Standard von Neumann Algebras and Tomita–Takesaki Theory

For a von Neumann algebra $\mathcal M\subset\mathcal B(\mathcal H)$ and a cyclic separating vector $\Omega$, the adjoint operation on $\mathcal M$ defines a densely defined closable antilinear operator. Its polar factors are the modular conjugation $J$ and modular operator $\Delta$. Tomita–Takesaki theory then proves $J\mathcal M J=\mathcal M'$ and $\Delta^{it}\mathcal M\Delta^{-it}=\mathcal M$. These are abstract algebraic statements; identifying them with spacetime symmetries requires another theorem. The wedge-vacuum application is developed on Tomita–Takesaki modular operators and flow.

Required background. States, GNS representations, and folia supplies the represented algebra and vector state; representation types, factors, and local-algebra structure supplies commutants; and unbounded operators, domains, closure, and adjoints supplies the operator-theoretic language.

Helpful background. Tomita–Takesaki modular operators and flow gives the physical interpretation; standard form, cyclic and separating vectors gives finite and field-theoretic comparisons; and modular Hamiltonians: definitions and domains treats the logarithm of $\Delta$.

The Tomita operator and its domain

The pair $(\mathcal M,\Omega)$ is standard when $\Omega$ is both cyclic, $\overline{\mathcal M\Omega}=\mathcal H$, and separating, $A\Omega=0\Rightarrow A=0$. On the dense domain $\mathcal D_0=\mathcal M\Omega$, define

$$S_0(A\Omega)=A^*\Omega.$$

Separation makes this definition unambiguous. Cyclicity makes its domain dense. The same two conditions imply that $\Omega$ is cyclic and separating for the commutant with the roles exchanged, so

$$F_0(B'\Omega)=B'^*\Omega, \qquad B'\in\mathcal M',$$

is also densely defined. The commutation of $A$ and $B'$ gives the adjoint relation $F_0\subset S_0^*$ and $S_0\subset F_0^*$, with the antilinear-adjoint convention fixed consistently. Hence $S_0$ and $F_0$ are closable. Write

$$S=\overline{S_0}, \qquad \mathcal D(S)=\mathcal D(\Delta^{1/2}).$$

The polar decomposition of the closed antilinear operator is

$$S=J\Delta^{1/2}, \qquad \Delta=S^*S.$$

Here $\Delta$ is positive, self-adjoint, and injective; $J$ is antiunitary. In general $\Delta^{1/2}$ and $\log\Delta$ are unbounded, so their domains cannot be suppressed. By contrast, $\Delta^{it}$ is a bounded unitary for every real $t$. The foundational construction and commutation theorem are developed in Takesaki 1970, pp. 1–120.

Tomita–Takesaki conclusions

The theorem gives

$$\begin{gathered} J^2=1,\qquad J\Omega=\Omega,\qquad J\Delta J=\Delta^{-1},\\ J\mathcal M J=\mathcal M',\qquad \sigma_t^\Omega(A) =\Delta^{it}A\Delta^{-it}\in\mathcal M. \end{gathered}$$

Thus $t\mapsto\sigma_t^\Omega$ is a strongly continuous one-parameter automorphism group of $\mathcal M$. The group depends on the faithful normal state represented by $\Omega$, although its outer-automorphism class has additional invariance properties. The conjugation is not ordinary complex conjugation and not, without geometric input, charge conjugation.

For elements analytic under $\sigma^\Omega$, the vector state $\omega_\Omega(A)=\langle\Omega,A\Omega\rangle$ obeys the unit-strip KMS boundary relation. With our convention $\sigma_t^\Omega=\operatorname{Ad}\Delta^{it}$, take

$$F_{A,B}(t) =\omega_\Omega\!\left(\sigma_t^\Omega(B)A\right),$$

there is a bounded continuous function on the closed strip $0\leq\operatorname{Im}z\leq1$, analytic inside, such that

$$F_{A,B}(t+i) =\omega_\Omega\!\left(A\sigma_t^\Omega(B)\right).$$

The number $1$ here belongs to the modular parameter. Converting it into a physical inverse temperature requires a relation between modular flow and a physical time evolution. Borchers explains this distinction and the QFT applications in Borchers 2000, pp. 3604–3673.

The proof mechanism is operator-theoretic rather than a formal manipulation of $S=J\Delta^{1/2}$. One first proves the Tomita relations on the left Hilbert algebra $\mathcal M\Omega$, where products and adjoints are controlled. Entire analytic elements for the modular group provide a core on which powers of $\Delta$ can be moved through algebra elements. Closure then upgrades the core identities to $\Delta^{it}\mathcal M\Delta^{-it}=\mathcal M$, and the adjoint relations between $S_0$ and $F_0$ identify $J\mathcal M J$ with the commutant. This order matters: the formula $SAS$ is generally meaningless on arbitrary vectors because both occurrences of $S$ are unbounded antilinear operators.

There is also a useful domain check. Since $S=J\Delta^{1/2}$ and $J$ is bounded,

$$\|S\xi\|=\|\Delta^{1/2}\xi\|, \qquad \xi\in\mathcal D(S).$$

Thus the graph norms of $S$ and $\Delta^{1/2}$ agree. A proposed core for one is a core for the other, whereas no such statement follows for $\log\Delta$ without additional spectral control near zero and infinity.

The relation $J\Delta J=\Delta^{-1}$ supplies an independent spectral check. If $E_\Delta$ is the spectral measure, then $JE_\Delta(B)J=E_\Delta(B^{-1})$ for Borel sets away from zero. Modular spectral weight at $\lambda$ is therefore paired with weight at $\lambda^{-1}$; an unpaired proposed finite-dimensional spectrum cannot be the modular spectrum of the stated standard pair.

Wedge algebra and vacuum

Let $\mathcal M=\mathcal A(W_R)$ be a right-wedge algebra in a vacuum net. Reeh–Schlieder cyclicity for $W_R$, together with locality and cyclicity for the opposite wedge, makes $\Omega$ separating for $\mathcal M$. Therefore

$$S_W(A\Omega)=A^*\Omega, \qquad A\in\mathcal A(W_R),$$

closes and has polar factors $(J_W,\Delta_W)$. Without using any boost formula, Tomita–Takesaki already verifies

$$J_W\mathcal A(W_R)J_W=\mathcal A(W_R)', \qquad \Delta_W^{it}\mathcal A(W_R)\Delta_W^{-it} =\mathcal A(W_R).$$

The later Bisognano–Wichmann theorem identifies these operators geometrically under Wightman or suitable net hypotheses. It is not part of the abstract polar-decomposition theorem.

An independent check follows from $S(A\Omega)=A^*\Omega$. On the core, $S^2(A\Omega)=A\Omega$; polar decomposition then forces $J\Delta^{1/2}J\Delta^{1/2}=1$, consistent with $J\Delta J=\Delta^{-1}$. This core identity does not say that $S$ is bounded.

Adversarial test: cyclic but not separating

Let $\mathcal M=\mathcal B(\mathbb C^2)$ act on $\mathbb C^2$ and take $\Omega=e_1$. The vector is cyclic because the full matrix algebra sends it to every vector, but it is not separating. For $A=|e_1\rangle\langle e_2|$,

$$A\Omega=0, \qquad A^*\Omega=e_2.$$

The vector $0$ is represented both as $A\Omega$ and as $0\Omega$, while the proposed rule sends these representatives to $e_2$ and $0$. Thus $S_0$ is not even well-defined. Cyclicity alone licenses a dense domain but not the Tomita map.

Exercises

Show that $\Omega$ is separating for $\mathcal M$ if and only if it is cyclic for $\mathcal M'$.

Solution

If $\Omega$ is cyclic for $\mathcal M'$ and $A\Omega=0$, then $AB'\Omega=B'A\Omega=0$ for every $B'\in\mathcal M'$. Density of $\mathcal M'\Omega$ gives $A=0$, so $\Omega$ is separating for $\mathcal M$. Conversely, if the closure of $\mathcal M'\Omega$ were a proper subspace, its projection would be a nontrivial element of $(\mathcal M')'=\mathcal M$ annihilating $\Omega$, contradicting separation.

References

• Borchers, Hans-Jürgen. 2000. “On Revolutionizing Quantum Field Theory with Tomita’s Modular Theory.” Journal of Mathematical Physics 41: 3604–3673. DOI.
• Takesaki, Masamichi. 1970. Tomita’s Theory of Modular Hilbert Algebras and Its Applications. Lecture Notes in Mathematics 128. Springer. DOI.