Pol Sanz Berman

The design of low-temperature fuel cells and electrolyzers involving proton exchange membranes (PEMs) has been dominated by the exceptional performance and industrial dominance of Nafion. With emerging regulations banning the use of fluorinated compounds, the search for F-free alternatives must be accelerated. Here, we optimize a fluorine-free, sustainable, and low-cost liquid crystalline poly(epichlorohydrin)membrane, based on the identification of key descriptors for proton mobility obtained using machine learning interatomic potentials (MLIPs) derived from first-principles calculations for extensive molecular dynamics. Proton transfer is controlled by the interplay between water confinement and internal pore chemistry. This allows us to modify the active functional groups so that they directly participate in the proton diffusion process, increasing three times the diffusion rate in the membrane, thus being closer to Nafion standards. This work provides key descriptors for a targeted design approach essential for catalyzing the commercial development of next-generation fluorine-free PEMs.

Machine learning interatomic potentials enable atomistic simulations at near first-principles accuracy. As model architectures mature, their reliability is increasingly constrained by training data quality. Here, we introduce Automated Training with Latent-space-Aware Sampling, ATLAS, a unified and multifunctional framework based on reversibly compressed latent space and online reliability verification. By co-designing diversity and efficiency criteria with a diversity-aware database generator and a manifold-aware active learning workflow, we steer exploitation toward sparse configurational regions in the structural space while preventing premature convergence. ATLAS produces compact datasets that allow MLIP training with high accuracy reducing up to 300 times fewer structures. To test ATLAS-trained MLIPs, we looked for high-temperature properties such as melting (physical) and decomposition (involving chemical reactions), demonstrating that our methodology consistently brings down prediction errors for temperature below 175 K, for Cu, CuZn and IrO ₂ . ATLAS provides a principled and transferable strategy for building robust, similarly accurate and well-balanced MLIPs across diverse material classes.

Machine learning interatomic potentials (MLIPs) enable atomistic simulations at near first-principles accuracy. As model architectures mature, their reliability is increasingly constrained by training data quality. Here, we introduce ATLAS (Automated Training with Latent-space Aware Sampling), a unified and multifunctional framework based on the reversibly compressed latent space and online reliability verification. By co-designing diversity and efficiency criteria with a diversity-aware database generator and a manifold-aware active learning workflow, we steer exploitation toward sparse configurational regions while preventing premature convergence. Validated on metals, alloys, and metal oxides, ATLAS produces compact datasets exceeding foundation model training chemical spaces by orders of magnitude with up to 317 times fewer structures. ATLAS-MLIPs outperform universal MLIPs by up to 425 K in melting point accuracy for CuZn, while surface energy errors for IrO _x drop from 1.21 to 0.04 J/m ² . ATLAS provides a principled and transferable strategy for building robust MLIPs across diverse material classes.

Assessing the sustainability of plastic chemical recycling requires realistic feedstocks and catalysts designed within sustainability-led frameworks (Plastic-to-X). We link catalyst design and systems analysis to study hydrogenolysis of high-density polyethylene (virgin and bottle caps; M_w = 100–200 kDa). We report Ru–Ni alloy nanoparticles (3–4 nm) supported on titania that yield up to 55% liquid C₆–C₄₅ products under optimized conditions, whereas monometallic Ru produces virtually no liquids Operando spectroscopy and simulations reveal structure sensitivity: backbone scission follows dehydrogenation and hydrogenation cycles at defective alloy sites formed in situ. Integrating these mechanistic insights with life cycle and techno-economic analyses indicates profitable processing of plastic caps over the optimal catalyst (2.5 wt% Ru, 5 wt% Ni) with substantially lower CO₂ emissions even when using green H₂. Furthermore, within the Plastic-to-X framework, we identify a minimum average chain length threshold of C₁₁ for product distributions as a critical design metric to reconcile environmental and economic objectives.

Process optimization in heterogeneous catalysis relies on the control of competing reactions. The reaction mechanisms based on chemical knowledge can be evaluated via density functional theory unveiling experimental catalytic trends. However, this approach finds its limits when applied to complex reaction networks or large molecules, disregarding alternative paths and rare events. Here we present CARE, a foundational model for catalysis on metal surfaces with a rule-based reaction network generator for CxHyOz species built with GAME-Net-UQ, a graph neural network with uncertainty quantification targeting thermodynamic and kinetic parameters, coupled to microkinetic modeling. CARE reproduces experimental activity trends in methanol decomposition, selectivity to C3 products in electrochemical reduction processes, and models the Fischer-Tropsch synthesis to C6 products, including 370k reactions, breaking the current limits of network exploration. This comprehensive model opens the path towards the exploration of thermal and electrocatalytic surface processes previously not amenable to atomistic simulations.

The oxygen evolution reaction is the bottleneck to energy-efficient water-based electrolysis for the production of hydrogen and other solar fuels. In proton exchange membrane water electrolysis (PEMWE), precious metals have generally been necessary for the stable catalysis of this reaction. In this work, we report that delamination of cobalt tungstate enables high activity and durability through the stabilization of oxide and water-hydroxide networks of the lattice defects in acid. The resulting catalysts achieve lower overpotentials, a current density of 1.8 amperes per square centimeter at 2 volts, and stable operation up to 1 ampere per square centimeter in a PEMWE system at industrial conditions (80°C) at 1.77 volts; a threefold improvement in activity; and stable operation at 1 ampere per square centimeter over the course of 600 hours.