problem converging molecule in doublet state
garold
I am attempting to calculate the energy of a molecule (dicarboyxlate
anion with a lithium cation; net charge is zero) in its doublet
state. I have tried using both restricted and unrestricted formalisms
(LSD and ROKS, respectively). However, I cannot get the SCF to
converge with either. I have been using both B3LYP and BLYP and
various basis sets. Ideally, I would prefer to use B3LYP/GAPW/6-31+G*
but have been unable to obtain convergence even with BLYP/GAPW/6-31G*
(no diffuse functions) or BLYP/GPW/DZVP-MOLOPT-SR-GTH (my original
assumption was that diffuse functions were source of the problem).
The input for my preferred method/basis set is below and the output is
attached. Please let me know if you have any suggestions. Thank you
in advance.
Note that I am using a recent tested version of cp2k compiled with
libint. Both BLYP/ and B3LYP/GAPW/6-31+G* work well for the closed-
shell, singlet version of this system (e.g., with one electron added,
net charge of -1, same geometry). However, a preconditioner (OT/
OUTER_SCF option) does not seem to work well for this system so I am
not using it.
Best regards,
Garold Murdachaew
ps: Input file (gzipped output file is attached):
&FORCE_EVAL
METHOD Quickstep
&DFT
LSD
CHARGE 0
MULTIPLICITY 2
BASIS_SET_FILE_NAME ../EMSL_BASIS_SETS_mine
POTENTIAL_FILE_NAME ../POTENTIAL
&MGRID
CUTOFF 280
NGRIDS 5
&END MGRID
&QS
METHOD GAPW
&END QS
&POISSON
PERIODIC NONE
PSOLVER MT
&END
&SCF
SCF_GUESS ATOMIC
EPS_SCF 1.0E-7
MAX_SCF 20
! &OUTER_SCF
! EPS_SCF 1.0E-7
! MAX_SCF 20
! &END
! &OT
! PRECONDITIONER FULL_ALL
! &END
&END SCF
&XC
&XC_FUNCTIONAL
&LYP
SCALE_C 0.81
&END
&BECKE88
SCALE_X 0.72
&END
&VWN
FUNCTIONAL_TYPE VWN3
SCALE_C 0.19
&END
&XALPHA
SCALE_X 0.08
&END
&END XC_FUNCTIONAL
&HF
&SCREENING
EPS_SCHWARZ 1.0E-10
&END
&MEMORY
MAX_MEMORY 1200
&END
FRACTION 0.20
&END
&END XC
&END DFT
&SUBSYS
&CELL
ABC 20.0 20.0 20.0
PERIODIC NONE
&END CELL
&COORD
C -0.793023 -0.703538 0.527367
C -0.046767 -0.838845 -0.834379
C -0.109334 0.381305 -1.793731
O -0.533605 0.203716 -2.940264
C 0.080655 -0.434112 1.783207
O -0.014088 -1.209796 2.739982
O 0.835715 0.609771 1.736225
O 0.313892 1.501426 -1.316006
H -1.529788 0.107036 0.458189
H -1.351096 -1.620668 0.723529
H 1.012459 -1.057227 -0.647331
H -0.452088 -1.690968 -1.382698
Li 0.909030 1.669426 0.332287
&END COORD
&KIND C
BASIS_SET 6-31+Gx
POTENTIAL ALL
&END KIND
&KIND O
BASIS_SET 6-31+Gx
POTENTIAL ALL
&END KIND
&KIND H
BASIS_SET 6-31Gx
POTENTIAL ALL
&END KIND
&KIND Li
BASIS_SET 6-31+Gx
POTENTIAL ALL
&END KIND
&END SUBSYS
&END FORCE_EVAL
&GLOBAL
PROJECT d_m
PRINT_LEVEL MEDIUM
&END GLOBAL
pps: I constructed the 6-31+G* basis set simply by adding the diffuse
functions to C, O, and Li, the basis set file is attached.
garold
file with whoever wishes to see it.
Garold
Ross, Sun
Firstly, MAX_SCF 20 in your input file. I think 20 steps can not
guarantee any system to converge at such small value.
Secondly, you can add XC_SMOOTH in the XC domain, which may lead to
the convergence faster.
&XC_GRID
XC_SMOOTH_RHO NN10
XC_DERIV SPLINE2_SMOOTH
&END XC_GRID
garold
Thank you for the suggestions. I had tried with larger MAX_SCF but I
see I was not going far enough. Now, after changing also some other
settings (based on sample inputs in ~/cp2k/tests/benchmarks*), I am
getting efficient convergence for B3LYP/6-31+G*. Energies agree well
with molpro. BLYP I cannot get to converge at all with this basis set
(but fortunately I don't need this).
As for use of smoothing, I believe this is not recommended when using
GAPW. (If I do use smoothing, convergence is reached in fewer
iterations but with errors of few tenths of hartree in the energies.)
I am still curious about why I cannot use OT for my system....
Garold
hut...@pci.uzh.ch
if you have problems converging OT you should use safe options.
&OUTER_SCF ON
EPS_SCF ...
&END
&OT ON
Minimizer CG
LINESEARCH 3PNT
PRECONDITIONER FULL_ALL
&END OT
Outer SCF helps to get a good preconditioner. Keep the number of SCF
cycle in the inner loop small (10-20 max).
regards
Juerg
--------------------------------------------------------------
Juerg Hutter Phone : ++41 44 635 4491
Physical Chemistry Institute FAX : ++41 44 635 6838
University of Zurich E-mail: hut...@pci.uzh.ch
Winterthurerstrasse 190
CH-8057 Zurich, Switzerland
---------------------------------------------------------------
-----cp...@googlegroups.com wrote: -----
To: cp2k <cp...@googlegroups.com>
From: garold
Sent by: cp...@googlegroups.com
Date: 04/09/2011 01:33AM
Subject: [CP2K:3193] Re: problem converging molecule in doublet state
Hi Sun,
Garold
--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To post to this group, send email to cp...@googlegroups.com.
To unsubscribe from this group, send email to cp2k+uns...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/cp2k?hl=en.