Holstein → CRETIN
Eight layers of radiation trapping. From a single atomic resonance to a CRETIN-class 3D NLTE solver. Then a LightCell 819 nm Na extraction calculator falls out.
You can read Molisch & Oehry's 443-page Radiation Trapping in Atomic Vapours (Oxford 1998) end to end. Most people don't. The mathematics is not the obstacle — it's that nobody has ever made it manipulable. Bret Victor would tear into this book in a paragraph. Howard Scott already did the production-scale version (CRETIN, JQSRT 71:689, 2001) but the binary is export-controlled and the code is hard to teach with.
So we built it — interactively, in eight layers, with the constants from Molisch & Oehry Appendix D and the architectural shape of CRETIN. Every parameter is a slider. Every equation has a worked numerical case. Each layer compounds: Layer N reuses the objects from Layers 1..N-1.
The terminal layer is the engineering case study that motivated this whole effort: LightCell Energy's sodium-vapour 819 nm extraction problem. You'll be able to drag bath temperature, optical depths, cavity-Q, and recuperator efficiency and watch the design surface live.
The eight layers
Open questions surfaced along the way
- Na(3d) molecular quenching coefficient. M&O §3.3 gives a 4-decade range (10⁻¹⁴ to 10⁻¹⁸ cm²) for molecular quench cross sections. The Na(3d) + {N₂, O₂, CO₂, CO, H₂, H₂O} matrix is missing from the open literature. Confirming LTE in the 3p/3d manifold under combustion-product conditions requires one experimental measurement.
- BEC density limit revisited. Sesko et al. 1991 used seven simplifications that modern MOTs violate. Bezuglov-Molisch-Fuso-Allegrini-Ekers (PRA 77, 063414, 2008) predicts subnatural decay; no clean experimental confirmation in the open literature as of 2026.
- Generalization of the Hg-196 isotope-doping trick. Works for Hg. Cs is mono-isotopic — must use Zeeman splitting (Sommerer 1993). Na has no stable rare isotope to dope. What about ⁴⁰K at 0.012 % natural abundance? Optimum predicted ~1–3 % but unmeasured.
- RadJAX authorship. Gemini deep-research cited a 10⁴× speedup over RADMC-3D. Authorship unverified at time of writing.
Team:
Feynman (theorist-lead) ·
Bret Victor (UX-lead) ·
Alan Kay (architecture) ·
Howard Scott (CRETIN computational) ·
Andreas Molisch + Bernhard Oehry (canon) ·
N.N. Bezuglov (Russian analytical extension) ·
Holstein / Biberman / Milne (historical anchors)
Source material:
M&O 1998 (OCR'd via Gemini 3.1 Pro — see the bench essay) ·
6 CRETIN PDFs (Scott 2001, Sequoia-Tillack-Scott 2006, Langer-Scott-Marinak-Landen 2002, Scott 2011/2017, fusion-plasma Ch II) ·
3 deep-research runs (Claude, Gemini, ChatGPT — cross-referenced for trust)
Built by an LLM and a human in a few sessions, ~2026-05-26.