39.06.02 · operator-algebras / spectral-triples-ncg

The Connes Distance Formula

shipped3 tiersLean: none

Anchor (Master): Connes *Noncommutative Geometry* (1994) Ch. VI; Connes 'Compact metric spaces, Fredholm modules, and hyperfiniteness' (1989); Rieffel 'Metrics on state spaces' (1999)

Intuition Beginner

Distance is usually something you measure with a ruler laid along the ground. Noncommutative geometry measures it the opposite way: not by walking between two places, but by asking how far apart you can make the readings of a thermometer at those two places, while forbidding the temperature from changing too fast as you move. If the temperature is only allowed to rise by at most one degree per metre, then the difference in readings between two spots can never exceed the number of metres between them. Push that difference as high as it will go, and you have measured the distance.

The "speed limit on change" is supplied by one special operator, the same Dirac operator that ran the spectral triple 39.06.01. Comparing a function against that operator tells you the function's steepest slope. So the recipe is: among all functions whose slope stays under control, find the pair of readings at the two points that are spread as far apart as possible. That largest spread is the distance.

The payoff is that this recipe never mentions a path, a curve, or even points. You can run it on a "space" that has no points at all — separating two states of a noncommutative algebra instead — so distance survives into worlds where ordinary geometry breaks down.

Visual Beginner

The distance between two points is the largest gap in readings you can open up using a function whose slope never exceeds one.

The dictionary reads: the operator caps how fast a function may change; capping the change is the same as a speed limit; a traveller obeying the speed limit covers at most the distance between the points; and the function that uses the full speed limit the whole way opens a gap exactly equal to the distance.

Worked example Beginner

Take the line segment from $0$ to $1$ with ordinary distance, and cap the slope of every function at one. Pick the two endpoints, $0$ and $1$ . Which function spreads their readings the furthest while keeping its slope no larger than one?

Try the function that just copies the position: its value at a point $x$ is $x$ itself. Its slope is exactly $1$ everywhere, right at the cap. Its reading at the point $1$ is $1$ , and its reading at the point $0$ is $0$ , so the gap between the two readings is $1 - 0 = 1$ .

Can any allowed function do better? A function with slope capped at $1$ can climb by at most $1$ unit of height over the $1$ unit of distance from $0$ to $1$ . So no allowed function opens a gap larger than $1$ . The position function already reaches that ceiling.

What this tells us: the largest achievable gap is $1$ , which is exactly the distance from $0$ to $1$ . The formula returned the right distance, and it did so by squeezing functions against a slope cap rather than by measuring along the segment.

Check your understanding Beginner

Formal definition Intermediate+

Let $(A, H, D)$ be a spectral triple 39.06.01: a unital $*$ -algebra $A$ represented faithfully by bounded operators on a Hilbert space $H$ , with $D$ self-adjoint, of compact resolvent, and $[D, a]$ bounded for every $a \in A$ . Write $A = \overline{A}$ for the C*-algebra completion, and let $S (A)$ denote its state space: the set of positive linear functionals $φ : A \to C$ with $φ (1) = 1$ , a weak- $*$ compact convex set whose extreme points are the pure states 39.01.02.

Define the Lipschitz seminorm $L : A \to [0, \infty]$ by $$ L(a) = \big| [D, a] \big|, $$ the operator norm of the commutator. Because $[D, λ 1] = 0$ for scalars and $[D, \cdot]$ is a derivation, $L$ is a seminorm with kernel containing the scalars $C 1$ . The Connes distance between two states $φ, ψ \in S (A)$ is $$ d_D(\varphi, \psi) ;=; \sup\Big{, |\varphi(a) - \psi(a)| ;:; a \in \mathcal{A},\ L(a) = \big| [D, a] \big| \le 1 ,\Big} ;\in; [0, +\infty]. $$ The supremum may be $+ \infty$ (for instance when $φ, ψ$ lie in different connected pieces of a disconnected geometry), so $d_{D}$ is an extended metric: a function valued in $[0, + \infty]$ satisfying symmetry, the triangle inequality, and $d_{D} (φ, ψ) = 0 ⟺ φ = ψ$ whenever the representation is such that ${a : L (a) = 0} = C 1$ (the irreducible / connected case) ^{[Connes 1989]}.

Two reductions are routine. First, the supremum is unchanged if restricted to self-adjoint $a = a^{*}$ : writing $a = a_{1} + i a_{2}$ with $a_{j}$ self-adjoint, $φ (a) - ψ (a)$ has real and imaginary parts coming from $a_{1}, a_{2}$ , each with $L (a_{j}) \leq L (a)$ since $[D, a^{*}] = - [D, a]^{*}$ gives $∥ [D, a_{j}] ∥ \leq ∥ [D, a] ∥$ . Second, the kernel of $L$ being the scalars means adding a constant to $a$ changes neither $L (a)$ nor $φ (a) - ψ (a)$ , so one may normalise $φ (a) = 0$ .

The motivating commutative case is the canonical triple $(C^{\infty} (M), L^{2} (M, S), \neq D)$ of a closed Riemannian spin manifold $(M, g)$ 39.06.01. The pure states of $A = C (M)$ are exactly the evaluations $ev_{p}$ at points $p \in M$ (Gelfand duality 39.01.02); these are the Dirac masses $δ_{p}$ . The commutator $[\neq D, f] = c (df)$ is Clifford multiplication by the differential, and $∥ c (df) ∥ = ∥∣ df ∣_{g} ∥_{\infty}$ , so $L (f) = ∥∣ df ∣ ∥_{\infty} = Lip_{g} (f)$ is the Lipschitz constant of $f$ for the geodesic distance.

Counterexamples to common slips

The constraint is $∥ [D, a] ∥ \leq 1$ , not $∥ a ∥ \leq 1$ . Bounding $∥ a ∥$ would bound $∣ φ (a) - ψ (a) ∣ \leq 2$ and produce a fixed diameter rather than a metric; it is the commutator norm — the "slope" — that must be capped, and the value $∣ φ (a) - ψ (a) ∣$ is then free to be large.
$d_{D}$ is an extended metric, not a metric: $+ \infty$ is a legitimate value. For a disconnected manifold no $1$ -Lipschitz function is constrained between the components, so points in different components are at infinite distance. Finiteness on all of $S (A)$ is an extra hypothesis (Rieffel's compact-quantum-metric-space condition), not automatic.
The distance is computed between states, not algebra elements. The points $p, q \in M$ enter as the pure states $ev_{p}, ev_{q}$ ; the algebra elements $a$ are the test functions in the supremum. Confusing the two — treating $a$ as a point — inverts the roles in the formula.

Key theorem with proof Intermediate+

Theorem (Connes distance formula; commutative reduction). Let $(C^{\infty} (M), L^{2} (M, S), \neq D)$ be the canonical spectral triple of a closed Riemannian spin manifold $(M, g)$ , and let $p, q \in M$ with associated pure states (Dirac masses) $δ_{p} = ev_{p}$ , $δ_{q} = ev_{q}$ . Then $$ d_{\not{!D}}(\delta_p, \delta_q) ;=; \sup\Big{, |f(p) - f(q)| ;:; f \in C^\infty(M),\ \big| [\not{!D}, f] \big| \le 1 ,\Big} ;=; d_g(p, q), $$ the geodesic distance between $p$ and $q$ . The metric on pure states recovered from $D$ is exactly the Riemannian distance. ^{[Connes Ch. VI §1; Connes 1989]}

Proof. By the self-adjoint reduction it suffices to take $f$ real-valued. The key identity is that the commutator of the Dirac operator with a function is Clifford multiplication by its differential, $$ [\not{!D}, f] = c(df), $$ because $\neq D = c \circ \nabla^{S}$ obeys the Leibniz rule and the connection terms cancel against each other in the commutator 39.06.01. Clifford multiplication by a one-form has operator norm equal to its pointwise Riemannian length, $∥ c (ω_{x}) ∥ = ∣ ω_{x} ∣_{g}$ , fibre by fibre, so $$ \big| [\not{!D}, f] \big| = \sup_{x \in M} |df_x|g = \big| |df| \big|\infty = \operatorname{Lip}_g(f). $$ Hence the constraint $∥ [\neq D, f] ∥ \leq 1$ is precisely the constraint that $f$ be $1$ -Lipschitz for the geodesic distance.

Upper bound. Let $f$ be $1$ -Lipschitz and let $γ : [0, ℓ] \to M$ be any unit-speed path from $p$ to $q$ of length $ℓ$ . Then $$ |f(p) - f(q)| = \Big| \int_0^\ell \frac{d}{dt} f(\gamma(t)), dt \Big| = \Big| \int_0^\ell df_{\gamma(t)}(\dot\gamma(t)), dt \Big| \le \int_0^\ell |df_{\gamma(t)}|_g, dt \le \ell. $$ Taking the infimum over paths gives $∣ f (p) - f (q) ∣ \leq d_{g} (p, q)$ , so the supremum over $1$ -Lipschitz $f$ is at most $d_{g} (p, q)$ .

Lower bound. Consider the distance-to- $p$ function $f_{0} (x) = d_{g} (p, x)$ . The triangle inequality $∣ d_{g} (p, x) - d_{g} (p, y) ∣ \leq d_{g} (x, y)$ shows $f_{0}$ is $1$ -Lipschitz, and $f_{0} (p) = 0$ , $f_{0} (q) = d_{g} (p, q)$ . The function $f_{0}$ is smooth away from $p$ and the cut locus; mollifying it on that small set produces a smooth $1$ -Lipschitz $f_{ϵ} \in C^{\infty} (M)$ with $∣ f_{ϵ} (p) - f_{ϵ} (q) ∣ \geq d_{g} (p, q) - ϵ$ . Letting $ϵ \to 0$ shows the supremum is at least $d_{g} (p, q)$ .

Combining the two bounds, the supremum equals $d_{g} (p, q)$ . $□$

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 integration data that complements this degree-zero metric. The foundational reason it works is exactly the identity $∥ [\neq D, f] ∥ = Lip_{g} (f)$ : the single operator $D$ converts the metric question "how far apart are $p$ and $q$ ?" into the algebraic question "how far apart can a slope-capped function spread their values?", and this is exactly the dual of the supremum that defines the Lipschitz norm in metric geometry. The construction generalises the Gelfand picture of 39.01.02: there the pure states recover the points of a commutative algebra as a topological space, and here the operator $D$ upgrades that topological recovery to a metric one, so putting these together the algebra supplies the points and the commutator with $D$ supplies the distances. The bridge is that $D$ is to the Riemannian metric what the algebra is to the topology — the same slogan that organises the reconstruction theorem 39.06.01, now made quantitative.

Exercises Intermediate+

Exercise 4 (medium, numeric).

Consider the two-point space with algebra $A = C \oplus C$ , $H = C^{2}$ , and Dirac operator $D = (0 \overset{m}{ˉ} m 0)$ with $m \in C$ , $m \neq = 0$ , acting with $a = diag (a_{1}, a_{2})$ . Compute $∥ [D, a] ∥$ in terms of $∣ a_{1} - a_{2} ∣$ and $∣ m ∣$ , then give the Connes distance between the two pure states $δ_{1}, δ_{2}$ .

Hint

The commutator $[D, a]$ is off-diagonal with entries $\pm m (a_{2} - a_{1})$ (up to conjugation); its operator norm is the magnitude of that entry. Then maximise $∣ a_{1} - a_{2} ∣$ subject to $∥ [D, a] ∥ \leq 1$ .

Answer

A direct computation gives $[D, a] = (0 \overset{m}{ˉ} (a_{1} - a_{2}) m (a_{2} - a_{1}) 0)$ , an off-diagonal matrix whose operator norm is $∥ [D, a] ∥ = ∣ m ∣ ∣ a_{1} - a_{2} ∣$ . The pure states are $δ_{i} (a) = a_{i}$ , so $∣ δ_{1} (a) - δ_{2} (a) ∣ = ∣ a_{1} - a_{2} ∣$ . Maximising under $∣ m ∣ ∣ a_{1} - a_{2} ∣ \leq 1$ gives $∣ a_{1} - a_{2} ∣ \leq 1/∣ m ∣$ , hence $$ d_D(\delta_1, \delta_2) = \frac{1}{|m|}. $$ The discrete two-point geometry has a finite distance set by the off-diagonal "mass" $m$ . Rubric: full credit for $∥ [D, a] ∥ = ∣ m ∣∣ a_{1} - a_{2} ∣$ and $d = 1/∣ m ∣$ .

Exercise 5 (medium, short-answer).

Show that $d_{D}$ satisfies the triangle inequality on the state space.

Hint

For any fixed admissible $a$ , the quantity $∣ φ (a) - ψ (a) ∣$ already obeys the triangle inequality; take the supremum afterward.

Answer

Fix states $φ, ψ, χ$ and any $a$ with $∥ [D, a] ∥ \leq 1$ . Then $$ |\varphi(a) - \psi(a)| \le |\varphi(a) - \chi(a)| + |\chi(a) - \psi(a)| \le d_D(\varphi, \chi) + d_D(\chi, \psi), $$ the last step because each term on the right is bounded by the corresponding supremum over admissible $a$ . The right-hand side is independent of $a$ , so taking the supremum of the left-hand side over admissible $a$ preserves the inequality: $d_{D} (φ, ψ) \leq d_{D} (φ, χ) + d_{D} (χ, ψ)$ .

Rubric: full credit for using the pointwise triangle inequality before passing to the supremum.

Exercise 6 (hard, short-answer).

For the two-point space of Exercise 4, replace the pure states by mixed states $φ_{t} = t δ_{1} + (1 - t) δ_{2}$ and $ψ_{s} = s δ_{1} + (1 - s) δ_{2}$ , $t, s \in [0, 1]$ . Show that $d_{D} (φ_{t}, ψ_{s}) = ∣ t - s ∣/∣ m ∣$ , and interpret the result.

Hint

$φ_{t} (a) = t a_{1} + (1 - t) a_{2}$ , so $φ_{t} (a) - ψ_{s} (a) = (t - s) (a_{1} - a_{2})$ ; reuse the commutator-norm computation.

Answer

With $a = diag (a_{1}, a_{2})$ , $φ_{t} (a) = t a_{1} + (1 - t) a_{2}$ and likewise for $ψ_{s}$ , so $$ \varphi_t(a) - \psi_s(a) = (t - s)(a_1 - a_2). $$ From Exercise 4, $∥ [D, a] ∥ = ∣ m ∣ ∣ a_{1} - a_{2} ∣ \leq 1$ allows $∣ a_{1} - a_{2} ∣$ up to $1/∣ m ∣$ , whence $$ d_D(\varphi_t, \psi_s) = |t - s| \cdot \frac{1}{|m|}. $$ Interpretation: the whole state space of $C \oplus C$ is the segment of probability mixtures ${(t, 1 - t)}$ , and the Connes metric makes it an honest interval of length $1/∣ m ∣$ — a line segment, not just two points. The internal "two-point" geometry is a continuum once mixed states are included, which is exactly the picture behind the NCG Standard Model's internal Higgs distance.

Rubric: full credit for $d = ∣ t - s ∣/∣ m ∣$ and the segment interpretation.

Exercise 7 (hard, short-answer).

State the Kantorovich–Rubinstein duality for the Wasserstein- $1$ distance between probability measures on a metric space $(X, d)$ , and explain in what sense the Connes distance restricted to all states (not just pure states) is its noncommutative analogue.

Hint

Kantorovich–Rubinstein writes the optimal-transport cost as a supremum over $1$ -Lipschitz functions; compare its form to the Connes supremum.

Answer

The Kantorovich–Rubinstein theorem gives, for probability measures $μ, ν$ on $(X, d)$ , $$ W_1(\mu, \nu) = \sup\Big{ \int f, d\mu - \int f, d\nu ;:; \operatorname{Lip}(f) \le 1 \Big}, $$ the Wasserstein- $1$ (earth-mover) distance written as a supremum over $1$ -Lipschitz test functions. In the commutative case $A = C (X)$ , a state is exactly a probability measure $μ$ (Riesz representation), $φ (f) = \int f d μ$ , and the Lipschitz constraint $Lip (f) \leq 1$ is $∥ [D, f] ∥ \leq 1$ . So the Connes distance on all states is $W_{1}$ : $$ d_D(\mu, \nu) = \sup{ |\textstyle\int f,d\mu - \int f,d\nu| : |[D,f]| \le 1 } = W_1(\mu, \nu). $$ The Connes formula is therefore the noncommutative Monge–Kantorovich dual: the supremum-over-Lipschitz-functions side of optimal transport, with the transport (Monge) side replaced by states on a possibly noncommutative algebra. On pure states it restricts to the ground metric $d$ itself ( $W_{1} (δ_{p}, δ_{q}) = d (p, q)$ ).

Rubric: full credit for the Kantorovich–Rubinstein formula and the identification of states with measures making $d_{D} = W_{1}$ .

Lean formalization Intermediate+

lean_status: none — Mathlib has metric spaces, ℝ≥0∞-valued extended pseudometrics, the operator norm, the state space of a C*-algebra as a WeakDual subset, and the basic theory of bounded and unbounded operators, but it does not assemble the Lipschitz-seminorm $L (a) = ∥ [D, a] ∥$ and the associated state-space metric into a single named construction, nor does it carry Rieffel's compact-quantum-metric-space axioms or the commutative identification $∥ [\neq D, f] ∥ = Lip_{g} (f)$ .

The intended statement reads schematically:

import Mathlib.Topology.MetricSpace.Basic
import Mathlib.Analysis.CStarAlgebra.ContinuousFunctionalCalculus.Basic

variable {A H : Type*} [CStarAlgebra A]
  [NormedAddCommGroup H] [InnerProductSpace ℂ H]

/-- The Connes distance on the state space induced by a Lipschitz seminorm
`L a = ‖[D, a]‖`. (Schematic: `L` and the unbounded `D` need the not-yet-bundled
NCG infrastructure; the result is an `ℝ≥0∞`-valued extended metric.) -/
noncomputable def connesDist (L : A → ℝ≥0∞) (φ ψ : stateSpace A) : ℝ≥0∞ :=
  ⨆ (a : A) (_ : L a ≤ 1), ENNReal.ofReal |φ.toFun a - ψ.toFun a|

The commutative theorem — that for the canonical triple of a spin manifold this recovers the geodesic distance between Dirac masses — needs the spin-geometry and elliptic stack and is a far-horizon target alongside the spectral-triple bundling of 39.06.01.

Advanced results Master

The metric topology and compact quantum metric spaces. The extended metric $d_{D}$ on $S (A)$ need not be finite, and even when finite it need not induce the weak- $*$ topology. Rieffel isolated the conditions under which it does: a Lip-norm is a seminorm $L$ on a dense subspace of the self-adjoint part of $A$ , with kernel exactly the scalars, such that the associated $d_{D}$ is finite and metrizes the weak- $*$ topology on $S (A)$ . The pair $(A, L)$ is then a compact quantum metric space ^{[Rieffel 1999]}. For the canonical triple of a closed manifold $L (f) = Lip_{g} (f)$ is a Lip-norm, recovering the Wasserstein- $1$ metric on probability measures and, on pure states, the geodesic distance. The Lip-norm axiom is the correct noncommutative formulation of "bounded diameter plus the metric recovers the topology", and it is what makes Gromov–Hausdorff convergence of noncommutative spaces meaningful.

Pure states, mixed states, and Monge–Kantorovich duality. On the commutative algebra $C (X)$ the states are the probability measures on $X$ (Riesz representation), pure states are the Dirac masses, and the distance formula reads $$ d_D(\mu, \nu) = \sup\Big{ \big| \textstyle\int f, d\mu - \int f, d\nu \big| : |[D, f]| \le 1 \Big} = W_1(\mu, \nu), $$ the Kantorovich–Rubinstein dual of the Wasserstein- $1$ optimal-transport cost. The Connes distance is therefore the genuine noncommutative generalisation of optimal transport: the Lipschitz-test-function (Kantorovich) side survives verbatim when $A$ is noncommutative, while the Monge side — couplings of measures — has no literal analogue because there are no points to transport between. On pure states the duality restricts to the ground geodesic distance, and the convex geometry of $S (A)$ extends this to all mixtures.

Finite spectral triples and the internal distance. For a finite-dimensional triple $(A_{F}, H_{F}, D_{F})$ with $A_{F}$ a sum of matrix algebras, $S (A_{F})$ is a finite-dimensional convex body and $d_{D_{F}}$ is computed by finite-dimensional optimisation. The two-point case $A_{F} = C \oplus C$ with off-diagonal $D_{F} = (0 \overset{m}{ˉ} m 0)$ gives $d (δ_{1}, δ_{2}) = 1/∣ m ∣$ , turning the discrete pair into a segment of length $1/∣ m ∣$ once mixed states are admitted. In the noncommutative Standard Model the internal algebra $A_{F} = C \oplus H \oplus M_{3} (C)$ carries a $D_{F}$ built from the fermion Yukawa matrices, and the resulting internal distances are governed by the inverse mass scales; the Higgs field appears as the "coordinate" along the internal segment, and the spectral distance between the two sheets of $M \times {1, 2}$ is set by the top-quark Yukawa coupling ^{[Iochum-Krajewski-Martinetti 2001]}. The Higgs potential is then the curvature cost of varying this internal distance over $M$ .

Distance in the product geometry. For an almost-commutative triple $C^{\infty} (M) \otimes A_{F}$ with product Dirac operator $D = \neq D \otimes 1 + γ \otimes D_{F}$ , the spectral distance between two points-times-internal-states is not in general the Pythagorean sum of the geodesic and internal distances. The orthogonality of the manifold and internal directions in $D$ does yield a Pythagorean lower bound, and equality holds when the optimal test function factorises; but the constraint $∥ [D, a] ∥ \leq 1$ couples the two commutators $[\neq D, \cdot]$ and $[γ \otimes D_{F}, \cdot]$ through the single operator-norm bound, so the exact distance can be strictly larger than the naive sum and requires solving the coupled variational problem. This subtlety is the reason explicit Standard-Model distance computations are delicate.

Failure of finiteness and the role of irreducibility. If the kernel of $L$ exceeds the scalars — equivalently if some non-scalar $a$ has $[D, a] = 0$ — then $d_{D}$ fails to separate states and is only a pseudometric: states agreeing on the commutant of $D$ are at distance zero. Irreducibility of the representation, or more precisely the condition ${a \in A : [D, a] = 0} = C 1$ , restores the separation property. This is the metric counterpart of the connectedness needed for the reconstruction theorem 39.06.01: a reducible $D$ describes a disconnected or degenerate geometry on which distances collapse or diverge.

Synthesis. The distance formula is the central insight that the metric of a geometry is encoded in a single operator, and putting these together with the reconstruction theorem 39.06.01 gives the full dictionary: the algebra recovers the topology, the Dirac operator recovers the metric, and the supremum form is exactly the Kantorovich dual that makes "distance" survive the loss of points. The foundational reason the construction is robust is that it never refers to paths — it is built from $∥ [D, a] ∥ \leq 1$ and the linear functionals $φ$ , both of which generalise verbatim — so the formula is dual to optimal transport on the commutative side and generalises it on the noncommutative side. The Lip-norm axioms of Rieffel are what guarantee this metric recovers the weak- $*$ topology, so the commutative case is exactly Gelfand duality 39.01.02 with a metric refinement; and the finite-dimensional internal distances are the bridge to physics, where the inverse Yukawa masses become the internal lengths and the Higgs is the coordinate measuring them. The central insight is that one self-adjoint operator simultaneously carries dimension (its eigenvalue growth), integration (its Dixmier trace), and now metric (its commutators) — three faces of the same spectral datum.

Full proof set Master

Proposition (the distance formula defines a genuine extended metric on the state space). Let $ρ (φ, ψ) = sup {∣ φ (a) - ψ (a) ∣ : a \in A, ∥ [D, a] ∥ \leq 1} \in [0, + \infty]$ for states $φ, ψ$ on $A$ . Symmetry is immediate, since the admissible set ${a : ∥ [D, a] ∥ \leq 1}$ is closed under $a \mapsto - a$ and $∣ φ (a) - ψ (a) ∣ = ∣ φ (- a) - ψ (- a) ∣$ with the roles of $φ, ψ$ exchangeable. The triangle inequality holds because for each fixed admissible $a$ , $∣ φ (a) - ψ (a) ∣ \leq ∣ φ (a) - χ (a) ∣ + ∣ χ (a) - ψ (a) ∣ \leq ρ (φ, χ) + ρ (χ, ψ)$ , and the right side, independent of $a$ , dominates the supremum. Non-negativity is clear, and $φ = ψ$ forces $ρ = 0$ . Conversely $ρ (φ, ψ) = 0$ means $φ (a) = ψ (a)$ for every $a$ with $∥ [D, a] ∥ \leq 1$ ; rescaling, $φ (a) = ψ (a)$ for every $a$ with $[D, a]$ bounded, i.e. for all of $A$ , and since $A$ is weak- $*$ dense in $A$ and states are weak- $*$ continuous, $φ = ψ$ . The value $+ \infty$ is admissible, so $ρ$ is an extended metric. $□$

Proposition ( $∥ [\neq D, f] ∥ = Lip_{g} (f)$ for the canonical triple). Let $f \in C^{\infty} (M)$ be real-valued on a closed spin manifold and $ψ \in Γ (S)$ . The Leibniz rule for the spin connection gives $\nabla^{S} (f ψ) = df \otimes ψ + f \nabla^{S} ψ$ , hence $\neq D (f ψ) = c (df) ψ + f \neq D ψ$ , so $[\neq D, f] = c (df)$ . Clifford multiplication by a one-form $ω$ satisfies $c (ω)^{2} = - ∣ ω ∣_{g}^{2}$ pointwise, so $c (ω_{x})$ has operator norm $∣ ω_{x} ∣_{g}$ on each fibre, and $∥ c (df) ∥ = sup_{x} ∣ d f_{x} ∣_{g} = ∥∣ df ∣ ∥_{\infty}$ . The right side is the least $L$ with $∣ f (x) - f (y) ∣ \leq L d_{g} (x, y)$ for all $x, y$ — by integrating $df$ along minimising geodesics for the upper bound and by differentiating along a geodesic realising the supremum of $∣ df ∣$ for the lower bound — which is the geodesic Lipschitz constant $Lip_{g} (f)$ . Hence $∥ [\neq D, f] ∥ = Lip_{g} (f)$ . $□$

Proposition (recovery of the geodesic distance between Dirac masses). With $δ_{p}, δ_{q}$ the evaluation pure states, $ρ (δ_{p}, δ_{q}) = sup {∣ f (p) - f (q) ∣ : Lip_{g} (f) \leq 1}$ by the previous proposition. For $1$ -Lipschitz $f$ and a minimising geodesic $γ$ of length $d_{g} (p, q)$ , $∣ f (p) - f (q) ∣ \leq \int_{γ} ∣ df ∣ \leq d_{g} (p, q)$ , giving $ρ \leq d_{g} (p, q)$ . The function $f_{0} = d_{g} (p, \cdot)$ is $1$ -Lipschitz with $f_{0} (p) = 0$ , $f_{0} (q) = d_{g} (p, q)$ ; smoothing on the cut locus produces admissible $f_{ϵ}$ with $∣ f_{ϵ} (p) - f_{ϵ} (q) ∣ \geq d_{g} (p, q) - ϵ$ , so $ρ \geq d_{g} (p, q)$ . Therefore $ρ (δ_{p}, δ_{q}) = d_{g} (p, q)$ . $□$

Proposition (two-point distance equals $1/∣ m ∣$ ). For $A = C \oplus C$ acting diagonally on $H = C^{2}$ with $D = (0 \overset{m}{ˉ} m 0)$ and $a = diag (a_{1}, a_{2})$ , the commutator is $[D, a] = (0 \overset{m}{ˉ} (a_{1} - a_{2}) m (a_{2} - a_{1}) 0)$ . The singular values of an anti-diagonal $2 \times 2$ matrix with entries $u, v$ are $∣ u ∣, ∣ v ∣$ , so $∥ [D, a] ∥ = max (∣ m (a_{2} - a_{1}) ∣, ∣ \overset{m}{ˉ} (a_{1} - a_{2}) ∣) = ∣ m ∣ ∣ a_{1} - a_{2} ∣$ . The pure states are $δ_{i} (a) = a_{i}$ , so $∣ δ_{1} (a) - δ_{2} (a) ∣ = ∣ a_{1} - a_{2} ∣$ , maximised subject to $∣ m ∣ ∣ a_{1} - a_{2} ∣ \leq 1$ at $∣ a_{1} - a_{2} ∣ = 1/∣ m ∣$ . Hence $d_{D} (δ_{1}, δ_{2}) = 1/∣ m ∣$ . The same computation with mixed states $φ_{t}, ψ_{s}$ gives $∣ (t - s) (a_{1} - a_{2}) ∣ \leq ∣ t - s ∣/∣ m ∣$ , so $d_{D} (φ_{t}, ψ_{s}) = ∣ t - s ∣/∣ m ∣$ , exhibiting the state space as a segment of length $1/∣ m ∣$ . $□$

Proposition (commutative Connes distance equals the Wasserstein- $1$ distance). For $A = C (X)$ , $X$ compact metric, with a Dirac operator giving $∥ [D, f] ∥ = Lip (f)$ , states are probability measures by the Riesz representation theorem, $φ_{μ} (f) = \int f d μ$ . Then $d_{D} (μ, ν) = sup {∣ \int f d μ - \int f d ν ∣ : Lip (f) \leq 1}$ , which is exactly the Kantorovich–Rubinstein formula for $W_{1} (μ, ν)$ . The dropping of the absolute value (replacing $∣ \int f d μ - \int f d ν ∣$ by $\int f d μ - \int f d ν$ ) does not change the supremum because the constraint set is symmetric under $f \mapsto - f$ . Restricting to $μ = δ_{p}$ , $ν = δ_{q}$ recovers $W_{1} (δ_{p}, δ_{q}) = d (p, q)$ , the ground metric. $□$

Connections Master

Spectral triples and the reconstruction theorem 39.06.01 — the distance formula is the metric half of the spectral-triple package: where the reconstruction theorem recovers the smooth manifold and its dimension from $(A, H, D)$ , the formula $d_{D} (φ, ψ) = sup {∣ φ (a) - ψ (a) ∣ : ∥ [D, a] ∥ \leq 1}$ recovers its Riemannian distance, so this unit supplies the quantitative metric that the reconstruction theorem only asserts exists.
Commutative C-algebras and Gelfand duality 39.01.02* — Gelfand duality identifies the pure states of $C (M)$ with the points of $M$ and gives the topology; the Connes distance refines that topological recovery to a metric one, so the two together upgrade "algebra recovers space" to "algebra-with- $D$ recovers metric space".
Cyclic cohomology and the Chern character 39.07.01 — the distance is the degree-zero metric datum, while cyclic cohomology supplies the higher-degree integration current; the noncommutative integral $\fint = Tr_{ω} (\cdot ∣ D ∣^{- n})$ and the distance formula are the two extreme-degree readings of the same operator $D$ .
Optimal transport and the Wasserstein metric 23.07.04 pending — the commutative distance on all states is exactly the Kantorovich–Rubinstein dual of the Wasserstein- $1$ optimal-transport cost, so the Connes formula is the noncommutative generalisation of optimal transport, with the Lipschitz-dual side surviving and the coupling side dissolving when points disappear.
Riemannian geodesic distance 03.06.02 pending — the commutative theorem identifies the spectral distance between Dirac masses with the geodesic distance $d_{g}$ , so this unit is the spectral re-encoding of the classical length-minimising-path definition of Riemannian distance.

Historical & philosophical context Master

The distance formula first appears in Connes' 1989 paper Compact metric spaces, Fredholm modules, and hyperfiniteness ^{[Connes 1989]}, where he observed that a Fredholm module — and in particular a spectral triple — equips the state space of the algebra with a metric defined purely by the commutator constraint $∥ [D, a] ∥ \leq 1$ , and that for the canonical triple of a Riemannian manifold this metric reproduces the geodesic distance. It is consolidated in Chapter VI of Noncommutative Geometry (1994) ^{[Connes Ch. VI §1]} as the metric pillar of the program, complementing the dimension (eigenvalue growth) and integration (Dixmier trace) pillars. The slogan "the metric is encoded in the operator $D$ " dates from this period and is the precise sense in which Riemannian geometry becomes spectral.

The recognition that the formula is the Kantorovich–Rubinstein dual of optimal transport, and the axiomatic isolation of when the resulting metric recovers the weak- $*$ topology, are due to Marc Rieffel, whose Metrics on state spaces (1999) ^{[Rieffel 1999]} introduced Lip-norms and compact quantum metric spaces and connected the construction to Gromov–Hausdorff convergence. The explicit computation of distances in finite and almost-commutative geometries — the two-point segment of length $1/∣ m ∣$ , the internal Standard-Model distances governed by Yukawa masses — was carried out by Iochum, Krajewski, Martinetti and others around 2000 ^{[Iochum-Krajewski-Martinetti 2001]}, establishing the picture in which the Higgs field is the coordinate measuring the spectral distance between the two sheets of the almost-commutative manifold.

Bibliography Master

Connes, A., "Compact metric spaces, Fredholm modules, and hyperfiniteness", Ergodic Theory and Dynamical Systems 9 (1989), 207–220.
Connes, A., Noncommutative Geometry, Academic Press, 1994. Ch. VI §1.
Rieffel, M. A., "Metrics on state spaces", Documenta Mathematica 4 (1999), 559–600.
Rieffel, M. A., "Gromov-Hausdorff distance for quantum metric spaces", Memoirs of the American Mathematical Society 168 (2004), no. 796.
Iochum, B., Krajewski, T. & Martinetti, P., "Distances in finite spaces from noncommutative geometry", Journal of Geometry and Physics 37 (2001), 100–125.
Martinetti, P. & Wulkenhaar, R., "Discrete Kaluza-Klein from scalar fluctuations in noncommutative geometry", Journal of Mathematical Physics 43 (2002), 182–204.
Villani, C., Optimal Transport: Old and New, Grundlehren der mathematischen Wissenschaften 338, Springer, 2009. Ch. 6 (Kantorovich-Rubinstein duality).

Operator-algebras spine, noncommutative-geometry chapter. The metric heart of NCG: the Connes distance formula $d_{D} (φ, ψ) = sup {∣ φ (a) - ψ (a) ∣ : ∥ [D, a] ∥ \leq 1}$ on the state space, its commutative reduction to the geodesic distance via $∥ [\neq D, f] ∥ = Lip_{g} (f)$ , finite/two-point examples, and Monge-Kantorovich duality. Builds on spectral triples (39.06.01) and Gelfand duality (39.01.02).

Prerequisites

39.06.01
39.01.02

Tier anchors

beginner: Connes-style 'distance is the most you can spread two functions apart while keeping their slope under control'; the road-grading analogy of a speed-limited traveller
intermediate: Gracia-Bondía, Várilly & Figueroa *Elements of Noncommutative Geometry* Ch. 9–10; Connes *Noncommutative Geometry* (1994) Ch. VI §1
master: Connes *Noncommutative Geometry* (1994) Ch. VI; Connes 'Compact metric spaces, Fredholm modules, and hyperfiniteness' (1989); Rieffel 'Metrics on state spaces' (1999)

References

Connes, A. — Noncommutative Geometry · Ch. VI §1, the metric on the state space, the distance formula, the commutative reduction to geodesic distance
Connes, A. — Compact metric spaces, Fredholm modules, and hyperfiniteness · Ergodic Theory Dynam. Systems 9 (1989) 207–220 — the distance formula and the metric heart of NCG
Rieffel, M. A. — Metrics on state spaces · Doc. Math. 4 (1999) 559–600 — Lipschitz seminorms, compact quantum metric spaces, weak-* topology recovery
Iochum, B., Krajewski, T. & Martinetti, P. — Distances in finite spaces from noncommutative geometry · J. Geom. Phys. 37 (2001) 100–125 — the two-point space and matrix-algebra distance computations

Estimated time

beginner: 18m
intermediate: 55m
master: 90m