Index Theorems

⚡ Key Takeaways

Index Theorems equate an analytic index (a difference of solution-space dimensions for an elliptic operator) with a topological index (a characteristic-class integral over the manifold). The flagship result, the Atiyah–Singer Index Theorem (1963), states $\operatorname{ind}(D) = \int_M \operatorname{ch}(\sigma(D))\,\operatorname{Td}(TM\otimes\mathbb{C})$ . They apply to elliptic (pseudo)differential operators on compact manifolds, with vast reach in geometry, topology, and physics. Originated by Michael Atiyah and Isadore Singer; precursors due to Gauss, Riemann, Hirzebruch, and Gel'fand.

💡 The Core Insight

Two numbers that look completely different are forced to coincide. One — the analytic index — measures the failure of an elliptic operator to be invertible: how many independent solutions its kernel has, minus how many obstructions to solvability live in its cokernel. The other — the topological index — is a sum of integrals of curvature/characteristic data, depending only on the manifold's smooth topology and the operator's leading symbol. Because the analytic index is a homotopy invariant (stable under deformations), it can only depend on coarse data, and that coarse data turns out to be exactly the topological index. Symbolically: deforming the operator doesn't change the index, so the index must be topological.

DefinitionElliptic Operator and its Symbol

Let $E, F \to M$ be smooth complex vector bundles over a compact oriented Riemannian manifold $M$ of dimension $n$ , without boundary. A linear differential operator $D : \Gamma(E) \to \Gamma(F)$ of order $m$ has a principal symbol $\sigma_m(D)(x,\xi) \in \operatorname{Hom}(E_x, F_x)$ for $(x,\xi) \in T^*M$ , obtained by replacing each derivative $\partial/\partial x^j$ with $i\xi_j$ and keeping only top-order terms:

1

\quad \sigma_m(D)(x,\xi) = \sum_{|\alpha|=m} a_\alpha(x)\,(i\xi)^\alpha .

Derivation hint: write $D = \sum_{|\alpha|\le m} a_\alpha(x)\partial^\alpha$ ; the symbol is the Fourier-transform leading part, $e^{-i\langle x,\xi\rangle} D e^{i\langle x,\xi\rangle}$ truncated to degree $m$ . $D$ is elliptic if $\sigma_m(D)(x,\xi)$ is invertible for every $x \in M$ and every nonzero $\xi \in T_x^*M$ .

DefinitionAnalytic Index

For an elliptic operator $D$ on a compact manifold, $\ker D$ and $\operatorname{coker} D = (\operatorname{ran} D)^\perp \cong \ker D^*$ are finite-dimensional. The analytic index is

2

\quad \operatorname{ind}_a(D) = \dim_{\mathbb{C}} \ker D - \dim_{\mathbb{C}} \operatorname{coker} D = \dim \ker D - \dim \ker D^* .

Derivation hint: ellipticity gives elliptic regularity and the Fredholm property via parametrices — there exists $Q$ with $QD - I$ and $DQ - I$ compact (smoothing) operators, so $D$ is Fredholm and the index is well-defined and integer-valued.

TheoremAtiyah–Singer Index Theorem, 1963

Let $D : \Gamma(E) \to \Gamma(F)$ be an elliptic (pseudo)differential operator on a compact smooth manifold $M$ of even dimension $n$ . Then

3

\quad \operatorname{ind}_a(D) = (-1)^n \int_M \operatorname{ch}\!\big(\sigma(D)\big)\,\operatorname{Td}(TM \otimes \mathbb{C}),

where $\operatorname{ch}(\sigma(D)) \in H^*(M;\mathbb{Q})$ is the Chern character of the symbol class (an element of $K$ -theory $K(T^*M)$ pushed to $H^*(M)$ ) and $\operatorname{Td}$ is the Todd class of the complexified tangent bundle. The right side is the topological index $\operatorname{ind}_t(D)$ .

TheoremLocal Index Theorem — Dirac Operator Form

For the Atiyah–Singer Dirac operator $D = D^+ \oplus D^-$ on a spin manifold $M^n$ twisted by a vector bundle $W$ with connection,

4

\quad \operatorname{ind}(D^+) = \int_M \hat{A}(M)\,\operatorname{ch}(W), \qquad \hat A(M)=\prod_j \frac{x_j/2}{\sinh(x_j/2)} ,

where $\hat A(M)$ is the $\hat A$ -genus built from the Pontryagin roots $x_j$ of $TM$ and $\operatorname{ch}(W) = \operatorname{tr}\exp(F_W/2\pi i)$ is the Chern character of the twisting curvature $F_W$ . Derivation hint: specialize [3] to the symbol of the twisted Dirac operator; the Todd class combines with the Chern character of the spinor bundle to produce $\hat A(M)$ via the identity $\operatorname{Td}\cdot\operatorname{ch}(S^+ - S^-) = \hat A \cdot e^{(\cdots)}$ .

ProofStrategy: Cobordism / K-theory Proof of Atiyah–Singer

The original 1963 proof followed Hirzebruch's cobordism method; the cleaner 1968 proof uses $K$ -theory. The key steps:

Step 1 — Both indices are invariants of the symbol class in $K(T^*M)$ . The symbol of an elliptic operator defines a compactly-supported $K$ -theory class $[\sigma(D)] \in K(T^*M)$ . The analytic index depends only on this class (homotopy invariance + the fact that lower-order terms don't affect ellipticity). The topological index $\operatorname{ind}_t$ is a homomorphism $K(T^*M)\to\mathbb{Z}$ defined functorially.

Step 2 — Both are functorial under embeddings. Embed $M \hookrightarrow \mathbb{R}^N$ for large $N$ . There is a Thom isomorphism $K(T^*M) \xrightarrow{\sim} K(T^*\mathbb{R}^N)$ (restricted to a tubular neighborhood), and both indices are compatible with the pushforward $i_!$ induced by this embedding.

Step 3 — Reduce to a single point. After embedding, the problem reduces to computing the index map on $K(T^*\mathbb{R}^N) \cong K(\mathbb{R}^{2N})$ . By Bott periodicity, $K(\mathbb{R}^{2N}) \cong \mathbb{Z}$ , and the generator is the Bott class.

5

\quad K(T^*M) \xrightarrow{\ i_!\ } K(T^*\mathbb{R}^N) \xleftarrow[\sim]{\ j_!\ } K(\mathrm{pt})=\mathbb{Z}.

Derivation hint: $i_!$ is the Gysin/Thom pushforward in K-theory along the proper embedding; $j_!$ is the Bott class of the trivial bundle over $\mathbb{R}^{2N}$ ; Bott periodicity identifies the target with $\mathbb{Z}$ .

Step 4 — Normalization. Both the analytic index map and the topological index map are uniquely determined by three axioms (functoriality, normalization on a point, multiplicativity). On the point both give the same value, $1$ . Since both maps satisfy the same axioms and agree on the generator, they are equal:

6

\quad \operatorname{ind}_a = \operatorname{ind}_t : K(T^*M) \longrightarrow \mathbb{Z}.

Derivation hint: this is the uniqueness step — Atiyah–Singer's axiomatic characterization of index maps shows any natural transformation $K(T^*M)\to\mathbb{Z}$ satisfying the excision and multiplicativity properties and normalized on a point must be the topological index.

Step 5 — Convert to cohomology. Apply the Chern character $\operatorname{ch}: K(T^*M)\to H^*_c(T^*M;\mathbb{Q})$ and integrate, producing the Todd-class correction from the difference between the K-theoretic and cohomological Thom isomorphisms. This yields formula [3]. $\blacksquare$

∎

ProofAlternative: Heat Kernel / McKean–Singer

A purely analytic proof gives the local index theorem. For $D = D^+\oplus D^-$ with $\Delta^\pm = D^\mp D^\pm$ :

7

\quad \operatorname{ind}(D^+) = \operatorname{Tr}\big(e^{-t\Delta^+}\big) - \operatorname{Tr}\big(e^{-t\Delta^-}\big) \quad\text{for all } t>0.

Derivation hint: nonzero eigenvalues of $\Delta^+$ and $\Delta^-$ pair up bijectively via $D^+$ , so their heat-trace contributions cancel; only the kernels (eigenvalue $0$ ) survive, giving $\dim\ker D^+ - \dim\ker D^-$ independent of $t$ . Taking $t\to 0^+$ , the heat-kernel asymptotic expansion has coefficients that are local curvature invariants. The miraculous cancellation (proved cleanly by Getzler's rescaling / Mehler-formula argument, or Gilkey's invariant theory) leaves precisely the $\hat A$ -genus integrand, recovering [4]. $\blacksquare$

∎

mathematician · Michael Atiyah & Isadore Singer | 1929–2019 / 1924–2021

Atiyah, an Oxford/Cambridge geometer and 1966 Fields Medalist, and Singer, an MIT analyst, met in 1962 when Atiyah was puzzling over why the $\hat A$ -genus of a spin manifold is always an integer. Singer recognized this as the index of a Dirac operator, and within weeks they had conjectured the general theorem. Their collaboration spanned decades and four landmark papers (1968–1971). The index theorem won them the Abel Prize jointly in 2004. Atiyah's later work bridged the theorem to instantons, gauge theory, and physics; Singer's pushed into operator algebras and the eta invariant. Their partnership is often cited as the model for deep mathematical collaboration across the analysis–topology divide.

🧠 Intuition

Think of an elliptic operator as a system of equations. Solving $Du = 0$ counts symmetries/zero modes ( $\ker D$ ); solvability of $Du = f$ is obstructed by $\operatorname{coker} D$ . The index is the net count: solutions minus obstructions. Now imagine smoothly deforming the operator — eigenvalues drift, but an eigenvalue can only cross zero by pairing: a kernel mode in $\ker D$ disappears exactly when a cokernel obstruction also disappears (or both appear together). So $\dim\ker D$ and $\dim\operatorname{coker} D$ jump simultaneously and the difference stays fixed. This rigidity is why the index is topological. The Atiyah–Singer theorem then names the invariant: it's a curvature integral, the same kind of "total curvature = topology" statement as Gauss–Bonnet, but for an arbitrary elliptic operator.

Figure

Spectral flow: kernel and cokernel modes pair up as an elliptic operator is deformed, leaving the index invariant.

ExampleGauss–Bonnet via the Euler Operator

Given: A compact oriented surface $M$ of genus $g$ (e.g. a genus- $2$ surface), with the de Rham operator $D = d + d^*$ mapping even forms to odd forms.

Setup: The index of $D^+ : \Omega^{\text{even}}\to\Omega^{\text{odd}}$ is the alternating sum of Betti numbers:

\operatorname{ind}(D^+) = b_0 - b_1 + b_2 = \chi(M).

Topological side: Specializing [3], the integrand reduces to the Euler class $e(TM)$ , the Pfaffian of curvature. By the Chern–Gauss–Bonnet theorem,

\operatorname{ind}(D^+) = \int_M e(TM) = \frac{1}{2\pi}\int_M K\,dA.

Numbers: For genus $g=2$ , $\chi = 2 - 2g = 2 - 4 = -2$ . Therefore the total Gaussian curvature is

\frac{1}{2\pi}\int_M K\,dA = -2 \implies \int_M K\,dA = -4\pi.

Boxed answer: $\boxed{\operatorname{ind}(D^+)=\chi=-2,\quad \int_M K\,dA = -4\pi \text{(radians}\,\cdot\,\text{area, dimensionless total curvature)}}$

ExampleIndex of the Dirac Operator on a K3 Surface

Given: A K3 surface $M$ — a compact complex surface (real dimension $4$ ) that is spin, with $\hat A$ -genus needed for $\operatorname{ind}(D^+)$ of the untwisted Dirac operator ( $W$ trivial).

Topological data of K3: signature $\tau(M) = -16$ ; Euler characteristic $\chi = 24$ . In dimension $4$ , the $\hat A$ -genus integrates to

\int_M \hat A(M) = -\frac{1}{24}\,p_1[M], \qquad p_1[M] = 3\,\tau(M).

Compute Pontryagin number: $p_1[M] = 3\tau = 3(-16) = -48.$

Index:

\operatorname{ind}(D^+) = \int_M \hat A(M) = -\frac{1}{24}(-48) = 2.

Check: A K3 surface admits a Ricci-flat (hyperkähler) metric with holonomy $SU(2)$ , giving $2$ covariantly constant spinors — exactly $\dim\ker D^+ = 2$ , $\dim\ker D^- = 0$ .

Boxed answer: $\boxed{\operatorname{ind}(D^+)=2 \text{ (net number of harmonic spinors, dimensionless integer)}}$

⚠️ Common mistake

The index is not the dimension of the kernel. Many beginners conflate $\operatorname{ind}(D) = \dim\ker D - \dim\ker D^*$ with $\dim\ker D$ alone. Only the difference is topologically rigid; $\dim\ker D$ itself depends on the metric and can jump. The theorem computes the difference, not the individual dimensions.

The theorem requires a closed manifold. On manifolds with boundary, the formula [3] fails and must be corrected by a boundary term — the eta invariant — giving the Atiyah–Patodi–Singer (APS) theorem. Forgetting this is a classic error.

Ellipticity is essential. Hyperbolic or parabolic operators (e.g. the wave operator, the heat operator) are not Fredholm on closed manifolds; their "index" is not defined by [2]. The symbol must be invertible off the zero section.

The dimension parity matters. On odd-dimensional closed manifolds the index of any elliptic differential operator vanishes (by a symbol argument), so the interesting differential-operator content lives in even dimensions; odd dimensions carry spectral invariants (eta) instead.

Counterexample

Consider the operator $D = d/dx$ on the circle $S^1$ , acting on smooth complex functions $C^\infty(S^1)\to C^\infty(S^1)$ . Its kernel is the constants (dimension $1$ ) and its cokernel is also the constants (dimension $1$ , since $\int_{S^1} f = 0$ is the solvability condition). So $\operatorname{ind}(D) = 1 - 1 = 0$ . This illustrates the odd-dimensional vanishing: even though $D$ is elliptic on $S^1$ , the index is forced to be zero. One cannot extract topological information about $S^1$ from this index — confirming that the rich content of Atiyah–Singer lives in even dimensions, and that odd-dimensional refinement requires the eta invariant and spectral flow, not the naive index.

connection

The index theorem unifies a remarkable web of classical results, each recovered by choosing a particular elliptic operator:

Gauss-Bonnet Theorem — the Euler operator $d+d^*$ gives $\operatorname{ind} = \chi(M)$ , integrating the Euler class.
Hirzebruch Signature Theorem — the signature operator gives $\operatorname{ind} = \tau(M) = \int_M L(M)$ , the $L$ -genus.
Riemann-Roch Theorem — the Dolbeault operator $\bar\partial + \bar\partial^*$ on a complex manifold gives $\operatorname{ind} = \chi(M,\mathcal{O})$ , the Todd-genus / holomorphic Euler characteristic.
K-Theory — the natural home of the symbol class and the cleanest proof framework via Bott periodicity and the Thom isomorphism.
<a class="wiki-link" href="/wiki/characteristic-classes">Characteristic Classes</a> — Chern, Pontryagin, Euler, Todd, and $\hat A$ classes are the building blocks of the topological index.
<a class="wiki-link" href="/wiki/heat-kernel-methods">Heat Kernel Methods</a> — provide the local proof and the bridge to spectral geometry and zeta-function regularization.
<a class="wiki-link" href="/wiki/chiral-anomaly">Chiral Anomaly</a> — the physics incarnation: the anomalous non-conservation of axial current is the Dirac-operator index density.

🔬 Where this appears

⏳ History

- 1827–1848 — Gauss & Bonnet. The Gauss–Bonnet theorem links total curvature to the Euler characteristic — the first "index theorem" in spirit.

- 1850s — Riemann & Roch. The Riemann–Roch theorem for curves equates

\dim H^0 - \dim H^1

of a line bundle with topological data, the prototypical analytic-minus-cokernel formula.

- 1954 — Hirzebruch. Proves the signature theorem and the general Hirzebruch–Riemann–Roch theorem using cobordism and his multiplicative genera; introduces the Todd and

L

classes systematically.

- 1959–1960 — Gel'fand. Poses the general problem: compute the index of an elliptic operator from topological data, conjecturing it is a homotopy invariant.

- 1963 — Atiyah & Singer. Announce and prove the general index theorem (cobordism proof), unifying Gauss–Bonnet, Hirzebruch–Riemann–Roch, and the signature theorem.

- 1968 — Atiyah, Singer, Segal. The

K

-theory proof (papers I–III), plus the equivariant Atiyah–Bott–Segal–Singer fixed-point theorem.

- 1971 — Atiyah, Singer. Paper IV–V extend to families (the index in

K(B)

) and real/

KR

-theory.

- 1975 — Atiyah, Patodi, Singer. The APS theorem for manifolds with boundary, introducing the eta invariant and spectral flow.

- 1983–1986 — Getzler, Bismut, Berline–Vergne. Heat-kernel/supersymmetric proofs giving the local index theorem; physics-inspired (Alvarez-Gaumé) supersymmetric quantum mechanics derivation.

open-problem

Beyond compact manifolds. For non-compact or open manifolds, $D$ need not be Fredholm and the index in the classical sense may not exist; one needs $L^2$ -indices (Atiyah's $L^2$ -index theorem for covering spaces), the Connes–Moscovici higher index, or coarse index theory. Determining the right invariants and their computability is an active program.

The Novikov conjecture — homotopy invariance of higher signatures — is essentially a far-reaching index-theoretic statement; it remains open in general and is one of the central motivating problems of noncommutative geometry and the Baum–Connes program.

The Gromov–Lawson–Rosenberg conjecture on existence of positive scalar curvature metrics is governed by index obstructions ( $\hat A$ and $\alpha$ -invariants in $KO$ -theory); the full conjecture is false in general (counterexamples by Schick) and the precise correct statement is open.

Foliations and singular spaces. Extending index theory to foliated manifolds (Connes' index theorem for foliations), stratified/singular spaces, and orbifolds (Kawasaki) raises ongoing analytic and operator-algebraic difficulties.

Effective/local refinements. While Getzler's proof localizes the index, fully understanding the heat-kernel cancellations as a manifestation of supersymmetry, and rigorously connecting them to path-integral derivations, continues to stimulate work at the math–physics interface.

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also