.png){kind=icon}
.png){kind=icon}
Return
Predicting superconducting temperatures with new hierarchical neural network AI model
.png){kind=icon}
PRE
AI
X
C
Q
J
X
DOI：10.15302/frontphys.2025.014205.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Superconducting critical temperature is the most attractive material property due to its impact on the applications of electricity transmission, railway transportation, strong magnetic fields for nuclear fusion and medical imaging, quantum computing, etc. The ability to predict its value is a constant pursuit for condensed matter physicists. We developed a new hierarchical neural network (HNN) AI algorithm to resolve the contradiction between the large number of descriptors and the small number of datasets always faced by neural network AI approaches to materials science. With this new HNN-based AI model, a much-increased number of 909 universal descriptors for inorganic compounds, and a dramatically cleaned database for conventional superconductors, we achieved high prediction accuracy with a test R-2 score of 95.6%. The newly developed HNN model accurately predicted T-c of 45 new high-entropy alloy super-conductors with a mean absolute percent error below 6% compared to the experimental data. This demonstrated a significant potential for predicting other properties of inorganic materials.
Keywords:
conventional superconducting critical temperature
hierarchical neural network
universal descriptors
artificial intelligence
Journal
IF:
5.3
Papers:
1.4K
Citations:
3.7K
Organization
Cited Papers