Theorem. The finite surface character formula [fdip-006W]

For the surface data and coefficients, \[Z_\Sigma (f)=\sum _{\rho \in \widehat G} d_\rho ^{\,2-2g}\left (\frac {c_\rho }{d_\rho }\right )^P.\] The formula uses normalized counting on every edge. In particular, \(f\equiv 1\) gives \(Z_\Sigma (1)=1\); the unnormalized sum is \(|G|^{|E|}\).

The exceptional CW description of a sphere with one vertex, no edges and one face attached by a constant map lies outside the nonempty polygon convention. Its partition sum is directly \(f(1)\). The same expression still holds, since \(g=0\), \(P=1\) and \(f(1)=\sum _\rho d_\rho c_\rho \).

For \(G=\{1,-1\}\) and \(f(h)=e^{-\beta (1-h)}\), the two coefficients are \(c_+=e^{-\beta }\cosh \beta \) and \(c_-=e^{-\beta }\sinh \beta \). On the torus the normalized sum is \(c_+^P+c_-^P\). Multiplication by \(2^{|E|}\) recovers the exact square-torus partition function, since \(|E|=2P\). For a disk of \(A\) faces, insertion of its boundary Wilson sign multiplies the face weight by the sign on each interior face. This swaps \(c_+\) and \(c_-\) there. The same edge contraction then gives numerator \[c_-^A c_+^{P-A}+c_+^A c_-^{P-A}.\] Division by the partition sum recovers \((t^A+t^{P-A})/(1+t^P)\), with \(t=\tanh \beta \), under either counting convention.

For character convolution with normalized group integration, see equation (9) in [migdal1975recursion, p. 414]. Witten's Section 2.3 uses normalized Haar measure and character contractions in equations (2.44), (2.48), (2.50)–(2.51) [witten1991quantum, pp. 163–166]. His continuum construction uses heat-kernel face weights with area-dependent Casimir factors. The proof here is for a finite group and an arbitrary fixed central face weight. It does not assert subdivision invariance for the Wilson weight or transfer continuum conclusions to these finite models.