Quantum Variational Solving of Nonlinear and Multi-Dimensional Partial Differential Equations

Abhijat Sarma,Thomas W. Watts,Mudassir Moosa,Yilian Liu,Peter L. McMahon
DOI: https://doi.org/10.1103/PhysRevA.109.062616
2024-06-25
Abstract:A variational quantum algorithm for numerically solving partial differential equations (PDEs) on a quantum computer was proposed by Lubasch et al. In this paper, we generalize the method introduced by Lubasch et al. to cover a broader class of nonlinear PDEs as well as multidimensional PDEs, and study the performance of the variational quantum algorithm on several example equations. Specifically, we show via numerical simulations that the algorithm can solve instances of the Single-Asset Black-Scholes equation with a nontrivial nonlinear volatility model, the Double-Asset Black-Scholes equation, the Buckmaster equation, and the deterministic Kardar-Parisi-Zhang equation. Our simulations used up to $n=12$ ansatz qubits, computing PDE solutions with $2^n$ grid points. We also performed proof-of-concept experiments with a trapped-ion quantum processor from IonQ, showing accurate computation of two representative expectation values needed for the calculation of a single timestep of the nonlinear Black--Scholes equation. Through our classical simulations and experiments on quantum hardware, we have identified -- and we discuss -- several open challenges for using quantum variational methods to solve PDEs in a regime with a large number ($\gg 2^{20}$) of grid points, but also a practical number of gates per circuit and circuit shots.
Quantum Physics
What problem does this paper attempt to address?
The problem that this paper attempts to solve is to use variational quantum algorithms to solve nonlinear and multi - dimensional partial differential equations (PDEs) on quantum computers. Specifically, the goals of the paper are: 1. **Generalize existing methods**: Extend the method proposed by Lubasch et al. so that it can handle a wider range of nonlinear and multi - dimensional PDEs. 2. **Performance evaluation**: Study the performance of variational quantum algorithms on several example equations through numerical simulations, including the single - asset Black - Scholes equation (with a complex nonlinear volatility model), the two - asset Black - Scholes equation, the Buckmaster equation, and the deterministic Kardar - Parisi - Zhang (KPZ) equation. 3. **Experimental verification**: Use IonQ's ion - trap quantum processor for a proof - of - concept demonstration to calculate two representative expected values required for a single time step of the nonlinear Black - Scholes equation. 4. **Challenge analysis**: Identify and discuss several open challenges in using quantum variational methods to solve PDEs under the condition of large - scale grid points (much larger than \(2^{20}\)) while keeping the number of circuit gates and the number of circuit shots reasonable. ### Main contributions of the paper 1. **Method generalization**: The paper proposes a general quantum method that can train intermediate quantum states to represent a large class of nonlinear and inhomogeneous terms, making the method applicable to most nonlinear and multi - dimensional PDEs of interest, especially first - order PDEs in time. 2. **Numerical simulation**: Demonstrate the application of this method on several important equations, such as the Black - Scholes equation in the financial field, the Buckmaster equation in fluid mechanics, and the KPZ equation in surface growth problems. 3. **Hardware implementation**: Conduct preliminary experimental verification on actual quantum hardware, proving the feasibility of this method for small - scale problems. 4. **Challenges and prospects**: Discuss in detail the technical challenges faced in applying this method to large - scale problems and propose future research directions. ### Mathematical description of specific problems The PDEs considered in the paper are in the form: \[ \frac{\partial u}{\partial t}=\hat{O}[u] \] where \(\hat{O}\) is a possibly nonlinear differential operator that depends on \(u\) and its derivatives. The initial condition is: \[ u(x, t = 0)=u_0(x) \] ### Framework of the solution 1. **Quantum state representation**: Represent the solution \(u(x, t)\) as a quantum state \(|u(t)\rangle=\sum_{k = 0}^{2^n-1}u(x_k, t)|k\rangle\), where \(n\) is the number of qubits. 2. **Cost function construction**: Replace the time derivative with a discrete form using a finite - difference approximation (e.g., the backward Euler method) to construct the cost function: \[ C_u(\lambda_0, \lambda)=\left\|(1-\tau\hat{O})|u(\lambda_0, \lambda)\rangle-|\tilde{u}(\tilde{\lambda}_0, \tilde{\lambda})\rangle\right\|^2 \] 3. **Optimization process**: Minimize the cost function through a classical optimization algorithm, update the parameters \(\lambda_0\) and \(\lambda\), and thus find the optimal solution for the next time step. ### Key technologies 1. **Adder operator**: Used to construct spatial derivative terms. 2. **Diagonal operator**: Used to handle nonlinear terms. 3. **Variational parameter selection**: Design two variational parameterization schemes, Real - Valued ZGR - QFT and Universal Layered Ansatz (ULA), which are respectively suitable for smooth functions and discontinuous distributions. Through these techniques and methods, the paper successfully demonstrates how to efficiently solve complex nonlinear and multi - dimensional PDEs on quantum computers.