Error reading file with the skip routine

General discussion of the Cambridge quantum Monte Carlo code CASINO; how to install and setup; how to use it; what it does; applications.
Post Reply
Christopher_Eames
Posts: 8
Joined: Sun Dec 22, 2013 1:36 pm

Error reading file with the skip routine

Post by Christopher_Eames »

Dear CASINO community

I'm trying to calculate the formation energy of lithium peroxide. This is my first real go with CASINO and so I'm on a steep learning curve.

For the O2 calculation I'm using a primitive cell of four O2 molecules from the ICSD (I used the same one with DFT to determine the O2 chemical potential). Using CASTEP I calculated the wavefunction with a 3x3x3 k-point mesh and now with CASINO I'm using a 3x3x3 supercell. When I run vmc_opt I get an error message;

Reading plane-wave function and associated data
===============================================

ERROR : SKIP
Error reading file with the skip routine.

I don't really understand this message? My input and correlation.data files are attached. If anyone can spot any elementary mistakes I would be most grateful.

I have managed to calculate the VMC total energy of Li2O2 satisfactorily (-1.9453 au/atom) and I can calculate the vmc energy of lithium metal but here the energy seems way too low (-0.22 au/atom). My inputs for lithium are attached and I would again appreciate any help.

Best Wishes

Chris Eames.

P.S. So far I haven't used any p-terms in the Jastrow factor and I haven't tried any finite size corrections (but I will do both for production runs).

P.P.S. Mike - As you can see I have switched system from LiVO2 to Li2O2. I decided to use my computer time on a high profile system.
inputs.tgz
(5.88 KiB) Downloaded 821 times
Mike Towler
Posts: 239
Joined: Thu May 30, 2013 11:03 pm
Location: Florence
Contact:

Re: Error reading file with the skip routine

Post by Mike Towler »

Dear Chris,

I'll investigate this for you - give me a few minutes.

By the way, don't use a plane-wave basis set directly in CASINO because that will add an extra factor of N to the scaling system size (i.e. be very much more expensive to evaluate than it need be). You need to re-expand the orbitals in localized 'blip' functions using the blip utility and then CASINO will run orders of magnitude faster. Basically just type 'blip' in any directory containing a pwfn.data file - see the manual. This will give you a 'bwfn.data' file instead - change the 'atom_basis_type' keyword at the top of the input file from 'plane-wave' to 'blip'.

Of, if you use PWSCF rather than CASTEP to generate the trial wave function then it can be told to generate blip files directly without going through an intermediate plane-wave stage.

Note that as a consequence of nobody using the old plane-wave orbital facility in CASINO any more, errors are less likely to be routinely detected; I hope you haven't discovered one!

Best wishes,
Mike
Christopher_Eames
Posts: 8
Joined: Sun Dec 22, 2013 1:36 pm

Re: Error reading file with the skip routine

Post by Christopher_Eames »

Hi Mike

Thanks for the reply.

While converting to blips I got an error message about the pwfn.data file being incomplete for oxygen. I remade pwfn.data and now things are fine. The reason I'm using CASTEP here is that I couldn't find a compiled casino2upf on hector so I couldn't use PWSCF. But the convenience of the CASINO pseudopotentials being CASTEP compatible is a fair swap for ready-made blips in PWSCF.

Hopefully, using blips will now make things run more efficiently. I'm still a bit unsure about the generating wavefunctions part but as it says in the manual this is quite tricky to learn.

The one thing that doesn't sit right with me is using an lxmxn kpoint grid in dft to make a wavefunction for a lxmxn supercell in casino. For a metal this will mean the k-point sampling is far too sparse and result in a very poor input wavefunction?

Best Wishes

Chris.
Mike Towler
Posts: 239
Joined: Thu May 30, 2013 11:03 pm
Location: Florence
Contact:

Re: Error reading file with the skip routine

Post by Mike Towler »

OK, so something had just interrupted the write of your pwfn.data file? (disk space???) This doesn't appear to be my problem then, so I won't investigate further..
I'm using CASTEP here is that I couldn't find a compiled casino2upf on hector so I couldn't use PWSCF.
Yep, that sounds about right - casino2upf is a PWSCF utility and isn't included with the CASINO distribution. To quote the README file:

Code: Select all

CASINO2UPF
----------

The fortran routines here are copies of the relevant routines of the CASINO
pseudopotential utility supplied with the PWSCF/Quantum Espresso distribution.
These are provided for information only, and unlike casino2gon this is not a
standalone functional utility (since it INCLUDEs lots of espresso modules not
given here).

You must download and install Quantum Espresso in order to use this utility:

http://www.quantum-espresso.org/

Full instructions for use are given in the INSTRUCTIONS file in this directory
- this is a partial copy of the README file in the uptools directory of the
  Espresso distribution (which may have been updated since I copied it - do
check..).

MDT 3.2011
The one thing that doesn't sit right with me is using an lxmxn kpoint grid in dft to make a wavefunction for a lxmxn supercell in casino. For a metal this will mean the k-point sampling is far too sparse and result in a very poor input wavefunction?
Fair point. Some converters (like the CRYSTAL to CASINO one) will allow you to use e.g. a 6x6x6, 9x9x9, 12x12x12 grid in the DFT calculation and 'pluck' out a subset of orbitals from 3x3x3 subset of k points for use in the QMC. There is an argument to say that this is a bad idea since it corresponds to using suboptimal orbitals for the 3x3x3 cell; however, hard numerical evidence for how much this actually matters is lacking.

I'm not a CASTEP user, so I just asked Neil (our CASTEP expert) whether it will do that, and he says no. "Only if you did it by hand, or wrote a program to doctor the pwfn.data file, or adapted the castep2casino utility".

So just try it anyway, using as big a cell as you can, and you should probably use twist-averaging as well (randomizing the offset of the k point of the origin every so often in the DMC). Remember you only need to get the zeroes of the many-body wave function (the nodal surface) more or less correct, and DMC will correct the wave function everywhere else, so DMC is supposedly much less sensitive to the overall quality of the wavefunction than the DFT calc.

There are various papers on metals (on sodium and stuff like that) in the archive on the website where people probably discuss these issues.

Mike
Post Reply