The JARVIS Infrastructure is All You Need for Materials Design

Kamal Choudhary · 2025

Read The JARVIS Infrastructure is All You Need for Materials Design by Kamal Choudhary (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-04133.txt.

Introduction

Kamal Choudhary surveys JARVIS, a unified platform for designing materials by computer, built to carry a project from first guess to tested candidate. His 2025 paper describes the Joint Automated Repository for Various Integrated Simulations. It spans the full toolkit: quantum methods at several depths, classical force fields, and machine learning.

Experimental threads plug in as well, from microscopy and diffraction to cryogenics. Datasets, tools, benchmarks, and web apps sit open to the community, with reproducibility as a stated aim. The platform covers both forward design, predicting a material's traits, and the inverse problem of naming a material for wanted traits. Choudhary frames it as a bridge between computation and the lab bench.

Catalog record

CREATORKamal Choudhary
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE44 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-04133

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

The JARVIS Infrastructure is All You Need for Materials Design Kamal Choudhary Abstract Joint Automated Repository for Various Integrated Simulations (JARVIS) is a comprehensive infrastructure offering databases, tools, tutorials, and benchmarks for multiscale, multimodal, forward, and inverse materials design. Emphasizing open access principles and reproducibility, it integrates theoretical and experimental methodologies such as density functional theory, quantum Monte Carlo, tight-binding, classical force fields, and machine-learning approaches-including fingerprinting, graph neural networks, and transformer models. Its experimental data collection spans cryogenics, microscopy, and diffraction, covering materials like metals, semiconductors, insulators, superconductors, carbon capture systems, high-strength compounds, and low-dimensional materials, heterostructures and defects. JARVIS disseminates resources via open datasets, web applications, executable scripts, and peer-reviewed publications, ensuring broad accessibility and reproducibility. Widely adopted worldwide, it has facilitated millions of data and tool downloads. By unifying diverse methods and data under one platform, JARVIS drives both fundamental discoveries and real-world innovations, advancing conventional and data-driven materials design. 1 Introduction Materials design has long been a cornerstone of technological progress, underpinning advancements in energy, computing, healthcare, and manufacturing [ 1 ] . Traditionally, materials discovery followed an empirical, trial-and-error approach, often requiring decades to identify and optimize materials for specific applications. The advent of computational materials science, particularly with the rise of density functional theory (DFT), molecular dynamics (MD)/force-field (FF) and machine learning (ML) has significantly accelerated the pace of discovery [ 2 , 3 , 4 ] . However, the growing complexity of materials problems, the vast chemical and structural search space, and the increasing demand for high-performance materials necessitate more sophisticated, integrated frameworks that can streamline the entire process from fundamental property prediction to real-world implementation. An integrated materials design infrastructure bridges the gap between theoretical models, computational simulations, and experimental validation, ensuring that materials research is reproducible, scalable, and accessible. Moreover, multiscale modeling [ 5 ] capabilities available in these infrastructures enable realistic materials design. Such frameworks incorporate multimodal datasets, automated workflows, and ML models, enabling both forward design (predicting properties from structures) [ 6 , 7 , 8 , 9 ] and inverse design (identifying structures with desired properties) [ 10 , 11 , 12 , 13 ] . Forward design remains challenging due to the need for high-throughput calculations across multiple material classes, while inverse design requires advanced optimization techniques, generative models, and robust experimental validation. One of the critical challenges in forward materials design is the computational cost associated with methods such as first-principles simulations. While DFT and beyond methods are widely used for predicting electronic and structural properties, it is limited by system size and accuracy trade-offs, particularly for strongly correlated materials and high-temperature phases [ 14 ] . In classical approaches, empirical force fields enable large-scale simulations but often lack transferability across different materials and conditions [ 15 ] . More recently, machine learning property prediction models and machine learning force fields (MLFFs) have emerged as a promising alternative, offering near-DFT accuracy at significantly lower computational cost, yet challenges remain in ensuring generalizability and robustness [ 16 , 17 ] . For experimental methods, availability of standardized benchmarks, integration of analytical and computational methods at various lengths and time scales and framework to enhance reproducibility is still a critical challenge [ 18 ] . Inverse design, on the other hand, relies heavily on data-driven methodologies, requiring extensive databases of materials properties and efficient search algorithms [ 13 , 19 ] . Traditional approaches, such as genetic algorithms and evolutionary strategies, have been complemented by deep learning models, which can generate candidate materials based on target functionalities [ 20 ] . However, a major bottleneck is the integration of experimental validation with computational predictions, as synthetic feasibility and stability assessments are often overlooked in algorithmic designs [ 21 ] . Additionally, the lack of standardized benchmarking protocols makes it difficult to compare different methods and validate new materials against experimental findings [ 22 ] . Given these challenges, an effective materials design infrastructure must not only provide computational tools but also enable reproducibility, findability, accessibility, interoperability, and reusability (FAIR) [ 23 ] data practices, and seamless integration of theoretical, computational, and experimental methods. The development of platforms such as Joint Automated Repository for Various Integrated Simulations (JARVIS), which integrates quantum and classical calculations, machine learning models, and experimental datasets, addresses these gaps by offering a unified ecosystem for forward and inverse materials design. By standardizing workflows, automating simulations, and facilitating community-driven data sharing, such infrastructures pave the way for accelerated discoveries and technological breakthroughs in materials science. The JARVIS infrastructure [ 4 , 24 ] distinguishes itself from other materials design frameworks by offering a multimodal, scalable, and reproducible ecosystem that integrates first-principles calculations, machine learning models, experimental datasets, and web-based tools into a unified platform. Databases such as Automatic FLOW for Materials Discovery (AFLOW) [ 25 ] , Open Quantum Materials Database (OQMD) [ 10 ] , Novel Materials Discovery (NOMAD) [ 26 ] , Materials Cloud [ 27 ] and open catalyst project (OC20) [ 28 ] focus primarily on density functional theory (DFT)-based materials screening. These initiatives have fostered a materials-by-design paradigm, where advanced computation, automation, and artificial intelligence (AI) accelerate the transition from fundamental discoveries to industrial applications for next-generation materials innovation. JARVIS extends its capabilities beyond DFT calculations to include classical force-field, machine learning force fields (MLFF), beyond DFT methods such as quantum Monte Carlo and dynamical mean field theory (DMFT). It also covers all major classes of materials such as superconductors, solar cells, topological materials, low dimensional, catalytic, high-strength, thermoelectric, piezoelectric, dielectric discovery, quantum computing algorithms, and experimental validation datasets, making it a truly multimodal and a multiscale infrastructure. Another key differentiator is reproducibility, which is central to JARVIS through the use of open-access datasets, FAIR-compliant data sharing, and the distribution of workflows via interactive web apps and notebooks, benchmarking platform such as JARVIS-Leaderboard ensuring that users can verify and extend existing research [ 29 ] . Additionally, JARVIS is designed for scalability, providing not only extensive high-throughput datasets but also machine learning models capable of accelerating materials discovery at significantly reduced computational cost compared to traditional methods [ 30 ] . Unlike various other platforms, which often focus on a single domain (e.g., semiconductors or metals), JARVIS covers a broad range of materials, including superconductors, carbon capture materials, and low-dimensional heterostructures, facilitating both forward and inverse design approaches. Its seamless integration of theoretical and experimental techniques-spanning quantum Monte Carlo, classical force fields, microscopy, and diffraction databases ensures that the infrastructure is not only predictive but also experimentally validated, bridging the gap between computational discoveries and real-world applications. With thousands of users, millions of dataset downloads, and growing adoption across academia, industry, and government research, JARVIS has established itself as a comprehensive infrastructure that accelerates materials innovation through automation, interoperability, and AI-driven insights. In this article, we briefly discuss some of the components, impacts, and outreach of the JARVIS infrastructure. For more detailed information, we refer to our latest references such as [ 4 , 24 ] . 2 Overview of the JARVIS Infrastructure Figure 1: Core Components of the NIST JARVIS Infrastructure. This schematic highlights key elements of JARVIS, including databases for structural, electronic, mechanical, and topological properties, tools for AI/ML

Read the complete text