EBSD parameters

[ronnie-guymon]

@claude using Edison Scientific, explain how parameters are decided for EBSD. I don't want an answer of "consult existing standards", I want to understand how the parameters for those standards were chosen.

[ronnie-guymon]

@claude using Edison Scientific, explain how parameters are decided for EBSD. I don't want an answer of "consult existing standards", I want to understand how the parameters for those standards were chosen.

[claude]

Claude finished @ronnie-guymon's task in 5m 35s —— View job

---

EBSD Parameter Selection

Edison Scientific status — blocked (empty API key). I installed the client and attempted to submit the task, but authentication fails because the EDISON_PLATFORM_API_KEY secret is present-but-empty in this environment.

What I ran and observed:

• pip install edison-client → succeeded (edison-client 0.16.0)
• EdisonClient(api_key=EDISON_PLATFORM_API_KEY).create_task(...) → failed
  • Root cause: HTTPStatusError: Client error '422 unknown' for url 'https://api.platform.edisonscientific.com/auth/login'
  • Diagnosis: len(os.environ['EDISON_PLATFORM_API_KEY']) == 0 (the var exists, value is an empty string). The login payload sends an empty api_key, which the server rejects with 422.
• Checked for alternate variable names (EDISON*, PLATFORM*, FUTUREHOUSE*): only EDISON_PLATFORM_API_KEY exists, and it is empty.

To unblock Edison: add a non-empty EDISON_PLATFORM_API_KEY to the repo/org Actions secrets and confirm the workflow maps it into the job's env:. Re-comment @claude and I'll submit the same query to the LITERATURE_HIGH (paperqa3-high) job, which returns a fully citation-grounded answer, and commit the trajectory artifacts.

Since I can't reach Edison, below is a first-principles explanation from established EBSD physics and metrology (not Edison-sourced). Where it names a standard, it explains why the value was chosen, not just that the standard says so.

---

How EBSD parameters are actually decided (the reasoning behind the standards)

1. Specimen tilt ≈ 70°. The backscatter coefficient η rises steeply with tilt, and — more importantly — the fraction of backscattered electrons that retain coherent diffraction information (the Kikuchi signal, versus diffuse background) peaks in roughly the 60–75° range. Tilting shortens the escape depth normal to the surface, so the emerging electrons sample a thin near-surface layer where their exit trajectories still carry channeling/diffraction contrast. Above ~70° the pattern contrast keeps improving marginally, but the projected beam footprint elongates (spatial resolution becomes strongly anisotropic), geometric foreshortening/distortion grows, and shadowing worsens. 70° is the empirical optimum balancing Kikuchi-band contrast against spatial resolution and geometric distortion — that is why the standards fixed on it, not the reverse.

2. Accelerating voltage (typ. 10–30 kV, 20 kV default). Two competing physics chains set this:

• Signal/pattern sharpness: electron wavelength λ ∝ 1/√V, and Kikuchi band width scales as ~2λ/d (Bragg angle). Higher kV → shorter λ → thinner, sharper bands and higher BSE yield → better statistics and, if the detector can resolve the thinner bands, better angular precision.
• Spatial resolution: higher kV enlarges the interaction/excitation volume (deeper, wider), degrading resolution and blurring measurements near boundaries.

So 20 kV is a general-purpose compromise; you drop to ~10 kV for nanoscale features, thin films, or beam-sensitive/low-Z materials (smaller interaction volume), and push to 30 kV for weakly scattering light-element phases that need more signal. Low-Z elements have low BSE yield → need higher kV/current; high-Z phases scatter strongly → tolerate lower kV.

3. Beam/probe current. Pattern SNR ∝ √N (electrons per pattern), so more current gives cleaner patterns and shorter exposure (faster maps). The cost is probe size: for a given column, higher current means a larger probe → worse spatial resolution (less severe on FEG than thermionic sources). You raise current for speed/SNR on coarse microstructures and lower it (small probe) to resolve fine grains and boundaries.

4. Step size — the "≥ ~10 points per grain" rule. This comes from sampling statistics, not convention. If a grain is covered by N measurement points, the fractional error in its reconstructed area/diameter from pixel discretization falls roughly as ~1/√N, and boundary/grain-count reconstruction becomes unreliable when the smallest grains contain only a handful of points. Choosing step ≤ (smallest grain of interest)/10 puts ≥ ~10 points across a grain, bringing pixelation error down to a few percent and letting a grain-reconstruction algorithm (misorientation-threshold flood fill, typ. 5°) distinguish real grains from noise. ASTM E2627 codifies exactly this logic (minimum points-per-grain and a misorientation threshold for defining a grain) so that grain-size measurements are reproducible. Lower bound on step size is physical: nothing is gained by stepping below the effective spatial resolution (~20–50 nm for FEG at 20 kV) or below what intragranular-gradient (GND) sensitivity requires.

5. Working distance, detector distance, and pattern centre. WD (~15–20 mm) is set so the tilted region of interest projects onto the phosphor with correct geometry. Detector insertion distance trades angular capture against gnomonic magnification: a closer detector captures a wider slice of the diffraction sphere (more bands → more over-determined, more robust indexing) but lower magnification per band. The pattern centre must be calibrated because indexing solves orientation from band positions relative to the PC — a PC error maps almost linearly into an orientation error, which is why calibration (known-orientation standard, moving-screen, or iterative refinement) is mandatory rather than optional.

6. Camera binning, exposure, gain. Detector pixel count sets the ceiling on angular resolution: band-position precision improves with more pixels across the pattern, so orientation precision does too — but at lower per-pixel SNR and slower readout. Binning (4×4, 8×8) sums pixels for higher SNR and frame rate at the cost of angular resolution: heavy binning for fast mapping, low/no binning for high-precision or HR-EBSD (resolving <0.1° misorientations). Exposure/frame-averaging improves SNR as ~√(frames), sharpening Hough peaks and indexing reliability; you set it to the minimum that reliably meets your indexing-quality target (CI/MAD) at the desired speed. Gain amplifies signal and noise, so it doesn't substitute for real exposure.

7. Hough transform parameters. More detected bands over-determine the orientation fit → higher precision and fewer misindexes, with diminishing returns and slower indexing (typ. 7–12 bands). Finer Hough (θ, ρ) binning improves band-position precision (→ better angular resolution) but is noisier and slower; the butterfly-mask size is tuned to the expected band width, which itself depends on kV and detector distance. These are chosen to sit at the knee of the speed / indexing-success / precision trade-off for the material.

8. Minimum grains / scan area for representative statistics. The counts in standards are confidence-interval arithmetic. The standard error of the mean grain size ∝ σ/√(n_grains), so to bound the mean to ±X% at 95% confidence you need n ≥ (1.96·σ / (X·mean))² grains — a few hundred to a few thousand depending on size spread. Texture (ODF) representativeness is stronger still (typically thousands of grains for a stable ODF, more for weak/random textures). The required scan area then follows from (grains needed) × (mean grain area), subject to the step-size constraint — which is what ultimately sets map size and acquisition time.

9. CI / MAD / band-contrast thresholds for clean-up. These cutoffs were set empirically by correlating each metric against independently confirmed correct indexing. Confidence Index (TSL) = (V₁−V₂)/V_ideal from the band-voting scheme; CI > 0.1 is the conventional "reliable" line because, across many materials, >99% of points above it index correctly. MAD (mean angular deviation, Oxford) is the mean misfit between detected and simulated bands; MAD < ~1° marks a good fit. Band-contrast/image-quality thresholds separate poorly diffracting points (boundaries, deformed regions, unindexed phases). Clean-up routines (neighbour-orientation correlation, grain dilation) use these thresholds to drop/replace unreliable points — and standards require reporting the clean-up because over-aggressive filtering biases grain-size and texture results.

The through-line: every EBSD parameter is a point on an explicit trade-off curve — pattern signal/contrast vs. spatial resolution vs. angular precision vs. acquisition speed vs. statistical representativeness — where the physics (backscatter yield, interaction volume, λ ∝ 1/√V, SNR ∝ √N, gnomonic geometry) fixes the shape of the curve and sampling/confidence statistics fix how far along it you need to go. The standards simply record the settled optima of those curves for common materials.

Verify the exact numeric clauses of ASTM E2627 / ISO 13067 against the current published editions before using them for compliance; the values above are the widely-used practitioner optima and their derivations. Re-run with Edison once the API key is set for a fully citation-backed version.

[ronnie-guymon]

@claude create a PR for this issue

[claude]

Claude finished @ronnie-guymon's task in 1m 54s —— View job • claude/issue-2-20260724-1601 • Create PR ➔

---

PR created for issue No. 2

I captured the EBSD parameter-selection reasoning as project documentation and pushed it to claude/issue-2-20260724-1601.

• Read repo context (README, CLAUDE.md, structure)
• Capture the EBSD parameter-selection explanation as a documentation file
• Commit and push to branch
• Provide PR creation link

What changed: added docs/ebsd-parameter-selection.md (152 lines), a first-principles walkthrough of why each EBSD parameter takes its standard value — tilt ≈ 70°, accelerating voltage, beam current, step size (≥ ~10 points/grain), working/detector distance and pattern centre, binning/exposure/gain, Hough parameters, minimum grains for representative statistics, and CI/MAD/band-contrast clean-up thresholds. Each section explains the physics or sampling statistics that fixed the value, rather than deferring to ASTM E2627 / ISO 13067.

Edison status — still blocked. EDISON_PLATFORM_API_KEY remains present-but-empty in this environment, so I could not produce a citation-grounded version via the LITERATURE_HIGH job. The doc notes this and the path to re-run once a non-empty key is added to Actions secrets. This is flagged inline in the document so it's tracked with the content.

· Branch