Theory Department - FHI der MPG

👋Today we welcome Sebastian Rejman, starting as a #Postdoc in Christoph´s group working on Automation for Experimental #Catalysis Research. Sebastian did his #PhD at @utrechtuniversity.bsky.social and will defend his thesis 🔜!
When not at work, he enjoys powerlifting🏋️& playing guitar🎸!

👏 Hands up for @zausibergmann.bsky.social, Nico and collaborators from the THEOS lab at @materials-epfl.bsky.social, whose paper is NOW OUT on @physrevlett.bsky.social by @apsphysics.bsky.social!

📝Paper Digest: bsky.app/profile/thde...
🔗Paper Link: journals.aps.org/prl/abstract...

👋Today we welcome Alexander von Rueden, who starts as a new #Postdoc in Vanessa´s group, working on "Atomic modeling of electrocatalysts". Alex obtained his #PhD in #ChemE from the University of Wisconsin-Madison and did research at PNNL too.
When not at work, he enjoys playing jazz piano🎹! 👋

3️⃣ Why it matters?

Leads to more accurate predictions of #surfacechemistry—key for better #catalysts, corrosion resistance, and materials design.

2️⃣ What is new?

The authors use AI-powered simulations to explore many possible surface structures, not just a few ideal ones. This gives a more realistic view of how surfaces behave during oxidation.

1️⃣ What is the paper about?

Modeling how Cu(111) surfaces oxidize, capturing the disorder and complexity that typical methods miss.

Out of the Crystalline Comfort Zone: Sampling the Initial Oxide Formation at Cu(111) Oxidizing transition metal surfaces are generally characterized by an increasing heterogeneity at simultaneous lowering of crystalline order. This complexity eludes present-day first-principles…

👏 Congrats to Felix, @zausibergmann.bsky.social, and Hendrik, whose paper "Out of the Crystalline Comfort Zone: Sampling the Initial Oxide Formation at Cu(111)" made it into Advanced Science!

🔗Link to preprint on #chemrxiv: chemrxiv.org/engage/chemr...
📝Paper digest 👇

#compchem #machinelearning

3️⃣ What are the implications for the future?
#AI can speed up the discovery of advanced photoswitches with custom properties. It opens the door to faster and smarter molecular design in materials science, with potential impacts in fields like smart materials, sensors, and molecular computing.

2️⃣ What is new?
- A trained recurrent #neuralnetwork (RNN) to deisgn light-sensitive molecules;
- Novel, diverse, and high-performing molecules;
- Some of the AI-designed molecules were synthesized and succesfully tested.

1️⃣ What is the paper about?
It presents a generative #machinelearning model designed to create new spiropyran photoswitches — molecules that change structure under light. The goal is to improve their switching speed and addressability - how clearly the two forms can be controlled using light.

👏 Congrats to Robert and Hannes, whose paper "Machine learning driven design of spiropyran photoswitches" is now online on Digital Discovery by @rsc.org!

🔗 Link to paper: pubs.rsc.org/en/content/a...
📝 Paper digest 👇

3️⃣What are the implications for the future?
It shows that machine learning works much better when it follows the rules of #physics. This could revolutionize how we design materials for everything from batteries to medicines — with AI that respects the rules of physics.

2️⃣What is new?
- Tools to better measure how well machine learning predicts NMR data;
- A large, realistic dataset to train and test models;
- Full shape and symmetry of the data gives much better results than treating each part separately — especially for predicting direction.

1️⃣What is the paper about?
It explores how to make #machinelearning better at predicting complex physical properties used in #NMR spectroscopy by comparing two methods: one that understands the shape and symmetry of the data, and one that doesn’t.

Performance metrics for tensorial learning: prediction of Li4Ti5O12 nuclear magnetic resonance observables at experimental accuracy Machine learning (ML) surrogate modeling is a powerful approach to reduce the computational cost of first-principles calculations. While well established for the prediction of scalar observables like ...

👏Congrats to Ange, Simone and Christoph, whose paper "Performance Metrics for Tensorial Learning: Prediction of Li4Ti5O12 Nuclear Magnetic Resonance Observables at Experimental Accuracy" made it into the J Mater Chem A by @rsc.org!
Preprint on #chemrxiv: doi.org/10.26434/che...
Paper digest 👇

Pushing charge equilibration-based machine learning potentials to their limits - npj Computational Materials npj Computational Materials - Pushing charge equilibration-based machine learning potentials to their limits

💫 Massive news to start your week – don’t miss this! 💫

The paper from @mvondrak.bsky.social and Hannes just got published on npj computational materials by @nature.com.

📝Paper Digest: bsky.app/profile/thde...
🔗Paper Link: www.nature.com/articles/s41...

🥳 It´s #PAW2025 part two!

Postdocs are more than their research—habits, routines and personal rituals keep their engines running. Check out below 👇 what are some of the daily non-negotiables of our #Postdocs. Happy #PAW2025!

Share with us what are the little things that power your day! ✅

Postdoc Appreciation Week Germany – PAWde Postdocs become the center and focus! It is a special week to highlight, recognise and celebrate all their contributions to research and academic life!

2️⃣ Held each year in the third week of September, #PAW features seminars, workshops, socials, and more—a chance to say thank you and to celebrate the people driving research forward. 💡🌍

Check out their fantastic program for this week paw-germany.de!

#PostdocAppreciationWeek #PAW2025 #PostdocLife

1️⃣ #PAW is a global celebration of #Postdocs and their vital contributions to #science and #innovation.
First launched in 2009 by the National Postdoctoral Association (NPA), #PAW has grown into an international event with institutions across the US, UK, Canada, Ireland, Germany, and beyond.

🥳 It´s #PostdocAppreciationWeek!

Postdocs are more than their research—habits, routines and personal rituals keep their engines running. Check out below 👇 what are some of the daily non-negotiables of our #Postdocs. Happy #PAW2025!

Share with us what are the little things that power your day! ✅

👋 Today we welcome Bowen Cheng, new #PhDstudent in Hendrik´s group working on "High Throughput Enabled Optimization of #machinelearning for Property Targeted Spiropyran Design".
Bowen obtained his MChem degree from the Trinity College @ox.ac.uk and when not at work he enjoys playing 🎸 and cooking! 👋

3️⃣What are the implications for the future?
This study highlights the need for new ML frameworks that can handle electrostatics and charge transfer accurately, key for simulating complex systems in the future.

2️⃣What is new?
While promising, ML-based charge equilibration models still face issues like spurious charge transfers and overpolarization, even with enhancements—highlighting gaps in current methods.

1️⃣What is this paper about?
The paper explores #machinelearning (ML) potentials with charge equilibration (kQEq) models for atomistic simulations, focusing on how well they handle charge distribution and long-range electrostatic interactions in molecular systems.
#compchem #chemsky #physicssky

Pushing Charge Equilibration Based Machine Learning Potentials to their Limits Machine learning (ML) has demonstrated its potential in atomistic simulations to bridge the gap between accurate first-principles methods and computationally efficient empirical potentials. This is ac...

👏Congrats @mvondrak.bsky.social and Hannes, whose paper "Pushing Charge Equilibration Based Machine Learning Potentials to their Limits" made it into npj computational materials by @nature.com!
While we wait for it to be online, check the #preprint on @chemrxiv.org or read the digest 👇

3️⃣What are the implications for the future?
This could speed up simulations for #batteries, fuel cells & #catalysts. It helps connect lab conditions (like pH or voltage) to atomic behavior—paving the way for smarter material design.

🔋🔬 Bridging experiments & atomistic models.

2️⃣What´s new?
The model is bias-aware: it includes electric bias directly, using local atomic descriptors and Born charges. This lets it capture realistic voltage effects—something previous #machinelearning potentials missed.

⚡️ More accurate, more physical.

1️⃣What is the paper about?
A new #machinelearning model predicts how charging metal surfaces in liquids changes their energy—key for understanding electrochemical reactions. It links atomic-scale changes to applied voltage, using Cu(100)/OH as a case study.

#compchem #chemsky #physicssky

Machine Learning the Energetics of Electrified Solid/Liquid Interfaces We present a response-augmented machine learning (ML) approach to the energetics of electrified metal surfaces. We leverage local descriptors to learn the work function as the first-order energy chang...

👏Congrats to Zausi, Nico and our collaborators from @icepfl.bsky.social, whose paper "Machine Learning the Energetics of Electrified Solid/Liquid Interfaces" made it into @physrevlett.bsky.social by @apsphysics.bsky.social.
While it´s coming online, check out the #preprint on #arXiv and the digest👇

🙏 We would like to say a big thank you to our fundings from @dfg.de Cluster of Excellence @e-conversion.bsky.social EXC 2089/1 and Max Planck Computing and
Data Facility (MPCDF) by @maxplanck.de for providing computing time.