Efficient Tensor Network Low-Rank Approximation for Solving High-Dimensional PDEs

Abstract

Tensor networks (TNs) were introduced more than 30 years ago to solve high-dimensional many-body quantum problems. In the past decade, methods involving TNs have been utilized across a rapidly growing and diverse range of application domains, far beyond the initial focus on quantum problems. TNs decompose high-dimensional functions into sparsely interconnected, lower-dimensional functions and can break the curse of dimensionality in many application areas, including solving high-dimensional partial differential equations (PDEs). However, the computational cost of solving nonlinear PDEs on low-rank tensor manifolds can exceed that of solving the full-order model for a large class of problems. This issue is the primary barrier to adopting TNs for a large class of computational science and engineering problems because reducing the computational cost is the primary reason low-rank approximations are utilized.

Document Details

Document Type
DoD Grant Award
Publication Date
Feb 06, 2025
Source ID
FA95502510039

Entities

People

  • Hessam Babaee

Organizations

  • Air Force Office of Scientific Research
  • United States Air Force
  • University of Pittsburgh

Tags

Fields of Study

  • Computer science

Readers

  • Distributed Systems and Data Platform Development
  • Linear Algebra

Technology Areas

  • Quantum Computing