Canonical: https://knowlpedia.com/differential-geometry/atiyah-guillemin-sternberg-convexity-theorem/

# Atiyah–Guillemin–Sternberg convexity theorem

The moment-map image of a compact connected Hamiltonian torus space is the convex hull of its fixed-point images, and its fibers are connected.

Let a compact torus \(T\) act through a
[Hamiltonian torus action](https://knowlpedia.com/differential-geometry/hamiltonian-torus-action/)
on a compact connected [symplectic manifold](https://knowlpedia.com/differential-geometry/symplectic-manifold/) \((M,\omega)\), with
[moment map](https://knowlpedia.com/fiber-bundles/moment-map/)
\(\mu:M\to\mathfrak t^*\). The
**Atiyah–Guillemin–Sternberg convexity theorem** states that:

1. every fiber \(\mu^{-1}(\alpha)\) is connected; and
2. the image is the [convex hull](https://knowlpedia.com/convex-analysis/convex-hull/) of the fixed-point image,
\[
\mu(M)=\operatorname{conv}\bigl(\mu(M^T)\bigr).
\]

Consequently, \(\mu(M)\) is a convex polytope. The moment map is constant on
each [connected component](https://knowlpedia.com/topology/connected-component/) of \(M^T\), so the convex hull is generated by
finitely many values. Neither effectiveness of the action nor isolatedness
of the fixed points is required.

## Proof idea

For each \(\xi\in\mathfrak t\), the component
\[
\mu^\xi=\langle\mu,\xi\rangle
\]
is a Morse–Bott function whose critical set is the fixed set of the subtorus generated by \(\xi\). Its critical submanifolds have even index and coindex. Morse-theoretic analysis of these components gives connected level sets and controls extrema in every direction. Separating-hyperplane arguments then identify the full image with the convex hull of fixed-point values.

Guillemin and Sternberg obtain the convexity statement through closely related Hamiltonian and symplectic arguments.

## Consequences

Every linear functional on \(\mathfrak t^*\) attains its maximum and minimum on \(\mu(M)\) at images of torus-fixed components. Thus the fixed-point data determine the entire moment image. For an effective \(T^n\)-action on a \(2n\)-manifold, the resulting polytope is the starting point of Delzant’s classification of compact symplectic toric manifolds.

The image \(\mu(M)\), together with its integral-affine structure, is called
the [moment polytope](https://knowlpedia.com/differential-geometry/moment-polytope/) of the
Hamiltonian torus space.

Connectedness of the fibers is stronger than convexity of the image. It helps identify reduced spaces as parameters vary and rules out disconnected inverse images that a generic map onto the same polytope could have.

## Scope and near misses

Compactness cannot simply be dropped: noncompact [Hamiltonian actions](https://knowlpedia.com/differential-geometry/hamiltonian-lie-group-action/) can have nonclosed or nonconvex moment images unless the moment map satisfies appropriate properness hypotheses. Connectedness of \(M\) is also needed for the single-polytope conclusion.

For a nonabelian compact group, the raw moment image in \(\mathfrak g^*\) need not be convex. The corresponding theorem concerns its intersection with a positive Weyl chamber and requires a different formulation.

## References

1. M. F. Atiyah, “Convexity and Commuting Hamiltonians,” *Bulletin of the London Mathematical Society* 14 (1982), 1–15. [DOI record](https://doi.org/10.1112/blms/14.1.1). Relevant: §§2–3, connected level sets and convexity.
2. V. Guillemin and S. Sternberg, “Convexity Properties of the Moment Mapping,” *Inventiones Mathematicae* 67 (1982), 491–513. [DOI record](https://doi.org/10.1007/BF01402100). Relevant: the fixed-point convex-hull theorem for Hamiltonian torus actions.
