[Pw_forum] PWSCF QMC interface

Mike Towler mdt26 at cam.ac.uk
Fri Apr 15 12:59:08 CEST 2011



Dear all,

I've been asked to provide a brief overview of the new interface to the 
CASINO quantum Monte Carlo program available in the new release of PWSCF.

For those of you who don't know, QMC is a highly-accurate technique with 
similar or better accuracy than the best high-order quantum chemistry 
methods (such as CCSD(T) etc.). Its advantage lies in the fact that the 
required CPU time scales only quadratically with system size (rather than 
as N^7 or whatever) meaning that it is potentially applicable to large 
systems - periodic or otherwise - of up to several thousand electrons. 
It's the same technique that Ceperley and Alder used to do their 
essentially exact calculations of the homogeneous electron gas thirty 
years ago that led to the development of the LDA exchange-correlation 
functionals etc., but we use it to study real systems with atoms.

It's significantly more expensive than DFT, of course, but unlike PW-DFT 
or standard quantum chemistry techniques the QMC algorithm can be made to 
have essentially perfect parallel efficiency and our CASINO code can 
easily exploit as many cores as you like on the biggest machines 
available. For example, it was recently tested on 120000 nodes of the ORNL 
Jaguar machine showing perfect parallel scaling.  For more details, 
together with a recent accessible review of the technique, see last 
February's Psi-k highlight written by me, Dario Alfe and Mike Gillan - 
available here:

http://www.tcm.phy.cam.ac.uk/~mdt26/papers/petascale_psik.pdf

Why the interface with PWSCF? Because QMC requires a 'starting guess' for 
the many-body wave function - which can be, for example, a determinant of 
Kohn-Sham orbitals multiplied by a so-called 'Jastrow factor' (the latter 
introduces an explicit dependence on the distances between pairs of 
particles). The KS orbitals need to be obtained from a DFT calculation, 
for example, with PWSCF (though we also support CASTEP, ABINIT, GP etc.. 
as well as various programs using other basis sets such as Gaussians).

Full instructions are provided in the 'Direct interface with CASINO' 
section of the latest Espresso user guide, and in the documentation 
provided with the 2.9 version of CASINO. But basically it involves little 
more than adding a '-pw2casino' flag when invoking pw.x, thus:

pw.x -pw2casino < input_file > output_file

will produce a wave function file readable by CASINO.

For more information, the CASINO web site is here:

http://www.tcm.phy.cam.ac.uk/~mdt26/casino2.html

If anybody wants to try some QMC calculations, the code is given away for 
free to academics (to see how to get it, click the 'Introduction' link in 
the left-hand sidebar).

Finally, PWSCF users wishing to learn how to use CASINO may like to attend 
one of our annual CASINO summer schools in my "Apuan Alps Centre for 
Physics" institute - a 15th century monastery in a beautiful mountain 
village in Tuscany, Italy. The sixth event in this series will be held 
this year from 30th July to 6th August - there are still some place 
available so if you wish to attend then let me know.. The official ad for 
this is below, and more details are available on the website:

http://www.vallico.net/tti/tti.html (Click 'Public Events').

I hope that the Espresso community will find the new QMC interface useful 
(and thanks to Norbert Nemec, Layla Martin-Samos, and Simon Binnie for 
help with creating it). Have fun!

Best wishes,
Mike

+------------------------------------------------------------------------+
|Dr. Mike Towler (mdt26 at cam.ac.uk) Theory of Condensed Matter (Rm 529)|
|  Cavendish Laboratory, University of Cambridge, Cambridge  CB3 0HE, UK |
|Tel. +44-(0)1223-337378 OR -334256 (College)     Fax. +44-(0)1223-337356|
+----------------------: www.tcm.phy.cam.ac.uk/~mdt26 :------------------+



                    SUMMER SCHOOL ANNOUNCEMENT

          "Quantum Monte Carlo and the CASINO program VI"
          ===============================================

               Sat 30th July - Sat 6th August 2011
  Apuan Alps Centre for Physics @ TTI, Vallico Sotto, Tuscany, Italy

              http://www.vallico.net/tti/tti.html

            A4 POSTER FOR THE SCHOOL AVAILABLE HERE :
         http://www.tcm.phy.cam.ac.uk/~mdt26/poster.png


The sixth international summer school in the series "Quantum Monte Carlo 
and the CASINO program" will take place during August 2011 at the TTI 
monastery in the Tuscan Apuan Alps in Italy, organized and hosted by 
members of Cambridge University physics department's Theory of Condensed 
Matter Group. The aim of the school is to give students a thorough 
introduction to quantum Monte Carlo as a method for performing 
high-quality calculations of the electronic structure of atoms, molecules, 
and materials. The course is designed for young quantum chemists or 
theoretical physicists who have no previous experience with this 
technique, though anyone interested is welcome to take part, including old 
people.

The monastery is a unique venue where the community spirit and magnificent 
location have inspired memorable workshops in the past. It is a delightful 
16th century building incorporating an ancient church, and is situated in 
the isolated but spectacular setting of the Tuscan mountain village of 
Vallico Sotto. The church is fully equipped with relevant presentation and 
computer technology, and all accommodation is on-site.  As with all events 
at the Institute, formal lectures are restricted to the mornings, and 
participants are given the freedom and space to think and to contemplate 
and discuss the issues at hand. In addition to hands-on exercises, a 
programme of healthy recreational activities will be organized in the 
afternoons, and it is hoped that by following this strict regime, together 
with breathing clean mountain air and by preparing and sampling fine 
Tuscan cuisine, the participant will be able to return home mentally and 
physically refreshed as well as better informed.

Describing the complex behaviour of materials at the atomic level requires 
a sophisticated description of the correlated motion of the electrons. 
Quantum Monte Carlo (QMC) is an increasingly popular and explicitly 
many-body method with the unusual capability of yielding highly accurate 
results whilst also exhibiting a very favourable scaling of computational 
cost with system size. Over the last eighteen years, the Cambridge group 
have been researching QMC methods and we have created a powerful, general 
computer program - CASINO - to carry out the calculations. The school will 
focus both on the basic theory of QMC and on more advanced practical 
techniques, and will include a thorough introduction to the CASINO 
program. A background in density functional theory or similar - though not 
essential - is normally thought to be useful.

Instructors at the school will include the main authors of the CASINO program
(Dr. Mike Towler, Dr. Neil Drummond and Dr. Pablo Lopez Rios) and possibly
others.

Participants would normally need to book a flight to Pisa airport from
where onward transportation will be arranged (though other destinations
are possible). Details of previous schools - including photographs - are
available under the PUBLIC EVENTS link on the TTI web site.

Interested students should email Mike Towler (mdt26 at cam.ac.uk) for
registration and further details.



More information about the Pw_forum mailing list