Inverse molecular design and parameter optimization with Hückel theory using automatic differentiation

Abstract

Semi-empirical quantum chemistry has recently seen a renaissance with applications in high-throughput virtual screening and machine learning. The simplest semi-empirical model still in widespread use in chemistry is Hückel's $\pi$-electron molecular orbital theory. In this work, we implemented a Hückel program using differentiable programming with the JAX framework, based on limited modifications of a pre-existing NumPy version. The auto-differentiable Hückel code enabled efficient gradient-based optimization of model parameters tuned for excitation energies and molecular polarizabilities, respectively, based on as few as 100 data points from density functional theory simulations. In particular, the facile computation of the polarizability, a second-order derivative, via auto-differentiation shows the potential of differentiable programming to bypass the need for numeric differentiation or derivation of analytical expressions. Finally, we employ gradient-based optimization of atom identity for inverse design of organic electronic materials with targeted orbital energy gaps and polarizabilities. Optimized structures are obtained after as little as 15 iterations, using standard gradient-based optimization algorithms.

In Silico Chemical Experiments in the Age of AI: From Quantum Chemistry to Machine Learning and Back

“…Several strategies leverage the synergy between differentiable physics and data-driven schemes. These include improvements to the Hamiltonian facilitated by Δ-ML, exemplified by the Artificial Intelligence-Quantum Mechanical Method 1 (AIQM1), [304,305] backpropagation through Hamiltonian parameters, as seen in the ML-enhanced extended Hückel model (ML-EHM), [306,307] fitting of repulsive contributions and electronic integrals in density functional tight-binding (DFTB), [308][309][310] predicting the energy from DFTB solution as in OrbNet, [311] and full analytical mapping of the electronic Hamiltonian, as enabled by the ML-augmented atomic cluster expansion (ML-ACE). [312] For further insight, readers are directed to Ref.…”

Size Isn’t Everything: Geometric Tuning in Polycyclic Aromatic Hydrocarbons and Its Implications for Carbon Nanodots

“…The Hückel method is one of the older and more famous methods, which is essentially based on connectivity between atoms . Its use is typically relegated to an educational tool, or for low accuracy calculations; however, due to its relative efficiency, Hückel’s method continues to see some interest for large data set screening . Most other models are based on DFT results, ,,− with outcomes largely in agreement with CSR and LLS for transition energies.…”

Refining DIIS algorithms for Si and GaAs solar cells: incorporation of weight regularization, conjugate gradient, and reverse automatic differentiation techniques

“…Moreover, the innovative approach by Yoshikawa et al 30 in applying reverse-mode AD to the Hartree–Fock method showcases the method's stability and efficiency improvements over traditional SCF methods. Vargas–Hernández et al 's 31 utilization of differentiable programming within the Hückel's π-electron molecular orbital theory emphasizes AD's capacity for parameter optimization and its potential in materials design. Tan et al 's 32 development of PROFESS-AD in the context of orbital-free density functional theory underscores the broader applicability of AD in materials simulation, enabling direct evaluation of complex derivatives and supporting the development of new functionals.…”