Scholarlink

<p>Powder X-ray diffraction (PXRD) is a vital technique for the structural characterization of crystalline compounds. However, this analysis is challenged by practically encountered multi-phase systems, whose mixed PXRD patterns necessitate prior phase decomposition. Traditional approaches require multiple mixture samples or prior knowledge of the constituent phases, limiting their applicability in complex or high-throughput scenarios. Here we show that <i>multi-phase PXRD patterns from a single observation can be automatically decomposed into their constituent single-phase patterns using machine learning</i>. The proposed solution—Phase Decomposition Diffusion Transformers (PhaseDifformer)—reinterprets the denoising process of diffusion models as a probabilistic regressor, enabling recursive extraction of unknown constituent phases. We validate the method on both synthetic mixtures and experimental measurements, achieving accurate phase decomposition in both cases. Given ongoing progress in single-phase PXRD-to-structure methods, our study fills a critical gap toward realizing fully automated, end-to-end analysis of complex PXRD patterns commonly encountered in materials research.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enabling single-observation decomposition of multi-phase X-ray diffraction patterns via generative deep learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月8日

查看全文
  • Yusei Ito,
  • Naoya Chiba,
  • Tatsunori Taniai,
  • Ryo Igarashi,
  • Yuta Suzuki,
  • Kotaro Saito,
  • Yoshitaka Ushiku,
  • Kanta Ono

摘要

Powder X-ray diffraction (PXRD) is a vital technique for the structural characterization of crystalline compounds. However, this analysis is challenged by practically encountered multi-phase systems, whose mixed PXRD patterns necessitate prior phase decomposition. Traditional approaches require multiple mixture samples or prior knowledge of the constituent phases, limiting their applicability in complex or high-throughput scenarios. Here we show that multi-phase PXRD patterns from a single observation can be automatically decomposed into their constituent single-phase patterns using machine learning. The proposed solution—Phase Decomposition Diffusion Transformers (PhaseDifformer)—reinterprets the denoising process of diffusion models as a probabilistic regressor, enabling recursive extraction of unknown constituent phases. We validate the method on both synthetic mixtures and experimental measurements, achieving accurate phase decomposition in both cases. Given ongoing progress in single-phase PXRD-to-structure methods, our study fills a critical gap toward realizing fully automated, end-to-end analysis of complex PXRD patterns commonly encountered in materials research.