Gibbs States and Finite-Temperature Simulation
Preparing thermal states of a Hamiltonian on a quantum computer to compute finite-temperature properties.
Why finite temperature
Ground-state simulation captures zero-temperature physics, but many properties, phase diagrams, free energies, transport coefficients, magnetic susceptibility, occur at finite temperature. These require the Gibbs state rho = e^(-beta H) / Z, where beta is inverse temperature and Z = tr(e^(-beta H)) is the partition function. Preparing and measuring Gibbs states is the finite-temperature analog of ground-state preparation.
The challenge
The Gibbs state is a mixed state, not a pure state, and e^(-beta H) is not unitary. Preparing it therefore requires either purification (a larger pure state whose reduced density matrix is thermal) or non-unitary techniques, the same machinery used for open systems and matrix functions.
Preparation approaches
- Thermofield double: prepare a pure state on doubled registers whose partial trace is the Gibbs state.
- Quantum Metropolis and quantum thermalization: quantum analogs of Markov-chain sampling that converge to the thermal state.
- Imaginary-time evolution: apply e^(-beta H) via QSVT block-encoding or a variational surrogate.
- Dissipative preparation: engineer a Lindbladian whose steady state is the Gibbs state.
Imaginary-time evolution
Replacing real time it with imaginary time beta turns the propagator into e^(-beta H), which damps high-energy components and, applied to a suitable initial state, projects toward low-energy or thermal states. Because e^(-beta H) is non-unitary, it is implemented as a matrix function via QSVT or approximated variationally (variational imaginary-time evolution), each with its own overhead.
Computing thermal observables
Once a Gibbs state is prepared (or sampled), thermal expectation values follow from measurement. Free energies and partition functions are harder, often requiring integration over temperature or specialized estimators. Finite-temperature quantum simulation is less mature than ground-state simulation but is essential for connecting to experiment, since real materials and plasmas exist at finite temperature. It draws on the same block-encoding and QSVT toolkit, extended from unitary evolution to the non-unitary thermal operator, and it remains an active area where quantum and classical (quantum Monte Carlo) methods compete on different regimes.