Preconditioned low-rank Riemannian optimization for linear systems with tensor product structure
Предобусловленная низкоранговая риманова оптимизация для линейных систем со структурой тензорного произведения
2015-08-12
SCID: 54.1/rjammkr3
Discuss with AI
Tucker formatapproximate Riemannian Newtonlow-rank Riemannian optimizationpreconditioned Richardsontensor train
Figures from the paper
Abstract (AI)
The numerical solution of partial differential equations on high-dimensional domains gives rise to computationally challenging linear systems. When using standard discretization techniques, the size of the linear system grows exponentially with the number of dimensions, making the use of classic iterative solvers infeasible. During the last few years, low-rank tensor approaches have been developed that allow to mitigate this curse of dimensionality by exploiting the underlying structure of the linear operator. In this work, we focus on tensors represented in the Tucker and tensor train formats. We propose two preconditioned gradient methods on the corresponding low-rank tensor manifolds: A Riemannian version of the preconditioned Richardson method as well as an approximate Newton scheme based on the Riemannian Hessian. For the latter, considerable attention is given to the efficient solution of the resulting Newton equation. In numerical experiments, we compare the efficiency of our Riemannian algorithms with other established tensor-based approaches such as a truncated preconditioned Richardson method and the alternating linear scheme. The results show that our approximate Riemannian Newton scheme is significantly faster in cases when the application of the linear operator is expensive.
Key Findings
1
Developed an efficient approach for solving the Newton equation arising from the Riemannian Hessian in the approximate Newton scheme.
2
Numerical experiments compare the proposed Riemannian algorithms to truncated preconditioned Richardson and alternating linear scheme baselines.
3
Proposed two preconditioned gradient methods on low-rank tensor manifolds for Tucker and tensor train formats: a Riemannian preconditioned Richardson and an approximate Riemannian Newton scheme.
4
The approximate Riemannian Newton scheme is significantly faster than competing tensor-based approaches when applying the linear operator is expensive.
Research Object
High-dimensional linear systems with tensor product structure represented in low-rank Tucker and tensor-train formats
Research Subject
Preconditioned low-rank Riemannian optimization methods (preconditioned Richardson and approximate Riemannian Newton) and their efficiency/solver behavior for solving these tensor-structured linear systems, including efficient solution of the Riemannian Newton equation
Publication Details
Publication Date
2015-08-12
Journal
Publisher
ISSN
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest