Spectral Triples and the Reconstruction Theorem

Intuition Beginner

Geometry usually starts with points, distances, and curves. Noncommutative geometry starts somewhere stranger: with an algebra of functions on a space, a place for those functions to act, and one special operator. The claim is that this small package already contains all the geometry — the dimension, the distances, and the shape — and you read it off from the operator instead of from a map.

The special operator is a curved-space cousin of the Dirac operator 03.09.08, the device physicists wrote down for the electron. Its eigenvalues are a list of numbers attached to the space, like the resonant frequencies of a drum. The bigger the space, the slower those numbers grow, and that growth rate turns out to be the dimension. The gaps between the algebra and the operator measure how far apart two points are.

The surprising payoff is a theorem: when the algebra of functions happens to be commutative — ordinary multiplication, order does not matter — this spectral package is nothing other than an ordinary smooth space in disguise. Drop the requirement that order does not matter, and you have a "space" with no points at all, described entirely by spectra.

Visual Beginner

A spectral triple has three ingredients that talk to each other: an algebra of functions, a space of states for them to act on, and one operator whose eigenvalues encode the geometry.

The dictionary reads: the algebra is the functions; the operator is the geometry; the eigenvalue list of the operator, read at its growth rate, is the dimension; and a formula comparing the algebra against the operator returns the distance between two points.

Worked example Beginner

Take the simplest curved shape with a known spectrum: the circle of circumference $2\pi$. The functions are the smooth complex functions on the circle. The operator is one derivative, the circle's Dirac operator, and its eigenvalues are the whole numbers, one for each wave $e^{in\theta }$:

$$\text{eigenvalues of }D=\{\dots ,-2,-1,0,1,2,\dots \}.$$

Count how many eigenvalues have size at most $N$: there are about $2N$ of them, growing in proportion to $N$ to the first power. The dimension is exactly that power: $1$. The circle is one-dimensional, and the eigenvalue growth said so without ever drawing the circle.

Now read off a distance. The distance between two points on the circle equals the largest gap in function values you can create while keeping the function's derivative no bigger than one in size. A function whose slope never exceeds one can climb at most by the arc length between the two points, so the largest gap you can make is the arc length itself.

What this tells us: the eigenvalue growth of the operator gave the dimension, and comparing functions against the operator gave the distance — geometry recovered from spectrum alone.

Check your understanding Beginner

Formal definition Intermediate+

A spectral triple $(\mathcal{A},\mathcal{H},D)$ consists of:

• a unital $\ast$-algebra $\mathcal{A}$ (a dense $\ast$-subalgebra of a C*-algebra $A$ 39.01.01), represented faithfully by bounded operators $\pi :\mathcal{A}\to B(\mathcal{H})$ on a Hilbert space $\mathcal{H}$ 02.11.08;
• a self-adjoint operator $D$ on $\mathcal{H}$, densely defined and generally unbounded 02.11.03, called the Dirac operator of the triple;

subject to two conditions:

1. (Bounded commutators) for every $a\in \mathcal{A}$, the operator $a$ preserves $\mathrm{dom}(D)$ and the commutator $[D,\pi (a)]=D\pi (a)-\pi (a)D$ extends to a bounded operator on $\mathcal{H}$;
2. (Compact resolvent) for some — equivalently every — $\lambda \notin \sigma (D)$, the resolvent $(D-\lambda {)}^{-1}$ is a compact operator.

Write $\pi (a)=a$ when no confusion arises. The triple is even if there is a $\mathbb{Z}/2$-grading $\gamma \in B(\mathcal{H})$, $\gamma ={\gamma }^{\ast }$, ${\gamma }^2=1$, with $\gamma a=a\gamma$ for $a\in \mathcal{A}$ and $\gamma D=-D\gamma$; the bracket $\gamma D=-D\gamma$ records that $D$ is odd. A triple with no such grading is odd.

The triple is $p^{+}$-summable (or has metric dimension $p$) if the eigenvalues ${\mu }_n$ of $|D|=(D^{\ast }D{)}^{1/2}$, listed with multiplicity in increasing order, grow so that ${\mu }_n\sim c{n}^{1/p}$; equivalently $|D{|}^{-1}$ lies in the weak Schatten ideal ${\mathcal{L}}^{p,\infty }$, the operators whose singular values satisfy ${\mu }_n=O(n^{-1/p})$. The dimension spectrum is the set of poles of the family of zeta functions ${\zeta }_b(s)=\mathrm{Tr}(b|D{|}^{-s})$ for $b$ ranging over the algebra generated by $\mathcal{A}$, $[D,\mathcal{A}]$, and $|D{|}^z$; for a $p$-dimensional manifold it is contained in $\{1,2,\dots ,p\}$.

A real structure of $KO$-dimension $n\mathrm{mod}8$ is an antiunitary $J:\mathcal{H}\to \mathcal{H}$ with $J^2=\epsilon$, $JD={\epsilon }^{\prime }DJ$, and (in the even case) $J\gamma ={\epsilon }^{\prime \prime }\gamma J$, where the signs $(\epsilon ,{\epsilon }^{\prime },{\epsilon }^{\prime \prime })\in \{\pm 1{\}}^3$ are fixed by $n\mathrm{mod}8$ through the Clifford-algebra table 03.09.08. The real structure imposes the commutant (zeroth-order) and first-order conditions $$ [a,, J b^* J^{-1}] = 0, \qquad \big[[D, a],, J b^* J^{-1}\big] = 0 \qquad (a, b \in \mathcal{A}), $$ which encode that $D$ is a first-order differential operator and that $\mathcal{A}$ acts as functions on both sides.

The motivating example is commutative. Let $(M,g)$ be a closed Riemannian spin manifold of dimension $n$ with spinor bundle $S$ and Dirac operator $/D$ 03.09.08. Set $$ \mathcal{A} = C^\infty(M), \qquad \mathcal{H} = L^2(M, S), \qquad D = \not{!D}, $$ with $\pi$ pointwise multiplication. Then $[/D,f]=c(df)$ is Clifford multiplication by the differential of $f$, bounded because $\Vert c(df)\Vert =\Vert df{\Vert }_{\infty }$; and $/D$ has compact resolvent because it is elliptic on a closed manifold 03.09.08. The grading $\gamma$ in even dimensions is the chirality operator, and $J$ is charge conjugation. This is the canonical spectral triple of $M$.

Counterexamples to common slips

• Boundedness of $D$ is not required and usually fails. A bounded $D$ on an infinite-dimensional $\mathcal{H}$ cannot have compact resolvent, so the metric dimension would be $0$; the Dirac operator of a positive-dimensional manifold is genuinely unbounded. What must be bounded is each commutator $[D,a]$, not $D$.
• Compact resolvent is a condition on $D$ relative to $\mathcal{H}$, not on $\mathcal{A}$. The pair $(\mathbb{C},{\ell }^2(\mathbb{N}),D)$ with $D=\mathrm{diag}(n)$ has compact resolvent and bounded commutators (the scalars commute with $D$, so $[D,\lambda ]=0$), but the algebra is too small to be a geometry; the reconstruction axioms (orientation, finiteness) rule such degenerate triples out.
• The metric (spectral) dimension is the eigenvalue-growth exponent of $|D|$ and need not equal the Hausdorff dimension of any underlying set, nor be an integer. Fractal and quantum examples have non-integer dimension spectra; only the manifold axioms force the integer value.

Key theorem with proof Intermediate+

Theorem (Connes' distance formula). Let $(\mathcal{A},\mathcal{H},D)$ be the canonical spectral triple of a closed Riemannian spin manifold $(M,g)$, and let $\phi ,\psi$ be two points of $M$, viewed as the characters ${\mathrm{ev}}_{\phi },{\mathrm{ev}}_{\psi }:C^{\infty }(M)\to \mathbb{C}$. Then the geodesic distance is recovered purely from the algebra and $D$: $$ d_g(\varphi, \psi) ;=; \sup\Big{, |a(\varphi) - a(\psi)| ;:; a \in \mathcal{A},\ | [D, a] | \le 1 ,\Big}. $$ ^{[Connes Ch. VI; Connes 1996]}

Proof. Since $\mathcal{A}$ is a $\ast$-algebra, both sides are unchanged if we restrict the supremum to real-valued $a$ (replace $a$ by its real part; the commutator norm does not increase and $|a(\phi )-a(\psi )|$ is achieved on a real or imaginary part). For $a=f\in C^{\infty }(M)$ real-valued, the commutator with the Dirac operator is Clifford multiplication by the differential, $$ [\not{!D}, f] = c(df), $$ because the first-order operator $/D=c\circ {\nabla }^S$ obeys the Leibniz rule and the connection terms cancel in the commutator 03.09.08. The operator norm of Clifford multiplication by a one-form $\omega$ at a point is its pointwise length, $\Vert c({\omega }_x)\Vert =|{\omega }_x{|}_g$, so $$ | [\not{!D}, f] | = \sup_{x \in M} |df_x|g = |,|df|,|\infty = \operatorname{Lip}_g(f), $$ the Lipschitz constant of $f$ with respect to the geodesic distance. Thus the constraint $\Vert [D,f]\Vert \le 1$ is exactly the constraint that $f$ be $1$-Lipschitz.

Now compare with the geodesic distance. For any $1$-Lipschitz $f$ and any path $\gamma$ from $\phi$ to $\psi$, $$ |f(\varphi) - f(\psi)| \le \int_\gamma |df| \le \operatorname{length}(\gamma), $$ and taking the infimum over paths gives $|f(\phi )-f(\psi )|\le d_g(\phi ,\psi )$. Hence the supremum on the right is $\le d_g(\phi ,\psi )$. Conversely the function $f_0(x)=d_g(\phi ,x)$ is $1$-Lipschitz (triangle inequality) and smooth away from the cut locus; smoothing it slightly keeps it $1$-Lipschitz with $f_0(\phi )=0$ and $f_0(\psi )=d_g(\phi ,\psi )$ to within $ϵ$, so the supremum is $\ge d_g(\phi ,\psi )$. Equality follows. $\square$

Bridge. The distance formula builds toward the entire metric layer of noncommutative geometry, and it appears again in 39.07.01 where cyclic cohomology supplies the higher-degree analogue (the metric is the degree-zero data, the integration current the top-degree data). The foundational reason it works is exactly the identity $\Vert [D,f]\Vert =\mathrm{Lip}(f)$: the single unbounded operator $D$ encodes the metric through its commutators, so distance becomes an algebraic supremum that makes sense even when $\mathcal{A}$ is noncommutative and there are no points $\phi ,\psi$ to separate — one then separates states on $A$. This is exactly the dual of the construction in 39.01.01, where Gelfand-Naimark recovers the topological points of a commutative C*-algebra as characters; here $D$ upgrades that topological recovery to a metric one. Putting these together, the commutative Gelfand picture supplies the points and the Connes formula supplies the distances, and the bridge is that $D$ is to the metric what the algebra is to the topology.

Exercises Intermediate+

Lean formalization Intermediate+

lean_status: none — Mathlib has unbounded self-adjoint operators, the spectrum and resolvent, compact operators, and the C*-algebra continuous functional calculus, but it does not bundle these into a spectral triple, and it lacks the Schatten–Dixmier, dimension-spectrum, regularity, Hochschild-orientation, and real-structure layers the axioms require.

The intended statement reads schematically:

import Mathlib.Analysis.InnerProductSpace.l2Space
import Mathlib.Analysis.CStarAlgebra.ContinuousFunctionalCalculus.Basic

/-- A spectral triple: a unital *-subalgebra A of bounded operators on H,
together with a self-adjoint D with compact resolvent and bounded
commutators [D, a]. (Schematic; the unbounded D and the compact-resolvent /
bounded-commutator predicates need the not-yet-bundled NCG infrastructure.) -/
structure SpectralTriple (A : Type*) (H : Type*)
    [CStarAlgebra A] [NormedAddCommGroup H] [InnerProductSpace ℂ H] where
  rep        : A →⋆ₐ[ℂ] (H →L[ℂ] H)
  D          : H →ₗ[ℂ] H          -- morally unbounded, densely defined
  selfAdjoint : True              -- IsSelfAdjoint D
  compactResolvent : True         -- (D - λ)⁻¹ ∈ CompactOperators
  boundedCommutator : ∀ a, True   -- ∃ C, ‖D ∘ rep a - rep a ∘ D‖ ≤ C

Connes' reconstruction theorem — a commutative regular oriented finite real spectral triple is the canonical triple of a smooth compact spin manifold — is a far-horizon target requiring the full spin-geometry and elliptic-operator stack alongside the NCG axioms.

Advanced results Master

The axioms below promote a bare spectral triple to a noncommutative spin geometry. They are stated for the commutative model and are the hypotheses of the reconstruction theorem ^{[Connes 2013]}.

Regularity. Let $\delta (T)=[|D|,T]$. The triple is regular if every $a\in \mathcal{A}$ and every $[D,a]$ lies in the smooth domain ${\bigcap }_{k\ge 1}\mathrm{dom}({\delta }^k)$. For the canonical triple this is the statement that $a$ and its symbol are smooth functions; regularity is the spectral surrogate for $C^{\infty }$. It is what makes the zeta functions ${\zeta }_b(s)=\mathrm{Tr}(b|D{|}^{-s})$ meromorphic with the discrete dimension spectrum.

Finiteness and absolute continuity. The smooth domain ${\mathcal{H}}^{\infty }={\bigcap }_k\mathrm{dom}(D^k)$ is a finitely generated projective $\mathcal{A}$-module (the noncommutative Serre-Swan statement: it plays the role of sections of the spinor bundle), and the noncommutative integral $\text{fint}={\mathrm{Tr}}_{\omega }(\cdot |D{|}^{-n})$ defined by the Dixmier trace ${\mathrm{Tr}}_{\omega }$ on ${\mathcal{L}}^{1,\infty }$ is a faithful trace recovering ${\int }_M(\cdot )d{\mathrm{vol}}_g$ up to the constant $c_n=(2\pi {)}^n/(n{\Omega }_{n-1})$ via Connes' trace theorem ${\mathrm{Tr}}_{\omega }(f|D{|}^{-n})=c_n{\int }_{M}fd{\mathrm{vol}}_g$.

Orientation (Hochschild cycle). There is a Hochschild $n$-cycle $c\in Z_n(\mathcal{A},\mathcal{A}\otimes {\mathcal{A}}^{\mathrm{op}})$ whose Connes representative ${\pi }_D(c)=\sum a^0[D,a^1]\cdots [D,a^n]$ equals the grading $\gamma$ (even) or $1$ (odd). This is the orientation/volume axiom; it fixes the metric dimension $n$ and, by the exercise above, forces $n=\dim M$ in the commutative case.

Poincaré duality. The index pairing of $K$-theory against $K$-homology, realised through the Fredholm module $(\mathcal{H},D,\gamma )$, is nondegenerate: the map $K_{\ast }(\mathcal{A})\to K^{\ast }(\mathcal{A})$ induced by cap product with the fundamental class $[D]\in K{R}_n(\mathcal{A}\otimes {\mathcal{A}}^{\mathrm{op}})$ is an isomorphism. This is the spectral form of manifold Poincaré duality and uses the real structure $J$ to produce the $\mathcal{A}$-bimodule fundamental class.

The dimension spectrum and the noncommutative integral. Regularity makes each ${\zeta }_b(s)=\mathrm{Tr}(b|D{|}^{-s})$ extend meromorphically; its pole set, gathered over all $b$ in the algebra $\mathcal{B}$ generated by ${\delta }^k(a)$ and ${\delta }^k([D,a])$, is the dimension spectrum $\mathrm{Sd}\subset \mathbb{C}$. For the canonical triple of an $n$-manifold $\mathrm{Sd}\subseteq \{n,n-1,\dots ,1\}$, all simple poles, by the heat-kernel expansion of $\mathrm{Tr}(e^{-t{D}^2})\sim {\sum }_{k\ge 0}{a}_k{t}^{(k-n)/2}$ whose Mellin transform produces poles at $s=n-k$. The residue at the top pole reconstructs the volume: ${\mathrm{Res}}_{s=n}{\zeta }_{|D|}(s)$ is proportional to $\mathrm{vol}(M)$, and more generally the noncommutative integral $\text{fint}b={\mathrm{Res}}_{s=n}\mathrm{Tr}(b|D{|}^{-s})$ agrees with the Dixmier-trace integral and equals $c_n{\int }_{M}bd{\mathrm{vol}}_g$ in the commutative case (Connes' trace theorem identifies ${\mathrm{Tr}}_{\omega }$ with the Wodzicki residue). When the dimension spectrum has points off the real axis or multiple poles, the geometry is genuinely fractal or noncommutative — the integer-and-simple structure is a manifold signature.

Inner fluctuations and almost-commutative geometry. The real structure $J$ lets one fluctuate the metric by inner perturbations $D\mapsto D_A=D+A+JA{J}^{-1}$, where $A={\sum }_i{a}_i[D,b_i]$ is a self-adjoint one-form; the fluctuations are the noncommutative analogue of the diffeomorphisms-and-gauge-fields freedom. For an almost-commutative triple — the product $C^{\infty }(M)\otimes {\mathcal{A}}_F$ of a manifold triple with a finite-dimensional triple $({\mathcal{A}}_F,{\mathcal{H}}_F,D_F)$ — the inner fluctuations are exactly the gauge potentials of a Yang-Mills-Higgs theory, with $D_F$ supplying the fermion mass matrix. The spectral action $S=\mathrm{Tr}\chi (D_A^2/{\Lambda }^2)$, a function of the spectrum alone, expands via the heat-kernel coefficients into the Einstein-Hilbert action plus the bosonic Standard-Model Lagrangian, recovering gravity coupled to matter from a finite spectral geometry ^{[Connes 1996]}. This is the program's physical payoff: the Standard Model is the inner geometry ${\mathcal{A}}_F=\mathbb{C}\oplus \mathbb{H}\oplus M_3(\mathbb{C})$ of an almost-commutative manifold.

Reconstruction (Connes 2013). Let $(\mathcal{A},\mathcal{H},D)$ be a real spectral triple of $KO$-dimension $n$ that is commutative (with $\mathcal{A}$ commutative), regular, finite, of metric dimension $n$, satisfying the orientation and Poincaré-duality axioms together with the first-order condition and a "no junk" multiplicity/irreducibility condition. Then there is a closed oriented smooth spin manifold $M$ with $\mathcal{A}\cong C^{\infty }(M)$, and the triple is unitarily equivalent to the canonical spectral triple $(C^{\infty }(M),L^2(M,S),/D)$ for a Riemannian metric and spin structure determined by $(\mathcal{H},D)$. In words: manifolds are exactly commutative spectral geometries, and a noncommutative algebra obeying the same axioms is, by definition, a noncommutative manifold. ^{[Connes 2013; Gracia-Bondía-Várilly-Figueroa]}

Synthesis. The axioms are not an arbitrary checklist; each one is the foundational reason a piece of classical geometry survives the passage to spectra, and putting these together is exactly the content of the reconstruction theorem. Regularity is dual to smoothness; the orientation cycle is dual to the volume form and is what forces the metric dimension; the Dixmier-trace integral is the central insight that turns $|D{|}^{-n}$ into the Riemannian volume; and Poincaré duality through the real structure $J$ is the spectral face of the manifold's fundamental class. The reconstruction theorem is exactly the statement that these spectral conditions, in the commutative case, generalise nothing — they recover the manifold on the nose — while in the noncommutative case the same conditions generalise manifold geometry to objects with no points. This is dual to the Gelfand-Naimark theorem of 39.01.01: that result reconstructed the topological space from a commutative C*-algebra, and the reconstruction theorem reconstructs the smooth Riemannian spin structure from the commutative spectral triple, so the bridge from topology to geometry is precisely the upgrade from the algebra alone to the algebra-with-Dirac-operator.

Full proof set Master

Proposition (compact resolvent forces $|D|$ to have discrete, divergent spectrum). If $(D-\lambda {)}^{-1}$ is compact for one $\lambda \notin \sigma (D)$, then $D$ has purely discrete spectrum with finite multiplicities accumulating only at $\pm \infty$, and the eigenvalues ${\mu }_n$ of $|D|$ satisfy ${\mu }_n\to \infty$. Indeed, a compact self-adjoint operator $R=(D-\lambda {)}^{-1}$ has spectrum a sequence $\{r_k\}$ accumulating only at $0$ with finite-dimensional eigenspaces (spectral theorem for compact operators 02.11.03). The spectral mapping for resolvents, $\sigma (D)=\{\lambda +r_k^{-1}\}$, transports this to a sequence of eigenvalues of $D$ accumulating only at infinity; since $0$ is the only accumulation point of $\{r_k\}$, $|r_k^{-1}|\to \infty$, so ${\mu }_n=|\lambda +r_k^{-1}|\to \infty$. $\square$

Proposition (the commutator $[/D,f]$ is bounded and equals Clifford multiplication). For $f\in C^{\infty }(M)$ on a closed spin manifold and $\psi \in \Gamma (S)$, the Leibniz rule for ${\nabla }^S$ gives ${\nabla }^S(f\psi )=df\otimes \psi +f{\nabla }^S\psi$, whence $/D(f\psi )=c(df)\psi +f/D\psi$, i.e. $[/D,f]=c(df)$. Since $\Vert c(df)\Vert =\Vert |df|{\Vert }_{\infty }<\infty$ on a compact manifold, the commutator is bounded. Thus every $f\in C^{\infty }(M)$ has bounded commutator with $/D$, verifying the spectral-triple axiom for the canonical triple. $\square$

Proposition (metric dimension of the canonical triple equals $\dim M$). By Weyl's law for the elliptic operator $D^2=/D^2$ on a closed $n$-manifold, the eigenvalue counting function obeys $N(\Lambda )=\#\{{\mu }_k^2\le \Lambda \}\sim C_n\mathrm{vol}(M){\Lambda }^{n/2}$ as $\Lambda \to \infty$. Hence the eigenvalues of $|D|$ satisfy $N(\mu \le t)\sim C_n^{\prime }{t}^n$, so ${\mu }_k\sim c{k}^{1/n}$ and $|D{|}^{-1}\in {\mathcal{L}}^{n,\infty }$: the metric dimension is exactly $n=\dim M$. The Weyl asymptotics also pin the leading singularity of ${\zeta }_{|D|}(s)=\mathrm{Tr}|D{|}^{-s}$ at $s=n$, the top of the dimension spectrum. $\square$

Proposition (heat-kernel poles place the dimension spectrum on $\{n,n-1,\dots \}$). For the canonical triple, the small-time expansion $\mathrm{Tr}(b{e}^{-t{D}^2})\sim {\sum }_{k\ge 0}{a}_k(b)t^{(k-n)/2}$ holds with locally computable Seeley–DeWitt coefficients $a_k(b)={\int }_{M}b{\alpha }_{k}d\mathrm{vol}$. The Mellin transform $\mathrm{Tr}(b|D{|}^{-s})=\frac{1}{\Gamma (s/2)}{\int }_0^{\infty }{t}^{s/2-1}\mathrm{Tr}(b{e}^{-t{D}^2})dt$ converts each term $a_k(b)t^{(k-n)/2}$ into a simple pole of the meromorphic continuation at $s=n-k$, with residue $\frac{2}{\Gamma ((n-k)/2)}{a}_k(b)$. Hence $\mathrm{Sd}\subseteq \{n,n-1,\dots \}$, simple, and the top residue at $s=n$ is proportional to ${\int }_{M}bd\mathrm{vol}$. $\square$

Proposition (inner fluctuations preserve the spectral-triple axioms). Let $A=A^{\ast }={\sum }_i{a}_i[D,b_i]$ be a self-adjoint one-form and $D_A=D+A+{\epsilon }^{\prime }JA{J}^{-1}$. Then $D_A$ is self-adjoint (each summand is, using $JA{J}^{-1}$ self-adjoint and the sign ${\epsilon }^{\prime }$ from $JD={\epsilon }^{\prime }DJ$); $D_A$ differs from $D$ by a bounded operator, so $(D_A-\lambda {)}^{-1}=(D-\lambda {)}^{-1}(1+(A+{\epsilon }^{\prime }JA{J}^{-1})(D_A-\lambda {)}^{-1}{)}^{-1}$ remains compact by compactness of $(D-\lambda {)}^{-1}$ and boundedness of the second factor; and $[D_A,a]=[D,a]+[A+{\epsilon }^{\prime }JA{J}^{-1},a]$ is bounded because $A$ is bounded and, by the first-order condition, $[JA{J}^{-1},a]=0$. Thus $(\mathcal{A},\mathcal{H},D_A)$ is again a spectral triple — the fluctuations move within the space of geometries. $\square$

Proposition (the distance formula defines a genuine metric on the state space). Let $\rho (\phi ,\psi )=\sup \{|\phi (a)-\psi (a)|:\Vert [D,a]\Vert \le 1\}$ for states $\phi ,\psi$ on $A$. Symmetry and the triangle inequality are immediate from the supremum form. Non-negativity is clear; $\rho (\phi ,\psi )=0$ implies $\phi (a)=\psi (a)$ for all $a$ with $\Vert [D,a]\Vert \le 1$, hence (rescaling) for all $a$ with $[D,a]$ bounded, a set dense enough to separate states when the triple is irreducible, so $\phi =\psi$. The value may be $+\infty$ in general (the metric topology need not match the weak-$\ast$ topology unless the triple is, in Rieffel's sense, a compact quantum metric space). $\square$

Connections Master

• C-algebras and Gelfand-Naimark 39.01.01* — the commutative reconstruction of topology from a C*-algebra is the template; the spectral triple adds the Dirac operator $D$ to upgrade topological recovery to Riemannian and spin recovery, so $D$ supplies precisely the metric data Gelfand-Naimark omits.

• Dirac operator on a spin manifold 03.09.08 — the canonical commutative example is $(C^{\infty }(M),L^2(M,S),/D)$; the bounded-commutator axiom is $[/D,f]=c(df)$ and the compact-resolvent axiom is ellipticity, so the entire spin-geometry unit is the worked classical case of this one.

• Unbounded self-adjoint operators 02.11.03 — $D$ is unbounded and self-adjoint, and the compact-resolvent condition is exactly the spectral-theorem hypothesis that makes $|D|$ diagonalisable with discrete spectrum; the resolvent and spectral measure machinery is imported wholesale.

• Cyclic cohomology 39.07.01 — the Chern character of the Fredholm module $(\mathcal{H},D,\gamma )$ lands in cyclic cohomology, computing the index pairing that the Poincaré-duality axiom asserts; the orientation Hochschild cycle and the noncommutative integral live in this homological framework.

• Atiyah-Singer index theorem 03.09.10 — the local index formula of Connes-Moscovici is the noncommutative-geometry generalisation, expressing the index pairing through the dimension spectrum and the Wodzicki residue; the commutative case reduces to the Â-genus computation of the classical theorem.

• Yang-Mills action 03.07.05 — the inner fluctuations $D\mapsto D+A+JA{J}^{-1}$ of an almost-commutative triple are exactly gauge potentials, and the spectral action $\mathrm{Tr}\chi (D_A^2/{\Lambda }^2)$ expands into the Yang-Mills-Higgs Lagrangian; the gauge-field curvature of that unit is the bosonic content of the fluctuated Dirac operator here.

Historical & philosophical context Master

The program began with Connes' work on the classification of factors and foliations in the 1970s, where the index theory of operators forced an algebraic substitute for badly behaved quotient spaces. The decisive abstraction — a $\ast$-algebra, a Hilbert space, and an unbounded operator $D$ with the two axioms — appears in Connes' 1985 paper on noncommutative differential geometry and is consolidated in his 1994 book Noncommutative Geometry ^{[Connes Ch. VI]}, where the distance formula and the axioms of a noncommutative spin geometry are laid out. The seven axioms (regularity, dimension, finiteness, reality, first order, orientation, Poincaré duality) were stated in the 1996 paper Gravity coupled with matter ^{[Connes 1996]}, written partly to formulate the Standard Model as the inner geometry of an almost-commutative triple $C^{\infty }(M)\otimes {\mathcal{A}}_F$.

The converse — that the axioms characterise manifolds and nothing else — was the open problem the program had assumed. Connes proved the reconstruction theorem in On the spectral characterization of manifolds (2013) ^{[Connes 2013]}, showing that a commutative triple meeting the axioms is the canonical triple of a smooth compact spin manifold, after a correction supplied by Connes and others to the orientation/finiteness hypotheses identified by Rennie and Várilly. The textbook treatment by Gracia-Bondía, Várilly and Figueroa develops the real-spectral-triple framework and the $KO$-dimension table that organises the sign rules $(\epsilon ,{\epsilon }^{\prime },{\epsilon }^{\prime \prime })$ ^{[Gracia-Bondía-Várilly-Figueroa]}. The lineage runs from Kac's 1966 question of whether one can hear the shape of a drum, through Weyl's law and the Atiyah-Singer index theorem, to Connes' answer that one can hear far more than the shape: the entire spin Riemannian structure is audible in the Dirac spectrum.

Bibliography Master

• Connes, A., Noncommutative Geometry, Academic Press, 1994. Ch. VI.
• Connes, A., "Noncommutative differential geometry", Publ. Math. IHÉS 62 (1985), 257–360.
• Connes, A., "Gravity coupled with matter and the foundation of non-commutative geometry", Communications in Mathematical Physics 182 (1996), 155–176.
• Connes, A., "On the spectral characterization of manifolds", Journal of Noncommutative Geometry 7 (2013), 1–82.
• Connes, A. & Marcolli, M., Noncommutative Geometry, Quantum Fields and Motives, AMS Colloquium Publications 55, 2008. §1.
• Gracia-Bondía, J. M., Várilly, J. C. & Figueroa, H., Elements of Noncommutative Geometry, Birkhäuser, 2001. Ch. 9–11.
• Connes, A. & Moscovici, H., "The local index formula in noncommutative geometry", Geometric and Functional Analysis 5 (1995), 174–243.

---

Operator-algebras spine, noncommutative-geometry chapter. The central definitional unit of NCG: spectral triples, the distance formula, the seven axioms, and Connes' reconstruction theorem identifying commutative spectral geometries with spin manifolds. Builds on C-basics (39.01.01) and the classical Dirac operator (03.09.08).*