[QE-users] 回复: 回复: 回复: Is it necessary to do "vc-relax" with pw.x before doing CPMD with cp.x ?

Sun Apr 10 00:42:37 CEST 2022

Dear Hsin-Yu,

I really appreciate your patience and detailed explanation. Now I understood
my questions much better. 

Anyway, I will do a series of tests to choose these parameters as per the
direction you introduced.

Thanks a lot for your kind help.

All the best,
Kejiang
-------------------------------------------------
Dr. Kejiang Li
School of Metallurgical and Ecological Engineering,
University of Science and Technology Beijing
Xueyuan Rd., Haidian District, Beijing 100083, P. R. China
https://kejiangli.com/

发件人: users-bounces at lists.quantum-espresso.org
<users-bounces at lists.quantum-espresso.org> 代表 Hsin-Yu Ko
发送时间: 2022年4月10日 0:05
收件人: 'Quantum ESPRESSO users Forum' <users at lists.quantum-espresso.org>;
users-request at lists.quantum-espresso.org
主题: Re: [QE-users] 回复: 回复: Is it necessary to do "vc-relax" with pw.x
before doing CPMD with cp.x ?

Dear Kejiang,

You are welcome. For the first question, emas_cutoff is used to scale the
mass of Fourier components of the wavefunction during the propagation of
CPMD dynamics so as to compress the electronic power spectrum (i.e., the
span of the lowest to the highest frequency modes of the wavefunction);
doing so allows one to choose a larger timestep during the propagation.
Please see [Tassone, Mauri, and Car, Phys. Rev. B 50, 10561 (1994).] for
details. When larger emass_cutoff is given (make sense upto ecutwfc), less
mass preconditioning is introduced so closer to the original CPMD result
(though might require a smaller dt).

For the second, the core idea of CPMD is essentially to reproduce BOMD
without the need to perform self-consistent field calculations beyond the
first step (thus reducing the computational cost). For metallic systems, the
validity of the BO approximation becomes questionable with the closing
bandgap. As such, this topic is still a branch of active research and
requires careful understanding of the simulation results even though both
CPMD (may need electronic thermostats) and BOMD machinery can be practically
performed. You might want to consult a colleague familiar with ab initio
simulations of iron oxides and iron surfaces to make these decisions.

Best,

Hsin-Yu

-- 

Hsin-Yu Ko

Postdoctoral Research Fellow

Department of Chemistry and Chemical Biology
Cornell University

  _____  

From: users <users-bounces at lists.quantum-espresso.org
<mailto:users-bounces at lists.quantum-espresso.org> > on behalf of
likejiang at ustb.edu.cn <mailto:likejiang at ustb.edu.cn>  <likejiang at ustb.edu.cn
<mailto:likejiang at ustb.edu.cn> >
Sent: Friday, April 8, 2022 7:19 PM
To: 'Quantum ESPRESSO users Forum' <users at lists.quantum-espresso.org
<mailto:users at lists.quantum-espresso.org> >;
users-request at lists.quantum-espresso.org
<mailto:users-request at lists.quantum-espresso.org>
<users-request at lists.quantum-espresso.org
<mailto:users-request at lists.quantum-espresso.org> >
Subject: [QE-users] 回复: 回复: Is it necessary to do "vc-relax" with pw.x
before doing CPMD with cp.x ? 

Dear Hsin-Yu,

Thanks a lot for your explanation and sharing the document.

Now I understand the two questions much better, but I have two more
questions to consult with you.

The first is a small question:
In the CPMD tutorial, it also suggests to tune ‘emass_cutoff’. What is the
physical philosophy to tune ‘emass_cutoff’ ?  Will a higher emass_cutoff
produce a result more accurately? 

The second is might a bigger question:
I noticed in many places that we should use BOMD if we need to simulate
metal system or systems with a band gap. What is the physical philosophy
behind this practice? In fact, I am simulating the reduction of iron oxides
by H2 to produce iron. There are band gaps for various kinds of iron oxides
and metal (iron) will be produced after reaction. Do I need to used BOMD
instead of CPMD?

Sorry that my academic background is not related with quantum physics or
chemistry, I am not good at understand the complex equations related with
quantum chemistry.

Best regards,
Kejiang

-------------------------------------------------
Dr. Kejiang Li
School of Metallurgical and Ecological Engineering,
University of Science and Technology Beijing
Xueyuan Rd., Haidian District, Beijing 100083, P. R. China
https://kejiangli.com/

发件人: users-bounces at lists.quantum-espresso.org
<mailto:users-bounces at lists.quantum-espresso.org>
<users-bounces at lists.quantum-espresso.org
<mailto:users-bounces at lists.quantum-espresso.org> > 代表 Hsin-Yu Ko
发送时间: 2022年4月9日 1:10
收件人: 'Quantum ESPRESSO users Forum' <users at lists.quantum-espresso.org
<mailto:users at lists.quantum-espresso.org> >;
users-request at lists.quantum-espresso.org
<mailto:users-request at lists.quantum-espresso.org> 
主题: Re: [QE-users] 回复: Is it necessary to do "vc-relax" with pw.x before
doing CPMD with cp.x ?

Dear Kejiang,

Good questions! The physical idea of emass is to slow down the timescale of
the fictitious electronic motions (to allow larger integration timestep, dt,
for propagating CPMD). However, emass has to be chosen to maintain good
adiabatic separation between fictitious electronic and ionic degrees of
freedom such that the electronic wavefunctions remain close to the ground
state (BO surface) during the CPMD propagation. A very nice explanation
(including how parameters should be chosen) by Marx and Hutter can be found
in Secs. 2.4.3-2.4.4 at
https://juser.fz-juelich.de/record/44687/files/NIC-Band-1.pdf#page=311. 

For your second question, it depends on what properties one wants to sample
via CPMD. For instance, the structural properties sampled in the canonical
(NVT) ensemble is independent of ionic mass (assuming that the nuclei are
classical particles) since the resulting/equilibrated trajectory follows the
Boltzman distribution (independent of ionic mass). However, arbitrary
choices of ionic mass certainly defeat the purpose if the goal is to extract
dynamical properties.

Hope that helps,

Hsin-Yu

-- 

Hsin-Yu Ko

Postdoctoral Research Fellow

Department of Chemistry and Chemical Biology
Cornell University

  _____  

From: users <users-bounces at lists.quantum-espresso.org
<mailto:users-bounces at lists.quantum-espresso.org> > on behalf of
likejiang at ustb.edu.cn <mailto:likejiang at ustb.edu.cn>  <likejiang at ustb.edu.cn
<mailto:likejiang at ustb.edu.cn> >
Sent: Friday, April 8, 2022 12:04 PM
To: 'Quantum ESPRESSO users Forum' <users at lists.quantum-espresso.org
<mailto:users at lists.quantum-espresso.org> >;
users-request at lists.quantum-espresso.org
<users-request at lists.quantum-espresso.org
<mailto:users-request at lists.quantum-espresso.org> >
Subject: [QE-users] 回复: Is it necessary to do "vc-relax" with pw.x before
doing CPMD with cp.x ? 

Dear Hsin-Yu,

Thanks a lot for your explanation and suggestion.

As per your comments, it might be due to high emass since I used the default
emass (400). I will try to decrease emass (300, 200, 100) to see whether the
force can converge.

Regarding ‘emass’, I have one more question:
Is it a common practice or acceptable to change the value of ‘emass’ in
different stages of the simulation, i.e., emass (50) in the electronic
minimization stage, emass (200) in the ionic minimization stage and emass
(400) in the CPMD stage.  If yes, what is the physical philosophy behind or
support this practice? 

I addition, the official guide also suggests to decrease the ionic mass the
speed up the computation. Why is this physically acceptable since a specific
kind of element should has a specific mass?

Thanks again for your kind help.

Best regards,
Kejiang

发件人: users-bounces at lists.quantum-espresso.org
<mailto:users-bounces at lists.quantum-espresso.org>
<users-bounces at lists.quantum-espresso.org
<mailto:users-bounces at lists.quantum-espresso.org> > 代表 Hsin-Yu Ko
发送时间: 2022年4月8日 21:29
收件人: users at lists.quantum-espresso.org
<mailto:users at lists.quantum-espresso.org> ;
users-request at lists.quantum-espresso.org
<mailto:users-request at lists.quantum-espresso.org> 
主题: Re: [QE-users] Is it necessary to do "vc-relax" with pw.x before doing
CPMD with cp.x ?

Hi Kejiang,

Structural relaxation in cp.x (step 2 as you mentioned) can involve the cell
degrees of freedom (in addition to the ionic ones) and thus is in principle
equivalent to doing "vc-relax" in pw.x. As such, it is unnecessary to do a
"vc-relax" with pw.x.

In practice, proper parameter setting relies on some intuition of the
targeted physical system. I think the difficulty in force convergence
probably is related to "emass" and "dt" (e.g., when the emass is set too
large, ionic motion during relaxation could fictitiously heat up the
electronic degrees of freedom and lead to fluctuations in forces).
Nevertheless, detailed input/output information would be needed for more
definitive comments.

Best,

Hsin-Yu

-- 

Hsin-Yu Ko

Postdoctoral Research Fellow

Department of Chemistry and Chemical Biology
Cornell University

  _____  

From: users <users-bounces at lists.quantum-espresso.org
<mailto:users-bounces at lists.quantum-espresso.org> > on behalf of
likejiang at ustb.edu.cn <mailto:likejiang at ustb.edu.cn>  <likejiang at ustb.edu.cn
<mailto:likejiang at ustb.edu.cn> >
Sent: Friday, April 8, 2022 2:44 AM
To: users at lists.quantum-espresso.org
<mailto:users at lists.quantum-espresso.org>  <users at lists.quantum-espresso.org
<mailto:users at lists.quantum-espresso.org> >;
users-request at lists.quantum-espresso.org
<mailto:users-request at lists.quantum-espresso.org>
<users-request at lists.quantum-espresso.org
<mailto:users-request at lists.quantum-espresso.org> >
Subject: [QE-users] Is it necessary to do "vc-relax" with pw.x before doing
CPMD with cp.x ? 

Dear QE-Users,

I am doing CPMD with cp.x for iron oxide system following the official
tutorial which introduced two steps (Step1: Reaching the electronic ground
state; Step 2: Relax the system) before doing cpmd.

However, in both the electronic minimization and ionic minimization
processes, I found that the force cannot converge (forc_conv_thr=1.0D-3
cannot be satisfied), while it is relative easy for electronic energy and
total energy to converge (ekin_conv_thr=1.0D-6, etot_conv_thr= 1.0D-4). 

I am wondering whether this is caused by that the initial crystal structure
is not fully relaxed with the current pseudopotential. Is it necessary to do
"vc-relax" with pw.x before doing CPMD with cp.x ?  If yes, which parameters
should be kept to be the same in both pw.x and cp.x?

Any comment or suggestion will be much appreciated.

Thanks,
Kejiang

-------------------------------------------------
Dr. Kejiang Li
School of Metallurgical and Ecological Engineering,
University of Science and Technology Beijing
Xueyuan Rd., Haidian District, Beijing 100083, P. R. China
https://kejiangli.com/

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20220410/61fddb90/attachment.html>