Geometry Optimization fails without proper messages
Rad
rad.... at arl.army.mil
Mon Oct 15 12:51:06 UTC 2007
Thanks Teo.
Here is the output after enabling the trace and using smaller unit
cell:
========================================================================
DISTRIBUTION OF THE PARTICLES (COLUMNS)
Process col Number of particles Number of matrix columns
0 174 -1
Sum 174 -1
*******************************************************************************
*** STARTING GEOMETRY
OPTIMIZATION ***
*** CONJUGATE
GRADIENTS ***
*******************************************************************************
*** STARTING CG Geo
Main ***
*** STRATING
cp_eval_at ***
*** STARTING
subsys_get ***
*** STARTING Force/Energy
Calc ***
*** ^AöAbout to calc
energy ***
*** about to actually calc
force ***
DISTRIBUTION OF THE NEIGHBOR LISTS
Process Number of particle pairs Number of matrix elements
0 58025 7040113
Sum 58025 7040113
DISTRIBUTION OF THE OVERLAP MATRIX
Process Number of matrix blocks Number of matrix elements
0 15225 1774689
Sum 15225 1774689
of 15225 (100.0 % occupation)
Number of electrons: 720
Number of occupied orbitals: 360
Number of orbital functions: 1878
*****************************************************************************
*** 08:40:49 ERRORL2 in cp_fm_cholesky:cp_fm_cholesky_decompose
processor ***
*** 0 err=-300 condition FAILED at line
116 ***
*****************************************************************************
===== Routine Calling Stack =====
9 cp_fm_cholesky_decompose
8 init_scf_run
7 qs_energies
6 qs_forces
5 cp_eval_at
4 cp_cg_main
3 geoopt_cg
2 qs_geo_opt
1 CP2K
CP2K| Stopped by process
number 0
CP2K| Abnormal program termination
Fortran abort routine called
IOT Trap
============================================================================================
I will getting access to super compueters sometime this week. After
that only I can try this system on bigger machines.
Here is the input I am using, could you please take a look at it and
let me know if something stands out:
=============================================================================================
&MOTION
&GEO_OPT
OPTIMIZER CG
MAX_ITER 200
USE_FIT_LINE_SEARCH T
&PRINT
&PROGRAM_RUN_INFO
&END PROGRAM_RUN_INFO
&END PRINT
&END GEO_OPT
&END MOTION
&FORCE_EVAL
METHOD QS
&DFT
&QS
&LOCALIZE
METHOD JACOBI
&END LOCALIZE
&END QS
&SCF
SCF_GUESS ATOMIC
&END SCF
&XC
DENSITY_CUTOFF 1.0000000000000000E-10
GRADIENT_CUTOFF 1.0000000000000000E-10
DENSITY_SMOOTH_CUTOFF_RANGE 0.0000000000000000E+00
TAU_CUTOFF 1.0000000000000000E-10
FUNCTIONAL_ROUTINE NEW
&XC_GRID
XC_SMOOTH_RHO NONE
XC_DERIV PW
USE_FINER_GRID F
&END XC_GRID
&XC_FUNCTIONAL BLYP
&END XC_FUNCTIONAL
&XC_POTENTIAL
ENERGY NONE
&END XC_POTENTIAL
&END XC
&POISSON
POISSON_SOLVER PERIODIC
PERIODIC XYZ
&EWALD
EWALD_TYPE EWALD
ALPHA .44
GMAX 64
NS_MAX 11
O_SPLINE 6
EPSILON 9.9999999999999995E-07
RS_GRID REPLICATED
&PRINT
&PROGRAM_RUN_INFO LOW
EACH 1
ADD_LAST NUMERIC
COMMON_ITERATION_LEVELS 0
FILENAME __STD_OUT__
&END PROGRAM_RUN_INFO
&END PRINT
&END EWALD
&END POISSON
&END DFT
&SUBSYS
&CELL
ABC 9.3027 9.3027 6.6403
UNITS ANGSTROM
PERIODIC XYZ
&END CELL
&COORD
C 0.499964 0.499964 0
C 0.622509 0.561774 0.128007
===================================================================================
Rad
On Oct 11, 3:58 pm, Teodoro Laino <teodor... at gmail.com> wrote:
> Ciao Rad,
>
> you can get more diagnostic enabling the keyword TRACE in &GLOBAL..
> you will get a *lot* of information.. everytime entering and exiting
> from a routine.. memory allocated (if enabled for that platform) and
> cpu time..
>
> Now few questions:
> 1) Did you try to run the same job on another platform?
> I suspect that the problem could be the architecture/compiler.. If
> it's not running on a simple
> linux box than could you please send me the input and all necessary
> files? there could always be
> a bug around.. but if it runs smoothly on a LINUX or CRAY or IBM box
> I would more suspect that
> it's a miscompilation problem..
>
> ----------
>
> Regarding the linear scaling geometry optimizer there's the LBFGS and
> the CG..
>
> LBFGS is somehow pathological in the line search algorithm..
> CG is more robust..
>
> You can have an idea of how to use them looking at the regtest..
>
> for LBFGS:
> find cp2k/test -name "*.inp" -exec grep -Hi "LBFGS" {} \;
>
> for CG:
> find cp2k/test -name "*.inp" -exec grep -Hi "CG" {} \;
>
> Anyway for any doubt just ask..
> cheers
> teo
>
> On 11 Oct 2007, at 21:44, Rad wrote:
>
>
>
>
>
> > Hi,
>
> > I am trying to optimize the geometry of a crystal with about 900 atoms
> > in the unit cell. I tried to run the calculation on an SGI ALTRIX m/c
> > with CP2K compiled for serial execution. The calculation crashed with
> > a core dump.
>
> > How do I get any diagnostic message in a CP2K run? I did add PRINT
> > command in all the sections of the input (Global, DFT, Motion etc) and
> > here is the output (I added few additional write statements to the
> > code and compiled):
> > ====================================================================
> > DISTRIBUTION OF THE PARTICLES (ROWS)
>
> > Process row Number of particles Number of matrix rows
>
> > 0 870 -1
>
> > Sum 870 -1
>
> > DISTRIBUTION OF THE PARTICLES (COLUMNS)
>
> > Process col Number of particles Number of matrix columns
>
> > 0 870 -1
>
> > Sum 870 -1
>
> > **********************************************************************
> > *********
> > *** STARTING GEOMETRY
> > OPTIMIZATION ***
> > *** CONJUGATE
> > GRADIENTS ***
>
> > **********************************************************************
> > *********
> > *** STARTING CG Geo
> > Main ***
> > *** STRATING
> > cp_eval_at ***
> > *** STARTING
> > subsys_get ***
> > *** STARTING Force/Energy
> > Calc ***
> > *** ^AöAbout to calc
> > energy ***
> > *** about to actually calc
> > force ***
> > ================================================
> > The execution fails when qs_forces subroutine is called from
> > force_env_calc_energy_force.
>
> > Is there a way I can use the linear scaling algorithm implemented in
> > CP2K for this optimization? If so can I have a sample input to set up
> > such a run?
>
> > Thanks
>
> > Radhakrishnan Balu, PhD
>
> > US Army Research Lab
>
> > Aberdeen , MD, USA- Hide quoted text -
>
> - Show quoted text -
More information about the CP2K-user
mailing list