[QE-developers] subroutine "xc"

[Lorenzo Paulatto]

Hello,
this is the email I was talking before at the developer meeting, please 
take no offense if it seems a bit rude. It was meant like an incitement 
to be bold and change others people code if it looks better.

--

Compiling the d3q code I discovered today that the xc subroutine from 
the module "funct" does not exist anymore, so I went and checked the 
changes trying to make it work again.

I'm quite surprised because the new code adds quite a lot of 
complication for not an obvious reason. I.e. this code 
(LR_Modules/setup_dmuxc.f90):

   dmuxc(:) = 0.d0
   DO i = 1,dfftp%nnr
      rhotot = rho%of_r(i,1) + rho_core(i)
      IF ( rhotot> 1.d-30 ) dmuxc(i)= dmxc( rhotot)
      IF ( rhotot<-1.d-30 ) dmuxc(i)=-dmxc(-rhotot)
   ENDDO

becomes

   CALL init_lda_xc()
   !
   rhotot(:) = rho%of_r(:,1) + rho_core(:)
   !
   sign_r = 1.0_DP
   DO i = 1, dfftp%nnr
      IF ( rhotot(i) < -1.d-30 ) THEN
         sign_r(i) = -1.0_DP
         rhotot(i) = -rhotot(i)
      ELSEIF ( rhotot(i)<1.d-30 .AND. rhotot(i)>-1.d-30 ) THEN
         sign_r(i) = 0.0_DP
         rhotot(i) = 0.5_DP
      ENDIF
   ENDDO
   !
   CALL dmxc_lda( dfftp%nnr, rhotot, dmuxc )
   !
   dmuxc = dmuxc * sign_r

The new code requires:

1. to create, allocate and deallocate two new arrays

2. to call an initialization routine

I can take some more complexity, but only if it provides a real 
performance improvement (I don't recall doing local density functional 
being expensive in a long while) or it can solve a conceptual problem.

Here there is quite an obvious conceptual problem: how to treat the 
negative charge, and the fact that it could be done differently in 
different parts of the code. But the new implementation does not solve 
it, on the contrary it makes it more complicated !

And, if we have to call an initialization subroutine EVERY time we call 
dmxc, why is it not called INSIDE dmxc?

cheers


-- 

Lorenzo Paulatto - Paris