Variational simulation of imaginary time evolution


It would be great to have the protocol described in arXiv:1804.03023 implemented in Penny Lane. It was implemented in QuEST by the authors, but the automatic differentiation of Penny Lane would work for the method.

This protocol gives a different trajectory across the state space than QAOA and numerical results show that it avoids local minima that could trap a gradient-based VQE.


Thanks Peter for letting us know! We’ll look into implementing it in PennyLane.

From a quick read through, it looks like the only codebase change required would be the addition of a new optimizer in the pennylane.optimize sub-module?

I will look into this!

That’s an excellent suggestion!

Here’s a follow-up paper that extends the same idea to approximate a thermal state: arXiv:1812.08767. It is similar to the quantum approximate thermalization protocol of arXiv:1712.05304.