Geometry optimization by using LIBXC caused no SCF iteration with error NaN

[Geoffrey Wood]

Hi Marco,
 
I haven't had much success with this. I have compiled everything from 
scratch including gcc (v 4.8.3) lapack (v 3.1.4) atlas (v 3.10.1) scalapack 
(installer v 1.0.2)  libint (v 1.1.5) libxc (both v 2.1.0 and v 2.2.0). 
 
CP2K <CP at K> (SVN 14141) compiles and links but I get a warning at the end 
saying: 
 


/usr/bin/ld: Warning: size of symbol `build_deriv1_eri' changed from 10368 
in 
/home/woodg07/work/software/cp2k_trunk/cp2k/lib/Linux-x86-64-gfortran/popt/libcp2k.a(hfx_libint_wrapper.o) 
to 2048 in 
/home/woodg07/work/software//libintBuild/lib/libderiv.a(init_libderiv.o)
/usr/bin/ld: Warning: size of symbol `build_eri' changed from 19208 in 
/home/woodg07/work/software/cp2k_trunk/cp2k/lib/Linux-x86-64-gfortran/popt/libcp2k.a(hfx_libint_wrapper.o) 
to 5000 in 
/home/woodg07/work/software//libintBuild/lib/libint.a(init_libint.o)
/usr/bin/ld: Warning: size of symbol `build_deriv1_eri' changed from 10368 
in 
/home/woodg07/work/software/cp2k_trunk/cp2k/lib/Linux-x86-64-gfortran/popt/libcp2k.a(hfx_libint_wrapper.o) 
to 2048 in 
/home/woodg07/work/software//libintBuild/lib/libderiv.a(init_libderiv.o)
/usr/bin/ld: Warning: size of symbol `build_eri' changed from 19208 in 
/home/woodg07/work/software/cp2k_trunk/cp2k/lib/Linux-x86-64-gfortran/popt/libcp2k.a(hfx_libint_wrapper.o) 
to 5000 in 
/home/woodg07/work/software//libintBuild/lib/libint.a(init_libint.o)

 


 

If I run the tests for libxc (i.e. B3LYP and PBE0) these fail at the 
initial SCF iteration giving the error:

 

Program received signal SIGSEGV: Segmentation fault - invalid memory 
reference.

Backtrace for this error:

Program received signal SIGSEGV: Segmentation fault - invalid memory 
reference.

Backtrace for this error:

Program received signal SIGSEGV: Segmentation fault - invalid memory 
reference.

Backtrace for this error:

Program received signal SIGSEGV: Segmentation fault - invalid memory 
reference.

Backtrace for this error:
#0  0x2BD271D in _gfortran_backtrace at backtrace.c:258
#1  0x2BAF3E0 in _gfortrani_backtrace_handler at compile_options.c:129
#2  0x3C7BA3291F
#3  0x0
#0  0x2BD271D in _gfortran_backtrace at backtrace.c:258
#1  0x2BAF3E0 in _gfortrani_backtrace_handler at compile_options.c:129
#2  0x3C7BA3291F
#3  0x0

However if I just run a standard BLYP calculation then there are no 
problems and the job completes.

On Monday, July 21, 2014 7:13:48 PM UTC-4, Geoffrey Wood wrote:

> Thanks Marco,
>
> I'll compile, test and give you my feedback.
>
> Geoff.
>
>
>
> On Monday, July 21, 2014 11:32:58 AM UTC-4, Marco wrote:
>>
>> Hi Geoffrey,
>>
>> The libxc v2.1 and v2.2 were added to the SVN version of cp2k (
>> https://groups.google.com/forum/#!searchin/cp2k/Marco/cp2k/kyPFdrTV4uc/-skBy_pEMYUJ). 
>> The xc_libxc.F was updated. Using the latest libxc versions may help. In my 
>> previous post regarding meta-GGA functionals, I may have made an incorrect 
>> statement. It may not be possible to perform cp2k calculations with 
>> meta-GGAs which require 2nd derivatives but single-point and geometry 
>> optimization calculations should run. Can you compile CP2K-2.5.1 with libxc 
>> version 2.1.0 and test your calculations again. The XC_MGGA_X_M06 and 
>> XC_MGGA_C_M06 components should be combined.
>>
>> Best regards,
>> Marco
>>
>> On Monday, July 21, 2014 9:16:39 AM UTC-4, Geoffrey Wood wrote:
>>>
>>>  
>>> I was wondering if this problem is resolvable or will be addressed in 
>>> version 2.6?  I can run Hybrid-GGA functionals (eg B3LYP and PBE0) using 
>>> the libxc library but hybrid-meta-GGAs all fail at the SCF (i.e not a 
>>> geometry optimization just a single-point). I've compiled CP2k version 
>>> 2.5.1 svn:13632 and libxc version 2.01.  I've tried to run M06 and BR89 
>>> both of with stop at the SCF initiation. However the TPSS functional, which 
>>> doesn't have an HF section as given in the CP2K tests does run (i.e  
>>> H2O-tpssx_libxc.inp). As a side question the libxc library contains both 
>>> exchange and correlation parts of these functionals, i.e. XC_MGGA_X_M06 and 
>>> XC_MGGA_C_M06 are these suppossd to be combined somehow?
>>>
>>> Thanks
>>>  
>>>  
>>>
>>> On Sunday, April 13, 2014 2:33:23 PM UTC-4, Jingyun Ye wrote:
>>>
>>>> Hi, all
>>>>
>>>> I am trying to use M06-2X functional in CP2K (PBE potential)  to 
>>>> compare the binding energy of CO2  with Gaussian results( also use M06-2X 
>>>> functional, same basis sets). 
>>>>
>>>> But the calculation stops  at the SCF every time with the following 
>>>> information. But if I change the functional M06-2X to PBE, the job runs 
>>>> well. 
>>>>
>>>> Anyone know what's the problem? Any suggestion will be help. 
>>>> Thanks very much.
>>>>
>>>>  
>>>>  SCF WAVEFUNCTION OPTIMIZATION
>>>>
>>>>   ----------------------------------- OT 
>>>> ---------------------------------------
>>>>
>>>>   Allowing for rotations:  F
>>>>   Optimizing orbital energies:  F
>>>>   Minimizer      : CG                  : conjugate gradient
>>>>   Preconditioner : FULL_ALL            : diagonalization, state 
>>>> selective
>>>>   Precond_solver : DEFAULT
>>>>   Line search    : 2PNT                : 2 energies, one gradient
>>>>   stepsize       :    0.15000000
>>>>   energy_gap     :    0.00100000
>>>>
>>>>   eps_taylor     :   0.10000E-15
>>>>   max_taylor     :             4
>>>>
>>>>   mixed_precision    : F
>>>>
>>>>   ----------------------------------- OT 
>>>> ---------------------------------------
>>>>
>>>>   Step     Update method      Time    Convergence         Total energy 
>>>>    Change
>>>>   
>>>> ------------------------------------------------------------------------------
>>>>
>>>>   Trace(PS):                                  209.9999999999
>>>>   Electronic density on regular grids:                   NaN           
>>>>       NaN
>>>>   Core density on regular grids:              210.0000000000       
>>>>  0.0000000000
>>>>   Total charge density on r-space grids:                 NaN
>>>>   Total charge density g-space grids:                    NaN
>>>>
>>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20140724/a198670d/attachment.htm>