Commutative C*-Algebras and Gelfand Duality

Intuition Beginner

A space of continuous functions remembers the shape it lives on. If you know every continuous function on a circle, and you know how to add and multiply them, then you can reconstruct the circle itself — the points of the circle are hidden inside the algebra of functions. Gelfand duality is the precise statement that nothing is lost: the shape and its function algebra carry exactly the same information.

How do you recover a point from the functions alone? A point is the same thing as a rule that takes a function and reports its value there. Picking the point $p$ on the circle is the same as picking the rule "evaluate at $p$". These evaluation rules respect addition and multiplication, and it turns out that every such rule comes from a point. So the points of the shape are exactly the multiplication-respecting evaluation rules on the algebra.

This is the conceptual seed of noncommutative geometry: if a commutative function algebra is secretly a space, then a noncommutative algebra can be treated as the functions on a "space" that has no ordinary points at all.

Visual Beginner

A commutative C*-algebra and a compact shape are two views of one object. Going right, you read functions off the shape; going left, you recover the points as evaluation rules.

The dictionary reads: a point becomes an evaluation rule; a continuous map between shapes becomes a backward substitution of functions; and the whole shape is rebuilt from the algebra as the set of all its evaluation rules with their natural notion of closeness.

Worked example Beginner

Take the shape to be just two points, named $1$ and $2$. A continuous complex-valued function on two points is simply a choice of two complex numbers: its value at point $1$ and its value at point $2$. So the algebra of functions is all pairs $(z_1,z_2)$, added and multiplied slot by slot.

Now recover the points from the algebra of pairs. An evaluation rule must respect multiplication. The rule "report the first slot", sending $(z_1,z_2)$ to $z_1$, respects both addition and multiplication: the first slot of a product is the product of first slots. The rule "report the second slot" does the same. These are the only two such rules.

So the algebra of pairs has exactly two evaluation rules, and they correspond to the two points $1$ and $2$ we started with. Count them: two points on the shape, two evaluation rules in the algebra.

What this tells us: the number of points in the shape equals the number of multiplication-respecting evaluation rules in the function algebra, and each rule pins down one point. The shape was never lost; it was encoded in the algebra all along.

Check your understanding Beginner

Formal definition Intermediate+

Let $A$ be a commutative C*-algebra 39.01.01. A character of $A$ is a nonzero algebra homomorphism $\tau :A\to \mathbb{C}$, equivalently a nonzero linear functional with $\tau (ab)=\tau (a)\tau (b)$. The set of all characters is the character space (or spectrum, or maximal ideal space) $\hat{A}$. In the unital case a character automatically satisfies $\tau (1)=1$ and is a $\ast$-homomorphism ($\tau (a^{\ast })=\overline{\tau (a)}$), and characters are automatically continuous with $\Vert \tau \Vert =1$. The kernel of a character is a maximal ideal of codimension one, and in a commutative unital C*-algebra this correspondence between characters and maximal ideals is a bijection — hence the name maximal ideal space ^{[Murphy Ch. 2]}.

Give $\hat{A}$ the weak-$\ast$ topology inherited from the dual space $A^{\ast }$, the topology of pointwise convergence: ${\tau }_i\to \tau$ iff ${\tau }_i(a)\to \tau (a)$ for every $a\in A$. With this topology $\hat{A}$ is a compact Hausdorff space when $A$ is unital (it is a weak-$\ast$ closed subset of the unit ball of $A^{\ast }$, compact by Banach-Alaoglu 02.11.04), and a locally compact Hausdorff space when $A$ is non-unital.

The Gelfand transform is the map $$ \Gamma : A \longrightarrow C_0(\widehat A), \qquad \Gamma(a)(\tau) = \hat a(\tau) = \tau(a), $$ sending an algebra element $a$ to the function $\hat{a}$ on $\hat{A}$ that evaluates each character at $a$. The transform is an algebra homomorphism, and $\hat{a}(\hat{A})\cup \{0\}=\sigma (a)\cup \{0\}$, so the range of $\hat{a}$ is essentially the spectrum of $a$ ^{[Davidson Ch. I]}. In the unital case the codomain is $C(\hat{A})$ (continuous functions on a compact space); in the non-unital case it is $C_0(\hat{A})$ (continuous functions vanishing at infinity).

Counterexamples to common slips

• A character is not just any linear functional. On $A={\mathbb{C}}^2$ the functional $(z_1,z_2)\mapsto z_1+z_2$ is linear but is not a character: it fails $\tau (ab)=\tau (a)\tau (b)$, since $(1,0)\cdot (0,1)=(0,0)$ maps to $0$ while $\tau (1,0)\tau (0,1)=1\cdot 1=1$. Only the two coordinate projections are characters.
• The character space is not discrete in general. For $A=C[0,1]$ the character space is $[0,1]$ with its usual topology, not a discrete set of points; the weak-$\ast$ topology recovers the original topology of the shape, which is the whole content of duality.
• Non-unital algebras lose compactness, not the theory. For $A=C_0(\mathbb{R})$ the character space is $\mathbb{R}$, which is locally compact but not compact; adjoining a unit corresponds to one-point compactification $\mathbb{R}\cup \{\infty \}$, and characters of the unitisation that kill $A$ are the point at infinity.

Key theorem with proof Intermediate+

Theorem (commutative Gelfand-Naimark). Let $A$ be a commutative unital C-algebra. The Gelfand transform* $$ \Gamma : A \longrightarrow C(\widehat A), \qquad \Gamma(a) = \hat a, \quad \hat a(\tau) = \tau(a), $$ is an isometric $$-isomorphismonto$C(\widehat A)$,where$\widehat A$isthecompactHausdorffcharacterspace.Inthenon-unitalcase$\Gamma$isanisometric$$-isomorphismof$A$onto$C_0(\widehat A)$with$\widehat A$ locally compact Hausdorff. ^{[Davidson Ch. I; Murphy Ch. 2]}

Proof. Assume $A$ unital. Each character $\tau$ is a unital $\ast$-homomorphism, so $\Gamma$ is a unital $\ast$-homomorphism into $C(\hat{A})$: $\Gamma (a^{\ast })(\tau )=\tau (a^{\ast })=\overline{\tau (a)}=\overline{\hat{a}(\tau )}$, and products and the unit are preserved pointwise.

Range and spectrum. For $a\in A$ and $\lambda \in \mathbb{C}$, the value $\lambda$ lies in the range of $\hat{a}$ iff some character has $\tau (a)=\lambda$ iff $a-\lambda 1$ lies in a maximal ideal iff $a-\lambda 1$ is non-invertible (in a commutative unital algebra a non-invertible element generates a proper ideal, contained in a maximal ideal, which is the kernel of a character). Hence $\hat{a}(\hat{A})=\sigma (a)$, and in particular $\Vert \hat{a}{\Vert }_{\infty }={\sup }_{\tau }|\tau (a)|=r(a)$, the spectral radius.

Isometry. For self-adjoint $a=a^{\ast }$, the C*-axiom forces $\Vert a^2\Vert =\Vert a^{\ast }a\Vert =\Vert a{\Vert }^2$, so by induction on powers of two $\Vert a^{2^n}\Vert =\Vert a{\Vert }^{2^n}$, and the spectral radius formula gives $r(a)={\lim }_n\Vert a^{2^n}{\Vert }^{1/2^n}=\Vert a\Vert$. Thus $\Vert \hat{a}{\Vert }_{\infty }=r(a)=\Vert a\Vert$ on self-adjoint elements. For general $b$, applying this to the self-adjoint $b^{\ast }b$ and using that $\Gamma$ is a $\ast$-homomorphism, $$ |\hat b|\infty^2 = |\overline{\hat b},\hat b|\infty = |\widehat{b^*b}|\infty = |b^*b| = |b|^2, $$ so $|\hat b|\infty = |b|$.Thetransformisisometric,henceinjective,anditsrangeisaclosedsubalgebraof$C(\widehat A)$.

Surjectivity. The range $\Gamma (A)$ is a closed $\ast$-subalgebra of $C(\hat{A})$ that contains the constants (the unit maps to $1$) and separates points: if ${\tau }_1\ne {\tau }_2$ are distinct characters, some $a$ has ${\tau }_1(a)\ne {\tau }_2(a)$, i.e. $\hat{a}({\tau }_1)\ne \hat{a}({\tau }_2)$. By the Stone-Weierstrass theorem a point-separating, conjugation-closed, unital subalgebra of $C(\hat{A})$ is dense; being closed, $\Gamma (A)=C(\hat{A})$. So $\Gamma$ is an isometric $\ast$-isomorphism onto $C(\hat{A})$.

For non-unital $A$, adjoin a unit to form $\tilde{A}=A\oplus \mathbb{C}1$, a commutative unital C*-algebra. Its character space $\hat{\tilde{A}}$ is the one-point compactification of $\hat{A}$, the extra character being the one that kills $A$. Restricting the unital isomorphism $\tilde{A}\cong C(\hat{\tilde{A}})$ to the ideal $A$ identifies it with the functions vanishing at the point at infinity, namely $C_0(\hat{A})$. $\square$

Bridge. This theorem builds toward the entire dictionary of noncommutative topology, and it appears again in 39.01.01 from the other side: the continuous functional calculus is exactly the single-generator case of this duality, applied to the commutative subalgebra $C^{\ast }(a,1)$ generated by one normal element, whose character space is the spectrum $\sigma (a)$. The foundational reason the transform is isometric is the C*-axiom on self-adjoint elements, identical to the mechanism in 39.01.01; this is exactly why $\Vert \hat{a}{\Vert }_{\infty }=\Vert a\Vert$ rather than merely $\le$. The duality generalises the finite case of 01.01.13, where a diagonalisable operator's commutative algebra has a finite character space equal to its eigenvalue set. Putting these together, the bridge is that "commutative C*-algebra" and "compact Hausdorff space" name one object viewed two ways, and the functional calculus is duality read for the smallest interesting algebra.

Exercises Intermediate+

Lean formalization Intermediate+

lean_status: none — Mathlib has the character space WeakDual.characterSpace, the gelfandTransform, and gelfandStarTransform proving the unital commutative transform is a $\ast$-isometric isomorphism onto C(characterSpace ℂ A, ℂ). What is not consolidated as one named theorem is the contravariant equivalence of categories — the spectrum functor on morphisms with its quasi-inverse $X\mapsto C(X)$, the unit/counit natural isomorphisms, and the $C_0$ non-unital version via one-point compactification.

The intended statement reads schematically:

import Mathlib.Analysis.CStarAlgebra.GelfandDuality

variable {A : Type*} [CommCStarAlgebra A]

/-- Commutative Gelfand-Naimark: the Gelfand transform is a *-isometric
isomorphism onto continuous functions on the character space. -/
theorem gelfand_naimark_commutative :
    ∃ Φ : A ≃⋆ₐ[ℂ] C(characterSpace ℂ A, ℂ),
      ∀ a, ‖Φ a‖ = ‖a‖ :=
  sorry  -- gelfandStarTransform packaged as a bundled *-isometric equiv

Advanced results Master

The duality is not a single isomorphism but a contravariant equivalence of categories, and the structural payoff is organised by reading the dictionary in both directions.

Functoriality of the spectrum. Let $\mathsf{Comm}{\mathsf{C}}^{\ast }$ be the category of commutative C*-algebras with $\ast$-homomorphisms, and $\mathsf{CHaus}$ the category of compact Hausdorff spaces with continuous maps (for the unital case; locally compact Hausdorff spaces with proper continuous maps for the non-unital $C_0$ case). The character-space construction $A\mapsto \hat{A}$ is a contravariant functor: a $\ast$-homomorphism $\phi :A\to B$ induces $\hat{\phi }:\hat{B}\to \hat{A}$, $\hat{\phi }(\tau )=\tau \circ \phi$, continuous for the weak-$\ast$ topologies. The continuous-function construction $X\mapsto C(X)$ is the contravariant functor in the other direction, $f\mapsto (g\mapsto g\circ f)$.

Equivalence. These two functors are mutually quasi-inverse. The counit is the Gelfand isomorphism ${\Gamma }_A:A\stackrel{\sim }{\to }C(\hat{A})$; the unit is the evaluation homeomorphism ${\epsilon }_X:X\stackrel{\sim }{\to }\widehat{C(X)}$, $x\mapsto {\mathrm{ev}}_x$. Both are natural in their arguments. Thus ${\mathsf{Comm}{\mathsf{C}}^{\ast }}^{\mathrm{op}}\simeq \mathsf{CHaus}$: commutative C*-algebra theory and the topology of compact Hausdorff spaces are the same mathematics. Algebraic operations translate exactly — quotients of $A$ correspond to closed subspaces of $\hat{A}$, tensor products to products of spaces, direct limits to inverse limits, and the ideals of $A$ to open subsets of $\hat{A}$.

Spectral synthesis of self-adjoint operators. Specialising to the commutative C*-algebra generated by a single self-adjoint or normal operator recovers the spectral theorem: the character space is the spectrum $\sigma (a)\subseteq \mathbb{C}$, and $\Gamma$ is the continuous functional calculus $C(\sigma (a))\cong C^{\ast }(a,1)$. The duality is therefore the conceptual container for the functional calculus 39.01.01: one normal element is one commutative algebra is one compact subset of $\mathbb{C}$.

Noncommutative topology. Because every topological construction on $\hat{A}$ has an algebraic shadow on $A$, one defines the analogues for noncommutative C*-algebras by their algebraic form. A noncommutative C*-algebra is read as the algebra of "functions" on a virtual noncommutative space with no points; connectedness, dimension, $K$-theory, and homology all migrate to the algebra. This is the founding move of Connes' noncommutative geometry ^{[Connes Ch. I]}, where a spectral triple equips such an algebra with the metric data a Riemannian manifold would supply.

Synthesis. The commutative Gelfand-Naimark theorem is the foundational reason the spectrum deserves to be called a space: it shows the character space, equipped with the weak-$\ast$ topology, is the compact Hausdorff shape whose functions reconstruct $A$, and this is exactly the equivalence ${\mathsf{Comm}{\mathsf{C}}^{\ast }}^{\mathrm{op}}\simeq \mathsf{CHaus}$. The construction is dual to the functional calculus of 39.01.01 — one passes from algebra to space, the other applies functions on the space back to the algebra — and putting these together gives the central insight that points, characters, and maximal ideals are three names for the same data. The duality generalises in two directions that appear again across operator algebras: downward to the single-generator case, where it is the continuous functional calculus and the character space is $\sigma (a)$, and upward to the measurable analogue, where von Neumann algebras 39.03.01 replace $C(X)$ by $L^{\infty }(X,\mu )$ and weak-$\ast$ closure replaces norm closure. The bridge to the rest of the subject is that "noncommutative C*-algebra" means "functions on a noncommutative space", and the entire program of noncommutative geometry is the systematic exploitation of that single reframing.

Full proof set Master

Proposition (characters are automatically $\ast$-preserving on a commutative unital C*-algebra). Let $\tau$ be a character and $a=a^{\ast }$ self-adjoint. Write $\tau (a)=\alpha +i\beta$. For real $t$, the element $a+it1$ is normal, and $\tau (a+it1)=\alpha +i(\beta +t)$, so $|\alpha +i(\beta +t){|}^2\le \Vert a+it1{\Vert }^2=\Vert (a-it1)(a+it1)\Vert =\Vert a^2+t^{2}1\Vert \le \Vert a{\Vert }^2+t^2$. Expanding, ${\alpha }^2+{\beta }^2+2\beta t+t^2\le \Vert a{\Vert }^2+t^2$, so ${\alpha }^2+{\beta }^2+2\beta t\le \Vert a{\Vert }^2$ for all $t\in \mathbb{R}$, forcing $\beta =0$. Hence $\tau$ maps self-adjoint elements to $\mathbb{R}$, and writing a general $b=b_1+i{b}_2$ in self-adjoint parts gives $\tau (b^{\ast })=\tau (b_1)-i\tau (b_2)=\overline{\tau (b)}$.

Proposition (maximal ideals are kernels of characters). Let $A$ be commutative unital and $M$ a maximal ideal. Then $M$ is closed (its closure is a proper ideal, since the invertibles form an open set disjoint from $M$, so $\overline{M}=M$ by maximality), and $A/M$ is a unital Banach algebra in which every nonzero element is invertible, hence a division algebra. By Gelfand-Mazur $A/M\cong \mathbb{C}$, and the quotient map $A\to A/M\cong \mathbb{C}$ is a character with kernel $M$. Conversely the kernel of any character is a maximal ideal of codimension one. This bijection identifies $\hat{A}$ with the maximal ideal space.

Proposition (the Gelfand transform has range the spectrum). For $a\in A$ commutative unital, $\hat{a}(\hat{A})=\sigma (a)$. If $\lambda \in \sigma (a)$ then $a-\lambda 1$ is non-invertible, so it generates a proper ideal contained in some maximal ideal $M=\ker \tau$; then $\tau (a)=\lambda$, so $\lambda \in \hat{a}(\hat{A})$. Conversely if $\tau (a)=\lambda$ then $a-\lambda 1\in \ker \tau$, a proper ideal, so $a-\lambda 1$ is non-invertible and $\lambda \in \sigma (a)$.

Proposition (the unit and counit are natural isomorphisms). Naturality of the counit: for $\phi :A\to B$ a $\ast$-homomorphism, the square relating ${\Gamma }_A,{\Gamma }_B$ and the induced map $C(\hat{\phi }):C(\hat{A})\to C(\hat{B})$ commutes because both routes send $a\in A$ to the function $\tau \mapsto \tau ({\phi }^{-1}\dots )$ — concretely, $C(\hat{\phi })({\Gamma }_{A}a)(\tau )={\Gamma }_A(a)(\hat{\phi }(\tau ))=(\tau \circ \phi )(a)=\tau (\phi (a))={\Gamma }_B(\phi (a))(\tau )$. Naturality of the unit ${\epsilon }_X:X\to \widehat{C(X)}$ for $f:X\to Y$ is the identity ${\mathrm{ev}}_{f(x)}={\mathrm{ev}}_x\circ (f^{\ast })$ from Exercise 6. Both ${\Gamma }_A$ (isometric $\ast$-isomorphism, by the commutative Gelfand-Naimark theorem) and ${\epsilon }_X$ (homeomorphism, by Urysohn separation and the maximal-ideal description of $C(X)$) are isomorphisms, completing the equivalence ${\mathsf{Comm}{\mathsf{C}}^{\ast }}^{\mathrm{op}}\simeq \mathsf{CHaus}$.

Proposition (non-unital duality via unitisation). Let $A$ be a non-unital commutative C*-algebra and $\tilde{A}=A\oplus \mathbb{C}1$ its unitisation. Then $\hat{\tilde{A}}=\hat{A}\bigsqcup \{{\tau }_{\infty }\}$ where ${\tau }_{\infty }(a+\lambda 1)=\lambda$ is the character vanishing on $A$, and $\hat{\tilde{A}}$ is the one-point compactification of the locally compact $\hat{A}$. The unital isomorphism $\tilde{A}\cong C(\hat{\tilde{A}})$ restricts on the ideal $A$ to an isomorphism onto $\{g\in C(\hat{\tilde{A}}):g({\tau }_{\infty })=0\}=C_0(\hat{A})$. Hence $A\cong C_0(\hat{A})$ with $\hat{A}$ locally compact Hausdorff, and proper continuous maps correspond to $\ast$-homomorphisms respecting the structure at infinity.

Connections Master

• C-algebras: axioms, spectrum, and the continuous functional calculus 39.01.01* — the continuous functional calculus is the single-generator case of this duality: the commutative subalgebra $C^{\ast }(a,1)$ generated by one normal element has character space $\sigma (a)$, so $C(\sigma (a))\cong C^{\ast }(a,1)$ is Gelfand-Naimark for the smallest interesting commutative algebra.

• Spectral theorem for normal operators 01.01.13 — in finite dimensions the commutative algebra generated by a diagonalisable operator has a finite character space equal to its eigenvalue set, and the Gelfand transform is the diagonalisation $T={\sum }_i{\lambda }_i{P}_i$; this unit is the coordinate-free, infinite-dimensional generalisation.

• Von Neumann algebras and the double commutant 39.03.01 — strengthening norm closure to weak-operator closure replaces the commutative model $C(X)$ by $L^{\infty }(X,\mu )$ and the character space by a measure space; the commutative von Neumann algebras are exactly the $L^{\infty }$ algebras, the measurable analogue of Gelfand duality.

• Spectral triples and noncommutative geometry 39.06.01 — the equivalence ${\mathsf{Comm}{\mathsf{C}}^{\ast }}^{\mathrm{op}}\simeq \mathsf{CHaus}$ is the entry point: a noncommutative C*-algebra is treated as functions on a pointless noncommutative space, and a Dirac operator restores the metric data that the topology alone omits.

Historical & philosophical context Master

The transform and its commutative representation theorem originate with Israel Gelfand's theory of normed rings in the late 1930s, where he attached to a commutative Banach algebra its space of maximal ideals and represented elements as functions on that space ^{[Gelfand-Naimark 1943]}. The decisive sharpening for C*-algebras came in the 1943 paper of Gelfand and Naimark, which proved that a commutative C*-algebra is isometrically $\ast$-isomorphic to the continuous functions on its maximal ideal space, and that an abstract C*-algebra embeds into the bounded operators on a Hilbert space. The isometry, as opposed to a mere contractive representation, rests on the C*-axiom through the spectral radius identity for self-adjoint elements, which is why the same algebra admits at most one C*-norm.

Marshall Stone's representation of Boolean algebras as clopen sets of a compact totally disconnected space (Stone duality, 1936) is the prototype of the same idea one categorical level down, and the appearance of Stone-Weierstrass in the surjectivity argument is not accidental. The reading of duality as a definition rather than a theorem belongs to Alain Connes, whose noncommutative geometry program from the 1980s onward takes the commutative dictionary as the specification for the noncommutative case, replacing the space by the algebra and equipping it with a spectral triple to recover differential and metric structure.

Bibliography Master

• Davidson, K. R., C-Algebras by Example*, Fields Institute Monographs 6, American Mathematical Society, 1996. Ch. I.
• Murphy, G. J., C-Algebras and Operator Theory*, Academic Press, 1990. Ch. 2.
• Gelfand, I. and Naimark, M., "On the imbedding of normed rings into the ring of operators in Hilbert space", Mat. Sbornik 12 (1943), 197–213.
• Dixmier, J., C-Algebras*, North-Holland, 1977. §1.
• Connes, A., Noncommutative Geometry, Academic Press, 1994. Ch. I.
• Stone, M. H., "The theory of representations for Boolean algebras", Transactions of the American Mathematical Society 40 (1936), 37–111.

---

Operator-algebras spine, commutative-theory companion to 39.01.01. Produced as the Gelfand-duality anchor: the contravariant equivalence between commutative C-algebras and (locally) compact Hausdorff spaces, and the conceptual basis of noncommutative topology.*