9
1
mirror of https://github.com/QuantumPackage/qp2.git synced 2024-12-21 11:03:29 +01:00

Fixed division by zero in RSDFT

This commit is contained in:
Anthony Scemama 2020-05-20 11:45:41 +02:00
parent 764aed423e
commit 512525508b
3 changed files with 34 additions and 25 deletions

View File

@ -138,6 +138,8 @@ subroutine ex_lda_sr(mu,rho_a,rho_b,ex,vx_a,vx_b)
!Density and kF !Density and kF
rho_a_2=rho_a*2.D0 rho_a_2=rho_a*2.D0
akf = ckf*(rho_a_2**f13) akf = ckf*(rho_a_2**f13)
! Avoid division by zero
if (akf == 0.d0) akf = 1.d-20
a = mu/(z2*akf) a = mu/(z2*akf)
a2 = a*a a2 = a*a
a3 = a2*a a3 = a2*a
@ -169,6 +171,7 @@ subroutine ex_lda_sr(mu,rho_a,rho_b,ex,vx_a,vx_b)
!Density and kF !Density and kF
rho_b_2= rho_b * 2.d0 rho_b_2= rho_b * 2.d0
akf = ckf*(rho_b_2**f13) akf = ckf*(rho_b_2**f13)
if (akf == 0.d0) akf = 1.d-20
a = mu/(z2*akf) a = mu/(z2*akf)
a2 = a*a a2 = a*a
a3 = a2*a a3 = a2*a

View File

@ -92,42 +92,47 @@
end end
double precision function erf0(x) double precision function erf0(x)
implicit double precision (a-h,o-z) implicit none
if(x.lt.0.d0)then double precision, intent(in) :: x
erf0=-gammp(0.5d0,x**2) double precision, external :: gammp
if(x < 0.d0)then
erf0=-gammp(0.5d0,x*x)
else else
erf0=gammp(0.5d0,x**2) erf0=gammp(0.5d0,x*x)
endif endif
end end
double precision function gammp(a,x) double precision function gammp(a,x)
implicit double precision (a-h,o-z) implicit none
if(x.lt.0..or.a.le.0.)stop 'error in gammp' double precision, intent(in) :: a, x
if(x.lt.a+1.)then double precision :: gln, gammcf
if(x<0.d0.or.a<=0.d0) then
stop 'error in gammp'
endif
if(x < a+1.d0)then
call gser(gammp,a,x,gln) call gser(gammp,a,x,gln)
else else
call gcf(gammcf,a,x,gln) call gcf(gammcf,a,x,gln)
gammp=1.-gammcf gammp=1.d0-gammcf
endif endif
return
end end
subroutine gser(gamser,a,x,gln) subroutine gser(gamser,a,x,gln)
implicit double precision (a-h,o-z) implicit double precision (a-h,o-z)
parameter (itmax=100,eps=3.e-7) parameter (itmax=100,eps=3.d-7)
gln=gammln(a) gln=gammln(a)
if(x.le.0.)then if(x.le.0.d0)then
if(x.lt.0.) stop 'error in gser' if(x.lt.0.d0) stop 'error in gser'
gamser=0. gamser=0.d0
return return
endif endif
ap=a ap=a
sum=1./a sum=1.d0/a
del=sum del=sum
do 11 n=1,itmax do 11 n=1,itmax
ap=ap+1. ap=ap+1.d0
del=del*x/ap del=del*x/ap
sum=sum+del sum=sum+del
if(abs(del).lt.abs(sum)*eps)go to 1 if(abs(del).lt.abs(sum)*eps)go to 1
@ -139,14 +144,14 @@
subroutine gcf(gammcf,a,x,gln) subroutine gcf(gammcf,a,x,gln)
implicit double precision (a-h,o-z) implicit double precision (a-h,o-z)
parameter (itmax=100,eps=3.e-7) parameter (itmax=100,eps=3.d-7)
gln=gammln(a) gln=gammln(a)
gold=0. gold=0.d0
a0=1. a0=1.d0
a1=x a1=x
b0=0. b0=0.d0
b1=1. b1=1.d0
fac=1. fac=1.d0
do 11 n=1,itmax do 11 n=1,itmax
an=float(n) an=float(n)
ana=an-a ana=an-a
@ -155,8 +160,8 @@
anf=an*fac anf=an*fac
a1=x*a0+anf*a1 a1=x*a0+anf*a1
b1=x*b0+anf*b1 b1=x*b0+anf*b1
if(a1.ne.0.)then if(a1.ne.0.d0)then
fac=1./a1 fac=1.d0/a1
g=b1*fac g=b1*fac
if(abs((g-gold)/g).lt.eps)go to 1 if(abs((g-gold)/g).lt.eps)go to 1
gold=g gold=g

View File

@ -24,8 +24,9 @@ double precision function primitive_value_explicit(power_prim,center_prim,alpha,
end end
double precision function give_pol_in_r(r,pol,center, alpha,iorder, max_dim) double precision function give_pol_in_r(r,pol,center, alpha,iorder, max_dim)
double precision :: r(3), center(3), alpha,pol(0:max_dim,3) implicit none
integer, intent(in) :: iorder(3), max_dim integer, intent(in) :: iorder(3), max_dim
double precision :: r(3), center(3), alpha,pol(0:max_dim,3)
integer :: i,m integer :: i,m
double precision :: gauss(3), x double precision :: gauss(3), x
gauss = 0.d0 gauss = 0.d0
@ -33,7 +34,7 @@ double precision function give_pol_in_r(r,pol,center, alpha,iorder, max_dim)
do m = 1, 3 do m = 1, 3
x = r(m) - center(m) x = r(m) - center(m)
do i = 0, iorder(m) do i = 0, iorder(m)
gauss(m) += pol(i,m) * dexp(-alpha *x**2 ) * x**i gauss(m) += pol(i,m) * dexp(-alpha *x*x ) * x**i
enddo enddo
enddo enddo
give_pol_in_r = gauss(1) * gauss(2) * gauss(3) give_pol_in_r = gauss(1) * gauss(2) * gauss(3)