Type-III Local Algebras, Entropy, and Cutoff Limits

Exact continuum localization generally supplies a type-III von Neumann algebra, not a tensor factor with an intrinsic reduced density matrix. Consequently the sharp-region expression $-\operatorname{Tr}\rho_{\mathcal O}\log\rho_{\mathcal O}$ is not defined by the local algebra alone. The mathematically controlled replacements are algebraic relative entropy, mutual information for separated algebras when a normal product state exists, and entropy computed only after specifying a regulator or a split collar. The type-III local-algebra analysis makes this distinction concrete for a free scalar Rindler cut: the lattice entropy diverges as the cutoff is removed, separated mutual information can remain finite, and a split entropy depends on the collar and the intermediate type-I factor.

Required background. Representation types and local factors supplies the Murray–von Neumann classification used below; standard von Neumann algebras and Tomita–Takesaki theory supplies natural-cone representatives and relative modular operators; split inclusions and statistical independence supplies the intermediate type-I factor used to introduce a collar-dependent tensor product.

Helpful background. Type-III local algebras supplies the continuum subsystem interpretation; factorization failure separates commuting algebras from Hilbert-space tensor factors; continuum subsystem choice explains why the algebra is part of the subsystem specification; regulated subregion entropy, replica branched geometries, twist and replica defects, Gaussian correlation-matrix entropy, and free-field entropy provide regulated calculations; zero modes, boundaries, and infrared limits and ultraviolet divergences and the area law identify the two principal limiting hazards.

What type III forbids

Let $\mathcal M$ be a factor. A projection $E\in\mathcal M$ is finite if every partial isometry $V\in\mathcal M$ satisfying $V^*V=E$ and $VV^*\leq E$ actually has $VV^*=E$. A type-III factor has no nonzero finite projection. Equivalently, it admits no nonzero normal semifinite trace. This is the precise obstruction behind the density-matrix warning: there is no intrinsic trace $\operatorname{Tr}_{\mathcal M}$ with which to write every normal state as

$$\omega(A)=\operatorname{Tr}_{\mathcal M}(\rho_{\mathcal M}A), \qquad S(\rho_{\mathcal M})=-\operatorname{Tr}_{\mathcal M} (\rho_{\mathcal M}\log\rho_{\mathcal M}).$$

A representation-dependent subtlety matters. If $\mathcal M\subset B(\mathcal H)$ is concrete, a normal functional on $\mathcal M$ can be represented, nonuniquely, by a trace-class operator on the ambient $\mathcal H$. That extension does not turn $\mathcal M$ into $B(\mathcal H_{\mathcal O})$, does not produce a canonical factorization $\mathcal H\simeq\mathcal H_{\mathcal O}\otimes\mathcal H_{\mathcal O'}$, and does not define an intrinsic local entropy. Type is an algebraic statement, not the claim that normal local states cease to exist.

Under additional locality, covariance, scaling, and phase-space assumptions, local algebras in relativistic QFT are often isomorphic to the unique hyperfinite type-$\mathrm{III}_1$ factor. This is a theorem for specified classes of nets, not a consequence of isotony and locality alone: Buchholz, D’Antoni, and Fredenhagen 1987, pp. 123–135 state hypotheses leading to the universal hyperfinite type-$\mathrm{III}_1$ structure, while Yngvason 2005, pp. 8–11 reviews which structural inputs enter. Dropping those inputs permits other factor types and nonfactorial centers.

Relative modular entropy is intrinsic

Let $(\mathcal M,\mathcal H,J,\mathcal P)$ be a standard form and let $\varphi,\psi$ be faithful normal states with their unique natural-cone vectors $\xi_\varphi,\xi_\psi\in\mathcal P$. On the dense domain $\mathcal M\xi_\varphi$, define the antilinear relative Tomita operator

$$S_{\psi\mid\varphi,0}(A\xi_\varphi)=A^*\xi_\psi, \qquad A\in\mathcal M.$$

It is closable; its closure has polar decomposition

$$S_{\psi\mid\varphi} =J_{\psi\mid\varphi}\Delta_{\psi\mid\varphi}^{1/2}.$$

The Araki relative entropy is

$$S_{\mathcal M}(\varphi\Vert\psi) =-\big\langle\xi_\varphi, \log\Delta_{\psi\mid\varphi}\,\xi_\varphi\big\rangle,$$

provided the logarithmic quadratic form is defined; the extended value $+\infty$ is allowed. Nonfaithful states require support projections and the extended definition. In a type-I matrix algebra this reduces to $\operatorname{Tr}\rho_\varphi(\log\rho_\varphi-\log\rho_\psi)$, but the modular expression remains meaningful without a trace. Araki proves positivity, monotonicity under restriction, and the relation to relative modular operators in Araki 1976, §§ 1–8, pp. 809–833.

For commuting separated algebras $\mathcal A$ and $\mathcal B$ with a normal product state $\varphi_{\mathcal A}\otimes\varphi_{\mathcal B}$ on $\mathcal A\vee\mathcal B$, the algebraic mutual information may be defined by

$$I_\varphi(\mathcal A:\mathcal B) =S_{\mathcal A\vee\mathcal B} \bigl(\varphi\Vert \varphi_{\mathcal A}\otimes\varphi_{\mathcal B}\bigr).$$

The split property licenses the normal product state and a type-I realization; it does not by itself prove that this relative entropy is finite. Finiteness needs further state-dependent ultraviolet estimates. Monotonicity immediately gives a useful check: shrinking either algebra cannot increase $I_\varphi$.

A free scalar across a Rindler cut

Take the Minkowski vacuum of a free scalar field and regulate a spatial half-space by a lattice spacing $a$. The regulated Hilbert space factorizes and its Gaussian covariance matrix defines a reduced density operator. In $d$ spacetime dimensions the leading entropy behaves schematically as

$$S_a\sim c_{d-2}\frac{\operatorname{Area}(\partial R)}{a^{d-2}} +\text{subleading terms},$$

with regulator- and theory-dependent coefficient $c_{d-2}$; in two dimensions the leading behavior is logarithmic instead. The divergence is not a failed numerical limit. It records that the sharp continuum wedge algebra is type III and supplies no limiting trace-class reduced density operator. Free-field mode and correlation-matrix derivations, including the ultraviolet and infrared qualifications, are reviewed in Casini and Huerta 2009, §§ 2–3, pp. 4–37.

Now separate the wedge from its complement by a collar of width $\varepsilon>0$. If

$$\mathcal A(R_{-\varepsilon})\subset \mathcal F_\varepsilon\subset \mathcal A(R), \qquad \mathcal F_\varepsilon\simeq B(\mathcal K_\varepsilon),$$

is a split inclusion, the restriction of the state to $\mathcal F_\varepsilon$ has a density matrix and a von Neumann entropy. But $\mathcal F_\varepsilon$ is not unique. Its entropy may depend on that choice and normally diverges as $\varepsilon\downarrow0$. By contrast, relative entropy and mutual information assigned to two fixed, positively separated algebras do not require choosing $\mathcal F_\varepsilon$; their possible finiteness is a separate analytic result, not a converse to the split property.

Proof mechanism and independent checks

The structural argument has three distinct steps. First, the absence of finite projections rules out an intrinsic normal trace on a type-III factor. Second, standard form replaces trace formulas by relative modular operators and their logarithmic quadratic forms. Third, an explicit regulator or split factor temporarily returns to type I, where ordinary density matrices exist; removing that auxiliary structure is a nonuniform ultraviolet limit.

Two checks keep the claims separated. For $\mathcal M=M_n(\mathbb C)$, the ordinary matrix trace exists and the relative modular formula exactly reproduces Umegaki relative entropy. For a type-III factor, assuming a faithful normal density-matrix trace would give finite spectral projections of nonzero weight, contradicting the defining absence of nonzero finite projections. Thus the same formula cannot simply be carried across the type boundary.

Adversarial test. Demand a trace-one reduced density operator for the exact continuum Rindler wedge, with no lattice, collar, or chosen type-I factor. The demand is ill-posed: the wedge algebra and its commutant do not provide the required type-I tensor factorization. A formal thermal expression for boosts may encode the KMS condition, but it is not a trace-class Gibbs density matrix for the wedge algebra.

Exercise

Let $\rho$ and $\sigma$ be faithful density matrices on a finite-dimensional Hilbert space. Represent $B(\mathcal K)$ on the Hilbert–Schmidt space and show that the relative modular formula gives the usual relative entropy.

Solution

Use $\xi_\rho=\rho^{1/2}$ and identify the relative modular operator as

$$\Delta_{\sigma\mid\rho}=L_\sigma R_{\rho^{-1}},$$

where $L$ and $R$ denote left and right multiplication. The two commuting multiplication operators allow

$$\log\Delta_{\sigma\mid\rho} =L_{\log\sigma}-R_{\log\rho}.$$

With the Hilbert–Schmidt inner product,

$$-\langle\rho^{1/2}, \log\Delta_{\sigma\mid\rho}\,\rho^{1/2}\rangle =\operatorname{Tr}\rho(\log\rho-\log\sigma).$$

The calculation relies on the finite-dimensional trace. The modular definition, not that trace computation, is what survives for a type-III algebra.

References

• Araki, Huzihiro. “Relative Entropy of States of von Neumann Algebras.” Publications of the Research Institute for Mathematical Sciences 11, no. 3 (1976): 809–833. DOI.
• Buchholz, Detlev, Claudio D’Antoni, and Klaus Fredenhagen. “The Universal Structure of Local Algebras.” Communications in Mathematical Physics 111, no. 1 (1987): 123–135. DOI.
• Casini, Horacio, and Marina Huerta. “Entanglement Entropy in Free Quantum Field Theory.” Journal of Physics A: Mathematical and Theoretical 42, no. 50 (2009): 504007. DOI. Open PDF.
• Yngvason, Jakob. “The Role of Type III Factors in Quantum Field Theory.” Reports on Mathematical Physics 55, no. 1 (2005): 135–147. DOI. Open PDF.