Chemically Controllable Magnetic Transition Temperature and Magneto‐Elastic Coupling in MnZnSb Compounds

Abstract

Magneto‐caloric materials offer the possibility to design environmentally friendlier thermal management devices compared to the widely used gas‐based systems. The challenges to develop this solid‐state based technology lie in the difficulty of finding materials presenting a large magneto‐caloric effect over a broad temperature span together with suitable secondary application parameters such as low heat capacity and high thermal conductivity. A series of compounds derived from the PbFCl structure is investigated using a combination of computational and experimental methods focusing on the change of cell volume in magnetic and non‐magnetic ground states. Scaling analysis of the magnetic properties determines that they are second order phase transition ferromagnets and that the magnetic entropy change is driven by the coupling of magneto‐elastic strain in the square‐net through the magnetic transition determined from neutron and synchrotron X‐ray diffraction. The primary and secondary application related properties are measured experimentally, and the c/a parameter is identified as an accurate proxy to control the magnetic transition. Chemical substitution on the square‐net affords tuning of the Curie temperature over a broad temperature span between 252 and 322 K. A predictive machine learning model for the c/a parameter is developed to guide future exploratory synthesis.

Random projections and kernelised leave one cluster out cross validation: universal baselines and evaluation tools for supervised machine learning of material properties

“…A further consideration in use of LOCO-CV is that K-means does not guarantee the size of any clusters, nor does it guarantee that clusters would be deemed chemically sensible (this is discussed further in section 2.4). It has been observed that clusters taken on materials data can vary in size by multiple orders of magnitude, which hinders the application of LOCO-CV 9 .…”

“…It has been observed that clusters taken on materials data can vary in size by multiple orders of magnitude, which hinders the application of LOCO-CV. 9…”

“…While differences in cluster sizes in this domain are expected, it has been observed that clusters found with K-means can differ in size by orders of magnitude 9 , which can pose a practical challenge to adoption of this method. With such differences in cluster size, LOCO-CV measurements can represent the performance of an algorithm on a small training set rather than the performance of an algorithm in extrapolation.…”

Random projections and Kernelised Leave One Cluster Out Cross-Validation: Universal baselines and evaluation tools for supervised machine learning for materials properties

“…A further consideration in use of LOCO-CV is that K-means does not guarantee the size of any clusters, nor does it guarantee that clusters would be deemed chemically sensible (this is discussed further in section 2.4). It has been observed that clusters taken on materials data can vary in size by multiple orders of magnitude, which hinders the application of LOCO-CV 7 .…”

“…While differences in cluster sizes in this domain are expected, it has been observed that clusters found with K-means can differ in size by orders of magnitude 7 , which can pose a practical challenge to adoption of this method. With such differences in cluster size, LOCO-CV measurements can represent the performance of an algorithm on a small training set rather than the performance of an algorithm in extrapolation.…”