Table of Contents
Fetching ...

Identification of an Unreported Structure Type in GdNiSn4 and Its Implications for Materials Prediction

Xin Zhang, Scott B. Lee, Sudipta Chatterjee, Hanqi Pi, Yi Yang, Fatmagül Katmer, Emily G. Ward, Daniel E. Widdowson, Charles C. Tam, Sarah Schwarz, Connor J. Pollak, Jaime M. Moya, Grigorii Skorupskii, Vitaliy A. Kurlin, Stephen D. Wilson, B. Andrei Bernevig, Leslie M. Schoop

Abstract

Crystal structures define how matter is organized at the atomic level. In the realm of crystalline inorganic materials, new structure types are rarely found, and most experimentally-realized structural motifs were established decades ago. Considerable efforts are underway to discover new crystalline inorganic compounds, often aided by artificial intelligence (AI). However, thus far, these methods have not yielded convincing new structure types, but rather substitutional variations of existing compounds. Here we introduce a new structure type adopted by the compound GdNiSn4, discovered the old-fashioned way. We test whether current state-of-the-art AI-based material generation models can predict this material in its correct structure and find that they fail to do so. We carefully analyze the new structure and argue that it can be viewed as a stack of two known structure types. We explore electronic and steric factors underlying its stability and propose strategies to improve future AI-guided materials discovery. Furthermore, we report complex magnetic properties in GdNiSn4, highlighting its potential interest for future studies of unconventional magnetism.

Identification of an Unreported Structure Type in GdNiSn4 and Its Implications for Materials Prediction

Abstract

Crystal structures define how matter is organized at the atomic level. In the realm of crystalline inorganic materials, new structure types are rarely found, and most experimentally-realized structural motifs were established decades ago. Considerable efforts are underway to discover new crystalline inorganic compounds, often aided by artificial intelligence (AI). However, thus far, these methods have not yielded convincing new structure types, but rather substitutional variations of existing compounds. Here we introduce a new structure type adopted by the compound GdNiSn4, discovered the old-fashioned way. We test whether current state-of-the-art AI-based material generation models can predict this material in its correct structure and find that they fail to do so. We carefully analyze the new structure and argue that it can be viewed as a stack of two known structure types. We explore electronic and steric factors underlying its stability and propose strategies to improve future AI-guided materials discovery. Furthermore, we report complex magnetic properties in GdNiSn4, highlighting its potential interest for future studies of unconventional magnetism.
Paper Structure (15 sections, 6 figures, 2 tables)

This paper contains 15 sections, 6 figures, 2 tables.

Figures (6)

  • Figure 1: Monoclinic GdNiSn4 crystal structure.(a) The overall GdNiSn4 structure ($C2/m$, No. 12) composed of alternating GdSn2 and NiSn2 units. The inset shows a single crystal of GdNiSn4. (b) The ZrGa2-type GdSn2 unit consists of uncoordinated Gd atoms occupying sites adjacent to the Sn zigzag chains that run along the crystallographic $a$ axis, forming a corrugated GdSn2 sheet. (c) In-plane projection down $b$ of the PdSn2-type/CoGe2-type NiSn2 unit, highlighting the connectivity between the $4^{4}$ Sn square-net and $3^{2}434$ Sn-dimer sublayers. The Sn-dimer sublayer links the square-net sublayers above and below and is interleaved with an array of Ni dimers. (d) Projection down $c^*$ (normal to $ab$) showing the $4^{4}$ Sn square-net sublayer. (e) Projection down $c^*$ showing the $3^{2}434$ Sn-dimer sublayer. Dashed lines help visualize the $3^{2}434$ vertex configuration. Bonds are drawn for all contacts shorter than 3.15$\,$Å. In panels where axes are labeled $a$, $b$, and $c^*$, $c^*$ denotes the reciprocal-lattice vector normal to the $ab$-plane as $c$ is not orthogonal to $ab$ in the monoclinic cell.
  • Figure 2: GdNiSn4 and LuNiSn4 crystal structures.(a)GdNiSn4 crystallizes in the monoclinic space group $C2/m$ (No. 12) and can be described as an alternation of ZrGa2-type GdSn2 units and PdSn2/CoGe2-type NiSn2 units. (b) The reportedSkolozdra2000NewPropertiesLuNiSn4 structure crystallizes in the orthorhombic space group $Cmmm$ (No. 65) and can be described as an alternation of ZrGa2-type LuSn2 units and PtHg2-type NiSn2 units. Bonds are drawn for all contacts shorter than 3.15$\,$Å.
  • Figure 3: Electron localization function (ELF) isosurfaces and in-plane cutouts for monoclinic and orthorhombic GdNiSn4.(a) Monoclinic GdNiSn4 with an in-plane isosurface cut through the $3^{2}434$ Sn-dimer sublayer. The Sn zigzag chain is delocalized along $a$, and the connected $4^{4}$ Sn square-nets are delocalized along the $ab$ plane. Comparatively, the $3^{2}434$ Sn-dimer sublayer exhibits enhanced localization between the Sn--Sn dimers. The isosurface level was set to $\eta=0.45$ for the monoclinic structure. (b) Orthorhombic GdNiSn4 with an in-plane isosurface cut through the middle $4^{4}$ square-net layer. Similarly, the Sn zigzag chain, along with the top/bottom $4^{4}$ square-nets, is delocalized in the same way as the monoclinic structure. The middle $4^{4}$ square-net exhibits slightly enhanced localization between the Sn--Sn bonds in comparison to the top/bottom $4^{4}$ square-nets, but substantially weaker than in the monoclinic $3^{2}434$ Sn-dimer sublayer. The isosurface level was set to $\eta=0.40$ for the orthorhombic structure. All ELF results shown here were computed and visualized in the primitive cells of the monoclinic/orthorhombic structures.
  • Figure 4: Compositional phase diagram visual of Gd-Ni-Sn. The GdNiSn4 structure can be described as alternating units of CoGe2-type NiSn2 and ZrSi2-type GdSn2 fragments. The binary endpoints are shown in their experimentally realized structures in the Gd--Ni--Sn system (e.g., the GdSn2 binary adopts the ZrSi2 structure-type.)
  • Figure 5: Temperature-dependent magnetic susceptibility and zero-field resistivity of GdNiSn4.(a) Measurements were performed at $0.1\,\mathrm{T}$ under zero-field cooled (ZFC), field-cooled cooling (FCC), and field-cooled warming (FCW) conditions for three orientations: $H\perp ab$ (top), $H\parallel a$ (middle), $H\parallel b$ (bottom). Figures on the right magnify the temperature region to 2--30$\,\mathrm{K}$. Two antiferromagnetic (AFM) transitions are observed at $T_{\mathrm{N}}=25.8\,\mathrm{K}$ and $T_{\mathrm{2}}=15.4\,\mathrm{K}$ (dashed lines) for all orientations. A small FCC--FCW splitting of $\approx1\,\mathrm{K}$ is observed around $T_{\mathrm{2}}$ for all orientations. (b) Temperature-dependent resistivity measurements were performed from $300\,\mathrm{K}$ to $1.8\,\mathrm{K}$ with current applied along the crystallographic $b$-axis ($\rho_{yy}$, $I \parallel y$). The residual resistivity ratio is calculated as $\mathrm{RRR}=\rho(300\,\mathrm{K})/\rho(4\,\mathrm{K})$. The figure on the right highlights the low-temperature region, which exhibits anomalies near $T_{\mathrm{N}}$ and $T_{\mathrm{2}}$. The asterisks indicate a superconducting transition attributed to residual tin flux in the measured crystal.
  • ...and 1 more figures