Phase determination with and without deep learning.

Area of Science:

• Statistical Physics
• Machine Learning Applications

Background:

• Phase transition detection is crucial in statistical physics.
• Machine learning offers novel approaches beyond traditional methods.
• Unsupervised learning is a key area for these new techniques.

Purpose of the Study:

• To evaluate unsupervised learning for phase transition detection in the J1-J2 Ising model.
• To compare a simple configuration comparison method with variational autoencoders.
• To demonstrate efficient machine learning applications in statistical physics.

Main Methods:

• Utilized the J1-J2 Ising model on a square lattice.
• Developed a simple method comparing configurations using reconstruction error.
• Contrasted results with variational autoencoder-generated configurations.

Main Results:

• The simple configuration comparison method accurately identified critical temperatures.
• This straightforward approach yielded results comparable to complex neural networks.
• Effectiveness demonstrated in both simple and complex scenarios.

Conclusions:

• Simple machine learning methods can be highly effective for phase transition detection.
• This work provides an efficient machine determination technique for critical phenomena.
• Highlights the potential of accessible ML tools in statistical physics research.