Chern–Simons action and level quantization

Work on an oriented closed three-dimensional spacetime with path-integral weight exponential of imaginary action. Let a be a compact U(1) connection with curvature periods two pi times integers and integer Wilson charges. Choose a minus sign in front of the level-k action so the positive oriented linking convention below gives a positive braiding phase. Locally, a gauge variation is a total derivative, but nontrivial bundles prevent treating the potential as one global smooth form.

One way to define the exponent uses an extension to a four-manifold whose oriented boundary is the physical spacetime, with the connection extended as well. Comparing two extensions gives an integral over a closed four-manifold. That difference must change the action only by a multiple of two pi. This test detects global information that a local integration by parts can miss, and it specifies which values of the nominal coefficient define a quantum theory.

On a general oriented closed four-manifold, the square of the first Chern class can be an odd integer. The extension ambiguity is therefore pi times k times that integer, with the chosen overall sign. To define the ordinary bosonic U(1) theory without choosing spin data on arbitrary oriented manifolds, k must be even. On spin four-manifolds this characteristic number is even, so integer k is sufficient for the corresponding spin theory.

Odd level is thus not universally inconsistent; it requires the appropriate spin-dependent definition and line interpretation. Conversely, saying only that the level is integer silently omits information if a bosonic nonspin theory was intended. More general gauge groups have their own normalization and global-form restrictions. Quantum regularization can also introduce framing dependence and a gravitational phase, so metric independence of the classical density does not mean every quantum partition function is an unframed topological number.

For a nonzero even level, Wilson charges are identified modulo the absolute level in the ordinary bosonic topological theory. Fusion adds the labels. Two distinct oriented worldlines with linking number one acquire a mutual braiding phase proportional to the product of their charges. Our positive source insertion gives $da=(2\pi q/k)\delta_C$, fixing the sign seen by the other line. A self-twist additionally requires a framing of a line; its phase is the topological spin. The sign convention chosen for the action fixes both phases consistently.

These phases are exact within the topological sector. A pure abelian theory has no local propagating bulk photon, while a Maxwell completion can contain massive non-topological excitations outside its infrared regime. Boundaries require appropriate edge degrees of freedom or other boundary data to handle gauge variation. The torus ground-state space has dimension equal to the absolute level in this simple theory, but that does not imply every microscopic Hamiltonian with the same local action has precisely this finite-size spectrum at all energies.