SLAC researchers help organize community challenge to benchmark molecular simulations with experiments

Illustration of a prediction challenge, in which “blind” simulations of a light-activated reaction were compared to experimental data. (Top) 15 different theory groups predicted the experimental signatures of a photochemical reaction and published their results in separate papers. (Center, left to right) The reaction under investigation was the breakdown of cyclobutanone after it was hit by a flash of light. (Bottom) In an ultrafast electron diffraction experiment, in which highly energetic electrons scatter off a sample, producing a scattering pattern like the one shown here, SLAC researchers were then able to determine which simulation approaches worked best. @ Greg Stewart/SLAC National Accelerator Laboratory

Certain molecules rearrange their structure when exposed to light, which makes them useful in energy applications, pharmaceuticals and more. Yet, researchers struggle to predict exactly how light transforms these molecules, and that makes it difficult to design better ones.

Now, a chemistry community challenge aims to figure out how to make better predictions about these reactions. First, theorists predicted how the structure of one such light-activated molecule, cyclobutanone, changes after light exposure. Then, a research team at the Department of Energy’s SLAC National Accelerator Laboratory was entrusted with collecting the data that was used to evaluate the 15 submitted simulations. “There are a lot of different simulation methods that make different approximations, and they all provide slightly different results,” said Thomas Wolf, head of the Chemicals Sciences Department at SLAC’s Linac Coherent Light Source (LCLS) and principal investigator at the Stanford PULSE Institute and of the experimental study. “The excitement is that this data can be used to unambiguously evaluate simulations.”

The researchers used an instrument for ultrafast electron diffraction (MeV-UED) at LCLS to resolve the split-second transformation of cyclobutanone with atomic-scale resolution, which they reported in a special issue of The Journal of Chemical Physics alongside the predictions and discussed further at a recent workshop. Todd Martinez, professor of photon science at SLAC and David Mulvane Ehrsam and Edward Curtis Franklin professor of chemistry at Stanford University, helped come up with the idea to hold a prediction challenge at a 2023 workshop hosted by CECAM (Centre Européen de Calcul Atomique et Moléculaire), a theoretical chemistry organization based in Switzerland.

“This effort highlights the strategic leadership role SLAC – and particularly Todd – has taken in bringing the research community together around a shared scientific challenge,” said Kelly Gaffney, professor of photon science and, by courtesy, of chemistry at SLAC and Stanford University, who was not involved in the challenge. “Community-wide coordination of this kind is rare, and SLAC’s approach serves as a powerful template for amplifying the impact of science enabled by the lab.”

Complex simulations face off with precise experimental data

Cyclobutanone is a relatively small molecule and potentially a good model system for studying light-activated molecules in general. “The hope is that if we understand simple species in detail, we can extrapolate this to more complicated molecules,” said Alice Green, lecturer at the University of Edinburgh, United Kingdom, former Marie-Curie fellow and lead author of the experimental study.

However, despite its relatively small size, cyclobutanone is still tricky to simulate. When hit with light, the electrons and atomic nuclei inside the molecule move around, and it’s currently not clear which calculations correctly describe the complicated motions and quantum phenomena that occur.

Because of this complexity, simulations are also often done after rather than before an experiment, with the experimental results guiding the calculations and introducing a “data bias.” The prediction challenge initiated by Martinez is unique because it motivated a larger number of theory groups to perform “unbiased” calculations without experimental input first. “With the blind prediction challenge, we get information about which methods work and which methods don’t because different people will make different choices, and every choice they make is documented,” Martinez said. He contrasts this with published simulations, which typically only mention the decisions that yielded success.

But which predictions are closest to what actually happens in the molecule on the atomic level in response to light? Here is where the SLAC experiment comes in. Researchers there exposed gas molecules of cyclobutanone to a pulse of light, then shot electrons from the MeV-UED instrument at them. The pattern created by the electrons scattering off cyclobutanone revealed how the electrons and atomic nuclei inside the molecule were moving – data that can be used to narrow down the simulations submitted to the challenge. Comparing theory and experiment leads to new insights

At a second CECAM workshop earlier this year, experimentalists and theorists finally met to compare the data and simulations. While they found that most of the simulations were able to predict general features of the reaction, such as what products formed, some predictions agreed with the measured timescale of the reaction better than others. Green said this is likely due to differences in calculating the reaction’s energy barrier: “The formation timescale of products, which is directly accessed in the experiment, is a very sensitive measure of the energetic barrier that the reaction must overcome to occur, and so can be used to benchmark the different quantum chemistry methods for calculating this barrier.”

Additionally, most of the predictions of the experimental results only included signatures from the motions of cyclobutanone’s atomic nuclei, which is usually a good approximation for other types of reactions, and missed properly accounting for the signatures from electron motions. However, the comparison with actual data suggests that simulations of light-activated reactions need to predict signatures from electron motions as well. Future rounds of prediction challenges and data collection could further refine these simulations. Experimental techniques beyond ultrafast electron diffraction, such as time-resolved X-ray spectroscopy at LCLS’s superconducting X-ray free-electron laser, will be even more sensitive to the motions of electrons in cyclobutanone, which will bring the chemistry community closer to understanding light-activated reactions.

“The hope is that combining the insights from different experimental techniques can help build up a more complete understanding of the molecule’s light-activated dynamics,” Green said. Reference Imaging the photochemistry of cyclobutanone using ultrafast electron diffraction: Experimental results A. E. Green, Y. Liu, F. Allum, M. Graßl, P. Lenzen, M. N. R. Ashfold, S. Bhattacharyya, X. Cheng, M. Centurion, S. W. Crane, R. Forbes, N. A. Goff, L. Huang, B. Kaufman, M.-F. Kling, P. L. Kramer, H. V. S. Lam, K. A. Larsen, R. Lemons, M.-F. Lin, A. J. Orr-Ewing, D. Rolles, A. Rudenko, S. K. Saha, J. Searles, X. Shen, S. Weathersby, P. M. Weber, H. Zhao, T. J. A. Wolf https://pubs.aip.org/aip/jcp/article-abstract/162/18/184303/3346674/Imaging-the-photochemistry-of-cyclobutanone-using SLAC National Accelerator Laboratory