Abstract:We consider the approximate solution of parametric PDEs using the low-rank Tensor Train (TT) decomposition. Such parametric PDEs arise for example in uncertainty quantification problems in engineering applications. We propose an algorithm that is a hybrid of the alternating least squares and the TT cross methods. It computes a TT approximation of the whole solution, which is beneficial when multiple quantities of interest are sought. This might be needed, for example, for the computation of the probability density function (PDF) via the maximum entropy method [Kavehrad and Joseph, IEEE Trans. Comm., 1986]. The new algorithm exploits and preserves the block diagonal structure of the discretized operator in stochastic collocation schemes. This disentangles computations of the spatial and parametric degrees of freedom in the TT representation. In particular, it only requires solving independent PDEs at a few parameter values, thus allowing the use of existing high performance PDE solvers. In our numerical experiments, we apply the new algorithm to the stochastic diffusion equation and compare it with preconditioned steepest descent in the TT format, as well as with (multilevel) quasi-Monte Carlo and dimension-adaptive sparse grids methods. For sufficiently smooth random fields the new approach is orders of magnitude faster.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to efficiently solve parameter - dependent partial differential equations (PDEs) in engineering applications, especially in uncertainty quantification problems. Specifically, the paper focuses on using low - rank Tensor Train (TT) decomposition to approximately solve such PDEs. Such problems are very important in practical applications, for example, in stochastic groundwater flow modeling, parameterization is achieved through Karhunen - Loève expansion: \[ c(x,y) = c_0(x) + \sum_{k = 1}^{d} y_k\psi_k(x), \quad y_k\sim U(-\sqrt{3}, \sqrt{3}) \] Or for the log - normal and log - uniform cases: \[ c(x,y) = \exp\left(\sum_{k = 1}^{d} y_k\psi_k(x)\right), \quad y_k\sim N(0,1) \text{ or } y_k\sim U(-\sqrt{3}, \sqrt{3}) \] ### Main Challenges 1. **High - Dimensional Parameter Space**: The parameter dimension \( d \) can range from tens to thousands. The classical tensor - product one - dimensional quadrature rule is not feasible in this case because the number of quadrature points will increase exponentially with \( d \). 2. **Computational Complexity**: Traditional solution methods such as the Monte Carlo and Quasi - Monte Carlo (QMC) methods can handle high - dimensional problems, but they have a slow convergence rate or high computational cost. 3. **Quantities of Interest (QoI)**: Sometimes it is necessary to calculate multiple QoI, for example, calculating the probability density function (PDF) through the maximum entropy method, which requires an efficient approximation of the entire solution. ### Solutions Proposed in the Paper The paper proposes a new method that combines the Alternating Least Squares (ALS) and TT Cross algorithms. The main features of this method include: - **Maintaining Block - Diagonal Structure**: Utilize the block - diagonal structure of the discretization operator in the stochastic collocation scheme to separate the calculations of spatial and parametric degrees of freedom. - **Efficiently Solving Independent PDEs**: Only need to solve independent PDEs at a few parameter values, thus allowing the use of existing high - performance PDE solvers. - **Low - Rank Approximation**: Achieve a low - rank approximation of the entire solution through TT decomposition, thereby greatly reducing storage and computational requirements. ### Results of Numerical Experiments The paper verifies the effectiveness of the new algorithm through numerical experiments. In particular, when dealing with sufficiently smooth random fields, the new method is several orders of magnitude faster than existing methods such as QMC and sparse grids.

A hybrid Alternating Least Squares -- TT Cross algorithm for parametric PDEs

Distributed memory parallel adaptive tensor-train cross approximation

Tensor Network Space-Time Spectral Collocation Method for Time Dependent Convection-Diffusion-Reaction Equations

Tensor Network Space-Time Spectral Collocation Method for Time-Dependent Convection–Diffusion–Reaction Equations

Computation of the Response Surface in the Tensor Train data format

Tensor Train Decomposition for Solving Large-Scale Linear Equations.

Randomized algorithms for computing the tensor train approximation and their applications

Tensor Networks for Solving the Time-Independent Boltzmann Neutron Transport Equation

Tensor Networks for Solving Realistic Time-independent Boltzmann Neutron Transport Equation

Kriging in Tensor Train data format

Parametric PDEs: Sparse or Low-Rank Approximations?

Cross Interpolation for Solving High-Dimensional Dynamical Systems on Low-Rank Tucker and Tensor Train Manifolds

Nonnegative tensor train for the multicomponent Smoluchowski equation

Preconditioned low-rank Riemannian optimization for linear systems with tensor product structure

APTT: An accuracy-preserved tensor-train method for the Boltzmann-BGK equation

Adaptive Cross Tubal Tensor Approximation

A semi-Lagrangian Vlasov solver in tensor train format

Randomized Alternating Least Squares for Canonical Tensor Decompositions: Application to a PDE with Random Data

Tensor Network Space-Time Spectral Collocation Method for Solving the Nonlinear Convection Diffusion Equation

Low-rank tensor approximation for Chebyshev interpolation in parametric option pricing