arXiv subjects

Oleksandr Voznyy

Publications and source records attributed to Oleksandr Voznyy.

11 recordsLinked to original sources

arXiv2026

The Mechanics of Delta Learning: Target Design for Generalizable Scientific Machine Learning

In scientific machine learning, $Δ$-learning trains models on residual errors relative to physical baselines, assuming that more accurate baselines with smaller residual scales inherently improve downstream performance. Here, we demonstrate that residual scale alone is an insufficient heuristic for learnability. Evaluating molecular graph neural networks on total energy targets, we show that complex local descriptor baselines can yield small residual targets that are disproportionately rough within architecture-informed proxy spaces and harder to learn relative to their scale. Conversely, semi-empirical baseline reduces both scale and normalized roughness, improving in-domain and out-of-domain prediction. We introduce scale-normalized graph Dirichlet roughness ($D_{\text{IQR}}$) as a pre-training diagnostic for residual learnability and establish baseline complementarity as a core target-design principle, elevating target space formulation alongside model architecture as a key axis for scientific machine learning.

cs.LG↗

arXiv2026

Multitask learning with semiempirical orbital charges enables sample-efficient MLIPs

Machine learning interatomic potentials (MLIPs) require generating computationally expensive, large-scale training datasets to accurately simulate materials and molecules. Incorporating electronic structure information using multitask learning improves sample efficiency, however, training on full Hamiltonian matrices, which scale quadratically with the number of atoms, is intractable for large datasets. In this work, we show that multitask learning utilizing orbitally resolved semiempirical charges significantly improves sample efficiency and accuracy in MLIPs. To efficiently predict orbital charges, we implement a specialized equivariant model, reducing charge prediction error compared to an invariant baseline. By augmenting training with computationally inexpensive GFN1-xTB orbital charges, which scale linearly with the number of atoms, our model achieves a 46\% reduction in energy mean absolute error and requires five times less data to match the performance of energy-only models. Furthermore, our approach outperforms models trained on expensive density functional theory (DFT) atomic charges, capturing orbitally resolved electronic complexity and forcing the network to learn a physically accurate latent space that spontaneously clusters metals by shared chemical properties. Because orbital charges are only required during training, this approach preserves inference efficiency, providing a scalable recipe for developing accurate, data-efficient foundation models for complex chemical systems.

physics.chem-ph↗

arXiv2025

Adsorption energies are necessary but not sufficient to identify good catalysts

As a core technology for green chemical synthesis and electrochemical energy storage, electrocatalysis is central to decarbonization strategies aimed at combating climate change. In this context, computational and machine learning driven catalyst discovery has emerged as a major research focus. These approaches frequently use the thermodynamic overpotential, calculated from adsorption free energies of reaction intermediates, as a key parameter in their analysis. In this paper, we explore the large-scale applicability of such overpotential estimates for identifying good catalyst candidates by using datasets from the Open Catalyst Project (OC20 and OC22). We start by quantifying the uncertainty in predicting adsorption energies using \textit{ab initio} methods and find that $\sim$0.3-0.5 eV is a conservative estimate for a single adsorption energy prediction. We then compute the overpotential of all materials in the OC20 and OC22 datasets for the hydrogen and oxygen evolution reactions. We find that while the overpotential allows the identification of known good catalysts such as platinum and iridium oxides, the uncertainty is large enough to misclassify a broad fraction of the datasets as ``good'', which limits its value as a screening criterion. These results question the reliance on overpotential estimation as a primary evaluation metric to sort through catalyst candidates and calls for a shift in focus in the computational catalysis and machine learning communities towards other metrics such as synthesizability, stability, lifetime or affordability.

cond-mat.mtrl-sci↗

arXiv2024

Using GNN property predictors as molecule generators

Graph neural networks (GNNs) have emerged as powerful tools to accurately predict materials and molecular properties in computational discovery pipelines. In this article, we exploit the invertible nature of these neural networks to directly generate molecular structures with desired electronic properties. Starting from a random graph or an existing molecule, we perform a gradient ascent while holding the GNN weights fixed in order to optimize its input, the molecular graph, towards the target property. Valence rules are enforced strictly through a judicious graph construction. The method relies entirely on the property predictor; no additional training is required on molecular structures. We demonstrate the application of this method by generating molecules with specific DFT-verified energy gaps and octanol-water partition coefficients (logP). Our approach hits target properties with rates comparable to or better than state-of-the-art generative models while consistently generating more diverse molecules.

cs.LG↗

arXiv2023

The Open Catalyst 2022 (OC22) Dataset and Challenges for Oxide Electrocatalysts

The development of machine learning models for electrocatalysts requires a broad set of training data to enable their use across a wide variety of materials. One class of materials that currently lacks sufficient training data is oxides, which are critical for the development of OER catalysts. To address this, we developed the OC22 dataset, consisting of 62,331 DFT relaxations (~9,854,504 single point calculations) across a range of oxide materials, coverages, and adsorbates. We define generalized total energy tasks that enable property prediction beyond adsorption energies; we test baseline performance of several graph neural networks; and we provide pre-defined dataset splits to establish clear benchmarks for future efforts. In the most general task, GemNet-OC sees a ~36% improvement in energy predictions when combining the chemically dissimilar OC20 and OC22 datasets via fine-tuning. Similarly, we achieved a ~19% improvement in total energy predictions on OC20 and a ~9% improvement in force predictions in OC22 when using joint training. We demonstrate the practical utility of a top performing model by capturing literature adsorption energies and important OER scaling relationships. We expect OC22 to provide an important benchmark for models seeking to incorporate intricate long-range electrostatic and magnetic interactions in oxide surfaces. Dataset and baseline models are open sourced, and a public leaderboard is available to encourage continued community developments on the total energy tasks and data.

cond-mat.mtrl-sci↗

arXiv2021

Interpretable discovery of new semiconductors with machine learning

Machine learning models of materials$^{1-5}$ accelerate discovery compared to ab initio methods: deep learning models now reproduce density functional theory (DFT)-calculated results at one hundred thousandths of the cost of DFT$^{6}$. To provide guidance in experimental materials synthesis, these need to be coupled with an accurate yet effective search algorithm and training data consistent with experimental observations. Here we report an evolutionary algorithm powered search which uses machine-learned surrogate models trained on high-throughput hybrid functional DFT data benchmarked against experimental bandgaps: Deep Adaptive Regressive Weighted Intelligent Network (DARWIN). The strategy enables efficient search over the materials space of ~10$^8$ ternaries and 10$^{11}$ quaternaries$^{7}$ for candidates with target properties. It provides interpretable design rules, such as our finding that the difference in the electronegativity between the halide and B-site cation being a strong predictor of ternary structural stability. As an example, when we seek UV emission, DARWIN predicts K$_2$CuX$_3$ (X = Cl, Br) as a promising materials family, based on its electronegativity difference. We synthesized and found these materials to be stable, direct bandgap UV emitters. The approach also allows knowledge distillation for use by humans.

cond-mat.mtrl-sci↗

arXiv2019

Direct CO2 Electroreduction from Carbonate

The process of CO2 valorization, all the way from capture of CO2 to its electrochemical upgrade, requires significant inputs in each of the capture, upgrade, and separation steps. The gas phase CO2 feed following the capture-and-release stage and into the CO2 electroreduction stage produce a large waste of CO2, between 80 and 95% of CO2 is wasted due to carbonate formation or electrolyte crossover, that adds cost and energy consumption to the CO2 management aspect of the system. Here we report an electrolyzer that instead directly upgrades carbonate electrolyte from CO2 capture solution to syngas, achieving 100% carbon utilization across the system. A bipolar membrane is used to produce proton in situ, under applied potential, which facilitates CO2 releasing at the membrane:catalyst interface from the carbonate solution. Using an Ag catalyst, we generate pure syngas at a 3 to 1 H2 to CO ratio, with no CO2 dilution at the gas outlet, at a current density of 150 mA cm-2, and achieve a full cell energy efficiency of 35%. The direct carbonate cell was stable under a continuous 145 h of catalytic operation at ca. 180 mA cm-2. The work demonstrates that coupling CO2 electrolysis directly with a CO2 capture system can accelerate the path towards viable CO2 conversion technologies.

physics.chem-ph↗

arXiv2016

Passivation-sensitive exciton finestructure produces excess Stokes shifts in colloidal quantum dots

The excitonic finestructure of colloidal quantum dots (CQDs) is comprised of a manifold of transitions, of which only the lowest are populated and contribute to photoluminescence. This leads to a Stokes shift in emission relative to absorption. Here we show experimentally that the Stokes shift in Pb and Cd-based chalcogenide CQDs is correlated with the degree of surface passivation, and develop a model that explains how coupling to the surface affects the core electronic states. Dark and bright transitions can reorder and split, increasing the Stokes shift even without the formation of deep traps. Our findings resolve the highly-debated topic of excess Stokes shifts in PbS nanocrystals as due to parity-forbidden transitions instead of traps. We predict that the Stokes shift in PbS can be eliminated via core stoichiometry control, a critical step towards enhancing the open circuit voltage in quantum dot solar cells.

cond-mat.mes-hall↗

arXiv2011

Theory of highly excited semiconductor nanostructures including Auger coupling: exciton-bi-exciton mixing in CdSe nanocrystals

We present a theory of highly excited interacting carriers confined in a semiconductor nanostructure, incorporating Auger coupling between excited states with different number of excitations. The Coulomb matrix elements connecting exciton, bi-exciton and tri-exciton complexes are derived and an intuitive picture of breaking neutral multi-exction complexes into positively and negatively charged multi-exciton complexes is given. The general approach is illustrated by analyzing the coupling of biexciton and exciton in CdSe spherical nanocrystals. The electron and hole states are computed using atomistic $sp^3d^5s^*$ tight binding Hamiltonian including an effective crystal field splitting and surface passivation. For each number of electron-hole pairs the many-body spectrum is computed in the configuration-interaction approach. The low-energy correlated biexciton levels are broken into charged complexes: a hole and a negatively charged trion and an electron and a positively charged trion. Out of a highly excited exciton spectrum a subspace coupled to bi-exciton levels via Auger processes is identified. The interaction between correlated bi-exciton and exciton states is treated using exact diagonalization techniques. This allows to extract the spectral function of the biexciton and relate its characteristic width and amplitude to the characteristic amplitude and timescale of the coherent time evolution of the coupled system. It is shown that this process can be described by the Fermi's Golden Rule only if a fast relaxation of the excitonic subsystem is accounted for.

cond-mat.mes-hall↗

arXiv2010

Fine structure and size dependence of exciton and bi-exciton optical spectra in CdSe nanocrystals

Theory of electronic and optical properties of exciton and bi-exciton complexes confined in CdSe spherical nanocrystals is presented. The electron and hole states are computed using atomistic $sp^3d^5s^*$ tight binding Hamiltonian including an effective crystal field splitting, spin-orbit interactions, and model surface passivation. The optically excited states are expanded in electron-hole configurations and the many-body spectrum is computed in the configuration-interaction approach. Results demonstrate that the low-energy electron spectrum is organized in shells ($s$, $p$, ...), whilst the valence hole spectrum is composed of four low-lying, doubly degenerate states separated from the rest by a gap. As a result, the bi-exciton and exciton spectrum is composed of a manifold of closely lying states, resulting in a fine structure of exciton and bi-exciton spectra. The quasi-degenerate nature of the hole spectrum results in a correlated bi-exciton state, which makes it slowly convergent with basis size. We carry out a systematic study of the exciton and bi-exciton emission spectra as a function of the nanocrystal diameter and find that the interplay of repulsion between constituent excitons and correlation effects results in a change of the sign of bi-exciton binding energy from negative to positive at a critical nanocrystal size.

cond-mat.mes-hall↗

arXiv2010

Edge stability, reconstruction, zero-energy states and magnetism in triangular graphene quantum dots with zigzag edges

We present the results of ab-initio density functional theory based calculations of the stability and reconstruction of zigzag edges in triangular graphene quantum dots. We show that, while the reconstructed pentagon-heptagon zigzag edge structure is more stable in the absence of hydrogen, ideal zigzag edges are energetically favored by hydrogen passivation. Zero-energy band exists in both structures when passivated by hydrogen, however in case of pentagon-heptagon zigzag, this band is found to have stronger dispersion, leading to the loss of net magnetization.

cond-mat.mes-hall↗