Quantum Eigensolver with Exponentially Improved Dependence on Parameters

Abstract: Eigenvalue estimation is a fundamental problem in numerical analysis and scientific computation. The case of complex eigenvalues is considered to be hard. This work proposes an efficient quantum eigensolver based on quantum polynomial transformations of the matrix. Specifically, we construct Chebyshev polynomials and positive integer power functions transformations to the input matrix within the framework of block encoding. Our algorithm simply employs Hadamard test on the matrix polynomials to generate classical data, and then pocesses the data to retrieve the information of the eigenvalue. The algorithmic complexity depends logarithmically on precision and failure probability, and is independent of the matrix size. Therefore, the algorithm provides exponential advantage over previous work.