Absorption Spectra with Kernel Polynomial Neural Quantum States.

Area of Science:

• Quantum Chemistry
• Computational Physics
• Materials Science

Background:

• Predicting optical absorption spectra for correlated quantum systems is computationally intensive due to the exponential scaling of many-body wave functions.
• Existing methods often require costly explicit calculations of excited states, limiting their applicability to larger systems.

Purpose of the Study:

• To introduce a novel generative framework, kernel polynomial neural quantum states (KPNQS), for efficient spectral property prediction.
• To overcome the computational limitations of traditional methods for correlated quantum systems.

Main Methods:

• KPNQS unifies the kernel polynomial method (KPM) with autoregressive neural wave functions.
• The framework computes spectral properties directly from the ground state by efficiently evaluating KPM moments.
• This approach avoids explicit excited-state calculations, enabling scalable linear response evaluations.

Main Results:

• KPNQS achieves exact agreement with full configuration interaction for molecular systems up to 52 electrons.
• The method demonstrates effective polynomial scaling, overcoming exponential limitations.
• The framework is architecture-agnostic, offering broad applicability.

Conclusions:

• KPNQS provides a scalable and broadly applicable paradigm for excited-state-free spectral modeling in correlated matter.
• This advancement significantly reduces the computational cost of predicting optical absorption spectra.
• The KPNQS framework opens new possibilities for studying complex quantum systems.