A database-wide atlas of single-crystal filters for thermal neutron beams

Single-crystal filters that pass thermal neutrons and remove epithermal and fast neutrons and γ rays have been chosen from a handful of materials. For 18 449 inorganic crystals of the Materials Project, the thickness L₀.₈ at which a large single crystal oriented away from Bragg reflections transmits 80% of a 300 K Maxwellian beam over 0.5–3 Å was computed, together with its rejection −ln T of a 1/E neutron spectrum (1 eV–100 keV), of a fission spectrum and of 1 and 6 MeV photons at that thickness; 3 437 crystals have density-functional and 1 279 machine-learned phonons. A composition-only proxy reproduces the bulk order of L₀.₈ (Spearman ρ = 0.83 at 293 K, 0.95 at 77 K) but not the top of the ranking, which a single elastic Debye temperature largely recovers. Against measured total cross-sections the acceptance criterion fixed in advance was not met (6 of 13 crystals fail); restricted to single-crystal data, 4 of 11 fail, or 6 of 11 without an allowance for the spread between datasets; and the model lies above all four sapphire datasets. With a model uncertainty of 0.138 in ln L₀.₈ propagated by 20 000 Monte Carlo draws, the Pareto front of commercially grown crystals (epithermal against 1 MeV photon rejection) consists of known filters alone with probability 0.10 at 293 K and 0.04 at 77 K, and cooling reorders it. No grown crystal that is not a known filter beats sapphire with probability above 0.90, the level fixed for a statement. 6H-SiC and chrysoberyl are candidates on the epithermal axis: the probability that 6H-SiC rejects the 1/E spectrum better than sapphire is 0.85 at 293 K and 0.88 at 77 K (chrysoberyl: 0.66 and 0.85); with an error component shared by all compounds it rises to 0.93 and 0.95, and with sapphire corrected by its measured residual it falls to 0.69 and 0.73. On the fission spectrum neither candidate is distinguishable from sapphire. Counting every transmitted particle instead of the uncollided ones changes the epithermal rejection by −0.23 to −6.28 and reorders the front, so uncollided figures of merit rank filters optimistically for uncollimated beams. The pipeline that produces the atlas is openly available; the atlas and its uncertainties will be archived openly on acceptance.