RuleStatics
Mesh band of the thrust range (Richardson, grid convergence index)
statics.tna.mesh-band
Our derivation from a construction or rule in the sources.
Statement
The same row is solved with 8, 16 and 32 courses per compartment. From the three values f₁, f₂, f₃ of an extreme and their changes ε₂ = f₂ − f₁, ε₃ = f₃ − f₂, the ratio R = ε₃/ε₂ gives the trend. If the changes shrink with one sign (0 < R < 1) the observed order is p = log₂(1/R) and Richardson’s extrapolation f₃ + ε₃/(2^p − 1) estimates the value for an infinitely fine pattern. When p lies between ½ and the formal order 2, the band is f₃ ± 1.25·|ε₃|/(2^p − 1) (Roache’s grid convergence index); otherwise, and for oscillating or growing changes, the conservative f₃ ± 3·|ε₃|. The finest proof’s gap widens the band on its open side. Where a finer proof leaves more open than the mesh changes, the values are not known well enough and no band is given.
Formula
Parameters
- course_counts
- 8, 16, 32
- safety_factor
- 1.25 in the asymptotic range (½ ≤ p ≤ 2), else 3 with p = 1
- formal_order
- 2
- max_nodes
- 2,000
- ib_32_courses
- greatest H : V 2.579, proved to 0.11 % (until wave 6: 2.338 found, 17 % open)
Modern note
An estimate, not a proof: Richardson extrapolation assumes the error falls as a power of the course spacing, and the refined patterns also have more force parameters, so their feasible sets change both ways. The studied sequences do not show the formal order (Ib’s greatest thrust converges faster, p ≈ 2.9; the others more slowly or oscillating), so their bands use the factor 3. Ib’s least thrust changes less from 16 to 32 courses than its 0.1 % proof leaves open, so it gets no band.
Sources
Where implemented
@ungewitter/statics#meshBand