📝 SummarXiv

Abstract

Simulating time evolution under fermionic Hamiltonians is a compelling application of quantum computers because it lies at the core of predicting the properties of materials and molecules. Fermions can be simulated on qubit-based quantum computers using a fermion-to-qubit mapping, subject to an overhead -- the circuit depth on a qubit quantum computer divided by that on a quantum computer built from native fermionic modes -- at worst scaling linearly with the number of modes $N$. Existing approaches that lower this depth overhead usually trade it for space, using $O(N)$ ancilla qubits. We exponentially reduce the worst-case overhead of ancilla-free fermion-to-qubit mappings to $O(\log^2 N)$ by constructing circuits that perform any fermionic permutation on qubits in the Jordan-Wigner encoding in depth $O(\log^2 N)$. We also show that our result generalizes to permutations in any product-preserving ternary tree fermionic encoding. When introducing $O(N)$ ancillas and mid-circuit measurement and feedforward, the overhead reduces to $O(\log N)$. Finally, we show that our scheme can be used to implement the fermionic fast Fourier transform, a key subroutine in chemistry simulation, with overhead $\Theta(\log N)$ without ancillas and $\Theta(1)$ with ancillas, improving exponentially over the best previously known ancilla-free algorithm with overhead scaling linearly with $N$. Our results show that simulating fermions with qubit quantum computers comes at a much lower asymptotic overhead than previously thought.