Pareto front analysis and multi-objective Bayesian optimization for (R, Z)(Fe,Co,Ti)12 (R = Y, Nd, Sm; Z = Zr, Dy)

TL;DR

This paper employs multi-objective Bayesian optimization to analyze the Pareto front of ThMn12-type compounds, revealing trade-offs among magnetization, Curie temperature, and cost, and demonstrating MBO's effectiveness in approximating the Pareto front.

Contribution

It introduces a multi-objective Bayesian optimization framework to study trade-offs in complex magnetic compounds, providing insights into controllable variables and cost-performance relationships.

Findings

Trade-off relationships between cost, Curie temperature, and composition identified.

MBO effectively approximates the Pareto front when true Pareto front is difficult to obtain.

Reduction in Dy and increase in Co favor low cost and high Curie temperature.

Abstract

We propose a scheme for investigating the correlation and trade-off among target variables using a multi-objective Bayesian optimization (MBO). We discuss the features of the Pareto front (PF) of ThMn12-type compounds, (R, Z)(Fe,Co,Ti)12 (R = Y, Nd, Sm; Z = Zr, Dy) in terms of magne- tization, Curie temperature, and a price index by using data from first-principles calculations, and we extract the trade-off relations from the analysis. We show that the trade-off relationships can be used to determine changes in the controllable variables by using partial least squares regression. For example, the tendency toward low cost and high Curie temperature is related to the reduction in Dy and increase in Co. We also discuss the efficiency of MBO as a practical scheme to obtain the features of the PF. We show that MBO can offer an approximated set for the PF even when obtaining the true PF is difficult.