Fig. 1: QiankunNet: Neural Network Quantum State (NNQS) Transformer pipeline.
From: Solving the many-electron Schrödinger equation with a transformer-based framework
a The decoder-only Transformer architecture for the electron wave function ansatz, where the Transformer generates the amplitude ∣ψ(x)∣ and a multi-layer perceptron (MLP) generates the phase ϕ(x). The wave function is expressed as ψ(x) = ∣ψ(x)∣eiϕ(x). b Training workflow of the variational Monte Carlo (VMC) optimization. Starting from physics-informed initialization, the model parameters are updated through gradient descent to minimize the energy expectation value. The process involves autoregressive sampling of electron configurations, local energy calculation, and parameter updates via backpropagation.
