Anubhav Jain
@anubhavjain.bsky.social
Mainly research group updates
Staff Scientist at Lawrence Berkeley National Laboratory
https://hackingmaterials.lbl.gov
All views my own
Staff Scientist at Lawrence Berkeley National Laboratory
https://hackingmaterials.lbl.gov
All views my own
Laser-written rotating-lattice crystals of Sb₂S₃ enable microscale orientation-dependent thermal conductivity patterning; κ from 0.6 → 2.5 W m⁻¹ K⁻¹. DFT + Wigner transport show lone-pair-induced anisotropy.
Isotta et al.,
Adv. Funct. Mater.
doi.org/10.1002/adfm...
Isotta et al.,
Adv. Funct. Mater.
doi.org/10.1002/adfm...
Local Thermal Conductivity Patterning in Rotating Lattice Crystals of Anisotropic Sb2S3
Microscale control of thermal conductivity in Sb2S3 is demonstrated via laser-induced rotating lattice crystals. Thermal conductivity imaging reveals marked thermal transport anisotropy, with the c a...
doi.org
October 29, 2025 at 5:10 PM
Laser-written rotating-lattice crystals of Sb₂S₃ enable microscale orientation-dependent thermal conductivity patterning; κ from 0.6 → 2.5 W m⁻¹ K⁻¹. DFT + Wigner transport show lone-pair-induced anisotropy.
Isotta et al.,
Adv. Funct. Mater.
doi.org/10.1002/adfm...
Isotta et al.,
Adv. Funct. Mater.
doi.org/10.1002/adfm...
Happy to collaborate on hashin_shtrikman_mp, a Python tool that combines theoretical bounds, genetic ML optimization, and Materials Project data to design optimal composite formulations from desired properties.
Becker et al., J. Open Source Softw.
doi.org/10.21105/jos...
Becker et al., J. Open Source Softw.
doi.org/10.21105/jos...
hashin_shtrikman_mp: a package for the optimal design and discovery of multi-phase composite materials
Becker et al., (2025). hashin_shtrikman_mp: a package for the optimal design and discovery of multi-phase composite materials. Journal of Open Source Software, 10(114), 8412, https://doi.org/10.21105/...
doi.org
October 16, 2025 at 6:52 PM
Happy to collaborate on hashin_shtrikman_mp, a Python tool that combines theoretical bounds, genetic ML optimization, and Materials Project data to design optimal composite formulations from desired properties.
Becker et al., J. Open Source Softw.
doi.org/10.21105/jos...
Becker et al., J. Open Source Softw.
doi.org/10.21105/jos...
Can machines learn microscopy without labels?
Work with KIT/UCB on self-supervised ConvNeXtV2 achieves ~41% error reduction over untrained models (15% vs ImageNet) for particle segmentation using 25k SEM images.
Rettenberger et al., npj Comp Mater
doi.org/10.1038/s415...
Work with KIT/UCB on self-supervised ConvNeXtV2 achieves ~41% error reduction over untrained models (15% vs ImageNet) for particle segmentation using 25k SEM images.
Rettenberger et al., npj Comp Mater
doi.org/10.1038/s415...
Leveraging unlabeled SEM datasets with self-supervised learning for enhanced particle segmentation - npj Computational Materials
npj Computational Materials - Leveraging unlabeled SEM datasets with self-supervised learning for enhanced particle segmentation
doi.org
October 15, 2025 at 8:53 PM
Can machines learn microscopy without labels?
Work with KIT/UCB on self-supervised ConvNeXtV2 achieves ~41% error reduction over untrained models (15% vs ImageNet) for particle segmentation using 25k SEM images.
Rettenberger et al., npj Comp Mater
doi.org/10.1038/s415...
Work with KIT/UCB on self-supervised ConvNeXtV2 achieves ~41% error reduction over untrained models (15% vs ImageNet) for particle segmentation using 25k SEM images.
Rettenberger et al., npj Comp Mater
doi.org/10.1038/s415...
U.S. PhD students: interested in spending time at Berkeley Lab working with us on AI agents, computational materials design, data-driven synthesis, or the Materials Project?
Check out the DOE SCGSR program: science.osti.gov/wdts/scgsr
If interested and eligible, please reach out!
Check out the DOE SCGSR program: science.osti.gov/wdts/scgsr
If interested and eligible, please reach out!
LinkedIn
This link will take you to a page that’s not on LinkedIn
lnkd.in
September 30, 2025 at 4:32 PM
U.S. PhD students: interested in spending time at Berkeley Lab working with us on AI agents, computational materials design, data-driven synthesis, or the Materials Project?
Check out the DOE SCGSR program: science.osti.gov/wdts/scgsr
If interested and eligible, please reach out!
Check out the DOE SCGSR program: science.osti.gov/wdts/scgsr
If interested and eligible, please reach out!
🚀 We’re hiring a Materials AI Postdoc at Berkeley Lab! Join us in building the next generation of AI for materials discovery, spanning simulations, autonomous labs & DOE supercomputers via AI agents.
Apply here 👉 jobs.lbl.gov/jobs/postdoc...
#AI #MaterialsScience #PostdocJobs
Apply here 👉 jobs.lbl.gov/jobs/postdoc...
#AI #MaterialsScience #PostdocJobs
Postdoctoral Scholar - AI-Driven Materials Discovery in Bay Area, California, United States
Lawrence Berkeley National Lab’s (LBNL) Energy Storage & Distributed Resources Division has an opening for a Postdoctoral Scholar in AI-Driven...
jobs.lbl.gov
September 4, 2025 at 4:42 PM
🚀 We’re hiring a Materials AI Postdoc at Berkeley Lab! Join us in building the next generation of AI for materials discovery, spanning simulations, autonomous labs & DOE supercomputers via AI agents.
Apply here 👉 jobs.lbl.gov/jobs/postdoc...
#AI #MaterialsScience #PostdocJobs
Apply here 👉 jobs.lbl.gov/jobs/postdoc...
#AI #MaterialsScience #PostdocJobs
Electrocatalysts can treat tough water contaminants, but discovery is slow. We review how ML potentials + autonomous screening platforms can accelerate catalyst design for next-gen water purification.
Wang et al., AI for Sci.
doi.org/10.1088/3050...
Wang et al., AI for Sci.
doi.org/10.1088/3050...
Computational catalysis and machine learning applications to water treatment technologies - IOPscience
Computational catalysis and machine learning applications to water treatment technologies, Wang, Duo, Xie, Ao, Ma, Shengcun, Tong, Wei, Zou, Shiqiang, Jain, Anubhav
doi.org
September 3, 2025 at 6:32 PM
Electrocatalysts can treat tough water contaminants, but discovery is slow. We review how ML potentials + autonomous screening platforms can accelerate catalyst design for next-gen water purification.
Wang et al., AI for Sci.
doi.org/10.1088/3050...
Wang et al., AI for Sci.
doi.org/10.1088/3050...
With ~180K materials and millions of calculated properties, the Materials Project enables inverse design, synthesis screening, and discovery. Examples include phosphors, thermoelectrics, electrides, and battery electrolytes.
Horton et al, Nature Materials
doi.org/10.1038/s415...
Horton et al, Nature Materials
doi.org/10.1038/s415...
Accelerated data-driven materials science with the Materials Project - Nature Materials
Materials design and informatics have become increasingly prominent over the past several decades. Using the Materials Project as an example, this Perspective discusses how properties are calculated a...
doi.org
July 11, 2025 at 11:19 PM
With ~180K materials and millions of calculated properties, the Materials Project enables inverse design, synthesis screening, and discovery. Examples include phosphors, thermoelectrics, electrides, and battery electrolytes.
Horton et al, Nature Materials
doi.org/10.1038/s415...
Horton et al, Nature Materials
doi.org/10.1038/s415...
Atomate2 is a fully modular workflow platform for high-throughput DFT and MLIP calculations. Supports ~30 workflows, hybrid DFT/MLIP chaining, defect and phonon automation, & more - collaboration amongst multiple groups!
@virtualatoms.bsky.social et al., Digital Discovery
doi.org/10.1039/D5DD...
@virtualatoms.bsky.social et al., Digital Discovery
doi.org/10.1039/D5DD...
Atomate2: modular workflows for materials science
High-throughput density functional theory (DFT) calculations have become a vital element of computational materials science, enabling materials screening, property database generation, and training of...
doi.org
July 1, 2025 at 10:17 PM
Atomate2 is a fully modular workflow platform for high-throughput DFT and MLIP calculations. Supports ~30 workflows, hybrid DFT/MLIP chaining, defect and phonon automation, & more - collaboration amongst multiple groups!
@virtualatoms.bsky.social et al., Digital Discovery
doi.org/10.1039/D5DD...
@virtualatoms.bsky.social et al., Digital Discovery
doi.org/10.1039/D5DD...
MLIP evaluation: Matbench Discovery focuses on predicting stability; universal interatomic potentials (UIPs) are top performers w/ ~5X improvement in discovery efficiency. Regression accuracy not the same as discovery!
Riebesell et al., Nat. Mach. Intell.
doi.org/10.1038/s422...
Riebesell et al., Nat. Mach. Intell.
doi.org/10.1038/s422...
A framework to evaluate machine learning crystal stability predictions - Nature Machine Intelligence
Riebesell et al. introduce Matbench Discovery, a framework to compare machine learning models used to identify stable crystals. Out of several architectures, they find that universal interatomic poten...
doi.org
June 26, 2025 at 6:51 PM
MLIP evaluation: Matbench Discovery focuses on predicting stability; universal interatomic potentials (UIPs) are top performers w/ ~5X improvement in discovery efficiency. Regression accuracy not the same as discovery!
Riebesell et al., Nat. Mach. Intell.
doi.org/10.1038/s422...
Riebesell et al., Nat. Mach. Intell.
doi.org/10.1038/s422...
PV-Pro detects off-MPP behavior in solar arrays using real-time modeling that accounts for system degradation. Analyzing a 271 kW array, ~5% of points are detected as off-MPP, largely due to current loss.
Li et al, IEEE PVSC
https://doi.org/10.1109/PVSC48320.2023.10359868
Li et al, IEEE PVSC
https://doi.org/10.1109/PVSC48320.2023.10359868
June 9, 2025 at 10:54 PM
PV-Pro detects off-MPP behavior in solar arrays using real-time modeling that accounts for system degradation. Analyzing a 271 kW array, ~5% of points are detected as off-MPP, largely due to current loss.
Li et al, IEEE PVSC
https://doi.org/10.1109/PVSC48320.2023.10359868
Li et al, IEEE PVSC
https://doi.org/10.1109/PVSC48320.2023.10359868
RuO₂-based catalysts remove >90% Se(IV) in wastewater (8 hours). DFT shows Sn doping lowers the energy barrier for reduction by stabilizing intermediates, explaining the superior activity of Ru₀.₉Sn₀.₁Oₓ/TP over pure RuO₂.
Hao et al, Nano Lett.
https://doi.org/10.1021/acs.nanolett.4c06344
Hao et al, Nano Lett.
https://doi.org/10.1021/acs.nanolett.4c06344
June 9, 2025 at 10:54 PM
RuO₂-based catalysts remove >90% Se(IV) in wastewater (8 hours). DFT shows Sn doping lowers the energy barrier for reduction by stabilizing intermediates, explaining the superior activity of Ru₀.₉Sn₀.₁Oₓ/TP over pure RuO₂.
Hao et al, Nano Lett.
https://doi.org/10.1021/acs.nanolett.4c06344
Hao et al, Nano Lett.
https://doi.org/10.1021/acs.nanolett.4c06344
BiFeO3 synthesis: simulations indicate that Bi nitrate + 2ME form stable dimers via nitrite bridges, contrary to the assumed full solvation route. Text mining shows precursors most often leading to phase-purity.
Baibakova & Cruse et al, Digital Discovery
https://doi.org/10.1039/d5dd00160a
Baibakova & Cruse et al, Digital Discovery
https://doi.org/10.1039/d5dd00160a
June 9, 2025 at 10:54 PM
BiFeO3 synthesis: simulations indicate that Bi nitrate + 2ME form stable dimers via nitrite bridges, contrary to the assumed full solvation route. Text mining shows precursors most often leading to phase-purity.
Baibakova & Cruse et al, Digital Discovery
https://doi.org/10.1039/d5dd00160a
Baibakova & Cruse et al, Digital Discovery
https://doi.org/10.1039/d5dd00160a
Using 492 text-mined AuNP syntheses, we show that precursor choice (e.g., CTAB vs citrate) can accurately classify final NP morphology (e.g., rod, cube). But even “identical” recipes can yield 86% difference in aspect ratio.
Lee et al, Digital Discovery
https://doi.org/10.1039/d4dd00158c
Lee et al, Digital Discovery
https://doi.org/10.1039/d4dd00158c
June 9, 2025 at 10:54 PM
Using 492 text-mined AuNP syntheses, we show that precursor choice (e.g., CTAB vs citrate) can accurately classify final NP morphology (e.g., rod, cube). But even “identical” recipes can yield 86% difference in aspect ratio.
Lee et al, Digital Discovery
https://doi.org/10.1039/d4dd00158c
Lee et al, Digital Discovery
https://doi.org/10.1039/d4dd00158c
AlabOS is a Python-based framework for managing autonomous materials labs. Supports modular DAG workflows, device/resource coordination, and real-time tracking; used to synthesize >3500 samples at LBNL in 1.5 years.
Fei & Rendy et al, Digital Discovery
https://doi.org/10.1039/d4dd00129j
Fei & Rendy et al, Digital Discovery
https://doi.org/10.1039/d4dd00129j
June 9, 2025 at 10:54 PM
AlabOS is a Python-based framework for managing autonomous materials labs. Supports modular DAG workflows, device/resource coordination, and real-time tracking; used to synthesize >3500 samples at LBNL in 1.5 years.
Fei & Rendy et al, Digital Discovery
https://doi.org/10.1039/d4dd00129j
Fei & Rendy et al, Digital Discovery
https://doi.org/10.1039/d4dd00129j
A curated review of >50 open PV degradation data sets: environmental, performance, imaging, and materials. We highlight ML-ready(ish) sets, image benchmarks, analysis tools, and where fragmentation still hinders progress.
Chen & Li et al, Appl. Energy
https://doi.org/10.1016/j.apenergy.2025.126132
Chen & Li et al, Appl. Energy
https://doi.org/10.1016/j.apenergy.2025.126132
June 9, 2025 at 10:54 PM
A curated review of >50 open PV degradation data sets: environmental, performance, imaging, and materials. We highlight ML-ready(ish) sets, image benchmarks, analysis tools, and where fragmentation still hinders progress.
Chen & Li et al, Appl. Energy
https://doi.org/10.1016/j.apenergy.2025.126132
Chen & Li et al, Appl. Energy
https://doi.org/10.1016/j.apenergy.2025.126132
A review of recent developments in machine learning in materials science (growing at ~1.67 yearly for the last decade). Focus on tools/data sets/improvements particularly for inorganic materials property prediction.
Jain, Curr Opinion Sol State & Mat Sci
https://doi.org/10.1016/j.cossms.2024.101189
Jain, Curr Opinion Sol State & Mat Sci
https://doi.org/10.1016/j.cossms.2024.101189
June 9, 2025 at 10:54 PM
A review of recent developments in machine learning in materials science (growing at ~1.67 yearly for the last decade). Focus on tools/data sets/improvements particularly for inorganic materials property prediction.
Jain, Curr Opinion Sol State & Mat Sci
https://doi.org/10.1016/j.cossms.2024.101189
Jain, Curr Opinion Sol State & Mat Sci
https://doi.org/10.1016/j.cossms.2024.101189
DFT-based phonon calculations are expensive particularly for higher-order interactions. Our recent paper shows that it is possible to automate them at 100-1000X speedup using recent fitting tools (originally HipHive, now pheasy):
Zhu et al, npj Comp Mat
https://doi.org/10.1038/s41524-024-01437-w
Zhu et al, npj Comp Mat
https://doi.org/10.1038/s41524-024-01437-w
June 9, 2025 at 10:54 PM
DFT-based phonon calculations are expensive particularly for higher-order interactions. Our recent paper shows that it is possible to automate them at 100-1000X speedup using recent fitting tools (originally HipHive, now pheasy):
Zhu et al, npj Comp Mat
https://doi.org/10.1038/s41524-024-01437-w
Zhu et al, npj Comp Mat
https://doi.org/10.1038/s41524-024-01437-w
Announcement: Are you a current PhD candidate interested in working in our group at Berkeley Lab? You can be funded to do so via the DOE SCGSR program - please contact me if this is of interest.
https://science.osti.gov/wdts/scgsr/
(program restricted to U.S. citizen & permanent residents)
https://science.osti.gov/wdts/scgsr/
(program restricted to U.S. citizen & permanent residents)
June 9, 2025 at 10:54 PM
Announcement: Are you a current PhD candidate interested in working in our group at Berkeley Lab? You can be funded to do so via the DOE SCGSR program - please contact me if this is of interest.
https://science.osti.gov/wdts/scgsr/
(program restricted to U.S. citizen & permanent residents)
https://science.osti.gov/wdts/scgsr/
(program restricted to U.S. citizen & permanent residents)
If anyone is interested in learning pymatgen in 2025, I posted a series of video tutorials on it. You can go from beginner to pymatgen wizard in just a couple of hours:
https://www.youtube.com/playlist?list=PL7gkuUui8u7_M47KrV4tS4pLwhe7mDAjT
https://www.youtube.com/playlist?list=PL7gkuUui8u7_M47KrV4tS4pLwhe7mDAjT
June 9, 2025 at 10:54 PM
If anyone is interested in learning pymatgen in 2025, I posted a series of video tutorials on it. You can go from beginner to pymatgen wizard in just a couple of hours:
https://www.youtube.com/playlist?list=PL7gkuUui8u7_M47KrV4tS4pLwhe7mDAjT
https://www.youtube.com/playlist?list=PL7gkuUui8u7_M47KrV4tS4pLwhe7mDAjT
Predicting the expected power output of solar PV modules as they degrade can be challenging. The PV-Pro tool can model the internal state of a degraded module to provide accurate estimates of expected power (>17% improvement).
Li et al, Renewable Energy
https://doi.org/10.1016/j.renene.2024.121493
Li et al, Renewable Energy
https://doi.org/10.1016/j.renene.2024.121493
June 9, 2025 at 10:54 PM
Predicting the expected power output of solar PV modules as they degrade can be challenging. The PV-Pro tool can model the internal state of a degraded module to provide accurate estimates of expected power (>17% improvement).
Li et al, Renewable Energy
https://doi.org/10.1016/j.renene.2024.121493
Li et al, Renewable Energy
https://doi.org/10.1016/j.renene.2024.121493
A short perspective (I was a co-author) on opportunities for LLMs in alloy design, headed by Zongrui Pei. Even seemingly small things like alloy naming standardization may be useful!
Pei et al, Nature Reviews Materials
https://doi.org/10.1038/s41578-024-00726-6
Pei et al, Nature Reviews Materials
https://doi.org/10.1038/s41578-024-00726-6
June 9, 2025 at 10:54 PM
A short perspective (I was a co-author) on opportunities for LLMs in alloy design, headed by Zongrui Pei. Even seemingly small things like alloy naming standardization may be useful!
Pei et al, Nature Reviews Materials
https://doi.org/10.1038/s41578-024-00726-6
Pei et al, Nature Reviews Materials
https://doi.org/10.1038/s41578-024-00726-6
with Wei Tong & @EmoryChanNano. Also shout-out to @zackulissi for the collaborative computational screening effort for identifying nitrate reduction electrocatalysts prior to this work.
June 9, 2025 at 10:54 PM
with Wei Tong & @EmoryChanNano. Also shout-out to @zackulissi for the collaborative computational screening effort for identifying nitrate reduction electrocatalysts prior to this work.
We've been computationally screening electrocatalysts for oxyanion remediation in water. Our collaborators at LBL describe an experimental platform for testing, with some results on Ag-Ni alloys for nitrate reduction.
Ma et al, ACS Appl Energy Mater
https://doi.org/10.1021/acsaem.4c00631
Ma et al, ACS Appl Energy Mater
https://doi.org/10.1021/acsaem.4c00631
June 9, 2025 at 10:54 PM
We've been computationally screening electrocatalysts for oxyanion remediation in water. Our collaborators at LBL describe an experimental platform for testing, with some results on Ag-Ni alloys for nitrate reduction.
Ma et al, ACS Appl Energy Mater
https://doi.org/10.1021/acsaem.4c00631
Ma et al, ACS Appl Energy Mater
https://doi.org/10.1021/acsaem.4c00631
We demonstrate that LLMs can accelerate data extraction from unstructured text (journal articles) using an iterative training procedure. Since the preprint, we show that Llama-2 results are close to GPT-3 results for the task.
Dagdelen et al, Nat Comm
https://doi.org/10.1038/s41467-024-45563-x
Dagdelen et al, Nat Comm
https://doi.org/10.1038/s41467-024-45563-x
June 9, 2025 at 10:54 PM
We demonstrate that LLMs can accelerate data extraction from unstructured text (journal articles) using an iterative training procedure. Since the preprint, we show that Llama-2 results are close to GPT-3 results for the task.
Dagdelen et al, Nat Comm
https://doi.org/10.1038/s41467-024-45563-x
Dagdelen et al, Nat Comm
https://doi.org/10.1038/s41467-024-45563-x
By text mining BiFeO3 synthesis, we quantify details typically missing from published synthesis descriptions (e.g. mixing temp & time). We next test if we can achieve phase pure synthesis using imputed values for past experiments.
Cruse et al, Chem Mat
https://doi.org/10.1021/acs.chemmater.3c02203
Cruse et al, Chem Mat
https://doi.org/10.1021/acs.chemmater.3c02203
June 9, 2025 at 10:54 PM
By text mining BiFeO3 synthesis, we quantify details typically missing from published synthesis descriptions (e.g. mixing temp & time). We next test if we can achieve phase pure synthesis using imputed values for past experiments.
Cruse et al, Chem Mat
https://doi.org/10.1021/acs.chemmater.3c02203
Cruse et al, Chem Mat
https://doi.org/10.1021/acs.chemmater.3c02203