AI to make quantum physics more accessible and sustainable

When researchers today search for new materials for green energy or medicine, it no longer happens solely with flasks and microscopes. Increasingly, it takes place on supercomputers that simulate the behaviour of atoms and electrons. But the cost is high—not just financially, but also in terms of electricity use and wasted computing time.

That is what Line Jelver, who has just taken up a position as assistant professor at the Centre for Polariton-driven Light-Matter Interactions (POLIMA) at SDU, wants to change. With a grant of DKK 13.3 million from the Novo Nordisk Foundation’s Data Science Investigator – Emerging 2025 programme, she will lead the project ADAM.

The goal is to democratise access to advanced quantum mechanics by letting artificial intelligence assist researchers in managing heavy computational workloads.

Hidden resource consumption

At the core of Line Jelver’s research is so-called Density Functional Theory (DFT)—a method used to describe material properties all the way down to the atomic level. It is a discipline that typically requires many years of experience and a deep understanding of complex numerical physics.

- It is extremely difficult to become good at this type of calculation, and the learning curve is very steep. Often, up to 90 per cent of computer resources are spent systematically tuning and testing the right parameters before the actual calculation even begins, explains Line Jelver.

This is where the waste occurs. Although parameter tuning is a necessary and systematic part of the work, it requires numerous repeated calculations, resulting in vast amounts of energy consumption on supercomputers. When this process is carried out thousands of times worldwide, the costs quickly add up. Jelver’s project estimates that the technology could save up to 88 million CPU hours annually.

- If we can use language models to guess the right starting parameters, we can save enormous amounts of resources. It’s about making the process more efficient, so that computing power is spent on results rather than preparation, she says.

AI as autopilot

Line Jelver’s solution is not merely a chatbot one can consult for advice. It involves so-called agentic language models—a form of AI that can act autonomously. The system is designed to function as an intelligent assistant that controls the supercomputer, corrects errors as they occur, and guides the researcher through complex setups.

Drawing on extensive materials databases and existing research, the system can generalise from a far broader knowledge base than any single researcher could ever build.

This will enable many more researchers to conduct calculations that are currently reserved for a small elite of experts, while maintaining the quality of the results through built-in guidelines and continuous validation of the system’s proposed choices.

Specifically, the technology will be used to study the optical properties of materials—for example, how light interacts with matter. This has direct applications in the development of next-generation solar cells, where the hunt is on for materials that absorb sunlight more efficiently, as well as in biological sensors.

- The grant from the Novo Nordisk Foundation comes at exactly the right time, says Horst-Günter Rubahn, Head of the Mads Clausen Institute.

- At the institute, we are strongly committed to developing radically new solar cells, and in recent years we have focused intensively on advanced materials technologies. In particular, we need advanced theoretical methods of the kind that Line will apply in her project. I am therefore extremely pleased that the Novo Nordisk Foundation has decided to provide such generous support to one of our young top researchers, he concludes.

He is echoed by Professor N. Asger Mortensen, Head of the POLIMA centre funded by the Danish National Research Foundation:

- At the POLIMA Centre of Excellence, we are also delighted to see one of our young research group leaders now receiving additional funding to pursue her own independent research activities.

The dream of the virtual laboratory

In the long term, Line Jelver sees a perspective that goes beyond simply speeding up calculations. She envisions a future in which artificial intelligence becomes a research partner capable of predicting answers without first having to compute them.

- The dream is that language models become so good that they can predict the properties of a material without us having to perform the heavy calculations. That you can say, ‘I need a molecule that absorbs light at precisely this wavelength,’ and the model produces a suggestion that you can then go and test in the laboratory, says Line Jelver.

The grant also marks a personal milestone for her, as it enables her to establish her own research group at the Mads Clausen Institute.