Spectral Triples and the Index Pairing

2 The Fredholm index pairing

This chapter develops the index attached to an even spectral triple. The graded-kernel index (Section 2.2) — the special case of the index pairing at the unit projection \(p = 1\) — is formalized, building on a minimal notion of Fredholm linear map (Section 2.1). The general index pairing against an arbitrary \(K_0\)-class \([p]\) (Section 2.3) is not yet formalized; it corresponds to Phase 2 of the project plan.

2.1 Fredholm operators

This section records a minimal, self-contained notion of Fredholm linear map, scoped to exactly what the index pairing needs. Mathlib has ongoing, as yet unmerged, work towards a general theory of Fredholm operators on topological vector spaces; the definitions here are deliberately narrower — stated for plain linear maps rather than continuous linear maps, so that they apply uniformly to bounded operators and to the restriction of an unbounded operator to its domain, as needed for the chiral Dirac operator \(D^{+}\) — and are expected to be replaced by an import once the upstream theory lands.

Definition 13 Fredholm linear map
#

A linear map \(f \colon E \to F\) is Fredholm if \(\ker f\) is finite-dimensional, the range of \(f\) is closed, and the range of \(f\) has finite codimension (the cokernel \(F / \operatorname {ran} f\) is finite-dimensional).

Definition 14 Fredholm index
#

The Fredholm index of a linear map \(f \colon E \to F\) is the integer

\[ \operatorname {ind}(f) = \dim (\ker f) - \dim (F / \operatorname {ran} f). \]

A bijective linear map with closed range is Fredholm, of index \(0\).

2.1.1 Compact perturbations of the identity

This subsection proves the structural part of the Hilbert-space case of the classical Riesz–Schauder theorem: \(1 - K\) is Fredholm whenever \(K\) is compact (finite kernel, closed range, finite-dimensional cokernel). The further classical fact that the index is \(0\) is not proved here. None of this is yet in Mathlib (only the weaker spectral dichotomy for compact operators is), so it is built here from scratch, as the analytic ingredient needed to eventually show the chiral Dirac operator \(D^{+}\) is Fredholm.

Theorem 16 Compact operators are approximable by finite rank

A compact operator \(K\) on a Hilbert space is approximable, in operator norm, by finite-rank operators: for every \(\varepsilon {\gt} 0\) there is a finite-rank \(F\) with \(\| K - F\| \leq \varepsilon \). The approximant is built by composing \(K\) with the orthogonal projection onto the span of a finite \(\varepsilon \)-net of the (relatively compact) image of the closed unit ball.

Lemma 17 Finite-rank operators have finite-rank adjoints

If a continuous linear map on an inner product space has finite-dimensional range, so does its adjoint.

Theorem 18 The adjoint of a compact operator is compact
#

If \(K\) is a compact operator on a Hilbert space, then its adjoint \(K^{\dagger }\) is also compact: the adjoints of finite-rank approximants of \(K\) (Theorem 16) are themselves finite-rank (Lemma 17), hence compact, and converge to \(K^{\dagger }\) in norm since the adjoint operation is norm-preserving.

Theorem 19 Compact perturbations of the identity are Fredholm

The structural part of the classical Riesz–Schauder theorem: if \(K\) is compact on a Hilbert space, then \(1 - K\) is Fredholm. The kernel is finite-dimensional (it is the eigenspace of \(K\) at the eigenvalue \(1\)); the range is closed, via a bounded-below estimate on the orthogonal complement of the kernel established by a contradiction argument using compactness of \(K\); and the cokernel is finite-dimensional, via the adjoint identity \((1-K)^{\dagger } = 1 - K^{\dagger }\) (Theorem 18) together with \((\operatorname {ran}(1-K))^{\perp } = \ker ((1-K)^{\dagger })\). (The further classical fact that the index is \(0\) is not proved here.)

2.2 The graded-kernel index

Definition 20 Kernel of the Dirac operator
#

For a (partially defined) linear operator \(D \colon \operatorname {dom}(D) \to \mathcal{H}\), the kernel \(\ker D \subseteq \mathcal{H}\) is the subspace of vectors \(\xi \in \operatorname {dom}(D)\) with \(D\xi = 0\).

Lemma 21 The grading preserves the kernel

If \((A, \mathcal{H}, \pi , D, \gamma )\) is an even spectral triple and \(\xi \in \ker D\), then \(\gamma \xi \in \ker D\). Consequently \(\ker D\) splits as \(\ker D = (\ker D)^{+} \oplus (\ker D)^{-}\) into the \((\pm 1)\)-eigenspaces of \(\gamma \) restricted to \(\ker D\).

Definition 22 Graded-kernel index
#

The (super)index of an even Dirac datum \((D, \gamma )\) is the integer

\[ \operatorname {index}(D, \gamma ) \; =\; \dim (\ker D)^{+} - \dim (\ker D)^{-}. \]

This is the Fredholm index of the chiral operator \(D^{+}\) from Definition 24 in the special case \(p = 1\). The corresponding invariant of an even spectral triple is .

Theorem 23 Finite-dimensionality of the kernel

If \(D\) is self-adjoint with compact resolvent at \(i\), then \(\ker D\) is finite-dimensional. Consequently \(\operatorname {index}(D, \gamma )\) (Definition 22) is a difference of genuine finite dimensions: for \(\xi \in \ker D\), the resolvent \(R(i, D)\) satisfies \(R(i,D)(i\xi ) = \xi \), so \(\ker D\) embeds into the \(i^{-1}\)-eigenspace of the compact operator \(R(i, D)\), which is finite-dimensional by Riesz theory.

2.3 The general index pairing

This section is not yet formalized; it corresponds to Phase 2 of the project plan. It generalizes Section 2.2 from the unit projection \(p = 1\) to an arbitrary self-adjoint idempotent representing a class in \(K_0(A)\).

Definition 24 Index pairing

Let \((A, \mathcal{H}, \pi , D, \gamma )\) be an even spectral triple and let \(p \in M_n(A)\) be a self-adjoint idempotent (a projection representing a class in \(K_0(A)\)). The index pairing of \(p\) with \((A, \mathcal{H}, \pi , D, \gamma )\) is the Fredholm index

\[ \langle [p], (\mathcal{H}, \pi , D, \gamma ) \rangle \; =\; \operatorname {ind}\big( \pi (p) D^{+} \pi (p) \big), \]

where \(D^{+}\) denotes the restriction of \(D\) to the \((+1)\)-eigenspace of \(\gamma \).

Theorem 25 The index pairing is well defined

In the situation of Definition 24, the operator \(\pi (p) D^{+} \pi (p)\), restricted to \(\pi (p)\mathcal{H}\cap \operatorname {dom}(D)\), is Fredholm, so that \(\langle [p], (\mathcal{H}, \pi , D, \gamma ) \rangle \in \mathbb {Z}\) is well defined.