Dong An,Andrew M. Childs,Lin Lin,Lexing Ying
Abstract:Eigenvalue transformations, which include solving time-dependent differential equations as a special case, have a wide range of applications in scientific and engineering computation. While quantum algorithms for singular value transformations are well studied, eigenvalue transformations are distinct, especially for non-normal matrices. We propose an efficient quantum algorithm for performing a class of eigenvalue transformations that can be expressed as a certain type of matrix Laplace transformation. This allows us to significantly extend the recently developed linear combination of Hamiltonian simulation (LCHS) method [An, Liu, Lin, Phys. Rev. Lett. 131, 150603, 2023; An, Childs, Lin, <a class="link-https" data-arxiv-id="2312.03916" href="https://arxiv.org/abs/2312.03916">arXiv:2312.03916</a>] to represent a wider class of eigenvalue transformations, such as powers of the matrix inverse, $A^{-k}$, and the exponential of the matrix inverse, $e^{-A^{-1}}$. The latter can be interpreted as the solution of a mass-matrix differential equation of the form $A u'(t)=-u(t)$. We demonstrate that our eigenvalue transformation approach can solve this problem without explicitly inverting $A$, reducing the computational complexity.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is to develop an efficient method to represent and implement matrix eigenvalue transformations on quantum computers, especially those transformations that cannot be processed by the singular value transformation method. Specifically, the paper proposes a Laplace - based Linear Hamiltonian Simulation combination (Lap - LCHS) method, which can be extended to a broader class of eigenvalue transformations, such as the power of the matrix inverse \( A^{-k} \) and the exponent of the matrix inverse \( e^{-A^{-1}} \), etc. This method allows problems to be solved without explicitly inverting the matrix \( A \), thereby reducing the computational complexity.
### Main contributions of the paper
1. **Proposing a new method**: The paper proposes a Laplace - based Linear Hamiltonian Simulation combination (Lap - LCHS) method for implementing a class of eigenvalue transformations.
2. **Theoretical basis**: Through the Laplace transform, the eigenvalue transformation is represented as a linear combination of matrix exponential functions, and then it is further interpreted as a linear combination of Hamiltonian simulation problems.
3. **Algorithm implementation**: The corresponding quantum algorithm is constructed, and the eigenvalue transformation is implemented by discretizing the integral and using the Quantum Linear Combination Unit (LCU) technique.
4. **Complexity analysis**: The complexity of the algorithm is analyzed in detail, and its optimality or near - optimality in terms of the number of queries and the cost of initial state preparation is proven.
5. **Application examples**: The advantages of the Lap - LCHS method in solving specific problems such as non - homogeneous differential equations, the power of the matrix inverse, linear differential equations with non - diagonal mass matrices, and second - order differential equations are demonstrated.
### Key techniques in the paper
- **Laplace transform**: Represent the eigenvalue transformation \( h(A) \) in the form of a Laplace transform, that is, \( h(A)=\int_{0}^{\infty} g(t) e^{-At}dt \).
- **Linear Hamiltonian Simulation combination**: Utilize the Linear Hamiltonian Simulation combination (LCHS) method to represent the matrix exponential function \( e^{-At} \) as a linear combination of continuously parameterized Hamiltonian simulation problems.
- **Quantum Linear Combination Unit (LCU)**: By discretizing the integral and using the LCU technique, the linear combination of multiple Hamiltonian simulation problems is efficiently implemented.
### Application examples in the paper
1. **Non - homogeneous differential equations**: Consider the non - homogeneous differential equation of the form \( \frac{du(t)}{dt}=-Au(t)+|\psi\rangle \), the Lap - LCHS method can effectively solve this type of equation.
2. **The power of the matrix inverse**: For \( h(z) = (\eta + z)^{-p} \), the Lap - LCHS method can efficiently implement the power of the matrix inverse \( (A+\eta I)^{-p} \).
3. **Linear differential equations with non - diagonal mass matrices**: Consider the linear differential equation of the form \( A\frac{du(t)}{dt}=-u(t) \), the Lap - LCHS method can solve it without explicitly inverting the matrix \( A \).
4. **Second - order differential equations**: For the second - order differential equation of the form \( \frac{d^{2}u}{dt^{2}} = Au \), the Lap - LCHS method can simulate the decay branch of the solution.
### Summary
This paper solves the problem of efficiently implementing a wide range of eigenvalue transformations on quantum computers by proposing the Laplace - based Linear Hamiltonian Simulation combination (Lap - LCHS) method. This method is not only of great theoretical significance but also shows significant advantages in practical applications.