Discovery of Physics From Data: Universal Laws and Discrepancies

Area of Science:

• Physics
• Data Science
• Computational Science

Background:

• Machine learning (ML) and artificial intelligence (AI) are increasingly used for automated scientific discovery.
• Inferring universal physical laws from data alone is challenging due to measurement noise and unmodeled physics.

Purpose of the Study:

• To investigate the challenges in automated physics discovery using real-world data.
• To evaluate the effectiveness of data-driven methods in uncovering fundamental physical laws.

Main Methods:

• Utilized the sparse identification of non-linear dynamics (SINDy) method to identify governing equations.
• Applied SINDy to real-world measurement data and simulated trajectories of falling objects.
• Incorporated assumptions of shared physical laws across different objects to enhance model robustness.

Main Results:

• Measurement noise and complex secondary mechanisms, such as unsteady fluid drag, can obscure the underlying law of gravitation.
• While shared physical law assumptions improve model robustness, discrepancies persist due to drag dynamics subtleties.
• Naive application of ML/AI is insufficient for inferring universal physical laws without modifications.

Conclusions:

• Automated physics discovery requires more than just data-driven algorithms; discrepancy modeling is crucial.
• Subtleties in physical phenomena like fluid dynamics pose significant challenges for current ML/AI approaches.
• Future research should focus on integrating physical constraints and advanced modeling techniques for robust scientific discovery.