A Tensor-Network Chebyshev-Galerkin Method for Boundary-Value Problems in Computational Mechanics
math.NA, cs.NA, quant-ph
Submitted: 2026-11-04
Updated: 2026-11-04
Comments: 27 pages, 11 figures, 3 tables. Code and data: https://doi.org/10.5281/zenodo.23931798
License: http://creativecommons.org/licenses/by/4.0/
The gist: Quantum-inspired numerical methods for partial differential equations represent discretized fields and operators as quantics tensor trains (QTT)…
Terminology
Abstract
Quantum-inspired numerical methods for partial differential equations represent discretized fields and operators as quantics tensor trains (QTT), so that their cost depends on how compressible the problem is rather than on the number of unknowns. Global spectral formulations in this setting have so far relied on Fourier bases, restricted to periodic problems, or on collocation, which does not preserve the variational structure exploited by tensor-network solvers. We introduce a QTT Chebyshev-Galerkin method that assembles and solves global spectral systems directly in compressed form, without forming the underlying arrays. Starting from the weak formulation, the method handles coupled elliptic systems with variable coefficients, smoothly mapped geometries, and mixed Dirichlet-Robin boundary conditions, and yields symmetric positive-definite systems suited to variational tensor-network solvers. Parameters and random inputs enter as additional coordinates, so the same construction returns the solution over an entire parameter space in a single solve. In linear elasticity, the method retains spectral convergence and reaches errors below 10-8 on a three-dimensional benchmark, beyond those attained by QTT finite differences and Q2/Q4 FEniCSx finite elements over the tested range. A six-parameter cantilever design study covering 1.7 times10 7 designs is solved in a single 92s run, and a stochastic problem with ten random coordinates is represented with 2.4 times10 6 times less memory than its full coefficient array.
Sources
- SeeMPS: A Python-based Matrix Product State and Tensor Train Library
- Comparative study of matrix product state/quantized tensor-train algorithms for solving time-independent partial differential equations
- Stable full-field simulation of a multiscale elliptic equation by means of Quantized Tensor Trains
Related papers
- Do physics-informed neural networks (PINNs) need to be deep? Shallow PINNs using the Levenberg-Marquardt algorithm
- A Neural-preconditioned Poisson Solver for Mixed Dirichlet and Neumann Boundary Conditions
- Advanced Linear Algebra with Applications - Part I (Numerical linear algebra for PDEs, machine learning, and data assimilation)
- Operator learning meets inverse problems: A probabilistic perspective
- Second-order consistency for learning chaotic dynamics via randomized Jacobian matching
- Windowed thinning and query complexity for the bouncy particle and Zigzag samplers