Round up of 2022. What is 2023 bringing?

It is time for drawing up conclusions for the year and here is my brief overview of our group’s third year. 2022 will come down in history as a tumultuous year with major global events which couldn’t not affect me as a Ukrainian. I should, however, not complain while sitting on a sunny balcony in Xiamen just after recovering from Covid caught after sudden relaxation of the three-year-long zero-Covid policy. Next year is my year of a rabbit and it will be good to spend it in the origin country of the lunar calendar, although I will finally visit other countries for conferences and private affairs, hopefully, also Ukraine (see my already known schedule).

Looking at the start of 2022, I was coordinating the final stages of multi-author drafting of the book ‘Quantum chemistry in the age of machine learning’ which got finally published in September. That was the biggest project I overtook, it had 65+ people involved and lasted for more than two years! At the end, I am happy that the final book included everything essential to dive into the wonderful world of machine learning and quantum chemistry with practical hands-on tutorials and codes for teaching. The book is not a loose collection of chapters, it is a real textbook! My group members were involved very heavily throughout, all chapters were read and exercises were done by at least one-two of group members and many chapters were co-written by Bao-Xin (graduated), Fuchun, Yi-Fan, Lina, and Arif.

Finalizing a book comes with a downside that it heavily bites into time for research. Nevertheless, the group did well not just in staying afloat during this period but actually making great progress in a number of directions. I am particularly happy that we introduced several new ways of performing dynamics simulations by leveraging the power of AI. We demonstrated that quantum dynamics trajectories can be accurately and efficiently simulated as a function of time without the need of recursive propagation up to asymptotic limit at infinite time. This AI-QD work was done by Arif on an example of exciton energy transfer in light-harvesting complex and published in Nature Communications. Related works by us showed that further improvements and extensions are feasible, i.e., we could predict the entire trajectories at once (Arif’s OSTL approach) and predict molecular dynamics as a function of time by building a molecular model in 4D-spacetime (mainly done by Fuchun and with contributions from a growing list of group members, Lina, Yifan, Arif, Xinxin, …). The last 4D approach still can only be found in a preprint, since then we made lots of improvements to our AI models and hope that 2023 will see a paper published.

While above research is free exploratory research, our groups also done many important practical developments for making fast and accuracy quantum chemistry more accessible with AI. A huge development is that now basically all our implementations are not just available open source and free but that these calculations can be run online via a webbrowser. This has becomes possible by bringing our MLatom program to the Xiamen Atomistic Computing Suite (XACS) cloud ecosystem founded this year. XACS also contains tons of other features such as valence bond theory and energy decomposition analyses in programs mainly developed by groups of our colleagues Wei Wu and Peifeng Su in Xiamen. We are also happy that XACS has many internationally well-recognized partners and, e.g., Mario Barbatti’s group is using MLatom for nonadiabatic dynamics with ML. Fuchun from our group is very helpful in maintaining MLatom@XACS version. MLatom itself has undergone transition to a package focusing on simulating molecular properties with methods like AIQM1 (whose main developer is Peikun) and we increasingly use MLatom for materials design. Next year will see more improvements of MLatom@XACS, with details on molecular dynamics (Yifan’s work), 4D models, two-photon absorption cross section simulations (preprintavailable in collaboration with Cheng Wang in Xiamen University), interface to Arif’s MLQD program, and many more coming up.

This year, we have been developing many other new methods such as KREG (preprint is available), but also use them and explore their capabilities and weaknesses as well as openly release data. For example, we showed that AIQM1 can produce very fast and very accurate heats of formation while also using its built-in uncertainty quantification to detect errors either in AIQM1 or experiment (work in JPCL by Peikun and Wudi in collaboration with Olexandr Isayev). Wudi is finalizing his simulations of some very big systems (up to thousands of atoms) with AIQM1, including MD with one work already submitted. In collaboration with Alexei Kananenka, we with Arif explored the limitations of various ML methods for recursive quantum dynamics and gave recommendations, see our study published in MLST. Several new big data sets were made available: VIB5 of molecular PESs including data beyond coupled-cluster (by Lina and Shuang in collaboration with Alex Owens), WS22 of molecular PES with wider distribution of energies than MD17 (at preprint stage, collaboration led by Max Pinheiro Jr in France), and one more data set to be submitted in 2023. The only non-machine learning study was done by Lina, to be resubmitted in 2023.

2022 also saw the graduation of Bao-Xin who is now happily employed elsewhere. New members joined in September: Yaohuan is our research assistant who actually does research in addition to all the essential administrative support, Xinxin – new master student who really started very well, and Yeyun – a co-supervised PhD student with the AI institute of our University who will do lots of multi-disciplinary investigations.

As being part of the University, it is our responsibility to share our knowledge. Internationally, our group participated in the CECAM school, where I, Fuchun, Yifan, and Shuang were making lectures and tutorials on how to use ML for computational chemistry and dynamics. Domestically, we have held our first XACS workshop in hybrid online/offline format and introduced the use of MLatom on XACS cloud platform. We also visited our only in-person conference this year with many group members and their posters (Peikun about AIQM1, Fuchun about MLatom, Lina about VIB5, Arif about ML for quantum dynamics, Yifan about IR, Wudi and Xinxin about their unpublished studies). At the university level, we were teaching machine learning in (computational) chemistry where some of my group members (Shuang, Fuchun, Yifan) helped along. 2023 hopefully will allow us to be more active on social media and also finally organize our long-planned SMLQC-biweekly online talks and I would love to visit SMLQC-2023 symposium in Uppsala organized by Roland Lindh.

On a journal level, I have served as one of the invited guest editors of special issues in JCP (about semi-empirical methods) and PCCP (about insightful ML), and next year we are looking at rounding up with editorials all the beautiful contributions. Related to this, I am happy to announce that I have joined the new open journal Artificial Intelligence Chemistry as an Associate Editor and we are looking forward to receiving new submissions in 2023!

It was not all work though, we have had many wholesome moments in hiking and other group activities (hopefully, no videos of us in karaoke will go public though), see some photos.

That’s a wrap, next year will bring more news from our group!