The EFT of Large Spin Mesons

arXiv:2412.08691v1 [hep-th] 11 Dec 2024

Gabriel Cuomo,a,b Sergei Dubovsky,a,c Guzmán Hernández-Chifflet,d Alexander Monin,e

and Shahrzad Zarea
a
Center for Cosmology and Particle Physics, Department of Physics, New York University, New York,
NY 10003, USA
b
Department of Physics, Princeton University, Princeton, NJ 08544, USA
c
Institute for Advanced Study, Princeton, NJ 08540, USA
d
Instituto de Fı́sica, Facultad de Ingenierı́a, Universidad de la Republica, J.H.y Reissig 565, 11300
Montevideo, Uruguay
e
Department of Physics and Astronomy, University of South Carolina, Columbia, SC 29208, USA
E-mail: [email protected], [email protected], [email protected],
[email protected], [email protected]

Abstract: We use effective string theory to study mesons with large spin J in large Nc QCD
as rotating open strings. In the first part of this work, we formulate a consistent effective field
theory (EFT) for open spinning strings with light quarks. Our EFT provides a consistent
treatment of the endpoints’ singularities that arise in the massless limit. We obtain results, in
a systematic 1/J expansion, for the spectrum of the leading and daughter Regge trajectories.
Interestingly, we find that the redshift factor associated with the quarks’ acceleration implies
that the applicability regime of the EFT is narrower compared to that of static flux tubes.
In the second part of this work, we discuss several extensions of phenomenological interests,
including mesons with heavy quarks, the quarks’ spin and the daughter Regge trajectories
associated with the worldsheet axion, a massive string mode identified in lattice simulations
of 4d flux tubes. We compare our predictions with 4d QCD spectroscopy data, and suggest
potential stringy interpretations of the observed mesons. We finally comment on the relation
between the EFT spectrum and the Axionic String Ansatz, a recently proposed characterization
of the spectrum of Yang-Mills glueballs.
Contents

1 Introduction 1
1.1 Summary of Part I 4
1.2 Summary of Part II 6

I The EFT with light quarks 8

2 Lightning review of effective string theory 8

2.1 The effective theory of long strings 8
2.2 EST from lattice data 10

3 Endpoint singularities for spinning strings 11

4 Effective endpoints: power-counting and results 13

4.1 Warm-up: massless scalar field on a non-dynamical rotating string 14
4.2 The general boundary action 18
4.2.1 The construction 18
4.2.2 Alternative parametrization and the effective string length 21
4.3 Fluctuations 24
4.3.1 Leading order analysis 24
4.3.2 Subleading orders: massless endpoints’ scheme 27
4.3.3 Subleading orders: masssive endpoints’ scheme 29
4.4 Comments on one-loop corrections 31
4.5 Different endpoints 33

5 Endpoint renormalization from bulk operators 35

5.1 Warm-up: massless scalar field with curvature coupling 35
5.2 Endpoints’ renormalization in the dynamical string: an example 38

II Extensions and phenomenology 39

6 Additional degrees of freedom in 4d 39

6.1 Endpoints with spin 40
6.2 Worldsheet axion states on the rotating string 42

–i–
7 Heavy quarks 45
7.1 Double-scaling limit 45
7.2 String excitations and quantum effects in the non-relativistic limit 48
7.3 Quarks’ spin and spin-orbit inversion 51

8 A stringy view on meson spectroscopy 51

8.1 Discrete quantum numbers 52
8.2 Light mesons 54
8.3 Comments on other mesons 60

9 Beyond EFT: inequivalent quantizations and glueballs in 3d 61

Appendix 67

A Geometric identities and notation 67

A.1 Bulk geometry 67
A.2 Boundary geometry 68

B Near boundary expansion with the (K 2 )2 term 69

C Bosonic action for a point particle with spin 71

C.1 Coadjoint orbit through spinors 71
C.2 Spin from Goldstones 73

D Worldsheet axion energy levels from the Mathieu equation 76

E Daughter Regge trajectories for heavy quarks 77

F Tables of mesons 79

1 Introduction

It is believed that large N confining gauge theories, such as SU (N ) Yang-Mills or QCDN ,

admit a dual formulation in terms of a weakly coupled string theory. The concept of confining
strings dates back to the observation that the lightest hadronic resonances organize into
families following the scaling relation M 2 ∝ J. This empirical relation inspired early string
models, which ultimately evolved into the modern framework of fundamental string theory
aiming to describe all particles and interactions.
Despite decades of research, our understanding of the worldsheet dynamics of confining
strings remains incomplete. Nonetheless, recent years have witnessed substantial progress,

–1–
Figure 1: Left: Spin J meson as a rotating open string. Right: Spin J glueball as a rotating closed
string.

driven by a synergy of theoretical insights and lattice simulations.1 Much of this progress
concerns the dynamics of long, static flux tubes in SU (N ) Yang-Mills theory at large N , both
in d = 3 and d = 4 spacetime dimensions (see [4] for a recent account).
In particular, lattice measurements have confirmed the validity of the effective string theory
(EST) description of long strings in terms of d−2 Goldstones [5–7], providing quantitative values
for the basic parameters of the worldsheet theory (such as the string tension). Additionally,
Monte-Carlo simulations provide a glimpse on the UV dynamics of the flux tube beyond
effective field theory.
Surprisingly, all observed string states in three-dimensional SU (N ) gauge theories are
well described by a single Goldstone, with no additional massive string degrees of freedom.
In contrast, in four-dimensional Yang-Mills theory, lattice data revealed the existence of a
massive pseudoscalar mode on the worldsheet, referred to as the “worldsheet axion”, besides
the two Goldstones [8, 9]. Both lattice measurements and theoretical arguments hypothesis
that no additional degrees of freedom exist on the worldsheet of the confining string of SU (N )
Yang-Mills theory in d = 3 and d = 4.2
Given these advancements in our understanding of confining strings, it is natural to revisit
the calculation of the hadron spectrum in large N confining theories from the dual string
perspective. This endeavor is further motivated by both experimental data from QCD and
recent lattice simulations of the glueball spectrum in 3d and 4d Yang-Mills theory [11–13].
In particular, identifying signatures of the 4d worldsheet axion in these datasets would be a
fascinating development.
A natural starting point for this analysis is particles with large spin J. Mesons and
1
There has also been recent progress in the bootstrap approach to the QCD flux tube, see e.g. [1–3].
2
Intriguingly, there exist integrable Lorentz-invariant worldsheet S-matrices precisely in terms of a single
massless mode in three dimensions, and in terms of two Goldstone modes and a massless worldsheet pseudoscalar
in four dimensions [10]. While the data show that the Yang-Mills string is only approximately described by
such integrable theories, one might speculate that the UV dynamics of the flux tube restores integrability at
high energies for certain inclusive observables, potentially reflecting the asymptotic freedom of partons.

–2–
glueballs are described by, respectively, open and closed rotating strings. The string stretches
between a quark and an anti-quark for mesons, and it is folded between two cusps for glueballs
(see fig. 1). As the angular momentum increases, the string length grows, inducing thus
√
a parametric separation between its size L ∼ J ℓ̄s and the inverse confinement scale ℓ̄s .
This separation, similarly to long flux tubes, enables the systematic calculation of the mass
spectrum of the theory, in the vicinity of the leading Regge trajectory, via a semiclassical
expansion in powers of ℓ̄2s /L2 ∼ 1/J.3 The primary goals of this work are to provide a
systematic formulation of the effective field theory for rotating open strings (in any d > 2),
and to compare its predictions with QCD meson spectroscopy data.
The effective field theory (EFT) argument outlined above offers a modern explanation
for the mass relation M 2 ∼ J/ℓ̄2s obeyed by mesons on the leading Regge trajectory.4 The
spectrum of daughter Regge trajectories is similarly obtained by quantizing the worldsheet
Goldstone modes. However, the systematic calculation of 1/J corrections to the spectrum of
the leading and daughter Regge trajectories is hindered by a technical challenge: the presence
of singularities in the worldsheet metric. These singularities are localized at the cusps for
closed strings, and at the string endpoints for mesons composed of massless quarks. Even for
mesons made of light, but not necessarily massless quarks, the derivative expansion breaks
down near the string endpoints due to their large acceleration.
In the first part of this paper we provide a consistent EFT treatment of the string endpoints’
singularities for mesons composed of light quarks. To this aim, we augment the bulk EST
action with a suitable boundary action that describes the endpoints. The Wilson coefficients
of this boundary action absorb the singular contributions of bulk operators, ensuring that the
derivative expansion remains well-posed. Our analysis is largely inspired by [16, 17], which first
addressed this issue focusing on the leading Regge trajectory. Here we extend the framework
of those works to the calculation of the spectrum of daughter Regge trajectories.
A surprising outcome of our analysis is that the effective field theory cutoff is parametri-
cally smaller than the naive expectation, ∼ ℓ̄−1 s . This phenomenon arises because the high
acceleration of the quarks causes frequencies at the string endpoints to appear significantly
redshifted relative to those at the string’s center. Consequently, the derivative expansion at
the boundary is formulated in terms of this reduced cutoff, and the expansion for the mass
√
spectrum progresses in inverse powers of 1/ J rather than 1/J. A more detailed summary of
the results of Part I of this paper is given below in Section 1.1.
In the second part of this paper, we explore several extensions of the effective theory that
are of phenomenological interest: the inclusion of quark spin in the EFT, the daughter Regge
trajectories associated with the worldsheet axion, and the extension of our approach to mesons
containing heavy quarks. We then compare our predictions with QCD spectroscopy data from
the Particle Data Group (PDG) [18], focusing on mesons with light quarks. From fits of the
leading Regge trajectories, we estimate the leading boundary Wilson coefficient for different
3
Recall that string splitting effects are negligible for large number of colors.
4
Under some technical assumptions, it may be proven that the relation M 2 ∝ J is a universal result for
particles on the leading Regge trajectories in large N theories [14, 15].

–3–
quark families. We also suggest potential interpretations of the observed mesons beyond the
leading Regge trajectory in terms of states predicted by EST (extrapolating to low values of
J); fascinatingly, we suggest a candidate for a worldsheet axion excitation, though further
experimental data are required to confirm or disprove this interpretation.
Finally, we close this work commenting on the relation between the effective field theory
spectrum and the Axionic String Ansatz (ASA) [19, 20], a recently proposed recipe that
successfully interprets (and predicts) the quantum numbers of all the glueballs measured on
the lattice in 3d Yang-Mills theory. A more detailed account of the results of this work’s Part
II can be found in Section 1.2 below.
An important next step is to extend our approach to glueballs. While the physical picture
is similar for mesons and glueballs, with the role of the endpoints replaced by the cusps for
glueballs, the closed string setup seems to pose additional technical challenges. In particular,
we expect that a systematic treatment of the cusps will require more Wilson coefficients than
the case of open string endpoints. We are currently investigating this issue. An important
concrete target is the calculation of the energy splittings between particles just above the
leading Regge trajectory, which have been measured on the lattice for glueballs with angular
momentum up to J = 4 in 3d Yang-Mills theory [11, 12]. It would also be interesting to
identify axion excitations in the recent simulations of the mass spectrum of 4d glueballs in
SU (N ) gauge theory. We plan to report on progress in these directions in the future.
In this work, we focus on classical corrections to the mass spectrum of large J particles.
It would also be valuable to analyze quantum corrections to the EFT spectrum of rotating
strings. While these corrections are subleading compared to the leading contributions from
the string endpoints, they are particularly interesting because they do not depend on new
Wilson coefficients. For the leading Regge trajectory, the quantum Regge intercept for mesons
was computed in [16], but performing similar calculations for the daughter Regge trajectories
(and for closed strings) does not appear to be entirely straightforward.
Finally, it would be interesting to identify sufficiently inclusive observables on the rotating
string that are not sensitive to the physics near the string endpoints and are thus unaffected
by the redshift of the cutoff described above. Such quantities should allow for more precise
EFT predictions, purely in terms of the Wilson coefficients of the bulk EST action that have
been measured in simulations of static flux tubes.

1.1 Summary of Part I

Effective String Theory (EST) describes the low energy dynamics of the chromodynamic
flux tube when this is much longer than the length scale associated with confinement ∼ Λ−1QCD .
The EST action is organized as a series in terms of geometric invariants of the worldsheet
metric, with the leading term given by the Nambu-Goto (NG) action.
In the first part of this work we use EST to study mesons in large Nc QCD-like theories
with massless or light quarks in d = 3 and d = 4. We think of a meson as composed of a flux
tube stretching between a quark and an anti-quark. We ignore string breaking effects, that
are suppressed at large Nc for a finite number of flavors. As it is well known, the length of a

–4–
 δ δ

Figure 2: Graphical representation of a meson as a rotating open string with effective

endpoints of size δ.

classical rotating NG string is proportional to the square root of its angular momentum J, see
fig. 2. Therefore effective string theory (EST) should allow for the calculation of the spectrum
of mesons in a 1/J expansion. Famously, this logic predicts the existence of Regge trajectories
satisfying M 2 ∝ J.
To successfully implement this program, we need to address an important technical
point: the effective theory close to the string endpoints, which is famously a subtle issue.
For instance, when the quarks are taken to be massless, the worldsheet metric at the string
endpoints becomes singular (since the projection onto the quark worldline is null), and all
higher derivative operators in EST diverge at the endpoints.5 More generally, for light quark
masses m ≲ ΛQCD the derivative expansion breaks down close to the endpoints, and all
higher derivative operators contribute at the same order in the 1/J expansion to the mesons’
spectrum–seemingly defying the derivative expansion.
The physical origin of the apparent breakdown of the derivative expansion is that, within
EFT, we cannot exactly resolve the positions nor the velocities of the quarks. Rather, we
need to supplement the bulk EST with a suitable effective boundary action that describes
the endpoints in a multipole expansion. Physically, this is similar to the worldline EFT
description for black holes as seen from afar [21]. The Wilson coefficients of the boundary
action naturally depend on a cutoff δ ∼ Λ−1 QCD , representing the size of the effective quarks.
Within this EFT, the contributions from bulk operators that are singular in the limit δ → 0
can simply be understood as a renormalization of the boundary coefficient of the endpoints’
action. Once these singular contributions are reabsorbed in the boundary Wilson coefficients,
higher derivative bulk operators contribute to observables according to the expected EFT
counting.
The formulation and study of the large spin mesons EFT for light quarks, including the
endpoints action, is the main topic of the first part of this work. We find that the masses of
the particles on the leading and daughter Regge trajectories can be computed in terms of
5
For fundamental strings, for which we only need the Nambu-Goto action, the exact quantization of the
theory resolves these divergences.

–5–
the string tension 1/ℓ2s ≡ 1/(2π ℓ̄2s ), and two boundary Wilson coefficients µ1 and µ2 , up to
relative order ∼ J −3/2 . For instance, the mass of the leading Regge trajectory is given by:
1 2µ1 1/4 1 2µ2 −1/4 3 2 −1/2
M2 = J + J − + J + µ1 J + ... . (1.1)
ℓ̄2s ℓ̄s ℓ̄2s ℓ̄s 4
Daughter Regge trajectories are associated with excitations of the string and have a gap
√
≳ 1/L ∼ ℓ̄−1
s / J on top of the leading trajectory. We provide explicit results for the spectrum
in the main text.
Interestingly, from the study of daughter Regge trajectories we find that we can reliably
compute the mass of only those mesons whose gap above the leading Regge trajectory is
smaller than ΛQCD /J 1/4 , rather than the expected cutoff ΛQCD . This is due to the redshift
of the proper time at the location of the effective quarks with respect to the time coordinate
at the center of the string:
dtquark ∼ J −1/4 dtbulk , (1.2)
which is a result of the large acceleration. Therefore, the local cutoff at the endpoints is
smaller than in the bulk by a factor of J 1/4 . In other words, Poincaré’ invariance implies that
the boundary action, is organized in a derivative expansion with cutoff Λb.dry ∼ ΛQCD /J 1/4 ,
when written in terms of the bulk time coordinate.
We remark that, once the boundary singularities are renormalized away, higher derivative
bulk operators are always at least O(J −2 ) suppressed, and therefore contribute at quite
subleading order compared to the boundary Wilson coefficients.
The results of the first part of this work agree with former investigations of large spin
mesons’ in the literature [16, 22–28]. In particular, Hellerman and Swanson formulated the
endpoints EST and derived eq. (1.1) working in the Polyakov formalism in [16, 27]. We instead
work in Nambu-Goto formalism, which does not require dealing with constraints and allows
obtaining results for the daughter Regge trajectories as well in a straightforward manner. We
also note that EFTs with nontrivial space-dependent local cutoffs were recently analyzed in
the context of finite density systems [29–31].

1.2 Summary of Part II

In the second part of this work we discuss extensions and applications of the EFT developed
in the first part that are of phenomenological interest.
First we discuss how to describe quarks’ spin in d = 4 EFT. We find that, generically,
states with spin wave-functions that are different from those of the leading order Regge
trajectory have a large gap, of the order of the boundary cutoff. Intuitively, this result can be
understood as a consequence of the Thomas’ spin precession, which results from the quarks’
accelaration; furthermore, interactions with the string renormalize the spins’ gap at leading
order. We conclude that within EFT, we cannot reliably analyze the energy splitting between
different spin states, and the spin variables must be integrated out.
We then analyze the daughter Regge trajectories associated with the QCD worldsheet
axion. Massive worldsheet particles on the rotating strings localize near the endpoints of

–6–
the string due to a centrifugal force. We argue that as a consequence of the centrifugal
barrier, the wave-functions of the corresponding modes are approximately localized near the
two endpoints, and the axion’s single particle spectrum organizes in a set of approximately
doubly degenerate states on top of the leading Regge trajectory (at large J). We also point
out a puzzle concerning the boundary conditions at the string endpoints, that are needed to
compute the mass spectrum of daughter trajectories associated with the worldsheet axion.
Contrary to the EFT expectation, based on genericity arguments and symmetries, simulations
of open static flux tubes suggest that Neumann boundary conditions provide a more accurate
descriptions of the observed data [32, 33]. If this conclusion were to extend to the rotating
string setup, it would suggest that the daughter trajectories of the worldsheet axion may be
relatively light at moderate J. Due to the physical relevance of this point, in the future it
would be interesting to analyze further the axion boundary conditions, perhaps extending the
flux tube bootstrap of [1–3] to include the boundary reflection S-matrix.
Afterwards, we generalize the mesons’ EFT to the conceptually simpler case of heavy
quarks. We argue that the EFT is properly formulated as a double-scaling limit J → ∞
with J/(m2 ℓ̄2s ) = fixed, where m is the quarks’ mass. We find that pfor the flux tube to be
sufficiently long and EST to apply, we need angular momenta J ≫ mℓ̄sp . We show that the
particle’s spectrum in the non-relativistic regime, which corresponds to mℓ̄s ≪ J ≪ m2 ℓ̄2s ,
can be computed by semiclassically quantizing a simple quantum-mechanical potential model,
and we obtain analytical results for the quantum corrections to the leading Regge-trajectory in
the heavy quark regime. We finally analyze the quarks’ spins, which are within EFT for heavy
endpoints, and provide predictions for p the gap of the corresponding states. In particular, we
predict that in the EFT regime J ≫ mℓ̄s , the state with minimal energy among the different
spin wavefunctions is the one in which the quarks’ spins are aligned with the orbital angular
momentum, therefore rigorously establishing the so-called spin-orbit inversion scenario [34].
We then compare our findings with real world mesons spectroscopy data from the Particle
Data Group (PDG) [18], focusing on light quark mesons. We estimate the string tension
and the coefficient µ1 (for u, d and s quarks) fitting the formula (1.1). For daughter Regge
trajectories, the scarcity of data points for J ≳ 2 does not allow for a quantitative comparison.
We nonetheless discuss potential interpretations of the observed data in terms of states
predicted by EST extrapolating to low values of J. To this aim, we discuss how to compute
parity and charge conjugation assignments of the particles within EFT. To discard states that
do not project onto open strings at Nc → ∞ (such as glueballs and molecules of two mesons),
we compare different isospin channels and identify a small subset of states that may be reliably
interpreted as mesons. Intriguingly, we argue that one of the observed states with J = 2 might
correspond to a worldsheet axion excitation on top of the leading Regge trajectory. Future
experimental analyses of J = 2 and J = 3 mesons, perhaps along the lines of [35], might prove
or disprove this interpretation.
We finally comment on some subtleties that arise when one attempts to extrapolate the
EFT Hilbert space to low values of J, even when one assumes that the worldsheet does not
admit any additional fields besides those included in EFT (as lattice data suggest that is the

–7–
case in d = 3). Intuitively, the subtleties arise because knowing the dimensions and geometry
of the phase space at large J near the leading Regge trajectory is not enough to constrain the
shape and topology of the phase space at all values of J. Our discussion is motivated by a
recently proposed recipe to reproduce the spectrum of closed effective strings – the axionic
string Ansatz (ASA) – that qualitatively matches lattice data for glueballs in d = 3 all the
way down to J = 0 and disagrees with other quantization prescriptions such as lightcone
quantization [19]. Our discussion illustrates the physical origin of the discrepancies between
different quantization procedures in the simpler context of open strings.

Part I
The EFT with light quarks
2 Lightning review of effective string theory

2.1 The effective theory of long strings

As is well known, both in d = 3 and d = 4, in the limit where the length of the compact
dimension is much larger than the inverse QCD scale, inserting a Polyakov loop in Yang-Mills
(YM) theory creates a long, closed, string-like object: the QCD flux tube.
Any string-like object in field theory spontaneously breaks the d-dimensional Poincaré
group ISO(d-1,1) down to ISO(1, 1) × SO(d − 2), where the first factor represents the 2d
Poincaré group, and the second one corresponds to rotations in the transverse directions to
the string.
Therefore, the low-energy dynamics of a flux tube of length L, much larger than the
size of the lightest glueball m−1g , is universally described in terms of the d − 2-dimensional
Goldstones associated with the broken translations.6 The corresponding effective theory is
usually referred to as effective string theory (EST).
Let us review the construction of the EST action in the Nambu-Goto formalism. For
clarity, we consider an infinitely long string.
The action for a string is written in terms of the coordinates X µ of the string worldsheet’s
embedding into the target space, which are conveniently parametrized as follows in the static
gauge:
X µ = τ, σ, X i (τ, σ) ,


(2.1)
where σ α = (τ, σ) are the worldsheet coordinates, and X i denotes the transverse string
coordinates, which are the d − 2 dynamical Goldstone fields. Our conventions and some useful
geometric identities are summarized in appendix A.
6
Note that no additional Goldstones are required to account for the breaking of boosts and rotations; such
a mismatch between the number of broken generators and Goldstones is common for spontaneously broken
spacetime symmetries, see e.g. [36, 37].

–8–
 The action is constrained by the reparametrization invariance of the worldsheet coordinates
and at the leading order it is given by the well known Nambu-Goto (NG) term:

1
Z √ h 
µ 4
i
Sbulk = − 2 d2 σ G 1 + O ℓ̄4s (Kαβ ) , (2.2)
ℓs

where 1/ℓ2s is the string tension and G = − det(Gαβ ) is the determinant of the induced metric
defined by
Gαβ = ∂α X µ ∂β X ν ηµν , (2.3)
where ηµν is the spacetime metric in mostly minus signature. For future convenience, we also
define the (reduced) string length ℓ̄s as
1 1
2
= . (2.4)
ℓs 2π ℓ̄2s

Note that 1/ℓ¯2s coincides with the α′ parameter in fundamental string theory; more generally,
we expect that the reduced string length parametrically coincides with the mass of the lightest
glueball ℓ̄−1
s ∼ mg , and thus sets the cutoff of the theory according to generalized dimensional
analysis [38].
Higher derivative contributions to the action (2.2) are given in terms of curvature invariants
and, because of the Gauss-Codazzi relation, they can all be written in terms of the second
fundamental form of the worldsheet,
µ
Kαβ = ∇α ∂β X µ , (2.5)

and derivatives thereof. It turns out that, upon using the equations of motion (EOM) to
remove redundant operators and discarding total derivatives, the first nontrivial terms arise at
fourth subleading order in derivatives. There is therefore a high degree of universality in the
low energy predictions of EST. It is also straightforward to include additional, non-Goldstone,
fields in the action (2.2). These can be analyzed consistently within EFT as long as they are
sufficiently light and weakly-coupled.
As is well known from the study of fundamental strings, the NG action is integrable
at the classical level. Integrability is broken by quantum effects away from the critical
dimensions d = 26 and (somewhat surprisingly) d = 3. Nonetheless, since the contribution of
loops is subleading in the derivative expansion, the low energy dynamics of the flux tube is
approximately integrable. As explained in [8, 39], this property can be used to improve the
convergence of the EFT results for the spectrum of flux tubes wrapping a nontrivial space
cycle of length L.

–9–
2.2 EST from lattice data
Let us now review what is known about EST in Yang-Mills theory from lattice data.
First of all, the value of ℓ̄s can be expressed in terms of the mass mg of the lightest glueball,
comparing the mass spectrum with the ground-state energy of a long string. The result in
d = 3 [11] and d = 4 [13] reads, in the Nc −→ ∞ extrapolation,

mg |d=3 = 1.646ℓ̄−1
s , (2.6)
mg |d=4 = 1.229ℓ̄−1
s . (2.7)

In d = 3 there is a unique higher derivative operator at fourth order in derivatives, which

reads: Z √ Z √ h µ ν  i2
d σ GR = d2 σ G Kαβ
2 2
Kγδ − Kαγµ
Kβδν
Gαβ Gγδ ηµν . (2.8)

The coefficient of this term was measured by comparing its contribution to the string fluctuation
S-matrix with the phase shift extracted via the Lüscher method from lattice calculations of
the spectrum of strings of finite length L. Rather surprisingly, it turns out that, using the
latest available data [40, 41] (which go up to Nc = 8), the value of the measured coefficient
is zero (within the uncertainties) in dimensional regularization in the minimal subtraction
scheme [39, 42].
In d = 4, lattice data surprisingly revealed the existence of a pseudoscalar resonance on
the worldsheet [8], which we shall denote a. While its mass ma is of the order of the cutoff ℓ̄−1
s ,
it turns out that its width is quite small: Γa /ma ≃ 0.21ma for Nc = 3. The basic properties
of this mode (at least at the qualitative level) can be understood by including a pseudoscalar
field in the EST and pretending that its mass ma is much smaller than the cutoff, as if a were
endowed with an approximate shift symmetry broken by effects of order (ma ℓ̄s )2 . The action
of this field, including the first nontrivial coupling to the string coordinates, reads
√ m2a 2
Z  
2 1 2 Qa
Sa = d σ G (∂a) − a + aK · K̃ . (2.9)
2 2 32π

The first two terms are the free theory ones. Because of parity, the first nontrivial operator to
which a can couple is

εαβ ρ εγη
K · K̃ = √ ∂α X µ ∂β X ν εµνρσ Kγδ σ
Kηλ Gδλ √ . (2.10)
G G
1
√
The integral of 8π GK · K̃ is a topological invariant that counts the sign-weighted self-
intersection number of the string worldsheet [43, 44]. Therefore the coupling between a and
K · K̃ is analogous to the QCD axion coupling - for this reason a is sometimes referred to
as a worldsheet axion. This coupling yields the leading contribution to the width of a. The
numerically measured values of ma and Qa for Nc = 3 are

ma = 0.74ℓ̄−1
s , Qa = 0.37 ; (2.11)

– 10 –
these numbers do not appear to change significantly in the large Nc limit [9]. Notice that ma
is slightly below the mass of the lightest glueball, which partially justifies its inclusion within
the EFT.7

3 Endpoint singularities for spinning strings

Before presenting a proper formulation of the endpoints’ EFT, the goal of this section is to
explain the breakdown of the bulk derivative expansion for spinning strings, that we alluded
to in the introduction. To this aim, it is simplest to consider the explicit calculation of the
leading Regge trajectory. We consider the same bulk action as in the previous Section (2.2).
Here, the spatial worldsheet coordinate runs over a finite range:

σ ∈ [−1, 1] . (3.1)

For now, we do not commit to any specific boundary action at the endpoints, that we assume
to be identical. In arbitrary d ≥ 3 spacetime dimensions, the classical profile describing a
spinning string is:
1
Xµ =


τ, αf σ cos τ, αf σ sin τ, 0, 0, . . . . (3.2)
ω | {z }
d−3 times

This is guaranteed to be a solution of the bulk EOMs to all orders in the derivative expansion.
This is because eq. (3.2) is the most general ansatz invariant under translations in the transverse
directions and the linear combination H − ωJ of time translations and rotations in the first
two spatial coordinates. This profile corresponds to the following background metric:
1 − σ 2 αf2 dσ 2
ds2 = dτ 2 − . (3.3)
ω2 ω2
In eqs. (3.2) and (3.3), ω is the angular velocity and we work in units such that τ and σ are
dimensionless. The parameter αf in eq. (3.2) is the velocity of the endpoints and depends
upon the precise boundary conditions. It determines physical length of the string L, which is
given by Z p 2αf
L = dσ Gσσ = . (3.4)
ω
We see that the string is long for small angular velocity ω ℓ̄s ≪ 1; as we review below,
somewhat counter-intuitively this is precisely the regime that describes a string with large
angular momentum. For a pure NG action with free endpoints one finds αf = 1; we will see
that αf ≃ 1 up to small corrections, even for more general boundary conditions in the large
spin limit.
To obtain the mass of the leading Regge trajectory, it is convenient to compute the
R
Lagrangian L(ω) on the background profile (3.2), where S = dtL is the bulk action (2.2) and
7
Notice that eqs. (2.6) and (2.7) parametrically agree with the pi-ology expected from generalized dimensional
analysis [38], according to which the UV cutoff ∼ mg is expected to be set by ℓ̄−1s , which is slightly larger that
√

the square root of the tension 1/ℓs = 1/ 2π ℓ̄s .

– 11 –
t = τ /ω. Using Noether’s theorem, the Lagrangian is related to the energy M and angular
momentum J of the rotating string via8 [23, 24]
∂L ∂L
J= , M =ω − L. (3.5)
∂ω ∂ω
From these equations, we obtain the well known result for the pure NG action with free
endpoints (αf = 1):

π π π2 J
L=− , J= =⇒ M2 = = 2. (3.6)
2ωℓ2s 2ω 2 ℓ2s 2
ℓs ω 2 ℓ̄s
As anticipated, the relation between angular velocity and J implies that ω becomes small
√
at large J: ω ∼ ℓ̄−1 s / J. Therefore, the derivative expansion in the bulk runs in powers of
(ℓ̄s /L)2 ∼ (ω ℓ̄s )2 , which is equivalent to the 1/J expansion.
Let us now discuss the expectation for the scaling of subleading corrections to eq. (3.6).
According to the naive power counting, the NG term scales as ∼ 1/(ℓ̄s ω)2 ∼ J, which thus
provides the loop counting parameter. Therefore, we expect a quantum J 0 correction to M 2 in
eq. (3.6). Higher derivative corrections are further suppressed and, as explained in Section 2,
start at fourth-order in derivatives. Therefore, their contribution to M is expected to be
suppressed by a relative ∼ (ℓ̄s /L)4 ∼ 1/J 2 factor with respect to the leading order.
As anticipated, this naive counting is incomplete. This is manifested by looking at the
metric (3.3); setting momentarily αf = 1, we see that there is an infinite redshift factor at
σ = 1, and the metric is singular at the string endpoints. This implies that higher derivative
curvature invariants also become singular, e.g.:
h
µ 2
in 2n ℓ̄2n
s ω
2n
ℓ̄2s (Kαβ ) = for αf = 1 , (3.7)
(1 − σ 2 )2n
which diverges at σ = 1.
Within EFT, this means that we need to avoid going all the way to σ = 1. Instead, we
should consider some region around the endpoints and replace it by some effective boundary.
It is natural to cutoff the integration where higher derivative operators become of order one,
µ 2
ℓ̄2s (Kαβ ) ∼ O(1). To this aim, we consider endpoints at σ = ±(1 − z ℓ̄s ω) for some z > 0 that
does not scale with ω. In this way, the contribution of the operators (3.7) to the Lagrangian
reads
ℓ̄2n−2 1−z ℓ̄s ω
Z √ h µ 2 in
L⊃ s dσ G (Kαβ )
2π −1+z ℓ̄s ω
(3.8)
3 3
ω 2 2 −n z 2 −2n
 

(4n − 3)(4n − 1)
r 
2 
≃ 1 + z ℓ̄s ω + O z ℓ̄s ω .
ℓ̄s π(4n − 3) 4(4n − 5)
8
To obtain the result for J, note that a rotation in the X1 -X2 plane by a time dependent angle δϕ = ϵt is
equivalent to a shift ω → ω + ϵ, and that Noether’s theorem implies that the variation of the action under a
R
rotation is δS = dtJδ ϕ̇. The expression for the mass is obtained similarly using that the same rotation is
R t=t
equivalent to a time shift δt = tϵ/ω, and that by Noether’s theorem we have δS = dtM δ ṫ + L|t=tfi .

– 12 –
As we will see in the next section, a similar result can be obtained in a manifestly gauge
invariant approach upon introducing a mass term at the endpoints, which makes the difference
1 − αf nonzero.
Contrary to the naive expectation outlined before, eq. (3.8) shows that all higher derivative
terms contribute at the same order to the leading Regge trajectory. More precisely, each
contribution is suppressed only by ∼ (ℓ̄s ω)3/2 with respect to the NG contribution, with
further subleading terms organized in a series in ℓ̄s ω rather than the expected (ℓ̄s ω)2 . Similar
results are found in the study of other observables such as the energy of daughter Regge
trajectories. Notice also that eq. (3.8) does not contain any integer power of ω ℓ̄s - this fact
will be important when we will discuss quantum corrections in the next sections.
It seems therefore that we lack an organizing principle for the derivative expansion beyond
the leading order. This is not just because all operators contribute at the same order, but
equally dramatically because the expansion in eq. (3.8) contains unexpected fractional powers
of (ℓ̄s ω)2 ∼ ℓ̄2s /L2 . It is clear that these issues have to do with the physics of the endpoints,
and as such we should expect a more careful treatment and EFT description of the endpoints
to resolve the confusion. The formulation of such boundary EFT will be the topic of next two
sections.

4 Effective endpoints: power-counting and results

In this section, we discuss the general construction of the endpoint effective action,
neglecting bulk higher derivative terms. We will explain how to use this action to absorb the
singular contributions from bulk operators in Section 5.
Before delving into the technicalities, it is useful to explain the physical picture. Since
the size of the region in which the derivative expansion breaks down is small compared to the
string extension (3.4), it should be possible to treat it as an effective endpoint within EFT,
whose interaction with the bulk is described in a multipole expansion. It is however important
to note that the worldsheet metric (3.3) varies nontrivially along the string. EFT is applicable
only at time and spatial differences much larger than the string length:

∆τ 2 (1 − αf2 σ 2 )/ω 2 ≫ ℓ̄2s and ∆σ 2 /ω 2 ≫ ℓ̄2s . (4.1)

Since (1 − αf2 σ 2 ) ∼ ℓ̄s ω at the location where we should impose boundary conditions, we infer
that in terms of the physical time t = τ /ω at the center of the string, the frequency cutoff at
the boundary appears redshifted
p q
∂/∂t ≪ Λb.dry ∼ ℓ̄s ωΛbulk = ω/ℓ̄s . (4.2)

In other words, Poincaré invariance and the large acceleration of the quarks imply
p that the
boundary conditions for the bulk fields admit an expansion in even powers of ℓ̄s ω (notice
that this matches the form of the expansion in eq. (3.8)). The cutoff over spatial momenta is
not affected by these considerations in the coordinates (3.3).

– 13 –
 The most important physical consequence of the redshift concerns the calculability of
observables within EFT that are sensitive to the boundary conditions. For such quantities,
the derivative expansion breaks down at smaller momenta than in the long string setup. Most
notably, this observation implies than the number of daughter Regge trajectories that can be
studied within EFT is smaller than the naive expectation, as we will see explicitly in the next
sections.

4.1 Warm-up: massless scalar field on a non-dynamical rotating string

As a warm-up, let us consider an external scalar field on a rigid, i.e. non dynamical,
rotating string. We work to quadratic order in the field, with bulk action given by
Z Z 1 √
1
Sbulk = dτ dσ G (∂a)2 , (4.3)
−1 2

where we called a the scalar field in preparation of the analysis of the 4d worldsheet axion,
and (∂a)2 = Gαβ ∂α a∂β a. For the moment, we neglect higher derivative corrections whose
role we will analyze in Section 5. The worldsheet metric is given by (3.3) where we choose
αf = 1 − z ℓ̄s ω for some z > 0 which does not scale with ω.
The free bulk EOM reads
1
′′
ä − 2 1 − α 2 2
f σ a + σa′ = 0 , (4.4)
αf

where the dots and the primes respectively stand for time and spatial derivatives. Upon
changing coordinates via
s = arcsin(αf σ) , (4.5)
we recast eq. (4.4) into a standard Klein-Gordon equation, whose general solution is
Z
dϵ  −iϵτ +iϵs
f+ (ϵ) + e−iϵτ −iϵs f− (ϵ) ,

a(τ, σ) = e (4.6)
2π

where f+ (ϵ) = f+∗ (−ϵ) and f− (ϵ) = f−∗ (−ϵ) denote the Fourier components.
To study the spectrum, we need to supplement eq. (4.4) with the boundary conditions.
The simplest way to do so is to write the most general boundary action for a at σ = ±1 in a
derivative expansion. We suppose that the boundary conditions preserve the shift-invariance of
a. In this case, we can self-consistently write the boundary action in terms of time derivatives
of a. This is because, as we will show, operators built out of derivatives normal to the boundary
are redundant due to the EOM and the leading order Neumann boundary conditions. Working
to quadratic order in the field and assuming identical endpoints, we consider the boundary
action to be
X 1Z q h i
Sbdry = − ˆ t a)2 + ℓ̄3s da (∇
dτ Ĝτ τ ℓ̄s ba (∇ ˆ 2t a)2 + . . . , (4.7)
2
σ=±1

– 14 –
where ba and da are dimensionless Wilson coefficients, and we defined the endpoints’ worldline
metric and proper time derivatives as

Ĝτ τ = Gτ τ |σ=±1 , ˆ t = Ĝ−1/2 ∇

∇ ˆτ , (4.8)
ττ

where ∇ˆ τ is the covariant derivative at the endpoints (see appendix A). These definitions
ensure that the action (4.7) is invariant under reparametrizations of τ . For simplicity, we
neglected operators built out of the acceleration or higher derivatives of the string coordinates
X’s, whose power counting is more subtle and will be discussed in the next subsection.
Since we work in terms of dimensionless coordinates, in order to power-count, it is enough
to count the powers of the metric. Bulk terms (up to the renormalization issues that we will
address later) are counted as in the long string background, i.e. using Gαβ ∼ 1/ω 2 . Therefore,
√
the bulk action (2.2) is of order GG−1 0
αβ ∼ ω , so a is canonically normalized and we count it
as a ∼ O(1). At the boundary, we power-count according to the rules
r
ℓ̄s ˆt ∼ ω ,
Ĝτ τ ∼ , ∇ (4.9)
ω ℓ̄s

which followp
from the redshift factor previously discussed. Therefore, the first term in eq. (4.7)
scales as ∼ ℓ̄s ω, the second as (ℓ̄s ω)3/2 , etc. We can see this more explicitly by writing the
boundary conditions that follow from variation of (4.3) and (4.7):

ˆ 2 a − da ℓ̄3 ∇
∂n a|σ=±1 = ba ℓ̄s ∇ ˆ4
t s ta + ... , (4.10)

where ∂n = nα ∂α is the normal derivative (see app. A for notation). The aforementioned
scalings can be made manifest by working in terms of the coordinate s introduced in eq. (4.5)
p
(ℓ̄s ω)3/2
 
ℓ̄s ω 1
±∂s a|σ±1 = ba √ 1 + z ℓ̄s ω + . . . ä − da 3/2
(1 + . . .) ∂τ4 a + . . . , (4.11)
2z 4 (2z)
p p 

where the parentheses arise from the expansion of Ĝτ τ ≃ 2ℓ̄s z/ω 1 + O ℓ̄s ω .
It is instructive to use eq. (4.10) to extract the spectrum of single-particle states. In terms
of the coordinate (4.5), the endpoints are located at
 
π p 1
3/2
σ = ±1 =⇒ s=± − 2z ℓ̄s ω − √ z ℓ̄s ω + ... . (4.12)
2 6 2
Because of worldsheet parity s → −s, the solutions satisfying the boundary conditions must
take the form
a = e−iϵτ cos(ϵs) or a = e−iϵτ sin(ϵs) . (4.13)
Up to the leading order, we can identify the boundaries with s = ±π/2 and the boundary
conditions (4.10) with Neumann ∂s a = 0. Therefore, we find that the allowed frequencies are

ϵn ≃ n ∈ N , (4.14)

– 15 –
with even (odd) eigenfunction for even (odd) n. In physical units t = τ /ω, single-particle
states have energy ωϵn . Solving eqs. (4.10) to the subleading order, we find
( √ )
2Ba2 3
 
p 2Ba 3/2 6Da + Ba 2 0
ϵn = n 1 − ω ℓ̄s + ω ℓ̄s 2 + (ω ℓ̄s ) √ n + O(n ) + . . . , n ∈ N,
π π 3 2π
(4.15)
where we introduced the coefficients Ba and Da as follows for future convenience:
3da + 2z 3b2a + 6ba z + 4z 2


ba + 2z
Ba = − √ , Da = , (4.16)
z 6z 3/2
supposing Ba ∼ Da ∼ O(1). In the parenthesis of eq. (4.15), we neglected a (ℓ̄s ω)3/2 n0 term,
since this receives contributions from an operator built of higher derivatives of the X’s, which
we neglected in (4.7).9
Eq. (4.15) clearly illustrates the power-counting rules. The first correction
p to the dispersion
relation arises from the first term in eq. (4.7) and is suppressed by ∼ ωp ℓ̄s . We also find
a calculable ∼ ω ℓ̄s term determined by the same coefficient Ba of the ∼ ω ℓ̄s correction.
Finally, the second term in eq. (4.7) determines the (ω ℓ̄s )3/2 n2 correction to the dispersion
relation. By comparing the terms proportional to Ba and Da in eq. (4.15) we conclude that
the derivative expansion is under control for

ω ℓ̄s n2 ≪ 1 . (4.17)

The regime specified by (4.17) is smaller than the one that one would have guessed neglecting
the redshift factor, namely ϵ2 ω 2 ≪ ℓ̄−2 2 2
s , i.e. n (ω ℓ̄s ) ≪ 1. As we will see, the derivative
expansion assumes a similar structure in the calculation of the energy for the daughter Regge
trajectories.
We conclude this section with two important technical remarks for what follows.
• The expected expansion arises as a consequence of the cutoff on the string length at
(1 − αf ) = z ℓ̄s ω. Since the endpoint metric Ĝτ τ becomes singular for αf → 1, we may
think of a positive nonzero z as a regulator of the classical boundary divergences. While
working with a fixed z > 0 is justified on physical grounds, it is also possible to achieve
the same results by working directly in the limit z → 0, working thus with null endpoints.
To this aim, we introduce the following alternative metric at the endpoints:
h    i1/2
γ̂τ τ = ℓ̄s − ∇ˆ τ ∂τ X µ ηµν ∇ˆ τ ∂τ X ν , (4.18)

which is the square-root of the endpoints’ acceleration and therefore constitutes a singular
operator on a static background. However, it is simple to check that γ̂τ τ scales as ℓ̄s /ω
on the rotating background, like the wordline metric:10
ℓ̄s 1
γ̂τ τ = (1 − z ℓ̄s ω) ≃ Ĝτ τ . (4.19)
ω z
9 ˆ t a)2 .
This contribution arises from (∇3t X)2 (∇
10
Note that the worldline connection Γ̂τ τ = 12 Ĝ−1
τ
τ τ ∂τ Ĝτ τ vanishes for any z on the background solution.

– 16 –
 We will see in the next section that γ̂τ τ and Ĝτ τ are in fact equivalent and can be
used interchangeably around the rotating background also when the string fluctuates.
It is also possible to check that all the other operators with the right transformation
properties for defining a worldline metric are subleading in ω. Therefore, we may write
an action equivalent to (4.7) as
X 1Z h i
ˆ τ a)2 γ̂τ−1 3 ˜ ˆ 2 2 −2
p
Sbdry = − dτ γ̂τ τ ℓ̄s b̃a (∇ τ + ℓ̄ d
s a ( ∇τ a) γ̂ ττ + . . . . (4.20)
2
σ=±1

The action (4.20) is manifestly reparametrization invariant. The main advantage of the
formulation (4.20) is that it holds for arbitrary z and it allows us to work directly with
null endpoints setting z = 0, thus getting rid of the terms resulting from the expansion
of 1 − αf in eqs. (4.11) and (4.12).
Note that for null boundaries one cannot define a normalized normal vector, therefore ∂n a
in the left-hand side of eq. (4.10) is not well defined. To write the boundary conditions
in a geometric form, we note that in terms of the coordinate (4.5) the metric takes the
form (for αf = 1):
cos2 (s)
ds2 = 2 2


dτ − ds , (4.21)
ω2
which shows that at the endpoint we have a natural conformal geometric structure. This
structure is preserved by diffeomorphism of the form τ ± s → f (τ ± s), under which
both ∂/∂τ and ∂/∂s transform as vectors. Therefore, for null endpoints we may write
the boundary conditions that follow from the action (4.3)+(4.20) in geometric form as
   
ˆ τ γ̂ −1/2 ∇
±∂s a|s± π2 = ℓ̄s b̃a ∇ ˆ 2 γ̂ −3/2 ∇
ˆ τ a + ℓ̄3 d˜a ∇ ˆ 2a + . . . . (4.22)
ττ s τ ττ τ

We see that both sides manifestly transform homogeneously under diffeomorphism of

the form τ ± s → f (τ ± s).11
It is simple to solve (4.22) and check that the energy levels following from (4.20) take
the form (4.15) with the identifications
√ √
Ba = 2 b̃a , Da = 2 d˜a . (4.23)

which are significantly simpler than (4.16) since we work in the limit z → 0.

• The boundary conditions (4.10) are imposed at the string endpoint σ = 1. Since the
solutions of the EOMs are functions of αf σ, when solving the boundary condition (4.10)
perturbatively in ω for z > 0, we further need to expand (1 − αf ) using (4.12); as a
result, the physical parameters Ba and Da that enter the dispersion relation (4.15) are
nontrivial linear combinations of the ones multiplying derivatives of a in eq. (4.10). It is
11
Equivalently, in appendix A we show that the left-hand side of eq. (4.22) may be written in terms of a
Weyl invariant normal Nτα , that makes sense for both timelike and null boundaries.

– 17 –
 possible and technically more convenient to write the boundary conditions directly in
terms of the expansion of the field around the singular point αf σ = 1 of the EOMs. To
this aim, one considers the following (near boundary) expansion of the the solutions to
eq. (4.4):12

d2
 2
d4
  
1 2 d 3


a(τ, σ) = 1 + (1 − αf σ) 2 + (1 − αf σ) + + O (1 − αf σ) β1 (τ )
dτ 6 dτ 2 dτ 4
1 d2
   
p 1 2


+ 1 − αf σ 1 + (1 − αf σ) + + O (1 − αf σ) β2 (τ ) ,
12 3 dτ 2
(4.24)

where β1 (τ ) and β2 (τ ) are the boundary data which specify the solution. The expansion
for 1 + αf σ → 0 is identical. The boundary condition (4.10) is conveniently solved
perturbatively in terms of the boundary modes β1 and β2 as (at both endpoints)

d4 β1
p     
3/2
β2 = ω ℓ̄s Ba β̈1 + (ω ℓ̄s ) Da 4 + O β̈1 + O ω 5/2 ℓ̄s5/2 . (4.25)
dτ

Notice that eq. (4.25) manifestly depends solely upon the coefficients Ba and Da (and
not on z) that determine the dispersion relation (4.15). This is because the boundary
conditions (4.25) (at both endpoints) completely determine the profile of the field in
the bulk of the string with no reference to the regulator parameter z. We will use near
boundary expansions of the form (4.24) in the next sections to study fluctuations of the
string coordinates.

4.2 The general boundary action

4.2.1 The construction
Let us now discuss the endpoints of a dynamical string. Namely, we want to write boundary
actions that encode the boundary conditions for the X µ (τ, σ) fields at σ = ±1 in a consistent
derivative expansion. As in the scalar example previously discussed, the boundary conditions
must be obtained from the variation of a Poincaré invariant functional made of derivatives of
the string coordinates at the endpoints. First, let us consider the simplest possible endpoint
action: Z q
X
Sbdry = − m dτ Ĝτ τ , (4.26)
σ=±1

where, here and for most of this section, we assume identical endpoints. Eq. (4.26) models
mass m quarks at the endpoints. More generally, we expect that such a term will always be
present in an effective description of the endpoints. In this section, we assume the minimal
scenario where the quark masses are comparable or smaller than the Yang-Mills scale, i.e.,
m ∼ 1/ℓ̄s . We postpone the discussion of heavy quarks m ≫ 1/ℓ̄s to Section 7.
12
This expansion is analogous to the mode expansion of bulk fields near the boundary of spacetime in the
AdS/CFT.

– 18 –
 The boundary conditions that follow from (4.26) and (2.2) read
q
1
Ĝτ τ ∂n X µ = −m Ĝτ−1/2 ˆ µ
τ ∇τ ∂τ X . (4.27)
ℓ2s

Eq. (4.27) on the ansatz (3.2) imposes

αf = 1 − πmω ℓ̄2s + O ℓ̄2s ω 2 .

(4.28)

As announced, for m ∼ 1/ℓ̄s the quark mass term shortens the string similarly to the cutoff
prescription adopted inpSection 3. Plugging back in the action, we see that the boundary
action (4.26) scales as ω/ℓ̄s , while the NG term scales as 1/(ℓ2s ω 2 ) (recall X ∼ 1/ω from
eq. (3.2)). We may therefore think
p of the quark mass term (4.26) as a perturbation of the
Neumann boundary conditions Ĝτ τ ∂n X µ = 0.
Additional boundary operators are further suppressed. To see this, notice that leading
boundary conditions (4.27) imply that, working in a derivative expansion, we can remove
redundant operators according to the relations13
1 ˆ 2X µ
∂n X µ ≈ m∇ t and Ĝτ τ ≈ 2π ℓ̄s m γ̂τ τ , (4.29)
ℓ2s

where the second is obtained by squaring eq. (4.27) and using ∂n X · ∂n X = −1. Eq. (4.29)
implies a somewhat unusual equivalence between the square-root of the quark accelaration
and the worldline metric. This is a special property of the rotating background and does not
hold for a static string. Using the relations (4.29) and the discussed geometric arguments, we
power-count boundary operators according to the rules
r
1 ℓ̄s ˆt ∼ ω .
X ∼ , Ĝτ τ ∼ γ̂τ τ ∼ , ∂n X ∼ 1 , ∇ (4.30)
ω ω ℓ̄s

These rules allow us to estimate the relative size of a given operator contribution to an
arbitrary observable.
It is now straightforward to write the first correction to the endpoint action (4.27).
The analogy with the massless scalar example presented in eq. (4.7) is the simplest if we
momentarily choose a basis where according to (4.29), we systematically get rid of operators
ˆ kt X, in favor of those with normal derivatives ∇
with multiple time derivatives ∇ ˆ k−2 ∂n X µ . To
t
the first subleading order,
X Z q h   i
Sbdry = −m dτ Ĝτ τ 1 + ℓ̄2s b̄1 ∇ˆ t ∂n X · ∇
ˆ t ∂n X + . . . , (4.31)
σ=±1
13
For bulk operators it is well known that operators vanishing on the leading EOMs can be eliminated via a
field redefinition [45]. A similar statement holds also for boundary operators provided we allow for singular
field redefinitions, that can be treated with standard methods within EFT, see e.g. [46]. We checked explicitly
that different choices of the basis of operators for the boundary action yield the same results for the leading
and the daughter Regge trajectories.

– 19 –
where m ∼ 1/ℓ̄s and b̄1 ∼ O(1). We assumed time reversal and neglected terms with more
than overall 2 time derivatives (also accounting
p for partial integration). According p to the
power counting, the leading order scales as 1/ ω ℓ̄s , while the subleading scales as ω ℓ̄s . In
general, the expansion runs in single powers of ω ℓ̄s as expected.
Let us illustrate our discussion by looking at the Lagrangian and the Regge intercept for
the theory defined by the NG action and the b.dry action (4.31). To this aim, we compute
the Lagrangian L for arbitrary αf on the background (3.2)
 
1 h q i q 2
ℓ̄ ω 2
L = − 2 αf 1 − αf2 + arcsin(αf ) − 2m  1 − αf2 − b̄1 q s , (4.32)
ℓs ω 1−α 2
f

where the first parenthesis arises from the bulk action, the second one from the boundary
terms. We then extremize with respect to αf :

∂L 1 2 2 2
αf = 1 − πmω ℓ̄2s + π ℓ̄s m − b̄1 ℓ̄2s ω 2 + O ℓ̄3s ω 3 .



=0 =⇒ (4.33)
∂αf 2

Plugging in eq. (4.32), we find

√ r
1 4 2π 3/2 √ 1 2√
mℓ̄s 5b̄1 + 2π 2 m2 ℓ̄2s ω 3/2 + . . . .


L=− 2 − m ℓ̄s ω + (4.34)
4ℓ̄s ω 3 5 π
√
Eq. (4.34) contains terms at order ∼ ω and ω 3/2 with independent coefficients (m and
b̄1 ), as it was expected from the discussion in Section 3. We then obtain the relation between
ω and J from (3.5)
" p #
2(mℓ̄s )3/2 3 ℓ̄s m 5b̄1 + 2π 2 ℓ̄2s m2


4(mℓ̄s )3 π

1 1
ω= √ 1− + √ + +O , (4.35)
2ℓ̄s J 3J 3/4 20 π J 5/4 9J 3/2 J 7/4

from which we get the Regge intercept (neglecting quantum corrections):

√ √
m 5b̄1 + 2π 2 ℓ̄2s m2 −1/4


2 J 8 π m3/2 1/4
Mclassical = 2 + p J − √ 3/2 J
ℓ̄s 3 ℓ̄s 5 π ℓ̄s (4.36)
4  
+ π ℓ̄s m3 J −1/2 + O J −3/4 .
3
The existence of a J 1/4 correction from the boundary mass term is a well known result.
The existence of a J −1/4 term (and in general of ∼ J 1/4−n/2 with n ∈ N contributions) was
previously discussed in [16, 17] in the Polyakov formalism using a seemingly different EFT for
the endpoints. We will present an equivalent formulation of the action (4.31) which is more
closely related to that of those works in the next section. Finally the J −1/2 contribution arises
when using (4.35) to relate ω and J at subleading orders.
In eq. (4.36) we neglected quantum corrections. These provide a ∼ 1/J relative correction
to the result. We will discuss this point further in Section 4.4.

– 20 –
 We will soon show that the expansion for daughter Regge trajectories is similar to
eq. (4.36). The small ℓ̄s ω expansion for the length of the string L = 2αf /ω takes instead a
√
different form, with the first correction at order ℓ̄s ω ∼ 1/ J, see (4.33), therefore seemingly
violating the power-counting. The reason why this is allowed is that the string length, at least
as defined in eq. (3.4), is not a physical observable within EFT. Physically, this is because the
string length depends upon the precise size of the endpoints, which are fuzzy objects within
EFT. Technically, this is because the definition (3.4) is not invariant under field redefinitions,
which are instead crucial to get rid of redundant operators via (4.29). We will indeed see that
a different choice for the basis of operators of the boundary action leads to a different value
for 1 − αf without affecting on-shell observables.

4.2.2 Alternative parametrization and the effective string length

In principle, we could now proceed and use the action (4.31) to compute the spectrum of
the rotating string up to relative order ∼ J −3/2 . However, doing so would require us to deal
with some technical complications at intermediate steps of the calculation. This is because the
expansion of the operators in eq. (4.37) yields inverse powers of 1 − αf , that in turn introduce
spurious powers of ω due to eq. (4.33). While these spurious powers eventually cancel from
the final result, the power counting and the expansion discussed in the previous sections are
obscured in intermediate steps.
To remedy this state of affair, in this section we use the redundancies (4.29) to write
the boundary action in a different basis of operators, where the effective quarks move at the
speed of light αf = 1 to all orders in the 1/J expansion. In this basis the power counting
discussed in the previous section remains manifest also at intermediate steps. Additionally,
the derivation of this simplified approach will also clarify the physical reasons for the validity
of the EFT approach.
The discussion proceed in two steps. First, we note that the relations (4.29) allow to
choose a different basis of operators, where ∂n X does not appear at all, and where we trade
Ĝτ τ for γ̂τ τ . From these observations, we infer that the action (4.31) is equivalent to the
following
X  Z q Z p
Sbdry = −m bm dτ Ĝτ τ − bγ dτ γ̂τ τ
σ=±1
Z   (4.37)
˜ 2 −3
p
+ ℓ̄4s b1 3
dτ γ̂τ τ (∇τ X) γ̂τ τ + O (ℓ̄s ω)3/2
,

˜ τ is defined using the following connection

where in the second line the covariant derivative ∇
1
Γ̃ττ τ = γ̂τ−1 ∂τ γ̂τ τ . (4.38)
2 τ
Note that the usual connection Γ̂ττ τ = 12 Ĝ−1
τ τ ∂τ Ĝτ τ still appears in the definition (4.18) of γ̂τ τ .
p
According topthe power-counting, the first line of eq. (4.37) scales as ∼ 1/ ℓ̄s ω, while the
second as ∼ ℓ̄s ω.

– 21 –
 The action (4.37) is not yet the most convenient for our purposes. Nonetheless, it is useful
to discuss some of its features before performing further manipulations. Note in particular
that in the first line of (4.37) we retained an arbitrary linear combination of Ĝτ τ and γ̂τ τ .
However, due to the discussed redundancy only two combinations of the coefficients bm , bγ ,
and b1 enter in physical observables. We will find below that these combinations are
√ 4 √ 3/2 3/2 1/2
µ1 = − 2bγ m + π bm m ℓ̄s , (4.39)
3
b1 bγ bm π π 3/2 5/2
µ2 = −m √ + ℓ̄s m2 √ − ℓ̄s3/2 m5/2 b . (4.40)
2 2 2 5 m
Thus µ1 and µ2 are the true non-redundant Wilson coefficient: on dimensional grounds we
expect µ1 ∼ µ2 ∼ 1/ℓ̄s for a theory with light quarks. Nonetheless, the leading correction to
the string length, which is not an on-shell observable as previously observed, is controlled only
by bm :
p !
2 bm ℓ̄3s m3 π 3 π 2 2 2 2 2 2
αf = 1 − bm mω ℓ̄s π + bγ √ − bm ℓ̄s m ℓ̄s ω
2 2
p ! (4.41)
π2 2 2 2 bm ℓ̄3s m3 π 3 3 3 4 4


+ b ℓ̄ m − b1 √ ℓ̄s ω + O ℓ̄s ω .
4 γ s 2
This observation has an important physical consequence: by choosing a sufficiently large
coefficient bm while keeping fixed the coefficients µ1 and µ2 in eqs. (4.39) and (4.40), we can
effectively cutoff the string before the derivative expansion breaks down without modifying
the spectrum. We may therefore think of bm as a regulator for the endpoints singularities.14
In some sense, the freedom of changing the effective string length via a redundant coupling
provides a technical explanation for the validity of the EFT approach.
Retaining a finite value of 1 − αf > 0 will be important when discussing bulk operators
in Section 5, but, as announced at the beginning of this section, to make the power counting
manifest in our calculations we would like to work in a scheme in which the endpoints move at
the speed of light exactly. Naively, we might try to achieve this by setting bm = 0 in eq. (4.37),
but this is not enough due to the subleading terms in the expansion (4.41). Instead, it is
convenient to perform a further step and introduce an auxiliary einbein field e to rewrite the
first term in eq. (4.37) as
!
Ẋ 2
Z q Z
X m X
−m bm dτ Ĝτ τ → − dτ + b2m e , (4.42)
2 e
σ=±1 σ=±1

while leaving the rest of the action unchanged. Solving for e on its EOMs for bm > 0 we
recover the previous expression (assuming e ≥ 0). If we instead set the redundant coefficient
to zero, bm = 0, we find that the EOM for the einbein requires the string endpoints to be
precisely null:
Ẋ 2 |σ=±1 = 0 . (4.43)
14
As usual, quantum corrections require an additional, different, regulator - such as dimensional regularization.

– 22 –
Note that this constraint implies that the modified boundary metric (4.18) can be simply
written as  1/2
γ̂τ τ = ℓ̄s −Ẍ µ ηµν Ẍ ν , (4.44)

with no reference to a connection.15 We therefore conclude that we can work in a scheme

where the endpoints’ effective action reads
X  Z Z  
˜ 2 −3
p 4
p 3 3/2
Sbdry = m bγ dτ γ̂τ τ − ℓ̄s b1 dτ γ̂τ τ (∇τ X) γ̂τ τ + O (ℓ̄s ω) , (4.45)
σ=±1

with the constraint (4.43) that the endpoints move on lightlike trajectories. The action (4.45)
is analogous to the one discussed in [17], where the authors worked in Polyakov formalism.
Note that eq. (4.45) does not include operators containing ∂s X µ , where s is the natural normal
coordinate in the null case according to the discussion around (4.21), since these vanish on
the leading order boundary conditions that follow from the Nambu-Goto action.
As an illustration, let us discuss the calculation of the leading Regge trajectory again,
using the action (4.45). The constraint (4.43) imposes that on the background profile (3.2)
we must set
αf = 1 . (4.46)
The Lagrangian is then straightforward to compute. Using (4.44) and noticing that Γ̃ττ τ = 0
on the solution (3.2), we find
1 √ p √  
L = − 2 − 2µ1 ℓ̄s ω − 2 2µ2 (ℓ̄s ω)3/2 + O ℓ̄s5/2 ω 5/2 , (4.47)
4ℓ̄s ω

where the Wilson coefficients are the ones given in (4.39) and (4.40) in the limit bm → 0:
√
bm →0 bm →0 1
µ1 = − 2mbγ , µ2 = − √ mb1 . (4.48)
2
We then find ω as
1 −1/2 µ1 −5/4 3µ2 −7/4  
ω= J − J − J + O J −9/4 , (4.49)
2ℓ̄s 4 4
from which we recover the same expansion in eq. (4.36) (neglecting the Casimir energy)
√
J µ1 µ2 µ2 ℓ̄s  
Mclassical = + 1/4 + 3/4 + 1 J −2 + O J −5/4 . (4.50)
ℓ̄s J J 8
The same result is obtained using the action (4.37), with the Wilson coefficients given by
eq. (4.39) and (4.40). On physical grounds, we expect that the effective endpoints contribute
to the energy with a positive amount, hence µ1 > 0; this can indeed be proven rigorously
using Regge theory and the analytic S-matrix bootstrap [15].
15
To derive eq. (4.18) we used that the connection Γ̂ττ τ remains finite in the limit Ẋ 2 → 0, as it can be
checked solving for (4.43) explicitly as in eq. (4.53) below.

– 23 –
 In the following sections, we will discuss the calculation of the spectrum of the daughter
Regge trajectories. We will mostly work with exactly massless endpoints, but we will also
comment on the calculation using the action (4.37), as this will be important when discussing
the renormalization of bulk operators. An important subtlety when working with massive
endpoints with the action (4.37) is that it is important to keep bm > 0 at intermediate
steps. This is because keeping bm > 0 ensures that the first two corrections to 1 − αf in
eq. (4.41) do not vanish and, as we will see, the expansion into fluctuations of the operators
in eq. (4.37) yields inverse powers of 1 − αf , which obscure the power-counting as mentioned
in the introduction to this section. We checked that the final results for the spectrum are
independent of the scheme when expressed in terms of the physical Wilson coefficients (4.39)
and (4.40).

4.3 Fluctuations
4.3.1 Leading order analysis
We parametrize fluctuations around the background (3.2) with d − 1 fields {α, ϕ, ρi } as
1
Xµ = {τ, [αf σ + α(τ, σ)] cos (τ + ϕ(τ, σ)) , [αf σ + α(τ, σ)] sin (τ + ϕ(τ, σ)) , ρi (τ, σ)} ,
ω
(4.51)
where i = 1, . . . , d−3. One might think that α(τ, σ) can be removed by a gauge transformation.
This is true everywhere but at σ = ±1, since we are only allowed to consider diffeomorphisms
that preserve the position of the boundaries. We will see that α(τ, σ) enters in the bulk
action only in total derivative terms, and only the α(τ, ±1) boundary modes are physical. The
parametrization (4.51) was formerly used in [28].
Let us consider first the NG action, for arbitrary αf , neglecting the boundary terms.
Neglecting total time derivatives, the expansion to quadratic order yields:
" #
Z 2 ′ )2 α 2 σ 2 ϕ̇2 2 (ϕ′ )2
αf q (ρ̇i ) (ρ f σ
Sbulk ≃ d2 σ 1 − αf2 σ 2 − i2 + −
2ω 2 ℓ2s 1 − αf2 σ 2 αf (1 − αf2 σ 2 )2 1 − αf2 σ 2

2 σ 2 ϕ̇ α
 (4.52)
Z
α σα 2 α
1 2 ∂ 
q
f f
+ 2 2 d σ − 1 − αf2 σ 2 α + q +q .
ω ℓs ∂σ 2 1 − αf2 σ 2 1 − αf2 σ 2

As anticipated, α enters only in a total derivative term, and is thus purely a boundary mode.
As formerly noted, the constraint (4.43) implies αf = 1 at leading order in fluctuations;
this sets to zero the term linear in α in eq. (4.52). At subleading order, we may solve
perturbatively for the null constraint in terms of the boundary mode α:
 
2 1 2 1 2
α|σ=±1 = −σ ϕ̇ + σ ϕ̇ − ϕ̈ − ρ̇ + . . . , (4.53)
2 2

where the term that is quadratic in the fields will be important to obtain the expansion of the
boundary action (4.45) to quadratic order, that we report in the next subsection. Plugging

– 24 –
back into eq. (4.52) we obtain
" #
(ρ̇i )2 σ 2 ϕ̇2 σ 2 (ϕ′ )2
Z
1 2
p
′ 2
Sbulk ≃ 2 2 d σ 1 − σ2 − (ρi ) + −
2ℓs ω 1 − σ2 (1 − σ 2 )2 1 − σ2
(4.54)
σ 2 ϕ̇2
Z
1 X
− 2 2 lim dτ √ .
2ℓs ω ± σ→±1 1 − σ2

The boundary term in the second line ensures that the action is finite for configurations such
that the angular mode ϕ does not vanish at the endpoints. Within the standard approach, in
which eq. (4.43) is not imposed as a constraint, the action (4.54) is obtained noticing that
there is no kinetic term for α in eq. (4.52) and we can therefore integrate it out on its EOM.
We are now ready to extract the leading order spectrum from the action (4.54). The
action for the transverse mode(s) ρi is identical to that of a massless scalar in eq. (4.3), yielding
the EOM (4.4) with Neumann boundary conditions at the leading order:
√
∓ 1 − σ2 ′
lim ρi = 0 . (4.55)
σ→±1 ℓ2s ω 2

Proceeding as in Section 4.1, we conclude that each of the transverse modes is associated
with a spectrum of excitations whose energy levels are integerly spaced in units of the angular
velocity:
ϵn = n ∈ N . (4.56)
At a quantum level, the modes with n ≥ 1 in eq. (4.56) are associated with creation and
annihilation operators â†ρi ,n , âρi ,n , and generate the corresponding Fock space. The n = 0
mode as usual needs to be treated differently and is associated with translations of the string.
Since we can always work in the rest-frame of the particle this mode is irrelevant for our
purposes. Also the n = 1 mode is special. To understand why, it is simpler to specialize
momentarily to d = 4 where there is a single transverse coordinate. The creation operator
associated with the n = 1 mode acts on the vacuum as

â†ρ,1 |J, J⟩ = |J + 1, J⟩ , (4.57)

where |J, J⟩ indicates the EFT ground state, which is a highest weight state of the spin J
representation [47]. Therefore â†ρ,1 creates a non-highest weight state. The same is true for all
the n = 1 modes in d ≥ 4. In more physical terms, the n = 1 modes create states which can
be obtained by applying a rotation to states with a different J on the leading Regge trajectory.
Hence, to study the spectrum of particles in spin-J representations it is enough to focus on
highest weight states. Therefore only the n ≥ 2 modes are relevant for our discussion.
Let us now discuss the angular variable ϕ. The EOM reads

ϕ̈ − (1 − σ 2 )ϕ′′ − 2 − σ 2 ϕ′ = 0 .


(4.58)

– 25 –
The most general solution may be written as a superposition of plane waves of the form (see
also [28])
2 3/4
 
−iϵτ (1 − σ )
3 3
2 2
ϕ=e c1 Pϵ− 1 (σ) + c2 Qϵ− 1 (σ) , (4.59)
σ 2 2

where P and Q are Legendre functions. For αf < 1, the solution is given by eq. (4.59) replacing
σ → αf σ. The overall 1/σ prefactor is due to the parametrization (4.51), which implies that
δX ∝ σϕ to linear order. The ratio c2 /c1 and the allowed values of ϵ in eq. (4.59) are found
imposing the boundary conditions at both endpoints:

∓ϕ′ + σ 2 ϕ̈
lim √ = 0. (4.60)
σ→±1 ω 2 ℓ2 1 − σ 2
s

The limit in this expression naively looks singular, but it is not. This is because, as for
the transverse fluctuations, the endpoints σ = ±1 are singular points of the differential
equation (4.58). To see that the limit is indeed well defined, it is convenient to consider the
near boundary expansion for ϕ:

ϕ = β1ϕ (τ ) − (1 − σ)β̈1ϕ (τ ) + (1 − σ)3/2 β2ϕ (τ ) + O (1 − σ)2 ,

(4.61)

where we use a superscript to distinguish the transverse and angular boundary modes. The
expansion for 1 + σ → 0 is identical. Plugging (4.61) in (4.60), we see that the first two terms
cancel and in the limit σ → ±1 we are left with a nonsingular result

β2ϕ = 0 . (4.62)

Imposing eq. (4.62) at both endpoints implies that the solutions (4.59) must satisfy

c2 = 0 and ϵn = n ∈ N . (4.63)

Eq. (4.63) yields a Fock space spectrum identical to that of the transverse modes, as expected
from the standard results in critical string theory (this calculation is independent of the
number of spacetime dimensions).
As for the transverse coordinates, the n = 0 and n = 1 modes are associated, respectively,
with rigid rotations and translations; therefore they must be excluded when constructing the
spectrum of physical particles at fixed angular momentum. In particular, highest-weight states
in different representations are related via the action of the 0-mode of ϕ, which enters the
mode decomposition via
ϕ = ϕ̂0 + p̂0 τ + wave-modes , (4.64)
where p̂0 ∝ ⟨J⟩ − Jˆ and ϕ̂0 is a h2π-periodic
i quantum mechanical variable. The canonical
ˆ
commutation relation imply that ϕ̂0 , J = i, therefore the operator eiϕ̂0 increases the angular
momentum; e.g. in four dimensions in the same notation of eq. (4.57) we have

eiϕ̂0 |J, J⟩ = |J + 1, J + 1⟩ . (4.65)

– 26 –
Note that eqs. (4.57) and (4.65) imply that the lowering operator of the SU (2) algebra is
given by Jˆ− = e−iϕ̂0 â†ρ,1 , as it can be checked via Noether’s procedure.16
Upon quantizing the system we conclude that daughter Regge trajectories corresponding
to multi-particle states created by ρi and ϕ have a gap over the ground-state (4.50) given by
X X (ρ ) (ρ ) X (ϕ) (ϕ)
δM = ω nk i ϵk i + ω nk ϵk
i k≥2 k≥2
" # (4.66)
(ρ ) (ϕ)
X X
≃ω nk i + nk k,
k≥2 i

where the nk ’s denote the occupations numbers and we used superscripts to distinguish between
the modes in obvious notation. We stress again that in eq. (4.66) there is no contribution
from the k = 0 and k = 1 modes, for the hPreasons explainedi above. Using (4.49) we recover
2 −2
P (ρi ) (ϕ)
the well known result δM = ℓ̄s k≥2 i nk + nk k to leading order in J.
Finally, we remark that upon canonically normalizing the bulk fields we infer that nonlinear
terms are suppressed in ωℓs , with the loop counting parameter given by ω 2 ℓ2s as anticipated.
We will discuss loop corrections further in Section 4.4.

4.3.2 Subleading orders: massless endpoints’ scheme

Let us now study the corrections to the energy spectrum that arise from the boundary
action (4.45). Upon expressing α through (4.53) and neglecting total derivatives, the expansion
to quadratic order of eq. (4.45) yields
(
1 ... 2
X Z  
bγ 2 2 1 2 2
Sbdry = mℓ̄s dτ p (ρ̇i ) + (ρ̈i ) − ϕ̇ + ϕ̈ − ϕ
4 ω ℓ̄s
2 2
σ=±1
p
7 ... 2 .... 2
 
b1 ℓ̄s ω 2 2 ... 2 3 2 2
+ (ρ̇i ) − 5(ρ̈i ) + 4( ρ i ) + ϕ̇ + 4ϕ̈ + ϕ − ϕ
4 2 2
)
 
+ O (ℓ̄s ω)3/2 .

(4.67)

Note that because of the O(d − 3) unbroken symmetry acting on the ρi there is no quadratic
mixing between the transverse modes and ϕ. We therefore discuss transverse and angular
modes separately below.
We start from the transverse modes. From the variation of the boundary action (4.67),
16
This is analogous to the role of gapped Goldstones [48] in nonlinear sigma models at finite chemical potential
and finite volume [47].

– 27 –
we find that the leading order boundary condition (4.55) is corrected as
√ " ! !
± 1 − σ2 ′ mℓ̄s bγ p ∂ 2 ρi bγ p ∂ 4 ρi
lim ρi = − b1 ω ℓ̄ s + − 5b1 ω ℓ̄ s
ℓ2s ω 2 ∂τ 2 ∂τ 4
p p
σ→±1 2 ω ℓ̄s ω ℓ̄s
# (4.68)
∂ 6 ρi 8

3/2 ∂ ρi
p
− 4b1 ω ℓ̄s 6 + O (ω ℓ̄s ) .
∂τ ∂τ 8
This condition may be written in a manifestly finite form using the near boundary expan-
sion (4.24) of the solution:
 2 ρ
d4 β1ρ

ρ 3/2 5/2 d β1
β2 = 0 + πµ1 ω ℓ̄s +
dτ 2 dτ 4
 2 ρ
d4 β1ρ
 4 ρ
d6 β1ρ 8 ρ
   
5/2 7/2 d β1 d β1 7/2 9/2 d β1
− 2πµ2 ω ℓ̄s + +4 + + O ω ℓ̄s ,
dτ 2 dτ 4 dτ 4 dτ 6 dτ 8
(4.69)
where we used eq. (2.4).
Imposing the boundary condition (4.69) on plane wave solutions of the form (4.13), we
find that the eigenfrequencies are corrected as
h √ √ i
ϵn = n 1 + 2(ω ℓ̄s )3/2 µ1 ℓ̄s n2 − 1 + 2 2(ω ℓ̄s )5/2 µ2 ℓ̄s (n2 − 1)(4n2 − 1) + . . . , (4.70)

where n ∈ N. Similarly to eq. (4.15), the derivative expansion breaks down for n2 ω ℓ̄s ∼ 1.
Eq. (4.70) neglects O n6 (ω ℓ̄s )7/2 corrections from higher derivative boundary terms. For the

moment we are also neglecting ω 2 ℓ̄2s one-loop corrections to the spectrum. Notice that ϵn is
unmodified by the higher derivative corrections for n = 0 and n = 1, since as explained these
are protected modes implementing translations and rotations.
The calculation of the corrections to the leading order boundary condtions (4.60) for ϕ
proceeds analgously. We directly report the result in terms of the boundary modes defined in
eq. (4.61):
!
π d 2βϕ d4βϕ d6βϕ
β2ϕ = 0 + µ1 ω 3/2 ℓ̄5/2
s
1
+ 2 41 + 1
3 dτ 2 dτ dτ 6
" ! !#
2π d2βϕ d4βϕ d6βϕ d4βϕ d6βϕ d8βϕ
− µ2 ω 5/2 ℓ̄7/2
s 3 1
+ 2 41 + 1
+2 1
+ 2 61 + 1
(4.71)
3 dτ 2 dτ dτ 6 dτ 4 dτ dτ 8
!
d 10 β ϕ
+ O ω 7/2 ℓ̄9/2
s
1
.
dτ 10
From eq. (4.71) we find the corrections to eq. (4.60):
√
 
1 3/2 2 5/2 2 2
c2 = c1 0 + √ µ1 ℓ̄s (ω ℓ̄s ) n(n − 1) + 2µ2 ℓ̄s (ω ℓ̄s ) n(2n − 3)(n − 1) + . . . , (4.72)
2

√
 
1
ϵn = n 1 − √ µ1 ℓ̄s (ω ℓ̄s )3/2 n2 − 1 − 2µ2 ℓ̄s (ω ℓ̄s )5/2 (2n2 − 3)(n2 − 1) + . . . , (4.73)
2

– 28 –
where n ∈ N. Comments analogous to those below eq. (4.70) apply here. In particular,
notice again that the result for the n = 0 and n = 1 modes is unmodified by higher-derivative
corrections due to their protected nature. We also remark that the first correction to the leading
order result in eq. (4.73) takes the opposite sign compared to the transverse modes (4.70). As
formerly commented, we must have µ1 > 0, and therefore the daughter trajectories associated
with the angular mode are lighter than the ones for the transverse modes.

4.3.3 Subleading orders: masssive endpoints’ scheme

Let us now briefly comment on the calculation using the action (4.37) in which the
endpoints are not constrained by eq. (4.43), and the quarks velocity αf is given in eq. (4.41)
(this ensures that in the expansion of the action, the terms linear in the fluctuations vanish).
We focus first on the transverse modes ρi . Expanding (4.37), we get
 s " #
X Z  b (ρ̇ )2 3
2
m i bγ ℓ̄s 2 ℓ̄s ω d ρi 5 2

Sbdry ⊃ m dτ + (ρ̈i ) + b1 ℓ̄s − (ρ̈i ) .
αf5 dτ 3
q q
 2ω 1 − α2 4α α ℓ̄ ω 4 
σ=±1 f f f s

q (4.74)
p
2
We see that, differently than eq. (4.67), since 1 − αf ∼ ω ℓ̄s , the first term in eq. (4.74)
is proportional to ω −3/2 rather than the expected power ω −1/2 . This might seem to suggest
that
p the leading-order boundary conditions for ρ in eq. (4.55) is corrected already at order
ω ℓ̄s , therefore contradicting the EFT power-counting and the results of the previous section.
Luckily, it turns out that upon carefully using the value of αf in eq. (4.41), this contribution
cancels out upon summing the variation of the bulk and the boundary action, and the final
result agrees with the one we found in eq. (4.69).
Let us show this up to order O(ω). We write the boundary term in the variation of the
bulk action (4.52) using the near boundary expansion (which is identical to (4.24))
√
δβ1ρ β2ρ
X Z 
mbm  ρ ρ ρ ρ

δSbulk = bulk EOMs + dτ √ + 2δβ 1 β̈1 + δβ2 β̈ 2
σ=±1
2ω 2 ℓ2s 2ℓs ω 3/2
!# (4.75)
bm m  ρ ρ ρ ρ
 1
+√ δβ2 β̈1 + δβ1 β̈2 + O p ,
2ω ℓ̄s ω
where we suppressed the O(d − 3) indices and retained terms up to order 1/ω. Expanding
similarly the variation of eq. (4.74) we obtain
"√ !#
X Z mbm ρ ρ bm m  ρ ρ ρ ρ
 1
δSbdry = − dτ 3/2
δβ1 β̈1 + √ δβ2 β̈1 + δβ1 β̈2 + O p , . (4.76)
σ=±1
ℓs ω 2ω ℓ̄s ω

Summing eqs. (4.75) and (4.76) we see that the subleading terms proportional to δβ1ρ and β1ρ
cancel out and we obtain
" √ !#
X Z δβ1ρ β2ρ mbm ρ ρ 1
δSbulk +δSbdry = bulk EOMs+ dτ √ + δβ2 β̈2 + O p , (4.77)
σ=±1
2ℓ2s ω 2 2ℓs ω 3/2 ℓ̄s ω

– 29 –
which is compatible with the leading order condition (4.55) up to order (ω ℓ̄s )3/2 as expected:17
 
β2ρ = 0 + O (ω ℓ̄s )3/2 . (4.78)
Similar cancellations occur at subleading order. The reason for the existence of these
seemingly enhanced terms in the boundary action is that Ĝτ τ , and therefore also the boundary
connection Γ̂ = 12 Ĝ−1
τ τ ∂τ Ĝτ τ , are not analytic for αf → 1. At some intuitive level, such terms
arise because the redundant operator in eq. (4.37) is effectively equivalent to a cutoff regulator
near the endpoints. While conceptually satisfying, it is often the case that cutoff schemes
yield enhanced contributions at intermediate steps in EFT.
The technical reason why such spurious terms in the expansion do not contribute to
observables, is that they always depend on the specific value of the regulator coefficient bm
(and are often singular in the limit bm → 0), which determines the leading correction to
the string length (4.41). Based on the arguments presented at the beginning of this section,
only the combinations in eq. (4.39) and (4.40), which are regular in the limit bm → 0, can
contribute to physical observables. Unfortunately, retaining bm > 0 in the action (4.37) is
nonetheless important at intermediate steps if we do not impose the condition (4.43).
Despite these complications, it is nonetheless straightforward to compute the subleading
corrections to the boundary conditions for the transverse modes. By expanding the variation of
the bulk and boundary actions as in eqs. (4.75) and (4.76), we eventually recover the previous
result (4.69) in terms of the physical Wilson coefficients in eqs. (4.39) and (4.40).
The analysis of angular fluctuations from the action (4.37) proceeds similarly to that of
the transverse modes. The main difference is that we now need to compute also subleading
corrections to the expression for the boundary mode α when we integrate it out.
It turns out that the cancellations described above are even more dramatic for the boundary
EOMs of α and ϕ than for the ρi ’s. For instance the expansion of eq. (4.37) produces the
following terms

X Z bm (α ± ϕ̇)2 bγ ℓ̄s (α̇ ± ϕ̈)2 bm (3α2 + 8α̇2 ± 14αϕ̇ − 5ϕ̇2 )
Sbdry ⊃ m dτ  √ + q + p
σ=±1
4 2ω(1 − αf )3/2 8 ℓ̄s ω(1 − α ) 12ω 2(1 − αf )
f

3/2 √
!#
5b1 ℓ̄s ω(α̇ ± ϕ̈)2 1
+ +O , (4.79)
4(1 − αf )2
p
ω ℓ̄s
where the first term scales as 1/ω 5/2 and the last three terms are all of order 1/ω 3/2 using
1 − αf ∼ ω ℓ̄s . Luckily again, the first term vanishes on the leading order solution for α, given
by α|σ=±1 = ∓ϕ̇. Upon carefully expanding both the bulk and boundary actions we can solve
perturbatively for α and β2ϕ (cf. (4.61)), and we find similar cancellations at subleading order.
The expansion for α takes a form similar to eq. (4.41)
h
i
α|σ=±1 = ∓ β̇1ϕ − (ω ℓ̄s )2 α2 − (ω ℓ̄s )3 α3 + O (ℓ̄s ω)4 . (4.80)
17
Notice that when solving (4.77) with the ansatz β2ρ = F d
β1ρ we also need to express the variation of β2ρ


dτ
in terms of that of β1ρ as δβ2ρ = F dτ
d

ρ
δβ1 .

– 30 –
The precise value of α2 and α3 is scheme-dependent for the same reason as the background
string length. We show for illustration the result for α2 obtained from the action (4.37)
√ !
d5 β1ϕ d3 β1ϕ dβ1ϕ
 2 
π 2 2 2 bγ bm 3/2 3/2 3/2
α2 = b ℓ̄ m − √ π ℓ̄s m +2 3 + . (4.81)
2 m s 2 dτ 5 dτ dτ

The expression for α3 is lengthy and we do not report it. Using this result and the near
boundary expansion to the third subleading order, as reported in appendix B, we eventually
recover the result (4.71) given in the previous subsection.

4.4 Comments on one-loop corrections

As already commented, the loop counting parameter is 1/ω 2 ∼ J. A detailed account of
one-loop effects is beyond the scope of this work. Here we limit ourselves to some comments.
As well known, the NG action at the classical level is equivalent to the Polyakov action,
which consists of d free fields subject to the Virasoro constraints. The Polyakov formulation
is convenient for the purposes of this section, since it makes the classical integrability of the
NG action manifest. As emphasized in [6, 27], at one-loop level, the same equivalence holds
provided one adds the so called Polchinski-Strominger (PS) term to the Polyakov action [5]. In
modern language, the PS term is a Wess-Zumino term compensating for the anomalous Weyl
transformation properties of the path-integral measure away from d = 26 [27]. To compute
one-loop corrections in Polyakov formalism, one therefore simply needs to sum the result from
the free bosons’s action (which is simple to obtain), to the correction from a single insertion
of the PS term.
For the leading Regge trajectory, the calculation was done in [16], where it was found
that the result is independent of the number of dimensions and reads18

2 1 
−3/4

δMquantum =− + O J . (4.82)
ℓ̄2s

The O(J −3/4 ) correction is expected to arise from the deviations from Neumann boundary
conditions. We remark that, as well known, in the covariant formalism one directly computes
the mass square rather than the mass.
To perform a similar calculation for excited states is less straightforward unfortunately.
This is because the Polchinski-Strominger term corrects the structure of the constraints
nontrivially. Nevertheless, there are specific regimes where the free theory result, that
coincides with the critical string theory expression:
" #
1 X X (ρ ) (ϕ)
δM 2 = 2 nk i + nk k , (4.83)
ℓ̄s i
k≥2

18
Ref. [24] obtained a different result working in the Nambu-Goto formalism; we suspect that the mismatch
is due to subtleties in dealing with UV divergences compatibly with Poincaré symmetry in the Nambu-Goto
formalism.

– 31 –
offers an improvement over the naive expansion (4.66). This is because the formula (4.83)
is exact in the classical limit, and therefore accounts for infinitely many corrections from
nonlinear vertices (which are not suppressed in derivatives) in the NG formalism. In other
words, the classically exact Polyakov result should provide a better approximation than the
(ρ ) (ϕ)
naive expansion (4.66) for large occupation numbers: nk i ≫ 1 and/or nk ≫ 1.
In practice, the only difference between the formula for the NG expansion (4.66) and the
result of the Polyakov formalism (4.83) is that the latter directly yields to M 2 rather than M .
Therefore, upon accounting for the corrections that we computed in (4.70) and (4.73), our
final result for the spectrum is written as

M = MP ol + δMN G , (4.84)

where MP ol is the exact result from the free Polyakov action

v " #
u
1u X X (ρ ) (ϕ)
MP ol = t(J − 1) + nk i + nk k , (4.85)
ℓ̄s i
k≥2

and in the second term we include the correction in the usual derivative expansion

µ21 ℓ̄s
 
µ1 µ2
δMN G = + − + ...
J 1/4 J 3/4 8J
X X (ρ ) (ρ ) X (ϕ) (ϕ) (4.86)
+ω nk i (ϵk i − k) + ω nk (ϵk − k) + . . . ,
i k≥2 k≥2

where we have subtracted the free part of the excitations’ energy, which is already included in
MP ol , and ω is given in eq. (4.49). In general, δMN G is constructed such that when expanding
the square root, MP ol + δMN G agrees with the result of a naive calculation in NG formalism.
Let us finally comment that the square-root formula (4.85) might improve the convergence
of the derivative expansion already for low-momentum few-particle states. This is because
actual one-loop corrections are expected to be further supressed by factors of 2π in the
denominator with respect to the terms arising from the expansion of (4.85). While this is
admittedly just optimistic numerology in the absence of one-loop results to compare to, we
remark that an analogous phenomenon was analyzed to study long flux tubes in [39]. In that
case, arranging the derivative expansion similar to (4.84) (using the classical GGRT result for
the string spectrum [49, 50]) rather than the naive EFT expansion, dramatically improves the
agreement between lattice results and theoretical predictions.

– 32 –
4.5 Different endpoints
We finally consider mesons made of different quarks, that we model via different boundary
conditions at the endpoints. We introduce different Wilson coefficients at the two ends of the
string:  Z q Z 
X p
Sbdry = − m bm,σ dτ Ĝτ τ − bγ,σ dτ γ̂τ τ + ... , (4.87)
σ=±1

where we work to relative order ∼ ω 3/2 . Similarly to eq. (4.37), the operators multiplied
by bm,σ are redundant and may be removed if we work with null endpoints as in eq. (4.43).
Analogolously to eq. (4.39), the physical Wilson coefficients that control the first corrections
to the spectrum are given by
√ 4p 3/2
µ1,±1 = − 2mbγ,±1 + ℓ̄s π m3/2 bm,±1 . (4.88)
3
The calculation of the mass of the leading Regge trajectory and the spectrum of the
daughter particles proceeds analogously to the case of identical points. The only difference is
that we have to modify the background solution (3.2) to account for the asymmetry of the
string:
µ 1

Xclass. = τ, (αf σ + δαcm ) cos τ, (αf σ + δαcm ) sin τ, 0, 0, . . . , (4.89)
ω | {z }
d−3 times
where σ ∈ (−1, 1) as before and δαcm represents a shift in the center of mass of the string
with respect to the symmetric case. Extremizing the classical action we can solve for αf and
δαcm perturbatively in ω
bm,1 + bm,−1
αf =1 − πmω ℓ̄2s
" 2 #
π 2 m 2 ℓ̄2 π 3/2 (ℓ̄ m)3/2
s 2 s
− ω 2 ℓ̄2s (bm,1 + b2m,−1 ) −
p p
√ (bγ,1 bm,1 + bγ,−1 bm,−1 ) + . . . ,
4 2 2
(4.90)
bm,1 − bm,−1
δαcm = − πmω ℓ̄2s
" 2 #
2 2 2 π 3/2 (ℓ̄s m)3/2
2 2 π m ℓ̄s 2 2
p p
− ω ℓ̄s (bm,1 − bm,−1 ) − √ (bγ,1 bm,1 − bγ,−1 bm,−1 ) + . . . .
4 2 2
(4.91)
Proceeding as in Section 4, we arrive at the final result for the mass of the leading Regge
trajectory: √
J µ1,1 + µ1,−1 
−3/4

Mclassical = + + O J . (4.92)
ℓ̄s 2J 1/4
To analyze fluctuations we use a parametrization similar to (4.51):
1
Xµ = {τ, [αf σ + δαcm + α(τ, σ)] cos (τ + ϕ(τ, σ)) ,
ω
[αf σ + δαcm + α(τ, σ)] sin (τ + ϕ(τ, σ)) , ρi (τ, σ)} . (4.93)

– 33 –
The analysis of the bulk and boundary equations proceeds as before. We find that the boundary
conditions at the two endpoints of the string depend on the boundary coefficients (4.88) exactly
as in the former results (4.69) and (4.71). We spare the details of the derivation to the reader,
and simply provide the results for the spectrum and wave-functions of the single-particle states
below.
Let us start from the transverse mode. As explained around (4.13), the solution to the
bulk equations of motion is given by plane waves of the form

ρi = e−iϵτ [c̃1 cos(ϵs̃) + c̃2 sin(ϵs̃)] , (4.94)

where s̃ = arcsin(αf σ + δαcm ), a modified version of eq. (4.5) due to the shift in the center of
mass. For equal endpoints, worldsheet parity (i.e. charge conjugation) guarantees that the
wave-function must be either even or odd under s → −s, hence setting one between c̃1 and c̃2
to zero. For different endpoints instead the boundary conditions break worldsheet parity, and
even and odd plane waves mix at subleading order. The ratio c̃2 /c̃1 and the allowed values for
the energy are found by imposing the boundary conditions at both σ → 1 and σ → −1. We
find for n even
 
ℓ̄s π 3/2 2

5/2
c̃2 = c̃1 0 − √ (ω ℓ̄s ) (µ1,1 − µ1,−1 )n(n − 1) + O (ω ℓ̄s ) , (4.95)
2 2
 
ℓ̄s 
ϵn = n 1 + √ (ω ℓ̄s )3/2 (µ1,1 + µ1,−1 ) n2 − 1 + O (ω ℓ̄s )5/2


. (4.96)
2
For n odd, the energy levels are given by the same expression but the ratio should be inverted,
which means
 
ℓ̄s π 3/2 2

5/2
c̃2 = c̃1 0 + √ (ω ℓ̄s ) (µ1,1 − µ1,−1 )n(n − 1) + O (ω ℓ̄s ) . (4.97)
2 2
Note that the mixing between even and odd wavefunctions disappears for µ1,1 = µ1,−1 , as
expected, as well as for n = 0, 1. This is because the first two modes correspond to the action
of symmetry generators, as explained in Section 4.3.1.
For angular fluctuations, the solution to the EOM reads as in eq. (4.59) with the replace-
ment σ → αf σ + δαcm /αf . Imposing the boundary conditions we find that the coefficients of
the wave-functions (4.59) and the energy levels are given by:
 
ℓ̄s 3/2 2

5/2
c2 = c1 0 + √ (ω ℓ̄s ) µ1,1 n(n − 1) + O (ω ℓ̄s ) , (4.98)
2
 
ℓ̄s 
ϵn = n 1 − √ (ω ℓ̄s )3/2 (µ1,1 + µ1,−1 ) n2 − 1 + O (ω ℓ̄s )5/2


. (4.99)
2 2
3
2
Note that ratio c2 /c1 depends only on µ1,1 since the Legendre function Qϵ− 1 (x) in general
2
does not have any symmetry property as a function of x.

– 34 –
5 Endpoint renormalization from bulk operators

Given the construction of the boundary action in the previous section, we are now ready
to address the issue raised in Section 3: the breakdown of the derivative expansion near the
string endpoints. In this section we show that higher derivative bulk operators renormalize,
already at the classical level, the boundary action (4.37). As a zeroth order check, we observe
that the contribution (3.8) of an higher derivative operator to the leading Regge trajectory
has the same ω-scaling as that of the boundary operators in eq. (4.37). Therefore, we may
think that each bulk operator is associated with an infinite tower of boundary operators that
cancel its singular contributions.
The physical picture underlying this renormalization procedure is simple. Within EFT we
resolve the location of the effective endpoints only up to a certain uncertainty, parametrized
by the boundary cutoff. By lowering such cutoff we also effectively shorten the bulk string
length and therefore modify the contribution of the bulk operators near the boundary. For
physical quantities to remain invariant, such contributions must thus be absorbed in a change
of the boundary Wilson coefficients.
In practice we technically implement this strategy by retaining a finite value of the
redundant coefficient bm in eq. (4.37); this ensures that the contributions from bulk operators
remain finite. This also implies that, when dealing with bulk operators, we cannot work with
null endpoints as in the previous sections, and therefore we encounter unphysical powers of
ω ℓ̄s in the calculations. It would clearly be desirable to have a more convenient calculational
scheme, more analogous to a mass independent regulator, which would make power-counting
manifest at all stages of the calculation. Unfortunately, we were not able to devise such a
refined approach.19

5.1 Warm-up: massless scalar field with curvature coupling

As in Section 4, it is convenient to first illustrate the renormalization procedure in the toy
example of a massless scalar field on a rigid rotating string. We consider the system defined
by the bulk action (4.3) and the boundary action (4.20), perturbed by the following coupling
to the extrinsic curvature
Z Z 1 √
2 c
Sbulk,2 = ℓ̄s dτ dσ G K 2 (∂a)2 , (5.1)
−1 4
µ
where c is some O(1) coefficient and K 2 = Kαβ ν Gαγ Gβδ is the extrinsic curvature
ηµν Kγδ
squared. As in Section 4.1, the worldsheet metric is given by (3.3) with αf = 1 − z ℓ̄s ω.
We obtained the boundary conditions and the spectrum for the free scalar in eqs. (4.14)
3/2
and (4.25) up to order O(ω 3/2 ℓ̄s ). By power-counting, the operator (5.1) scales as ω 2 ℓ̄2s and
19
For instance, it is simple to check that considering a D = 2 + ε dimensional rotating brane and working in
dimensional regularization does not regulate the endpoint singularities discussed in Section 3. In [17] a different
scheme, which does not involve introducing a redundant operator, was used to compute the leading Regge
trajectory; that scheme does not present any technical advantage compared to ours, and it is unclear to us how
to use it in calculations involving fluctuations around the background.

– 35 –
thus it should not affect these results. As formerly explained, because of eq. (3.7) K 2 however
is not suppressed near σ = ±1. This implies that the physical boundary Wilson coefficients in
eq. (4.23) receive a contribution also from the bulk coefficient c. Intuitively, this is because
we can lower our boundary cutoff increasing z → z + δz with δz > 0 and reabsorb in the
boundary action the contribution of the bulk operators between σ = 1 and σ = 1 − δz ℓ̄s ω/αf .
Notice that already in the definition (4.23) the physical Wilson coefficients depend upon the
value of z. Additionally, the operator in eq. (5.1) contributes as
h i
Z Z 1 cℓ̄2s ω 2 αf2 ȧ2 − (1 − αf2 σ 2 )(a′ )2
δSbdry = dτ dσ + (σ → −σ)
1−δz ℓ̄s ω/αf 2αf (1 − σ 2 αf2 )5/2
Z (5.2)
δz X p h
−1 ˆ 2 3 ˆ 2 2 −2
i
≈ c 5/2 dτ γ̂τ τ #ℓ̄s γ̂τ τ (∇τ a) + #ℓ̄s (∇τ a) γ̂τ τ + . . . ,
z σ=±1

where in the second line we Taylor expanded at leading order a near σ = 1 and we used the
leading order boundary conditions (4.25). We denoted with # some O(1) numerical coefficients
whose precise value is irrelevant for our discussion. The important message is that we expect
a contribution of order c/z 3/2 for z ≫ 1 to the coefficients (4.23), such that a variation of
z can be reabsrobed by a shift of the form (5.2) in the Wilson coefficients of the boundary
action (4.20).
Let us now show how to compute the boundary conditions in the presence of the bulk
higher derivative operator (5.1). First notice that the bulk EOMs in the presence of the
operator (5.1) can be solved perturbatively in the form
!
cℓ̄ 2ω2 c2 ℓ̄4 ω 4
s s
a(τ, σ) = a(0) (τ, σ) + a(1) (τ, σ) + O , (5.3)
(1 − αf2 σ 2 )2 (1 − αf2 σ 2 )4

where a(0) is the leading order solution (4.13) and we defined the prefactor of the perturbation
such that a(1) is regular for αf σ → ±1. In the case at hand a(1) can be written explicitly as a
product of hypergeometric functions and trigonometric functions. For our purposes it suffices
to know the expansion of a(1) for 1 − αf σ → 0:
d2
 2
d4
  
4 8 d
a(1) (τ, σ) = (1 − αf σ) 2 − (1 − αf σ)2 log(1 − αf σ) + + . . . β1 (τ )
3 dτ 3 dτ 2 dτ 4
(5.4)
11 7 d2
   
p 1
+ 1 − αf σ + (1 − αf σ) + + . . . β2 (τ ) ,
3 12 3 dτ 2
where β1 and β2 are the boundary modes defined in eq. (4.24). Using peq. (5.4) in eq. (5.3)
we see that the higher derivative correction to the solution is only ∼ ω ℓ̄s suppressed with
respect to the leading order for |σ| ≃ 1. To formally retain control of the expansion we may
imagine taking z ≫ 1. In this way the field at the endpoints admits a double-expansion for
(z ℓ̄s ω) → 0 and z ≫ 1, whose schematic form up to possible logarithmic corrections is
X  2 
(0) c (1) c
a(τ, ±1) = cn (τ ) + 2 cn (τ ) + O (z ℓ̄s ω)n/2 , (5.5)
z z4
n≥0

– 36 –
 (k)
where the cn (τ )’s are functions of β1 and β2 as in eqs. (4.24) and (5.4).
To proceed we simply need to compute the variation of the bulk and boundary actions in
a double-expansion of the form (5.5). This procedure is justified since we can always choose
z ≫ 1, but in practice can be implemented for any value of z > 0. It is however important to
retain z ̸= 0 to regulate the endpoint singularities. The boundary term in the variation of the
bulk action (4.3)+(5.1) is
( p
X Z δβ1 β2 ω ℓ̄ z h
δSbulk + δSbulk,2 = bulk EOMs + dτ √ + √ s δβ2 β2 + 2δβ1 β̈1
σ=±1
2 2
 2   (5.6)
c   c

+ δβ2 β2 − 2δβ1 β̈1 + O + O zω ℓ̄s .
12z 4 z8

The variation of the boundary action is particularly simple to this order

X Z p

δSbdry = dτ b̃a ℓ̄s ωδβ1 β̈1 + O ℓ̄s ω . (5.7)
σ=±1

Summing eqs. (5.6) and (5.7) we find the solution

p
β2 ≃ ω ℓ̄s Ba(new) β̈1 , (5.8)
(new)
where Ba can be written in terms of the coefficient defined in (4.23) as
 2 
(new) c c
Ba = Ba + 3/2 + O
6z z 7/2
(5.9)
√
 2 
√ c c
= − 2 b̃a − 2 z + 3/2 + O .
6z z 7/2

As promised, the contribution of the bulk higher derivative operator can be absorbed in a
√
shift of the boundary Wilson coefficient b̃a → b̃a + c/( 2 6z 3/2 ).
Restoring the subleading orders in eqs. (5.6) and (5.7), we find that the boundary
conditions take exactly the same form as in eq. (4.25). As in eq. (5.9), the only difference is
that the coefficient Da in (4.16) is replaced by
 √ 
c 2z − 4 2z b̃a + b̄2a  2 
c
(new)
Da = Da + 3/2
+O
6z z 5/2
 √  (5.10)
c 2z − 4 2z b̃ + b̃2
√ √ 3/2
 2 
√ 4z a a c
= 2 d˜a + 2b̃2a z + 2 2 b̃a z + + 3/2
+O .
3 6z z 5/2

This is equivalent to a renormalization of the boundary Wilson coefficient d˜a .

Finally we comment that it is tempting to interpret the z-dependence of the coefficients
(new) (new)
Ba and Da as a boundary RG flow for the operators multilplied by b̃a and d˜a in the
action (4.20), with z setting the boundary cutoff. While conceptually appealing, we did not

– 37 –
find any practical use of this viewpoint. This is because all the operators of the boundary
action can be thought as irrelevant deformation of the rotating NG string with free endpoints,
and therefore, similarly to the chiral Lagrangian, no interesting RG flow takes place. Note also
that bulk and boundary operators always have different scalings in ω ℓ̄s , and therefore we do
not expect any logarithmic running for the boundary Wilson coefficients at the classical level.

5.2 Endpoints’ renormalization in the dynamical string: an example

The discussion of boundary conditions for the rotating string coordinates in the presence
of bulk higher derivative operators is completely analogous to the example of a massless scalar
that we just discussed. Intuitively, this can be understood using arguments analogous to those
presented around eq. (5.2), as formerly discussed in [17].20
Let us illustrate this point by considering the NG action perturbed by a nontrivial bulk
operator: Z Z 1 √
1
dσ G 1 − c4 ℓ̄4s (K 2 )2 .
 
Sbulk = − 2 dτ (5.11)
ℓs −1

In the bulk, the operator multiplied by c4 is suppressed by a factor (ℓ̄s ω)4 with respect to the
leading order action.
Let us consider first the leading Regge trajectory. As in the previous section, it is formally
advantageous to consider a large value of z ≡ (1 − αf )/(ℓ̄s ω). Because of eq. (4.41), this is
equivalent to taking a large value of the redundant coefficient bm . In this case, we can compute
the classical action and extremize over αf to obtain

c24
  
c4 1
αf =1 − ωbm πmℓ̄2s + 4 +O
bm 4π 3 m3 ℓ̄2s b8m
"
2
√ p
2  2 #
π bγ c4 5 2mbm bγ − 12 π ℓ̄s mbm c
+ ω 2 b2m m2 ℓ̄4s − √ 3/2 π 3/2 m3/2 ℓ̄7/2
s + 4 p + O 84
2 2bm bm 16π 5/2 ℓ̄s m3 b2m bm
+ O ℓ̄3s ω 3 ,

(5.12)

where we work in a double-expansion analogous to eq. (5.5). Namely, at each order in ω ℓ̄s we
have a series in c4 /b4m .
The rest of the calculation proceeds as in Section 4.2. Eventually we recover the result
20
Unfortunately, it seems hard to turn such physical arguments based on explicitly shortening the integration
region of σ into concrete calculational schemes. For instance, in [17] the authors use the leading order boundary
conditions to rewrite operators integrated along the small strip as in eq. (5.2), while as we have seen the string
length is not invariant under such replacements.

– 38 –
(4.50) with the coefficients µ1 and µ2 now replaced by
!
(new) c4 1 c24
µ1 = µ1 − 5/2 7/2
+O 13/2
,
bm 10 π 7/2 m5/2 ℓ̄s bm ℓ̄s
p ! !
(new) c4 bγ 5 ℓ̄s m c24
µ2 = µ2 + 3/2 √ 3/2
− 5/2 3 2
+O 11/2
, (5.13)
bm 8 2 π 3 bm ℓ̄3s m2 24 π ℓ̄s m bm ℓ̄s
where µ1 and µ2 were given in eqs. (4.39) and (4.40). As in the previous section, the shifts in
eqs. (5.13) can be understood as a renormalization of the coefficients bγ and b1 in eq. (4.37).
It is also possible to check that the boundary conditions (4.69) and (4.71) for the fluctua-
tions are unchanged up to the shifts in eq. (5.13). To see this one proceeds as in the previous
section. For instance, the new operator (5.11) modifies the solutions to the bulk equations
similarly to eq. (5.3), e.g. for the transverse modes we have
(0) ℓ̄4s ω 4 (1)
ρi (τ, σ) = ρi (τ, σ) + c4 2 2 4
ρi (τ, σ) + . . . , (5.14)
(1 − αf σ )

where the ρ(i) are normalized such that they are regular for αf σ → ±1:
X
ρ(i) (τ, σ) = cn(i) (τ )(1 − αf σ)n/2 , (5.15)
n≥0

which again holds up to logarithms. Analogous expressions hold for ϕ. We provide the explicit
results for the near boundary expansion in the presence of the operator (K 2 )2 in appendix B.
For bm ≫ 1 we impose the boundary conditions at a point (1 − αf σ) = zω ℓ̄s ≫ ω ℓ̄s . We
thus again organize the near boundary expansion for ρ (and ϕ) as of a double expansion
analogous to eq. (5.5) (neglecting O(d − 3) indices):
Xh c4 (1) i
ρ(τ, 1) = c(0)
n (τ ) + c (τ ) + . . . (zω)n . (5.16)
z4 n
n≥0

We checked that, writing the variation of the bulk and boundary action using such expansions
for ρ and ϕ as in eqs. (5.6) and (5.7), we eventually recover the former results (4.69) and (4.71)
for the boundary conditions (and thus for the spectrum, cf. (4.70) and (4.73)), with the
(new) (new)
replacement µ1 → µ1 and µ2 → µ2 .

Part II
Extensions and phenomenology
6 Additional degrees of freedom in 4d

To apply our results to mesons with light quarks in real world QCD, it is important to
discuss two important extensions of the formalism that we have presented so far: the role
of spin at the endpoints and the inclusion of the pseudoscalar resonance on the worldsheet
discussed in Section 2.2. In this section we therefore specialize to d = 4.

– 39 –
6.1 Endpoints with spin
In QCD, the string endpoints are expected to be endowed with a 2-dimensional quantum
mechanical system describing different spin states of a quark. As we review in appendix C,
it is possible to describe the spin degrees of freedom for a point particle of arbitrary spin
s using the coadjoint orbit technique. This amounts to considering two angles (θs , ϕs ) that
parametrize the orientation of the particle spin in the rest frame via
⃗rest = s (sin θs cos ϕs , sin θs sin ϕs , cos θs ) .
S (6.1)

To write the spin-component of the Lorentz generator in an arbitrary frame, it is convenient

to introduce a bosonic Dirac spinor living on the worldline as
! !
1 z √ e−iϕs /2 cos θ2s
ψ = √ Λu · where z = 2 , (6.2)
2 z eiϕs /2 sin θ2s
p
where Λu is the Dirac boost matrix associated with the four-velocity uµ = Ẋ µ / Ẋ 2 of the
particle. The spin components of the Lorentz generators is then written in terms of ψ as
i
Sµν = sψ̄[γµ , γν ]ψ , (6.3)
4
from which we obtain the Pauli-Lubanski vector: S µ = 12 εµνρσ uν Sρσ . Notice that these satisfy
Sµ uµ = Sµν uν = 0. Finally, in the free limit, the worldline action for the spin degrees of
freedom reads: Z
Sspin = s dτ iψ̄ ψ̇ . (6.4)

It is possible to show that the action (6.4) leads to the expected canonical commutation
i , S j ] = iεijk S k
relations [Srest ⃗2
rest rest and the constraint Srest = s(s + 1).
Before discussing additional interactions, it is interesting to add the term (6.4) at the
string endpoints. Indeed eq. (6.4) already encodes interactions between the spin degrees of
freedom, (θs , ϕs ), and the string coordinates via the boost matrix in eq. (6.2). We work in the
massive endpoint formalism (4.37) for simplicity. Expanding (6.4) around the background (3.2)
we find
Z Z  
Ω
s dτ iψ̄ ψ̇ = s dτ cos θs ϕ̇s + cos θs , (6.5)
ω
where we defined a positive frequency as
 
r
1 1 ω
Ω ≡ ω q − 1 ≃ ; (6.6)
1 − αf2 ℓ̄s 2πbm m

in the second line we used (4.41) and expanded for mω ≪ 1/ℓ2s . From the EOMs of the spin
angles, we find that at the classical level the spin moves on trajectories specified by

θs = const. , ϕs = −Ωt + const. (6.7)

– 40 –
where t = τ /ω. This is just the well known Thomas spin’s precession for an accelerated
particle.21 Quantum-mechanically, since S 3 = s cos θs , eq. (6.5) implies that (in physical units
t = τ /ω) the Hamiltonian for the spin variables reads

H = −ΩS 3 . (6.9)
p
Therefore different spin states are separated by a gap Ω ∼ ω/ℓ̄s . Recalling eq. (4.2), we
conclude that the splitting between the different spin states is of the order of the strong
coupling scale at the boundary.
Being at the cutoff scale, the gap between the lowest spin state and the excited ones is not
calculable within EFT. Notice that indeed eq. (6.8) depends on the redundant coefficient bm .
Additionally, we show below that the splitting between the spin states receives contributions
from infinitely many operators.
Let us consider couplings between the spin tensor Sµν , or equivalently S µ , and the endpoint
worldline. To estimate the size of the interaction terms, it is convenient to write S µ as

S µ = huµ + δS µ , (6.10)

where h = ⃗u · S⃗rest /u0 ∼ O(1) is the helicity, uµ ∼ 1/√ω and δS µ ∼ O(1). The first term in
eq. (6.10) is large despite the constraint S · S = −s2 (at the classical level) due to the boost
factor.
Inverting eq. (6.10) via Sρσ = εµνρσ S µ uν and recalling Sµν uν = 0, we identify the
following leading interaction term compatibly with parity and time reversal (notice that S µ is
a pseudovector)
Z
dτ γ̂τ τ f (Os ) where Os = Sµν ∇2τ X µ ∇3τ X ν γ̂τ−5/2
p
τ . (6.11)

Here f is an arbitrary function.22 Evaluating (6.11) on the background (3.2) we find

f (S 3 )
Z
dτ p . (6.12)
ω ℓ̄s

Therefore the interaction (6.11) contributes to the splitting between different spin states at
the the same order of the free term. Since f is an arbitrary function, the splittings between
different spin states does not need to scale linearly with S 3 (for endpoints in spin s > 1/2
21
Thomas precession is usually stated in terms of the following equation [51]

Ṡ µ = −uµ u̇ · S . (6.8)
i
This equation can be derived noticing that the fact that Ṡrest = 0 in the rest frame for a free particle is
equivalent to Ṡ ∝ u in covariant form. Eq. (6.8) follows then using that S µ uµ = 0 implies dτ
µ µ d
(S · u) = 0 (see
e.g. [52]). It is simple to check that eq. (6.7) is indeed a solution of eq. (6.8).
22
For f (x) = x, the coupling (6.11) (and its equivalent formulation in eq. (7.27)) induces the long-range
interaction between spin at different endpoints first discussed in [53].

– 41 –
representations). In practice, EFT breaks down at energies smaller than the ones needed to
excite the first nontrivial spin state.
In conclusion, the different spin states of the quarks are p separated by (uncalculable)
splittings of the order of the cutoff at the endpoints, Λbdry ∼ ω/ℓ̄s . Therefore, within EFT
we are forced to integrate out all the spin states but the lowest one. We conclude that the
only effect of spin degrees of freedom at the endpoints is to renormalize the boundary Wilson
coefficients in eq. (4.31), which are anyway uncalculable from the low energy viewpoint.
Physically, the degeneracy between the spin states is broken due to the accelerated motion
of the endpoints, that implies a nontrivial spin precession. Due to the sign of the Thomas’
force (i.e. the minus sign in eq. (6.9)), this argument suggests that in the lowest energy state
the quarks’ spin align with the orbital angular momentum of the meson. This possibility is
sometimes referred to as spin-orbit inversion [34], in comparison with the usual spin-orbit
coupling in atomic physics. We remark however that the interaction terms (6.11) do not have
a fixed sign and may therefore affect this conclusion. In fact, it is also possible that for light
quarks the spin states become infinitely heavy and completely decouple for sufficiently large J.
We will nonetheless argue in favor of the spin orbit inversion phenomenon in Section 7 for
heavy quarks’ mesons with large orbital angular momentum.
Finally, we remark that, from the viewpoint of symmetry breaking, it should not come
as a surprise that the spin degrees of freedom are gapped. Indeed, for a point particle, the
spin angles (θs , ϕs ) can be thought as Goldstone fields parametrizing the symmetry breaking
pattern SO(3) → SO(2), where SO(3) is the rotation group in the rest frame of the particle.
The rotating string breaks spontaneously the rotation group even for spinless endpoints, with
the string coordinates accounting for its nonlinear realization. Therefore the spin angles at
the endpoints provide redundant Goldstone fields. This is not a new phenomenon: it is often
the case that in the presence of spontaneously broken spacetime symmetries one can introduce
redundant Goldstones, see e.g., [37, 54, 55]. In all the examples of this sort that we are aware
of, such additional Goldstones behave similarly to generic matter fields, and acquire a gap.
This is also the fate of the spin states at the string endpoints.

6.2 Worldsheet axion states on the rotating string

As we reviewed in Section 2, lattice data revealed the existence of a massive pseudoscalar
resonance on the worldsheet. As explained in Section 2.2, in EFT we describe it via the
action (2.9), pretending that its mass is sufficiently small ma ≪ ℓ̄−1
s . In this section we analyze
this massive scalar mode on the rotating string background (3.2). The regime of interest to us
is
ω ≪ ma ≪ ℓ̄−1 s . (6.13)
Due to the nontrivial dependence on σ of the worldsheet metric (3.3), the energy levels
of a free massive particle have a nontrivial dependence on ma and ω. To see this, consider
the free EOM for the worldsheet axion in frequency space a(τ, σ) = e−iϵτ a(σ) (neglecting the

– 42 –
higher-derivative coupling to K · K̃):

m2
− 1 − σ 2 a′′ (σ) + σa′ (σ) + 2a 1 − σ 2 a(σ) = ϵ2 a(σ) ,



(6.14)
ω
where we set αf = 1 for simplicity since we are dealing with light quarks. Changing coordinates
as in eq. (4.5), we recast (6.14) as

m2a
−a′′ (s) + cos2 (s)a(s) = ϵ2 a(s) . (6.15)
ω2
In appendix D we discuss the general solution of (6.15) in terms of Mathieu functions. Here
we present an approximate analysis in the regime (6.13), that captures the essential results.
Eq. (6.15) has the same structure of the Schrödinger equation for a particle in a potential
m2a
ω2
cos2 (s). In the limit of interest ma ≫ ω, such potential localizes the field near the endpoints
s = ± π2 . To look for approximate solutions, it is convenient to rewrite s = ±(π/2 − ω/ma x),
p

so that near each endpoint we obtain the following equation to leading order in ω/ma

1 x2 ω 2
− a′′ (x) + a(x) ≃ ϵ a(x) with x > 0 . (6.16)
2 2 2ma
Notice that in the approximation we are working with we obtain two copies of eq. (6.16) at the
two endpoints. We thus conclude that the particle spectrum of the axion is doubly-degenerate
up to corrections exponentially suppressed in ma /ω, physically associated with the tunneling
between the localized states.
Eq. (6.15) is the Schrödinger equation for the harmonic oscillator in half-space with energy
ω 2
levels 2m a
ϵ . The allowed values of ϵ depend on the boundary conditions at x = 0;23 we will
discuss below the possibilities for the example at hand. For the moment we simply notice that
the eigenfunctions for Dirichlet and Neumann boundary conditions are obtained restricting
the usual wave-functions for the harmonic oscillator to, respectively, even and odd quantum
numbers. Therefore we have solutions for
(
ω 2 2k + 21 for Neumann
ϵ = with k ∈ N , (6.17)
2ma 2k + 3 for Dirichlet
2

where N stands for natural numbers including zero. In any case the energy of a single-particle
p
axion mode scales as ϵ ∼ ma /ω. This is an important conclusion. In fact, this result shows
that the axion states are below cutoff for ma ≪ ℓ̄−1
s whenp we account for the redshift factor
at the boundary. Indeed eq. (4.2) demands ωϵ ≪ Λbdry ∼ ω/ℓ̄s .
Let us now briefly discuss the boundary conditions for a. In the regime (6.13), a is
approximately invariant under constant shifts. We would like to understand if the string
endpoints also preserve such approximate shift symmetry, in which case the Neumann condition
would be a good approximation as in Section 4.1; conversely, if the (approximate) shift
23
In the approximation leading to (6.16) we simply demand regularity at x → ∞.

– 43 –
symmetry is explicitly broken at the string endpoints, one expects a Dirichlet condition to
provide a better approximation.
One the one hand, even for ma = 0, the axion-like coupling in eq. (2.9) always breaks the
shift invariance of a in the presence of a boundary. This is because K · K̃, despite being a
topological invariant, cannot be written as a total derivative of a gauge-invariant invariant
quantity, as showed in eq. (A.10) of the appendix. This suggests that the string endpoints
will always break the approximate shift symmetry. Therefore, a priori nothing prevents us
from considering an arbitrary potential V (a) at the string endpoints, where V (a) = V (−a)
due to parity. Under the assumption that a = 0 ispa global minimum of the potential, this
would set a|σ=±1 = 0 up to corrections of order ∼ ω ℓ̄s .24
Note that the above theoretical argument suggests that the approximate shift symmetry
for a should also be broken at the boundaries of a long static flux tube. This would imply
that the gap of the worldsheet axion in a long open flux tube of length L should be larger by
a factor ∼ π 2 /(ma L2 ) compared to a closed string due to the Dirichlet boundary condition.
Perhaps surprisingly, lattice data suggest instead that the lightest worldsheet axion state in
a static open flux tube has approximately the same gap as in a closed flux tube, even for
relatively short strings [32, 33]. This surprising result suggests that the above theoretical
argument, hinting at Dirichlet-like boundary conditions for a, should be taken with caution.
For this reason, we prefer not to commit to a specific choice of boundary conditions.
In summary, the low energy excitations of the worldsheet axion in the rotating string
background are localized at distance (1 − |σ|) ≲ ω/ma from the endpoints. This implies that
at large J the corresponding modes have an approximately doubly degenerate spectrum. The
result may be summarized by the formula
q  
n
ma  4 2 + 3 for Dirichlet
r 
ϵn ≃ × q   with n ∈ N , (6.19)
ω 
 4 n2 + 1 for Neumann

where we reported the result both for Dirichlet and Neumann boundary conditions and [n/2]
denotes the integer part of n/2 (notice that the lowest energy state corresponds to n = 0).
In appendix D we provide the exact result for arbitrary values of ma /ω. Since in practice in
24
As well known, for Dirichlet boundary conditions it is convenient to write the boundary actions in terms of
normal derivatives (see e.g. appendix B of [56]). In the case at hand, one could consider an action of the form
X Z
Sbdry = − dτ σ a∂s a
σ=±1
X Z  h i2  (6.18)
c1 ℓ̄s (∂s a)2 γ̂τ−1 −2
p 3 ˜
− dτ γ̂τ τ τ + c2 ℓ̄ s ∇ τ (∂s a) γ̂ ττ + . . . ,
σ=±1

where σ∂s a|σ=±1 = Nτα ∂α a in terms of the covariant normal defined in eq. (A.19). The first term in eq. (6.18)
is designed to cancel the boundary term in the variation of the bulk action ∼ δa∂s a. The boundary conditions
then arise from the variation with respect to ∂s a (which is independent of a at the endpoints). The firstp term
in eq. (6.18) then scales like ω 0 . The terms proportional to c1 and c2 in eq. (6.18) scale as, respectively, ω ℓ̄s
and (ω ℓ̄s )3/2 .

– 44 –
QCD ma is close to the cutoff, we do not consider higher derivative corrections in ℓ̄s ma to this
√ p
−1/4
result. Note that the axion states have a gap δM ∼ ωϵ ∼ ma ω ∼ ma /ℓ̄s J . Therefore,
for ma ∼ ℓ̄−1
s , the gap of the axion scales as the contribution of an endpoint mass term (4.26)
to M .

7 Heavy quarks

In order to understand how EST relates to heavy quarkonium states and what are the
different regimes as a function of the quarks’ masses, in this section we develop the EFT for
mesons with heavy quarks, i.e., with mass m ≫ ℓ̄−1 s . The EFT of mesons with heavy quarks is
not plagued by the endpoints’ singularities that we mentioned in Section 3, and it is therefore
technically straightforward. (In fact, some of the calculations that we present have already
appeared in [28]). Nonetheless, the existence of an additional large scale, the quark mass m,
makes power counting more involved compared
p to mesons with light quarks.
We will find that one needs J ≫ ℓ̄s m for the flux tube to be much longer than ℓ̄s ,
pushing the validity regime of the EFT to higher values of J compared to light quarks. We
will also show that the parameter y ≡ J/(m2 ℓ̄2s ) controls the EFT spectrum. In the relativistic
regime y ≫ 1, the mass becomes a small perturbation and we recover results similar to those
that we discussed so far. In the opposite regime, y ≪ 1, the lowest energy states of the
effective theory coincide with the energy levels of two heavy particles in a linear potential.
Notably, in this regime the quarks’ spin states are below the cutoff, and can be analyzed
within EFT differently pthan for light quarks (cf. Section 6.1). We will use this to show that in
the EFT regime (J ≫ ℓ̄s m), the quarks’ Thomas precession leads to spin-orbit inversion of
the energy levels.

7.1 Double-scaling limit

Let us again consider the NG action supplemented with the endpoint mass term (4.26).
For a heavy quarkonium meson the quarks velocity αf does not need to be close to the speed
of light for the string to be long. Recalling eq. (4.28), we may therefore formulate the EFT in
the following double-scaling limit

ℓ̄s ω → 0 with m̃ ≡ mωℓ2s = 2πmω ℓ̄2s = fixed , (7.1)

which is equivalent to
J
J → ∞ with y ≡ = fixed . (7.2)
m2 ℓ̄2s
In other words, we can use EFT to obtain result for the spectrum in a series at large J with
coefficients that are functions of J/(m2 ℓ̄2s ).
For heavy quarks, the endpoints’ redshift factor is controlled by the fixed parameter
mωℓ2s and is not necessarily large. Therefore we can use the bulk power counting also at
the endpoints, i.e. X ∼ 1/ω and Gαβ ∼ 1/ω 2 . Due to the factor of m upfront, in the
double-scaling limit (7.1), the quark mass term (4.26) scales as the NG bulk action and cannot

– 45 –
be thought as a small perturbation of the Neumann boundary conditions. The scaling of
higher derivative boundary terms depends on the Wilson coefficients’ dependence on m. In
the simplest scenario, which is also the one of phenomenological interests in QCD, we think
of the endpoints as fundamental particles whose free action is exactly given by (4.26). Note
that all higher derivative terms for a free point particle are redundant due to the equations
of motion [55]. In this case, higher derivative boundary terms can be thought as the result
of the interactions with the string. Therefore, we expect them to be suppressed simply by
ˆ t with respect to the free-particle action.
powers of ℓ̄s ∇
In summary, using the equivalence (4.29) between ℓ−2 ˆ 2
s ∂n X and m(∇t X) dictated by the
leading order boundary conditions (now being careful about the factors of m’s), we conclude
that to the first nontrivial subleading order, the boundary action can be written in the same
form as eq. (4.31) after rescaling b̄1 → b̄1 /(ℓ̄2s m2 ):
X Z q  
b̄1  ˆ ˆ

Sbdry = −m dτ Ĝτ τ 1 + 2 ∇t ∂n X · ∇t ∂n X + . . . , (7.3)
m
σ=±1

where we used the fact implied by eq. (4.27) that (∇ ˆ 2t X)2 is a redundant operator and the last
term in eq. (7.3) is equivalent to ∼ ℓ̄4s (∇ ˆ 3t X)2 . As formerly mentioned, in the double-scaling
limit (7.1), the leading term in eq. (7.3) contributes at order m/ω ∼ 1/(ℓ̄s ω)2 , which is the
same as the NG action, while the last term in eq. (7.3) is suppressed by a relative factor of
ω 2 /m2 ∼ (ℓ̄s ω)4 , i.e. it is fourth order in derivatives according to the power-counting that we
just discussed.
Let us now illustrate our discussion concretely by calculating the energy of the leading
Regge trajectory. Taking the bulk action to be just the NG one25 , from the boundary conditions
we obtain the value of αf as
αf = αf,0 + (ℓ̄s ω)4 αf,1 + . . . , (7.4)
where
1 p  4π 2 b̄
αf,0 = 4 + m̃2 − m̃ , αf,1 = − √ 1 . (7.5)
2 m̃2 4 + m̃2
Expressing m̃ in terms of αf,0 and using eq. (3.5), the angular momentum and the mass are
conveniently expressed as
q  
2 + arcsin(α )
αf,0 1 − αf,0 2ω2α 2
f,0 4π b̄ ℓ̄
1 s f,0 2 + α f,0 6 6


J= 2 2
+ 3/2   + O ℓ̄s ω , (7.6)
2π ℓ̄s ω 2 2
1 − αf,0 1 + αf,0
q
2 /α
1 − αf,0 f,0 + arcsin(αf,0 ) 4π b̄1 ℓ̄2s ω 3 αf,0 6 7


M= + + O ℓ̄ s ω . (7.7)
π ℓ̄2s ω 3/2
   
1 − αf,0 2 2
1 + αf,0
25
At fourth order in derivatives, there are also bulk operators that contribute at the same order of b̄1 /m2 in
eq. (7.3) in the double-scaling limit (7.1); we nonetheless neglect such bulk contributions since our goal here is
simply illustrating the structure of the expansion.

– 46 –
Note that when the endpoints have spin, eq. (7.6) is just the orbital contribution to the angular
momentum. From eqs. (7.5) and (7.6) we can solve for ω in terms of J (now using m̃ = mωℓ2s ).
We find
1 b̄1
ℓ̄s ω = √ g1 (y) + 5/2 g2 (y) + . . . , (7.8)
J J
where we defined two functions g1 (y) and g2 (y). In the two opposite limits these are given by:
√
3π 3/2

1 π
−
 2 3y 3/4 20y 5/4 + . . .
+ for y ≫ 2π



g1 (y) = (7.9)
 y 1/6 7y 5/6
− + . . . for y ≪ 1 ,



(2π 2 )1/3 36(2π 2 )2/3
 3/4 √
3y 5 π 1/4
 8√π + 32 y + . . . for y ≫ 2π



g2 (y) = (7.10)
 24/3 y 5/6 871y 13/6
−
+ ... for y ≪ 1 ,


3π 4/3 19440 21/3 π 8/3


where y ≫ 1 corresponds to the ultra-relativistic limit J ≫ ℓ̄2s m2 , while y ≪ 1 is the

non-relativistic (NR) limit J ≪ ℓ̄2s m2 . Plugging back into eq. (7.5), we extract αf,0
 !
1 π2 b̄1 3 √ 1/4 π 3/2
r 
 1
 1 − 2 π y + 8y + . . . − J 2 8 πy + 32y 1/4 + . . . + . . . for y ≫ 1



αf,0 = ! !
 y 1/3 11y b̄1 2y 1022/3 y 5/3
 22/3 π 1/3 − 72π + . . . − J 2 3π − 27π 5/3 + . . . + . . . for y ≪ 1 .




(7.11)
Finally, using eq. (7.11) we obtain M from eq. (7.7). In the relativistic limit, after
re-arranging terms , we recover the previous result (4.36) up to the rescaling b̄1 → b̄1 /(m2 ℓ̄2s )
that we performed in going from the boundary action for light quarks (4.31) to our current
conventions in eq. (7.3). Note that although the expansion takes the same form in terms
of J for both light and heavy quarks, the quark mass provides the leading contributions to
several terms in the large J expansion in the case of heavy endpoints. E.g. the coefficient
of the term proportional to ∼ J −1/4 in eq. (4.36) is ∼ (m4 + 10b̄1 /ℓ̄4s )/m3/2 (recalling the
rescaling of b̄1 ), and therefore is dominated by the mass contribution for heavy quarks. In
other words, for heavy quarks all the 1/J (classical) corrections to the relativistic Regge slope
are approximately determined only by the quark mass, as expected.
The result takes a qualitatively different form in the NR limit J ≪ (ℓ̄2s m2 ). First, notice
that in this limit the length of the string (3.4) is given by
1/3
4πJ 2

2αf,0
L≃ ≃ ℓ̄s . (7.12)
ω ℓ̄s m
Therefore, the EFT is reliable only as long as
p
J ≫ ℓ̄s m . (7.13)

– 47 –
Physically, eq. (7.12) arises because in this regime the meson can be thought as a bound state of
two NR particles in a linear potential. In such a system the Virial theorem relates the potential
and the kinetic energy L/ℓ2s ∼ mω 2 L2 , while the angular momentum is J ≃ mωL2 , which
√ −4/3
results in eq. (7.12) and ω ∼ ℓ−1 s / Lm ∼ ℓs /(mJ)1/3 as formerly found. In Section 7.2,
we will discuss the derivation of the NR quantum-mechanical Hamiltonian from EST in more
detail. In the NR limit, the mass of a state on the leading Regge trajectory reads
"  2/3 # " 4/3 #
3 J 2b̄1 m J
M = 2m 1 + + ... − + . . . + . . . , (7.14)
4(2π 2 )1/3 ℓ̄2s m2 (2π 2 )2/3 J 2 ℓ̄2s m2

where we arranged the result in a double-series according to eq. (7.2). We see that the leading
contribution in the NR limit is the rest mass of the quarks. The second term in the first square
parenthesis of eq. (7.14) is the first nontrivial contribution and arises both from the string
tension and the kinetic energy of the quarks. Finally, the term proportional to b̄1 in eq. (7.14)
scales as ℓ̄4s mω 6 L2 and is suppressed with respect to the first nontrivial contribution by a
factor of ∼ 1/(L2 m2 ) in terms of the length (7.12). In the next section, we will discuss the
scaling of corrections in the NR regime in more detail. In particular, we will discuss quantum
corrections that we neglected so far and are larger than the b̄1 term in eq. (7.14).

7.2 String excitations and quantum effects in the non-relativistic limit

The spectrum of fluctuations in the NR regime J ≪ m2 ℓ̄2s takes a qualitatively different
form compared to the relativistic limit J ≫ m2 ℓ̄2s . A general analysis was done in [28]; we
checked the results and reported some details in appendix E. Here, we provide a simple
interpretation of the results in the non-relativistic limit.
We find that in the relativistic limit, the spectrum takes the same form as for light quarks,
as one would expect. In the NR regime, on the other hand, due to the slow motion, most of
the modes behave as in a static string with Dirichlet boundary conditions and therefore have
energy ϵ ≳ π/(ωL) ∼ 1/αf,0 (in units of ω). The only states which make exception are the
protected states whose frequencies are fixed to be ϵ = 0, 1 and do not contribute to the mass
spectrum as formerly explained, and the lightest nontrivial state associated with the angular
√
mode, which has a gap ϵ2 ≃ 3. This mode is physically associated with the stretching of the
string and is parametrically lighter than all the others nontrivial states (because αf,0 ≪ 1 in
the NR limit).
To illustrate the physics of the NR regime clearly, it is instructive to derive the result for
the leading Regge trajectory and the light excitations’ gap ϵ2 in the language of a potential
quark model. To this aim, we first restore powers of the speed of light c explicitly and then
take the limit c → ∞ in the leading order action. In this limit, the Dirichlet-like modes have
energy πc/L → ∞ and therefore decouple. Using ηµν = diag(c2 , −1, −1, . . .) and X µ = (t, X), ⃗

– 48 –
the leading order action is
T
Z √ X Z p
S=− d2 σ G − mc dτ Ẋ 2
c
σ=±1
Z " Z 1 q  # (7.15)
m X ˙
⃗2 −T ⃗ +O 1
⃗ · ∂σ X
→ dt −2mc2 + X dσ ∂σ X ,
2 −1 c2
σ=±1

where we defined T = c/ℓ2s . More precisely, the corrections to the above results are suppressed
by inverse powers of the double-scaling parameter αf,0 2 ∼ v 2 /c2 ∼ T 2 /(m2 ω 2 c2 ) = m̃−2 .
Equivalently, in terms of the angular momentum, corrections are proportional to inverse
powers of y 1/3 , where y is the large double-scaling parameter in eq. (7.2); note that after
restoring the speed of light, y 1/3 is given by
1/3 1/3
m2 m2 ℓ̄2s c2
 
1/3
y =c = . (7.16)
2πJT J

As expected in the NR limit, only the boundary fields have a kinetic term. Solving the
bulk equations of motion, we therefore arrive at the expected result
Z
⃗˙ CM + m ∆X⃗˙ 2 − T |∆X|
h i
S = dt −2mc2 + mX ⃗ , (7.17)
4
where we wrote the action in terms of the center of mass coordinates in obvious notation. In
the following discussion, we neglect the center of mass variables.
At large J, the system (7.17) can be quantized semiclassically similarly to the rigid
rotor or the hydrogen atom (see e.g. [47, 57]). To this aim, we work in spherical coordinates
∆X⃗ = r(sin θ cos φ, sin θ sin φ, cos θ) and expand the field around the minimal energy solution
of the EOMs with fixed angular momentum J:
π
r = r0 , θ= , φ = ωt , (7.18)
2
where the EOMs and the angular momentum fix r0 and ω as
 13  31
2J 2 2T 2
 
m m 2
r0 ω 2 = T , r ω=J =⇒ r0 = , ω= . (7.19)
2 2 0 mT Jm

The result (7.14) for the leading Regge trajectory is then obtained by evaluating the classical
energy on the solution (7.18):
  
2 3 2 2 1
Mclassical = 2mc + mr0 ω 1 + O 2
4 c
 2 2  13 " 2/3 T 2/3
!# (7.20)
2 J T J
= 2mc + 3 1+O .
4m m4/3 c2

– 49 –
To compute the spectrum of fluctuations, we simply need to quantize the theory at leading
order in the 1/J expansion. This is accomplished expanding ∆X ⃗ around the solution (7.18).
We obtain the following quadratic action
m m m
L ≃ (δ ṙ2 − 3ω 2 δr2 ) + r02 (δ θ̇2 − ω 2 δθ2 ) + (r0 δ φ̇ + 2ωδr)2 . (7.21)
4 4 4
Eq. (7.21) describes a massless mode and two harmonic oscillators, with frequencies ω and
√
3ω. The massless mode and the mass ω oscillator correspond to symmetry generators
(equivalent to the n = 0 and n = 1 modes analyzed in Section 4.3.1) and thus do not create
new particles. Therefore, at a quantum level the nontrivial states are made of n ≥ 0 quanta of
the radial oscillator, each with energy (in units of ω)
√
ϵ2 = 3 . (7.22)
√
The corresponding particles have mass Mclassical + n 3ω.
Finally, let us analyze the structure of quantum corrections. We focus on the leading
Regge trajectory (7.20). To the first subleading order, the corrections arise from the one-loop
Casimir energy. This consists of two contributions. The first is due to the heavy Dirichlet-like
modes, that at one-loop order simply modify the linear potential by the Casimir contribution
of (d − 2) massless scalar modes:
d−2 X∞     
⃗ ⊃
X πcn 1 π(d − 2) 1
V (|∆X|) 1+O 2 = −c +O . (7.23)
|∆ ⃗
X| c 12|∆ ⃗
X| c
k=1 n=1

This term is suppressed by ℓ̄2s /L2 ∼ c/(T L2 ) with respect to the leading order potential T L;
note the explicit factor of c due to the relation 2πT /c = ℓ̄−2
s . The other contribution is simply
due to the oscillators in eq. (7.21) and trivially reads
√   
1+ 3 1
ω 1+O 2 . (7.24)
2 c
Summing up, we arrive at
√  
(1−loop) π(d − 2) 1 + 3 1
δM = −c + ω+O
12r0 2 c
1/3 " !# (7.25)
√  JT 1/3 J 2/3 T 2/3
  
mT π(d − 2) 
=− c − 1+ 3 +O ,
2J 2 12 2m2 m4/3 c

which is suppressed by a factor y 1/3 /J compared to the kinetic term in eq. (7.20). As formerly
√
noted, the expansion is well posed only as long as y 1/3 /J ≪ 1, i.e. J 2 ≫ mc3/2 T ∼ mcℓ̄s .
The structure of higher derivative and loop corrections is analogous to the one-loop
result (7.25). At each order, the most important contribution arises from integrating out
the heavy modes, which yield to contributions to the potential starting at relative order
(ℓ̄s /L)2n ∼ y n/3 /J n (due to the Dirichlet-like modes Casimir energy) with additional terms
suppressed in y. Bulk higher derivative terms result into corrections of order ∼ y k/3 /J n with
k < n, while quantum corrections from the fluctuations of the point-like particles are always
suppressed by 1/J n . Boundary higher derivative terms are further suppressed in y.

– 50 –
7.3 Quarks’ spin and spin-orbit inversion
Let us analyze the quarks’ spin states for heavy quarks. The kinetic term for the spin
is again given by eq. (6.4), and yields the Hamiltonian (6.9). The main difference between
light and heavy quarks is that in the latter case the gap Ω is not large and, in the NR limit
J ≪ m2 ℓ̄2s c2 , is given by
ω 2 cJ 1/3 /ℓ̄s
Ω ≃ αf,0 = , (7.26)
2 (4π)4/3 (mc ℓ̄s )5/3
where we keep track of the speed of light c as in the previous section.
All the other couplings between the string coordinates and the spin variables involve
higher derivative terms and therefore are subleading compared to the kinetic term contribution.
The most important nontrivial contribution arises at first order in derivatives from an operator
similar to eq. (6.11), explicitly given by

mℓ̄3s
Z q Z q
ˆ 2 µˆ3 ν 1 ˆ t ∂n X ν ,
dτ Ĝτ τ Sµν ∇t X ∇t X ∼ dτ Ĝτ τ Sµν ∂n X µ ∇ (7.27)
c4 mℓ̄s c2
where we used eq. (4.29) keeping track of the factors of m (and c) as commented above
eq. (7.3).26 Evaluating the operator (7.27) on the classical background, we find that its
contribution to the gap between the spin states in the NR limit is proportional to
 1/3
ω mℓ̄s c
δΩ ∝ ∼Ω× , (7.28)
mℓ̄s c J2

which is always smaller than eq. (7.26) in the regime of applicability of EFT (7.13).
In conclusion, the leading contribution to the energy splitting between the different spin
states within EFT arises from the free Thomas precession effect and is given by eq. (7.26).
This implies that in the state that minimizes the energy the quarks’ spin are aligned with
the orbital angular momentum, therefore realizing the spin-orbit inversion scenario
p alluded
in Section 6.1. Note however that this prediction is reliable only for J ≫ mcℓ̄s , as the
extrapolation of the interaction contribution in eq. (7.28) makes clear.

8 A stringy view on meson spectroscopy

In this section, we compare the predictions of EST with real world data, both for light and
heavy quarks’ mesons. Unfortunately, for most mesons’ families we barely have data beyond
J = 2, making a quantitative comparison impossible. Nonetheless, motivated by the success
of similar descriptions for glueballs in 3d Yang-Mills [19, 20], we will attempt at a qualitative
description of the observed spectrum in terms of stringy states, i.e. as particles predicted by
EST.
26
The power of m is also justified on physical grounds if we interpret eq. (7.27) as a generalized version of
⃗ · S/m,
the Pauli interaction term B ⃗ ⃗ ∼ ℓ̄−1
with the magnetic field identified as B ⃗ ⃗˙
s ∂n X ∧ ∂n X [53].

– 51 –
8.1 Discrete quantum numbers
Let us consider a meson made of a string connecting a quark and anti-quark. Besides
its mass and spin J, we can measure its quantum numbers under parity (P ) and charge
conjugation (C). Therefore, in order to compare our analysis to the experimental data, we
need to discuss the assignments under P and C for the daughter Regge trajectories (including
the ones associated with the worldsheet axion) in d = 4.
Parity acts in the standard way X i → −X i . In terms of the string fluctuations (4.51) and
the worldsheet axion, therefore

P ϕ(τ, σ)P = ϕ(τ, σ) + π , P ρ(τ, σ)P = −ρ(τ, σ) , P a(τ, σ)P = −a(τ, σ) . (8.1)

Charge conjugation turns the quark at the endpoint into an antiquark, and viceversa.
When the quarks at the endpoints have different flavors, we obtain a meson with opposite
charge. Therefore, since C is a conserved quantum number in QCD, mesons with different
endpoints are not eigenstates of C and yield two identical towers of states with opposite
charges.
When the quark and the anti-quark have the same flavor, charge conjugation is equivalent
to reversing the string orientation and performing a rotation by a factor of π, such that the
background (3.2) remains invariant. Therefore, working at linear order in the fluctuations, we
find

Cϕ(τ, σ)C = ϕ(τ, −σ) + π , Cρ(τ, σ)C = ρ(τ, −σ) , Ca(τ, σ)C = a(τ, −σ) , (8.2)

which is equivalent to worldsheet parity up to the nonlinear action on the zero-mode of ϕ.

Note that for long strings worldsheet parity has a different interpretation. Namely, the
action σ → −σ corresponds to spacetime CP for long strings, and the worldsheet axion flips
sign under that, differently than in eq. (8.2).
Consider the string ground-state with angular momentum J. This has quantum numbers
(PJ , CJ ), which are arbitrary within EFT. Because of eq. (4.65), since the zero mode of ϕ
shifts by π under both P and C, we conclude that the next state in the Regge trajectory has
opposite quantum numbers (PJ+1 , CJ+1 ) = (−PJ , −CJ ). More in general we must have

(PJ+δJ , CJ+δJ ) = (−1)δJ (PJ , CJ ) . (8.3)

For the same reasons, this relation holds also for all the daughter Regge trajectories. For
heavy quarks, this relation reflects the fact that, in a NR potential model, the wave-function
of the quarks is even/odd for even/odd orbital angular momentum.
It is also straightforward to compute the quantum numbers of the daughter Regge
trajectories in terms of the leading one. For parity, this follows directly from eq. (8.1):
daughter Regge trajectories obtained by acting on the vacuum with an even (odd) number of
creation operators associated with ρ and a, have the same (opposite) parity compared to the
state in the leading Regge trajectory with the same angular momentum.

– 52 –
 ϕ ρ a
n even P C (−P ) C (−P ) C
n odd P (−C) (−P ) (−C) (−P ) (−C)

Table 1: P and C assignments for the modes of the string fluctuations ϕ and ρ, and the
worldsheet axion a. The notation compares the quantum numbers P and C of the leading
Regge trajectory with those of the daughter trajectory obtained by exciting the corresponding
mode on the string.

For mesons with identical quarks at the endpoints, eq. (8.2) implies that a single-particle
state on the string is C even (odd) if its wavefunction e−iϵτ fϵ (σ) is even (odd) under σ → −σ.
From our results eqs. (4.59) and (4.63) for ϕ, we conclude that angular modes with n even
are C-even, while those with n odd are C-odd. A similar result is found for ρ and a (see
Section 4.1 and app. D). The C assignment of a daughter Regge trajectory is that of the
leading Regge trajectory times the product of the C-charges of all the single particles.
In table 1, we summarize the quantum numbers of the single-modes created by the string
fluctuations and the worldsheet axion. Notice that the lowest energy states are expected to
be the daughter Regge trajectories associated with a single n = 2 mode of ϕ or ρ, while the
lightest axion mode has n = 0, but this is expected to be almost degenerate with the n = 1
axionic state for light quarks.
Let us conclude this section discussing the action of P and C on the different spin states
(assuming these are not infinitely heavy) for mesons made of spin one-half quarks. It is
convenient to think of a meson as a string of orbital angular momentum ℓ tensored with
the Hilbert space of the spins. Therefore for each string state (either in the leading or in
a daughter Regge trajectory), we obtain a spin-family of four states. The Hilbert space of
the two spins decomposes into an s = 1 triplet representation with symmetric wave-function
and an s = 0 singlet with an antisymmetric wavefunction. The tensor product of the triplet
states with a string state with (P, C) charges yields three states with the same charges under
parity and charge conjugation: (PJ , CJ ) = (P, C) with J = ℓ − 1, ℓ, ℓ + 1. The state obtained
considering the product of the string state and the spin singlet instead has the opposite value
of C due to the antisymmetric wavefunction: (PJ , CJ ) = (P, −C) with J = ℓ. Note again that
the charge conjugation assignment makes sense only for mesons with identical endpoints.

– 53 –
 M2
*
6 ++ **
ρ/a (J ) *
ρ/a (J-- ) *

5 ω/f (J++ ) **

ω/f (J-- )
K * (J+)
4 K * (J-)
ϕ/f'(J++ )
3 ϕ/f'(J-- )

0 1 2 3 4 5 6
J

-1

Figure 3: Leading Regge trajectories: the plot shows the mass of the lightest particles for each
value of J ≥ 1 for all possible light quark compositions. The ω/f and ϕ/f ′ trajectory have I = 0
and (to a very good approximation) correspond, respectively, to Ū U and s̄s states. The K trajectory
has I = 1/2, corresponding to s̄U , and the I = 1 trajectory is made of ρ/a particles, with quark
wave-function Ū⃗σ U . The legend details the discrete quantum numbers of the particles. Data points
with one star are listed in the summary table of PDG but they are not yet firmly established, while
data with two stars are considered poorly established and are listed in the further states section of
PDG. The continuous lines are the result of the best fit using the model (8.4).

8.2 Light mesons

In this section, we discuss mesons made of light quarks (u, d and s). These are conveniently
grouped according to their isospin I = 0, 1/2, 1. We recall the reader that the u and d quarks
are approximately degenerate and, in the approximation in which they are approximately
massless, form a doublet under isospin U = √12 (u, d). In the following, all the data are obtained
from the current PDG listings [58]; we provide summary tables in appendix F.
Let us begin by discussing the leading Regge trajectories, that we plot in fig. 3. Note
that we plot two different trajectories for I = 0, where the ω/f particles are the ground
states for Ū U mesons, and the ϕ/f ′ particles correspond instead to s̄s mesons.27 Parity and
charge conjugation quantum numbers, that we indicate with the standard notation J P C (C
is measured only for I = 0, 1), follow the predicted pattern (8.3). We fitted each trajectory
(boldly using all the data, including unconfirmed ones, down to J = 1) with the formula

2 1 2µ1 1/4
Mleading = 2
(J − 1) + J , (8.4)
ℓ̄s ℓ̄s
27
In principle Ū U and s̄s mesons may mix with each other, but the mixing angle is numerically small.

– 54 –
 ℓ̄−2
s (GeV )
2
2µ1 /ℓ̄s (GeV2 )
ω/f 1.09 0.52
ρ/a 1.06 0.59
K 1.10 0.76
ϕ/f ′ 1.02 1.06

Table 2: Results of the fit of the leading Regge trajectories using eq. (8.4).

that is obtained truncating eq. (4.50) to the first nontrivial correction and including the
quantum predicted Regge intercept (4.82). The fits are shown in fig. 3 and the results for the
coefficients are reported in table 2.
We remark that the data in fig. 3 include some particles that are not yet firmly established,
in particular all the ones with J > 4. Also because of this, we did not attempt at a systematic
analysis of the errors in the fit, but we expect the results in table 2 to be subject to significant
uncertainties. Nonetheless, we find the consistency of the results obtained from different
families encouraging, as we discuss more in detail below.
The Regge slope is approximately the same ℓ̄−2 2
s ≈ 1.06 GeV for all families. This is
remarkable, especially since for the ϕ/f ′ trajectory we only have three data points. The value
of µ1 is very similar for the ω/f and the ρ/a trajectories, whose curves almost overlap in
plot 3. These mesons indeed differ only in the quarks’ flavor content and one expects them
to be degenerate at large Nc . The K and ϕ/f ′ trajectories have increasingly larger values of
µ1 , as expected since the s quark is heavier than u and d.28 Note also that the prediction for
different endpoints (4.92) implies (in obvious notation)

µ1 |ρ/a,ω/f + µ1 |ϕ/f ′
µ1 |K = , (8.5)
2
which is in good agreement with the results of the fit in table 2.
Let us now consider states beyond the leading Regge trajectory. Due to the low values of
J, typically J = 0, 1, 2, we cannot expect a quantitative agreement with the EST predictions.
Nonetheless, we find it instructive to analyze the data from a stringy perspective, and discuss
their potential interpretations as EST states based on their quantum numbers.
In figs. 4 we plot the mass square as a function of J for most observed established mesons29 ,
as well as some unestablished ones, that are made of light quarks (u, d and s) and whose
quantum numbers are known. The plot stops at J = 3, since for higher values of J the
28
If we use µ1 to naively determine the quark mass from eq. (4.36), we obtain mu ≃ md ≈ 240 MeV from
the ρ/a and the ω/f trajectories, and ms ≈ 370 MeV from the ϕ/f ′ trajectory. These values are somewhat
smaller than the constituent quark masses quoted in PDG: mu ≃ md ≈ 340 MeV and ms ≈ 486 MeV. This
is not suprising since, as we have explained, the coefficient µ1 provides a reliable determination of the quark
masses only for heavy quarks.
29
We do not show the π(1800), f0 (1710), f0 (2020), ϕ(2170), f2 (2300), and f2 (2340), which appear around
the third daughter Regge trajectory and are thus much heavier than the others particles shown in figs. 4.

– 55 –
 M2
6 η/h (J-+)
ω/f or ϕ/f' (J++ )
ω/f (J-- ) **
5 **
η/h (J-+)
η/h (J+-)
4 ω/f or ϕ/f' (J-- ) **
**
** **
**
**
ω/f (J++ ) *
**

*
*
3 **

*
*
*
*
2 *

0 1 2 3
J

-1

(a) I=0

M2 M2
6 6 π /b (J-+)
- ρ/a (J++ )
K (J )
ρ/a (J-- )
*

5 K * (J+) 5
**
* **
π /b (J-+)
K (J+)
π /b (J+-)
K * (J-)
*

4 4 ρ/a (J-- ) **
** *
**
*
**
**
ρ/a (J++ ) ** **
**

3 3
* *

2 2

1 1

0 1 2 3
J 0 1 2 3
J

-1 -1

(b) I=1/2 (c) I=1

Figure 4: Mesons’ spectrum for light quarks: The three plots show the masses and the quantum
numbers for mesons with isospon I = 0, I = 1/2 and I = 1. As in fig. 3, the legend details the discrete
quantum numbers of the particles, data points with one star are listed in the summary table of PDG
but they are not yet firmly established, while data with two stars are considered poorly established
and are listed in the further states section of PDG. The solid lines are the fits for the leading Regge
trajectories, and the dashed lines represent the curves Mn2 (J) = Mleading
2
(J) + nℓ̄−2
s with n = 1, 2.

only confirmed particles are those on the leading Regge trajectories; we also excluded some
unconfirmed very heavy states.
A naive extrapolation of the EST predictions would suggest that the lightest daughter
Regge trajectories should be found around 2ℓ̄−2s above the leading Regge trajectory in the
absence of an axion. This prediction spectacularly fails in all plots: in the I = 1/2, 1 plots, we

– 56 –
 M2
5
π /b (J-+)
ρ/a (J++ )
4 ρ/a (J-- )
π /b (J+-)
ρ/a (J-- )
3 ρ/a (J++ )

0 1 2 3
J

-1

Figure 5: Reliable mesons’ states: the plot shows a subset of I = 1 states that appear also in the
other isospin channels (twice at I = 0) and that therefore we expect to be continuously connected to
the true mesons’ spectrum at Nc → ∞.

see that the distance between the leading Regge trajectory and the other particles with the
same spin is roughly ℓ̄−2
s , while for I = 0 it is sometimes even smaller. We also observe many
unconfirmed states as well as sometimes partially overlapping resonances (we did not report
the widths for simplicity); finally, we might be missing states because we do not consider
particles listed in PDG whose quantum numbers are unknown. It appears therefore very hard
to provide a rationale for all the observed data points given these features and uncertainties.
To make progress, we focus on a subset of data which is more promising from our
perspective. In experiments, it is hard to distinguish mesons with light quarks from other
kinds of particles, such as molecular bound states of mesons, tetraquarks, glueballs, and
mixtures thereof. To identify states which are continuously connected to meson particles in
the limit Nc → ∞ we therefore compare different isospin channels. Since at Nc = ∞ the
meson spectrum is expected to take qualitatively the same form for all possible endpoints,
we consider (confirmed) states that appear once, with the same quantum numbers (or just
the same P and J when comparing to I = 1/2) and roughly similar mass, in the I = 1/2
and I = 1 plots, and appear twice in the I = 0 listing (twice to account for both Ū U and s̄s
mesons).
Proceeding in this way, we conservatively identify a set of 10 states that we believe should
admit an interpretation as proper meson particles. We show them in fig. 5, focusing for
concreteness on I = 1. Note that for J ≥ 3 only the particles on the leading Regge trajectories
showed before can be considered reliable meson candidates according to this rationale. The
J = 0 and J = 1 candidates selected with this rationale agree with the states found in the

– 57 –
large Nc lattice study in [59], that measured masses and quantum numbers of the lightest
mesons with small angular momentum.
Among the reliable mesons we still observe several particles at ∼ ℓ̄−2 s distance from
the leading Regge trajectory. (However, all the I = 0 particles with mass smaller than
2
Mleading + ℓ̄−2 ′
s , besides the ω/f and ϕ/f leading Regge trajectory, have no analogues at
I = 1/2 and I = 1, suggesting that these should be more properly interpreted as molecular
bound states of two lighter mesons). To interpret these unexpectedly light states, the most
natural scenario is to suppose that, for some reason, we observe states with different spin
wavefunctions in the low J part of the spectrum, as if the meson behaved similarly to the
NR heavy quarks’ regime. This is because the orbital angular momentum on the leading
Regge trajectory is ℓ = J − 1, and therefore at J = 1 the string would classically be static.
The two lightest states are the ρ at J = 1 and the pion at J = 0, and, as well known [60],
these can be interpreted, respectively, as having a symmetric s = 1 and an antisymmetric
s = 0 spin wavefunction for the quarks, which are connected by a static flux tube with ℓ = 0
orbital angular momentum. The pion is much lighter than the ρ due to (approximate) chiral
symmetry breaking. Similarly, the grey and orange states at J = 1, and the empty blue point
at J = 0, have the right quantum numbers to arrange in a spin quartet together with the J = 2
a2 particle in the leading Regge trajectory. Note that these four particles are approximately
mass degenerate, in agreement with the assumption that the spins’ energy splittings are small.
At this point we might expect to see also the states corresponding to different spin
wave-functions of the J = 3 particle on the leading Regge trajectory. However, at J = 2 there
is only a candidate for the antisymmetric state, which has 2−+ quantum numbers, but no
2−− candidate for the symmetric state. The disappearence of this spin state at J = 2 seems
a robust conclusion and is a well known puzzle in mesons’ phenomenology (see e.g. [61] and
refs. therein). Indeed in the complete plots 4 we see that at I = 1 the only particle with the
right quantum numbers to be interpreted as the missing spin state is poorly established and is
quite heavier than the other potential spin state; similarly, at I = 0 we have only one (poorly
established) particle, rather than two as required, with 2−− quantum numbers. Additionally,
even if we insist in interpreting the heaviest 2−+ and 1−− particles as spin partners of the
3−− ρ3 , we do not have an understanding of the heaviest 0+− state in plot 5.
EST suggests also a potentially different interpretation of the data. We may speculate
that between J = 2 and J = 3 a transition occurs, from a regime in which the meson behaves
as if the quarks were heavy to one that follows more closely the expectations from the light
quarks’ EFT that we described in the first part of this work. According to this scenario,
for J ≥ 3 all the nontrivial spin states, i.e., those with J < ℓ + 1, acquire a large (infinite?)
gap above the leading Regge trajectory. Note that indeed the energy of the states in the
spin family of the leading Regge trajectory increases with J (at J = 0 the π is much lighter
than the ρ obviously, while at J = 1 the four states in the a2 spin family are approximately
degenerate). The disappearence of spin states with J ≥ 2 would also be compatible with
the spectrum being analytic for J ≥ 2, as expected from general arguments. According to
this scenario, the heaviest 2−+ and 1−− particles in fig. 5 are naturally interpreted as some

– 58 –
 M2
5 M2
π /b (J-+ ) Missing spin state?
5
ρ/a (J++ )
π /b (J-+ )
4 ρ/a (J-- ) ρ/a (J++ )
Axion?
π /b (J+- ) ρ/a (J-- )
4
ρ/a (J-- )
ρ/a (J++ )
| ↑ ↓ ⟩S ? π /b (J+- )
3 ρ/a (J-- )
ρ/a (J++ )
| ↑ ↓ ⟩A | ↑ ↑ ⟩S 3
| ↑ ↓ ⟩S
2
? 2 | ϕ ↑ ↓ ⟩A
| ϕ ↑ ↑ ⟩S | a ↑ ↑ ⟩S | ↑ ↑ ⟩S

| ↑ ↓ ⟩S | ↑ ↓ ⟩A | ↑ ↑ ⟩S
| ↑ ↓ ⟩S | ↑ ↓ ⟩A | ↑ ↑ ⟩S
1
1
| ↓ ↓ ⟩S | ↓ ↓ ⟩S
| ↑ ↑ ⟩S
J | ↑ ↑ ⟩S
0 | ↑ ↓ ⟩A 1 2 3 J
0
| ↑ ↓ ⟩A 1 2 3

-1 -1

Figure 6: Potential interpretations of the states: Graphic representations of the two scenarios
described in the main text. The colored kets label the states, the subscript denoting whether the spin
wave-function is symmetric or antisymmetric. Kets with the same color correspond to the same leading
Regge trajectory or excitations thereof.

anomalously light string excitations, with the heaviest 0+− state obtained from the 1−− one
modifying the spin wave-function. In particular, the 1−− particle could be a n = 2 ϕ single
particle state, which is indeed the expected lightest excitation in the heavy quark regime.
Appealingly, the 2−+ particle one has instead the right quantum numbers to be interpreted as
a worldsheet axion n = 0 mode or a ρ n = 2 state. In particular, the extrapolation of our
formulas suggests that the gap δM 2 ≈ 2/ℓ̄2s of the n = 2 ρ state is larger than the one of the
first axion daughter trajectory at low values of J. Indeed, given the lattice measured value
of the axion mass (2.11), we can estimate the mass of the associated daughter trajectories
√
using the formula δMn2 ≃ 2 ℓωϵn /ℓ̄s and eqs. (D.5) and (D.6) of the appendix. A naive
extrapolation to ℓ = J − 1 = 1 gives δM02 ≈ 1/ℓ̄2s and δM12 ≈ 1.3/ℓ̄2s for the first two axion
energy levels assuming Neumann boundary conditions, and δM02 ≈ 1.6/ℓ̄2s and δM12 ≈ 2.25/ℓ̄2s
assuming Dirichlet boundary conditions. The prediction δM02 ≈ 1/ℓ̄2s with Neumann boundary
conditions is in remarkable agreement with the experimental measure δM 2 ≃ 0.99/ℓ̄2s . Notice,
however, that if we were to take the numerical predictions with Neumann boundary conditions
seriously, we would also expect a slightly heavier 2−− partner axion particle, which is nowhere
to be seen as commented above. We therefore do not take this numerical agreement at low
values of J as an indication that Neumann boundary conditions are preferred for the axion
state.
The two potential interpretations of the data that we described are schematically repre-
sented in fig. 6. Clearly, at this stage none of the described scenarios is particularly compelling,
and it is well possible that the right interpretation of the data requires more refined ideas.
Nonetheless, our analysis provides motivation for experiments to look for more mesons, in

– 59 –
all isospin channels, at J = 3. In particular, if the 2−+ state in fig. 5 indeed corresponds
to a wordlsheet axion mode, it would be natural to expect at 3+− meson, again due to a
axion excitation, somewhere in between ℓ̄−2 −2
s and 2ℓ̄s above the leading Regge trajectory.
Alternatively, if the states with different spin wavefunctions do not completely disappear and
only slightly increase their gap from J = 1 to J = 2 (after all, we are at low values of J),
we should expect two approximately degenerate particles with quantum numbers 3+− and
3++ - and the absence of the 2−− particle would then need to be explained by some other
mechanism.
Before concluding, we would like to make a remark about the so-called hybrid mesons.
These are the two π/b particles and the candidate η/h mesons with 1−+ quantum numbers,
denoted by plain black markers in plot 4. These are called hybrid because their quantum
numbers cannot be reproduced in a potential quark model, and are therefore usually interpreted
as mesons in which the glue binding the quarks is excited [60]. From our viewpoint, these
should therefore be string excitations of the leading Regge trajectory. Consider for instance
the lightest 1−+ π/b in the I = 1 plot in fig. 4. In the interpretation in which the lightest 2−+
π/b is an axion excitation, then the lightest hybrid 1−+ π/b and one of the nearby 1−− ρ/a
could be interpreted as different spin wavefunctions of the J = 2 worldsheet axion. However,
as noted also in [62], for the interpretation of hybrid mesons as string excitations to be reliable
one would expect iso-partner states at I = 0 and I = 1/2 of the two hybrid π/b’s; instead, the
only potential candidate at I = 0 is much heavier than the two I = 1 hybrids, while there is at
most one particle candidate for an isopartner at I = 1/2 (the heaviest K 1+ state). Therefore,
at this stage, pending further experimental data, the interpretation of the currently observed
hybrid mesons remains puzzling.
Let us conclude by remarking that the discussion in this section is only a first attempt at
mesons’ phenomenology from a stringy viewpoint. More data and more ideas are undoubtedly
needed to improve our understanding of QCD from the point of view of the confining string.

8.3 Comments on other mesons

We close this section with some brief comments on other mesons’ families.
We did not attempt a quantitative analysis of the heavy charmonium and bottomium
states due to the reduced validity window of EST (and the scarcity of data for J ≳ 1). All
confirmed states in heavy quarkonium are believed to admit an interpretation in terms of the
quantization of a phenomenological potential model of quarks moving in a central potential
(qualitatively, and suprisingly often also quantitatively; see e.g.[60, 63]). Since, as explained in
Section 7.2, for heavy quarks EST reduces to a NR potential model, and the quantum numbers
of the bound states do not depend on the details of the potential (as long as it is spherically
symmetric), we do not have anything to add to the existing literature on the subject. In other
words, all observed heavy mesons admit a reasonable qualitative interpretation in terms of
states of the QM potential model (7.17), including the different spin wavefunctions. The states
corresponding to string excitations are expected to be quite heavy (with a ∼ π/ℓ̄s gap on top

– 60 –
of the leading Regge trajectory for the low values of J that we can access in experiments) and
presumably have not yet been observed.30
It might be easier to identify candidates for string excitations and axion states in heavy-
light mesons. The application of EST to these particles is a straightforward extension of our
work. Unfortunately, very few confirmed particles are listed in PDG, for instance, for both
charmed and bottom particles we have data only for J < 2 after excluding the leading Regge
trajectory and unconfirmed states.
In the future, additional data for heavy-heavy and heavy-light mesons might provide a
testing
p ground for our predictions. In particular, it would be interesting to confront data at
J ≳ mℓ̄s for heavy quarkonium with the prediction of spin-orbit inversion and the result for
the leading Regge trajectory in Section 7.

9 Beyond EFT: inequivalent quantizations and glueballs in 3d

Let us now briefly consider what happens when we try to extend our results beyond the
vicinity of the leading order Regge trajectory, where we expect the large J expansion to be
valid. The discussion will not focus on whether the expansion is under control but rather
on the qualitative features of the leading order spectrum, such as energy degeneracies, when
extended beyond this regime. As we will review, our motivation for extrapolating the EFT
predictions is the study of the glueball spectrum in d = 3, 4 Yang-Mills theory. Indeed, lattice
simulations show that the string transverse coordinate qualitatively describes all the observed
flux tube states, up to very high energies, also for short strings in 3d. Similarly, the string
coordinates and the worldsheet axion describe all observed flux tube states in 4d. This suggests
that the string coordinates can be thought as the only fundamental worldsheet fields in 3d, and
similarly in 4d including the axion. In the context of the rotating string, these observations
suggest that it should be possible to, at least qualitatively, describe all the glueballs measured
on the lattice in 3d and 4d Yang-Mills theory in terms of states of the rotating string Hilbert
space that we described in this work.
Perhaps surprisingly, it turns out that, even after specifying the field content of the theory,
the extrapolation of the Hilbert space of the theory to low values of J is not unambiguous
beyond the leading Regge trajectory. We will see that the natural semiclassical quantization
procedure that emerges when considering the large J effective field theory expansion differs,
even when considering critical strings, with respect to what one would obtain from lightcone
quantization. Indeed, this fact was already observed in [19], where a recipe for quantization of
closed effective strings at large J—the axionic string Ansatz (ASA)—was used to describe
qualitative features of the lattice glueball spectrum all the way to J −→ 0. In this section, we
30
On the other hand, assuming Neumann boundary conditions, as suggested by simulations of open static
flux tubes [32, 33], the lightest daughter trajectory that is not captured by the quark model might be the one
associated with a single worldsheet axion state on the string, whose gap is simply given by its mass (2.11) and
is thus not enhanced by a factor of π, for sufficiently low values of J.

– 61 –
identify the physical origin of this discrepancy in the simpler context of open strings—we plan
to investigate this issue further in the context of closed strings in a future publication.
We will start with a brief recap of how the closed string spectrum of [19, 20] is obtained,
where we will consider only d = 3 for now. There, one starts with the leading order effective
string theory description at large J for open strings, which is in essence the leading order
part of the theory described in this work, although there it was obtained by working with the
Polyakov action and solving the Virasoro constraints. From the semiclassical quantization
of fluctuations around the classical rotating rod solution at a given J, one may construct a
Hilbert space of states which we will label Hopen,effective and within this Hilbert space, one can
find the energy degeneracies—the levels—at leading order in J. The resulting Hilbert space is
spanned by vectors of the form
Y  mn
|Ψ⟩open,effective = a†n |J⟩ (9.1)
n≥2

where an ’s have the standard commutation relations and are the only oscillators that survive
after imposing the Virasoro constraints on top of the fluctuations on the rotating rod back-
ground. The an can be schematically thought as the annihilation operators for the fluctuations
of the angular coordinate ϕ in Polyakov formalism. The an are built in such a way that they
don’t change the value of J. At leading order in J, all states with the same level N
X
N =J+ na†n an (9.2)
n≥2

are mass degenerate: M 2 = N/ℓ̄2s .31

As explained in [19], within the ASA the closed string spectrum is postulated to be, at
each value of the level N , the tensor product of left-moving and right-moving Hilbert spaces
X
Hclosed,ASA = HL (N ) ⊗ HR (N ) (9.3)
N

where in analogy with what happens in the critical case, the left- and right-moving sectors
are postulated to be identical to the open string Hilbert space, which is taken to be the one
obtained from the large J effective string description just described

HL = HR = Hopen,effective . (9.4)

The closed string quantization obtained from this prescription was found in [19] to qualitatively
match the features of the d = 3 glueball spectrum. Furthermore, a number of predictions
made in [19] were later confirmed by [12].
Interestingly, as discussed in [20], one may also use the closed string large J semiclassical
quantization directly to construct the “effective closed string” spectrum. The result is that at
31
In the present discussion, we ignore all ordering issues, as they are irrelevant for the key point we are trying
to make.

– 62 –
d = 3 the Hilbert space Hclosed,ef f ective is spanned by vectors of the form
Y  mn Y  † mn′
|Ψ⟩closed,ef f ective = a†n an′ |J⟩ (9.5)
n≥2 n′ ≥2

where again the an , an′ are oscillators that do not alter J. The level is given by
X X
N =J+ na†n an + n′ a†n an′ (9.6)
n≥2 n′ ≥2

As explained in [20], there is crucially the extra subtlety that, due to the imposition
of reparametrization invariance, the states in Hclosed,ef f ective are those of the form of (9.5)
subject to the extra condition that N must be even. This, in particular, restricts us to states
of even J at the leading Regge trajectory.
Interestingly, the two prescriptions to build the closed strings quantizations, Hclosed,ASA
and Hclosed,ef f ective do not match beyond the leading Regge trajectory when extrapolated to
low values of J. This is seen from the fact that they give different degeneracies: at a given
value of J and N the number of states in Hclosed,ef f ective is larger than the number of states
in Hclosed,ASA . For instance we see that at J = 1 and N = 2, Hclosed,ASA is empty, while
Hclosed,ef f ective yields 2 states [20]. As discussed in [20], the two prescriptions match close to
the leading Regge trajectory for large J. It is still surprising however that the “closed effective
string” prediction so heavily over-predicts the number of states at higher levels.
Furthermore, both prescriptions also differ from what is obtained from the lightcone
quantization. To see this, it is simpler to work in an arbitrary number of dimensions d > 3.
We also consider for simplicity only states with zero spatial momentum. Then the Hilbert
space is spanned by vectors of the form
Y  mI,n
|Ψ⟩open,lightcone = aI†
n |0⟩ (9.7)
I,n>0

for open strings and

Y  mI,n Y  ′
mI ′ ,n′
|Ψ⟩closed,lightcone = aI†
n aIn′† |0⟩ (9.8)
I,n>0 I ′ ,n′ >0

for closed strings. Here the label I = 1 . . . d − 2 denotes the fact that the αI oscillators generate
fluctuations in the d − 2 directions transverse to the lightcone plane. The corresponding masses
(squared) are obtained from the zero mode of the Virasoro constraints and yield (again we
ignore constant terms that arise from ordering)
X
2
Mopen ∼N = nanI† aIn (9.9)
n,I

and X X
2
Mclosed ∼ naI† I
n an + na′I†
n an
′I
(9.10)
n,I n,I

– 63 –
where the ∼ symbol denotes up to some dimensionful constant factor and the closed spectrum
is subject to the level-matching condition that imposes the total worldsheet spatial momentum
to be zero: X X
naI† I
n an = na′I†
n an
′I
(9.11)
n,I n,I

As usual in lightcone quantization, the states obtained with this procedure come out in O(d−2)
representations, and should then be arranged into irreps. of the massive little group O(d − 1).
As is well known, such a rearrangement fails for N = J = 1 in open strings, and for N = 1
and J = 2 in closed strings (away from the critical dimension d = 26 where it is possible to
bypass this requirement by making the corresponding states massless), but it can be done for
all other states.32
It can be seen that the d = 3 spectrum obtained in this way differs from both the “ASA”
and “effective string” spectra described above. For instance, besides the aforementioned issue
at N = 1, at N = 2 lightcone quantization predicts two scalars and a J = 4 particle at
N = 4; “ASA” instead predicts three scalars, two J = 2 particles, and one J = 4; finally, the
extrapolation of closed effective string theory yields the same particles as “ASA”, plus two
J = 1 particles and two additional scalars. In general, the full situation is summarized by the
following schematic inequalities
! !2 !2 !
closed open open closed
= < < . (9.12)
lightcone lightcone effective effective

The first equality denotes the fact that the tensor product procedure outlined above, which
accounts for level matching (this is what is meant from “squaring” in this picture), applied to
the open string lightcone spectrum yields the closed string lightcone spectrum (indeed, this
fact was the inspiration for performing this procedure within the ASA). The first inequality
describes the fact that open string lightcone quantization yields a smaller number of states at
fixed N and J than ASA, which is just the open effective string theory spectrum squared. The
second inequality further indicates that the closed effective string theory spectrum predicts an
even larger number of particles.
Let us attempt to identify the physical origin of how these discrepancies come about. In
order to do this let us consider the situation in d = 4, where the analysis is simpler, although
the key subtlety we will point out is also present in the d = 3 case we have described. The
usefulness of considering the d = 4 case is that here one can start by gauge fixing the lightcone
gauge, and then decide whether to perform lightcone quantization or to expand around a fixed
angular momentum background. This is because in d = 4, it is possible to take a background
where the string rotates in the plane transverse to the lightcone coordinates, which is not
possible in d = 3.
32
In d = 3, we are interested in the spectrum that is obtained by analytic continuation of the results of
the lightcone quantization procedure outlined in the main text for d > 3 —(see also [19]). This spectrum is
different than the one that is obtained applying lightcone quantization directly in d = 3, whose Hilbert space
consists of particles with fractional angular momentum and anyonic statistics [64].

– 64 –
 To illustrate he point, we will consider only open strings. If 2, 3 denote the two transverse
coordinates to the lightcone plane and we define (the ± symbol here should not be confused
with the ± associated to lightcone coordinates)

√
a± 2 3
n = an ± ian / 2 , (9.13)

then in lightcone gauge the zero mode of the Virasoro constraints imply that the number
operator defined above is given by
X X
N= na+† +
n an + na−† −
n an . (9.14)
n n

The angular momentum operator for rotation in the (2, 3) plane is given by
X X
J= a+† +
n an − a−† −
n an . (9.15)
n n

We will now see how different inequivalent quantizations arise when the theory is quantized
around the a± = 0 background versus when one quantizes around a rotating rod background
with non zero J. This subtlety is already captured when one considers just the two a+
oscillators with lower frequency, so in order to show our point we may restrict ourselves to the
simpler quantum system with two oscillators and two commuting charges, the “Hamiltonian”

H = ω1 a†1 a1 + ω2 a†2 a2 , (9.16)

and the “angular momentum”

J = a†1 a1 + a†2 a2 , (9.17)
where the a1 and a2 oscillators should be thought of as the two a+ oscillators with lowest
frequency and the ωi represent the frequencies in N of these oscillators (their values of n in
formula for N above). The standard quantization – that is, around the trivial a1 = a2 = 0
background – obviously yields states of the form

(a† )n (a† )m
|n, m⟩ = √1 √2 |0⟩ . (9.18)
n! m!
Then if we take the case ω2 > ω1 , at fixed J the states with smallest energy are of the form
|J, 0⟩. This would be the analogue of our leading Regge trajectory. At fixed J, there is a finite
tower of states on top, of the form |J − 1, 1⟩ , |J − 2, 2⟩ . . . |0, J⟩.
Let us now consider, in this simple system, the analogue of our effective field theory
quantization around the non-zero J background. In order to do this, we will define a new set
of canonical variables {δ, J, α, α† } by the canonical transformation
 
1/2
a1 = e−iω1 (t+δ) J0 + β(J) , (9.19)
a2 = e−iω1 (t+δ) α , (9.20)

– 65 –
where J0 is the background value of the angular momentum and α and β describe the
fluctuations around this background. In order to work at fixed angular momentum, the β
fluctuation is written in terms of our new canonical variable J by plugging these expressions
into (9.17) and solving for β. In these terms, the “Hamiltonian” becomes

H = ω1 J + (ω2 − ω1 )α† α (9.21)

while the symplectic form yields

ω = ida†1 ∧ da1 + ida†2 ∧ da2

= dδ ∧ dJ + idα† ∧ dα . (9.22)

When we quantize this canonical system we realize that the eigenvalues of the operator
J are integer because this is the conjugate variable to δ, which is 2π periodic. Furthermore,
since δ does not appear in the Hamiltonian, J is conserved, and so the Hilbert space can be
split into the direct sum of sectors with constant integer J. The key difference with the trivial
quantization in terms of a1 and a2 outlined above is that now for each value of J we have
an infinite tower of α oscillators, so that at fixed J there are in this quantization many more
states than in the standard one. Namely, we can consider states of the form

(α† )n
|J0 , n⟩ = √ |J0 , 0⟩ , (9.23)
n!
for every integer n ≥ 0, where the leading Regge trajectory state is defined by

J|J0 , 0⟩ = J0 |J0 , 0⟩ , α|J0 , 0⟩ = 0 . (9.24)

The existence of this infinite tower of states is precisely the reason why the “effective string”
quantizations outlined above have many more states at fixed J and N than the lightcone
quantization.
How can it be, however, that such a simple system—two harmonic oscillators—has two
inequivalent quantizations? The reason is that the phase spaces spanned by the canonical
variables {a1 , a†1 , a2 , a†2 , } and {δ, J, α, α† } are globally inequivalent, given that the first is merely
R4 , whereas the second one is actually R3 × S 1 . In other words, the canonical transformation
that relates {a1 , a†1 , a2 , a†2 , } and {δ, J, α, α† } is singular at the origin. Given the full theory, it
is always possible to identify the correct set of global coordinates for the phase space, but
if one starts from the EFT spectrum, that only describes the Hilbert space locally near the
leading Regge trajectory, one is faced with a choice when extrapolating to low values of J.
Surprisingly, our discussion suggests that the right choice for QCD strings results in a different
phase space structure compared to the natural choice in the context of fundamental strings.

– 66 –
Acknowledgements

We thank O. Aharony, V. Gorbenko, S. Hellerman, Z. Komargodski, and R. Rattazzi for useful

discussions. During this work GC, SD and SZ were supported by the Simons Foundation
grant 994296 (Simons Collaboration on Confinement and QCD Strings) and by the BSF grant
2018068. SD is also supported in part by the NSF grant PHY-2210349, and by the IBM
Einstein Fellow Fund at the IAS. During this work GHC was supported by CSIC I+D grant
number 583 and ANII grant number FCE 1 2023 1 175902. The work of AM is supported in
part by the National Science Foundation under Award No. 2310243.

Appendix
A Geometric identities and notation

A.1 Bulk geometry

We denote the string coordinates as X µ and assume the ambient space to be flat: ηµν =
ηµν = diag(1, −1, −1, . . .). The worldsheet coordinates are denoted by σ α and the induced
metric then takes the form
Gαβ = ∂α X µ ∂β X ν ηµν . (A.1)
The associated connection is denoted Γαβγ and ∇α is the associated covariant derivative. We
also define G = − det(Gαβ ).
We define the second fundamental form and the worldsheet curvature via
µ
Kαβ = ∇α ∂β X µ , Rαβγδ = ∂γ Γαβδ − ∂δ Γαβγ + Γαγη Γηβδ − Γαδη Γηγβ . (A.2)

The second fundamental form satisfies:

µ
Kαβ ηµν ∂γ X ν = 0 . (A.3)

The Riemann tensor and the second fundamental form are related by the Gauss-Codazzi
formula:
µ ν µ ν


Rαβγδ = Kαγ Kβδ − Kαδ Kβγ ηµν . (A.4)
Sometimes it is useful to choose a basis of d − 2 (spacelike) normal vectors {nµA }.33
Normalizing them such that nµA nνB ηµν = −δAB , we can decompose the bulk metric as

η µν = ∂α X µ ∂β X ν Gαβ − nµA nνB δ AB . (A.5)

33
√
In d = 3 there is a unique normal vector and it can be written as nµ = 12 εµνρ ∂α X µ ∂β X ν εαβ / G, where
we normalized it such that nµ nµ = −1.

– 67 –
There are clearly many equivalent choices for the normals. Indeed we can change the basis of the
normal bundle with an arbitrary local orthogonal transformation of the normals: nµA → RAB nµB
where R ∈ SO(d − 2). We may thus define a connection on the normal bundle [7]:

AAB ,α = nµA ∂α nνB ηµν = −ABA ,α . (A.6)

Under a local change of basis, parametrized by R ∈ SO(d − 2), AAB ,α transforms as (omitting
the SO(d − 2) indices):
Aα → RAα RT − R∂α RT . (A.7)
The connection can be used to define covariant derivatives of the normal vectors, that are
parallel to the string and are related to the second fundamental form as

Dα nµA ≡ ∂α nµA − AAB, α δ BC nµC = −∂β X µ Gβγ Kαγ

ν
nρA ηνρ . (A.8)

Finally, we notice that in d = 4 the connection on the normal bundle is Abelian:

1
Aα = εAB nµA ∂α nνB ηµν (d = 4) . (A.9)
2
The topological invariant introduced in eq. (2.10) is nothing but the field strength of this
Abelian connection
√
GK · K̃ = εαβ Fαβ where Fαβ = ∂α Aβ − ∂β Aα . (A.10)

A.2 Boundary geometry

We denote the endpoint coordinate with τ and σ α (τ ) denotes the boundary wordline on
the worldsheet. The induced metric is

Ĝτ τ = ∂τ σ α ∂τ σ β Gαβ = ∂τ X µ ∂τ X ν ηµν . (A.11)

Let us first discuss the case of a time-like boundary Ĝτ τ > 0. In this case we call
= 12 Ĝ−1
Γ̂ττ τ ˆ
τ τ ∂τ Ĝτ τ the worldline connection and ∇τ is the associated covariant derivative. We
can decompose the inverse worldsheet metric at the boundary as

Gαβ = ∂τ σ α ∂τ σ β Ĝ−1 α β
τ τ − n̂ n̂ , (A.12)

where ∂τ σ α is the vector parallel to the boundary, and n̂α denotes instead the normal. The
normal can be written explicitly as
√
G
n̂α = p εαβ ∂τ σ β =⇒ n̂α n̂α = −1 . (A.13)
Ĝτ τ

From (A.13) it follows that

(n̂α ∂α X µ )2 = n̂α ∂α X µ n̂β ∂β X ν ηµν = −1 . (A.14)

– 68 –
The second fundamental form associated with the embedding of the endpoint in the ambient
ˆ τ ∂τ X µ . We can also consider the second fundamental
space coincides with the acceleration ∇
form associated with the embedding of the endpoint in the worldsheet,

ˆ τ ∂τ σ α .
k̂τατ = ∇ (A.15)

It turns out however that k̂τατ is not an independent object, since it satisfies:
 
k̂τατ = −n̂α ∇ˆ τ ∂τ X µ ηµν n̂β ∂β X ν . (A.16)

Let us now discuss a null boundary: Ĝτ τ = 0. For a null worldline, we may define a
reparametrization invariant metric as [65]:
 1/2
γ̂τ τ = ℓ̄s −Ẍ µ ηµν Ẍ ν , (A.17)

that allows to define the following natural connection

1
Γ̃ττ τ = γ̂τ−1 ∂τ γ̂τ τ . (A.18)
2 τ
The main difference compared to the timelike case is that it is not possible to define a normal
which is invariant under reparametrizations of the boundary coordinate τ as in eq. (A.13).
We may however define a covariant normal as
√
Nτα = lim G Gαβ εβγ ∂τ σ γ , (A.19)
σ α →σ α (τ )

where the limit is well defined under the assumption that the boundary admits a natural
conformal structure as in eq. (4.21); note indeed that the definition (A.19) is Weyl invariant.
It is simple to check that the left-hand side of eq. (4.22) coincides with Nτα ∂α a. Note that
p the
definition (A.19) makes sense for both timelike and null boundaries, and reduces to n α Ĝτ τ
in the former case.

B Near boundary expansion with the (K 2 )2 term

As explained in Section 5.2, the solutions of the quadratic EOMs deriving from eq. (5.11)
admit the following expansion:

(0) ℓ̄4s ω 4 (1)

ρi (τ, σ) = ρi (τ, σ) + c4 ρ (τ, σ) + . . . , (B.1)
(1 − αf2 σ 2 )4 i
ℓ̄4s ω 4
ϕ(τ, σ) = ϕ(0) (τ, σ) + c4 2 2 4
ϕ(1) (τ, σ) + . . . , (B.2)
(1 − αf σ )

(0) (1)
where both the {ρi , ϕ(0) } and the {ρi , ϕ(1) } are regular for αf σ → ±1.

– 69 –
 We used the following expansions for 1 ∓ αf σ ≡ δ → 0 of the leading order solutions:

d2 δ 2 d2 d4 δ3
 2
d4 d6
    
(0) d 4

ρ
ρi = 1+δ 2 + + + 4 2 +5 4 + 6 +O δ β1 (τ )
dτ 6 dτ 2 dτ 4 90 dτ dτ dτ
1 d2 1 d2 1 d4
    
1/2 1 2 3
+δ 1+δ + +δ + +
12 3 dτ 2 160 12 dτ 2 30 dτ 4
37 d2 1 d4 1 d6
  
5
ρ
+δ 3 + + + + O δ 4
β2 (τ ) , (B.3)
896 1440 dτ 2 72 dτ 4 630 dτ 6
d2 2 1 d4 2 11 d4 1 d6
    
2 3 d 3 14 d
ϕ(0) = 1−δ 2 −δ + −δ + +
dτ 2 dτ 2 2 dτ 4 9 dτ 2 18 dτ 4 18 dτ 6
2 229 d4 13 d6 1 d8
  
4 47 d 5

ϕ
−δ + + + +O δ β1 (τ )
30 dτ 2 360 dτ 4 180 dτ 6 360 dτ 8
21 1 d2 33 d2 1 d4
    
3/2 2 237
+δ 1+δ + +δ + +
20 5 dτ 2 224 140 dτ 2 70 dτ 4
7387 d2 143 d4 1 d6
  
3 407 4

ϕ
+δ + + + +O δ β2 (τ ) . (B.4)
384 30240 dτ 2 7560 dτ 4 1890 dτ 6

For the second correction we used

24 2 d2 d4 2 512 d4 224 d6
     
(1) 3 96 d 4

ρ
ρi = 0 + 0 + δ + +δ + + +O δ β1 (τ )
15 dτ 2 dτ 4 5 dτ 2 15 dτ 4 15 dτ 6
41 36 d2 187 d2 158 d4
    
1/2 2 247
+δ −4 + δ + +δ + +
5 5 dτ 2 120 15 dτ 2 15 dτ 4
2 4 6
  
3 8419 26171 d 1295 d 754 d 4

ρ
+δ + + + +O δ β2 (τ ) , (B.5)
480 360 dτ 2 18 dτ 4 45 dτ 6
512 3 d2 d4 d6
  
(1)
ϕ = 0+0+0− δ +2 4 + 6
15 dτ 2 dτ dτ
 2 4 6 d8
 
256 4 d d d 5

ϕ
+ δ log δ 4 2 + 9 4 + 6 6 + 8 + O δ β1 (τ )
15 dτ dτ dτ dτ
37 124 d2 713 d2 242 d4
    
3/2 84 2 1223
+δ − −δ − −δ − −
5 5 5 dτ 2 40 5 dτ 2 5 dτ 4
55493 d2 10081 d4 766 d6
  
3 18399 4

ϕ
+δ − − − +O δ β2 (τ ) . (B.6)
160 120 dτ 2 30 dτ 4 15 dτ 6

Notice the appearence of a δ 4 log δ term. In general, an operator of order 2n in derivatives

leads to δ 2n+k log δ terms, with k = 0, 1, . . ., in the near boundary expansion.

– 70 –
C Bosonic action for a point particle with spin

C.1 Coadjoint orbit through spinors

Let us consider a relativistic-point particle with spin s. We are interested in writing a
worldline action which describes its spin. In the standard approach, one introduces some
suitably coupled Grassmanian variables on the worldline, whose Hilbert space describes the
different spin states [66–68], see [69] for a review. This approach relies on the introduction of
supersymmetric gauge redundancies, and the number of Grassmanian fields needed depends
upon the size of the spin representation.
A more straighftorward approach, which encompasses all spin representations, is the so-
called coadjoint orbit technique [70, 71]. Below, we review how to implement this construction
in terms of spinors following [72]. In the next subsection we explain how this approach can be
understood from the viewpoint of spontaneous symmetry breaking.
Let us first review the action for a non-relativistic spin in the 2s + 1-dimensional repre-
sentation at a point in space. We consider a bosonic su(2) spinor z = {z1 , z2 }, subject to the
constraint z̄z = 2. In this formulation, the action reads (see e.g. [73])34
Z
S = is dτ z̄ ż , z̄z = 2 . (C.1)

The action (C.1) is invariant under U (1) gauge transformations z → eiα(τ ) z. Invariance under
large gauge-transformations requires
s ∈ Z/2 . (C.2)
The target space is CP 1 ≃ S 2 , physically parametrizing the orientation of the spin:

σi
S i = sz̄ z. (C.3)
2
The canonical commutation relations [z, z̄] = 1/s imply that the spin satisfies the su(2)
algebra [S i , S j ] = iεijk S k , while the constraint implies S i S i = s(s + 1).35 Because of the U (1)
gauge-invariance and the constraint, all the interactions with other fields on the worldline are
straightforwardly written in terms of S i .
To make contact with the standard literature on the subject [73–75], it is convenient to
write z in terms of polar and azimuthal angles θs and ϕs :
!
√ e−iϕs /2 cos θ2s
z= 2 (C.4)
eiϕs /2 sin θ2s
34
The action (C.1) has been studied in several different contexts, see e.g. [74–76]. The constrained spinor z
is also equivalent to a charged particle moving on a sphere with a charge s monopole at the center in the lowest
Landau level [77–79].
35
To check this, it is important to use the correct ordering of the operators; we refer the reader to [80] for a
more detailed discussion of ordering issues in the path-integral formulation.

– 71 –
Then the action (C.1) becomes
Z Z
S = s dτ cos θs ϕ̇s = −s ϵijk n̂i dn̂j dn̂k , n̂i = S i /s , (C.5)
M

where in the second line we wrote the spin kinetic term as a Wess-Zumino integral, such that
∂M coincides with the worldline.
Let us now discuss the generalization to a relativistic point particle with worldline X µ (τ ).36
R p
The motion of the particle is described via the standard action −m dτ Ẋ 2 . To account for
the spin, we introduce a bosonic Dirac spinor ψ satisfying

ψ̄ψ = 2 , u
/ψ = ψ , (C.6)
p
where uµ = Ẋ µ / Ẋ 2 . ψ can be written explicitly in terms of a two-component spinor z like
before as (in Weyl basis)
1
ψ = √ Λu · {z, z} , (C.7)
2
where Λu is the Dirac boost matrix with velocity uµ . Then we include on the worldline an
action analogous to eq. (C.1): Z
Sspin = s dτ iψ̄ ψ̇ . (C.8)

This action is manifestly Lorentz invariant and admits a U (1) gauge-invariance similarly to
eq. (C.1), which fixes s ∈ Z/2.
It is straightforward to check that in the nonrelativistic limit uµ ≃ δ0µ , the action (C.8)
reduces to (C.1). To prove that the action (C.8) describes a relativistic spin-s particle in general,
it is convenient to quantize ψ as an unconstrained variable, and impose the constraints (C.6)
on-shell. One finds that the canonical commutation relations
1
[ψ, ψ̄] = 1 , (C.9)
s
imply that the spin part of the Lorentz generators,
i
Sµν = sψ̄[γµ , γν ]ψ , (C.10)
4
satisfies the Lorentz algebra:

[Sµν , Sρσ ] = −iηµσ Sνρ + permutations . (C.11)

36
In the relativistic case, the coadjoint orbit technicque is equivalent to a certain nonlinear relalization of
the coset ISO(3, 1)/ (U (1) × R), as we will see below. This coset spans an eight dimensional phase space,
parametrized by the tree spatial positions X i , the corresponding momenta and the two angles describing the
spin variable [70–72].

– 72 –
One also finds that the on-shell constraint ψ̄ψ implies S µ Sµ |phys⟩ = −s(s + 1)|phys⟩,37 where
S µ is the Pauli-Lubanski vector defined by
1
Sµ = εµνρσ uν Sρσ . (C.13)
2
Finally, all interactions with other worldline fields are written in terms of Sµν . This is
because of the U (1) gauge-invariance and the following relation that follows from (C.6)

ψ ψ̄ = Pu + Pu γ5 γµ Pu n̂µ , (C.14)

where n̂µ = S µ /s and we introduced the following projector:

1
Pu = / ) = Pu2 .
(1 + u (C.15)
2
Notice also that Sµν uν = 0 and Sµν S µν = −2Sµ S µ = 2s2 (at the classical level).

C.2 Spin from Goldstones

As emphasized in [55], a point-particle, or more in general any relativistic object which
appears point-like from large distances, defines a state which breaks spontaneously the Poincaré
group, similarly to a string (as described in Section 2) or a membrane. A spinless particle in
its rest frame, breaks the Poincaré group to time translations and transverse rotations SO(3).
It is well known that we can think of the particle’s spatial coordinates X i (τ ) as the Goldstone
fields associated with such symmetry breaking pattern. Accordingly, the standard worldline
action can be shown to be the most general action compatible with the nonlinear realization of
the symmetry via the spacetime generalization of the Callan-Coleman-Wess-Zumino (CCWZ)
construction [82, 83]. A particle with spin further breaks the transverse SO(3) rotations to
SO(2) in the rest frame, due to the spin orientation. Here we show that the action (C.8)
can also be derived insisting on the nonlinear realization of the symmetry via the CCWZ
construction.38
It is again instructive to derive first from this perspective the non-relativistic action (C.1).
To this aim we consider the symmetry breaking pattern SO(3) → SO(2). Assuming that the
spin points in the direction “1”, we consider the coset
π
ΩR = e−iϕs (τ )J3 e−i(θs (τ )− 2 )J2 , (C.16)
37
To prove this it is useful to write S µ as [81]:
!
u·S
⃗ ⃗rest
µ
S = ⃗ ⃗ ⃗
u · Srest , Srest + ⃗
u , (C.12)
1 + u0

and use that S ⃗rest

2
|phys⟩ = s(s + 1)|phys⟩.
38
[69] proposed a generalization of the CCWZ construction to describe particles with spin in the Grassmanian
formalism alluded before. The approach of [69] however relies on the introduction of supsersymmetric gauge
redundancies, while the construction in this section instead does not require any extra ingredients compared to
the usual CCWZ procedure for broken spacetime symmetries.

– 73 –
where we denoted with J2 and J3 the broken generators, while J1 is the unbroken one. To
construct SO(3) invariants, we are instructed to consider the Maurer-Cartan one form

Ω−1 a
R Ω̇R = iDΘs Ja + iAJ1 , (C.17)

where a = 2, 3 and

DΘ2s = −θ̇s , DΘ3s = − sin θs ϕ̇s , A = − cos θs ϕ̇s . (C.18)

The SO(3) group acts linearly via local SO(2) rotations on the vector DΘas , while A transforms
as a U (1) connection, which may be used to define a covariant derivatives acting on the DΘas
and matter fields.
We now have two choices for the kinetic term of the angular Goldstones compatibly with
the SU (2) symmetry. The simplest one is to consider a quadratic kinetic term
Z Z
I I
S= dτ DΘas DΘas = dτ n̂˙ i n̂˙ i , (C.19)
2 2

where n̂i = S i /s is a unit vector. This choice gives the standard rigid rotor and, at the
quantum level, describes an infinite tower of states arranged in all the representations of
SO(3). Eq. (C.5) corresponds instead to a Chern-Simons like term,
Z
S = −s dτ A , (C.20)

which is first order in derivatives and it is allowed since A shifts by a total derivative under
SO(3) transformations.
From the viewpoint of symmetries, the choice between eq. (C.19) or (C.20) is dictated by
the transformation properties of the angles under time reversal T (assuming it is a symmetry).
T
Indeed eq. (C.20) is allowed only if the action of T is accompanied by {θs , ϕs } → {π−θs , ϕs +π}
(up to SO(3) rotations), that physically corresponds to the spin being a pseudovector.39
Notice also that (C.20), due to the transformation properties of A, is the only allowed
term that cannot be written as a local integral in terms of the unit vector n̂i . To write A in a
convenient form, it is useful to evaluate Ω in a spin j representation. Working in a basis that
diagonalizes J1 we have, in obvious notation,
1
iA = ⟨j, m|Ω−1
R Ω̇R |j, m⟩ . (C.21)
m
The spinor representation in eq. (C.1) is the minimal choice, corresponding to j = m = 1/2,
so that the spinor (C.4) is obtained from
!
√ 3
−iϕs σ2 −i(θs − π2 ) σ2
2
1
z = 2 Ω|1/2, 1/2⟩ = e e · . (C.22)
1
39
Note that, within EFT, in the simplest option, given the action (C.20), terms with two or more derivatives,
such as (C.19), are treated perturbatively and do not yield new states when included.

– 74 –
 The generalization to the relativistic case is conceptually straightforward. We refer
the reader to [55, 84] for an introduction to the CCWZ technology for nonlinearly realized
spacetime symmetries and its application to the spinless relativistic point-particle. Here we
focus directly on a particle with spin. The associated symmetry breaking pattern is
(
P0 time translations ,
unbroken =
J1 spatial rotation ,

(C.23)
Pi spatial translations ,


broken = Ki boosts ,


J spatial rotations ,
a

where i = 1, 2, 3 and a = 2, 3 as before. The coset can be parametrized as

µ (τ )P i (τ )K
Ω = eiX µ
eiη i
ΩR , (C.24)

where ΩR was defined in eq. (C.16). The Maurer-Cartan one-form is given by

Ω−1 Ω̇ = iE P0 + ∇π i Pi + ∇η i Ki + ∇Θas Ja + AP J1 .


(C.25)

Here E transforms as a worldline vielbein, while the coefficients of the broken generators can
be thought as the covariant derivatives of the Goldstones and, under the action of any element
of the full group, they transform as a linear (local) representation of the unbroken subgroup
SO(2). Finally AP transforms as U (1) connection similarly to A in eq. (C.17). We can write
the covariant derivatives and the connection explicitly in terms of the following matrices
iK
(e−iη i
)µν = (Λ−1 )µν = Λνµ , (ΩR )µν = Rνµ , Λ̄νµ = Rνρ Λρµ . (C.26)

We find
E = Ẋ ν Λν0 , ∇π i = E −1 Ẋ ν Λ̄νi , ∇η i = −E −1 Λ̄ν0 Λ̄˙ νi ,
1 (C.27)
∇Θas = − εaij E −1 Λ̄νi Λ̄˙ νj , AP = −E −1 Λ̄ν2 Λ̄˙ ν3 .
2
As in the spinless case, the boost Goldstones η i are redundant [36]. This means that we can
eliminate them imposing an Inverse Higgs Constraint [54], namely a covariant condition which
allows to express the redundant Goldstones in terms of other Goldstone fields. For the case at
hand the relevant condition is
ηi Ẋ i
∇π i = 0 =⇒ tanh η = . (C.28)
η Ẋ 0
As explained in [55], eq. (C.28) implies Λµ0 = uµ , so that the vectors {Λµi } form an orthonormal
basis for the normal vectors on the worldline.
We may finally write the action for the point-particle. To lowest order in derivatives the
kinetic term for its coordinates is simply given by the vielbein
Z Z p
S = −m dτ E = −m dτ Ẋ 2 , (C.29)

– 75 –
where we used eq. (C.28). It is possible to check that, upon using eq. (C.28), the ∇η i become
proportional to the particle’s acceleration and therefore are of higher order in derivative
expansion. As in the non-relativistic case, we have two choices for the leading order action for
the rotational Goldstones (θs , ϕs ) analogously to the non-relativistic case. If we assume that
they do not transform under T , we consider a rigid-rotor like kinetic term:
Z
I
dτ E ∇Θas ∇Θas . (C.30)
2
Eq. (C.30) has physical applications in the description of astrophysical objects [55]. Unsur-
prisingly, the spin action (C.8) is given instead by
Z
Sspin = −s dτ E AP , (C.31)

which as eq. (C.20) is compatible with time reversal only if the spin flips sign under T .
Finally, we remark that (C.31) is the only allowed term that cannot be written as a local
integral in terms of the vector n̂µ = S µ /s, which satisfies n̂µ n̂µ = −1 and n̂µ uµ = 0. For
instance, we have
˙ ˙
E∇η 1 = −u̇µ n̂µ , E 2 ∇η a ∇η a = u̇µ Ŝµν Ŝ νρ u̇ρ ,
E 2 ∇Θas ∇Θas = Ŝ µν Ŝµν + u̇µ Ŝµν Ŝ νρ u̇ρ ,
(C.32)
µ ν
where Ŝρσ = Sρσ /s = εµνρσ n̂ u . To write the connection AP we instead need to evaluate the
coset matrix in a specific representation, as in eq. (C.21). The Dirac spinor in eq. (C.8) is
the minimal choice, that allows writing E AP without picking additional terms (e.g. a Weyl
spinor always picks up a term proportional to E ∇η 1 ).

D Worldsheet axion energy levels from the Mathieu equation

Eq. (6.15) can be written as a Mathieu equation in the form

m2a m2a
a′′ + [R − 2Q cos(2s)] a = 0 , R = ϵ2 − , Q= . (D.1)
2ω 2 4ω 2
The most general solution is a linear combination of the two Mathieu functions. For fixed Q,
only a series of discrete values of R labeled by k ∈ N yields solutions satisfying Dirichlet or
Neumann boundary conditions. These are
(
Bk (Q) k = even > 0
Dirichlet =⇒ R = Dk (Q) ≡ (D.2)
Ak (Q) k = odd > 0 ,
(
Ak (Q) n = even ≥ 0
Neumann =⇒ R = Nk (Q) ≡ (D.3)
Bk (Q) n = odd > 0 ,
where Ak and Bk are the so called Mathieu characteristics (MathieuCharacteristicA and
MathieuCharacteristicB in Mathematica), and we defined Dk and Nk for future reference.

– 76 –
(Notice that k = 0 is allowed only for Neumann). For Q = 0 we have Ak (0) = Bk (0) = k 2 ,
and we recover thus the massless solution discussed in Section 4.1.
As explained in Section 6.2, we are interested in the large Q ∼ m2a /ω 2 limit. In this limit
the Mathieu Characteristics admit the following expansion [85]
p
Ak (Q) ∼ Bk+1 (Q) = −2Q + 2(2k + 1) Q + O(Q0 ) . (D.4)
√
Additionally, the difference Bk+1 (Q) − Ak (Q) ∼ e−4 Q is nonperturbative in Q. This is
because the Mathieu equation has the structure of a Schrödinger equation in a double-well
potential, as explained below eq. (6.16).
In conclusion, depending on the boundary conditions, the single-particle energy levels are
given by
 2 (
m 2 ma (2n + 3) mωa n even
ϵ2n = a2 + Dn+1 ≃ (D.5)
2ω 4ω 2 (2n + 1) ma n odd , ω

for Dirichlet, and by

(
m2 m2a (2n + 1) mωa
 
n even
ϵ2n = a2 + Nn ≃ (D.6)
2ω 4ω 2 (2n − 1) mωa n odd ,

for Neumann, where n ∈ N and we ordered such that ϵn < ϵn+1 . Eq. (D.5) and (D.6) imply
a double-degeneracy to all orders in 1/Q ∼ ω/ma . Notice that this result coincides with
eq. (6.19) to leading order in ω/ma . Under σ → −σ, the wave-functions are even (MathieuC
in Mathematica) for n =even, and odd (MathieuS in Mathematica) for n =odd.

E Daughter Regge trajectories for heavy quarks

In this appendix we briefly analyze the spectrum of fluctuations for heavy quarks. We limit
ourselves to the leading order in derivatives, therefore neglecting the contribution proportional
to b̄1 = 0 in eq. (7.3). The analysis is largely analogous to that of Section 4. The main
difference is that the boundary mode α cannot be thought of as an auxiliary field anymore.
This is because the expansion of the boundary action (7.3) results in the following contribution
to the quadratic action
X 1/(αf,0 ω 2 ) Z h i
2
ϕ̇2 + 1 − αf,0
2

2
α̇ + α2 ± 2αf,0 2 − αf,02 2
)ρ̇2i ,


q dτ αf,0 αϕ̇ + (1 − αf,0
2 2
σ=±1 2ℓs 1 − αf,0

(E.1)

where we wrote m̃ in terms of αf,0 to make manifest that this contribution is of the same
order as the NG action. Therefore we cannot integrate out α as for light quarks. Note that
α mixes nontrivially with the bulk field ϕ, and therefore it does not correspond to a special
boundary mode at generic values of αf,0

– 77 –
 The spectrum of fluctuations is then extracted from the solutions of the EOMs that follow
from the action (E.1)+(4.52). The analysis is straightforward and we only report the results.
We refer the reader to [28] for details.
For the transverse modes ρi , in units of ω the frequencies are the solutions of the following
equation
q
2 cos (2ϵ arcsin(α )) + α2 + ϵ2 1 − α2


2ϵ αf,0 1 − αf,0 f,0 f,0 f,0 sin (2ϵ arcsin(αf,0 )) = 0 . (E.2)

Eq. (E.2) is solved by ϵ = 0, 1 for all values of αf,0 , as expected due to the protected nature of
these modes. In the relativistic regime αf,0 ≃ 1 the other modes have frequencies given by

3/2
2 mωℓ2s
n n2 − 1 + . . .


ϵn = n + n = 2, 3, . . . , (E.3)
3π
in agreement with eq. (4.70). In the NR regime αf,0 ≪ 1 there is a parametric separation
between the first two protected modes ϵ = 0, 1 and the other string excitations, whose gap
scales as 1/αf,0 ∼ mωℓ2s ≫ 1 and is given by
π(n − 1) π 2 (n − 1)2 /4 − 6
ϵn ≃ − αf,0 + ...
2αf,0 3π(n − 1)
(E.4)
5π (n − 1)2 /4 + 6 ℓ−2
 2 
π(n − 1)mω s
= + + ... n = 2, 3, . . . .
2ℓ−2
s 3π(n − 1)mω
The leading order term in eq. (E.4) coincides with the answer for the energy levels on a static
string of length 2αf,0 /ω and Dirichlet boundary conditions.
The eigenfrequencies of the α/ϕ system are the solutions of the following equation:
h
2
2 4 i
2 2 4

2 2
1 + αf,0 − 2 1 + 2αf,0 − 3αf,0 ϵ + 1 − αf,0 ϵ sin (2ϵ arcsin(αf,0 ))
q
2 2
2

2
+ 4ϵ αf,0 1 − αf,0 1 + αf,0 − 1 − αf,0 ϵ cos (2ϵ arcsin(αf,0 )) = 0 . (E.5)

As before, the lowest modes are protected and have ϵ = 0, 1, that solve eq. (E.5) for any value
of αf,0 . In the relativistic regime the other modes have frequency given by
1
3/2
mωℓ2s n n2 − 1 + . . .


ϵn = n − n = 2, 3, . . . , (E.6)
3π
in agreement with the result for light quarks eq. (4.73). In the NR regime the first unprotected
mode has frequency given by
√ 2
αf,0 √ 1
ϵ2 = 3 + √ + ... = 3 + √ + ... . (E.7)
3 3 3 3m2 ω 2 ℓ4s
The other modes are parametrically heavier and, similarly to eq. (E.4), their gap at leading
order is as for a static string of length 2αf,0 /ω and Dirichlet boundary conditions:
π(n − 2) αf,0 π 2 (n − 2)2 − 48
 
ϵn = − + ...
2αf,0 12π(n − 2)
(E.8)
5π (n − 2)2 + 48 ℓ−2
 2 
π(n − 2)mω s
≃ + + ... n = 3, 4, . . . .
2ℓ−2
s 12π(n − 2)mω

– 78 –
Note that the nth mode of ρi and the (n + 1)th ϕ/α mode have the same energy at leading
order in the NR limit.
A similar analysis can be performed for the axion. In the relativistic regime mωℓ2s ≪ 1 we
recover the results of Section 6.2, while for a NR rotating string its energy levels approximately
coincide with those of a massive particle on a static string:
r q
m2a π 2 (n+1)2 m2a 2

 ω2
 + 4α2f,0
= ω2
+ π4 (n + 1)2 m2 ω 2 ℓ4s for Dirichlet
ϵn ≃ r q n = 0, 1, . . . ,
m2 π 2 n2 m2a π2 2 2 2 4
 ω2a + 4α = + n m ω ℓ for Neumann

 2 ω2 4 s
f,0

(E.9)
where we reported the result both for Dirichlet and Neumann boundary conditions according
to the discussion in Section 6.2.
In summary, in the relativistic limit, we recover results analogous to those for light quarks,
while in the NR regime the spectrum coincides with that of a static string with Dirichlet
boundary conditions, but for the protected modes and the excitation (E.7), that creates the
only parametrically light states. As explained in the main text in Section 7.2, these light
states admit a simple interpretation in terms of a quark model with linear potential.

F Tables of mesons

This appendix reports the data that we used for mesons and lists the potential stringy
interpretation of each particle purely based on its quantum numbers. Fig. 4 summarizes our
predictions for the lightest excited states on top of the leading Regge trajectory, and tables 3
to 5 list and compare the mesons from PDG with these predictions.
At each spin J, we take the lightest ρ/a meson for I = 1 and the lightest ω/f meson
for I = 0 to be the string gorund-state. Fig. 3 and fig. 4 show that the parity assignment of
these states follow the same behavior expressed in eq. (8.3). As discussed in the first part
of this paper, we expect to have two string excitations at the second energy level on top of
each state on the Regge trajectory. Looking at PDG, we see candidates for this prediction
with parity assignments that match with table 1. Due to the subleading order corrections,
these two degenerate excited states would have an energy splitting that results in the angular
fluctuation to be lighter than the transverse one. The first plot in fig. 8 represents a subset of
particles with I = 1 that approximately follow this mass hierarchy.
Based on the discussion on parity assignments for spin states explained in Section (8.1),
for each string ground-state, we might have up to three associated spin states in the vicinity
of leading Regge trajectory. The second plot in fig. 8 shows the I = 1 PDG candidates for
these spin states. Following the parity assignment for pseudoscalar in table 1, we also expect
to see up to three axionic excitations among the particles around the first two daugther Regge
trajectories; we included three because the gap of the axion is not determined within EFT,
and it is a priori possible that the resulting modes are parametrically light. In fig. 8, the third
plot shows the PDG I = 1 candidates for these axionic excitations.

– 79 –
 axionic spin axionic spin
excitations excitations excitations excitations

n=2

t
n=2
t tt
t n=1 s
t
a

t n=1
n=0
s
t

t
n=0

J=2k J=2k+1

Figure 7: Qualitative presentation of the predictions for the lightest daughter Regge
trajectories: For each spin J, the states are classified into three categories, separated by gray dashed
lines; we know the energy hierarchy in each category, but we don’t know the hierarchy between different
categories. The first column shows the axionic excitations (n = 0, n = 1, and two n = 0). The second
one shows the string excitations around second daughter Regge trajectory. The last column shows the
spin excitations. The double-side arrows are included because the hierarchy between those two states
is not determined from EFT.

Other than the single excitations and the double-axion of first-level (presented in fig. 7),
there are some multiple excitations that could be detectable in the range of our study. Potential
PDG candidates for such states are marked on the last two columns of tables 3 to 5. The
I = 1 candidates for such excitations are demonstrated in fig. 9. They all have overlaps with
the particles shown in fig. 8 except for π1 (1400) and π1 (1600)40 , known as isovector exotic
mesons. As mentioned in the main text, in our framework, their quantum numbers could only
match with an axionic excitation of a spin state.

40
In fig. 9, they are represented in black.

– 80 –
Table 3: Flavourless light mesons with isospin one, I = 1: Numbers in parenthesis are particle
masses in MeV. In the first column, “?” means that the particle is on Meson Summary Table but has
yet to be established; the particles that PDG refers to as Further States are marked by “??”. The
particles are first ordered by spin and then, by mass. The spin and discrete quantum numbers are given
in the third column. In the remaining columns, we put a check symbol if the particle has the right
quantum numbers to be interpreted as respectively a string ground-state, a transverse string excitation,
an angular string excitation, a spin state, an axionic excitation of the respective ground-state, an
axionic excitation of a spin state, or the angular fluctuation of a spin state. For the axionic states, the
level of axion excitation n is also determined. The symbol “∗” stands double-axion excitations. The
color codes are the same as those used in our plots.

JPC GS ρ ϕ S na S, a S, ϕ
π 0 (135) 0−+ ✓ 0
a0 (980) 0++ ✓ ✓ 0∗
π(1300) 0−+ ✓ ✓ ✓
a0 (1450) 0++ ✓ 0∗ ✓
ρ(770) 1−− ✓
b1 (1235) 1+− ✓ 0
a1 (1260) 1++ ✓ 1
π1 (1400) 1−+ ✓
ρ1 (1450) 1−− ✓ ✓
? ρ1 (1570) 1−− ✓ ✓ 0∗ ✓
π1 (1600) 1−+ ✓
a1 (1640) 1++ 1 ✓
ρ(1700) 1−− ✓ ✓ 0∗ ✓
?? π1 (1960) 1+− ✓ ✓
a2 (1320) 2++ ✓
π2 (1670) 2−+ ✓ 0
a2 (1700) 2++ ✓ ✓
π2 (1880) 2−+ ✓ ✓
?? ρ2 (1940) 2−− ✓ 1 ✓
?? a2 (1950) 2++ ✓ ✓ 0∗ ✓
?? a2 (1990) 2++ ✓ ✓ 0∗ ✓
? π2 (2005) 2−+ ✓ ✓
?? a2 (2030) 2++ ✓ 0∗ ✓
? π2 (2100) 2−+ ✓ ✓
ρ3 (1690) 3−− ✓

– 81 –
 JPC GS ρ ϕ S na S, a S, ϕ
?? a3 (1875) 3++ ✓
? ρ3 (1990) 3−− ✓
?? b3 (2030) 3+− ✓ 0
?? a3 (2030) 3++ ✓ 1 ✓
?? b3 (2245) 3+− ✓ ✓
? ρ3 (2250) 3−− ✓ ✓ 0∗ ✓
?? a3 (2275) 3++ 1 ✓
a4 (1970) 4++ ✓
? ρ5 (2350) 5−− ✓
?? a6 (2450) 6++ ✓

Table 4: Flavourless light mesons with isospin zero, I = 0

JPC GS ρ ϕ S na S, a S, ϕ
f0 (500) 0++ ✓
η(548) 0−+ ✓ 0
η(958) 0−+ ✓ 0 ✓
f0 (980) 0++ ✓ ✓ 0∗
η(1295) 0−+ ✓ ✓ ✓
f0 (1370) 0++ ✓ 0∗ ✓
η(1405) 0−+ ✓
η(1475) 0−+ ✓
f0 (1500) 0++ ✓ 0∗ ✓
ω(782) 1−− ✓
ϕ(1020) 1−− ✓
η(1170) 1+− ✓ 0
f1 (1285) 1++ ✓ 1
h1 (1415) 1+− ✓ 0 ✓
f1 (1420) 1++ ✓ 1 ✓
ω(1420) 1−− ✓ ✓ 0∗
? f1 (1510) 1++ ✓ 1 ✓
? h1 (1595) 1+− ✓ ✓

– 82 –
 JPC GS ρ ϕ S na S, a S, ϕ
ω(1650) 1−− ✓ ✓ 0∗ ✓
ϕ(1680) 1−− ✓ 0∗ ✓
? η1 (1855) 1−+ ✓
f2 (1270) 2++ ✓
? f2 (1430) 2++ ✓
f2′ (1525) 2++ ✓
? f2 (1565) 2++ ✓
? f2 (1640) 2++ ✓
η2 (1645) 2−+ ✓ 0
?? f2 (1750) 2++ ✓ ✓
? f2 (1810) 2++ ✓ ✓
η2 (1870) 2−+ ✓ ✓
? f2 (1910) 2++ ✓ ✓ 0∗
f2 (1950) 2++ ✓ ✓ 0∗ ✓
?? ω2 (1975) 2−− ✓ 1 ✓
?? f2 (2000) 2++ ✓ ✓ 0∗ ✓
f2 (2010) 2++ ✓ ✓ 0∗ ✓
?? η2 (2030) 2−+ ✓
ω3 (1670) 3−− ✓
ϕ3 (1850) 3−− ✓
?? ω3 (1945) 3−− ✓
?? h3 (2025) 3+− ✓ 0
?? f3 (2050) 3++ ✓ 1
?? ω3 (2255) 3−− ✓ ✓ 0∗ ✓
?? h3 (2275) 3+− ✓ ✓
f4 (2050) 4++ ✓
?? ω5 (2250) 5−− ✓
? f6 (2510) 6++ ✓

– 83 –
 Table 5: Strange mesons with isospin half, I = 1/2

JPC GS ρ ϕ S na S, a S, ϕ
K 0 (498) 0− ✓ 0
K0∗ (700) 0+ ✓
K0∗ (1430) 0+ ✓ 0∗ ✓
K(1460) 0− ✓ ✓ 1
K ∗ (892) 1− ✓
K1 (1270) 1+ ✓ 0
K1 (1400) 1+ ✓ 0/1
K ∗ (1410) 1− ✓ ✓
K1 (1650) 1+ ✓ 1 ✓
K ∗ (1680) 1− ✓ ✓ 0∗ ✓
K2∗ (1430) 2+ ✓
? K2 (1580) 2− 0
K2 (1770) 2− ✓ 0/1
K2 (1820) 2− ✓ 0/1
K2∗ (1980) 2+ ✓ ✓ 0∗ ✓
K3∗ (1780) 3− ✓
? K3 (2320) 3+ ✓ ✓ 1
K4∗ (2045) 4+ ✓
? K5∗ (2380) 5− ✓

– 84 –
 M2
π /b (J-+)
ρ/a (J++ ) **
5 π /b (J+-) * **

ρ/a (J-- )
*
** **
4 ** * *
** **
**

0 1 2 3
J

-1

(a) PDG candidates for string excitations

M2 M2
-+
π /b (J ) π /b (J-+, n=0)
6 6
ρ/a (J++ ) ρ/a (J++ , n=0)
ρ/a (J-- ) ρ/a (J-- , n=1)
π /b (J+-) π /b (J+-, n=0)
** **
5 *
5 *

ρ/a (J-- ) ρ/a (J-- , n=0)

ρ/a (J++ ) ** ** **
ρ/a (J++ , n=1) ** ** **
4 **
** * 4 **
** *
** ** ** **
** **

3 3
* *

2 2

1 1

0 1 2 3
J 0 1 2 3
J

-1 -1

(b) PDG candidates for spin states (c) PDG candidates for axionic excitations

Figure 8: I=1 PDG candidates for our predictions: The solid line shows the leading Regge
trajectory, which is associated with ρ/a meson. The dashed lines show the first two daughter Regge
trajectories: Mn2 (J) = M 2 (J)leading + n/ℓ̄2s with n = 1, 2. In (a), the blue markers represent the
candidates for the angular excitation, while particles in orange are candidates for the transverse
excitation. In (b), the orange and gray markers stand for candidates for the first two spin states in
plot 7. The last, expectedly heavier, spin state with the same quantum numbers as the ground-state, is
shown in blue. In (c), the candidates for the lightest axionic excitation, associated with level n = 0,
are shown in orange. The particles in blue are candidates for double-axion excitation with n = 0. An
axionic excitation with n = 1, heavier than the first n = 0 excitation, could appear either above or
below the second n = 0 excitation in gray.

– 85 –
 M2 M2
** **
5 *
5 **
++ π /b (J-+)
ρ/a (J )
ρ/a (J-- )
π /b (J+-) *
π /b (J+-)
4 ρ/a (J-- ) ** ** **
4 *
** **
**
**
**
*
ρ/a (J-- ) **
**
** **

3 3

* *

2 2

1 1

0 1 2 3
J 0 1 2 3
J

-1 -1

(a) PDG candidates for axionic excitations of (b) PDG candidates for angular string fluctua-
spin states tions of spin states

Figure 9: I=1 PDG candidates for mixed excitations: In (a), we represent the candidates
for n = 0 and n = 1 axionic excitations of spin states, which could be light enough to be observable
in the range of our study. Noting that the angular string excitation is lighter than transverse string
excitation, we only show the candidates for angular fluctuations of spin states in (b).

– 86 –
References

[1] J. Elias Miró, A. L. Guerrieri, A. Hebbar, J. a. Penedones and P. Vieira, Flux Tube S-matrix
Bootstrap, Phys. Rev. Lett. 123 (2019) 221602 [1906.08098].
[2] J. Elias Miró and A. Guerrieri, Dual EFT bootstrap: QCD flux tubes, JHEP 10 (2021) 126
[2106.07957].
[3] A. Gaikwad, V. Gorbenko and A. L. Guerrieri, QCD worldsheet axion from the bootstrap, JHEP
01 (2024) 090 [2310.20698].
[4] A. Athenodorou, S. Dubovsky, C. Luo and M. Teper, Confining Strings and the Worldsheet
Axion from the Lattice, 2411.03435.
[5] J. Polchinski and A. Strominger, Effective string theory, Phys. Rev. Lett. 67 (1991) 1681.
[6] S. Dubovsky, R. Flauger and V. Gorbenko, Effective String Theory Revisited, JHEP 09 (2012)
044 [1203.1054].
[7] O. Aharony and Z. Komargodski, The Effective Theory of Long Strings, JHEP 05 (2013) 118
[1302.6257].
[8] S. Dubovsky, R. Flauger and V. Gorbenko, Evidence from Lattice Data for a New Particle on the
Worldsheet of the QCD Flux Tube, Phys. Rev. Lett. 111 (2013) 062006 [1301.2325].
[9] A. Athenodorou and M. Teper, On the mass of the world-sheet ’axion’ in SU (N ) gauge theories
in 3+1 dimensions, Phys. Lett. B 771 (2017) 408 [1702.03717].
[10] S. Dubovsky and V. Gorbenko, Towards a Theory of the QCD String, JHEP 02 (2016) 022
[1511.01908].
[11] A. Athenodorou and M. Teper, SU(N) gauge theories in 2+1 dimensions: glueball spectra and
k-string tensions, JHEP 02 (2017) 015 [1609.03873].
[12] P. Conkey, S. Dubovsky and M. Teper, Glueball spins in D = 3 Yang-Mills, JHEP 10 (2019) 175
[1909.07430].
[13] A. Athenodorou and M. Teper, SU(N) gauge theories in 3+1 dimensions: glueball spectrum,
string tensions and topology, JHEP 12 (2021) 082 [2106.00364].
[14] S. Caron-Huot, Z. Komargodski, A. Sever and A. Zhiboedov, Strings from Massive Higher Spins:
The Asymptotic Uniqueness of the Veneziano Amplitude, JHEP 10 (2017) 026 [1607.04253].
[15] A. Sever and A. Zhiboedov, On Fine Structure of Strings: The Universal Correction to the
Veneziano Amplitude, JHEP 06 (2018) 054 [1707.05270].
[16] S. Hellerman and I. Swanson, String Theory of the Regge Intercept, Phys. Rev. Lett. 114 (2015)
111601 [1312.0999].
[17] S. Hellerman and I. Swanson, Boundary Operators in Effective String Theory, JHEP 04 (2017)
085 [1609.01736].
[18] Particle Data Group collaboration, S. Navas et al., Review of particle physics, Phys. Rev. D
110 (2024) 030001.
[19] S. Dubovsky and G. Hernandez-Chifflet, Yang–Mills Glueballs as Closed Bosonic Strings, JHEP
02 (2017) 022 [1611.09796].

– 87 –
[20] S. Dubovsky, G. Hernández-Chifflet and S. Zare, 3D Yang-Mills glueballs vs closed effective
strings, JHEP 07 (2021) 216 [2104.02154].
[21] W. D. Goldberger, Les Houches lectures on effective field theories and gravitational radiation, in
Les Houches Summer School - Session 86: Particle Physics and Cosmology: The Fabric of
Spacetime, 1, 2007, hep-ph/0701129.
[22] A. Chodos and C. B. Thorn, Making the Massless String Massive, Nucl. Phys. B 72 (1974) 509.
[23] M. Baker and R. Steinke, Effective string theory of vortices and Regge trajectories, Phys. Rev. D
63 (2001) 094013 [hep-ph/0006069].
[24] M. Baker and R. Steinke, Semiclassical quantization of effective string theory and Regge
trajectories, Phys. Rev. D 65 (2002) 094042 [hep-th/0201169].
[25] F. Wilczek, Diquarks as inspiration and as objects, in Deserfest: A Celebration of the Life and
Works of Stanley Deser, pp. 322–338, 9, 2004, hep-ph/0409168, DOI.
[26] M. Kruczenski, L. A. Pando Zayas, J. Sonnenschein and D. Vaman, Regge trajectories for mesons
in the holographic dual of large-N(c) QCD, JHEP 06 (2005) 046 [hep-th/0410035].
[27] S. Hellerman, S. Maeda, J. Maltz and I. Swanson, Effective String Theory Simplified, JHEP 09
(2014) 183 [1405.6197].
[28] J. Sonnenschein and D. Weissman, Quantizing the rotating string with massive endpoints, JHEP
06 (2018) 148 [1801.00798].
[29] S. Hellerman and I. Swanson, Droplet-Edge Operators in Nonrelativistic Conformal Field
Theories, 2010.07967.
[30] S. Hellerman, D. Orlando, V. Pellizzani, S. Reffert and I. Swanson, Nonrelativistic CFTs at large
charge: Casimir energy and logarithmic enhancements, JHEP 05 (2022) 135 [2111.12094].
[31] G. Cuomo and Z. Komargodski, Giant Vortices and the Regge Limit, JHEP 01 (2023) 006
[2210.15694].
[32] K. J. Juge, J. Kuti and C. Morningstar, Excitations of the static quark anti-quark system in
several gauge theories, in International Conference on Color Confinement and Hadrons in
Quantum Chromodynamics - Confinement 2003, pp. 221–232, 12, 2003, hep-lat/0312019, DOI.
[33] A. Sharifian, A. Athenodorou and P. Bicudo, The spectrum of open confining strings in the
large-Nc limit, 11, 2024, 2411.07412.
[34] N. Isgur, Spin orbit inversion of excited heavy quark mesons, Phys. Rev. D 57 (1998) 4041.
[35] A. Guerrieri, K. Häring and N. Su, From data to the analytic S-matrix: A Bootstrap fit of the
pion scattering amplitude, 2410.23333.
[36] I. Low and A. V. Manohar, Spontaneously broken space-time symmetries and Goldstone’s
theorem, Phys. Rev. Lett. 88 (2002) 101602 [hep-th/0110285].
[37] A. Nicolis, R. Penco, F. Piazza and R. A. Rosen, More on gapped Goldstones at finite density:
More gapped Goldstones, JHEP 11 (2013) 055 [1306.1240].
[38] H. Georgi, Generalized dimensional analysis, Phys. Lett. B 298 (1993) 187 [hep-ph/9207278].
[39] S. Dubovsky, R. Flauger and V. Gorbenko, Flux Tube Spectra from Approximate Integrability at
Low Energies, J. Exp. Theor. Phys. 120 (2015) 399 [1404.0037].

– 88 –
[40] A. Athenodorou, B. Bringoltz and M. Teper, Closed flux tubes and their string description in
D=2+1 SU(N) gauge theories, JHEP 05 (2011) 042 [1103.5854].
[41] A. Athenodorou and M. Teper, Closed flux tubes in D = 2 + 1 SU(N ) gauge theories: dynamics
and effective string description, JHEP 10 (2016) 093 [1602.07634].
[42] C. Chen, P. Conkey, S. Dubovsky and G. Hernández-Chifflet, Undressing Confining Flux Tubes
with T T̄ , Phys. Rev. D 98 (2018) 114024 [1808.01339].
[43] A. M. Polyakov, Fine Structure of Strings, Nucl. Phys. B 268 (1986) 406.
[44] P. O. Mazur and V. P. Nair, Strings in QCD and θ Vacua, Nucl. Phys. B 284 (1987) 146.
[45] S. Weinberg, The quantum theory of fields. Vol. 2: Modern applications. Cambridge University
Press, 8, 2013.
[46] S. Garcia-Saenz, E. Mitsou and A. Nicolis, A multipole-expanded effective field theory for vortex
ring-sound interactions, JHEP 02 (2018) 022 [1709.01927].
[47] G. F. Cuomo, Large charge, semiclassics and superfluids: from broken symmetries to conformal
field theories, Ph.D. thesis, EPFL, 2020. 10.5075/epfl-thesis-8397.
[48] A. Nicolis and F. Piazza, Implications of Relativity on Nonrelativistic Goldstone Theorems:
Gapped Excitations at Finite Charge Density, Phys. Rev. Lett. 110 (2013) 011602 [1204.1570].
[49] P. Goddard, J. Goldstone, C. Rebbi and C. B. Thorn, Quantum dynamics of a massless
relativistic string, Nucl. Phys. B 56 (1973) 109.
[50] J. F. Arvis, The Exact q q̄ Potential in Nambu String Theory, Phys. Lett. B 127 (1983) 106.
[51] M. Rivas, Classical relativistic spinning particles, Journal of Mathematical Physics 30 (1989) 318
[https://pubs.aip.org/aip/jmp/article-pdf/30/2/318/8158674/318 1 online.pdf].
[52] G. Cognola, “General relativity and cosmology.”
http://www.science.unitn.it/~cognola/DIDATTICA/RGeng.pdf.
[53] J. B. Kogut and G. Parisi, Long Range Spin Spin Forces in Gauge Theories, Phys. Rev. Lett. 47
(1981) 1089.
[54] E. A. Ivanov and V. I. Ogievetsky, The Inverse Higgs Phenomenon in Nonlinear Realizations,
Teor. Mat. Fiz. 25 (1975) 164.
[55] L. V. Delacrétaz, S. Endlich, A. Monin, R. Penco and F. Riva, (Re-)Inventing the Relativistic
Wheel: Gravity, Cosets, and Spinning Objects, JHEP 11 (2014) 008 [1405.7384].
[56] G. Cuomo, M. Mezei and A. Raviv-Moshe, Boundary conformal field theory at large charge,
JHEP 10 (2021) 143 [2108.06579].
[57] A. Monin, D. Pirtskhalava, R. Rattazzi and F. K. Seibold, Semiclassics, Goldstone Bosons and
CFT data, JHEP 06 (2017) 011 [1611.02912].
[58] Particle Data Group collaboration, R. L. Workman and Others, Review of Particle Physics,
PTEP 2022 (2022) 083C01.
[59] G. S. Bali, F. Bursa, L. Castagnini, S. Collins, L. Del Debbio, B. Lucini et al., Mesons in large-N
QCD, JHEP 06 (2013) 071 [1304.4437].
[60] C. Amsler, The Quark Structure of Hadrons: An Introduction to the Phenomenology and
Spectroscopy, vol. 949. Springer, Cham, 2018, 10.1007/978-3-319-98527-5.

– 89 –
[61] L. M. Abreu, F. M. da Costa Júnior and A. G. Favero, Revisiting the tensor J P C = 2−− meson
spectrum, Phys. Rev. D 101 (2020) 116016 [2004.10736].
[62] C. A. Meyer and E. S. Swanson, Hybrid Mesons, Prog. Part. Nucl. Phys. 82 (2015) 21
[1502.07276].
[63] H.-X. Chen, W. Chen, X. Liu, Y.-R. Liu and S.-L. Zhu, A review of the open charm and open
bottom systems, Rept. Prog. Phys. 80 (2017) 076201 [1609.08928].
[64] L. Mezincescu and P. K. Townsend, Anyons from Strings, Phys. Rev. Lett. 105 (2010) 191601
[1008.2334].
[65] W. B. Bonnor, Curves with null normals in Minkowski space-time., in A Random Walk in
Relativity and Cosmology (M. Dadhich, J. Krishna Rao, J. V. Narlikar and C. V. Vishveshwara,
eds.), pp. 33–47, Jan., 1985.
[66] F. A. Berezin and M. S. Marinov, Particle Spin Dynamics as the Grassmann Variant of Classical
Mechanics, Annals Phys. 104 (1977) 336.
[67] A. Barducci, R. Casalbuoni and L. Lusanna, Supersymmetries and the Pseudoclassical Relativistic
electron, Nuovo Cim. A 35 (1976) 377.
[68] L. Brink, P. Di Vecchia and P. S. Howe, A Lagrangian Formulation of the Classical and Quantum
Dynamics of Spinning Particles, Nucl. Phys. B 118 (1977) 76.
[69] M. J. Landry and G. Sun, The coset construction for particles of arbitrary spin, JHEP 05 (2021)
040 [2010.11191].
[70] A. Y. Alekseev and S. L. Shatashvili, Propagator for the Relativistic Spinning Particle via
Functional Integral Over Trajectories, Mod. Phys. Lett. A 3 (1988) 1551.
[71] P. B. Wiegmann, Multivalued Functionals and Geometrical Approach for Quantization of
Relativistic Particles and Strings, Nucl. Phys. B 323 (1989) 311.
[72] S. Forte, Spin in quantum field theory, Lect. Notes Phys. 712 (2007) 67 [hep-th/0507291].
[73] S. Sachdev, Quantum Phase Transitions. Cambridge University Press, 2 ed., 2011,
10.1017/CBO9780511973765.
[74] E. H. Lieb, The classical limit of quantum spin systems, Communications in Mathematical
Physics 31 (1973) 327.
[75] E. Rabinovici, A. Schwimmer and S. Yankielowicz, Quantization in the Presence of Wess-Zumino
Terms, Nucl. Phys. B 248 (1984) 523.
[76] M. P. Clark, A Semi-classical analysis of the Wilson Loop in a 2+1 Dimensional Yang-Mills
theory with a monopole gas, Ph.D. thesis, University of British Columbia, 1997.
http://dx.doi.org/10.14288/1.0085080.
[77] T. T. Wu and C. N. Yang, Dirac monopole without strings: Monopole harmonics, Nuclear
Physics B 107 (1976) 365.
[78] G. V. Dunne, R. Jackiw and C. A. Trugenberger, Topological (Chern-Simons) Quantum
Mechanics, Phys. Rev. D 41 (1990) 661.
[79] K. Hasebe, Hopf Maps, Lowest Landau Level, and Fuzzy Spheres, SIGMA 6 (2010) 071
[1009.1192].

– 90 –
[80] G. Cuomo, Z. Komargodski, M. Mezei and A. Raviv-Moshe, Spin impurities, Wilson lines and
semiclassics, JHEP 06 (2022) 112 [2202.00040].
[81] H. K. Dreiner, H. E. Haber and S. P. Martin, Two-component spinor techniques and Feynman
rules for quantum field theory and supersymmetry, Phys. Rept. 494 (2010) 1 [0812.1594].
[82] S. R. Coleman, J. Wess and B. Zumino, Structure of phenomenological Lagrangians. 1., Phys.
Rev. 177 (1969) 2239.
[83] C. G. Callan, Jr., S. R. Coleman, J. Wess and B. Zumino, Structure of phenomenological
Lagrangians. 2., Phys. Rev. 177 (1969) 2247.
[84] V. I. Ogievetsky, Nonlinear realizations of internal and space-time symmetries, Proceedings of the
Xth Winter School of Theoretical Physics in Karpacz 1 (1974) 117.
[85] “NIST Digital Library of Mathematical Functions 5.4.” https://dlmf.nist.gov/28.8.

– 91 –