Postdoctoral Researcher · Mathematics

Abhishek Singh

Optimization — multi-objective, vector, and generalized Nash equilibrium problems

I work on convergent numerical methods for multi-objective and vector optimization, building quasi-Newton, trust-region, and conjugate gradient algorithms and pushing them from the single-objective setting into the vector-valued one. My doctoral work developed inexact Newton and BFGS-type frameworks for generalized Nash equilibrium problems, with global and superlinear convergence results for multi-player games. That line of work now extends into unconstrained and constrained multi-objective optimization, aimed at methods that hold up both in theory and at scale.

Editorial roles
01

Research interests

Multi-objective & vector optimization Pareto stationarity & scalarization Generalized Nash equilibrium problems Trust-region methods Quasi-Newton / BFGS Nonlinear conjugate gradient Inexact Newton methods Nonmonotone line search
02

Academic appointments

03

Education

04

Publications

  1. [1]

    A trust-region algorithm with a \(T_k\)-based nonmonotone strategy for unconstrained multiobjective optimization

    J.W. Peng, J. Ren, A. Singh\(^*\), E. Köbis

    Journal of Nonlinear and Variational Analysis

    (Accepted for publication)

    Chinese Academy of Sciences - Journal Division: Tier 1 SCIE · Q2 IF 1.5
  2. [2]

    Improved nonmonotone quasi-Newton method for multiobjective optimization problems

    K. Kumar, A. Singh, A. Upadhayay, D. Ghosh\(^*\)

    Journal of Nonlinear and Convex Analysis, 27(1), 119–142 · 2026

    SCIE · Q2 IF 1.5 BFGS Nonmonotone
    Abstract

    Surveys the components of quasi-Newton methods for multiobjective optimization — Hessian approximations, search directions, and step lengths — then proposes an improved nonmonotone BFGS-based scheme. Establishes well-posedness and global convergence under mild assumptions, with numerical results on standard test problems.

  3. [3]

    A novel modified Liu–Storey nonlinear conjugate gradient method for solving vector optimization problems

    J.-W. Peng, D.-H. Zhong, A. Singh\(^*\)

    Optimization, 1–30 · 2025 (Taylor & Francis)

    SCIE · Q1 IF 1.8 Conjugate gradient Wolfe line search
    Abstract

    Addresses a gap in Gonçalves et al.'s Liu–Storey extension for vector optimization, where descent is not guaranteed even under exact line search. Proposes a modified Liu–Storey scheme with a standard Wolfe line search that restores descent and proves global convergence for nonconvex vector optimization problems, backed by numerical experiments.

  4. [4]

    Inexact Newton method for solving generalized Nash equilibrium problems

    A. Singh\(^*\), D. Ghosh, Q. H. Ansari

    Journal of Optimization Theory and Applications, 201(3), 1333–1363 · 2024 (Springer)

    SCIE · Q1 GNEP Q-quadratic convergence
    Abstract

    Develops an inexact Newton method for player-convex and jointly convex generalized Nash equilibrium problems, reformulated via a complementarity function. Proves global convergence for both problem classes, with Q-quadratic convergence under strong BD-regularity and a suitable forcing sequence. Tested on an internet-switching game, outperforming an existing semi-smooth Newton method.

  5. [5]

    Improved nonmonotone adaptive trust-region method to solve generalized Nash equilibrium problems

    A. Singh, K. Kumar, D. Ghosh\(^*\)

    Journal of Nonlinear and Convex Analysis, 25(1), 11–29 · 2024

    SCIE · Q2 IF 1.5 Trust-region Error-bound
    Abstract

    Proposes an improved nonmonotone adaptive trust-region method for constrained optimization under loose error-bound conditions, applied to GNEPs where Newtonian methods often struggle with nonunique local solutions. Retains the local convergence properties of the nonmonotone trust-region method while adding global convergence, and outperforms it numerically.

  6. [6]

    A globally convergent improved BFGS method for generalized Nash equilibrium problems

    A. Singh, D. Ghosh\(^*\)

    SeMA Journal, 81(2), 235–261 · 2024 (Springer)

    ESCI · Q2 Armijo–Goldstein Global convergence
    Abstract

    Solves a player-convex GNEP with an improved BFGS method using an Armijo–Goldstein line search rather than the costlier Wolfe-type search. Reworks the Hessian update so it keeps positive definiteness under Armijo-type search, proves global convergence, and tests the method on three standard GNEPs plus two internet-switching problems.

  7. [7]

    Extended Karush–Kuhn–Tucker condition for constrained interval optimization problems and its application in support vector machines

    D. Ghosh\(^*\), A. Singh, K. K. Shukla, K. Manchanda

    Information Sciences, 504, 276–292 · 2019 (Elsevier)

    SCIE · Q1 IF 6.8 KKT conditions SVM
    Abstract

    Derives extended Fritz John and Karush–Kuhn–Tucker optimality conditions for constrained interval optimization problems, built from the geometric fact that the cone of feasible directions and the set of descent directions cannot overlap at an optimum. Applies the result to binary classification with interval-valued data via support vector machines.

05

Conferences & workshops

06

Service & awards

Academic service

Awards & honors

  • 2016CSIR-JRF, Mathematical Sciences — All India Rank 22
  • 2015GATE, Mathematical Sciences — All India Rank 624
07

Computer skills

C C++ MATLAB Python (NumPy, SciPy) LaTeX / BibTeX
Copied BibTeX to clipboard.