Excited state geometry optimization in the presence of external charges


Click here for full thread
Just Got Here
Hello,

I'm fairly new to NWChem and I apologize if this has been asked nefore (I searched for similar topic, but didn't find one).

I'm trying to do excited state geometry optimization of a dye embedded in an environment consisting of a lipid bilayer and water molecules. We want the represent the water and lipids (a bit more than 35000 atoms) as static charges (I'm using the Bq block for this). My input looks somewhat like this:

start proba

memory total 11000 MB

geometry
x y z coordinates of 42 atoms
end

charge 0

basis print
 * library 3-21G
end

scf
 vectors input atomic output scf.movecs
maxiter 100
end

dft
 vectors input scf.movecs
xc b3lyp
noio
end

tddft
 nroots 2
algorithm 1
notriplet
target 1
targetsym none
civecs
grad
root 1
end
end

driver
 maxiter 100
trust 0.1
end

set bq:max_nbq 40000

bq
 load /home/hristina/local/nwchem-6.5/script/bq.xyz format 1 2 3 4
end

title "HF/3-21G energy"
task scf energy

title "B3LYP/3-21G* opt"
task tddft optimize


I've played around with various options - I've tried b3lyp, cam-b3lyp and pbe, grid xfine, all possible algorithms (1 to 4), different basis sets (3-21G, 6-31G), different trust (0.3, 0.1), optimization with cdartesian coordinates only (noautoz). Most of the time the calculations start but then the excited state gradients are doing some really strange things:

@ Step Energy Delta E Gmax Grms Xrms Xmax Walltime
@ 1 -1030.02588412 1.2D+00 3.67988 1.00017 0.80467 2.56912 2786.5
@ 2 -1036.88721545 -6.9D+00 9.64027 2.17237 0.37773 1.18885 4403.1
@ 3 -1038.45412595 -1.6D+00 16.60024 2.82696 0.10545 0.41192 5946.9
@ 4 -1103.81946714 -6.5D+01891.12293143.95009 0.37701 1.20258 7494.5
@ 5 -1072.58990887 3.1D+01326.57817 33.08277 0.05390 0.14372 9016.6
...

and they keep going this way for whatever the maxiter is set to in the driver block. I ran the optimization of the dye in vacuum and it took some time (50+ iterations) but it converged at the end. However, the moment I include the Bq charges the optimization goes awry within the first few steps. Do you have any ideas what might be wrong? I can upload input and output files if necessary.