The Modular Automorphism Group and the KMS Condition

Intuition Beginner

The previous unit handed the algebra a clock: from one honest arrow came a single dial you can turn by any amount of time, and turning it carries every operator in the algebra to another operator still inside the algebra. This unit asks a sharper question. Of all the clocks one could imagine attaching to the algebra, why is this one special — and is there only one?

The answer hides in a balance condition. Picture two operators and a state, which you can think of as a way of averaging — like reading off the temperature-weighted expectation of a measurement. Take the average of "the first operator, run forward on the clock, then the second." Now take the average with the two operators swapped: "the second, then the first run forward." In general these differ. The balance condition says the two are not unrelated: one becomes the other if you let the clock run forward by a fixed imaginary amount of time. Swapping the order costs exactly one tick down into imaginary time.

That one rule is the signature of thermal equilibrium — a system sitting at a fixed temperature, settled, no net flow. The remarkable fact is that it pins the clock down completely: there is exactly one dial for which the state is in balance, and it is the modular clock. So the algebra's internal time is not a choice. It is forced by the demand that the state be at rest.

Visual Beginner

Run the first operator forward on the clock, then average with the second; swapping the order is the same average shifted by one full tick into imaginary time. That equality across the strip is the balance rule.

The dictionary: the clock is the modular flow ${\sigma }_t$; the average is the state $\phi$; the bottom edge is $\phi ({\sigma }_t(a)b)$ and the top edge is $\phi (b{\sigma }_t(a))$; the shaded sheet is a single function analytic in the strip of width one; and the balance rule that joins the edges is the KMS condition, satisfied by exactly one clock.

Worked example Beginner

Take the smallest thermal system: a single two-level cell — spin up or spin down — at a fixed temperature. The state weights "up" and "down" by two positive numbers that add to one; call them the up-weight and the down-weight, with up heavier than down. The clock is the natural one for this cell: it leaves the two levels alone but spins their relative phase, faster for the gap between the levels.

Now test the balance rule on the two operators "raise" (turn down into up) and "lower" (turn up into down). Average "raise run forward, then lower": running raise forward only adds a phase, and the average of raise-then-lower reads off the down-weight, because lowering after raising returns you to a level the state must already occupy as "down." Average the swapped order, "lower then raise": this reads off the up-weight instead. The two averages differ by exactly the ratio of the up-weight to the down-weight.

What this tells us: that ratio is precisely what one tick of imaginary time on this clock multiplies by. So the swapped average equals the original average pushed one tick into imaginary time — the balance rule holds, on the nose, for this clock and no other. With equal weights the ratio is one, the two averages agree with no shift, and the clock stands still: that is the infinite-temperature, no-flow case. The moment the weights differ, the clock must tick, and the balance rule is what forces its exact rate.

Check your understanding Beginner

Formal definition Intermediate+

Let $M\subseteq B(H)$ be a von Neumann algebra 39.03.01 with a cyclic and separating vector $\xi$ 39.04.01 implementing the faithful normal state $\phi ={\omega }_{\xi }$, $\phi (a)=⟨a\xi ,\xi ⟩$. Let $S=J{\Delta }^{1/2}$ be the polar decomposition of the closure of $S_0:a\xi \mapsto a^{\ast }\xi$, with modular operator $\Delta =S^{\ast }S$ and modular conjugation $J$ 39.04.02.

The modular automorphism group of $\phi$ is the one-parameter group $$ \sigma_t^\varphi : M \to M, \qquad \sigma_t^\varphi(a) = \Delta^{it}, a, \Delta^{-it} \qquad (t \in \mathbb{R}), $$ where ${\Delta }^{it}=\exp (it\log \Delta )$ is the strongly continuous unitary group from the Borel functional calculus 02.11.03. Tomita's theorem 39.04.02 is exactly the assertion that ${\sigma }_t^{\phi }(M)=M$, so each ${\sigma }_t^{\phi }$ is a $\ast$-automorphism of $M$ and $t\mapsto {\sigma }_t^{\phi }$ is a homomorphism $\mathbb{R}\to \mathrm{Aut}(M)$, $\sigma$-weakly continuous in $t$. The state is invariant, $\phi \circ {\sigma }_t^{\phi }=\phi$, because ${\Delta }^{it}\xi =\xi$.

An element $a\in M$ is $\sigma$-analytic (entire analytic) if $t\mapsto {\sigma }_t^{\phi }(a)$ extends to an entire $M$-valued function $z\mapsto {\sigma }_z^{\phi }(a)$; such elements form a $\sigma$-weakly dense $\ast$-subalgebra (obtained by smearing, $a_n=\sqrt{n/\pi }\int e^{-n{t}^2}{\sigma }_t^{\phi }(a)dt$). For analytic $a$, ${\sigma }_z^{\phi }(a)={\Delta }^{iz}a{\Delta }^{-iz}$ makes sense on a dense domain.

Definition (KMS condition). A $\sigma$-weakly continuous one-parameter automorphism group $\alpha =({\alpha }_t{)}_{t\in \mathbb{R}}$ of $M$ and a normal state $\phi$ satisfy the modular KMS condition (Kubo-Martin-Schwinger) at inverse temperature $\beta =1$ if for every $a,b\in M$ there is a function $F_{a,b}$, bounded and continuous on the closed strip $\overline{D_1}=\{z\in \mathbb{C}:0\le \mathrm{Im}z\le 1\}$ and holomorphic in its interior, with boundary values $$ F_{a,b}(t) = \varphi\big(\alpha_t(a),b\big), \qquad F_{a,b}(t + i) = \varphi\big(b,\alpha_t(a)\big) \qquad (t \in \mathbb{R}). $$ The state $\phi$ is then called an $(\alpha ,\beta )$-KMS state, with the convention $\beta =1$ absorbing the rescaling ${\alpha }_t⇝{\alpha }_{-\beta t}$ for general $\beta$. The boundary swap $\phi ({\alpha }_t(a)b)\leftrightarrow \phi (b{\alpha }_t(a))$ across the strip is the analytic interpolation that replaces commutativity.

Counterexamples to common slips

• ${\sigma }_t^{\phi }$ depends on $\phi$, not on $M$ alone. Two faithful normal states give different modular flows; only the class $[{\sigma }^{\phi }]\in \mathrm{Out}(M)=\mathrm{Aut}(M)/\mathrm{Inn}(M)$ is intrinsic. Treating "the modular group of $M$" as well defined as an automorphism (rather than as an outer class) is the standard error.
• The sign and width conventions vary. The convention here is $\beta =1$, strip $0\le \mathrm{Im}z\le 1$, boundary swap $\phi ({\sigma }_t(a)b)\to \phi (b{\sigma }_t(a))$ at $\mathrm{Im}z=1$. Bratteli-Robinson and Haag-Hugenholtz-Winnink use $\beta$ explicitly with ${\sigma }_t={\alpha }_{-t}$; mismatched signs invert $\Delta$ and flip the strip. State the convention before computing.
• KMS is not just invariance. Every ${\sigma }_t^{\phi }$ leaves $\phi$ invariant, but invariance under a flow is far weaker than the analytic boundary swap; many flows fix $\phi$ and only one satisfies KMS. The interpolation across the strip, not the fixed-point property, is the content.

Key theorem with proof Intermediate+

Theorem (the modular flow is KMS, and is the unique such flow). Let $M$ be a von Neumann algebra with faithful normal state $\phi$ and modular automorphism group ${\sigma }^{\phi }$. Then $\phi$ satisfies the KMS condition at $\beta =1$ with respect to ${\sigma }^{\phi }$. Conversely, if $\alpha =({\alpha }_t)$ is any $\sigma$-weakly continuous one-parameter automorphism group of $M$ for which $\phi$ is an $(\alpha ,1)$-KMS state, then ${\alpha }_t={\sigma }_t^{\phi }$ for all $t$. ^{[Bratteli-Robinson §2.5.3; Takesaki Ch. VIII]}

Proof. Existence. Fix $a,b\in M$. The vector $a\xi$ lies in the domain of ${\Delta }^{1/2}$ (it lies in the domain of $S$), so $z\mapsto {\Delta }^{iz}a\xi$ is defined and holomorphic on $0\le \mathrm{Im}z\le 1/2$ with ${\Delta }^{1/2}a\xi =JSa\xi =J{a}^{\ast }\xi$ at the far edge. Define $$ F_{a,b}(z) = \langle \Delta^{iz} a, \Delta^{-iz}\xi,; b^\xi\rangle. $$ Since ${\Delta }^{-iz}\xi =\xi$ for real $z$ and ${\Delta }^{it}\xi =\xi$ everywhere, ${\Delta }^{-iz}\xi =\xi$ extends to $\xi$ on the strip, so $F_{a,b}(z) = \langle \Delta^{iz}a\xi, b^\xi\rangle$.Thevector$a\xi \in \mathcal{D}(\Delta^{1/2})$makes$z \mapsto \Delta^{iz}a\xi$bounded-holomorphicon$0 \le \operatorname{Im} z \le 1/2$;thecompanionestimateon$b^\xi \in \mathcal D(\Delta^{1/2})$extendsthepairingtothefullstrip$0 \le \operatorname{Im} z \le 1$.Ontherealaxis$F_{a,b}(t) = \langle \Delta^{it}a\Delta^{-it}\xi, b^\xi\rangle = \langle \sigma_t^\varphi(a)\xi, b^\xi\rangle = \langle b,\sigma_t^\varphi(a)\xi, \xi\rangle = \varphi(b,\sigma_t^\varphi(a))$—tomatchthestatednormalisationweinsteadtake$F_{a,b}(z) = \langle \Delta^{iz}a\xi, b^\xi\rangle$with$F(t) = \varphi(\sigma_t^\varphi(a)b)$aftermoving$b$across;werecordthecleancomputation.At$z = t + i$,write$\Delta^{i(t+i)} = \Delta^{it}\Delta^{-1}$anduse$\Delta^{1/2}a\xi = Ja^\xi$,$\Delta^{1/2}b^\xi = Jb\xi$togetherwithantiunitarityof$J$: $$ F_{a,b}(t+i) = \langle \Delta^{1/2}\Delta^{it}a\xi,; \Delta^{-1/2}b^\xi\rangle = \langle J\sigma_t^\varphi(a)^\xi,; \Delta^{-1/2}b^\xi\rangle = \varphi\big(b,\sigma_t^\varphi(a)\big), $$ the last equality unwinding $\Delta^{-1/2}b^\xi = F b^*\xi = Jb\xi$andtheantiunitaryinner-productflip.Thusthesingleboundedholomorphic$F_{a,b}$hasboundaryvalues$\varphi(\sigma_t^\varphi(a)b)$and$\varphi(b,\sigma_t^\varphi(a))$, which is the KMS condition.

Uniqueness. Suppose $\phi$ is also $(\alpha ,1)$-KMS for a $\sigma$-weakly continuous automorphism group $\alpha$. Both flows fix $\phi$, so both are implemented on $H_{\phi }=H$ by unitary groups fixing $\xi$: ${\sigma }_t^{\phi }=\mathrm{Ad}({\Delta }^{it})$ with ${\Delta }^{it}\xi =\xi$, and ${\alpha }_t=\mathrm{Ad}(u_t)$ with $u_t\xi =\xi$ (the canonical implementation in the cone, $u_t=$ the standard unitary of ${\alpha }_t$ 39.04.01). Set $w_t=u_t{\Delta }^{-it}$, a unitary with $w_t\xi =\xi$. For $a,b\in M$ the two KMS functions for $a,b$ agree because a bounded holomorphic function on the strip is determined by either boundary; matching the boundary identity $\phi ({\alpha }_t(a)b)=\phi ({\sigma }_t^{\phi }(a)b)$ obtained by comparing the two interpolations forces $⟨u_{t}a{u}_t^{\ast }\xi ,b^{\ast }\xi ⟩=⟨{\Delta }^{it}a{\Delta }^{-it}\xi ,b^{\ast }\xi ⟩$ for all $a,b$, hence ${\alpha }_t={\sigma }_t^{\phi }$ on the $\sigma$-weakly dense set, hence everywhere. Concretely: the function $g(z)=⟨(u_t{\Delta }^{-it}){\sigma }_z^{\phi }(a)\xi ,\eta ⟩-⟨{\sigma }_z^{\phi }(a)\xi ,\eta ⟩$ is bounded holomorphic on the strip and vanishes on both boundaries by the matching KMS interpolations, so vanishes identically; with $a$ analytic and $\eta$ dense this gives $u_t{\Delta }^{-it}=1$, i.e. $u_t={\Delta }^{it}$ and ${\alpha }_t={\sigma }_t^{\phi }$. $\square$

Bridge. This uniqueness builds toward the Connes classification of type III factors and appears again in 39.04.02's Bisognano-Wichmann identification, where the geometrically defined boost flow must coincide with the abstract modular flow precisely because the vacuum is KMS for the boost. The foundational reason the proof closes is exactly that a bounded holomorphic function on the strip is fixed by its two boundary values, so two flows sharing the KMS interpolation of one state share the state's analytic fingerprint and therefore coincide — this is exactly the rigidity Haag, Hugenholtz, and Winnink isolated. The modular flow generalises the finite-dimensional Gibbs evolution $a\mapsto {\rho }^{it}a{\rho }^{-it}$ to the trace-free world, and the KMS condition is dual to the variational characterisation of equilibrium in statistical mechanics. Putting these together, the modular time is the unique intrinsic dynamics a state confers on its algebra, and the bridge is that this uniqueness is what makes the outer class $[{\sigma }^{\phi }]$ — examined next — an invariant of $M$ itself.

Exercises Intermediate+

Lean formalization Intermediate+

lean_status: none — Mathlib has the Borel functional calculus giving ${\Delta }^{it}$, strongly continuous unitary groups, and strip-holomorphy machinery, but no VonNeumannAlgebra predicate (a gap inherited from 39.03.01 / 39.04.01), no modularAutomorphismGroup, no IsKMS predicate for the analytic boundary condition, no uniqueness theorem (modular flow = unique KMS flow), and nothing of the Connes cocycle $(D\psi :D\phi {)}_t$, the relative modular operator, or the canonical $\delta :\mathbb{R}\to \mathrm{Out}(M)$.

The intended statements read schematically:

import Mathlib.Analysis.InnerProductSpace.Spectrum
import Mathlib.Analysis.Complex.Strip

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

/-- The modular automorphism group σ_t(a) = Δ^{it} a Δ^{-it} of a state φ. -/
def modularFlow (M : Set (H →L[ℂ] H)) (ξ : H) (t : ℝ) : (H →L[ℂ] H) → (H →L[ℂ] H) := sorry

/-- φ satisfies the KMS condition at β = 1 for a flow α: for all a b there is F
    holomorphic on {0 ≤ im z ≤ 1} with F(t)=φ(α t a · b), F(t+i)=φ(b · α t a). -/
def IsKMS (M : Set (H →L[ℂ] H)) (φ : (H →L[ℂ] H) → ℂ)
    (α : ℝ → (H →L[ℂ] H) → (H →L[ℂ] H)) : Prop := sorry

/-- The modular flow is the unique β=1 KMS flow for φ (Tomita-Takesaki-Winnink). -/
theorem modularFlow_unique_kms
    (M : Set (H →L[ℂ] H)) (hM : IsVonNeumannAlgebra M) (ξ : H)
    (hcyc : IsCyclic M ξ) (hsep : IsSeparating M ξ)
    (α : ℝ → (H →L[ℂ] H) → (H →L[ℂ] H))
    (hα : IsKMS M (fun a => ⟪a ξ, ξ⟫_ℂ) α) :
    α = modularFlow M ξ :=
  sorry  -- bounded strip-holomorphic function fixed by both boundaries

Advanced results Master

The KMS condition turns the modular flow from a construction into a characterisation: the flow is what the equilibrium state knows, and the results below organise the consequences once that identification is fixed.

The Connes cocycle Radon-Nikodym theorem. For faithful normal states (or weights) $\phi ,\psi$ on $M$ there is a $\sigma$-strongly continuous family of unitaries $(D\psi :D\phi {)}_t\in M$, the cocycle derivative, with ${\sigma }_t^{\psi }=\mathrm{Ad}((D\psi :D\phi {)}_t)\circ {\sigma }_t^{\phi }$, the cocycle identity $(D\psi :D\phi {)}_{t+s}=(D\psi :D\phi {)}_t{\sigma }_t^{\phi }((D\psi :D\phi {)}_s)$, and the chain rule $(D\chi :D\phi {)}_t=(D\chi :D\psi {)}_t(D\psi :D\phi {)}_t$ ^{[Connes 1973]}. It is constructed from the modular flow of the balanced weight $\theta =\phi \oplus \psi$ on $M\otimes M_2(\mathbb{C})$ by $(D\psi :D\phi {)}_t={\sigma }_t^{\theta }(e_{21})e_{12}$ where $e_{ij}$ are the matrix units, or equivalently from the relative modular operator ${\Delta }_{\psi ,\phi }$ of the antilinear $S_{\psi ,\phi }:a{\xi }_{\phi }\mapsto a^{\ast }{\xi }_{\psi }$ by $(D\psi :D\phi {)}_t={\Delta }_{\psi ,\phi }^{it}{\Delta }_{\phi }^{-it}$. The cocycle is the noncommutative Radon-Nikodym derivative $d\psi /d\phi$, refined to a one-parameter object because $\psi$ and $\phi$ need not commute.

The canonical outer time. Because $(D\psi :D\phi {)}_t\in M$ is inner-implementing, the class $[{\sigma }^{\phi }]$ in $\mathrm{Out}(M)=\mathrm{Aut}(M)/\mathrm{Inn}(M)$ is independent of $\phi$, yielding a homomorphism $\delta :\mathbb{R}\to \mathrm{Out}(M)$ canonical to $M$ — the modular spectrum / canonical flow. This is Connes' "Tomita-Takesaki gives a von Neumann algebra a canonical time evolution, defined up to inner automorphism." For semifinite $M$ a trace gives $\Delta =1$, so $\delta$ vanishes; $\delta$ is nonzero exactly in type III, where no trace exists and the modular flow is properly outer. The $T$-invariant $T(M)=\{t:{\sigma }_t^{\phi }\in \mathrm{Inn}(M)\}=\ker \delta$ and the Connes spectrum $S(M)={\bigcap }_{\phi }\mathrm{Spec}({\Delta }_{\phi })$ are the two invariants read off $\delta$, splitting the type III factors into III${}_0$, III${}_{\lambda }$ ($0<\lambda <1$), and III${}_1$.

The KMS variational and stability characterisations. Beyond the boundary condition, KMS states are singled out among invariant states by passivity (no cyclic process extracts work: $-i{\sum }_k\omega (u_k^{\ast }[\cdot ,u_k])\ge 0$ for unitaries built from the dynamics, the Pusz-Woronowicz second law) and by return to equilibrium / stability under local perturbations. For a perturbation $\phi ⇝{\phi }^h$ by a self-adjoint $h=h^{\ast }\in M$, the perturbed state is again KMS for the perturbed flow ${\sigma }_t^{{\phi }^h}$ with cocycle $(D{\phi }^h:D\phi {)}_t$ given by an Araki-Dyson expansion ${\sum }_n{\int }_{0\le s_1\le \cdots \le s_n\le t}{\sigma }_{s_1}^{\phi }(h)\cdots {\sigma }_{s_n}^{\phi }(h)ds$, the noncommutative analogue of the interaction-picture / Duhamel series. This is the operator-algebraic form of the statement that equilibrium is thermodynamically stable.

Continuous decomposition and Takesaki duality. The crossed product $M{⋊}_{{\sigma }^{\phi }}\mathbb{R}$ of a type III factor by its modular flow is a type II${}_{\infty }$ von Neumann algebra $N$ carrying a faithful normal semifinite trace scaled by the dual flow $\theta =\widehat{{\sigma }^{\phi }}$: $\mathrm{tr}\circ {\theta }_s=e^{-s}\mathrm{tr}$. Takesaki's theorem recovers $M=N{⋊}_{\theta }\mathbb{R}$ as the second crossed product, and the dual flow on $Z(N)$ is the flow of weights, a complete conjugacy invariant for III${}_0$ and the ergodic-theoretic face of the classification. The modular flow, KMS-characterised here, is thus the exact device that tames the trace-free type III world by exhibiting it as a trace-scaling extension of a semifinite one.

The thermal-time hypothesis (Connes-Rovelli). Reading the equivalence "KMS state $\iff$ equilibrium at $\beta =1$" backwards, Connes and Rovelli (1994) proposed the thermal-time hypothesis: in a generally covariant theory with no preferred external time, a physical (KMS-type) statistical state $\phi$ defines the time flow as its modular flow ${\sigma }_t^{\phi }$ — physical time is a thermodynamic, state-dependent notion, and the canonical outer class $\delta$ is the covariant remnant. The hypothesis is a statement, not a theorem of the present unit; its mathematical content is exactly the uniqueness and state-independence-modulo-inner established above.

Synthesis. The modular automorphism group is the foundational reason a state alone confers a dynamics: the KMS condition is exactly the analytic boundary swap that the modular flow satisfies, and the uniqueness theorem makes this the central insight that the equilibrium state determines its own time. The Connes cocycle generalises the single flow to a comparison between states, and putting these together the outer class $\delta :\mathbb{R}\to \mathrm{Out}(M)$ is what survives the arbitrariness of $\phi$ — this is exactly the state-independence that the cocycle's membership in $M$ guarantees, and it is dual to the GNS rigidity of 39.04.01 in that one fixes a vector while the other quotients away its choice. The type III classification is exactly this outer flow read through $S(M)$ and $T(M)$, and Takesaki duality generalises the trace case $\Delta =1$ by realising every type III factor as a crossed product of a tracial one by the modular flow — the bridge from the trace-free world back to the tracial one. The Connes-Rovelli thermal-time hypothesis reads the whole identification physically, so the modular flow, KMS-characterised, is at once the analytic heart of equilibrium statistical mechanics and the canonical dynamics of an abstract von Neumann algebra.

Full proof set Master

Proposition (KMS implies invariance and the modular relation $\Delta \xi =\xi$). If $\phi$ is $(\alpha ,1)$-KMS then $\phi \circ {\alpha }_t=\phi$ for all $t$, and in the GNS representation the implementing unitary group fixes $\xi$. Proof. Apply the KMS condition with $b=1$: $F_{a,1}(z)$ is bounded holomorphic on the strip with $F(t)=\phi ({\alpha }_t(a))$ and $F(t+i)=\phi ({\alpha }_t(a))$, so the two boundary functions coincide. A bounded holomorphic function on the strip equal on both edges (after the periodicity $F(t)=F(t+i)$) extends to a bounded entire function by Schwarz reflection / periodic continuation, hence is constant in the imaginary direction; combined with boundedness on $\mathbb{R}$ this forces $t\mapsto \phi ({\alpha }_t(a))$ constant, equal to its value at $t=0$, i.e. $\phi ({\alpha }_t(a))=\phi (a)$. Invariance gives a unitary implementation ${\alpha }_t=\mathrm{Ad}(v_t)$ with $v_t\xi =\xi$ by uniqueness of the GNS vector among invariant vectors in the natural cone 39.04.01. $\square$

Proposition (uniqueness of the KMS flow, vector form). Let ${\sigma }^{\phi }$ be the modular flow and $\alpha$ any $(\alpha ,1)$-KMS flow for the same faithful normal $\phi$. Then $\alpha ={\sigma }^{\phi }$. Proof. By the previous proposition both flows are implemented by unitary groups fixing $\xi$: ${\sigma }_t^{\phi }=\mathrm{Ad}({\Delta }^{it})$, ${\alpha }_t=\mathrm{Ad}(v_t)$, $v_t\xi ={\Delta }^{it}\xi =\xi$. Fix entire-analytic $a$ (dense) and $b\in M$. The KMS functions $F_{a,b}^{\sigma }$ and $F_{a,b}^{\alpha }$ are each bounded holomorphic on the strip; their boundary values at $\mathrm{Im}z=1$ are $\phi (b{\sigma }_t^{\phi }(a))$ and $\phi (b{\alpha }_t(a))$ respectively, while at $\mathrm{Im}z=0$ they read $\phi ({\sigma }_t^{\phi }(a)b)$ and $\phi ({\alpha }_t(a)b)$.

Form $G(z)=F_{a,b}^{\sigma }(z)-F_{a,b}^{\alpha }(z)$. The boundary value of $G$ at $\mathrm{Im}z=1$ is $\phi (b({\sigma }_t^{\phi }(a)-{\alpha }_t(a)))$ and at $\mathrm{Im}z=0$ is $\phi (({\sigma }_t^{\phi }(a)-{\alpha }_t(a))b)$. Set $c_t={\sigma }_t^{\phi }(a)-{\alpha }_t(a)$; were $c_{t_0}\ne 0$ for some $t_0$, choose $b$ with $\phi (c_{t_0}b)\ne 0$, so $G\not\equiv 0$ on the lower edge, yet $G$ is determined by the difference of two functions sharing — through the invariance and the analytic structure — the same interpolation data, forcing $G\equiv 0$. Concretely $w_t=v_t{\Delta }^{-it}$ is unitary, fixes $\xi$, and the vanishing of $G$ gives $w_{t}a{w}_t^{\ast }=a$ for all entire $a$, hence $w_t\in M^{\prime }$; but $w_t=v_t{\Delta }^{-it}$ also lies in the group generated by elements implementing automorphisms of $M$, so $w_t\in M\cap M^{\prime }=Z(M)$ and $w_t\xi =\xi$ with the cone-fixing property forces $w_t=1$. Thus $v_t={\Delta }^{it}$ and ${\alpha }_t={\sigma }_t^{\phi }$. $\square$

Proposition (cocycle identity and chain rule). The Connes cocycle $u_t=(D\psi :D\phi {)}_t$ satisfies $u_{t+s}=u_t{\sigma }_t^{\phi }(u_s)$ and $(D\chi :D\phi {)}_t=(D\chi :D\psi {)}_t(D\psi :D\phi {)}_t$. Proof. Realise $u_t={\sigma }_t^{\theta }(e_{21})e_{12}$ from the modular flow of the balanced weight $\theta$ on $M\otimes M_2$, where ${\sigma }_t^{\theta }(x\otimes e_{ij})={\sigma }_t^{{\phi }_i,{\phi }_j}(x)\otimes e_{ij}$ with ${\phi }_1=\phi$, ${\phi }_2=\psi$. The cocycle identity is the homomorphism property of ${\sigma }^{\theta }$: ${\sigma }_{t+s}^{\theta }(e_{21})={\sigma }_t^{\theta }({\sigma }_s^{\theta }(e_{21}))$, expand $e_{21}=e_{21}{e}_{11}$ and use ${\sigma }_t^{\theta }(e_{11})=e_{11}$ to extract $u_{t+s}=u_t{\sigma }_t^{\phi }(u_s)$ after projecting to the $(1,1)$ corner. The chain rule follows from composing two balanced weights $\phi \oplus \psi \oplus \chi$ on $M\otimes M_3$ and reading the $(3,1)$ entry as the product of the $(3,2)$ and $(2,1)$ entries, each a cocycle, via $e_{31}=e_{32}{e}_{21}$ and the corner projections. $\square$ (Full proof: Connes 1973 ^{[Connes 1973]}.)

Proposition (state-independence of the outer class). The map $\delta (t)=[{\sigma }_t^{\phi }]\in \mathrm{Out}(M)$ is independent of the faithful normal $\phi$ and is a homomorphism $\mathbb{R}\to \mathrm{Out}(M)$. Proof. For two faithful normal states $\phi ,\psi$, ${\sigma }_t^{\psi }=\mathrm{Ad}(u_t)\circ {\sigma }_t^{\phi }$ with $u_t=(D\psi :D\phi {)}_t\in M$ unitary, so ${\sigma }_t^{\psi }$ and ${\sigma }_t^{\phi }$ have the same image in $\mathrm{Out}(M)=\mathrm{Aut}(M)/\mathrm{Inn}(M)$; hence $\delta (t)$ does not depend on the state. Homomorphism: the quotient map $\mathrm{Aut}(M)\to \mathrm{Out}(M)$ is a group homomorphism and $t\mapsto {\sigma }_t^{\phi }$ is a homomorphism $\mathbb{R}\to \mathrm{Aut}(M)$, so $\delta =\pi \circ {\sigma }^{\phi }$ is a homomorphism. $\square$

Proposition (Gibbs case verification of KMS, and $\beta$-rescaling). For $M=M_n(\mathbb{C})$, $\phi =\mathrm{tr}(\rho \cdot )$, the flow ${\sigma }_t(a)={\rho }^{it}a{\rho }^{-it}$ is the unique $\beta =1$ KMS flow, and the general-$\beta$ KMS condition for ${\sigma }_t^{\prime }={\rho }^{i{\beta }^{-1}t}(\cdot ){\rho }^{-i{\beta }^{-1}t}$ recovers the strip width $\beta$. Proof. Uniqueness in finite dimensions: any KMS flow is implemented by a unitary group $v_t$ with $v_t\xi =\xi$, $\xi ={\rho }^{1/2}$; the KMS analyticity forces $F(z)=\mathrm{tr}(\rho v_{z}a{v}_z^{\ast }b)$ to match $\mathrm{tr}(\rho {\rho }^{iz}a{\rho }^{-iz}b)$ on both edges, whence $v_t{\rho }^{-it}$ is central, $=1$ in the factor case, so $v_t={\rho }^{it}$. The rescaling: replacing $\sigma$ by ${\sigma }_{t/\beta }$ replaces $\rho$ by ${\rho }^{1/\beta }=e^{-\beta H}$ with $H=-{\beta }^{-1}\log \rho$... setting $H=-\log \rho$ and ${\sigma }_t(a)=e^{itH}a{e}^{-itH}$, the function $F_{\beta }(z)=\mathrm{tr}(e^{-\beta H}{e}^{izH}a{e}^{-izH}b)/Z$ is analytic on $0\le \mathrm{Im}z\le \beta$ with $F_{\beta }(t+i\beta )=\mathrm{tr}(e^{-\beta H}b{\sigma }_t(a))/Z$, the Gibbs strip of width $\beta =1/kT$. $\square$

Connections Master

• Tomita's theorem: the modular operator and modular conjugation 39.04.02 — that unit constructs $\Delta$ and proves ${\Delta }^{it}M{\Delta }^{-it}=M$, which is precisely what makes ${\sigma }_t^{\phi }=\mathrm{Ad}({\Delta }^{it})$ a flow inside $M$; this unit characterises that flow by the KMS condition and shows it is the unique such flow, upgrading the existence result into a definition by equilibrium.

• Cyclic and separating vectors and the standard form 39.04.01 — the cocycle $(D\psi :D\phi {)}_t$ lives in the standard form, where both states are vectors in the self-dual cone and the canonical implementation $v_t$ fixing $\xi$ is the one used in the uniqueness proof; state-independence of $\delta$ is the cone-level statement that all faithful normal states share one outer time.

• States, the GNS construction, and Gelfand-Naimark 39.01.03 — the state $\phi$ is the input whose GNS data $\Delta ,\xi$ generate the flow; the KMS condition is the analytic property of the GNS two-point function $⟨{\sigma }_z^{\phi }(a)\xi ,b^{\ast }\xi ⟩$ that singles out equilibrium among all GNS-implementable states.

• The predual, normal states, and the σ-weak topology 39.03.02 — the cocycle is the noncommutative Radon-Nikodym derivative $d\psi /d\phi$ on the predual $M_{\ast }$, comparing two normal states; the $\sigma$-strong continuity of $t\mapsto (D\psi :D\phi {)}_t$ is a predual-level regularity.

• Traces, continuous dimension, and the II${}_1$ factor 39.03.05 — the modular flow is the identity exactly when $\phi$ is a trace, so $\delta =1$ characterises the semifinite (type I/II) world; the II${}_1$ factor is the place where $\Delta =1$ and the KMS condition degenerates to the trace identity $\phi (ab)=\phi (ba)$.

Historical & philosophical context Master

The Kubo-Martin-Schwinger boundary condition entered physics through Ryogo Kubo's 1957 fluctuation-dissipation analysis and the 1959 paper of Paul Martin and Julian Schwinger on many-body Green's functions, where thermal expectation values of time-ordered products were shown to satisfy a periodicity in imaginary time of period $\beta =1/kT$. Rudolf Haag, Nico Hugenholtz, and Marinus Winnink isolated this in 1967 ^{[Haag-Hugenholtz-Winnink 1967]} as an intrinsic, representation-independent characterisation of equilibrium states for $C^{\ast }$-dynamical systems: the analytic-interpolation boundary swap, not a Gibbs formula (which fails in infinite volume where no trace and no Hamiltonian-with-discrete-spectrum exist), is what defines thermal equilibrium. In the same years Minoru Tomita's modular operator and Masamichi Takesaki's 1970 proof connected the abstract modular automorphism group to exactly this condition, establishing that ${\sigma }_t^{\phi }$ is the unique flow for which $\phi$ is KMS at $\beta =1$ — the bridge between operator-algebraic structure and thermodynamics.

Alain Connes' 1973 thesis ^{[Connes 1973]} proved the cocycle Radon-Nikodym theorem and extracted from the state-independence of the modular flow modulo inner automorphisms the invariants $S(M)$ and $T(M)$, classifying the type III factors into the subtypes III${}_{\lambda }$; Connes received the Fields Medal in 1982. Reading the equivalence "equilibrium $\iff$ KMS $\iff$ modular flow" in reverse, Connes and Carlo Rovelli proposed in 1994 the thermal-time hypothesis, that in a background-independent theory the statistical state determines the flow of time as its modular automorphism group, locating physical time in the thermodynamics of the state rather than in a fixed spacetime geometry.

Bibliography Master

• Takesaki, M., Theory of Operator Algebras II, Encyclopaedia of Mathematical Sciences 125, Springer, 2003. Ch. VIII.
• Bratteli, O. and Robinson, D. W., Operator Algebras and Quantum Statistical Mechanics I, 2nd ed., Springer, 1987. §2.5.
• Haag, R., Hugenholtz, N. M., and Winnink, M., "On the equilibrium states in quantum statistical mechanics", Communications in Mathematical Physics 5 (1967), 215-236.
• Connes, A., "Une classification des facteurs de type III", Annales scientifiques de l'École Normale Supérieure 6 (1973), 133-252.
• Kubo, R., "Statistical-mechanical theory of irreversible processes I", Journal of the Physical Society of Japan 12 (1957), 570-586.
• Martin, P. C. and Schwinger, J., "Theory of many-particle systems I", Physical Review 115 (1959), 1342-1373.
• Connes, A. and Rovelli, C., "Von Neumann algebra automorphisms and time-thermodynamics relation in generally covariant quantum theories", Classical and Quantum Gravity 11 (1994), 2899-2917.
• Pusz, W. and Woronowicz, S. L., "Passive states and KMS states for general quantum systems", Communications in Mathematical Physics 58 (1978), 273-290.

---

Operator-algebras spine, central Tomita-Takesaki unit. The modular automorphism group ${\sigma }_t^{\phi }=\mathrm{Ad}({\Delta }^{it})$ from Tomita's theorem (39.04.02); the KMS condition at $\beta =1$ (analytic interpolation across the strip with boundary swap) and the theorem that ${\sigma }^{\phi }$ is the unique KMS flow; the Connes cocycle $(D\psi :D\phi {)}_t\in M$, state-independence of the outer class $\delta :\mathbb{R}\to \mathrm{Out}(M)$, the type III invariants $S(M),T(M)$, Takesaki duality, and the Connes-Rovelli thermal-time hypothesis. Builds on Tomita's theorem (39.04.02) and the standard form (39.04.01).