arXiv subjects

Yamil J. Colón

Publications and source records attributed to Yamil J. Colón.

10 recordsLinked to original sources

arXiv2026

A Machine-Learned Interatomic Potential Free-Energy Surface for the First Step of MIL-101(Cr) Secondary Building Unit Formation

Metal-organic framework (MOF) self-assembly mechanisms remain poorly characterized because ab initio molecular dynamics (AIMD) are accurate but too computationally expensive to converge the free-energy surfaces (FES) that govern secondary building unit (SBU) nucleation and growth. This work addresses that limitation for the first step of MIL-101(Cr) SBU formation. Using MACE-POLAR-M, a long-range-aware equivariant machine-learned interatomic potential fine-tuned on a compact DFT reference dataset built from exploratory metadynamics, this work obtains a converged 2D FES for this step at a level of theory (ωB97M-V/def2-TZVPP) at which AIMD would be prohibitively expensive. The pre-trained model samples relevant configurations but gets the thermodynamics wrong, predicting an endothermic reaction and a false global minimum. Fine-tuning corrects both, recovering the expected exothermic process and the correct product basin in agreement with proposed literature mechanisms. The resulting FES also refines the mechanism. It shows that water release from the chromium center proceeds stepwise, through two sequential energy barriers, rather than the single concerted step originally proposed. We find that model accuracy is specific to the reaction coordinate and does not extend to the dissociated fragment configurations excluded from fine-tuning, a direct consequence of the training-data choice that protects accuracy along the reaction path for the FES prediction task. Together, these results establish a validated, computationally tractable route to obtain free-energy surfaces for MOF formation chemistry, transferable to other reactions where AIMD remains prohibitive, and identify training-data composition as the key determinant of which reaction properties a fine-tuned checkpoint can reliably describe.

physics.chem-ph↗

arXiv2026

DynaMate2: runtime registration of expert-defined tools for agentic scientific workflow automation

Agentic large-language-model systems can coordinate scientific tools, but many implementations remain difficult for domain scientists to extend without modifying the source orchestration code or relying on unconstrained code generation. DynaMate2 is a LangGraph-based multi-agent framework for converting expert-defined Python functions into persistent AI-callable tools. The architecture separates domain execution from LLM supervision: registered tools perform scientific operations, while a supervisor LLM decomposes goals, selects specialist agents, routes inputs, and propagates outputs across steps. DynaMate2 supports: runtime tool registration from inline code, source files, and explicitly requested natural-language specifications; persistent storage of tools, agents, and conversation state; and a web interface for interactive workflow assembly. We demonstrate the framework on a molecular simulation workflow in which a single instruction retrieves a MACE foundation model, builds a NaCl-water configuration, runs an ASE molecular dynamics trajectory, and generates energy and temperature diagnostics. The demonstration illustrates how validated workflow components can be composed into a supervised agentic pipeline without rewriting the framework. DynaMate2 therefore provides a reusable template for extending LLM-based automation to research groups with existing Python workflows, while preserving the need for explicit tool validation, reproducibility logs, and deployment-specific safeguards.

physics.chem-ph↗

arXiv2025

Hybrid Atomistic-Parametric Decoherence Model for Molecular Spin Qubits

Solid-state molecular qubits with open-shell ground states have great potential for addressability, scalability, and tunability, but understanding the fundamental limits of quantum coherence in these systems is challenging due to the complexity of the qubit environment. To address this, we develop a random Hamiltonian approach where the molecular $g$-tensor fluctuates due to classical lattice motion obtained from molecular dynamics simulations at constant temperature. Atomistic $g$-tensor fluctuations are used to construct Redfield quantum master equations that predict the relaxation $T_1$ and dephasing $T_2$ times of copper porphyrin qubits in a crystalline framework. Assuming one-phonon spin-lattice interaction processes, $1/T$ temperature scaling and $1/B^3$ magnetic field scaling of $T_1$ are established using atomistic bath correlation functions. Atomistic $T_1$ predictions overestimate the available experimental data by orders of magnitude. Quantitative agreement with measurements at all magnetic fields is restored by introducing a magnetic field noise model to describe lattice nuclear spins, with field-dependent noise amplitude in the range $δB\sim 10\,μ{\rm T}- 1\,{\rm mT}$ for the copper porphyrin system. We show that while $T_1$ scales as $1/B$ experimentally due to a combination of spin-lattice and magnetic noise contributions, $T_2$ scales strictly as $ 1/B^2$ due to low-frequency dephasing processes associated with magnetic field noise. Our work demonstrates the potential of dynamical methods for modeling the open quantum system dynamics of molecular spin qubits.

quant-ph↗

arXiv2025

Considerations in the use of ML interaction potentials for free energy calculations

Machine learning force fields (MLFFs) promise to accurately describe the potential energy surface of molecules at the ab initio level of theory with improved computational efficiency. Within MLFFs, equivariant graph neural networks (EQNNs) have shown great promise in accuracy and performance and are the focus of this work. The capability of EQNNs to recover free energy surfaces (FES) remains to be thoroughly investigated. In this work, we investigate the impact of collective variables (CVs) distribution within the training data on the accuracy of EQNNs predicting the FES of butane and alanine dipeptide (ADP). A generalizable workflow is presented in which training configurations are generated with classical molecular dynamics simulations, and energies and forces are obtained with ab initio calculations. We evaluate how bond and angle constraints in the training data influence the accuracy of EQNN force fields in reproducing the FES of the molecules at both classical and ab initio levels of theory. Results indicate that the model's accuracy is unaffected by the distribution of sampled CVs during training, given that the training data includes configurations from characteristic regions of the system's FES. However, when the training data is obtained from classical simulations, the EQNN struggles to extrapolate the free energy for configurations with high free energy. In contrast, models trained with the same configurations on ab initio data show improved extrapolation accuracy. The findings underscore the difficulties in creating a comprehensive training dataset for EQNNs to predict FESs and highlight the importance of prior knowledge of the system's FES.

physics.chem-ph↗

arXiv2024

First-Principles Screening of Metal-Organic Frameworks for Entangled Photon Pair Generation

The transmission of strong laser light in nonlinear optical materials can generate output photons sources that carry quantum entanglement in multiple degrees of freedom, making this process a fundamentally important tool in optical quantum technology. However, the availability of efficient optical crystals for entangled light generation is severely limited in terms of diversity, thus reducing the prospects for the implementation of next-generation protocols in quantum sensing, communication and computing. To overcome this, we developed and implemented a multi-scale first-principles modeling technique for the computational discovery of novel nonlinear optical devices based on metal-organic framework (MOF) materials that can efficiently generate entangled light via spontaneous parametric down-conversion(SPDC). Using collinear degenerate type-I SPDC as a case study, we computationally screen a database of 114,373 synthesized MOF materials to establish correlations between the structure and chemical composition of MOFs with the brightness and coherence properties of entangled photon pairs. We identify a subset of 49 non-centrosymmetric mono-ligand MOF crystals with high chemical and optical stability that produce entangled photon pairs with intrinsic $G^{(2)}$ correlation times $τ_c\sim 10-30$ fs and pair generation rates in the range $10^4-10^{8}$ s$^{-1}$mW$^{-1}$mm$^{-1}$ at 1064 nm. Conditions for optimal type-I phase matching are given for each MOF and relationships between pair brightness, crystal band gap and optical birefringence are discussed. Correlations between the optical properties of crystals and their constituent molecular ligands are also given. Our work paves the way for the computational design of MOF-based devices for optical quantum technology.

quant-ph↗

arXiv2023

Giant Generation of Polarization-Entangled Photons in Metal Organic Framework Waveguides

Parametric nonlinear optical processes are instrumental in optical quantum technology for generating entangled light. However, the range of materials conventionally used for producing entangled photons is limited. Metal-organic frameworks (MOFs) have emerged as a novel class of optical materials with customizable nonlinear properties and proven chemical and optical stability. The large number of combinations of metal atoms and organic ligand from which bulk MOF crystals are known to form, facilitates the search of promising candidates for nonlinear optics. To accelerate the discovery of next-generation quantum light sources, we employ a multi-scale modeling approach to study phase-matching conditions for collinear degenerate type-II spontaneous parametric down conversion (SPDC) with MOF-based one dimensional waveguides. Using periodic-DFT calculations to compute the nonlinear optical properties of selected zinc-based MOF crystals, we predict polarization-entangled pair generation rates of $\sim 10^3-10^6$ s$^{-1}$mW$^{-1}$mm$^{-1}$ at 1064 nm, which are comparable with industry materials used in quantum optics. We find that the biaxial MOF crystal Zn(4-pyridylacrylate)$_2$ improves two-fold the conversion efficiency over a periodically-poled KTP waveguide of identical dimensions. This work underscores the great potential of MOF single crystals as entangled light sources for applications in quantum communication and sensing.

physics.optics↗

arXiv2023

Semi-Empirical Haken-Strobl Model for Molecular Spin Qubits

Understanding the physical processes that determine the relaxation $T_{1}$ and dephasing $T_2$ times of molecular spin qubits is critical for envisioned applications in quantum metrology and information processing. Recent spin-echo $T_1$ measurements of solid-state molecular spin qubits have stimulated the development of quantum mechanical models for predicting intrinsic spin qubit timescales using first-principles electronic structure methods. We develop an alternative semi-empirical approach to construct Redfield quantum master equations for molecular spin qubits using a stochastic Haken-Strobl model for a central spin with a fluctuating gyromagnetic tensor due to spin-lattice interaction and a fluctuating local magnetic field due to interactions with other lattice spins. Using a vanadium-based spin qubit as a case study, we compute qubit population and decoherence timescales as a function of temperature and magnetic field using a bath spectral density parametrized with a small number of $T_{1}$ measurements. The theory quantitatively agrees with experimental data over a range of conditions beyond those used to parametrize the model, demonstrating the generalization potential of the method. The ability of the model to describe the temperature dependence of the ratio $T_2/T_1$ is discussed and possible applications for designing novel molecule-based quantum magnetometers are suggested.

quant-ph↗

arXiv2023

Transfer Learning Facilitates the Prediction of Polymer-Surface Adhesion Strength

Machine learning (ML) accelerates the exploration of material properties and their links to the structure of the underlying molecules. In previous work [J. Shi, M. J. Quevillon, P. H. A. Valença, and J. K. Whitmer, \textit{ACS Appl. Mater. Interfaces.}, 2022, 14, 32, 37161--37169], ML models were applied to predict the adhesive free energy of polymer--surface interactions with high accuracy from the knowledge of the sequence data, demonstrating successes in inverse-design of polymer sequence for known surface compositions. While the method was shown to be successful in designing polymers for a known surface, extensive datasets were needed for each specific surface in order to train the surrogate models. Ideally, one should be able to infer information about similar surfaces without having to regenerate a full complement of adhesion data for each new case. In the current work, we demonstrate a transfer learning (TL) technique using a deep neural network to improve the accuracy of ML models trained on small datasets by pre-training on a larger database from a related system and fine-tuning the weights of all layers with a small amount of additional data. The shared knowledge from the pre-trained model facilitates the prediction accuracy significantly on small datasets. We also explore the limits of database size on accuracy and the optimal tuning of network architecture and parameters for our learning tasks. While applied to a relatively simple coarse-grained (CG) polymer model, the general lessons of this study apply to detailed modeling studies and the broader problems of inverse materials design.

cond-mat.soft↗

arXiv2022

Model reduction for molecular diffusion in nanoporous media

Porous materials are widely used for applications in gas storage and separation. The diffusive properties of a variety of gases in porous media can be modeled using molecular dynamics simulations that can be computationally demanding depending on the pore geometry, complexity and amount of gas adsorbed. We explore a dimensionality reduction approach for estimating the self-diffusion coefficient of gases in simple pores using Langevin dynamics, such that the three-dimensional (3D) atomistic interactions that determine the diffusion properties of realistic systems can be reduced to an effective one-dimensional (1D) diffusion problem along the pore axis. We demonstrate the approach by modeling the transport of nitrogen molecules in single-walled carbon nanotubes of different radii, showing that 1D Langevin models can be parametrized with a few single-particle 3D atomistic simulations. The reduced 1D model predicts accurate diffusion coefficients over a broad range of temperatures and gas densities. Our work paves the way for studying the diffusion process of more general porous materials as zeolites or metal-organics frameworks with effective models of reduced complexity.

physics.chem-ph↗

arXiv2022

Machine learning identification of organic compounds using visible light

Identifying chemical compounds is essential in several areas of science and engineering. Laser-based techniques are promising for autonomous compound detection because the optical response of materials encodes enough electronic and vibrational information for remote chemical identification. This has been exploited using the fingerprint region of infrared absorption spectra, which involves a dense set of absorption peaks that are unique to individual molecules, thus facilitating chemical identification. However, optical identification using visible light has not been realized. Using decades of experimental refractive index data in the scientific literature of pure organic compounds and polymers over a broad range of frequencies from the ultraviolet to the far-infrared, we develop a machine learning classifier that can accurately identify organic species based on a single-wavelength dispersive measurement in the visible spectral region, away from absorption resonances. The optical classifier proposed here could be applied to autonomous material identification protocols or applications.

cs.LG↗