arXiv subjects

Mohammad Mostafanejad

Publications and source records attributed to Mohammad Mostafanejad.

6 recordsLinked to original sources

arXiv2026

BERTology of Molecular Property Prediction

Chemical language models (CLMs) have emerged as promising competitors to popular classical machine learning models for molecular property prediction (MPP) tasks. However, an increasing number of studies have reported inconsistent and contradictory results for the performance of CLMs across various MPP benchmark tasks. In this study, we conduct and analyze hundreds of meticulously controlled experiments to systematically investigate the effects of various factors, such as dataset size, model size, and standardization, on the pre-training and fine-tuning performance of CLMs for MPP. In the absence of well-established scaling laws for encoder-only masked language models, our aim is to provide comprehensive numerical evidence and a deeper understanding of the underlying mechanisms affecting the performance of CLMs for MPP tasks, some of which appear to be entirely overlooked in the literature.

cs.LG↗

arXiv2025

SEAMM: A Simulation Environment for Atomistic and Molecular Modeling

The Simulation Environment for Atomistic and Molecular Modeling (SEAMM) is an open-source software package written in Python that provides a graphical interface for setting up, executing, and analyzing molecular and materials simulations. The graphical interface reduces the entry barrier for the use of new simulation tools, facilitating the interoperability of a wide range of simulation tools available to solve complex scientific and engineering problems in computational molecular science. Workflows are represented graphically by user-friendly flowcharts which are shareable and reproducible. When a flowchart is executed within the SEAMM environment, all results, as well as metadata describing the workflow and codes used, are saved in a datastore that can be viewed using a browser-based dashboard, which allows collaborators to view the results and use the flowcharts to extend the results. SEAMM is a powerful productivity and collaboration tool that enables interoperability between simulation codes and ensures reproducibility and transparency in scientific research.

physics.chem-ph↗

arXiv2024

Unification of popular artificial neural network activation functions

We present a unified representation of the most popular neural network activation functions. Adopting Mittag-Leffler functions of fractional calculus, we propose a flexible and compact functional form that is able to interpolate between various activation functions and mitigate common problems in training neural networks such as vanishing and exploding gradients. The presented gated representation extends the scope of fixed-shape activation functions to their adaptive counterparts whose shape can be learnt from the training data. The derivatives of the proposed functional form can also be expressed in terms of Mittag-Leffler functions making it a suitable candidate for gradient-based backpropagation algorithms. By training multiple neural networks of different complexities on various datasets with different sizes, we demonstrate that adopting a unified gated representation of activation functions offers a promising and affordable alternative to individual built-in implementations of activation functions in conventional machine learning frameworks.

cs.LG↗

arXiv2024

Numerical Analysis of the Complete Active-Space Extended Koopmans's Theorem

We investigate the numerical accuracy of the extended Koopmans's theorem (EKT) in reproducing the full configuration interaction (FCI) and complete active-space configuration interaction (CAS-CI) ionization energies (IEs) of atomic and molecular systems calculated as the difference between the energies of N and (N - 1) electron states. In particular, we study the convergence of the EKT IEs to their exact values as the basis set and the active space sizes vary. We find that the first FCI EKT IEs approach their exact counterparts as the basis set size increases. However, increasing the basis set or the active space sizes do not always lead to more accurate CAS-CI EKT IEs. Our investigation supports the Davidson et al.'s observation [E. R. Davidson, et al., J. Chem. Phys. 155, 051102 (2021)] that the FCI EKT IEs can be systematically improved with arbitrary numerical accuracy by supplementing the basis set with diffuse functions of appropriate symmetry which allow the detached electron to travel far away from the reference system. By changing the exponent and the center of the diffuse functions, our results delineate a complex pattern for the CAS-CI EKT IE of LiH which can be important for the spectroscopic studies of small molecules.

physics.chem-ph↗

arXiv2019

Global hybrid multiconfiguration pair-density functional theory

A global hybrid extension of variational two-electron reduced-density matrix (v2RDM)-driven multiconfiguration pair-density functional theory (MCPDFT) is developed. Using a linear decomposition of the electron-electron repulsion term, a fraction λ of the nonlocal exchange interaction, obtained from v2RDM-driven complete active-space self-consistent field (CASSCF) theory, is combined with its local counterpart, obtained from an on-top pair-density functional. The resulting scheme (called λ-MCPDFT) inherits the benefits of MCPDFT (e.g., its simplicity and the resolution of the symmetry dilemma), and, when combined with the v2RDM approach to CASSCF, λ-MCPDFT requires only polynomially scaling computational effort. As a result, it can efficiently describe static and dynamical correlation effects in strongly correlated systems. The efficacy of the approach is assessed for several challenging multiconfigurational problems, including the dissociation of molecular nitrogen, the double dissociation of a water molecule, and the 1,3-dipolar cycloadditions of ozone to ethylene and ozone to acetylene in the O3ADD6 benchmark set.

physics.chem-ph↗

arXiv2018

Combining pair-density functional theory and variational two-electron reduced-density matrix methods

Complete active space self-consistent field (CASSCF) computations can be realized at polynomial cost via the variational optimization of the active-space two-electron reduced-density matrix (2-RDM). Like conventional approaches to CASSCF, variational 2-RDM (v2RDM)-driven CASSCF captures nondynamical electron correlation in the active space, but it lacks a description of the remaining dynamical correlation effects. Such effects can be modeled through a combination of v2RDM-CASSCF and on-top pair-density functional theory (PDFT). The resulting v2RDM-CASSCF-PDFT approach provides a computationally inexpensive framework for describing both static and dynamical correlation effects in multiconfigurational and strongly correlated systems. On-top pair-density functionals can be derived from familiar Kohn-Sham exchange-correlation (XC) density functionals through the translation of the v2RDM-CASSCF reference densities [Li Manni et al., J. Chem. Theory Comput. 10, 3669-3680 (2014)]. Translated and fully-translated on-top PDFT versions of several common XC functionals are applied to the potential energy curves of N2, H2O, and CN-, as well as to the singlet/triplet energy splittings in the linear polyacene series. Using v2RDM-CASSCF-PDFT and the translated PBE functional, the singlet/triplet energy splitting of an infinitely-long acene molecule is estimated to be 4.87 kcal/mol.

physics.chem-ph↗