AI-Agent-Orchestrated DFT Screening of Conjugated Donor and Electron-Acceptor Building Blocks for Organic Photovoltaics
DOI:
https://doi.org/10.31224/8804Abstract
This study develops and evaluates an AI-agent-orchestrated computational workflow for the preliminary electronic-structure screening of conjugated molecular building blocks relevant to organic photovoltaics (OPVs). A computer-use agent interacts with The structure files and audit spreadsheet are stored in the project archive on the computer. Windows workstation through Remote Desktop Commander (RDC), while RDKit is employed for molecular representation validation, three-dimensional structure generation and molecular mechanics geometry preparation. Quantum-chemical calculations are executed using PySCF within a Linux environment provided by Windows Subsystem for Linux 2 (WSL2). The initial molecular set comprises three oligothiophene-based donor fragments and three electron-deficient heteroaromatic acceptor building blocks, defining nine possible donor–acceptor fragment combinations. Six neutral molecular structures were successfully generated and validated using the ETKDGv3 and MMFF94s procedures. Subsequently, six restricted Kohn–Sham single-point self-consistent-field (SCF) calculations were completed at the B3LYP/6-31G(d) level of theory on the MMFF94s-prepared geometries. All six calculations achieved electronic SCF convergence. The calculated Kohn–Sham HOMO–LUMO gaps ranged from 3.431 to 4.261 eV for the donor fragments and from 3.933 to 4.074 eV for the acceptor fragments. The nine isolated-fragment donor–acceptor comparisons produced acceptor-minus-donor LUMO offsets ranging from −0.702 to −1.736 eV. The six-molecule single-point calculation batch required approximately 388.68 s of wall-clock time. However, no DFT geometry optimization achieved verified convergence, and no validated TD-DFT excited-state results were obtained. Consequently, the reported electronic properties are preliminary single-point descriptors evaluated at molecular mechanics-prepared starting geometries rather than final optimized molecular properties. The study demonstrates the practical integration of AI-assisted computer operation, molecular preparation, quantum-chemical execution, data extraction and preliminary orbital-level analysis. It also establishes the computational foundation for subsequent geometry optimization and excited-state investigations without inferring photovoltaic device performance directly from isolated-molecule calculations.
Downloads
Downloads
Posted
License
Copyright (c) 2026 Ejikeme Ezo Igbokwe, Emeka Friday Onu

This work is licensed under a Creative Commons Attribution 4.0 International License.