[CPMD-list] BLYP instead of B3LYP

[Reinout Declerck]

This question passed almostly exactly one year ago. See below the reply of 
Axel Kohlmeyer.

BLYP is said to reproduce hydrogen bonds rather well, in my view you could 
therefore use BLYP just as well.

regards,

Reinout Declerck


----- Original Message ----- 
From: "Axel Kohlmeyer" <akohlmey at cmm.chem.upenn.edu>
To: "Lan-Feng Yuan" <yuanlf at ustc.edu.cn>
Cc: <cpmd-list at cpmd.org>
Sent: Saturday, January 06, 2007 11:18 PM
Subject: Re: [CPMD-list] B3LYP with CPMD?


> Lan-Feng Yuan wrote:
>>   Dear CPMD people,
>>
>>   I heard that hybrid functionals like B3LYP are terribly slow in CPMD. 
>> While the
>> paper "molecular dynamics simulation of liquid water: hybrid density 
>> functionals"
>> (JPCB, 2006, 110, 3685) adopted a bunch of methods to make B3LYP water 
>> simulation
>> possible, which is very interesting. As they show, hybrid is now only 40 
>> times
>> slower than pure DFT. Are these methods incorporated in current version 
>> 3.11? What
>
> yes. hartree-fock exchange with distance screening is available in 3.11.
> please see the version history in the manual:
>
> - Distance screening for HFX calculation and hybrid funcional for PBC
>
>
>> do you think about the perspective of using hybrid based on plane wave?
>
> this is very much a matter of taste and of how much CPU time you are
> willing to spend on this method. the improvement is undeniable, however
> for most practical applications, you have to compare these improvements
> to other sources of errors, e.g. finite size effects, insufficient 
> statistics, the CP-method itself (when running CP-dynamics, that is). 
> personally, i'd
> rather use the hybrid functional results as an estimate of the error for
> the 'pure DFT' results and invest the additional CPU time to get better
> statistics or larger systems or use a better converged (=larger) basis set
> (note that the authors of the paper point out the 70ry cutoff is at the
> lower end of being converged and that results change with larger basis 
> sets)
> or a combination of it.
>
> cheers,
>      axel.