Relating the modular Hamiltonian to two-point

functions
arXiv:2501.09669v2 [math-ph] 22 Feb 2025

Markus B. Fröb
Institut für Theoretische Physik, Universität Leipzig, Brüderstraße 16, 04103 Leipzig, Germany

E-mail: mfroeb@[Link]

Abstract: We consider the modular Hamiltonian associated to standard subspaces for

a free scalar field in a globally hyperbolic spacetime in an arbitrary Gaussian state. We
show how the modular Hamiltonian is related to the two-point function of the theory. For
the restriction of the modular Hamiltonian to the subspace, we recover formulas that were
obtained previously by Peschel, Casini and Huerta. We also show how the same results can
be obtained more directly from the KMS condition, and generalize our results to general
CCR algebras.
Contents

1 Introduction 1

2 Free scalar fields 4

2.1 Modular data 5
2.2 An alternative formula for the modular Hamiltonian 7
2.3 Relation to the two-point function 8
2.4 General CCR algebras 12
2.5 KMS condition 13

References 19

1 Introduction

The Tomita–Takesaki theory of modular flows of von Neumann algebras [1, 2] has found
a multitude of applications in quantum field theory. In particular, it allows to generalize
certain concepts and quantities that are defined in terms of density matrices and traces
for finite-dimensional quantum systems to relativistic field theory, where they would be
otherwise ill-defined without imposing an ultraviolet cutoff. A prominent example is relative
entropy, which given a Hilbert space H and two density matrices ρ and σ is defined as

Srel (ρ∥σ) = trH (ρ ln ρ − ρ ln σ) , (1.1)

and which is a measure for the amount of information needed to distinguish two states.
More precisely, given a system in a state described by ρ, the probability of mistakenly
ascribing to it the state described by σ after a large number N of measurements has been
performed is proportional to exp[−N Srel (ρ∥σ)].
While each of the ingredients in Eq. (1.1) is on its own ill-defined in a quantum field
theory (or more specifically for a theory whose local von Neumann algebras are of type III),
it is possible to rewrite the equation in such a way that it remains well-defined also in the
latter case [3–5]. Namely, one considers instead of H the bipartite Hilbert space H ⊗ H, and
derives the density matrices ρ and σ as partial traces of two cyclic and separating states
|Φ⟩, |Ψ⟩ ∈ H⊗H (such states can be constructed by purification). The relative entropy (1.1)
can then be written as
   
Srel (ρ∥σ) = − tr ρ ln ∆ρ|σ = −ωρ ln ∆ρ|σ , (1.2)

where
ln ∆ρ|σ = ln σ ⊗ 1 − 1 ⊗ ln ρ (1.3)

–1–
is the so-called relative modular Hamiltonian. Tomita–Takesaki theory shows that this
operator is well-defined also in quantum field theory, and thus the definition
 
Srel (Φ∥Ψ) = −ωΦ ln ∆Φ|Ψ = −⟨Φ| ln ∆Φ|Ψ |Φ⟩ (1.4)

makes sense for two arbitrary cyclic and separating states |Φ⟩ and |Ψ⟩ in field theory.
It is not our task nor our goal to give a lengthy introduction to Tomita–Takesaki mod-
ular theory and the many beautiful identities and properties that the modular operators
have. Instead, we focus on the part that seems most important for practical applications,
namely the actual computation of ln ∆Φ|Ψ . An important case for applications is if the
states |Φ⟩ and |Ψ⟩ are obtained as unitary excitations of a common state |Ω⟩, in which
case the relative modular operator ln ∆Φ|Ψ can be obtained by a unitary conjugation of the
modular operator ln ∆Ω of the common state. This in turn is known in a variety of cases,
with the most famous one probably being for the algebra of fields restricted to a wedge in
Minkowski spacetime and the Minkowski vacuum state (the Bisognano–Wichmann theo-
rem [6, 7]). While this result holds for arbitrary Wightman theories, other examples only
apply to free theories or in lower dimensions. In even dimensions, we mention massless free
scalar fields in the Minkowski vacuum state and the algebra of fields inside the future light
cone [8] or a double cone [9, 10], and in general dimensions scalar fields in the de Sitter
Euclidean (or Bunch–Davies) vacuum state and the algebra of fields inside the static patch
of de Sitter spacetime (the de Sitter wedge) [11–18], as well as the analogous result for
anti-de Sitter spacetime [19] or FLRW spacetimes [20, 21]. For the Schwarzschild black
hole, the modular Hamiltonian is known for free scalar fields in the Hartle–Hawking state
of the Kruskal extension of the Schwarzschild metric, with the algebra of fields localized in
either the right or left Kruskal wedge [22–25]. The results for massless free fields restricted
to either the future light cone or a double cone in Minkowski spacetime can also be gener-
alized to conformal fields in conformally flat spacetimes [26], and many more examples are
known in one or two dimensions, both for free fields (scalars and fermions) and conformal
field theories in various quantum states and for the algebras of fields restricted to compli-
cated subregions (also including defects and boundaries), see for example Refs. [27–59] and
references therein.
Let us also mention a technical point: while (smeared) fermion fields are bounded
operators, bosonic fields are unbounded even after smearing. To obtain a von Neumann
algebra of bounded operators and apply the tools of Tomita–Takesaki theory, one thus has
to consider exponentiated fields, the Weyl operators W (f ) which satisfy the product rule

W (f )W (g) = e−iσ(f,g) W (f + g) (1.5)

for a symplectic form σ. In a Fock representation, these are obtained as eiϕ(f ) where ϕ
is the representation of the usual scalar field, and the product rule substitutes for the
canonical commutation relations. If f is real (and ϕ a Hermitean field), they are unitary
operators and thus bounded, but if f is permitted to be complex even the Weyl operators
are unbounded. We thus have to restrict to real f and Hermitean ϕ, but since we can
decompose a complex bosonic field (or a complex f ) into two independent real parts, this
is fortunately no restriction.

–2–
 In this work, we consider a free scalar field theory in a general curved spacetime. We
work with the formulation in terms of initial-value data on a Cauchy surface Σ, and the von
Neumann algebra that we consider is generated by Weyl operators W (f ) with f restricted
to some region R ⊂ Σ. For free theories, everything descends to the one-particle Hilbert
space H and the subspace L corresponding to initial-value data supported in R. We show
how one can relate the abstract expression for the modular Hamiltonian ln ∆ in terms of the
projector P on the standard subspace L elaborated in detail in Refs. [60–64], namely (2.19)

ln ∆ = 2 arcoth(1 − P + IP I) , (1.6)

to a concrete formula for the integral kernel of its restriction to the subspace in terms of
the two-point function G. This formula reads (2.40)
!
0 2M
I ln ∆ R
= (1.7)
−2N 0

with (2.41)
1 1 1 1
M = Π 2 B −1 arcoth(2B)Π 2 , N = Π− 2 B arcoth(2B)Π− 2 , (1.8)
q
1 1
where B = Π 2 XΠ 2 and X and Π are the correlation functions of the field and its
conjugate momentum in the region R, seen as convolution operators. A very similar formula
is known from the work of Casini and Huerta [31], namely
1 2C + 1 1 2C + 1
   
M =Π ln , N= ln X, (1.9)
2C 2C − 1 2C 2C − 1
√
where C = XΠ. For the discretized theory where X and Π are bounded operators and
which is treated in Ref. [31], we show that both are equivalent. For fermions, a similar
connection was shown very recently [65], but for bosons such a proof is still missing. We
also generalize our results to general CCR algebras, and finally show that the main relations
can alternatively be obtained from the KMS condition that the state ω satisfies with respect
to the modular flow.
While the formula (1.6) can be used for numerical computations [64], to determine the
modular Hamiltonian analytically it seems much more expedient to use the relation (1.7).
Indeed, using the spectral decomposition of the two-point function (seen as a convolution
operator) and functional calculus, Casini and Huerta [30] computed the explicit form of
the modular Hamiltonian for massless fermions in 1+1 dimensions and the Minkowski
vacuum state, in the case where the subspace L corresponds to initial data with support
in a collection of intervals. Later on, the same method was used to compute corrections
for small mass [36, 65], and similarly the analogue of Eq. (1.7) for the U (1) current on the
line was used to compute the modular Hamiltonian for various intervals [39]. Instead of
performing the spectral decomposition explicitly, it is also possible to define the logarithm
as an integral over the resolvent (our result (2.27)), which seems to be a very efficient
method of obtaining the modular Hamiltonian in a variety of settings, see for example the
recent works [48, 66, 67] and references therein that treat fermions. Moreover, it is possible

–3–
to consider a finite-dimensional approximation of the right-hand side (e.g., by applying
a lattice discretization), determining the corresponding modular Hamiltonian and finally
taking the limit where the approximation becomes exact. Results using this approach can
be found in Refs. [52, 53, 56, 68] and references therein.

2 Free scalar fields

We consider a globally hyperbolic spacetime with Cauchy surface Σ. The quantization of a

free scalar field on such a spacetime is well-known (see for example Ref. [25] and references
therein), and we only recapitulate the main features. Initial data for the Klein–Gordon
equation is given by a pair of real-valued, smooth and compactly supported functions
f = (f1 , f2 ) ∈ S = C0,R ∞ (Σ) ⊕ C ∞ (Σ), and the classical time evolution preserves the
0,R
(non-degenerate) symplectic form
1
Z
σ(f, g) = (f1 g2 − g1 f2 ) dx . (2.1)
2 Σ

The quantization of this system proceeds as follows [25, Prop. 3.1]: consider a bilinear
positive symmetric form µ on S × S satisfying the inequality

σ 2 (f, g) ≤ µ(f, f )µ(g, g) (2.2)

for all f, g ∈ S. The inequality is strict if the state is mixed, but for pure states equality
might be obtained. Since a mixed state can always be represented as a pure state on a
larger Hilbert space (so-called purification) [69, Ch. 4], we restrict in the following to pure
states. Then there exists a complex Hilbert space H with complex structure I (a bounded
operator satisfying I 2 = −1 and I † = −I with respect to µ) such that S is dense in H and
that
⟨f, g⟩ = µ(f, g) + iσ(f, g) (2.3)
for all f, g ∈ S. Complex multiplication in H is defined by if = If , which entails in
particular that µ(f, g) = σ(f, Ig). The full Hilbert space of the theory is then constructed
as the usual symmetric Fock space over this one-particle Hilbert space, but we only need
the one-particle Hilbert space H in the following. The symmetry of µ and the antisymmetry
of σ also entail that

σ(f, g) = −µ(f, Ig) = −µ(Ig, f ) = −σ(Ig, If ) = σ(If, Ig) , µ(If, Ig) = µ(f, g) . (2.4)

The relations (2.4) easily yield that I † = −I, where the adjoint is computed either with
respect to the full complex scalar product or its real part µ. We note that the construction
of a state (and thus a complex structure) in a general spacetime is a non-trivial task, see
for example the recent Refs. [70–74] and the many references therein.
Consider now a closed real-linear subspace L ⊂ H. We assume that L is standard and
factorial, which means that L ∩ IL = {0} (separating), that L + IL is dense in H (cylic),
such that H = L ⊕ IL, and that L ∩ (IL)⊥ = {0}, where the orthogonal complement is
taken with respect to the real part µ of the scalar product. It follows straightforwardly

–4–
that also L⊥ ∩ IL = {0} and L⊥ ∩ (IL)⊥ = L⊥ ∩ IL⊥ = {0}1 . Namely, for f ∈ L⊥ ∩ IL
we have f = Ih for some h ∈ L and µ(f, g) = 0 ∀ g ∈ L, hence µ(h, Ig) = 0 ∀ g ∈ L
using that I † = −I and thus h ∈ L ∩ (IL)⊥ = {0} and f = I0 = 0. The second relation
is shown analogously, and the four subspaces L, IL, L⊥ and (IL)⊥ are thus in generic
position [75, 76]. In applications, such a subspace corresponds to functions with support
inside a region R ⊂ Σ, for example a half-space (the initial data for a wedge) or a sphere
(the initial data for a double cone). In order for L to be cylic, I must then be an anti-local
operator [77–79], which in particular entails that If has support in all of Σ even if f is
supported only in R. Let P be the operator that multiplies h ∈ H with the characteristic
function of R such that im P = L. We compute

µ((1 − P )f, IP g) = −σ((1 − P )f, P g) = 0 , (2.5)

where the last equality holds because of the explicit expression (2.1) of the symplectic form.
It follows that im(1−P ) = ker P = (IL)⊥ , and thus P is not an orthogonal but a symplectic
projection (called “cutting projection” in [63]). Since H = L ⊕ IL, any vector h ∈ H can
be uniquely written as h = f + Ig with f, g ∈ L. Using the symplectic projection P , we
also obtain the decomposition h = f + g with f = P h ∈ L and g = (1 − P )h ∈ (IL)⊥ ,
such that H = L ⊕ (IL)⊥ also holds. Multiplying by the complex structure, we further
obtain the decomposition h = f + g with f = −IP Ih ∈ IL and g = (1 + IP I)h, and since
µ(f, (1 + IP I)h) = µ(If, (1 − P )Ih) = 0 ∀ f ∈ L because of im(1 − P ) = (IL)⊥ , we have
g = (1 + IP I)h ∈ L⊥ and it holds that H = L⊥ ⊕ IL. To determine the adjoint P † of the
symplectic projection with respect to µ, we take f + g ∈ L ⊕ (IL)⊥ and h + k ∈ L⊥ ⊕ IL,
and compute
 
µ P † (h + k), f + g + µ(IP I(h + k), f + g)
= µ(h + k, P (f + g)) − µ(P I(h + k), I(f + g)) (2.6)
= µ(h + k, f ) − µ(Ik, I(f + g)) = 0 ,

where we used that µ(If, Ig) = µ(f, g) and that Ih ∈ (IL)⊥ such that P Ih = 0. It follows
that P † = −IP I, and clearly P is not an orthogonal projection. Moreover, with the above
results we have im P † = IL and ker P † = L⊥ .

2.1 Modular data

For a free scalar field in the Fock representation, the modular data is obtained as the
second quantization of the one-particle modular data [80, 81], such that we can restrict to
the one-particle level. In the above standard subspace setting, all the modular objects have
been determined and their properties elaborated [60–64], and we will give just a summary
of the main formulas and derivations.
Using that any vector h ∈ H can be uniquely written as h = f + Ig with f, g ∈ L, the
Tomita operator S is well-defined as the closure of the densely defined anti-linear involution

S : f + Ig 7→ f − Ig . (2.7)
1
For the first equality, consider g ∈ (IL)⊥ ⇔ µ(g, If ) = 0 ∀ f ∈ L ⇔ µ(Ig, f ) = 0 ∀ f ∈ L ⇔ Ig ∈ L⊥ ⇔
g ∈ IL⊥ , using that I 2 = −1 and µ(If, Ig) = µ(f, g).

–5–
Its polar decomposition results in the modular conjugation J and the modular operator ∆,
and we would like to obtain a formula for ∆ in terms of the projector P on the standard
subspace L. For this, we write ∆ = S † S and consider

⟨S(f + Ig), h + Ik⟩ = ⟨f − Ig, h + Ik⟩ = µ(f − Ig, h + Ik) + iσ(f − Ig, h + Ik)
(2.8)
= µ(h + Ik, f − Ig) + iµ(h + Ik, If + g)

for h + Ik ∈ (IL)⊥ ⊕ L⊥ and f + Ig ∈ L ⊕ IL, where we used the symmetry of µ and that
σ(f, g) = µ(f, −Ig). Using that h, k ∈ (IL)⊥ and f, g ∈ L as well as µ(f, g) = µ(If, Ig),
this reduces to

⟨S(f + Ig), h + Ik⟩ = µ(h, f ) − µ(k, g) + iµ(h, g) + iµ(k, f )

= µ(h − Ik, f + Ig) + iµ(h − Ik, g − If ) (2.9)
= µ(h − Ik, f + Ig) + iσ(h − Ik, Ig + f ) = ⟨h − Ik, f + Ig⟩ .

On the other hand, for an antilinear operator we have

D E∗ D E
⟨Sf, g⟩ = f, S † g = S † g, f (2.10)

for all f, g ∈ H, and thus can identify the action

S † (h + Ik) = h − Ik , (2.11)

recalling that h + Ik ∈ (IL)⊥ ⊕ L⊥ .

To derive the relation between ∆ and P , we first would like to show that [61]

P (1 − ∆) = 1 + S . (2.12)

Using that for f + Ig ∈ L ⊕ IL we have

[P (1 − ∆) − 1 − S](f + Ig) = P (f + Ig) − P S † (f − Ig) − 2f

(2.13)
= −P (1 + S † )(f − Ig) ,

we compute for h ∈ H that

⟨h, [P (1 − ∆) − 1 − S](f + Ig)⟩

= − h, P (1 + S † )(f − Ig)
D E

(2.14)
= −µ h, P (1 + S † )(f − Ig) − iµ Ih, P (1 + S † )(f − Ig)
   

= µ IP Ih, (1 + S † )(f − Ig) − iµ IP h, (1 + S † )(f − Ig) ,

   

where we used that µ(f, P g) = µ(P If, Ig) and that µ(If, Ig) = µ(f, g). Using furthermore
that for the real part of the scalar product we have
   
µ f, S † g = µ S † g, f = µ(Sf, g) = µ(g, Sf ) (2.15)

–6–
and that SIh = −ISh for all h ∈ H, it follows that

⟨h, [P (1 − ∆) − 1 − S](f + Ig)⟩

= µ((1 + S)IP Ih, f − Ig) − iµ((1 + S)IP h, f − Ig) (2.16)
= µ(I(1 − S)P Ih, f − Ig) − iµ(I(1 − S)P h, f − Ig) = 0 ,

where the last equality holds because P f ∈ L such that SP f = P f .

The analogous computation shows that

IP I(1 − ∆) = −1 + S , (2.17)

and combining both we obtain

1 − P + IP I = −(1 + ∆)(1 − ∆)−1 , (2.18)

from which it follows that [64]

ln ∆ = 2 arcoth(1 − P + IP I) . (2.19)

This equation needs to be understood from spectral calculus. In particular, since ∆ is a

positive self-adjoint operator the relation (2.18) shows that the spectrum of 1 − P + IP I
is contained in (−∞, −1] ∪ [1, ∞), such that the arcoth is well-defined.

2.2 An alternative formula for the modular Hamiltonian

While the formula (2.19) is short, it is difficult to use in practice, and we would like to
derive a different one. For complex numbers, we have the integral representation
Z ∞
z
arcoth z = dt , (2.20)
1 t2 z 2 −1
which we would like to extend to self-adjoint operators, following [82]. Let thus A be a
self-adjoint operator on H, whence by the spectral theorem there exists a unique spectral
measure EA on the Borel sigma-algebra of R, supported on the spectrum σ(A), such that
Z
A= λ dEA (λ) . (2.21)
σ(A)

If σ(A) ⊆ (−∞, −1] ∪ [1, ∞), spectral calculus defines

Z
arcoth(A) = arcoth(λ) dEA (λ)
σ(A)
Z ∞ (2.22)
λ
Z
= 2 2
dt dEA (λ) .
σ(A) 1 t λ −1

For x ∈ D(A) and 0 < ϵ < 12 , consider the vector

2 1
   
y = Pϵ x , Pϵ = EA − , −1 − ϵ + EA 1 + ϵ, . (2.23)
ϵ ϵ

–7–
For all such ϵ, y lies in the domain of A, and we have
Z ∞
λ
Z
arcoth(A)y = dt dEA (λ)y
σ(A) 1 t2 λ2 − 1
Z ∞Z
λ
= 2 2
dEA (λ)y dt (2.24)
1 σ(A) t λ −1
Z ∞ −1
A t 2 A2 − 1

= y dt ,
1

where we could interchange the integrals for ϵ > 0 by Fubini’s theorem for Bochner in-
tegrals. Since the set of all such y (for all 0 < ϵ < 21 ) is a core for A, the conclusion
follows.
Applied to our result (2.19), we thus obtain
Z ∞ i−1
(1 − P + IP I) t2 (1 − P + IP I)2 − 1
h
ln ∆ = 2 dt
1
Z ∞ i−1 (2.25)
1 + t2 P IP IP + t2 (1 − P )I(1 − P )I(1 − P )
h
= 2I (I − IP − P I) dt .
1

We see that A2 = P IP IP + (1 − P )I(1 − P )I(1 − P ) is block diagonal, i.e., that it leaves

invariant the closed subspaces im P = L and im(1 − P ) = (IL)⊥ which together sum to
the full Hilbert space. It follows that also the resolvent is block diagonal, such that
i−1
1 + t2 P IP IP + t2 (1 − P )I(1 − P )I(1 − P )
h

  −1
=P 1 + t IP I 2
P (2.26)
L
 −1
+ (1 − P ) 1 + t2 I(1 − P )I (1 − P ) ,

(IL)⊥

where the inverse is computed on each subspace. Therefore, we obtain

Z ∞   −1
I ln ∆ = 2 P IP 1 + t2 IP I P dt
1 L
Z ∞  −1 (2.27)
(1 − P )I(1 − P ) 1 + t I(1 − P )I (1 − P ) dt .

2
−2
1 (IL)⊥

We see that I ln ∆ leaves not only the subspace L invariant (which is known), but also the
real orthogonal complement (IL)⊥ .

2.3 Relation to the two-point function

To connect with other formulas in the literature which only involve functions supported in
the region R ⊂ Σ, we need to express the projection of the complex structure P IP using
the two-point function. For this, we need to consider a different Hilbert space, namely
Ĥ = L2C (R) ⊕ L2C (R) with the L2 scalar product
Z
(f, g) = (f1∗ g1 + f2∗ g2 ) dx , (2.28)
R

–8–
which we see as the completion of SR,C = C0,C ∞ (R) ⊕ C ∞ (R) with respect to this scalar
0,C
product. Both the symplectic form σ (2.1) and the bilinear form µ can be straightforwardly
∞ (R) ⊕ C ∞ (R). Namely, we write
extended to real functions f, g ∈ SR,R = C0,R 0,R

1
σ(f, g) = (f, ϵg) (2.29)
2
!
0 1
with the constant matrix ϵ = fulfilling ϵ2 = −1 and
−1 0
1 1
µ(f, g) = σ(P f, IP g) = (f, P ϵIP g) = (f, ϵP IP g) . (2.30)
2 2
Since I is an antilocal operator and thus in particular does not preserve the support of
functions, we had to introduce the projection P , which is the multiplication with the
characteristic function of the region R (and thus clearly commutes with ϵ). This is also the
reason why we consider the usual complex-valued L2 space instead of simply restricting
I to R: the restriction does not fulfill anymore the requirements for a complex structure,
in particular (P IP )2 ̸= −1. Whereas P is an unbounded operator on the original Hilbert
space H, it is the identity on Ĥ and thus trivially bounded. The price we have to pay for
this simplification is that the projection of the complex structure P IP is unbounded on Ĥ,
while I was bounded on H. However, since µ is symmetric and positive definite, it follows
that (f, ϵP IP g) = (ϵP IP f, g) for real f and g. We can thus extend ϵP IP to a symmetric,
positive and densely defined operator on SR,C , and we obtain a self-adjoint operator on
Ĥ by taking the Friedrichs extension (denoted by the same symbol). On the other hand,
the polarization identity shows that ϵ cannot be extended as a linear operator to SR,C , but
only as an antilinear operator. This is however inconvenient for later use, and so we instead
consider iϵ as a linear, symmetric and bounded operator on SR,C . One easily verifies that
this entails
(f + ig, iϵ(f + ig)) = −2σ(f, g) for f, g ∈ SR,R , (2.31)
from which arbitrary matrix elements can be computed using the polarization identity.
Moreover, it is clear that while iϵ is symmetric, its matrix elements do not have a defi-
nite sign. (In fact, it is easy to see that Ĥ decomposes into a direct sum of eigenspaces
corresponding to the eigenvalues ±1 of iϵ.)
The original scalar product ⟨·, ·⟩ on H, restricted to f, g ∈ SR,R , then defines an oper-
ator G on SR,R via
1 1
(f, Gg) = ⟨f, g⟩ = (f, ϵP IP g) + (f, iϵg) . (2.32)
2 2
Using the extensions of ϵP IP and iϵ as above, we extend G to a symmetric, densely defined
and semibounded operator on SR,C , and denote its Friedrichs extension to a self-adjoint
operator on Ĥ by the same symbol.2 While we know that ϵP IP is positive, iϵ can have
2
Since the form associated to the bounded operator iϵ is closed, and the sum of two closed semibounded
forms is closed, the Friedrichs extension of G coincides with the sum of the Friedrichs extension of ϵP IP
and iϵ. G is thus a self-adjoint operator
on Ĥ, whose domain of definition is the domain of the Friedrichs
1
extension of ϵP IP , namely D (ϵP IP ) 2 .

–9–
either sign, and so it is not a priori clear that G is actually a positive operator. Let us show
this: by using the relation (2.31) and the inequality (2.2), we compute that

(f + ig, iϵ(f + ig))2 = 4σ(f, g)2 ≤ 4µ(f, f )µ(g, g)

= [µ(f, f ) + µ(g, g)]2 − [µ(f, f ) − µ(g, g)]2 (2.33)
2 2
≤ [µ(f, f ) + µ(g, g)] = (f + ig, ϵP IP (f + ig))

for f, g ∈ SR,R , such that

[(h, ϵP IP h) − (h, iϵh)][(h, ϵP IP h) + (h, iϵh)] ≥ 0 (2.34)

1

for all h ∈ D (ϵP IP ) 2 . It follows that G ≥ 0, and we obtain an isometry
√ √
U : H → Ĥ , f + Ig 7→ Gf + i G g for f, g ∈ L , (2.35)

where we used that H = L ⊕ IL such that U is uniquely defined. Rewriting the rela-
tion (2.32) as relation between operators on Ĥ according to

2G = ϵP IP + iϵ ⇔ P IP = 2i(iϵ)−1 G − i1 , (2.36)

we may replace the projection of the complex structure P IP in our result (2.27) to obtain
2(iϵ)−1 G − 1
  Z ∞
U I ln ∆ U −1 = iK = 2i dt (2.37)
R
1 1 − t2 [2(iϵ)−1 G − 1]2
on a suitable dense domain in Ĥ, which defines the modular Hamiltonian K on Ĥ. We
may interpret both G and K as convolution operators acting on functions with support in
R, and then the formula (2.37) relates their integral kernels. The analogous computation
establishes a similar formula for the restriction of the modular Hamiltonian to the com-
plement region Σ \ R, which only depends on the two-point function G restricted to the
complement. We omit the straightforward details, which essentially only differ in an overall
minus sign coming from the result (2.27).
To obtain a more concrete expression for the result (2.37), we note that of the initial
data f = (f1 , f2 ) the first component f1 is the initial data for the field ϕ itself, while the
second component f2 is the initial data for its normal derivative, the conjugate momen-
tum π. The integral kernel of G is thus nothing else but the two-point function, which
decomposes as
X 2i 1
!
G= . (2.38)
− 2i 1 Π
The off-diagonal entries are one-half of the equal-time commutator between ϕ and π, while
the diagonal entries are the equal-time correlation functions of ϕ with itself (X) or π with
itself (Π), restricted to R, which are both symmetric. Since the symmetric part of G is
equal to the positive operator ϵP IP , it follows that also X and Π are positive operators,
and in particular invertible on a suitable domain dense in Ĥ. From the relation (2.36) we
further obtain !
0 −2Π
P IP = 2i(iϵ) G − i1 =
−1
, (2.39)
2X 0

– 10 –
and thus the result (2.37) reads
!Z  !2 −1
0 −Π ∞ 0 −Π
iK = 4 1 + 4t2  dt
X 0 1 X 0
 −1 
1−
1 R ∞ 1 1 1
0 −4Π 2
1 4t2 Π 2 XΠ 2 dt Π 
2
=

−1
1 − 4t2 Π XΠ
1 R ∞ 1 1
− 12

4XΠ 2
1
2 2 dt Π 0 (2.40)
!
0 2M
=
−2N 0

with
1 1
M = Π 2 B −1 arcoth(2B)Π 2 , (2.41a)
− 12 − 12
N =Π B arcoth(2B)Π . (2.41b)
q
1 1
Here we defined B = Π 2 XΠ 2 , and employed the integral representation (2.20) of the
1 1
arcoth, extended to self-adjoint operators. For this, we used that since Π 2 XΠ 2 is positive,
symmetric and densely defined we can again take its Friedrichs extension, which we denote
by the same symbol. To ensure that Eqs. (2.41) are valid definitions, we need to verify
that B ≥ 12 , which follows from positivity of the state [31]. To show this, we write the
decomposition (2.38) as

1 2i Π−1 1
!   ! !
XΠ − 1
4 Π−1 0 0
G= , (2.42)
0 1 0 Π − 2i Π−1 1
 1 1

where the inverse Π−1 is well-defined (at least) on C0∞ (R), such that for f = Π 2 f1 , 2i Π− 2 f1
we obtain
1
   
1 1
(f, Gf ) = f1 , Π 2 XΠ 2 − f1 ≥ 0 . (2.43)
4
1 1
Therefore, Π 2 XΠ 2 is densely defined and lower semibounded with bound 41 , and its
Friedrichs extension satisfies the same bound, such that B ≥ 12 .
The result (2.40) together with the explicit expressions (2.41) for M and N is very
similar to the result of Casini and Huerta [31], who give the modular Hamiltonian as a
second-quantized operator on Fock space. (Similar but less explicit formulas were also ob-
tained by Peschel [28] and Araki [83].) However, Casini and Huerta work with a discretized
theory, where X and Π are actually bounded operators. In this case, it is possible to ex-
press the operators M and N (2.41) in a different way. Namely, XΠ as the product of two
positive operators is weakly positive in the sense of Wigner [84], which entails that it is
1 1
similar to a positive operator. In fact, we see immediately that Π 2 XΠ 2 is similar to XΠ
1
with intertwining operator Π− 2 , hence both have the same spectrum. It follows that XΠ
has a spectral decomposition of the form [84]
Z M
1 1
 
XΠ = λ d Π− 2 Eλ Π 2 , (2.44)
0

– 11 –
 1 1
where Eλ are the self-adjoint projections in the spectral decomposition of Π 2 XΠ 2 and
1 1
M = Π 2 XΠ 2 . This allows us to define functions of XΠ by spectral calculus, and we
obtain
1 2C + 1
 
M = ΠC −1 arcoth(2C) = Π ln , (2.45a)
2C 2C − 1
1 2C + 1
 
N = C −1 arcoth(2C)X = ln X (2.45b)
2C 2C − 1
√
with C = XΠ. These now exactly coincide with the result of Casini and Huerta [31].

2.4 General CCR algebras

Our results can be easily generalized to more general commutation relations, as long as the
theory remains free. Relative entropy for general CCR algebras was studied in Refs. [39, 62],
using both the standard subspace formalism and the generalization of formulas (2.40)
and (2.45). We also refer to [27], where the modular Hamiltonian of massless generalized
free fields in two dimensions has been determined both for thermal states and the vacuum
in various regions.
A generalized free field is characterized by a symplectic form σ different from (2.1).
However, in the standard subspace formulation the concrete form of σ is irrelevant, such
that the result (2.27) for I ln ∆ still holds. Moreover, σ can still be written in the form (2.29),
where however now ϵ is a (possibly unbounded) densely defined and invertible antisym-
metric operator on SR,R (since σ is non-degenerate by assumption). We assume that ϵ
and P commute, and again extend iϵ to a linear, symmetric and densely defined operator
on Ĥ. Denoting its closure by the same symbol, its polar decomposition reads iϵ = J|iϵ|
with J † = J and J 2 = 1 since ϵ is invertible. The relation (2.36) between the two-point
function G, iϵ and the projection of the complex structure P IP then still holds, and thus
our result (2.37). Let us define
1 1 1 1
E = −i|iϵ| 2 P IP |iϵ|− 2 = J|iϵ|− 2 ϵP IP |iϵ|− 2
(2.46)
= J|iϵ|− 2 (2G − iϵ)|iϵ|− 2 = 2J|iϵ|− 2 G|iϵ|− 2 − 1 ,
1 1 1 1

which is a symmetric and densely defined operator in Ĥ, and where we used the rela-
tion (2.36). By uniqueness of the polar decomposition and using that ϵP IP ≥ 0, we see
1 1
that the polar decomposition of E reads E = J|E| with |E| = |iϵ|− 2 ϵP IP |iϵ|− 2 .
In terms of E, the modular Hamiltonian K (2.37) on Ĥ reads
Z ∞
1 E 1
K = 2|iϵ|− 2 dt |iϵ| 2
1 1−t E
2 2
Z ∞
1 |E| 1
= 2|iϵ|− 2 J dt |iϵ| 2 (2.47)
1 1 − t2 |E| 2
1 1
= −2|iϵ|− 2 J arcoth(|E|) |iϵ| 2 ,

where we again employed the integral representation (2.20) of the arcoth. To ensure that
this is a sensible result, we must have |E| ≥ 1, which again follows from positivity of G.

– 12 –
Namely, the result (2.34) still holds, and using that iϵ = J|iϵ| with J 2 = 1 we conclude
that
ϵP IP − |iϵ| ≥ 0 (2.48)
on the intersection of their domains of definition, and thus

|E| = |iϵ|− 2 (ϵP IP − |iϵ|)|iϵ|− 2 + 1 ≥ 1 .

1 1
(2.49)

The result (2.47) is again very similar to the result of Arias et al. [39], who give the
modular Hamiltonian as a second-quantized operator on Fock space. However, they also
work with a discretized theory, where all operators are bounded. We then define
1 1 1 1
V = |iϵ|− 2 E|iϵ| 2 = (iϵ)−1 G − , (2.50)
2 2
which is weakly positive in the sense of Wigner, and hence has the same spectrum as 21 E.
As in the scalar field case, V thus has a spectral decomposition of the form [84]
Z M
1  1 1

V = λ d |iϵ|− 2 Eλ |iϵ| 2 , (2.51)
2 0

where Eλ are the self-adjoint projections in the spectral decomposition of E and M =

∥E∥. This allows us to define functions of V by spectral calculus, and we can write our
result (2.47) as
2V − 1
 
K = −2 arcoth(2V ) = ln , (2.52)
2V + 1
which coincides with the result of Arias et al. [39], except for a factor of |iϵ|−1 multiplying
the expression (2.52) on the right.

2.5 KMS condition

Lastly, we would like to show that the expression (2.37) for the modular Hamiltonian also
follows from the KMS condition. This condition reads as follows: Given two elements a, b in
the von Neumann algebra, the function F (t) = ω(σt (a)b) admits an analytic continuation
to the strip −1 < ℑm t < 0 with a continuous extension to the boundary ℑm t = −1, and
its boundary value there is given by F (t − i) = ω(bσt (a)). Here, σt (a) = ∆it a∆−it is the
modular flow generated by the modular Hamiltonian ln ∆, and ω is the state on the algebra
that enters the construction. In our case, the von Neumann algebra is the algebra of Weyl
operators with support inside the region R ⊂ Σ, and the state is the quasifree state on
Fock space whose restriction to the one-particle level has the two-point function G.
Since the modular Hamiltonian on Fock space is the second quantization of the one-
particle modular Hamiltonian [80, 81], for a Weyl operator W (f ) in the Fock representation
we have the modular flow ∆it W (f )∆−it = W (f t ), where f t describes the flow at the one-
particle level. Concretely, for a vector f ∈ SR,R with support inside the region R ⊂ Σ, we
have
2 Z
X
fat (x) = Lab (t, x, y)fb (y) dy , (2.53)
b=1 R

– 13 –
where L is an integral kernel that is related to the integral kernel of K (2.37) according to

∂
 
iKab (x, y) = Lab (t, x, y) . (2.54)
∂t t=0

From the group property σt (σs (a)) = σs+t (a) of the modular flow, we also obtain the
condition
2 Z
X
Lab (t + s, x, y) = Lac (t, x, z)Lcb (s, z, y) dz , (2.55)
c=1 R

which together with the relation (2.54) uniquely determines L. If we take a = W (f ) and
b = W (g) and use that
1 1 1
ω(W (f )W (g)) = e−iσ(f,g) e− 2 µ(f +g,f +g) = e− 2 G(f,f )− 2 G(g,g)−G(f,g) , (2.56)

the KMS condition results in

G(f t−i , f t−i ) + 2G(f t−i , g) = G(f t , f t ) + 2G(g, f t ) . (2.57)

Consider the terms linear in g, for which this reduces to

2
X ZZZ
Gab (x, y)Lac (t − i, x, z)fc (z)gb (y) dx dy dz
a,b,c=1 R

2 ZZZ (2.58)
X
= Gab (x, y)ga (x)Lbc (t, y, z)fc (z) dx dy dz ,
a,b,c=1 R

and since f and g are arbitrary we obtain the relation

2 Z
X 2 Z
X
Gba (y, x)Lbc (t − i, y, z) dy = Gab (x, y)Lbc (t, y, z) dy (2.59)
b=1 R b=1 R

that connects the integral kernel G of the two-point function restricted to R with L.
Interpreting L as a convolution operator on Ĥ, the relation (2.59) reads GT L(t − i) =
GL(t), and the condition (2.55) translates into L(s + t) = L(t)L(s). It is easy to see that
the unique solution of this system reads
h i
L(t) = eitK with K = − ln G−1 GT . (2.60)

From the decomposition (2.38), we also obtain

Gba (y, x) = Gab (x, y) − iϵab δ(x, y) (2.61)

where we used that X and P are symmetric operators, or in operator notation GT = G−iϵ,
such that the result (2.60) for K can be written as
 i−1 
1 − G−1 iϵ = − ln 1 − (iϵ)−1 G
h i h
K = − ln . (2.62)

– 14 –
On the other hand, changing the integration variable in (2.37) as t → 1/t and performing
a partial fraction decomposition (using the second resolvent identity), we obtain

2(iϵ)−1 G − 1
Z 1
K=2 2 dt
0 t2 − [2(iϵ)−1 G − 1]
Z 1
1 1

= − dt (2.63)
0 t − 2(iϵ)−1 G + 1 t + 2(iϵ)−1 G − 1
Z 1
1 1
 
=− dt = − ln 1 − ,
0 t + (iϵ) G − 1
−1 (iϵ)−1 G

where we performed the additional variable changes t → 1 − 2t in the first integral and
t → 2t − 1 in the second one. We have thus full agreement with the relation (2.62) derived
from the KMS condition. The connection between the KMS condition and the modular
flow was previously derived for chiral conformal fields, in particular the U(1) current [42],
but we see that it also holds for non-conformal free scalar fields.

Acknowledgments

This work has been funded by the Deutsche Forschungsgemeinschaft (DFG, German Re-
search Foundation) — project no. 396692871 within the Emmy Noether grant CA1850/1-1.
It is a pleasure to thank Horacio Casini, Igor Khavkine and Christoph Minz for discussions,
and Leonardo Sangaletti for discussions and a critical reading of the manuscript.

References

[1] M. Tomita, On canonical forms of von Neumann algebras (in Japanese), in Fifth Functional
Analysis Symposium, Tôhoku Univ., Sendai, (Sendai, Japan), pp. 101–102, Tôhoku Univ.,
Math. Inst., 1967.
[2] M. Takesaki, Tomita’s Theory of Modular Hilbert Algebras and its Applications, vol. 128 of
Lecture Notes in Mathematics, Springer-Verlag, Berlin, Heidelberg, Germany (1970),
10.1007/BFb0065832.
[3] H. Araki, Inequalities in von Neumann algebras, Rencontr. phys.-math. de Strasbourg –
RCP25 22 (1975) 1.
[4] H. Araki, Relative Entropy of States of von Neumann Algebras, Publ. RIMS, Kyoto Univ. 11
(1976) 809.
[5] A. Uhlmann, Relative Entropy and the Wigner-Yanase-Dyson-Lieb Concavity in an
Interpolation Theory, Commun. Math. Phys. 54 (1977) 21.
[6] J.J. Bisognano and E.H. Wichmann, On the duality condition for a Hermitian scalar field, J.
Math. Phys. 16 (1975) 985.
[7] J.J. Bisognano and E.H. Wichmann, On the duality condition for quantum fields, J. Math.
Phys. 17 (1976) 303.
[8] D. Buchholz, On the structure of local quantum fields with non-trivial interaction, in
Proceedings of the International Conference on Operator Algebras, Ideals, and Their
Applications in Theoretical Physics: Leipzig, September 12.-20., 1977, H. Baumgärtel,

– 15 –
 G. Laßner, A. Pietsch and A. Uhlmann, eds., vol. 67 of Teubner-Texte zur Mathematik,
(Leipzig, GDR), pp. 146–153, BSB B. G. Teubner-Verlag, 1978.
[9] P.D. Hislop and R. Longo, Modular structure of the local algebras associated with the free
massless scalar field theory, Commun. Math. Phys. 84 (1982) 71.
[10] P.D. Hislop, Conformal Covariance, Modular Structure, and Duality for Local Algebras in
Free Massless Quantum Field Theories, Annals Phys. 185 (1988) 193.
[11] J. Bros, H. Epstein and U. Moschella, Analyticity Properties and Themal Effects for General
Quantum Field Theory on de Sitter Space-Time, Commun. Math. Phys. 196 (1998) 535
[gr-qc/9801099].
[12] O. Dreyer, Das Prinzip der geometrischen modularen Wirkung im de Sitter-Raum, Master’s
thesis, Universität Hamburg, 1996.
[13] M. Florig, Geometric modular action, Ph.D. thesis, University of Florida, 1999.
[14] H.J. Borchers and D. Buchholz, Global properties of vacuum states in de Sitter space, Ann.
Inst. H. Poincaré Phys. théor. 70 (1999) 23 [gr-qc/9803036].
[15] D. Buchholz, O. Dreyer, M. Florig and S.J. Summers, Geometric modular action and
space-time symmetry groups, Rev. Math. Phys. 12 (2000) 475 [math-ph/9805026].
[16] R. Figari, R. Höegh-Krohn and C.R. Nappi, Interacting Relativistic Boson Fields in the De
Sitter Universe with Two Space-Time Dimensions, Commun. Math. Phys. 44 (1975) 265.
[17] G.W. Gibbons and S.W. Hawking, Cosmological event horizons, thermodynamics, and
particle creation, Phys. Rev. D 15 (1977) 2738.
[18] V. Chandrasekaran, R. Longo, G. Penington and E. Witten, An algebra of observables for de
Sitter space, JHEP 02 (2023) 082 [2206.10780].
[19] D. Buchholz, M. Florig and S.J. Summers, Hawking-Unruh temperature and Einstein
causality in anti-de Sitter space-time, Class. Quant. Grav. 17 (2000) L31 [hep-th/9905178].
[20] D. Buchholz, J. Mund and S.J. Summers, Transplantation of local nets and geometric
modular action on Robertson-Walker space-times, Fields Inst. Commun. 30 (2001) 65
[hep-th/0011237].
[21] D. Buchholz, J. Mund and S.J. Summers, Covariant and quasi-covariant quantum dynamics
in Robertson-Walker space-times, Class. Quant. Grav. 19 (2002) 6417 [hep-th/0207057].
[22] G.L. Sewell, Quantum fields on manifolds: PCT and gravitationally induced thermal states,
Annals Phys. 141 (1982) 201.
[23] B.S. Kay, Purification of KMS States, Helv. Phys. Acta 58 (1985) 1030.
[24] B.S. Kay, The Double Wedge Algebra for Quantum Fields on Schwarzschild and Minkowski
Spacetimes, Commun. Math. Phys. 100 (1985) 57.
[25] B.S. Kay and R.M. Wald, Theorems on the uniqueness and thermal properties of stationary,
nonsingular, quasifree states on spacetimes with a bifurcate killing horizon, Phys. Rept. 207
(1991) 49.
[26] M.B. Fröb, Modular Hamiltonian for de Sitter diamonds, JHEP 12 (2023) 074 [2308.14797].
[27] H.J. Borchers and J. Yngvason, Modular groups of quantum fields in thermal states, J. Math.
Phys. 40 (1999) 601 [math-ph/9805013].

– 16 –
[28] I. Peschel, Calculation of reduced density matrices from correlation functions, J. Phys. A 36
(2003) L205 [cond-mat/0212631].
[29] H. Casini and M. Huerta, Analytic results on the geometric entropy for free fields, J. Stat.
Mech. 0801 (2008) P01012 [0707.1300].
[30] H. Casini and M. Huerta, Reduced density matrix and internal dynamics for multicomponent
regions, Class. Quant. Grav. 26 (2009) 185005 [0903.5284].
[31] H. Casini and M. Huerta, Entanglement entropy in free quantum field theory, J. Phys. A 42
(2009) 504007 [0905.2562].
[32] R. Longo, P. Martinetti and K.-H. Rehren, Geometric modular action for disjoint intervals
and boundary conformal field theory, Rev. Math. Phys. 22 (2010) 331 [0912.1106].
[33] H. Casini and M. Huerta, Entanglement entropy for the n-sphere, Phys. Lett. B 694 (2011)
167 [1007.1813].
[34] K.-H. Rehren and G. Tedesco, Multilocal fermionization, Lett. Math. Phys. 103 (2013) 19
[1205.0324].
[35] J. Cardy and E. Tonni, Entanglement Hamiltonians in two-dimensional conformal field
theory, J. Stat. Mech. 1612 (2016) 123103 [1608.01283].
[36] R. Arias, D. Blanco, H. Casini and M. Huerta, Local temperatures and local terms in modular
Hamiltonians, Phys. Rev. D 95 (2017) 065005 [1611.08517].
[37] H. Casini, E. Teste and G. Torroba, Modular Hamiltonians on the null plane and the Markov
property of the vacuum state, J. Phys. A 50 (2017) 364001 [1703.10656].
[38] E. Tonni, J. Rodríguez-Laguna and G. Sierra, Entanglement hamiltonian and entanglement
contour in inhomogeneous 1D critical systems, J. Stat. Mech. 1804 (2018) 043105
[1712.03557].
[39] R.E. Arias, H. Casini, M. Huerta and D. Pontello, Entropy and modular Hamiltonian for a
free chiral scalar in two intervals, Phys. Rev. D 98 (2018) 125008 [1809.00026].
[40] R. Abt and J. Erdmenger, Properties of Modular Hamiltonians on Entanglement Plateaux,
JHEP 11 (2018) 002 [1809.03516].
[41] V. Eisler, E. Tonni and I. Peschel, On the continuum limit of the entanglement Hamiltonian,
J. Stat. Mech. 1907 (2019) 073101 [1902.04474].
[42] S. Hollands, On the Modular Operator of Multi-component Regions in Chiral CFT, Commun.
Math. Phys. 384 (2021) 785 [1904.08201].
[43] D. Blanco and G. Pérez-Nadal, Modular Hamiltonian of a chiral fermion on the torus, Phys.
Rev. D 100 (2019) 025003 [1905.05210].
[44] P. Fries and I.A. Reyes, Entanglement Spectrum of Chiral Fermions on the Torus, Phys. Rev.
Lett. 123 (2019) 211603 [1905.05768].
[45] P. Fries and I.A. Reyes, Entanglement and relative entropy of a chiral fermion on the torus,
Phys. Rev. D 100 (2019) 105015 [1906.02207].
[46] G. Di Giulio and E. Tonni, On entanglement Hamiltonians of an interval in massless
harmonic chains, J. Stat. Mech. 2003 (2020) 033102 [1911.07188].
[47] V. Eisler, G. Di Giulio, E. Tonni and I. Peschel, Entanglement Hamiltonians for non-critical
quantum chains, J. Stat. Mech. 2010 (2020) 103102 [2007.01804].

– 17 –
[48] J. Erdmenger, P. Fries, I.A. Reyes and C.P. Simon, Resolving modular flow: a toolkit for free
fermions, JHEP 12 (2020) 126 [2008.07532].
[49] M. Mintchev and E. Tonni, Modular Hamiltonians for the massless Dirac field in the
presence of a boundary, JHEP 03 (2021) 204 [2012.00703].
[50] M. Mintchev and E. Tonni, Modular Hamiltonians for the massless Dirac field in the
presence of a defect, JHEP 03 (2021) 205 [2012.01366].
[51] T. Grava, A.P. Kels and E. Tonni, Entanglement of Two Disjoint Intervals in Conformal
Field Theory and the 2D Coulomb Gas on a Lattice, Phys. Rev. Lett. 127 (2021) 141605
[2104.06994].
[52] N. Javerzat and E. Tonni, On the continuum limit of the entanglement Hamiltonian of a
sphere for the free massless scalar field, JHEP 02 (2022) 086 [2111.05154].
[53] V. Eisler, E. Tonni and I. Peschel, Local and non-local properties of the entanglement
Hamiltonian for two disjoint intervals, J. Stat. Mech. 2208 (2022) 083101 [2204.03966].
[54] M. Mintchev and E. Tonni, Modular conjugations in 2D conformal field theory and
holographic bit threads, JHEP 12 (2022) 149 [2209.03242].
[55] N. Abate, D. Blanco, M. Koifman and G. Pérez-Nadal, Modular conjugation for
multicomponent regions, Phys. Rev. D 107 (2023) 045015 [2209.10711].
[56] F. Rottoli, S. Murciano, E. Tonni and P. Calabrese, Entanglement and negativity
Hamiltonians for the massless Dirac field on the half line, J. Stat. Mech. 2301 (2023) 013103
[2210.12109].
[57] G. Di Giulio and J. Erdmenger, Symmetry-resolved modular correlation functions in free
fermionic theories, JHEP 07 (2023) 058 [2305.02343].
[58] M. Huerta and G. van der Velde, Modular Hamiltonian of the scalar in the semi infinite line:
dimensional reduction for spherically symmetric regions, JHEP 06 (2023) 097 [2301.00294].
[59] M. Huerta and G. van der Velde, Modular Hamiltonian in the semi infinite line, Part II:
dimensional reduction of Dirac fermions in spherically symmetric regions, 2307.08755.
[60] M. Rieffel and A. Van Daele, A bounded operator approach to Tomita-Takesaki theory, Pac.
J. Math. 69 (1977) 187.
[61] F. Ciolli, R. Longo and G. Ruzzi, The Information in a Wave, Commun. Math. Phys. 379
(2019) 979 [1906.01707].
[62] H. Bostelmann, D. Cadamuro and S. Del Vecchio, Relative Entropy of Coherent States on
General CCR Algebras, Commun. Math. Phys. 389 (2022) 661.
[63] R. Longo, Modular Structure of the Weyl Algebra, Commun. Math. Phys. 392 (2022) 145
[2111.11266]. Correction: Commun. Math. Phys. 401 (2023) 1061.
[64] H. Bostelmann, D. Cadamuro and C. Minz, On the Mass Dependence of the Modular
Operator for a Double Cone, Ann. H. Poincaré 24 (2023) 3031 [2209.04681].
[65] D. Cadamuro, M.B. Fröb and C. Minz, Modular Hamiltonian for fermions of small mass,
Ann. H. Poincaré (2024) [2312.04629].
[66] D. Blanco, A. Garbarz and G. Pérez-Nadal, Entanglement of a chiral fermion on the torus,
JHEP 09 (2019) 076 [1906.07057].

– 18 –
[67] D. Cadamuro, M.B. Fröb and G. Pérez-Nadal, Modular Hamiltonian and modular flow of
massless fermions on a cylinder, 2406.19360.
[68] D. Katsinis and G. Pastras, Entanglement in (1 + 1)-dimensional Free Scalar Field Theory:
Tiptoeing between Continuum and Discrete, 2406.11031.
[69] D. Petz, An Invitation to the Algebra of Canonical Commutation Relations, vol. A2 of
Leuven Notes in Mathematical and Theoretical Physics, Leuven University Press, Leuven,
Belgium (1990).
[70] C. Gérard and M. Wrochna, The massive Feynman propagator on asymptotically Minkowski
spacetimes II, Int. Math. Res. Not. 2020 (2020) 6856 [1806.05076].
[71] A. Much and R. Oeckl, Complex structures for Klein–Gordon theory on globally hyperbolic
spacetimes, Class. Quant. Grav. 39 (2022) 025015 [1812.00926].
[72] Y. Sanchez Sanchez and E. Schrohe, Adiabatic Ground States in Non-smooth Spacetimes,
Ann. H. Poincaré 24 (2023) 2929 [2207.01429].
[73] O. Islam and A. Strohmaier, On microlocalisation and the construction of Feynman
Propagators for normally hyperbolic operators, Commun. Anal. Geom. 32 (2024) 1811
[2012.09767].
[74] J. Dereziński and C. Gaß, Propagators in curved spacetimes from operator theory,
2409.03279.
[75] J. Dixmier, Position relative de deux variétés linéaires fermées dans un espace de Hilbert,
Rev. Sci. 86 (1948) 387.
[76] P.R. Halmos, Two subspaces, Trans. Amer. Math. Soc. 144 (1969) 381.
[77] I.E. Segal and R.W. Goodman, Anti-Locality of Certain Lorentz-Invariant Operators, J.
Math. Mech. 14 (1965) 629.
[78] K. Masuda, Anti-Locality of the One-half Power of Elliptic Differential Operators, Publ. Res.
Inst. Math. Sci. Kyoto 8 (1972) 207.
[79] M. Murata, Anti-locality of certain functions of the Laplace operator, J. Math. Soc. Japan 25
(1973) 556.
[80] F. Figliolini and D. Guido, The Tomita operator for the free scalar field, Ann. I.H.P. Phys.
théor. 51 (1989) 419.
[81] F. Figliolini and D. Guido, On the type of second quantization factors, J. Operat. Theor. 31
(1994) 229.
[82] K. Schmüdgen, Unbounded Self-adjoint Operators on Hilbert Space, vol. 265 of Graduate
Texts in Mathematics, Springer-Verlag, Dordrecht, Netherlands (2012),
10.1007/978-94-007-4753-1.
[83] H. Araki, On Quasifree States of the Canonical Commutation Relations (II), Publ. RIMS,
Kyoto Univ. 7 (1971) 121.
[84] E.P. Wigner, On weakly positive matrices, Can. J. Math. 15 (1963) 313.

– 19 –