What is the difference between variational and diffusion Monte Carlo?

Variational Monte Carlo evaluates and optimizes the energy of a fixed-form trial wavefunction, so its accuracy is limited by that form. Diffusion Monte Carlo goes further by projecting onto the true ground state through imaginary-time evolution, giving lower, often near-exact energies for systems without a sign problem.

What is the fermion sign problem?

For fermions, the wavefunction changes sign under particle exchange, so the quantities sampled can be positive or negative and tend to cancel, making the statistical error grow exponentially with system size. It is the central obstacle to exact quantum Monte Carlo for many fermionic systems.

Quantum Monte Carlo

Quantum Monte Carlo brings stochastic sampling to the many-body Schrodinger equation, computing ground-state energies and correlations of interacting quantum systems with an accuracy that scales far better than brute-force diagonalization.

Definition

Quantum Monte Carlo is a family of stochastic methods that evaluate expectation values and project ground states of quantum many-body systems by interpreting the squared wavefunction or imaginary-time propagator as a probability distribution to be sampled.

Scope

This topic covers the main quantum Monte Carlo flavors: variational Monte Carlo, which optimizes a trial wavefunction by sampling its probability density, and projector methods such as diffusion Monte Carlo, which filter out the ground state by imaginary-time evolution. It also addresses the fermion sign problem that limits these methods.

Core questions

How does variational Monte Carlo evaluate the energy of a trial wavefunction by sampling?
How does diffusion Monte Carlo project out the ground state through imaginary-time evolution?
Why does the fermion sign problem make many quantum systems hard to simulate?
How does the fixed-node approximation control the sign problem at the cost of a bias?

Key theories

Variational Monte Carlo: A parametrized trial wavefunction is sampled by Metropolis according to its squared amplitude, and the variational energy and its parameter gradients are estimated as Monte Carlo averages and minimized.
Diffusion and projector Monte Carlo: Treating imaginary-time evolution as a diffusion-plus-branching process projects an initial trial state onto the ground state, giving in principle exact ground-state energies for bosonic and sign-problem-free systems.
Fixed-node approximation: To control the fermion sign problem, the nodes of a trial wavefunction are fixed and the ground state is found within that nodal structure, yielding a variational upper bound whose quality depends on the trial nodes.

Clinical relevance

Quantum Monte Carlo provides benchmark ground-state energies for the electron gas, molecules and solids, informs and tests density-functional approximations, and treats strongly correlated systems where mean-field methods fail.

History

The 1980 Ceperley-Alder Monte Carlo calculation of the electron gas ground state provided the correlation energy that underpins modern density-functional theory; subsequent decades developed diffusion, fixed-node and continuum quantum Monte Carlo into high-accuracy tools for electronic structure.

Debates

Severity of the fermion sign problem: Whether the sign problem can be solved efficiently in general is unresolved and is believed to be computationally hard, so practical fermionic quantum Monte Carlo relies on approximations such as fixed nodes that trade exactness for tractability.

Key figures

David Ceperley
Berni Alder
Matthew Foulkes

Seminal works

ceperleyalder1980
foulkes2001

Frequently asked questions

What is the difference between variational and diffusion Monte Carlo?: Variational Monte Carlo evaluates and optimizes the energy of a fixed-form trial wavefunction, so its accuracy is limited by that form. Diffusion Monte Carlo goes further by projecting onto the true ground state through imaginary-time evolution, giving lower, often near-exact energies for systems without a sign problem.
What is the fermion sign problem?: For fermions, the wavefunction changes sign under particle exchange, so the quantities sampled can be positive or negative and tend to cancel, making the statistical error grow exponentially with system size. It is the central obstacle to exact quantum Monte Carlo for many fermionic systems.