[claude]

Adds docs/mgo_cube_standard.md for the SPI 01845-AB MgO cube, the Mg/O EDS standard. It records:

• the failed cleave attempt, with video
• SPI's own cleaving guidance and the likely causes of the failure
• the diamond saw → Bakelite → polish plan, with the water-free changes MgO needs in the byu-vcl polishing SOP

Related to #12.

Transferred from #12

@ronnie-guymon (original comment):

@claude create a PR for the MgO cube, noting that it did not cleave when I tried placing a razor on the {1 0 0} plane and hitting it with a hammer. I even tried sawing a little bit to get a notch cut, didn't help at all. I will probably just use the diamond saw to cut it and put it into bakelite to polish. Transfer this comment, and your response to the new PR. Here's the video of my attempt https://youtu.be/TNEI0Uvtbqg

Claude (original response):

Short answer: Recorded in docs/mgo_cube_standard.md with your video. The saw → Bakelite → polish route works for an EDS standard. The big change from the metal SOP is no water at any step. SPI's own note for this crystal says "For polishing or repolishing MgO, non-aqueous media should be used at all times". Otherwise the polished face hydrates (SPI's "white haze"), which is the surface film the cube was bought to avoid.

Why it probably didn't cleave

SPI says this crystal "is normally cleaved very easily" with a single-edge GEM razor blade placed parallel to an existing edge and a light hammer tap. Per SPI, "the only time there is a problem is when the razor blade being used gets dull", and 2 mm is about the thinnest slab you can get. Likely contributors:

1. Blade edge. MgO is Mohs 6, about as hard as the blade. Sawing the notch with steel dulls the edge and leaves a blunt groove, not a sharp crack starter.
2. Slab thickness. My earlier advice in Standards #12 was that a re-cleave "costs about 1 mm" and to mount a "1–2 mm slab". That is below SPI's 2 mm minimum. With the blade near an edge, the crack tends to curve out and chip. Splitting the cube in half is the most forgiving geometry.
3. Support. If the strike was on the textured rubber bench mat seen in the video, much of the tap went into the mat. SPI calls for a clean, lint-free work surface.

I couldn't watch the full video because YouTube blocks this runner, so I only saw its three auto-generated stills. Treat the list above as inference. If you want one more try before sawing: use a new GEM blade (not the one used for the notch), a hard bench with a lint-free wipe, the blade across the middle of a face, and one light tap.

Changes to the byu-vcl polishing SOP for MgO

The full step-by-step table is in the doc.

• Saw: optional. The cube already matches the SOP's 1 cm³ target size, so only saw if you want to keep a spare half.
• Mount: put an undamaged original face down as the analysis face, not the hammer-struck face or a saw face. It has far less damage to grind off.
• Grind and polish without water:
  • Skip the belt sander.
  • Grind from 600 grit with ethanol/IPA or a glycol lubricant (e.g., Allied GreenLube) instead of water.
  • Polish with diamond in a non-aqueous carrier down to 1 µm (0.25 µm if available).
  • Rinse with ethanol, not DI water.
• Pad: use a dedicated, labeled MgO pad. MgO carried onto an AlSi10Mg pad would add Mg to the unknowns.
• Final polish: skip the colloidal silica / vibratory step. It's aqueous and leaves Si behind.
• After polishing:
  • Dry it, store it in the desiccator or under vacuum, and carbon-coat soon. SPI says haze appears after about three days in air.
  • Bakelite and MgO are both insulators, so run carbon tape from the coated face to the holder.
  • Enter the coating in DTSA-II as planned.

🤖 Generated with Claude Code

[ronnie-guymon]

@claude what are the sources for the MgO cleaving? And what is considered "sharp"? Use Edison to confirm this will work the way it is supposed to, and find a video and SOP for cleaving the MgO cube.

[claude]

Claude finished @ronnie-guymon's task in 37m 5s —— View job

---

MgO cleaving: sources, "sharp", Edison check, video + SOP

Short answer: My earlier cleaving advice rested on one procedure source, SPI's own note, plus a Mohs value from webmineral. The rest was my own inference. Edison and a wider search partly confirm SPI's method:

• Blade-and-tap cleaving of MgO is documented lab practice.
• No source gives a success rate for a 10 mm cube.
• Two cleaver papers found that loading a single point works better than a straight blade on hard rock-salt oxides like MgO.

I found no video of MgO being cleaved and no formal SOP beyond SPI's note. Everything is now in docs/mgo_cube_standard.md.

Sources for the cleaving method

Source	What it says
SPI note (I read the full text)	GEM blade and small hammer, "tap lightly". The only problem it names is a dull blade, which "should be changed after every several cleavings". A 2 mm slab is the practical limit, and some chips are expected
Futagami & Akashi 1973 (full text read)	7 mm MgO blocks "cleaved into specimens of 1 mm thickness at room temperature by a chisel with a sharp knife edge", by "a blow on the chisel"
Schmid et al. 2006 (full text read)	NiO, which has the same rock-salt structure as MgO: "cleaving NiO with a razor blade that touches the sample along a line is rather difficult, while cleaving it with a wire cutter is simple"
Sander et al. 2022	A single-point tungsten carbide blade is "more suitable for cleaving hard crystals such as MgO ... the cleavage requires less force"
Yang et al. 2025 (Oak Ridge group)	SPI MgO "cleaved with a razor blade along the (100) surface immediately before reaction"

Corrections to my earlier answer:

• Slab thickness: I said a 1–2 mm slab was "below SPI's 2 mm minimum". That was overstated, because Futagami & Akashi took 1 mm slabs with a chisel. Splitting the cube in half is still the most forgiving geometry.
• Rubber mat: this is still only my inference. The literature recommends a thin pad, not a hard anvil: Edison cites Langdon & Pask (1968) cleaving MgO "on a soft pad of tissue paper".

What counts as "sharp"

There is no standard definition for cleaving. Measured edge sizes:

Edge	Size	Source
New razor blade, SEM edge width	0.35–0.45 µm	Verhoeven 2004
New utility blades, including Olfa	< 200 nm	Science of Sharp
Unused scalpel	about 1 µm radius; > 5 µm is "unrealistically blunt"	McCarthy et al. 2010
Notch that behaves like a crack in ceramics	root below about 10–20 µm	Kübler 2000, NASA/TM-2006-214090
Notch cut with a diamond disk / diamond wire	250 / 70–80 µm	Palacios et al. (measured in tungsten)

In practice:

• Use a new blade and check it. Take a blade straight from the pack and hold it edge-on under a lamp. A sharp edge looks like "an almost invisibly smooth black line"; any glints are "dull, bent or chipped areas" (Peachey 2016).
• Never reuse a blade on MgO. This cube is harder than the blade (SPI lists Vickers 910; razor steel is about 620–850 HV). Any blade that has struck or sawed MgO is dull for this purpose.
• A sawed notch doesn't start a crack. Its root is far too blunt unless you sharpen it, for example with a razor blade and diamond paste.

Edison results

I ran two LITERATURE_HIGH queries.

• Procedure: the method is "generally reliable" for a 10 mm cube, because MgO's {100} cleavage toughness is low (0.81 MPa·m½). It is not guaranteed: expect chips, cleavage steps, slip under a blunt edge, and subgrain effects. Edison recommends a fresh blade, a split near the middle, one light sharp tap, and a thin compliant pad.
• Sharpness: the edge radius is what matters. A steel edge can dull on MgO, so treat blades as consumables. Sawed notches are too blunt to act as cracks.
• Where I disagree: Edison says pre-scoring "generally helps", but its evidence is notches that were sharpened afterwards.

Video and SOP

• Video: I found none of MgO itself. I searched YouTube in English, Japanese, Chinese, and German, plus Vimeo, JoVE, and paper supplements. The closest matches:
  • The same blade-and-hammer method on rock salt (NaCl), which cleaves far more easily: Science Museum, Tokyo and ASNR. YouTube playback is blocked on this runner, so I checked these by title and thumbnails only.
  • Schmid et al. (2006) published movies of KBr and NiO cleaving with their cleaver (EPAPS E-RSINAK-77-209602). I haven't opened them.
• SOP: I found none for MgO cubes beyond SPI's note. The doc combines SPI's note and the Futagami & Akashi methods into a five-step retry list:
  1. A new blade, checked edge-on under a lamp.
  2. A hard bench with a lint-free wipe as the pad, not the rubber mat.
  3. The blade across the middle of an undamaged face, so the cube splits in half.
  4. One light, sharp tap.
  5. If that still fails, try point contact, such as a wire cutter. In the papers this worked on cross sections of a few mm, not 10 mm.

Question: what did you saw the notch with? The razor, a hacksaw, and the diamond saw leave very different notches.

Commands run

python - <<EOF … client.create_task({"name": JobNames.LITERATURE_HIGH, …}) … EOF   # 2 tasks submitted 21:38 UTC
python - <<EOF … get_task() every 60 s … EOF                                    # q1 success 22:02, q2 success 22:08
python - <<EOF … get_task(verbose=True); list_files(); fetch_data_from_storage() … EOF   # artifacts saved
git-push.sh origin HEAD                                                         # b64bc5e, 2a577d0, a87cd8d, 5d58172 pushed

Not done

• I couldn't play any YouTube video on this runner, including yours.
• I didn't open the Schmid movies, Langdon & Pask, or Preuss et al.; those are Edison citations only.
• Edison's two internal document caches (44 MB and 56 MB pickles) are listed in files.json but not committed. The answers, verbose trajectories, and tables are in outputs/edison_mgo_cleaving/. Say if you want the caches too.
  · Branch

[ronnie-guymon]

I got it to work! I'll post the video and explanation in #12

[ronnie-guymon]

@claude using edison, how much does charging effect EDS? We don't really care about how high quality the micrograph is, just how accurate the EDS results are. We will be working at 3.2nA, so a relatively high current for an insulating sample. The only problem I could see charging causing is not allowing as many electrons into the sample as with a conductive sample at a comparable beam current. This would just increase the needed acquisition time, but I don't see how this would affect the x-ray count accuracy at all. Is there something I am missing here? I'm asking this because I realized we might not want to coat our MgO standard since that would affect the spectra we are wanting to use as a standard.

[claude]

Claude finished @ronnie-guymon's task in 15m 53s —— View job

---

Charging vs. coating for the MgO EDS standard

Edison didn't run. Edison's login rejects the repo's EDISON_PLATFORM_API_KEY with 403 {"detail":"Forbidden"}. A made-up key gets the same response, and Edison's /health endpoint returns ok, so the key has been revoked or has expired. To fix it, a repo admin creates a new token (Edison: Account → Profile → API Tokens → Create New Token, per Edison's docs) and replaces the secret. Then comment @claude again and I'll run the Edison check. Everything below is from sources I read myself and is not Edison-checked.

Short answer: Yes, there's something missing. Charging doesn't mainly reduce how many electrons arrive. It reduces the energy they arrive with, and that changes the X-ray counts, not just the time it takes to collect them:

• The trapped charge sets up a negative surface potential that slows every incoming electron.
• Each electron then makes fewer Mg and O X-rays than it would at E₀, while DTSA-II still assumes E₀.
• The MgO standard reads low per nA·s, so Mg and O in the AlSi10Mg read high by the same ratio. A 20%-low standard gives results 25% high.
• A longer acquisition only makes that biased number more precise.

A thin carbon coat costs a few % of O Kα, the loss is predictable, and DTSA-II corrects for it. Charging errors are typically tens of %, drift over time, and DTSA-II can't correct them. Keep the coat in the plan.

What the longer-acquisition picture misses

1. Landing energy. In high vacuum, the bremsstrahlung cutoff (the Duane–Hunt limit, DHL) equals the electrons' energy at the surface: "the PE 'landing energy' = DHL" (Toth et al. 2002). On uncoated insulators, "charge accumulation on the sample surface produces a repulsive potential to the incident electron beam, decreasing its energy" (Galván Josa et al. 2013).

2. Measured size of the error. Galván Josa et al. analyzed an uncoated insulator of known composition (a milled ceramic sherd) at 15 kV and only 0.5 nA:

   • The DHL was 13.2 keV, so the surface potential was 1.8 kV.

   • Quantified at the nominal 15 keV, every major element came out 25–29% low:

     Element	Reference (wt%)	Quantified at 15 keV (nominal)	Quantified at 13.2 keV (DHL)
     O	47.57	35.1	46.17
     Si	29.07	21.77	28.92
     Al	10.98	7.85	10.28

   • Across their other uncoated samples, the DHL ranged from 12.3 to 14.8 keV.

3. Rule of thumb. Line intensity goes roughly as (E₀ − E_c)^1.7 (U. Oregon EPMA lab, ch. 5). At 15 kV, that is about 1% fewer Mg Kα and O Kα counts per electron for every 100 V of surface potential. My estimate below covers generated intensity only:

   Surface potential	Mg Kα and O Kα per electron, 15 kV	20 kV
   0.1 kV	−1%	−1%
   0.3 kV	−3 to −4%	−3%
   1 kV	−11 to −12%	−9%
   1.8 kV	−20 to −21%	−15 to −16%
   3 kV	−33 to −34%	−25 to −26%

4. Even "fewer electrons" would bias the result. DTSA-II normalizes each spectrum by dose = live time × probe current (getDose), and the probe current comes from the Faraday cup, not from what lands on the sample. Fewer counts per nA·s on the standard is a bias, not a slowdown.

5. Normalizing to 100% won't hide it. Mg (a few tenths of a wt%) and O are minor in AlSi10Mg, so the error lands almost entirely on them.

6. Stray X-rays and drift.

   • "When insulating materials were electrically charged, the obtained spectra contained characteristic X-rays … of materials other than the samples such as the sample stage and the collimator in the X-ray detector. This is because electrons from the electron beam were decelerated by and deflected from the insulating materials" (Imashuku et al. 2013).
   • "The dynamic nature of charging can introduce time dependent effects" (Newbury 2002), so a charged standard won't be reproducible between sessions.

7. DTSA-II can't correct it. It stops with "The beam energy for the standard and unknown must match." (source). So you can't enter the DHL as the standard's beam energy, and a drifting potential has no single value to enter anyway.

What the carbon coat costs

Calculated with xraylib attenuation coefficients for amorphous C at 1.9 g/cm³ and a 35° take-off angle:

C thickness	O Kα lost	Mg Kα lost	Electron energy lost at 15 keV
5 nm	2.0%	0.2%	13 eV
10 nm	3.9%	0.4%	27 eV
20 nm	7.6%	0.8%	54 eV

• Cross-checks:
  • NIST's NeXL coating example: about 17 nm of C on SiO₂ (15 kV, 40°) raised the O k-ratio 5.8% after correction (0.992 → 1.050) and Si 0.2%. My calculation gives 5.8% for 17 nm at 40°.
  • Kerrick et al. 1973: "For light elements, X-ray intensity errors of up to 4 percent are possible with film thickness differences of 200 A between sample and standard."
• Energy loss is negligible. The 27 eV lost in 10 nm of C acts like 27 V of charging, about 0.3%.
• DTSA-II corrects the coat. Set Conductive Coating (material, thickness in nm) in the MgO spectrum's properties (panel).
  • DTSA-II then multiplies the standard's expected intensity by the coat's X-ray transmission (trStd) and handles the unknown's coat separately.
  • It models X-ray absorption only, which is the dominant term.
  • With a 10 nm coat known to ±3 nm, the leftover O Kα error is about ±1%.

Recommendation

1. Coat the MgO with 10–20 nm of C and enter it in DTSA-II. Kerrick et al. had conduction problems below about 20 nm on their microprobe.
   • Put a polished brass witness in the coater. Interference colors from Kerrick et al.: 150 Å orange, 200 Å indigo-red, 250 Å blue.
   • Measuring the C Kα k-ratio also works (NeXL).
2. More robust: coat an AlSi10Mg unknown in the same run and enter the same coat on both. The coat then mostly cancels, and its exact thickness matters much less. Both NeXL and Kerrick describe this as the usual practice.
3. To test it on your SEM, take spectra on the uncoated MgO and on a conductor (the Al alloy or Cu tape) at your kV and 3.2 nA, and compare Duane–Hunt limits. DTSA-II's quant report prints D-H = … keV (source).
   • Take a few back-to-back spectra to catch drift.
   • A DHL more than about 0.1 keV below the conductor's means a bias of more than about 1%.
   • Galván Josa saw 1.8 kV at 0.5 nA. I expect much more than 0.1 keV for a bulk crystal in Bakelite at 3.2 nA, but that's inference; the measurement settles it.
   • Run the same check after coating, to confirm the coat is thick enough.

Low vacuum and ionic-liquid coatings aren't good substitutes for an O standard. Both put O in the beam path: air or water vapor in the chamber, or the acetate in EMI-CH₃COO. In low vacuum, the gas ions also shift the landing energy (Toth et al. 2002).

Not done

• Edison: key rejected (see top).
• No file changes. If you want, I can add a short "why coat" note and the DHL check to the "After polishing" list in docs/mgo_cube_standard.md.

Commands run

pip install edison-client                                    # 0.18.0 installed
python - <<EOF … client.create_task({"name": JobNames.LITERATURE_HIGH, …}) … EOF
                                                             # Exception: Failed to authenticate (403 on /auth/login)
POST /auth/login with the repo key → 403 {"detail":"Forbidden"}   # made-up key → same 403; GET /health → 200 "ok"
pip install xraylib; python - <<EOF … xraylib.CS_Total(6, E) … EOF     # coat, energy-loss and landing-energy tables
gh api repos/usnistgov/EPQ/contents/… | grep -n -i "coating\|duane\|dose"   # DTSA-II source checks

· Branch