SCDM method in Wannier90 #167

VVitale · 2018-02-14T15:54:51Z

Major commit for the SCDM method. Files modified are: parameters.F90 kmesh.F90 and pw2wannier90.F90, the latter being the one with the actual code for the projections.
I have also added the necessary info in the documentation (user guide) on how to use the new projection method.

… the SCDM method Ref.[Damle et al. ArXiv 2017 eprint arXiv:1703.06958]

…e user guide

giovannipizzi

As an additional comment - no need to recompile the docs, as the repository becomes very big - we can recompile it when we will do a release.

giovannipizzi · 2018-02-15T08:21:32Z

doc/user_guide/parameters.tex

@@ -611,7 +614,8 @@ \subsection{Shells}
 \section{Projection}

 The projections block defines a set of localised functions used to
- generate an initial guess for the unitary transformations. 
+ generate an initial guess for the unitary transformations. The projection block can be specified
+in conjunction with {\tt scdm\_proj=true} (see belove). This is only used to read the centers of the projections, which in some cases could help the optimisation if {\tt guiding\_centres=true} is added to the input file.


belove->below

giovannipizzi · 2018-02-15T08:24:02Z

doc/user_guide/parameters.tex

@@ -736,6 +740,31 @@ \section{Job Control}

 The default value of this parameter is $\verb#false#$.

+\subsection[scdm\_proj]{\tt logical :: scdm\_proj}
+If {\tt scdm\_proj=true} then the $A_{mn}^{(\mathbf{k})}$ matrices are generated with the SCDM-k method of Ref.~\cite{LinLin-ArXiv2017}. In this case, one also needs to specify the other three {\tt scdm} keywords. One then needs to run {\tt wannier90.x -pp seedname} to generate the {\tt seedname.nnkp} file, to be used by a first-principle code (at the moment only interface available is with the QuantumEspresso code). It is not possible to set {\tt scdm\_proj=true} and simultaneously specify a projection block.


Are mu and sigma really always needed to be specified, also in the isolated case?

No they don't you are right. I have modified the text according to the logic.

giovannipizzi · 2018-02-15T08:27:03Z

src/parameters.F90

+    if(found .and. scdm_proj .and. spinors) &
+        call io_error('Error: SCDM method is not compatible with spinors yet.')
+    if(found .and. scdm_proj .and. guiding_centres) &
+        call io_error('Error: guiding_centre is not compatible with the SCDM method yet.')


guiding_centres

giovannipizzi · 2018-02-15T08:28:11Z

src/parameters.F90

+       elseif(ctmp=='erfc') then
+         scdm_entanglement = 1
+       elseif(ctmp=='gaussian') then
+         scdm_entanglement = 1


should be 2!

Oops! This one is more serious.

giovannipizzi · 2018-02-15T08:29:49Z

doc/user_guide/parameters.tex

+\subsection[scdm\_sigma]{\tt real(kind=dp) :: scdm\_sigma}
+The value of the $\sigma$ parameter in the formulas for the occupation numbers matrix. It defines the spread of the occupation numbers matrix around $\mu$, and as such it cannot be negative. It must be given in units of eV. 
+
+The default value is {\tt 0.0 eV} 


Not sure if setting the default to zero is a good idea, will give rise to division by zero. Better to give e.g. a default of 2eV, or 4eV? Or leave to zero, but then in parameters.F90 check if it is zero and complain that it has to be specified or > 0.

Now sigma is 1.0 eV by default. Then in parameters.F90 it checks if sigma is specified and it returns an error if it is not positive.

giovannipizzi · 2018-02-15T08:31:37Z

doc/user_guide/parameters.tex

+\subsection[scdm\_mu]{\tt real(kind=dp) :: scdm\_mu}
+The value of the $\mu$ parameter in the formulas above. It defines the characteristic energy for the occupation numbers matrix, in units of eV. If {\tt scdm\_entanglement=1}, it gives the mean value of the complementary error function. If {\tt scdm\_entanglement=2}, it gives the mean value of the gaussian instead. 
+
+The default value is {\tt fermi\_energy eV}


However, from the logic in parameters.F90, if the fermi_energy is not specified, the default remains at zero eV... maybe better to simplify the logic, and leave this value by default at 0 eV? If the user specifies by hand the fermi_energy flag, he can also set it twice here. Also, in our experience mu=fermi is not always the optimal choice.

Now mu is 0.0 eV by default.

giovannipizzi · 2018-02-15T08:32:05Z

doc/user_guide/parameters.tex

+
+\subsection[scdm\_entanglement]{\tt integer :: scdm\_entanglement}
+Select the functional form for the occupation number matrix $f(\epsilon_{n\mathbf{k}})$ for the SCDM-k method.
+Only three integer values are allowed:


This is not currently true, in the input the user specifies three strings - internally they are converted to an integer

giovannipizzi · 2018-02-15T08:35:45Z

pwscf/v6.2.1/pw2wannier90.f90

+           READ (iun_nnkp,*) scdm_mu, scdm_sigma
+        ENDIF
+     ELSE
+        WRITE(stdout,'(//," ****** begin WARNING ****** ",/)') 


Do we really want to issue a Warning? If the user does not want to use SCDM, then he does not need to upgrade to a recent version. If he wants to, then he will have to have already upgraded.
This warning might "scare" users but I don't think it is needed (most probably they will read this and don't know what SCDM is).

…er90 into SCDM_WFs

hjunlee · 2018-02-16T11:50:43Z

Dear developers:

Although I tried to do the similar things in both USPP in QE and PAW in VASP, I am still wondering about the extension of SCDM to USPP and PAW where the norm of PW parts only is not equal to 1.

In the pw2wannier90, you represent only (soft) PW parts of wave functions on the real grid and you enforce normalization of psic_all by dividing norm_psi. As a consequence, since in general norm_psi is different between different band indices and thus, psic_all with different band indices are scaled by the different factors.

My question is whether or not piv is affected by these things.

I know well that even in NCPP case, psi is not meaningful near the nucleus of the atoms and I also guess that in PAW and USPP cases, we don't need to construct psi exactly considering augmentation parts. But, I am still wondering whether the enforcement of normalization in USPP and PAW cases works in all (or most) cases.

Sincerely,

Hyungjun Lee
EPFL SB Institute of Physics

giovannipizzi · 2018-02-20T18:31:47Z

Hi, my guess is that the current implementation is (very often) ok (at least for the cases that we have tested). BTW, I think this is the same the authors of the original method did in their Matlab implementation. We're doing a more thorough validation study with more systems in the meantime.

@VVitale There's a typo in the code that prevents compilation I think: https://travis-ci.org/wannier-developers/wannier90/jobs/343964349#L600 could you fix it?

…r happy

…SCDM_WFs

codecov · 2018-02-20T19:15:05Z

Codecov Report

Merging #167 into develop will decrease coverage by <.01%.
The diff coverage is 58.33%.

@@             Coverage Diff             @@
##           develop     #167      +/-   ##
===========================================
- Coverage    57.42%   57.41%   -0.01%     
===========================================
  Files           27       27              
  Lines        15538    15573      +35     
===========================================
+ Hits          8922     8942      +20     
- Misses        6616     6631      +15

Impacted Files	Coverage Δ
src/kmesh.F90	`60.19% <50%> (-0.11%)`	⬇️
src/parameters.F90	`77.55% <60.71%> (-0.16%)`	⬇️

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 6507313...ad15a23. Read the comment docs.

hjunlee · 2018-02-21T11:14:43Z

It seems to me that there is a very minor bug in pw2wannier90:
In a serial case, the phase array has a wrong dimension:
phase(iw,iw)

Sincerely,

Hyungjun Lee
EPFL SB IoP

SCDM method in Wannier90

Valerio Vitale (PDRA added 3 commits February 14, 2018 14:53

Changes to the source code and documentation for the SCDM method

bafccb5

Major changes to pw2wannier90 in order to generate the Amn matrix via…

e9d26ca

… the SCDM method Ref.[Damle et al. ArXiv 2017 eprint arXiv:1703.06958]

Changes to wannier.bib to include citations for the SCDM method in th…

7608a88

…e user guide

giovannipizzi requested changes Feb 15, 2018

View reviewed changes

giovannipizzi self-assigned this Feb 15, 2018

Valerio Vitale (PDRA added 3 commits February 15, 2018 10:36

Bugfix to scdm logic and documentation

9ec1a36

Merge branch 'develop' of https://github.com/wannier-developers/wanni…

06b31ec

…er90 into SCDM_WFs

Minor changes to SCDM documentation and logic

0c620fc

giovannipizzi mentioned this pull request Feb 15, 2018

Removed deallocate spn from compute_orb, causing crash at compilation… #169

Closed

Merge branch 'develop' into SCDM_WFs

eed061d

giovannipizzi approved these changes Feb 20, 2018

View reviewed changes

Valerio Vitale (PDRA added 2 commits February 20, 2018 18:59

Bugfix in parameters.f90. Line has been truncated to keep the compile…

1e40800

…r happy

Merge branch 'SCDM_WFs' of https://github.com/VVitale/wannier90 into …

ad15a23

…SCDM_WFs

giovannipizzi merged commit 85f143f into wannier-developers:develop Feb 21, 2018

manxkim pushed a commit to manxkim/wannier90 that referenced this pull request Jan 10, 2021

Merge pull request wannier-developers#167 from VVitale/SCDM_WFs

d2c6eb8

SCDM method in Wannier90

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

SCDM method in Wannier90 #167

SCDM method in Wannier90 #167

VVitale commented Feb 14, 2018

giovannipizzi left a comment

giovannipizzi Feb 15, 2018

VVitale Feb 15, 2018

giovannipizzi Feb 15, 2018

VVitale Feb 15, 2018

giovannipizzi Feb 15, 2018

VVitale Feb 15, 2018

giovannipizzi Feb 15, 2018

VVitale Feb 15, 2018

giovannipizzi Feb 15, 2018

VVitale Feb 15, 2018

giovannipizzi Feb 15, 2018

VVitale Feb 15, 2018

giovannipizzi Feb 15, 2018

giovannipizzi Feb 15, 2018

hjunlee commented Feb 16, 2018

giovannipizzi commented Feb 20, 2018

codecov bot commented Feb 20, 2018 •

edited

Loading

hjunlee commented Feb 21, 2018

SCDM method in Wannier90 #167

SCDM method in Wannier90 #167

Conversation

VVitale commented Feb 14, 2018

giovannipizzi left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

hjunlee commented Feb 16, 2018

giovannipizzi commented Feb 20, 2018

codecov bot commented Feb 20, 2018 • edited Loading

Codecov Report

hjunlee commented Feb 21, 2018

codecov bot commented Feb 20, 2018 •

edited

Loading