From NWChem
Viewed 923 times, With a total of 2 Posts

Clicked A Few Times
Threads 5
Posts 10


4:04:37 AM PST  Tue, Feb 9th 2016 

Dear all,
I'm doing geometry optimization of 46 atoms molecule in two steps  at first I do it with small basis set (321G) and then I proceed to larger one (631+G*). The problem I'm facing is that for the larger basis set I'm getting "Load of old vectors failed. Forcing atomic density guess" error for each GO step. My input is following:
start dimerBO4MM1c2_GOB3LYP631+Gstar
MEMORY stack 5200 heap 1000 global 1600 mb
title "betaO4dimer in ccl4, initial structure from oBabel weighted conf search (MM1)  c2, geometry optimization B3LYP 631+G*"
geometry units an print xyz noautosym noautoz
H 2.74728191 1.87018077 0.49356313
C 1.72555291 1.97047424 0.15863115
C 1.12899352 3.24218872 0.17964039
O 1.96933124 4.25797191 0.60162743
C 0.19656223 3.38790055 0.24929258
O 0.79192717 4.65958655 0.30669082
C 0.91324611 2.24393212 0.67249208
O 2.22077430 2.48421901 1.08008852
C 0.31411507 0.98971986 0.67583103
H 0.86583681 0.10748981 0.96093248
C 1.01908701 0.85750889 0.26635916
C 1.69478932 0.50210090 0.26732380
C 1.22178953 1.37964278 0.89640709
H 1.34188160 0.84978156 1.84355039
O 0.22475252 1.64120707 0.71069941
H 1.79480222 2.30647875 0.89961233
C 0.65466895 2.96150475 0.56918867
C 0.10474974 3.85199495 0.38003637
O 0.94030871 3.35659764 1.15210260
C 0.62972838 5.13932065 0.49507947
H 0.21664636 5.82929823 1.21832745
C 1.69899429 5.54906150 0.30861234
H 2.09428614 6.55189615 0.19986531
C 2.24828452 4.66929881 1.23895458
H 3.07503245 4.97682120 1.86749103
C 1.71336063 3.38480684 1.36765346
H 2.11074093 2.67964948 2.08652543
C 3.00477286 1.34005795 1.55237925
C 1.95620029 4.84465172 0.59664725
C 1.44330427 5.55509670 1.05471651
H 0.98066598 6.10689918 0.23912070
H 0.71601445 5.42796803 1.86326679
H 2.32230566 6.08044581 1.43156313
H 2.73451143 4.12344464 0.35070904
H 1.65328033 4.73886515 1.64432851
H 2.30053072 5.86292824 0.40972777
H 3.13695738 0.59222366 0.76298002
H 3.97084098 1.75979957 1.83304676
H 2.53541718 0.86941768 2.42299522
C 1.56208176 4.25650905 2.12795948
H 1.99236445 5.13724910 1.63921061
H 0.84135268 4.57615503 2.88859268
H 2.35177257 3.66495837 2.59038307
O 3.13735166 0.39463213 0.05091155
H 3.51914450 0.06909010 0.84439266
H 1.45671359 1.04924943 1.18524217
end
#basis small
# * library 321G
#end
basis
* library 631+G*
end
#set "ao basis" large
dft
xc b3lyp
CONVERGENCE density 1e8
CONVERGENCE energy 1e8
GRID lebedev 90 14 ssf euler
MAXITER 60
semidirect filesize 5000000 memsize 200000000
end
driver
EPREC 1e8
GRMS 0.000004
MAXITER 900
XYZ geomSteps
end
task dft optimize
and here is the corresponding part of the output:
WARNING : Found 9 linear dependencies
S eigenvalue threshold: 1.00000E05
Smallest S eigenvalue : 3.82890E07
Largest S eigenvalue : 9.55154E06
!! The overlap matrix has 9 vectors deemed linearly dependent with
eigenvalues:
3.83D07 1.03D06 2.05D06 2.71D06 3.60D06 4.79D06 6.20D06 8.02D06
9.55D06
!! nbf/nmo/basisname mismatch
nbf= 500 nbf_file= 500
nmo= 491 nmo_file= 490
basis="ao basis" basis_file="ao basis"
Either an incorrect movecs file was
specified, or linear dependence has changed,
or the basis name was changed.
Loading old vectors from job with title :
betaO4dimer in ccl4, initial structure from oBabel weighted conf search (MM1)  c2, geometry optimization B3LYP 631+G*
Load of old vectors failed. Forcing atomic density guess
I also tried restarting the calculation from the previous optimization in smaller basis set projecting to larger basis with the same result.
Do you have anybody idea why is this happening?
thank you
stanislav.




Gets Around
Threads 1
Posts 165


11:27:40 AM PST  Tue, Feb 9th 2016 

Most likely the number of linearly dependent vectors is changing from one step to the next as the geometry changes. Essentially this means that the size of your basis set changes from one step to the next, so the code disregards the old vectors and starts the SCF for the current step from scratch. You can just leave it alone and it will continue on just fine, albeit a little slower since it is starting from scratch rather than using the previous step as a guess for the current step. Alternatively you can add
set lindep:n_dep n
to your input, where n is an integer that forces the code to always consider n vectors as linearly dependent. With this setting, the code will always be able to use the previous set of vectors as a guess for the current step since you have fixed the size of the basis. You just need to be cautious of numerical stability problems if you set n too small and include vectors that are (nearly) linearly dependent in your SCF calculation.




Clicked A Few Times
Threads 5
Posts 10


3:09:55 AM PST  Thu, Feb 11th 2016 

Thank you. Also, other calculations (other conformers) are doing better.



AWC's:
2.5.10 MediaWiki  Stand Alone Forum Extension
Forum theme style by: AWC