MLeaP

Machine Learning Pseudopotentials (Māšinmācīšanās pseidopotenciālu konstruēšanai in Latvian) or simply MLeaP addresses the following unsolved issues regarding pseudopotentials in electronic structure calculations:

• insufficient reference data for testing and training pseudopotentials,
• insufficient precision,
• there are numerous exchange-correlation functionals, and using them requires an especially constructed pseudopotential,
• preparation of pseudopotentials is a time-consuming and highly non-trivial process.

MLeaP aims at solving the described problems employing machine-learning techniques. The objectives are as follows:

• generate highly precise all-electron reference data (addressed in WP1),
• find a set of essential descriptive parameters that allow one to predict the quality of a pseudopotential (WP2),
• use machine-learning techniques for finding the optimal set of parameters (WP3).

Project Leader: Prof. Andris Ambainis (Center for Quantum Computing Science, Faculty of Computing)
Investigator: As. Prof. Andris Gulans
Administrative coordinator: Agnese Ozoliņa
External collaborator: Prof. Stefan Goedecker (University of Basel)

Posted on 30.03.2022

• Scientific activities.
  • We have designed a test set of molecules with chemical species from the first for rows of the periodic tables and computed atomisation atomisation energies using the PBE exchange-correlation functional.

• Seminars and lectures:
  • A public lecture “Precizitāte un atkārtojamība pirmo principu aprēķinos” (Precision and reproducibility in first principles calculations) at the LU Professor Council meeting on February 21, 2022. As the result, Andris Gulans was appointed as an associate professor at the faculaty of Physics, Mathematics and Optometry at LU.
  • A seminar “Skaitliskie izaicinājumi pirmo principu aprēķinos” (Numerical challenges in first principles calculations) at the Center for Quantum Computing Science (Faculty of Computing, LU) on February 24, 2022.

• Outreach and popularisation of science
  • Organisational duties during the preparation for the National Physics Olympiad,
  • Questions and Answers session at the Young Physicists School (Jauno Fiziķu Skola) event on March 19, 2022. (https://youtu.be/v5uBT_QyJWA).

Updated on 06.07.2022

• Seminars and lectures
  • Hylleraas seminar “High precision from augmented plane waves” at The Arctic University of Tromsø via Zoom on April 22, 2022.
• Secondments, visits
  • Collaboration visit at The Arctic University of Tromsø on May 17-20, 2022
• Knowledge transfer, teaching duties
  • Chairing the bachelor defense committee at the Physics Department, LU on June 8-10, 2022
  • Bachelor thesis defended by Kristians Kacars supervised by Andris Gulans, June 10, 2022. Thesis title: "Puasona vienādojuma risināšana elektronu struktūras simulācijās ar dažādu periodiskumu" (Poisson equation solvers in electronic-structure simulations with different periodicities). All obtained data are freely accessible in an open-access repository.

Updated on 13.10.2022

• Secondments, visits
  • Secondment at the University of Basel on August 17 - September 2, 2022,
  • Collaboration visit at the Technical University of Vienna on September 14-16, 2022, seminar talk.
• Conferences, workshops
  • NOMAD Industry Workshop on June 24-25, 2022,
  • the PSI-K 2022 conference on August 22-25, 2022, contributed talk Benchmarking pseudopotentials for all-electron precision in chemically diverse materials (abstract book),
  • HPC for Research and Development on September 30 - October 2, 2022, invited talk Towards exascale in computational materials science.
• Outreach
  • Participation at the European Researchers' Night on September 30, 2022.

Updated on 11.04.2023

• Seminars
  • (Quantum) machine-learning seminar organised by the Center for Quantum Computer Science, Faculty of Computing, University of Latvia, March 21, 2023.
• Conferences, workshops
  • The CECAM workshop Challenges and Advances in Solving Eigenproblems for Electronic-Structure Theory, November 15-18, 2022, invited talk Iterative eigensolver for augmented plane waves revisited (abstract),
  • The DPG Spring Meeting of the Condensed Matter Section (SKM) on March 26 - March 31, 2023, contributed talk Performance of pseudopotentials in chemically diverse materials (abstract).
• Datasets consisting of 683 molecular and 65 atomic species are published at an open data repository. These datasets contain inputs and outputs of PBE calculations performed with the electronic-structure code exciting. The basis settings are chosen to guarantee the precision level of at least 0.1 kcal/mol.
  • A. Gulans, S. Goedecker, "All-electron spin-restricted atomic reference",
  • A. Gulans, S. Goedecker, "All-electron reference for a chemically diverse molecule set".
• Knowledge transfer, teaching duties
  • Course Numerical Methods (Bachelor of Physics programme),
  • Course High-Performance Computing (Master of Physics programme).

Updated on 30.11.2023

• Conferences, workshops
  • Numerical Methods in Quantum Chemistry 2023, UiT - The Arctic University of Norway, Tromsø, June 5-8 2023. An invited talk (abstract).
  • HoW exciting! 2023, Humboldt-Universität zu Berlin, Berlin, August 2-10 2023. A talk and organisation duties.
  • Towards exascale solutions in Green function methods and advanced DFT, October 3-8 (2023), Paphos, Cyprus. Organization duties and tutorial sessions.
• The molecule sets were updated due to issues with the self-consistency converging to different solutions. The dataset initially consisted of 683 molecules, whereas now it is reduced 667. The cell shape and molecular geometry were symmetrized for a majority of molecules with transition-metal atoms. Data sets are published in the NOMAD repository. This time data are available also for two further exchange/correlation functionals. 
  • All-electron reference for a chemically diverse molecule set (PBE, version 0.2),
  • All-electron reference for a chemically diverse molecule set (PBEsol, version 0.2),
  • All-electron reference for a chemically diverse molecule set (LDA, version 0.2).
• Publications with a funding acknowledgment

  • M. Azizi, J. Wilhelm, D. Golze, M. Giantomassi, R. L. Panades-Barrueta, F. A. Delesma, A. Buccheri, A. Gulans, P. Rinke, C. Draxl, and X. Gonze “Time-frequency component of the GreenX library: minimax grids for efficient RPA and GW calculations,” Journal of Open Source Software 8, 5570 (2023) 

  • Vikram Gavini et al., “Roadmap on electronic structure codes in the exascale era,” Modelling Simul. Mater. Sci. Eng. 31, 063301 (2023) 

  • C. Vona, S. Lubeck, H. Kleine, A. Gulans, C. Draxl, Accurate and efficient treatment of spin-orbit coupling via second variation employing local orbitals, arXiv:2306.02965, accepted in Physical Review B

  • A. Brakestad, S. R. Jensen, C. Tantardini, Q. Pitteloud, P. Wind, J. Uzulis, A. Gulans, K. H. Hopmann, and Luca Frediani, Scalar relativistic effects with Multiwavelets: Implementation and benchmark, arXiv: 2310.02149, submitted to The Journal of Chemical Theory and Computation

• Knowledge transfer, teaching duties
  • Quantum Mechanics (Bachelor of Physics programme).