Psi-k 2025 interviews

These are some of the questions that we asked to a group of top experts in theory and simulation of materials — some MARVEL members, some members of the wider global community — when we met them at the Psi-k conference that was held in Lausanne from 25 to 28 August 2025. 

Giulia Galli is the Liew Family Professor of Electronic Structure and Simulations in the Pritzker School of Molecular Engineering and the Department of Chemistry at the University of Chicago. She also holds a senior scientist position at Argonne National Laboratory, where she is a group leader and the director of the Midwest Integrated Center for Computational Materials. She is an expert in the development of theoretical and computational methods to predict and engineer material and molecular properties from first principles. Her research focuses on problems relevant to the development of sustainable energy sources and quantum technologies. She is a member of the NCCR MARVEL Scientific Advisory Board.

In the interview she reflects on how the progress of computational materials science over the last decades has exceeded her expectations. "The field has made a lot of progress towards open science, and this has pushed the community very far in terms of applying the calculations to realistic systems. And there has been a lot of development of new methods". But she also notes how the collaboration with experimentalists and with the industry can be improved, and she warns AI and machine learning should be applied to materials simulation with caution. "We need metrics to understand if we really remain predictive when we apply AI methods". 

Georg Kresse heads the Computational Materials Physics group at the University of Vienna. In addition to his research contributions, he is known in the computational materials science community for developing the Vienna ab initio simulation package (VASP), which is the leading code for first-principles calculations of solids and liquids and is used by 4,000 research groups worldwide. His group’s current is on modern quantum field theory and quantum chemistry methods, which allow more accurate predictions than density functional theory. 

Here, he takes stock of how machine learning is transforming the field and looks at the challenges that lie ahead for first-principle simulations, starting with accuracy. "We need to shift to more accurate methods that actually approximate the Schrödinger equation". He also warns young scientists who are entering the field of the perils of hyper-specialization. "Materials chemistry is one of the broadest fields that exist. You need thermodynamics, quantum mechanics, Maxwell equations. You need all this background information that you've learned in physics and you can forget very quickly when you start your PhD. So, broaden your knowledge". 

In his interview, the Psi-k 2025 conference chair Peter Haynes notes how the community has made impressive progress in terms of reliability and interoperability of the software tools available for simulating materials. But simulation methods – even those based on density functional theory, that owes its success to a good trade-off between accuracy and computational load  – remain computationally expensive. “We’re only scratching the surface” of what we can achieve using artificial intelligence to improve that trade off, says Haynes, who also notes that collaboration between academia and industry on computational materials is still a work in progress.

Peter Haynes is Professor of Theory and Simulation of Materials at Imperial College London. His research interests focus on the development of new methods for performing first-principles quantum-mechanical simulations and their application to materials science, nanotechnology and biological systems.

Applications of quantum computing and quantum machine learning to materials are hot areas of research surrounded by great expectations. Zoë Holmes, a group leader in the quantum simulation project in MARVEL, explains that whereas quantum hardware has seen impressive advancements over the last decade, software still needs to catch up. “We have a whole zoo of algorithms now, but we don't have that many end to end analysis which use different algorithmic primitives showing that things can work”, she says. As for quantum machine learning, in principle it has a great potential for studying materials, but so far most results have happened in niche areas. Her top tip for young scientists is to strive for originality, and try to introduce new algorithms rather than tweaking what others have done.

Zoë Holmes received in 2015 her MPhil degree in Physics and Philosophy from the University of Oxford. In 2016 she obtained her MRes (Master of Research) from the Imperial College London, where in 2019 she got her PhD in quantum thermodynamics. In 2020 she started as a Postdoctoral Researcher at Los Alamos National Laboratory (USA) working on quantum algorithms and quantum machine learning methods for Noisy Intermediate-Scale Quantum (NISQ) computers. In 2021 she became the Mark Kac Fellow at Los Alamos National Lab. Since August 2022 she is Tenure Track Assistant Professor of Physics at EPFL.

Michele Ceriotti calls the impact of machine learning on the field “revolutionary”, and says that machine learning not only speeds up calculations, but can actually bring improved understanding of quantum systems by allowing scientists to manipulate parts of their models to test hypothesis. One big challenge for the future, he says, is that the community has learned different flavours of electronic structure approximations that work for transition metals or transition metal oxides or or for reactions in the gas phase, but a single framework that is practical and can be applied across the board is still lacking. and without these reference data, something that limits the usefulness of universal machine learning models. As for the impact of MARVEL, Ceriotti says it mainly lies in the software infrastructure. “Even though it will not be trivial to keep maintaining these software now that MARVEL comes to an end, I think that it has definitely created a huge amount of value” he says.

Michele Ceriotti has been a professor at the department of Materials Science at EPFL since autumn 2013, establishing the Laboratory of Computational Science and Modeling (COSMO). Within MARVEL phase I, he was a group leader in Horizontal Project 4. Then in phase II, he has been the project leader of Design & Discovery Project 1 and also a group leader in Design & Discovery Project 2 and Incubator Project 2. Within MARVEL phase III, since May 2022, he is project leader of Pillar 2, Machine Learning Platform for Molecules and Materials. He is also one of the deputy directors of the NCCR MARVEL and member of the Executive Committee.

Michele Kotiuga is a great example of a MARVEL junior member who went on to build a career in the industry. Formerly in the THEOS group at EPFL, where we worked with Nicola Marzari and Giovanni Pizzi, she is now Support and Application Scientist at Materials Design, a company that provides software and services for atomistic simulations of materials to industries such as energy, transportation, chemistry and microelectronics.

In the interview Michele Kotiuga reflects on how her time with MARVEL prepared her for a career in the private sector, and reflects on to create bridges between academic research and industrial applications.

Shobhana Narasimhan is a professor of theoretical sciences at the Jawaharlal Nehru Centre for Advanced Scientific Research in Bengaluru, India and a member of the SNSF's MARVEL Review Panel.

Speaking to the MARVEL team during the 2025 Psi-k conference in Lausanne, Shobhana Narasimhan comments on the challenges of integrating machine learning in materials prediction without sacrificing actual understanding of physical systems, of improving the interface between industry and academia, and how the computational materials science community should continue its efforts towards more diversity.

Giovanni Pizzi is the group leader of the Materials Software and Data Group in the Laboratory for Materials Simulation at the Paul Scherrer Institute in Villigen, Switzerland. In the third phase of NCCR MARVEL, he has been a project leader for the pillar on Digital Infrastructure of Open Simulations and Data.

In this interview, Giovanni Pizzi talks about the importance of the standardization and verification of computer codes in computational materials science. This is more important than ever in the era of machine learning, because the investment on building and training new machine learning models only pays off if the datasets used for training are accurate. He also talks about a crucial legacy of the NCCR MARVEL: the combination of new simulation methods and open-source tools that allow to handle complex computational workflows.

Virginie de Mestral is a doctoral student in Mathieu Luisier’s group at ETH Zurich, and a member of NCCR MARVEL. During her PhD, funded by Innosuisse, she has focussed  on a collaborative project with a Swiss startup to simulate promising materials for applied optoelectronics.

In this interview Virginie de Mestral reflects on how participating in MARVEL as a junior member has opened many doors for her, and on the lessons she’s learned doing research at the interface of academia and industry. She has some key recommendations for other early-career scientists in materials simulation: stay curious, be responsible, and always ask that extra question.

Xingao Gong is a professor of physics at Fudan University in Shanghai, China, where he leads the Institute for Computational Physical Sciences. He is a member of the Chinese Academy of Sciences and a fellow of the American Physical Society.  His research has focussed on explorations of multiferroic materials, magnetic phase transitions, and dynamic simulations of ion and electron behaviors in electrolytes, and has included the development of machine learning potentials for materials simulation.

In his interview, he comments on how AI is changing computational materials science “even too dramatically”, and lays out his expectation that within a decade machine learning tools will make it possible to simulate realistic physical systems like electronic devices, including their manufacturing process. He also describes the current AI boom in China and the strong backing it receives from the government.

Finally, we hear from Nicola Marzari, Director of NCCR MARVEL, Professor and Chair of Theory and Simulation of Materials at the École polytechnique fédérale de Lausanne (EPFL), and head of the Laboratory for Materials Simulations at the Paul Scherrer Institute (PSI), and the next Cavendish Professor of Physics at the University of Cambridge, starting in 2026.

In this interview, Nicola Marzari notes that a huge change for the field of computational materials science in the last 10 years has been the interest – and the investments – of major tech companies, now aware that materials and devices can “make or break their future”. He describes the important challenge of increasing the capacity for experiments and automate them to some extent, and more fundamentally of creating computational tools with better predictive accuracy. Looking at the legacy of NCCR MARVEL, that is drawing to a close in 2026, he praises the SNSF funding agency for supporting not only science, but also outreach, technology transfer and equal opportunities.