Cluster sizing from an interaction radius: the mesh-ceiling rule¶
AICCM/CCM remains experimental. VibeQCAICCM/MeshCeiling.lean adds seven
kernel-checked claims for Paper 1 eq:kspacing and the sizing rule stated
beneath it. The cluster is sized from a requested interaction radius Rc
and the interplanar spacing d of each primitive axis as
N = ceil(2 Rc / d).
Lengths are in bohr and wave vectors in inverse bohr; Lean uses real magnitudes.
Claim |
Statement |
|---|---|
|
|
|
|
|
Any count |
|
|
|
Equality holds exactly when |
|
A vector whose inner product with a nonzero |
|
If `2 Rc <= 1/ |
eq:kspacing is therefore an upper bound on the built mesh spacing, as the
manuscript says, and an equality only without rounding. In
mesh_vector_length the vector g is the dual basis vector of one supercell
axis in the convention g . A_j = delta_ij (no factor of two pi), so 1/|g|
is the interplanar spacing. Applying mesh_shortest_vector to every axis
gives the shortest-vector bound and hence an inscribed Wigner-Seitz radius of
at least Rc.
The implementation is the sizing loop in
python/vibeqc/periodic/ccm/wigner_seitz.py, which evaluates
max(1, ceil(2 Rc / d - 1e-9)). The subtracted rounding slack means its
count is one below meshCount when 2 Rc / d exceeds an integer by less
than 1e-9. In that case the covering and spacing bounds hold only up to a
relative 1e-9, not exactly.
Not covered: the correspondence between the parser’s lattice and these spacings, the selection of nearest images, and any statement that a mesh of this spacing converges a property.
Check: python formal/check_proofs.py with the pinned Lake (685 claims).