About

I work in the calculus of variations, nonlinear partial differential equations, and formalized mathematics at the University of Pittsburgh.

I am completing an M.S. in Mathematics with a thesis on stochastic tug-of-war games and the regularized p-Laplacian, advised by Juan J. Manfredi. I also work with Irene Fonseca at Carnegie Mellon University on variational problems under variable-coefficient differential constraints.

I am also a Ph.D. student in Economics at Pittsburgh, and studied Mathematical Sciences at Sharif University of Technology, Tehran, Iran.

Research

Variational problems under variable-coefficient differential constraints

Current project, supervised by Irene Fonseca. Manuscript in preparation.

I study weak lower semicontinuity of integral energies whose fields satisfy a linear differential constraint with variable coefficients. The condition that replaces convexity is 𝒜-quasiconvexity of the integrand with the coefficients frozen at each point. For Lipschitz coefficients, I prove that this condition is sufficient and necessary under separate rank assumptions, using a blow-up argument for sufficiency and, for necessity, a moving-kernel construction adapted from Davoli and Fonseca.

Project details

P. M. Santos characterized lower semicontinuity in this setting for smooth coefficients of uniform constant rank, using pseudodifferential operators and Young measures. For sufficiency, the blow-up reduces the problem to a single frozen, constant-coefficient operator, so constant rank is needed only pointwise and the rank may vary from point to point. For necessity, the construction needs the rank to be locally constant along each lattice direction.

Stochastic tug-of-war games and the regularized p-Laplacian

M.S. thesis, advised by Juan J. Manfredi.

The value functions of a tug-of-war game with noise converge, as the step size vanishes, to solutions of a p-Laplace-type equation. The thesis rests on an affine lift that turns the regularized p-Laplace equation into the ordinary one in one more dimension, and develops it in two directions: a projected game whose values converge to the viscosity solution, and a pointwise asymptotic mean-value characterization of classical solutions. Both appear among the papers below.

Moving-kernel construction, schematic. Each arrow lies in the kernel of the symbol at its own point, so the arrows turn as the coefficients vary; the inset freezes the coefficients at one point.
A sample path of a tug-of-war game with noise, run until the token leaves the domain at the marked point; see the published paper.

Mathematics papers

  1. 2026

    A formally verified convergence theorem for the tug-of-war scheme of the normalized p-Laplacian

    Preprint

    A Lean 4 formalization of existence, uniqueness, and uniform convergence for the tug-of-war scheme of the normalized p-Laplacian, for 2 ≤ p < ∞.

    Paper details

    For bounded open domains satisfying an exterior sphere condition and fixed bounded continuous data, the discrete solutions converge uniformly on the closure to the unique viscosity solution. The development includes the scheme expansion, half-relaxed limits, boundary estimates, comparison, Aleksandrov’s theorem, and Jensen’s lemma. The contribution is a formalization of these established results in Lean 4 and Mathlib.

  2. 2026

    A projected tug-of-war game for the regularized p-Laplacian

    Published in Journal of Elliptic and Parabolic Equations

    A stochastic-game interpretation of the regularized p-Laplace equation for p ≥ 2, built on an affine lift that turns it into the ordinary p-Laplace equation one dimension higher.

    Paper details

    Projecting the higher-dimensional game back to the original domain gives a dynamic programming principle with strip boundary data. I prove existence, uniqueness and Borel measurability for it, identify its solution with the value of the projected game, and prove convergence to the viscosity solution as the step size tends to zero. The lift introduces artificial caps on the cylinder; the compactness estimates are arranged to stay away from them.

  3. 2026

    An asymptotic mean value characterization for the regularized p-Laplacian

    Preprint

    For C2 functions, solving the regularized p-Laplace equation is equivalent to a pointwise asymptotic mean-value identity, throughout 1 < p < ∞.

    Paper details

    The identity combines tilted extrema with a weighted local average. Because the lifted gradient never vanishes, no separate treatment of critical points is needed, unlike the classical p-harmonic theory. An example built from Aronsson's function shows that the identity fails pointwise at the formal endpoint p = ∞. Convergence of the associated game for p ≥ 2 is the subject of the companion paper.

  4. 2025

    Extended set difference: an inverse operation of Minkowski summation

    Accepted in Journal of Convex Analysis

    Minkowski addition has no exact inverse in general. We define a difference of compact convex sets by minimizing the Hausdorff distance between A and B + X.

    Paper details

    We prove that a minimizer exists for every pair of nonempty compact convex sets. Minimizers need not be unique, so we study the whole collection: bounds on its size, stability under transformations and perturbations, and a linear-optimization method for computing it.

  5. 2025

    Optimal control of an epidemic with intervention design

    Preprint

    Optimal vaccination in an SEIR model with a hard healthcare-capacity constraint, an infinite horizon, and a delay between decision and implementation.

    Paper details

    Moreau–Yosida regularization approximates the hard constraint by penalized problems, and Γ-convergence relates finite- and infinite-horizon solutions. Pontryagin conditions characterize the intervention strategies, including boundary arcs on which infections stay at the capacity limit, and numerical examples measure the cost of delay.

Download BibTeX for all papers

Lean formalization

From the tug-of-war scheme to the normalized p-Laplacian

Joint work with Juan J. Manfredi.

The companion website for our convergence paper pairs mathematical statements with their Lean proofs and shows the dependencies between results. The development starts from the dynamic programming principle and follows the analytic route to convergence using Lean 4 and Mathlib.

Education

  • Expected 2027 M.S. in Mathematics, University of Pittsburgh. Thesis advisor: Juan J. Manfredi.
  • 2023 – present Ph.D. student in Economics, University of Pittsburgh (M.A. 2025).
  • 2016 – 2022 B.S. in Mathematical Sciences, Sharif University of Technology.

Talks and meetings

  • Dec 2026 Contributed talk (scheduled), CFE–CMStatistics 2026, Berlin: Asymptotically optimal test statistics for simulation-based finite-sample inference.
  • Fall 2026 Participant, working group on the De Giorgi conjecture, Center for Nonlinear Analysis, Carnegie Mellon University.
  • Oct 2026 Participant, Frontiers of Mathematical and Computational Aspects of Materials Science, Carnegie Mellon University.
  • 2025, 2026 Contributed talks, Midwest Econometrics Group Annual Meeting.
  • Jun 2026 Participant, summer school on Topics at the Interface of Partial Differential Equations and Probability, Carnegie Mellon University.

Teaching

  • 2022 – 2026 Teaching assistant, University of Pittsburgh: graduate stochastic processes, probability and statistics, microeconomics, and applied econometrics.
  • 2020 – 2022 Teaching assistant, Sharif University of Technology: abstract algebra, topology, and measure theory.
  • 2019 Instructor, Introduction to Linear Algebra for Engineers, Sharif University of Technology, with sole responsibility for lectures and assessment.

Lecture notes

Other papers

Econometrics and statistics.

  1. 2026

    Partial identification with auxiliary moment restrictions

    Preprint

    How known population aggregates restrict the sharp identification region of the best linear predictor when outcomes are observed only as intervals.

    Paper details

    The added information constrains the admissible completions of the data without changing the reported intervals. Some restrictions reduce the dimension of the region; others narrow it without reducing its dimension. The paper derives directional measures of identifying value and applies them to interval-valued wages in the Current Population Survey.

  2. 2026

    Public good provision under locally private signals

    Preprint

    Optimal public-good provision when preferences are observed only through a privacy channel that randomizes individual reports.

    Paper details

    The optimal allocation is determined by an aggregate posterior score combining expected valuations and virtual values. The implementing transfers solve a Fredholm integral equation: completeness gives uniqueness when a solution exists, while existence needs a separate range condition. Comparisons between privacy channels depend on how their privacy levels are calibrated.

  3. 2025

    Bounds for restricted selections of random sets

    Preprint

    Sharp bounds on the means, medians and event probabilities attainable by selections of a random interval that satisfy aggregate restrictions.

    Paper details

    On a non-atomic probability space we characterize the restricted selection sets and derive the bounds by rearrangement and convex duality. For random compact convex sets in higher dimensions, an approximation by unions of random cubes reduces the problems to coordinate-wise interval problems.

Download BibTeX for all papers

Software

LangevinFlow is an R package for sampling probability distributions using the unadjusted and Metropolis-adjusted Langevin algorithms. It has a C++ backend and is available on CRAN.

I also work in Python, Julia, MATLAB, Lean 4 and LaTeX.