Torus

Torus

The ground-state degeneracy of the Levin-Wen model on a torus, for every category it was run on, with its checks and timings. Its stages follow this description; the checks are defined at the bottom of the page.

The Levin-Wen string-net model on a 2×22\times2 honeycomb lattice on the torus, with a vertex projector AvA_v on every vertex and a plaquette projector BpB_p on every hexagon,

H=−∑vAv−∑pBp,Bp=1dim⁡(C)∑sds Bps.H = -\sum_{v} A_v - \sum_{p} B_p,\qquad B_p = \frac{1}{\dim(\mathcal{C})}\sum_{s} d_s\, B_p^{s} .

The vertex projectors are imposed exactly by restricting to the physical basis of fusion-allowed edge labellings; the plaquette operators are eq. 20 of PhysRevB.102.115154, with block inverses F−1F^{-1} at the inverse moves. Every ground state has energy E0=−(∣V∣+∣P∣)E_0 = -(|V| + |P|), and the dimension of the ground space is the number of simple objects of the center:

GSD⁡(T2)=rank⁡(Z(C)).\operatorname{GSD}(T^2) = \operatorname{rank}(\mathcal{Z}(\mathcal{C})) .

The BpB_p are commuting projectors, so their product P=∏pBpP = \prod_p B_p projects onto the joint Bp=1B_p = 1 eigenspace, which is the ground space, and the degeneracy is its trace,GSD⁡=Tr⁡P\operatorname{GSD} = \operatorname{Tr}P. Each BpB_p has between 6 and a few hundred nonzeros per column, so PP is never formed: the trace is a sum over columns of three sparse matrix-vector products against a row of B4B_4, and the cost is linear in the basis dimension. This works for unitary and non-unitary input alike, and it is the same for both.

InputMethodDelivers
any categorythe trace of the sparse commuting projector P=∏pBpP = \prod_p B_p; P2=PP^2 = P is checked on a random vector and the trace must be an integer.rank
unitary, physical basis of at most 3000 statesas a cross-check, HH is also assembled densely and its lowest eigenspace counted; the count must equal the trace, and this is where E0=−(∣V∣+∣P∣)E_0 = -(|V|+|P|) is read off directly.agrees with the trace
the four B_p would not fit in 16 GB, or the basis exceeds 2 million statesskipped, not estimated, with the reason recorded (fill is sampled on 2000 columns before anything is built).nothing
non-unitary, ill-conditioned at 16 figuresthe projector check fails (Tr⁡P\operatorname{Tr}P off an integer, P2≠PP^2 \neq P) and the category is refused rather than rounded.not computed
blocks, dimensions, fusion, braiding of Z(C)\mathcal{Z}(\mathcal{C})a trace on one torus carries none of this; it is a count.not a torus quantity
Stages
StageBuildsChecks
1physical basis: the fusion-allowed edge labellings of the 2×22\times2 honeycomb torus, so every AvA_v is imposed exactlyits size: above 2 million states the category is skipped, with the reason recorded
2plaquette operators BpB_p, eq. 20 of PhysRevB.102.115154, sparse, with block inverses F−1F^{-1} at the inverse movestheir fill, sampled on 2000 columns before anything is built: if the four BpB_p would not fit in 16 GB the category is skipped
3GSD⁡=Tr⁡P\operatorname{GSD} = \operatorname{Tr}P, P=∏pBpP = \prod_p B_p, as a sum over columns of three sparse matrix-vector products against a row of B4B_4∥P2v−Pv∥/∥Pv∥\|P^2 v - P v\| / \|P v\| on a random vector and the distance of Tr⁡P\operatorname{Tr}P from an integer, both within 10−610^{-6}, or the category is refused
4for unitary input with a basis of at most 3000 states, HH assembled densely and its lowest eigenspace countedthe count equals the trace; E0=−(∣V∣+∣P∣)E_0 = -(|V|+|P|)

Loading the index…