Tomita–Takesaki Modular Theory

Tomita–Takesaki theory assigns an intrinsic one-parameter automorphism group to a von Neumann algebra together with a faithful state in standard form. The construction starts from a cyclic and separating vector, closes an antilinear operator, and takes its polar decomposition. Its output is exact and representation-independent up to unitary equivalence, but it is state–algebra dependent and need not coincide with spacetime evolution or flow along a local vector field.

Required background. Quantum Information supplies faithful density matrices, purification, and finite-dimensional thermal evolution.

Helpful background. Flat-Spacetime Quantum Field Theory provides operator representations; Partial Differential Equations and Green Functions defines operator domains and closures in the needed analytic setting; Causality and Global Structure establishes causal complements; Quantum Fields in Curved Spacetime provides local von Neumann algebras and standard state–region pairs.

In finite dimensions, “faithful” means $\rho>0$: no nonzero positive observable has zero expectation. A purification of such a matrix has full Schmidt rank. The continuum replacement is a vector that is separating as well as cyclic for the algebra; this is a hypothesis, not an automatic property of an arbitrary vector state.

The standard pair and the Tomita operator

Let $\mathcal M\subset\mathcal B(\mathcal H)$ be a von Neumann algebra and let $\Omega\in\mathcal H$ be cyclic and separating for $\mathcal M$. On the dense domain $\mathcal D(S_0)=\mathcal M\Omega$, define the antilinear operator

$$S_0(A\Omega)=A^*\Omega, \qquad A\in\mathcal M.$$

Separatingness makes this definition unambiguous: if $A\Omega=B\Omega$, then $(A-B)\Omega=0$ and hence $A=B$. Cyclicity makes its domain dense. Defining the analogous antilinear operator on $\mathcal M'\Omega$ shows that $S_0$ is closable. Denote its closure by $S$; this closure and its domain are essential, because $S$ is generally unbounded.

The polar decomposition of the closed antilinear operator is

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

Here $\Delta$ is the positive self-adjoint modular operator and $J$ is the antiunitary modular conjugation. For a standard pair, $J^2=\mathbf1$, $J\Omega=\Omega$, and $\Delta\Omega=\Omega$. Takesaki develops the Tomita operator and this polar decomposition with their domain requirements in Takesaki 1970, pp. 31–44.

The modular theorem

The fundamental identities are

$$J\mathcal M J=\mathcal M', \qquad \Delta^{it}\mathcal M\Delta^{-it}=\mathcal M \quad(t\in\mathbb R).$$

The second identity means that

$$\sigma_t^\Omega(A)=\Delta^{it}A\Delta^{-it}$$

is a one-parameter group of $*$-automorphisms of $\mathcal M$. The first exchanges the algebra and its commutant. This is the Tomita–Takesaki theorem in the standard-vector setting; the structural steps and the automorphism result occupy Takesaki 1970, pp. 44–52 and 60–63.

One proof idea is to begin with the algebraic identities relating $S_0$ and the corresponding operator for the commutant, then use polar decomposition and spectral calculus for $\Delta$. Analytic elements—operators for which $t\mapsto\sigma_t(A)$ continues holomorphically—form a strongly dense subalgebra. Relations first established there extend by strong continuity to all of $\mathcal M$. The theorem is consequently much stronger than formal manipulation of $\log\Delta$.

The vector state $\omega_\Omega(A)=\langle\Omega,A\Omega\rangle$ is invariant under the modular group. Separatingness makes it faithful on $\mathcal M$, and cyclicity supplies the standard representation. If either property fails, the construction must be reduced to the state’s support or reformulated in a standard form; the displayed $S_0$ on $\mathcal M\Omega$ is then not the same globally defined standard-pair object.

KMS structure and sign conventions

With the convention $\sigma_t(A)=\Delta^{it}A\Delta^{-it}$, the state is KMS at dimensionless inverse temperature one. For suitable $A,B$ there is a function analytic in $0<\operatorname{Im}z<1$, continuous on the boundary, such that

$$F(t)=\omega_\Omega\!\left(\sigma_t(A)B\right), \qquad F(t+i)=\omega_\Omega\!\left(B\sigma_t(A)\right).$$

This boundary exchange is the thermal-looking content of modular evolution Takesaki 1970, pp. 63–72. It does not assign a physical temperature until a separately identified physical time relates to the modular parameter.

Define the full modular Hamiltonian by $K=-\log\Delta$. Then $\Delta^{it}=e^{-itK}$ and

$$\sigma_t(A)=e^{-itK}Ae^{itK}.$$

Some literature reverses the flow parameter or defines $K=+\log\Delta$; mixing those conventions reverses every modular-flow sign. Additive constants in a one-sided finite-system modular Hamiltonian do not affect its adjoint action, while the full standard-form operator $-\log\Delta$ fixes the algebra–commutant difference.

Faithful matrix state as an exact model

Let $\rho=\sum_i p_i|i\rangle\langle i|$ with every $p_i>0$. Represent $\mathcal M=\mathcal B(\mathbb C^n)$ by left multiplication on the Hilbert space of Hilbert–Schmidt matrices, with inner product $\langle X,Y\rangle=\operatorname{tr}(X^*Y)$ and cyclic vector $\Omega_\rho=\rho^{1/2}$. Right multiplication gives the commutant. Full rank makes $\Omega_\rho$ cyclic and separating.

For a Hilbert–Schmidt matrix $X$,

$$S(X)=\rho^{-1/2}X^*\rho^{1/2}, \qquad J(X)=X^*, \qquad \Delta(X)=\rho X\rho^{-1}.$$

Indeed, $J\Delta^{1/2}(X)=\rho^{-1/2}X^*\rho^{1/2}=S(X)$. On matrix units,

$$\Delta(E_{ij})=\frac{p_i}{p_j}E_{ij}.$$

Conjugating a left multiplication operator gives

$$\sigma_t(A)=\rho^{it}A\rho^{-it}.$$

The KMS boundary condition can now be checked entry by entry or by cyclicity of the trace. This finite model fixes all signs and shows that modular flow is not an additional Hamiltonian chosen from outside: it is determined by the pair $(\mathcal M,\rho)$.

For a Gibbs matrix $\rho=Z^{-1}e^{-\beta H}$, the constants cancel and

$$\sigma_t(A)=e^{-i\beta tH}Ae^{i\beta tH} =\alpha_{-\beta t}(A), \qquad \alpha_s(A)=e^{isH}Ae^{-isH}.$$

Thus increasing the chapter’s modular parameter runs opposite to physical Heisenberg time. This is why the upper-strip KMS relation above places the evolving operator first. The full operator on Hilbert–Schmidt matrices is $K=L_{-\log\rho}-R_{-\log\rho}$; the one-sided matrix $-\log\rho$ alone generates the same action on the left algebra, but it is a different operator. Keeping the two meanings separate avoids a sign or factor-of-two error when passing to a region and its complement.

What failure of faithfulness changes

If one eigenvalue of $\rho$ vanishes, let $P_0$ project onto its kernel. Then left multiplication by $P_0$ is a nonzero algebra element but

$$P_0\rho^{1/2}=0.$$

Thus $\Omega_\rho$ is not separating. The inverse $\rho^{-1}$ in $\Delta=L_\rho R_{\rho^{-1}}$ is also undefined off the support. The remedy is not to pretend that zero eigenvalues are small positive numbers; one restricts to the support algebra or uses the general standard-form theory for nonfaithful functionals.

In local QFT, a Reeh–Schlieder vector for a region with a nonempty causal complement is cyclic and separating, so it supplies a standard pair. Tomita–Takesaki theory then guarantees an abstract modular automorphism group. Whether that flow moves local operators geometrically is a further dynamical and spacetime question. Geometric action means, in addition, that the automorphism transports supports by a spacetime diffeomorphism while preserving the region algebra. That much stronger property requires separate assumptions about the field domain, state, region, and causal action Sorce 2024, §2. Only special symmetric algebra–state–region triples admit a local geometric interpretation; the modular theorem itself makes no such promise.

Solved diagnostic

For a qubit take $\rho=\operatorname{diag}(p,1-p)$ with $0<p<1$. Compute the modular eigenvalue and flow of $E_{01}=|0\rangle\langle1|$. What fails at $p=1$?

Solution

As a Hilbert–Schmidt vector, $E_{01}$ obeys $\Delta(E_{01})=p(1-p)^{-1}E_{01}$. As an element of the matrix algebra, it obeys $\sigma_t(E_{01})=[p/(1-p)]^{it}E_{01}$. These are different uses of the same matrix. More explicitly, conjugation by $e^{-itK}$ on the Hilbert–Schmidt space acts on the left-multiplication operator $L_{E_{01}}$ and produces $L_{\sigma_t(E_{01})}$. At $p=1$, $|1\rangle\langle1|$ annihilates $\rho^{1/2}$, so separatingness and the global inverse of $\rho$ fail; the standard-pair formula must be restricted to the support.

The two nonzero probabilities therefore encode the relative weighting of transitions between the two levels. Their ratio sets the modular phase, while a common normalization cancels. At the endpoint, the state cannot detect an operator acting solely on the unoccupied level. This loss of information is exactly the obstruction to separatingness; restricting the algebra to the occupied support removes that undetectable operator.

References

• Sorce, Jonathan. “Analyticity and the Unruh Effect: A Study of Local Modular Flow.” Journal of High Energy Physics 09 (2024): 040. DOI; Open PDF.
• Takesaki, Masamichi. Tomita’s Theory of Modular Hilbert Algebras and Its Applications. Lecture Notes in Mathematics 128. Berlin: Springer, 1970. DOI.