Two neighboring pixels can differ by half a degree because a crystal is bending, because the detector geometry drifts across the scan, or because orientation noise is being differentiated over one step. Two grains can share the same minimum rotation angle while rotating about entirely different axes and meeting on different interface planes. Misorientation analysis becomes useful when it identifies exactly which orientations are compared, applies crystal symmetry consistently, retains the rotation axis and reference frame, and connects angular statistics to spatial scale and independent physical evidence.

Misorientation is a relative rotation, while disorientation is a symmetry-reduced representative. An orientation maps a crystal frame into a specimen frame. Combining two orientations eliminates the common specimen frame and produces a rotation from one lattice frame to the other, with multiplication order determined by the adopted convention. Crystal symmetry generates many mathematically different rotations that describe the same physical relationship. The disorientation is commonly chosen as the proper-symmetry equivalent with the smallest rotation angle. Software often reports that minimum by default and calls it “misorientation angle,” so terminology and conventions must be declared.

If $g_1$ and $g_2$ map crystal coordinates into a common specimen frame, one relative-rotation convention is $\Delta g=g_1^{-1}g_2$. A symmetry-reduced disorientation angle may then be written

for proper crystal-symmetry operations in $\mathcal{G}$. Other valid conventions reverse multiplication order or map the second crystal into the first, changing the reported axis frame or sign while preserving physical equivalence when handled consistently. Same-phase grain boundaries also have grain-exchange symmetry: a boundary has no preferred first side, so a rotation and its inverse describe the same unordered relationship.

The comparison pair defines the statistic before any color map is drawn. Grain-to-grain analysis may compare reconstructed mean orientations, orientations immediately adjacent to each boundary segment, or selected interior reference points. Those choices differ when grains contain gradients or boundary patterns are mixed. Point-to-point line scans measure increments between successive positions, while point-to-origin scans accumulate deviation from a fixed position. The same orientation field can therefore yield small incremental angles and a large end-to-end rotation.

Kernel average misorientation compares one site with a specified set of neighbors. For valid neighbors $j$ and nonnegative weights $w_{ij}$, a generic form is

where $K_i$ depends on grid topology, neighbor order, grain or phase masking, and often an exclusion threshold that removes angles interpreted as boundaries. Changing any of those settings changes the measured length scale and value. A square first-neighbor kernel, a hexagonal first shell, and a multi-shell physical-radius kernel are not equivalent. The number of accepted neighbors should be retained because edge, pore, and unindexed sites otherwise appear deceptively comparable to interior sites.

GROD compares each orientation with a reference assigned to its grain. The reference may be the symmetry-aware mean, a selected undeformed pixel, the grain center, a low-KAM point, or a pre-deformation state. Each answers a different question. The mean can move as deformation becomes heterogeneous; a chosen point may contain strain or indexing error; pre/post comparison adds registration and remapping uncertainty. Grain orientation spread and grain average misorientation collapse a spatial field to one scalar and cannot show whether rotation is smooth, localized, or split into subgrains.

Define whether the question concerns boundaries, twins, deformation, transformation, or precision
  -> Choose the orientation pairs, phase combinations, reference state, and spatial scale
  -> Establish crystal symmetry, specimen frame, rotation convention, and grain-exchange rule
  -> Acquire raw diffraction patterns with calibrated geometry and representative sampling
  -> Estimate angular precision, systematic drift, and spatial response on standards
  -> Index phases and orientations while preserving alternatives and unindexed sites
  -> Reconstruct grains with declared thresholds and compare sensitivity cases
  -> Compute full relative rotations before reducing to angle-only summaries
  -> Calculate boundary, KAM, GROD, GOS, line-profile, or MDF products as appropriate
  -> Report neighbor kernels, reference choices, cutoffs, weights, and valid counts
  -> Separate crystal-frame axes from specimen-frame rotation axes
  -> Compare distributions with texture-aware and adjacency-aware baselines
  -> Test step-size, noise, cleanup, and segmentation dependence
  -> Correlate with TEM, HR-EBSD, mechanics, processing, or device behavior
  -> Archive raw maps, patterns, scripts, conventions, and uncertainty

Rotation axis and angle carry different uncertainty and frame dependence. A proper rotation can be represented by a unit quaternion $q=(q_0,q_1,q_2,q_3)$. After symmetry reduction and a sign convention, its angle is

and the vector part gives the axis when $\sin(\theta/2)$ is sufficiently separated from zero. As $\theta$ approaches zero, the axis becomes ill-conditioned: small orientation errors produce large axis-direction changes even when the angle remains fairly precise. An attractive low-angle axis map may therefore display noise direction more strongly than physical rotation axes.

The axis can be expressed in the crystal frame of either grain or in the specimen frame. Crystal-frame axes are useful for crystallographic relationships, slip, twins, and transformation variants; specimen-frame axes reveal rotation relative to loading, growth, current, or device directions. Converting between them requires the associated orientation, which may be lost if only a reduced misorientation object is exported. Axis pole figures must state frame, symmetry, grain order, antipodal treatment, and any minimum-angle filter.

Angular precision, spatial resolution, and scan geometry set the floor for local metrics. Conventional Hough indexing, dictionary or template matching, and cross-correlation produce different orientation precision. Pattern center error, detector distortion, beam position, stage motion, surface relief, charging, pattern binning, signal-to-noise, pseudosymmetry, and phase competition can create systematic or random apparent rotations. Precision from repeated measurements on a stable single crystal does not prove absolute orientation accuracy, but it reveals a noise floor and spatial correlation.

Map-wide projection-center variation can create smooth phantom gradients. Scan-line noise can create directional KAM bands. Drift and charging can turn time into apparent position-dependent rotation. Calibration should be tested across the map, not only at its center. Repeating a standard in both scan directions, rotating the raster, acquiring fast frames, and comparing neighboring-distance statistics can separate material curvature from instrument structure.

Step size is part of the metric. For a fixed physical gradient, orientation difference between immediate neighbors tends to shrink with smaller step, while orientation noise may not. KAM can therefore rise, fall, or reach a noise plateau as the step changes. Interaction volume and probe size create spatial averaging that the nominal step does not remove. Cross-study comparison requires matched or explicitly normalized spatial scale, angular method, neighbor distance, and signal quality.

Cleanup directly modifies derivatives. Wild-spike removal can suppress noise, but neighbor filling and smoothing can manufacture continuous gradients, erase subgrain boundaries, or spread one orientation across a real interface. KAM, GOS, GROD, boundary fractions, and GND estimates should be computed on raw and controlled derivative maps. Unindexed pixels should remain visible because they may identify high deformation, boundary overlap, a second phase, surface damage, or loss of pattern quality.

Misorientation distributions require the correct random and textured baseline. The Mackenzie distribution describes the symmetry-reduced disorientation angle distribution for independent randomly oriented cubic crystals. It is not a universal random curve for all point groups, and it is not the expected neighbor distribution for a textured material. Texture changes the probability that two randomly drawn orientations have a given relationship; processing can also create correlations between adjacent grains beyond the one-point orientation distribution.

A misorientation distribution function is a probability density over full relative-rotation space, not only an angle histogram. For a phase pair with normalized density $M(\Delta g)$,

over the symmetry-reduced misorientation domain $\mathcal{F}_{\Delta}$. Boundary-segment weighting estimates trace- or area-related populations; one vote per grain pair estimates a boundary-count population; random pairs drawn from the ODF define a texture-only reference. These are different estimands. A measured excess over the texture-only baseline can reveal neighbor correlation, but it still needs uncertainty and network context.

CSL and twin classification adds a distance from an ideal rotation and a tolerance. Near-$\Sigma3$, for example, is a misorientation statement, not proof of coherent boundary plane, low energy, electrical benefit, or mechanical behavior. Transformation variants likewise require phase-specific orientation relationships and treatment of parent-product symmetry. Reporting only the nearest named relationship forces every rotation into a class; maximum-distance or outlier rules must preserve unmatched data.

An MDF does not contain boundary positions, plane normals, connectivity, or triple junctions. Two networks can share the same angle-axis distribution while having different percolation paths. Grain-boundary engineering and reliability studies need the spatial graph and, where properties require it, five-parameter boundary character and chemistry in addition to misorientation statistics.

KAM, GROD, and orientation gradients are deformation proxies rather than direct strain meters. Plastic deformation can produce lattice curvature and substructure, so local misorientation often correlates with dislocation content or accumulated deformation under controlled conditions. Elastic strain changes lattice spacings and pattern geometry but is not generally equal to a finite orientation difference. KAM is dimensionless angular contrast; labeling it “strain” or percent deformation without calibration and a material model is incorrect.

A dimensional scaling sometimes used to interpret a simple orientation gradient is

where $\theta$ is a small rotation across distance $L$ and $b$ is an applicable Burgers-vector magnitude. This is an order-of-magnitude relation, not a complete inversion. A surface orientation map provides only some lattice-curvature components; noise is amplified by differentiation; multiple dislocation types can produce the same measurable curvature; and statistically stored dislocations may not contribute to net curvature. Full GND estimates require coordinate-consistent derivatives, slip or Burgers-vector choices, regularization, boundary handling, and uncertainty.

HR-EBSD measures relative pattern shifts with much higher sensitivity than conventional orientation indexing when patterns share phase and sufficiently similar orientation. It can separate elastic strain and lattice rotation under a calibrated projection and reference model, yet the reference pattern may itself be strained. Cross-grain comparison, large rotations, remapping, pattern-center error, and surface relaxation require special treatment. High precision does not eliminate reference-state uncertainty.

Physical interpretation needs correlative validation and representative sampling. In semiconductor manufacturing, misorientation analysis can distinguish epitaxial variants in GaN, SiC, and oxide films; track mosaicity and tilt boundaries; map rotation near vias, bonds, cracks, or stressed interconnects; identify twin-related populations in copper and solder; characterize recrystallization after anneal; and examine orientation gradients around electromigration or packaging failures. The relevant axes should be tied to wafer normal, device line, current, interface, or loading direction rather than only the screen frame.

A process conclusion needs multiple fields, dies, wafer positions, process splits, and lots at the level claimed. Pixels within one grain and boundary segments along one interface are correlated observations. Resampling and confidence intervals should use grains, boundaries, fields, specimens, or wafers as appropriate. Rare twin or variant classes need adequate independent counts, while targeted failure sites should be reported separately from unbiased population sampling.

TEM diffraction or imaging can validate twins, dislocations, and subgrain structures; HR-EBSD can test small rotations and strain; digital image correlation supplies mechanical strain; EDS or EELS constrains phase and chemistry; XRD evaluates wider-area mosaicity and texture; and in-situ loading or annealing tests temporal sequence. Registration error and foil-preparation relaxation must be included when comparing fields across methods.

A reproducible deliverable preserves raw patterns and orientations, phase symmetry, specimen axes, rotation order, symmetry reduction, grain-exchange convention, axis frame, acquisition precision, spatial response, step, neighbor kernel, exclusion cutoff, grain reconstruction, reference orientation, weights, cleanup, random baseline, software, and scripts. It separates a relative rotation from its minimum-angle representative, an orientation proxy from strain, and a distribution from a connected network. Read misorientation analysis through the symmetry-reference-neighborhood-scale-uncertainty-and-mechanism lens.