Machine Learning Study of Highly Spin-Polarized Heusler Alloys at Finite Temperature

Strong reduction of magnetoresistance (MR) ratio in Heusler alloys based MR devices at finite temperature raises the need for the development of new highly spin-polarized Heusler alloys. However, recent high throughput expoloration and machine learning studies combined with first principles calculations to find prospective Heusler alloys are only performed with 0 K assumption, which lead to the significant discrepancy of material prediction with experiments. In this work, we carried out the finite temperature first-principles calculation combined with machine learning to find new highly spin-polalized Heusler alloys. We employed Bayesian optimization for the machine learning algorithm and the disordered local moment method for the finite temperature first-principles calculations, respectively. We successfully found several new prospective Heusler alloys with high spin polarization at finite temperature such as Co
2
MnGa
0.2
As
0.8
and Co
2
FeAl
0.4
Sn
0.6
. Additionally, the effect of composition modulation on the temperature dependence of Co
2
MnGa
y
As
1-y
and Co
2
FeAl
y
Sn
1-y
was also studied.