The periodic Poisson pairing in Γ and χ¶
Paper 1 writes the neutral four-centre integral in two ways (labels
eq:eri4, eq:eriG). Paper 2 contracts the same reciprocal interaction in
a transfer sector. This tranche identifies their real-space meaning for the
unfitted periodic Green operator. The fitting and tail bounds in
the Coulomb error derivation compare approximations
to this operator.
Conventions and derivation¶
Take lattice columns A, positive volume Ω=|det A|, reciprocal columns B=2π A⁻ᵀ, and reduced transfer t. Write x=Aξ and p=B(n+t). Divide each quasiperiodic pair by its common phase exp(i Bt·x) before Fourier analysis on the fractional unit torus. Its normalized Haar measure is dξ=dx/Ω. If f denotes the resulting envelope, the physical coefficient is Ω times its Haar coefficient f̂(n). Use the same twist in both slots, reversing the second orbital pair where the implementation takes its conjugate. Set a(n)=|p|². The Coulomb symbol w(n) is 4π/a(n) except that a physical zero has weight zero. An integer label zero in a nonzero transfer sector normally survives.
Gap and existence (
poisson_dual_gap,poisson_mesh_gap,poisson_weight_bound). For a commensurate mesh with repetitions N, write n+t=h/N componentwise after relabelling, h an integer vector. The scaled dual row Dᵢ=Nᵢ Aᵢᵀ/(2π) maps p to hᵢ. If h≠0, some |hᵢ|≥1, so |p|≥1/L where L bounds the three dual row norms. No orthogonal nearest-image rule is involved. A gap δ in squared momentum bounds the multiplier by 4π/δ. Multiplying the square-summable Fourier coefficients of any L² source g by this symbol leaves them in ℓ². The inverse Fourier isometry therefore defines a potential u.Equation and gauge (
poisson_coeff,poisson_spec_iff,poisson_exists_unique). For a(n)≠0 the character equation is a(n) û(n)=4π ĝ(n). At a zero set û(n)=0, projecting out the source constant rather than prescribing the unsolvable equation 0=4π ĝ(0). The solution is unique because its coefficients determine its L² class. There is no arbitrary constant in a nontrivial shifted sector with no physical zero. This neutral operator does not supply the exchange seam.Pairing (
poisson_physical_pairing,poisson_absolute_pairing). Parseval pairs f with u: Ω ∫ f conj(u) dξ equals the sum of (w/Ω)(Ω f̂) conj(Ω ĝ). The sum is absolutely convergent, including for complex pairs. In atomic units p has dimensions of inverse length; using pair densities of inverse-volume dimension gives a Coulomb energy.Real-space kernel (
poisson_finite_integral,poisson_cont_coeff,poisson_green_limit). For continuous envelopes and any finite index set S, put K_S(ξ,η)=Σ w(n)exp(−2πin·ξ)exp(+2πin·η). Integrating first against conj(g(η)), then against f(ξ), gives Σ w f̂ conj(ĝ). This conjugation convention uses the test density in the first slot. At Γ the full symmetric spectrum also permits the opposite sign convention. In physical coordinates the Green kernel carries 1/Ω; the two volume elements restore a net Ω. The directed finite-set limit equals the potential pairing. No pointwise sum at ξ=η or exchange of a singular pointwise limit with an integral is needed.
The kernel-checked statements are in formal/rhf/VibeQCAICCM/PeriodicPoisson.lean.
They use the pinned mathlib Fourier isometry on the unit additive torus.
The correspondence between an AO basis and a continuous periodic envelope,
and the independent equality of a particular Ewald real-space evaluator to
this weak Green operator, are external to these statements. This proof does
not give convergence rates for truncated higher-angular-momentum AO images.
Implementation check¶
In the standalone archive run
python formal/check_aiccm_poisson.py poisson-input.json replay.json
with NumPy 2.2.6. The input records the lattice, centred functions, primitive
coefficients, integer labels and image list. To regenerate it, run
formal/capture_aiccm_poisson.py through the qualified engine. It calls
the classic reciprocal enumerator and physical fit weights, the C++ Bloch
pair transform, and the AO scaling used by the fit. For a pure p shell the
raw transform needs a factor √(4π/3); the s partner has unit scale. This
is a boundary between conventions, not a change to the production method.
The fixture combines a contracted p shell and a displaced s shell in a skew cell, using a deliberately asymmetric relative-image list. The probe is a specified finite trigonometric polynomial; its physical coefficients include the cell volume. The independent oracle evaluates the probe potential in real space on tensor Gauss-Hermite nodes and integrates all three polynomial p components after Gaussian product completion. The common lattice sum is unfolded to all space; both routes have the same four relative images and the witness makes no assertion about the discarded images. It checks the real Γ and complex shifted pairings, including errors from dropping conjugation, volume, the nonzero central transfer, or the AO normalization. The classic enumerator rounds reduced transfers to twelve decimal places; comparison on both nominal and stored momenta separates that change from AO quadrature error. This is a numerical check on known inputs, not a proof of the C++ or floating-point routines. A molecular total energy would not supply this twisted, finite-mode matrix element.
For the 3D Ewald representation of this weak operator, see the Ewald derivation.