mat-xrd-phase-analysis

XRD Phase Analysis (DARA Tree Search)

Goal

Identify crystalline phases present in an XRD pattern by using DARA's phase search (parallelized tree search with BGMN), following the DARA tutorial (Tutorial 2: Phase analysis with tree search).

This skill:

• Fetches candidate CIFs for a given chemical system (e.g. Ge-O-Zn) using either the COD or ICSD database.
• Runs DARA's search_phases to search mixtures of phases.
• Saves plots and a human-readable report alongside the pattern.

Database Sources

DARA searches for structures in experimental databases. It supports:

1. CODDatabase (default, --database cod): Open-access. DARA will attempt to download CIFs from crystallography.net. If the website is offline or unreachable, DARA will automatically look for a local copy at ~/COD_2024 (or the path defined in ~/.dara.yaml).
2. ICSDDatabase (--database icsd): Commercial database. Requires a local copy, default path ~/ICSD_2024 (or configured in ~/.dara.yaml). Online download is not supported for ICSD.

Requirements

• Conda environment: xrd-agent (see conda-envs/xrd-agent).
• Dependencies: dara-xrd (and its Ray/BGMN stack).
• BGMN: DARA will prompt to download or use a local BGMN installation.
• For clusters where only some nodes have internet:
  • You can optionally use a two-step workflow (download CIFs on a node with internet, run Ray search elsewhere).

Script

scripts/phase_search.py

• Uses:
  • CODDatabase.get_cifs_by_chemsys(chemical_system, dest_dir=...)
  • search_phases(pattern_path, phases, wavelength, instrument_profile)
• Output directory (default):
  • phase_analysis_results/ under the same directory as the XRD pattern.

Usage

Use this script when the node where you run it can reach the database servers (for automatic COD CIF download), or when you have a local directory of CIFs/local database installed.

# Env: xrd-agent
python .agents/skills/mat-xrd-phase-analysis/scripts/phase_search.py \
  --xrd_data .agents/skills/mat-xrd-phase-analysis/examples/GeO2-ZnO/GeO2-ZnO_700C_60min.xrdml \
  --chemical_system "Ge-O-Zn" \
  --database icsd

This will:

1. Create phase_analysis_results/ next to the XRD file.
2. Search and filter CIFs for Ge-O-Zn from the ICSD into phase_analysis_results/cifs/.
3. Start a local Ray cluster.
4. Run search_phases(...).
5. Write plots and reports into phase_analysis_results/.

You can also provide your own CIFs:

python .../phase_search.py \
  --xrd_data pattern.xrdml \
  --cif_dir /path/to/my_cifs

In this case no database lookup is used; only --cif_dir is searched.

Specific example:

python .../phase_search.py \
  --xrd_data .agents/skills/mat-xrd-phase-analysis/examples/GeO2-ZnO/GeO2-ZnO_700C_60min.xrdml \
  --cif_dir .agents/skills/mat-xrd-phase-analysis/examples/GeO2-ZnO/phase_analysis_results/cifs

Arguments (phase_search.py)

• --xrd_data (required):
  • Path to XRD pattern (.xy, .xrdml, or .raw).
• --chemical_system:
  • String like "Ge-O-Zn". Used with COD or ICSD to fetch CIFs when --cif_dir is not given.
• --cif_dir:
  • Directory containing .cif files to search. When set, skips COD/ICSD lookup.
• --database (optional):
  • cod (default) or icsd. Specifies which database to use when using --chemical_system.
• --output_dir:
  • Custom output directory. Default: <xrd_dir>/phase_analysis_results/.
• --wavelength (optional):
  • X-ray source: Cu, Co, Cr, Fe, Mo, or a numeric wavelength in nm. Default: Cu.
• --instrument_profile (optional):
  • BGMN instrument profile. Default: Aeris-fds-Pixcel1d-Medipix3.
• --quiet:
  • Suppress verbose logging/prints.

Outputs

All outputs go under <output_dir> (default: phase_analysis_results/ under the XRD directory):

• CIFs

  • cifs/: Candidate phases for the chemical system (when downloaded).

• Search results

  • results_summary.json:
    • pattern, chemical_system, cif_dir
    • num_solutions, best_rwp
    • solutions: list of {rank, rwp, phase_files}.
  • phase_search_report.txt:
    • Text report similar to the DARA tutorial:
      • Total number of solutions and best Rwp.
      • Rwp of solution i = ... % for each solution.
      • Phases found in solution 0: with grouped phase file names.
      • Paths to the results directory and summary JSON.

• Plots

  • solution_0_refinement.html, solution_0_refinement.png
  • solution_1_refinement.html, solution_1_refinement.png, etc. (if multiple solutions)
  • Plots come from SearchResult.visualize() (observed / calculated / difference pattern, like the tutorial).

Notes and Constraints

• Ray / cluster config:
  • This script assumes you can start a local Ray cluster with ray.init(address="local", ...).
  • On HPC systems with a preconfigured Ray/GCS, you may need a clean environment or a job script that does not inherit cluster-level Ray environment variables.
• Internet vs no-internet:
  • On clusters with internet on compute nodes, use Mode A (single-step).
  • On clusters with internet only on login nodes, use Mode B (two-step).
• Performance:
  • Phase search can be CPU-intensive; adjust num_cpus in the script’s ray.init(...) if needed.

Related Skills

• mat-xrd-digitizer:
  • Use this skill first to convert an image/screenshot of an XRD plot into the .xy file required by this phase analysis skill.
• mat-xrd-refinement:
  • For known-phase Rietveld refinement (given specific CIFs).
• mat-xrd-calculator:
  • For calculating theoretical XRD patterns from crystal structures.

---

Author: Nofit Segal Contact: GitHub @nofitsegal