When gravity became a gauge theory

There are discoveries that add a new ingredient to a theory. There are others that reveal that the ingredients already on the table can be assembled into something you recognize. The Chern–Simons description of three-dimensional gravity belongs to the second kind.

Start with the familiar geometric language, but unpack the metric. A frame field \(e\) tells you how local measuring rods fit into spacetime. A spin connection \(\omega\) tells you how local frames rotate as you move. In first-order gravity, these are the variables you work with.

For Lorentzian AdS₃ gravity, put them together like this:

Two combinations, with just a sign between them. Suddenly, the variables look like two gauge connections.

The equation that changes the conversation

With a standard normalization of the Lie-algebra trace, the bulk Einstein action becomes a difference of Chern–Simons actions:

Appropriate boundary terms still have to be supplied. The relevant Lie algebra is \(\mathfrak{sl}(2,\mathbb R)\oplus\mathfrak{sl}(2,\mathbb R)\). The two flatness equations encode constant curvature and vanishing torsion in the invertible-frame sector.

Achúcarro and Townsend gave the AdS supergravity construction in 1986.^[1] Witten’s 1988 work developed the relation into a broad program for understanding and quantizing 2+1-dimensional gravity.^[2] These are related landmarks, not interchangeable credits for a single paper.

Why is the rewriting exciting? Because it changes which questions are convenient to ask. A connection invites you to study parallel transport around loops. Flatness says small contractible loops are locally uninteresting, while noncontractible loops can retain holonomy. The global information that survived the disappearance of gravitational waves now has a natural mathematical language.

A beautiful shortcut with a border crossing

It would be wonderful if the next sentence were simply: “Chern–Simons theory is understood, so quantum gravity is solved.” But that sentence silently crosses a border.

The identity between classical actions does not automatically identify the domains of the quantum path integrals. A pair of gauge connections can produce a degenerate frame through their difference. Such a frame does not define an ordinary invertible spacetime metric. Do we include it? If not, is our restriction preserved by the transformations we intend to quotient out?

There are other choices too. A Lie algebra does not fix the global gauge group. Allowed bundles, large transformations, boundary conditions, and the integration contour all affect the quantum problem. Writing the same local expression in two sets of variables does not answer those questions for us.

Imagine two maps with the same street pattern. Before using one to navigate the other, you still need to know whether the roads connect across the edge of the map, whether a bridge is open, and which apparent intersections are actually overpasses. Local agreement is a powerful start. Global instructions complete the route.

A clue that keeps returning

The lasting gift of the CS reformulation is a language: connections, holonomies, moduli spaces, boundary states, and gluing. Later approaches can use that language while being more selective about the quantum construction.

The Virasoro TQFT proposal we will reach at the end explicitly distinguishes itself from ordinary \(SL(2,\mathbb R)\) Chern–Simons theory.^[3] That distinction is part of the inheritance of the early observation, not a reason to forget it.

For the next chapter, keep one lesson close: a locally simple connection can carry important global information. In 1992, a particularly dramatic piece of global information entered the story. It had an event horizon.