skip to main content

This content will become publicly available on June 14, 2023

Title: Harnessing interpretable and unsupervised machine learning to address big data from modern X-ray diffraction
The information content of crystalline materials becomes astronomical when collective electronic behavior and their fluctuations are taken into account. In the past decade, improvements in source brightness and detector technology at modern X-ray facilities have allowed a dramatically increased fraction of this information to be captured. Now, the primary challenge is to understand and discover scientific principles from big datasets when a comprehensive analysis is beyond human reach. We report the development of an unsupervised machine learning approach, X-ray diffraction (XRD) temperature clustering (X-TEC), that can automatically extract charge density wave order parameters and detect intraunit cell ordering and its fluctuations from a series of high-volume X-ray diffraction measurements taken at multiple temperatures. We benchmark X-TEC with diffraction data on a quasi-skutterudite family of materials, (Ca x Sr 1 − x ) 3 Rh 4 Sn 13 , where a quantum critical point is observed as a function of Ca concentration. We apply X-TEC to XRD data on the pyrochlore metal, Cd 2 Re 2 O 7 , to investigate its two much-debated structural phase transitions and uncover the Goldstone mode accompanying them. We demonstrate how unprecedented atomic-scale knowledge can be gained when human researchers connect the X-TEC results to more » physical principles. Specifically, we extract from the X-TEC–revealed selection rules that the Cd and Re displacements are approximately equal in amplitude but out of phase. This discovery reveals a previously unknown involvement of 5 d 2 Re, supporting the idea of an electronic origin to the structural order. Our approach can radically transform XRD experiments by allowing in operando data analysis and enabling researchers to refine experiments by discovering interesting regions of phase space on the fly. « less
; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ;
Award ID(s):
Publication Date:
Journal Name:
Proceedings of the National Academy of Sciences
Sponsoring Org:
National Science Foundation
More Like this
  1. Thermoelectric materials can convert heat into electricity. They are used to generate electricity when other power sources are not available or to increase energy efficiency by recycling waste heat. The Yb 21 Mn 4 Sb 18 phase was previously shown to have good thermoelectric performance due to its large Seebeck coefficient (∼290 μV K −1 ) and low thermal conductivity (0.4 W m −1 K −1 ). These characteristics stem respectively from the unique [Mn 4 Sb 10 ] 22− subunit and the large unit cell/site disorder inherent in this phase. The solid solutions, Yb 21 Mn 4− x Cd x Sb 18 ( x = 0, 0.5, 1.0, 1.5) and Yb 21− y Ca y Mn 4 Sb 18 ( y = 3, 6, 9, 10.5) have been prepared, their structures characterized and thermoelectric properties from room temperature to 800 K measured. A detailed look into the structural disorder for the Cd and Ca solid solutions was performed using synchrotron powder X-ray diffraction and pair distribution function methods and shows that these are highly disordered structures. The substitution of Cd gives rise to more metallic behavior whereas Ca substitution results in high resistivity. As both Cd and Ca aremore »isoelectronic substitutions, the changes in properties are attributed to changes in the electronic structure. Both solid solutions show that the thermal conductivities remain extremely low (∼0.4 W m −1 K −1 ) and that the Seebeck coefficients remain high (>200 μV K −1 ). The temperature dependence of the carrier mobility with increased Ca substitution, changing from approximately T −1 to T −0.5 , suggests that another scattering mechanism is being introduced. As the bonding changes from polar covalent with Yb to ionic for Ca, polar optical phonon scattering becomes the dominant mechanism. Experimental studies of the Cd solid solutions result in a max zT of ∼1 at 800 K and, more importantly for application purposes, a ZT avg ∼ 0.6 from 300 K to 800 K.« less
  2. The copper corrosion was studied for 30 days in two alkaline electrolytes: saturated Ca(OH)2 and cement extract, employed to simulate concrete-pore environments. Electrochemical Impedance Spectroscopy (EIS) and Cyclic Voltammetry were carried out at the open circuit potential (OCP), and potentiodynamic polarization (PDP) curves were performed for comparative purposes. Electrochemical current fluctuations, considered as electrochemical noise (EN), were employed as non-destructive methods. The tests revealed that sat. Ca(OH)2 is the less aggressive to the Cu surface, mainly because of the lower in one order pH. In consequence, the OCP values of Cu were more positive, the polarization resistance values were higher by one order of magnitude, and the anodic currents of Cu were lower than those in the cement extract. The analyzed EN indicated that the initial corrosion attacks on the Cu surface are quasi-uniform, resulting from the stationary persistent corrosion process occurring in both model solutions. XPS analysis and X-ray diffraction (XRD) patterns revealed that in sat. Ca(OH)2, a Cu2O/CuO corrosion layer was formed, which effectively protects the metallic Cu-surface. We present evidence for the sequential oxidation of Cu to the (+1) and (+2) species, its impact on the corrosion layer, and also its protective properties.
  3. Controlling the growth of complex relaxor ferroelectric thin films and understanding the relationship between biaxial strain–structural domain characteristics are desirable for designing materials with a high electromechanical response. For this purpose, epitaxial thin films free of extended defects and secondary phases are urgently needed. Here, we used optimized growth parameters and target compositions to obtain epitaxial (40–45 nm) 0.67Pb(Mg 1/3 Nb 2/3 )O 3 –0.33PbTiO 3 /(20 nm) SrRuO 3 (PMN–33PT/SRO) heterostructures using pulsed-laser deposition (PLD) on singly terminated SrTiO 3 (STO) and ReScO 3 (RSO) substrates with Re = Dy, Tb, Gd, Sm, and Nd. In situ reflection high-energy electron diffraction (RHEED) and high-resolution X-ray diffraction (HR-XRD) analysis confirmed high-quality and single-phase thin films with smooth 2D surfaces. High-resolution scanning transmission electron microscopy (HR-STEM) revealed sharp interfaces and homogeneous strain further confirming the epitaxial cube-on-cube growth mode of the PMN–33PT/SRO heterostructures. The combined XRD reciprocal space maps (RSMs) and piezoresponse force microscopy (PFM) analysis revealed that the domain structure of the PMN–33PT heterostructures is sensitive to the applied compressive strain. From the RSM patterns, an evolution from a butterfly-shaped diffraction pattern for mildly strained PMN–33PT layers, which is evidence of stabilization of relaxor domains, to disc-shaped diffraction patterns formore »high compressive strains with a highly distorted tetragonal structure, is observed. The PFM amplitude and phase of the PMN–33PT thin films confirmed the relaxor-like for a strain state below ∼1.13%, while for higher compressive strain (∼1.9%) the irregularly shaped and poled ferroelectric domains were observed. Interestingly, the PFM phase hysteresis loops of the PMN–33PT heterostructures grown on the SSO substrates (strain state of ∼0.8%) exhibited an enhanced coercive field which is about two times larger than that of the thin films grown on GSO and NSO substrates. The obtained results show that epitaxial strain engineering could serve as an effective approach for tailoring and enhancing the functional properties in relaxor ferroelectrics.« less
  4. CuZrO 3 has been hypothesized to be a catalytic material with potential applications for CO 2 reduction. Unfortunately, this material has received limited attention in the literature, and to the best of our knowledge the exact crystal structure is still unknown. To address this challenge, we utilize several different structural prediction techniques in concert, including the Universal Structure Predictor: Evolutionary Xtallography (USPEX), the Materials Project Structure Predictor, and the Open Quantum Materials Database (OQMD). Leveraging these structural prediction techniques in conjunction with Density-Functional Theory (DFT) calculations, we determine a possible structure for CuZrO 3 , which resembles a “sandwich” morphology. Our calculations reveal that this new structure is significantly lower in energy than a previously hypothesized perovskite structure, albeit it still has a thermodynamic preference to decompose into CuO and ZrO 2 . In addition, we experimentally tried to synthesize CuZrO 3 based on literature reports and compared computational to experimental X-ray Diffraction (XRD) patterns confirming that the final product is a mixture of CuO and ZrO 2 . Finally, we conducted a computational surface energetics and CO 2 adsorption study on our discovered sandwich morphology, demonstrating that CO 2 can adsorb and activate on the material. However, these COmore »2 adsorption results deviate from previously reported results further confirming that the CuZrO 3 is a metastable form and may not be experimentally accessible as a well-mixed oxide, since phase segregation to CuO and ZrO 2 is preferred. Taken together, our combined computational and experimental study provides evidence that the synthesis of CuZrO 3 is extremely difficult and if this oxide exists, it should have a sandwich-like morphology.« less
  5. Abstract

    We present an alternative and, for the purpose of non-crystalline materials design, a more suitable description of covalent and ionic glassy solids as statistical ensembles of crystalline local minima on the potential energy surface. Motivated by the concept of partially broken ergodicity, we analytically formulate the set of approximations under which the structural features of ergodic systems such as the radial distribution function (RDF) and powder X-ray diffraction (XRD) intensity can be rigorously expressed as statistical ensemble averages over different local minima. Validation is carried out by evaluating these ensemble averages for elemental Si and SiO2over the local minima obtained through the first-principles random structure sampling that we performed using relatively small simulation cells, thereby restricting the sampling to a set of predominantly crystalline structures. The comparison with XRD and RDF from experiments (amorphous silicon) and molecular dynamics simulations (glassy SiO2) shows excellent agreement, thus supporting the ensemble picture of glasses and opening the door to fully predictive description without the need for experimental inputs.