<div dir="ltr"><div>Hello,</div><div><br></div><div>This is Adam Denchfield, PhD student of Hyowon Park at UIC. My advisor suggested there's a bug in QE's projections for PAW pseudopotentials and that I should reach out to the QE devs. <br></div><div><br></div><div>Your time is valuable, so here is the short version: our main system has qualitative differences between the default and PAWproj=.true. projection schemes:<br></div><div><div><img src="cid:ii_lhr8qb0t4" alt="image.png" style="margin-right: 0px;" width="269" height="255"><br></div></div><div>(The bottom is zoomed in for visibilities' sake)<br></div><div>We are not sure which to trust for physical insight, considering test calculations on
simpler systems like LuH3 seem to suggest the PAWproj=.true. option is
faulty. A comparison to VASP, which also uses PAW projectors by default, seems to see disagreement with the QE default and QE PAWproj=.true. options as well. See below for an extended account & details on pseudopotentials, etc. This is all with QE version 7.1. <br></div><br><div>----------------------------------</div><div><br></div><div>Myself and my advisor have compared PDOS calculations of cubic LuH3 with QE [default], QE [pawproj=.true.], and VASP default [PAW] projection schemes. There's some disagreement here. Would love your input as we're trying to prepare to submit this but can't seem to agree on what's going wrong here.<br></div><div><br></div><div>Below is an image comparing a hydride (LuH3) with the Wentzovich PS for Lu, and the H.pbe-kjpaw_psl.1.0.0.UPF for H. I use PBE here. There shouldn't be any PAW sphere overlap issues since this structure isn't under pressure, and I've used large PW cutoffs and k-meshes to confirm this isn't just a convergence issue. Also using the optimized tetrahedron method here.<br></div><div><img src="cid:ii_lhr8k02v2" alt="image.png" width="514" height="194"><br><br></div><div>It appears PAWproj=.true. massively underestimates the PDOS contributions of the PBE case, from all orbitals. I may have expected it to occur for some orbitals (H-s), but even localized ones like Lu-f are very misrepresented. The error in occupations isn't a simple scale factor either, it's orbital dependent. I would have suspected the Wentzovich pseudopotentials but the problem occurs for the hydrogen too. <br></div><div><br></div><div><div>My advisor ran a VASP-PBE calculation on the same material for
comparison and found the VASP default projection scheme (which as far as
I can tell also uses the PAW projectors) doesn't have this massive
underestimation. <br></div><div><br></div><div><img src="cid:ii_lhr807611" alt="image.png" style="margin-right: 0px;" width="288" height="269"><br></div><div>While
the VASP results for this system are close enough to the QE default
scheme, this isn't true for our actual system of study, which includes
Nitrogen too. This is just a reference system. </div></div><div><br></div><div>The above is a reference system. Here is a snippet of our main system and its qualitative discrepancy between the two methods. <br></div><div><img src="cid:ii_lhr8qb0t4" alt="image.png" style="margin-right: 0px;" width="269" height="255"><br></div><div>As you can see, one method predicts one orbital character dominant at Ef over another.<br></div><div><br></div><div>Back to the reference system; When I ran an HSE06 calculation on LuH3 using the input_dft='hse06' (as well as the ACE), the problem seemed to mostly go away for PAWproj=.true., and both projection schemes basically agree other than a small normalization to be expected from the PAW spheres. My advisor was skeptical of the HSE06 results for other reasons (increasing the f-orbital energies), but that's another story [would love a comment on this from you though if you have the time]. See below.</div><div><img src="cid:ii_lhr8l1nc3" alt="image.png" width="514" height="158"><br></div><div><br></div><div><br></div><div>Essentially, my advisor suspects a bug in the QE projections for PAW projectors, and possibly also in the default projection scheme [since it disagrees with VASP's output]. <br></div><div><br></div><div>I can send over the input files if needed. <br></div><div><br><span class="gmail_signature_prefix">-- </span><br><div dir="ltr" class="gmail_signature" data-smartmail="gmail_signature"><div dir="ltr"><div>Regards,</div><div>Adam Denchfield</div><div>UIC Physics Ph.D Student</div></div></div></div></div>