Publications

Recent research output, papers, and preprints from our group.

Binary Optimization of Measurement Groupings for Quantum Energy Estimation

Isaac L. Huidobro-Meezs and Rodrigo A. Vargas-Hernández (2026)

Repeated measurements can dominate the resources required for quantum energy estimation, making the choice of which Pauli observables to measure together a central optimization problem for variational quantum algorithms. We formulate fully commuting measurement grouping as a classical binary optimization problem based on clique selection and construct non-overlapping groups using mixed-integer linear programming (MILP). For a benchmark of molecular Hamiltonians, groupings optimized with approximate covariances reduce the non-overlapping measurement requirement ε²M by 51.8% on average relative to sorted insertion (SI). Using the MILP groups to initialize iterative coefficient splitting (ICS), denoted MILP-ICS, yields an average 24.3% reduction relative to ICS initialized from SI (SI-ICS). The optimized groups also transfer across nearby molecular geometries while preserving substantial measurement savings. We further introduce O-clique, which directly optimizes overlapping commuting supports through candidate-clique selection and coefficient profiles. Although O-clique provides only modest additional reductions beyond MILP-ICS, it reduces the measurement requirement by 27.3% on average relative to SI-ICS with 100 iterations, despite using only five final coefficient refinement iterations, indicating improved support quality. Finally, we extend the comparison to Fermi–Hubbard, Kitaev–Heisenberg–Γ, and XYZ lattice Hamiltonians, where MILP-based groupings substantially outperform the corresponding SI-based strategies. Together, these results show that variance-informed optimization of measurement-group structure can substantially reduce sampling costs across molecular and lattice Hamiltonians, with optimized non-overlapping groups providing strong, transferable initializations and direct overlapping optimization offering further gains.

arXiv

j-IR-vis: Vision model for infrared spectroscopy embeddings

Rudra Sondhi, Edwin Chacko, Rodrigo A. Vargas-Hernández (2026)

j-IR-vis is a vision-based neural network model designed to interpret infrared (IR) spectroscopy data. The model functions as a general-purpose spectral encoder that learns chemically interpretable representations directly from raw IR spectroscopy plots. Inspired by the ResNet family of vision models, it takes an IR spectrum image as input and performs multi-label classification of functional groups. Beyond classification, j-IR-vis embeds molecules into a spectroscopically informed similarity space, capturing relationships between molecules that are not always evident in standard fingerprint representations.

Paper GitHub

ParetoMol: a free web application for multi-objective Pareto analysis of molecular properties

Ilkham Yabbarov, Rodrigo A. Vargas-Hernández (2026)

ParetoMol is an open-source web application designed for the multi-objective Pareto analysis of molecular properties, specifically focusing on safety and pharmacokinetics. The tool helps researchers navigate the trade-offs involved in molecular optimization—such as balancing potency, selectivity, pharmacokinetics, and safety. ParetoMol accepts molecules via SMILES strings, CSV/SDF file uploads, ChEMBL target identifiers, or PubChem name resolution. It provides synchronized visualization tabs, including configurable Pareto scatter plots, BOILED-Egg partitioning, scaffold analysis, and chemical space projections.

Paper Website GitHub

Thumbnail for Experimental Validation of Universal Machine Learning Interatomic Potentials for Lithium-Ion Dynamics in Solid Electrolytes via 7Li NMR

Experimental Validation of Universal Machine Learning Interatomic Potentials for Lithium-Ion Dynamics in Solid Electrolytes via 7Li NMR

Cameron A. Gurwell, Taiana L. E. Pereira, Mengyang Cui, Carlos A. Martins Junior, Gillian R. Goward, and Rodrigo A. Vargas-Hernández (2026)

A significant step toward electrification is the adoption of more sustainable energy storage technologies such as batteries. Safer alternatives to conventional solution-state lithium-ion batteries (LIBs) are solid-state batteries, which employ solid-state electrolytes (SSEs). These SSEs can exhibit lithium conductivities comparable to those of liquid electrolytes but are often discovered through experimental trial and error. Computational pipelines powered by universal machine-learning interatomic potentials (uMLIPs) offer a cheaper and scalable route to screen large datasets of candidate SSEs. This recent class of models enables rapid materials exploration compared to traditional quantum chemistry methods and ab initio molecular dynamics (AIMD). In this work, we assess the current capabilities of uMLIPs for solid-state lithium battery applications by systematically benchmarking their predictions against experimental solid-state nuclear magnetic resonance (ssNMR) diffusion data and density functional theory (DFT) calculations. We considered twelve different Li-based compounds, five ordered electrolytes, three disordered electrolytes, and four cathodes. This study includes eighteen different uMLIPs categorized into four main families, CHGNet, M3GNet, MACE, and ORB. Our results demonstrate that model performance is strongly dependent on both the underlying architecture and the chemical system under investigation. MACE- and ORB-based models consistently achieve higher accuracy across structural, energetic, and diffusion-related properties. However, the reliability of lithium diffusion predictions remains system-dependent, highlighting current limitations in the transferability of uMLIPs for complex materials. These findings provide a critical assessment of the readiness of uMLIPs for integration into computational pipelines aimed at accelerating the discovery of next-generation solid-state battery materials.

ChemRxiv

Bayesian Optimization for High-Dimensional Coarse-Grained Model Parameterization: A Case Study on Pebax Polymer

Carlos A. Martins Jr, Daniela A. Damasceno, Keat Yung Hue, Caetano Rodrigues Miranda, Erich A. Müller, and Rodrigo A. Vargas-Hernández (2026)

In this work, we present a Bayesian optimization framework for parameterizing high-dimensional coarse-grained (CG) force field models, challenging the conventional assumption that such approaches are limited to low-dimensional problems. Using the tree-structured Parzen estimator (TPE), we successfully optimize a 41-parameter CG model of Pebax-1657, a polyamide-polyether copolymer, simultaneously targeting structural and thermodynamic properties including density, radius of gyration, and glass transition temperature. Our approach not only converges faster than traditional search algorithms but also delivers consistent improvements over standard parametrization strategies, opening the door to scalable Bayesian optimization of complex molecular simulations.

DOI

LivePyxel: accelerating image annotations with a Python-integrated webcam live streaming

Uriel Garcilazo-Cruz, Joseph O. Okeme, Rodrigo A. Vargas-Hernández (2026)

In this work, we introduce LivePixel, a Python-based graphical user interface that integrates with imaging systems—such as webcams and microscopes—to enable on-site image annotation in laboratory environments. LivePyxel addresses the limitations of existing annotation tools by supporting on-demand pipelines without requiring pre-collected datasets. The software features an intuitive interface with precision annotation tools, including Bézier splines and binary masks, alongside non-destructive layer editing for high-performance workflows. Built on OpenCV and NumPy, LivePyxel offers broad video device compatibility and is optimized for object detection tasks, accelerating the development of AI models in experimental settings.

DOI GitHub

MOLPIPx: An end-to-end differentiable package for permutationally invariant polynomials in Python and Rust

Manuel S. Drehwald, Asma Jamali, Rodrigo A. Vargas-Hernández (2025)

In this work, we present MOLPIPx, a versatile library designed to seamlessly integrate permutationally invariant polynomials with modern machine learning frameworks, enabling the efficient development of linear models, neural networks, and Gaussian process models. These methodologies are widely employed for parameterizing potential energy surfaces across diverse molecular systems. MOLPIPx leverages two powerful automatic differentiation engines—JAX and EnzymeAD-Rust—to facilitate the efficient computation of energy gradients and higher-order derivatives, which are essential for tasks such as force field development and dynamic simulations.

DOI GitHub

Normalizing flows and orbital-free DFT

A. de Camargo et al. (2024)

Orbital-free density functional theory (OF-DFT) for real-space systems has historically depended on Lagrange optimization techniques, primarily due to the inability of previously proposed electron density ansatze to ensure the normalization constraint. This study illustrates how leveraging contemporary generative models, notably normalizing flows (NFs), can surmount this challenge. We pioneer a Lagrangian-free optimization framework by employing these machine learning models as ansatze for the electron density. This novel approach also integrates cutting-edge variational inference techniques and equivariant deep learning models, offering an innovative alternative to the OF-DFT problem.

arXiv GitHub

Softmax parameterization of the occupation numbers for natural orbital functionals

L. Franco et al. (2024)

Within the framework of natural orbital functional theory, having a convenient representation of the optimization function becomes critical for the computational performance of the calculation. Recognizing this, we propose an innovative parametrization of the occupation numbers that takes advantage of the electron-pairing approach used in Piris natural orbital functionals, through the adoption of the softmax function, a pivotal component in modern deep-learning models. Our approach not only ensures adherence to the N-representability of the first-order reduced density matrix (1RDM) but also significantly enhances the computational efficiency of 1RDM functional theory calculations. The effectiveness of this alternative parameterization approach was assessed using the W4-17-MR molecular set, which demonstrated faster and robust convergence compared to previous implementations.

arXiv

PennyLane: Automatic differentiation of hybrid quantumclassical computations

V. Bergholm et al. (2022)

Lorem ipsum dolor sit amet consectetur adipisicing elit. Error culpa modi hic nihil accusamus? Accusantium dolore dignissimos dolorum ratione nihil deserunt unde maxime nisi eos, veniam vel, molestiae cum accusamus? Ducimus quibusdam repellendus ea reiciendis assumenda pariatur dolores sit voluptatum quasi. Unde facilis excepturi, voluptatem hic iste optio eius est fugit accusamus ratione nihil fugiat, tempora architecto, soluta magnam laborum.

Ref1