[Paper Review] Quantum Multigrid Algorithm for Finite Element Problems

Osama M. Raisuddin, Suvranu De|arXiv (Cornell University)|Apr 11, 2024

Quantum Computing Algorithms and Architecture4 citations

TL;DR

This paper proposes a Quantum Multigrid Algorithm (qMG) that applies multigrid iteration operations directly on quantum states to solve finite element-based linear systems. By encoding the sequence of iterates in a quantum state, qMG achieves exponential speedup in solving such systems, with runtime scaling as O(poly log(N/ε′)) and requiring only O(poly log(N/ε′)) qubits, while leveraging an initial guess to improve convergence efficiency.

ABSTRACT

Quantum linear system algorithms (QLSAs) can provide exponential speedups for the solution of linear systems, but the growth of the condition number for finite element problems can eliminate the exponential speedup. QLSAs are also incapable of using an initial guess of a solution to improve upon it. To circumvent these issues, we present a Quantum Multigrid Algorithm (qMG) for the iterative solution of linear systems by applying the sequence of multigrid operations on a quantum state. Given an initial guess with error e_0, qMG can produce a vector encoding the entire sequence of multigrid iterates with the final iterate having a relative error e'=e/e_0, as a subspace of the final quantum state, with exponential advantage in O( poly log (N/e') ) time using O( poly log (N/e') ) qubits. Although extracting the final iterate from the sequence is efficient, extracting the sequence of iterates from the final quantum state can be inefficient. We provide an analysis of the complexity of the method along with numerical analysis.

Motivation & Objective

• To address the loss of exponential speedup in Quantum Linear System Algorithms (QLSAs) due to growing condition numbers in finite element problems.
• To overcome QLSA limitations in utilizing initial solution guesses to accelerate convergence.
• To develop a quantum algorithm that integrates multigrid iteration into quantum computation for finite element systems.
• To achieve exponential speedup in solving large-scale linear systems arising from finite element discretizations.

Proposed method

• The algorithm applies multigrid operations—restriction, smoothing, and prolongation—directly on quantum states representing the solution iterates.
• It encodes the entire sequence of multigrid iterates into a single quantum state, with the final iterate embedded as a subspace.
• The method leverages quantum amplitude amplification and quantum state preparation to accelerate convergence toward the solution.
• It uses a recursive quantum circuit structure to implement multigrid cycles on quantum states, maintaining logarithmic qubit scaling.
• The algorithm ensures that the final iterate has a relative error ε′ = ε / ε₀, where ε₀ is the initial error, improving efficiency from an initial guess.
• It assumes access to quantum oracles for the finite element matrix and its hierarchical coarse-grid representations.

Experimental results

Research questions

• RQ1 Can multigrid iteration be efficiently implemented on quantum states to solve finite element linear systems with exponential speedup?
• RQ2 How does the quantum multigrid approach handle the condition number growth that undermines QLSA performance?
• RQ3 To what extent can an initial guess be encoded and exploited in a quantum iterative solver to accelerate convergence?
• RQ4 What is the quantum resource cost (qubits and gate count) of maintaining the sequence of multigrid iterates in superposition?
• RQ5 Is the final solution extractable efficiently from the final quantum state while preserving the exponential speedup?

Key findings

• The quantum multigrid algorithm achieves a runtime of O(poly log(N/ε′)) for solving finite element linear systems, offering exponential speedup over classical multigrid methods.
• The algorithm requires only O(poly log(N/ε′)) qubits, making it resource-efficient for large-scale problems.
• The final quantum state encodes the entire sequence of multigrid iterates, with the final iterate having a relative error ε′ = ε / ε₀, where ε₀ is the initial error.
• Extracting the final iterate from the quantum state is efficient, but retrieving the full sequence of iterates is not guaranteed to be efficient.
• Numerical analysis confirms the theoretical complexity scaling and validates the convergence behavior of the algorithm on model finite element problems.
• The method overcomes the condition number barrier that typically negates exponential speedups in QLSA for finite element problems.