The ABINIT Users Mailing List ( CLOSED )

[Masayoshi Mikami <mmikami@yk.rim.or.jp>]

Dear Prasenjit Sen,

From the observation of the forum ML, more specific information 
could be disclosed to grasp what was going on ? 
(the 4th or 5th case in 
 http://www.abinit.org/ABINIT/Infos/FAQ.html#NoResponce )

Namely, the following could be disclosed :
  (Full) input file, information about Si/H Pseudopotential
  All the geometries (1st, 2nd, ..., 5th Broyden Step; 
  your system seems a bit too large to be tried by ourselves)

Attachment of the output file may be too large in your case. 
Attachment of large-sized files should be avoided on the ML.
(Should we set up a rule ("netiquette") about file attachment ?)

So, the above disclosure is just suggested. 
I heard a case similar with yours; after some MD steps of 
his/her Si surface model, suddenly SCF could not be converged.
But I have not yet received his/her input case (unfortunately...), 
so I have been wondering if what kind of error could happen.

So far, I suppose that SCF may fail for impractical geometries.
(Still there may be chances to have convergence for such cases)

Good continuation,
Masayoshi

From: prasen@mri.ernet.in
Subject: [abinit-forum] Convergence issues for Si(100)
Date: Tue, 1 Jun 2004 14:21:18 +0200
Message-ID: sympa.1086092321.23860.965@abinit.org

> Dear Abinitioners:
> 
> I am doing a calculation on Si(100) surface. First, I am trying to get
> the p(2x1) reconstruction, to be used for further calculations. I have a
> couple of questions about convergence of this system.
> 
> I am starting from atomic positions slightly disturbed from an ideal
> surface--alternate top layer atoms moved slightly (~0.1 A) towards the 
> possible
> dimerization direction. BTW, I have 5 Si layers 16 atoms per layer,
> bottom layer saturated by H, and 9 A vacuum. I use ionmov=2 (BFGS)
> scheme for atomic relaxation.
> 
> 1. For the first 4 Bryoden steps, SCF converges within 4-5 cycles
> (nstep).  However, from the 5th Broyden step onwards, SCF does not converge
> even within 60 cycles. The energy changes are quite large and there
> seems no tendency of convergence. Any help as to what might cause this?
> Unless SCF converges, probably the forces would be wrong.
> 
> May be related to this...
> 2. From the 3rd Broyden step onwards, some atoms are moved
> opposite to the direction of forces on them. Is this possible in BFGS
> scheme?
> 
> 3. Is it good to use ionmov=2 when there are a large no. of atoms (~96)
> to be relaxed, or should I use some other scheme?
> 
> Values of some parameters that may be related to convergence:
> diecut = 2.3        diemac = 12.0   diemix = 0.5   iprcel = 45  iscf = 5
> 
> Getting the p(2x1) reconstruction should not be too difficult I guess.
> Am I missing somethiing? Any help is appreciated, and thanks in advance.
> 
> Prasenjit Sen.
> Harish-Chandra Research Inst.
> Jhnsi, Allahabad 211019
> INDIA.