Estimating the properties of quantum thermal states is a fundamental task in quantum many-body physics and plays a central role in understanding the thermodynamic behavior of molecules and materials. While recent quantum algorithms have enabled efficient preparation of thermal states, predicting thermal properties, i.e., thermal expectation values of observables, remains computationally costly, as standard Gibbs sampling requires waiting for a full mixing time to generate each independent sample. In this work, we show that this sampling cost can be significantly reduced: by using a single trajectory, effectively independent samples can be generated in a time shorter than the mixing time, enabling a more efficient method for observable estimation. We illustrate the challenges and key strategies through the estimation of average energy and then generalize the method to arbitrary observables that are measurable in a thermodynamically reversible way (i.e., satisfy detailed balance). We also introduce a weighted operator Fourier transform technique to mitigate disturbances caused by measurements of general observables.
Copyright. All Rights Reserved.