Week 4 — KMS States, the Modular Interpretation, Type III Preview

Last week we classified factors and saw that type III admits no trace. The standard quantum-mechanical formula ${\rho }_{\beta }=e^{-\beta H}/Z$ therefore has no intrinsic local analogue there. This week we develop the algebraic substitute: the KMS condition. It is a condition on a state relative to a specified one-parameter automorphism group; it needs neither a Hamiltonian operator inside the algebra nor a density matrix. Tomita–Takesaki theory will then say that every faithful normal state supplies a canonical group of this kind—the modular flow—for which it is KMS at $\beta =1$ in our convention.

How to use this chapter

• In class: derive the finite-dimensional KMS boundary identity first, extract detailed balance, and only then state the abstract strip condition. End by using the Powers product state as a controlled preview of type III${}_{\lambda }$.
• For self-study: reproduce the two-level-system calculation and annotate every sign in the Fourier and modular conventions. Treat the thermal-time discussion as an interpretation after the KMS theorem, not as part of its proof.
• Instructor checkpoint: verify that students can explain why the exact $\lambda =0$ product state is nonfaithful and type I${}_{\infty }$, why III${}_0$ needs a different limiting construction, and why the course uses ${\sigma }_t=\mathrm{Ad}{\Delta }^{-it}$ at inverse modular temperature $+1$.

0. Reading

Primary:

• Bratteli & Robinson Vol. II, §5.3 (KMS condition and equilibrium states).
• Witten, “Notes on some entanglement properties of QFT,” arXiv:1803.04993, §3.

Secondary:

• Haag, Local Quantum Physics, ch. V.
• Powers, “Representations of uniformly hyperfinite algebras…,” Annals 86 (1967) — original construction of type III${}_{\lambda }$.
• Connes & Rovelli, “Von Neumann algebra automorphisms and time-thermodynamics relation in generally covariant quantum theories,” CQG 11 (1994) — for the “thermal time” interpretation we sketch in §6.

1. From Gibbs to KMS: a finite-dimensional warmup

In ordinary quantum mechanics on a finite-dimensional Hilbert space $\mathcal{H}$ with Hamiltonian $H=H^{\ast }$, the Gibbs state at inverse temperature $\beta$ is

$${\rho }_{\beta }=\frac{e^{-\beta H}}{Z(\beta )},Z(\beta )=\mathrm{Tr}(e^{-\beta H}).$$

The expectation value of any observable $A\in \mathcal{B}(\mathcal{H})$ is ${\omega }_{\beta }(A)=\mathrm{Tr}({\rho }_{\beta }A)$. Heisenberg evolution: $A_t:=e^{itH}A{e}^{-itH}$.

1.1 The KMS identity

The key calculation. Fix $\beta >0$. For any $a,b\in \mathcal{B}(\mathcal{H})$ define the function

$$F(z):={\omega }_{\beta }(a{\sigma }_z(b))=\frac{1}{Z}\mathrm{Tr}\left(e^{-\beta H}a{e}^{izH}b{e}^{-izH}\right),z\in \mathbb{C},$$

where ${\sigma }_z(b):=e^{izH}b{e}^{-izH}$ is the analytic continuation of the Heisenberg evolution. In finite dimensions $F$ is entire: the matrix exponentials are entire functions of $z$, and the trace is a continuous linear functional. On the real axis $F(t)={\omega }_{\beta }(a{\sigma }_t(b))$, recovering the usual expectation value.

The KMS identity. The boundary value at $z=t+i\beta$ is

$$F(t+i\beta )={\omega }_{\beta }({\sigma }_t(b)a).\text{\hspace{1em}(∗)}$$

Proof of ($\ast$). Substitute $z=t+i\beta$ and split the exponentials:

$$F(t+i\beta )=\frac{1}{Z}\mathrm{Tr}\left(e^{-\beta H}a{e}^{itH}{e}^{-\beta H}b{e}^{-itH}{e}^{\beta H}\right).$$

Move the rightmost $e^{\beta H}$ around the trace by cyclicity, where it cancels the leftmost $e^{-\beta H}$:

$$=\frac{1}{Z}\mathrm{Tr}\left(a{e}^{itH}{e}^{-\beta H}b{e}^{-itH}\right).$$

Now move $e^{-itH}$ around the trace and combine with $e^{itH}$ to form ${\sigma }_{-t}(a)=e^{-itH}a{e}^{itH}$:

$$=\frac{1}{Z}\mathrm{Tr}({\sigma }_{-t}(a)e^{-\beta H}b)=\frac{1}{Z}\mathrm{Tr}(e^{-\beta H}b{\sigma }_{-t}(a))={\omega }_{\beta }(b{\sigma }_{-t}(a)).$$

Finally, ${\omega }_{\beta }$ is ${\sigma }_t$-invariant (${\rho }_{\beta }$ commutes with $H$, so ${\omega }_{\beta }({\sigma }_t(\cdot ))={\omega }_{\beta }(\cdot )$), and ${\sigma }_t(b{\sigma }_{-t}(a))={\sigma }_t(b)a$, so

$${\omega }_{\beta }(b{\sigma }_{-t}(a))={\omega }_{\beta }({\sigma }_t(b)a).\square$$

1.2 The structure of the KMS identity

Equation ($\ast$) — together with the holomorphic interpolation $F$ — has three crucial features:

1. No reference to ${\rho }_{\beta }$ explicitly. Once the automorphism group $\sigma$ has been specified, the boundary values $F(t)={\omega }_{\beta }(a{\sigma }_t(b))$ and $F(t+i\beta )={\omega }_{\beta }({\sigma }_t(b)a)$ are expressed entirely in terms of the state and that dynamics — the density matrix has dropped out.

2. Holomorphic structure on the upper strip. $F(z)$ extends to a holomorphic function on $\{0\le \mathrm{Im}z\le \beta \}$ — entire in finite dimensions, holomorphic-on-the-strip in the general C*-algebraic setting (Definition 2.1).

3. Trace cyclicity is the engine. The proof uses only cyclicity of the trace and the fact that ${\rho }_{\beta }$ is a function of $H$. A type-III factor has no nonzero normal semifinite trace and hence no intrinsic Gibbs density matrix relative to such a trace, but the abstract KMS condition — the existence of a holomorphic $F$ with these boundary values — still makes sense. That is the abstraction we now make.

2. The KMS condition

Definition 2.1. Let $\mathcal{A}$ be a C*-algebra with a one-parameter automorphism group $\sigma :\mathbb{R}\to \mathrm{Aut}(\mathcal{A})$, $t\mapsto {\sigma }_t$, such that $t\mapsto {\sigma }_t(a)$ is norm-continuous for every $a\in \mathcal{A}$. A state $\omega$ on $\mathcal{A}$ is KMS at inverse temperature $\beta$ with respect to $\sigma$ if for every $a,b\in \mathcal{A}$, there is a function $F_{a,b}(z)$:

• holomorphic on the open strip $\{z\in \mathbb{C}:0<\mathrm{Im}z<\beta \}$,
• continuous on the closed strip $\{0\le \mathrm{Im}z\le \beta \}$,
• bounded on the strip,
• with boundary values $F_{a,b}(t)=\omega (a{\sigma }_t(b))$ and $F_{a,b}(t+i\beta )=\omega ({\sigma }_t(b)a)$ for $t\in \mathbb{R}$.

The condition encodes Eq. ($\ast$) abstractly: the function $t\mapsto \omega (a{\sigma }_t(b))$ on the real axis coincides at $t+i\beta$ with $\omega ({\sigma }_t(b)a)$, with a holomorphic interpolation in between.

Physical picture. The analytic strip is imaginary time. A physicist meets the KMS condition as the statement that thermal correlation functions are periodic in imaginary time with period $\beta$ — the Matsubara/Euclidean-path-integral structure: $⟨a(0)b(i\beta ){⟩}_{\beta }=⟨b(0)a(0){⟩}_{\beta }$ up to ordering, i.e. translating one operator all the way across the thermal circle swaps the operator order. Definition 2.1 is exactly this, stated so that no path integral, trace formula, or intrinsic density matrix is required: what remains is the analyticity of the $\sigma$-evolved correlators in a strip of width $\beta$ plus the boundary identification. The state supplies the expectation functional, while the chosen automorphism group supplies the notion of time translation. The strip width being $\beta$ also explains physically why higher temperature is “less analytic”: at $T\to \infty$ the strip collapses and the condition degenerates to traciality (§5), while at $T\to 0$ the strip becomes a half-plane — the boundary of the KMS family is the spectrum condition of the vacuum.

Detailed balance, derived

The slogan “KMS = detailed balance” can be made exact in two lines of Fourier analysis. Define the two correlation functions

$$G_{+}(t):=\omega (a{\sigma }_t(b)),G_{-}(t):=\omega ({\sigma }_t(b)a),$$

physically: the amplitude for the system to absorb the perturbation $b$ and later release it into $a$, in the two possible operator orders. KMS says $G_{-}(t)=G_{+}(t+i\beta )$ via the holomorphic interpolant. Writing $G_{\pm }(t)=\frac{1}{2\pi }\int d\nu e^{+i\nu t}{\hat{G}}_{\pm }(\nu )$ and shifting through the strip (first for analytic observables, and then in the distributional sense),

$$\boxed{{\hat{G}}_{-}(\nu )=e^{-\beta \nu }{\hat{G}}_{+}(\nu ).}$$

The Fourier components at frequency $\nu$ — the rates for processes in which the system gives up energy $\nu$ versus absorbs it — are related by the Boltzmann factor $e^{-\beta \nu }$. This is detailed balance, the Einstein relation between emission and absorption, derived from the strip analyticity alone. In the finite-dimensional model it reduces to $\omega (E_{jj})/\omega (E_{ii})=e^{-\beta (E_j-E_i)}$, the Step-1 computation of Theorem 3.1 below. In QFT, this identity applied to the Rindler modular flow is the statement that an accelerated detector’s excitation and de-excitation rates have thermal ratios — the operational form of the Unruh effect.

2.1 Remarks and conventions

• Sign convention. Many references (including Bratteli–Robinson) use $\beta >0$ and an “upper strip” (as written above). Some references invert. Be careful.
• Strip width = inverse temperature. $\beta =1/k_{B}T$. Higher temperature → narrower strip.
• Locality in $a,b$. Just one pair of analytic functions per pair of observables. Hence verifying KMS is a “local” condition on observable pairs.
• Continuity depends on the setting. Definition 2.1 is the C*-dynamical formulation and assumes point-norm continuity. A modular automorphism group on a von Neumann algebra is, in general, only $\sigma$-weakly continuous on the whole algebra; its analytic elements form a $\sigma$-weakly dense subalgebra. The corresponding W*-KMS formulation uses this weaker continuity. One should not claim norm continuity for every modularly evolved observable.

2.2 What KMS captures

The KMS condition is the algebraic content of thermal equilibrium. To see this, recall that in a Gibbs state, the response to a small perturbation $V$ at time $0$ is given by linear response theory:

$$\delta \omega (A;t)\propto {\int }_0^t\omega ([{\sigma }_s(V),A_t])ds+\text{higher order}.$$

The KMS condition ($\ast$) makes this response time-translation invariant and passively determined: the state cannot be exploited as an energy source. (This is the passivity of thermal states; see Pusz–Woronowicz.)

The KMS condition is also the algebraic content of detailed balance: the relative populations of different states are determined by their energy difference and the temperature, regardless of the path between them.

3. KMS in finite dimensions = Gibbs

Theorem 3.1 (KMS uniqueness in finite dimensions; proved below). Let $\mathcal{A}=\mathcal{B}(\mathcal{H})$ with $\dim \mathcal{H}<\infty$, $H=H^{\ast }\in \mathcal{A}$, and ${\sigma }_t=\mathrm{Ad}{e}^{itH}$. Then a state $\omega$ on $\mathcal{A}$ is KMS at $\beta >0$ for $\sigma$ iff $\omega ={\omega }_{{\rho }_{\beta }}$ where ${\rho }_{\beta }=e^{-\beta H}/\mathrm{Tr}(e^{-\beta H})$.

Proof. Direction (⇐). Suppose $\omega ={\omega }_{{\rho }_{\beta }}$. The function $F_{a,b}(z):=\frac{1}{Z}\mathrm{Tr}(e^{-\beta H}a{e}^{izH}b{e}^{-izH})$ is the entire holomorphic extension constructed in §1.1, with boundary values $F_{a,b}(t)={\omega }_{{\rho }_{\beta }}(a{\sigma }_t(b))$ and $F_{a,b}(t+i\beta )={\omega }_{{\rho }_{\beta }}({\sigma }_t(b)a)$ established there. Boundedness on the strip $0\le \mathrm{Im}z\le \beta$ holds because ${\rho }_{\beta }$, $e^{-\beta H}$, $a$, $b$, and the Heisenberg evolution are all bounded operators on the finite-dimensional $\mathcal{H}$.

Direction (⇒). Suppose $\omega$ is KMS at $\beta$ for ${\sigma }_t=\mathrm{Ad}{e}^{itH}$. Diagonalize $H$ in an orthonormal basis $\{|i⟩{\}}_{i=1}^n$ with $H|i⟩=E_i|i⟩$, and let $E_{ij}:=|i⟩⟨j|$ be the matrix units. Then ${\sigma }_t(E_{ij})=e^{it(E_i-E_j)}{E}_{ij}$.

Step 1: diagonal weighting. Fix $i\ne j$ and take $a=E_{ij}$, $b=E_{ji}$. Then $ab=E_{ii}$ and $ba=E_{jj}$, and

$$F_{a,b}(t)=\omega (a{\sigma }_t(b))=e^{it(E_j-E_i)}\omega (E_{ii}).$$

The unique entire function extending this is $F_{a,b}(z)=e^{iz(E_j-E_i)}\omega (E_{ii})$. The KMS boundary condition at $z=t+i\beta$ demands

$$e^{(it-\beta )(E_j-E_i)}\omega (E_{ii})=\omega ({\sigma }_t(b)a)=e^{it(E_j-E_i)}\omega (E_{jj}),$$

which after dividing through by $e^{it(E_j-E_i)}$ gives

$$e^{-\beta (E_j-E_i)}\omega (E_{ii})=\omega (E_{jj}).$$

Equivalently $\omega (E_{ii})\propto e^{-\beta E_i}$ — the thermal weighting.

Step 2: off-diagonal vanishing. Fix $i\ne l$ and take $a=E_{il}$, $b=E_{ll}$. Since $b$ is diagonal in the energy basis, ${\sigma }_t(b)=b$, and

$$F_{a,b}(t)=\omega (ab)=\omega (E_{il}{E}_{ll})=\omega (E_{il})$$

is constant in $t$. The unique entire extension is therefore the constant $\omega (E_{il})$, so the KMS boundary condition gives

$$\omega (E_{il})=\omega ({\sigma }_t(b)a)=\omega (E_{ll}{E}_{il})=0.$$

Hence $\omega (E_{il})=0$ for all $i\ne l$.

Step 3: assembly. Combining Steps 1 and 2, $\omega (E_{ij})={\delta }_{ij}{e}^{-\beta E_i}/Z$ with $Z={\sum }_k{e}^{-\beta E_k}$ fixed by normalization $\omega (I)=1$. By linearity, $\omega (\cdot )=\mathrm{Tr}({\rho }_{\beta }\cdot )={\omega }_{{\rho }_{\beta }}(\cdot )$. $\square$

Remark. Step 2 (off-diagonal vanishing) is the part most often glossed over. Without it, the diagonal weighting alone does not pin down $\omega$: a state can have the right diagonal entries but non-zero coherences. The KMS condition, applied to a non-diagonal $a$ paired with a diagonal $b$, forces those coherences to vanish.

Remark on energy degeneracies. If $E_i=E_l$ with $i\ne l$, Step 2 still works: $b=E_{ll}$ is diagonal, ${\sigma }_t(b)=b$, and the constant-$F$ argument gives $\omega (E_{il})=0$. Step 1’s exponential argument is unchanged whether or not $E_i\ne E_j$.

This theorem says: in finite dimensions, once the dynamical system $(\mathcal{A},\sigma )$ has been specified, the KMS condition picks out exactly the Gibbs state. Although the proof used a convenient matrix realization and its implementing Hamiltonian, the boundary condition itself is formulated in terms of $(\mathcal{A},\sigma ,\omega )$.

4. KMS in QFT: the example we need

In QFT — and especially for type III algebras — the KMS condition is far more general than Gibbs. The state $\omega$ in question is typically the vacuum, restricted to a wedge or other unbounded region; the flow ${\sigma }_t$ is typically the modular flow generated by Tomita-Takesaki.

The leading example, which we will work out fully in Week 10:

Example 4.1 (Bisognano–Wichmann, preview). Let $\mathcal{A}(W_R)$ be the local algebra of the right Rindler wedge in a Wightman QFT satisfying the hypotheses of the Bisognano–Wichmann theorem. In our convention ${\sigma }_t^{\Omega }=\mathrm{Ad}({\Delta }_{\Omega }^{-it})$, the modular automorphism is the boost automorphism with physical rapidity $+2\pi t$. The vacuum state ${\omega }_0$ restricted to $\mathcal{A}(W_R)$ is KMS at $\beta =1$ with respect to this flow. (With the more common convention $\mathrm{Ad}({\Delta }^{it})$, the rapidity is $-2\pi t$.)

Equivalently, in physical units: the vacuum looks thermal at temperature $T_U=a/2\pi$ to a uniformly accelerated observer with proper acceleration $a$. This is the Unruh effect, here re-expressed as a KMS condition. Bisognano–Wichmann by itself identifies the modular dynamics; it does not classify the factor. Standard wedge algebras are type III${}_1$ only after the separate type theorem and its hypotheses are supplied (Weeks 10 and 12). In that case there is no intrinsic Gibbs density matrix—only the algebraic KMS condition.

This example is the prototype for everything in Block C and beyond. The KMS condition is the standard algebraic characterization of thermal equilibrium when a Gibbs density matrix is unavailable. Tomita-Takesaki and Bisognano-Wichmann tell us where the relevant dynamics comes from: the boost is the canonical modular flow of the vacuum wedge algebra.

Example 4.2 (Two-sided Rindler — TFD as a regulated analogy). The right- and left-wedge algebras commute, and under wedge duality they are commutants. Their common-boundary situation does not furnish a literal type-I tensor factorization $\mathcal{H}={\mathcal{H}}_R\otimes {\mathcal{H}}_L$, nor a normal thermofield-double vector in such a factorization. A UV regulator, or a split inclusion with a collar between regions, produces a type-I approximation in which the familiar TFD formula is useful. In the continuum, the exact statement is instead the KMS/modular correlation structure of the vacuum. This distinction will matter in Semester II Block 2.

Example 4.3 (Tolman–Ehrenfest, scoped). In a stationary spacetime, a global KMS state for the Killing-time evolution is read by stationary local thermometers at a redshifted temperature. In static coordinates this gives $T_{\mathrm{loc}}(x)\sqrt{-g_{00}(x)}=\text{constant}$. This does not say that the restriction to every small region is exactly KMS for an independently defined local flow; the KMS property belongs to a specified global automorphism group. The local temperature is the operational, proper-time reading of that same equilibrium state.

5. The trace as a KMS state at all temperatures

Proposition 5.1 (proved below). Let $\tau$ be a faithful normal tracial state on a finite von Neumann algebra $\mathcal{M}$ (for example, a II${}_1$ factor). Let ${\sigma }_t=\mathrm{id}$. Then $\tau$ is KMS for every strip width $\beta >0$. At $\beta =0$, the degenerate boundary condition is precisely traciality.

Proof. With ${\sigma }_t=\mathrm{id}$, the function $t\mapsto \tau (a{\sigma }_t(b))=\tau (ab)$ is constant. By tracial property, $\tau (ab)=\tau (ba)$, which is the value at $t=0$ — and at $t+i\beta$ — and at every $t$. The constant function $F(z):=\tau (ab)$ is trivially holomorphic on any strip and has boundary values $F(t)=\tau (ab)=\tau (a{\sigma }_t(b))$ and $F(t+i\beta )=\tau (ab)=\tau (ba)=\tau ({\sigma }_t(b)a)$. $\square$

This is striking: the tracial state is KMS at all temperatures simultaneously, but with the trivial flow. Physically: type II factors with the trace flow have no notion of “thermal time” — the modular dynamics is degenerate. This is the algebraic content of “the trace is a uniform measure.”

Physical picture (heuristic). A normalized trace is the finite-system analogue of an infinite-temperature state: at $\beta =0$ the Gibbs weight is flat and detailed-balance ratios become $1$. This analogy is exact for finite algebras with a tracial state, such as II${}_1$ factors. A II${}_{\infty }$ factor has a faithful normal semifinite trace but no normalized tracial state, so the slogan needs that qualification. Type III factors have no faithful normal semifinite trace at all; density-matrix thermodynamics must then be replaced by the KMS/modular formulation.

For a type III factor, the modular automorphism group of a faithful normal state cannot be trivial: a trivial modular group would make the state tracial. Its image in the outer automorphism group carries state-independent information, but one must not say that ${\sigma }_t^{\omega }$ is outer for every $t\ne 0$. Exceptional inner times are measured by Connes’ $T$-invariant; in a type III${}_{\lambda }$ factor they form a non-zero discrete subgroup. Tomita-Takesaki constructs the flow; the finer inner/outer statement belongs to Connes’ classification.

6. The Tomita-Takesaki preview

Theorem 6.1 (Preview of Tomita–Takesaki). [Stated only — full statement/proof refs in Week 5.] Let $\mathcal{M}$ be a vN algebra and $\Omega$ a cyclic-separating vector for $\mathcal{M}$. Let $\omega (a):=⟨\Omega ,a\Omega ⟩$ be the corresponding faithful normal state. There is a canonical positive operator ${\Delta }_{\omega }$ on the Hilbert space — the modular operator — and a canonical one-parameter automorphism group of $\mathcal{M}$,

$${\sigma }_t^{\omega }(a):={\Delta }_{\omega }^{-it}a{\Delta }_{\omega }^{it},$$

the modular automorphism group, for which $\omega$ is KMS at $\beta =1$ in the sense of Definition 2.1.

Sign convention. We have chosen ${\sigma }_t^{\omega }=\mathrm{Ad}({\Delta }_{\omega }^{-it})$ specifically so that $\beta =+1$ in our upper-strip Definition 2.1. The opposite convention ${\sigma }_t^{\omega }=\mathrm{Ad}({\Delta }_{\omega }^{+it})$ also appears in the literature (e.g., Bratteli-Robinson Vol. II) and gives KMS at $\beta =-1$. The two are related by $t\to -t$: the same automorphism group running in opposite directions. Pick a convention and enforce it; we use $\beta =+1$ throughout this course.

This theorem is the central result of Block B. We highlight three of its features now:

6.1 The modular flow is canonical

The modular flow depends only on $\mathcal{M}$ and $\omega$ — no Hamiltonian needs to be chosen externally. In QFT, whenever the vacuum is cyclic and separating for the chosen local algebra (equivalently, its vector state is faithful and normal there), the algebra–vacuum pair therefore determines a canonical modular flow. Reeh–Schlieder supplies precisely these hypotheses for the regions treated in Block C. Calling this flow physical “thermal time” is the further interpretation discussed in §6.4.

The modular flow is intrinsic in the strongest sense: if ${\mathcal{M}}_1\cong {\mathcal{M}}_2$ as vN algebras and ${\omega }_1,{\omega }_2$ are corresponding states, then the modular flows are isomorphic. There is no choice of Hamiltonian or any external structure required.

6.2 It works in any type

The construction goes through for type I, II, and III. In type I:

• For $\mathcal{M}=\mathcal{B}({\mathcal{H}}_R)\otimes 1$ acting on ${\mathcal{H}}_R\otimes {\mathcal{H}}_L$, and $\Omega$ a generic cyclic-separating vector (a purification of a faithful density matrix ${\rho }_{\Omega }$ on ${\mathcal{H}}_R$), the modular operator is ${\Delta }_{\Omega }={\rho }_{\Omega }\otimes {\rho }_{\Omega }^{-1}$ (where ${\rho }_{\Omega }^{-1}$ acts on ${\mathcal{H}}_L$ via the GNS isomorphism).
• The modular flow on $\mathcal{B}({\mathcal{H}}_R)\otimes 1$ is ${\sigma }_t^{\omega }(a)={\rho }_{\Omega }^{-it}a{\rho }_{\Omega }^{it}=\mathrm{Ad}(e^{it{K}_{\omega }})(a)$, the Heisenberg evolution generated by the modular Hamiltonian $K_{\omega }:=-\ln {\rho }_{\Omega }$ (a positive operator). One checks directly that $\omega (a{\sigma }_t(b))$ analytically continues to $\omega ({\sigma }_t(b)a)$ at $\beta =1$, exactly as in §1.1.
• For the maximally mixed state (= tracial state), ${\rho }_{\Omega }\propto 1$, so $K_{\omega }\propto 1$ and the modular flow is trivial.

In type III, the modular flow is genuinely non-trivial. Its class modulo inner automorphisms is part of the structure used to classify type-III factors. Individual times may nevertheless be inner; this is why the $T$-invariant, rather than the blanket statement “every non-zero modular time is outer,” is the correct object.

6.3 It is non-trivial in type III

When $\mathcal{M}$ is type III, modular flow cannot be removed globally by choosing a trace: there is no faithful normal semifinite trace to provide a trivial modular reference state. Connes’ cocycle theorem says that modular groups associated with two faithful normal states differ by an inner cocycle. Consequently their images in $\mathrm{Out}(\mathcal{M})$ define a state-independent flow. The kernel and spectral data of this outer flow enter the type-III classification.

For type III${}_1$, Connes’ $S$-invariant is ${\mathbb{R}}_{\ge 0}$; equivalently, the logarithmic modular spectrum of every faithful normal state is unrestricted. One state with unrestricted spectrum is not enough, because $S$ is an intersection over all of them. This is the strongest of the type-III spectral possibilities. The statement concerns the classification invariant, not a claim that modular flow by itself supplies the physical Hamiltonian of every QFT problem.

6.4 The “thermal time” interpretation

Connes and Rovelli (1994) proposed the thermal time hypothesis: in a generally covariant quantum theory, where no external time is preferred, the modular flow of a chosen state may identify the physically relevant time evolution. The mathematical input is exact: every faithful normal state is KMS at $\beta =1$ for its own modular group. The further identification of that group with observed physical time is a hypothesis, not a consequence of Tomita-Takesaki theory.

We will use this as a guiding interpretation throughout the rest of the course, while keeping the exact modular theorem separate from the physical hypothesis.

7. Construction of a type III factor: Powers’ construction

We close the week by constructing a concrete type III factor, the Powers factor ${\mathcal{R}}_{\lambda }$. This will give us a hands-on example of a type III algebra with a non-trivial modular flow.

7.1 The setup

Fix $\lambda \in (0,1)$. On $M_2(\mathbb{C})$, define the state

$${\omega }_{\lambda }(a):=\frac{1}{1+\lambda }⟨e_1,a{e}_1⟩+\frac{\lambda }{1+\lambda }⟨e_2,a{e}_2⟩=\mathrm{Tr}\left({\rho }_{\lambda }a\right),{\rho }_{\lambda }:=\frac{1}{1+\lambda }\left(\begin{array}{cc}1& 0\\ 0& \lambda \end{array}\right).$$

This is a state with density matrix ${\rho }_{\lambda }$. Crucially, ${\omega }_{\lambda }$ is not the trace (unless $\lambda =1$).

The state ${\omega }_{\lambda }$ can be written as a Gibbs state for the Hamiltonian $H=\mathrm{diag}(0,-\log \lambda )$ at $\beta =1$: ${\rho }_{\lambda }=e^{-H}/Z$ with $Z=1+\lambda$. The modular flow on $M_2(\mathbb{C})$ is ${\sigma }_t^{{\omega }_{\lambda }}(a)=e^{itH}a{e}^{-itH}$, and it is periodic with period $\frac{2\pi }{|\log \lambda |}$ — the smallest $T$ with $e^{iTH}=1$ on the spectrum of $H$.

7.2 The infinite tensor product

Consider the inductive limit ${⨂}_{n=1}^N{M}_2(\mathbb{C})$ as $N\to \infty$, in the state ${\omega }_{\lambda }^{\otimes \infty }$ — the consistent product state. The C*-completion is the uniformly hyperfinite C*-algebra (UHF) of type $2^{\infty }$.

Form the GNS representation $\pi :(\text{UHF})\to \mathcal{B}(\mathcal{H})$ of the state ${\omega }_{\lambda }^{\otimes \infty }$. In the course convention (inner products linear in the second slot), the Hilbert space $\mathcal{H}$ is the completion of the UHF algebra in the inner product $⟨a,b⟩={\omega }_{\lambda }^{\otimes \infty }(a^{\ast }b)$. Take the WOT-closure ${\mathcal{R}}_{\lambda }:=\pi (\text{UHF}{)}^{\prime \prime }$.

This is the Araki–Woods construction (or, as Powers studied it, an ITPFI — infinite tensor product of finite type I — construction).

Physical picture (heuristic, not the classification proof). Think of ${\mathcal{R}}_{\lambda }$ as the weak closure of a half-infinite spin chain in which every site is placed in the same two-level Gibbs state. In a purification, each site is paired with an auxiliary copy and has Schmidt-weight ratio $\lambda$. Finite products only contain finitely many ratios; the infinite product contains ${\lambda }^n$ at arbitrarily large positive and negative $n$. Powers proved that the resulting factors for distinct parameters are non-isomorphic; Connes’ later modular classification packages the distinction in the invariant

$$S({\mathcal{R}}_{\lambda })=\{0\}\cup \{{\lambda }^n:n\in \mathbb{Z}\}.$$

This is a useful model for how short-distance structure can determine algebraic type. It is not a derivation of the QFT type-III${}_1$ result: that requires separate scaling and phase-space theorems, discussed in Week 12.

7.3 The Powers factor

Theorem 7.1 (Powers construction, in modern Connes notation). [Stated only.] For $\lambda \in (0,1)$, the algebra ${\mathcal{R}}_{\lambda }$ is a hyperfinite type III${}_{\lambda }$ factor. Different values of $\lambda \in (0,1)$ give non-isomorphic algebras.

Theorem 7.2 (standard UHF construction). [Stated only.] For $\lambda =1$ (the tracial case), ${\mathcal{R}}_1=\mathcal{R}$, the hyperfinite II${}_1$ factor.

Why ${\mathcal{R}}_{\lambda }$ is type III for $\lambda \in (0,1)$ — the invariant-level argument. Powers’ theorem is delicate to prove from scratch and we treat it as a black box (full proof: Powers 1967; Bratteli–Robinson Vol. II §6.2; Takesaki Vol. II Ch. XII). What we can do honestly is compute the distinguished product-state modular data and then state the later invariant that distinguishes ${\mathcal{R}}_{\lambda }$ from type II${}_1$ and from the other type-III subtypes—without simulating a classification proof we have not given.

The calculation available to us is the period of the distinguished product-state modular flow. It is state-level data, not by itself an isomorphism invariant. Apply Theorem 6.1 (Tomita–Takesaki preview) to the GNS representation of ${\omega }_{\lambda }^{\otimes \infty }$. The cyclic-separating vector is the GNS vector ${\Omega }_{\lambda }$, the modular Hamiltonian on a single tensor factor is $K_{{\omega }_{\lambda }}=-\ln {\rho }_{\lambda }=\mathrm{diag}(0,-\log \lambda )+(\log (1+\lambda ))I$, and the modular flow on that factor is

$${\sigma }_t^{{\omega }_{\lambda }}(a)={\rho }_{\lambda }^{-it}a{\rho }_{\lambda }^{it}=e^{itH}a{e}^{-itH},H=\mathrm{diag}(0,-\log \lambda ).$$

This is periodic in $t$ with period $T_{\lambda }=2\pi /|\log \lambda |$ — at $t=T_{\lambda }$ the unitary ${\rho }_{\lambda }^{-i{T}_{\lambda }}$ is a scalar phase, so $\mathrm{Ad}({\rho }_{\lambda }^{-i{T}_{\lambda }})=\mathrm{id}$.

The modular flow on the full ITPFI ${\mathcal{R}}_{\lambda }$ is the infinite tensor product of these single-site flows, and is therefore also periodic with period $T_{\lambda }$ (every site returns to the identity simultaneously). For $\lambda \in (0,1)$ this period is finite and non-zero — distinguishing ${\mathcal{R}}_{\lambda }$ from:

• Type II${}_1$ (e.g., the hyperfinite II${}_1$ factor on the tracial state): the modular flow is trivial, so every $t$ is a period — the periodicity invariant is degenerate.
• Type III${}_1$ (the QFT case): the modular flow has no period — its spectrum is all of $\mathbb{R}$.
• Type III${}_0$: degenerate limit, no useful period.

The finite, nonzero period is the visible clue. The actual conclusion that ${\mathcal{R}}_{\lambda }$ has type III${}_{\lambda }$ is the quoted factor theorem together with Connes’ later invariant theory; it cannot be inferred from the spectrum or period of one state alone. Distinct values of $\lambda \in (0,1)$ give non-isomorphic algebras: ${\mathcal{R}}_{\lambda }\cong {\mathcal{R}}_{{\lambda }^{\prime }}$ iff $\lambda ={\lambda }^{\prime }$.

On the absence of a faithful normal trace. The structural fact that no faithful normal trace exists on ${\mathcal{R}}_{\lambda }$ for $\lambda \ne 1$ is a separate part of Powers’ theorem and we do not prove it here. Naive arguments using the GNS vector state ${\omega }_{\lambda }^{\otimes \infty }$ as a “trace witness” fail — Murray-von Neumann equivalence of projections is internal to the algebra (existence of a partial isometry $u\in {\mathcal{R}}_{\lambda }$ with $u^{\ast }u=p$, $u{u}^{\ast }=q$), and a non-tracial vector state cannot certify or rule out such equivalences. Powers’ own proof distinguishes the ${\mathcal{R}}_{\lambda }$ by an asymptotic property of the algebra, his property $L_{\lambda }$. The modern route goes through Connes’ invariant $S(\mathcal{M})={\bigcap }_{\phi }\mathrm{Sp}{\Delta }_{\phi }$ (§6.3 and Week 12). For the product state ${\omega }_{\lambda }^{\otimes \infty }$ the centralizer is a factor, and in that case $S({\mathcal{R}}_{\lambda })$ equals the spectrum of this one modular operator, $\{0\}\cup {\lambda }^{\mathbb{Z}}$. Every semifinite factor has $S=\{1\}$, so no faithful normal semifinite trace can exist. Proofs: Powers 1967; Connes, Ann. Sci. ENS 6 (1973) 133; Takesaki Vol. II Ch. XII.

The contrast is striking:

• $\lambda =1$ (tracial state): ${\rho }_1=\frac{1}{2}I$, ${\omega }_1$ is the tracial state. The algebra is type II${}_1$, with a trace and continuous-dimension projections.
• $\lambda \in (0,1)$ (non-tracial state): ${\rho }_{\lambda }$ is non-tracial. The algebra is type III${}_{\lambda }$, with no trace and all projections infinite.

A small change in the state — moving away from the tracial point — qualitatively changes the algebraic type. This is one of the most surprising facts in operator algebra theory and has no analog in finite-dimensional quantum mechanics.

7.4 The modular flow on the Powers factor

By Tomita-Takesaki (Week 5), the modular automorphism group ${\sigma }_t^{\omega }$ on ${\mathcal{R}}_{\lambda }$ is a non-trivial one-parameter group. Its key invariant:

Theorem 7.3 (product-state modular flow). [Exact calculation.] For the distinguished product state on ${\mathcal{R}}_{\lambda }$, the modular flow is periodic with period $T_{\lambda }=\frac{2\pi }{|\log \lambda |}$. On each tensor factor it is the local Gibbs flow ${\sigma }_t^{{\omega }_{\lambda }}(a)={\rho }_{\lambda }^{-it}a{\rho }_{\lambda }^{it}$ (in the $\beta =+1$ convention of §6).

The full modular flow is the infinite product of these local flows. At time $T_{\lambda }$, each local flow has completed one period, returning to the identity. So the global modular flow is also periodic with period $T_{\lambda }$.

This computation exhibits the period of one particularly natural modular group. Powers’ classification theorem then identifies the factor as type III${}_{\lambda }$; in Connes’ language, the corresponding outer-period subgroup is

$$T({\mathcal{R}}_{\lambda })=\frac{2\pi }{|\log \lambda |}\mathbb{Z},$$

while $S({\mathcal{R}}_{\lambda })=\{0\}\cup {\lambda }^{\mathbb{Z}}$. The endpoint behavior must be read carefully:

• As $\lambda \to 1$, the single-site state tends to the trace and the modular flow tends pointwise to the identity. At the endpoint $\lambda =1$, the GNS closure is type II${}_1$.
• As $\lambda \to 0^{+}$, the period tends to zero, but this does not produce a type III${}_0$ factor at $\lambda =0$. The endpoint state is pure and non-faithful; its GNS representation is irreducible and its weak closure is type I${}_{\infty }$.
• Type III${}_0$ ITPFI factors require more general, typically non-stationary sequences of single-site weights. They are not the missing endpoint of the constant-$\lambda$ Powers family.

For a type III${}_1$ factor, the Connes $T$-invariant is trivial and the $S$-invariant is the full positive half-line. In the wedge example of Week 10, Bisognano–Wichmann gives an explicitly nonperiodic boost modular flow. That one calculation is consistent with the III${}_1$ classification, but the separate type theorem is still needed.

7.5 Interpolating between the types

The formula for the product states varies continuously with $\lambda \in (0,1]$, but the resulting factor type does not vary continuously in any naive topology on isomorphism classes. As $\lambda \to 1$:

• The state ${\omega }_{\lambda }$ approaches the trace.
• The modular flow period goes to $\infty$.
• The type changes from III${}_{\lambda }$ to II${}_1$.

Thus a continuous change in the local state parameter labels mutually non-isomorphic factors for $0<\lambda <1$ and reaches a different type at $\lambda =1$. This is the precise—and sufficiently surprising—statement; no topology on the space of factor types is being asserted.

8. Where this leaves us

We now have:

• The KMS condition as the algebraic notion of thermal equilibrium.
• The verification that KMS = Gibbs in finite dimensions.
• The preview that Tomita-Takesaki, applied to a faithful normal state $\omega$ (equivalently, a cyclic-separating vector in a standard representation), produces the state’s canonical modular flow ${\sigma }_t^{\omega }$; the state is KMS for this flow at $\beta =1$ in our convention.
• A concrete type III factor (the Powers factor), an exact calculation of its product-state modular period, and the later Connes invariants that turn the model data into algebra-level classification.
• The thermal time interpretation: modular flow is the algebraic clock.

Block B picks up exactly here: it constructs the modular operator $\Delta$ and modular conjugation $J$ from a cyclic-separating vector, proves Tomita-Takesaki, and develops the Connes cocycle and Araki-Uhlmann relative entropy. Block C applies all this to QFT.

9. Looking ahead: Week 5

Next week we begin Block B with the central technical construction: given a cyclic-separating vector $\Omega$ for a vN algebra $\mathcal{M}$, define the Tomita operator $S$ by $Sa\Omega =a^{\ast }\Omega$, polar-decompose $S=J{\Delta }^{1/2}$, and identify $\Delta$ as the modular operator and $J$ as the modular conjugation. We compute these explicitly for type I (everything reduces to density-matrix language) and prove the key parts of Tomita’s theorem in the type I case. The full type III statement is stated only, with references.

10. Problem set

Core problems.

1. KMS for the Gibbs state. Let $\mathcal{H}={\mathbb{C}}^2$, $H=\mathrm{diag}(0,E)$ with $E>0$, ${\sigma }_t=\mathrm{Ad}{e}^{itH}$. (a) Verify the KMS condition for the Gibbs state ${\omega }_{\beta }(a)=\mathrm{Tr}({\rho }_{\beta }a)$ explicitly: pick $a={\sigma }_x$, $b={\sigma }_y$, compute ${\omega }_{\beta }(a{\sigma }_t(b))$ on the real axis and verify the holomorphic extension to $t\to t+i\beta$. (b) Show that any other state on $M_2(\mathbb{C})$ that satisfies KMS at $\beta$ for this ${\sigma }_t$ must equal ${\omega }_{\beta }$. (Hint: follow the matrix-units argument of Theorem 3.1’s $(\Rightarrow )$ direction with $H=\mathrm{diag}(0,E)$ — diagonal weighting from $a=E_{12},b=E_{21}$, off-diagonal vanishing from $a=E_{12},b=E_{22}$.)

2. The trace as KMS for the trivial flow. On $M_n(\mathbb{C})$ with ${\sigma }_t=\mathrm{id}$, verify the KMS condition for the normalized trace $\tau$ at every $\beta >0$. Identify $F_{a,b}(z)$ explicitly and verify both boundary values. Then discuss $\beta =0$ only as the degenerate tracial limit of the strip condition; negative $\beta$ is outside the convention adopted in this course.

3. KMS detailed balance. On $M_n(\mathbb{C})$ with $H,{\sigma }_t=\mathrm{Ad}{e}^{itH}$: (a) For energy eigenstates $|\xi ⟩,|{\xi }^{\prime }⟩$ with $H\xi =E\xi ,H{\xi }^{\prime }=E^{\prime }{\xi }^{\prime }$, show that the KMS condition forces $\omega (|\xi ⟩⟨\xi |)/\omega (|{\xi }^{\prime }⟩⟨{\xi }^{\prime }|)=e^{-\beta (E-E^{\prime })}$ — the thermal weighting. (b) Use (a) to give a complete proof of Theorem 3.1 (that KMS at $\beta$ on $M_n(\mathbb{C})$ implies $\omega ={\omega }_{{\rho }_{\beta }}$).

4. Powers factor: non-traciality. For the Powers state ${\omega }_{\lambda }^{\otimes \infty }$ on ${⨂}_{n=1}^{\infty }{M}_2(\mathbb{C})$ with $\lambda \in (0,1)$: (a) Find specific $a,b\in {\mathcal{A}}_{\infty }$ with ${\omega }_{\lambda }^{\otimes \infty }(ab)\ne {\omega }_{\lambda }^{\otimes \infty }(ba)$, demonstrating non-traciality. (b) Compute ${\omega }_{\lambda }^{\otimes \infty }(a{\sigma }_t(b))$ for $a,b$ on a single tensor factor and identify the periodic oscillation period $T_{\lambda }=2\pi /|\log \lambda |$.

5. Product-state modular period and its endpoint. For the Powers factor ${\mathcal{R}}_{\lambda }$, the modular period is $T_{\lambda }=2\pi /|\log \lambda |$. Verify: (a) $T_{\lambda }\to \infty$ as $\lambda \to 1$ (i.e., the flow becomes trivial). (b) $T_{\lambda }\to 0$ as $\lambda \to 0$ (i.e., the flow becomes very fast). (c) $T_{1/2}=2\pi /\log 2\approx 9.06$. Compute explicitly the modular flow at $t=T_{1/2}/4\approx 2.26$ on the simplest non-trivial element (e.g., a single-site ${\sigma }_x$). (d) Explain why (b) is only an asymptotic statement inside the family $0<\lambda <1$: at $\lambda =0$ the product state is pure and non-faithful, and its GNS weak closure is type I${}_{\infty }$, not type III${}_0$.

6. KMS in QFT (warmup for Bisognano–Wichmann). Anticipating Week 10, consider the derivative field of a free two-dimensional massless scalar—or, equivalently for this formal exercise, restrict scalar smearings to test functions of zero integral so that the infrared zero mode is absent. The vacuum two-point distribution is logarithmic, with the appropriate $i0$ prescription. In units $\hslash =c=1$, the formal right-wedge modular Hamiltonian density on the $t=0$ half-line is $2\pi {\int }_0^{\infty }{x}^1{T}^{00}(0,x^1)d{x}^1$. Denote the standard global boost generator by $K$, so that the unitary below is $U(B_s)=e^{isK}$. On right-wedge observables the one-sided stress-tensor expression and the global boost implementer induce the same geometric automorphism, but they are not the same global operator formula. (a) Compute $W(x,y)$ for $x,y$ in the wedge. (b) With $U(B_s)=e^{isK}$ and scalar covariance $U(B_s)\varphi (f)U(B_s{)}^{\ast }=\varphi (f\circ B_{-s})$, verify formally that

$${\sigma }_t(W(f))=W(f_t),f_t(x)=f(B_{-2\pi t}x).$$

(c) Check that the vacuum state, restricted to wedge observables, is KMS at $\beta =1$ for this flow. (Just outline; full details Week 10.)

Starred problems.

7*. The KMS condition determines the dynamics. Suppose $\omega$ is a faithful state on $\mathcal{B}(\mathcal{H})$ ($\dim \mathcal{H}<\infty$) and $\omega$ is KMS at $\beta =1$ for a one-parameter automorphism group ${\sigma }_t$. Show that ${\sigma }_t$ is uniquely determined by $\omega$. (Hint: if $\omega (a)=\mathrm{Tr}(\rho a)$, compare the KMS condition with ${\sigma }_t^{\omega }(a)={\rho }^{-it}a{\rho }^{it}$; adding a scalar to the implementing Hamiltonian changes no automorphism.) Explain why there is no “finite-dimensional type-II factor” analogue: every finite-dimensional factor is type I. Then state the von Neumann algebraic uniqueness theorem for the modular group of a faithful normal state.

8*. KMS and complete passivity (Pusz–Woronowicz). For part (a), work in finite dimension so that no energy-domain issue is hidden. A state $\omega$ on $\mathcal{B}(\mathcal{H})$ is passive with respect to ${\sigma }_t=\mathrm{Ad}{e}^{itH}$ if $\omega (U^{\ast }HU)\ge \omega (H)$ for every unitary $U$—one cannot extract energy from $\omega$ by a cyclic unitary process. (a) Show that the Gibbs state at any $\beta >0$ is passive. (b) A state is completely passive if every tensor power is passive for the sum Hamiltonian. State, without proof, the Pusz–Woronowicz theorem: completely passive states are KMS states (including the tracial $\beta =0$ case) or ground states, under the theorem’s dynamical hypotheses. Explain why passivity of a single copy is weaker and does not imply KMS.

9*. Modular flow on a tensor product. Verify that for $\mathcal{M}=M_n(\mathbb{C})$ and a generic non-tracial state $\omega ={\omega }_{\rho }$, the modular flow in the course convention (${\sigma }_t^{\omega }=\mathrm{Ad}({\Delta }^{-it})$, §6.1) is ${\sigma }_t^{\omega }(a)={\rho }^{-it}a{\rho }^{it}$. Compute the period (if any) of this flow for $\rho =\mathrm{diag}({\rho }_1,\dots ,{\rho }_n)$.

10** (Optional, hard). Read Powers’ original construction and then a modern account of Connes’ later classification. Compute the product-state modular spectrum and use the later $S$-invariant

$$S({\mathcal{R}}_{\lambda })=\{0\}\cup \{{\lambda }^n:n\in \mathbb{Z}\}$$

to distinguish the factors for different $\lambda \in (0,1)$. Formulate $S$ using spectra of $\Delta$—or, equivalently, exponentials of the corresponding $\log \Delta$ spectral data. Do not attribute the Connes invariant, introduced after 1967, to Powers’ original proof.

Self-study answer checkpoints

These checkpoints cover the core calculations. The starred problems remain source-led extensions, particularly complete passivity and the Powers/Connes classification step.

1. Gibbs KMS state. With $p_0=Z^{-1}$ and $p_1=e^{-\beta E}{Z}^{-1}$,

   $$F_{{\sigma }_x,{\sigma }_y}(t)=i{p}_0{e}^{iEt}-i{p}_1{e}^{-iEt}.$$

   At $t+i\beta$, the relations $p_0{e}^{-\beta E}=p_1$ and $p_1{e}^{\beta E}=p_0$ give the reversed boundary ordering. Applying KMS to matrix units forces all off-diagonal state entries to vanish and the diagonal ratio $p_1/p_0=e^{-\beta E}$, hence the Gibbs state is unique.

2. Trivial flow. Take the constant entire function $F_{a,b}(z)=\tau (ab)$. Its upper boundary is $\tau (ba)=\tau (ab)$. This works for every $\beta >0$; $\beta =0$ is only the collapsed tracial limit, not a positive-width KMS strip.

3. Detailed balance. Matrix units between energy eigenspaces give

   $$\frac{\omega (P_E)}{\omega (P_{E^{\prime }})}=e^{-\beta (E-E^{\prime })}.$$

   Off-diagonal matrix elements vanish between distinct energies, while within degenerate eigenspaces KMS forces the normalized trace. Thus $\rho =Z^{-1}{e}^{-\beta H}$ block by block.

4. Powers non-traciality. At one site choose $a=E_{12}$ and $b=E_{21}$. Then $\omega (ab)=1/(1+\lambda )$ while $\omega (ba)=\lambda /(1+\lambda )$. Moreover,

   $$\omega \left(a{\sigma }_t(b)\right)=\frac{e^{-it\log \lambda }}{1+\lambda },$$

   whose fundamental period is $2\pi /|\log \lambda |$.

5. Period and endpoints. $T_{\lambda }\to \infty$ as $\lambda \to 1$ and $T_{\lambda }\to 0$ as $\lambda \to 0^{+}$. For $\lambda =1/2$, $T=2\pi /\log 2$ and

   $${\sigma }_{T/4}({\sigma }_x)={\sigma }_y.$$

   At $\lambda =0$ the state loses faithfulness and the GNS closure is type I${}_{\infty }$; there is no type-III${}_0$ endpoint.

6. Wedge KMS preview. In the infrared-safe two-dimensional theory the two-point distribution is the logarithmic Wightman boundary value with the time prescription $x^0-y^0-i0$. Scalar covariance gives

   $${\sigma }_t(W(f))=W(f\circ B_{-2\pi t}).$$

   Boost analyticity continues the correlator through the modular strip, and the upper boundary reverses the two wedge observables. This is the check; Week 10 supplies the theorem and its domain control.

Wiki connections. Liu’s lectures on entanglement and von Neumann algebras

---

Notes prepared for the algebraic-QFT course, Semester I Block A. Last revised 2026-09-29. End of Block A — the foundations of operator algebras.