Konstantin S. Jakob
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Konstantin S. Jakob has a total of 92 co-authors. Here are the top co-authors by number of shared publications:
- Johannes T. Margraf
Fritz-Haber-Institut der Max-Planck-Gesellschaft, Faradayweg 4-6, 14195 Berlin, Germany
8 shared publications - Karsten Reuter
Fritz-Haber-Institut der Max-Planck-Gesellschaft , Faradayweg 4-6
6 shared publications - Aron Walsh
Department of Materials, Imperial College London, London, United Kingdom
2 shared publications - Bettina Valeska Lotsch
Max Planck Institute for Solid State Research , Heisenbergstrasse 1
2 shared publications - Jakob Blahusch
Max Planck Institute for Solid State Research, Department of Nanochemistry , Heisenbergstraße 1
2 shared publications Karsten Reuter
Fritz Haber Institute of the Max Planck Society Berlin Germany
2 shared publications
Latest Publications
- Open Access
- 11 August 2026
- Published by AIP Publishing in The Journal of Chemical Physics
- Vol. 165 (6)
Benchmarking local geometry optimization algorithms for computational materials discovery
Most current computational materials discovery workflows are dependent on efficient and reliable structure relaxations in order to predict a hypothetical material’s equilibrium structure. Besides an accurate description of the potential energy surface, this requires efficient and reliable optimization algorithms. In this study, we investigate how different optimization algorithms affect the relaxation of inorganic crystal structures using general-purpose machine-learned interatomic potentials. Our findings emphasize the critical role of optimizer selection in large-scale computational materials science workflows, particularly in the context of element-substitution-based materials discovery. This can guide the community toward choosing appropriate algorithms for the efficient and reliable prediction of new crystalline materials.
- 11 August 2026
- Published by AIP Publishing in The Journal of Chemical Physics
- Vol. 165 (6)
Open AccessBenchmarking local geometry optimization algorithms for computational materials discovery
Most current computational materials discovery workflows are dependent on efficient and reliable structure relaxations in order to predict a hypothetical material’s equilibrium structure. Besides an accurate description of the potential energy surface, this requires efficient and reliable optimization algorithms. In this study, we investigate how different optimization algorithms affect the relaxation of inorganic crystal structures using general-purpose machine-learned interatomic potentials. Our findings emphasize the critical role of optimizer selection in large-scale computational materials science workflows, particularly in the context of element-substitution-based materials discovery. This can guide the community toward choosing appropriate algorithms for the efficient and reliable prediction of new crystalline materials.
- Top Cited Papers
- 13 November 2025
- Published by AIP Publishing in The Journal of Chemical Physics
- Vol. 163 (18)
A foundation model for atomistic materials chemistry
Atomistic simulations of matter, especially those that leverage first-principles (ab initio) electronic structure theory, provide a microscopic view of the world, underpinning much of our understanding of chemistry and materials science. Over the last decade or so, machine-learned force fields have transformed atomistic modeling by enabling simulations of ab initio quality over unprecedented time and length scales. However, early machine-learning (ML) force fields have largely been limited by (i) the substantial computational and human effort required to develop and validate potentials for each particular system of interest and (ii) a general lack of transferability from one chemical system to the next. Here, we show that it is possible to create a general-purpose atomistic ML model, trained on a public dataset of moderate size, that is capable of running stable molecular dynamics for a wide range of molecules and materials. We demonstrate the power of the MACE-MP-0 model—and its qualitative and at times quantitative accuracy—on a diverse set of problems in the physical sciences, including properties of solids, liquids, gases, chemical reactions, interfaces, and even the dynamics of a small protein. The model can be applied out of the box as a starting or “foundation” model for any atomistic system of interest and, when desired, can be fine-tuned on just a handful of application-specific data points to reach ab initio accuracy. Establishing that a stable force-field model can cover almost all materials changes atomistic modeling in a fundamental way: experienced users obtain reliable results much faster, and beginners face a lower barrier to entry. Foundation models thus represent a step toward democratizing the revolution in atomic-scale modeling that has been brought about by ML force fields.
- 13 November 2025
- Published by AIP Publishing in The Journal of Chemical Physics
- Vol. 163 (18)
Top Cited PapersA foundation model for atomistic materials chemistry
Atomistic simulations of matter, especially those that leverage first-principles (ab initio) electronic structure theory, provide a microscopic view of the world, underpinning much of our understanding of chemistry and materials science. Over the last decade or so, machine-learned force fields have transformed atomistic modeling by enabling simulations of ab initio quality over unprecedented time and length scales. However, early machine-learning (ML) force fields have largely been limited by (i) the substantial computational and human effort required to develop and validate potentials for each particular system of interest and (ii) a general lack of transferability from one chemical system to the next. Here, we show that it is possible to create a general-purpose atomistic ML model, trained on a public dataset of moderate size, that is capable of running stable molecular dynamics for a wide range of molecules and materials. We demonstrate the power of the MACE-MP-0 model—and its qualitative and at times quantitative accuracy—on a diverse set of problems in the physical sciences, including properties of solids, liquids, gases, chemical reactions, interfaces, and even the dynamics of a small protein. The model can be applied out of the box as a starting or “foundation” model for any atomistic system of interest and, when desired, can be fine-tuned on just a handful of application-specific data points to reach ab initio accuracy. Establishing that a stable force-field model can cover almost all materials changes atomistic modeling in a fundamental way: experienced users obtain reliable results much faster, and beginners face a lower barrier to entry. Foundation models thus represent a step toward democratizing the revolution in atomic-scale modeling that has been brought about by ML force fields.
- Open Access
- 23 October 2025
- Published by Wiley in Advanced Materials
- Vol. 38 (5) , e14226
Learning Crystallographic Disorder: Bridging Prediction and Experiment in Materials Discovery
Recent computational materials discovery efforts have led to an enormous number of predictions of previously unknown, potentially stable inorganic, crystalline compounds. In particular, both high‐throughput screenings and generative models have benefited tremendously from recent advances in computational resources and available data. However, these efforts are currently limited to predicting pristine crystalline materials. As a consequence, many of these predictions cannot be realized in experiments, where kinetic effects, defects, and crystallographic disorder can be crucial. To address this shortcoming, the current work aims to introduce disorder into computational materials discovery with machine learning (ML) based classification models. Trained on the inorganic crystal structure database (ICSD), these classifiers capture the chemical trends of crystallographic disorder and estimate the prevalence of disorder in computational databases produced by the Materials Project or Graph Networks for Materials Science (GNoME) initiatives. This opens the door toward disorder‐aware computational materials discovery workflows, bridging the gap between prediction and experiment.
- 23 October 2025
- Published by Wiley in Advanced Materials
- Vol. 38 (5) , e14226
Open AccessLearning Crystallographic Disorder: Bridging Prediction and Experiment in Materials Discovery
Recent computational materials discovery efforts have led to an enormous number of predictions of previously unknown, potentially stable inorganic, crystalline compounds. In particular, both high‐throughput screenings and generative models have benefited tremendously from recent advances in computational resources and available data. However, these efforts are currently limited to predicting pristine crystalline materials. As a consequence, many of these predictions cannot be realized in experiments, where kinetic effects, defects, and crystallographic disorder can be crucial. To address this shortcoming, the current work aims to introduce disorder into computational materials discovery with machine learning (ML) based classification models. Trained on the inorganic crystal structure database (ICSD), these classifiers capture the chemical trends of crystallographic disorder and estimate the prevalence of disorder in computational databases produced by the Materials Project or Graph Networks for Materials Science (GNoME) initiatives. This opens the door toward disorder‐aware computational materials discovery workflows, bridging the gap between prediction and experiment.
- Open Access
- 24 July 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Learning Crystallographic Disorder: Bridging Prediction and Experiment in Materials Discovery
Recent computational materials discovery efforts have led to an enormous number of predictions of previously unknown, potentially stable inorganic, crystalline compounds. In particular, both high-throughput screenings and generative models have benefited tremendously from recent advances in computational resources and available data. However, these efforts are currently limited to predicting pristine crystalline materials. As a consequence, many of these predictions cannot be realized in experiments, where kinetic effects, defects, and crystallographic disorder can be crucial. To address this shortcoming, the current work aims to introduce disorder into computational materials discovery with machine learning (ML) based classification models. Trained on the inorganic crystal structure database (ICSD), these classifiers capture the chemical trends of crystallographic disorder and estimate the prevalence of disorder in computational databases produced by the Materials Project or Graph Networks for Materials Science (GNoME) initiatives. This opens the door towards disorder-aware computational materials discovery workflows, bridging the gap between prediction and experiment
- 24 July 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Open AccessLearning Crystallographic Disorder: Bridging Prediction and Experiment in Materials Discovery
Recent computational materials discovery efforts have led to an enormous number of predictions of previously unknown, potentially stable inorganic, crystalline compounds. In particular, both high-throughput screenings and generative models have benefited tremendously from recent advances in computational resources and available data. However, these efforts are currently limited to predicting pristine crystalline materials. As a consequence, many of these predictions cannot be realized in experiments, where kinetic effects, defects, and crystallographic disorder can be crucial. To address this shortcoming, the current work aims to introduce disorder into computational materials discovery with machine learning (ML) based classification models. Trained on the inorganic crystal structure database (ICSD), these classifiers capture the chemical trends of crystallographic disorder and estimate the prevalence of disorder in computational databases produced by the Materials Project or Graph Networks for Materials Science (GNoME) initiatives. This opens the door towards disorder-aware computational materials discovery workflows, bridging the gap between prediction and experiment
- Open Access
- 9 July 2025
- Published by Wiley in Advanced Intelligent Discovery
- Vol. 2 (4)
Universally Accurate or Specifically Inadequate? Stress‐Testing General Purpose Machine Learning Interatomic Potentials
Machine learning interatomic potentials (MLIPs) have revolutionized the field of atomistic materials simulation, both due to their remarkable accuracy and their computational efficiency compared to established ab initio methods. Very recently, several general purpose MLIPs have been reported, which are broadly applicable across the periodic table. These represent a fascinating opportunity for materials discovery, provided that they are robust and transferable. In order to stress test current general purpose MLIPs, we evaluate the performance of M3GNet and MACE models in element‐substitution based structure prediction workflows for a diverse range of inorganic, crystalline materials. Importantly, these results are compared with a full density functional based workflow, shifting the focus from merely evaluating single‐point energy and force predictions of MLIPs toward an end‐to‐end perspective. We find that general purpose MLIPs are in general well‐suited to accelerate computational materials discovery and structure prediction, but also display certain systematic biases. To address these, a simple metric to quantify MLIP reliability for materials discovery is introduced. As a by‐product, we also predict novel ground state structures for 15 out of 100 analyzed compositions.
- 9 July 2025
- Published by Wiley in Advanced Intelligent Discovery
- Vol. 2 (4)
Open AccessUniversally Accurate or Specifically Inadequate? Stress‐Testing General Purpose Machine Learning Interatomic Potentials
Machine learning interatomic potentials (MLIPs) have revolutionized the field of atomistic materials simulation, both due to their remarkable accuracy and their computational efficiency compared to established ab initio methods. Very recently, several general purpose MLIPs have been reported, which are broadly applicable across the periodic table. These represent a fascinating opportunity for materials discovery, provided that they are robust and transferable. In order to stress test current general purpose MLIPs, we evaluate the performance of M3GNet and MACE models in element‐substitution based structure prediction workflows for a diverse range of inorganic, crystalline materials. Importantly, these results are compared with a full density functional based workflow, shifting the focus from merely evaluating single‐point energy and force predictions of MLIPs toward an end‐to‐end perspective. We find that general purpose MLIPs are in general well‐suited to accelerate computational materials discovery and structure prediction, but also display certain systematic biases. To address these, a simple metric to quantify MLIP reliability for materials discovery is introduced. As a by‐product, we also predict novel ground state structures for 15 out of 100 analyzed compositions.
- Open Access
- 3 July 2025
- Published by American Chemical Society (ACS) in Chemistry of Materials
- Vol. 37 (14) , 5027-5035
Predict before You Precipitate: Learning Templating Effects in Hybrid Antimony and Bismuth Halides
- 3 July 2025
- Published by American Chemical Society (ACS) in Chemistry of Materials
- Vol. 37 (14) , 5027-5035
Open AccessPredict before You Precipitate: Learning Templating Effects in Hybrid Antimony and Bismuth Halides
- Open Access
- 4 March 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Universally Accurate or Specifically Inadequate? Stress-testing General Purpose Machine Learning Interatomic Potentials
Machine learning interatomic potentials (MLIPs) have revolutionized the field of atomistic materials simulation, both due to their remarkable accuracy and their computational efficiency compared to established \textit{ab initio} methods. Very recently, several general purpose MLIPs have been reported, which are broadly applicable across the periodic table. These represent a fascinating opportunity for materials discovery, provided that they are robust and transferable. In order to stress test current general purpose MLIPs, we evaluate the performance of M3GNet and MACE models in element-substitution based structure prediction workflows for a diverse range of inorganic, crystalline materials. Importantly, these results are compared with a full density functional based workflow, shifting the focus from merely evaluating single-point energy and force predictions of MLIPs towards an end-to-end perspective. We find that general purpose MLIPs are in general well-suited to accelerate computational materials discovery and structure prediction, but also display certain systematic biases. To address these, a simple metric to quantify MLIP reliability for materials discovery is introduced. As a by-product, we also predict novel ground state structures for 15 out of 100 analysed compositions.
- 4 March 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Open AccessUniversally Accurate or Specifically Inadequate? Stress-testing General Purpose Machine Learning Interatomic Potentials
Machine learning interatomic potentials (MLIPs) have revolutionized the field of atomistic materials simulation, both due to their remarkable accuracy and their computational efficiency compared to established \textit{ab initio} methods. Very recently, several general purpose MLIPs have been reported, which are broadly applicable across the periodic table. These represent a fascinating opportunity for materials discovery, provided that they are robust and transferable. In order to stress test current general purpose MLIPs, we evaluate the performance of M3GNet and MACE models in element-substitution based structure prediction workflows for a diverse range of inorganic, crystalline materials. Importantly, these results are compared with a full density functional based workflow, shifting the focus from merely evaluating single-point energy and force predictions of MLIPs towards an end-to-end perspective. We find that general purpose MLIPs are in general well-suited to accelerate computational materials discovery and structure prediction, but also display certain systematic biases. To address these, a simple metric to quantify MLIP reliability for materials discovery is introduced. As a by-product, we also predict novel ground state structures for 15 out of 100 analysed compositions.
- Open Access
- 19 February 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Predict Before You Precipitate: Learning Templating Effects in HOI Antimony and Bismuth Halides
Hybrid organic-inorganic (HOI) antimony and bismuth halides exhibit diverse structural features and have been studied intensely for their promising electronic and optical properties. There are well-explored structure-property relations for these materials. However, a thorough understanding of synthesis routes and templating effects is lacking, turning their targeted synthesis into an open challenge. In this study, we assemble a literature data set of established HOI material candidates and train an explainable machine learning (ML) classification model to explore the templating effects in more detail. With a classification accuracy upwards of 70%, our model is effective in predicting HOI structure types based on the reactants and points out several structural and electrostatic design features for the organic cation that influence the inorganic substructure most strongly. We further demonstrate the validity of our classifier on 9 newly synthesized members of this materials class and propose incremental learning routes to expand the model in future research.
- 19 February 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Open AccessPredict Before You Precipitate: Learning Templating Effects in HOI Antimony and Bismuth Halides
Hybrid organic-inorganic (HOI) antimony and bismuth halides exhibit diverse structural features and have been studied intensely for their promising electronic and optical properties. There are well-explored structure-property relations for these materials. However, a thorough understanding of synthesis routes and templating effects is lacking, turning their targeted synthesis into an open challenge. In this study, we assemble a literature data set of established HOI material candidates and train an explainable machine learning (ML) classification model to explore the templating effects in more detail. With a classification accuracy upwards of 70%, our model is effective in predicting HOI structure types based on the reactants and points out several structural and electrostatic design features for the organic cation that influence the inorganic substructure most strongly. We further demonstrate the validity of our classifier on 9 newly synthesized members of this materials class and propose incremental learning routes to expand the model in future research.