Speaker
Aron Walsh, Professor, Thomas Young Centre for Theory and Simulation of Materials, Imperial College London, UK
Date
Wednesday February 16th, 10am (USA/Pacific)
Abstract
Density functional theory (DFT) has become a powerful tool for modelling the structures and properties of materials. The first role of a DFT practitioner is to choose an exchange-correlation functional that is appropriate for the chemical system and properties of interest. While there are still issues for describing materials that can be classified as highly correlated, such as those containing partially occupied d or f bands, generally DFT can provide an adequate description for the majority of systems. Indeed, there are a growing number of structure-property databases, including the Materials Project, that conveniently collect large numbers of DFT calculations. Such databases are now being widely used for training statistical machine learning models to reproduce the structure-property correlations of known crystals and also to extrapolate to new chemistries. There is a danger however of forgetting that even if an exact functional was known, the static predictions of DFT calculations for perfect structures do not represent the ground truth. Real crystals are excited and imperfect: the atoms vibrate around their ideal lattice sites; they contain point and extended defects; high energy configurations are accessed by temperature and light. In the context of my own research on semiconducting materials for clean energy applications, I will outline some challenges and possible directions for more accurate materials property predictions.
Recording
A recording of this seminar is available at this link.
Questions
If you are unable to ask questions live, please feel welcome to ask any questions following the talk here and we will ask the speaker to check afterwards. Whether they will be able to answer questions or not depends on the speaker’s availability.
Questions answered live
| Question | Asker Name |
|---|---|
| Can one predict, based on atomic number alone, which elements have chemically active versus inactive lone pairs? | Gabriel Man |
| There have been some very impressive advances in using ML to fit interatomic potentials or solve the many-body SE. These models, however, tend to be black-box, which – based on your talk today – isn’t always desired for an intuition for materials design. What are your perspectives / are you optimistic that simpler theories for materials design will be discovered, e.g., in the next decade? | Robert Wexler |
| can we use “synthetic data” to fill the missing pieces in data ? | Baha Khalil |
| Given a standard halide perovskite (e.g. MAPbBr3), what are the smallest and largest terms in the Hamiltonian? How large is the anharmonic term relative to the static crystal potential term? | Gabriel Man |
| Aron, thank you very much for a very thought provoking talk. With regard to the group theoretical analysis of the octahedra distortions in perovskites, you mentioned that these distortions are of Jahn-Teller or crystal field splitting type. It is known that these types of distortions are not reliably reproduced at the LDA/GGA level of DFT that are usually applied. Yet some of the disorted structures are represented very well. Is there a common wisdom on when GGA can or cannot be applied for these systems? | Jacob Gavartin |
| What’s your view on the machine learning’s future to provide more mechanistic/scientific details matching those from DFT, rather than prediction/engineering? | lei zhang |
| Are you aware of methods looking into the electron exictation or electron-phonon coupling effect on the energy of materials at finite temperatures? | Anonymous Attendee |
| Do you expect new unknown features (maybe a combination of known features or new ones) coming up as key parameters in predicting properties (as approximations to a reasonable range?) | Ebie Sam |
Additional questions asked
| Question | Asker Name |
|---|---|
| Molecular dynamics takes into account electron-phonon coupling naturally. How? | Namrata Jaykhedkar |
| You showed a very nice example to parametrize the perovskite distortions along 4 dimension. Do you think it is possible to parametrize much more complex motions, for example oxygen migration similar way to train machine learning force fields? | Gergely Juhasz |
| Hi… You just talked about compound semiconductors. What about ternary semiconducting compounds? Is it valid for them too? | Oday Hammadi |
| You talked about synthesizability and the burden of proof for computational material scientists. From your experience, what evidence do you need to provide to persuade a experimentalist that a previous unknown material is synthesizable? | Tian Xie |
| Is Inverse designing periodic crystals ambitious ? what are the main challenges there ? | Nawaf A |
| Currently, crystal structure prediction methods are able to predict materials at zero temperature, e.g. evolutionary crystal structure predictio or even random search approach is routinely used for materials prediction. However, there are not that many works devoted to the predictions at finite temperatures. Could you give a bit more color if you tried to do work on that direction? | Nikita Rybin |
| Do you see a value hierarchy in your list of stability criteria (convex hull, phonons, MD, etc.). If you had to pick 2 or 3 of these to be convinced something can be made, which ones would you prioritize? How much does it depend on the system of interest? | Janosh Riebesell |
| Thank you for a very nice talk. Do you have any further hypothesis for very system-dependent SOC band gaps? | Viet-Anh Ha |
| Can a post calculation correction methodology developed for addressing Van der Waals interaction will be as efficient as developing a Exchange- Co-relational functional specifically designed for incorporating Van der Waals interaction ? What is the best approach to follow for properly including Vander Waals interaction in DFT. | K M Rajashekhar Vaibhav |
| Based on your experience, which would be the best functional to model a transition metal oxide? In what terms could piezoelectric effect be studied from DFT? | Anonymous Attendee |
| Hi… You just talked about compound semiconductors. What about ternary semiconducting compounds? Is it valid for them too? | Anonymous Attendee |
| You discussed the ‘burden of proof’ and in the Q&A talked about NLP of the literature… what can experimentalists/ publishers do to improve what/ how they report data (particularly data that they might not think significant)? | Eric McClure |
From the above list, I removed duplicate questions or questions that were not directly relevant to the talk.
Finally, for everyone who is joining this forum for the first time, welcome! Thanks again for everyone who chose to join us today and who asked questions.