[gpaw-users] Generating a core-hole setup with generator2

[Eric Hermes]

On 3/23/2015 8:38 PM, Marcin Dulak wrote:
> On 03/23/2015 11:44 PM, Eric Hermes wrote:
>>
>> On 3/23/2015 5:21 PM, Marcin Dulak wrote:
>>> On 03/23/2015 10:37 PM, Eric Hermes wrote:
>>> > Marcin,
>>> >
>>> > Is there an optimized dataset for ruthenium? I wish to benchmark my
>>> > methodology against a literature result, and the paper to which I am
>>> > comparing is studying oxygen-covered ruthenium. If there isn't an
>>> > optimized dataset, do you believe it is safe to use the old generator
>>> > for the time being?
>>> the Ru from the gpaw-setups version 0.9 (the semicore one) is fine for
>>> the ground state, see for example:
>>> http://listserv.fysik.dtu.dk/pipermail/gpaw-users/2014-November/003097.html 
>>>
>>> If you have any curious results please share with us.
>>>
>>> Best regards,
>>>
>>> Marcin
>> I wish to ensure that I'm doing the same thing for both of my 
>> systems. If a core-hole setup for ruthenium generated by the old 
>> generator produces results that are consistent with the prior work to 
>> which I am comparing (in which all-electron FP-LAPW calculations are 
>> used), that does not necessarily indicate to me that I can achieve 
>> the same accuracy with generator2 and palladium. To that end, I would 
>> like to use either the old generator for everything, or generator2 
>> for everything. Which if these two approaches would you find the most 
>> trustworthy?
>>
> JJ mentioned that generator2 may not be usable for core-hole setups.
>
What do you mean by this? The only reason I am generating ground-state 
setups is so that I can guarantee that the ground-state and core-hole 
setups are as similar as possible. If the core-hole setups that 
generator2 produces are not trustworthy, then I will need to use the old 
generator.
> Comparing the 0.9 semi-core datasets and the generator2 ones:
> it seems that we are getting better datasets from generator2 so it 
> would be better to use it.
> In case you want to try a Ru and O from generator2, in addition to Pd, 
> here they are:
> python -c "from gpaw.atom.generator2 import main as g; g(['Ru', '-f', 
> 'PBE', '-s', '-P', '4s,5s,4p,5p,4d,0.2d,F', '-r', '2.36,2.2,2.31', 
> '-0', '5,2.09', '-pl', 'spdfg,-1.5:1.5:0.01'])"
> python -c "from gpaw.atom.generator2 import main as g; g(['O', '-f', 
> 'PBE', '-s', '-P', '2s,0.9s,2p,0.4p,0.5d,1.0d', '-r', 
> '1.24,1.3,1.3,1.3,1.3', '-0', '5,0.72', '-pl', 'spdfg,-1.5:1.5:0.01'])"
>
> I'm reporting this a bit "under a table" but we are not ready to make 
> the final parameters and the method of obtaining them public.
> I would use datasets 0.9 if they work for you because if you publish 
> anything you can refer to them,
> and the generator2 can change any day making your results 
> non-reproducible.
I appreciate that you are sharing this as-of-yet unpublished 
information. Though, given the WIP nature of generator2 datasets, I may 
wish to use the old generator simply to ensure that my results are 
publishable. Also given this fact, it would seem that it might be 
premature to consider the old generator and gpaw-setup obsolete. I was 
under the false impression that generator2 was the production-ready 
replacement for gpaw-setup, based on previous posts to this mailing 
list. Please correct me if I am misunderstanding something.

Thanks,
Eric
>
> On the other hand if we receive now any input about the generator2 
> datasets not performing well enough
> we can still try to fix them.
>
> Best regards,
>
> Marcin
>
>> Eric Hermes
>>
>>
>
>

-- 
Eric Hermes
J.R. Schmidt Group
Chemistry Department
University of Wisconsin - Madison