Sigbjørn Løland Bore
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Sigbjørn Løland Bore has a total of 141 co-authors. Here are the top co-authors by number of shared publications:
- Michele Cascella
University of Oslo
20 shared publications Giuseppe Milano
Department of Organic Material Science, Yamagata University, Yonezawa 992-8510, Japan
8 shared publications- Morten Ledum
University of Oslo
8 shared publications - Francesco Paesani
University of California San Diego
7 shared publications Hima Bindu Kolli
Department of Chemistry, and Hylleraas Centre for Quantum Molecular Sciences, University of Oslo, P.O. Box 1033, Blindern, 0315 Oslo, Norway
5 shared publications- Samiran Sen
Institut Pasteur, Université Paris Cité, CNRS UMR 3528, Computational Structural Biology Unit, Paris, France
5 shared publications
Latest Publications
- Open Access
- 18 November 2026
- Published by American Physical Society (APS) in Physical Review Letters
- Vol. 137 (21)
Why Ice Is So Slippery
We simulate nanoscale ice-glass friction from first principles and upscale it by modeling frictional heating. Nanoscale simulations alone overestimate friction and miss its velocity dependence. Frictional heating warms the contact toward the melting point, thickening the premelting film tenfold and cutting its viscosity a hundredfold; friction drops with velocity, matching experiment. Premelting forms the film without sliding or heating, but its slipperiness hinges on frictional heating, as Bowden and Hughes proposed in 1939, without melting. Snow, other materials, and coatings remain open.
- 18 November 2026
- Published by American Physical Society (APS) in Physical Review Letters
- Vol. 137 (21)
Open AccessWhy Ice Is So Slippery
We simulate nanoscale ice-glass friction from first principles and upscale it by modeling frictional heating. Nanoscale simulations alone overestimate friction and miss its velocity dependence. Frictional heating warms the contact toward the melting point, thickening the premelting film tenfold and cutting its viscosity a hundredfold; friction drops with velocity, matching experiment. Premelting forms the film without sliding or heating, but its slipperiness hinges on frictional heating, as Bowden and Hughes proposed in 1939, without melting. Snow, other materials, and coatings remain open.
- Open Access
- 8 December 2025
- Published by Elsevier in Journal of Catalysis
- Vol. 454
The influence of the organic residue and the solvent in the Schlenk equilibrium for Grignard reagents in THF. A molecular dynamics study with machine learning potentials
- 8 December 2025
- Published by Elsevier in Journal of Catalysis
- Vol. 454
Open AccessThe influence of the organic residue and the solvent in the Schlenk equilibrium for Grignard reagents in THF. A molecular dynamics study with machine learning potentials
- Open Access
- 1 July 2025
- Published by Springer Nature in npj Computational Materials
- Vol. 11 (1) , 1-9
Learning atomic forces from uncertainty-calibrated adversarial attacks
Adversarial approaches, which intentionally challenge machine learning models by generating difficult examples, are increasingly being adopted to improve machine learning interatomic potentials (MLIPs). While already providing great practical value, little is known about the actual prediction errors of MLIPs on adversarial structures and whether these errors can be controlled. We propose the Calibrated Adversarial Geometry Optimization (CAGO) algorithm to discover adversarial structures with user-assigned errors. Through uncertainty calibration, the estimated uncertainty of MLIPs is unified with real errors. By performing geometry optimization for calibrated uncertainty, we reach adversarial structures with the user-assigned target MLIP prediction error. Integrating with active learning pipelines, we benchmark CAGO, demonstrating stable MLIPs that systematically converge structural, dynamical, and thermodynamical properties for liquid water and water adsorption in a metal-organic framework within only hundreds of training structures, where previously many thousands were typically required.
- 1 July 2025
- Published by Springer Nature in npj Computational Materials
- Vol. 11 (1) , 1-9
Open AccessLearning atomic forces from uncertainty-calibrated adversarial attacks
Adversarial approaches, which intentionally challenge machine learning models by generating difficult examples, are increasingly being adopted to improve machine learning interatomic potentials (MLIPs). While already providing great practical value, little is known about the actual prediction errors of MLIPs on adversarial structures and whether these errors can be controlled. We propose the Calibrated Adversarial Geometry Optimization (CAGO) algorithm to discover adversarial structures with user-assigned errors. Through uncertainty calibration, the estimated uncertainty of MLIPs is unified with real errors. By performing geometry optimization for calibrated uncertainty, we reach adversarial structures with the user-assigned target MLIP prediction error. Integrating with active learning pipelines, we benchmark CAGO, demonstrating stable MLIPs that systematically converge structural, dynamical, and thermodynamical properties for liquid water and water adsorption in a metal-organic framework within only hundreds of training structures, where previously many thousands were typically required.
- Top Cited Papers
- 2 May 2025
- Published by American Chemical Society (ACS) in Journal of Chemical Theory and Computation
- Vol. 21 (9) , 4375-4385
DeePMD-kit v3: A Multiple-Backend Framework for Machine Learning Potentials
In recent years, machine learning potentials (MLPs) have become indispensable tools in physics, chemistry, and materials science, driving the development of software packages for molecular dynamics (MD) simulations and related applications. These packages, typically built on specific machine learning frameworks, such as TensorFlow, PyTorch, or JAX, face integration challenges when advanced applications demand communication across different frameworks. The previous TensorFlow-based implementation of the DeePMD-kit exemplified these limitations. In this work, we introduce DeePMD-kit version 3, a significant update featuring a multibackend framework that supports TensorFlow, PyTorch, JAX, and PaddlePaddle backends, and demonstrate the versatility of this architecture through the integration of other MLP packages and of differentiable molecular force fields. This architecture allows seamless back-end switching with minimal modifications, enabling users and developers to integrate DeePMD-kit with other packages using different machine learning frameworks. This innovation facilitates the development of more complex and interoperable workflows, paving the way for broader applications of MLPs in scientific research.
- 2 May 2025
- Published by American Chemical Society (ACS) in Journal of Chemical Theory and Computation
- Vol. 21 (9) , 4375-4385
Top Cited PapersDeePMD-kit v3: A Multiple-Backend Framework for Machine Learning Potentials
In recent years, machine learning potentials (MLPs) have become indispensable tools in physics, chemistry, and materials science, driving the development of software packages for molecular dynamics (MD) simulations and related applications. These packages, typically built on specific machine learning frameworks, such as TensorFlow, PyTorch, or JAX, face integration challenges when advanced applications demand communication across different frameworks. The previous TensorFlow-based implementation of the DeePMD-kit exemplified these limitations. In this work, we introduce DeePMD-kit version 3, a significant update featuring a multibackend framework that supports TensorFlow, PyTorch, JAX, and PaddlePaddle backends, and demonstrate the versatility of this architecture through the integration of other MLP packages and of differentiable molecular force fields. This architecture allows seamless back-end switching with minimal modifications, enabling users and developers to integrate DeePMD-kit with other packages using different machine learning frameworks. This innovation facilitates the development of more complex and interoperable workflows, paving the way for broader applications of MLPs in scientific research.
- Open Access
- 11 April 2025
- Published by Royal Society of Chemistry (RSC) in Chemical Science
- Vol. 16 (19) , 8196-8216
The fellowship of the Grignard: 21st century computational tools for hundred-year-old chemistry
This perspective begins with the discovery of the Grignard reaction by a graduate student in the last years of the 19th century, followed by describing why it has remained largely unexplained for more than a century. From the summary of what has been achieved, focusing on the computational aspects, it is now clear that further studies of the chemistry of any chemical species that is highly sensitive to solvents, such as Group I and II elements, require a holistic approach that includes the solute and the solvent together. Ab initio molecular dynamics, which meets these requirements, has produced some results but has hit hard limits due to its relatively high computational costs. In these days, it is becoming clear that data-driven methods, including machine learning potentials and simulations driven by quantitative on-the-fly calculation of relevant observables, have the potential to better and more completely explore the very large chemical space associated with the presence of a large number of species in solution. These methodologies have the chance to give the keys to enter the challenging and still poorly explored world of chemical species whose behaviour and reactivity are strongly influenced by the solvent and the experimental conditions.
- 11 April 2025
- Published by Royal Society of Chemistry (RSC) in Chemical Science
- Vol. 16 (19) , 8196-8216
Open AccessThe fellowship of the Grignard: 21st century computational tools for hundred-year-old chemistry
This perspective begins with the discovery of the Grignard reaction by a graduate student in the last years of the 19th century, followed by describing why it has remained largely unexplained for more than a century. From the summary of what has been achieved, focusing on the computational aspects, it is now clear that further studies of the chemistry of any chemical species that is highly sensitive to solvents, such as Group I and II elements, require a holistic approach that includes the solute and the solvent together. Ab initio molecular dynamics, which meets these requirements, has produced some results but has hit hard limits due to its relatively high computational costs. In these days, it is becoming clear that data-driven methods, including machine learning potentials and simulations driven by quantitative on-the-fly calculation of relevant observables, have the potential to better and more completely explore the very large chemical space associated with the presence of a large number of species in solution. These methodologies have the chance to give the keys to enter the challenging and still poorly explored world of chemical species whose behaviour and reactivity are strongly influenced by the solvent and the experimental conditions.
- Top Cited Papers
- 3 February 2025
- Published by Springer Nature in Nature Physics
- Vol. 21 (3) , 480-485
Constraints on the location of the liquid–liquid critical point in water
- 3 February 2025
- Published by Springer Nature in Nature Physics
- Vol. 21 (3) , 480-485
Top Cited PapersConstraints on the location of the liquid–liquid critical point in water
- Open Access
- 12 November 2024
- Published by Royal Society of Chemistry (RSC) in Chemical Science
- Vol. 15 (48) , 20355-20364
Morphology of lithium halides in tetrahydrofuran from molecular dynamics with machine learning potentials
Using machine-learning potentials with ab initio accuracy, molecular dynamics simulations predict a diversity of structures for lithium halogen salts dissolved in tetrahydrofuran – from more compact LiCl, to more dispersed LiI.
- 12 November 2024
- Published by Royal Society of Chemistry (RSC) in Chemical Science
- Vol. 15 (48) , 20355-20364
Open AccessMorphology of lithium halides in tetrahydrofuran from molecular dynamics with machine learning potentials
Using machine-learning potentials with ab initio accuracy, molecular dynamics simulations predict a diversity of structures for lithium halogen salts dissolved in tetrahydrofuran – from more compact LiCl, to more dispersed LiI.
- Open Access
- 7 November 2024
- Published by American Chemical Society (ACS) in ChemRxiv
Morphology of Lithium Halides in Tetrahydrofuran from Molecular Dynamics with Machine Learning Potentials
The preferred structures of lithium halides (LiX, with X = Cl, Br, I) in organic solvents have been the subject of a wide scientific debate, and a large variety of forms has been isolated and characterized by X-ray diffraction. The identified molecular scaffolds for LiX are diverse, often built on (LiX)n rings with a prevalence of rhomboidal arrangements and an appropriate number of solvent or Lewis base molecules coordinating the lithium ions. Much less is known about the structures of LiX in solution, limiting the understanding of the synergistic role of LiX in reactions with various organometallic complexes, as prominently represented by the turbo Grignard reaction. Here, we trained a machine learning potential on ab initio data to explore the complex conformational landscape for systems comprising 4 LiX moieties in tetrahydrofuran (THF). For all the considered halogens a large number of scaffolds were found at thermally accessible free energy values, indicating that LiX in solution are a diverse ensemble constituted by (LiX)n moieties of various size, completed by the appropriate number of coordinating THF. LiCl shows preference for compact, pseudo-cubane [(LiCl)(THF)]4 structures, coexisting with open rings. At concen- trations close to the solubility limit, LiCl forms hexagonal structures, in analogy with literature observations on pre-nucleating NaCl. LiBr tends to favor less compact, more solvated aggregates. LiI significantly differs from the two other cases, producing highly solvated, monomeric, dimeric or linear structures. This study provides a comprehensive view of LiX in organic solvent, revealing dynamical polymorphism that is not easily experimentally observable.
- 7 November 2024
- Published by American Chemical Society (ACS) in ChemRxiv
Open AccessMorphology of Lithium Halides in Tetrahydrofuran from Molecular Dynamics with Machine Learning Potentials
The preferred structures of lithium halides (LiX, with X = Cl, Br, I) in organic solvents have been the subject of a wide scientific debate, and a large variety of forms has been isolated and characterized by X-ray diffraction. The identified molecular scaffolds for LiX are diverse, often built on (LiX)n rings with a prevalence of rhomboidal arrangements and an appropriate number of solvent or Lewis base molecules coordinating the lithium ions. Much less is known about the structures of LiX in solution, limiting the understanding of the synergistic role of LiX in reactions with various organometallic complexes, as prominently represented by the turbo Grignard reaction. Here, we trained a machine learning potential on ab initio data to explore the complex conformational landscape for systems comprising 4 LiX moieties in tetrahydrofuran (THF). For all the considered halogens a large number of scaffolds were found at thermally accessible free energy values, indicating that LiX in solution are a diverse ensemble constituted by (LiX)n moieties of various size, completed by the appropriate number of coordinating THF. LiCl shows preference for compact, pseudo-cubane [(LiCl)(THF)]4 structures, coexisting with open rings. At concen- trations close to the solubility limit, LiCl forms hexagonal structures, in analogy with literature observations on pre-nucleating NaCl. LiBr tends to favor less compact, more solvated aggregates. LiI significantly differs from the two other cases, producing highly solvated, monomeric, dimeric or linear structures. This study provides a comprehensive view of LiX in organic solvent, revealing dynamical polymorphism that is not easily experimentally observable.
- Open Access
- 27 August 2024
- Published by American Chemical Society (ACS) in ChemRxiv
Pinpointing the Location of the Elusive Liquid-Liquid Critical Point in Water
The fascinating hypothesis that supercooled water may segregate into two distinct liquid phases, each with unique structures and densities, was first posited in a seminal study by Poole et al. in 1992. This idea, initially based on numerical analyses with the ST2 water-like empirical potential, challenged conventional understanding of water’s phase behavior at the time and has since intrigued the scientific community. Over the past three decades, advancements in computational modeling – particularly through the advent of data-driven many-body potentials rigorously derived from “first principles” and augmented by the efficiency of neural networks – have significantly enhanced the accuracy of molecular simulations, enabling the exploration of the phase behavior of water with unprecedented realism. Our study leverages these computational advances to probe the elusive liquid-liquid tran- sition in supercooled water. Microsecond-long simulations with chemical accuracy, conducted over several years, provide compelling evidence that water indeed exists in two discernibly distinct liquid states at low temperature and high pressure. By pinpointing a realistic estimate for the location of the liquid-liquid critical point at ∼200 K and ∼1050 atm, our study not only advances current understanding of water’s anomalous behavior but also establishes a basis for experimental validation. Importantly, our simulations indicate that the liquid-liquid critical point falls within temperature and pressure ranges that could potentially be experimentally probed in water nanodroplets, opening the possibility for direct measurements.
- 27 August 2024
- Published by American Chemical Society (ACS) in ChemRxiv
Open AccessPinpointing the Location of the Elusive Liquid-Liquid Critical Point in Water
The fascinating hypothesis that supercooled water may segregate into two distinct liquid phases, each with unique structures and densities, was first posited in a seminal study by Poole et al. in 1992. This idea, initially based on numerical analyses with the ST2 water-like empirical potential, challenged conventional understanding of water’s phase behavior at the time and has since intrigued the scientific community. Over the past three decades, advancements in computational modeling – particularly through the advent of data-driven many-body potentials rigorously derived from “first principles” and augmented by the efficiency of neural networks – have significantly enhanced the accuracy of molecular simulations, enabling the exploration of the phase behavior of water with unprecedented realism. Our study leverages these computational advances to probe the elusive liquid-liquid tran- sition in supercooled water. Microsecond-long simulations with chemical accuracy, conducted over several years, provide compelling evidence that water indeed exists in two discernibly distinct liquid states at low temperature and high pressure. By pinpointing a realistic estimate for the location of the liquid-liquid critical point at ∼200 K and ∼1050 atm, our study not only advances current understanding of water’s anomalous behavior but also establishes a basis for experimental validation. Importantly, our simulations indicate that the liquid-liquid critical point falls within temperature and pressure ranges that could potentially be experimentally probed in water nanodroplets, opening the possibility for direct measurements.
- Open Access
- 4 July 2024
- Published by American Chemical Society (ACS) in Journal of Chemical Information and Modeling
- Vol. 64 (14) , 5510-5520
Learning Force Field Parameters from Differentiable Particle-Field Molecular Dynamics
We develop ∂-HylleraasMD (∂-HyMD), a fully end-to-end differentiable molecular dynamics software based on the Hamiltonian hybrid particle-field formalism, and use it to establish a protocol for automated optimization of force field parameters. ∂-HyMD is templated on the recently released HylleraaasMD software, while using the JAX autodiff framework as the main engine for the differentiable dynamics. ∂-HyMD exploits an embarrassingly parallel optimization algorithm by spawning independent simulations, whose trajectories are simultaneously processed by reverse mode automatic differentiation to calculate the gradient of the loss function, which is in turn used for iterative optimization of the force-field parameters. We show that parallel organization facilitates the convergence of the minimization procedure, avoiding the known memory and numerical stability issues of differentiable molecular dynamics approaches. We showcase the effectiveness of our implementation by producing a library of force field parameters for standard phospholipids, with either zwitterionic or anionic heads and with saturated or unsaturated tails. Compared to the all-atom reference, the force field obtained by ∂-HyMD yields better density profiles than the parameters derived from previously utilized gradient-free optimization procedures. Moreover, ∂-HyMD models can predict with good accuracy properties not included in the learning objective, such as lateral pressure profiles, and are transferable to other systems, including triglycerides.
- 4 July 2024
- Published by American Chemical Society (ACS) in Journal of Chemical Information and Modeling
- Vol. 64 (14) , 5510-5520
Open AccessLearning Force Field Parameters from Differentiable Particle-Field Molecular Dynamics
We develop ∂-HylleraasMD (∂-HyMD), a fully end-to-end differentiable molecular dynamics software based on the Hamiltonian hybrid particle-field formalism, and use it to establish a protocol for automated optimization of force field parameters. ∂-HyMD is templated on the recently released HylleraaasMD software, while using the JAX autodiff framework as the main engine for the differentiable dynamics. ∂-HyMD exploits an embarrassingly parallel optimization algorithm by spawning independent simulations, whose trajectories are simultaneously processed by reverse mode automatic differentiation to calculate the gradient of the loss function, which is in turn used for iterative optimization of the force-field parameters. We show that parallel organization facilitates the convergence of the minimization procedure, avoiding the known memory and numerical stability issues of differentiable molecular dynamics approaches. We showcase the effectiveness of our implementation by producing a library of force field parameters for standard phospholipids, with either zwitterionic or anionic heads and with saturated or unsaturated tails. Compared to the all-atom reference, the force field obtained by ∂-HyMD yields better density profiles than the parameters derived from previously utilized gradient-free optimization procedures. Moreover, ∂-HyMD models can predict with good accuracy properties not included in the learning objective, such as lateral pressure profiles, and are transferable to other systems, including triglycerides.