Amit Samanta
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Amit Samanta has a total of 69 co-authors. Here are the top co-authors by number of shared publications:
- Vincenzo Lordi
Lawrence Livermore National Laboratory
10 shared publications - Axel van de Walle
Brown University
3 shared publications Babak Sadigh
3 shared publications- Ellad B. Tadmor
University of Minnesota
3 shared publications - Joshua A. Vita
Lawrence Livermore National Laboratory, 7000 East Ave, Livermore, CA 94550, United States of America
3 shared publications - Sebastien Hamel
Lawrence Livermore National Laboratory, Livermore, USA
3 shared publications
Latest Publications
- Open Access
- 22 October 2026
- Published by Elsevier in SSRN Electronic Journal
Analysis of Dynamical Transitions at Symmetric and Asymmetric Tilt Grain Boundaries in W, Mo and Ta at High Temperature
The structure of grain boundaries (GBs) in refractory body-centered-cubic metals has been studied extensively at 0~K, but their behavior under finite-temperature conditions has received comparatively little attention. Using an interatomic potential designed for the W--Ta--Mo ternary alloy system, we illustrate two contrasting types of dynamical transitions that occur at GBs in W, Mo, and Ta at temperatures above half their melting temperatures. In the first type of dynamical transitions, we consider the $\Sigma{27}[1\bar{1}0](552)$ symmetric tilt boundary, which stays geometrically flat while its atomic columns continuously exchange atoms: ring-like structural units repeatedly nucleate and dissolve, and the per-column atom count fluctuates through diffusion along the boundary. We use a pseudo-Markovian state-space framework in which the height of each atomic column along the tilt axis is treated as a discrete state, and time-averaged populations and continuous-time transition-rate matrices are used to quantify the stability of the GB. In the second type of dynamical transitions, we consider the $\Sigma{5}[100]$ asymmetric tilt boundary, which relaxes into a rough, faceted profile of $\{310\}$, $\{210\}$, and $\{100\}\!\parallel\!\{110\}$ segments. Diffusion along it is not uniform but localized at facet junctions and under-occupied columns. The dominant facet type and the overall mobility are governed by chemistry, with W forming sharp $\{210\}/\{310\}$ bends whereas Mo and Ta favor flatter $\{100\}\!\parallel\!\{110\}$ segments. This faceted boundary is characterized through the transverse motion of the boundary line, whose mean-square displacement crosses over from sub-diffusive to diffusive behavior depending on chemistry and sample thickness.- 22 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessAnalysis of Dynamical Transitions at Symmetric and Asymmetric Tilt Grain Boundaries in W, Mo and Ta at High Temperature
The structure of grain boundaries (GBs) in refractory body-centered-cubic metals has been studied extensively at 0~K, but their behavior under finite-temperature conditions has received comparatively little attention. Using an interatomic potential designed for the W--Ta--Mo ternary alloy system, we illustrate two contrasting types of dynamical transitions that occur at GBs in W, Mo, and Ta at temperatures above half their melting temperatures. In the first type of dynamical transitions, we consider the $\Sigma{27}[1\bar{1}0](552)$ symmetric tilt boundary, which stays geometrically flat while its atomic columns continuously exchange atoms: ring-like structural units repeatedly nucleate and dissolve, and the per-column atom count fluctuates through diffusion along the boundary. We use a pseudo-Markovian state-space framework in which the height of each atomic column along the tilt axis is treated as a discrete state, and time-averaged populations and continuous-time transition-rate matrices are used to quantify the stability of the GB. In the second type of dynamical transitions, we consider the $\Sigma{5}[100]$ asymmetric tilt boundary, which relaxes into a rough, faceted profile of $\{310\}$, $\{210\}$, and $\{100\}\!\parallel\!\{110\}$ segments. Diffusion along it is not uniform but localized at facet junctions and under-occupied columns. The dominant facet type and the overall mobility are governed by chemistry, with W forming sharp $\{210\}/\{310\}$ bends whereas Mo and Ta favor flatter $\{100\}\!\parallel\!\{110\}$ segments. This faceted boundary is characterized through the transverse motion of the boundary line, whose mean-square displacement crosses over from sub-diffusive to diffusive behavior depending on chemistry and sample thickness. - Open Access
- 21 September 2026
- Published by Elsevier in SSRN Electronic Journal
Effects Of Solute Chemistry and Local Arrangements on Screw-Dislocation Cores and Glide Barriers in BCC Alloys
Understanding how alloy chemistry alters screw-dislocation cores is vital to plasticity in body-centered cubic (BCC) alloys. Here, we develop density-functional-theory-trained interatomic potentials within the generalized embedded atom method framework for three representative binary systems, Mo--V, W--Ti, and W--Cu, and a quaternary system, Cr--Mo--Nb--V, to investigate how solute chemistry and local arrangement govern the structure and glide behavior of $\frac{1}{2}\langle 111\rangle$ screw dislocations. By combining differential-displacement maps, Nye-tensor analysis, nudged elastic band calculations, and long-line relaxations, we establish a clear hierarchy of disorder sensitivity across these alloys. Mo--V retains a compact, low-polarity non-degenerate core, whereas W--Ti remains comparatively compact but develops increasing polarity and approaches a degenerate-like compact configuration. In contrast, W--Cu exhibits both compact and non-compact cores at the same global composition, with polarity and spatial extent varying independently across realizations. Canonical degenerate and non-degenerate labels thus tail to capture the full range of alloy-induced core structures. To quantify this behavior, we complement the polarity metric with the second-moment of the Nye tensor that captures spatial delocalization of the core. The calculated Peierls barriers further show that glide energetics do not map uniquely onto a single static core structure; instead, pathway-to-pathway variability increases markedly from Mo--V to W--Ti to W--Cu. Consistent with this trend, long dislocation-line relaxations reveal minimal roughening in Mo--V and W--Ti but pronounced waviness and kink-like deviations in W--Cu. Local chemical disorder thus governs both core morphology and glide energetics, motivating ensemble-based descriptions of dislocation behavior in chemically complex alloys.- 21 September 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessEffects Of Solute Chemistry and Local Arrangements on Screw-Dislocation Cores and Glide Barriers in BCC Alloys
Understanding how alloy chemistry alters screw-dislocation cores is vital to plasticity in body-centered cubic (BCC) alloys. Here, we develop density-functional-theory-trained interatomic potentials within the generalized embedded atom method framework for three representative binary systems, Mo--V, W--Ti, and W--Cu, and a quaternary system, Cr--Mo--Nb--V, to investigate how solute chemistry and local arrangement govern the structure and glide behavior of $\frac{1}{2}\langle 111\rangle$ screw dislocations. By combining differential-displacement maps, Nye-tensor analysis, nudged elastic band calculations, and long-line relaxations, we establish a clear hierarchy of disorder sensitivity across these alloys. Mo--V retains a compact, low-polarity non-degenerate core, whereas W--Ti remains comparatively compact but develops increasing polarity and approaches a degenerate-like compact configuration. In contrast, W--Cu exhibits both compact and non-compact cores at the same global composition, with polarity and spatial extent varying independently across realizations. Canonical degenerate and non-degenerate labels thus tail to capture the full range of alloy-induced core structures. To quantify this behavior, we complement the polarity metric with the second-moment of the Nye tensor that captures spatial delocalization of the core. The calculated Peierls barriers further show that glide energetics do not map uniquely onto a single static core structure; instead, pathway-to-pathway variability increases markedly from Mo--V to W--Ti to W--Cu. Consistent with this trend, long dislocation-line relaxations reveal minimal roughening in Mo--V and W--Ti but pronounced waviness and kink-like deviations in W--Cu. Local chemical disorder thus governs both core morphology and glide energetics, motivating ensemble-based descriptions of dislocation behavior in chemically complex alloys. - Open Access
- 29 August 2026
- Published by Springer Nature in Journal of Materials Science: Materials Theory
- Vol. 10 (1) , 16
Inverse design of bespoke interatomic potentials via active learning by information-matching
- 29 August 2026
- Published by Springer Nature in Journal of Materials Science: Materials Theory
- Vol. 10 (1) , 16
Open AccessInverse design of bespoke interatomic potentials via active learning by information-matching
- Open Access
- 23 March 2026
- Published by IOP Publishing in Machine Learning: Science and Technology
- Vol. 7 (2) , 025048
Federated learning for 2D synchrotron X-ray diffractometry: A cross-institutional approach for phase quantification of Ti-6Al-4V alloy
High-energy 2D synchrotron X-ray diffractometry provides important insights into the atomistic structure and phase evolution of materials, yet traditional analysis methods remain complex, knowledge-intensive, and computationally demanding. Deep- learning models offer a powerful alternative for automating their analysis. Institutions that hold these datasets may be unwilling to share their data due to privacy and security policies, as well as the challenges associated with large-scale data transfer. As a result, models trained on local datasets often perform well only on their own data but exhibit bias and poor generalization across different instruments or facilities. To overcome these limitations, we explore federated learning for 2D synchrotron diffractograms, enabling collaborative model training without exchanging raw data. In this study, 2D synchrotron diffractograms of Ti-6Al-4V alloy collected from two independent facilities are used to train convolutional neural networks for predicting the β-phase volume fraction. Experimental results show that federated global models significantly outperform locally trained models in terms of generalization and achieve accuracy comparable to centralized trained models. These findings demonstrate the potential of federated learning to enable secure, cross-institutional collaboration and enhance the scalability of deep-learning-based materials characterization.
- 23 March 2026
- Published by IOP Publishing in Machine Learning: Science and Technology
- Vol. 7 (2) , 025048
Open AccessFederated learning for 2D synchrotron X-ray diffractometry: A cross-institutional approach for phase quantification of Ti-6Al-4V alloy
High-energy 2D synchrotron X-ray diffractometry provides important insights into the atomistic structure and phase evolution of materials, yet traditional analysis methods remain complex, knowledge-intensive, and computationally demanding. Deep- learning models offer a powerful alternative for automating their analysis. Institutions that hold these datasets may be unwilling to share their data due to privacy and security policies, as well as the challenges associated with large-scale data transfer. As a result, models trained on local datasets often perform well only on their own data but exhibit bias and poor generalization across different instruments or facilities. To overcome these limitations, we explore federated learning for 2D synchrotron diffractograms, enabling collaborative model training without exchanging raw data. In this study, 2D synchrotron diffractograms of Ti-6Al-4V alloy collected from two independent facilities are used to train convolutional neural networks for predicting the β-phase volume fraction. Experimental results show that federated global models significantly outperform locally trained models in terms of generalization and achieve accuracy comparable to centralized trained models. These findings demonstrate the potential of federated learning to enable secure, cross-institutional collaboration and enhance the scalability of deep-learning-based materials characterization.
- Open Access
- 13 February 2026
- Published by IOP Publishing in Machine Learning: Science and Technology
- Vol. 7 (2) , 025003
Unsupervised atomic data mining via multi-kernel graph autoencoders for machine learning force fields
Constructing a chemically diverse dataset while avoiding sampling bias is critical to training efficient and generalizable force fields. However, in computational chemistry and materials science, many common dataset generation techniques are prone to oversampling regions of the potential energy surface. Furthermore, these regions can be difficult to identify and isolate from each other or may not align well with human intuition, making it challenging to systematically remove bias in the dataset. While traditional clustering and pruning (down-sampling) approaches can be useful for this, they can often lead to information loss or a failure to properly identify distinct regions of the potential energy surface due to difficulties associated with the high dimensionality of atomic descriptors. In this work, we introduce the Multi-kernel Edge Attention-based Graph Autoencoder (MEAGraph) model, an unsupervised approach for analyzing atomic datasets. MEAGraph combines multiple linear kernel transformations with attention-based message passing to capture geometric sensitivity and enable effective dataset pruning without relying on labels or extensive training. Demonstrated applications on niobium, tantalum, and iron datasets show that MEAGraph efficiently groups similar atomic environments, allowing for the use of basic pruning techniques for removing sampling bias. This approach provides an effective method for representation learning and clustering that can be used for data analysis, outlier detection, and dataset optimization.
- 13 February 2026
- Published by IOP Publishing in Machine Learning: Science and Technology
- Vol. 7 (2) , 025003
Open AccessUnsupervised atomic data mining via multi-kernel graph autoencoders for machine learning force fields
Constructing a chemically diverse dataset while avoiding sampling bias is critical to training efficient and generalizable force fields. However, in computational chemistry and materials science, many common dataset generation techniques are prone to oversampling regions of the potential energy surface. Furthermore, these regions can be difficult to identify and isolate from each other or may not align well with human intuition, making it challenging to systematically remove bias in the dataset. While traditional clustering and pruning (down-sampling) approaches can be useful for this, they can often lead to information loss or a failure to properly identify distinct regions of the potential energy surface due to difficulties associated with the high dimensionality of atomic descriptors. In this work, we introduce the Multi-kernel Edge Attention-based Graph Autoencoder (MEAGraph) model, an unsupervised approach for analyzing atomic datasets. MEAGraph combines multiple linear kernel transformations with attention-based message passing to capture geometric sensitivity and enable effective dataset pruning without relying on labels or extensive training. Demonstrated applications on niobium, tantalum, and iron datasets show that MEAGraph efficiently groups similar atomic environments, allowing for the use of basic pruning techniques for removing sampling bias. This approach provides an effective method for representation learning and clustering that can be used for data analysis, outlier detection, and dataset optimization.
- 23 December 2025
- Published by Elsevier in Computational Materials Science
- Vol. 263
Development of an interatomic potential for the Ta–Li system
- 23 December 2025
- Published by Elsevier in Computational Materials Science
- Vol. 263
Development of an interatomic potential for the Ta–Li system
- Open Access
Unsupervised Atomic Data Mining via Multi-Kernel Graph Autoencoders for Machine Learning Force Fields
Constructing a chemically diverse dataset while avoiding sampling bias is critical to training efficient and generalizable force fields. However, in computational chemistry and materials science, many common dataset generation techniques are prone to oversampling regions of the potential energy surface. Furthermore, these regions can be difficult to identify and isolate from each other or may not align well with human intuition, making it challenging to systematically remove bias in the dataset. While traditional clustering and pruning (down-sampling) approaches can be useful for this, they can often lead to information loss or a failure to properly identify distinct regions of the potential energy surface due to difficulties associated with the high dimensionality of atomic descriptors. In this work, we introduce the Multi-kernel Edge Attention-based Graph Autoencoder (MEAGraph) model, an unsupervised approach for analyzing atomic datasets. MEAGraph combines multiple linear kernel transformations with attention-based message passing to capture geometric sensitivity and enable effective dataset pruning without relying on labels or extensive training. Demonstrated applications on niobium, tantalum, and iron datasets show that MEAGraph efficiently groups similar atomic environments, allowing for the use of basic pruning techniques for removing sampling bias. This approach provides an effective method for representation learning and clustering that can be used for data analysis, outlier detection, and dataset optimization.
Open AccessUnsupervised Atomic Data Mining via Multi-Kernel Graph Autoencoders for Machine Learning Force Fields
Constructing a chemically diverse dataset while avoiding sampling bias is critical to training efficient and generalizable force fields. However, in computational chemistry and materials science, many common dataset generation techniques are prone to oversampling regions of the potential energy surface. Furthermore, these regions can be difficult to identify and isolate from each other or may not align well with human intuition, making it challenging to systematically remove bias in the dataset. While traditional clustering and pruning (down-sampling) approaches can be useful for this, they can often lead to information loss or a failure to properly identify distinct regions of the potential energy surface due to difficulties associated with the high dimensionality of atomic descriptors. In this work, we introduce the Multi-kernel Edge Attention-based Graph Autoencoder (MEAGraph) model, an unsupervised approach for analyzing atomic datasets. MEAGraph combines multiple linear kernel transformations with attention-based message passing to capture geometric sensitivity and enable effective dataset pruning without relying on labels or extensive training. Demonstrated applications on niobium, tantalum, and iron datasets show that MEAGraph efficiently groups similar atomic environments, allowing for the use of basic pruning techniques for removing sampling bias. This approach provides an effective method for representation learning and clustering that can be used for data analysis, outlier detection, and dataset optimization.
- Top Cited Papers
- 9 June 2025
- Published by American Chemical Society (ACS) in Journal of the American Chemical Society
- Vol. 147 (25) , 22104-22114
Atomic-Scale Dynamics of Five-Fold Twin Mediated Coalescence: Pathway-Dependent and Defect-Governed Nonclassical Growth Mechanisms
Defective crystals with distinct properties have been discovered in many systems. However, the growth mechanism of defective crystals is still poorly understood. Here, using a 5-fold twinned gold (Au) nanocrystal (NC) as a model system, three new coalescence pathways involving detwinning or twinning have been uncovered through atomic-scale dynamic observations in an aberration-corrected transmission electron microscope coupled with atomistic simulations. This demonstrates that beyond crystal size, coalescence growth dynamics involving 5-fold twins (5-FTs) are highly dependent on crystal defect density and the approach pathways of the crystals. When a 5-FT encounters a smaller 5-FT or a smaller NC in a face-to-face way, a new, larger 5-FT is produced at a relatively fast coalescence growth rate; while in a corner-to-corner way, the coalescence dynamics are more retarded and sluggish, which is conducive to the formation of complex multitwined structures rather than 5-FTs. This highlights that the planar defect density and crystal approach pathway influence the coalescence dynamics of crystals containing 5-FT. Moreover, a column-by-column grain boundary (GB) migration mechanism, which results in bent GBs, was also unveiled during the crystal coalescence process. These results enrich the general understanding of the crystal growth theory and provide new insights into the controllable fabrication of 5-FTs by crystal coalescence mechanisms.
- 9 June 2025
- Published by American Chemical Society (ACS) in Journal of the American Chemical Society
- Vol. 147 (25) , 22104-22114
Top Cited PapersAtomic-Scale Dynamics of Five-Fold Twin Mediated Coalescence: Pathway-Dependent and Defect-Governed Nonclassical Growth Mechanisms
Defective crystals with distinct properties have been discovered in many systems. However, the growth mechanism of defective crystals is still poorly understood. Here, using a 5-fold twinned gold (Au) nanocrystal (NC) as a model system, three new coalescence pathways involving detwinning or twinning have been uncovered through atomic-scale dynamic observations in an aberration-corrected transmission electron microscope coupled with atomistic simulations. This demonstrates that beyond crystal size, coalescence growth dynamics involving 5-fold twins (5-FTs) are highly dependent on crystal defect density and the approach pathways of the crystals. When a 5-FT encounters a smaller 5-FT or a smaller NC in a face-to-face way, a new, larger 5-FT is produced at a relatively fast coalescence growth rate; while in a corner-to-corner way, the coalescence dynamics are more retarded and sluggish, which is conducive to the formation of complex multitwined structures rather than 5-FTs. This highlights that the planar defect density and crystal approach pathway influence the coalescence dynamics of crystals containing 5-FT. Moreover, a column-by-column grain boundary (GB) migration mechanism, which results in bent GBs, was also unveiled during the crystal coalescence process. These results enrich the general understanding of the crystal growth theory and provide new insights into the controllable fabrication of 5-FTs by crystal coalescence mechanisms.
- Open Access
- 20 February 2025
- Published by IOP Publishing in Machine Learning: Science and Technology
- Vol. 6 (1) , 015048
LTAU-FF: Loss Trajectory Analysis for Uncertainty in atomistic Force Fields
Model ensembles are effective tools for estimating prediction uncertainty in deep learning atomistic force fields. However, their widespread adoption is hindered by high computational costs and overconfident error estimates. In this work, we address these challenges by leveraging distributions of per-sample errors obtained during training and employing a distance-based similarity search in the model latent space. Our method, which we call LTAU (Loss Trajectory Analysis for Uncertainty), efficiently estimates the full probability distribution function of errors for any test point using the logged training errors, achieving speeds that are 2–3 orders of magnitudes faster than typical ensemble methods and allowing it to be used for tasks where training or evaluating multiple models would be infeasible. We apply LTAU towards estimating parametric uncertainty in atomistic force fields (LTAU-FF), demonstrating that it produces well-calibrated confidence intervals and predicts errors that correlate strongly with the true errors for data near the training domain. Furthermore, we show that the errors predicted by LTAU-FF can be used in practical applications for detecting out-of-domain data, tuning model performance, and predicting failure during simulations. We believe that LTAU will be a valuable tool for uncertainty quantification in atomistic force fields and is a promising method that should be further explored in other domains of machine learning.
- 20 February 2025
- Published by IOP Publishing in Machine Learning: Science and Technology
- Vol. 6 (1) , 015048
Open AccessLTAU-FF: Loss Trajectory Analysis for Uncertainty in atomistic Force Fields
Model ensembles are effective tools for estimating prediction uncertainty in deep learning atomistic force fields. However, their widespread adoption is hindered by high computational costs and overconfident error estimates. In this work, we address these challenges by leveraging distributions of per-sample errors obtained during training and employing a distance-based similarity search in the model latent space. Our method, which we call LTAU (Loss Trajectory Analysis for Uncertainty), efficiently estimates the full probability distribution function of errors for any test point using the logged training errors, achieving speeds that are 2–3 orders of magnitudes faster than typical ensemble methods and allowing it to be used for tasks where training or evaluating multiple models would be infeasible. We apply LTAU towards estimating parametric uncertainty in atomistic force fields (LTAU-FF), demonstrating that it produces well-calibrated confidence intervals and predicts errors that correlate strongly with the true errors for data near the training domain. Furthermore, we show that the errors predicted by LTAU-FF can be used in practical applications for detecting out-of-domain data, tuning model performance, and predicting failure during simulations. We believe that LTAU will be a valuable tool for uncertainty quantification in atomistic force fields and is a promising method that should be further explored in other domains of machine learning.
- 13 January 2025
- Published by Elsevier in Acta Materialia
- Vol. 286
Fundamental microscopic properties as predictors of large-scale quantities of interest: Validation through grain boundary energy trends
- 13 January 2025
- Published by Elsevier in Acta Materialia
- Vol. 286
Fundamental microscopic properties as predictors of large-scale quantities of interest: Validation through grain boundary energy trends