Refactor: modify EXX realx/md calculation framework

[1041176461]

The new EXX relax/md framework does not need to repeat the PBE calculation to build the density matrix for subsequent ion steps except step 0. Instead, it directly reads the density matrix of the previous step to perform EXX self-consistent calculation, reducing unnecessary self-consistent calculations.

Here are some comparsion

1. For relax, exx_separate_loop 1, left -> new, right -> old
   

2. For relax, exx_separate_loop 0, left -> new, right -> old
   

3. For md, exx_separate_loop 1, left -> new, right -> old
   

4. For md, exx_separate_loop 0, left -> new, right -> old
   

[1041176461]

For EXX relax/md Integration Test, exx_separate_loop 1 and exx_hybrid_step 2 settings are used. However, the first two EXX outer steps do not fully converge, and the results of the new and old frameworks are different. According to my test above, although the first few outer steps are different, they finally converged to consistent results. Do you have any suggestions on integration testing? @maki49

[maki49]

@1041176461 You can change the reference results of the related cases.
I find nothing wrong with your code, so I'm confused about why the energy difference is larger than before in first exx-loop (compared to the following exx-loop) in the separate-loop case ( as figure 1 and 3)...
"The density matrix difference resulting from the structure difference contributes more than the difference between pbe and hybrid functional"? hmm... seems make sense

[1041176461]

@maki49 Thanks! I changed "result.ref" of tests "285_NO_KP_RE_HSE", "286_NO_KP_CR_HSE", "385_NO_GO_RE_S1_HSE" and "386_NO_GO_MD_S1_HSE". I guess density matrix extrapolation may be a good way to reduce the difference, and corresponding implementations are planned.

[WHUweiqingzhou]

I notice you change the ref. I think these changes should be reviewed by more relevant developers. Could you make a review @PeizeLin ?

[PeizeLin]

I notice you change the ref. I think these changes should be reviewed by more relevant developers. Could you make a review @PeizeLin ?

Since all the Integration Tests are unconverged, the older refs are meaningless when the loop is changed.