria Converge to the Mean Field Equilibrium at Rate 1/NResearch Paper
Mean field games describe Nash equilibria of stochastic differential games with very many symmetric players, each of whom interacts with the others only through the empirical distribution of their states. The limiting model, introduced by Lasry and Lions and by Huang, Malhamé and Caines, replaces the population by a measure flow and the N-player equilibrium…
A Squared Smoothing Newton Method for Semidefinite Programming: Under Primal and Dual Constraint Nondegeneracy the Huber-Based Method Converges Superlinearly with Order 1+τResearch Paper
Semidefinite programming (SDP) minimizes a linear function of a symmetric matrix subject to linear equations and positive semidefiniteness. It is a standard modelling tool in control, combinatorial optimization, statistics and quantum information. Interior point methods solve SDPs accurately but each iteration factorizes a dense m×m Schur complement,…
Performance of the Smallest-Variance-First Rule in Appointment Sequencing: under the dilation ordering, svf with mean-based schedules costs at most four times the optimal sequenceResearch Paper
In outpatient clinics, surgery blocks and other appointment systems, a provider sees n clients one after another, each with a random service time. Two decisions determine how long clients wait and how long the provider sits idle: the sequence (who goes in which slot) and the schedule (how long each slot is). Finding the optimal sequence is hard even with…
Designing Information to Engage Customers: Under Observable Engagement the Seller-Optimal Signal Reveals Everything, Exaggerates a Low Prior or Downplays a High OneResearch Paper
Firms increasingly let customers take part in designing the product they will buy: configurators, prototypes, trial versions and co-design platforms. Two features of these settings interact. The firm controls what the customer learns about the product's basic value (it can tune a trial, a simulation or a test), and the customer can invest costly engagement t…
Compact Extended Formulations for Low-Rank Functions with Indicator Variables: clconv(Q) from the Pieces V(I) with ∣I∣≤k and the Ray Set RResearch Paper
Many optimization models in statistics, machine learning and finance minimize a convex function of a continuous vector x∈Rn under logical conditions on its support. Best subset selection, sparse regression and sparse portfolio selection are standard cases. A binary indicator variablezi models such a condition: zi=0 forces xi=0, and zi=1 leave…
A Superlinearly Convergent Subgradient Method for Sharp Semismooth Problems: One PolyakBundle Step Gives f(x̃) − f* ≤ Cₛ(f(x) − f*)^(1+η) Near a Sharp (b)-Regular MinimizerResearch Paper
Subgradient methods are the basic tool for minimizing a nonsmooth function f:Rd→R from first-order information alone. At a sharp minimum, where f grows at least linearly away from its solution set, the Polyak step x+=x−∥v∥2f(x)−f∗v converges linearly (Polyak 1969; in the convex and weakly convex settings, Davis,…
Static Pricing for Multi-unit Prophet Inequalities: The Static Price Equalizing E[min(Xₚ,k)]/k and P[Xₚ ≤ k − 1] Earns at Least φₖ, Its Poisson Value, of the Expected Optimal WelfareResearch Paper
A seller has k identical units and faces n buyers who arrive one at a time. Each buyer's value is drawn independently from a known distribution, but the arrival order is adversarial and unknown. The simplest selling policy posts one static price to every buyer until the units run out. Prophet inequalities compare such a policy with a prophet who sees all…
Diffusion Approximations for a Class of Sequential Experimentation Problems: The Optimal Stopping Value of the Belief Diffusion Is the Maximum of the Closed-Form Two-Action ValuesResearch Paper
A firm about to launch a product, a platform choosing which items to display, or a regulator testing a treatment must pick one action from a finite menu while uncertain about which of two states of the world holds. Before committing, it can run experiments, each of which shifts its belief. Experimenting has a cost in time, modelled by discounting, and the…
Stochastic Subgradient Descent Escapes Active Strict Saddles on Weakly Convex Functions: The Iterates Almost Surely Avoid Active Strict Saddles Satisfying the Verdier and Angle ConditionsResearch Paper
Training a neural network or fitting a nonsmooth statistical model usually means running the stochastic subgradient method on a function that is neither smooth nor convex. Convergence theory gives limit points that are Clarke critical (Davis et al., 2020; Bolte and Pauwels, 2021), but Clarke critical points include points that no one wants, and a…
Fast Convex Optimization via Time Scale and Averaging of the Steepest Descent: Inertial Dynamics with Implicit Hessian Damping Converge Weakly with Gradient Rate o(1/s²)Research Paper
Second-order dynamics with vanishing damping are the continuous-time models of accelerated first-order methods, so a rate proved for such a dynamic predicts the behaviour of an algorithm class used across large-scale convex optimization. The line of work runs as follows.
Playing Divide-and-Choose Given Uncertain Preferences: For Independent Normal Priors, Algorithm 1 Is a Fully Polynomial-Time Approximation Scheme for the Divider's UtilityResearch Paper
Divide-and-choose is the oldest fair-division protocol: one player cuts, the other picks. With known preferences its analysis is classical: the chooser always gets at least half of her value (proportionality), and the divider, knowing her values, cuts so that she is indifferent and keeps everything above half that the difference in tastes allows. In…
Diffusion Limit of Fair Resource Control: Under Aθ < 0 and Bounded Workload, Stationary Workloads Converge to the Diffusion Limit's Stationary Law, with Moments of Order m < p − 1Research Paper
Bandwidth-sharing networks model data transfers that each occupy several links of a communication network at once. Under proportional fair (PF) allocation, introduced by Kelly, Maulloo and Tan (J. Oper. Res. Soc. 1998), the capacity of the links is split among the active transfers by maximizing a weighted sum of logarithms. Exact steady-state distributions…
Model-Based Reinforcement Learning for Offline Zero-Sum Markov Games: VI-LCB-Game Returns an ε-Nash Policy Pair with Probability 1−δ from c₁·C⋆_clipped·S(A+B)·log(N/δ)/((1−γ)³ε²) Offline SamplesResearch Paper
Many sequential decision problems with two adversaries are modelled as zero-sum Markov games (Shapley, 1953; Littman, 1994): two players act simultaneously in a shared state, one maximizes and the other minimizes the same discounted reward. In applications such as games, security and robust control, interaction with the system is expensive or unsafe, and…
Transparency and Control in Platforms for Networked Markets: Open-Access Cournot Competition Has Price of Anarchy at Most (3/2)(1−1/(3n+6)), Tight When All Markets Share One Willingness to PayResearch Paper
Online marketplaces such as eBay, Etsy, Amazon and Alibaba, ride-sharing services and electricity markets are platforms: they stand between many producers and many geographically or logically separate markets, and they decide which producer may serve which market. A basic design question is how much welfare is lost when the platform simply lets everyone…
Conditional Distributionally Robust Functionals: The Conditional Risk Measure of Atom-Supported Sets 𝔄ᵢ Equals the Conditional Distributionally Robust Functional of Their Mixture 𝔐_ν = Σ νᵢ𝔄ᵢResearch Paper
A distributionally robust functional evaluates a random cost Z by its worst expectation over a set M of probability measures, the ambiguity set: R(Z)=supQ∈MEQ[Z]. Functionals of this form are the dual representation of coherent risk measures and the objective of distributionally robust optimization. Multistage…