Jassmaira Singh

Research Intern @Graphic Era Hill University

IN

Excited to share research investigating a fundamental challenge in neural network quantum simulation; why do these models fail when applied to excited quantum states?Working with the 1D quantum harmonic oscillator as a controlled testbed, this study shows that variational collapse toward ground state solutions is not a capacity problem, but a structural one. Neural networks, without physical guidance, simply lack the inductive biases needed to represent excited states reliably.By incrementally introducing physics informed constraints; parity, nodal structure and orthogonality, the work isolates what actually matters: symmetry alone does not prevent collapse. Explicit orthogonality enforcement is what preserves correct energy ordering and wave function fidelity.Findings were validated through energy convergence analysis, wave function overlap diagnostics and nodal position accuracy metrics, and presented at ICHM 2026.The next phase extends this framework to deeper architectures, higher excited states and more complex quantum systems, from the harmonic oscillator to the hydrogen atom, building toward a generalizable approach for neural network ansätze in excited state quantum simulation.

Manipal Institute of Technology

Bachelor of Technology, Electronics and Instrumentation