How can I estimate the uncertainty on a calculated energy value?

Filipe Teixeira filipe.t... at gmail.com
Thu May 11 11:59:30 CEST 2017


Hello

In the early 2000's Irikura introduced the concept of "virtual 
meassurement" (J. Phys. Chem. A, 2005, 109, 8430). In this article one of 
the sources of error he explored is the error associated with the SCF of 
optimied geometries (which may depend, for example on the initial geometry 
being optimized and the propagation of numerical errors in the SCF 
procedures). It is an interesting reading if you are interested on that 
matter, but (spoiller allert) those errors (even in 2005 when 32-bit 
machines were common) are minor compared to the errors of the methodologies 
themselves. Nowadays, with 64-bit machines being the norm, I doubt machine 
round-up will be a big source of error for most applications... 

I hope this helps.

Best regards,
Filipe Teixeira

On Wednesday, May 10, 2017 at 12:42:45 PM UTC+1, Cristiana Passiu wrote:
>
> Dear Vladimir,
> thank you for your answer. It will be helpful for knowing how to express 
> the results of the calculations.
>
> Best regards,
> Cristiana
>
>
>
>
> Il giorno mercoledì 10 maggio 2017 10:07:57 UTC+2, Vladimir Rybkin ha 
> scritto:
>>
>> Dear Cristiana,
>>
>> the question has no correct answer. To give a few points:
>> 1) the energy here is in Hartree (that's the only thing for sure here)
>> 2) the energy itself is meaningless, what is reported is the energy 
>> difference. In kJ/mol people report them with one digit, which, however 
>> does not mean that the uncertainty is 0.05. It's just a convention.
>> 3) try to find benchmarks on your systems and use the recommended methods 
>> with estimated accuracy.
>>
>> Yours,
>>
>> Vladimir
>>
>> понедельник, 24 апреля 2017 г., 10:55:29 UTC+2 пользователь Cristiana 
>> Passiu написал:
>>>
>>> Dear all,
>>> assuming that a model is correct and that the functional chosen is 
>>> suitable for the system, is there a standard uncertainty that can be 
>>> attributed to the final energy values? In other words, how many 
>>> significative digits can I report in the results?
>>>
>>> What is the default unit of measurement for the energy in the output? 
>>> For example, if I have this in the output:
>>>
>>>   Total energy:                                             
>>>  -158.42425955059213
>>>  ENERGY| Total FORCE_EVAL ( QS ) energy (a.u.):             
>>> -158.424259550592126
>>>
>>> are the values in Rydberg or in Hartree? In order to convert them to eV, 
>>> do I have to multiply them by 27.2113838668?
>>>
>>> Many thanks,
>>> Cristiana
>>>
>>>
>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20170511/8bd86acb/attachment.html>


More information about the CP2K-user mailing list