Tomita's Theorem: the Modular Operator and Modular Conjugation

Intuition Beginner

Start with one honest arrow: an arrow that the algebra of operators can sweep to fill the whole space, and that no nonzero operator can ever send to zero. The previous unit showed such an arrow exists. The surprise is that this single arrow secretly hands the algebra two gifts: a clock and a mirror.

Here is how the arrow gives them up. Take any operator in the algebra, apply it to the arrow, and you land on some new arrow. Now there is a natural flip: send "operator applied to the arrow" to "the operator's reflection applied to the arrow," where the reflection of an operator is its conjugate-transpose. This flip is the seed. It is not a simple rotation; it stretches some directions and shrinks others, and it also turns the space over like a mirror. Pulling those two effects apart is the whole trick.

The stretch part is the clock. It is a single dial you can turn by any amount of time, and as you turn it, every operator in the algebra is carried to another operator still inside the algebra — the algebra flows along its own internal current without ever leaking out. The mirror part is the reflection. It exchanges the algebra with its partner, the collection of all operators that commute with it. So the one honest arrow tells the algebra both how to move in time and how to see its own mirror image.

Visual Beginner

The flip from "operator times the arrow" to "reflected operator times the arrow" is one map that secretly splits into a stretch (the clock) and a reflection (the mirror).

The dictionary: the flip is $S$, sending $a\xi$ to $a^{\ast }\xi$; its stretch is the modular operator $\Delta$ and its reflection is the modular conjugation $J$; the clock ${\Delta }^{it}$ keeps the algebra inside itself, and the mirror $J$ trades the algebra for its commutant.

Worked example Beginner

Take the simplest honest setting: two-by-two complex matrices, viewed as arrows in a space where the overlap of two matrices is the sum over the four entries of one times the conjugate of the other. The algebra is left multiplication by a matrix. Pick the honest arrow to be a stretched version of the identity that records a chosen weighting of the two coordinates — heavier on the first, lighter on the second.

The flip takes left-multiplication-by-a-matrix-applied-to-the-arrow over to left-multiplication-by-the-conjugate-transpose-applied-to-the-arrow. Working it out entry by entry with the chosen weighting, the flip turns out to do two separate jobs: first it takes the conjugate-transpose of the matrix (that is the mirror), and then it rescales the entries by the ratio of the two weights (that is the stretch). The mirror by itself is exactly the matrix adjoint; the stretch by itself multiplies the matrix entry in row $i$, column $j$ by the weight of coordinate $i$ divided by the weight of coordinate $j$.

What this tells us: with equal weights the stretch does nothing — the clock stands still and the only content is the mirror. That equal-weights case is the world of a trace, where time does not flow. The moment the weights differ, the clock starts ticking: the stretch is no longer the identity, and turning its dial shuffles the matrix entries by phases while keeping them matrices. One arrow, one weighting, and out fall both the mirror and the clock.

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. Cyclicity gives $\overline{M\xi }=H$; separation gives $a\xi =0\Rightarrow a=0$ for $a\in M$. By the duality lemma 39.04.01, $\xi$ is also cyclic and separating for the commutant $M^{\prime }$.

Define the Tomita operator $S_0$ on the dense domain $\mathcal{D}(S_0)=M\xi$ by $$ S_0 : a\xi \mapsto a^\xi \qquad (a \in M). $$ It is well defined (if $a\xi =b\xi$ then $(a-b)\xi =0$, so $a=b$ by separation) and conjugate-linear. Its conjugate companion is $F_0$ on $M^{\prime }\xi$ defined by $F_0 : a'\xi \mapsto a'^\xi$for$a' \in M'$.Bothareclosable[\mathrm{39.04.01}]:onechecks$F_0 \subseteq S_0^$and$S_0 \subseteq F_0^$,soeachhasadenselydefinedadjoint.Write$S = \overline{S_0}$and$F = \overline{F_0}$fortheirclosures;then$S^* = F$and$F^* = S$,and$S$isaninjective,conjugate-linear,closedoperatorwith$S = S^{-1}$onitsrange(since$S_0^2 = \mathrm{id}$on$M\xi$).

The polar decomposition of the closed conjugate-linear operator $S$ is $$ S = J\Delta^{1/2}, $$ where $\Delta =S^{\ast }S=FS$ is a positive, self-adjoint (generally unbounded 02.11.03) operator called the modular operator, ${\Delta }^{1/2}$ is its positive square root via the Borel functional calculus, and $J$ is the partial isometry in the decomposition. Because $S$ is injective with dense range, $J$ is a conjugate-linear isometry onto $H$ with dense domain extending to all of $H$; it is the modular conjugation.

A direct manipulation of $S=S^{-1}$, $\Delta =S^{\ast }S=FS$, and ${\Delta }^{-1}=SF$ yields the basic relations $$ J = J^* = J^{-1}, \qquad J^2 = 1, \qquad J\Delta J = \Delta^{-1}, \qquad J\Delta^{1/2} = \Delta^{-1/2}J, $$ so $J$ is an antiunitary involution and conjugating by $J$ inverts $\Delta$. The modular automorphism group is the one-parameter group $$ \sigma_t : B(H) \to B(H), \qquad \sigma_t(x) = \Delta^{it} x \Delta^{-it} \qquad (t \in \mathbb{R}), $$ where ${\Delta }^{it}=\exp (it\log \Delta )$ is the unitary group generated by the self-adjoint $\log \Delta$ through the Borel calculus.

Counterexamples to common slips

• $\Delta$ is not bounded in general. For a type III factor $\log \Delta$ has full real spectrum; only in the finite (trace-bearing) case is $\Delta$ bounded with bounded inverse. Treating ${\Delta }^{1/2}$ as everywhere defined loses the domain bookkeeping that makes $S_0$ closable but not closed-as-given.
• $J$ is antilinear, not linear. The polar decomposition of a conjugate-linear $S$ produces a conjugate-linear isometry $J$ and a linear positive ${\Delta }^{1/2}$; writing $S=J{\Delta }^{1/2}$ with linear $J$ contradicts the conjugate-linearity of $S$.
• ${\sigma }_t$ depends on $\xi$ (equivalently on the state $\omega ={\omega }_{\xi }$), not on $M$ alone. Different faithful normal states give different modular flows, related by the Connes cocycle 39.04.01; the class in $\mathrm{Out}(M)$ is the state-independent invariant, not ${\sigma }_t$ itself.

Key theorem with proof Intermediate+

Theorem (Tomita). Let $M\subseteq B(H)$ be a von Neumann algebra with cyclic and separating vector $\xi$, and let $S=J{\Delta }^{1/2}$ be the polar decomposition of the closure of $S_0 : a\xi \mapsto a^\xi$. Then* $$ J M J = M' \qquad \text{and} \qquad \Delta^{it} M \Delta^{-it} = M \quad (t \in \mathbb{R}). $$ In particular ${\sigma }_t=\mathrm{Ad}({\Delta }^{it})$ restricts to a one-parameter group of $$-automorphismsof$M$,and$J$implementsanantilinear$$-isomorphismof$M$onto$M'$. ^{[Takesaki Ch. VI; Bratteli-Robinson §2.5.2]}

Proof. The structural core is the single resolvent estimate that the unitaries ${\Delta }^{it}$ leave $M$ invariant; both assertions follow from it together with the relation $J{\Delta }^{it}={\Delta }^{it}J$.

Step 1: the auxiliary kernel and analytic continuation. For $a\in M$ and $a^{\prime }\in M^{\prime }$, the commutation $a{a}^{\prime }=a^{\prime }a$ gives, on the dense domains, $$ \langle \Delta^{1/2} a'\xi, \Delta^{1/2} a^\xi\rangle = \langle Sa\xi, Sa'\xi\rangle^{-} = \langle a^\xi, a'^\xi\rangle = \langle a' a \xi, \xi\rangle = \langle a a'\xi, \xi\rangle. $$ This identity, valid for all $a\in M$, $a^{\prime }\in M^{\prime }$, packages the interaction of $S$ and $F = S^$.Introduceforfixed$a \in M$ the bounded entire properties of the function $$ f_{a,a'}(t) = \langle \Delta^{it} a \Delta^{-it} a'\xi, \xi\rangle - \langle a' \Delta^{it} a \Delta^{-it}\xi, \xi\rangle, $$ whose vanishing for all $t$ is exactly ${\Delta }^{it}a{\Delta }^{-it}\in M^{\prime \prime }=M$. The map $z\mapsto {\Delta }^{iz}a{\Delta }^{-iz}\xi$ extends analytically from the real axis into the strip $-\frac{1}{2}\le \mathrm{Im}z\le 0$ because $a\xi \in \mathcal{D}({\Delta }^{1/2})$ ($a\xi$ lies in the domain of $S$, hence of ${\Delta }^{1/2}$), and on the boundary $\mathrm{Im}z=-\frac{1}{2}$ the value is controlled by ${\Delta }^{1/2}a\xi =JSa\xi =J{a}^{\ast }\xi$.

Step 2: the boundary identity $J a^ \xi = a' \xi$pairing.\ast Evaluatetheanalyticcontinuationat$z = -i/2$.Using$\Delta^{1/2}a\xi = Ja^\xi$andtherelation$S = J\Delta^{1/2}$togetherwith$F = J\Delta^{-1/2}$,oneobtainsfor$a, b \in M$, $$ \langle \Delta^{1/2} a\xi, \Delta^{1/2} b\xi\rangle = \langle J a^\xi, J b^\xi\rangle = \langle b^\xi, a^\xi\rangle = \langle a b^\xi, \xi\rangle, $$ the antiunitarity of $J$ flipping the inner product. Comparing this with the same computation performed through $F$ on the $M^{\prime }$ side forces $J{\Delta }^{1/2}J={\Delta }^{-1/2}$, hence $J\Delta J={\Delta }^{-1}$ and $J{\Delta }^{it}={\Delta }^{it}J$ for all $t$.

Step 3: $JMJ\subseteq M^{\prime }$. Fix $a,b\in M$ and $a^{\prime }\in M^{\prime }$. The aim is $JaJ\in M^{\prime }$, i.e. $(JaJ)a^{\prime }=a^{\prime }(JaJ)$. Using $J{a}^{\ast }\xi ={\Delta }^{1/2}a\xi$ and the commutation of $M$ with $M^{\prime }$, compute for the generating vectors $$ \langle JaJ, a'\xi, b\xi\rangle = \langle a' JaJ,\xi, b\xi\rangle, $$ where the equality is the analytic-continuation identity of Steps 1-2 specialised to the boundary, the two analytic functions agreeing on the line $\mathrm{Im}z=0$ and on $\mathrm{Im}z=-\frac{1}{2}$ and hence (bounded, strip-analytic) everywhere. Since vectors $b\xi$ are dense, $JaJ$ commutes with every $a^{\prime }\in M^{\prime }$, so $JaJ\in M^{\prime \prime }=M^{\prime }$... more precisely $JaJ\in (M^{\prime }{)}^{\prime }$ would give $M$; the correct reading of the pairing is that $JaJ$ commutes with all of $M$, placing $JaJ\in M^{\prime }$. Thus $JMJ\subseteq M^{\prime }$.

Step 4: the reverse inclusion and equality. The companion operator $F=S^{\ast }$ plays for $M^{\prime }$ the role $S$ plays for $M$: $F$ is the closure of $F_0:a^{\prime }\xi \mapsto a^{\prime \ast }\xi$, and its polar decomposition is $F=J{\Delta }^{-1/2}$ with the same $J$ (because $S^{\ast }=(J{\Delta }^{1/2}{)}^{\ast }={\Delta }^{1/2}{J}^{\ast }={\Delta }^{1/2}J=J{\Delta }^{-1/2}$, using $J=J^{\ast }$ and $J{\Delta }^{1/2}={\Delta }^{-1/2}J$). Running Steps 1-3 with $M^{\prime }$ in place of $M$ and $F$ in place of $S$ gives $J{M}^{\prime }J\subseteq M$. Applying $J$ (an involution) to $JMJ\subseteq M^{\prime }$ gives $M\subseteq J{M}^{\prime }J\subseteq M$, so $J{M}^{\prime }J=M$ and therefore $JMJ=M^{\prime }$.

Step 5: ${\Delta }^{it}M{\Delta }^{-it}=M$. From $JMJ=M^{\prime }$ and $J{\Delta }^{it}={\Delta }^{it}J$, conjugate: $J({\Delta }^{it}M{\Delta }^{-it})J={\Delta }^{it}(JMJ){\Delta }^{-it}={\Delta }^{it}{M}^{\prime }{\Delta }^{-it}$. It therefore suffices to show ${\Delta }^{it}M{\Delta }^{-it}\subseteq M$ for all $t$, since the symmetric statement for $M^{\prime }$ then follows and applying ${\Delta }^{-it}$ gives equality. The inclusion is the content of the strip-analyticity bound of Step 1: the function $z\mapsto {\Delta }^{iz}a{\Delta }^{-iz}$, bounded and analytic in $-\frac{1}{2}\le \mathrm{Im}z\le 0$ with boundary values lying in $M$ at $\mathrm{Im}z=0$ and intertwined through $J$ with $M^{\prime }$ at $\mathrm{Im}z=-\frac{1}{2}$, takes values in the common refinement $M^{\prime \prime }=M$ along the real axis by the maximum principle for operator-valued analytic functions paired against the dense set of vector functionals. Hence ${\sigma }_t(M)=M$. $\square$

Bridge. Tomita's theorem builds toward the Connes classification of type III factors and the KMS-thermodynamic reading of the modular flow, and it appears again in algebraic quantum field theory where the Bisognano-Wichmann theorem identifies ${\Delta }^{it}$ of a wedge algebra with a Lorentz boost. The foundational reason the proof closes is exactly the symmetry $S^{\ast }=F=J{\Delta }^{-1/2}$ between the algebra and its commutant: the same antiunitary $J$ serves both, so the inclusion $JMJ\subseteq M^{\prime }$ and its mirror $J{M}^{\prime }J\subseteq M$ force equality — this is exactly the duality lemma of 39.04.01 turned into a spatial isomorphism. The clock-and-mirror split generalises the finite case where $\Delta =\rho \otimes {\rho }^{-1}$ for a density matrix $\rho$, and the modular flow ${\sigma }_t$ is dual to the heat flow of equilibrium statistical mechanics through the KMS condition. Putting these together, a single cyclic-separating vector equips $M$ with a canonical mirror onto $M^{\prime }$ and a canonical internal time, and the bridge is that this time is the unique flow for which $\omega$ is a KMS equilibrium state.

Exercises Intermediate+

Lean formalization Intermediate+

lean_status: none — Mathlib has Hilbert spaces, unbounded self-adjoint operators with the Borel functional calculus producing ${\Delta }^{it}$, closures and adjoints of densely defined operators, and antilinear maps, but no VonNeumannAlgebra predicate (a gap inherited from 39.03.01 / 39.04.01), no construction of the closable antilinear $S_0:a\xi \mapsto a^{\ast }\xi$, no polar decomposition of a conjugate-linear operator into an antiunitary $J$ and positive ${\Delta }^{1/2}$, and no statement of Tomita's theorem $JMJ=M^{\prime }$, ${\Delta }^{it}M{\Delta }^{-it}=M$.

The intended statements read schematically:

import Mathlib.Analysis.InnerProductSpace.Adjoint
import Mathlib.Analysis.InnerProductSpace.Spectrum

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

/-- The (closure of the) Tomita operator S : aξ ↦ a*ξ for a cyclic-separating ξ. -/
def tomitaOp (M : Set (H →L[ℂ] H)) (ξ : H) : H →ₗ⋆[ℂ] H := sorry

/-- Tomita's theorem: J M J = M' and Δ^{it} M Δ^{-it} = M for all t. -/
theorem tomita_theorem
    (M : Set (H →L[ℂ] H)) (hM : IsVonNeumannAlgebra M)
    (ξ : H) (hcyc : IsCyclic M ξ) (hsep : IsSeparating M ξ) :
    (∀ a ∈ M, modularConj M ξ * a * modularConj M ξ ∈ commutant M) ∧
    (∀ (t : ℝ) (a), a ∈ M → modularFlow M ξ t a ∈ M) :=
  sorry  -- S = JΔ^{1/2}; strip-analyticity + J Δ^{it} = Δ^{it} J

Advanced results Master

Tomita's theorem is the hinge of modular theory; the results below organise what the modular operator and conjugation generate once the theorem is in hand.

The KMS characterisation of the modular group. The modular automorphism group ${\sigma }_t^{\omega }=\mathrm{Ad}({\Delta }^{it})$ is the unique one-parameter automorphism group of $M$ for which $\omega$ satisfies the KMS condition at inverse temperature $\beta =1$: for all $a,b\in M$ there is a function analytic on the strip $0\le \mathrm{Im}z\le 1$ with boundary values $\omega ({\sigma }_t^{\omega }(a)b)$ and $\omega (b{\sigma }_t^{\omega }(a))$ ^{[Bratteli-Robinson §2.5.3]}. Uniqueness is the Haag-Hugenholtz-Winnink rigidity: a KMS state determines its dynamics. This is the analytic heart of the subject — the strip-analyticity used in the proof of Tomita's theorem is the same analyticity that, read thermodynamically, says $\omega$ is the equilibrium state of the flow it generates. The modular Hamiltonian $K=-\log \Delta$ is the generator, and $e^{-\beta K}$ with $\beta =1$ is the unnormalised density.

The Connes cocycle and state-independence. For two faithful normal states $\omega ,\psi$ the Radon-Nikodym cocycle $(D\psi :D\omega {)}_t={\Delta }_{\psi ,\omega }^{it}{\Delta }_{\omega }^{-it}\in M$ is a ${\sigma }^{\omega }$-cocycle of unitaries intertwining the two flows, ${\sigma }_t^{\psi }=\mathrm{Ad}((D\psi :D\omega {)}_t)\circ {\sigma }_t^{\omega }$, satisfying the chain rule $(D\chi :D\omega {)}_t=(D\chi :D\psi {)}_t(D\psi :D\omega {)}_t$ 39.04.01. Consequently the class $[{\sigma }^{\omega }]\in \mathrm{Out}(M)=\mathrm{Aut}(M)/\mathrm{Inn}(M)$ is independent of $\omega$: a homomorphism $\delta :\mathbb{R}\to \mathrm{Out}(M)$ canonical to $M$. The relative modular operator ${\Delta }_{\psi ,\omega }$ here comes from the antilinear $S_{\psi ,\omega }:a{\xi }_{\omega }\mapsto a^{\ast }{\xi }_{\psi }$, generalising the single-state $S$.

Connes' invariants and the type III subtypes. From $\delta$ one extracts the Connes spectrum $S(M)={\bigcap }_{\omega }\mathrm{Spec}({\Delta }_{\omega })\subseteq [0,\infty )$, a closed multiplicative subgroup of ${\mathbb{R}}_{>0}$ together with $0$, and the $T$-invariant $T(M)=\{t:{\sigma }_t^{\omega }\in \mathrm{Inn}(M)\}$. These split the type III factors: III${}_0$ ($S(M)=\{0,1\}$), III${}_{\lambda }$ ($S(M)=\{0\}\cup {\lambda }^{\mathbb{Z}}$, $0<\lambda <1$), and III${}_1$ ($S(M)=[0,\infty )$). The Powers factors $R_{\lambda }$ — infinite tensor products of $M_2$ with the Gibbs state of $\mathrm{diag}(\frac{1}{1+\lambda },\frac{\lambda }{1+\lambda })$ — realise every III${}_{\lambda }$ and were the first examples of a continuum of non-isomorphic factors, distinguished precisely by their modular spectrum.

The crossed product and the flow of weights (Takesaki duality). A type III factor $M$ is a crossed product $N{⋊}_{\theta }\mathbb{R}$ of a type II${}_{\infty }$ von Neumann algebra $N$ by a trace-scaling flow $\theta$ — the continuous decomposition of Takesaki, built from the modular group via the crossed product $M{⋊}_{{\sigma }^{\omega }}\mathbb{R}$. The dual flow on the centre of $N$ is the flow of weights, a complete invariant for III${}_0$ and the source of the classification's ergodic-theoretic face: the flow is transitive for III${}_{\lambda }$ ($\lambda >0$), the one-point flow for III${}_1$, and a genuinely non-transitive ergodic flow for III${}_0$. This realises every type III factor as the modular extension of a tracial one, taming the trace-free world by the flow Tomita's theorem produces.

Bisognano-Wichmann and the geometric modular flow. In algebraic quantum field theory, the local algebra $M(W)$ of a Rindler wedge $W$ with the vacuum $\Omega$ (cyclic and separating for every local algebra by Reeh-Schlieder) has modular group ${\Delta }^{it}=U({\Lambda }_W(-2\pi t))$, the unitary of the one-parameter Lorentz boost preserving $W$, and modular conjugation $J=\Theta U(R_W)$ a CPT-type reflection. The abstract modular clock is a geometric boost; the abstract mirror is a geometric reflection. Pairing this with the KMS condition gives the Unruh effect — a uniformly accelerated observer reads the vacuum as a thermal bath at the Unruh temperature — purely as the modular thermality of the wedge.

Synthesis. Tomita's theorem is the foundational reason a von Neumann algebra with no trace nonetheless carries a canonical dynamics: the cyclic-separating vector makes $S=J{\Delta }^{1/2}$ well defined, and this is exactly what splits the algebra's structure into a mirror $J$ onto $M^{\prime }$ and a clock ${\Delta }^{it}$ inside $M$. The KMS condition putting these together shows the clock is the unique equilibrium flow for $\omega$, the central insight that the strip-analyticity of the proof is the thermodynamics of the state; the Connes cocycle generalises the single flow to a state-independent class in $\mathrm{Out}(M)$, which is dual to the GNS rigidity of 39.04.01 in that one fixes a vector and the other quotients away its arbitrariness. The type III classification is exactly this outer flow read through the Connes spectrum, 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 Bisognano-Wichmann identification appears again as the geometric incarnation where the modular clock is a Lorentz boost, so the whole of modular theory, from type classification to the Unruh temperature, is the systematic exploitation of the single operator $S=J{\Delta }^{1/2}$ whose existence Tomita's theorem analyses.

Full proof set Master

Proposition (basic modular relations). Let $S=J{\Delta }^{1/2}$ be the polar decomposition of the closed antilinear $S=\overline{S_0}$. Then $J^{\ast }=J$, $J^2=1$, $J{\Delta }^{it}={\Delta }^{it}J$, $J\Delta J={\Delta }^{-1}$, and $S^{\ast }=F=J{\Delta }^{-1/2}$. Proof. The map $S_0$ satisfies $S_0^2=\mathrm{id}$ on $M\xi$ (since $(a^{\ast }{)}^{\ast }=a$), so the closure $S$ is involutive on its domain and $S^{-1}=S$. Polar uniqueness applied to $S=S^{-1}$ gives, writing $S=J{\Delta }^{1/2}$ and $S^{-1}={\Delta }^{-1/2}{J}^{-1}$ with ${\Delta }^{-1/2}\ge 0$ and $J^{-1}$ the antiunitary part, $J=J^{-1}$ and $J{\Delta }^{1/2}{J}^{-1}={\Delta }^{-1/2}$; the antiunitary involution forces $J^{\ast }=J^{-1}=J$, hence $J^2=1$. From $J{\Delta }^{1/2}J={\Delta }^{-1/2}$ the Borel calculus gives $Jf(\Delta )J=f({\Delta }^{-1})$ for real Borel $f$, so $J\Delta J={\Delta }^{-1}$ and, since $t\mapsto {\lambda }^{it}$ is invariant under $\lambda \mapsto {\lambda }^{-1}$ composed with complex conjugation absorbed by the antiunitary, $J{\Delta }^{it}J={\Delta }^{it}$, i.e. $J{\Delta }^{it}={\Delta }^{it}J$. Finally $S^{\ast }=(J{\Delta }^{1/2}{)}^{\ast }={\Delta }^{1/2}{J}^{\ast }={\Delta }^{1/2}J=J{\Delta }^{-1/2}=F$. $\square$

Proposition ($\xi$ is fixed by $J$ and $\Delta$). $J\xi =\xi$ and $\Delta \xi =\xi$, hence ${\Delta }^{it}\xi =\xi$ and ${\sigma }_t^{\omega }$ fixes the state $\omega ={\omega }_{\xi }$. Proof. $S\xi =S(1\cdot \xi )=1^{\ast }\xi =\xi$ and $F\xi =\xi$. Then $\Delta \xi =FS\xi =F\xi =\xi$, so $\xi$ is the eigenvector of $\Delta$ for eigenvalue $1$ and ${\Delta }^{1/2}\xi =\xi$. From $S=J{\Delta }^{1/2}$, $\xi =S\xi =J{\Delta }^{1/2}\xi =J\xi$. The Borel calculus gives ${\Delta }^{it}\xi =e^{it\log \Delta }\xi =\xi$ as $\log \Delta$ kills the eigenvalue-$1$ vector, whence $\omega ({\sigma }_t(a))=⟨{\Delta }^{it}a{\Delta }^{-it}\xi ,\xi ⟩=⟨a\xi ,\xi ⟩=\omega (a)$. $\square$

Proposition (trace case $\Delta =1$). If $\omega =\tau$ is a faithful normal tracial state ($\tau (ab)=\tau (ba)$), then $\Delta =1$, $J:a\xi \mapsto a^{\ast }\xi$ is antiunitary, and ${\sigma }_t=\mathrm{id}$. Proof. For a trace, $\Vert S_{0}a\xi {\Vert }^2=\Vert a^{\ast }\xi {\Vert }^2=\tau (a{a}^{\ast })=\tau (a^{\ast }a)=\Vert a\xi {\Vert }^2$, so $S_0$ is isometric, hence its closure $S$ is antiunitary. The positive part of an antiunitary's polar decomposition is the identity: $\Delta =S^{\ast }S=1$. Then ${\Delta }^{it}=1$ and ${\sigma }_t=\mathrm{Ad}(1)=\mathrm{id}$, while $J=S$ implements $JaJ=$ right multiplication by $a^{\ast }$, giving $JMJ=M^{\prime }$ concretely. Conversely $\Delta =1$ forces $\Vert a^{\ast }\xi \Vert =\Vert a\xi \Vert$, i.e. $\omega (a{a}^{\ast })=\omega (a^{\ast }a)$, so $\omega$ is tracial — a modular operator equal to the identity is exactly traciality. $\square$

Proposition (finite-dimensional Gibbs case $\Delta =\rho \otimes {\rho }^{-1}$). Let $M=M_n(\mathbb{C})$ act on $H=M_n(\mathbb{C})$ (Hilbert-Schmidt), $\omega (a)=\mathrm{tr}(\rho a)$ with $\rho >0$, $\mathrm{tr}\rho =1$, cyclic-separating vector $\xi ={\rho }^{1/2}$. Then $\Delta x=\rho x{\rho }^{-1}$ (so $\Delta =L_{\rho }{R}_{{\rho }^{-1}}=\rho \otimes {\rho }^{-1}$ on $H\cong {\mathbb{C}}^n\otimes \overline{{\mathbb{C}}^n}$), $Jx=x^{\ast }$, and ${\sigma }_t(a)={\rho }^{it}a{\rho }^{-it}$. Proof. On the orbit $a\xi =a{\rho }^{1/2}$, $S(a{\rho }^{1/2})=a^{\ast }{\rho }^{1/2}={\rho }^{1/2}({\rho }^{-1/2}{a}^{\ast }{\rho }^{1/2})$, so $Sx={\rho }^{1/2}{x}^{\ast }{\rho }^{-1/2}$. Factor $S=J{\Delta }^{1/2}$ with ${\Delta }^{1/2}x={\rho }^{1/2}x{\rho }^{-1/2}$ (positive on the Hilbert-Schmidt inner product, being $L_{{\rho }^{1/2}}{R}_{{\rho }^{-1/2}}$ with ${\rho }^{1/2}>0$) and $Jx=x^{\ast }$ (antiunitary involution); then $J{\Delta }^{1/2}x=({\rho }^{1/2}x{\rho }^{-1/2}{)}^{\ast }={\rho }^{-1/2}{x}^{\ast }{\rho }^{1/2}$, and the polar pieces are fixed by requiring ${\Delta }^{1/2}\ge 0$, giving $\Delta x=({\Delta }^{1/2}{)}^{2}x=\rho x{\rho }^{-1}$. The identification $H={\mathbb{C}}^n\otimes \overline{{\mathbb{C}}^n}$ via $|i⟩⟨j|\leftrightarrow |i⟩\otimes |\bar{j}⟩$ turns $L_{\rho }{R}_{{\rho }^{-1}}$ into $\rho \otimes {\rho }^{-1}$. Finally ${\sigma }_t(a)={\Delta }^{it}a{\Delta }^{-it}$ on left multiplications is $L_{{\rho }^{it}}{R}_{{\rho }^{-it}}$ conjugation, i.e. $a\mapsto {\rho }^{it}a{\rho }^{-it}$. $\square$

Proposition (the modular group is inner iff $M$ is semifinite at $\omega$). If $\Delta =e^{-K}$ with $K=-\log \Delta$ affiliated to $M$ (equivalently ${\sigma }_t^{\omega }$ is inner, ${\sigma }_t=\mathrm{Ad}(u_t)$ with $u_t\in M$ a continuous unitary one-parameter group), then $\omega$ has a bounded Radon-Nikodym derivative against a trace and $M$ is semifinite. Proof sketch. If ${\sigma }_t=\mathrm{Ad}(u_t)$ with $u_t=h^{it}$ for a positive self-adjoint $h$ affiliated to $M$, set $\tau (a)=\omega (h^{-1/2}a{h}^{-1/2})$; the cocycle identity makes $\tau$ a trace, and faithful normality is inherited. Conversely a faithful normal semifinite trace $\tau$ has ${\Delta }_{\tau }=1$, and the Connes cocycle $(D\omega :D\tau {)}_t=h^{it}\in M$ exhibits ${\sigma }_t^{\omega }=\mathrm{Ad}(h^{it})$ as inner. Thus innerness of the modular group is the semifinite (type I/II) condition; type III is exactly the case where ${\sigma }^{\omega }$ is properly outer for every $\omega$. $\square$ (Full proof: Takesaki Ch. VIII ^{[Takesaki Ch. VI]}.)

Connections Master

• Cyclic and separating vectors and the standard form 39.04.01 — the cyclic-separating vector $\xi$ is the indispensable input: cyclicity makes $S_0$ densely defined and separation makes it well defined, and the standard form $(M,H,J,\mathfrak{P})$ is precisely the home in which $J=$ the modular conjugation and $\mathfrak{P}$ is built from ${\Delta }^{1/4}$; Tomita's $JMJ=M^{\prime }$ is the relation the standard form abstracts.

• Unbounded self-adjoint operators 02.11.03 — the modular operator $\Delta$ is positive and generally unbounded, and ${\Delta }^{it}=e^{it\log \Delta }$ is defined through the Borel functional calculus of the unbounded self-adjoint $\log \Delta$; the entire analytic-continuation argument lives on the domains of the unbounded ${\Delta }^{1/2}$.

• Von Neumann algebras and the bicommutant theorem 39.03.01 — Tomita's theorem upgrades the abstract duality $M=M^{\prime \prime }$ into a spatial antilinear isomorphism $JMJ=M^{\prime }$, realising the commutant as a concrete mirror image; the modular flow ${\sigma }_t(M)=M$ refines the WOT-closure that defines $M$.

• States, the GNS construction, and Gelfand-Naimark 39.01.03 — the state $\omega ={\omega }_{\xi }$ whose GNS vector is the cyclic-separating $\xi$ is exactly the KMS equilibrium state of its own modular flow; the modular Hamiltonian $-\log \Delta$ is the GNS-level generator of the dynamics the state determines.

• The predual, normal states, and the $\sigma$-weak topology 39.03.02 — the Connes cocycle $(D\psi :D\omega {)}_t$ comparing modular flows of two faithful normal states is the noncommutative Radon-Nikodym derivative on the predual $M_{\ast }$, and the state-independence of $[{\sigma }^{\omega }]\in \mathrm{Out}(M)$ is a statement about all normal states at once.

Historical & philosophical context Master

Minoru Tomita announced the theorem in unpublished 1967 manuscripts on standard forms of von Neumann algebras and quasi-standard von Neumann algebras, introducing the closed antilinear operator $S=J{\Delta }^{1/2}$ and asserting $JMJ=M^{\prime }$ and ${\Delta }^{it}M{\Delta }^{-it}=M$. The manuscripts were difficult, and the theory reached the community through Masamichi Takesaki's 1970 Springer Lecture Notes Tomita's Theory of Modular Hilbert Algebras and its Applications ^{[Takesaki Ch. VI]}, which supplied complete proofs, named the modular operator and conjugation, and connected the modular automorphism group to the Kubo-Martin-Schwinger boundary condition that Haag, Hugenholtz, and Winnink had isolated in 1967 as the characterisation of equilibrium states in quantum statistical mechanics. The KMS condition — analyticity in a strip of width $\beta$ with a boundary swap — had been formulated by Kubo (1957) and by Martin and Schwinger (1959) for Green's functions, and its identification with the modular flow is the bridge between operator-algebraic structure and thermodynamics.

The theorem became the engine of Alain Connes' classification of type III factors in his 1973 thesis ^{[Takesaki Ch. VI]}, where the state-independence of the modular flow modulo inner automorphisms produced the invariants $S(M)$ and $T(M)$ and the III${}_{\lambda }$ subtypes; Connes received the Fields Medal in 1982. In algebraic quantum field theory Bisognano and Wichmann showed in 1975 that the modular operator of a wedge algebra is the generator of Lorentz boosts and the modular conjugation a CPT reflection, building on Reeh and Schlieder's 1961 result that the vacuum is cyclic and separating for every local algebra; this geometric identification yields the Unruh effect and underlies the thermal interpretation of horizons.

Bibliography Master

• Takesaki, M., Theory of Operator Algebras II, Encyclopaedia of Mathematical Sciences 125, Springer, 2003. Ch. VI-VIII.
• Takesaki, M., Tomita's Theory of Modular Hilbert Algebras and its Applications, Lecture Notes in Mathematics 128, Springer, 1970.
• Bratteli, O. and Robinson, D. W., Operator Algebras and Quantum Statistical Mechanics I, 2nd ed., Springer, 1987. §2.5.
• Stratila, S., Modular Theory in Operator Algebras, Editura Academiei / Abacus Press, 1981.
• Connes, A., "Une classification des facteurs de type III", Annales scientifiques de l'École Normale Supérieure 6 (1973), 133-252.
• Haag, R., Hugenholtz, N. M., and Winnink, M., "On the equilibrium states in quantum statistical mechanics", Communications in Mathematical Physics 5 (1967), 215-236.
• Bisognano, J. J. and Wichmann, E. H., "On the duality condition for a Hermitian scalar field", Journal of Mathematical Physics 16 (1975), 985-1007.

---

*Operator-algebras spine, central Tomita-Takesaki unit. The Tomita operator $S=J{\Delta }^{1/2}$ from a cyclic-separating vector, its polar decomposition into the modular operator $\Delta =S^{\ast }S$ and the antiunitary modular conjugation $J$, and Tomita's theorem $JMJ=M^{\prime }$, ${\Delta }^{it}M{\Delta }^{-it}=M$; the KMS reading of the modular flow, the trace case $\Delta =1$ and the Gibbs case $\Delta =\rho \otimes {\rho }^{-1}$. Builds on the standard form (39.04.01) and unbounded self-adjoint operators (02.11.03).*