Searcharxiv⌕ Search

arXiv subjects

Andrii Tuzhykov

Publications and source records attributed to Andrii Tuzhykov.

2 recordsLinked to original sources

Edge-scaling softening of a universal machine-learned potential: point-defect formation energies in a coronene graphene quantum dot

Universal machine-learning interatomic potentials (MLIPs) are increasingly used as computationally efficient surrogates for density functional theory (DFT) in high-throughput nanomaterials screening. Systematic softening has been reported for several universal MLIPs when high-energy and far-from-equilibrium configurations are underrepresented in training data. We examine how this behavior appears at under-coordinated edge sites in graphene quantum dots (GQDs), where a vacancy can have pronounced molecular-radical character. Using coronene (C$_{24}$H$_{12}$), a hexagonal GQD with $D_{6h}$ symmetry and three carbon classes, we benchmark SevenNet (7net-omni-i12) against spin-polarized PBE DFT (GPAW) for monovacancy, divacancy, and Stone-Wales defects. Beyond the full site- and bond-resolved landscapes, we perform a controlled matched-reconstruction comparison in which 28 vacancy and divacancy configurations are scored by both engines using self-consistent chemical-potential reservoirs. In this dataset, SevenNet underestimates all 28 vacancy and divacancy formation energies, and the mean deficit increases with under-coordination within each defect family: 0.30-0.74 eV for monovacancies and 0.65-1.11 eV for divacancies as the number of broken edge C-H bonds increases. By contrast, the mean Stone-Wales energies differ by only 0.013 eV. SevenNet also preserves the ordering of three hydrogen-dependent reconstruction basins and reproduces $D_{6h}$-equivalent sites to within a few meV. The within-class symmetry spread therefore provides a practical diagnostic of basin-sensitive relaxations. Within the present coronene benchmark, these results support MLIP-based structure screening and preliminary ranking, while absolute vacancy energies and multi-basin edge configurations still require DFT validation.

cond-mat.mtrl-sci↗

SITO: Sifting a Commercial Text-Safety API for Missed Harm and False Flags in English, Ukrainian and Surzhyk

Platforms that host young users screen messages with commercial text-safety APIs, but independent evidence on how often such systems miss harm or flag harmless text is scarce, especially outside English, and the documentation of the one studied here leaves its false-positive rate unquantified. We evaluated one such API, Tuteliq Detect as deployed under the release identifier tuteliq-detect-v5.20261011 (not the separately published v6 model), on six harm categories: authorised push-payment fraud, money-mule recruitment, social engineering, romance scams, gambling harm and bullying. We built SITO (Synthetic Independent Text-safety benchmark with Overflagging tests; sito is Ukrainian for 'sieve'), 2,332 synthetic cases in English, Ukrainian and mixed Ukrainian, with each harmful case matched by hard harmless ones: same-topic messages, minimal-edit contrasts, warnings and retrospective accounts. On English text the system flagged 84.9-94.1% of harmful cases and 0-4.3% of harmless ones per category. It missed third-party reports of ongoing harm far more often than perpetrators' messages (63.6% against 95.1% recall, all languages), when sent without the settings the API offers for reports of harm. Ukrainian and mixed-language input was detected at least as often as English, recall was lower on text from a held-out generator family, so headline recall may be optimistic, and 3.2% of cases received a different decision in repeated runs. A platform should measure false positives on its own traffic and check how bystander reports are handled. The datasets, generation prompts and code are released.

cs.CL↗