Transfer learning from Hermitian to non-Hermitian quantum many-body physics

Area of Science:

• Quantum many-body physics
• Machine learning applications
• Condensed matter theory

Background:

• Identifying phase boundaries is crucial for understanding quantum many-body models.
• Progress has been made in Hermitian systems, but non-Hermitian models face numerical and analytical challenges.
• Machine learning offers a new approach to predict phase boundaries from observables.

Purpose of the Study:

• To investigate if machine learning models trained on Hermitian systems can identify phase boundaries in non-Hermitian models.
• To demonstrate the effectiveness of transfer learning in non-Hermitian physics.
• To establish a versatile strategy for analyzing non-Hermitian phenomena.

Main Methods:

• Utilizing a machine learning methodology.
• Training the model solely on Hermitian correlation functions.
• Applying the trained model to non-Hermitian interacting models.

Main Results:

• The machine learning methodology successfully identified phase boundaries of non-Hermitian interacting models.
• Hermitian-trained machine learning algorithms were redeployed to non-Hermitian models without retraining.
• Non-Hermitian phase diagrams were revealed using this transfer learning approach.

Conclusions:

• Hermitian machine learning algorithms can be effectively applied to non-Hermitian models.
• Transfer learning is a versatile strategy for leveraging Hermitian physics in machine learning non-Hermitian phenomena.
• This approach overcomes limitations in analyzing non-Hermitian many-body systems.