计算化学公社

标题: Help: Suggestion for gaussian calculation setup involving nanoparticles and DNA [打印本页]

作者
Author: rpestana94 时间: 2020-11-13 05:00
标题: Help: Suggestion for gaussian calculation setup involving nanoparticles and DNA
Greetings, I want to use gaussian to see the interaction between DNA and nanoparticles(NPs) of gold and silver, for these, I was thinking on calculate the binding energy between the DNA and the NPs and also make a cytosine(C)-guanine(G) and put one NPs in different places and measure the binding energy between C-G. For this, the cluster of gold and silver were made by putting some atoms and do an optimization using Lanl2dz, the C-G was optimized using B3LYP.

I'm not pretty sure how to calculate the energy between the C-G and the NPs, maybe using Gen keyword or even if this kind of approximation makes sense, so I open to suggestions, changes in calculation, improvement, advice, etc.

Thanks in advance.

作者
Author: sobereva 时间: 2020-11-13 06:52
You can refer to the binding energy calculation for benzene - Ag surface described in this post: http://bbs.keinsci.com/thread-16532-1-1.html
The difference is that you may want to use a spherical metal cluster instead of a finite slab model.

The binding energy can be calculated as E(GC-NP) - E(GC) - E(NP). For optimization purpose, Lanl2DZ for Au/Ag and 6-31G* for C-G is acceptable, however for evaluating binding energy, better basis set should be used, for example, SDD for Au/Ag and 6-311+G** for C-G, and its is best to consider Counterpoise correction. The recommended DFT functional for this case is PBE0-D3(BJ). B3LYP without dispersion correction should never be used for evaluating the binding energy because it is well known that B3LYP completely fails to represent dispersion effect.

作者
Author: wzkchem5 时间: 2020-11-13 10:50
Please note that LANL2DZ is a basis set while B3LYP is a functional. Every calculation requires both a functional and a basis set, so sentences like "do an optimization using Lanl2dz, the C-G was optimized using B3LYP" don't really make sense. Functionals are global, so one cannot specify different functionals for different atoms, but it is possible to assign different basis sets to different atoms.

作者
Author: rpestana94 时间: 2020-11-21 00:40

sobereva 发表于 2020-11-12 17:52
You can refer to the binding energy calculation for benzene - Ag surface described in this post: htt ...

Thanks, I will try it and let you know, any suggestion to create the spherical metal cluster?

作者
Author: rpestana94 时间: 2020-11-21 00:46

wzkchem5 发表于 2020-11-12 21:50
Please note that LANL2DZ is a basis set while B3LYP is a functional. Every calculation requires both ...

Thanks, you're right, I apologize I check and we use pbepbe/def2sv for both, the C-G and the metal cluster.

作者
Author: wzkchem5 时间: 2020-11-22 12:14

rpestana94 发表于 2020-11-21 00:46
Thanks, you're right, I apologize I check and we use pbepbe/def2sv for both, the C-G and the metal ...

def2sv is too small a basis set to be of publication quality, because it doesn't have any polarization functions. If you want to stick to the def2 series, you should use at least def2svpp.

作者
Author: sobereva 时间: 2020-11-27 11:08

rpestana94 发表于 2020-11-21 00:40
Thanks, I will try it and let you know, any suggestion to create the spherical metal cluster?

You can use VMD program to dig out a spherical cluster from supercell of metal

作者
Author: rpestana94 时间: 2021-3-30 06:02

sobereva 发表于 2020-11-12 17:52
You can refer to the binding energy calculation for benzene - Ag surface described in this post: htt ...

Prof Sob, for this question about the nanoparticle, following the tutorial you suggest and what you comment here, should I use this approximation?
! PBE0 D3 def2-TZVP(BJ) def2-SV(P) RIJCOSX opt noautostart miniprint nopop

Using the same functional for optimization and for single point calculation? That is publication quality? In my case is a nanoparticle I think I don't need to freeze atoms as you do in the tutorial but not totally sure.

作者
Author: sobereva 时间: 2021-3-30 12:52

rpestana94 发表于 2021-3-30 06:02
Prof Sob, for this question about the nanoparticle, following the tutorial you suggest and what yo ...

Your keywords are incorrect.

If your nanoparticle is very large, due to consideration of computational cost, I suggest using pure DFT functional, it is significantly cheaper than hybrid-functional in ORCA when RI approximation is employed.

I suggest using the following keywords for optimization:
! PBE D3 def2-SVP def2/J noautostart miniprint nopop opt

For single point calculation:
! PBE D3 def2-TZVP def2/J noautostart miniprint nopop

Probably there are better choices, but I cannot give you more suggestion without detailed information of your nanoparticle (number of atoms, element...) and your computational resource.

作者
Author: wzkchem5 时间: 2021-3-30 17:03

rpestana94 发表于 2021-3-30 06:02
Prof Sob, for this question about the nanoparticle, following the tutorial you suggest and what yo ...

BJ is a modifier of D3, not a modifier of the basis set. Writing BJ directly after the basis set is a syntax error. On the other hand, however, ORCA automatically uses D3BJ when you write D3, so that's why Sob omitted the BJ altogether.

作者
Author: rpestana94 时间: 2021-3-30 21:35

sobereva 发表于 2021-3-29 23:52
Your keywords are incorrect.

If your nanoparticle is very large, due to consideration of comput ...

Prof Sob, the nanoparticle is about 43 Au atoms, the two DNA pair each one has around 29 to 32 atoms, so in the system, I will have the nanoparticle and a base pair of DNA, (43+32+32=107), this kind of calculation is suitable for this system, I think is a little big or what you recommend?

作者
Author: rpestana94 时间: 2021-3-30 21:36

wzkchem5 发表于 2021-3-30 04:03
BJ is a modifier of D3, not a modifier of the basis set. Writing BJ directly after the basis set i ...

I see, thanks, I didn't know that, I'm really new in this kind of calc and still is a bit confusing to me

作者
Author: sobereva 时间: 2021-3-30 21:42

rpestana94 发表于 2021-3-30 21:35
Prof Sob, the nanoparticle is about 43 Au atoms, the two DNA pair each one has around 29 to 32 ato ...

My recommendation of keywords works for your system.

作者
Author: rpestana94 时间: 2021-3-30 21:53

sobereva 发表于 2021-3-30 08:42
My recommendation of keywords works for your system.

thanks

作者
Author: rpestana94 时间: 2021-3-30 21:58

sobereva 发表于 2021-3-30 08:42
My recommendation of keywords works for your system.

Prof Sob another question, make sense to optimize the nanoparticle by it self and the optimize again with the base pair or just optimize the base pair with the nanoparticle immediately? I get the base pair using gaussian (gauss view) and the nanoparticle was made using charm-gui

作者
Author: sobereva 时间: 2021-3-31 01:58

rpestana94 发表于 2021-3-30 21:58
Prof Sob another question, make sense to optimize the nanoparticle by it self and the optimize aga ...

If there is no special reason, just optimize the complex.

作者
Author: rpestana94 时间: 2021-3-31 19:46

sobereva 发表于 2021-3-30 12:58
If there is no special reason, just optimize the complex.

thanks

作者
Author: rpestana94 时间: 2021-4-6 23:43

sobereva 发表于 2020-11-26 22:08
You can use VMD program to dig out a spherical cluster from supercell of metal

Prof Sob, i have a question about the construction of the nanoparticle, I'm trying to use charmm-gui to build it but I get an odd number of gold atoms, this gives problem with orca calculation due to the multiplicity, so how I can solve this multiplicity problem or better try to get one nanoparticle with even number of gold atoms?

作者
Author: sobereva 时间: 2021-4-7 00:31

rpestana94 发表于 2021-4-6 23:43
Prof Sob, i have a question about the construction of the nanoparticle, I'm trying to use charmm-g ...

Use GaussView to properly edit your nanoparticle (e.g. delete an atom or add a atom)

作者
Author: rpestana94 时间: 2021-4-7 01:05

sobereva 发表于 2021-4-6 11:31
Use GaussView to properly edit your nanoparticle (e.g. delete an atom or add a atom)

if I remove or add one atom it can be randomly or for example I should remove the center one, the interaction with this will be really low or zero but not sure if this can destabilize the whole structure or better remove one on the surface?

作者
Author: sobereva 时间: 2021-4-7 01:20

rpestana94 发表于 2021-4-7 01:05
if I remove or add one atom it can be randomly or for example I should remove the center one, the ...

I can't give you answer because I know nothing about character of your practical system.

作者
Author: rpestana94 时间: 2021-4-7 22:37

sobereva 发表于 2021-4-6 12:20
I can't give you answer because I know nothing about character of your practical system.

Thanks for the advice I solve the problem

作者
Author: rpestana94 时间: 2021-6-25 11:20

sobereva 发表于 2021-3-29 23:52
Your keywords are incorrect.

If your nanoparticle is very large, due to consideration of comput ...

Professor Sob thanks for this recommendation for this basis set in gaussian16 how it would be? Here I will try to guess the correct keywords but let me know if I'm wrong:

For opt
opt pbepbe/3-21+g/def2sv geom=connectivity

For single point
pbepbe/3-21+g/def2tzv geom=connectivity

I don't know if the dispersion correction is already added by default in gaussian

Thanks in advance

作者
Author: zjxitcc 时间: 2021-6-25 11:29

rpestana94 发表于 2021-6-25 11:20
Professor Sob thanks for this recommendation for this basis set in gaussian16 how it would be? Her ...

These keywords will not provide any meaningful result.

For practical computation, a double-zeta basis set is at least required, such as def2SVP, 6-31G(d,p), 6-31G(d), etc. So 3-21+g is useless.

For RI (or density fitting), you should use ORCA instead of Gaussian, since density fitting in Gaussian does not accelerate the computation (or acceleration is little). But this is significant in ORCA.

作者
Author: rpestana94 时间: 2021-6-25 11:36

zjxitcc 发表于 2021-6-24 22:29
These keywords will not provide any meaningful result.

For practical computation, a double-zeta ...

orca is significantly faster than gaussian for general purposes or just with the dispersion?

作者
Author: zjxitcc 时间: 2021-6-25 11:43
本帖最后由 zjxitcc 于 2021-6-25 11:46 编辑

rpestana94 发表于 2021-6-25 11:36
orca is significantly faster than gaussian for general purposes or just with the dispersion?

ORCA is much faster than Gaussian for RI (or density fitting) related computations. If RI is turned off, ORCA may be slower than Gaussian (but it also depends on which method you use).

By the way, the dispersion correction keyword in Gaussian is em=GD3 or em=GD3BJ, where 'BJ' means with Becke-Johnson damping. The dispersion correction is not default, and you should add it explicitly. You can find this on http://gaussian.com/dft. Dispersion correction costs little time (less than 1s, or only several seconds) in any quantum chemistry software.

作者
Author: rpestana94 时间: 2021-6-25 12:09

zjxitcc 发表于 2021-6-24 22:43
ORCA is much faster than Gaussian for RI (or density fitting) related computations. If RI is turne ...

Thanks

作者
Author: sobereva 时间: 2021-6-25 18:09

rpestana94 发表于 2021-6-25 11:36
orca is significantly faster than gaussian for general purposes or just with the dispersion?

ORCA is significantly faster than Gaussian for single point and optimization tasks when RIJ (for pure DFT functionals) or RIJCOSX/RIJK (for hybrid DFT functionals) or RI (MP2 and double-hybrid functionals) is enabled for medium and large-sized systems.

However, speed of frequency analysis of current version of ORCA (4.2.1) doesn't outperform Gaussian even if RI related techniques is enabled.

Gaussian also has RIJ for pure DFT functionals, however it significantly speeds up calculation only when you use relatively large basis sets (e.g. def2-TZVP) or in frequency analysis process.

Overall, you can benefit from RI related techniques in ORCA much more than Gaussian.

作者
Author: wzkchem5 时间: 2021-6-25 18:49

sobereva 发表于 2021-6-25 11:09
ORCA is significantly faster than Gaussian for single point and optimization tasks when RIJ (for p ...

As a side note: ORCA 5.0 will be released on next Thursday. The frequency calculation of ORCA 5.0 will be noticeably faster than in ORCA 4.2.1. But due to some (in)famous reasons I cannot comment on whether ORCA 5.0 will become faster than Gaussian in this aspect.

作者
Author: rpestana94 时间: 2021-6-25 20:50

sobereva 发表于 2021-6-25 05:09
ORCA is significantly faster than Gaussian for single point and optimization tasks when RIJ (for p ...

Three questions, about the basis sets in gaussian how it would be? I have your recommendation in ORCA but I don't know exactly how to translate this to gaussian, is there a way to do this automatically?

If I'm right gaussian16 have some GPU acelleration for some process, ORCA have something like this?

The RIJCOSX and this kind of keywords actually accelerate the calculation but I understand that this also give some error in the measurement, It will be better to do a calculation with RIJCOSX activate and then a new calculation deactivating RIJCOSX or the error is too low

作者
Author: sobereva 时间: 2021-6-25 23:20

rpestana94 发表于 2021-6-25 20:50
Three questions, about the basis sets in gaussian how it would be? I have your recommendation in O ...

If gridx setting is not quite small (i.e. the grid for COSX part is not too coarse), the numerical error caused by RIJCOSX is fully negligible. So you do not need to worry about this.

I have a blog article about choice of basis set, see http://sobereva.com/336 (in Chinese). The recommendations in this article are very suitable for Gaussian users.

GPU acceleration performance in G16 is quite poor, simply ignore it. ORCA currently doesn't support GPU acceleration.

作者
Author: rpestana94 时间: 2021-6-26 00:14

sobereva 发表于 2021-6-25 10:20
If gridx setting is not quite small (i.e. the grid for COSX part is not too coarse), the numerical ...

Thanks professor Sob

作者
Author: rpestana94 时间: 2021-7-24 00:52

sobereva 发表于 2021-3-29 23:52
Your keywords are incorrect.

If your nanoparticle is very large, due to consideration of comput ...

Professor I running the system with this configuration using orca 5, I'm reading about RI and auxiliary basis sets to speed up the calculation and help to converge, but I'm kind of lost of which one I should use, I think RI or RIJCOSX can be worth using but not sure which one I should choose or there are a better one to consider

作者
Author: wzkchem5 时间: 2021-7-24 01:15

rpestana94 发表于 2021-7-23 17:52
Professor I running the system with this configuration using orca 5, I'm reading about RI and auxi ...

RI is for pure functionals, while RIJCOSX is for hybrid functionals. However in ORCA 5 there is relatively little need to worry about this, since the correct one is turned on by default. This is different from ORCA 4 where RI is turned on by default for pure functionals, but hybrid functionals by default do not use any approximation.
Besides, please note that RI only speeds up individual SCF iterations, but does not reduce the number of iterations, so they cannot be said to "help to converge".

作者
Author: rpestana94 时间: 2021-7-24 01:47

wzkchem5 发表于 2021-7-23 12:15
RI is for pure functionals, while RIJCOSX is for hybrid functionals. However in ORCA 5 there is re ...

I see, so in orca 5 I can leave out that keyword and by defaults orca decided the best things to do

欢迎光临计算化学公社 (http://ccc.keinsci.com/)