The Connes-Moscovici Local Index Formula

Intuition Beginner

An index is a whole number that counts solutions of an equation minus the obstructions to solving it. You met it as a robust integer that does not move when you wiggle the operator. The question this unit answers is brutally practical: given an abstract geometry encoded by a single operator, how do you actually compute that integer? Integrating a curvature form over a manifold is the classical recipe, but there may be no manifold here, no points, no curvature to integrate.

The trick comes from physics. When a physicist meets a divergent sum, the useful information often lives in how it diverges — the coefficient of the leading blow-up. Connes and Moscovici turned this into a theorem. Attach to the operator a function of one complex variable, a kind of generating function built from the operator's eigenvalues, and watch where it blows up. The locations of the blow-ups form a short list of numbers called the dimension spectrum, and the strengths of the blow-ups — the residues — assemble into exactly the integer you want.

So the index is read off the way you read off the residue of a pole: a local, finite recipe replacing a global integral. In the familiar case of a curved space this recipe reproduces the celebrated index theorem, but it keeps working when the space dissolves into pure algebra.

Visual Beginner

A zeta-type function of a complex variable, plotted as a landscape, has spikes at a short list of points; the heights of the spikes combine into the index.

The dictionary reads: the flat sheet is the complex plane of the variable $z$; the spikes are the poles of the zeta-type function built from the operator; the marked points are the dimension spectrum; the residue tags are the strengths of the poles; and the assembly box adds them with fixed weights to return the index. The picture to keep is a divergent quantity whose manner of diverging carries the answer.

Worked example Beginner

Take the simplest geometry: the circle, with the operator that measures how fast a wave wiggles. Its eigenvalues are the whole numbers, one wave for each whole number. Build the generating function by raising the size of each eigenvalue to a negative power $z$ and adding all the terms together:

$$\zeta (z)=2(1+2^{-z}+3^{-z}+4^{-z}+5^{-z}+\cdots ).$$

The factor of $2$ is there because each nonzero whole number is matched by its mirror image of the same size. This running total settles to a finite value when $z$ is bigger than $1$, and it blows up as $z$ comes down to $1$. So the function has exactly one spike, located at $z=1$. The location $1$ is the dimension of the circle — the recipe has detected that the circle is one-dimensional, with no mention of points or charts.

The strength of the spike, its residue at $z=1$, comes out to the value $2$ (twice the standard residue of the basic zeta function, because each nonzero whole number is counted with its mirror image). That number $2$ is, up to the fixed normalisation built into the recipe, the length of the circle measured the noncommutative way.

What this tells us: a single sum over eigenvalues found the dimension of the space as the place where it blows up, and a geometric size as the strength of the blow-up. The full local index formula is this same move, run with a few extra eigenvalue-sums, delivering an integer instead of a length.

Check your understanding Beginner

Formal definition Intermediate+

Let $(\mathcal{A},\mathcal{H},D)$ be a spectral triple 39.06.01: $\mathcal{A}$ a unital $\ast$-algebra represented on $\mathcal{H}$, $D=D^{\ast }$ with compact resolvent, and $[D,a]$ bounded for $a\in \mathcal{A}$. Assume it is finitely summable of metric dimension $p$, so $|D{|}^{-1}\in {\mathcal{L}}^{p,\infty }$ 39.06.05. Let $\delta (T)=[|D|,T]$ be the unbounded derivation of $B(\mathcal{H})$. The triple is regular (or smooth) if $a$ and $[D,a]$ lie in ${\bigcap }_{k\ge 0}\mathrm{dom}{\delta }^k$ for every $a\in \mathcal{A}$. Regularity is the noncommutative analogue of the symbols of a differential operator having all derivatives.

The algebra of pseudodifferential operators $\Psi (\mathcal{A})$ is generated by the operators ${\delta }^k(a)$, ${\delta }^k([D,a])$, and the complex powers $|D{|}^z$. An operator $P$ has order $\le m$ if $P|D{|}^{-m}$ is bounded; ${\mathrm{OP}}^m$ denotes operators of order $\le m$, and ${\mathrm{OP}}^{-\infty }={\bigcap }_m{\mathrm{OP}}^m$ is the smoothing ideal. This filtration mirrors the classical order filtration on differential operators, with $|D|$ playing the role of $(1+\Delta {)}^{1/2}$.

For $b\in \Psi (\mathcal{A})$ the zeta function is $$ \zeta_b(z) = \operatorname{Tr}\big(b,|D|^{-z}\big), \qquad \Re z > p, $$ holomorphic where the trace converges. The triple has discrete dimension spectrum $\mathrm{Sd}\subset \mathbb{C}$ if for every $b$ built from the ${\delta }^k(a)$ the function ${\zeta }_b$ extends to a meromorphic function on $\mathbb{C}$ with all poles contained in the discrete set $\mathrm{Sd}$. The spectrum is simple if all poles are simple. The noncommutative residue (the abstract Wodzicki residue, extending the Dixmier trace of 39.06.05) is $$ \tau_0(b) = \operatorname{Res}{z=0},\Gamma(z),\zeta_b(z) \quad\text{or, in the simple case,}\quad \fint b := \operatorname{Res}{z=0}\operatorname{Tr}\big(b,|D|^{-z}\big), $$ a trace on $\Psi (\mathcal{A})$ vanishing on ${\mathrm{OP}}^{-\infty }$.

Definition (the residue cocycle). Write $a^{(k)}={\nabla }^k(a)$ for the $k$-fold commutator with $D^2$, $\nabla (T)=[D^2,T]$, and for a multi-index $k=(k_1,\dots ,k_n)$ set $|k|=k_1+\cdots +k_n$. In the even case (with grading $\gamma$) the Connes-Moscovici cochain is the family $({\phi }_n{)}_{n\mathrm{even}}$, $$ \varphi_n(a^0, a^1, \dots, a^n) = \sum_{k}, c_{n,k}\ \operatorname{Res}{z=0},\operatorname{Tr}\Big(\gamma, a^0, [D,a^1]^{(k_1)} \cdots [D,a^n]^{(k_n)}, |D|^{-2|k|-n-2z}\Big), $$ with universal constants $$ c{n,k} = \frac{(-1)^{|k|}}{k_1!\cdots k_n!},\frac{\Gamma!\big(|k| + \tfrac{n}{2}\big)}{\big(k_1+1\big)\big(k_1+k_2+2\big)\cdots\big(k_1+\cdots+k_n+n\big)}, $$ together with a degree-zero term ${\phi }_0(a^0)$ given by a regularised trace (the constant term of $\mathrm{Tr}(\gamma a^0|D{|}^{-2z})$ at $z=0$ plus the contribution of $\ker D$). The odd case drops $\gamma$, runs over odd $n$, and pairs with $K_1$. The cochain lives in the $(b,B)$-bicomplex of cyclic cohomology 39.07.01.

Counterexamples to common slips

• The dimension spectrum is a property to be verified, not assumed: a finitely summable triple need not have meromorphic zeta functions at all. The hypothesis $\mathrm{Sd}$ discrete is exactly what licenses taking residues. Without it the residue cochain is undefined; the theta-summable JLO cocycle is the fallback (see Advanced results).
• The residue $\text{fint}$ is not the Dixmier trace in general; it agrees with it on the leading order $|D{|}^{-p}$ but the higher residue terms (poles below the top of $\mathrm{Sd}$) carry the lower-order curvature data the Dixmier trace alone cannot see. Equating the two collapses the whole cochain to its $n=p$ term and loses the lower components.
• The sum over $k$ in ${\phi }_n$ is finite on each pair $(a^0,\dots ,a^n)$ because regularity forces $[D,a^j{]}^{(k_j)}\in {\mathrm{OP}}^{2k_j-1+1}$ and the order eventually pushes the operator into ${\mathrm{OP}}^{-\infty }$, where the residue vanishes; truncating the sum prematurely, or treating it as infinite, both fail.

Key theorem with proof Intermediate+

Theorem (Connes-Moscovici local index formula, even case). Let $(\mathcal{A},\mathcal{H},D)$ be a regular even spectral triple of metric dimension $p$ with simple discrete dimension spectrum. Then the cochain $({\phi }_n{)}_{n\mathrm{even}}$ above is a $(b,B)$-cocycle in the cyclic complex of $\mathcal{A}$, its class equals the Chern character $\operatorname{ch}^[D] \in HP^{\mathrm{ev}}(\mathcal{A})$oftheK-homologyclass$[D]$[\mathrm{39.06.03}],andforanyprojection$e \in M_q(\mathcal{A})$representing$[e] \in K_0(\mathcal{A})$,* $$ \langle [D], [e]\rangle ;=; \operatorname{index}\big(e D^+ e\big) ;=; \sum_{n\ \mathrm{even}} (\varphi_n ,#,\operatorname{tr})(e, e, \dots, e), $$ a finite sum of residues equal to the Fredholm index. ^{[Connes-Moscovici 1995; Higson 2004]}

Proof. The strategy is to start from the McKean-Singer expression for the index and resum it into residues. Let $F=D|D{|}^{-1}$ be the bounded phase 39.06.03, $\gamma$ the grading, and $e\in \mathcal{A}$ a projection (amplify by $\mathrm{tr}$ on $M_q$ at the end). By the Calderón-Fedosov formula, for any even integer $m\ge p$, $$ \operatorname{index}(eF^+e) = \operatorname{Tr}\big(\gamma,e,(F[F,e])^{m}\big)\cdot \kappa_m, $$ a supertrace of a sufficiently high power of the commutator $[F,e]$, finite because $[F,e]\in {\mathcal{L}}^p$ 39.06.03. Express $F=D|D{|}^{-1}$ and expand each $[F,e]$ using $[D|D{|}^{-1},e]=[D,e]|D{|}^{-1}+D[|D{|}^{-1},e]$. The second term is lower order by regularity, and iterating the expansion writes the supertrace as a sum of terms $$ \operatorname{Tr}\big(\gamma, e,[D,e]^{(k_1)}\cdots[D,e]^{(k_n)},|D|^{-2|k|-n}\big), $$ each a value of a trace of an operator in ${\mathrm{OP}}^{-2|k|-n}$.

These traces are individually only conditionally convergent, so regularise: replace $|D{|}^{-2|k|-n}$ by $|D{|}^{-2|k|-n-2z}$, giving the zeta function $\zeta (z)$ whose value at $z=0$ is the desired (regularised) trace. By the dimension-spectrum hypothesis each such $\zeta$ is meromorphic with at worst a simple pole at $z=0$. The McKean-Singer index is $z$-independent, so the sum over all terms is regular at $z=0$; but the individual Laurent expansions have poles that cancel. Collecting the residues and the constant terms, and tracking the combinatorial weights produced by the resummation (the $c_{n,k}$, which arise from integrating the heat-kernel / resolvent expansion $\frac{1}{2\pi i}\oint e^{-\lambda }{\lambda }^{-z}(\lambda -D^2{)}^{-1}d\lambda$ term by term), one obtains precisely $$ \operatorname{index}(eF^+e) = \sum_{n\ \mathrm{even}}\sum_k c_{n,k}\operatorname{Res}{z=0}\operatorname{Tr}\big(\gamma,e,[D,e]^{(k_1)}\cdots[D,e]^{(k_n)}|D|^{-2|k|-n-2z}\big) = \sum_n \varphi_n(e,\dots,e). $$ Finiteness of the $n$-sum: each $[D,e{]}^{(k_j)}$ raises order by $2k_j$ but $[D,e]\in {\mathrm{OP}}^0$, and once $n$ exceeds $p$ the operator $e[D,e{]}^{(k)}\cdots |D{|}^{-\cdots }$ lies in ${\mathrm{OP}}^{-\infty }$ where the residue vanishes, so only finitely many $n$ contribute. That $({\phi }_n)$ is a $(b,B)$-cocycle representing ${\mathrm{ch}}^{\ast }[D]$ is the cohomological packaging: $b\varphi_n + B\varphi{n+2} = 0$followsfromthetracepropertyof$\fint$andintegrationbypartsfor$\delta$,andthepairingitcomputesis,by[\mathrm{39.07.01}],theindexpairing.$\square$

Bridge. The local index formula builds toward the full computational engine of noncommutative geometry, and it appears again in the transverse index theory for foliations and the Hopf-cyclic cohomology that Connes and Moscovici built to organise its combinatorics. The foundational reason the residues do the work is exactly the cancellation forced by the $z$-independence of the McKean-Singer index: each individual zeta function has a pole at $z=0$, but their assembled sum is regular, so the poles' residues — local spectral data — must encode the global integer. This is exactly the move that turns the Dixmier-trace integral of 39.06.05 into a full cochain rather than a single functional, and it generalises the heat-kernel proof of Atiyah-Singer 03.09.10, where the short-time asymptotics of $\mathrm{Tr}(\gamma e^{-t{D}^2})$ play the role the residues play here. Putting these together, the bridge is that the Chern character ${\mathrm{ch}}^{\ast }[D]$ of 39.06.03, an abstract cyclic class, acquires a concrete local representative $({\phi }_n)$, so the index pairing of 39.07.01 becomes a finite computation in residues of zeta functions over the dimension spectrum.

Exercises Intermediate+

Lean formalization Intermediate+

lean_status: none — Mathlib has neither the pseudodifferential calculus on a spectral triple, the zeta functions $\mathrm{Tr}(b|D{|}^{-z})$ and their meromorphic continuation, the dimension spectrum, the noncommutative residue, nor the $(b,B)$-bicomplex of cyclic cohomology. The intended statement is schematic:

-- Schematic. Mathlib lacks: regular spectral triples, the operator-order
-- filtration OPᵐ, the zeta function Tr(b |D|^{-z}) with meromorphic
-- continuation, the dimension spectrum Sd, the noncommutative residue ∮,
-- the (b,B)-bicomplex, and the K-homology index pairing.

/-- The residue cochain of a regular spectral triple with discrete dimension
    spectrum: φ_n is a sum over multi-indices of residues of zeta functions. -/
noncomputable def residueCochain
    {A H : Type*} (T : RegularSpectralTriple A H) (n : ℕ) :
    (Fin (n + 1) → A) → ℂ :=
  sorry  -- Σ_k c_{n,k} · Res_{z=0} Tr(γ a⁰ [D,a¹]^{(k₁)} ⋯ |D|^{-2|k|-n-2z})

/-- The local index formula: the residue cochain is a (b,B)-cocycle whose
    pairing with K₀ computes the Fredholm index of e D⁺ e. -/
theorem local_index_formula
    {A H : Type*} (T : RegularSpectralTriple A H)
    (e : A) (he : IsIdempotentElem e) :
    True :=   -- index (compress T.D e) = Σ_n (residueCochain T n) (e, …, e)
  trivial

The honest gap spans the entire analytic stack: the order filtration and pseudodifferential algebra, the meromorphy hypothesis defining the dimension spectrum, the residue functional extending the Dixmier trace of 39.06.05, the cyclic $(b,B)$-bicomplex of 39.07.01, and the K-homology pairing of 39.06.03. Each is a long-horizon formalisation target on its own.

Advanced results Master

The reduced (resolvent) cocycle. The original Connes-Moscovici cochain mixes degrees through the constants $c_{n,k}$; Higson and, in the semifinite setting, Carey-Phillips-Rennie-Sukochev recast it as a resolvent cocycle, replacing $|D{|}^{-z}$ by contour integrals of the resolvent $(\lambda -D^2{)}^{-1}$ ^{[Carey-Phillips-Rennie-Sukochev]}. The resolvent form makes the $(b,B)$-cocycle identities transparent: $B$ of one component is $b$ of the next, verified by a single integration by parts inside the contour integral, and it extends the formula to spectral triples in a semifinite von Neumann algebra where $\mathrm{Tr}$ is replaced by a faithful normal semifinite trace $\tau$ and the index lands in $\mathbb{R}$ rather than $\mathbb{Z}$ — the spectral-flow / Fredholm-in-a-${\mathrm{II}}_{\infty }$-factor refinement.

The odd formula and spectral flow. For an odd regular triple and a unitary $u\in M_q(\mathcal{A})$ the analogous cochain $({\phi }_n{)}_{n\mathrm{odd}}$ computes $$ \langle [D], [u]\rangle = -\operatorname{sf}(D, uDu^*) = \frac{1}{\sqrt{2\pi i}}\sum_{n,\mathrm{odd}} (\varphi_n,#,\operatorname{tr})(u^{-1}, u, \dots, u^{-1}, u), $$ the spectral flow 39.06.03 expressed as a residue sum. On $C^{\infty }(\mathbb{T})$ with $D=-id/d\theta$ and $u=e^{ik\theta }$, only ${\phi }_1(u^{-1},u)=\frac{1}{\sqrt{2\pi i}}\text{fint}{u}^{-1}[D,u]|D{|}^{-1}$ survives, evaluating to $-k$, the winding number, recovering the Noether-Gohberg-Krein index of 39.06.03.

The dimension spectrum as a refined invariant. When $\mathrm{Sd}$ has poles of order $>1$ (the non-simple case) the residue functional ${\tau }_0(b)={\mathrm{Res}}_{z=0}\Gamma (z){\zeta }_b(z)$ must be supplemented by higher-order residue functionals ${\tau }_j(b)={\mathrm{Res}}_{z=0}{z}^{-j-1}$-type terms, and the cochain acquires logarithmic corrections. This is the regime of singular geometries and the modular spectral triples of type-$\mathrm{III}$ examples, where the dimension spectrum sits off the real axis. The set $\mathrm{Sd}$ thereby encodes finer data than the metric dimension alone: it is a noncommutative analogue of the singularity structure of the heat-kernel expansion, with each pole a heat-coefficient $a_k$.

The theta-summable bridge: the JLO cocycle. When the triple is only $\theta$-summable — $\mathrm{Tr}(e^{-t{D}^2})<\infty$ for $t>0$ but no finite $p$ — the zeta functions need not be meromorphic and the residue cochain is undefined. The Chern character is instead the entire cyclic JLO cocycle 39.07.01 $$ \varphi_n^{\mathrm{JLO}}(a^0, \dots, a^n) = \int_{\Delta_n}\operatorname{Tr}\big(\gamma, a^0 e^{-s_0 D^2}[D,a^1]e^{-s_1 D^2}\cdots[D,a^n]e^{-s_n D^2}\big),ds, $$ integrated over the simplex $\sum s_i=1$. The local index formula is the asymptotic expansion of the JLO cocycle as one rescales $D\mapsto {\epsilon }^{-1}D$ and extracts the $\epsilon \to 0$ limit: the residues are the coefficients of the heat-kernel short-time asymptotics, so the finitely summable residue cocycle and the theta-summable JLO cocycle are two faces of the same Chern character, related by a regularisation of the heat-kernel integral.

Synthesis. The local index formula is the foundational reason noncommutative index theory is computable, and it is dual, term by term, to the heat-kernel proof of the classical theorem: where Atiyah-Singer 03.09.10 reads the index off the short-time asymptotics of $\mathrm{Tr}(\gamma e^{-t{D}^2})$, the residue cocycle reads it off the poles of $\mathrm{Tr}(b|D{|}^{-z})$, and these are exactly Mellin transforms of one another. This is exactly the completion of the program begun in 39.06.03: the abstract Chern character ${\mathrm{ch}}^{\ast }[D]$ there had no formula, and the central insight here is that regularity plus a discrete dimension spectrum supply a local representative, turning the cyclic pairing of 39.07.01 into residues. The noncommutative residue generalises the Dixmier trace of 39.06.05 from a single top-degree functional to the full cochain $({\phi }_n)$, and the JLO cocycle is dual to the residue cocycle through the heat-kernel rescaling. Putting these together, the same integer is computed four ways — as a Fredholm index of a compression 39.06.03, as a cyclic pairing of Chern characters 39.07.01, as a sum of residues over the dimension spectrum, and (commutatively) as the $\hat{A}\mathrm{ch}$ characteristic integral 03.09.10 — and the assertion that these coincide is the local index theorem of noncommutative geometry.

Full proof set Master

Proposition (the noncommutative residue is a trace). Let $(\mathcal{A},\mathcal{H},D)$ be a regular triple with discrete dimension spectrum. Then $\text{fint}b={\mathrm{Res}}_{z=0}\mathrm{Tr}(b|D{|}^{-z})$ is a trace on $\Psi (\mathcal{A})$: $\text{fint}(b_1{b}_2)=\text{fint}(b_2{b}_1)$, and it vanishes on ${\mathrm{OP}}^{-\infty }$.

Proof. Vanishing on ${\mathrm{OP}}^{-\infty }$: if $b\in {\mathrm{OP}}^{-\infty }$ then $b|D{|}^{-z}$ is trace-class for all $z$ in a neighbourhood of $0$ and $\mathrm{Tr}(b|D{|}^{-z})$ is holomorphic there, so its residue at $z=0$ vanishes. Traciality: for $b_1,b_2\in \Psi (\mathcal{A})$ consider $f(z)=\mathrm{Tr}([b_1,b_2]|D{|}^{-z})$. The operator $[b_1,b_2]$ has order strictly below $\mathrm{ord}(b_1)+\mathrm{ord}(b_2)$ by the symbol calculus (the leading symbols commute), and writing $\mathrm{Tr}(b_1{b}_2|D{|}^{-z})-\mathrm{Tr}(b_2{b}_1|D{|}^{-z})=\mathrm{Tr}(b_2[|D{|}^{-z},b_1])\cdot (\text{sign})$, the commutator $[|D{|}^{-z},b_1]=-z|D{|}^{-z}\delta (b_1)|D{|}^{-1}+O(z^2)$ carries an explicit factor of $z$. Hence $f(z)=zg(z)$ with $g$ having at worst a simple pole at $z=0$, so ${\mathrm{Res}}_{z=0}f={\lim }_{z\to 0}zf(z)=0$. Therefore $\text{fint}{b}_1{b}_2=\text{fint}{b}_2{b}_1$. $\square$

Proposition (finiteness of the residue cochain). For each fixed $(a^0,\dots ,a^n)$ the multi-index sum defining ${\phi }_n$ is finite, and ${\phi }_n=0$ for $n>p$.

Proof. The factor $[D,a^j{]}^{(k_j)}$ lies in ${\mathrm{OP}}^{2k_j}$ (each application of $\nabla =[D^2,\cdot ]$ raises order by $2$, starting from $[D,a^j]\in {\mathrm{OP}}^0$ in the convention where order counts powers of $|D|$ in $D=|D|\cdot F$). Thus $a^0[D,a^1{]}^{(k_1)}\cdots [D,a^n{]}^{(k_n)}|D{|}^{-2|k|-n}$ has order $\le 2|k|-2|k|-n=-n$. The residue of $\mathrm{Tr}(\cdot |D{|}^{-2z})$ of an order-$(-n)$ operator is nonzero only when $-n\ge -p$, i.e. $n\le p$; for $n>p$ the operator is in ${\mathrm{OP}}^{-\infty }$ relative to the trace and the residue vanishes by the previous Proposition. Within fixed $n$, increasing $|k|$ lowers the order further, so beyond a finite range the residue again vanishes; the surviving terms are those with $-2|k|-n\ge -p$, a finite set. $\square$

Proposition (the residue cochain is a $(b,B)$-cocycle). The family $({\phi }_n)$ satisfies $b{\phi }_n+B{\phi }_{n+2}=0$, hence defines a class in $H{P}^{\mathrm{ev}}(\mathcal{A})$.

Proof. Expand $(b{\phi }_n)(a^0,\dots ,a^{n+1})$ using the Hochschild coboundary 39.07.01 and the Leibniz rule $[D,a^i{a}^{i+1}]=[D,a^i]a^{i+1}+a^i[D,a^{i+1}]$ inside each residue. The cross terms reorganise, using traciality of $\text{fint}$ (previous Proposition) to cyclically permute factors under the residue, into the expression $-(B{\phi }_{n+2})(a^0,\dots ,a^{n+1})$, where $B$ inserts a unit and antisymmetrises. The combinatorial constants $c_{n,k}$ are defined precisely so that the weights match across the two sides: the recursion $c_{n,k}$ vs. $c_{n+2,k^{\prime }}$ encodes the integration-by-parts identity $\text{fint}[D^2,T]=0$ (a consequence of traciality applied to $[D^2,\cdot ]$) that converts a $\nabla$-derivative in degree $n+2$ into a Hochschild face in degree $n$. Assembling all terms gives $b{\phi }_n+B{\phi }_{n+2}=0$. $\square$

Proposition (the residue cochain represents ${\mathrm{ch}}^{\ast }[D]$). The class of $({\phi }_n)$ in $H{P}^{\mathrm{ev}}(\mathcal{A})$ equals the Chern character of the K-homology class $[D]=[F]$, $F=D|D{|}^{-1}$.

Proof. Both classes pair with $K_0(\mathcal{A})$ to give the Fredholm index $\mathrm{index}(e{F}^{+}e)$: for $({\phi }_n)$ this is the content of the Key theorem (the McKean-Singer resummation), and for ${\mathrm{ch}}^{\ast }[F]$ it is the defining property of the Chern character of a finitely summable Fredholm module 39.06.03. Since the pairing $H{P}^{\mathrm{ev}}(\mathcal{A})\times K_0(\mathcal{A})\to \mathbb{C}$ is, on the image of the index map, nondegenerate enough to separate the two only if they differed on some class, and they agree on every class, the two periodic cyclic classes coincide. (More precisely: both are entire cyclic cocycles whose paired values with the dense set of K-theory classes agree, and the Chern character of a fixed module is unique up to $S$-equivalence, so the residue cocycle is its local representative.) $\square$

Proposition (commutative reduction). For $\mathcal{A}=C^{\infty }(M)$, $M$ a closed even-dimensional spin manifold, $D=/D$, the residue cochain integrates to the Atiyah-Singer integrand: ${\sum }_n{\phi }_n(e,\dots ,e)={\int }_M\hat{A}(M)\mathrm{ch}(E)$ for $E=e\underline{{\mathbb{C}}^q}$.

Proof. The noncommutative residue $\text{fint}$ on the algebra of classical pseudodifferential operators on $M$ is the Wodzicki residue ${\mathrm{res}}_W(P)={\int }_{S^{\ast }M}\mathrm{tr}{\sigma }_{-n}(P)$, the integral of the order-$(-n)$ component of the symbol over the cosphere bundle. Each ${\phi }_n(a^0,\dots ,a^n)$ becomes the integral of a local density built from $a^0,d{a}^1,\dots ,d{a}^n$ and the curvature of $M$. The Connes-Moscovici constants $c_{n,k}$ match the Seeley-DeWitt heat coefficients, and the assembled sum is the Gilkey local index density, which by the local Atiyah-Singer theorem 03.09.10 integrates to ${\int }_M\hat{A}(M)\mathrm{ch}(E)$. Hence the residue cocycle reproduces the twisted-Dirac index. $\square$

Connections Master

• Fredholm Modules and the Index Pairing 39.06.03 — the Chern character ${\mathrm{ch}}^{\ast }[D]$ of a finitely summable Fredholm module, defined there abstractly through the supertrace of commutators of $F=D|D{|}^{-1}$, receives in this unit its concrete local representative; the McKean-Singer resummation of the Calderón-Fedosov index formula of that unit is the engine of the proof here, and the odd local formula computes the spectral flow / winding-number pairing introduced there.

• The Dixmier Trace and the Noncommutative Integral 39.06.05 — the noncommutative residue $\text{fint}b={\mathrm{Res}}_{z=0}\mathrm{Tr}(b|D{|}^{-z})$ is the Wodzicki/zeta extension of the Dixmier trace of that unit; the bottom component ${\phi }_0$ of the index cochain is exactly the Dixmier-trace pairing, and the higher ${\phi }_n$ are the lower-order curvature corrections the single Dixmier functional cannot resolve.

• Cyclic Cohomology and the Pairing with K-Theory 39.07.01 — the residue cochain $({\phi }_n)$ is a cocycle in the $(b,B)$-bicomplex of that unit, its class lives in $H{P}^{\mathrm{ev}}(\mathcal{A})$, and the value it computes against a projection is the K-theory pairing constructed there; the JLO cocycle for the theta-summable case is the entire-cyclic-cohomology bridge that unit introduces.

• Atiyah-Singer Index Theorem 03.09.10 — the commutative case of the local index formula is the heat-kernel / Wodzicki-residue proof of Atiyah-Singer: the residue cochain integrates to ${\int }_M\hat{A}(M)\mathrm{ch}(E)$, so this unit is the noncommutative envelope of that theorem, with residues of zeta functions replacing short-time heat asymptotics.

• Spectral Triples and the Reconstruction Theorem 39.06.01 — the regularity and finite-summability axioms, the derivation $\delta =[|D|,\cdot ]$, the metric dimension, and the dimension spectrum are all data of the spectral triple defined there; the local index formula is the analytic payoff that makes those axioms compute an index.

• The Noncommutative Torus as a Noncommutative Geometry 39.06.04 — the flat triple on $A_{\theta }^{\infty }$ has dimension spectrum $\{2\}$ and a two-term residue cochain $({\phi }_0,{\phi }_2)$ whose pairing with $K_0(A_{\theta })\cong {\mathbb{Z}}^2$ computes the integer detecting the Powers-Rieffel projection, a fully manifold-free instance of the formula of this unit.

Historical & philosophical context Master

The local index formula was proved by Alain Connes and Henri Moscovici in The local index formula in noncommutative geometry (Geom. Funct. Anal. 5, 1995) ^{[Connes-Moscovici 1995]}, the culmination of Connes' program, begun in the 1994 book Noncommutative Geometry ^{[Connes 1994]}, to compute the Chern character of a spectral triple from spectral data. The immediate motivation was the transverse index theorem for foliations, where the relevant operator is hypoelliptic rather than elliptic and the classical heat-kernel methods do not directly apply; the residue calculus and the dimension spectrum were the tools that survived the loss of ellipticity. The combinatorial complexity of the constants $c_{n,k}$ led Connes and Moscovici to introduce Hopf-cyclic cohomology and the Hopf algebra ${\mathcal{H}}_n$ of transverse geometry to organise them.

Nigel Higson gave an influential streamlined proof through the zeta-function and resolvent calculus ^{[Higson 2004]}, clarifying the role of the dimension spectrum and the comparison with the heat kernel. Carey, Phillips, Rennie and Sukochev extended the formula to semifinite von Neumann algebras with a faithful normal trace, replacing the integer index by a real-valued spectral flow and recasting the cochain as a resolvent cocycle ^{[Carey-Phillips-Rennie-Sukochev]}. The noncommutative residue itself originates in the work of Wodzicki and Guillemin on the unique trace on the algebra of classical pseudodifferential operators, which Connes recognised as the abstract integral of noncommutative geometry. The lineage runs from the heat-kernel proof of Atiyah-Singer, through Connes' identification of the Dixmier trace as the noncommutative integral, to the residue cocycle that computes any index pairing from the poles of zeta functions.

Bibliography Master

• Connes, A. & Moscovici, H., "The local index formula in noncommutative geometry", Geometric and Functional Analysis 5 (1995), 174–243.
• Connes, A., Noncommutative Geometry, Academic Press, 1994. Ch. IV.
• Higson, N., "The local index formula in noncommutative geometry", in Contemporary Developments in Algebraic K-Theory, ICTP Lecture Notes 15, 2004, 443–536.
• Carey, A. L., Phillips, J., Rennie, A. & Sukochev, F. A., "The local index formula in semifinite von Neumann algebras I, II", Advances in Mathematics 202 (2006), 451–516 and 517–554.
• Connes, A. & Moscovici, H., "Hopf algebras, cyclic cohomology and the transverse index theorem", Communications in Mathematical Physics 198 (1998), 199–246.
• Wodzicki, M., "Noncommutative residue, Part I: Fundamentals", in K-theory, Arithmetic and Geometry, Lecture Notes in Mathematics 1289, Springer, 1987, 320–399.
• Gracia-Bondía, J. M., Várilly, J. C. & Figueroa, H., Elements of Noncommutative Geometry, Birkhäuser, 2001. Ch. 10.
• Connes, A. & Marcolli, M., Noncommutative Geometry, Quantum Fields and Motives, AMS Colloquium Publications 55, 2008. §1.3–1.4.

---

Operator-algebras spine, noncommutative-geometry chapter. The computational engine of noncommutative index theory: the dimension spectrum $\mathrm{Sd}$ as the poles of ${\zeta }_b(z)=\mathrm{Tr}(b|D{|}^{-z})$, the pseudodifferential calculus and noncommutative (Wodzicki) residue on a regular spectral triple, the Connes-Moscovici residue $(b,B)$-cocycle $({\phi }_n)$ representing the Chern character $\operatorname{ch}^[D]$, the local index formula computing the index pairing as a finite sum of residues, the odd/spectral-flow and semifinite/resolvent refinements, the JLO theta-summable bridge, and the commutative reduction to the heat-kernel proof of Atiyah-Singer. Builds on Fredholm modules (39.06.03), the Dixmier trace (39.06.05), and cyclic cohomology (39.07.01).*