Scale-Invariant Quantum Machine Learning: Zero-Shot Transfer Across Molecular Dimensions via Information-Theoretic Relational Calculus

Key Points

• This research aims to enhance molecular energy predictions using a novel information-theoretic approach.
• Developed an information-theoretic relational calculus for projecting molecular energies.
• Utilized a lightweight XGBoost model trained on H₂ and tested zero-shot on LiH.
• Focused on reducing absolute prediction error while addressing dimensionality drift.
• Achieved a reduction in absolute prediction error by 80% using the trained model.
• Successfully eliminated dimensionality drift without employing deep neural networks.