arXiv subjects

Zsuzsanna Koczor-Benda

Publications and source records attributed to Zsuzsanna Koczor-Benda.

7 recordsLinked to original sources

arXiv2026

Beyond Drug Discovery: The Nanotechnology Molecular Optimization (NMO) Benchmark

Generative molecular design is shaped by simple proxy benchmarks for drug-like properties and models pretrained on large pharmaceutical datasets. This combination yields strong benchmark metrics but limits transferability to domains structurally distinct from drug discovery. To overcome this limitation and drive discovery toward real, scientifically grounded targets, we introduce the Nanotechnology Molecular Optimization (NMO) Benchmark, which bridges machine learning (ML) and quantum materials science. NMO acts simultaneously as a rigorous testbed for the ML community and a discovery engine for nanotechnology research. The suite replaces proxy oracles with quantum simulations and introduces strict protocols that prioritize scientific utility over leaderboard-oriented overfitting. The physics-based NMO tasks impose hard structural constraints and rugged fitness landscapes, posing fundamentally new requirements on generative models. Notably, advanced molecular optimization methods underperform much simpler approaches on the NMO tasks. We develop a new baseline method identifying the critical components to solve the NMO tasks, including a novel representation for modeling structural constraints and a domain-agnostic pretraining strategy to eliminate pharmaceutical dataset bias. Our results surpass state-of-the-art physical properties and reveal previously unknown structural motifs, offering new insights for the nanotechnology community and demonstrating that ML can drive genuine scientific discovery.

cs.LG↗

arXiv2025

Generative design of functional organic molecules for terahertz radiation detection

Plasmonic nanocavities are molecule-nanoparticle junctions that offer a promising approach to upconvert terahertz radiation into visible or near-infrared light, enabling nanoscale detection at room temperature. However, the identification of molecules with strong terahertz-to-visible frequency upconversion efficiency is limited by the availability of suitable compounds in commercial databases. Here, we employ the generative autoregressive deep neural network, G-SchNet, to perform property-driven design of novel monothiolated molecules tailored for terahertz radiation detection. To design functional organic molecules, we iteratively bias G-SchNet to drive molecular generation towards highly active and synthesizable molecules based on machine learning-based property predictors, including molecular fingerprints and state-of-the-art neural networks. We study the reliability of these property predictors for generated molecules and analyze the chemical space and properties of generated molecules to identify trends in activity. Finally, we filter generated molecules and plan retrosynthetic routes from commercially available reactants to identify promising novel compounds and their most active vibrational modes in terahertz-to-visible upconversion.

physics.chem-ph↗

arXiv2025

Structural bias in three-dimensional autoregressive generative machine learning of organic molecules

A range of generative machine learning models for the design of novel molecules and materials have been proposed in recent years. Models that can generate three-dimensional structures are particularly suitable for quantum chemistry workflows, enabling direct property prediction. The performance of generative models is typically assessed based on their ability to produce novel, valid, and unique molecules. However, equally important is their ability to learn the prevalence of functional groups and certain chemical moieties in the underlying training data, that is, to faithfully reproduce the chemical space spanned by the training data. Here, we investigate the ability of the autoregressive generative machine learning model G-SchNet to reproduce the chemical space and property distributions of training datasets composed of large, functional organic molecules. We assess the elemental composition, size- and bond-length distributions, as well as the functional group and chemical space distribution of training and generated molecules. By principal component analysis of the chemical space, we find that the model leads to a biased generation that is largely unaffected by the choice of hyperparameters or the training dataset distribution, producing molecules that are, on average, more unsaturated and contain more heteroatoms. Purely aliphatic molecules are mostly absent in the generation. We further investigate generation with functional group constraints and based on composite datasets, which can help partially remedy the model generation bias. Decision tree models can recognize the generation bias in the models and discriminate between training and generated data, revealing key chemical differences between the two sets. The chemical differences we find affect the distributions of electronic properties such as the HOMO-LUMO gap, which is a common target for functional molecule design.

physics.chem-ph↗

arXiv2023

Analytic evaluation of non-adiabatic couplings within the complex absorbing potential equation-of-motion coupled-cluster method

We present the theory for the evaluation of non-adiabatic couplings (NACs) involving resonance states within the complex absorbing potential equation-of-motion coupled-cluster (CAP-EOM-CC) framework implemented within the singles and doubles approximation. Resonance states are embedded in the continuum and undergo rapid decay through autodetachment. In addition, nuclear motions can facilitate transitions between different resonances and between resonances and bound states. These non-adiabatic transitions affect the chemical fate of resonances and have distinct spectroscopic signatures. The NAC vector is a central quantity needed to model such effects. In the CAP-EOM-CC framework, resonance states are treated on the same footing as bound states. Using the example of fumaronitrile, which supports a bound radical anion and several anionic resonances, we analyze the non-adiabatic coupling between bound states and pseudocontinuum states, between bound states and resonances and between two resonances. We find that the NAC between a bound state and a resonance is nearly independent of the CAP strength and thus straightforward to evaluate whereas the NAC between two resonance states or between a bound state and a pseudocontinuum state is more difficult to evaluate.

physics.chem-ph↗

arXiv2021

Molecular Vibration Explorer: an online database and toolbox for surface-enhanced frequency conversion, infrared and Raman spectroscopy

We present Molecular Vibration Explorer, a freely accessible online database and interactive tool for exploring vibrational spectra and tensorial light-vibration coupling strengths of a large collection of thiolated molecules. The `Gold' version of the database gathers the results from density functional theory calculations on 2'800 commercially available thiol compounds linked to a gold atom, with the main motivation to screen the best molecules for THz and mid-infrared to visible upconversion. Additionally, the `Thiol' version of the database contains results for 1'900 unbound thiolated compounds. They both provide access to a comprehensive set of computed spectroscopic parameters for all vibrational modes of all molecules in the database. The user can simultaneously investigate infrared absorption, Raman scattering and vibrational sum- and difference frequency generation cross sections. Molecules can be screened for various parameters in custom frequency ranges, such as large Raman cross-section under specific molecular orientation, or large orientation-averaged sum-frequency generation (SFG) efficiency. The user can select polarization vectors for the electromagnetic fields, set the orientation of the molecule and customize parameters for plotting the corresponding IR, Raman and sum-frequency spectra. We illustrate the capabilities of this tool with selected applications in the field of surface-enhanced spectroscopy.

physics.optics↗

arXiv2021

Detecting mid-infrared light by molecular frequency upconversion with dual-wavelength hybrid nanoantennas

Coherent interconversion of signals between optical and mechanical domains is enabled by optomechanical interactions. Extreme light-matter coupling produced by confining light to nanoscale mode volumes can then access single mid-infrared (MIR) photon sensitivity. Here we utilise the infrared absorption and Raman activity of molecular vibrations in plasmonic nanocavities to demonstrate frequency upconversion. We convert λ~10 μm incoming light to visible via surface-enhanced Raman scattering (SERS) in doubly-resonant antennas that enhance upconversion by >10^10. We show >200% amplification of the SERS antiStokes emission when a MIR pump is tuned to a molecular vibrational frequency, obtaining lowest detectable powers ~1 μW/μm^2 at room temperature. These results have potential for low-cost and large-scale infrared detectors and spectroscopic techniques, and bring single-molecule sensing into the infrared

physics.optics↗

arXiv2020

Breaking the Selection Rules of Spin-Forbidden Molecular Absorption in Plasmonic Nanocavities

Controlling absorption and emission of organic molecules is crucial for efficient light-emitting diodes, organic solar cells and single-molecule spectroscopy. Here, a new molecular absorption is activated inside a gold plasmonic nanocavity, and found to break selection rules via spin-orbit coupling. Photoluminescence excitation scans reveal absorption from a normally spin-forbidden singlet to triplet state transition, while drastically enhancing the emission rate by several thousand fold. The experimental results are supported by density functional theory, revealing the manipulation of molecular absorption by nearby metallic gold atoms.

physics.optics↗