Hi Adam,<= o:p>

I agree that your observations are not what y= ou would expect to get. I have noticed looking at the code that the presenc= e of fractional occupations interferes with a number of things (the fractio= nal occupations may be negligibly small). The things it interacts with incl= ude: Level shifters and DIIS. So there is potential that this may mess up t= he optimization of solution you are after. If you still happen to have exam= ples of calculations that demonstrate these issues I am happy to have a loo= k and see if I can figure out what is going on.

Best wishes,

= ;

= ; Huub

&n= bsp;

From: Adam Pelzer [mailto:adampelzer2010===u.no= rthwestern.edu]
Sent: Wednesday, April 06, 2011 4:46 PM
To= : Govind, Niri
Cc: Van Dam, Hubertus J; CCL Subscribers; nwch= em-users===emsl.pnl.gov
Subject: Re: [NWCHEM] RE: smearing with NWC= hem

Hello all. I asked a simi= lar question a few months back. We found that:

a) some calcul= ations performed with smearing did not converge to the same state as calcul= ations with no smearing even when the HOMO/LUMO gap was much much larger th= an the energy of the extra term. That is, there were no fractional occupati= ons and so smearing "should" have had no effect.

b) the s= olution that we converged to with smearing was different from the one arriv= ed at without smearing. With smearing we arrived at a higher energy solutio= n with a different distribution of charge and spin density than we arrived = at with no smearing. This was not a numerical error, it was an entirely dif= ferent solution that was 0.25 eV higher. This was for single point calculat= ions. When we allowed the system to relax, we again arrived at two differen= t states with slightly different geometries and charge distributions with t= he smeared case 0.03 eV above the no smear solution.

c) the use of = unconstrained smearing (without the spinset T keyword) allowed the calculat= ion to change spin state, but it did not always change it to a lower energy= spin state. I had been told that it would only change spins if it could fi= nd one of lower energy, but this turned out not to be the case.

The= idea of just using it for everything so that any issues wash out when you = take the difference works fine if you're still guaranteed to get the ground= state solution, but if you're not then it's more problematic.

Any = thoughts on these issues?

Cheers,
= Adam

On Wed, Apr 6, 2011 at 6:25 PM, Govind, Ni= ri <niri.govind===pnl.gov> w= rote:

Hi Frank,

Try the following= if you want to preserve the multiplicity with smear.

dft
m= ult 3
...
...
end
set dft:spinset t
task dft
=
Best regards,
-Niri<= /p>

-----Original Message-----
From: owner-nwchem-users===emsl.pnl.gov [mailto:owner-nwchem-users===emsl.pnl.gov] On B= ehalf Of Van Dam, Hubertus J
Sent: Wednesday, April 06, 2011 4:08 PM
= To: CCL Subscribers
Cc: nwc= hem-users===emsl.pnl.gov
Subject: [NWCHEM] RE: smearing with NWChem
Hi Frank,

This is an interesting question. There are a few spec= ific points here:
1. The energy that is associated with the smearing is = normally supposed to be negative, so the total energies with smearing shoul= d be lower than the ones without smearing.
2. The smearing functionality= may change the spin state of your system. As you are working on transition= metal complexes I am guessing you might have open shell systems. I can pro= vide some special directives to control this if needed.
3. The code will= actually print the energy that it adds on as a result of the smearing. Sea= rch for "eFermi=3D" in the output. (see also 4.)
4. However, i= f you subtract the energy from 3. you will not get the energy you are looki= ng for. The problem is that the DFT energy for a system with fractionally o= ccupied orbitals is higher than the energy of the same system with integer = occupations. So the eFermi term that is added on corrects for that differen= ce and adds a bit more on top of that.
5. Concluding it seems to me that= your best bet is to run all calculations with the same smearing parameter = for the final energies and compare the resulting energies.
I am sorry if= this is not the answer you were looking for. If you want any more specific= information please feel free to contact me.

Best wishes,

&nb= sp; Huub
________________________________________
From:= owner-chemistry+hubertus.vandam=3D=3Dpnl.gov===ccl.net [owner-chemistry+hubertus.vandam=3D=3D pnl.gov===ccl.ne= t] On Behalf Of Xiaofeng Frank Duan duanx:_:asc.hpc.mil [owner-chemistry===ccl.net]
Sent: Wednesday, April 06, 2011 1:12= PM
To: Van Dam, Hubertus J
Subject: CCL: smearing with NWChem
Sent to CCL by: "Xiaofeng Frank Duan" [duanx^asc.hpc.mil]
Hello,

I optimized= some transitional metal compounds with DFT in NWChem. For some compounds, = I had to use "smear" to get SCF converged even though I decreased= "smear" values gradually to the minimum. Since 'smear' adds an a= dditional energy term, the total energy is higher than that of the "ze= ro smear" for the same structure.
My question is: How to correct or= scale the "smeared" total energy to the "zero smeared"= one so that I can compare energies among different compounds.

Thank= s,
Frank

-=3D This is automatically added to each message= by the mailing script =3D-
To recover the email address of the author o= f the message, please change
the strange characters on the top line to t= he === sign. You can also
look up the X-Original-From: line in the mail he= ader.

E-mail to subscribers: CH= EMISTRY===ccl.net or use:
http://www.ccl.net/cg= i-bin/ccl/send_ccl_message

E-mail to administrators: CHEMISTRY-REQUEST===ccl.net or use
&= nbsp; http://www.ccl.net/cgi-bin/ccl/send_ccl_message
Subscribe/Unsubscribe:
http://www.ccl.net/chem= istry/sub_unsub.shtml

Before posting, check wait time at: http://www.ccl.net

Jo= b: http://www.ccl.net= /jobs
Conferences: http://server.ccl.net/chemistry/ann= ouncements/conferences/

Search Messages: http://www.ccl.ne= t/chemistry/searchccl/index.shtml

If your mail bounces from CCL = with 5.7.1 error, check:
http://www.ccl.net/spammers.txt
<= br>RTFI: http://www.ccl.net/chemistry/aboutccl/instructions/
=