Abstract:We propose a physics-informed quantum algorithm to solve nonlinear and multidimensional differential equations (DEs) in a quantum latent space. We suggest a strategy for building quantum models as state overlaps, where exponentially large sets of independent basis functions are used for implicitly representing solutions. By measuring the overlaps between states which are representations of DE terms, we construct a loss that does not require independent sequential function evaluations on grid points. In this sense, the solver evaluates the loss in an intrinsically parallel way, utilizing a global type of the model. When the loss is trained variationally, our approach can be related to the differentiable quantum circuit protocol, which does not scale with the training grid size. Specifically, using the proposed model definition and feature map encoding, we represent function- and derivative-based terms of a differential equation as corresponding quantum states. Importantly, we propose an efficient way for encoding nonlinearity, for some bases requiring only an additive linear increase of the system size $\mathcal{O}(N + p)$ in the degree of nonlinearity $p$. By utilizing basis mapping, we show how the proposed model can be evaluated explicitly. This allows to implement arbitrary functions of independent variables, treat problems with various initial and boundary conditions, and include data and regularization terms in the physics-informed machine learning setting. On the technical side, we present toolboxes for exponential Chebyshev and Fourier basis sets, developing tools for automatic differentiation and multiplication, implementing nonlinearity, and describing multivariate extensions. The approach is compatible with, and tested on, a range of problems including linear, nonlinear and multidimensional differential equations.

Coarse-Gridded Simulation of the Nonlinear Schrödinger Equation with Machine Learning

Learning effective stochastic differential equations from microscopic simulations: Linking stochastic numerics to deep learning.

Numerical Solution of Two-Dimensional Nonlinear Schrödinger Equation Using a New Two-Grid Finite Element Method

Enhancing Open Quantum Dynamics Simulations Using Neural Network-Based Non-Markovian Stochastic Schrödinger Equation Method

Learning effective stochastic differential equations from microscopic simulations: combining stochastic numerics and deep learning

Neural Galerkin Schemes with Active Learning for High-Dimensional Evolution Equations

Deep Stochastic Mechanics

Enhancing Low-Order Discontinuous Galerkin Methods with Neural Ordinary Differential Equations for Compressible Navier--Stokes Equations

Data-driven Vector Soliton Solutions of Coupled Nonlinear Schrödinger Equation Using a Deep Learning Algorithm

Soliton, Breather and Rogue Wave Solutions for Solving the Nonlinear Schrödinger Equation Using a Deep Learning Method with Physical Constraints

Machine Learning Optimization of Compact Finite Volume Methods on Unstructured Grids

Accelerating wavepacket propagation with machine learning

Analytical and numerical aspects of certain nonlinear evolution equations. II. Numerical, nonlinear Schrödinger equation

Enhancement of shock-capturing methods via machine learning

Physics-informed Neural Network Method for the Modified Nonlinear Schrödinger Equation

Physics-Informed Quantum Machine Learning: Solving nonlinear differential equations in latent spaces without costly grid evaluations

A Novel Multilevel Finite Element Method for a Generalized Nonlinear Schrödinger Equation

Time integration schemes based on neural networks for solving partial differential equations on coarse grids

Stable numerical methods for a stochastic nonlinear Schrödinger equation with linear multiplicative noise

Nonlinear Schrödinger Network

Learning Rogue Waves of Nonlinear Schrödinger Equation with Enhanced Physics-Informed Neural Networks Simulator.