QMCPACK
Excited state energy of Silicon crystal is wrong when the trial wavefnction generate from PySCF
The system in study is a diamond-structure silicon crystal with a 2×1×1 supercell. The PySCF input is modeled after the test case in qmcpack-4.0.0/tests/solids/diamondC_1x1x1-Gaussian_pp_MSD/twf_input/Carbon1x1x1-tw0.py. However, the multistate calculation included in QuantumPackage was omitted, and only a single-determinant calculation was performed. During the calculation, we observed that the excited-state energy was lower than the ground-state energy. This excited state was constructed by promoting an electron from the 8th energy band to the 9th energy band—an electronic transition that has been verified as valid in previous Quantum ESPRESSO (QE) calculations. The corresponding VMC input files (*.wfj.xml) and output results for the ground and excited states are provided below:
@anbenali , @kgasperich , @Hyeondeok-Shin : Any thoughts on this? I think you are the only people who have done periodic PySCF-QMC calculations to any degree. This should "just work".
@huangxxxxxxxx : Do you know if calculations with splines give the expected result? If so, can put the output printed by QMCPACK in both cases here? (The periodic PySCF route has been only used by a few people while e.g. we have tutorials on the spline route and many more people have published papers. The spline route should be completely robust. That said, the implementation should be general but presumably there is an issue somewhere in the plumbing for PySCF. )
In case it matters for reproducing the problem, which version of PySCF are you using?
I don't recall any relevant changes between v4.0, v4.1, and the development version.
The calculation with splines give a reliable results , and the input and output of VMC for ground and excited state are showed below:
ground state