Computing a Sparse Approximate Inverse on Quantum Annealing Machines

Sanjay Suresh,Krishnan Suresh

2023-10-04

Abstract:Many engineering problems involve solving large linear systems of equations. Conjugate gradient (CG) is one of the most popular iterative methods for solving such systems. However, CG typically requires a good preconditioner to speed up convergence. One such preconditioner is the sparse approximate inverse (SPAI). In this paper, we explore the computation of an SPAI on quantum annealing machines by solving a series of quadratic unconstrained binary optimization (QUBO) problems. Numerical experiments are conducted using both well-conditioned and poorly-conditioned linear systems arising from a 2D finite difference formulation of the Poisson problem.

Numerical Analysis,Emerging Technologies

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to calculate the Sparse Approximate Inverse (SPAI) on a quantum annealer to accelerate the solution process of large - scale linear equations. Specifically, the paper proposes a new method that uses a quantum annealer to calculate SPAI by solving a series of Quadratic Unconstrained Binary Optimization (QUBO) problems. This SPAI can be used as a pre - conditioner and combined with the classical Conjugate Gradient (CG) method to quickly solve large - scale linear equations on a classical computer.

### Background and Motivation
1. **Solving large - scale linear equations**:
- Many engineering problems require solving large - scale linear equations \( \mathbf{Ku}=\mathbf{f} \), where \( \mathbf{K} \) is a symmetric positive - definite sparse matrix.
- The conjugate gradient method (CG) is a commonly used iterative method, but usually requires a good pre - conditioner to accelerate convergence.

2. **Importance of pre - conditioners**:
- Sparse approximate inverse (SPAI) is an effective pre - conditioner that can significantly improve the convergence speed of the CG method.
- Traditional SPAI calculation methods have high computational costs for large - scale problems and are difficult to implement efficiently on classical computers.

3. **Advantages of quantum computing**:
- Quantum annealers (such as the D - Wave system) have the ability to handle large - scale optimization problems and are insensitive to noise.
- By transforming the SPAI calculation into a QUBO problem, it can be efficiently solved on a quantum annealer.

### Method Overview
1. **Construction of QUBO problems**:
- Transform the SPAI calculation problem into a series of QUBO problems.
- Use a quantum annealer to solve these QUBO problems and obtain each column vector of SPAI.

2. **Utilization of sparse patterns**:
- By taking advantage of the structural characteristics of the finite - difference grid, the number of SPAI columns that need to be calculated can be significantly reduced.
- For a single - material domain, only 9 independent columns need to be calculated; for a dual - material domain, only 21 independent columns need to be calculated.

3. **Algorithm steps**:
- Initialize the sparse pattern of the SPAI matrix.
- For each column vector, if the column has been calculated, directly copy the result; otherwise, calculate it through the QUBO solver.
- Force the symmetry of the SPAI matrix to reduce numerical errors.

### Experimental Results
1. **Performance evaluation**:
- In the single - material domain, the Q - PCG method using SPAI as a pre - conditioner does not have a significant improvement in convergence speed compared to the standard CG method because the matrix is relatively well - conditioned.
- In the multi - material domain, the convergence speed of the Q - PCG method is significantly improved, especially when the condition number is large.

2. **Influence of parameters**:
- The influence of box tolerance and box length on the convergence of the Q - PCG method was studied.
- The results show that appropriate box tolerance and box length can significantly improve the convergence speed and accuracy of Q - PCG.

### Conclusion
This paper proposes a method for calculating SPAI using a quantum annealer and applies it to the solution of large - scale linear equations. The experimental results show that this method has significant performance advantages in the multi - material domain, especially when the matrix condition number is large. Future research directions include extending to quantum - gate computers, exploring greater sparsity, and studying adaptive sparse patterns.