Dark Field X-ray Microscopy (DFXM)

• Dark Field X-ray Microscopy (DFXM) is a diffraction-based imaging technique that non-destructively maps lattice orientation, strain, and defects in bulk crystalline materials by isolating a selected diffracting volume.
• It employs weak-beam and strong-beam configurations where angular rocking-curve scans encode information on elastic strain and mosaicity through intensity variations.
• Recent advancements include high-resolution objectives, integration with high-energy and time-resolved sources, and automated machine-learning workflows for quantitative 3D defect mapping.

Dark Field X-ray Microscopy (DFXM) is a diffraction-based, full-field X-ray imaging technique for non-destructive mapping of orientation, strain, domains, and defects inside bulk crystalline materials. Its defining configuration places an objective lens in the Bragg-diffracted beam, producing a magnified real-space image of a selected diffracting volume rather than of the transmitted beam. In this way, DFXM combines hard-X-ray penetration depth with microscopy-style spatial selectivity, and has been developed for three-dimensional mapping of embedded microstructures that are inaccessible to surface-sensitive or thin-foil methods such as conventional transmission electron microscopy (Isern et al., 2024, Poulsen et al., 2020, Pal et al., 2024).

1. Physical basis and contrast mechanisms

DFXM operates in Bragg geometry. A crystalline region contributes to the image only when its local lattice spacing and orientation satisfy the diffraction condition, conventionally expressed as

$2d_{hkl}\sin\theta = \lambda$

or, in related formulations,

$2d_{hkl}\sin\theta = n\lambda .$

Local strain and lattice rotation shift the Bragg condition, so spatial variations in orientation and deformation appear as intensity variations in the diffracted image (Shukla et al., 25 Aug 2025, Isern et al., 2024).

The method is especially useful for defects whose long-range elastic fields perturb the local lattice. In DFXM images of single-crystal aluminum, dislocations appear as asymmetric joined bright and dark regions rather than as direct images of the dislocation core. The contrast arises from subtle displacement gradients in the crystal lattice, to which the microscope is sensitive as a reciprocal-space filter (Gonzalez et al., 2020). For near-perfect crystals and weakly perturbed regions, DFXM can also display dynamical diffraction phenomena such as Borrmann fringes (Carlsen et al., 2022).

A central operational distinction is between weak-beam and strong-beam conditions. In DFXM, weak-beam images are acquired at the tails of a rocking curve, where kinematic scattering dominates and dislocation contrast is enhanced. Strong-beam conditions occur near the Bragg peak and are dominated by multiple scattering or dynamical diffraction effects, which can obscure defect contrast and complicate interpretation (Benhadjira et al., 5 Sep 2025). This distinction, long familiar in electron microscopy, has become equally important in DFXM because defect visibility, Burgers-vector analysis, and automated image interpretation all depend on the chosen angular setting.

Rocking-curve imaging extends this principle into a mapping modality. As the sample is rocked through the reflection, one acquires a sequence of images whose pixelwise intensity maxima or centers of mass encode local mosaicity, orientation spread, and, depending on geometry, elastic strain. In one formulation used at ID03, local strain along the scattering vector is related to the Bragg-angle shift by

$$\epsilon = \frac{\Delta d_{hkl}}{d_{hkl}} = -\cot(\theta_B)\Delta\theta ,$$

providing a direct link between angular scans and deformation fields (Isern et al., 2024).

2. Optical architecture, beamlines, and imaging geometries

The canonical DFXM instrument illuminates a sample with monochromatic or broader-bandwidth hard X-rays, selects a Bragg-diffracted beam, and images that beam with an X-ray objective onto a detector. Compound refractive lenses (CRLs) have been the standard objective, but recent work has expanded the optical basis of the technique. The ESRF ID03 instrument integrates a CPMU undulator source, a channel-cut crystal monochromator for conventional monochromatic operation, a double multilayer monochromator for pink-beam operation, a transfocator with Be and polished diamond CRLs, an upgraded multi-axis goniometer, and near-field and far-field camera systems. The beamline supports both projection mode, using box-beam illumination, and section topography, using line-beam illumination for thin-section imaging (Isern et al., 2024).

Objective development has materially changed DFXM performance. A crossed pair of multilayer Laue lenses used as a DFXM objective at 19.0 keV, with a physical aperture of $50 \times 50~\mathrm{\mu m^2}$ and focal length of 14.25 mm, yielded a spatial resolution of $55.6 \pm 0.3$ nm by a 10% MTF criterion in bright-field mode; the dark-field resolution was reported as similar, and the measured efficiency was 26.7% (Staeck et al., 22 Apr 2026). This places DFXM in a sub-60 nm regime that had previously been limited by CRL figure errors.

At higher photon energies, diamond CRLs have been introduced as an alternative objective material. Characterization at 17, 33, and 37 keV showed that diamond objectives can support high-energy DFXM and enable studies of thicker or more absorbing samples. At 33 keV, the capability was demonstrated on two iron-based samples of approximately 0.5 mm thickness that could not be probed at 17 keV (Staeck et al., 25 May 2026). This high-energy extension is consequential because photon energy directly determines penetration, absorption, and the range of sample systems accessible to diffraction microscopy.

The method is also compatible with multimodal layouts. At XFELs, DFXM has been combined with simultaneous bright-field X-ray microscopy so that transmitted-beam and diffracted-beam images from the same volume are collected together. In that configuration, bright-field images provide density and absorption contrast, while dark-field images isolate lattice distortions, orientation, and defect contrast (Dresselhaus-Marais et al., 2022).

3. Forward models, diffraction theory, and tensor reconstruction

A large part of DFXM’s maturation has occurred through its forward and inverse theoretical formalisms. In the geometrical-optics approach, the local deformation is expressed through a deformation-gradient tensor $\mathbf{F}^g$, which maps the undeformed lattice to the deformed one. In reciprocal space,

$\mathbf{B} = (\mathbf{F}^g)^{-T}\mathbf{B}_0 ,$

and the detector intensity is written as an integral over real space mediated by the instrumental resolution function,

$$I(y_i', z_i') = C \int_{\vec{r}} \Phi_0(\vec{r}) \rho(\vec{r}) \mathrm{Res}(\vec{r}, \vec{q}(\vec{r}), y_i', z_i')\, d^3\vec{r} .$$

This formalism was developed for any crystallographic space group and explicitly supports tailoring of the reciprocal-space resolution function to map selected components of the deformation gradient tensor (Poulsen et al., 2020).

For complex discrete dislocation structures, a scalable DD-DFXM forward model has been built by coupling non-singular elasticity for the local strain field to an efficient geometrical-optics algorithm for image formation. In that model, the total deformation gradient is obtained by superposition of contributions from discretized dislocation segments, and pixel intensity is computed by integrating the local diffraction response over the gauge volume. The framework was applied to large-scale molecular-dynamics-derived dislocation structures in compressed single-crystal silicon and was shown to reproduce features such as contrast differences between edge and screw dislocations and broadening of rocking curves with dislocation density (Wang et al., 2024).

Geometrical optics does not exhaust DFXM theory. For near-perfect crystals and sharp defects, dynamical diffraction and coherence effects can dominate. A wavefront-propagation framework based on numerical integration of the Takagi–Taupin equations has therefore been developed for DFXM image formation. In one validation on a diamond crystal containing a single stacking fault, the model reproduced the observed image width, Borrmann fringes, and contrast location, showing that wave-optical treatments are required when multiple scattering and coherence cannot be neglected (Carlsen et al., 2022).

Quantitative defect characterization has likewise moved beyond qualitative contrast interpretation. A theoretical analysis of Burgers-vector measurement revisited the transmission-electron-microscopy invisibility criterion in DFXM geometry and proposed the generalized condition

$\mathbf{g}\cdot\mathbf{b}_{\text{proj}} = 0 ,$

showing how scans about a single $\{hkl\}$ reflection can encode the Burgers vector of edge, screw, and mixed dislocations through symmetry-dependent contrast modulation (Pal et al., 2024). More generally, an inverse modeling formalism has recently been introduced to reconstruct the full deformation-gradient tensor from DFXM measurements on symmetry-equivalent noncoplanar reflections. In that framework, the recovered tensor is

$2d_{hkl}\sin\theta = n\lambda .$0

and sensitivity matrices are used to propagate motor-angle uncertainties into componentwise errors in strain and lattice rotation (Kanesalingam et al., 23 Jul 2025).

4. Acquisition workflows, image analysis, and automation

DFXM generates large, multiscale datasets whose interpretation depends on analysis pipelines as much as on optics. Wavelet-based multiresolution analysis has been proposed as a general method for DFXM images because it preserves spatial localization while resolving features across multiple length scales. In demonstrated applications, wavelet transforms were used to extract and track twin-boundary contrast, reconstruct localized image content over selected scale bands, and determine focus by maximizing energy in the smallest-scale detail coefficients (Abulshohoud et al., 2022).

For dislocation dynamics, semi-automated workflows have combined 2D stationary wavelet transforms, fast marching segmentation, Kalman filtering, and the Munkres assignment algorithm to isolate and track composite bright-dark dislocation objects over time. In aluminum timescans acquired at 98% of the melting temperature, the reported image sequences contained 500 frames per scan with $2d_{hkl}\sin\theta = n\lambda .$1 s, and the workflow produced quantitative trajectories, velocities, and orientations for statistically analyzing dislocation motion and interactions (Gonzalez et al., 2020).

A different analytical direction is Bayesian superresolution localization. A physically informed statistical model, combining a DFXM forward model and realistic detector noise, was used to estimate dislocation positions with sub-pixel uncertainty quantification. On synthetic realistic DFXM images of edge dislocations in single-crystal aluminum, the primary MAP estimator achieved a median error of 5.3 nm in low-noise conditions and 14.2 nm in high-noise conditions, well below the instrumental pixel size used in the simulations (Brennan et al., 2022).

Automation has recently expanded from post-processing to experimental decision-making. A lightweight convolutional neural network based on depthwise separable convolutions has been trained to classify $2d_{hkl}\sin\theta = n\lambda .$2 DFXM image patches as weak-beam or strong-beam. Using six hand-labeled images per class to generate 6144 patches per class, the reported LCNN reached 92.7% test accuracy with 879k parameters and 24 s training time, compared with 97.1% for ResNet18 and 89.1% for VGG16. The same study reports close agreement between LCNN-selected weak-beam frames and specialist curation while using approximately 90% less data (Benhadjira et al., 5 Sep 2025).

Automation also operates at the scale of instrument navigation. A multiscale workflow bridging 3DXRD, DCT, or LabDCT with DFXM translates grain orientation and position data directly into goniometer settings without dismounting or reorienting the sample. In an iron polycrystal containing 1100 grains, DFXM motor positions for all grains were calculated within seconds, enabling on-the-fly targeting and later high-resolution maps with 36 nm pixel size (Shukla et al., 25 Aug 2025).

5. Experimental regimes and scientific applications

DFXM has diversified beyond room-temperature, monochromatic synchrotron microscopy. A cryogenic implementation using a low-vibration closed-cycle helium cryostat extended the method to temperatures down to 3.2 K. In NaMnO$2d_{hkl}\sin\theta = n\lambda .$3, the experiment combined a Be-CRL objective, 38× total magnification, and a diffraction-limited sampling of approximately 170 nm per pixel; the final feature resolution was on the order of 1 $2d_{hkl}\sin\theta = n\lambda .$4m. Rocking-curve imaging mapped local mosaic spread with sensitivity to orientation changes as fine as 0.001°, and a temperature-dependent logistic fit of the $2d_{hkl}\sin\theta = n\lambda .$5 Bragg peak intensity showed a 20% step centered at 44 K, with some regions of interest exhibiting increases up to 55% and 25% (Plumb et al., 2022).

Photon-flux engineering has similarly expanded the range of accessible microstructures. Pink-beam DFXM at ESRF ID03 uses a broader bandwidth than conventional monochromatic mode and was reported to deliver a 27-fold increase in diffracted intensity while maintaining 100 nm spatial resolution. The method resolved subgrain structures in partially recrystallized aluminum, enabled full-field mapping of a highly deformed ferritic iron grain that was inaccessible in monochromatic mode without focusing optics, and supported in-situ tracking of grain growth during annealing with approximately 400 ms per frame and hundred-millisecond temporal resolution (Yildirim et al., 7 Mar 2025).

Time-resolved DFXM has been demonstrated with synchrotron pump-probe methods. In a germanium single crystal at APS, a 10 ns, 1064 nm laser pump was synchronized with hybrid-mode X-ray probe pulses, and the relaxation of laser-induced lattice deformation was modeled as

$2d_{hkl}\sin\theta = n\lambda .$6

with the characteristic thermal decay time reported as 12.94 ns within 1% error. The experiment also observed strain formation and propagation of an acoustic wave generated by laser-induced lattice deformation (Poudyal et al., 2022).

At XFELs, DFXM has entered an ultrafast regime. Simultaneous bright-field and dark-field microscopy has been implemented at PAL-XFEL and LCLS using pulses of up to $2d_{hkl}\sin\theta = n\lambda .$7 photons and durations of approximately 32–50 fs, enabling structural characterization down to 100 fs resolution and single-shot imaging of reversible or irreversible lattice dynamics (Dresselhaus-Marais et al., 2022). At the European XFEL, DFXM was used to visualize an optically driven longitudinal strain wave in diamond with measured spatial resolution of 0.74 $2d_{hkl}\sin\theta = n\lambda .$8m and temporal resolution below 100 fs, while introducing 3D structure scans and photon-energy-based axial-strain scans adapted to XFEL operation (Irvine et al., 2023). Related kinematic-diffraction theory has further extended DFXM formalism from coherent GHz phonons to incoherent phonons and thermal diffuse scattering, with explicit discussion of optimal sampling in real space, reciprocal space, and time (Chalise et al., 2024).

6. Performance limits, trade-offs, and current direction of the field

DFXM performance is governed by coupled trade-offs in flux, spatial resolution, reciprocal-space resolution, and interpretability. Pink-beam operation increases intensity and enables weakly diffracting or highly deformed samples, but it broadens the Bragg peak and reduces longitudinal $2d_{hkl}\sin\theta = n\lambda .$9-resolution, so precise elastic-strain mapping remains a strength of monochromatic DFXM (Yildirim et al., 7 Mar 2025). Likewise, higher numerical aperture improves spatial resolution and acquisition speed, but with MLL objectives the reciprocal-space resolution in the longitudinal and rolling directions becomes approximately three times broader than with CRLs under the reported conditions (Staeck et al., 22 Apr 2026).

Interpretation also depends on the scattering regime. Strong-beam conditions are dominated by multiple scattering and dynamical diffraction, whereas weak-beam conditions are designed to enhance defect contrast (Benhadjira et al., 5 Sep 2025). This is one source of a common misconception: DFXM contrast is not determined solely by local defect geometry, but by the combined action of deformation fields, instrument resolution, and the specific reciprocal-space slice selected by the optics and scan strategy. In near-perfect crystals, this means that wavefront-propagation and Takagi–Taupin calculations may be necessary; in more strongly deformed systems, geometrical-optics forward models offer a scalable approximation for experiment design and interpretation (Carlsen et al., 2022, Poulsen et al., 2020).

Another misconception is that DFXM is only a qualitative orientation-mapping tool. The recent development of Burgers-vector formalisms, full deformation-gradient reconstruction, and explicit sensitivity analysis shows that quantitative tensorial inversion is now a central research direction, although it requires carefully chosen reflections, accurate goniometer metrology, and control of numerical conditioning (Pal et al., 2024, Kanesalingam et al., 23 Jul 2025).

The current direction of the field is toward integration rather than specialization. High-energy diamond CRLs expand the accessible materials set; MLL objectives push spatial resolution below 60 nm; pink-beam and XFEL implementations extend temporal reach from hundred-millisecond to sub-100-fs scales; open-source multiscale workflows bridge grain mapping and DFXM without remounting; and machine-learning models automate weak-beam selection for high-throughput studies (Staeck et al., 25 May 2026, Shukla et al., 25 Aug 2025). A plausible implication is that DFXM is evolving from a specialized synchrotron microscope into a multiscale diffraction-imaging platform that couples instrument design, forward modeling, inverse reconstruction, and automated analysis within a single experimental workflow.