Compositional certificates for graph rotations #
Tensor and zig-zag rotations are straight-line programs calling their constituent rotations. Their certificates charge the complete vertex and edge-label encodings, including copying and intermediate pairs. The affine-plane rotation similarly composes multiplication and subtraction.
These rules certify a rotation call. They do not enumerate a graph or certify an unbounded recursive tower. In particular, a certificate for a fixed finite base is not a uniform efficiency theorem for a varying field size or tower depth.
Main results #
Turing.MultiTapeTM.ComputableTidilyInPolyTime.rotationMap_tensor: tensor rotations.Turing.MultiTapeTM.ComputableTidilyInPolyTime.rotationMap_zigzag: zig-zag rotations.Turing.MultiTapeTM.ComputableTidilyInPolyTime.rotationMap_affinePlane: arithmetic rotations.Turing.MultiTapeTM.ComputableTidilyInPolyTime.rotationMap_contract: contraction through efficiently encoded and decoded slots, including the unused-slot branch.Turing.MultiTapeTM.ComputableTidilyInPolyTime.rotationMap_reindex: change the vertex encoding using efficient forward and inverse maps.
References #
A tensor rotation makes two certified calls and rearranges their complete encoded results.
A zig-zag rotation makes two small-graph calls and one large-graph call, retaining labels.
Field multiplication and subtraction suffice to certify the affine-plane rotation.
Contraction charges slot decoding, the occupied-slot rotation and re-encoding, and the unused-slot fallback, with all complete vertex and label encodings.
Vertex renaming charges the forward and backward maps and the complete rotation.