Giant room-temperature magnetocaloric effect MM′X alloys explored by machine learning

Magnetic refrigeration is emerging as a promising alternative to conventional gas compression refrigeration. The key to advancing this technology lies in identifying materials with high magnetocaloric effect (MCE). However, traditional experimental methods require time-consuming and labor-intensive experimentation to screen compositions. Herein, we developed a machine learning (ML) model utilizing the Random Forest algorithm to efficiently identify MM'X alloy compositions with giant MCE near room temperature based on a small dataset (<200). The ML-predicted results were perfectly confirmed by the validation experiments. Notably, while most typical magnetic refrigeration materials near room temperature show a maximum ∆S around or <20 kg^-1 K^-1(0–5 T), our predicted alloys show a remarkable ∆S of up to 51 kg^-1 K^-1, which is more than four times higher than that of the benchmark material, Gd. The origin of this giant MCE is linked to atomic disorder and lattice ∆S predominantly from Ni atoms, as confirmed through spherical aberration corrected transmission electron microscopy, neutron powder diffraction and element-resolved vibrational density of states analysis. This work accelerates the discovery of magnetic refrigerants and significantly promotes the development of magnetic refrigeration.