Clearing up the Strong CP problem

Joshua N. Benabou,1, 2 Anson Hook,3 Claudio Andrea Manzari,4 Hitoshi Murayama,2, 5, 1 and Benjamin R. Safdi1, 2, 6
1
Theoretical Physics Group, Lawrence Berkeley National Laboratory, Berkeley, CA 94720, U.S.A.
2
Leinweber Institute for Theoretical Physics, University of California, Berkeley, CA 94720, U.S.A.
3
Maryland Center for Fundamental Physics, Department of Physics,
University of Maryland, College Park, MD 20742, U.S.A.
4
School of Natural Sciences, Institute for Advanced Study, Princeton, NJ 08540, USA
5
Kavli Institute for the Physics and Mathematics of the Universe (WPI),
University of Tokyo Institutes for Advanced Study, University of Tokyo, Kashiwa 277-8583, Japan
6
Theoretical Physics Department, CERN, 1211 Geneva, Switzerland
(Dated: October 23, 2025)
The absence of a neutron electric dipole moment (EDM) constrains the quantum chromodynamics
(QCD) theta angle to be less than one part in ten billion, posing the Strong C P problem. We revisit
two classes of proposed solutions. First, we show that when P or CP is realized as a gauged discrete
arXiv:2510.18951v1 [hep-ph] 21 Oct 2025

symmetry—as can arise in quantum gravity—the vacuum necessarily preserves CP , contrary to

recent claims that discrete-symmetry solutions fail. Gauged discrete models face model-building
challenges, such as avoiding contributions to the neutron EDM after spontaneous P or CP breaking,
but in principle have no fundamental obstructions. Second, we critically examine recent arguments
that the Strong CP problem is illusory, demonstrating that a nonzero neutron EDM at finite θ̄
follows directly from well-understood QCD dynamics. Taken together, our results reinforce the
reality of the Strong CP problem and highlight gauged discrete-symmetry realizations of P or CP
as plausible solutions.

I. INTRODUCTION lem whatsoever. These authors claim that traditional

calculations suggesting a θ̄-dependent neutron EDM are
mirages arising from an incorrect order of limits in the
The Strong CP problem provides one of the most
process of taking the infinite volume limit. Conversely,
promising avenues at present for learning about physics
Refs. [35–38] argue that the Strong CP problem does ex-
beyond the Standard Model (SM). The problem is ex-
ist but can only be solved by axions or related physics,
perimentally manifest by the absence of a neutron elec-
as discrete symmetries can remove CP -violating terms
tric dipole moment (EDM), which constrains the θ̄ pa-
from the Lagrangian or Hamiltonian but are not enough
rameter of quantum chromodynamics (QCD), which we
to guarantee that the QCD vacuum preserves CP .
describe in detail below, to be |θ̄| ≲ 10−10 [1] (see [2–
5] for reviews). Given that a priori θ̄ could take val- In this work we attempt to clear up the above confu-
ues between −π and π and no additional symmetry sion about the Strong CP problem. We argue that the
of the SM is restored at the point θ̄ = 0, this repre- Strong CP problem does exist and that it may be solved
sents a fine-tuning problem. Moreover, this fine-tuning by discrete symmetries, though we argue that gauged P
problem does not appear to have an anthropic explana- /CP are the most motivated ultraviolet (UV) discrete-
tion [6, 7], which strongly motivates mechanisms that symmetry-based solutions. This is because, as we show,
set the neutron EDM to zero. Broadly speaking, there gauged P or CP in the UV enforces the vacuum to be
are two classes of solutions to the Strong CP problem CP preserving, while also ensuring no P or CP violat-
that evade fine tuning: (i) the axion, which is a new ing operators in the Lagrangian. The gauged discrete
light degree of freedom that dynamically sets the neu- symmetry must be spontaneously broken in the infrared
tron EDM to zero [8–11]; and (ii), symmetry-based solu- (IR), but all contributions to the neutron EDM are cal-
tions that use spontaneously-broken P or CP symmetry culable from the dynamics responsible for spontaneous
in the ultraviolet (UV) to keep θ̄ ∼ 0 in the infrared P / CP breaking; in particular, there is no dependence
(IR) while also generating the CP violation observed of the neutron EDM on an otherwise unknowable su-
in the Cabibbo–Kobayashi–Maskawa (CKM) matrix [12– perselection parameter for the QCD vacuum. Theories
26]. The second class of solutions benefit from the fact of gauged P or CP emerge naturally in the context of
that θ̄ evolves very mildly in the SM under the renormal- quantum gravity theories that involve compactified ex-
ization group (RG) [27]. In both scenarios the Strong CP tra dimensions. The discrete symmetries arise below the
problem is associated with new physics, though only in compactification scale as the unbroken subgroups of the
case (i) does the new physics need to be light enough to continuous, identity-connected Lorentz transformations
be observable. of the higher-dimensional theory, in some cases in com-
A number of works have recently cast doubts on the bination with unbroken internal symmetries. We illus-
Strong CP problem and its possible solutions. Refs. [28– trate this idea through a simple 5D toy model of QCD
34], for example, argue that the neutron EDM is zero that leads to gauged P in 4D, and we also illustrate how
even for θ̄ ̸= 0, such that there is no Strong CP prob- gauged P and CP may arise in string theory.
 2

The rest of this work is organized as follows. In Sec. II Lagrangian contains a single dim-4 CP -violating opera-
we review the traditional approaches to describing the tor, namely
Strong CP problem, theta vacua, and computing the neu-
tron EDM. Then, in Sec. III we show that to solve the θL
LQCD ⊃ − Ga G̃a µν , (1)
Strong CP problem, P /CP have to be symmetries of 32π 2 µν
all quantum correlation functions and not simply at the
Lagrangian level, highlighting the case of gauged P /CP . where θL is a dimensionless parameter taking values be-
We illustrate how gauged P can arise in the context of tween 0 and 2π, Gaµν is the gluon field strength tensor,
gauged Lorentz symmetry of a higher-dimensional theory and G̃a µν = 21 ϵµνρσ Gaρσ its dual. Additionally, through a
through a simple 5D example in Sec. IV, and we discuss chiral rotation of the quark fields, which is anomalous, we
how gauged P / CP arise from the basic principle il- may remove a phase in the quark mass matrix to form the
lustrated in the toy example in the context of more UV physical parameter combination θ̄ ≡ θL − arg det(Yu Yd )
complete string theory constructions. In Sec. V we show in (1), where Yu and Yd are the up and down-type Yukawa
definitively, for example using the Witten-Veneziano re- matrices, respectively.
lation, that the Strong CP problem exists, and we point The theta term in QCD affects expectation values O
out the flaws in recent works stating that there is no such in a straightforward way:
problem. Lastly, in Sec. VI we speculate that the entire
ei(θ̄+θ)ν ν [dA] exp(iS[A])O[A]
P R
discussion of e.g. Ref. [38] about discrete symmetries not ν
solving the Strong CP problem is in the swampland. We ⟨O⟩ = P i(θ̄+θ)ν R , (2)
νe ν
[dA] exp(iS[A])
argue that in theories that emerge from quantum gravity
the neutron EDM is always dynamically determined and where we sum over gauge configurations labeled by their
does not depend on an otherwise unknowable superselec- instanton number
tion sector parameter. This is in part because quantum Z Z
gravity theories are conjectured to gauge the Chern–Weil 1 1 4

µν

ν= tr (G ∧ G) = d xtr Gµν G̃
global symmetry associated with trF ∧ F [39], with F 8π 2 16π 2
(3)
the non-abelian field strength, for example through an
Z
1 4 a aµν
axion-like-particle or p-form field, and gauging this cur- = d x Gµν G̃ .
32π 2
rent yields the QCD superselection parameter unobserv-
able through gauge redundancy. Note that in writing (2) we do not include the theta term
We include a number of Appendices that provide ad- in the action S[A], as its contribution is accounted for in
ditional insight into the problem of gauged P / CP the exponential prefactors. The parameter θ appearing
and, more broadly, the Strong CP problem. In Ap- in (2) is a free and otherwise undetermined parameter
pendix. A we work through 1+1D quantum electrody- labeling the superselection sector of the theory. The θ
namics (QED), which is an exactly solvable theory that parameter is commonly associated to the theta vacua of
has a θ term. We discuss the interpretation of theta as the theory, but it may also be interpreted as the choice of
a boundary condition and as a Lagrangian parameter in boundary condition for the QCD wavefunctional under
this theory, and show how theta may emerge from an large gauge transformations (see App. A). However, θ
extra dimensional UV completion in analogy with the and θ̄ always appear in the linear combination θ̄ +θ, so in
Aharonov-Bohm effect. We also highlight that this the- the path integral formulation of quantum field theory it
ory does indeed exhibit a Strong CP problem in that θ is unphysical and unnecessary to distinguish them. Still,
contributes to physical observables analogous to the neu- in the next section we keep these terms distinct to make
tron EDM, and we show that this problem is solved by contact with the claims of Refs. [35–38], which we discuss
gauging P . Appendix B reviews Type I string theory, in the next section.
where worldsheet parity is gauged; we draw analogies We emphasize that θ is part of the definition of the the-
between this well-understood theory and the theories of ory. Distinct values of θ correspond to different theories.
gauged P / CP in 4D that we discuss in the main body of One might wonder, for example, whether the Universe
this work. Appendices C, D, E provide additional tech- could contain causally disconnected regions characterized
nical calculations that are referenced in the main work. by different values of θ; in the Standard Model, the an-
swer is no. If one extends the Standard Model by the
inclusion of e.g. an axion, then there can be an effective
II. DEFINITION OF THE PROBLEM θ̄ + θ that varies on superhorizon scales, but without a
new degree of freedom like an axion this is not possible.
The Strong CP Problem refers to the absence of a satis- The θ̄ dependence of the neutron EDM may then be
factory theoretical explanation for why P and CP appear computed through a number of techniques.1 The most
to be extremely well conserved in the strong interactions.
In order to address this question, it is useful to first iden-
tify the possible sources of P /CP violation in QCD that
1 Here, we absorb θ into the definition of θ̄.
could induce a non-vanishing neutron EDM. The QCD
 3

straightforward, textbook approach is to use chiral per- 3. A gauged discrete symmetry—this is a local identi-
turbation theory (χ-PT), which returns the estimate (see, fication of configurations related by the symmetry
e.g., [3, 4, 40]) action.

dn |χPT ≈ 3 × 10−16 θ̄ e · cm , (4) The requirement that P /CP is the first type of symme-
try, that is a classical one, is clearly not sufficient to solve
though within χ-PT it is difficult to estimate the uncer- the Strong CP Problem, since the terms involving θ and
tainties on this prediction. A more precise calculation θ̄ in the partition function do not enter into the classical
may be done using QCD sum rules [41], yielding equations of motion.2 That is, even at θ + θ̄ ̸= 0, one
would conclude that P is a classical symmetry. How-
dn |QCD sum rule ≈ 2.4(7) × 10−16 θ̄ e · cm . (5) ever, expectation values of operators would not obey P
because in this case P is anomalous. That is, the vac-
The θ̄ dependence of the neutron EDM has also recently uum expectation value of parity odd operators would not
been computed on the lattice, with the most recent esti- need to vanish in this case. Here, we define an anomalous
mate [42] symmetry to be a symmetry of the classical equations of
motion that is broken by quantum effects; this is, a sym-
dn |lattice QCD ≈ 1.48(14)(31) × 10−16 θ̄ e · cm , (6) metry of the classical but not quantum theory.
Now, let us suppose that P is a non-anomalous global
with the uncertainties statistical and systematic, respec- symmetry (option 2 above). We claim that this is only
tively. The best measurements of dn set the bound [1, 43– possible if θ + θ̄ = 0 or π. For P to be an exact symme-
45] try of the quantum theory, the partition function should
be invariant under orientation reversals of the underlying
|dn | ≲ 1.8 × 10−26 e · cm , (7) manifold. Under such an orientation reversal, the instan-
ton number ν changes sign: ν → −ν. Since the instanton
at 90% confidence, which constraints |θ̄| ≲ 10−10 . number enters into the partition function as ei(θ+θ̄)ν , for
Thus, it is clear that θ̄ ∼ 0 to very good accuracy in the theory to be invariant under orientation reversals we
nature. On the other hand, why should this be the case, need ei(θ+θ̄)ν = e−i(θ+θ̄)ν . Equivalently, e2i(θ+θ̄)ν = 1,
given that the SM violates P and CP? This is the Strong which implies that θ + θ̄ = 0 or π, since ν is an integer.
CP problem. Of course, it is not so much a problem as Let us introduce a Z2 -valued classical background field
a hint of possible new dynamics. In the next section we a that is a function of spacetime that implements lo-
address recent claims [35–38] that discrete symmetries cal orientation reversals.3 A theory with global, non-
in the UV cannot be used as a mechanism to explain to anomalous P symmetry should be well defined in the
θ̄ ∼ 0. Following this discussion, in Sec. V we address the presence of a non-trivial background for a. A configu-
even more provocative claims that dn is actually zero in ration for a may be specified by the co-dimension 1 sur-
the SM for any θ̄, in contrast with the calculation results faces across which the orientation changes (see, e.g., [46]);
quoted in (4), (5), and (6). we refer to these co-dimensional 1 surfaces as parity do-
main walls. Note that the parity domain walls, as we
define them, are not physical objects if parity is an ex-
III. SYMMETRY SOLUTIONS TO THE act symmetry; if parity is spontaneously broken these
STRONG CP PROBLEM

In the previous section we argue that θ̄ and θ, the pa- 2 Note that throughout the rest of this section we discuss P for
rameter characterizing the superselection sector of the simplicity and not CP , since we focus purely on the gluon sector
theory, always show up in the combination θ̄ + θ, so that without fermions in this section. We comment specifically on CP
we may absorb θ into the definition of θ̄. While this is later in this work, though the general statements of this section
the standard story, Refs. [35–38] claim that P or CP can apply to both P and CP symmetry solutions.
3 To be more precise, we can introduce a background gauge field
in principle be used to remove the θ̄ term from the La-
as a holonomy, i.e., a map W : C → Z2 between closed loops C
grangian, since this term is odd under P /CP , but cannot in spacetime and the Z2 , such that a particle traversing C will
be used to remove θ. Let us now discuss why their in- return to the starting point with the same (opposite) handedness
terpretation is overly simplistic. First, however, we must if W (C) = 1 (W (C) = −1). Alternatively, for a manifold covered
define what we mean by a discrete symmetry. We give by charts Ui , each with a choice of coordinates xνi , we can define
a background field as the transition function between overlaps
three definitions, following from the usual description of ∂xν
j
symmetries in quantum field theory: ai→j = sign det ∂xi
µ ∈ Z2 . This means that ai→j = 1 on
overlaps between charts which have the same orientation and −1
1. A classical, discrete, global symmetry—this is a where the orientations differ. For a non-orientable manifold such
symmetry of the classical equations of motion. as a Möbius strip or a Klein bottle, any choice of charts will have
some overlap with ai→j = −1. This is related to the definition
2. A quantum, discrete, global symmetry—this is a in terms of holonomies because for any closed loop C, W (C) can
non-anomalous symmetry of all quantum correla- be defined as the product of the ai→j over all chart overlaps
tion functions. intersecting C.
 4

correspond to physical configurations of localized energy Note that, more precisely, gauging P requires a theory
density [47]. As we illustrate below, configurations of a of quantum gravity [47] (see also App. E), so that the
with non-trivial parity domain walls lead to color gauge partition function above should be embedded within a
anomalies unless θ + θ̄ = 0 or π. full gravitational path integral, which would likely re-
Suppose that we have arbitrary θ + θ̄ and are con- quire a sum over all manifolds M, as happens in e.g.
sidering the partition function of QCD with a classical Type-I string theory (see App. B). However, only the
background field a that implements orientation reversals contribution shown in (11) is relevant for this discussion,
across parity domain walls. We refer to this partition and so we suppress the full gravitational path integral
function as Za . An example of such a domain wall config- throughout this work for simplicity. Explicitly, the sum
uration would be the case where the orientation of space in (11) is only over manifolds related by non-trivial par-
flips as we cross r = rΣ in the spatial radial direction for ity holonomies. Parity holonomy implies that after be-
all time; that is, the parity domain wall Σ is a sphere of ing parallel transported along a non-contractable curve
radius rΣ .4 We may describe the presence of Σ through a field comes back to itself with a parity transform. An
a unit section ε of the orientation line bundle, which lo- example of a set of manifolds related by the insertion of
cally describes the orientation of any point of spacetime. a non-contractable parity domain wall are the torus and
Explicitly, let ε = +1 on M+ and −1 on M− , with the Klein bottle.
M+ describing Minkowski space outside Σ and M− the Note that the cohomology group H 1 (M, Z2 ) is trivial
inside. Then, we may write (see also [48] and [49] for for 4D Minkowski space M; that is, there are no nontriv-
related discussions) ial parity holonomies because every parity domain wall
 Z  in Minkowski space can be contracted to a point by par-
(θ + θ̄)

ity gauge transformations. Thus, the partition function
exp i 2
ε tr(G ∧ G) = exp i(θ + θ̄)ν ×
8π M of QCD with gauged parity in 4D Minkowski space looks
  (8)
identical to that of QCD with parity as an exact global
Z
1
exp i2(θ + θ̄) 2 CS3 (A) , symmetry, with both cases requiring θ + θ̄ = 0 or π. On
8π Σ
the other hand, option 2—the case of global P symme-
with try5 —is less satisfactory than option 3—the case of local
 
2ig P symmetry—because, as we expand upon below, of the
CS3 (A) ≡ tr A ∧ dA + A∧A∧A (9) philosophy that all global symmetries, anomalous or non-
3
anomalous, are accidental symmetries in the IR and are
the Chern-Simons 3-form localized on Σ, which is topo- not fundamental in the UV. On the other hand, option
logically equivalent to an S 3 , and ν the usual instanton 3 provides such a UV motivation for non-anomalous P —
number, coming from the difference in winding numbers this is the idea that P is gauged. In the next section we
at t = ±∞. Under large color gauge transformations provide a simple example where gauged P symmetry in
with support on Σ, ν is invariant but 4D Minkowski space descends from a continuous gauge
1
Z
1
Z symmetry in a higher dimensional theory with a compact
CS3 (A) → CS3 (A) + n , (10) extra dimension.
8π 2 Σ 8π 2 Σ
Before moving on, we give an additional argument for
for n ∈ Z. To preserve QCD gauge invariance in the θ = 0 or π for gauged parity, this time in the Hamiltonian
partition function Za , we must enforce ei2n(θ+θ̄) = 1, formulation. For concreteness, we work in Euclidean sig-
which restricts θ+ θ̄ to 0 and π. Once we enforce θ+ θ̄ = 0 nature and compactify time, so that M = S 3 × S 1 ; in
or π, the partition function Za of QCD is well defined for this case, M supports non-trivial parity holonomy. We
any background field a. take the interval [0, 1] to represent the Euclidean time
When θ + θ̄ = 0 or π we may gauge parity (option direction, with t = 0 and t = 1 identified. However, with
3 above), which corresponds to summing over all back- gauged parity we must sum over the inclusion of par-
ground fields a in the partition function, modulo par- ity domain walls, which here will be inserted at slices of
ity gauge transformations that correspond to continuous constant t:
deformations of the parity domain walls. After mod-
ding out by parity gauge transformations the resulting Z = Tre−βH + TrP e−βH , (12)
P -gauged partition function is given by [47, 50, 51] (see
also, e.g., [50, 52] for similar constructions) where H is the Hamiltonian, P is the parity operator
X inserted at t = 0, and β = 1 is the length of the Euclidean
Zgauged ⊃ Za . (11) time interval. Since the parity operator P can be inserted
[a]∈H 1 (M,Z2 )

4 To avoid subtleties about boundaries at t = ±∞, let us assume 5 See also Ref. [53], which defines a symmetry in this way, and also
that around t = ±tmax , for some large time tmax , the sphere is examines which additional assumptions, e.g., on the embedding
continuously grown / shrunk from / to a point, so that Σ has of the Standard Model in a larger gauge group, would further
the topology of S 3 . select θ + θ̄ = 0.
 5

at any time t, it should commute with the Hamiltonian, with fa the axion decay constant. Suppose that a has a
which sets θ̄ = 0 or π. On the other hand, we may potential of the form
rewrite (12) as
λ 4
V (a) = − 12 µ2 a2 + a . (15)
Z = Tr(1 + P )e−βH , (13) 4
This theory is consistent with gauged P with the trans-
which makes it clear that the states need to be even un- formation a → −a under P . On the other hand, P is
der parity as the odd components are projected out. This spontaneously
√ broken by the choice of vacuum ⟨a⟩ =
restricts the states to the superselection sectors θ = 0 or ±µ/ λ; in this case,√ the neutron EDM is calculable and
π, such that in total θ + θ̄ = 0 or π. To make this discus- proportional to µ/ λ.
sion concrete, in Appendix. A we compute the partition Gauged P or CP provides a clear counterexample to
function for 1+1D QED with parity gauged, and arrive the claims in [38]. Moreover, as discussed in e.g. [55],
at the same conclusion that the θ angle in that theory is most discrete solutions to Strong CP discussed previ-
restricted to 0 or π. ously in the literature can simply be embedded in UV
Note that our discussion of gauged parity is completely theories that realize P /CP as a gauge symmetry, though
analogous to how Type I string theory arises from gaug- there are cosmological complications related to stable
ing worldsheet parity in Type IIB string theory. In that parity domain walls that should be addressed, for exam-
context, anomaly cancellation leads to the restriction of ple by limiting the reheat temperature from inflation [47].
the gauge group to SO(32) via the Green–Schwarz mech- From a UV perspective gauged P /CP is also more mo-
anism. Here, gauging spacetime parity similarly restricts tivated than global P /CP , given the philosophy that all
θ to 0 or π (see, e.g., [54]). The orientifold planes in- global symmetries are accidental IR symmetries. In par-
troduced in the case of worldsheet parity are analogous ticular, quantum gravity is expected to violate all global
to the parity domain walls discussed above with gauged symmetries (see, e.g., [56, 57]). If global symmetries
spacetime parity. We explain this in more detail in Ap- are accidental IR symmetries, then there is no easily-
pendix B. justifiable reason to expect θ to be non-zero. Perhaps it
A UV theory that has gauged P (or gauged CP ) will is possible to find UV completions where θ + θ̄ = 0 with-
then automatically give the ‘initial condition’, from a UV out tuning, without imposing global symmetries in the
perspective, of θ + θ̄ = 0 or π. Of course, P /CP must be UV, and without promoting P or CP to gauge symme-
spontaneously broken at an intermediate energy scale, tries, but such UV completions are not obvious. On the
and it is a model building exercise to ensure that af- other hand, it certainly is the case that gauged P /CP
ter spontaneous symmetry breaking the neutron EDM is can make this cancellation automatic.
kept small enough for phenomenological purposes while
generating the CP violation in the CKM matrix. On
the other hand, this is now an IR model building ques- IV. GAUGED P FROM QUANTUM GRAVITY
tion that can be addressed at the level of the Lagrangian IN HIGHER DIMENSIONS
and not a question about an ambiguity in the superselec-
tion sector. Explicitly, if P or CP is gauged in the UV, It is well established that in string theory construc-
then every contribution to the neutron EDM is calculable tions 4D P or CP symmetry arises as a discrete gauge
in terms of the IR physics responsible for spontaneously symmetry [47, 55, 58, 59]. This is perhaps unsurpris-
breaking the discrete symmetry. ing since it is expected that quantum gravity violates all
Note that the spontaneous breaking of P /CP can global symmetries [57, 60–65]. Thus, in embedding P or
be decoupled from the dynamics involved in gauging CP solutions to the Strong CP problem in the context
the symmetry and take place at much lower energy of quantum gravity, it makes sense to suppose that these
scales. For example, as we discuss in the next section symmetries are gauged in the UV and spontaneously bro-
gauged P /CP can arise as the discrete remnant of gauged ken at intermediate energy scales. As we review here, in
Lorentz transformations of a higher dimensional gravita- the context of extra dimensions there are natural embed-
tional theory compactified on compact extra dimensions. dings of P and CP into continuous gauge symmetries of
On the other hand, P /CP can be spontaneously bro- the higher dimensional theory. We will begin by present-
ken at an energy scale parametrically smaller than the ing an explicit 5D example in which gauged P is realized
size of those extra dimensions through purely field theo- by embedding P in the proper Lorentz group before re-
retic mechanisms, though of course P /CP could also be viewing how gauged P and CP arise in specific string
spontaneously broken during the compactification. As a compactifications.
simple field theory realization of spontaneously broken
P symmetry, let us supplement our theory of pure QCD
with an axion field a(x) with the Lagrangian term A. Gauging P : a simple 5D example

−1
 
a Here we give a simple 5D example where 4D P arises
L⊃ Gaµν G̃a µν , (14)
32π 2 fa in the IR as a discrete, unbroken gauge symmetry of a
 6

continuous gauge symmetry in the UV through dimen- can be thought of as a gauge symmetry,6 with an associ-
sional reduction. Consider a pure 5D SU (3) gauge theory ated gauge field in the form of the spin connection ωµab (x)
coupled to gravity, with no charged matter, compactified taking values in in the Lie algebra so(1, d − 1). In par-
on a circle of radius R. The Lorentz transformation ticular, we can define a covariant derivative on vectors in
the tangent space V a = eaµ V µ ,
Λ5 = diag(1, −1, −1, −1, −1) , (16)
Dν V a = ∂ν V a + ω a bν V b , (19)
is continuously connected to the identity. We may con-
struct Λ5 through a sequence of rotations. First, we per- provided that the spin connection transforms under Λ as
form a rotation by ϕ = π in the x1 -x5 plane, giving the
Lorentz transformation matrix Λ = diag(1, −1, 1, 1, −1). ω 7→ ΛωΛ−1 + ΛdΛ−1 . (20)
Next, we perform a rotation by π in the x2 -x3 plane,
giving the rotation Λ5 . Note that we may write Λ5 = From the spin connection we can compute the curvature
P4 × P5 , with P4 = diag(1, −1, −1, −1, 1) the 4D parity
operator and P5 = (1, 1, 1, 1, −1) 5D parity. Ωαβ IJ = 2∂[α ωβ] IJ + 2ω[α IK ωβ]K J . (21)
Now let us dimensionally reduce the theory on S 1 .
Considering only the Kaluza-Klein (KK) zero modes of This is the tetrad formulation of GR [72–74]. In this
the 5D gauge field, this leads to the 4D SU (3) gauge formulation the gravitational action reads
theory with no θ-term in the Lagrangian. (As we later Z
discuss, in the presence of the 5D Chern–Simons term, β IJ
Stetrad = dd x e eαI eJ Ωαβ , (22)
θ = π is allowed.) The symmetry Λ5 is preserved by the
compactification, acting as xµ → −xµ and simultane- √
ously x5 → −x5 , with µ = 1, 2, 3. Since no fields depend with e = −g, and g is a function of the vielbein. Note
on x5 , as we are only considering the KK zero modes, Λ5 that unlike in Yang-Mills theories, the spin connection
descends to 4D parity; that is, P5 acts trivially on the is not associated to any propagating degrees of freedom
states of the EFT. (In theories with fermions this rela- because it is not an independent field; it is determined by
tion is more complicated, as we discuss below and as is requiring zero torsion. However, observables must still be
discussed in e.g. [55, 58, 59].) gauge invariant in the sense that they must be invariant
Let us consider how gauged P arises at the level of under local Lorentz transformations.
the path integral in this scenario using the fact that the The path integral of quantum gravity in 5D is of course
Lorentz group of the 5D spacetime is gauged by grav- a difficult object to construct, and at the moment the
ity. In fact, gauging P requires a gravitational theory. A only known way of making sense of such quantum grav-
rather elegant way to see this was presented in Ref. [47]: ity in 5D and 4D is through the dimensional reduction
the Z2 symmetry group associated to parity is a non- of string theory starting from a higher dimensional the-
normal subgroup of the full Lorentz group O(1, d − 1), ory. However, what is clear is that the gravitational
so if parity is to be gauged and the (proper) Lorentz path integral on a fixed manifold will include a sum
symmetry SO(1, d − 1) also preserved, the full Lorentz over vielbein fields eaµ , up to gauge equivalent identifi-
group must be gauged. We justify this in Appendix. E cations. Those gauge equivalent identifications include
According to general relativity (GR), the metric gµν (x) local Lorentz transformations, including Λ5 = P4 P5 . In
varies spatially but is locally Minkowski by the equiva- the case in which we only focus on zero modes of the met-
lence principle, such that there is a local Lorentz symme- ric and the QCD gauge field, then the gravitational path
try. Of course, for a generic curved spacetime, Lorentz integral in 5D reduces, in part, to the sum over gauge-
symmetry is not a global symmetry (note also that even inequivalent actions of the operator P4 , which gives the
for a flat spacetime, global Lorentz symmetry is broken partition function written in (11).
by the compactification in our 5D example). The local From a Hamiltonian perspective, the sum over
Lorentz symmetry acts in the tangent space of a given Lorentz-equivalent configurations ensures that the par-
point, and is captured by introducing the vielbein fields tition function is a sum of two pieces that differ by an
eaµ (x) such that we may write the metric as

gµν (x) = eaµ (x)ebν (x)ηab , (17)

6 Note that GR is also a gauge theory in a different sense; smooth
with ηab the Minkowski metric (the indices µ, ν, a, b run coordinate-changes xµ → x′µ (diffeomorphisms) of the space-
from 0 to d − 1, with d the number of spacetime dimen- time manifold leave the action invariant and can be viewed as a
gauge symmetry [66]; in linearized GR, small diffeomorphisms
sions). The local inertial frame is rotated by Lorentz give a gauge transformation of metric excitation h, hµν −→
transformations hµν + ∂µ ξν + ∂ν ξµ . The graviton, quantum of h, is the gauge
boson. In this case the gauge group is the infinite-dimensional
eaµ (x) → Λa b (x)ebµ (x) , (18) group of diffeomorphisms, which is not a Lie group as in Yang-
Mills. The understanding of GR as a gauge theory has a long
where Λ(x) ∈ SO(1, d − 1). This local Lorentz symmetry and interesting history, see [67–71] for details.
 7

action of the Lorentz transformation Λ5 = P4 P5 : the mass m breaks the 5D parity and must be real
to ensure the hermiticity of the Lagrangian. It is
Z = Tre−βH + TrΛ5 e−βH impossible to define charge conjugation symmetry in
= TrS 1 e−βH5 TrM4 e−βH4 + TrS 1 P5 e−βH5 TrM4 P4 e−βH4 . 5D. The Λ5 acts on the Dirac spinor ψ(x0 , x1,2,3,5 ) →
(23) iγ 1 γ 2 γ 3 γ5 ψ(x0 , −x1,2,3,5 ) where we use the 4D notation
with anti-hermitian γ 1,2,3 and hermitian γ5 . The product
If the fifth dimension is invariant under P5 , (23) reduces is nothing but γ 0 which defines the usual P4 . Note that
to the fermion Lagrangian could have contained a complex
mass term −mψ̄ψ cos ϕ−imψ̄γ5 ψ sin ϕ if only 4D Lorentz
Z = TrS 1 e−βH5 TrM4 e−βH4 + TrM4 P4 e−βH4 , (24) invariance is imposed, while the invariance under Λ5 for-


bids the second term. If there were the second term, a
where the 4D part is the sum of orientable and non- chiral rotation would eliminate it and also induce the θ-
orientable manifolds. term. Once again the invariance under Λ5 achieves the
Let us look at the consequence of Λ5 more concretely. 4D parity symmetry. Note, however, that the resulting θ
In 5D, there is no θ-term but there can be the Chern– may be 0 or π depending on the sign of the 5D mass.
Simons (CS) term in addition to the usual gauge kinetic This construction can be generalized to higher odd di-
term, mensions, for instance on M4 × S 1 × S 2n , if the 4D Dirac
1
fermion is obtained due to the flat connection H on S and
1 n 2n
the index cn ≡ (2π)n n! tr F on S . When S 1 A = π, we

1 k µνρσκ 1
Lgauge = − 2 trGµν Gµν + ϵ tr Aµ Gνρ Gσκ
2g5 24π 2 4 find the CS term ω2n+5 , defined via dω2n+5 = cn+3 , to be
1 1
 ω2n+5 = 12 cn c3 , and again the 4D θ angle is 0 or π. The
− Aµ Aρ Aσ Gσκ + Aµ Aν Aρ Aσ Aκ 4D mass term is also real. In higher 4 + 2n dimensions,
4 10
we can have a c2+n term with an arbitrary coefficient in
+ dimension six operators and above , (25) the Lagrangian. In order to obtain fermion zero modes,
we need an index cn , and hence a 4D θ term.
where here Aµ the SU (3) gauge field and Gµν its field
strength. The CS term is the only one in the pure Yang–
Mills theory that can potentially lead to the θ term in B. Gauging P and CP in string theory
4D. The coefficient k ∈ Z above must be an integer to
maintain the gauge invariance as the CS term changes by We have seen that gauging P requires a theory of quan-
an integer under large gauge transformations. The CS tum gravity. In string theory, if P or CP are symmetries,
term breaks the 5D parity explicitly, but of course is 5D they are expected to be gauged. In the following we re-
Lorentz invariant and hence preserves Λ5 . Compactifying view how P and CP can survive as 4D gauge symmetries
on M4 × S 1 , where S 1 has the circumference R, the first after compactification in string theory. (See also Refs.
term in (25) gives the usual 4D Lagrangian with g42 = [75, 76], who recently suggested a string-inspired solution
g52HR−1 . If the gauge group breaks by a Wilson line W = to the Strong CP problem using modular invariance.)
a a 5
ei S1 A5 T dx ̸= I for one of the generators a, the CS term As we see in the example of the previous subsection, P
can induce the θ-term in 4D and hence violate the 4D is gauged if it may be embedded as an unbroken element
parity P4 . However, if we assume the compactification of a higher dimensional Lorentz group. We have provided
preserves P5 , under which W → W −1 , then we need a simple construction to perform this embedding for odd-
every component of W = W −1 = (−1)n and hence A5 = dimensional spacetimes. Generalizing slightly the discus-
π
nR for n ∈ Z. The only term that survives in (25) is the sion in IV A, the trick was to note that in 2n-dimensional
first one in the parentheses where A5 may be part of Aµ Minkowski space
or Gνρ or Gσκ with three possible choices. Integrating
x0 , x1 , . . . , x2n−1 −→ x0 , −x1 , . . . , −x2n−1



over S 1 yields (28)
I
k µνρσκ

1 1 is in the component of the Lorentz group which is dis-
2
ϵ tr Aµ Gνρ Gσκ − Aµ Aν Aρ Gσκ connected from the identity, which is in the same discon-
S 1 24π 4 4
 nected component as the transformation x1 → −x1 . If an
1 knπ νρσκ 1 extra spatial dimension x2n is introduced, this Z2 sym-
+ Aµ Aν Aρ Aσ Aκ = ϵ tr Gνρ Gσκ , (26)
10 8π 2 4 metry is embedded in continuous SO(2) rotation group
of (x2n−1 , x2n ), so it is a proper Lorentz transformation
which implies θ = knπ, preserving P4 . A5 will acquire in 2n + 1 dimensions. The same trick works in some
mass from higher order corrections and it is quite possible string constructions. For example, consider a toroidal
that the minima are at A5 = 0, π/R given that they are compactification in 10D string theory on M4 × K with
symmetry-enhanced points. M4 4D Minkowski space and the internal manifold given
When considering fermions by K = T 2 × T 2 × T 2 , with T the torus. Grouping the
internal coordinates as z j = xj + ixj+1 for j = 4, 6, 8,
Lfermion = ψ̄(iγ µ Dµ − m)ψ, (27) the transformation Λ which combines 4D parity with the
 8

product of the conjugations z j → z̄ j is a proper Lorentz gauge CP as above becomes 10 8 . In general GYM does
transformation of the 10D theory. Note that it reverses not admit inner automorphisms which flip the sign of the
the orientation of K because an odd number of internal entire Cartan subalgebra 9 ; the only Lie groups which do
coordinates flip sign, and also of M4 , but preserves the are E8 , E7 , SO(2n + 1), SO(4n), Sp(2n), G2 , F4 , and
orientation of the full manifold. Λ also commutes with products of these groups do [79, 80]. Remarkably, this
the Gliozzi–Scherk–Olive condition [77], which ensures means that heterotic or Type I string theory with gauge
consistency with the wordlsheet supersymmetry and the group E8 × E8 or SO(32) can realize C as a gauge sym-
physical spectrum. By examining the 10D Dirac alge- metry of the 10D theory. In particular, compactifying
bra Λ can be shown to flip the handedness of massless on a CY 3-fold K, an orientation-reversing isometry IK
fermions, so it acts as 4D parity P [58]. This embed- of K can sometimes be identified as C. For example, in
ding of P is interesting because it applies to, e.g., 10D E8 × E8 heterotic string theory, the spin connection on
heterotic string theory, which is not invariant under 10D the tangent bundle of K, which has complex structure
parity to begin with (the theory contains Majorana-Weyl group SU (3), can be embedded in the gauge connection
spinors of a definite handedness). For a compactification of a single E8 via SU (3) ,→ SU (3) × E6 ,→ E8 . The
on a more complicated manifold such as a Calabi-Yau result is that in 4D the E8 gauge group is broken to E6 ,
(CY) 3-fold, the same embedding of parity above can and IK exchanges the 27 and 27 representations of E6
work: it would suffice to find an isometry IK of K (i.e a [81], so it acts as C 10 . In this case combining IK with
diffeomorphism of K which preserves the metric) which 4D parity realizes CP as a gauge symmetry in 4D, with
reverses the orientation of K. When combined with 4D CP of the form XL XYM as discussed above.
parity, the orientation of the 10D manifold is preserved. A common place where gauged parity is discussed is
For a CY 3-fold any such IK must be anti-holomorphic, in the context of Type I string theory, which implements
such as z j → z̄ j . In general z j → z̄ j is not a isometry of gauged worldsheet parity. We discuss this case in de-
the internal space, though for certain CY it can be [59]. tail in App. B and draw parallels between that well-
For example, the quintic hypersurface in CP 4 understood case and our discussion above about gauged
X parity in larger spacetime dimensions.
zi5 + εz1 z2 z3 z4 z5 = 0 (29)
i=1
V. THERE IS A STRONG CP PROBLEM
is a CY 3-fold. Then zi → z¯i is a symmetry if and only if ε
is real. Lastly, note that in some string theory examples,
In this Section we respond to recent claims in the lit-
such as 11D M-theory [49], the higher-dimensional theory
erature that there is no Strong CP problem, i.e that the
is already parity-symmetric.
neutron EDM vanishes even if θ̄ ̸= 0. We focus on the
Let us now ask how CP can arise as a gauged symme-
claim of Ref. [32] that boundary conditions in the path
try in string theory. As discussed in Ref. [55] (see also
integral set the neutron EDM to zero.
[78] for a more mathematical perspective), suppose we
wish to gauge CP by realizing it as a product of trans-
formations CP = XL XYM with XL an element of the
d-dimensional Lorentz group, and XY M an element of 8 We believe the discussion in Ref. [55] contains errors. First,
the internal Yang-Mills group GYM (which must be non- there it is claimed that gauging CP by realizing it as a product
trivial to have chiral fermions). Note that there may be CP = XL XYM requires 4D fields to be in irreducible representa-
tions of GYM × SO(d − 1, 1). However, a priori there is no reason
other ways to gauge CP . Supposing that all 4D fields live to not consider reducible representations instead. Ref. [55] also
in an irreducible representation of GYM × SO(d − 1, 1), implicitly assumes that self-conjugate irreducible spinor repre-
these representations must be self-conjugate. This is sentations must be Majorana, and further that Majorana spinors
possible if XL (XYM ) is an inner automorphism 7 of exist in 8k + 1 dimensions. This is incorrect: a self-conjugate
representation is either real or pseudo-real; in Minkowski space
SO(d − 1, 1) (GYM ) which exchanges an element of a
Majorana spinors exist only in 8k+2, 8k+3, and 8k+4 spacetime
representation with its complex conjugate. This requires dimensions [79].
irreducible spinor representations to be real or pseudo- 9 A group element g ∈ GYM acts on the Lie algebra via the ad-
real. As spinor representations are complex in 4D, at joint action Adg (X) = gXg −1 . We are interested in a g for
least five spacetime dimensions are needed to gauge CP which Adg (X) transforms the generators of the Cartan subalge-
bra as Hi → −Hi , and the step operators as Eα → E−α . This
in this way. Restricting to real representations requires ensures that the action of g flips the signs of weights λ → −λ,
8k + 1, 8k + 2, or 8k + 3 spacetime dimensions (for k meaning that the action of g realizes an isomorphism between
integer). If one allows for Majorana spinors, which ex- any representation of GYM to its conjugate representation.
10 This deserves some explanation. Under E8 ⊃ E6 × SU (3), the
ist in 8k + 2 dimensions (further, in these dimensions
Majorana-Weyl spinors exist), the minimal dimension to adjoint decomposes as 248 → (78, 1) ⊕ (1, 8) ⊕ (27, 3) ⊕ (27, 3).
Since K has SU (3) holonomy, the spin connection on the tangent
bundle of K (see (20)) is reduced to values in su(3) ⊂ so(6).
It can thus be identified as the gauge connection of the SU (3)
subgroup of E8 . In this case, an orientation-reversing isometry,
7 i.e is a map from the group to itself x → gxg −1 for some group being anti-holomorphic, exchanges the 3 ↔ 3̄ and thus also the
element g gauge group representations 27 ↔ 27.
 9

A. Order of limits in QCD partition function where ⟨O⟩Q indicates the expectation value fixing the
topological sector Q, and where the topological charge
In Ref. [32] (see also [28, 30]) it is claimed that CP density is
is conserved in QCD, and therefore the neutron EDM Z π
1
vanishes, due to a subtlety in the order of limits one takes p(Q) = dθZ(θ)e−iθQ (35)
when computing the expectation value of local operators. 2πZ(0) −π
2mπ 2 2
r  
In (2), we sum over topological sectors labeled by ν and 1 2mπ
send ν → ∞; implicitly we also take the limit of infinite = exp − Q . (36)
Z(0) β β
volume, V → ∞. The claim of Ref. [32] is that the only
mathematically sound procedure is to first take V → If we compute the topological susceptibility in this the-
∞ when computing the action in a given sector ν and ory,
then ν → ∞. This would lead to a vanishing neutron P 2
EDM, as well as a vanishing topological susceptibility of 2 1 |Q|<N Q p(Q)
χt = ⟨Q ⟩ = lim lim , (37)
QCD, independently of θ̄ 11 . Recall that the topological
P
N →∞ β→∞ β |Q|<N p(Q)
susceptibility of QCD is given by
taking first the limit β → ∞, we obtain χt = 0. On the
∂ 2 E(θ)
Z
χQCD ≡ = d4 x⟨q(x)q(0)⟩ , (30) other hand there is no issue with computing the topolog-
∂θ2 θ=0 ical charge at finite β; the sum over Q in (37) is perfectly
2
well-defined for any value of β, and taking first the limit
g
with E(θ) the vacuum energy and q(x) = 32π 2 F F̃ the N → ∞ we get χt = 1/(4π 2 m), which is the correct re-
topological charge density. sult that may be computed directly from the two-point
Ref. [82] pointed out that this claim can be understood function of Q, using (32), without any ambiguous order
to be incorrect by analogy to a simpler setting, the quan- of limits [82].
tum mechanical example of the particle on a ring (see also As in the example above, the order of limits claimed to
App. A). They consider a particle of mass m moving on be correct in Ref. [32] would set the topological suscepti-
a circle of unit radius, with Lagrangian bility to zero in pure Yang-Mills, just as it would in QCD
(this is claimed to be the case in [32]), which is clearly
1 θ incorrect. In the limit of large Nc (number of colors), it
L= mϕ̇2 + ϕ̇ , (31)
2 2π can be proved that at leading order, the mass squared
where ϕ is the coordinate of the particle on the ring and of the η ′ meson is proportional to the pure Yang-Mills
there is a total derivative in analogy with the θ term of topological susceptibility through the Witten-Veneziano
QCD. In this system time plays the role of the volume relation [83, 84]
for QCD and one could work in Euclidean time τ = it
4Nf
with periodic boundary conditions τ ∼ τ + β. We can m2η′ ≈ m2η + m2η′ − 2m2K = χYM , (38)
define a topological charge fπ2
Z β with mK and mη respectively the mass of the Kaon
1 dϕ
Q= dτ ∈ Z. (32) and η meson. The mass of the η ′ is measured to be
2π 0 dτ mη′ ≈ 957 MeV, and thus χYM ̸= 0 12 . The relation in
The partition function and energy levels of the system (38) has been confirmed by lattice calculations [86, 87].
are While this holds in the large Nc limit, it reproduces re-
 2 markably well the measured value of mη′ , using the value
X
−βEn 1 θ of χYM extracted from pure SU (3) lattice computations,
Z(θ) = e , En = n− . (33)
2m 2π χYM ≈ 190 MeV [86]. Note that (38) can be derived
n∈Z
without needing to discuss the order of the limits in the
The expectation value of an operator O is given by partition function (and also independently of the chiral
X Lagrangian), as we summarize in Appendix. C.
⟨O⟩ = lim lim ⟨O⟩Q p(Q) , (34)
N →∞ β→∞
|Q|<N
B. θ̄ in the Chiral Lagrangian

That θ̄ has physical effects, e.g., on the neutron EDM

11 In later work by the same authors Ref. [31], a similar argument or the vacuum energy of QCD, is well established in χ-
is presented that does not rely on taking the limit of infinite-
PT [40, 88, 89]. The claim of Ref. [32] would imply
volume. There it is claimed that the theta-parameter of pure
Yang–Mills theory on the 4D Euclidean torus does not have
physical effects. This claim is incorrect for the same reasons
we discuss. Note that in Appendix A we present a finite-volume
construction where the theta-term clearly has physical effects, 12 A argument using current algebra for why a massive η ′ directly
1+1D QED on a circle. implies violation of CP in QCD is given in [85].
 10

that this textbook result is incorrect, despite χ − PT, are nine Goldstone bosons associated with the spon-
being an effective theory of pseudoscalar mesons, know- taneous symmetry breaking to the diagonal subgroup
ing nothing of the topological sectors of QCD other than U (3)V . These Goldstone bosons can be parametrized by
through Wilson coefficients. Below, we summarize the the 3 × 3 unitary matrix
standard derivation of this result, and then point out the   
inconsistency in Ref. [32], which argues that the chiral i η1 λi
U = U0 exp √ I3 + √ ϕi , (40)
Lagrangian is independent of θ̄. fπ 3 2
Recall that in the limit of massless quarks, QCD with
Nf flavors possesses the flavor symmetries SU (Nf )L × where λi are the Gell-Mann matrices and the U (1) ×
SU (Nf )R ×U (1)V . At low energies, the chiral symmetry SU (3) content of the broken U (3)A is explicit. Under
is broken down to the diagonal subgroup SU (Nf )L × the chiral group, U transforms as gR U gL †
, with gR,L ∈
SU (Nf )R → SU (Nf )diag by the vacuum expectation U (3)R,L . The quark mass matrix M explicitly breaks
value (VEV) of the quark condensate ⟨q̄q⟩ ̸= 0. The U (3)A .
leading order terms of the chiral Lagrangian consistent However, the U (1)A part of the chiral symmetry group
with these symmetries are is anomalous in QCD. A naive approach to account
for the chiral symmetry breaking effect induced by the
anomaly in the effective low-energy theory, would be
fπ2 f 2 B0
LEFT Tr ∂µ U ∂ µ U † + π Tr M U + U † M † , through the term


M =
4 2
|λ|e−iξ fπ4 det U + h.c . (41)
with U ∈ SU (Nf ) the matrix of pseudo-Goldstone
bosons of the chiral symmetry breaking, fπ the pion de-
This is the route followed by the authors of Ref. [32].
cay constant, and M = diag (mq ) the mass matrix, with
Treating θ and M as spurions, they should transform
mq the quark masses, and B0 a constant. Note that we
properly under U (1)A , as
may parametrize U = exp (iπ a τ a ) with π a the Nf2 − 1
Goldstone fields and τ a the generators of SU (Nf ) nor-
M → e−2iβ M, θ → θ + 2Nf β ,
malized such that Tr τ a τ b = 2δ ab . Consider the topo- (42)
logical term in the QCD Lagrangian L ⊃ − 32π θ̄
2 GG̃.
det U → e2iNf β det U .
The chiral anomaly ensures that this term is removed
after performing a chiral rotation of the quark fields Ref. [32] claims that setting ξ = −arg detM is consistent
q → eiγ5 θ̄/4 q, corresponding to a rotation of the mass ma- with these transformations and is in fact the only op-
trix M → eiθ̄/2 M . Restricting to the two lightest quarks tion for ξ that respects their conclusion about the order
u, d for simplicity, a quick computation shows that of limits in the partition function, leaving the chiral La-

U ac-
quires a VEV of the form ⟨U ⟩ = diag eiϕ , e−iϕ with grangian independent of θ. In fact, a chiral rotation on
U that would make M real, would remove every phase
md − m u θ̄ from the Lagrangian. However, this choice for ξ is not
tan ϕ = tan . consistent with QCD as it implies that the EFT is invari-
md + m u 2
ant under the simultaneous rotations of M and U alone
The vacuum energy then depends on θ̄ as (given in (42)) without any shift in θ, whereas this is not
a symmetry of QCD.
Moreover, that this conclusion is incorrect can be seen
s
2 2 mu md 2 θ̄
E(θ̄) = −fπ mπ 1 − 4 2 sin 2 .
via a more consistent (but still heuristic) construction of
(mu + md ) the low energy Lagrangian from QCD. In fact, requiring
the correct transformation under a U (1) axial rotation,
The chiral Lagrangian contains terms of the form the absence of color degrees of freedom (confinement)
and demanding that the effective Lagrangian follows from
L ⊃ −c3 tr M U + M † U † N̄ U † PL + U PR N , (39)
 

QCD in the large-Nc limit, at leading order in 1/Nc , one
gets [93]
with N = (p, n)T the proton-neutron doublet under
SU(2) isospin and PL (PR ) left (right) handed projectors. fπ2 f 2 B0
Tr ∂µ U ∂ µ U † + π Tr M U + U † M †


This term clearly depends on θ̄ after the rotation of M ; LEFT ⊃
expanding at small θ̄ we obtain a term L ∼ θ̄π + p̄n which 4 2
i Nc
at one loop gives the neutron EDM dn ≈ 3×10−16 θ̄ e· cm. q(x) Tr log U − log U † + 2 q 2 (x) − θ̄q(x) ,
 
To argue instead that this θ̄-dependence is not present, 2 afπ
Ref. [32] follows an alternate, heuristic approach to ex- (43)
tend the chiral Lagrangian to include the η ′ (see e.g, 2
g µν
[84, 90–92]), which we now discuss. Let us consider where q(x) = 64π 2G G̃µν , a is a constant that is of
Nf = 3. At the classical level, the massless QCD La- order one as Nc → ∞, the normalization is chosen for
grangian has a U (3)L ×U (3)R chiral symmetry and there convenience and we work in the basis where M is real,
 11

so that all phases are encoded in θ̄. Removing q(x) with from the UV theory. Our reasoning for this conjecture is
its equation of motion, we get the following.
The key point is that the existence of theta vacua in
fπ2 f 2 B0 a non-abelian gauge theory is tied to the existence of a
Tr ∂µ U ∂ µ U † + π Tr M U + U † M †


LEFT ⊃
4 2 global symmetry; in particular, a Chern-Weil global sym-
a fπ2  2 metry [39]. The operator tr(F ∧ F ) is a closed form and
 
i  †
− θ̄ − Tr log U − log U . thus generates a generalized global symmetry; in the case
4Nc 2
(44) of 4D Yang-Mills theory this is a (−1)-form global sym-
metry. Since quantum gravity is conjectured to have no
Consequently after expanding, we see that m2η′ scales as exact global symmetries, it follows that this (−1)-form
global symmetry should be gauged or broken. Gauging
1/Nc . Note that (44) can brought in a form similar to
the Chern-Weil global symmetry corresponds to intro-
(41) since Tr(log U ) = log det U . The logarithm in (44)
ducing an axion field a, which has an exact shift sym-
is needed as we know from 1/Nc expansion counting
metry (which is the gauge symmetry). The axion is the
rules that the anomaly-induced interaction is, to leading
gauge boson of the (−1)-form gauge symmetry [39]. Since
order in 1/Nc , quadratic in η1 only [91]. It is therefore
the axion shift symmetry is a gauge redundancy, the con-
evident that the low energy theory which encodes all
cept of the theta vacua is also redundant; we can shift the
features of QCD has a physical dependence on θ̄. A
axion field to cancel any base contribution θ + θ̄. Thus,
chiral rotation on U would only shift it between M and
a unique and calculable value of θ + θ̄ is generated in the
the last term in (44).
IR.13
Of course, just because the theory has an axion does
not mean that the axion solves the Strong CP problem.
VI. DISCUSSION For example, the axion could spontaneously acquire a po-
tential through e.g. Euclidean D-branes wrapping extra
dimensions in a compactification of a higher dimensional
In this work we show explicitly that (i) the Strong CP theory or through the confinement of a hidden gauge sec-
problem is a real problem and not trivially solved by e.g. tor with a confinement scale well above that of QCD. This
the neutron EDM having no dependence on θ̄ (it does), would make the axion heavy and, generically, introduce a
and (ii) that models based on gauged but spontaneously non-zero θ + θ̄ in the IR EFT.14 However, in this case the
broken P or CP can solve this problem. Note that our value of θ + θ̄ in the IR is fully calculable from the physics
statement (ii) is an “in principle” statement in that we that sets the axion’s potential and does not depend on
do not develop new models solving the Strong CP prob- an unknowable superselection sector. Thus, in principle,
lem through discrete symmetries but rather we show that with creative model building the phase of this potential
many existing solutions can be embedded in UV models could be aligned with that of QCD, such that ⟨a/fa ⟩ ∼ 0.
of gauged P /CP . While this has largely been known
(To emphasize the point, though, just because this is in
for decades (see, e.g., [55]), it bears repeating in light
principle possible does not mean that constructing such
of the recent work [38]. Contrary to the claims in [38]
a model is easy.)
(who claim that only the axion can solve the Strong CP
Indeed, in the one theory of quantum gravity that we
problem), models in which P or CP is gauged in the UV
have—string theory—axions are abundant in 4D EFTs,
do not suffer from any ambiguity in theta state versus
where they often arise from the dimensional reduction of
Lagrangian contributions to the neutron EDM; in these
higher-form gauge fields during compactification to 4D.
cases, θ + θ̄ = 0 or π in the UV, with θ representing the
Those higher-dimensional higher-form gauge fields can
contribution to the neutron EDM from the theta vacuum
and θ̄ denoting the Lagrangian contribution. After spon-
taneous P or CP symmetry breaking, the contributions
to the neutron EDM are calculable from the physics re- 13 There is however an alternate possibility, emphasized in [39, 94,
sponsible for the breaking. (This, of course, does not 95], which is to break the Chern-Weil global symmetry. This
mean that constructing such models based off of gauged corresponds to the case where θ is “frozen” in the IR to one of
P or CP in the UV that satisfy all phenomenological a discrete set of possible values, without any modulus (such as
requirements is easy.) an axion) that can vary it away from those values. For example,
in Type IIB string theory compactified on a rigid Calabi-Yau 3-
On the other hand, at a deeper level it is unsatisfying fold, there is no light axion. Even in this example, the physical
to imagine a theory where the neutron EDM depends on theta parameter in the IR is calculable, and in fact must be 0
a parameter θ describing the theta state that is other- or π. Ref. [94] however argues that all known examples which
wise incalculable by the UV completion of the theory. break the Chern-Weil symmetry lack light charged matter and
thus cannot be phenomenologically viable.
We conjecture that in theories that arise from quantum 14 It is plausible that the Strong CP problem is solved by a
gravitational theories in the UV, the IR value of θ + θ̄, symmetry-based solution even if a heavy QCD axion is present.
which is the only physically observable combination of For example, corrections to the axion potential may preserve
theta angles (including, of course, the contribution from P /CP due to the geometry of the compactification, see e.g, [96–
the quark mass phases in θ̄), is dynamically determined 98].
 12

often be thought of as gauging the Chern-Weil global ACKNOWLEDGEMENTS

symmetries that would otherwise be present. The higher-
form gauge symmetries of these p-form fields imply that We thank Quentin Bonnefoy, Itay Bloch, Nathaniel
all contributions to θ + θ̄ in the IR, and thus the neutron Craig, Gian Francesco Giudice, Lawrence Hall, Daniel
EDM, are calculable from the dynamics of the theory. Harlow, David B. Kaplan, David E. Kaplan, Tom Melia,
Given that all models beyond the SM should be embed- Matthew McCullough, Jacob McNamara, Surjeet Ra-
ded in quantum gravity, this then leads us to question jendran, Maria Ramos, Matthew Reece, Mario Reig,
whether the entire discussion of global P and CP sym- Nicholas Rodd, Carlos Tamarit, David Tong, Gabriele
metries and their implications in [38] is not in the swamp- Veneziano and Alexander Zhiboedov for useful discus-
land. sions. We thank the Pollica Physics Center and CERN
for hospitality during the writing of this manuscript. J.B.
and B.R.S. are supported in part by the DOE award
DESC0025293. C.A.M. is supported by the U.S. Depart-
ment of Energy (DE-SC0009988) and the Sivian Fund.
AH is supported by NSF grant PHY-2514660 and the
Maryland Center for Fundamental Physics.

[1] C. Abel et al., “Measurement of the Permanent Electric [15] K. S. Babu and Rabindra N. Mohapatra, “A Solution
Dipole Moment of the Neutron,” Phys. Rev. Lett. 124, to the Strong CP Problem Without an Axion,” Phys.
081803 (2020), arXiv:2001.11966 [hep-ex]. Rev. D 41, 1286 (1990).
[2] Jihn E. Kim and Gianpaolo Carosi, “Axions and the [16] Stephen M. Barr, D. Chang, and G. Senjanovic,
Strong CP Problem,” Rev. Mod. Phys. 82, 557–602 “Strong CP problem and parity,” Phys. Rev. Lett. 67,
(2010), [Erratum: [Link]. 91, 049902 (2019)], 2765–2768 (1991).
arXiv:0807.3125 [hep-ph]. [17] L. Lavoura, “A New type of spontaneous CP break-
[3] Anson Hook, “TASI Lectures on the Strong CP ing,” Phys. Lett. B 400, 152–156 (1997), arXiv:hep-
Problem and Axions,” PoS TASI2018, 004 (2019), ph/9701221.
arXiv:1812.02669 [hep-ph]. [18] Luca Vecchi, “Spontaneous CP violation and the strong
[4] Benjamin R. Safdi, “TASI Lectures on the Parti- CP problem,” JHEP 04, 149 (2017), arXiv:1412.3805
cle Physics and Astrophysics of Dark Matter,” PoS [hep-ph].
TASI2022, 009 (2024), arXiv:2303.02169 [hep-ph]. [19] Michael Dine and Patrick Draper, “Challenges for
[5] Claudio Bonanno, Claudio Bonati, and Massimo the Nelson-Barr Mechanism,” JHEP 08, 132 (2015),
D’Elia, “Strong CP problem, theta term and QCD topo- arXiv:1506.05433 [hep-ph].
logical properties,” (2025), arXiv:2510.03059 [hep-lat]. [20] Lawrence J. Hall and Keisuke Harigaya, “Implications
[6] Lorenzo Ubaldi, “Effects of theta on the deuteron bind- of Higgs Discovery for the Strong CP Problem and Uni-
ing energy and the triple-alpha process,” Phys. Rev. D fication,” JHEP 10, 130 (2018), arXiv:1803.08119 [hep-
81, 025011 (2010), arXiv:0811.1599 [hep-ph]. ph].
[7] Michael Dine, Laurel Stephenson Haskins, Lorenzo [21] David Dunsky, Lawrence J. Hall, and Keisuke Harigaya,
Ubaldi, and Di Xu, “Some Remarks on Anthropic Ap- “Higgs Parity, Strong CP, and Dark Matter,” JHEP 07,
proaches to the Strong CP Problem,” JHEP 05, 171 016 (2019), arXiv:1902.07726 [hep-ph].
(2018), arXiv:1801.03466 [hep-th]. [22] Nathaniel Craig, Isabel Garcia Garcia, Giacomo
[8] R. D. Peccei and Helen R. Quinn, “CP Conservation in Koszegi, and Amara McCune, “P not PQ,” JHEP 09,
the Presence of Instantons,” Phys. Rev. Lett. 38, 1440– 130 (2021), arXiv:2012.13416 [hep-ph].
1443 (1977). [23] Quentin Bonnefoy, Lawrence Hall, Claudio Andrea
[9] R. D. Peccei and Helen R. Quinn, “Constraints Im- Manzari, and Christiane Scherb, “Colorful Mirror Solu-
posed by CP Conservation in the Presence of Instan- tion to the Strong CP Problem,” Phys. Rev. Lett. 131,
tons,” Phys. Rev. D16, 1791–1797 (1977). 221802 (2023), arXiv:2303.06156 [hep-ph].
[10] Steven Weinberg, “A New Light Boson?” Phys. Rev. [24] Lawrence Hall, Claudio Andrea Manzari, and
Lett. 40, 223–226 (1978). Bea Noether, “Strong CP and flavor in multi-
[11] Frank Wilczek, “Problem of Strong p and t Invariance Higgs theories,” Phys. Rev. D 111, 115012 (2025),
in the Presence of Instantons,” Phys. Rev. Lett. 40, arXiv:2407.14585 [hep-ph].
279–282 (1978). [25] Ferruccio Feruglio, Matteo Parriciatu, Alessandro Stru-
[12] Ann E. Nelson, “Naturally Weak CP Violation,” Phys. mia, and Arsenii Titov, “Solving the strong CP
Lett. B 136, 387–391 (1984). problem without axions,” JHEP 08, 214 (2024),
[13] Stephen M. Barr, “Solving the Strong CP Problem arXiv:2406.01689 [hep-ph].
Without the Peccei-Quinn Symmetry,” Phys. Rev. Lett. [26] Quentin Bonnefoy, Lawrence J. Hall, Claudio An-
53, 329 (1984). drea Manzari, and Bea Noether, “Two Higgs Dou-
[14] K. S. Babu and Rabindra N. Mohapatra, “CP Violation blet Solutions to the Strong CP Problem,” (2025),
in Seesaw Models of Quark Masses,” Phys. Rev. Lett. arXiv:2506.13853 [hep-ph].
62, 1079 (1989).
 13

[27] John R. Ellis and Mary K. Gaillard, “Strong and Weak (2016), [Erratum: [Link]. 119, 119901 (2017)],
CP Violation,” Nucl. Phys. B 150, 141–162 (1979). arXiv:1601.04339 [[Link]-ph].
[28] Wen-Yuan Ai, Juan S. Cruz, Björn Garbrecht, and Car- [46] Davide Gaiotto, Anton Kapustin, Nathan Seiberg, and
los Tamarit, “Consequences of the order of the limit of Brian Willett, “Generalized Global Symmetries,” JHEP
infinite spacetime volume and the sum over topologi- 02, 172 (2015), arXiv:1412.5148 [hep-th].
cal sectors for CP violation in the strong interactions,” [47] Jacob McNamara and Matthew Reece, “Reflections on
Phys. Lett. B 822, 136616 (2021), arXiv:2001.07152 Parity Breaking,” (2022), arXiv:2212.00039 [hep-th].
[hep-th]. [48] Dmitri Vassilevich, “Index Theorems and Domain
[29] Y. Nakamura and G. Schierholz, “The strong CP Walls,” JHEP 07, 108 (2018), arXiv:1805.09974 [hep-
problem solved by itself due to long-distance vac- th].
uum effects,” Nucl. Phys. B 986, 116063 (2023), [49] Edward Witten, “The ”Parity” Anomaly On An Un-
arXiv:2106.11369 [hep-ph]. orientable Manifold,” Phys. Rev. B 94, 195150 (2016),
[30] Wen-Yuan Ai, Juan S. Cruz, Björn Garbrecht, and arXiv:1605.02391 [hep-th].
Carlos Tamarit, “The limits of the strong CP [50] T. Daniel Brennan and Sungwoo Hong, “Introduction
problem,” PoS DISCRETE2020-2021, 084 (2022), to Generalized Global Symmetries in QFT and Particle
arXiv:2205.15093 [hep-th]. Physics,” (2023), arXiv:2306.00912 [hep-ph].
[31] Wen-Yuan Ai, Bjorn Garbrecht, and Carlos Tamarit, [51] Daniel Harlow and Tokiro Numasawa, “Gauging
“The QCD theta-parameter in canonical quantization,” spacetime inversions in quantum gravity,” (2023),
(2024), arXiv:2403.00747 [hep-th]. arXiv:2311.09978 [hep-th].
[32] Wen-Yuan Ai, Bjorn Garbrecht, and Carlos Tamarit, [52] Robbert Dijkgraaf and Edward Witten, “Topological
“CP Conservation in the Strong Interactions,” Universe Gauge Theories and Group Cohomology,” Commun.
10, 189 (2024), arXiv:2404.16026 [hep-ph]. Math. Phys. 129, 393 (1990).
[33] Gerrit Schierholz, “Absence of strong CP violation,” J. [53] Luca Vecchi, “When CP requires θ̄ = 0, not θ̄ = π,”
Phys. G 52, 04LT01 (2025), arXiv:2403.13508 [hep-ph]. (2025), arXiv:2507.10680 [hep-ph].
[34] Gerrit Schierholz, “Absence of CP Violation in the [54] Joseph Polchinski and Edward Witten, “Evidence for
Strong Interaction: Vacuum thwarts Axion,” PoS LAT- heterotic - type I string duality,” Nucl. Phys. B 460,
TICE2024, 398 (2025), arXiv:2502.04092 [hep-lat]. 525–540 (1996), arXiv:hep-th/9510169.
[35] Gia Dvali, “A Vacuum accumulation solution to the [55] Ki-woon Choi, David B. Kaplan, and Ann E. Nelson,
strong CP problem,” Phys. Rev. D 74, 025019 (2006), “Is CP a gauge symmetry?” Nucl. Phys. B 391, 515–530
arXiv:hep-th/0510053. (1993), arXiv:hep-ph/9205202.
[36] Gia Dvali and Glennys R. Farrar, “Strong CP Problem [56] Tom Banks and Nathan Seiberg, “Symmetries and
with 10**32 Standard Model Copies,” Phys. Rev. Lett. Strings in Field Theory and Gravity,” Phys. Rev. D 83,
101, 011801 (2008), arXiv:0712.3170 [hep-th]. 084019 (2011), arXiv:1011.5120 [hep-th].
[37] Gia Dvali, “Strong-CP with and without gravity,” [57] Daniel Harlow and Hirosi Ooguri, “Symmetries in quan-
(2022), arXiv:2209.14219 [hep-ph]. tum field theory and quantum gravity,” Commun.
[38] David E. Kaplan, Tom Melia, and Surjeet Rajendran, Math. Phys. 383, 1669–1804 (2021), arXiv:1810.05338
“What can solve the Strong CP problem?” (2025), [hep-th].
arXiv:2505.08358 [hep-ph]. [58] Michael Dine, Robert G. Leigh, and Douglas A. MacIn-
[39] Ben Heidenreich, Jacob McNamara, Miguel Montero, tire, “Of CP and other gauge symmetries in string the-
Matthew Reece, Tom Rudelius, and Irene Valen- ory,” Phys. Rev. Lett. 69, 2030–2032 (1992), arXiv:hep-
zuela, “Chern-Weil global symmetries and how quan- th/9205011.
tum gravity avoids them,” JHEP 11, 053 (2021), [59] A. Strominger and Edward Witten, “New Manifolds for
arXiv:2012.00009 [hep-th]. Superstring Compactification,” Commun. Math. Phys.
[40] M. Srednicki, Quantum field theory (Cambridge Univer- 101, 341 (1985).
sity Press, 2007). [60] Tom Banks and Lance J. Dixon, “Constraints on String
[41] Maxim Pospelov and Adam Ritz, “Theta induced elec- Vacua with Space-Time Supersymmetry,” Nucl. Phys.
tric dipole moment of the neutron via QCD sum rules,” B 307, 93–108 (1988).
Phys. Rev. Lett. 83, 2526–2529 (1999), arXiv:hep- [61] Daniel Harlow and Hirosi Ooguri, “Constraints on Sym-
ph/9904483. metries from Holography,” Phys. Rev. Lett. 122, 191601
[42] Jian Liang, Andrei Alexandru, Terrence Draper, Keh- (2019), arXiv:1810.05337 [hep-th].
Fei Liu, Bigeng Wang, Gen Wang, and Yi-Bo Yang [62] L. F. Abbott and Mark B. Wise, “Wormholes and
(χQCD), “Nucleon electric dipole moment from the θ Global Symmetries,” Nucl. Phys. B 325, 687–704
term with lattice chiral fermions,” Phys. Rev. D 108, (1989).
094512 (2023), arXiv:2301.04331 [hep-lat]. [63] Renata Kallosh, Andrei D. Linde, Dmitri A. Linde,
[43] C. A. Baker et al., “An Improved experimental limit on and Leonard Susskind, “Gravity and global symme-
the electric dipole moment of the neutron,” Phys. Rev. tries,” Phys. Rev. D 52, 912–935 (1995), arXiv:hep-
Lett. 97, 131801 (2006), arXiv:hep-ex/0602020. th/9502069.
[44] J. M. Pendlebury et al., “Revised experimental upper [64] Yiming Chen and Henry W. Lin, “Signatures of global
limit on the electric dipole moment of the neutron,” symmetry violation in relative entropies and replica
Phys. Rev. D 92, 092003 (2015), arXiv:1509.04411 [hep- wormholes,” JHEP 03, 040 (2021), arXiv:2011.06005
ex]. [hep-th].
[45] B. Graner, Y. Chen, E. G. Lindahl, and B. R. [65] Edward Witten, “Anti de Sitter space and holography,”
Heckel, “Reduced Limit on the Permanent Electric Adv. Theor. Math. Phys. 2, 253–291 (1998), arXiv:hep-
Dipole Moment of Hg199,” Phys. Rev. Lett. 116, 161601 th/9802150.
 14

[66] David Tong, “String Theory,” (2009), arXiv:0908.0333 [88] Varouzhan Baluni, “CP Violating Effects in QCD,”
[hep-th]. Phys. Rev. D 19, 2227–2230 (1979).
[67] Ryoyu Utiyama, “Invariant theoretical interpretation of [89] R. J. Crewther, P. Di Vecchia, G. Veneziano, and Ed-
interaction,” Phys. Rev. 101, 1597–1607 (1956). ward Witten, “Chiral Estimate of the Electric Dipole
[68] D. W. Sciama, “On the analogy between charge and Moment of the Neutron in Quantum Chromodynamics,”
spin in general relativity,” in Recent Developments in Phys. Lett. B 88, 123 (1979), [Erratum: [Link].B 91,
General Relativity (1962) p. 415. 487 (1980)].
[69] T. W. B. Kibble, “Lorentz invariance and the gravita- [90] Edward Witten, “Instantons, the Quark Model, and the
tional field,” J. Math. Phys. 2, 212–221 (1961). 1/n Expansion,” Nucl. Phys. B 149, 285–320 (1979).
[70] Jason Bennett, “A pedagogical review of gravity as a [91] Edward Witten, “Large N Chiral Dynamics,” Annals
gauge theory,” (2021), arXiv:2104.02627 [gr-qc]. Phys. 128, 363 (1980).
[71] E. Cartan, “Sur les variétés à connexion affine et la [92] Csaba Csáki, Raffaele Tito D’Agnolo, Rick S. Gupta,
théorie de la relativité généralisée. (première partie),” Eric Kuflik, Tuhin S. Roy, and Maximilian Ruhdorfer,
Annales Sci. Ecole Norm. Sup. 40, 325–412 (1923). “On the dynamical origin of the η ′ potential and the
[72] Robert M. Wald, General Relativity (Chicago Univ. Pr., axion mass,” JHEP 10, 139 (2023), arXiv:2307.04809
Chicago, USA, 1984). [hep-ph].
[73] Attilio Palatini, “Deduzione invariantiva delle equazioni [93] P. Di Vecchia and G. Veneziano, “Chiral Dynamics in
gravitazionali dal principio di Hamilton,” Rend. Circ. the Large n Limit,” Nucl. Phys. B 171, 253–272 (1980).
Mat. Palermo 43, 203–212 (1919). [94] Matthew Reece, “TASI Lectures: (No) Global Symme-
[74] Edward Witten, “(2+1)-Dimensional Gravity as an Ex- tries to Axion Physics,” PoS TASI2022, 008 (2024),
actly Soluble System,” Nucl. Phys. B 311, 46 (1988). arXiv:2304.08512 [hep-ph].
[75] Ferruccio Feruglio, Alessandro Strumia, and Arsenii [95] Sergio Cecotti and Cumrun Vafa, “Theta-problem and
Titov, “Modular invariance and the QCD angle,” JHEP the String Swampland,” (2018), arXiv:1808.03483 [hep-
07, 027 (2023), arXiv:2305.08908 [hep-ph]. th].
[76] Alexander Baur, Moritz Kade, Hans Peter Nilles, [96] Keiya Ishiguro, Tatsuo Kobayashi, and Hajime Ot-
Saul Ramos-Sanchez, and Patrick K. S. Vau- suka, “Spontaneous CP violation and symplectic mod-
drevange, “Siegel modular flavor group and CP from ular symmetry in Calabi-Yau compactifications,” Nucl.
string theory,” Phys. Lett. B 816, 136176 (2021), Phys. B 973, 115598 (2021), arXiv:2010.10782 [hep-th].
arXiv:2012.09586 [hep-th]. [97] Kilian Bönisch, Mohamed Elmi, Amir-Kian Kashani-
[77] F. Gliozzi, Joel Scherk, and David I. Olive, “Super- Poor, and Albrecht Klemm, “Time reversal and CP
symmetry, Supergravity Theories and the Dual Spinor invariance in Calabi-Yau compactifications,” JHEP 09,
Model,” Nucl. Phys. B 122, 253–290 (1977). 019 (2022), arXiv:2204.06506 [hep-th].
[78] Brett McInnes, “Gauging discrete symmetries,” J. [98] Tatsuo Kobayashi and Hajime Otsuka, “Challenge
Math. Phys. 36, 5414–5430 (1995). for spontaneous CP violation in Type IIB orien-
[79] R. Slansky, “Group Theory for Unified Model Building,” tifolds with fluxes,” Phys. Rev. D 102, 026004 (2020),
Phys. Rept. 79, 1–128 (1981). arXiv:2004.04518 [hep-th].
[80] Brian Henning, Xiaochuan Lu, Tom Melia, and Hitoshi [99] Guy Bonneau, Jacques Faraut, and Galliano Valent,
Murayama, “Outer automorphism anomalies,” JHEP “Selfadjoint extensions of operators and the teaching
02, 094 (2022), arXiv:2111.04728 [hep-th]. of quantum mechanics,” Am. J. Phys. 69, 322 (2001),
[81] P. Candelas, Gary T. Horowitz, Andrew Strominger, arXiv:quant-ph/0103153.
and Edward Witten, “Vacuum configurations for super- [100] David Tong, “Lectures on gauge theory,” https://
strings,” Nucl. Phys. B 258, 46–74 (1985). [Link]/user/tong/[Link]
[82] David Albandea, Guilherme Catumba, and Alberto (2018), lecture notes, University of Cambridge.
Ramos, “Strong CP problem in the quantum rotor,” [101] Claude W. Bernard and Erick J. Weinberg, “The Inter-
Phys. Rev. D 110, 094512 (2024), arXiv:2402.17518 pretation of Pseudoparticles in Physical Gauges,” Phys.
[hep-lat]. Rev. D 15, 3656 (1977).
[83] Edward Witten, “Current Algebra Theorems for the [102] Erick J. Weinberg, “Classical solutions in quantum
U(1) Goldstone Boson,” Nucl. Phys. B 156, 269–283 field theories,” Ann. Rev. Nucl. Part. Sci. 42, 177–210
(1979). (1992).
[84] G. Veneziano, “U(1) Without Instantons,” Nucl. Phys. [103] R. Jackiw, “TOPOLOGICAL INVESTIGATIONS OF
B 159, 213–224 (1979). QUANTIZED GAUGE THEORIES,” Conf. Proc. C
[85] Mikhail A. Shifman, A. I. Vainshtein, and Valentin I. 8306271, 221–331 (1983).
Zakharov, “Can Confinement Ensure Natural CP In- [104] C. P. Burgess and T. R. Morris, “Open and Unoriented
variance of Strong Interactions?” Nucl. Phys. B 166, Strings a La Polyakov,” Nucl. Phys. B 291, 256–284
493–506 (1980). (1987).
[86] Luigi Del Debbio, Leonardo Giusti, and Claudio Pica, [105] J. Polchinski, String theory. Vol. 1: An introduction to
“Topological susceptibility in the SU(3) gauge the- the bosonic string, Cambridge Monographs on Mathe-
ory,” Phys. Rev. Lett. 94, 032003 (2005), arXiv:hep- matical Physics (Cambridge University Press, 2007).
th/0407052. [106] J. Polchinski, String theory. Vol. 2: Superstring theory
[87] Krzysztof Cichy, Elena Garcia-Ramos, Karl Jansen, and beyond , Cambridge Monographs on Mathematical
Konstantin Ottnad, and Carsten Urbach (ETM), Physics (Cambridge University Press, 2007).
“Non-perturbative Test of the Witten-Veneziano For- [107] Carlo Angelantonj and Augusto Sagnotti, “Open
mula from Lattice QCD,” JHEP 09, 020 (2015), strings,” Phys. Rept. 371, 1–150 (2002), [Erratum:
arXiv:1504.07954 [hep-lat]. [Link]. 376, 407 (2003)], arXiv:hep-th/0204089.
 15

[108] Luis E. Ibanez and Angel M. Uranga, String theory and

particle physics: An introduction to string phenomenol-
ogy (Cambridge University Press, 2012).
[109] Joseph Polchinski, “What is string theory?” in NATO
Advanced Study Institute: Les Houches Summer School,
Session 62: Fluctuating Geometries in Statistical Me-
chanics and Field Theory (1994) arXiv:hep-th/9411028.
[110] Wieland Staessens and Bert Vercnocke, “Lectures on
Scattering Amplitudes in String Theory,” in 5th Mo-
dave Summer School in Mathematical Physics (2010)
arXiv:1011.0456 [hep-th].
 16

Appendix A: QED in 1+1D It is easy to see that this condition is the gauge invariance
of the state under “small” gauge transformation
Here we discuss QED in 1+1D; which is an exactly Z
solvable QFT that contains a θ term in the Lagrangian. dx ω(x)⟨A(x)|∂x Ex |phys⟩
This allows us to study the vacuum of this theory, which Z
∂
is in some ways analogous to that of QCD, by direct com- = dx ω(x)∂x i ⟨A(x)|phys⟩
putation. We begin by discussing the vacuum structure ∂A(x)
Z
after fully gauge fixing, and then proceed to study θ as ∂
= −i dx ∂x ω(x) ⟨A(x)|phys⟩ = 0. (A9)
a Lagrangian parameter as well as the associated bound- ∂A(x)
ary condition phase, via both canonical quantization and
the path integral. We also draw an analogy between the To make it clear what the physical state condition means,
Strong CP problem and the the presence of an electric we expand the vector potential at a fixed time slice as
dipole moment of an e+ e− bound state that arises when ∞ 
X 
massive fermions are included. Lastly we discuss gauging A(x) = a0 + an e2πinx/L + a∗n e−2πinx/L , (A10)
parity in this theory and show by explicit calculation that n=1
the partition function has contributions from orientable where a0 is real and an (n > 1) are complex. The wave
and non-orientable manifolds. functional ψ[A] = ⟨A(x)|phys⟩ is in principle a function
of infinite number of variables a0 , an , a∗n . However, the
physical state condition (A9) means that it does not
1. Perspectives on theta vacua
depend on any an , a∗n (n > 0). It is a function of a0 only,
ψ[A] = ψ(a0 ).
The Lagrangian density is the familiar one If we consider a “fully gauge fixed” space of all gauge
1 µν θ0 fields, we consider the wave functional to live on the space
L=− F Fµν + ϵµν F µν . (A1) A/G, where G is a group of all possible gauge transfor-
4e2 4π
mations on S 1 . Keeping only a0 eliminates redundancies
We consider the space to be a circle x ∈ [0, L] where x = 0 up to gauge transformations that are continuously con-
and x = L are identified. We can always choose the Weyl nected to the identity. However, there are also “large”
(temporal) gauge A0 = 0 using the gauge parameter gauge transformations gn (x) = e−2πinx/L
Z t 2π
ω(x, t) = A0 (x, t′ )dt′ , (A2) A(x) → A(x) + ign−1 (x)∂x gn (x) = A(x) + n. (A11)
0
L
It is clear that gn winds U(1) n times as we move along
and perform a gauge transformation
x ∈ S 1 . n is the winding number
A0 (x, t) → A0 (x, t) − ∂t ω(x, t), (A3) Z L
i
1 1
A (x, t) → A (x, t) + ∂x ω(x, t). (A4) n= dxgn−1 (x)∂x gn (x). (A12)
2π 0
By definition, A0 (x, t) is transformed to zero, while A1 Therefore, G = ⊕n Gn where Gn is a set of all possible
is not. Now that A0 = 0, the Euler–Lagrange equation gauge transformations with winding number n. a0 ∈
with respect to the variation of A0 becomes a constraint R is the coordinate of A/G0 , while a0 ∈ [0, 2π L ] is the
(namely, Gauss’s law), coordinate of A/G. Because of the identification by G,
the space A/G has a non-contractible loop. The wave
∂x Ex = −∂x Ȧ1 = 0. (A5) functional can have a non-zero phase when the gauge
field goes around this loop. This phase is not determined
Here and below, we use the simplified notation A = A1 . by the action.
The Lagrangian density in this gauge is The fact that a0 is defined only modulo 2π/L is evident
1 2 θ0 when we look at gauge-invariant observable such as the
L= Ȧ + Ȧ. (A6) Wilson loop in the spatial direction,
2e2 2π H
Now A is the dynamical variable in the theory and the W = ei A1 dx
= eia0 L . (A13)
canonical quantization condition is that
  It is clear that changing a0 → a0 + 2πL n does not change
1 the Wilson loop.
Ex = −Ȧ, A(x, t), 2 Ȧ(y, t) = iδ(x − y) , (A7) The “fully gauge fixed” a0 lives on a one-dimensional
e
circle, and the theory is identical to that of a quantum
and its wave functions ψ[A(x)] = ⟨A(x)|ψ⟩ are subject to particle on 1D circle with the action
the “physical state condition” Z Z  
1 2 θ0
S = dtdxL = L dt ȧ + ȧ0 . (A14)
∂x Ex |phys⟩ = 0. (A8) 2e2 0 2π
 17

1.0 Physical observables depend only on the combination

θ0 − θ. One way of understanding this is to consider ex-
pectation values of the Hamiltonian, which in this case,
0.8 as we discuss further below, is a function of the canonical
momentum operator Π = −i∂a0 in the specific combina-
tion Π − Lθ 2π . The boundary condition in (A17) implies
0

0.6 that eigenstates of the momentum operator have eigen-

H/(e2L)

values Πn = θL/(2π) + nL, for integer n ≥ 0, which

implies that eigenstates of the Hamiltonian only depend
0.4 on the combination θ − θ0 .
Below, we compute the transition amplitude
⟨af , tf |ai , ti ⟩ both with the canonical formulation
0.2 of quantum field theory and path integral formulation
to understand the consistency as well as how each
formulation handles the Lagrangian parameter and
0.0 the boundary condition. These results reinforce the
−3 −2 −1 0 1 2 3
(θ − θ0)/2π conclusion that physical observables only depend on
the combination θ − θ0 and cannot distinguish the two
contributions separately.
Figure 1. The eigenvalues of the Hamiltonian of 1+1D QED
on S 1 as a function of the angle θ −θ0 (black) with the ground Before turning to these calculations, however, let us
state indicated in red. comment briefly on how to generalize the lessons from
this example to QCD in 3+1D and the extra-dimensional
constructions, for example in 5D, that we discuss in the
It is clear that θ0 is analogous to the Aharonov–Bohm main text. First, let us understand the following conun-
phase. On the other hand, just like the quantum particle drum. To an observer living in 1+1D on a circle, we show
on a 1D circle, there is also an arbitrariness in gluing the above that there is a non-unique way of defining the pe-
wave function where a0 = 2π L (see, e.g., the discussions riodic boundary conditions for wavefunctions, character-
in [99, 100]),15 ized by the angle θ as given in (A17). Different values
  of θ lead to physically distinct observables, such as spec-
2π
ψ = eiθ ψ(0). (A17) tra, though θ is only observable in the combination θ−θ0 ,
L with θ0 the Lagrangian term. Now let us imagine that our
1+1D example arises from the dimensional reduction of
QED in 3+1D for particles that are localized on a circle.
15
In 3+1D Minkowski space the deficiency index for e.g.
The general boundary condition for a particle on a circle given
in (A17) can be understood through the following argument. the momentum operator is trivial, meaning that there is
Quantum operators should be self-adjoint rather than merely no ambiguity in defining boundary conditions for wave-
symmetric in order to give real values of observables and pre- functions in QED. How, then, do we understand where
serve unitarity under time evolution. However, some symme- the combination θ − θ0 arises from the UV perspective?
try operators admit multiple self-adjoint extensions, which are
The answer is that through the Aharonov-Bohm effect
quantified through the deficiency index, and which are repre-
sented by different, independent boundary conditions that pre- the combination θ − θ0 is proportional to the magnetic
serve the self-adjoint requirement. Consider the momentum op- flux through the circle on which our 1+1D QED lives.
erator P̂ = −i∂x in quantum mechanics on a circle of circumfer- The 1+1D observable is not able to access the interior of
ence L, with two states in the Hilbert space |ψ⟩ and |χ⟩. The this circle and so is unaware of the concept of magnetic
self-adjoint requirement enforces ⟨ψ|P̂ χ⟩ = ⟨P̂ ψ|χ⟩, which in po- flux; to this observer, θ0 is simply a Lagrangian parame-
sition space implies
Z L ter allowed by the theory and θ is a generalized boundary


dx ψ̄∂x χ + ∂x ψ̄χ = 0 . (A15) condition that is allowed by quantum mechanics. But to
0 the 3+1D observer who has access to the UV theory, the
Integrating by parts the equation above trivially vanishes, except physical combination θ − θ0 is simply set by the magnetic
for the boundary contributions, giving the self-adjoint require-
flux through the circle.
ment
Let us now apply this analogy to the theory of QCD
χ(L)ψ̄(L) = χ(0)ψ̄(0) . (A16)
in 3+1D and a UV completion that arises from the di-
This condition must be satisfied for all states in the domain. This mensional reduction of a 4+1D theory on a circle. Here,
requirement is not just satisfied by the normal periodic bound- we want to ask how the wave-functional ψ(A′ ) is related
ary condition but, more generally, by the boundary conditions
to ψ(A) for A′ that are given by gauge transformations
ψ(L) = eiα ψ(0) for any choice of α, so long as this boundary con-
dition is applied consistently to all states. The choice of bound- of the QCD vector potential A (here, our discussion mir-
ary condition can be thought of as a choice of domain (or a choice rors that of [100]). For A′ that are given by gauge trans-
of superselection sector) that is needed to make the momentum formations that are continuously deformable to the iden-
operator well defined on this compact space [99]. tity (i.e., topologically trivial gauge transformations), we
 18

must take ψ(A′ ) = ψ(A). On the other hand, for topolog- we derive the canonical momentum operator
ically non-trivial gauge transformations, which are char-  
1 θ0
acterized by a winding number n, we are allowed to take π̂ = L 2 ȧ0 + . (A18)
ψ(A′ ) = e−iθn ψ(A), in analogy to the 1+1D QED ex- e 2π
ample discussed above and the quantum mechanics of a Solving for ȧ0 , we find
particle on a ring (see [100]). This choice of domain is
precisely the choice of superselection sector in the theta e2 e2 θ 0
ȧ0 = π̂ − . (A19)
vacua, but just as in 1+1D QED θ only enters in com- L 2π
bination with the Lagrangian parameter θ0 for physical The Hamiltonian is
observables. 2
e2
  
Now, let us consider our 5D example, where we have 1 2 θ0 Lθ0
H = π̂ ȧ0 − L ȧ + ȧ0 = π̂ − .
non-abelian gauge theory in 5D that descends to QCD in 2e2 0 2π 2L 2π
4D by compactification on an S 1 of size R. At high ener- (A20)
gies (E ≫ R−1 ) spacetime looks locally like R4,1 , so that
the vacuum of the theory is characterized by π4 (SU (N )), As usual, the canonical momentum π̂ is represented as
which is trivial for N ≥ 3. Thus, there is no ambiguity ∂
in the UV in terms of the structure of the vacuum and π̂ = −i . (A21)
∂a0
also, as we have discussed, there is no theta term in the
Lagrangian in 5D. On the other hand, the 5D theory does A general solution to the boundary condition (A17) is
have a 5D CS term of the form A ∧ F ∧ F , and under di-  1/2
L
ei(n+ 2π )a0 L ,
θ
mensional reduction to 4D this can generate a theta term ψn (a0 ) = (A22)
in the Lagrangian if there are non-trivial fluxes across 2π
the S 1 . In this case, the combination θ + θ̄ in the IR
such that
EFT in 4D is precisely given by the magnetic fluxes of
  1/2
the 5D theory through the S 1 . This is in direct anal-

2π L
ei(n+ 2π )(a0 + L )L = eiθ ψn (a0 ).
θ 2π

ogy to our example discussed above of the dimensional ψn a0 + =

L 2π
reduction of QED in 4D to QED in 2D. The 4D theory (A23)
has a non-ambiguous vacuum, while the 2D theory has
an ambiguity in defining the vacuum along with a theta Therefore, the energy eigenvalues are
term—the combination of the two is physical. Through 2
e2 L θ0 − θ

the Aharonov-Bohm effect the combination of these two Hψn = n− ψn . (A24)
parameters in 1+1D is set by the magnetic flux through 2 2π
the spatial circle. 16 The Wilson loop operator takes one eigenstate to an-
other,
 1/2
L
ei(n+ 2π )a0 L
θ
2. Canonical and Path Integral Formulation of W ψn (a0 ) = eia0 L
QED in 1+1D 2π
 1/2
L
ei(n+1+ 2π )a0 L = ψn+1 (a0 ).
θ

Let us now return to the 1+1D QED example discussed =

2π
above to show explicitly that amplitudes depend on the (A25)
combination θ − θ0 , both for the canonical formulation of
quantum field theory and the path integral formulation. On the other hand, a “large gauge transformation” is
Starting with the canonical formulation, from the action given by the unitarity operator
U = e2πiπ̂/L . (A26)
It is easy to see
16 As a side-note, a crucial part of the arguments above is the com-
U ψn (a0 ) = e2πi(n+ 2π ) ψn (a0 ) = eiθ ψn (a0 ).
θ

pact nature of space, for example the compact circle in the 1+1D (A27)
example. In defining the winding numbers in QCD we also im-
plicitly compactify space at infinity. In contrast, the axial gauge The transition amplitude can be obtained readily as
introduced in Refs. [101, 102] cannot be defined on a compact
space; it must be formally infinite [103]. In this case, unsurpris- ⟨af , tf |ai , ti ⟩
ingly, there is no ambiguity in defining the wavefunction (i.e., no
X
superselection sectors for QCD). This result seems in contrast
= ψn (af )e−iEn (tf −ti ) ψn∗ (ai )
with the standard derivation in temporal gauge. We could not n
solve this puzzle. On the other hand, the Lagrangian still has a L X i(n+ 2π
θ 2 2
)(af −ai )L e−i e 2L (n− θ02π−θ ) (tf −ti ) .
theta parameter. In UV complete examples where θ + θ̄ is calcu- = e
2π n
lable there is no contradiction, since it is the combination θ + θ̄
that is set by the dynamics in the UV. (A28)
 19

Using the notation ∆θ = θ0 − θ, we obtain

L X i(n+ 2π
θ 2
)(af −ai )L e−i e 2L (n− ∆θ
2
2π ) (tf −ti ) .
⟨af , tf |ai , ti ⟩ = e
2π n
(A29)

Using T = tf − ti , we obtain

L X −i e22LT n2 +i((af −ai )L+e2 LT ∆θ

2π )n+i 2π (af −ai )L−i
θ e2 LT ∆θ 2
( 2π )
⟨af , tf |ai , ti ⟩ = e 2
2π n
L i 2π e2 LT ∆θ 2
e (af −ai )L−i 2 ( 2π ) ϑ(z; τ ),
θ
= (A30)
2π

where ϑ(z; τ ) is a Jacobi theta function We can also derive the path integral from (A30). We
need to work out its small T behavior. Using the modular
+∞ +∞
X 2 X 2 transformation,
ϑ(z; τ ) ≡ qn ηn = eπin τ +2πinz
. (A31)
n=∞ n=∞

z −1
 
Therefore for our case, ϑ ; = αϑ(z; τ ), (A34)
  τ τ
1 ∆θ 2
z= 2
(af − ai )L + e LT , (A32) α = (−iτ )1/2 eπiz /τ
, (A35)
2π 2π
e2 LT
τ =− . (A33)
2π we obtain

z −1
 
L i 2π e2 LT ∆θ 2 2
e (af −ai )L−i 2 ( 2π ) (−iτ )−1/2 e−πiz /τ ϑ
θ
⟨af , tf |ai , ti ⟩ = ;
2π τ τ
1/2 (a −a )2 L
z −1
  
L f i θ0
= 2
ei 2e2 T +i(af −ai )L 2π ϑ ; . (A36)
2πie T τ τ

For small τ in the upper half plane, the new q ′ = integral self-contained, we need to remove the boundary
e−πi/τ ≪ 1 and hence condition phase eiθ from the initial and final states. Then
we obtain the expression
z −1
 
ϑ ; = 1 + O(q ′ ). (A37) Z Rt h 2
θ0
i
τ τ θ
−i(af −ai )L 2π i t f dt ȧ2e2L +ȧL 2π
⟨af , tf |ai , ti ⟩ = e Da(t)e i

We find Z Rt h 2
i
i f dt ȧ2e2L +ȧL ∆θ
1/2 = Da(t)e ti 2π
. (A40)
(af −ai )2 L

L θ0
+i(af −ai )L 2π
⟨af , tf |ai , ti ⟩ ≃ ei 2e2 T
2πie2 T Thus, both in the canonical and path integral formula-
(A38) tions, amplitudes only depend on ∆θ.
for small T . Therefore, the path integral is given by
Z Rt h 2
θ0
i 3. An analogy to the Strong CP Problem
i f dt ȧ2e2L +ȧL 2π
⟨af , tf |ai , ti ⟩ = Da(t)e ti , (A39)
We may also identify a phenomenon in this theory
where the singular prefactor in (A38) is included into the analogous to the dependence of the neutron electric
path integral measure. Finally, in order to make the path dipole moment on the QCD vacuum angle. Let us now
 20

consider 1+1D QED coupled to massive matter particles After the center of mass motion is separated, we obtain
of charge e which we call electrons. If m ≫ e, the electron the Hamiltonian for the relative motion
is non-relativistic. Suppose there is a pair of an electron
and a positron. Their Lagrangian is
1 2 1 2
1 H= p + e |x| + E0 x ,
L = m ẋ21 + ẋ22 − A0 (x1 ) + a0 (x1 ) ẋ1

2µ 1 2
2
+ A0 (x2 ) − a0 (x2 ) ẋ2 . (A41) with µ = 12 m. With the term |x|, the elec-
tron and positron are confined. Yet when E0 =
Then the Gauss’ law constraint is modified to e2
(θ − θ = π, n = 0), it is deconfined in the x → −∞
2 0
e2
∂x E = −∂x2 A0 − ∂x ȧ0 = e2 δ (x − x1 (t)) − e2 δ (x − x2 (t)) , direction while when E0 = − 2 (θ0 − θ = −π, n = 0), it
is deconfined in the x → +∞ direction. More generally,
where x1 is the position of the positron and x2 that of E0 > 0 makes the confining potential steeper in positive
electron. The solution is x and shallower in negative x direction, and therefore the
center of the wave function is shifted to a negative value.
 1
It exhibits an electric dipole moment d = ⟨−ex⟩ ̸= 0.
(x2 − x1 ) x
 L 0 ≤ x ≤ x1
0 2
A (x) = e − (x − x1 ) + L1 (x2 − x1 ) x x1 ≤ x ≤ x2 ,
− (x2 − x1 ) + L1 (x2 − x1 ) x x2 ≤ x ≤ L

 4. Gauging parity in 1+1D QED
1
− L (x2 − x1 )
 0 ≤ x ≤ x1
2 1
E(x) = −ȧ0 + e 1 − L (x2 − x1 ) x1 ≤ x ≤ x2 QED on the torus T 2 is nothing but the partition func-
− 1 (x − x )

x2 ≤ x ≤ L tion
L 2 1
Z 2π/L
where we assume x1 < x2 . When x1 > x2 ,
Z = Tre−βH = da⟨a, tf |a, ti ⟩, (A44)
0

1
− L (x1 − x2 ) x
 0 ≤ x ≤ x2 where −β = −i(tf − ti ). We can also consider QED
0 2 1
A (x) = e (x − x2 ) − L (x1 − x2 ) x x2 ≤ x ≤ x1 on the Klein bottle K 2 , which is a non-orientable two-
(x − x ) − 1 (x − x ) x x ≤ x ≤ L

dimensional surface, where two ends of a cylinder are
1 2 L 1 2 1

1 identified after reversing the orientation. Namely that it
 L (x1 − x2 )
 0 ≤ x ≤ x1 is a torus with a parity twist. Therefore, we regard QED
E(x) = −ȧ0 + e2 −1 + L1 (x1 − x2 ) x2 ≤ x ≤ x1 . on K 2 as a theory on S 1 but take the partition function
 1 (x − x )

x1 ≤ x ≤ L with parity operator
L 1 2

The constant piece of the electric field depends on the Z− = TrP e−βH . (A45)
wave function (A22),
Then the total partition function with the gauged parity
Eψn (a0 ) = −ȧ0 ψn (a0 ) has a projection operator
θ0 − θ
 
2
= −e n − ψn (a0 ) = En ψn (a0 ) ; Z = Z+ + Z− = Tr(1 + P )e−βH . (A46)
2π
(A42)
Note that this result is analogous to the one-loop closed-
that is, the ground state energy is string contributions to the partition function of the Type
I string worldsheet theory discussed in B.
θ0 − θ
 
E0 = e2 . (A43) Since the parity operation can be done at any point
2π along the time axis with two patches overlapping, the
parity operator needs to commute with the Hamiltonian.
In the ground state, − 12 e2 < E0 < 12 e2 . We are inter- Namely that the Hamiltonian must be parity invariant.
ested in the limit |x1 − x2 | ≪ L → ∞. Then We first take the Lagrangian parameter θ0 = 0. We will
come back to the other possible case of θ0 = π later. The
Z
1 2 1 1 Hamiltonian eigenstates are (A22),
2
E dx = 2 E02 L + E0 (x2 − x1 ) + e2 |x2 − x1 | .
2e 2e 2  1/2
L
ei(n+ 2π )a0 L .
θ
Therefore, the electron and positron have the Hamilto- ψn (a0 ) = (A47)
2π
nian
1
1 To require the states survive under the projection 1 + P
H= p21 + p22 + e2 |x2 − x1 | + E0 (x2 − x1 ) . within the same Hilbert space, we must either take θ = 0
2m 2
 21

and have the Hamiltonian eigenstates or take θ = π and

 1/2  
L 1

L
1/2 ϕn (a0 ) = cos n + a0 L. (A50)
ϕ0 (a0 ) = , (A48) π 2
2π
 1/2 The sum is now reduced over the subsets.
L For the Hilbert space with θ = 0, the transition matrix
ϕn (a0 ) = cos(na0 L) (n ≥ 1), (A49)
π element is

∞
" #
L X 2
−i e 2LT n2
⟨af , tf |ai , ti ⟩ = 1+2 e cos(nai L) cos(naf L)
2π n=1
∞
" #
L 1 X −i e2 LT n2 in(af −ai )L in(a +a )L −in(a −a )L −in(a +a )L
= 1+ e 2 (e +e f i
+e f i
+e f i
)
2π 2 n=1
L 1 X −i e2 LT n2 −in(af −ai )L
= e 2 (e + e−in(af +ai )L )
2π 2 n
L 1
= (ϑ(z1 ; τ ) + ϑ(z2 ; τ )), (A51)
2π 2

where states related by Ω, e.g,

1 |a⟩L ⊗ |b⟩R , |a⟩R ⊗ |b⟩R (B1)
z1 = (af − ai )L, (A52)
2π
1 are considered equivalent, so in Type I Ω is a gauge sym-
z2 = (af + ai )L, (A53) metry. In addition to the closed unoriented strings, an
2π
e2 LT open string sector is required for tadpole and anomaly
τ =− . (A54) cancellation. Gauging Ω halves the Hilbert space, pro-
2π
jecting out states which are not invariant under Ω (e.g.,
For the Hilbert space with θ = π, a similar calculation the open string photon and the antisymmetric tensor
shows the transition matrix element is Bµν , which are Ω-odd; the dilaton and graviton survive).
In string perturbation theory, string S-matrix elements
L 1
⟨af , tf |ai , ti ⟩ = (ϑ10 (z1 ; τ ) + ϑ10 (z2 ; τ )). (A55) are computed via the Polyakov path integral, in which
2π 2 the sum over metrics and worldsheet topologies is orga-
For both cases, the result is indeed the average of that for nized by power counting in the string coupling gs , with
af and −af as expected from the operator 1+P inserted. the power set by the Euler characteristic χ of the world-
sheet topology:
X X Z
−Sstring −χ
Appendix B: Gauged worldsheet parity in Type-I Z= e ∼ gs DXDge−SPoly ,
string theory topologies topologies
metrics
(B2)
It is instructive to discuss an analogous example of a with Z the partition function, and
gauged spacetime inversion that is commonplace in string
theory: worldsheet-parity. The embedding of a funda- √
Z
1
mental string in spacetime is described by the worldsheet SPoly = T dσ 0 dσ 1 gg αβ ∂α X µ ∂β Xν (B3)
2 M
X µ (σ 0 , σ 1 ) with σ 1 the space-like coordinate along the
length of the string and σ 0 the time-like coordinate along is the Polyakov action (here T = 1/(2πα′ ) is the string
the string’s evolution. Type IIB string theory is a the- tension). The integration over worldsheet spacetime co-
ory of closed, oriented strings which is left-right symmet- ordinates X and worldsheet metrics g is modulo diffeo-
ric. That is, it is symmetric under the worldsheet parity morphisms and Weyl transformations (note that the tar-
transformation Ω : (σ 0 , σ 1 ) → (σ 0 , 2π − σ 1 ), which ex- get spacetime metric is fixed). For fixed χ there are a
changes left and right movers. Type I string theory is finite number of worldsheet topologies which contribute,
a theory of unoriented strings, and is obtained by mod- so the sum over topologies is perfectly well-defined and
ding out Type IIB by Ω (“orientifolding”), meaning that at low orders S-matrix elements can be easily computed.
 22

For example, in Type IIB the leading order (χ = 2) di- space. At tree-level and half-loop order, gauging world-
agram is S 2 , and the “one-loop” (χ = 0) contribution sheet parity therefore leaves the partition function un-
comes from the torus T 2 . changed, though the Hilbert space is still projected down
In Type I, the path integral is modified to include to Ω-invariant states (c.f. our previous discussion on
contributions from both orientable and non-orientable gauging spacetime parity in Sec. III).
worldsheets. Concretely, in the unoriented theory the
one-loop vacuum amplitude contribution from T 2 en-
codes closed strings evolving and gluing back to their Appendix C: The Witten-Veneziano relation
original state. The torus may be parametrized in the
complex plane by the modulus τ = τ1 + iτ2 with iden- Here we recall how the Witten-Veneziano formula (38)
tifications of the complex coordinate ω, ω ≃ ω + 2π ≃ for the mass of the η ′ can be derived via large N argu-
ω + 2πτ . The worldsheet theory in this setup is con- ments following the original work by Witten [83]. We
structed by considering a field theory on a spatial cir- stress that this derivation does not assume anything
cle with coordinate σ 1 = Re(ω), and then evolving about the order of limits in the partition function. We
σ 2 = Im(ω) for Euclidean time interval of length 2πτ2 , work in pure Yang-Mills, and assume the ’t Hooft limit,
while also shifting σ 1 by 2πτ1 , and then identifying the with λ = g 2 Nc fixed, such that the Lagrangian takes the
ends. The partition function is therefore a trace over form
the Hilbert space Hc of the closed string sector, which in  
1 µν θ µν
light-cone gauge takes the form [104–110] LYM = Nc − tr Fµν F + tr Fµν F̃ .
h i 2λ 16π 2 Nc
c c̃
ZT 2 = TrHc e2πiτ1 P −2πτ2 H = TrHc q L0 − 24 q̄ L̃0 − 24 ,
 
(C1)
(B4) The topological susceptibility is
with q = e 2πiτ ˜
and L0 (L0 ) are the left (right) moving Z
Virasoro zero-mode operators and c (c̃) are the left (right) χ(k) = d4 xeik·x ⟨q(x)q(0)⟩ , (C2)
moving central charges of the worldsheet conformal field
theory. Above, the momentum P = L0 − L˜0 and the
with q(x) = F F̃ , such that
Hamiltonian H = L0 + L˜0 − (c+c̃) 24 generate translations
1 2 2
along σ and σ respectively. Gauging Ω corresponds to d2 E

1
inserting the projector P = 12 (1 + Ω) = lim χ(k) . (C3)
dθ2 16π 2 Nc k→0
h i 1
Zclosed = TrHc P q L0 q̄ L̃0 = ZT 2 + ZK 2 , (B5) To leading order, the singularities of two-point function
2 of the topological charge density arise from glueballs and
such that there is ahcontribution from i the Klein bottle flavor-singlet pseudoscalar mesons,
c c̃
K 2 , ZK 2 = 12 TrHc Ωq L0 − 24 q̄ L̃0 − 24 , corresponding to
X a2n X b2n
closed strings evolving and gluing back to their original χ(k) = + , (C4)
state up to the action of Ω. The contributions in (B5) can k2 − Mn mesons N (k − m2n )
2 2
glueballs
be computed exactly (see e.g., [105, 106]). This discus-
sion is analogous to gauging parity in 1+1D QED, which with Mn (mn ) the glueball (meson) masses, and an (bn )
we discuss in Appendix A, where we explicitly compute the amplitudes for q to create these states from the vac-
the partition function of the theory on T 2 and K 2 . uum.
To account for the open-string sector in Type I string If massless quarks are introduced, then the topological
theory, the sum over worldsheets must include topolo- susceptibility must be zero, since a chiral rotation of the
gies with boundary. At “half-loop” order χ = 1 these quarks can make θ disappear, such that it cannot have
are the disk D2 and the real projective plane RP 2 , which any physical effects. However, quark loops contribute to
is unorientable. At one loop these are the cylinder and χ(k) only at order 1/Nc , whereas the gluons contribute
the Möbius strip. The vacuum amplitude contributions at order 1/Nc0 . The cancellation that must occur in (C4)
between the cylinder and the unorientable Möbius strip at k = 0 is between the leading order meson pole from
are related by parity projection similarly to the torus the η ′ and the glueball contribution (which is just the
and Klein bottle. It is natural to ask, given this pair- topological susceptibility of pure Yang-Mills χYM ). Note
ing between orientable and unorientable worldsheets at that this can only happen if m2η′ is of order 1/N . This
one-loop order associated to gauging parity, whether a cancellation tells us that
similar pairing happens at tree-level and half-loop order. b2η′
This cannot happen because there are an odd number of χYM ≡ χ(0)|Yang-Mills = . (C5)
N m2η′
topologies at these orders: S 2 , D2 , and RP 2 . The key
point is that these manifolds, unlike for those appear- The axial anomaly gives
ing at one-loop, do not have non-contractible loops, so
there are no Euclidean time-circles that allow the parti- p 8π 2 Nc µ ′
bη′ / Nc = ⟨0|F F̃ |η ′ ⟩ = ⟨0|∂µ JA |η ⟩ , (C6)
tion function to be written as a trace over the Hilbert λNf
 23

µ µ ′
with JA the axial current. Finally, using ⟨0|∂µ JA |η ⟩ = space-like separated and |x − y| → ∞,
µ µ ′
p ⟨0|JA |η ⟩ and
⟨α| A(x)B(y) |α⟩ = ⟨α| A(x) |α⟩ ⟨α| B(y) |α⟩
µ ′
p
⟨0|JA |η ⟩ = −i Nf fπ pµ (C7) 1
 
= ⟨θ1 | A(x) |θ1 ⟩ + ⟨θ2 | A(x) |θ2 ⟩
we arrive at the Witten-Veneziano relation 4
 
4Nf ⟨θ1 | B(y) |θ1 ⟩ + ⟨θ2 | B(y) |θ2 ⟩ .
m2η′ = χYM . (C8)
fπ2 (D4)
For a derivation of (38) including the dependence of χYM
Here we used that ⟨θ| A(x) |θ′ ⟩ ∝ δ(θ − θ′ ) for any gauge-
on the η and K masses, see Ref. [84]. Interestingly, since
invariant operator A. However, the relation above is
(C1) implies that the divergence of the axial current dis-
clearly not satisfied by the state in (D1), as we obtain
appears in the large Nc limit (as 1/Nc ), this relation sug-
gests that mη′ is not due to instantons, and instead comes Z
from fluctuations in the topological charge density. ⟨α| A(x)B(y) |α⟩ = ⟨α| A(x) |θ⟩ ⟨θ| B(y) |α⟩
θ
1
= ⟨θ1 | A(x) |θ1 ⟩ ⟨θ1 | B(y) |θ1 ⟩ (D5)
Appendix D: Superposition of θ states and cluster 2
decomposition principle 1
+ ⟨θ2 | A(x) |θ2 ⟩ ⟨θ2 | B(y) |θ2 ⟩ .
2
Here we argue that the vacuum of QCD cannot be a
superposition of |θ⟩ vacua, e.g, of the form As (D4) and (D5) differ, |α⟩ does not satisfy the clus-
ter decomposition principle and is therefore not a valid
|θ1 ⟩ + |θ2 ⟩ vacuum state.
|α⟩ = √ , (D1)
2
with θ1 ̸= θ2 . It is not clearly physically meaningful to Appendix E: Gauging non-normal subgroups of a
write a superposition of states which live in different su- symmetry group
perselection sectors; doing so would require the Hilbert
space to contain multiple |θ⟩ sectors. Suppose however Here we justify that the Z2 subgroup, corresponding to
that the Hilbert space does contain at least the sectors spacetime parity, of the full Lorentz group in d spacetime
|θ1 ⟩ and |θ2 ⟩. In this case, we may first note that un- dimensions O(1, d−1) cannot be gauged while preserving
der large gauge transformation the state (D1) does not Lorentz symmetry unless the full O(1, d−1) is gauged, as
transform by an overall phase, pointed out in Ref. [47]. Note that this implies that gaug-
ing the 4D P or CP requires gravity (which gauges the
eiθ1 |θ1 ⟩ + eiθ2 |θ2 ⟩ Lorentz symmetry, as we explain in Section IV). In fact
U |α⟩ = √ , (D2)
2 a more general claim holds 17 : if H is a non-normal sub-
group of a symmetry group G of a theory, then gauging
so (D1) breaks gauge invariance of the vacuum state. H is not possible without breaking the (global) symmetry
Though unusual, this is not so problematic as correla- G, unless G is gauged.
tion functions of gauge-invariant operators A are gauge-
To see why this holds, note that a global symme-
invariant; e.g, vacuum expectation values decompose as
try transformation must send gauge-equivalent states to
1 1 gauge-equivalent states. Let us consider the space H of
⟨α| A |α⟩ = ⟨θ1 | A |θ1 ⟩ + ⟨θ2 | A |θ2 ⟩ . (D3) states (where we have not yet declared states related by
2 2
gauge transformations to be equivalent), and for brevity
What is more problematic is that (D1) being the vac- we will write operators U (g) transforming in a represen-
uum is inconsistent with locality, and in particular the tation of G by the associated group element g ∈ G. For
cluster decomposition property, as we now show. We de- any g ∈ G, h ∈ H, and state |ψ⟩ ∈ H, g|ψ⟩ must be
note the eigenstates of the Hamiltonian as |θ⟩. They form gauge equivalent to gh|ψ⟩. That is, there exists some
′ ′
an orthonormal basis and thereforeR ⟨θ|θ ⟩ = δ(θ − θ ) and h′ ∈ H such that h′ g|ψ⟩ = gh|ψ⟩. As this reasoning
the identity decomposes as 1 = θ |θ⟩ ⟨θ|. To preserve the must hold for any other arbitrary state |ψ ′ ⟩ as well as for
locality of the theory, the cluster decomposition principle the sum |ψ⟩ + |ψ ′ ⟩, h′ must the same for every state |ψ⟩,
tells us that given two operators A(x) and B(y) invari- so h′ g = gh. Hence H is stable under conjugation by G,
ant under SU (3) gauge transformations, with x and y meaning H must be normal.

17 We thank Matthew Reece for discussions on this point.