Potential in NAH models not working properly

[Nhertl]

The potential energy in the Anderson Hamiltonian does not provide identical results with previous NQCD versions (tested against v.0.9.2). The NAH was constructed from the Erpenbeck-Thoss model. The parameters between the two versions were compared and found to be identical. Attached are the scripts that were used to conduct the simulations.

run_example_trajectory_NQCDv0.9.2.txt
run_example_trajectory_NQCDv.1.0.txt

[Alexsp32]

I've started looking at this a bit more. What's interesting is that in your code, you are able to initialize a Simulation{Classical} using a QuantumModel (ErpenbeckThoss). This gives undefined behaviour and @Ash-Baldwin and I didn't immediately see how this doesn't raise any errors.

I've recreated your scripts and tried out three cases:

Erpenbeck-Thoss parameters:

BCME

Just as the default example, with sufficiently high accuracy criteria, I can get smooth looking dynamics, with reference to #489.

Classical Dynamics with Quantum Model

Here, the behaviour for the potential is undefined. We need to figure out what numbers are even used in propagating dynamics, but they're almost definitely not the right ones.

Classical Dynamics with clamped state Quantum Model

This is the intended way of running classical dynamics with QuantumModels. By wrapping AdiabaticStateSelector(model, state), a QuantumModel works like a ClassicalModel with a fixed electronic state.

Here, I've run dynamics with your ErpenbeckThoss model parameters and a fixed electronic state of 2 after fixing a small issue in NQCD/NQCModels.jl#74.