Manta Ray AI optimizes crystal structure prediction with Levy Flight
New hybrid algorithm beats local optima in predicting formation energies of binary crystals.
A new paper by Adrian Rubio-Solis introduces EELM-MRFO-LF, an evolutionary extreme learning machine that uses Manta Ray Foraging Optimization (MRFO) with Levy Flight to predict crystal structure formation energies. The method improves on standard ELMs by using MRFO to select input weights and the Moore-Penrose generalized inverse to compute output weights analytically. Levy Flight adds random jumps to the optimization process, increasing population diversity and preventing the algorithm from getting stuck in local optima.
The EELM-MRFO-LF was tested on predicting both unrelaxed and relaxed formation energies of binary compounds, comparing against other nature-inspired algorithms. Results showed enhanced accuracy and robustness for crystal structure prediction, a key challenge in materials science. The approach bridges bio-inspired optimization and neural networks for computational chemistry.
For practical applications, this method could accelerate the discovery of new materials by predicting stable crystal structures without expensive quantum mechanical calculations. The 8-page paper (arXiv:2605.17148) demonstrates how hybrid AI techniques can tackle complex physics problems with improved convergence and solution quality.
- Combines Manta Ray Foraging Optimization with Levy Flight to train Extreme Learning Machines
- Uses Moore-Penrose pseudo-inverse for analytic output weight calculation, reducing training time
- Tested on binary crystal structure energy landscapes, outperforming other nature-inspired algorithms
Why It Matters
Speeds up materials discovery by predicting stable crystal structures without costly quantum simulations.