Determining QMC simulability with geometric phases

Itay Hen

Main idea

• Using a diagonal + off diagonal representation (permutation matrix representation, PMR) of the Hamiltonian $$H = D_0 + \sum_i D_i P_i$$ (for diagonal $$D_i$$ and permutation matrices $$P_i$$) to represent the partition function, this work studies the emergence of the sign problem, as a geometric property naturally arising from this decomposition.

Background

• Each term in the partition function expansion must be positive in order for there to be no sign problem (to interpret as legitimate probabilities).
• A stoquastic Hamiltonian (one with non-positive off-diagonals) is sign-problem-free (SPF), however the converse is not necessarily true. Often SPF and stoquastic are used synonomously, and this can therefore cause confusion (non-stoquastic does not imply hard to simulate!).
• The PMR defines permutations between basis states (similar to the time-slices in path integral Monte Carlo), which can be thought of as cycles on a graph (where nodes are the basis states).
• The positivity of each term relates to a “phase” acquired going around each path (each path is a product of complex numbers, the total phase is the sum of individual phases).

Main results

• In order for a Hamiltonian to be SPF, the phase acquired around any cycle must be a multiple of $$2\pi$$
• Using this result, an example of a non-stoquastic yet SPF Hamiltonian is constructed (i.e. it is easy to simulate with QMC).

tags: stoquastic  QMC