Hi Francesco,
First, let me refer to a point raised in the initial mail: DFTB is
*not* supposed to have the same memory requirements as DFT. DFTB is a
semi-empirical method, and will have memory requirements closer to
other semi-empirical methods (AM1, PM3, etc.)
Now, on to your problem... It is not uncommon for DFTB to have
convergence problems. We have some work in accelerating the DFTB
convergence, but that involves some major changes and will only be
available for Amber10. Meanwhile, if the problem persists, I'd
recommend you to try PDDG/PM3 instead.
The reason why DFTB-PME will not be added to Amber9 is similar. It
involves big changes and adds new functionality, really beloging in a
new version.
> Scripts used:
>
> Classical minimization
> &cntrl
> imin=1, maxcyc=500, ncyc=200,
> cut=25.0, ntb=0, ntc=2, ntf=2,
> ivcap=0, fcap=10
> /
I noticed that you are using shake (NTC=NTF=2) in the minimization.
Don't. That may be leading to some weird structures that can get
unnoticed by the classical MD run, but give bigger problems for the QM
run.
>
> 300K constant temp classical MD
> &cntrl
> imin=0, ntb=0,
> cut=25.0, ntc=2, ntf=2,
> tempi=300.0, temp0=300.0,
> ntt=3, gamma_ln=1.0,
> nstlim=10000, dt=0.001,
> ntpr=100, ntwx=100
> /
Should I assume that this is the order in which you are running your
calculations? If so, then it's actually luck that this is working with
MM. After an initial MM minimization, you should then slowly bring the
system to the final temperature (tempi=0, temp0=300). After that, you
should equilibrate the pressure before you can do any QM calculations.
(Look in the manual for temperature and pressure regulation.)
>
>
> Initial min for qmmm with dftb
> &cntrl
> imin=1, maxcyc=1000, ncyc=4000,
> cut=25.0, ntb=0, ntc=2, ntf=2,
> ivcap=0, fcap=10,
> ifqnt=1
> /
> &qmmm
> qmmask=':1',
> qmcharge=0,
> qmtheory=7,
> qmshake=0,
> qm_ewald=0
> /
Again, here you are using shake for a minimization run. Don't.
>
> 300K constant temp for qmmd with dftb
> &cntrl
> imin=0, ntb=0
> cut=25.0, ntc=2, ntf=2,
> tempi=300.0, temp0=300.0,
> ntt=3, gamma_ln=1.0,
> nstlim=20000, dt=0.002,
> ivcap=0, fcap=10,
> ntpr=100, ntwx=100,ifqnt=1
> /
> &qmmm
> qmmask=':1',
> qmcharge=0,
> qmtheory=7,
> qmshake=0,
> qm_ewald=0
> /
>
Same remarks as before for the classical one. Remember, if you wish to
run QM/MM at 300K, you should bring the system to 300K using MM first,
and only then turn on QM. If this is the case, you can skip the QM
minimization.
> Reminded here that the initial pdb was for a structure minimized by MMFF94
> global conformational space search, which agreed with the minimum structure
> from amber simulated annealing in vacuum. This was followed by DFT minimization
> with the most recent functionals in NWChem (were no crash ever occurred).
What exactly is you goal here? Real minimization or dynamics? I
noticed you get errors during the dynamics, not the minimization. That
could be a consequence of the bad minimizations (because of the shake
keyword) or that the system is just taking a really bad configuration
at some point during the dynamics. I'm not surprised that the DFT
minimizations worked, as they were probably done in the gas phase.
Assuming that your goal is the dynamics, try the sequence I suggested
above. If the problems continue, try writing the strucuture at every
step during the dynamics, then looking into the trajectory using a
graphical program such as VMD, to see if you can notice anything
wrong.
Gustavo.
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Wed Aug 29 2007 - 06:07:26 PDT
