The Noncommutative Torus and Its Geometry

Intuition Beginner

An ordinary torus is a square with opposite edges glued: a doughnut surface with two independent loops, one around the hole and one through it. You can name positions on it with two coordinate functions, and those two functions multiply in the usual way, where order does not matter.

The noncommutative torus keeps almost everything about this picture but changes one rule. It replaces the two coordinate functions by two abstract symbols, call them $u$ and $v$, that you can still multiply and add — but now $u$ times $v$ is not the same as $v$ times $u$. Going around the two loops in different orders leaves behind a fixed twist, a phase, set by a single number. When that number is a whole number the twist disappears and you are back to the ordinary torus; when it is irrational the twist never unwinds, and the surface has, in a precise sense, no points left on it at all.

The surprise is that this pointless object still has a full geometry. There are two directions you can differentiate in, a way to integrate, a notion of flatness and curvature, and even a Gauss-Bonnet theorem. You read the geometry off the algebra and two derivative rules instead of off a surface, and almost every classical formula has a faithful echo.

Visual Beginner

The noncommutative torus is the ordinary torus with one rule changed: its two coordinate loops fail to commute by a fixed phase, and two derivative operations give it a flat geometry.

The dictionary reads: $u$ and $v$ are the two coordinate loops; the phase $e^{2\pi i\theta }$ is the fixed twist you collect when you swap their order; ${\delta }_1$ and ${\delta }_2$ are the two ways to differentiate, one along each loop; and $\tau$ is the way to average a quantity over the whole surface, the noncommutative integral.

Worked example Beginner

Take the two symbols $u$ and $v$ with the twist rule $vu=e^{2\pi i\theta }uv$, where $\theta =1/4$ so the phase is $e^{2\pi i/4}=i$. So $vu=iuv$: swapping $u$ and $v$ costs a factor of $i$.

Now differentiate along the first loop. The first derivative rule ${\delta }_1$ is built to act like "count how many $u$'s there are": it sends $u$ to $u$ and leaves $v$ fixed, ${\delta }_1(u)=u$ and ${\delta }_1(v)=0$. So on the product $u^{2}v$ it counts the two $u$'s:

$${\delta }_1(u^{2}v)=2u^{2}v.$$

The second rule ${\delta }_2$ counts $v$'s the same way: ${\delta }_2(v)=v$, ${\delta }_2(u)=0$, so ${\delta }_2(u^{2}v)=u^{2}v$, one $v$.

Now average with the integral $\tau$. The rule for $\tau$ is simple: it keeps the constant part of an element and throws away everything with any $u$ or $v$ in it. So $\tau (u^{2}v)=0$, while $\tau (1)=1$. Differentiating first and then averaging gives $\tau ({\delta }_1(u^{2}v))=\tau (2u^{2}v)=0$.

What this tells us: the two derivative rules count powers along the two loops and the integral keeps only the constant part. Together they reproduce, on these twisted symbols, exactly the calculus you would do on a flat ordinary torus — and the average of any derivative is zero, the noncommutative version of "the integral of a derivative around a closed loop vanishes."

Check your understanding Beginner

Formal definition Intermediate+

Fix $\theta \in \mathbb{R}$. The noncommutative torus (or irrational rotation algebra when $\theta \notin \mathbb{Q}$) is the universal C*-algebra $$ A_\theta = C^*\big(u, v \ :\ u, v \text{ unitary},\ v u = e^{2\pi i \theta} u v\big), $$ introduced in 39.02.01 as the crossed product $C(\mathbb{T}){⋊}_{\theta }\mathbb{Z}$. Its canonical normalised trace $\tau$ is the unique tracial state with $\tau (u^m{v}^n)={\delta }_{(m,n),(0,0)}$; for $\theta$ irrational $A_{\theta }$ is simple with $\tau$ its unique trace.

The C*-algebra is too coarse for differential geometry, so one passes to the smooth subalgebra. Write a general formal element as a doubly-indexed sum $a={\sum }_{(m,n)\in {\mathbb{Z}}^2}{a}_{mn}{u}^m{v}^n$ with $a_{mn}\in \mathbb{C}$. The smooth noncommutative torus is $$ A_\theta^\infty = \Big{, a = \sum_{(m,n)} a_{mn}, u^m v^n \ :\ (a_{mn}) \in \mathcal{S}(\mathbb{Z}^2) ,\Big}, $$ the elements whose coefficient sequence is rapidly decaying: for every $k$, ${\sup }_{m,n}(1+|m|+|n|{)}^k|a_{mn}|<\infty$. This is a unital dense $\ast$-subalgebra of $A_{\theta }$, the noncommutative analogue of $C^{\infty }({\mathbb{T}}^2)\subset C({\mathbb{T}}^2)$; it is the smooth-vector subalgebra for the ${\mathbb{T}}^2$-action below and is closed under holomorphic functional calculus.

The torus action of ${\mathbb{T}}^2$ acts by ${\alpha }_{(s,t)}(u^m{v}^n)=e^{2\pi i(ms+nt)}{u}^m{v}^n$. Its two infinitesimal generators are the canonical derivations ${\delta }_1,{\delta }_2:A_{\theta }^{\infty }\to A_{\theta }^{\infty }$, the densely defined closed $\ast$-derivations determined on generators by $$ \delta_1(u) = 2\pi i, u,\quad \delta_1(v) = 0, \qquad \delta_2(u) = 0,\quad \delta_2(v) = 2\pi i, v, $$ extended by the Leibniz rule ${\delta }_j(ab)={\delta }_j(a)b+a{\delta }_j(b)$. (Some authors drop the $2\pi i$; this unit keeps it so that ${\delta }_j$ generates a $2\pi$-periodic flow.) They commute, ${\delta }_1{\delta }_2={\delta }_2{\delta }_1$, are trace-invariant, $\tau \circ {\delta }_j=0$, and satisfy the integration-by-parts identity $$ \tau\big(\delta_j(a), b\big) = -,\tau\big(a, \delta_j(b)\big), $$ the noncommutative analogue of ${\int }_{{\mathbb{T}}^2}({\partial }_{j}f)g=-{\int }_{{\mathbb{T}}^2}f({\partial }_{j}g)$. The pair $({\delta }_1,{\delta }_2)$ is the flat noncommutative vector fields generating the geometry.

The flat spectral triple 39.06.01 of $A_{\theta }$ packages $({\delta }_1,{\delta }_2)$ into a Dirac operator. Let ${\mathcal{H}}_0=L^2(A_{\theta },\tau )$ be the GNS Hilbert space of $\tau$ (completion of $A_{\theta }^{\infty }$ in the inner product $⟨a,b⟩=\tau (a^{\ast }b)$), let $\mathcal{H}={\mathcal{H}}_0\otimes {\mathbb{C}}^2$, and on it set $$ D = \delta_1 \otimes \gamma^1 + \delta_2 \otimes \gamma^2, $$ where ${\gamma }^1,{\gamma }^2$ are the $2\times 2$ Pauli-type matrices with ${\gamma }^j({\gamma }^k{)}^{\ast }+{\gamma }^k({\gamma }^j{)}^{\ast }=2{\delta }^{jk}$ representing the flat Clifford algebra of ${\mathbb{R}}^2$. With $A_{\theta }^{\infty }$ acting by left multiplication, $(A_{\theta }^{\infty },\mathcal{H},D)$ is an even spectral triple of metric dimension $2$: each commutator $[D,a]={\sum }_j[{\delta }_j,a]\otimes {\gamma }^j={\sum }_j{\delta }_j(a)\otimes {\gamma }^j$ is bounded, and $D$ has compact resolvent because $|D{|}^2=({\delta }_1^2+{\delta }_2^2)\otimes 1$ has eigenvalues $4{\pi }^2(m^2+n^2)$ on the orthonormal basis $u^m{v}^n$, growing without bound with finite multiplicities. A general flat metric is encoded by replacing ${\delta }_1,{\delta }_2$ with a complex structure: fix a modular parameter ${\tau }_0={\tau }_1+i{\tau }_2\in \mathbb{H}$ (upper half-plane, not to be confused with the trace $\tau$) and set $\partial ={\delta }_1+{\tau }_0{\delta }_2$, $\bar{\partial }={\delta }_1+{\bar{\tau }}_0{\delta }_2$; the Dirac operator becomes the boundary operator $\partial$ together with its adjoint.

Counterexamples to common slips

• The smooth subalgebra $A_{\theta }^{\infty }$ is genuinely needed; the derivations ${\delta }_j$ are unbounded and not everywhere defined on $A_{\theta }$. The element $a={\sum }_n{n}^{-2}{u}^n$ lies in $A_{\theta }$ but ${\delta }_1(a)={\sum }_{n}2\pi i{n}^{-1}{u}^n$ has coefficients that are not summable as needed, so $a\notin \mathrm{dom}{\delta }_1$. Rapid decay is exactly the smoothness condition that keeps every ${\delta }_1^j{\delta }_2^k(a)$ in the algebra.
• For $\theta \in \mathbb{Q}$ the algebra is not a noncommutative space in the strong sense: $A_{p/q}$ is the algebra of sections of an $M_q$-bundle over ${\mathbb{T}}^2$, has many traces and a centre of positive dimension, and is Morita equivalent to $C({\mathbb{T}}^2)$. The genuinely pointless behaviour — simplicity, unique trace, dense trace image $\mathbb{Z}+\theta \mathbb{Z}$ — requires $\theta$ irrational.
• The trace $\tau$ is not a vector state coming from a single eigenvector; it is the ${\mathbb{T}}^2$-average and is faithful and tracial, yet its GNS von Neumann closure is the hyperfinite II${}_1$ factor, not a type I algebra. Treating $\tau$ as evaluation at a point is the classical reflex that fails here.

Key theorem with proof Intermediate+

Theorem (Gauss-Bonnet for the noncommutative two-torus, Connes-Tretkoff). Let $A_{\theta }^{\infty }$ carry the flat conformal structure of modular parameter ${\tau }_0\in \mathbb{H}$, and let it be conformally rescaled by a positive Weyl factor $e^h$, $h = h^ \in A_\theta^\infty$,producingtheperturbedLaplacian$\Delta_h$actingon$\mathcal{H}0$.Let$\zeta{\Delta_h}(s) = \operatorname{Tr}'\big(\Delta_h^{-s}\big)$ be its zeta function with the kernel removed. Then the value* $$ \zeta_{\Delta_h}(0) + 1 ;=; \zeta_{\Delta_0}(0) + 1 $$ is independent of the Weyl factor $h$; equivalently, the constant term $a_2$ in the small-time heat expansion ${\mathrm{Tr}}^{\prime }(e^{-t{\Delta }_h})\sim a_0{t}^{-1}+a_2+O(t)$ does not depend on $h$. For the flat torus this constant is $0$, so the total scalar curvature of any conformally rescaled metric on $A_{\theta }$ vanishes. ^{[Connes-Tretkoff 2011]}

Proof. The value ${\zeta }_{\Delta }(0)$ is computed from the heat trace through the Mellin transform, ${\zeta }_{\Delta }(s)=\frac{1}{\Gamma (s)}{\int }_0^{\infty }{t}^{s-1}{\mathrm{Tr}}^{\prime }(e^{-t\Delta })dt$, whose meromorphic continuation has ${\zeta }_{\Delta }(0)+\dim \ker \Delta$ equal to the constant Seeley-DeWitt coefficient $a_2$ of the expansion ${\mathrm{Tr}}^{\prime }(e^{-t\Delta })\sim {\sum }_{k\ge 0}{a}_{2k}{t}^{k-1}$. The coefficients $a_{2k}$ are computed by the pseudodifferential calculus over $A_{\theta }^{\infty }$: one writes the resolvent $({\Delta }_h-\lambda {)}^{-1}$ as a symbol $\sigma (\xi ,\lambda )=b_0+b_1+b_2+\cdots$, where $b_0=(|\xi {|}_{{\tau }_0}^{2}k-\lambda {)}^{-1}$ with $k=e^h$ the rescaling, and the $b_j$ solve the recursive symbol equations ${\sum }_j{b}_j\#(p-\lambda )=1$ for the symbol $p$ of ${\Delta }_h$, with $\#$ the noncommutative symbol product that here reduces to ordinary multiplication plus correction terms involving the derivations ${\delta }_j(k)$.

The coefficient $a_2$ is the contour integral $a_2=\frac{1}{2\pi i}\tau (\int b_2(\xi ,\lambda )d\lambda d\xi )$. Carrying out the $\lambda$-integration term by term produces an expression in $k$, ${\delta }_j(k)$, and ${\delta }_j{\delta }_l(k)$. The decisive simplification is that $a_2$ collects into a sum of terms each of which is a total derivation applied to a function of $k$ and its first derivatives: after using the trace identities $\tau ({\delta }_j(x))=0$ and $\tau (x{\delta }_j(y))=-\tau ({\delta }_j(x)y)$, every contribution is of the form $\tau ({\delta }_1(X_1)+{\delta }_2(X_2))$ for elements $X_j\in A_{\theta }^{\infty }$, which vanishes by trace-invariance of the derivations. Hence $a_2$ takes the same value for ${\Delta }_h$ as for ${\Delta }_0$, and for the flat ${\Delta }_0$ on the two-torus the spectrum is $\{4{\pi }^2|m+n{\tau }_0{|}^2/{\tau }_2\}$ whose zeta value at $0$ gives $a_2=0$. Therefore ${\zeta }_{{\Delta }_h}(0)+1=1$ for every Weyl factor. $\square$

Bridge. The Gauss-Bonnet theorem builds toward the full local geometry of $A_{\theta }$ and appears again in the modular curvature of the Advanced results, where dropping the trace-integration that kills $a_2$ exposes a non-constant local scalar curvature even though its total integral vanishes. The foundational reason the total curvature is invariant is exactly the classical one transcribed to the noncommutative setting: the Euler characteristic is a topological invariant, here $\chi ({\mathbb{T}}^2)=0$, and the heat-coefficient $a_2$ computes it through the noncommutative integral $\tau$, so conformal rescalings — which are the noncommutative Weyl transformations — cannot change it. This is exactly the spectral-triple philosophy of 39.06.01 in its sharpest test case: the metric data lives in $D$ and the rescaled ${\Delta }_h$, the topological data lives in $\tau$ and the zeta value, and putting these together the noncommutative torus reproduces the Gauss-Bonnet pairing of a genuine surface, the bridge being that $a_2=\tau (\text{curvature})$ is simultaneously a spectral quantity and a topological one. The same heat-coefficient machinery generalises to the scalar curvature, the dual side being the conformal geometry that the modular automorphism of $\tau$ controls.

Exercises Intermediate+

Lean formalization Intermediate+

lean_status: none — Mathlib has the matrix C*-layer and continuous functional calculus but no noncommutative torus $A_{\theta }$, no smooth subalgebra $A_{\theta }^{\infty }$ of rapidly decaying coefficients, none of the differential data (${\delta }_1,{\delta }_2$, the trace $\tau$, the flat Dirac operator $D$), and none of the geometric machinery (the conformal Laplacian, its zeta/heat expansion, the Connes-Tretkoff Gauss-Bonnet theorem, the Fathizadeh-Khalkhali modular curvature, Heisenberg modules, the $S{L}_2(\mathbb{Z})$ Morita equivalence, or the Connes-Rieffel Yang-Mills functional).

The intended statement reads schematically:

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

variable (θ : ℝ)

/-- The smooth noncommutative torus: rapidly-decaying coefficient sequences
over the universal relation v u = e^{2πiθ} u v. (Schematic: the universal
C*-algebra, its Schwartz smooth subalgebra, and the closed derivations are not
yet in Mathlib.) -/
structure SmoothNCTorus where
  coeff : ℤ × ℤ → ℂ
  schwartz : ∀ k : ℕ, ∃ C, ∀ m n, (1 + |m| + |n| : ℝ)^k * ‖coeff (m,n)‖ ≤ C

/-- Gauss-Bonnet (Connes-Tretkoff): the constant heat coefficient of the
conformally rescaled Laplacian is independent of the Weyl factor. -/
theorem ncTorus_gaussBonnet (θ : ℝ) (hθ : Irrational θ) :
    ∀ h₁ h₂ : SmoothNCTorus θ, a₂ (confLaplacian θ h₁) = a₂ (confLaplacian θ h₂) :=
  sorry  -- pseudodifferential symbol calculus + τ ∘ δⱼ = 0 cancellation

Advanced results Master

The modular scalar curvature (Fathizadeh-Khalkhali). Dropping the trace that kills $a_2$ in Gauss-Bonnet exposes a genuine local scalar curvature. Writing the conformally rescaled metric with Weyl factor $k=e^h$, the second heat coefficient becomes, for $b\in A_{\theta }^{\infty }$, $$ a_2(b, \Delta_h) = \frac{1}{4\pi},\tau\Big( b,\big( K(\nabla)(\delta_j\delta_j h) + H(\nabla,\nabla)(\delta_j h, \delta_j h) \big)\Big), $$ where $K$ and $H$ are explicit functions of the modular operator $\nabla =-\log {\Delta }_k$ (${\Delta }_k=$ the modular automorphism implemented by $k$), and ${\delta }_{j}h$ enters through the noncommutative gradient. The functions $K(x)=\frac{2(x+\sinh x-x\cosh x)}{x^2\sinh (x/2)}$ (modular curvature function) and the two-variable $H$ were computed in closed form by Connes-Moscovici and Fathizadeh-Khalkhali ^{[Fathizadeh-Khalkhali 2013]}. They reduce to the classical constant when $\theta$ is rational and the algebra commutes ($\nabla \to 0$, $K\to 1$), recovering $R=e^{-h}(\Delta h)$; the modular correction $H\ne 0$ is the purely noncommutative signature of the non-tracial state $\tau (k\cdot )$. This is the deepest computation in the local geometry of $A_{\theta }$ and required the spectral functions of the modular automorphism group of Tomita-Takesaki theory 39.04.01.

Projective modules and the structure of $K$-theory. The finitely generated projective modules over $A_{\theta }^{\infty }$ are classified by Rieffel: a module is determined up to isomorphism by its class in $K_0(A_{\theta })\cong {\mathbb{Z}}^2$, with the trace giving a real-valued dimension ${\dim }_{\tau }(\mathcal{E})=c\theta +d$ for the module ${\mathcal{E}}_{c,d}$ corresponding to $(c,d)$. The basic non-free module is the Heisenberg module $\mathcal{E}=\mathcal{S}(\mathbb{R})$, the Schwartz functions on the line, with $A_{\theta }^{\infty }$ acting through the Heisenberg commutation relations $(u\xi )(x)=\xi (x-\theta )$, $(v\xi )(x)=e^{2\pi ix}\xi (x)$; its endomorphism algebra is $A_{{\theta }^{\prime }}^{\infty }$ for ${\theta }^{\prime }=-1/\theta$, the bimodule that implements Morita equivalence $A_{\theta }\sim A_{-1/\theta }$ ^{[Rieffel 1988]}.

The $S{L}_2(\mathbb{Z})$ action and Morita equivalence. Combining the flip $\theta \mapsto \theta +1$ (giving the isomorphic algebra) and the Heisenberg-module equivalence $\theta \mapsto -1/\theta$ generates the modular group $S{L}_2(\mathbb{Z})$ acting by $\theta \mapsto \frac{a\theta +b}{c\theta +d}$. The theorem of Rieffel and Connes is that $A_{\theta }$ and $A_{{\theta }^{\prime }}$ are Morita equivalent iff ${\theta }^{\prime }=\frac{a\theta +b}{c\theta +d}$ for some \begin{psmallmatrix} a&b\\c&d\end{psmallmatrix}\in SL_2(\mathbb{Z}), and isomorphic iff ${\theta }^{\prime }\equiv \pm \theta \mathrm{mod}1$ (the subgroup generated by the flip alone, 39.02.01). The Morita-equivalence classes are thus the orbits of the modular group on $\mathbb{R}\setminus \mathbb{Q}$, the same orbits that classify the cusps in the theory of modular forms — the arithmetic of the noncommutative torus is the arithmetic of $S{L}_2(\mathbb{Z})\backslash \mathbb{H}$.

Yang-Mills on the noncommutative torus (Connes-Rieffel). A Hermitian connection on a projective module $\mathcal{E}$ is a pair $\nabla =({\nabla }_1,{\nabla }_2)$ of operators with ${\nabla }_j(a\xi )={\delta }_j(a)\xi +a{\nabla }_j(\xi )$, and its curvature is $\Theta =[{\nabla }_1,{\nabla }_2]$. The Yang-Mills action is $\mathrm{YM}(\nabla )=-{\tau }_{\mathcal{E}}({\Theta }^2)$; minimising it within a fixed module (fixed topological degree ${\tau }_{\mathcal{E}}(\Theta )$) yields the constant-curvature connections $\Theta =2\pi i\mu \cdot \mathrm{id}$ with $\mu =\mathrm{deg}(\mathcal{E})/{\dim }_{\tau }(\mathcal{E})$, by the Cauchy-Schwarz argument of Exercise 7. The moduli space of Yang-Mills minimisers on ${\mathcal{E}}_{c,d}$ is itself a (commutative) torus of dimension equal to $2|\det |$ of the degree data — the noncommutative gauge theory has a classical moduli space, and this construction was the first concrete Yang-Mills theory on a noncommutative manifold ^{[Connes-Rieffel 1987]}. In the almost-commutative product $C^{\infty }(M)\otimes A_{\theta }$ the same fluctuations produce gauge fields on $M$ valued in the noncommutative torus, a mechanism studied in matrix-model and string compactifications on $A_{\theta }$.

Synthesis. The noncommutative torus is the central insight of noncommutative differential geometry made fully computable, and putting these together is exactly the demonstration that the spectral-triple axioms of 39.06.01 are not vacuous: a single irrational $\theta$ produces an algebra with no points whose every classical geometric structure has a faithful noncommutative counterpart. The derivations ${\delta }_1,{\delta }_2$ are dual to the vector fields ${\partial }_1,{\partial }_2$; the trace $\tau$ is the foundational reason an integral exists and is exactly the Dixmier-trace integral 39.06.05 of the flat triple; the flat Dirac operator $D=\sum {\delta }_j\otimes {\gamma }^j$ generalises the spinor Dirac operator and reproduces the Connes distance formula 39.06.02 on the state space. The Gauss-Bonnet theorem is dual to the topological invariance of the Euler characteristic; the modular curvature is the central insight that the noncommutative metric carries a Tomita-Takesaki modular automorphism 39.04.01 absent classically; and Morita equivalence under $S{L}_2(\mathbb{Z})$ is exactly the K-theoretic shadow of the modular group, so the arithmetic of $A_{\theta }$ is the arithmetic of the modular curve. The bridge is that the noncommutative torus simultaneously realises every layer of the program — metric, integral, curvature, gauge theory, and arithmetic — in a single example built from two unitaries and one number.

Full proof set Master

Proposition (the derivations are well-defined commuting trace-invariant $\ast$-derivations). Define ${\delta }_j$ on monomials by ${\delta }_1(u^m{v}^n)=2\pi im{u}^m{v}^n$ and ${\delta }_2(u^m{v}^n)=2\pi in{u}^m{v}^n$, extended linearly. For $a=\sum a_{mn}{u}^m{v}^n\in A_{\theta }^{\infty }$, rapid decay of $(a_{mn})$ forces rapid decay of $(m{a}_{mn})$ and $(n{a}_{mn})$, so ${\delta }_j(a)\in A_{\theta }^{\infty }$ and ${\delta }_j$ is everywhere defined on the smooth subalgebra. The Leibniz rule holds because on monomials ${\delta }_1(u^m{v}^n\cdot u^{m^{\prime }}{v}^{n^{\prime }})=2\pi i(m+m^{\prime })(u^m{v}^n{u}^{m^{\prime }}{v}^{n^{\prime }})$ — the twisting phase from $vu=e^{2\pi i\theta }uv$ is a scalar and does not affect the total $u$-degree — which equals ${\delta }_1(u^m{v}^n)u^{m^{\prime }}{v}^{n^{\prime }}+u^m{v}^n{\delta }_1(u^{m^{\prime }}{v}^{n^{\prime }})$. Commutativity ${\delta }_1{\delta }_2={\delta }_2{\delta }_1$ is immediate since both act diagonally with eigenvalues $2\pi im$ and $2\pi in$. The $\ast$-property ${\delta }_j(a^{\ast })={\delta }_j(a{)}^{\ast }$ follows from $(u^m{v}^n{)}^{\ast }=e^{2\pi i\theta mn}{u}^{-m}{v}^{-n}$ and the sign flip $m\mapsto -m$. Trace-invariance $\tau ({\delta }_j(a))=2\pi i{\sum }_{?}(\cdots )$ holds because ${\delta }_j$ multiplies the $(0,0)$ coefficient by $0$, so $\tau ({\delta }_j(a))=2\pi i\cdot 0\cdot a_{00}=0$. $\square$

Proposition (integration by parts). For $a,b\in A_{\theta }^{\infty }$, $\tau ({\delta }_j(a)b)=-\tau (a{\delta }_j(b))$. Apply $\tau \circ {\delta }_j=0$ to the product: $0=\tau ({\delta }_j(ab))=\tau ({\delta }_j(a)b+a{\delta }_j(b))=\tau ({\delta }_j(a)b)+\tau (a{\delta }_j(b))$ by linearity of $\tau$ and the Leibniz rule. Rearranging gives the identity, the skew-symmetry of ${\delta }_j$ for the inner product $⟨a,b⟩=\tau (a^{\ast }b)$. $\square$

Proposition (the flat Laplacian has metric dimension $2$). The Laplacian $\Delta =-({\delta }_1^2+{\delta }_2^2)$ acts on the orthonormal basis $\{u^m{v}^n\}$ of ${\mathcal{H}}_0$ with eigenvalue $4{\pi }^2(m^2+n^2)$. The eigenvalue-counting function is $N(\Lambda )=\#\{(m,n):4{\pi }^2(m^2+n^2)\le \Lambda \}=\#\{\text{lattice points in a disc of radius }\sqrt{\Lambda }/(2\pi )\}\sim \pi \cdot \frac{\Lambda }{4{\pi }^2}=\frac{\Lambda }{4\pi }$ by Gauss's circle problem. Hence $N(\Lambda )\sim c{\Lambda }^1$, so for the Dirac operator $D$ with $|D{|}^2=\Delta$ the singular values satisfy ${\mu }_k\sim c^{\prime }\sqrt{k}$, giving $|D{|}^{-1}\in {\mathcal{L}}^{2,\infty }$ and metric dimension $2=\dim {\mathbb{T}}^2$. The zeta function ${\zeta }_{\Delta }(s)={\mathrm{Tr}}^{\prime }{\Delta }^{-s}=(4{\pi }^2{)}^{-s}{\sum }_{(m,n)\ne 0}(m^2+n^2{)}^{-s}$ is an Epstein zeta function with its rightmost pole at $s=1$, confirming the top of the dimension spectrum at $2s=2$. $\square$

Proposition (the Heisenberg module implements Morita equivalence $A_{\theta }\sim A_{-1/\theta }$). Let $\mathcal{E}=\mathcal{S}(\mathbb{R})$ with the left $A_{\theta }^{\infty }$-action $(u\xi )(x)=\xi (x-\theta )$, $(v\xi )(x)=e^{2\pi ix}\xi (x)$. One checks $vu\xi (x)=e^{2\pi ix}\xi (x-\theta )$ and $uv\xi (x)=e^{2\pi i(x-\theta )}\xi (x-\theta )$, so $vu=e^{2\pi i\theta }uv$ on $\mathcal{E}$, a representation of $A_{\theta }$. Define operators $U,V$ commuting with the $A_{\theta }$-action by $(U\xi )(x)=\xi (x-1/\theta \cdot \theta )=\xi (x-1)$ rescaled and $(V\xi )(x)=e^{2\pi ix/\theta }\xi (x)$; they satisfy $VU=e^{-2\pi i/\theta }UV$, generating a right action of $A_{{\theta }^{\prime }}^{\infty }$ with ${\theta }^{\prime }=-1/\theta$. The $A_{\theta }$-valued and $A_{{\theta }^{\prime }}$-valued inner products built from $⟨\xi ,\eta {⟩}_{A_{\theta }}={\sum }_{m,n}⟨\xi ,u^m{v}^n\eta ⟩u^{-m}{v}^{-n}$ exhibit $\mathcal{E}$ as an $A_{\theta }$-$A_{{\theta }^{\prime }}$ equivalence bimodule (full and projective on both sides), so $A_{\theta }$ and $A_{-1/\theta }$ are Morita equivalent. The trace-dimension is ${\dim }_{\tau }\mathcal{E}=\theta$, the value realised by the Rieffel projection. $\square$

Proposition (Yang-Mills minimisers are constant-curvature connections). On a projective module $\mathcal{E}$ with compatible connection $\nabla =({\nabla }_1,{\nabla }_2)$ and curvature $\Theta =[{\nabla }_1,{\nabla }_2]$, the trace ${\tau }_{\mathcal{E}}(\Theta )$ depends only on the module class (it is $2\pi i$ times the topological degree, constant under deformation of $\nabla$ since a connection change shifts $\Theta$ by a commutator, which is ${\tau }_{\mathcal{E}}$-traceless). Minimise $\mathrm{YM}(\nabla )={\tau }_{\mathcal{E}}({\Theta }^{\ast }\Theta )$ at fixed ${\tau }_{\mathcal{E}}(\Theta )$. The Cauchy-Schwarz inequality for the faithful trace, ${\tau }_{\mathcal{E}}({\Theta }^{\ast }\Theta ){\tau }_{\mathcal{E}}(1)\ge |{\tau }_{\mathcal{E}}(\Theta ){|}^2$, with equality iff $\Theta$ is a scalar multiple of the identity endomorphism, gives $\mathrm{YM}(\nabla )\ge |{\tau }_{\mathcal{E}}(\Theta ){|}^2/{\dim }_{\tau }(\mathcal{E})$, attained exactly when $\Theta =2\pi i\mu \cdot \mathrm{id}$ with $\mu =\mathrm{deg}(\mathcal{E})/{\dim }_{\tau }(\mathcal{E})$. These are the constant-curvature connections; the Yang-Mills minimum is the topological lower bound, the noncommutative Bogomolny bound. $\square$

Connections Master

• AF algebras and the irrational rotation algebra 39.02.01 — this unit's underlying C*-algebra $A_{\theta }$, its universal presentation $vu=e^{2\pi i\theta }uv$, unique trace $\tau$, and trace image $\mathbb{Z}+\theta \mathbb{Z}={\tau }_{\ast }(K_0)$ are exactly the objects constructed there; the present unit adds the smooth subalgebra and the differential geometry that turns the C*-algebra into a noncommutative manifold.

• Spectral triples and the reconstruction theorem 39.06.01 — the flat triple $(A_{\theta }^{\infty },\mathcal{H},D)$ with $D={\sum }_j{\delta }_j\otimes {\gamma }^j$ is the prototype noncommutative (non-reconstructible) spectral triple; it satisfies the bounded-commutator and compact-resolvent axioms but, being genuinely noncommutative, falls outside the commutative reconstruction theorem, making it the canonical illustration of a "manifold without points."

• The Connes distance formula 39.06.02 — the seminorm $\Vert [D,a]\Vert =({\sum }_j\Vert {\delta }_j(a){\Vert }^2{)}^{1/2}$ of the flat triple is the Lipschitz seminorm entering the distance formula on the state space of $A_{\theta }$, making $A_{\theta }$ Rieffel's prototype compact quantum metric space.

• The Dixmier trace and the noncommutative integral 39.06.05 — the canonical trace $\tau$ is recovered as the Dixmier-trace integral $\tau (a)=\frac{1}{2\pi }{\mathrm{Tr}}_{\omega }(a|D{|}^{-2})$ of the flat triple, so the averaging functional of this unit is the metric-dimension-$2$ noncommutative integral of the spectral triple.

• Tomita-Takesaki modular theory 39.04.01 — the modular scalar curvature of Fathizadeh-Khalkhali is governed by the modular automorphism group of the non-tracial weight $\tau (k\cdot )$, so the local geometry of the conformally rescaled noncommutative torus is controlled by exactly the modular operator $\nabla =-\log {\Delta }_k$ of modular theory.

Historical & philosophical context Master

The noncommutative torus entered the subject as the simplest genuinely noncommutative space. Rieffel's 1981 paper on C*-algebras associated with irrational rotations constructed the projections of trace $\theta$ that fix its isomorphism class ^{[Rieffel 1981]}, and Connes had already singled out $A_{\theta }$ as the test object for noncommutative differential geometry, equipping it with the derivations ${\delta }_1,{\delta }_2$ and the trace in his work of the early 1980s, consolidated in Noncommutative Geometry (1994) Ch. VI §6 ^{[Connes Ch. VI]}. The Connes-Rieffel paper of 1987 built the first Yang-Mills theory over it, computing the constant-curvature minimisers and the moduli space ^{[Connes-Rieffel 1987]}, and Rieffel's 1988 study of projective modules over higher-dimensional tori gave the Heisenberg modules and the full $S{L}_2(\mathbb{Z})$ Morita-equivalence picture ^{[Rieffel 1988]}.

The local Riemannian geometry was opened much later. Connes and Tretkoff proved the Gauss-Bonnet theorem for the conformally rescaled $A_{\theta }$ in 2009-2011 ^{[Connes-Tretkoff 2011]}, and Fathizadeh and Khalkhali, in parallel with Connes and Moscovici, computed the modular scalar curvature in closed form in 2011-2013, discovering the modular curvature function $K$ and the appearance of the Tomita-Takesaki modular operator in a purely geometric quantity ^{[Fathizadeh-Khalkhali 2013]}. The noncommutative torus also became a fixture of physics: it is the algebra of functions on the worldvolume of D-branes in a constant $B$-field, the setting for the Seiberg-Witten map and matrix-model compactifications, so the irrational parameter $\theta$ reappears as a background flux. Its arithmetic — Morita orbits equal to $S{L}_2(\mathbb{Z})$ orbits on $\mathbb{R}\setminus \mathbb{Q}$ — ties it to modular forms and to Manin's program relating real multiplication to noncommutative geometry.

Bibliography Master

• Connes, A., Noncommutative Geometry, Academic Press, 1994. Ch. VI §6.
• Connes, A. and Rieffel, M. A., "Yang-Mills for noncommutative two-tori", in Operator Algebras and Mathematical Physics, Contemporary Mathematics 62 (1987), 237-266.
• Rieffel, M. A., "C*-algebras associated with irrational rotations", Pacific Journal of Mathematics 93 (1981), 415-429.
• Rieffel, M. A., "Projective modules over higher-dimensional noncommutative tori", Canadian Journal of Mathematics 40 (1988), 257-338.
• Connes, A. and Tretkoff, P., "The Gauss-Bonnet theorem for the noncommutative two torus", in Noncommutative Geometry, Arithmetic, and Related Topics, Johns Hopkins University Press (2011), 141-158.
• Fathizadeh, F. and Khalkhali, M., "Scalar curvature for the noncommutative two torus", Journal of Noncommutative Geometry 7 (2013), 1145-1183.
• Connes, A. and Moscovici, H., "Modular curvature for noncommutative two-tori", Journal of the American Mathematical Society 27 (2014), 639-684.
• Gracia-Bondía, J. M., Várilly, J. C. and Figueroa, H., Elements of Noncommutative Geometry, Birkhäuser, 2001. Ch. 12.

---

Operator-algebras spine, noncommutative-geometry chapter. The flagship worked example of NCG: the smooth noncommutative torus $A_{\theta }^{\infty }$, its derivations ${\delta }_1,{\delta }_2$, trace $\tau$, and flat spectral triple; the Gauss-Bonnet theorem and modular scalar curvature; Heisenberg modules and $S{L}_2(\mathbb{Z})$ Morita equivalence; and Connes-Rieffel Yang-Mills. Builds on spectral triples (39.06.01) and the irrational rotation algebra $A_{\theta }$ (39.02.01).