Hi all,

I am wondering if this question has already been asked, if someone has already tried to compare (energies) the same system using different potentials. I have a single wall carbon nanotube (zigzag,8,0) terminated by hydrogen at the ends. First I simulate the system in pcff (class2 forcefield as follows) .

8cnt24pa1_pristine_class2.lmp (305 KB)

8cnt24pa1_pristine_rebo.lmp (49 KB)

Comparing absolute energies makes no sense. You should be comparing the relative energy differences between several configurations for each ff or instead forces on unrelaxed identical geometries.

Carlos

Hi all,

I am wondering if this question has already been asked,

yes it has. and been answered, too.

if someone has already tried to compare (energies) the same system using

different potentials. I have a single wall carbon nanotube (zigzag,8,0)

terminated by hydrogen at the ends. First I simulate the system in pcff

(class2 forcefield as follows) .

[...]

Next, I do the same but with rebo potential as follows

[...]

The minimized energies in the twos cases are 30640.6494271 kcal/mol(

real units) for class2 and -4787.454358 ev (metal units) for rebo.

I know 1 Kcal/mol =0.04336 ev, but that does not make the minimized

energies comparable here, later I also initialize the structure at 300 K

and then perform NPT, NVT and NVE simulation. Since energies become

different (both value and sign!) at the very first step (minimization) I

dont bother comparing the energies during the equilibrium step.

My question is what is wrong here? The trajectory looks reasonable in both

cases. this is a very simple system. One should expect the energies to be

alteast closer than this if not exactly equal, Am I converting the energies

incorrectly from rel to metal units?

why on earth *should* they be equal?

if you build a model of something, lets say a castle, both from legos and

from sand, are they supposed to be the same? they both may look the same

but are very different models and neither is a *real* castle.

axel.

Thank you Dr. Kohlmeyer,

I understand your point. Actually my objective was to determine the suitability of these two forcefields in describing thermal transport through the cnt. As a later step, I subjected both of them to same delta_T and measured the energy flux through the system. So I was trying to compare those energy fluxes. I now understand that comparing absolute magnitudes of energy may not make sense at all, however, I am hoping to find a common basis based ( lets say thermal energy flux or thermal conductivity, or phonon spectrum in a region) on which I could say one ff is better applicable for describing heat transfer through the same structure. Would you please suggest what can be a meaningful comparison in determining that?

Thanks

Souvik.

While waiting for Axel to respond…

You have you make up your mind when it comes to deciding which are the properties of interest for which you aim to test the accuracy of the two FFs. Next, and more importantly, you have to ensure you have reference data at your disposal to compare your calculations to (ideally, experimental data, but first-principles QM calculations have become quite useful as reference too given the level of precision those techniques have achieved over the past few years).

Last, but not least, realize that the accuracy of some of the calculations you are referring to such as thermal conductivity not only depend on the suitability of the FF but on a whole lot of other parameters and effects, and only after making sure you understand the role they play in the calculations, you’ll be able to wisely judge the quality of each FF when doing the job of mimicking thermal transport for example. Check the archives for very many useful tips on this topic, look for Paul Saxe responses, he has been an excellent provider of data and advice, Axel and Steve for sure have some of their own in there as well and if I remember correctly Vikas has dropped quite a few hints about CNTs.

Now I let Axel to give you his dissertation on the topic.

Carlos

Thank you Dr. Kohlmeyer,

I understand your point. Actually my objective was to determine the

suitability of these two forcefields in describing thermal transport

through the cnt. As a later step, I subjected both of them to same delta_T

and measured the energy flux through the system. So I was trying to compare

those energy fluxes. I now understand that comparing absolute magnitudes of

energy may not make sense at all, however, I am hoping to find a common

basis based ( lets say thermal energy flux or thermal conductivity, or

phonon spectrum in a region) on which I could say one ff is better

applicable for describing heat transfer through the same structure. Would

you please suggest what can be a meaningful comparison in determining that?

the only way to determine whether a force field is suitable (at all, or

better than another) to compute a property is to compute said property with

it and compare those results with each other and to other available

reference data. there is not *the one true force field*, but only one that

is more or less suitable for a certain purpose and a certain

thermodynamical state.

with respect to thermal conductivity, it cannot be stressed enough, that

this is a property where the procedure and setup of the calculation is

about as important, if not more important, than a suitable choice of force

field parameters. you also need to be aware that LAMMPS would only compute

the ballistic contribution to the heat transfer.

since this is a frequently discussed topic, i strongly suggest to first

spend a significant amount to read and understand previous discussions on

the subject in the mailing list archives.

axel.

Hi Souvik,

Here are my two cents with respect to comparing accuracy of force fields w.r.t. heat transfer.

You calculate the vibrational density of states from each of them and then compare it with experimental spectra to compare the peak positions and relative amplitude. For heat transfer which occurs through atomic vibrations, it is my opinion that VDOS is the best bet to compare against. Other things you can compare against are elastic moduli.

Best Regards,

Vikas