CCP Flagship: Quasiparticle Self-Consistent GW for Next-Generation Electronic Structure

PI: Prof. Mark van Schilfgaarde, King's College London

Computational methods for electronic structure are dominated by density functional theory (DFT) and extensions to it. As these extensions are being steadily refined, inherent fundamental limitations become increasingly problematic. There is a growing need for a community code, that does not depend on DFT. Ideally it should possess the following properties:

  1. grounded in a true ab initio framework, systematically improvable in an unambiguous manner
  2. capabilities documented in a comprehensive and readily understandable manner
  3. small to moderate learning barriers
  4. able to address many kinds of properties
  5. means for verification, validation, source code documentation and troubleshootings
  6. scale reasonably with system size to enable study of complex materials

It is a very large challenge to meet all of these objectives at once. The most difficult is the underlying engine (a), as the many-body problem is very difficult to solve. Quasiparticle Self-Consistent GW (QSGW) theory forms a very promising framework, perhaps the most promising one, that is true ab initio, universally applicable, and systematically improvable. In formulating fundamental quantum theory as a many-body perturbation theory, GW is the lowest order term (diagram). If the perturbation could be carried out to all orders, it would solve the many-body problem exactly. Discrepancies with experiment are highly systematic and often can be explained in terms of particular diagrams that have been omitted. Strategies to systematically incorporate higher order diagrams have been formulated, that if successful will make it possible in principle to reliably determine the electronic structure of nearly any material.

While we believe QSGW constitutes the best engine for a next-generation community code, (b-f ) are equally important criteria for it to be a practical tool that appeals to a wide user base. These form the primary focus of CCP'9 flagship proposal.

A web site: https://www.questaal.org//

Recently the GW has be redesigned for greater efficiency; QSGW calculations can be easily be done on systems with 30 to 40 atoms.  A special version designed for the Xeon Phi has also been built.  It accelerates the critical step (QSGW self-energy) by a factor of about 4 compared to a single X86 node.  This speedup made facilitated a high-throughput study of the family of V-VI-VII compounds with 36 atom unit cells (Scott McKenzie).  To see about using this special branch, contact Dimitar Pashov at <dimitar.pashov@kcl.ac.uk>. An interface to Kristjian Haule's DMFT code has been completed. An MPI version, developed by Martin Lüders at Daresbury, is near the final stages of completion, This will extend the range of systems possible to calculate.  It will be posted on the master branch very soon. Also Brian Cunningham at Queen's University, Belfast has recently developed an extension of the RPA polarizability to include ladder diagrams. This will also be folded into the master branch very shortly.

The source codes are hosted on bitbucket (bitbucket.org/lmto/lm). Anyone can download the codes but you first must register.  To get access, contact Lewis Christiansen, at Lewis Christiansen <lewis.christiansen@gmail.com>.

Please look at the website, try the codes and give us some feedback!