Spectral Triples and Reconstruction¶
Job: state the full Connes axiom system for a spectral triple (regularity, finiteness, orientability, Poincaré duality, real structure / KO-dimension, first-order condition), the reconstruction theorem identifying commutative triples with spin\(^c\) manifolds, and the worked distance formula on \(S^2\) plus the NC-torus state-space distance.
Core¶
Axioms of a spectral triple (Connes)¶
A (real, even) spectral triple is data \((\mathcal{A},H,D,J,\gamma)\) where:
- \(\mathcal{A}\) is an involutive algebra represented faithfully on a separable Hilbert space \(H\);
- \(D=D^*\) is an unbounded operator on \(H\) with compact resolvent (\((1+D^2)^{-1/2}\) compact);
- \([D,a]\) extends to a bounded operator for all \(a\in\mathcal{A}\);
- \(\gamma=\gamma^*\), \(\gamma^2=1\), \(\gamma D=-D\gamma\), \(\gamma a=a\gamma\) (the even grading; omit for odd triples);
- \(J\) is an antilinear isometry with \(J^2=\varepsilon\), \(JD=\varepsilon' DJ\), \(J\gamma=\varepsilon''\gamma J\), \(JaJ^{-1}=a^*\) for commutative \(\mathcal{A}\) (the real structure / charge conjugation).
The seven axioms (Connes, NCG Ch. VI + later refinements):
- Dimension / summability. \((\mathcal{A},H,D)\) is \(p\)-summable for some finite \(p\): \(|D|^{-1}\in\mathcal{L}^{p+}\); the dimension spectrum \(\Sigma\subset\mathbb{C}\) is discrete.
- Regularity. \(\mathcal{A}\) and \([D,\mathcal{A}]\) belong to the domain of all powers of \(|D|\); the smooth algebra \(\mathcal{A}=\bigcap_k\mathrm{Dom}(|D|^k)\) is a pre-Fréchet algebra under the seminorms \(q_k(a)=\|[D,a]\|+\||D|^k a|D|^{-k}\|\) (Lipschitz + spectral seminorms).
- Finiteness. The smooth algebra \(\mathcal{A}\) is finitely generated projective as an \(\mathcal{A}\)-module (i.e. \(H^\infty=\bigcap\mathrm{Dom}(D^k)\) is a finite projective \(\mathcal{A}\)-module with an \(\mathcal{A}\)-valued inner product), and \(\mathcal{A}\) is dense in \(H\).
- Orientability. There exists a Hochschild cycle \(c\in Z_n(\mathcal{A},\mathcal{A}^*)\) with \(\pi_D(c)=\gamma\) (or \(1\) in the odd case); equivalently \(\gamma=\sum_i a_i[J,b_iJ^{-1}]\) for \(a_i,b_i\in\mathcal{A}\) — a volume form in NC language.
- Poincaré duality. The intersection form \(\cap:K^\bullet(\mathcal{A})\otimes K^\bullet(\mathcal{A}^{\mathrm{op}})\to\mathbb{Z}\), \(\langle[D],\mathrm{ch}(e)\cap\mathrm{ch}(f^{\mathrm{op}})\rangle\), is nondegenerate (after tensoring with \(\mathbb{Q}\)).
- First-order condition. \([[D,a],JbJ^{-1}]=0\) for all \(a,b\in\mathcal{A}\) — the Dirac is a first-order operator that commutes with the right (opposite) action.
- Absolute continuity. The trace \(\tau(a)=\mathrm{Tr}_\omega(a|D|^{-n})\) (Dixmier trace / residue) is absolutely continuous w.r.t. the Lebesgue class on \(\hat A\) when commutative.
KO-dimension¶
The signs \((\varepsilon,\varepsilon',\varepsilon'')=(J^2,JD J^{-1}D,\gamma J\gamma J^{-1})\in\{\pm1\}^3\) are constrained to lie on a specific \(8\)-periodic table — the KO-dimension \(n\pmod 8\) of the triple. For a commutative spin manifold of dimension \(n\), the table matches the usual Clifford-algebra reality of the spinor bundle.
| KO-dim \(n\pmod 8\) | \(J^2\) | \(JD=\varepsilon'DJ\) | \(\gamma J=\varepsilon''J\gamma\) | Clifford reality |
|---|---|---|---|---|
| 0 | \(+1\) | \(+1\) | \(+1\) | \(\mathrm{Cl}_{0,0}\cong\mathbb{R}\) |
| 1 | \(-1\) | \(+1\) | (odd: no \(\gamma\)) | \(\mathrm{Cl}_{0,1}\cong\mathbb{C}\) |
| 2 | \(-1\) | \(-1\) | \(-1\) | \(\mathrm{Cl}_{0,2}\cong\mathbb{H}\) |
| 3 | \(+1\) | \(-1\) | (odd) | \(\mathrm{Cl}_{0,3}\cong\mathbb{H}\oplus\mathbb{H}\) |
| 4 | \(+1\) | \(+1\) | \(+1\) | \(\mathrm{Cl}_{0,4}\cong M_2(\mathbb{H})\) |
| 5 | \(-1\) | \(+1\) | (odd) | \(\mathrm{Cl}_{0,5}\cong M_4(\mathbb{C})\) |
| 6 | \(-1\) | \(-1\) | \(-1\) | \(\mathrm{Cl}_{0,6}\cong M_8(\mathbb{R})\) |
| 7 | \(+1\) | \(-1\) | (odd) | \(\mathrm{Cl}_{0,7}\cong M_8(\mathbb{R})\oplus M_8(\mathbb{R})\) |
The signs are not free: a real spectral triple of KO-dimension \(n\) has exactly the row of the table. The KO-dimension governs which Clifford module the spinors form and is essential for the spectral action / Standard Model KO-dim \(6\) of the finite geometry.
Reconstruction theorem (Connes)¶
Theorem (Connes reconstruction). Let \((\mathcal{A},H,D,J,\gamma)\) be a commutative real even spectral triple satisfying (1)–(7) above with \(\mathcal{A}\) commutative. Then there exists a compact spin\(^c\) Riemannian manifold \(M\) (unique up to spin\(^c\) diffeomorphism) such that - \(\mathcal{A}\cong C^\infty(M)\) (as Fréchet algebras); - \(H\cong L^2(M,S)\) (spinor square-integrable sections of the spinor bundle \(S\to M\)); - \(D\) is a Dirac-type operator on \(S\), \(D=c\circ\nabla^S\) for a Clifford connection \(\nabla^S\); - \(J\) is the charge-conjugation / real structure on spinors, \(\gamma\) the chirality.
Conversely, the canonical triple of a compact spin\(^c\) manifold satisfies the axioms.
The commutative case is rigid — manifolds are exactly the commutative spectral triples satisfying the full axiom set.
Role of each axiom. Dropping any one permits non-manifold commutative triples: - Without Poincaré duality: commutative triples on singular spaces (orbifolds, weighted projective lines) satisfy the other axioms but not Poincaré duality. - Without the first-order condition \([[D,a],JbJ^{-1}]=0\): the Dirac could have non-first-order "potential" terms that break the geometric interpretation; the inner fluctuations would not close as a connection. - Without finiteness: \(H^\infty\) need not be a finite projective module, so the "spinors need not be a finite-rank bundle". - Without orientability: no volume form / no Hochschild cycle realizing \(\gamma\).
Examples at frontier depth¶
(a) Canonical triple of a compact spin manifold. For \((M^n,g)\) compact spin with spinor bundle \(S\to M\) and Dirac \(D\!\!\!/=c\circ\nabla^S\), $$ \mathcal{A}=C^\infty(M),\quad H=L^2(M,S),\quad D=D!!!/. $$ Reconstruction gives \(M\) back. The dimension spectrum is \(\Sigma=\{n,n-1,n-2,\dots,0\}\) with multiplicities (the eigenvalues of \(|D|^2\) are sums of eigenvalues of the Laplacian on spinors, growing as \(k^{1/n}\)). The distance formula \(d(p,q)=\sup\{|f(p)-f(q)|:\|[D,f]\|\le 1\}\) recovers geodesic distance.
(b) The noncommutative torus \(A_\theta\). Smooth subalgebra \(A_\theta^\infty\) of rapid-decay Fourier series; derivations \(\delta_1,\delta_2\); the Connes–Rieffel Dirac $$ D=\delta_1\otimes\gamma_1+\delta_2\otimes\gamma_2 $$ on \(H=A_\theta^\infty\otimes\mathbb{C}^2\) (the \(\gamma_i\) are Pauli matrices). Dimension spectrum \(\Sigma=\{2,1,0\}\) (simple). The triple is regular, finitely generated, orientable, satisfies Poincaré duality, real structure \(J\) by the involution \(U\mapsto U^{-1}\), \(V\mapsto V^{-1}\). Reconstruction fails (noncommutative \(\mathcal{A}\)) — this is the prototypical genuinely NC Riemannian geometry.
(c) Almost-commutative geometries. Let \(A_F\) be a finite-dimensional real \(C^*\)-algebra (e.g. \(A_F=M_N(\mathbb{C})\) or a sum). The product triple $$ \bigl(C^\infty(M)\otimes A_F,\ L^2(S)\otimes\mathbb{C}^N,\ D!!!/\otimes 1+\gamma_5\otimes D_F\bigr) $$ with finite Dirac \(D_F\) a self-adjoint matrix. Inner fluctuations (= gauge potentials) replace $$ D\to D_A=D+A+JAJ^{-1},\qquad A=\sum_i a_i[D,b_i] $$ recovering the Yang–Mills connection: \(A\) is the gauge potential, \(JAJ^{-1}\) its charge-conjugate, and the fluctuated Dirac \(D_A\) encodes the covariant derivative. This is the prototype for the NC Standard Model — see spectral action.
See it¶
Distance formula on \(S^2\) recovering geodesic distance. Let \(M=S^2\) with the round metric, \(\mathcal{A}=C^\infty(S^2)\), \(D=D\!\!\!/\) the canonical spin Dirac. For \(f\in C^\infty(S^2)\), $$ [D,f]=c(df) $$ is Clifford multiplication by the differential \(df\). Clifford multiplication on a Riemannian spin manifold is an isometry fiberwise: $$ |c(df)|=\sup_{\xi\in S,\,|\xi|=1}|c(df)\xi|=|df|g, $$ hence \(\|[D,f]\|=\|df\|_\infty=\mathrm{Lip}(f)\), the Lipschitz constant. The Connes distance between points \(p,q\in S^2\) (pure states \(f\mapsto f(p)\), \(f\mapsto f(q)\)) is therefore $$ d(p,q)=\sup{|f(p)-f(q)|:|[D,f]|\le 1}=\sup{|f(p)-f(q)|:\mathrm{Lip}(f)\le 1}=d(p,q), $$ the geodesic (great-circle) distance — the classical Hopf–Rinow/Kantorovich duality between Lipschitz functions and geodesic distance.}
Contrast: NC torus state-space distance. For \(A_\theta\) with the Connes–Rieffel Dirac, the distance is defined on the state space \(S(A_\theta)\) (no underlying points), \(d(\varphi,\psi)=\sup\{|\varphi(a)-\psi(a)|:\|[D,a]\|\le 1\}\), computed via the derivations \(\delta_i\) giving \(\|[D,a]\|\) in terms of Fourier coefficients. The trace \(\tau\) is at distance \(\sim\theta\) from the pure states; the geometry is genuinely non-pointwise but the formula remains computable (Rieffel, Rennie–Varilly).
Check¶
Show that dropping Poincaré duality allows a commutative spectral triple that is not a manifold.
Take \(M\) a compact orbifold with isolated singularities, \(\mathcal{A}=C^\infty(M)\) (orbifold smooth functions), \(H=L^2\) of orbifold spinors, \(D\) the orbifold Dirac. Axioms 1–4, 6, 7 hold (regularity, finiteness for the orbifold module, orientability via the orbifold volume form, first-order condition) but the intersection form on the orbifold \(K\)-homology is degenerate at the singular strata — Poincaré duality fails (orbifolds are rational homology manifolds, not integral ones). Hence reconstruction cannot apply and indeed \(M\) is not a smooth manifold.
Why is the first-order condition \([[D,a],JbJ^{-1}]=0\) essential for inner fluctuations to define a connection?
Inner fluctuations \(D\to D+A+JAJ^{-1}\) with \(A=\sum a_i[D,b_i]\) must commute with the right (opposite) action \(\mathcal{A}^{\mathrm{op}}\) (acting via \(J\)), so that \(A\) is a left connection 1-form. The first-order condition guarantees \([A,JbJ^{-1}]=0\), hence the fluctuated Dirac \(D_A\) is still a first-order operator with the correct commutator structure \([D_A,a]\) — without it the fluctuated \(D_A\) would have a non-connection piece breaking gauge invariance.
Pitfalls¶
- Reconstruction requires ALL the axioms — drop Poincaré duality or the first-order condition and non-manifold commutative triples exist. Do not cite "reconstruction" for an arbitrary commutative triple.
- The real structure \(J\) encodes KO-dimension (the signs \((J^2,\varepsilon,\varepsilon')\) are pinned to the \(n\bmod 8\) table); they are not free parameters and changing them changes the Clifford reality class.
- Almost-commutative \(\neq\) small perturbation. The finite algebra \(A_F\) changes the gauge group entirely (e.g. \(A_F=\mathbb{C}\oplus\mathbb{H}\oplus M_3(\mathbb{C})\) gives \(\mathrm{U}(1)\times\mathrm{SU}(2)\times\mathrm{SU}(3)\), not \(\mathrm{U}(1)\)). The "almost" refers to the product structure \(C^\infty(M)\otimes A_F\), not to a small deformation.
- Dimension spectrum must be simple/discrete for the local index formula (see cyclic cohomology); multiple eigenvalues of \(|D|\) or higher-order poles need the refined residue-with-log formula.
- Finiteness is "finitely generated projective as a module", not "finite-dimensional as a vector space" — for the canonical triple, \(H^\infty=C^\infty(M,S)\) is an infinite-dimensional vector space but a finite-rank projective \(C^\infty(M)\)-module (rank \(=\) spinor rank).
- The distance formula needs \([D,a]\) bounded for all \(a\) in the smooth subalgebra, which is a strong regularity condition; for a generic unbounded operator the Lipschitz seminorm may be infinite.
Refs¶
- Connes, Noncommutative Geometry, Ch. VI (definition of spectral triples), Ch. XI (reconstruction)
- Connes, "Gravity coupled with matter and the foundation of noncommutative geometry" (CMP 1996)
- Rennie–Varilly, Noncommutative Geometry and the Standard Model; Gracia-Bondía–Várilly–Figueroa, Elements of NCG
- Rieffel, "Metrics on state spaces" (Doc. Math. 1998); Rennie, "Smoothness and quotient equivalence of spectral triples"
Next¶
NCG index · Foundations · Cyclic Cohomology & Local Index · Baum–Connes Assembly · Spectral Action & Physics · Operator algebras · Differential geometry · K-theory