[Wien] Failed SCF calculation in the 1st iteration using 12.1 version of the WIEN2K

[Gavin Abo]

The attached two patches were tested with a TiC calculation, and the 
results were the same, but it was not tested thoroughly on other cases.  
The previously reported error for a MBJ calculation also went away.  If 
you want to use the fftpack library instead of FFTW2 or 3, you can try 
to use them.  Just place them in $WIENROOT/SRC_lapw0.  Enter in a terminal:

patch -b fft_modules.F fft_modules.patch
patch -b vresp.F vresp.patch

The -b should create fft_modules.F.orig, a backup of the fft_modules.F 
before the patch.  Then, recompile.  The patch is for the files in 
WIEN2k_12.tar currently on the website.

It is noted that TiC.output0 gave for the fftpack and fftw 2.1.5:

   v-mean, v0,u0 (4.569481054390656E-002,-4.067822355071745E-018)
  (5.894205821050297E-003,-4.067822355071745E-018)
  (69.7518919071863,0.000000000000000E+000)
  (-3.980060472285626E-002,0.000000000000000E+000)

fftw 3.3.2:

v-mean, v0,u0 (4.569481047410750E-002,-1.113100404819184E-034)
  (5.894205805621243E-003,-1.113100404819184E-034)
  (69.7518919071863,0.000000000000000E+000)
  (-3.980060466848626E-002,0.000000000000000E+000)

The mkl fftw2 and mkl fftw3 both had the same result, but the imaginary 
part was also different than the above cases.  I don't know why they are 
different or if they could significantly impact any of the calculations.

On 10/2/2012 1:27 AM, Peter Blaha wrote:
> So you find   NANs for the WARPING lines, which hints where the 
> problems come from.
>
> I suspect that case.vns will NOT contain readable numbers for the 
> fourier coefficients
> (search for "PW" in this file and check).
>
> In essence, lapw0 seems to have failed, and probably this is due to 
> problems with the fft-library.
>
> There were postings on the mailing list about the problem with the 
> fftwpack library (i.e. in
> case you are not using FFTW2 or FFTW3) and also a fix.
>
> Am 02.10.2012 09:11, schrieb yedu kondalu:
>> Dear Prof. Blaha,
>>
>>     I have checked case.scf1 and case.output1 and observed that no 
>> eigen values printed in case.scf1 after LAPW1 END and WARPING = NaN 
>> instead of a finite value in case.output1,
>> which leads to fail the job without performing LAPW2 step in the 
>> first iteration.  How to rectify this problem ?
>>
>> I didn't face this problem very few cases but most of the cases 
>> (about 15 compounds) I come across this problem.
>>
>> In the previous mail the matrix size is same in two versions of the 
>> WIEN2k, its a typo error when I copied from the terminal (sorry for 
>> the typo error).
>>
>> Thanking you sir,
>>
>>
>>     Regards
>> Yedukondalu
>>
>>
>>
>>
>> _______________________________________________
>> Wien mailing list
>> Wien at zeus.theochem.tuwien.ac.at
>> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>>
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20121002/774a236f/attachment.htm>
-------------- next part --------------
392,394c392,394
<         real(kind=8) :: DWORK(:)
<         complex(kind=8) :: CWORK(:)
<         complex(kind=8) :: C(LDC1,LDC2,N3,2)
---
>         real(kind=8) DWORK(*)
>         complex(kind=8) CWORK(*)
>         complex(kind=8) C(LDC1,LDC2,N3,2)
-------------- next part --------------
20a21,24
> #if defined (fftpack)
>       DOUBLE PRECISION   DWORK(*)
>       COMPLEX*16         CWORK(*)
> #else
22a27
> #endif