Officially launched in 2014, MARVEL aimed to fundamentally alter materials research by integrating quantum-mechanical simulations with the burgeoning power of high-performance computing and an emerging force in science: machine learning. This innovative integration sought not only to accelerate the pace of materials innovation but also to make it more predictive, systematic, reproducible, and inherently collaborative. The foresight behind this strategy has proved remarkably prescient, positioning computational materials science at the core of modern scientific inquiry and technological progress.

Nicola Marzari, MARVEL’s director and a professor at EPFL, reflects on the initiative’s continued relevance: major technology companies such as Google DeepMind, Microsoft, and Meta have recently ramped up efforts in materials design, paralleled by an influx of startups securing substantial investments exceeding hundreds of millions of dollars in early-stage funding. Moreover, the concept of agentic artificial intelligence, included in MARVEL’s initial vision, has now evolved into fully operational systems that are poised to revolutionize how materials are discovered and optimized.

Historically, materials science progressed through labor-intensive trial and error—synthesize, test, fail, and repeat. MARVEL disrupted this paradigm by weaving together expertise from physics, chemistry, computer science, and machine learning with experimental validation. This multidisciplinary approach enabled researchers to pre-select promising candidate materials computationally, dissect the mechanisms underpinning their behavior, and promote openness through sharing data and computational methods. As a result, MARVEL fostered a culture where discoveries could be systematically reproduced and extended across the global scientific community.

Across its twelve years, MARVEL contributed to key scientific advances in a variety of cutting-edge materials. Its researchers predicted and facilitated the experimental confirmation of novel quantum materials exhibiting unconventional electronic states highly promising for next-generation electronic devices and quantum information technologies. These breakthroughs provided not only isolated discoveries but also distilled complex phenomena into clear design principles, guiding the creation of materials with highly tailored properties for future technological applications.

At the software and methods level, MARVEL made enduring contributions by developing advanced electronic-structure techniques crucial for accurate quantum simulations. These methods were incorporated into open-source computational frameworks widely adopted by scientists worldwide, democratizing access to sophisticated simulation tools. In parallel, MARVEL was a trailblazer in embedding machine learning algorithms within materials modeling pipelines, creating AI systems capable of predicting properties from atomic scale structures with unprecedented accuracy.

The evolution of machine learning within MARVEL’s scope is particularly striking. Initially a promising adjunct, machine learning matured into a core computational instrument capable of accelerating atomistic simulations and forecasting complex materials properties, including spectroscopic signatures, chemical environments, electronic structures, diffusion dynamics, and more. This advancement laid the groundwork for today’s explosive interest in AI-powered materials research, elevating computational models from theoretical tools to practical engines of discovery.

Beyond fundamental science, MARVEL tackled materials challenges with tangible societal impact. The initiative’s portfolio spanned solar energy harvesting materials, catalysts for water splitting, solid-state battery electrolytes, and nanoporous materials for selective separations. It included molecular crystals relevant to pharmaceutical and chemical industries, ultrathin two-dimensional materials, and high-performance aerospace alloys. In its later phases, MARVEL expanded into spectroscopy, automated experimental platforms, and hybrid quantum-classical algorithms, continuously broadening the frontiers of computational materials research.

Integral to MARVEL’s philosophy was a commitment to open science and transparency. By prioritizing open-source software, verification protocols, and reproducibility standards, the initiative fostered greater reliability in computational results. It demonstrated that from simulations to experimental data, workflows could be integrated seamlessly into reproducible pipelines, sometimes even enabling computational steering of automated laboratory experiments. These pioneering efforts set new norms for how materials research should be conducted in the digital era.

A central pillar of MARVEL’s legacy is the establishment of a national Swiss digital ecosystem for materials science. This network, originally centered at EPFL and incorporating ETH Zurich, PSI, Empa, CSCS, and various universities, culminated in a robust platform supporting computational materials research nationwide. Critical components of this infrastructure include AiiDA and AiiDAlab, platforms facilitating reproducible workflows; the Materials Cloud, a portal for data dissemination and open access; and Lhumos, an innovative educational toolset designed to train new generations of computational materials scientists.

These digital infrastructures empower researchers to execute complex computational experiments with ease, meticulously track and record every analytical step, compare outputs from diverse simulation engines, publish datasets for reusability, and convert expert workflows into accessible analytical tools. This democratization and standardization have not only accelerated research but also fostered a global community united by shared resources and collaborative spirit.

MARVEL’s influence extends beyond academia, forging meaningful partnerships with a diverse industrial community spanning sectors such as energy, electronics, metallurgy, catalysis, and pharmaceuticals. The initiative cultivated collaborations with dozens of companies and shifted software development towards creating practical tools adaptable to industrial contexts. This transition from specialist-oriented software to industry-grade applications underscores MARVEL’s role in bridging scientific innovation with commercial utilization.

As the MARVEL program comes to a close, its twelve-year journey will be commemorated on 9 July at EPFL’s Rolex Forum in Lausanne with a full-day event gathering leading figures from academia and industry. Discussions will span quantum materials, design methodologies, machine learning advancements, and the expanding computational materials science community. The event will bring together luminaries from Europe, North America, China, and industry stakeholders such as Microsoft, BASF, and Stellantis, marking a celebration of both accomplishments and future prospects.

Ultimately, MARVEL has cemented a vision where materials of the future are increasingly imagined, optimized, shared, and validated digitally before ever undergoing synthesis in the laboratory. This digital-first paradigm has shifted how materials science is approached worldwide and has positioned Switzerland at the forefront of this scientific revolution. By championing open science, cutting-edge computational tools, and interdisciplinary collaboration, MARVEL has transformed materials research from an art of trial and error into a rigorous, data-driven endeavor poised to accelerate technological breakthroughs across a broad spectrum of fields.

Subject of Research: Computational design and discovery of novel materials.

Article Title: MARVEL at 12: Charting the Digital Revolution in Materials Science.

News Publication Date: Not specified in content.

Web References:

• https://nccr-marvel.ch/events/2026-07-marvel-epfl
• https://aiida.net/
• https://www.aiidalab.net/
• https://www.materialscloud.org/
• https://www.lhumos.org/

Keywords
Materials science, computational materials discovery, quantum materials, electronic structure, machine learning, artificial intelligence, quantum chemistry, simulation reproducibility, high-performance computing, energy materials, automated experiments, open-source software, digital scientific ecosystem.

Identifying materials that facilitate rapid ion conduction within solid electrolytes traditionally hinges on laborious experiments and computational strategies that struggle to contend with the dynamic, sometimes chaotic atomic environments in these compounds. Classical computational methods, despite their rigorous physical foundations, become extraordinarily resource-intensive and often impractical when applied to disordered or high-temperature ionic motions intrinsic to functioning solid electrolytes. This gap in accessible diagnostic tools has left a significant bottleneck in the exploration and rapid discovery of superionic conductors, materials where ions exhibit liquid-like mobility through otherwise crystalline lattices.

A transformative breakthrough emerges in the form of a novel machine learning (ML) accelerated framework, ingeniously designed to tackle the challenge of capturing and interpreting the subtle spectroscopic fingerprints of ion dynamics in solids. By synergizing ML-driven force fields with advanced tensor-based ML models trained to predict Raman spectra, researchers have unlocked a pathway to simulate vibrational characteristics of complex, dynamically disordered materials with near first-principles accuracy. This approach hulks over computational costs while preserving precision, facilitating rapid, predictive insights into ion conduction phenomena that were previously out of reach due to computational constraints.

The heart of this methodology lies in recognizing the unique impact of liquid-like ionic motion on the host material’s vibrational and symmetry properties. As mobile ions journey through the crystal lattice, their motion disrupts local symmetry patterns, leading to a relaxation of traditional Raman selection rules—a fundamental concept dictating which vibrational modes are active or inactive in Raman spectroscopy. This dynamical symmetry breaking manifests as pronounced low-frequency Raman scattering peaks, serving as direct, spectroscopic hallmarks of rapid ionic diffusion. The ability to correlate these low-frequency Raman features with ion mobility ushers in a new spectroscopic paradigm for diagnosing and understanding fast ion conduction in solid electrolytes.

In practical terms, this ML-accelerated Raman calculation workflow was rigorously tested on sodium-ion conductors exemplified by materials such as Na3SbS4. Extensive simulations revealed that systems with distinct, intense low-frequency Raman intensity features coincide with high ionic diffusivity, a signature of liquid-like conduction mechanisms and the underlying relaxational dynamics of the host lattice. Conversely, materials dominated by traditional hopping conduction of ions, lacking this dynamic lattice disruption, failed to exhibit these Raman signatures. This finding not only validates the computational approach but also bridges a crucial understanding gap between observable spectroscopic phenomena and the underlying ion transport physics.

Crucially, the framework transcends previous limitations confined to well-characterized superionic compounds, offering a unifying theory that extends the interpretation of diffusive Raman scattering to a wider spectrum of material classes. This generalization implicates that the breakdown of Raman selection rules, driven by complex ionic mobility and lattice dynamics, can be a universal descriptor of fast ion transport across disparate solid electrolytes. From a broader materials discovery perspective, this insight is highly potent, enabling the high-throughput screening of novel superionic materials through a spectroscopic lens, dramatically accelerating the pipeline from theoretical prediction to experimental realization.

Beyond its computational elegance, this work harmonizes theoretical atomistic models with experimental observables, forging a tight feedback loop that could revolutionize the characterization of solid electrolytes. By harnessing ML to handle vibrational spectral predictions at finite temperatures, researchers effectively decode complex dynamical behaviors intrinsic to working battery materials, opening a roadmap to design electrolytes with tailored ionic conductivities. This advance is pivotal for scaling solid-state battery technologies that promise safer, longer-lasting energy storage solutions critical for electric vehicles, portable electronics, and grid applications.

The validation of this approach within sodium-ion systems offers not just a proof of concept but a tangible toolset applicable to diverse battery chemistries, including promising lithium and other multivalent ion conductors. Since the ionic conduction mechanisms and lattice symmetries vary widely across material families, the ML model’s adaptability to these variances underscores its robustness and transformative potential. Researchers can now systematically screen large databases of candidate materials, filtering through vibrational spectral data to flag those with desired ionic mobility signatures—thereby prioritizing compounds for synthesis and experimental testing.

At its core, the research embodies a paradigm shift from conventional methods that rely heavily on direct computationally expensive molecular dynamics or experimental trial and error, towards data-driven insight powered by artificial intelligence. This shift is emblematic of a broader movement within materials science toward integrating AI and ML tools to accelerate discovery and deepen fundamental understanding. By extending this framework, scientists anticipate uncovering hidden correlations between ionic dynamics, lattice perturbations, and emergent material properties—insights that will feed back into improved material design principles.

In sum, this pioneering study illuminates the intricate tapestry of fast ionic conduction with unprecedented clarity, harnessing machine learning to unveil spectroscopic signatures that were previously elusive. The implications are far-reaching: from enabling safer, higher-performance solid-state batteries to inspiring new research directions that leverage AI for materials innovation. As energy storage technologies are thrust into ever-increasing demand by renewable energy integration and electrification trends, tools that bridge theory and experiment with such efficiency become indispensable cornerstones of the future scientific enterprise.

Published in the cutting-edge journal AI for Science, this work is poised to influence a wide community spanning computational chemists, materials scientists, and battery engineers. It sets a new benchmark by demonstrating how synergistic combinations of ML-accelerated simulations and experimental spectroscopy can decode the complexity of ion transport dynamics. The pathway forged here not only refines our microscopic understanding but also equips researchers with a practical instrument for the rapid evaluation and rational design of next-generation fast-ion conductors.

The authors behind this breakthrough—affiliated with the Technical University of Munich and collaborators—highlight the collaborative and interdisciplinary nature of modern materials research, where AI, physics, chemistry, and engineering converge. Their contributions propel the field into a new era where discovering the future’s battery materials is no longer bottlenecked by computational limitations or ambiguous experimental interpretations but is driven by intelligent, automated predictive tools. This heralds an exciting chapter in the journey to renewable energy solutions anchored by advanced solid-state battery platforms.

Subject of Research: Fast ionic conduction in solid electrolytes and machine learning-accelerated Raman spectral analysis

Article Title: Revealing fast ionic conduction in solid electrolytes through machine learning accelerated Raman calculations

News Publication Date: 18 February 2026

Web References: https://dx.doi.org/10.1088/3050-287X/ae411a

References: Manuel Grumet, Takeru Miyagawa, Olivier Pittet, Paolo Pegolo, Karin S Thalmann, Waldemar Kaiser, David A Egger. Revealing fast ionic conduction in solid electrolytes through machine learning accelerated Raman calculations[J]. AI for Science, 2026, 2(1): 011001. DOI: 10.1088/3050-287X/ae411a

Image Credits: Dr. Manuel Grumet, Dr. Waldemar Kaiser from Technical University of Munich

Keywords

Solid state chemistry, machine learning, ionic conduction, Raman spectroscopy, solid electrolytes, superionic conductors, battery materials, sodium-ion conductors, AI accelerated simulations, vibrational spectroscopy