WorksPhysics and Astronomy
HEXRD and HEXRDGUI: a flexible toolkit for X-ray diffraction analysis in dynamic compression experiments
Abstract
This article describes the application of HEXRD, an open-source Python library for X-ray diffraction analysis, and HEXRDGUI, its companion graphical application, to dynamic compression experiments. Built on a generic multi-detector geometric model, the toolkit supports arbitrary instrument configurations with independently positioned planar or cylindrical detector panels, arbitrary beam directions, and multiple X-ray sources. It calibrates instruments from powder rings, Laue spots or the rings of an unknown phase, fits whole patterns by Le Bail or Rietveld refinement, and imports data from the TARDIS, FIDDLE and PXRDIP platforms at the National Ignition Facility (NIF) and OMEGA EP. Although originally developed for high-energy diffraction microscopy at synchrotron sources, HEXRD has been adopted as an analysis tool for dynamic compression X-ray diffraction experiments at facilities including the NIF, the Matter in Extreme Conditions end-station at the Linac Coherent Light Source, the Dynamic Compression Sector (DCS) at the Advanced Photon Source, OMEGA EP at the Laboratory for Laser Energetics, and the High Energy Density (HED) instrument at the European XFEL. Calibrated from the same ring positions on a single planar detector, HEXRD and the independent code pyFAI predict the same scattering angle 2 to numerical precision. On a dual-source dataset from the NIF, calibrating the detectors against both X-ray sources at once fits the rings of both to an RMS of 0.18° in 2, about one image-plate pixel, whereas a fit to either source alone, with the plate tilts held at their nominal values, fits its own rings nearly as well but mispredicts the other's by about 0.6°. The software is freely available under a BSD license.
Introduction
In situ X-ray diffraction (XRD) is a primary diagnostic for probing the structure of matter under dynamic compression, where materials are subjected to extreme pressures and temperatures using laser or gas-gun drivers [72]. Over the past three decades, XRD measurements during dynamic loading have enabled a series of landmark observations. Early experiments demonstrated that the lattice of a shocked crystal could be probed directly on sub-nanosecond timescales [74], and later measurements resolved solid–solid phase transitions in situ, including the – transition in shock-compressed iron (Kalantar et al., 2005). The advent of X-ray free-electron lasers enabled femtosecond visualization of lattice dynamics and plasticity in shock-compressed copper [47] and tantalum [76], and shear-driven phase-transition lowering in silicon [45]. More recently, XRD has revealed the structures of planetary and exotic materials at extreme conditions, including a high-pressure phase of magnesium oxide at exoplanet pressures [19, 78], the formation of diamond and lonsdaleite from shocked graphite [36], superionic water ice [49], measurement of the high-pressure melt curve of iron [38], and the structure of liquid carbon [37]. These techniques have also extended crystallography into the terapascal regime, measuring solid tin to 1.2 TPa [41], revealing the exotic electride phase in elemental magnesium at TPa [25], and demonstrating the metastability of diamond ramp-compressed to 2 TPa [40].
These experiments typically employ non-standard instrument configurations: multiple detector panels at various orientations, pinhole collimation, curved image plates, and polychromatic or pink-beam X-ray sources. Facilities such as the National Ignition Facility (NIF) [61], the Matter in Extreme Conditions (MEC) end-station at the Linac Coherent Light Source (LCLS) [51, 11], the Dynamic Compression Sector (DCS) at the Advanced Photon Source (APS) [18], OMEGA EP at the Laboratory for Laser Energetics (LLE) [55], and the European X-ray Free Electron Laser (European XFEL) [26] each present unique instrumental challenges that general-purpose XRD software packages were not designed to address.
Established tools such as FIT2D [28], Dioptas [56], and pyFAI [1] provide excellent support for standard single and multi-detector setups at synchrotrons and XFEL sources, and DIALS [79] with its experiment toolbox
dxtbx (Parkhurst et al., 2014) models arbitrary multi-panel detector geometries and beam directions, though its data-reduction pipelines are designed primarily for single-crystal data. Comprehensive Rietveld refinement capabilities are also available in many open-source or commercial packages such as GSAS-II [67], MAUD (Lutterotti et al., 1999), FullProf [58], Profex [21], RIETAN [32], HighScore Plus (Degen et al., 2014), and TOPAS [16], to name a few. However, none of these packages natively supports the combination of powder and Laue calibration of multi-panel detectors with arbitrary orientations and beam directions, refinable detector distortion models, pinhole and target-assembly corrections, and facility-specific import workflows required for many dynamic compression experiments (Table 1). Many beamlines and research laboratories have developed home-grown software for their specific instruments, such as the IGOR Pro programs used at Lawrence Livermore National Laboratory (LLNL) for the Powder X-Ray Diffraction Image Plate (PXRDIP) and Target Diffraction In-Situ (TARDIS) platforms (Millot et al., 2026). Such programs can be effective, but each new user must learn a different tool for each facility. The TARDIS and PXRDIP analysis described by Rygg et al. (2020), for instance, already optimizes the image-plate geometry jointly with the direction of each X-ray source and corrects for the target assembly, but only for those platforms.
| Package | (1) | (2) | (3) | (4) | (5) | (6) | (7) |
| HEXRD / HEXRDGUI | • | • | • | • | ◦ | ◦ | • |
| FIT2D [28] | – | – | – | – | – | – | – |
| Dioptas [56] | ◦ | ◦ | – | – | – | – | ◦ |
| pyFAI (Ashiotis et al., 2015; Kieffer et al., 2020) | • | • | ◦ | – | – | – | ◦ |
| DIALS / dxtbx (Winter et al., 2018; Parkhurst et al., 2014) | • | ◦ | ◦ | – | – | – | • |
| GSAS-II [67] | ◦ | – | – | – | • | • | ◦ |
| MAUD (Lutterotti et al., 1999) | ◦ | ◦ | – | – | • | • | ◦ |
| FullProf [58] | – | – | – | – | • | • | – |
| TOPAS [16] | – | – | – | – | • | • | – |
Table 1. Capabilities relevant to dynamic compression XRD across representative open-source and commercial packages. Columns are: (1) multiple detector panels at arbitrary positions and orientations; (2) cylindrical detector panels; (3) multiple X-ray sources with independent energies and beam vectors; (4) pinhole and layered-target-assembly corrections, which account for the position of the diffracting layer within the layered target assembly (the “physics package”, § 5); (5) Le Bail fitting; (6) Rietveld refinement; (7) facility-specific import workflows. • denotes native support, ◦ partial or indirect support (for example, achievable through scripting or an external preprocessing step), and – no support. HEXRD’s Le Bail and Rietveld implementations are marked partial because they support a narrower range of structural and microstructural models than dedicated Rietveld packages such as GSAS-II or TOPAS; MAUD’s facility import is marked partial for its legacy LCLS import tool. Entries reflect the versions released as of October 2026. Platform-specific in-house programs, such as the LLNL IGOR Pro codes discussed in the text, are not listed.
HEXRD addresses these limitations through three key capabilities. First, its flexible geometry model (§ 2) supports arbitrary numbers of independently positioned and oriented detector panels with no assumptions about configuration, accommodating non-standard panel shapes, such as curved image plates, and unusual arrangements, such as arbitrarily oriented multi-panel arrays; the subpanels of a segmented detector can be modeled as independent panels so that their misalignment is refined, and detector distortion models are refined alongside the geometry. Second, corrections for pinhole collimation and layered target assemblies (§ 5.1) are integrated directly into the calibration process. Third, HEXRD provides native support for multiple X-ray sources with independent energies and beam vectors, enabling simultaneous calibration of detector position and orientation against diffraction from all beams, essential for instruments like the dual X-ray source (2XRS) configuration at NIF [61] where consistency between sources is critical. Dedicated import workflows for the NIF and OMEGA EP platforms, and a general one for other detectors (§ 4), then streamline preprocessing, after which the analysis is the same everywhere. Joint dual-source calibration and target-assembly corrections already exist in the platform-specific analyses noted above; HEXRD provides all three capabilities in one open, scriptable code that is not tied to any single platform. When a new platform comes online, the software can often be deployed with minimal or no modifications to the source code.
HEXRD (Highly Extensible X-Ray Diffraction toolkit) was originally developed at Lawrence Livermore National Laboratory for far-field high-energy diffraction microscopy (ff-HEDM) at synchrotron sources [9, 6, 7]. Its geometric model, written for multi-panel ff-HEDM detectors, also describes the flat and curved image plates and the multi-panel XFEL detectors used in dynamic compression experiments. The companion graphical application, HEXRDGUI, was subsequently developed in collaboration with Kitware, Inc. to provide interactive visualization, calibration, and analysis workflows accessible to experimentalists without programming expertise. The released software is in routine use by groups other than the developers; analyses led by users outside the development team include shock-compressed liquid tin at the European XFEL [26], shock-ramp-compressed gold at OMEGA EP [17], shock-compressed nanopolycrystalline diamond [29] and high-entropy alloys [43, 30] at the SPring-8 Angstrom Compact Free Electron Laser (SACLA), and the commissioning of the FIDDLE diagnostic at the NIF [68].
In this paper we describe the salient features and capabilities of both HEXRD and HEXRDGUI, with emphasis on capabilities most relevant to dynamic compression experiments. In particular, we focus on laser-driven platforms, where most dynamic compression XRD is currently performed. We note that while HEXRD retains its original ff-HEDM capabilities and continues to be widely used for that purpose, including by groups beyond the core developers [31, 27, 24] (Oh et al. 2025), the discussion here focuses on the dynamic compression workflows.
The remainder of the article follows the order of a typical analysis. The instrument geometry is defined (§ 2), together with the materials and their equations of state (§ 3); the images are imported § 4 and the target assembly is described, with the corrections it implies (§ 5); the data are inspected, masked and corrected (§ 6); the instrument is calibrated (§ 7); and the calibrated lineouts are refined by whole-pattern fitting (§ 8), whose lattice parameters the equation of state may convert to a pressure at a supplied temperature. These two sections also test HEXRD against independent codes and synthetic ground-truth data. Section 9 describes the software itself, and Section 10 applies it on six platforms. For a laser-driven shot the user supplies the detector images, a starting instrument configuration (for example, the defaults loaded by the import tools at the NIF and OMEGA EP), the energy of each X-ray source, the layer thicknesses of the target assembly, and, for calibration, a material of known lattice parameter in the same exposure, such as ambient diffraction from the pinhole, or, failing that, the rings alone for the structureless calibration of § 7.5. Every validation number reported here can be reproduced from the state files and scripts deposited on Zenodo ([3]).
Geometric model
The geometric model implemented in HEXRD is designed to be sufficiently general to accommodate any diffraction instrument composed of planar or cylindrical detector panels, with a common software interface (an abstract base class) from which other panel shapes can be derived. (Throughout this article, “geometry” refers to the position and orientation of the panels; the surface itself, planar or cylindrical, is the panel shape.) A collection of individual panels forms an instrument. An initial description was given by Bernier et al. [6]; here we summarize the essential features.
Coordinate systems
Five coordinate frames are used (Figure 1):

Figure 1. Diffraction geometry for (a) a planar and (b) a cylindrical detector panel, illustrating the laboratory (), beam (), detector (), sample (), and crystal () coordinate frames. Points P0–P4 define the geometric transfer function mapping reciprocal lattice vectors to detector coordinates; the reciprocal lattice vector and scattering angle are shown for the diffracted beam reaching P4 (magenta arrow: incident beam ; green: ; blue: ; red: ; yellow: diffracted ray). In (b), (violet) connects the diffracting volume at P3 to the cylinder axis (dashed), again points to the panel center P1, and the diffracted ray meets the cylinder at ; unwrapping the surface about gives its detector coordinates, the arc length (orange) and the axial coordinate (brown); the scalar and the cylinder polar angle are defined in § 2.4.
The laboratory frame , a stationary global reference with origin at the centroid of the beam cross-section, at an arbitrary position along the beam; with several X-ray sources, every beam passes through .
The beam frame, defined by the incident beam direction and a reference azimuth , providing natural representations for the scattering angle and azimuth , which is measured right-handedly about from the projection of onto the plane normal to the beam. The symbol for the ring azimuth is inherited from HEXRD’s HEDM origins and is used throughout the interface; the powder literature often denotes this angle .
The detector frame(s) , one per detector panel, with origin at the panel centroid and along the panel normal.
The sample frame , with origin P2 attached to the specimen, defining the sample coordinate system.
The crystal frame , with origin P3 defining the position and orientation of an
individual grain relative to the sample frame. This distinction is important for samples which are spatially extended such that the diffraction signal recorded does not necessarily originate from the same point.
In dynamic compression experiments, the points P0, P2, and P3 are typically all located at the same physical position (the sample/target location), but the framework retains the flexibility to separate them, as required for special circumstances in dynamic compression experiments such as diffraction from different layers offset in space, or significant motion of the sample induced by dynamic compression. The sample frame is placed in the laboratory frame by a translation and a rotation , and a crystal by a translation within the sample frame, so that
Detector parameterization
Each detector panel is parameterized by a translation vector and a rotation matrix , together defining the affine transformation between detector and laboratory coordinates:
The rotation is stored natively as the three components of an exponential map, a rotation vector whose direction is the rotation axis and whose length is the rotation angle; the interface can display and edit tilts in a choice of Euler conventions, but configuration files always record the exponential map. The pixel grid is defined by the number of rows and columns and the pixel pitch (the center-to-center pixel spacing) in each direction.
This per-panel parameterization allows HEXRD to represent instruments with any number of detectors at arbitrary positions and orientations. Detectors may also be grouped into physical units with known internal geometry (e.g. a Dexela 2923 detector composed of four subpanels), allowing subpanel misalignment within a detector to be taken into account. The grouping may be enforced as a rigid-body constraint during calibration. Cylindrical detectors, such as IMAGE-PLATE-3 on the TARDIS platform at NIF ([61]), are also supported. Each panel may also carry a distortion model, which maps measured pixel positions to their ideal ones and whose coefficients can be refined during calibration, such as the radial-polynomial GE_41RT model applied to the pincushion distortion of the GE detector in § 7.6. Detector distortion is distinct from the target-assembly corrections of § 5.1, which shift the apparent angles of diffraction from layers that are not at the nominal sample position.
Beam parameterization
The incident beam direction is specified by spherical polar angles in the laboratory frame, where the polar angle is measured from and the azimuth is measured from in the – plane:
so that the default recovers . These two angles appear in the interface and in instrument configuration files as the beam “polar angle” and “azimuth”; note that they parameterize the reversed direction , per the leading sign of equation (2). Every physical source is, of course, at a finite distance; when that distance is short enough to matter (e.g. pinhole geometries at the NIF), the source–sample distance becomes an additional parameter that enters into the pinhole distortion corrections described in § 5.1; it is taken from the facility metadata and is not among the refinable calibration parameters.
HEXRD also supports instruments with multiple X-ray sources, each with independent energy, beam vector, and source distance. Simulated diffraction patterns and calibration are associated with their respective sources, so that reflections from all beams are satisfied simultaneously during refinement (§ 7.4); this is the case, for example, for the dual-source (2XRS) configuration at the NIF [61].
Diffraction and the transfer function
Given a reciprocal lattice vector satisfying the Bragg condition, the unit diffracted beam vector is
The Bragg condition has both a directional and a magnitude part. The scattering vector must bisect the incident and diffracted beams,
and its length must match the interplanar spacing of the reflection,
for X-ray wavelength . Equation (4) constrains the orientation of relative to the beam, from which equation (3) gives the direction of the diffracted ray, and equation (5) fixes the Bragg angle at which the reflection is excited. The detector coordinates of the diffraction spot are then obtained by computing the ray–plane intersection of with each detector panel [6]:
Here denotes the in-plane coordinates of the intersection point measured relative to the panel origin P1 and expressed in the detector frame, which is the form in which pixel coordinates are stored. This transfer function is the core computational kernel of HEXRD. It maps reciprocal lattice vectors to detector coordinates for any instrument configuration, and its inverse provides the angular coordinates from measured detector positions. Both directions are implemented as compiled C and C++ kernels that operate on arrays of vectors; the cylindrical case is vectorized NumPy.
The same construction applies to cylindrical detectors, where the diffracted ray is intersected with a cylindrical surface rather than a plane. The cylinder can be unwrapped into a rectangle and this intersection point can be represented using planar Cartesian coordinates. A diffracted ray along , emanating from the diffracting volume at P3, meets a cylinder of radius at the point P3 + .
With reference to Figure 1(b), we define
is the cylinder axis, the tangential direction and the outward radial direction at the center of the panel; they are the detector-frame axes , and expressed in the laboratory frame. is the vector connecting the diffracting volume to the axis of the cylindrical detector. Note that the translation vector for the detector, , is defined to point to the center of the cylindrical panel, consistent with the planar detector. Since the diffracting volume lies inside the cylinder ( below), the two roots of the quadratic have opposite signs, and the physical intersection is the positive root:
with coefficients
Next we define a new vector, (not a unit vector), which points from the cylinder axis to the intersection point,
The intersection point is then described by the cylindrical coordinates , where the polar angle is measured from the center of the panel about and takes its sign from :
Unwrapping the cylinder about finally gives the Cartesian coordinates of the intersection point on the (flattened) detector, expressed in the detector frame,
so that cylindrical and planar panels are treated identically by the rest of the code. (For the planar case, intersections behind the diffracting volume are rejected in the projection itself, and rays parallel to the panel yield unbounded coordinates that the subsequent on-panel clipping discards.)
The symbol is used here for the cylinder polar angle to avoid confusion with the Bragg angle and the beam polar angle .
Because equation (11) recovers from an inverse cosine, it returns a principal value in to which the sign of is then applied, so and the unwrapped coordinate is single valued. A panel is therefore limited to a total subtended angle of less than , with the seam placed diametrically opposite the panel center at ; the image plates considered here subtend well under a full revolution. (The degenerate case , a ray parallel to the cylinder axis, cannot occur for a ray that intersects the panel.) As a numerical check of equations (8)–(12), forward projecting random diffraction directions onto a tilted cylindrical panel and inverting the resulting coordinates recovers the input angles to better than degrees in double precision. Checked against the independent cylindrical-detector model of pyFAI [1], with the sample on and off the cylinder axis and with the cylinder tilted, the two codes assign the same pixels the same to 0.01 mdeg and to 0.11 mdeg, and the forward projection places pyFAI’s angles on the panel to 0.01 m. The same comparison for planar panels, over 400 random geometries with panels 50–300 mm from the sample, tilted by up to and reaching , half of them with the beam along a non-default direction, agrees to better than mdeg in and mdeg in .
Crystallography and materials
HEXRD includes a crystallographic materials model for simulated pattern generation, calibration, and whole-pattern fitting.
Materials model
The materials model stores space group, lattice parameters, atomic site positions with fractional coordinates, site occupancies, and thermal displacement parameters (isotropic or factors, or anisotropic tensors). Space groups are selected by lattice type and identified by number, Hall symbol, or Hermann–Mauguin notation; lattice parameters are automatically constrained according to the lattice symmetry. Only the standard descriptions of the space groups are supported: those that admit two origin choices are represented in the first origin choice tabulated in International Tables (a CIF recorded in any other description must be transformed to the standard one before import), trigonal crystals are always represented in their hexagonal basis, and monoclinic groups in the unique-axis-, cell-choice-1 setting. The material model can also store the stiffness matrix, as well as other parameters related to the third-order Birch–Murnaghan thermal equation of state described later in the text.
Structure factors are computed using tabulated atomic scattering factors [71] with dispersion corrections [14, 15] and nuclear Thomson scattering (Berant et al., 1977). A reflections table (Figure 2) allows users to inspect d-spacings, values, structure factors, and intensities for each reflection, and to select which reflections are displayed as simulated patterns.

Figure 2. Reflections table for CeO, showing , d-spacing, (at a beam energy of 12.5 keV), structure factor , intensity without the multiplicity (computed with the powder Lorentz factor and the polarization factor for an unpolarized beam), powder intensity , and multiplicity for each reflection. The and columns are rescaled linearly so that the weakest reflection reads 0 and the strongest 100.
Materials can be imported from CIF files or entered manually through the GUI. A default library of materials commonly encountered in dynamic compression experiments (e.g. CeO, LaB, Pt, Ta, MgO, LiF, diamond) is bundled with the application.
Equation of state
For high-pressure experiments, HEXRD implements the thermal third-order Birch–Murnaghan equation of state (EOS) [8] of the JCPDS format, as used by Dioptas [46], so JCPDS parameter files are used directly. With and K, the thermal expansivity and the moduli at temperature are
and the pressure is the third-order Birch–Murnaghan isotherm about the room-temperature volume , evaluated with these moduli, plus the thermal pressure ,
with the Eulerian finite strain. The EOS is therefore parameterized by , , , their temperature derivatives, and the thermal expansion coefficients (an empirical form); parameters may be loaded from JCPDS files or entered manually. The EOS serves two roles. First, the interactive pressure–temperature slider: for the pressure and temperature it is set to, it solves equation (14) numerically for the volume and moves the simulated diffraction lines accordingly, so adjusting it until the lines match the data gives a quick estimate of the pressure. Second, the Python API works the other way round: the lattice parameters refined by whole-pattern fitting (§ 8) give the unit-cell volume , and equation (14) then gives the pressure directly.
Because the temperature enters equation (14) only through these parameters and the thermal pressure, it is an input rather than a fitted quantity: it must be supplied from an independent measurement or from hydrodynamic simulation of the drive. In dynamic compression this is often a major source of systematic uncertainty in a reported pressure, and it can exceed the uncertainty propagated from the refined lattice parameters themselves; the same machinery applied to static compression data, where the temperature is well known, is correspondingly more accurate. Reported pressures should therefore be quoted together with the assumed temperature and the EOS parameter set used. It also assumes hydrostatic strain: under uniaxial loading the rings are not circular, and fitting narrow azimuthal slices of the polar view (§ 6) instead resolves the strain along and across the load, as Hari et al. [29] did to measure the flow strength of diamond.
Data import and preprocessing
HEXRDGUI supports loading diffraction images in common formats (HDF5, TIFF, and HEXRD’s NumPy-based .npz frame caches) and additional formats via the FabIO library [35], with single- or multi-frame import for time-resolved experiments. Dark-frame subtraction, geometric transforms, and frame selection are configured during import.
Facility-specific import tools
Facility-specific import tools provide streamlined workflows for the TARDIS and FIDDLE (Flexible Imaging Diffraction Diagnostic for Laser Experiments) platforms at the NIF [68, 77] and the PXRDIP platform at OMEGA EP [60] (Figure 3). These tools load default instrument configurations, display interactive detector templates for boundary adjustment on image-plate scans, and handle facility-specific features such as instrument location and tilt calibration using the Advanced Tracking Laser Alignment System (ATLAS) on the NIF [12]. Data from other instruments, including synchrotron and XFEL detectors, go through a general import workflow that supports per-panel file assignment for multi-detector arrays (ePix10k, Rayonix, GE, Pilatus, Varex, etc.).
Figure 3. Importing TARDIS image-plate scans with the LLNL import tool. (a) The tool’s panel, with the instrument, configuration and current image plate selected. (b) The raw scan of that plate with the interactive detector boundary template (cyan outline), which is translated and rotated into alignment before cropping. (c) Raw view of the imported data with all three image plates cropped.
Physics package and target-assembly corrections
In dynamic compression experiments, the X-ray beam passes through multiple layers of the target assembly (the “physics package”) before and after diffracting from the sample. HEXRDGUI provides a physics-package editor (Figure 4) that allows users to define the target assembly geometry,

Figure 4. The physics-package editor showing an illustrative target assembly for a NIF experiment with all layers defined (ablator, heat shield, pusher, sample, reflective, window, and pinhole). The optional layers are enabled via the checked boxes (the sample layer is always present), and the assembly is illustrated by the corresponding diagram on the left.
including the following layers (in order of X-ray propagation): ablator, heat shield, pusher, sample, reflective coating, window, and pinhole. Each layer is specified by its material, density, and thickness. The pinhole layer additionally requires a diameter. Sample tilt can be specified using the globally configured Euler angle convention.
The physics-package definition serves multiple purposes within HEXRD: it provides the geometric parameters needed for pinhole distortion corrections (§ 5.1), enables correction for absorption of the scattered X-rays through the target stack, and is used for corrections that account for the offset between the different layers. The editor includes a visual diagram of the layer stack and automatically detects the instrument type (TARDIS, PXRDIP, or general) to apply appropriate defaults.
Layer and pinhole corrections
In many dynamic compression experiments at NIF and similar facilities, the diffracted X-rays leave the target assembly through a pinhole collimator, and the diffracting layer sits tens to hundreds of micrometers from the pinhole rather than at the point assumed by the instrument geometry. The apparent of its rings is then shifted, by up to about half a degree in the geometries tested below. HEXRD implements two correction models:
Layer distortion: the first-order correction of equation (8) of Rygg et al. [61], , where is the source-to-pinhole-center distance, the angle of the diffracted ray from the pinhole axis ( in the notation of Rygg et al.), and the offset of the layer’s midplane (e.g. sample, window, ablator) from the pinhole
center, measured along the pinhole axis and computed from the physics package as the layer’s standoff from the pinhole plus half the thicknesses of the layer and the pinhole. This is the point-pinhole approximation: the neglected second-order term, of order for a pinhole of diameter , is about for the standard TARDIS geometry [61]. For a driven sample, and the layer thickness at the time of the exposure can differ from their initial values and must then be supplied by the user, for example from the velocity-interferometer (VISAR) record [61].
Pinhole distortion: corrects for the geometric effect of the pinhole collimator, as described by Rygg et al. [61]. The implementation is a vectorized port of the reference implementation contributed to the HEXRD repository by the first author of Rygg et al. [61]; a test distributed with the validation scripts of § 7.6 drives the port and the reference on identical geometry and reproduces the reference to better than mdeg across 102 geometries with the X-ray source on the pinhole axis and 2520 with it – off the axis, at source and detector azimuths of both signs, for corrections as large as half a degree.
These corrections can be applied per overlay (allowing different corrections for different materials at different positions within the target assembly), and can optionally be applied directly to the polar view so that rings associated with the selected layer appear as straight vertical lines. The pinhole dimensions are taken from the physics-package definition above. Beyond the critical exit angle of the pinhole the corrections are undefined, and picks there are excluded from the calibration residual.
Visualization and data preparation
Once an instrument and its data are loaded, HEXRDGUI provides the interactive views, simulated overlays, and masking and correction tools used to inspect a dataset and prepare it for calibration and whole-pattern fitting.
View modes
HEXRDGUI provides four interactive view modes for inspecting diffraction data:
Raw view: displays detector images as acquired, with multi-detector data shown side-by-side or in tabs (Figure 3(c)).
Cartesian view: projects all detector panels onto a common virtual plane as seen from the sample, revealing gaps and overlaps between panels (Figure 5(a)); cylindrical panels are drawn flattened, so distances in this view are not physical for them and the view is intended for planar arrangements.

Figure 5. MEC data from four ePix10k quad detectors (LCLS experiment LV7318), with a simulated CeO powder overlay (red) shown in both views. (a) Cartesian view projecting all panels onto a common virtual plane, where the simulated Debye–Scherrer rings appear as concentric arcs. (b) Polar view ( vs. ) of the same data, in which those rings become vertical straight lines, with the azimuthally averaged lineout below (the axis spans only the range the panels cover; intensities are in arbitrary units here and in the later lineouts). The lineout shows the observed average (black) together with a one-dimensional azimuthal overlay (red) of the same phase, drawn as filled peak profiles on the lineout rather than as vertical lines, so that it simulates the relative peak intensities and profiles rather than only the line positions. The subscript on angular axes, here and in later figures, marks nominal angles, computed without the target-assembly corrections of § 5.1.
Polar view: resamples detector data onto a regular grid of (diffraction angle) vs. (azimuthal angle) with a user-set pixel size, where Debye–Scherrer rings appear as vertical
straight lines (Figure 5(b)). An azimuthally averaged lineout is shown below the image.
Stereographic view: projects all panels onto a stereographic circle, giving a single view of the full angular coverage of a multi-panel instrument (see the TARDIS example in § 10).
The polar view is the primary working mode for powder calibration and whole-pattern fitting. It supports configurable ranges in and for display and azimuthal averaging, selectable -axis units ( or momentum transfer, ), and optional layer or pinhole distortion corrections (§ 5.1) so that rings from the selected layer appear as straight vertical lines. Lineouts and whole-pattern fits are computed from this resampled grid, so its pixel size also sets how many polar pixels contribute to each lineout point (§ 8).
Overlays
Simulated diffraction patterns are displayed as overlays on the data in all view modes. Three overlay types used in dynamic compression are managed through the Overlay Manager, a dockable list with per-overlay controls, and drawn on the main canvas (the manager also offers rotation-series overlays for high-energy diffraction microscopy):
Powder overlays: Debye–Scherrer rings from polycrystalline phases, with selectable reflections, tolerance widths, and optional layer distortion corrections such as pinhole.
Laue overlays: single-crystal spots computed from the crystal orientation, position, and strain (stretch matrix), together with the lattice parameters, for a given energy range and minimum d-spacing. Spot ranges may be displayed as ellipses or rectangles, and spots can be optionally labeled by the diffracting plane or energy.
Constant- overlays: rings of constant angle (the angle between the diffraction plane normal and the sample normal; unrelated to the statistic), useful for identifying preferred orientation, fiber textures in particular.
Each of these overlay types supports interactive parameter editing and can be associated with different materials and X-ray sources. Overlay positions update in real time as parameters are adjusted. Powder overlays may use layer and pinhole distortion corrections from the physics-package definition, and Laue overlay parameters (crystal orientation, position, and stretch matrix) can be loaded from a file or adjusted interactively using sliders.
Additionally, one-dimensional azimuthally averaged overlays can be displayed on the azimuthal lineout (Figure 5(b)), with configurable FWHM and scale factor, for comparing simulated and observed relative peak intensities and profiles. These are managed separately through the polar view settings rather than the Overlay Manager.
Masking and intensity corrections
Masking
Regions of detector images may be masked using rectangular, elliptical, or hand-drawn polygonal selections in the raw or polar views, or by threshold-based intensity filtering. Masks can also be generated from powder or Laue overlays to exclude specific diffraction features, and from the pinhole geometry to mask regions that fall outside the pinhole aperture. All masks are managed through a mask manager that supports export/import in HDF5 format.
Intensity corrections
Intensity corrections available in HEXRD include:
Solid-angle corrections: accounting for the varying solid angle subtended by each pixel as a function of detector position and orientation.
Lorentz correction: dividing out the Lorentz factor , the geometric weighting of powder intensities measured on a stationary sample with an area detector.
Polarization correction: dividing out the polarization factor of the scattering, either for an unpolarized source such as a laser-plasma backlighter, or its -dependent form for a linearly polarized synchrotron or XFEL beam with configurable horizontal and vertical fractions.
Dark-frame subtraction: applied per-detector during import.
Absorption corrections: accounting for X-ray attenuation in the sample and window layers of the target assembly as defined in the physics package (§ 5), as well as through the detector-side filter and coating (each specified by material, density, and thickness) and the angle-dependent response of the image-plate phosphor (Rygg et al., [61]), as corrected in its erratum (Rygg et al., [62]).
Median filtering: for noise reduction in individual frames or frame sequences.
SNIP background subtraction: the Statistics-sensitive Non-linear Iterative Peak-clipping algorithm (Ryan et al., [59]) (two-dimensional form after Morháč et al. [50]) applied to the rebinned polar image for background estimation in the polar view. Three variants are available: fast SNIP 1D and SNIP 1D, which operate on each row of constant independently, and SNIP 2D, which operates on the full two-dimensional image.
Calibration
Accurate calibration of detector geometry is essential for quantitative diffraction analysis, particularly in dynamic compression experiments where small changes in lattice parameters correspond to large pressure variations. HEXRD provides several calibration methods suited to different experimental modalities. All methods minimize a residual between observed and simulated diffraction features using bounded nonlinear least-squares optimization via the lmfit library [52], with the Trust Region Reflective algorithm [10] as the default solver.
Refinable parameters include detector tilts (three components using the globally configured Euler angle convention) and translations (three components) for each panel, beam energy, beam vector orientation, material lattice parameters, the coefficients of a detector distortion model, and, for Laue overlays, the crystal orientation, position, and stretch. Each parameter can be individually toggled for refinement and bounded.
Parameter uncertainties, reported in the whole-pattern fitting dialog and available from the calibration API, are the standard errors returned by lmfit, obtained from the covariance matrix estimated from the Jacobian at the solution and scaled by the reduced of the fit. These are local, linearized estimates and should be interpreted with care in two common situations: when a parameter has converged onto one of its bounds, where the quadratic approximation underlying the covariance estimate no longer holds, and when parameters are strongly correlated, as detector distance, beam energy and lattice parameters inevitably are in a powder calibration. HEXRDGUI highlights parameters that have converged onto one of their bounds in the parameter tree (in red; an example appears in § 7.4) so that these cases are visible to the user.
Relative and engineering constraints
Two types of constraints help regularize calibration for instruments with known internal geometry:
Rigid-body constraints enforce that a group of detectors (either within a single physical unit or across the entire instrument) moves as a rigid body during refinement, preserving their known relative positions and orientations. For example, on the FIDDLE instrument at NIF [68], the relative positions and orientations of the individual Icarus cameras (gated hybrid-CMOS detectors) are measured by ATLAS, but the overall position and orientation of the assembly must be determined through calibration (§ 10.6).
Engineering constraints impose instrument-specific physical bounds (e.g. the distance between the top and bottom TARDIS image plates is constrained to a physically reasonable range given the known constraints of the NIF target positioner).
Both constraint types are set in the calibration dialog; in the TARDIS example of § 7.4, relative constraints are not used, but the engineering constraint on the plate separation is active.
Powder calibration
The powder calibration method operates against a known powder standard (Figure 6). Calibration points may be picked manually or auto-picked; auto-picking requires that simulated Debye–Scherrer lines are already close to the observed rings, which can be arranged by first adjusting the instrument parameters by hand with the calibration sliders.

Figure 6. “Fast powder” calibration of a DCS pink-beam dataset using a standard. (a) Calibration dialog with refinable parameters and constraints; with the “Use delta for boundaries” option checked, the Delta column bounds each parameter to about its current value. (b) Polar view showing auto-picked peak positions (magenta plus signs) overlaid on the powder rings, with simulated ring positions (vertical blue lines), and the azimuthal lineout below. In the configuration shown only the three detector translations are refined.
For auto-picking, the image data are projected into polar ( vs. ) coordinates, where ring positions are identified automatically by fitting user-specified peak profiles along radial slices at regular intervals; Gaussian, pseudo-Voigt, and pink-beam [70] peak profiles are available. User-configurable parameters include the annular search width in and , the minimum peak amplitude, and a maximum fit-error threshold for rejecting spurious picks. The objective function is the unweighted sum of squared differences between picked and predicted ring positions, measured in . A “Fast powder” calibration mode provides a streamlined workflow that operates on a single powder overlay and uses auto-picking exclusively, bypassing the manual point-picking step entirely.
Laue calibration
For experiments that produce single-crystal Laue diffraction, for example from a diamond or LiF window in the target assembly, the Laue calibration method can refine detector geometry and all crystal parameters (position, orientation, and stretch matrix) simultaneously. Laue spots may be picked manually or auto-picked; as with powder calibration, auto-picking requires that the simulated spots are already close to the observed positions, and both the crystal and the instrument sliders can be used to roughly align the simulation first. The refinement minimizes the discrepancy between picked and predicted spot positions on the detector.
Composite calibration
Composite calibration refines the instrument against any number of powder and Laue overlays at once, each tied to one material, one X-ray source and one layer of the target assembly. One mechanism thus covers both situations common in dynamic compression: diffraction from several layers of the physics package (for example a powder sample and a single-crystal window), each with its own correction (§ 5.1), and diffraction from several X-ray sources, for which the same material is simply given one overlay per source. The residuals of all overlays are joined into one unweighted least-squares objective: powder picks contribute their differences (in radians) and Laue picks their detector-plane position differences (in millimeters), with no relative weighting between the two; none of the calibrations in this paper mixes the two kinds of pick. As in the individual methods, points may be auto-picked or picked by hand; hand picking uses a two-click mode in which the user left-clicks on the main canvas to freeze the zoom dialog at that location, then left-clicks on the zoom dialog to place the point precisely (Figure 7). For instruments with multiple X-ray sources, the composite calibrator handles picks from each source independently, automatically switching the polar-view projection for picks associated with different beams (Figure 8).

Figure 7. Hand-picking points for composite calibration of TARDIS data. (a) Polar view across multiple image plates, with picked points shown as colored markers and simulated rings as blue dash-dot lines; the bold magenta line marks the ring currently being picked and the magenta rectangle the region shown magnified. The azimuthal lineout appears below; the axis label indicates that the layer correction of § 5.1 is applied, mapping angles to the sample layer at its 150 m standoff. (b) Zoom dialog, which provides the magnified view, corresponding to the magenta rectangle in (a), used to place each point precisely in two-click mode.

Figure 8. Composite calibration of a TARDIS dual-source (2XRS) dataset. The same material is assigned to two separate overlays, one per X-ray source, so that picks from both sources contribute to the overall residual and the instrument is calibrated against both beams simultaneously. (a) Calibration dialog showing refinable parameters for both X-ray sources (XRS1, XRS2) and the TARDIS engineering constraints. The “distance between plates” parameter constrains the Y-axis separation of IMAGE-PLATE-4 and IMAGE-PLATE-2 to a physically reasonable range; a parameter shown in red has reached its bound, signaling that the bound is active and should be reviewed. The “X-Ray Source” selector switches the polar view between sources, and the “Distortion” section applies a sample-layer correction. (b) Polar view with calibration picks (colored markers) and simulated rings (blue dash-dot lines), above the azimuthal lineout.
Simultaneous refinement against all sources is essential in practice: calibrating against a single source leaves the geometry underconstrained. Rotating a panel about the incident beam through the sample leaves every Debye–Scherrer cone of that beam unchanged, so one source cannot fix that rotation, while a second beam from another direction can. A single-source fit that appears excellent will therefore generally degrade the agreement for the other source. Only a joint refinement, in which picks from every beam contribute to a common residual, yields a geometry consistent with all sources at once.
Synthetic test. A ground-truth test, distributed with the validation scripts of § 7.6 [3], quantifies this. Powder picks were simulated for a two-source, two-panel instrument of known geometry; each panel was then displaced by mm and its normal tilted by about , and the composite calibrator was run on the picks alone, refining six degrees of freedom per panel with the beam parameters and lattice parameter held fixed. With noise-free picks, refining against both sources recovers the true geometry to numerical precision (below nm and mdeg); with Gaussian noise of half a pixel on every pick, it recovers it to better than m in translation and mdeg in panel-normal orientation. Refining against the first source alone reproduces that source’s noise-free picks exactly, yet settles and mm away from the true panel positions and mispredicts the second source’s rings by mdeg in : the fit drifts without penalty along the rotation that a single source cannot fix (the same freedom that forces one tilt to be fixed in § 7.6), and how far it drifts depends on where it starts. Alternating single-source refinements, the only option in HEXRD before native multi-source support, merely shift the error from one source to the other: recalibrating to the second source zeroes its own residual but displaces the first source’s rings by 97 mdeg, returning to the first displaces the second’s by 54 mdeg, and after three passes the geometry is still 0.25–0.38 mm from the truth.
NIF dual-source shot. The real dataset is NIF shot N160517-003 of Rygg et al. [61] (Figure 7 and Figure 8): a 6 µm Pt sample ramp-compressed to 180 GPa ( Å) between a diamond pusher and a 150 µm LiF window, exposed by two co-timed Ge He-α X-ray sources (10.25 keV) and recorded on three image plates through a Ta pinhole 75 µm thick and 350 µm in diameter. A simultaneous refinement of both sources converges on this dataset (116 hand-picked points across three image plates and both sources, with the sample-layer correction of § 5.1 applied for the sample’s 150 µm standoff from the pinhole, set by the LiF window between them) with an RMS residual of 0.18° in for each source, comparable with the typical uncertainty of 0.2° that Rygg et al. [61] report for such data. This is about one 100 µm pixel at the ≈27 mm plate distances, the precision with which positions can be picked by hand on the broad, noisy diffraction lines of these exposures (Figure 8). The refinement varies twelve parameters: the three translations and three tilts of IMAGE-PLATE-2, the same for IMAGE-PLATE-4 except its vertical position, and the plate separation, which sets that position through the TARDIS engineering constraint; IMAGE-PLATE-3 and the beam energies and directions are fixed. The plate separation converges onto the upper bound of its engineering range (23.43 mm, shown in red in Figure 8(a)); releasing the bound moves it by only 6 µm, to 23.44 ± 0.07 mm, and leaves the residuals unchanged. Refined against either source alone, even starting from this optimum, the same picks fit that source more closely (155 and 165 mdeg RMS, against 180 and 183 mdeg jointly) but mispredict the other source’s rings by 1.6 and 1.4° RMS, with the plates moved by up to 0.7 mm and tilted by 2.1–2.6°. One source leaves each plate’s rotation about its beam undetermined, so, as in the synthetic test, how far such a fit drifts depends on where it starts. Started instead from the nominal geometry loaded by the import tool, whose plate tilts differ from the optimum by about 0.5°, the joint refinement returns to the same optimum. A fit to either source alone must hold the tilts at assumed values; with them at their nominal values, it fits its own rings to 195 and 197 mdeg but mispredicts the other source’s by 0.63 and 0.58°, about three pixels. A script distributed with the validation scripts reruns this calibration from the geometry and picks stored in the distributed state file and confirms that the interactively obtained optimum is stationary.
The Pt lattice parameter is held throughout at 3.542 Å, the value Rygg et al. [61] obtained from their own analysis of this shot, so the calibration reproduces that analysis; where an absolute scale is needed, features of known angle, such as ambient diffraction from the pinhole substrate, provide it. The assumed standoff of the layer correction (§ 5.1) enters just as directly: refitting the lattice parameter alone at the converged geometry shows that each micrometer of change in moves it by 39 ppm, so must be known to within a few micrometers for the lattice parameter to be accurate to about . The calibration uses the as-built standoff. By the time of the exposure the compression wave has reached the LiF window and changed its thickness by a few micrometers, which is not accounted for here, since computing it needs hydrodynamic simulations of the drive. The error this introduces is far smaller than the effects of the spread of densities from the pressure distribution in the sample, the instrumental broadening and the high background of these exposures.
Structureless calibration
When the crystal structure of the sample is unknown at the high-pressure and temperature conditions achieved during dynamic compression, and there are no other diffraction features from a crystal with known structure and density, a structureless calibration can be performed using only the positions of observed Debye–Scherrer lines without reference to a structural model. The user
picks points along each ring on every panel with the two-click zoom tool of Figure 7. The objective is simply that all points on one ring map to the same , whichever panel or azimuth they came from: in the polar view, each ring must become a straight vertical line.
In addition to the instrument geometry, the positions of the ideal rings themselves are included as refinable parameters (DS_ring parameters), allowing the calibration to simultaneously determine both the detector geometry and the ring positions. Refining both simultaneously is degenerate in at least one direction (an overall detector-distance/-scale trade-off), so in practice the degeneracy is broken by keeping a subset of the geometric parameters fixed or tightly bounded (e.g. through the engineering constraints described above), with the ring positions started near the picked positions (by default at their mean ) or at values computed from the expected density of the compressed sample. When a ring of known angle is present, such as ambient diffraction from the pinhole, the powder or composite calibration of § 7.2 or § 7.4 is used instead. Rygg et al. [61] use the same equal- constraint when no reference lines are available.
This method is particularly valuable for dynamic compression experiments where the compressed sample may be in an unknown or novel phase, and only the requirement that rings be self-consistent is imposed. It implicitly assumes that all selected Debye–Scherrer lines originate from the same layer in the physics package.
Validation against independent implementations
To test the geometric model and the calibration against independent implementations, one calibration dataset was analyzed three times over: with HEXRD, with pyFAI 2026.9.0 ([1]), and with Dioptas 0.11.0 ([56]). The dataset is NIST standard reference material SRM 674b (CeO, certified Å, expanded uncertainty at ), measured at beamline 1-ID-E of the Advanced Photon Source with the monochromator set to the Au -absorption edge (80.725 keV), on a single planar GE panel (2048 × 2048 pixels of 200 µm) about 1051 mm from the sample. A single planar panel is a configuration that all three codes, including Dioptas, describe, so it allows a direct three-way comparison of the per-panel transfer function from which every HEXRD configuration is built. (The multi-source and target-assembly machinery, which those codes do not provide natively, is checked by the synthetic test of § 7.4 and the reference comparison of § 5.1; the planar transfer function at large tilts and other beam directions, and the cylindrical one, are checked against pyFAI in § 2.4.) The panel also has a pincushion distortion, described by HEXRD’s GE-41RT model (with its inverse evaluated by Newton iteration, the GE-41RT_newton variant, because the closed-form inverse handles distortion of only one sign and is exact only when all three radial exponents are 2), so the effect of modeling a detector distortion can be measured. That model maps the radius of each measured point about the panel center, at fixed azimuth , to the corrected radius with mm; here the exponents are and the three coefficients are the refinable parameters.
For the main comparison, a neutral extraction, independent of all three codes (a routine in the validation scripts that samples the image along radial lines across each ring, subtracts a linear background and locates the peak by a least-squares parabola within one pixel of its maximum), fitted 26,318 sub-pixel positions on 24 CeO rings directly from the image (its search windows are centered on the rings predicted by the stored calibration, but the position fitted within each window depends only on the image); these agree with HEXRD’s own pseudo-Voigt ring fits to a median of 0.027 pixel, with no systematic offset (median signed difference 0.002 pixel). Each package then refined the sample–detector distance, the beam center, and two detector tilts against those shared positions, with the same least-squares minimizer. Everything else was fixed: the beam energy at the monochromator setting, the lattice parameter at the certified value, and the third tilt, which axially symmetric rings cannot constrain. Dioptas was additionally run end to end with its own automatic ring search and refinement (2426 points), as a fully independent analysis. The two leftmost columns of Table 2 treat the distortion equivalently: HEXRD applies its model internally, and for the pyFAI “pre-corrected” control the same fixed polynomial was applied to the positions beforehand, so pyFAI refined its geometry without needing to know about the distortion. The distortion coefficients themselves were refined on this same image by HEXRD’s powder-calibration workflow, from its own auto-picked ring fits and with the beam energy and lattice parameter at the fixed values above, and were then held fixed everywhere; the deposited state file stores them.
| HEXRD | pyFAI | pyFAI | Dioptas | |
| ring positions | shared | shared | shared | own search |
| panel distortion | modeled | pre-corrected | not applied | not applied |
| distance (mm) | 1050.7356 ± 0.0004 | 1050.7356 | 1051.1617 | 1051.2654 |
| beam center, column (pixel) | 1030.770 | 1030.770 | 1030.773 | 1030.770 |
| beam center, row (pixel) | 1007.334 | 1007.334 | 1007.326 | 1007.309 |
| detector tilt (°) | 0.18955 ± 0.00034 | 0.18955 | 0.18819 | 0.18856 |
| RMS residual (°) | 0.00055 | 0.00055 | 0.00144 | 0.00170 |
Table 2. Detector geometry for the GE CeO dataset refined independently by three packages. HEXRD and pyFAI were refined against the same 26,318 independently extracted ring positions; Dioptas used its own automatic ring search. The last two columns have no distortion correction. Beam-center coordinates are fractional (column, row) pixel indices with the center of the first pixel at index 0 (0.5 was subtracted from pyFAI’s native values, which place that center at 0.5); the distance is measured along the panel normal, and the detector tilt is the angle between the panel normal and the beam. The beam energy and CeO lattice parameter were fixed at their physical values in every refinement; the quoted standard errors are conditional on those fixed values (see text), and are given for the HEXRD column.
Table 2 makes three separate comparisons:
Same positions, distortion on both sides (columns 1–2): HEXRD and pyFAI converge to the same geometry. Evaluated at all 26,318 positions, the two geometric models predict values agreeing to an RMS of mdeg, the level of numerical round-off, a direct check
of the transfer function of equation (6) against an independent implementation.
Withholding the distortion (column 3) moves the refined distance by 0.43 mm, the cost of having no distortion model, with everything else held constant.
An independent ring search (column 4): Dioptas shares pyFAI’s geometry engine, so the further 0.10 mm difference from column 3 measures how sensitive the poorly conditioned distance is to which peaks are picked, and to Dioptas’s own refinement. (Its search indexed 30 rings; the shared set has 24 because the six weakest, (444), (800), (733), (662), (840) and (842), fall below the amplitude threshold of the neutral extraction.)
In addition, the calibration stored in the state file, which a different user performed through the HEXRDGUI interface using HEXRD’s own auto-picked ring fits instead of the shared positions, agrees with the refined geometry to 1.7 µm in distance, 0.02 pixel in beam center and in tilt; that state file is the one deposited on Zenodo [3].
The distortion of this panel is a pincushion of about 0.5 pixel (100 µm) at a radius of 200 mm, rising to 1.1 pixels (220 µm) in the corners where the high-angle rings fall. pyFAI and Dioptas apply a detector distortion as a fixed, separately measured correction, such as a spline file, and none was used here. Without a distortion model, their refinement absorbs what it can into the geometry (the azimuthally symmetric part into the distance, the 0.43 mm above) and leaves the rest as the ring-by-ring trend of Figure 9. Supplying the same codes with pre-corrected positions removes the trend entirely and reproduces the HEXRD result, so the discrepancy comes from the missing distortion model and not from the underlying geometry. On this detector, the three-coefficient distortion model removes a ring-to-ring systematic error of up to 4.7 mdeg, eight times the 0.55 mdeg scatter of individual positions and far larger than the uncertainty of each ring mean. Its coefficients were determined on the same image, so this measures how much of the residual the model accounts for; the comparison of the geometric models themselves (columns 1 and 2 of Table 2) is independent of it.

Figure 9. Mean residual of each ring of the GE dataset with respect to the certified lattice parameter. With the panel distortion modeled, HEXRD is flat to within 0.23 mdeg across the whole range; supplying the same pre-corrected positions to pyFAI reproduces that curve exactly. Without a distortion model the residual acquires a systematic trend of −1.5 to +4.7 mdeg that no adjustment of the geometry can remove (one curve is plotted for pyFAI and Dioptas, which share the same geometry engine; it is the pyFAI refinement of the shared positions).
Two further points set the scope of the comparison. First, the lattice parameter, the beam energy and the distance are mutually degenerate for a monochromatic powder measurement (refining and the distance together gives a correlation coefficient of +1.00), so two of the three must be fixed externally. The comparison therefore tests the consistency of the geometric models, with the absolute scale set by the fixed values, and the standard errors in Table 2 are conditional on the fixed values (and somewhat optimistic, since neighboring positions along a ring are not independent); this is the same degeneracy discussed in § 7.5. Second, the residual floor of is set by the scatter of the extracted peak positions; the two geometric models agree to numerical round-off, far below it. The extraction and comparison scripts, the converters between HEXRD instrument files and pyFAI poni files, and the Dioptas calibration with its picked positions are deposited on Zenodo [3] so the comparison can be repeated.
Whole-pattern peak fitting
After instrument calibration, provides whole-pattern peak fitting (WPPF) for material characterization from the azimuthally averaged pattern (Figure 10). Two methods are implemented: Le Bail fitting (Le Bail et al. [42]), which extracts peak intensities without a crystal-structure model, and Rietveld refinement (Rietveld [57]), which uses a full structural model for phase quantification and structure refinement. The Rietveld intensities include the Lorentz–polarization factor, so the Lorentz and polarization corrections of § 6.3 should be left off a pattern that is to be refined this way.

Figure 10. Whole-pattern peak fitting of the GE CeO₂ dataset. (a) WPPF options dialog, with tabs for general settings, background, plots, amorphous contributions, texture and thermal diffuse scattering. The parameter tree lists the refinable parameters, whose number varies with the features enabled, with columns for value, refinement toggle, bounds and uncertainty (standard error). (b) Polar view with simulated line positions (vertical blue lines) and a message log reporting , and for each iteration of a Rietveld fit of this lineout (the closure test and the GSAS-II comparison in the text are Le Bail fits of the same data). The azimuthal lineout below it shows the measured intensities (black) overlaid with the fitted pattern (red), with the difference plot underneath.
Five pseudo-Voigt peak-shape models are available: the Thompson–Cox–Hastings (TCH) profile (Thompson et al. [66]), the Finger–Cox–Jephcoat (FCJ) profile with axial-divergence correction (Finger et al. [23]), a pink-beam pseudo-Voigt peak shape (Von Dreele et al. [70]), and two one-sided profiles in which the pseudo-Voigt is convolved with a single exponential or with a half Gaussian; the latter approximates the spread of lattice parameters in a sample with a temperature gradient, as when it is heated by the X-ray source. In all models the Gaussian width follows the Caglioti parameterization (Caglioti et al. [13]) (plus a Gaussian size term) and the Lorentzian width carries Scherrer size and microstrain terms. Anisotropic peak broadening may be modeled using the phenomenological approach of Stephens [65].
Refinement minimizes the weighted sum of squared residuals over the points of the azimuthally averaged lineout. The basic weighting scheme uses . Points with zero intensity are assigned and excluded from the number of observations, . A lineout containing negative intensities, for example after background subtraction, is first shifted by a constant offset to make its intensities non-negative. This shift is a numerical convention: it does not restore Poisson counting statistics and changes the intensity-based weights without changing the underlying noise variance.
The weighted profile -factor, expected -factor, reduced chi-squared, and goodness-of-fit are defined as
where is the number of independently refined parameters and the sums run over the included observations.
In high-energy-density experiments, diffraction lineouts frequently contain a large background from the target assembly and scattered drive radiation. This background increases the denominator of and can therefore obscure discrepancies between the calculated and observed Bragg peaks. To assess profile agreement relative to the intensity above background, also reports the background-adjusted statistic
where is the refined or estimated background and the weights are the same as in equation (15). The squared denominator terms accommodate locally negative background-subtracted intensities; is non-negative and finite provided that the denominator is nonzero. Its numerator retains the residuals over the full fitted profile. For patterns dominated by a positive background, is generally larger than and provides a more informative measure of the residual relative to the scattering above background. Available background representations include cubic spline interpolation through interactively selected control points, Chebyshev polynomials with refinable coefficients, and the SNIP algorithm ([59]).
The interpretation of as a statistical goodness-of-fit measure with an expected value near unity requires appropriate inverse-variance weights, approximately independent observations, and an adequate model. The basic weights generally do not meet these requirements for area-detector lineouts. Each lineout point averages contributions from multiple detector pixels, and the recorded intensities may be expressed in analog-to-digital units rather than photon counts. Interpolation can also introduce correlations between samples. Consequently, the variance of a lineout point is not generally equal to its intensity, and the basic weighting scheme can yield without implying an exceptionally accurate fit. For fixed preprocessing and weighting, and the background-adjusted defined above remain useful descriptive measures of profile agreement.
An optional correction, enabled by default in HEXRDGUI, multiplies the basic weight at each lineout point by the number of contributing pixels, , giving . For an uncorrected arithmetic average of independent Poisson counts, the estimated variance is , so this correction gives the appropriate inverse-variance weight. This correction remains approximate for integrating detectors and processed images. Detector gain, intensity corrections, and correlations must also be considered when estimating uncertainties. In particular, HEXRD takes as the number of valid polar-view pixels, and polar pixels finer than the detector pixels resample the same detector pixels, so can exceed the number of independent measurements; and therefore depend on the azimuthal bin size and are comparable only between fits made with the same binning. Scaling the weights by alone does not guarantee a statistically calibrated : for the Rietveld fit of the high-count data of Figure 10, with azimuthal bins, , because the residual is dominated by model error rather than counting noise. In practice we therefore recommend comparing fits by and computed with the same preprocessing and binning, and reading as a diagnostic of model error rather than as a calibrated statistic.
Amorphous scattering from the sample or target assembly can be modeled using split-Gaussian, split-pseudo-Voigt, or experimentally measured profiles. During refinement, HEXRD reports the degree of crystallinity (DOC), defined here as the integrated intensity assigned to the crystalline peaks divided by the sum of the integrated crystalline and amorphous intensities over the analyzed angular range. This quantity measures the relative contributions of the fitted scattering components. Its interpretation as a crystalline phase fraction requires appropriate normalization of their relative scattering intensities and accounting for other sources of diffuse scattering such as thermal diffuse scattering (TDS) and incoherent scattering, even when the crystalline and amorphous components have identical chemical compositions.
For Rietveld refinement, preferred orientation can be modeled using either the March–Dollase model [22] or a spherical-harmonic expansion of the general axis distribution function [69]. Thermal diffuse scattering (TDS) can be included using the analytical Warren model [75] for monatomic FCC and BCC structures or an experimentally measured diffuse-scattering profile. Models for more complex structures, such as that of Sakuma (1995), are being implemented. Accounting for TDS is particularly important at the high temperatures reached during dynamic compression experiments [73].
Two tests validate the fitting module. The first is a closure test against the calibration of § 7.6: a Le Bail refinement of the CeO lattice parameter against the azimuthally averaged lineout of the calibrated GE instrument, with the beam energy fixed and the lattice parameter started 1200 ppm away from its true value, recovers the certified value, at which the calibration held it, to 0.3 ppm at . The calibration, the polar mapping, and the peak-shape model are therefore mutually consistent.
The second test is a comparison with GSAS-II [67], version 5.8.4, in which both codes fitted the same two patterns by the Le Bail method. The first pattern is synthetic and noise-free, calculated with the symmetric Thompson–Cox–Hastings profile [66] that HEXRD fits, with a known lattice parameter, which HEXRD recovers to within 0.1 ppm. The profile of GSAS-II includes axial-divergence asymmetry [23], with a minimum of SH/L = 0.002, which shifts the low-angle peaks of this pattern by the equivalent of a 23 ppm change in wavelength. With the wavelength adjusted by this amount, the two lattice parameters differ by 0.6 ppm. The Caglioti width coefficients are defined differently in the two codes, so the peak widths were compared directly. The full widths at half maximum of 12 isolated reflections in the fitted patterns are within 0.02% (HEXRD) and 0.5% (GSAS-II) of those in the synthetic pattern.
The second pattern is the measured GE lineout used in the closure test. Here the lattice parameters from the two codes differ by 62 ppm. The minimum asymmetry, calculated for the peak widths of this pattern, accounts for 24 ppm of the difference. The remaining 38 ppm reflects other differences between the two codes’ profile models, so a lattice parameter from this pattern depends on the profile model at the level of a few tens of ppm. Both patterns and the fitting scripts are included with the validation scripts. Neither pattern tests the calculated intensities, on which Rietveld refinement depends. For these, the two codes were given the same CeO structure, and the relative intensities of all 18 reflections with Å agree to 0.4% at both 12.5 and 80.7 keV. The difference increases smoothly with scattering angle, consistent with the different atomic scattering-factor parameterizations used by the two codes. Preferred orientation and thermal diffuse scattering were not tested.
Program overview
This article describes HEXRD version 0.10.7 [2] and HEXRDGUI version 1.2.6 [4]; the archived releases corresponding to these versions are deposited on Zenodo at the DOIs given in the reference list, and each is also available as a tagged release on GitHub.
HEXRD is implemented as a Python library (requiring Python 3.11 or later; versions through 3.14 are supported) with performance-critical routines optimized using Numba [39] just-in-time compilation and core coordinate transform operations implemented in C/C++. HEXRDGUI provides an interactive graphical interface built on the Qt6 framework (PySide6) that uses the HEXRD library as its computational backend. This separation of library and application enables both programmatic analysis via the Python API for scripting and batch processing, and point-andclick operation via the GUI for interactive exploration.
Figure 11 shows how the two packages fit together. All of the scientific functionality (the materials and equation-of-state models of § 3, the geometric model, distortion and intensity corrections, and calibration engines of § 2, § 6.3 and § 7, and the whole-pattern fitting module of § 8) lives in the HEXRD library and is reachable both programmatically (Python scripts, Jupyter notebooks, and a command line interface) and through HEXRDGUI, which contributes the interactive views, overlay and mask management, point picking, workflow orchestration, and session state while calling into the same library routines that a script would. Analyses set up interactively can therefore be reproduced programmatically: the instrument configuration, materials, masks and fit parameters are all saved to files that a script can read, so nothing depends on the interface.
Figure 11. Architecture of the HEXRD toolkit. Raw data, instrument configurations, and material definitions (left) are consumed either programmatically or through the HEXRDGUI desktop application, both of which drive the same HEXRD scientific library. Within the library, the materials, instrument, and whole-pattern fitting components build on shared supporting libraries (NumPy, SciPy, lmfit, FabIO) with Numba and C/C++ kernels for performance-critical routines. Outputs (right) comprise the calibrated instrument geometry and reduced data, together with the derived thermodynamic state and microstructure: phase identification and transitions, density, microstrain and grain size, and, through the refined texture, Debye–Waller factors and thermal diffuse scattering, indicators of deformation mechanisms and temperature (the equation of state, in contrast, takes the temperature as an input). Following the focus of this article, the diagram shows the capabilities relevant to dynamic compression experiments; the ff-HEDM workflows that HEXRD also provides (see Introduction) are omitted.
Software architecture and dependencies
The HEXRD library is organized into subpackages by technique: hexrd.core holds the shared foundations described next, the powder workflows and whole-pattern fitting live in hexrd.powder, Laue simulation and fitting in hexrd.laue, and the pinhole and physics-package corrections in hexrd.hed, while the ff-HEDM workflows occupy their own hexrd.hedm subpackage. Within hexrd.core, the transforms module implements the core geometric calculations, including the transfer functions described in § 2 and coordinate mappings between laboratory, detector, and angular representations. Vectorized operations over arrays of diffraction vectors, including the transfer functions and their inverse, are compiled C and C++ kernels; as a representative figure, mapping every pixel of a panel to diffraction angles takes about 0.2 s on one core of an Intel Core Ultra 9 285K desktop processor. Numba (Lam et al., [39]) accelerates other vectorized routines (rotations, distortion models, fitting). The material module provides crystallographic definitions, structure factor calculations, and equation-of-state models. The fitting module implements the calibration algorithms and their parameter handling; whole-pattern fitting lives in hexrd.powder. The instrument module defines detector models through an abstract base class, allowing arbitrary panel shapes to be implemented by subclassing; planar and cylindrical detectors are provided, and the framework accommodates future extensions to other surfaces.
Core dependencies include NumPy and SciPy for numerical operations, lmfit (Newville et al., [52]) for the least-squares refinements of § 7 and § 8, and the FabIO library (Knudsen et al., [35]) for image formats. HEXRDGUI adds PySide6 for the interface framework and Matplotlib for plotting. A C/C++ compiler (gcc or Visual Studio) is required only when installing from source (e.g. via pip) to build the compiled extensions; the binary conda packages and standalone bundles described below ship prebuilt extensions.
Distribution and platform support
Binary packages for HEXRD and HEXRDGUI are distributed via conda from the hexrd channel, and as standalone application bundles for Windows (MSI installer and portable .zip), macOS (disk images for both Apple Silicon and Intel architectures), and Linux (tarball). Installation via conda requires approximately 2 GB of disk space for the full environment including dependencies. Standalone bundles include all dependencies and do not require a separate Python installation. The GUI is optimized for displays with resolution pixels; smaller displays are usable but may require scrolling within dialogs.
Both packages are under active development with stable releases tagged and distributed under the hexrd channel’s main label, and pre-release versions available from the hexrd/label/prerelease channel for early testing. The open-source development model allows community contributions through GitHub pull requests; all contributions must use the BSD 3-Clause license consistent with the upstream code (LLNL-CODE-819716). Documentation for HEXRDGUI is hosted at hexrdgui.readthedocs.io, including installation, configuration and calibration guides and tutorial workflows.
Availability
HEXRD ([2]) and HEXRDGUI ([4]) are released under a BSD 3-Clause license, and the source code for both is hosted in the HEXRD organization on GitHub (github.com/HEXRD).
Suite of examples
To illustrate the range of experimental configurations supported by HEXRD, we present examples from six different instrument platforms spanning several major dynamic compression facilities. Each highlights a different part of the workflow: pink-beam profiles at DCS, dual-source calibration with a cylindrical plate at TARDIS, liquid and amorphous scattering on multi-panel detectors at the European XFEL and MEC, the sample-layer and pinhole corrections at PXRDIP, and rigid-body calibration of a five-camera array at FIDDLE.
DCS at APS
The Dynamic Compression Sector beamline at the Advanced Photon Source provides pink-beam or monochromatic X-ray diffraction for gas-gun- and laser-driven experiments. Figure 12 shows a pink-beam calibration dataset acquired with a Rayonix detector and calibrated with the powder calibration workflow (Figure 6 shows the same dataset during calibration). For whole-pattern fitting of such data the WPPF module provides the pink-beam peak-profile model of Von Dreele et al. [70], which was used for the structural analysis of shock-compressed iron at DCS in Singh et al. [64].

Figure 12. DCS pink-beam calibration example with a Rayonix detector, shown as a polar view (azimuth vs. nominal momentum transfer , computed from the measured angles with the nominal beam energy) with simulated ring positions (vertical blue lines) and the azimuthal average below. The asymmetric peak shapes characteristic of the pink-beam spectrum (sharp onsets with extended tails toward higher ) are modeled by the pink-beam peak profile, used here for the calibration peak fits and also available for whole-pattern fitting. The simulated lines mark the ring positions at the nominal energy and therefore sit at the sharp low- onset of each peak, slightly below the apparent peak maximum, as expected for a pink beam.
TARDIS at NIF
The TARDIS platform at NIF ([61]) uses image plates arranged around the target with a pinhole collimator and quasi-monochromatic X-ray sources generated by laser irradiation of metal foils. At LLNL, TARDIS and PXRDIP data have also been analyzed with two platform-specific IGOR Pro programs, AnalyzeTARDIS and AnalyzePXRDIP ([48]), which share a common origin but have since diverged; HEXRD treats both platforms, and the others in this section, with one geometric model. Figure 13 shows data from a Pt sample ramp-compressed and probed with two X-ray sources, shown in three views: the polar view centered at X-ray source 1 (XRS1), the polar view centered at X-ray source 2 (XRS2), and the stereographic view, demonstrating HEXRD’s multi-beam calibration capability.
Figure 13. TARDIS examples showing ramp-compressed Pt at NIF. Blue and red lines are simulated Pt diffraction patterns from XRS1 and XRS2, respectively. Top left: polar view centered at XRS1. Top right: polar view centered at XRS2. Both polar views have a sample layer distortion applied, with IMAGE-PLATE-3 excluded from the azimuthal lineout; magenta dashed lines indicate manually masked regions. Bottom: stereographic view showing simulated patterns from both sources.
European XFEL
At the High Energy Density (HED) instrument of the European XFEL, shock-compressed liquid Sn was studied on two Varex detectors using the DiPOLE 100-X laser (Figure 14). This example demonstrates HEXRD’s ability to handle free-electron-laser diffraction data with minimal adaptation [26]. Amorphous and liquid samples give broad halos rather than sharp rings; once the detectors are calibrated, the azimuthal average gives their scattering profile directly.

Figure 14. Shock-compressed liquid Sn at the European XFEL using two Varex detectors [26], shown as a polar view ( vs. nominal momentum transfer ) with the azimuthal average below. Diffraction spots from solid Sn were masked (white ellipses) in the images before import, so that the average shows the broad scattering of the liquid.
MEC at LCLS
The Matter in Extreme Conditions end-station at LCLS uses femtosecond X-ray pulses to probe materials under extreme conditions. Figure 15 shows an ambient metallic glass sample recorded on two Varex detectors rather than the ePix10k detectors of the standard configuration (Figure 5), which shows that HEXRD adapts to non-standard configurations with minimal changes.

Figure 15. Ambient metallic glass with single-shot XRD at MEC (LCLS) using two Varex detectors, shown as a polar view with the azimuthal average below, in which the broad peaks are the amorphous scattering of the glass; unpublished data from LCLS proposal L1013376, courtesy of C. McGuire and D. Khaghani. A -pixel median filter was applied to the detector images to reduce noise.
PXRDIP at OMEGA EP
The PXRDIP platform at the OMEGA EP laser facility captures powder diffraction from shock-ramp-compressed samples on image plates, using quasi-monochromatic X-ray sources produced by laser irradiation of metal foils. Figure 16 shows data from a gold sample under shock-ramp compression [17].

Figure 16. Shock-ramp-compressed gold at OMEGA EP (LLE) on the PXRDIP platform. Red simulated pattern lines correspond to the high-pressure FCC Au sample and blue lines to the ambient Pt pinhole. Left: polar view with a sample distortion applied, which straightens the red sample lines while curving the blue pinhole lines. Right: stereographic view, with the range taken low enough to include the low-angle plates, so that the coverage extends into the center of the projection. The dashed line shows the edge of the stereographic circle. The missing high-angle data in the stereographic view are clipped due to the pinhole camera geometry.
FIDDLE at NIF
The FIDDLE diagnostic captures time-resolved XRD using arrays of Icarus detectors with 2 ns time resolution. Figure 17 shows an ambient dataset imaged using FIDDLE. It was recorded separately from the dataset of Vennari et al. [68] and analyzed with the same workflow. The positions and orientations of the five cameras relative to one another were determined by the NIF Advanced Tracking Laser Alignment System (ATLAS), and only the position and orientation of the instrument as a whole were then refined, as one rigid body, against hand-picked rings of both phases present. With those six parameters the simulated rings line up with the data on all five cameras and with the peaks of the lineout (Figure 17). The distributed state file carries the picks, each phase with its own target-assembly correction: 53 on the powder (36 of powder in epoxy against the source side of a Mo pinhole 110 thick and 380 in diameter), which takes the sample-layer correction, and 24 on the Mo pinhole substrate itself, which takes the pinhole correction. Holding that calibration fixed and allowing only the lattice parameter to vary, the composite calibrator returns (standard error from the picks), 290 ppm or 1.7 standard errors above the value certified for NIST SRM 674b, which is less than one 25 pixel (0.8 pixel for a ring at ) at these camera distances of 57–74 mm. The placement of the instrument was refined against these same rings, so this result tests the consistency of the calibration. An offset of this size could also come from the effective energy of the laser-plasma source alone: 290 ppm corresponds to about 3 eV at the 10.25 keV Ge He- line, much less than the energy spread of the He- line complex. The picks fit with an RMS residual of 75 mdeg for and 62 mdeg for Mo (the Mo rings do not enter the lattice-parameter result). A script distributed with the validation scripts of § 7.6 repeats this from the state file.

Figure 17. Ambient CeO₂ dataset from the FIDDLE instrument at NIF, showing multi-panel Icarus data. Cyan simulated pattern lines correspond to the CeO₂ sample and red lines to the Mo pinhole. The polar view has the pinhole correction applied (), and each set of simulated lines carries its own target-assembly correction (sample layer for CeO₂, pinhole for Mo).
Conclusions
Dynamic compression XRD relies on non-standard instrument configurations (arbitrarily oriented multi-panel detectors, pinhole collimation, curved image plates, polychromatic sources, and dual-beam geometries) that no single general-purpose package covers in full. HEXRD and HEXRDGUI address this gap through a flexible geometric framework that supports a broad range of detector arrangements and beam configurations.
The main advance of HEXRD over existing software is that it handles these instruments in one program, from calibration through whole-pattern refinement. The combination of three capabilities distinguishes HEXRD: (1) a generic geometric model supporting arbitrary multi-detector configurations including cylindrical surfaces, (2) integrated corrections for facility-specific geometries including pinhole distortion and layered target assemblies, and (3) native multi-beam calibration enabling simultaneous refinement against diffraction from independent X-ray sources. Coupled with whole-pattern peak fitting via Le Bail and Rietveld methods, these features allow experimentalists to carry out the analysis, from instrument calibration to whole-pattern refinement, within a single tool across multiple platforms. Comparisons with independent codes and synthetic ground-truth tests confirm the geometric model and the multi-source calibration. The software has been adopted for routine analysis of data recorded at several major dynamic compression platforms. Continued development focuses on automation of calibration workflows, support for emerging platforms, and expansion of the whole-pattern fitting module.
Acknowledgements
The authors thank Chris Harris, Marcus D. Hanwell and Donald E. Boyce for their contributions to the code, Paul Shade (AFRL), the many experimentalists who have provided feedback and testing, and the beamline and facility scientists at the facilities where the data shown here were recorded. We thank Chris McGuire (LLNL) and Dimitri Khaghani (SLAC) for providing the metallic glass data recorded at MEC using Varex detectors under LCLS proposal L1013376; the ePix10k data were recorded at MEC in LCLS experiment LV7318. Use of the Linac Coherent Light Source (LCLS), SLAC National Accelerator Laboratory, is supported by the U.S. Department of Energy (DOE), Office of Science, Office of Basic Energy Sciences under Contract No. DE-AC02-76SF00515. This research used resources of the Advanced Photon Source, a DOE Office of Science user facility operated for the DOE Office of Science by Argonne National Laboratory under Contract No. DE-AC02-06CH11357. This publication is partially based upon work performed at the Dynamic Compression Sector, which is operated by Washington State University under the U.S. DOE/National Nuclear Security Administration under Award No. DE-NA0003957. We acknowledge European XFEL in Schenefeld, Germany, for provision of X-ray free-electron laser beamtime at the HED instrument under community proposal 2720, and thank its staff for their assistance. We acknowledge the use of Anthropic’s Claude in revising the text during the preparation of this manuscript; the final contents remain the sole responsibility of the authors.
Funding
The development of HEXRD and HEXRDGUI has been supported through subcontracts with Lawrence Livermore National Laboratory and the Air Force Research Laboratory. This work was performed under the auspices of the US Department of Energy by Lawrence Livermore National Laboratory under Contract No. DE-AC52-07NA27344 (LLNL-JRNL-2024959).
Conflicts of interest: The authors declare no conflicts of interest.
Data availability: HEXRD and HEXRDGUI are freely available as open-source software at github.com/HEXRD. The HEXRDGUI state files of every dataset shown in this article (image data, calibration, materials, masks and, where used, the calibration picks), the validation scripts that reproduce the validation numbers of § 2.4, § 7.4, § 5.1, § 7.6 and § 8 and of the FIDDLE example, the converters between HEXRD instrument files and pyFAI poni files, and the Dioptas calibration with its picked positions are deposited on Zenodo ([3]). The GE CeO₂ dataset was recorded at beamline 1-ID-E of the Advanced Photon Source under an Air Force Research Laboratory Partner User Proposal.
References
References
- [1]Ashiotis, G., Deschildre, A., Nawaz, Z., Wright, J. P., Karkoulis, D., Picca, F. E. & Kieffer, J. (2015). Journal of Applied Crystallography, 48(2), 510–519.
- [2]Avery, P., Bernier, J. V., Boyce, D. E., Harris, C., Lewis, K., Singer, Z., Major, B., Tsolakis, C., Welsh, K., Lim, R., Pagan, D. C., Tourtellott, J., Villellas Guillén, Ó., Rygg, J. R., Gustafson, S., Budrow, C., Nygren, K., Shade, P. & Singh, S. (2026a). HEXRD: Highly Extensible X-Ray Diffraction. Version 0.10.7. Zenodo. https://doi.org/10.5281/zenodo.23426740
- [3]Avery, P., Bernier, J. V., Major, B. & Singh, S. (2026b). State files and validation scripts for “HEXRD and HEXRDGUI: a flexible toolkit for X-ray diffraction analysis in dynamic compression experiments”. Zenodo. https://doi.org/10.5281/zenodo.23440009
- [4]Avery, P., Major, B., Harris, C., Tourtellott, J., Gustafson, S., Budrow, C., Nygren, K., Shade, P., Bernier, J. V. & Singh, S. (2026c). HEXRDGUI. Version 1.2.6. Zenodo. https://doi.org/10.5281/zenodo.23428756
- [6]Bernier, J. V., Barton, N. R., Lienert, U. & Miller, M. P. (2011). The Journal of Strain Analysis for Engineering Design, 46(7), 527–547.
- [7]Bernier, J. V., Suter, R. M., Rollett, A. D. & Almer, J. D. (2020). Annual Review of Materials Research, 50, 395–436.
- [8]Birch, F. (1947). Physical Review, 71(11), 809–824.
- [9]Boyce, D. E. & Bernier, J. V. (2013). heXRD: Modular, Open Source Software for the Analysis of High Energy X-Ray Diffraction Data. Tech. Rep. LLNL-SR-609815. Lawrence Livermore National Laboratory.DOI
- [10]Branch, M. A., Coleman, T. F. & Li, Y. (1999). SIAM Journal on Scientific Computing, 21(1), 1–23.
- [11]Briggs, R., Gorman, M. G., Zhang, S., McGonegle, D., Coleman, A. L., Coppari, F., Morales-Silva, M. A., Smith, R. F., Wicks, J. K., Bolme, C. A., Gleason, A. E., Cunningham, E., Lee, H. J., Nagler, B., McMahon, M. I., Eggert, J. H. & Fratanduono, D. E. (2019). Applied Physics Letters, 115, 264101.
- [12]Brunton, G., Abed, Y., Fedorov, M., Fishler, B., Larson, D., Ludwigsen, A. P., Mathisen, D., Miller-Kamm, V., Paul, M., Reed, R., Speck, D., Stout, E., Townsend, S., Weaver, S., Wilson, R. & Van Wonterghem, B. (2018). In Proc. of International Conference on Accelerator and Large Experimental Control Systems (ICALEPCS’17), Barcelona, Spain, 8-13 October 2017, no. 16 in International Conference on Accelerator and Large Experimental Control Systems, pp. 14–18. Geneva, Switzerland: JACoW. http://jacow.org/icalepcs2017/papers/moap03.pdf
- [13]Caglioti, G., Paoletti, A. & Ricci, F. P. (1958). Nuclear Instruments, 3(4), 223–228.
- [14]Chantler, C. T. (1995). Journal of Physical and Chemical Reference Data, 24(1), 71–643. https://doi.org/10.1063/1.555974
- [15]Chantler, C. T. (2000). Journal of Physical and Chemical Reference Data, 29(4), 597–1056. https://doi.org/10.1063/1.1321055
- [16]Coelho, A. A. (2018). Journal of Applied Crystallography, 51(1), 210–218.
- [17]Coleman, A. L., Singh, S., Lockard, T. E., Ocampo, I. K., Lazicki, A. E., Gorman, M. G., Racioppi, S., Krygier, A. G., Wehrenberg, C. E., Ahmad, R., Millot, M., Hamel, S., Han, S., Ginnane, M. K., Swift, D. C., Bonev, S. A., Zurek, E., Duffy, T. S., Eggert, J. H., McNaney, J. & Smith, R. F. (2025). Physical Review Letters, 135, 186101.
- [18]Coleman, A. L., Singh, S., Vennari, C. E., Smith, R. F., Volz, T. J., Gorman, M. G., Clarke, S. M., Eggert, J. H., Coppari, F., Fratanduono, D. E. & Briggs, R. (2022). Journal of Applied Physics, 131(1), 015901.
- [19]Coppari, F., Smith, R. F., Eggert, J. H., Wang, J., Rygg, J. R., Lazicki, A., Hawreliak, J. A., Collins, G. W. & Duffy, T. S. (2013). Nature Geoscience, 6, 926–929.
- [21]Doebelin, N. & Kleeberg, R. (2015). Journal of Applied Crystallography, 48(5), 1573–1580. https://journals.iucr.org/j/issues/2015/05/00/kc5013/
- [22]Dollase, W. A. (1986). Journal of Applied Crystallography, 19(4), 267–272. https://onlinelibrary.wiley.com/doi/abs/10.1107/S002188986089458
- [23]Finger, L. W., Cox, D. E. & Jephcoat, A. P. (1994). Journal of Applied Crystallography, 27(6), 892–900.
- [24]Gordon, J. V., Lim, R. E., Wilkin, M. J., Pagan, D. C., Lebensohn, R. A. & Rollett, A. D. (2021). Acta Materialia, 215, 117120.
- [25]Gorman, M. G., Elatresh, S., Lazicki, A., Cormier, M. M. E., Bonev, S. A., McGonegle, D., Briggs, R., Coleman, A. L., Rothman, S. D., Peacock, L., Bernier, J. V., Coppari, F., Braun, D. G., Rygg, J. R., Fratanduono, D. E., Hoffmann, R., Collins, G. W., Wark, J. S., Smith, R. F., Eggert, J. H. & McMahon, M. I. (2022). Nature Physics, 18(11), 1307–1311.
- [26]Gorman, M. G., McGonegle, D., Smith, R. F. et al. (2024). Journal of Applied Physics, 135(16), 165902.
- [27]Gustafson, S. E., Ludwig, W., Shade, P. A., Naragani, D., Pagan, D. C., Cook, P., Yildirim, C., Detlefs, C. & Sangid, M. D. (2020). Nature Communications, 11, 3189.
- [28]Hammersley, A. P. (2016). Journal of Applied Crystallography, 49(2), 646–652.
- [29]Hari, A., Katagiri, K., Li, W., Luccioni, D. P., Lin, R., Parsons, S. E., Xu, Z., Hari, R., Reddy, T., Cubit, E. W., Amouretti, A., Eggert, J. H., Inubushi, Y., Irifune, T., Irvine, S. J., Kodama, R., Koenig, M., Madril, J., Matsuoka, T., Miyanishi, K., Nakamura, H., Nishiyama, N., Okuchi, T., Ota, M., Sekine, T., Seto, Y., Shinmei, T., Sueda, K., Tange, Y., Takagi, S., Togashi, T., Umeda, Y., Wang, Y., Yabashi, M., Yabuuchi, T., Ozaki, N. & Dresselhaus-Marais, L. E. (2026). Science Advances, 12(38), eead5546. https://www.science.org/doi/abs/10.1126/sciadv.aed5546
- [30]Huang, H. H., Ryu, M., Kamegaki, S., Stonyte, D., Malinauskas, T., Nishijima, Y., Hocking, R., Le, N. H. A., Katkus, T., Mu, H., Ng, S. H., Pinches, S., Ang, A. S. M., Mizeikis, V., Zatsepin, N., Miyanishi, K., Yabuuchi, T., Nakamura, H., Amouretti, A., Ozaki, N., Vinci, T., Vailionis, A., Gamaly, E. G., Hicks, D. G., Morikawa, J. & Juodkázis, S. (2026). High entropy alloy under shock compression: optical-pump X-ray-probe. arXiv:2603.21076. https://arxiv.org/abs/2603.21076
- [31]Hurley, R. C., Lind, J., Pagan, D. C., Akin, M. C. & Herbold, E. B. (2018). Journal of the Mechanics and Physics of Solids, 112, 273–290.
- [32]Izumi, F. (1989). Rigaku Journal, 6(1), 10–17.
- [35]Knudsen, E. B., Sørensen, H. O., Wright, J. P., Goret, G. & Kieffer, J. (2013). Journal of Applied Crystallography, 46(2), 537–539.
- [36]Kraus, D., Ravasio, A., Gauthier, M., Gericke, D. O., Vorberger, J., Frydrych, S., Helfrich, J., Fletcher, L. B., Schaumann, G., Nagler, B., Barbel, B., Bachmann, B., Gamboa, E. J., Göde, S., Granados, E., Gregori, G., Lee, H. J., Neumayer, P., Schumaker, W., Döppner, T., Falcone, R. W., Glenzer, S. H. & Roth, M. (2016). Nature Communications, 7, 10970.
- [37]Kraus, D., Rips, J., Schörner, M. et al. (2025). Nature, 642, 351–355.
- [38]Kraus, R. G., Hemley, R. J., Ali, S. J., Belof, J. L., Benedict, L. X., Bernier, J., Braun, D., Cohen, R. E., Collins, G. W., Coppari, F., Desjarlais, M. P., Fratanduono, D., Hamel, S., Krygier, A., Lazicki, A., McNaney, J. J., Millot, M., Myint, P. C., Newman, M. G., Rygg, J. R., Sterbentz, D. M., Stewart, S. T., Stixrude, L., Swift, D. C., Wehrenberg, C. & Eggert, J. H. (2022). Science, 375(6577), 202–205.
- [39]Lam, S. K., Pitrou, A. & Seibert, S. (2015). In Proceedings of the Second Workshop on the LLVM Compiler Infrastructure in HPC, pp. 1–6. ACM.
- [40]Lazicki, A., McGonegle, D., Rygg, J. R., Braun, D. G., Swift, D. C., Gorman, M. G., Smith, R. F., Heighway, P. G., Higginbotham, A., Suggit, M. J., Fratanduono, D. E., Coppari, F., Wehrenberg, C. E., Kraus, R. G., Erskine, D., Bernier, J. V., McManey, J. M., Rudd, R. E., Collins, G. W., Eggert, J. H. & Wark, J. S. (2021). Nature, 589, 532–535.
- [41]Lazicki, A., Rygg, J. R., Coppari, F., Smith, R., Fratanduono, D., Kraus, R. G., Collins, G. W., Briggs, R., Braun, D. G., Swift, D. C. & Eggert, J. H. (2015). Physical Review Letters, 115, 075502.
- [42]Le Bail, A., Duroy, H. & Fourquet, J. L. (1988). Materials Research Bulletin, 23(3), 447–452.
- [43]Lin, R., Hari, A., Cubit, E. W., Katagiri, K., Parsons, S., Xu, Z., Ren, J., Yang, W., Amouretti, A., Irvine, S. J., Kodama, R., Madril, L., Miyanishi, K., Nakamura, H., Ozaki, N., Reddy, T., Seto, Y., Sueda, K., Togashi, T., Umeda, Y., Yabashi, M., Yabuuchi, T., Vohra, Y. K., Chen, W. & Dresselhaus-Marais, L. E. (2026). Scripta Materialia, 281, 117347.
- [45]McBride, E. E., Krygier, A., Ehnes, A., Galtier, E., Harmand, M., Konôpková, Z., Lee, H. J., Liermann, H.-P., Nagler, B., Pelka, A., Rödel, M., Schropp, A., Smith, R. F., Spindloe, C., Swift, D., Tavella, F., Toleikis, S., Tschentscher, T., Wark, J. S. & Higginbotham, A. (2019). Nature Physics, 15, 89–94.
- [46]McHardy, J. D., Storm, C. V., Duff, M. J., MacLeod, S. G. & McMahon, M. I. (2023). High Pressure Research, 43(1), 40–57. Correction: High Pressure Research 43, 262 (2023), doi:10.1080/08957959.2023.2225889.
- [47]Milathianaki, D., Boutet, S., Williams, G. J., Higginbotham, A., Ratner, D., Gleason, A. E., Messerschmidt, M., Seibert, M. M., Swift, D. C., Hering, P., Robinson, J., White, W. E. & Wark, J. S. (2013). Science, 342, 220–223.
- [48]Millot, M., Coppari, F., Lazicki, A. & Eggert, J. H. (2026). Data analysis methods for powder x-ray diffraction under laser-driven dynamic compression at Omega and NIF laser facilities. arXiv:2606.23602 [cond-mat.mtrl-sci]. https://arxiv.org/abs/2606.23602
- [49]Millot, M., Coppari, F., Rygg, J. R., Correa Barrios, A., Hamel, S., Swift, D. C. & Eggert, J. H. (2019). Nature, 569, 251–255.
- [50]Morhác, M., Kliman, J., Matoušek, V., Veselý, M. & Turzo, I. (1997). Nuclear Instruments and Methods in Physics Research A, 401, 113–132.
- [51]Nagler, B., Arnold, B., Bouchard, G., Boyce, R. F., Boyce, R. M., Callen, A., Campell, M., Curiel, R., Galtier, E., Garofoli, J., Granados, E., Hastings, J., Hays, G., Heimann, P., Lee, R. W., Milathianaki, D., Plummer, L., Schropp, A., Wallace, A., Welch, M., White, W., Xing, Z., Yin, J., Young, J., Zastrau, U. & Lee, H. J. (2015). Journal of Synchrotron Radiation, 22(3), 520–525.
- [52]Newville, M., Stensitzki, T., Allen, D. B. & Ingargiola, A. (2014). LMFIT: Non-linear least-square minimization and curve-fitting for Python. https://doi.org/10.5281/zenodo.11813
- [55]Polsin, D. N., Rygg, J. R., Bishel, D. T., Chin, D. A., Coppari, F., Gong, X., Ivancic, S. T., Regan, S. P., Werellapatha, K., Collins, G. W., Marshall, M. C. & Benedetti, L. R. (2025). Review of Scientific Instruments, 96(6), 063510.
- [56]Prescher, C. & Prakapenka, V. B. (2015). High Pressure Research, 35(3), 223–230.
- [57]Rietveld, H. M. (1969). Journal of Applied Crystallography, 2(2), 65–71.
- [58]Rodríguez-Carvajal, J. (1993). Physica B: Condensed Matter, 192(1), 55–69.
- [59]Ryan, C. G., Clayton, E., Griffin, W. L., Sie, S. H. & Cousens, D. R. (1988). Nuclear Instruments and Methods in Physics Research Section B, 34(3), 396–402.
- [60]Rygg, J. R., Eggert, J. H., Lazicki, A. E., Coppari, F., Hawreliak, J. A., Hicks, D. G., Smith, R. F., Sorce, C. M., Uphaus, T. M., Yaakobi, B. & Collins, G. W. (2012). Review of Scientific Instruments, 83(11), 113904. https://doi.org/10.1063/1.47766464
- [61]Rygg, J. R., Smith, R. F., Lazicki, A. E., Braun, D. G., Fratanduono, D. E., Kraus, R. G., McNaney, J. M., Swift, D. C., Wehrenberg, C. E., Coppari, F., Ahmed, M. F., Barrios, M. A., Blobbaum, K. J. M., Collins, G. W., Cook, A. L., Di Nicola, P., Dzenitis, E. G., Gonzales, S., Heidl, B. F., Hohenberger, M., House, A., Izumi, N., Kalantar, D. H., Khan, S. F., Kohut, T. R., Kumar, C., Masters, N. D., Polsin, D. N., Regan, S. P., Smith, C. A., Vignes, R. M., Wall, M. A., Ward, J., Wark, J. S., Zobrist, T. L., Arsenlis, A. & Eggert, J. H. (2020). Review of Scientific Instruments, 91(4), 043902.
- [62]Rygg, J. R., Smith, R. F., Lazicki, A. E., Braun, D. G., Fratanduono, D. E., Kraus, R. G., McNaney, J. M., Swift, D. C., Wehrenberg, C. E., Coppari, F., Ahmed, M. F., Barrios, M. A., Blobbaum, K. J. M., Collins, G. W., Cook, A. L., Di Nicola, P., Dzenitis, E. G., Gonzales, S., Heidl, B. F., Hohenberger, M., House, A., Izumi, N., Kalantar, D. H., Khan, S. F., Kohut, T. R., Kumar, C., Masters, N. D., Polsin, D. N., Regan, S. P., Smith, C. A., Vignes, R. M., Wall, M. A., Ward, J., Wark, J. S., Zobrist, T. L., Arsenlis, A. & Eggert, J. H. (2026). Review of Scientific Instruments, 97(1), 019901.
- [64]Singh, S., Briggs, R., Gorman, M. G., Benedict, L. X., Wu, C. J., Hamel, S., Coleman, A. L., Coppari, F., Fernandez-Panella, A., McGuire, C., Sims, M., Wicks, J. K., Eggert, J. H., Fratanduono, D. E. & Smith, R. F. (2023). Physical Review B, 108, 184104.
- [65]Stephens, P. W. (1999). Journal of Applied Crystallography, 32(2), 281–289.
- [66]Thompson, P., Cox, D. E. & Hastings, J. B. (1987). Journal of Applied Crystallography, 20(2), 79–83.
- [67]Toby, B. H. & Von Dreele, R. B. (2013). Journal of Applied Crystallography, 46(2), 544–549.
- [68]Vennari, C. E., Palmer, N. E., Nyholm, P. R., Bhandarkar, N. S., Nagel, S. R., Petre, R. B., Stan, C. V., Eggert, J. H., Bradley, D. K., Ping, Y., Thomas, A., Swift, D. C., Carpenter, A. C., MacKinnon, A. J. & Benedetti, L. R. (2024). Review of Scientific Instruments, 95, 093534.
- [69]Von Dreele, R. B. (1997). Journal of Applied Crystallography, 30(4), 517–525. https://onlinelibrary.wiley.com/doi/abs/10.1107/S0021889897005918
- [70]Von Dreele, R. B., Clarke, S. M. & Walsh, J. P. S. (2021). Journal of Applied Crystallography, 54(1), 3–6.
- [71]Waasmaier, D. & Kirfel, A. (1995). Acta Crystallographica Section A, 51(3), 416–431. https://doi.org/10.1107/S0108767394013292
- [72]Wark, J. S., McMahon, M. I. & Eggert, J. H. (2022). Journal of Applied Physics, 132(8), 080902.
- [73]Wark, J. S., Peake, D. J., Stevens, T. et al. (2025). Journal of Applied Physics, 137(15), 155904. https://doi.org/10.1063/5.0256844
- [74]Wark, J. S., Whitlock, R. R., Hauer, A. A., Swain, J. E. & Solone, P. J. (1989). Physical Review B, 40, 5705–5714.
- [75]Warren, B. E. (1969). X-Ray Diffraction. Reading, Massachusetts: Addison-Wesley.
- [76]Wehrenberg, C. E., McGonegle, D., Bolme, C., Higginbotham, A., Lazicki, A., Lee, H. J., Nagler, B., Park, H.-S., Remington, B. A., Rudd, R. E., Sliwa, M., Suggit, M., Swift, D., Tavella, F., Zepeda-Ruiz, L. & Wark, J. S. (2017). Nature, 550, 496–499.
- [77]Werellapatha, K., Palmer, N. E., Gorman, M. G., Bernier, J. V., Bhandarkar, N. S., Bradley, D. K., Braun, D. G., Bruhn, M., Carpenter, A., Celliers, P. M., Coppari, F., Dayton, M., Durand, C., Eggert, J. H., Ferguson, B., Heidl, B., Heinbockel, C., Heredia, R., Huckins, J., Hurd, E., Hsing, W., Krauland, C. M., Lazicki, A. E., Kalantar, D., Kehl, J., Killebrew, K., Masters, N., Millot, M., Nagel, S. R., Petre, R. B., Ping, Y., Polsin, D. N., Singh, S., Stan, C. V., Swift, D., Tabimina, J., Thomas, A., Zobrist, T. & Benedetti, L. R. (2024). Review of Scientific Instruments, 95(1), 013903.
- [78]Wicks, J. K., Singh, S., Millot, M., Fratanduono, D. E., Coppari, F., Gorman, M. G., Ye, Z., Rygg, J. R., Hari, A., Eggert, J. H., Duffy, T. S. & Smith, R. F. (2024). Science Advances, 10(23), eadk0306. https://www.science.org/doi/abs/10.1126/sciadv.adk0306
- [79]Winter, G., Waterman, D. G., Parkhurst, J. M., Brewster, A. S., Gildea, R. J., Gerstel, M., Fuentes-Montero, L., Vollmar, M., Michels-Clark, T., Young, I. D., Sauter, N. K. & Evans, G. (2018). Acta Crystallographica Section D, 74(2), 85–97.


