Prepared for submission to JHEP MIT-CTP/5566

Towards natural and realistic E7 GUTs in F-theory

arXiv:2401.00040v1 [hep-th] 29 Dec 2023

Shing Yan Li and Washington Taylor

Center for Theoretical Physics
Department of Physics
Massachusetts Institute of Technology
77 Massachusetts Avenue
Cambridge, MA 02139, USA

E-mail: sykobeli at [Link], wati at [Link]

Abstract: We consider phenomenological aspects of a natural class of Standard Model-like

supersymmetric F-theory vacua realized through flux breaking of rigid E7 gauge factors. Three
generations of Standard Model matter are realized in many of these vacua. We further find
that many other Standard Model-like features are naturally compatible with these construc-
tions. For example, dimension-4 and 5 terms associated with proton decay are ubiquitously
suppressed. Many of these features are due to the group theoretical structure of E7 and as-
sociated F-theory geometry. In particular, a set of approximate global symmetries descends
from the E7 group, leading to exponential suppression of undesired couplings.
Contents

1 Introduction 2
1.1 Natural vs. tuned features 3
1.2 Review of previous work 5
1.3 Overview of results 7
1.4 Outline of paper 9

2 Fluxes in 4D F-theory models 10

2.1 Vertical and remainder fluxes 10
2.2 Flux breaking 13
2.3 Chiral matter 15
2.4 Vector-like matter 16

3 Review of E7 GUTs in F-theory 18

3.1 Flux breaking of rigid E7 factors 18
3.2 Chiral spectrum in flux-broken E7 models 19

4 Phenomenology of E7 GUTs 21
4.1 Approximate global symmetries 21
4.2 Vector-like exotics 22
4.3 List of Yukawa couplings 24
4.4 Proton decay 26
4.5 Higgs and Yukawa sectors 28
4.5.1 Doublet-triplet splitting and the Higgs masses 28
4.5.2 Structure of Yukawa couplings 30
4.6 Neutrino sector 31
4.7 Gauge coupling unification 32

5 Explicit global constructions of E7 GUTs 34

6 Comparing with other F-theory constructions in the literature 38

6.1 Tuned models 39
6.2 Rigid E6 GUTs 41

7 Conclusion 42

A Toric hypersurfaces 43

–1–
B Flux quantization 45

1 Introduction

String theory provides a consistent framework for a unified theory that combines gravity
with the other fundamental forces described by quantum field theory. To describe the real
world, however, ten-dimensional string theory must be compactified on a real six-dimensional
manifold, and various further objects like branes, fluxes, and orientifolds must be incorporated.
Such constructions give an enormous number (perhaps on the order of something like 10272000
[1]) of string theory vacua, known as the string landscape. As part of the program to realize our
Universe in string theory, it has been a long-standing and primary goal to find the structure
of the Standard Model (SM) of particle physics within the string landscape. While many
(supersymmetric) string vacua have been identified that share many of the principal features
of the Standard Model, there is as yet no single vacuum known in the string landscape that
reproduces all the observed phenomenological details of our world; for recent reviews of work
in this direction, see [2, 3].
Beyond the simple question of the existence of a vacuum matching observed physics,
it is perhaps even more important to understand the extent to which the physical features
of the Standard Model arise naturally in string theory. In other words, we would like to
understand the extent to which solutions like the Standard Model are widespread in the
string landscape or require extensive fine-tuning. This is a principal focus of this work and
the associated research program: we take a top-down perspective on the global set of string
vacua and attempt to identify realizations of the Standard Model that are compatible with
the most typical structures arising in string theory. We use F-theory [4–6] to study these
questions, as this approach gives a global and nonperturbative picture of the largest currently
understood set of string vacua. For reviews of F-theory and applications to Standard Model
constructions, see [7, 8]. This paper describes some more detailed phenomenological aspects
of SM constructions originally presented in [9, 10] that are realized through flux breaking of
rigid E6 and E7 gauge factors, which are relatively common features in F-theory geometries.
Constructing the detailed Standard Model requires many elements such as the gauge
group, the matter content including both chiral matter and the Higgs, the Yukawa couplings,
a supersymmetry (SUSY)-breaking mechanism, values of the 19 free parameters, and possibly
some room to address beyond-SM problems as well as cosmological aspects such as the density
of dark energy. Unfortunately, the current available string theory techniques are far from
enough to compute all these features precisely. Among the above SM features, string theory
techniques for constructing the gauge group and the chiral spectrum are well-established.
While there is some recent progress on the Higgs sector [11–15] and the Yukawa couplings
[16] in a large class of F-theory models, so far no fully precise statement on the realization of
these features in a way that matches observed physics has been made in this context. On the

–2–
other hand, incorporating these established features with e.g. SUSY breaking is far beyond
our current techniques. Although at this moment no complete realization of the Standard
Model has been constructed in any version of string theory, if we can identify a natural class
of models that realize a decent portion of the coarsest features of the SM, these structures
may naturally correlate with certain other features of the SM or beyond SM physics. We will
explore this philosophy in this paper.
One obvious way in which the models studied here (and elsewhere in much of the string
theory literature) differ from observed physics is that we focus on solutions with supersym-
metry. Supersymmetry has not yet been observed at low (TeV or below) energies in nature,
but as a theoretical tool it increases our level of analytic control. By studying solutions with
supersymmetry, we can gain some perspective on global aspects of the string landscape. Of
course, eventually we need to understand non-supersymmetric solutions to match observed
physics. One possibility is that the physics we see is in a broken-symmetry phase of a theory
with supersymmetry at energies beyond the TeV scale. Even if supersymmetry is broken at
the Planck or string scale, many insights gained by exploring the space of supersymmetric
vacua may be relevant to the less controlled non-supersymmetric vacua.

1.1 Natural vs. tuned features

Before describing our results, it is worth clarifying the concept of naturalness used in this
paper. To obtain vacua with all the SM features considered in this paper, quite a few specific
choices must be made in the construction of vacua. A list of such choices in the context of the
models studied in this paper is summarized in §1.3, and the mathematical conditions imposed
for such choices are given at the beginning of §5. The extent to which these different kinds of
choices are natural varies, within a hierarchy of naturalness/tuning. Roughly speaking, each
of the choices made in constructing a specific class of string vacua can be characterized as
belonging to one of the following categories:

1. Physical constraints: These constraints come from string theory itself and must be sat-
isfied in all string compactifications. These constraints ensure physically sensible vacua
that have, e.g., Poincaré (or AdS) invariance. Examples include tadpole cancellation
and primitivity of fluxes.

2. Ubiquitous/common conditions: Let us consider a reasonably large but presumably finite

set of string vacua or compactification geometries, such as N = 1 4D F-theory vacua or
the associated set of topologically distinct complex threefold bases that support elliptic
Calabi-Yau fourfolds. A condition is common if it holds for an O(1) fraction of the set
of vacua or geometries, considered with a simple counting measure. In particular, the
condition is ubiquitous if it holds for a substantial majority. As examples, the existence
of rigid E8 gauge factors in (known) F-theory base geometries is ubiquitous, and that
of E7 gauge factors appears to be common. (See, e.g., [17–19] and discussions below)

–3–
3. Fairly likely conditions: Sometimes, there are a family of similar conditions, such as
possible values of a discrete parameter. Each possible value may only hold for a relatively
small fraction of vacua within the above set, so that none of the conditions are ubiquitous.
We refer to a condition as being fairly likely if the fraction of vacua or geometries with
this property is considerably higher than for most of the other possibilities. As an
illustration, we would say that rolling a sum of 6 on a pair of six-sided dice is “fairly
likely,” although rolling a sum of 7 is slightly more likely. As another example, points
near the peaks in a distribution of some discrete parameters correspond to fairly likely
conditions. See [20] for more discussions along these lines. As a further example, [9, 10]
argued that three generations of chiral matter is fairly likely in this sense in our E7
model (although, for example, zero generations may be more likely). (See also, e.g.,
[21].)

4. Natural choices: These are choices for discrete parameters having many possible values
that are not (obviously) preferred in any way, but imposing a particular chosen value does
not require exponential amounts of tuning. Such choices may hold at the level of, e.g.,
0.1% of the given set of vacua. Such choices may be needed for obtaining some qualitative
features of observed phenomenology in some constructions. For example, obtaining the
SM gauge group and matter representations from flux breaking of E7 involves some
choices of fluxes given by mild linear constraints, which seem to be natural in this sense,
although they do not seem to be preferred in any particular way over other choices that
would give a variety of other possible groups and representations.

5. Fine-tuning: These are choices involving setting one or more continuous variables to
take specific values, or making an exponentially rare choice among discrete possibilities.
Vacua based on such choices are increasingly non-generic in the landscape as the number
of such tunings increases. In some constructions of string vacua, such choices are needed
to obtain certain qualitative features of observed phenomenology. For example, a tuned
SU(5) or SM gauge group in F-theory involves extensively fine-tuning many moduli to
specific values [21] (unless these moduli are somehow automatically tuned by a specific
class of flux choices). Notably, it seems that no such fine-tuning is involved in our E7
models.

6. Technical choices: To facilitate analytic control of the vacua and make some particular
calculations manageable, in some cases technical choices are made by restricting atten-
tion to some specific relatively simple choices of vacua. These choices are not necessary
either for physical or phenomenological consistency, but are made to illustrate specific
examples as simply as possible. The features of the models chosen in this way should be
representative of some larger class of vacua or geometries. In some situations, technical
choices can be made just to simplify calculations that are in principle possible and ex-
pected to give qualitatively similar results for all other choices. In other cases, technical
choices are made where it is not clear how to do the computation explicitly in general,

–4–
 and/or whether a completely general choice will give qualitatively similar results. If
not, some choices or tuning of one of the above types may be implicitly involved. For
the specific technical choices made here, we have some confidence that similar results
should also hold for a broader class of vacua without those technical choice. Neverthe-
less, some qualitative simplifications occur based on these choices, thus more explicit
further studies are required to understand the extent to which these technical choices
are relevant for phenomenologically interesting features. Examples of technical choices
include picking some certain topological types for the compactification, which we do in
this paper (specifically by choosing models where the gauge divisor is a del Pezzo sur-
face and the matter curve is a P1 ) to facilitate and simplify the analysis; these technical
choices made here fall in the latter category above that may implicitly involve some more
or less natural choice or tuning, as they may affect qualitative aspects of the low-energy
physics.

While the term “natural” is used widely in many different ways in the literature, we
attempt to use the above classification to be slightly more precise about the types of choices
involved in the construction of our models and the realization of phenomenological features.
This is a coarse characterization, however, as choices and tunings can occur across a broad
spectrum, and we do not attempt to make any precise division between the gradations of
“common,” “fairly likely,” and “natural” conditions. In particular, we do not have a perfect
understanding of the class of string geometries or F-theory compactifications, so any attempt
at classification of this type is necessarily quite imperfect given the current state of knowledge.
Moreover, the measure problem on the landscape is not at all understood, so we really do not
have any good sense of the proper probability measure to use on the landscape. Nonetheless,
in the absence of any known or conjectured dynamical mechanism that would modify these
considerations, features that seem to require exponentially large amounts of fine-tuning under
a simple counting measure seem likely to occur less frequently in a large string multiverse
than features that are ubiquitous, fairly likely, or even natural in the preceding terminology.
In principle, even without solving the measure problem, this may give us some insight into
the extent to which the Standard Model may be realized naturally in string theory, and what
BSM physics may be most naturally associated with those SM structures.

1.2 Review of previous work

In recent years, F-theory has become a particularly promising framework for studying many
aspects of string compactifications and phenomenology, as it provides a global description of
a large connected class of supersymmetric string vacua. (See [7] for a review.) In particular,
F-theory gives 4D N = 1 supergravity models when compactified on elliptically fibered Calabi-
Yau (CY) fourfolds Y , corresponding to non-perturbative compactifications of type IIB string
theory on general (non-Ricci flat) complex Kähler threefold base manifolds B. The number
of such threefold geometries B seems to already be on the order of 103000 for toric bases B
[22, 17, 18], without even considering the exponential multiplicity of fluxes possible for each

–5–
geometry, although the number of flop equivalence classes of bases is somewhat smaller [19].
F-theory is also known to be dual to many other types of string compactifications such as
heterotic models. Briefly, F-theory is a strongly coupled version of type IIB string theory with
non-perturbative configurations of 7-branes balancing the curvature of the compactification
space. The non-perturbative brane physics is encoded geometrically into the elliptically fibered
manifold, which can be analyzed using powerful tools from algebraic geometry. The gauge
groups and chiral matter content supported on these branes can then be easily determined
when combined with flux data.
Applying the above techniques, many SM-like constructions of 4D F-theory models with
the gauge group GSM = SU(3) × SU(2) × U(1)/Z6 have been achieved in the literature. The
early literature, starting from [23–26], focused on the breaking of GUT groups of SU(5) and
its U(1) extensions [27–31], while there has also been some study of SO(10) [32] and E6 [33–
35] GUTs. (See [7, 8] for more extensive reviews.) These constructions break the GUT group
using the so-called hypercharge flux further discussed in [36, 37], which is a kind of “remainder”
flux [38, 39] to be reviewed below. Some later constructions tried to construct GSM directly
without any symmetry breaking, with the recent culmination of finding 1015 explicit solutions
of directly tuned GSM with three generations of SM chiral matter (a “quadrillion Standard
Models” [40]), based on the “F11 ” fiber in [41]. These constructions are further generalized in
[42, 43]. Although these models nicely capture some of the most important phenomenological
features, they face one common issue: In terms of the notions discussed in §1.1 the gauge
groups in these constructions are highly “fine-tuned”, namely they are obtained by setting
specific values for many complex structure moduli. Furthermore, on most F-theory bases such
tuning of GSM is forbidden due to the presence of rigid gauge groups (to be discussed shortly).
Even if the tuning is available, it may not be compatible with moduli stabilization by fluxes
and/or nonperturbative effects.
A more natural class of SM-like constructions in F-theory comes from rigid gauge groups
such as E7 , E6 [44, 45]. These are gauge groups enforced by strong curvature (to be more
precise, very negative normal bundle) on the base, and are present throughout the whole
branch of moduli space over that base, hence avoid the issue of tuning moduli. Moreover,
statistical studies on (toric) F-theory bases have suggested that these rigid gauge groups are
fairly common in the landscape. While the specific base naively associated with the most
flux vacua [1] does not contain E7 or E6 factors, these gauge factors arise in a substantial
fraction of F-theory base geometries enumerated by a simple counting measure (which may
or may not distinguish bases related by a flop). The fraction of toric bases for 6D F-theory
models that contain rigid E7 and E6 factors is more than 50% [46]. The statistics of E7
and E6 factors in threefold bases for 4D F-theory models is less well understood; one study
found E7 factors in ∼ 20% of a limited simple of bases [22], and a more detailed analysis of
the prevalence of such factors is currently underway [19]. Nonetheless, breaking these gauge
groups to GSM should give us a very large set of SM-like constructions. Recently in [9, 10], we
have proposed a general class of SM-like models using rigid E7 , E6 GUT groups in F-theory,
with an intermediate SU(5) group. These models enjoy the advantages of being natural and

–6–
involving little or no fine-tuning. Specifically, a combination of “vertical” and “remainder”
fluxes can be used to break the rigid gauge groups in a way that is not transparent in the low-
energy field theory, but gives the correct SM gauge group and some chiral matter. Although
in many cases the breaking leads to exotic chiral matter, there are large families of models
in which the correct SM chiral matter representations are obtained through an intermediate
SU(5). The number of generations can easily be small and we have demonstrated that three
generations are fairly likely in many of these models. In particular, a fully global explicit
construction of such an E6 model has been given in [10].

1.3 Overview of results

As discussed before, one might hope that there are string realizations of the Standard Model
in which most or all of the features observed in nature arise in a relatively natural way. In this
paper, we show that apart from the SM gauge group and chiral spectrum, several additional
important SM features can be easily obtained in the E7 (but not E6 ) models, with some
additional but mild tuning on the geometry and flux background. Specifically, we obtain
the following features, each of which depends upon making choices with various extents of
naturalness:

• As mentioned above, rigid E7 factors are quite common in the F-theory landscape, and
may be natural or likely depending on the proper vacuum measure. For generic (non-
toric) bases, the E7 gauge group can be broken down to GSM by some natural choices
of vertical and remainder fluxes.

• For the models with flux breaking of E7 → GSM , a set of approximate global U(1)
symmetries descend from the E7 , leading to exponential suppression of certain couplings.

• With the E7 → GSM gauge-breaking fluxes, it appears to be fairly likely to have 3

generations of SM chiral matter (although 0, 1, or 2 generations may be more likely),
and fairly likely that the exotic (3, 2)−5/6 representation is removed from the spectrum.

• Due to the use of E7 , there are always candidate Higgs sectors with a string theory
origin different from that of chiral matter. Such a structure automatically leads to
distinct dynamics between the Higgs and chiral matter, and gives rise to unsuppressed
SM Yukawa couplings.

• Under this setup, dimension-4 and 5 proton decay is ubiquitously suppressed to phe-
nomenologically safe levels.

The distinction between the Higgs and chiral matter, the appearance of the approximate
global U(1) symmetries, and the ubiquitous suppression of dimension-4 and 5 proton decay
are the strongest features of these constructions, in which desirable properties associated with
observed physics arise essentially automatically. Most of the remaining features we explore
generally require small amounts of discrete tuning. They may involve common, fairly likely

–7–
or natural choices and do not arise automatically, but do not seem to require extensive fine-
tuning.

• There is some automatic splitting between the doublet and triplet masses, although the
amount of splitting and the exact masses are unknown.

• Although there are extra charged vector-like exotics in the spectrum, the Yukawa cou-
plings between most of these fields (all besides the triplet Higgs) and the SM matter are
exponentially suppressed through the above-mentioned approximate symmetries. We
call these fields inert vector-like exotics.

• It is plausible that there is some hierarchy in the SM Yukawa couplings, but the exact
values are unknown.

• With the setup so far, the model contains three right-handed neutrinos with masses
lower than the string/GUT scale. It is plausible but not fully clear that the seesaw
mechanism occurs.

To facilitate the discussions and calculations in this paper, we technically choose the gauge
divisor to be a del Pezzo surface, and the matter curve to be a P1 . Although we expect similar
results for many other choices, these choices do lead to some qualitative simplifications in the
analysis, and further work is needed to determine whether low-energy models with similar
structure arise for a broader class of gauge divisors and matter curves, and/or to determine
how natural or fine-tuned these geometric choices may be.
As an example, we work out an explicit global construction of the E7 models that realizes
all of the above SM features. We emphasize that many of these phenomenological advantages
are specific for the E7 models, and may be (much) harder to realize in other types of SM-like
constructions in F-theory. Some of the above features are inherited from the group structure
of E7 itself, regardless of the string theory physics. To the authors’ knowledge, however,
these group theoretical features have not been noticed in the field theory literature, probably
because E7 itself does not support any chiral matter, if there is no additional input like fluxes
from the UV.
While the E7 models considered in this paper have quite a few phenomenological advan-
tages over some other stringy realizations of the Standard Model, we note that these models
potentially still suffer from the following issues, in light of which extra care must be taken
when interpreting the results presented here. First, these models contain many vector-like
exotics that cannot be removed by fluxes (except the most dangerous (3, 2)−5/6 , which is
fairly likely to be absent). In particular, these exotics include other copies of the Higgs field.
From the effective field theory perspective, we generically expect these exotics to get heavy
masses near the GUT/string scale such that they do not affect the low-energy phenomenol-
ogy. On the other hand, this expectation in general may not be true in string theory, and it
is important to develop further techniques to ensure the right masses. Although it may be
possible, we do not see any reason in these models why one of the Higgs doublet pairs should

–8–
get much lower masses than the other copies. In other words, there is no totally clear solution
to the µ-problem in our setup. Next, these E7 models have codimension-3 (4, 6) singularities
on the base, which correspond to an extra family of flux and may be associated to an extra
sector of strongly coupled superconformal and chiral matter [47–51]. We can easily control the
flux such that this sector is non-chiral, but since we understand very little about this sector,
further studies are needed to ensure that this sector does not affect phenomenology.

1.4 Outline of paper

This paper is organized as follows. In Section 2, we review fluxes in F-theory, which are
the central tools in our construction of E7 models. We first discuss the notion of vertical
and remainder fluxes, and the constraints satisfied by these fluxes. We then describe how
these fluxes break a nonabelian gauge group (known as “flux breaking”), determine the chiral
spectrum, and tell us something about the vector-like spectrum. Despite the difficulty of
computing the vector-like spectrum in general, the vector-like spectrum can be completely
determined in some special cases. We discuss how our E7 models easily fit into these cases,
so that we can fully compute the matter spectrum in our models.
To initiate our discussions on semi-realistic E7 GUTs in F-theory, we first review our
previous work on the E7 models [9, 10] in Section 3. We describe the geometry and fluxes
needed to get the SM gauge group and three generations of SM chiral matter from a rigid E7 .
In Section 4, we discuss various phenomenological aspects of the E7 models, namely the
vector-like matter, Yukawa couplings, proton decay, the Higgs sector, the neutrino sector, and
gauge coupling unification. One of our main tools is the Stückelberg mechanism [52–54] used
in our flux breaking of E7 to GSM , which leaves several approximate U(1) global symmetries.
We discuss these symmetries in detail and study how they constrain the couplings and mass
terms1 in the low-energy theory. These constraints, plus some additional tuning on the fluxes,
lead to many of the above phenomenological advantages, especially the ubiquitous suppression
of proton decay. We also discuss the vector-like matter that appears in the spectrum. We
demonstrate how to easily remove the most dangerous (3, 2)−5/6 vector-like exotic, and discuss
why most other vector-like exotics are inert. Although we cannot make any precise statements,
we discuss various possible origins of the vector-like (including the Higgs) masses. Based on
such discussions, we make some brief comments about the Higgs sector, the neutrino sector,
and gauge coupling unification.
After describing the recipe of getting semi-realistic E7 GUTs F-theory, in Section 5 we
write down an explicit global construction of a E7 model that achieves all the above phe-
nomenological features. This example demonstrates the fact that these features can indeed be
obtained through some mild tuning on the geometry and the fluxes, but without the necessity
of fine-tuning any moduli. Therefore, it is reasonable to regard these features as being natural
in the string landscape. To emphasize various advantages and disadvantages of the E7 mod-
els, in Section 6 we briefly compare our models with other SM-like F-theory constructions in
1
Throughout the paper, “mass terms” refer to the µ-term and other similar terms in the superpotential,
which involve two different fields.

–9–
the literature. We finally conclude in Section 7. In Appendices A and B, we discuss several
technical tools that are useful in the construction in Section 5.

2 Fluxes in 4D F-theory models

In this section, we review vertical and remainder fluxes in 4D F-theory models, and how these
fluxes determine the gauge group and matter spectrum. Except in §2.4 on vector-like matter,
all the content in this section has been discussed in depth in [10]. Here we only recap the
essential facts for our construction of E7 models and set up the notation. Interested readers
can refer to [10] for more background information.

2.1 Vertical and remainder fluxes

To describe the flux backgrounds, we first need some basic geometric facts about the com-
pactifications. As mentioned in §1, we consider F-theory compactified on a CY fourfold Y ,
which is an elliptic fibration on a threefold base B. Nonabelian gauge groups arise when
sufficiently high degrees of singularities are developed in the elliptic fibers over divisors on B
(denoted by Dα ), called gauge divisors Σ. When this happens, Y itself is also singular and we
need to consider its resolution Ŷ such that we can study cohomology and intersection theory.
Let the total gauge group be G, where G has no U(1) factors before flux breaking. In the
main part of this paper, we always study the simple case where G = E7 , but for generality
we assume any simple Lie group G in this section. The nonabelian group G is supported on
a gauge divisor Σ, and the resolution results in exceptional divisors D1≤i≤rank(G) in Ŷ . For
ADE (or simply-laced) groups, the intersection structure of these divisors matches the Dynkin
diagram of G, where each exceptional divisor corresponds to a Dynkin node [55, 56]. By the
Shioda-Tate-Wazir theorem [57, 58], the divisors DI on Ŷ are spanned by the zero section D0
of the elliptic fibration, pullbacks of base divisors π ∗ Dα (which we also call Dα depending on
context), and the exceptional divisors Di .2 Although the choice of resolution is not unique,
our analysis and results are clearly resolution-independent [50].
Now we are ready to understand fluxes. These are most easily understood by considering
the dual M-theory picture of the F-theory models, that is, M-theory compactified on the
resolved fourfold Ŷ (reviewed in [7]). In the M-theory perspective, fluxes are characterized by
the three-form potential C3 and its field strength G4 = dC3 . The data of G4 flux, which can
be studied with well-established tools, is sufficient for constructing our E7 models.
In general, G4 is a discrete flux that takes values in the fourth cohomology H 4 (Ŷ , R).
The quantization condition on G4 is slightly subtle and is given by [59]
1
G4 + c2 (Ŷ ) ∈ H 4 (Ŷ , Z) , (2.1)
2
where c2 (Ŷ ) is the second Chern class of Ŷ . In general (and particularly for E7 models),
c2 (Ŷ ) can be odd (i.e., non-even), in which case the discrete quantization of G4 contains a
2
If G has U(1) factors, there are also divisors associated with these factors coming from the Mordell-Weil
group of rational sections with nonzero rank.

– 10 –
half-integer shift. When we construct an E7 model explicitly in §5, we will make use of an
odd c2 (Ŷ ) and half-integer fluxes. More details will be discussed in that section.
Next, to preserve the minimal amount of SUSY in 4D, G4 must live in the middle co-
homology i.e. G4 ∈ H 2,2 (Ŷ , R) ∩ H 4 (Ŷ , Z/2). Supersymmetry also imposes the condition of
primitivity [60, 61]:
J ∧ G4 = 0 , (2.2)

where J is the Kähler form of Ŷ . This is automatically satisfied when the geometric gauge
group is not broken, but when the gauge group is broken by vertical flux (to be discussed
below), this condition stabilizes some (but not all) Kähler moduli; stabilizing these moduli
within the Kähler cone imposes additional flux constraints.
We also have the D3-tadpole condition [62] that must be satisfied for a consistent vacuum
solution: Z
χ(Ŷ ) 1
− G4 ∧ G4 = ND3 ∈ Z≥0 , (2.3)
24 2 Ŷ
where χ(Ŷ ) is the Euler characteristic of Ŷ , and ND3 is the number of D3-branes, or M2-
branes in the dual M-theory. To preserve SUSY and stability, we forbid the presence of anti-
D3-branes i.e. ND3 ≥ 0. The integrality of ND3 is guaranteed by Eq. (2.1). This condition
has an immediate consequence on the sizes of fluxes. Since in general h2,2 > 2χ/3 ≫ χ/24, if
we randomly turn on flux in the whole middle cohomology such that the tadpole constraint
is satisfied, a generic flux configuration vanishes or has small magnitude in most of the h2,2
independent directions. In this sense, the tadpole contributed by fluxes along some particular
directions can be treated as a rough estimate on the amount of fine-tuning on fluxes. We leave
a more precise and detailed analysis of these considerations to future work.
We now consider the orthogonal decomposition of the middle cohomology [39]:
2,2
H 4 (Ŷ , C) = Hhor
4
(Ŷ , C) ⊕ Hvert 2,2
(Ŷ , C) ⊕ Hrem (Ŷ , C) . (2.4)

The horizontal subspace comes from the complex structure variation of the holomorphic 4-
form Ω, while the vertical subspace is spanned by products of harmonic (1, 1)-forms (which
are Poincaré dual to divisors, denoted by [DI ])
 
2,2
Hvert (Ŷ , C) = span H 1,1 (Ŷ , C) ∧ H 1,1 (Ŷ , C) . (2.5)

Components that do not belong to the horizontal or vertical subspaces are referred to as
remainder flux. In the construction here we use a specific type of remainder flux. Consider a
curve Crem ∈ H1,1 (Σ, Z) in Σ, such that its pushforward ι∗ Crem ∈ H1,1 (B, Z) is trivial, where
ι : Σ → B is the inclusion map. While such a curve cannot be realized on toric divisors on
toric bases, it has been suggested that such curves do exist on “typical” bases [39], so that
toric geometry may be insufficiently generic for this class of constructions; understanding this
question of typicality is an important problem for further study. In any case, we now restrict
each Di (considered as a fibration over Σ) onto Crem , giving a surface in Ŷ . Its Poincaré dual

– 11 –
(in Ŷ ) [Di |Crem ] is a (2, 2)-form, but since Crem cannot be obtained by intersections of base
divisors, we must have
2,2
[Di |Crem ] ∈ Hrem (Ŷ , C) . (2.6)

Here we explain more about vertical flux (denoted by Gvert 4 ), as there are more constraints
specifically on vertical flux such that G4 dualizes to a consistent F-theory background that
preserves Poincaré invariance. Combining Eqs. (2.5) and (2.1) gives the integral vertical
2,2
subspace Hvert (Ŷ , R) ∩ H 4 (Ŷ , Z).3 We focus primarily here on the vertical subspace spanned
by integer multiples of forms [DI ] ∧ [DJ ]
 
2,2
Hvert (Ŷ , Z) := spanZ H 1,1 (Ŷ , Z) ∧ H 1,1 (Ŷ , Z) . (2.7)

While this subspace does not necessarily include all lattice points in the full vertical coho-
2,2
mology Hvert (Ŷ , C) ∩ H 4 (Ŷ , Z) of the same dimension, this subspace is sufficient for us to
construct the E7 models, and we leave the analysis of the full vertical subspace to future
work.
Now we set up some notations for vertical flux. We expand

Gvert
4 = ϕIJ [DI ] ∧ [DJ ] , (2.8)

and work with integer (or half-integer if c2 is odd) flux parameters ϕIJ . We denote the
integrated flux as [63] Z
ΘΛΓ = G4 ∧ [Λ] ∧ [Γ] , (2.9)
Ŷ

where Λ, Γ are arbitrary linear combinations of DI ; subscripts 0, i, α, . . . refer to the basis

divisors D0 , Di , Dα , . . .. In this paper, we use the following resolution-independent formula to
relate Θiα to ϕiα [64, 50]:4
Θiα = −κij Σ · Dα · Dβ ϕjβ , (2.10)

where κij is the inverse Killing metric of G, and “dots” denote the intersection product. In the
E7 models, the only vertical flux parameters we turn on have indices of type ϕiα ; although it
is also possible to turn on nontrivial ϕij , we always turn them off for reasons to be discussed
below.
Now we write down the extra constraints for vertical flux. To preserve Poincaré symmetry
after dualizing, we require [65]
Θ0α = Θαβ = 0 . (2.11)

If the whole geometric gauge symmetry is preserved, a necessary condition is that

Θiα = 0 , (2.12)
3
We remind readers that the quantity Gvert
4 + c2 (Ŷ )/2 instead of Gvert
4 lives in this subspace. Note that
c2 (Ŷ ) is always vertical.
4
Indices appearing twice are summed over, while other summations are indicated explicitly.

– 12 –
for all i, α, otherwise flux breaking occurs. This condition is not sufficient when there is
nontrivial remainder flux, which will be discussed more in §2.2. When flux breaking occurs,
i.e. Eq. (2.12) is violated, the condition (2.11) for Poincaré symmetry is unchanged, but there
are extra constraints from primitivity, which are demonstrated in later sections.
Much of the above discussion on vertical flux extends naturally to the type of remainder
flux Grem
4 we need. Similarly, we expand

Grem
4 = ϕir [Di |Crem ] , (2.13)

and work with integer ϕir . In this paper, we always turn on remainder flux with a single Crem
only, so we do not specify the choice of Crem in the flux parameters; instead we just label
them by “r”. Eq. (2.10) straightforwardly generalizes to remainder flux by replacing the triple
intersection on the base with the intersection of two Crem ’s on Σ.

2.2 Flux breaking

With the knowledge of vertical and remainder fluxes, we now describe the breaking of geo-
metric gauge groups with these fluxes, a.k.a. flux breaking. This kind of breaking has been
used as early as [24] (see also [7]), and is recently developed in depth in [10]. In this paper,
we only list the results essential for our analysis on the E7 models, and we refer readers to
[10] for full technical details.
Let us first study vertical flux. Recall that we need Θiα = 0 for all i, α to preserve the
whole G. Now we break G into a smaller group G′ by turning on some nonzero ϕiα . Such flux
breaks some of the roots in G. It also induces masses for some Cartan gauge bosons by the
Stückelberg mechanism [53, 54], hence breaks some combinations of Cartan U(1)’s in G. Let
αi be the simple roots of G, and Ti be the Cartan generators associated with αi i.e. in the
co-root basis. The root bi αi is preserved under the breaking if
X
bi ⟨αi , αi ⟩ Θiα = 0 , (2.14)
i

for all α. Here ⟨., .⟩ denotes the inner product of root vectors. Moreover, the corresponding
linear combination of Cartan generators
X
bi ⟨αi , αi ⟩ Ti , (2.15)
i

is preserved. These generators form a nonabelian gauge group G′ ⊂ G after breaking. Note
that for ADE groups like E7 , ⟨αi , αi ⟩ is the same for all i, and we will drop this factor in the
above conditions. Below we will use a simple version of the breaking: we turn on Θi′ α ̸= 0 for
some set of Dynkin indices i′ ∈ I ′ and some α, in a generic way such that Eq. (2.14) is satisfied
only when bi′ = 0 for all i′ ∈ I ′ . Then G′ is given by removing the corresponding nodes in the
Dynkin diagram of G. The simple roots of G′ are directly descended from G and are given by
/ ′ . In the E7 models, one can show that all flux breaking routes to GSM (or SU(5) before
αi∈I
including remainder flux) are related to this simple version of breaking by automorphisms.

– 13 –
 There are additional constraints on vertical flux breaking coming from primitivity, since
Eq. (2.2) is not automatically satisfied when there is vertical flux breaking and Θiα ̸= 0 for
some i, α. To understand these constraints, we consider the F-theory limit where the fibers
shrink to zero volume. Hence we can expand J of Ŷ as J → π ∗ JB = tα [Dα ], where the Kähler
moduli tα are restricted to the positive Kähler cone. Now primitivity requires that
Z
[Di ] ∧ J ∧ G4 = tα Θiα = 0 , (2.16)
Ŷ

which is true only for specific choices of J when there is vertical flux breaking. The condition
of primitivity then stabilizes some but not all Kähler moduli in J; consistently stabilizing
these moduli within the Kähler cone imposes additional flux constraints.
As discussed in [10], the above flux constraints lead to an important necessary condition for
consistent vertical flux breaking: Let r be the number of linearly independent Dα ’s appearing
in the set of all homologically independent surfaces in the form of Siα = Di · Dα (for any i of
the given G). Now consider the (r × rank(G)) matrix Θ(αa )(i) (where a and i are the indices
for rows and columns respectively). The condition (2.16) asserts that tα Θαi = 0. Since the
solution to primitivity thus requires a nontrivial left null space of the matrix, the rank of the
matrix is at most r − 1. Moreover from Eq. (2.14), the rank of the matrix is also the change
in rank of G during flux breaking. Therefore, we require

r ≥ rank(G) − rank(G′ ) + 1 . (2.17)

Note that when remainder flux breaking is not available, and all divisors in Σ descend from
intersections in B, we have r = h1,1 (Σ). In the E7 models, however, we require the presence
of remainder flux and r is smaller than h1,1 (Σ). This condition means that we must have a
sufficiently large r in order to get a desired G′ , hence imposing constraints on the possible
geometries that support a given vertical flux breaking.
So far we have focused on the nonabelian part of the broken gauge group, but there may
also be U(1) factors in the broken gauge group like GSM . In the formalism of flux breaking,
there are two ways to get U(1) factors: for U(1) factors not along any roots, we can use vertical
flux to get these factors using the Stückelberg mechanism; for a recent application see [66].
On the other hand, U(1) factors like the hypercharge in GSM are along some roots of a higher
gauge group (SU(5) in the case of hypercharge), and remainder flux is necessary for obtaining
these factors. Therefore, we proceed as follows: If we turn on

Grem
4 = ϕir [Di |Crem ] , (2.18)

for some Crem satisfying the property mentioned in §2.1, G is broken into the commutant of
T = ϕir Ti within G. The difference is that the remainder flux does not turn on any Θiα , so
there is no Stückelberg mechanism and all the U(1) factors in the commutant are preserved.
In other words, breaking using remainder flux never decreases the rank of the gauge group,
while breaking using vertical flux always decreases the rank. Note that when G is a rigid

– 14 –
gauge group, Σ is a rigid divisor and supports remainder flux breaking only when embedded
into a non-toric base. This follows because for a toric base B, toric divisors span the cone of
effective divisors, so any rigid effective divisor Σ is toric, and toric curves in a toric Σ span
h1,1 (Σ).

2.3 Chiral matter

It is well known that vertical flux induces some chiral matter, and the same is true for vertical
flux that breaks the gauge group. Perhaps more strikingly, even if the unbroken gauge group
G does not have any complex representations and does not support chiral matter, there may
still be some chiral matter after vertical flux breaking [10]. The famous index formula states
that for a weight β in representation R, its chiral index χβ is [67, 68, 30, 63]
Z
χβ = Gvert
4 , (2.19)
S(β)

where S(β) is called the matter surface of β. When R is localized on a matter curve CR , S(β)
is the fibration of the blowup P1 corresponding to β over CR . Here we recall that a matter
curve is a curve on Σ where the fiber singularity is enhanced, resulting in additional fibral
curves in the resolution, which corresponds to matter multiplets in the 4D theory.
Since weights differ by roots, given a weight β in R of G, it is useful to expand β = −bi αi .
Hence we can decompose its matter surface S(β) as [7]

S(β) = S0 (R) + bi Di |CR , (2.20)

where S0 only depends on R but not β. S0 (R) corresponds to the flux that gives chiral matter
without breaking G, when G supports chiral matter. In contrast, when G itself does not
support chiral matter as in the E7 models, S0 (R) is trivial in homology. From now on we will
ignore S0 (R) and focus on the second term of Eq. (2.20). Matter curves in general can be
written as CR = Σ · DR for some divisor DR . Then,
Z
Gvert
4 = bi ΘiDR . (2.21)
S(β)

We can replace the i summation with i′ ∈ I ′ since the other terms vanish. When G is broken to
G′ , R decomposes into different irreducible representations R′ in G′ , which can be labelled by
bi′ . In general, different bi′ and different R can give rise to the same irreducible representation
R′ . The total chiral index χR′ for R′ is then
XX
χR ′ = bi′ Θi′ DR . (2.22)
R bi′

Note that this expression is nontrivial even when R is not complex.

There are also adjoint chiral multiplets (apart from the vector multiplet of the gauge
fields) living on the bulk of Σ, and matter curves or surfaces for this representation are not

– 15 –
well-defined. Nevertheless, it has been shown that adjoint matter can also become chiral after
flux breaking, and the chiral indices are given by setting S0 (Adj) = 0 and replacing CR by
KΣ [69]. Here K denotes the canonical class. By the adjunction formula, KΣ = Σ · (KB + Σ)
and we should set DR = KB + Σ.

2.4 Vector-like matter

To fully understand the phenomenology of these models, it is important to study the vector-
like spectrum in addition to the chiral spectrum. One of the main reasons for this is that
in a realistic model we need to realize the Higgs sector, while avoiding dangerous vector-
like exotics. While the techniques for computing the chiral spectrum are already at hand as
above, computing the vector-like spectrum in general requires not only the G4 flux, but the
full information of C3 in terms of line bundles on Σ and C.5 In many cases, these things are
hard to compute and some of the relevant technology has only been developed fairly recently
[11–14]. Fortunately, our models admit several important simplifications such that the G4
flux itself already determines the vector-like spectrum.
Let us first focus on vector-like matter that lives on the bulk of Σ. We follow the formalism
in [70, 69]. At least in most cases, the full C3 can be captured by an algebraic complex 2-
cycle A in the Chow group CH2 (Ŷ ) (algebraic cycles modulo rational equivalence instead of
homological equivalence), with homology class is [G4 ] [67]. We consider the restriction of A
onto a Cartan divisor Di , given by the intersection product A · Di ∈ CH2 (Di ). Its projection
onto Σ, given by π∗ (A · Di ) ∈ CH1 (Σ), is a curve on Σ associated with the line bundle

Li = OΣ (π∗ (A · Di )) . (2.23)

Now for each weight β = −bi αi of the adjoint, we define the line bundle

Lβ = ⊗i Lbi i . (2.24)

Then the chiral and anti-chiral multiplicities for β are counted by the following sheaf coho-
mologies [23, 24]

chiral : H 0 (Σ, Lβ ⊗ KΣ ) ⊕ H 1 (Σ, Lβ ) ⊕ H 2 (Σ, Lβ ⊗ KΣ ) , (2.25)

anti − chiral : 0 1 2
H (Σ, Lβ ) ⊕ H (Σ, Lβ ⊗ KΣ ) ⊕ H (Σ, Lβ ) . (2.26)

Notice that the sheaf cohomologies for chiral and anti-chiral matter are related by Serre
duality. To calculate their dimensions, we apply the following two simplifications [24]. First,
for fluxes with nontrivial Lβ and satisfying primitivity (i.e. preserving SUSY), we have
H 0 (Σ, Lβ ) = H 2 (Σ, Lβ ⊗ KΣ ) = 0. Next, we assume that Σ is a rational surface with
5
In general, these are described by sheaves when there are more severe singularities on Σ and/or C. In
this paper, we only consider completely smooth geometry on the bases, so the description by line bundles is
sufficient.

– 16 –
effective −KΣ ,6 in this case, we have H 2 (Σ, Lβ ) = H 0 (Σ, Lβ ⊗ KΣ ) = 0. Therefore, the
exact multiplicities nβ and n−β are fully determined by h1 (Σ, Lβ ) and h1 (Σ, Lβ ⊗ KΣ ) re-
spectively. Since only H 1 is nontrivial, the multiplicities are also captured by the topological
Euler characteristics χ (Σ, Lβ ) and χ (Σ, Lβ ⊗ KΣ ). These are fully determined by c1 (Lβ ),
given by the Hirzebruch-Riemann-Roch theorem: [23, 24, 71]

nβ = −χ (Σ, Lβ )
1 1
= [c1 (Lβ )] · KΣ − [c1 (Lβ )]2 − 1
2 2
1 1 2
= χβ − [c1 (Lβ )] − 1 , (2.27)
2 2
where χβ is the chiral index for β given in §2.2, and the Poincaré dual is taken with respect to
Σ. The expression for n−β is the same except that the sign of the first term is flipped; indeed
we get back nβ − n−β = χβ . In our models where only gauge-breaking fluxes are turned on,
we can read off c1 (Lβ ) from Eq. (2.10):

[c1 (Lβ )] = −bi κij (ϕjα Σ · Dα + ϕjr Crem ) . (2.28)

Combining Eqs. (2.27) and (2.28), this gives us a formula to compute the exact matter
multiplicities from the bulk of Σ in terms of the vertical and remainder flux parameters.
Now we turn to vector-like matter localized on matter curves. Similarly for a weight
β ∈ R supported on CR , we consider the pullback of A onto a matter surface S(β) given by
A · S(β) ∈ CH2 (S(β)). Its projection onto CR , given by π∗ (A · S(β)) ∈ CH1 (CR ) defines a
line bundle for β:
Lβ = OCR (π∗ (A · S(β))) . (2.29)
Then the chiral and anti-chiral multiplicities for β are counted by the following sheaf coho-
mologies
 
chiral : H 0 CR , Lβ ⊗ KCR , (2.30)
p
 
anti − chiral : H 1 CR , Lβ ⊗ KCR , (2.31)
p

where KCR is the spin bundle on CR . These sheaf cohomologies are more subtle than those
p

for the bulk of Σ. While they are well understood when the matter curve has genus 0 or 1,
for irreducible curves with higher genus, these cohomologies have complicated dependence on
moduli, and their dimensions can jump at special points in the moduli space. For reducible
curves, there can also be vector-like pairs between different irreducible components of the
curves, if the total chiral index is split into different components accordingly. Instead of
6
As shown below, the condition that −KΣ is effective is a reasonable simplifying assumption in the context
of rigid gauge groups. All toric surfaces are rational and have effective −KΣ . We assume these conditions
on Σ in the rest of the paper; in much of the paper we restrict attention to the special case where Σ is a
(generally non-toric) del Pezzo surface. Further work would be needed to understand the detailed structure of
the resulting models when these technical conditions are relaxed.

– 17 –
running into all these subtleties, below we just focus on a special case where the matter curve
is simply a P1 . In this case, there can never be any vector-like pairs from the matter curve,
since for P1 only one of the H 0 , H 1 is nontrivial, depending on the sign of the line bundle.
This means in particular that the vector-like spectrum is independent of the choice of spin
bundle, significantly simplifying the analysis. As shown below, this geometry is not hard to
achieve in our E7 models, and we leave the generalizations to more complicated matter curves
in future work.
So far, we have discussed the vector-like multiplicities for each weight separately. On the
other hand, we recall that weights from different bi′ and R can contribute to the same chiral R′
in Eq. (2.22) during flux breaking. Similarly, these different weights can form vector-like pairs
after flux breaking, even if each weight is purely chiral. This effect can occur on both matter
curves and the bulk of Σ. Such vector-like matter has qualitatively different behavior from
that obtained from sheaf cohomologies, and has interesting phenomenological implications.
More details will be discussed in later sections.

3 Review of E7 GUTs in F-theory

With the above background knowledge, now we are ready to describe the E7 models. For
completeness, first we briefly review our previous work [9, 10] on E7 models, namely describing
how the SM gauge group and chiral spectrum can be realized in a natural way through flux
breaking of a rigid E7 factor. We refer readers to those two papers for more details.

3.1 Flux breaking of rigid E7 factors

Recall that gauge groups in F-theory arise from sufficiently high degrees of fiber singularities on
a gauge divisor Σ. For an E7 gauge group, the (singular) elliptic CY fourfold Y is described
by a certain form of Weierstrass model [55, 56, 72]. Treating the elliptic curve as the CY
hypersurface in P2,3,1 with homogeneous coordinates [x : y : z], Y is given by the locus of

y 2 = x3 + s3 f3 xz 4 + s5 g5 z 6 , (3.1)

where s, f3 , g5 are sections of line bundles O(Σ), O(−4KB − 3Σ), O(−6KB − 5Σ) on the base
B, and the gauge divisor Σ supporting the E7 factor is given by s = 0. There is adjoint matter
133 arising from excitations localized around the bulk of Σ. There is also fundamental matter
56 localized on the curve s = f3 = 0, or C56 = −Σ · (4KB + 3Σ) in terms of the intersection
product, when the curve is nontrivial in homology. When Σ has a sufficiently negative normal
bundle NΣ , singularities of the elliptic fibration are enforced on Σ, and the Weierstrass model
for Y is automatically restricted to the form (3.1). A rigid E7 is then realized on Σ. To be
precise, we can consider the following divisors on Σ (not on B) [45]:

Fk = −4KΣ + (4 − k)NΣ ,
Gl = −6KΣ + (6 − l)NΣ , (3.2)

– 18 –
Figure 1. The Dynkin diagram of E7 . The Dynkin node labelled i corresponds to the exceptional
divisor Di . The solid nodes are the ones we break to get the Standard Model gauge group and chiral
matter. Node 3 (in gray) is broken by remainder flux while the others are broken by vertical flux.

where k, l are integers. Then there is a rigid E7 on Σ if Fk , Gl are effective for k ≥ 3, l ≥ 5

only. A simple way to satisfy this condition is to consider effective −KΣ and −NΣ , such that
−3KΣ + NΣ is not effective but −4KΣ + NΣ is effective. As discussed previously, the natural
choice of effective −KΣ matches nicely with the simplification we made in §2.4 for computing
vector-like spectrum, and we assume that this condition holds.
After setting up the geometry, we now turn on the flux background. We break E7 to GSM
in two steps. Since remainder flux preserves U(1)’s along the roots but vertical flux does not,
we first break E7 to SU(5) with vertical flux, then break SU(5) to GSM with remainder flux.
The latter flux is very similar to the hypercharge flux in traditional SU(5) GUTs in F-theory.
To perform the first step of breaking, we turn on nonzero Θi′ α for some α and i′ = 4, 5, 6
subject to the flux constraints listed in §2.1, see Figure 1. In terms of flux parameters ϕiα , we
turn on
ϕ1α = 2nα , ϕ2α = 4nα , ϕ3α = 6nα , ϕ4α = 5nα , ϕ7α = 3nα . (3.3)

The values of ϕ5α , ϕ6α , if sufficiently generic, do not affect the gauge group, but they will
be fixed by other flux and phenomenological constraints. Here we define a new set of flux
parameters nα , which can be integers or half-integers depending on the parity of c2 (Ŷ ). At
this point the gauge group has been broken to SU(5). To perform the second step of breaking,
we similarly turn on the remainder flux

ϕ1r = 2nr , ϕ2r = 4nr , ϕ3r = 6nr , ϕ7r = 3nr , (3.4)

and ϕ4r , ϕ5r , ϕ6r plays the same role as ϕ5α , ϕ6α . Here nr is always integer. Under the
construction of rigid E7 , we require a non-toric base B to ensure the existence of Crem , hence
this remainder flux. After the remainder flux breaking, the remaining unbroken gauge group
is
GSM = SU(3) × SU(2) × U(1)/Z6 . (3.5)

3.2 Chiral spectrum in flux-broken E7 models

It is straightforward to calculate the chiral spectrum given the above fluxes. Since only the
vertical flux induces chiral matter, we can analyze the matter content by breaking E7 →

– 19 –
SU (5), where the 56 breaks into a combination of 5, 10, uncharged singlets and conjugate
representations, and 133 includes these as well as the adjoint 24. Since the adjoint of SU(5)
is non-chiral, the only chiral representations we expect for GSM after the whole breaking are
the Standard Model representations, which descend from the 5, 10 of SU(5),

Q = (3, 2)1/6 , Ū = (3̄, 1)−2/3 , D̄ = (3̄, 1)1/3 , L = (1, 2)−1/2 , Ē = (1, 1)1 . (3.6)

Using Eq. (2.22), we indeed get the anomaly-free combination of SM chiral matter from vertical
flux. It will be useful to separate the contributions from 56 and 133 to the total chiral index,
i.e. χ(3,2)1/6 = χ56
(3,2)1/6 + χ(3,2)1/6 , where each contribution is anomaly-free by itself. The
133

fundamental 56 gives
χ56
(3,2) = Σ · (4KB + 3Σ) · Dα nα , (3.7)
1/6

and the adjoint 133 gives

χ133
(3,2) = 2Σ · (KB + Σ) · Dα nα . (3.8)
1/6

Note that the total chiral indices only depend on nα but not ϕ5α , ϕ6α . This is no longer true
when we look at the chiral indices for each weight in the phenomenological analysis below. An
important feature of these chiral indices is that they have a linear Diophantine structure in
the quantized flux parameters, with the coefficients not being very large. If we randomly pick
some small values of nα (bounded by the tadpole constraint (2.3)), generically different terms
in the chiral indices will cancel each other, resulting in small chiral indices. Therefore, small
chiral indices are preferred in these models, and it is not hard to achieve three generations of
SM chiral matter.
In most cases, the above Weierstrass model also has codimension-3 singularities at the
locus s = f3 = g5 = 0. Traditionally, codimension-3 singularities are interpreted as Yukawa
couplings in the low-energy theory. In E7 models, however, these are so-called non-minimal
singularities (with degrees (4, 6) or higher) where the fiber becomes non-flat, i.e. its dimension
jumps. Such singularities can no longer be interpreted as Yukawa couplings;7 this is also
manifest by noticing that 563 does not contain any singlets, hence cannot form any gauge-
invariant couplings. Instead, there is an extra family of vertical flux associating to the non-flat
fiber with nontrivial ϕij components [50]. Analogous to codimension-2 (4, 6) singularities in
6D F-theory models [73, 74], there has been evidence that this flux switches on an extra sector
of strongly coupled superconformal chiral matter, given by M2-branes wrapping curves on the
non-flat fiber [51]. For our phenomenological purpose, we can always set this flux to zero i.e.
ϕij = 0 for all i, j, such that the extra sector becomes non-chiral and probably does not affect
the Standard Model sector. We should warn readers, however, that without further studies
on these extra sectors, we cannot precisely rule out the possibility that these sectors ruin the
desired phenomenology.
7
It was pointed out in [49] that these singularities may give rise to quartic or higher order couplings.
Nevertheless, the singularities in [49] have an unusual local geometry where a curve intersects another curve
three times. We do not see any such intersections or any evidence of such higher order couplings in our models.

– 20 –
4 Phenomenology of E7 GUTs

So far, we have studied the gauge group and the chiral spectrum in the above class of E7
models. In this section, we start to analyze the phenomenological aspects of these models in
more detail. The presence of approximate global symmetries descending from the underlying
E7 group suppresses certain couplings, with significant implications for phenomenology of
these models; in particular, we show that proton decay is automatically suppressed. We focus
further on the Higgs sector and the interactions in these models. We also discuss vector-like
exotics in these models and the circumstances under which they can be phenomenologically
safe. We consider the extent to which the various features of the E7 flux-broken models
possibly, or even naturally, match with observed phenomenology. The analysis in this section,
together with an explicit construction of an example model in §5, are the main results of this
paper. These results provide evidence that natural and realistic E7 GUTs can be realized in
F-theory. On the other hand, due to limits on existing technologies for computing detailed
aspects of F-theory models (such as the specific values of couplings), most of the analysis in
this section is purely qualitative.

4.1 Approximate global symmetries

Our starting point is based on considering approximate U(1) global symmetries that arise in
the E7 flux-broken models. These approximate symmetries directly originate from vertical flux
breaking, and control the structure of interactions and mass terms in the low-energy theory,
as well as leading to many of the phenomenologically attractive features of these models such
as suppression of proton decay. Recall that during vertical flux breaking from E7 to GSM ,
the Cartan gauge bosons along the generators T4 , T5 , T6 get masses from the Stückelberg
mechanism [52–54]. These masses explicitly break the corresponding Cartan U (1)’s of the
nonabelian gauge symmetry, but the matter interactions descending from the unbroken gauge
group still respect the global parts of the broken U (1) gauge symmetries (at least for the
symmetries without mixed anomalies with the remaining gauge group). These U (1) global
symmetries are broken only by D3/M5-instanton effects, which are exponentially suppressed in
Kähler moduli [75–78]. In the low-energy theory, these effects turn on exponentially suppressed
mass terms and interactions that violate the global symmetries. Since these effects are small,
the symmetries still remain as approximate global symmetries in the theory. This scenario
is consistent with the No Global Symmetries Conjecture [79, 80]. Note that without more
details of the model, we cannot quantitatively specify the extent to which a certain quantity
is suppressed by these symmetries. Throughout this paper, we only adopt the qualitative
picture of exponential suppression, and leave efforts towards explicit calculations of these
quantities for future work.
To study the implications of these U (1) symmetries, it is important to understand the
branching rules from E7 to GSM in the presence of these additional U (1) charges. Below
we use the basis (Y, b4 , b5 , b6 ) for the U (1) charges, where Y is the SM hypercharge.8 The
8
One direction of the additional U(1) symmetries actually has mixed anomalies with the SM gauge group

– 21 –
branching rules are

56 → (1, 1)0,5/2,2,3/2 + (1, 1)0,5/2,2,1/2 + (1, 1)0,5/2,1,1/2 + (1, 1)1,3/2,1,1/2

+ (3, 2)1/6,3/2,1,1/2 + (3̄, 1)−2/3,3/2,1,1/2 + (3̄, 1)1/3,1/2,1,1/2 + (3̄, 1)1/3,1/2,0,1/2
+ (3̄, 1)1/3,1/2,0,−1/2 + (1, 2)−1/2,1/2,1,1/2 + (1, 2)−1/2,1/2,0,1/2 + (1, 2)−1/2,1/2,0,−1/2
+ conjugates , (4.1)

133 → (8, 1)0,0,0,0 + (1, 3)0,0,0,0 + 4 × (1, 1)0,0,0,0

h
+ (1, 1)0,0,0,1 + (1, 1)0,0,1,0 + (1, 1)0,0,1,1 + (1, 1)1,−1,0,0 + (1, 1)1,−1,−1,0
+ (1, 1)1,−1,−1,−1 + (3, 2)−5/6,0,0,0 + (3, 2)1/6,−1,0,0 + (3, 2)1/6,−1,−1,0 + (3, 2)1/6,−1,−1,−1
+ (3̄, 1)−2/3,−1,0,0 + (3̄, 1)−2/3,−1,−1,0 + (3̄, 1)−2/3,−1,−1,−1 + (3̄, 1)1/3,−2,−1,0
+ (3̄, 1)1/3,−2,−1,−1 + (3̄, 1)1/3,−2,−2,−1 + (3̄, 1)1/3,3,2,1 + (1, 2)−1/2,−2,−1,0
i
+ (1, 2)−1/2,−2,−1,−1 + (1, 2)−1/2,−2,−2,−1 + (1, 2)−1/2,3,2,1 + conjugates . (4.2)

It is then straightforward to apply the rule that only terms with all net U (1) charges vanishing
are unsuppressed in the superpotential of the low-energy theory. Notice that there are three
copies of D̄ = (3̄, 1)1/3 and L = (1, 2)−1/2 (or the SU(5) fundamental before remainder flux
breaking) in each of the decompositions 56, 133, which are distinguished by having different
U (1) charges. Without further information or inputs, three families of SM chiral matter arising
in a given model in general may be distributed within the three copies. As shown below, such
a distribution may lead to phenomenological inconsistencies, and some extra tuning must be
done to avoid those issues.

4.2 Vector-like exotics

As in all GUT models, the E7 models face the issue of having (many) vector-like exotics that
have not been observed in experiments. Although we have chosen a P1 matter curve to ensure
that there are no vector-like pairs on the matter curve, generically there are many vector-like
matter fields on the bulk of Σ. Therefore, all representations in Eq. (4.2), except the SM
adjoint representations, generically have nontrivial vector-like matter multiplicities. Interest-
ingly, these bulk vector-like fields involve the usual MSSM Higgs fields Hu , Hd , which indeed
play the role of a SM Higgs sector in the discussion below. This feature does not seem to hap-
pen in models with smaller GUT groups. On the other hand, there are also inert Higgs fields
Hu′ , Hd′ , which have the same representation (1, 2)±1/2 under the SM gauge group, but do not
have the right additional U(1) charges to form unsuppressed Yukawa couplings with SM mat-
ter (see §4.3). There are also similar sets of fields for the triplet Higgs Tu , Td , Tu′ , Td′ , as well as
(see also, e.g. [77], for a similar situation). In the example below, the anomaly-free directions are spanned by
−4b4 /5 + b5 and 2b4 /5 + b6 . It turns out that excluding the anomalous symmetry does not affect the selection
rules below at all, so for completeness we still use all the U(1) charges in labeling the charges.

– 22 –
vector-like fields in other exotic representations, namely (3, 2)−5/6 , (3, 2)1/6 , (3̄, 1)−2/3 , (1, 1)1 .
In particular, the exotic (3, 2)−5/6 ruins phenomenology by causing proton decay and spoiling
gauge coupling unification (see also §4.7), and must be removed from the spectrum. We will
also discuss the phenomenological safety of other vector-like exotics as we proceed in later
sections.
As seen in §2.4, the multiplicities of these vector-like fields are controlled by the fluxes.
Unfortunately, it has been shown in [25] that for GUT groups higher or equal to SO(10), it
is impossible to remove all the vector-like exotics by tuning the fluxes. Nevertheless, it was
pointed out in [10] that it is easy to remove the most dangerous (3, 2)−5/6 .
Now we show that in the E7 models, this representation is reasonably likely to be removed
from the spectrum, at least for certain kinds of gauge divisor Σ. First, we notice that the
representation and its conjugate have (b3 , b4 , b5 , b6 ) = (±1, 0, 0, 0). Recall that the vertical
flux we turn on breaks directions 4, 5, 6. We then see that χ(3,2)−5/6 = 0, since Θiα ̸= 0 only
for i = 4, 5, 6 in (2.14), and the vertical flux does not contribute in Eq. (2.28). In other words,
the multiplicity is purely controlled by the remainder flux in Eq. (3.4), given by9
1
n(3,2)−5/6 = − (5nr − ϕ4r )2 Crem
2
− 1. (4.3)
2
Therefore, n(3,2)−5/6 = 0 if 5nr − ϕ4r = ±1 and Crem
2 = −2. Interestingly, some choices of the
remainder flux with the smallest tadpole satisfy these conditions. Consider the tadpole
1 rem 2 1 2 ij
[G4 ] = − Crem κ ϕir ϕjr . (4.4)
2 2
As demonstrated in §5, in many cases Σ is a del Pezzo surface and the available Crem with the
least negative self-intersection has Crem
2 = −2. Going forward we assume these technical con-
ditions, which also imply the condition discussed earlier that −KΣ is effective. Further work
would be needed to generalize the analysis to the situation when these technical conditions
are relaxed. In this situation, κij ϕir ϕjr is minimized by e.g.

ϕir = (2, 4, 6, 4, 2, 1, 3) , (4.5)

which indeed leads to n(3,2)−5/6 = 0. From now on, we always assume this choice of remainder
flux, with tadpole
1 rem 2
[G ] = 4 . (4.6)
2 4
How about other vector-like exotics? Unlike the above, vertical flux also contributes to the
multiplicities of other vector-like exotics. Comparing to remainder flux, vertical flux satisfies
more constraints like primitivity. We also need vertical flux to get the right chiral spectrum
and, to be discussed below, the right interactions. After fulfilling these more important re-
quirements, we find no more room to remove the remaining vector-like exotics; i.e., generically
there is a nontrivial or even large contribution to nβ from vertical flux. Since remainder flux
9
When ϕ4r = 5nr , Lβ becomes trivial and Eq. (2.27) no longer applies.

– 23 –
is orthogonal to vertical flux, there is no remainder flux that can cancel the contribution from
vertical flux. Therefore, we expect that all the other vector-like exotics are present in our mod-
els. Fortunately, below we will show that these vector-like exotics, including the triplet Higgs
(3̄, 1)1/3 , which potentially mediates dangerous proton decay, can still be phenomenologically
acceptable if some additional assumptions and tunings are made.

4.3 List of Yukawa couplings

With the selection rules §4.1, we can now list the Yukawa couplings that are not suppressed
by the approximate global symmetries. Recall that in a general 4D F-theory model, there are
three types of Yukawa couplings on a gauge divisor Σ [24]. First, there are Yukawa couplings
between three fields on the bulk of Σ (denoted by ΣΣΣ), but it has been shown in [24] that
these couplings are all absent when −KΣ is effective, which is assumed in our E7 models as
discussed in §2.4. The second type of Yukawa couplings are between one field on the bulk
of Σ and two fields on matter curves (denoted by ΣCC). These couplings are generically
present, and we will simply assume that all couplings of this type that satisfy the symmetry
constraints are present. Finally, there are Yukawa couplings between three fields on matter
curves (denoted by CCC). These couplings are characterized by codimension-3 singularities
of the fibration. Nevertheless as discussed in last section, the codimension-3 singularities in
the E7 models cannot be interpreted as Yukawa couplings. In conclusion, there are only
ΣCC-type couplings in our theory.10 Note that this UV structure of Yukawa couplings is
quite different from that of other previous SM-like constructions in F-theory, and we expect
new phenomenological features in the IR to arise from this structure.
We now investigate how the SM Yukawa couplings can arise from ΣCC-type couplings.
In principle, we can localize the vector-like Higgs on matter curves more general than P1 .
If we make such a choice, however, the general (including both diagonal and off-diagonal)
Yukawa couplings will require the non-existent CCC- and/or ΣΣC-type couplings apart from
ΣCC-type couplings. Therefore, to reproduce the SM Yukawa couplings with mixing between
all three generations, it is necessary to localize the Higgs on the bulk of Σ, and all SM chiral
matter on the matter curve C56 . This choice of localization also matches with the fact that,
from the discussion of §2.4, generically there are many vector-like fields on the bulk of Σ, but
no such pairs on P1 matter curves. We also notice that this choice of Higgs is only available
when the GUT group is as large as E7 , such that the adjoint includes the Higgs after breaking.
On the other hand, the chiral matter induced by vertical flux breaking descends from both
133 and 56, and can be localized on both Σ and C56 . Therefore to reproduce the SM Yukawa
couplings, we need to impose the flux constraint

χ133
(3,2) = 2Σ · (KB + Σ) · Dα nα = 0 . (4.7)
1/6

10
This coupling structure, from the low-energy perspective, can also be understood as a kind of R-symmetry,
where the fields on the bulk of Σ have R-charge 1, and those on matter curves have R-charge 1/2. We thank
Jesse Thaler for this comment.

– 24 –
Below we will see that this constraint can be easily satisfied. It is worth emphasizing that
this choice of localization automatically implies very different low-energy physics between the
Higgs and chiral matter, due to their distinct geometric origins.
Now, assuming that the SM chiral spectrum is supported on C56 , we can easily list all
couplings that do not violate the approximate global symmetries. For simplicity, here we first
ignore the couplings involving uncharged singlets under the SM gauge group; these singlets
may play the role of right-handed neutrinos and will be studied in §4.6. We then have the SM
Yukawa couplings:11

Hu QŪ : (1, 2)1/2,−3,−2,−1 × (3, 2)1/6,3/2,1,1/2 × (3̄, 1)−2/3,3/2,1,1/2 ,


(1, 2)−1/2,−2,−2,−1 × (3, 2)1/6,3/2,1,1/2 × (3̄, 1)1/3,1/2,1,1/2 ,


Hd QD̄ : (1, 2)−1/2,−2,−1,−1 × (3, 2)1/6,3/2,1,1/2 × (3̄, 1)1/3,1/2,0,1/2 ,


−1/2,−2,−1,0 × (3, 2)1/6,3/2,1,1/2 × (3̄, 1)1/3,1/2,0,−1/2 ,
(1, 2)

(1, 2)−1/2,−2,−2,−1 × (1, 2)−1/2,1/2,1,1/2 × (1, 1)1,3/2,1,1/2 ,


Hd LĒ : (1, 2)−1/2,−2,−1,−1 × (1, 2)−1/2,1/2,0,1/2 × (1, 1)1,3/2,1,1/2 , (4.8)


−1/2,−2,−1,0 × (1, 2)−1/2,1/2,0,−1/2 × (1, 1)1,3/2,1,1/2 ,
(1, 2)

where the first representation in each product is the up and down Higgs Hu , Hd . We also have
a number of other exotic couplings. There are couplings involving the triplet Higgs (3, 1)−1/3 :

Tu QQ : (3, 1)−1/3,−3,−2,−1 × (3, 2)1/6,3/2,1,1/2 × (3, 2)1/6,3/2,1,1/2 ,

Tu Ū Ē : (3, 1)−1/3,−3,−2,−1 × (3̄, 1)−2/3,3/2,1,1/2 × (1, 1)1,3/2,1,1/2 ,

(3̄, 1)1/3,−2,−2,−1 × (3, 2)1/6,3/2,1,1/2 × (1, 2)−1/2,1/2,1,1/2 ,


Td QL : (3̄, 1)1/3,−2,−1,−1 × (3, 2)1/6,3/2,1,1/2 × (1, 2)−1/2,1/2,0,1/2 ,


1/3,−2,−1,0 × (3, 2)1/6,3/2,1,1/2 × (1, 2)−1/2,1/2,0,−1/2 ,
(3̄, 1)

(3̄, 1)1/3,−2,−2,−1 × (3̄, 1)−2/3,3/2,1,1/2 × (3̄, 1)1/3,1/2,1,1/2 ,


Td Ū D̄ : (3̄, 1)1/3,−2,−1,−1 × (3̄, 1)−2/3,3/2,1,1/2 × (3̄, 1)1/3,1/2,0,1/2 , (4.9)


1/3,−2,−1,0 × (3̄, 1)−2/3,3/2,1,1/2 × (3̄, 1)1/3,1/2,0,−1/2 .
(3̄, 1)

These couplings are always present together with the SM Yukawa couplings, but the ones with
triplet Higgs mediate dimension-5 proton decay and need extra attention. Note that there are
unique sets of additional U(1) charges for Hu , Tu . This uniquely identifies these fields in the
decomposition (4.2) as

Hu = (1, 2)1/2,−3,−2,−1 , Tu = (3̄, 1)−1/3,−3,−2,−1 . (4.10)

11
Similar to mass terms, the couplings described here are terms in the superpotential W ; Yukawa couplings
between one boson and two fermions come as usual from the contributions to the potential V of the form
(∂ 2 W/∂ϕi ∂ϕj )ψi ψj and its conjugate.

– 25 –
On the other hand, there are three possible fields with distinct approximate U(1) charges for
each of Hd , Td , each of which couples to D̄, L in one of the three possible copies of the SU(5)
fundamentals. The choices of these charges will be discussed below.
There are also couplings involving other types of vector-like exotics, namely (3, 2)1/6 , (3, 1)2/3 , (1, 1)1 :

(3, 2)1/6,−1,−1,−1 × (3̄, 1)1/3,1/2,1,1/2 × (1, 2)−1/2,1/2,0,1/2 ,

(3, 2)1/6,−1,−1,−1 × (3̄, 1)1/3,1/2,0,1/2 × (1, 2)−1/2,1/2,1,1/2 ,
(3, 2)1/6,−1,−1,0 × (3̄, 1)1/3,1/2,1,1/2 × (1, 2)−1/2,1/2,0,−1/2 ,
(3, 2)1/6,−1,−1,0 × (3̄, 1)1/3,1/2,0,−1/2 × (1, 2)−1/2,1/2,1,1/2 ,
(3, 2)1/6,−1,0,0 × (3̄, 1)1/3,1/2,0,1/2 × (1, 2)−1/2,1/2,0,−1/2 ,
(3, 2)1/6,−1,0,0 × (3̄, 1)1/3,1/2,0,−1/2 × (1, 2)−1/2,1/2,0,1/2 ,
(3̄, 1)−2/3,−1,−1,−1 × (3̄, 1)1/3,1/2,1,1/2 × (3̄, 1)1/3,1/2,0,1/2 ,
(3̄, 1)−2/3,−1,−1,0 × (3̄, 1)1/3,1/2,1,1/2 × (3̄, 1)1/3,1/2,0,−1/2 ,
(3̄, 1)−2/3,−1,0,0 × (3̄, 1)1/3,1/2,0,1/2 × (3̄, 1)1/3,1/2,0,−1/2 ,
(1, 1)1,−1,−1,−1 × (1, 2)−1/2,1/2,1,1/2 × (1, 2)−1/2,1/2,0,1/2 ,
(1, 1)1,−1,−1,0 × (1, 2)−1/2,1/2,1,1/2 × (1, 2)−1/2,1/2,0,−1/2 ,
(1, 1)1,−1,0,0 × (1, 2)−1/2,1/2,0,1/2 × (1, 2)−1/2,1/2,0,−1/2 . (4.11)

These couplings induce, e.g., additional proton decay and may not be compatible with phe-
nomenology. On the other hand, all these couplings mix distinct copies of (3̄, 1)1/3 and/or
(1, 2)−1/2 with different U(1) charges, while the couplings in Eqs. (4.8) and (4.9) relate Hd , Td
in a given copy with matter fields within the same corresponding copy. As shown in §4.5.2,
extra tuning of the model is available such that the couplings in Eq. (4.11) are absent, and
we assume such absence throughout the paper. All the other couplings without uncharged
singlets and not listed above, if present, are exponentially suppressed by the approximate
global symmetries.

4.4 Proton decay

Before building semi-realistic Higgs sector and Yukawa couplings, there is an important issue
to be resolved: as in every GUT model, there is a possibility of couplings that give a proton
decay rate that exceeds experimental limits. In particular, since the triplet Higgs cannot be
removed from the spectrum, the couplings in last subsection naively seem to suggest that the
E7 models will suffer from excess proton decay mediated by dimension-5 operators. We now
show that, fortunately, both dimension-4 and 5 proton decay are ubiquitously suppressed in
the E7 models. This feature in some sense “comes for free” with the construction of these
models.

– 26 –
 First, dimension-4 proton decay in the MSSM is driven by the R-parity violating terms
in the superpotential:
W ⊃ α1 QLD̄ + α2 LLĒ + α3 D̄D̄Ū , (4.12)
where we have used the notation in Eq. (3.6). These are couplings between three chiral fields,
which all descend from 56 under the assumption of Eq. (4.7). Hence from the geometric
perspective, dimension-4 proton decay requires CCC-type couplings, which are absent in the
E7 models. The absence of these couplings is also natural from the form of the fields in the
symmetry-broken theory: since the approximate global charges b4 , b6 of all fields (including
all copies of D̄, L) appearing in Eq. (4.12) are half-integers, none of the interactions of this
type have vanishing net b4 , b6 charges, so all such interactions violate one of the approximate
global symmetries. Thus, dimension-4 proton decay is automatically absent in the E7 models
(with the caveat that we do not completely understand the (4, 6) singularities; it is not fully
clear whether the interplay between these singularities and flux breaking would modify this
conclusion). In fact, it was already pointed out in [81] that dimension-4 proton decay can
be eliminated in this way only when the GUT group is E7 or E8 ; this suppression arises in
heterotic and M-theory models as well as in F-theory.
Now we turn to dimension-5 proton decay. Such decay in conventional supersymmetric
GUTs comes from the following terms

W ⊃ λ1 Tu QQ + λ2 Td QL + M Tu Td , (4.13)

where Tu , Td are the triplet Higgs, and M is a large mass close to the GUT scale MGUT . Inte-
grating out Tu , Td , we then get the dimension-5 operator QQQL/M , which is only suppressed
by 1/M and leads to an unacceptable rate of proton decay. Nevertheless, the E7 models are
different from conventional GUTs in the following sense: although the first two terms in Eq.
(4.13) are not suppressed, we see from Eq. (4.9) that Tu , Td never have opposite additional
U(1) charges, hence the mass term in Eq. (4.13) is exponentially suppressed. Instead, Tu , Td
have their own vector-like partners, denoted by Td′ , Tu′ , with opposite additional U(1) charges.
These “primed” fields are inert, i.e. their Yukawa interactions with SM chiral matter are expo-
nentially suppressed, but they give Tu , Td large masses by the conventional mass terms. Now
the superpotential schematically has the form of

W ⊃ λ1 Tu QQ + λ2 Td QL + M Tu Td′ + M Tu′ Td + mTu Td + mTu′ Td′ , (4.14)

where m is exponentially suppressed compared to M . The operator we get by integrating out

the triplet Higgs and the vector-like partners is roughly (m/M ) QQQL/M , which is indeed
further exponentially suppressed by the factor m/M . Therefore, as long as M is sufficiently
large (probably close to MGU T ) and m/M is sufficiently small, the E7 models are safe from
overly dangerous proton decay.
Regarding dimension-6 proton decay mediated by gauge bosons along the broken direc-
tions of the gauge group, we do not see an obvious mechanism of suppression. The gauge
bosons typically have masses around the KK/GUT scale, which may be sufficiently high to

– 27 –
evade the current experimental bounds [82]. Moreover, fluxes on Σ can make the ground
state wavefunction of the gauge bosons more localized, and suppress its wavefunction overlap
(hence the coupling) to the chiral matter on C56 [83]. Note that such suppression may not
be exponential [84], but may already be sufficient for our purposes due to the high GUT
scale. Despite all these heuristic arguments, more techniques and explicit calculations are still
required to determine the exact rate of proton decay, which is essential for realistic model
building.

4.5 Higgs and Yukawa sectors

Now we turn to the Higgs and Yukawa sectors in the E7 models. Even with the Yukawa
couplings in Eq. (4.8) that have the right representations, it is not guaranteed that those
couplings resemble the structure of Higgs and Yukawa sectors in the Standard Model, due to
various differences of the E7 models from the conventional Standard Model. Below we explain
each of these differences, and write down the necessary conditions for realizing the SM Higgs
and Yukawa sectors.

4.5.1 Doublet-triplet splitting and the Higgs masses

Apart from proton decay, one of the important questions in general GUT models is the doublet-
triplet splitting problem, or why the masses of doublet and triplet Higgs are separated by many
orders of magnitude. In F-theory GUTs, this splitting in principle can be explained by the
presence of hypercharge flux [25]. For the tuned SU(5) GUTs, however, the explicit realization
of such splitting can be difficult; see §6 for further discussion. In contrast, the doublet and
triplet Higgs in the E7 model live on the bulk of Σ and always receive mass splitting from
hypercharge flux, which is also localized on a (remainder) surface on Σ. Therefore, the doublet
and triplet masses are automatically split once we break E7 to GSM , although the amount
of splitting is still unknown and new techniques must be developed for finding out the Higgs
mass spectrum.
Still, what controls the mass terms before the splitting by hypercharge flux? Similar to the
triplet Higgs in §4.4, the conventional µ-term i.e. µHu Hd is exponentially suppressed. This
suppression, however, does not mean that the µ-problem is solved, since the Higgs can still
get large masses from terms Hu Hd′ , Hd Hu′ , when Hu , Hd have their own vector-like partners
Hd′ , Hu′ . On the other hand, it means that there is some vector-like matter with light masses
when such vector-like partners do not exist. Indeed, such a scenario generically arises for Hd ,
and the essence of this effect lies in vertical flux breaking: although we have imposed that
the total chiral index of fields arising from 133 vanishes, there can still be nontrivial chiral
surpluses for each of the three copies of doublet and triplet Higgs in this representation (see
§4.1). Suppose we have the following spectrum for the three copies of Higgs fields:

(1, 2)−1/2,−2,−2,−1 : n1 , (1, 2)1/2,2,2,1 : n′1 ,

(1, 2)−1/2,−2,−1,−1 : n2 , (1, 2)1/2,2,1,1 : n′2 ,
(1, 2)−1/2,−2,−1,0 : n3 , (1, 2)1/2,2,1,0 : n′3 , (4.15)

– 28 –
where ni , n′i denote the multiplicities. Eq. (4.7) implies that n1 +n2 +n3 = n′1 +n′2 +n′3 , hence
the spectrum is non-chiral under the SM gauge group. On the other hand, generically we have
ni ̸= n′i and the spectrum would be chiral if the additional U(1)’s were gauge symmetries, i.e.
the Stückelberg mechanism was absent. If ni ̸= n′i for some i, there must be a field direction in
the i-th copy that cannot acquire any mass terms within the same copy. It can only get mass
terms from fields in other copies. Since they do not have opposite additional U(1) charges,
the resulting mass terms are exponentially suppressed, leading to a light doublet Higgs Hd .
It is tempting to use the above mechanism to solve the µ-problem. Unfortunately, the
µ-problem cannot be solved in this way for two reasons. First, only Hd , but not Hu , has three
copies in the branching rule. In other words, this mechanism for producing light Hd cannot
produce a light Hu . Second, the above mechanism relies on vertical flux breaking, which only
breaks E7 to SU(5) instead of GSM . This means that whenever a light Hd is produced in
this way, there must also be a light Td . Although there may still be doublet-triplet splitting
from hypercharge flux, the mass of the Td is still exponentially suppressed. Such Td directly
interacts with SM chiral matter and ruins the argument in §4.4, i.e. there is still too much
dimension-5 proton decay even with the exponential suppression in §4.4. In this sense, we
should even avoid any light Hd or Td produced in this way. As discussed in §4.5.2, we will
arrange the fluxes such that only one copy of Hd interacts with SM chiral matter. Without
loss of generality, let us pick the copy with (b4 , b5 , b6 ) = (2, 2, 1). Then avoiding light Hd and
Td coming from the above mechanism is achieved by the flux constraint

χ133
(3̄,1)1/3,−2,−2,−1 = 0 . (4.16)

This is another linear constraint on the flux parameters, similar to the ones for breaking the
gauge group or inducing three generations of chiral matter. This new constraint, however, is
the first constraint that involves the previously unused flux parameters ϕ5α , ϕ6α . Therefore,
given the gauge group and total chiral spectrum, there is always still some room in the E7
models for satisfying this new constraint.
The above still does not explain the origin of light masses in the SM Higgs sector. Sadly, in
the current construction of our E7 models, there is still no obvious solution to the µ-problem.
This is understandable, however, since the Higgs masses in F-theory are very complicated
quantities to calculate. Traditionally, the Higgs masses come from the vevs of some fields
localized on divisors other than Σ but intersecting with Σ. These fields behave as singlets
and couple to the vector-like matter on Σ. Nevertheless, the vevs or potential of these fields
depends on many factors, including but not limited to the detailed couplings between these
fields, the D-term potential, the nonperturbative superpotential, and most importantly, soft
SUSY breaking [85]. Therefore without understanding more basic issues like moduli stabiliza-
tion and SUSY breaking in F-theory, no precise statements on these vector-like masses can be
made. On the other hand, given such a complicated origin of the Higgs masses, it is reasonable
to expect that some hierarchy is generated and brings some of the Higgs to light scales. At
the same time, we should not allow more than one pair of Higgs to be at the electroweak scale,
although generically there are many vector-like fields with the same representation. This is

– 29 –
because when more than one Higgs field couples to SM chiral matter in the same way, the
flavor basis generically does not align with the Higgs mass basis. Such misalignment produces
tree-level flavor-changing neutral currents (FCNCs), which are not observed in experiments.12
In conclusion, to reproduce the SM Higgs sector, it is far from clear how to realize exactly one
pair of light Higgs doublets among all the Higgs fields. This is a major shortcoming of the E7
models, and we hope to give a better explanation for this Higgs hierarchy in the future.
As a remark, there are still many inert Higgs fields in the other two copies. In particular,
there can be multiple light inert Higgs fields, coming from pairing chiral surpluses between
the copies or other ways. Fortunately since they are inert, there is no tight constraint on these
fields. We note that the current experimental lower bound on Hu′ , Hd′ masses is around 100
GeV [82].

4.5.2 Structure of Yukawa couplings

One of the most dangerous features in the E7 model is that there are three copies of (3̄, 1)1/3
and (1, 2)−1/2 in the branching rules in §4.1, with different additional U(1) charges. The
generic case where the three generations of chiral matter are distributed in all the copies
is not phenomenologically acceptable for various reasons. First, generically there are chiral
differences for each of the copies, as in (4.15). While these add up to the three generations
in the total chiral spectrum, as demonstrated in §4.5.1 they can also form light vector-like
exotics between different copies. Next, having different copies in the chiral spectrum turns
on an unsuppressed set of exotic couplings in Eq. (4.11), including additional proton decay.
More seriously, multiple light Higgs fields are required to generate unsuppressed SM Yukawa
couplings for all the copies, see Eq. (4.8).13 As discussed in §4.5.1, such a Higgs sector again
leads to unacceptable FCNCs. Therefore to avoid all the above issues, we must arrange all
three generations of chiral matter to be within the same copy. For the choice of light Higgs
with (b4 , b5 , b6 ) = (2, 2, 1) in §4.5.1, for example, we should choose the corresponding copy of
chiral matter with (b4 , b5 , b6 ) = (1/2, 1, 1/2) and impose the following flux constraints:

χ56 56
(3̄,1)1/3,1/2,0,1/2 = χ(3̄,1)1/3,1/2,0,−1/2 = 0 . (4.17)

Again, these constraints are mild tuning on the remaining flux parameters ϕ5α , ϕ6α , which
is generically achievable. Nevertheless, it will be interesting to see whether there is a more
fundamental reason that leads to such choice of fluxes.
After this choice of chiral matter, the remaining couplings in the low-energy theory are
just the SM Yukawa couplings and their counterparts with the triplet Higgs. It would be even
more informative if we can get the values of the Yukawa couplings. Although calculating those
values is beyond our current F-theory technologies, we can gather some of their qualitative
12
We thank Jesse Thaler for pointing out this issue.
13
While there is indeed some hierarchy between Yukawa couplings in the observed Standard Model, we do
not expect the hierarchy to be as large as the exponential suppression from the approximate global symmetries.
Therefore, if we only use one Higgs for multiple copies with exponentially suppressed couplings, very probably
it will not give the right flavor structure.

– 30 –
features. Unlike the conventional F-theory models with CCC-type couplings, the use of ΣCC-
type couplings means that the Yukawa couplings are supported on the whole C56 instead of
points on it. If the Higgs wavefunction is nearly uniform on C56 , the Higgs will interact
with all three generations of chiral matter in the same way, thus the Yukawa couplings will
be undesiredly close to an identity matrix.14 Nevertheless, especially with the presence of
bulk fluxes, we expect the Higgs wavefunction to be non-uniform and peak in some smaller
region. A simple scenario would be that the region intersects with C56 in a connected small
but finite range. This scenario is then similar to the case of a single Yukawa point studied
in e.g. [86, 87], where the small nonperturbative correction is now due to the finite size of
the interaction region. In this way, the Yukawa hierarchy is generated as in the SU(5) F-
theory GUTs. On the other hand, to really compute the Yukawa couplings, we first need to
understand the Higgs wavefunction profile and its possible correlations with the exponentially
low Higgs mass. Once we understand these issues, we may be able to use the ultra-local
approach developed in [88–91] to computing Yukawa couplings within the intersecting region,
but understanding those issues remains very challenging. In more complicated scenarios where
there are multiple disconnected interaction regions, they are similar to the case of multiple
Yukawa points. The arguments in [16] then suggest that there is also some Yukawa hierarchy,
although the methods in [16] do not straightforwardly generalize to our models due to the use
of flux breaking. In summary, it is possible that there is some hierarchy between the Yukawa
couplings. This hierarchy may match with the observed Yukawa hierarchy, but explicitly
computing the Yukawa matrix in our models will be an important future step for realistic
model building.

4.6 Neutrino sector

Here we turn to the neutrino sector and make some brief comments. From Eq. (4.1), we see
that 56 gives three copies of singlets that can be right-handed neutrinos. Since in the above
we have restricted the leptons into one copy, only one copy of the singlets (1, 1)0,5/2,1,1/2 have
unsuppressed Yukawa couplings with SM chiral matter:

(1, 2)1/2,−3,−2,−1 × (1, 2)−1/2,1/2,1,1/2 × (1, 1)0,5/2,1,1/2 ,

(3, 1)−1/3,−3,−2,−1 × (3̄, 1)1/3,1/2,1,1/2 × (1, 1)0,5/2,1,1/2 . (4.18)

The other two copies of singlets have nontrivial multiplicities but belong to inert matter.
We can obtain the multiplicity of right-handed neutrinos from its “chiral index”. It sounds
strange to calculate a “chiral index” for non-chiral matter; the correct interpretation is that
the right-handed neutrinos carry additional U(1) charges, hence “would be” chiral matter if
we ignore the Stückelberg mechanism. Since there are no vector-like exotics on the matter
curve, the chiral index is the same as the exact multiplicity, which remains unchanged under
the symmetry breaking. Now from the flux constraints we imposed in previous sections, we
14
We thank Jonathan Heckman for pointing this out.

– 31 –
see that the chiral index is fixed to be

χ(1,1)0,5/2,1,1/2 = 3 . (4.19)

Therefore, we have three right-handed neutrinos, which favorably combine with the left-handed
ones to give three Dirac neutrinos and a square PMNS matrix. The above Yukawa couplings
then give the usual Dirac mass terms after electroweak symmetry breaking. This scenario is
more or less the same as conventional GUTs with SO(10) gauge group or above.
There are also Majorana mass terms from the ΣCC-type couplings involving two right-
handed neutrinos and a bulk singlet. Since all right-handed neutrinos have the same addi-
tional U(1) charges, the Majorana masses are always exponentially suppressed (compared to
string/GUT scale) by the additional U(1) symmetries. In fact, similar suppression was al-
ready used in some early type II [76, 77] and F-theory [25] SM-like model. It was estimated
in those references that the exponential suppression factor might be around 10−6 to 10−4 ,
which is much more mild than the electroweak hierarchy. This is not incompatible with the
observational constraints on the seesaw mechanism. We emphasize that, however, these nu-
merical estimates are very crude, and without more explicit computations of the masses and
couplings, we cannot make fully precise statements on how the left-handed neutrinos get very
small masses.

4.7 Gauge coupling unification

Here we briefly comment on the possibility of gauge coupling unification in our models. Despite
the use of E7 in the construction of models, whether gauge coupling unification is present in
any useful sense is far from obvious. From the point of view of the world-volume theory on the
IIB 7-branes supporting the E7 gauge theory (as in, e.g., [24]), it should be possible to find a
classical description of flux breaking through turning on flux (T-dual to turning on an adjoint
scalar as in, e.g., [92]). From this perspective, at sufficiently high energies the world-volume
E7 gauge symmetry would be effectively restored, and the expected extra gauge bosons would
become relatively light, so there is some sense in which gauge coupling unification might be
expected. Note, however, that the quantization of flux means that the background flux will
give a mass scale mKK = 1/lKK , where lKK is the compactification scale, so that this unification
only occurs much above the KK scale. Furthermore, in the nonperturbative F-theory regime,
where there is no weakly coupled description, it is not clear that the 7-brane world-volume
theory can be meaningfully separated from string theory in the bulk space. Thus, we do not
necessarily expect unification even at the compactification scale. To understand some of the
issues, we first clarify the meaning of gauge coupling unification in our string theory context.
There are two separate aspects. First at the GUT scale MGUT ,15 the gauge couplings in
our models are clearly unified if flux breaking is absent. The coupling is given by the volume
15
In the string theory context, MGUT may be around the KK scale or string scale depending on model
details.

– 32 –
of the gauge divisor:
1
≃ vol (Σ) . (4.20)
g2
It is estimated from observations that 1/α ≃ 24 at MGUT [93]; we simply assume that the
divisor volume is stabilized to this particular value by certain mechanisms. On the other
hand, the remainder flux breaks SU(5) to GSM and induces some splitting of gauge couplings
at MKK . It is then important to understand such splitting and determine its size. Such
splitting has been understood in type IIB models [94, 36]: the splitting between the SU(3)
and SU(2) gauge couplings is
1 1 1 1
− ≃− [c1 (L3 )]2 = (nQ′ + 1) , (4.21)
α2 (MGUT ) α3 (MGUT ) 10gs 5gs
where gs is the string coupling, and nQ′ is the number of vector-like pairs in the exotic
representation (3, 2)−5/6 ; recall that we have set nQ′ = 0 in previous sections. There is also a
unification-like relation
1 5 1 1 2 1
= = + . (4.22)
αY (MGUT ) 3 α1 (MGUT ) α2 (MGUT ) 3 α3 (MGUT )
All the above, however, cannot be directly applied to F-theory models, especially when the
models, like our E7 models, are intrinsically strongly coupled and have no type IIB limit.
This is because the axio-dilaton varies over the internal space and the meaning of the 1/gs
correction is no longer clear. The worldvolume theory, which was used to derive the type
IIB result, also needs to be reconsidered in F-theory setups. In addition, there may be large
stringy threshold corrections to the gauge kinetic functions due to the strong coupling nature
of these models. All these subtleties imply that the splitting at MGUT may not be small even
if we set nQ′ = 0.
Next, at scales lower than MGUT , the RG flow of the SM gauge couplings are affected
by the vector-like exotics. The RG flow depends on both the representations and the masses
of the vector-like exotics. Since the remainder flux already breaks the GUT group at MGUT ,
it is possible that some vector-like exotics are light and do not form GUT multiplets. These
exotics seriously alter the RG flow and may ruin gauge coupling unification. The existence
of such exotics, however, depends crucially on uncontrolled aspects of the models such as
SUSY breaking. It is also possible that the presence of these exotics compensates the above
splitting at MGUT and makes the couplings apparently unified from the bottom-up perspective.
Therefore, so far we cannot make any definite statement on how the vector-like spectrum may
affect the RG flow.
In conclusion, the gauge couplings in our models are affected by a number of uncontrolled
aspects, thus gauge coupling unification is not guaranteed in our models. From this perspec-
tive, the unification of the observed gauge couplings in ordinary MSSM looks like an accident
if our models really describe our Universe. Nevertheless, this conclusion mainly comes from
our inability to compute non-topological details of our models. A more careful string theory
analysis in the future may reveal that the observed unification is in fact not an accident at
all.

– 33 –
5 Explicit global constructions of E7 GUTs

In all the above sections, we have written down many necessary constraints on the geometry
and fluxes for constructing semi-realistic E7 GUTs in F-theory. It remains important to see
whether all these constraints can be satisfied simultaneously within a 4D F-theory model. In
this section, we provide an explicit global construction of such a model, using the tools of
toric hypersurfaces. The construction here is a generalization of that in [37]. It is also the
first explicit example of a rigid E7 GUT (rigid E6 GUTs were presented in [10]). Although we
only present a single example here, the same construction can be generalized to large class of
F-theory compactifications. Before writing down such an explicit model, it is useful to first
review the geometric and flux constraints we want to achieve:

• Σ as a del Pezzo surface supporting both rigid E7 (with effective −KΣ ) and hypercharge
flux, and C56 = −Σ·(4KB + 3Σ) as a P1 , to enable explicit computations and interesting
phenomenology. The first requirement demands that Σ is a rigid divisor on a non-toric
base.

• The general flux constraints in §2.1: flux quantization, primitivity for vertical flux, and
tadpole cancellation. In particular, we should look for flux configurations with minimal
tadpole.

• A vertical flux breaking E7 → SU(5) and a remainder (hypercharge) flux breaking

SU(5) → GSM . In particular we need r ≥ 4, see §2.2.

• χ133
(3,2) = 0 and χ56
(3,2) = 3 for the total chiral spectrum.
1/6 1/6

• All three families of chiral (3̄, 1)1/3 and (1, 2)−1/2 coming from the same copy, i.e. Eq.
(4.17), to avoid exotic vector-like spectrum and couplings.

• The copy of bulk (3̄, 1)1/3 and (1, 2)−1/2 that interacts with the chiral matter being
itself non-chiral, i.e. Eq. (4.16), to avoid light vector-like exotics.

• [c1 (L3 )]2 = −2 for hypercharge flux, to remove the exotic (3, 2)−5/6 .

As always, we set the flux associated to the non-flat fiber to zero.

Now we write down an explicit F-theory model that satisfies all the above constraints. As
in [10], we choose the base B through the following procedure. We start with an auxilliary
toric threefold A with h1,1 (A) = 4. Then the ambient fourfold X is a P1 -bundle over A
with a certain normal bundle, and B is a certain hypersurface in X. The geometry of B
can be analyzed using the techniques in Appendix A. With appropriate choices in the above
procedure, we can construct B containing a rigid Σ with r = 4 and nontrivial remainder flux.
Let us first construct the ambient space X. We choose A to be a P1 -bundle over the del
Pezzo surface dP2 , which has a toric description. Let us first introduce the notations. Within
dP2 i.e. blowup of P2 at two generic points, let e1 , e2 be the exceptional curves from the

– 34 –
 Toric ray Divisor
(1, 0, 0, 0) FE1 + FF
(0, 1, 0, 0) FE2 + FF
(−1, −1, −1, 0) FF
(−1, 0, −1, −3) FE1
(0, −1, −1, 0) FE2
(0, 0, −1, −4) Fσ
(0, 0, 1, 0) Fσ + FE1 + FE2 + FF
(0, 0, 0, −1) σA
(0, 0, 0, 1) σA + 4Fσ + 3FE1

Table 1. The toric rays and the corresponding divisors in the toric construction of the ambient
fourfold X.

blowup, and h = f + e1 + e2 be the hyperplane. The intersection numbers are f 2 = e21 = e22 =
−1, f · e1 = f · e2 = 1, e1 · e2 = 0. Now on A, we denote σ as the dP2 section and E1 , E2 , F as
the P1 -fibers along e1 , e2 , f respectively. We choose the normal bundle of the P1 -bundle to be
Nσ = −h, and the anticanonical class is −KA = 2σ+3E1 +3E2 +4F . The intersection numbers
on A follow straightforwardly from those on dP2 and the relation σ · (σ + F + E1 + E2 ) = 0.
Finally we let the fourfold X be a P1 -bundle over A with normal bundle NA = −4σ − 3E1 .
We denote σA as the section and FI be the fiber along I ∈ {σ, E1 , E2 , F }. The anticanonical
class is −KX = 2σA + 6Fσ + 6FE1 + 3FE2 + 4FF . Again, the intersection numbers follow
from those on A and the relation σA · (σA + 4Fσ + 3FE1 ) = 0. Note that with these choices of
normal bundles, there is a unique triangulation such that X is a smooth and projective toric
variety. The toric rays of X are listed in Table 1.
We now choose the threefold base B as a hypersurface in X with irreducible class B =
σA + 5Fσ + 5FE1 + 2FE2 + 3FF . By abuse of notation, we use B to denote both the base
and its divisor class in X. By adjunction −KB = B · (σA + Fσ + FE1 + FE2 + FF ). Using the
techniques in Appendix A, one can check that h1,1 (B) = h1,1 (X) = 5. In particular, in this
situation the divisors of B are spanned by intersections in X. The intersection numbers of
these divisors relevant to our purpose are

 
−2 1 1 0
 1 −1 0 1 
B · σA · FI · FJ =  . (5.1)
 
 1 0 −1 1 
0 1 1 −1

Now we consider the divisor Σ = B · σA . It is also the hypersurface in A with class

σ + 2E1 + 2E2 + 3F . This class is irreducible and does not have any base locus (considered

– 35 –
as a hypersurface in A), so it is a well-defined irreducible gauge divisor. We compute

−KΣ = B · σA · (Fσ + FE1 + FE2 + FF )

= σA · (2Fσ · FE1 + 2Fσ · FE2 + 3Fσ · FF + 3FE1 · FF ) , (5.2)
2
NΣ = B · σA
= −σA · (7Fσ · FE1 + 4Fσ · FE2 + 8Fσ · FF + 3FE1 · FF ) . (5.3)

Therefore by Eq. (3.2), we see that Σ is indeed a rigid divisor supporting E7 . The matter
curve is

C56 = Σ · (−4KB − 3Σ) (5.4)

= B · σA · (σA + 4Fσ + 4FE1 + 4FE2 + 4FF )
= B · σA · (FE1 + 4FE2 + 4FF ) . (5.5)

Notice that the divisor E1 + 4E2 + 4F in A is also irreducible and does not have any base
locus. Therefore, C56 is also irreducible with genus
1
g = 1 + C56 · (C56 + KΣ ) = 0 , (5.6)
2
which means that the matter curve is simply a P1 .
We can now study the constraints on vertical flux. First, we study primitivity by expand-
ing the Kähler form of B using a basis of base divisors:

[JB ] =B · (t1 (FE1 + FF ) + t2 (FE2 + FF ) + t3 (FE1 + FE2 + FF )

+ t4 (Fσ + FE1 + FE2 + FF ) + t5 (σA + 4Fσ + 4FE1 + 4FE2 + 4FF ) , (5.7)

where t1 , t2 , t3 , t4 , t5 are linear combinations of Kähler moduli, and may be negative inside the
Kähler cone of B in general. While determining the exact Kähler cone of a hypersurface in a
toric variety can be subtle, the Kähler cone of B must contain that of X [95]. For simplicity,
we look for a solution of the primitivity constraints in the Kähler cone of X only. By a direct
toric computation, one can check that the Kähler cone of X is given by t1 , t2 , t3 , t4 , t5 > 0. The
independent Siα are Siσ , SiE1 , SiE2 , SiF , where we have simplified the notation and denoted
Si(B·FI ) as SiI . The primitivity condition is then

t1 (ϕiE2 + ϕiσ ) + (t2 + 3t5 ) (ϕiE1 + ϕiσ ) + (t3 + t5 ) (ϕiF + 2ϕiσ ) + t4 (ϕiF + ϕiE1 + ϕiE2 ) = 0 ,
(5.8)
for all i. A necessary but not sufficient condition for satisfying primitivity is that there must
be some coefficients in Eq. (5.8) with opposite signs for each i. Below we will find explicit
solutions to primitivity and check that the solutions are within the Kähler cone.
Next, we require the total chiral indices to be

χ133
(3,2)1/6 = −2 (nF + nE1 + nE2 ) = 0 , (5.9)
χ56
(3,2)1/6 = −nF − 3nE1 − 5nσ = 3 . (5.10)

– 36 –
Here the nI parameterize the fluxes through Eq. (3.3). To understand what values of nI we
should turn on to get the right total chiral spectrum, we should first look at flux quantization,
since for E7 models c2 (Ŷ ) is not necessarily even. We can calculate c2 (Ŷ ) using the techniques
in [50], which involve picking a particular resolution of the E7 models, but the parity of c2 (Ŷ )
is expected to be resolution-independent. We outline the procedure in Appendix B, while here
we only apply the result, which tells us that we can turn on half-integers nE1 , nF and integers
nE2 , nσ (similarly for ϕ6α ) to guarantee flux quantization. Note that these may not be the
only choices of nI , since the structure of H 4 (Ŷ , Z) is subtle and may include elements with
fractional coefficients. Also note that these choices of nI do not necessarily mean that the
vertical flux does not belong to H 4 (Ŷ , Z), since as we will see its tadpole is still integer. Now
with these choices of nI , we see that an almost minimal flux configuration (nσ , nE1 , nE2 , nF ) =
(0, −3/2, 0, 3/2) already gives the above two chiral indices and is consistent with primitivity.
There are more flux conditions that constraint the values of ϕ5α , ϕ6α . First, by flux quan-
tization we should turn on half-integers ϕ6E1 , ϕ6F and integers for the remaining parameters.
Primitivity still constraints their values nontrivially. Moreover, to put the chiral matter into
the copy (b4 , b5 , b6 ) = (1/2, 1, 1/2), we need to impose Eq. (4.17), or in terms of flux parameters

−ϕ5F − 3ϕ5E1 − 5ϕ5σ = 12 , −ϕ6F − 3ϕ6E1 − 5ϕ6σ = 6 . (5.11)

We also need to avoid light vector-like exotics in the bulk copy (b4 , b5 , b6 ) = (2, 2, 1). Eq.
(4.16) then leads to
ϕ5F + ϕ5E1 + ϕ5E2 = 0 . (5.12)
These are mild linear constraints on the flux parameters ϕ5α , ϕ6α . Although there are multiple
solutions to these linear constraints, we should seek for solutions with minimal tadpole. By a
brute force search, we find that one of the optimal solutions is

 
7 1
(ϕ5σ , ϕ5E1 , ϕ5E2 , ϕ5F , ϕ6σ , ϕ6E1 , ϕ6E2 , ϕ6F ) = 0, −5, 2, 3, 1, − , 1, − , (5.13)
2 2
which consistently stabilizes the Kähler moduli at t1 = t2 + 3t5 = t3 + t5 = 3t4 . Together with
nI , this vertical flux gives a tadpole
1  vert   vert 
G · G4 = 32 . (5.14)
2 4
As a comparison, if we do not impose Eqs. (4.16) and (4.17) i.e. primitivity is the only
constraint on ϕ5α , ϕ6α , the minimal tadpole is
1  vert   vert 
G · G4 = 20 , (5.15)
2 4
given by e.g.
 
5 3
(ϕ5σ , ϕ5E1 , ϕ5E2 , ϕ5F , ϕ6σ , ϕ6E1 , ϕ6E2 , ϕ6F ) = 0, −4, −1, 5, 1, − , −1, . (5.16)
2 2

– 37 –
Therefore, we see that the vertical flux we need to turn on is slightly non-generic.
Let us now turn to remainder flux. It can be shown that Σ is a del Pezzo surface dP6 and
supports remainder flux. First recall that Σ is a hypersurface in A with class σ+2E1 +2E2 +3F .
In other words, Σ is the vanishing locus

xP + yP ′ = 0 , (5.17)

in A, where P, P ′ are sections of OA (2E1 + 2E2 + 3F ), OA (E1 + E2 + 2F ) respectively, and

x, y are the homogeneous coordinates of the P1 in A. For generic points in the dP2 , Eq.
(5.17) has a unique solution, representing a single point in P1 . On the other hand, there are
(2e1 + 2e2 + 3f ) · (e1 + e2 + 2f ) = 4 points in dP2 such that P = P ′ = 0, and Eq. (5.17)
represents the whole P1 . Therefore, the geometry of Σ is dP2 blown up in 4 generic points i.e.
a dP6 .
To construct the remainder flux, notice that the four exceptional curves on Σ from blowing
up dP2 (denoted by e3 to e6 ) are all P1 fibers in A, hence all have the same class in B.
Therefore, we can choose e.g. Crem = e3 −e4 (or any difference ei −ej for distinct i, j = 3, 4, 5, 6)
with Crem
2 = −2, and turn on the remainder flux specified in §4.2. Therefore, we need a total
tadpole of 36 to satisfy all the flux constraints. Using the techniques in [50], we find that
χ(Ŷ ) = 1176 and χ(Ŷ )/24 = 49 > 36, so tadpole cancellation is satisfied. Unfortunately,
there seems to be not much room to achieve full moduli stabilization, but the situation should
improve if we consider more complicated geometries.
Having the full flux configuration, it is now straightforward to also compute the vector-like
spectrum. For simplicity again we ignore the uncharged singlets. Since C56 is a P1 , all vector-
like pairs comes from the bulk of Σ. Using the formula for nβ in §2.4, we get the vector-like
spectrum as in Table 2. It is clear that there are too many doublet and triplet Higgs that are
not inert, and it is important to understand how the mass hierarchy is produced, such that we
only see one doublet Higgs (pair) at the electroweak scale. There are also a number of light
and inert vector-like exotics from the table.
In conclusion, we have obtained an explicit F-theory model with the SM gauge group from
rigid E7 , three families of SM chiral matter with qualitatively standard Yukawa couplings
and suppressed proton decay, and excess numbers of heavy Higgs with some doublet-triplet
splitting. The flux configuration requires some but not too much fine-tuning. We emphasize
again that most analysis in this section depends on the local geometry only. We expect that
many of the F-theory threefold bases contain local geometries that are the same or similar to
the above, so this explicit construction can be easily generalized to large class of 4D F-theory
compactifications.

6 Comparing with other F-theory constructions in the literature

As we have pointed out a number of times in the previous sections, the E7 models have many
features that are distinct from previous SM-like constructions in F-theory. This distinction
makes the E7 models a new interesting class of models to be studied in depth in the future.

– 38 –
 (1, 2)−1/2,3,2,1 18 (1, 2)1/2,−3,−2,−1 18
(1, 2)−1/2,−2,−2,−1 4 (1, 2)1/2,2,2,1 4
(1, 2)−1/2,−2,−1,−1 19 (1, 2)1/2,2,1,1 16
(1, 2)−1/2,−2,−1,0 24 (1, 2)1/2,2,1,0 27
(3̄, 1)1/3,3,2,1 21 (3, 1)−1/3,−3,−2,−1 21
(3̄, 1)1/3,−2,−2,−1 3 (3, 1)−1/3,2,2,1 3
(3̄, 1)1/3,−2,−1,−1 16 (3, 1)−1/3,2,1,1 13
(3̄, 1)1/3,−2,−1,0 21 (3, 1)−1/3,2,1,0 24
(3̄, 1)−2/3,−1,−1,−1 6 (3, 1)2/3,1,1,1 3
(3̄, 1)−2/3,−1,−1,0 1 (3, 1)2/3,1,1,0 4
(3̄, 1)−2/3,−1,0,0 19 (3, 1)2/3,1,0,0 19
(3, 2)1/6,−1,−1,−1 5 (3̄, 2)−1/6,1,1,1 2
(3, 2)1/6,−1,−1,0 0 (3̄, 2)−1/6,1,1,0 3
(3, 2)1/6,−1,0,0 16 (3̄, 2)−1/6,1,0,0 16
(3, 2)−5/6,0,0,0 0 (3̄, 2)5/6,0,0,0 0
(1, 1)1,−1,−1,−1 6 (1, 1)−1,1,1,1 3
(1, 1)1,−1,−1,0 1 (1, 1)−1,1,1,0 4
(1, 1)1,−1,0,0 15 (1, 1)−1,1,0,0 15

Table 2. The representations and multiplicities of vector-like matter originated from the adjoint 133
on the bulk of gauge divisor. Only the bold multiplicities correspond to fields interacting with the SM
chiral matter without exponential suppression. All the other fields are inert vector-like exotics. Note
that there are nontrivial linear relations between these numbers implied by the formulas in §2.4.

In this section, we explain in more detail some of the specific differences between the models
presented here and the F-theory models with tuned GSM or SU(5) reviewed briefly in §1, as
well as the rigid E6 GUTs.

6.1 Tuned models

There have been many SM-like F-theory constructions using tuned gauge groups such as GSM
or SU(5) (again, for reviews see [7, 8, 2, 3]). The most obvious difference between those
constructions and the E7 models presented here has been discussed in §1: namely, fine-tuning
of many complex structure moduli is required to obtain GSM or SU(5) geometrically, while
the presence of rigid E7 only depends on the normal bundle of Σ instead of any moduli. It
seems that rigid E7 factors are relatively abundant in the landscape. Although the measure
on the landscape has never been clear, a naive counting measure on the (singular) geometries
suggests a large exponential dominance of geometries supporting rigid E7 factors over those
supporting tuned gauge factors. Note, however, that it is possible that certain flux choices
may in some situations force complex structure moduli to a tuned locus with an enhanced
gauge group; further investigation of this possibility is needed to clarify the level of tuning

– 39 –
really involved in geometrically tuned constructions beyond the level suggested by the analysis
of e.g., [21]. Due to the moduli-independence of the relevant gauge group, it may also be easier
to incorporate a full analysis of moduli stabilization in the rigid models than the tuned ones.
Another significant difference regards the Yukawa couplings. In many SM-like F-theory
constructions, some selection rules are required to get rid of exotic couplings. The Yukawa
couplings in tuned GSM or SU(5) models come from CCC-type couplings, and all the matter
fields are localized on matter curves. Therefore, the selection rules are usually obtained by
engineering a set of multiple matter curves where different types of matter localize separately,
or some additional U (1)’s by tuning the global geometry. In the E7 models, selection rules
that remove exotic couplings automatically follow from the use of ΣCC-type couplings, and
easily separate the chiral matter from vector-like matter including the Higgs. There are also
approximate U(1)’s from the Stückelberg mechanism that arrive without additional tuning.
Therefore, the selection rules needed to match expectations from observed physics are more
easily realized in the E7 models than in other constructions. An example is the proton decay
suppression described in §4.4.
The means of realizing the Higgs in the two classes of models is also qualitatively different.
In tuned GSM or SU(5), the Higgs comes from some vector-like matter on matter curves. Such
a construction requires explicit specification of the sheaf cohomology groups in Eq. (2.30),
which are in general very hard to compute since they are moduli-dependent quantities. More
exotic tools like root bundles [11–15] may also be needed in the construction. In many cases,
there is no Higgs in the low-energy theory unless some further tuning is done. In SU(5), we
also need the Higgs matter surfaces to have remainder components, such that the hypercharge
flux is present on the Higgs curves and doublet-triplet splitting can be achieved. Generic
matter surfaces, however, are purely vertical unless further tuning on moduli is done, and
global examples of such scenarios are rare in the literature (see e.g. [37]). In contrast, in the
E7 models we can instead realize the Higgs as bulk vector-like matter, which generically has
nonzero multiplicities that are easily calculated from the fluxes. In this situation, there are
already Higgs fields with some doublet-triplet splitting without any further tuning, but the
issue becomes having too many instead of too few Higgs fields. It is less clear how to make one
pair of the Higgs exponentially lighter in the E7 models, while in the tuned models there can
be exactly one pair of Higgs, and thus the way to obtain the Higgs hierarchy may be clearer.
Because of the use of flux breaking and E7 , there are many further differences between
these two classes of models in terms of computational abilities. First, in tuned GSM or SU(5)
the total chiral spectrum is controlled by one flux parameter only. In many cases the chi-
ral indices contain large prefactors, and three generations of chiral matter cannot easily be
obtained using integer fluxes, unless more nontrivial (and less completely understood) quan-
tization conditions are used, as in [40, 43]. In the E7 models, however, many flux parameters
from vertical flux breaking contribute to the chiral indices, giving a linear Diophantine struc-
ture. As a result, it is natural to get three generations of chiral matter just by generic integer
fluxes.
Specifically for SU(5), the removal of exotic (3, 2)−5/6 appears to be harder. This is

– 40 –
because the form of hypercharge flux is completely fixed to be ϕir ∝ (2, 4, 6, 3) and there is
no free flux parameter like ϕ4r as in the E7 models. Therefore, there must be a factor of 5
in c1 (L3 ), and we need to use fractional line bundles to satisfy the condition [c1 (L3 )]2 = −2
for removing the exotic (3, 2)−5/6 . In contrast, as in §4.2, this condition in E7 is already
satisfied by a fairly likely choice of integer remainder flux. On the other hand, there can be
some controlled scenario in SU(5) where all the vector-like exotics are removed, while in E7 we
cannot remove most of the vector-like exotics; we can at best arrange them into inert fields.
In conclusion, we see that the E7 models are not only more natural in the landscape, but
also possess a number of phenomenological advantages over the tuned models. These models
demonstrate how using naturalness as the guiding philosophy can help us discover more semi-
realistic features in the landscape. On the other hand, these models still have their own
shortcomings especially regarding the heavy mass spectrum, due to the lack of computational
technologies.

6.2 Rigid E6 GUTs

In [9, 10], it was proposed that the rigid construction of SM-like models works equally well for
both E7 and E6 , since these two gauge groups are similarly abundant in the landscape. While
the two GUT groups share some features such as the naturalness of the gauge group and three
generations of chiral matter, E6 behaves differently when coming to Yukawa couplings. While
E7 models do not have any CCC-type couplings, in E6 models the gauge group only gets
enhanced to E8 at codimension-3 singularities, which are well-defined Yukawa points giving
CCC-type couplings. Moreover, the branching rules from E6 to GSM including the additional
U(1) charges (b4 , b5 ) are

27 → (1, 1)0,5/3,4/3 + (1, 1)0,5/3,1/3 + (1, 1)1,2/3,1/3

+ (3, 2)1/6,2/3,1/3 + (3̄, 1)−2/3,2/3,1/3 + (3̄, 1)1/3,−1/3,1/3 + (3̄, 1)1/3,−1/3,−2/3
+ (3̄, 1)1/3,4/3,2/3 + (1, 2)−1/2,−1/3,1/3 + (1, 2)−1/2,−1/3,−2/3 + (1, 2)−1/2,4/3,2/3 , (6.1)
78 → (8, 1)0,0,0 + (1, 3)0,0,0 + 3 × (1, 1)0,0,0
+ [(1, 1)0,0,1 + (1, 1)1,−1,0 + (1, 1)1,−1,−1 + (3, 2)−5/6,0,0 + (3, 2)1/6,−1,0 + (3, 2)1/6,−1,−1
+ (3̄, 1)−2/3,−1,0 + (3̄, 1)−2/3,−1,−1 + (3̄, 1)1/3,−2,−1 + (1, 2)−1/2,−2,−1 + conjugates] .
(6.2)

Unlike E7 , one can check that there is no suitable field on the bulk of Σ that can play the
role of Higgs, so the Higgs must be localized on the matter curve. The Yukawa couplings in
E6 models are more similar to the tuned models and many results in §4 do not apply to E6
models, while E6 models also suffer from many vector-like exotics. In this sense, E7 models
are fundamentally different from any other F-theory GUT models.

– 41 –
7 Conclusion

In this paper, we have studied various phenomenological aspects of E7 GUTs in 4D F-theory

compactifications. These models were proposed in [9, 10] as a large class of natural SM-like
constructions, since rigid E7 gauge factors are moduli independent and common in the F-
theory landscape. Vertical and remainder fluxes are used to break E7 to the SM gauge group,
and appear fairly likely to induce three generations of SM chiral matter. Here we have shown
that the use of E7 and flux breaking also naturally implies several more phenomenologically
favorable features, including suppression of proton decay, doublet-triplet splitting, and Higgs
candidates with the right structure of SM Yukawa couplings due to approximate global sym-
metries descending from the E7 Cartan generators. For the first time, we have written down
an explicit global construction of such E7 models that achieve all the above features. The
construction of these features is qualitatively distinct from other SM-like constructions in the
previous F-theory literature. In particular, only mild tuning on the discrete data of the geom-
etry and the flux background is involved in the construction. The results in this paper give
us strong hints towards SM constructions in F-theory that are both realistic and natural. In
other words, these results appear to be compatible with the hypothesis that our Universe can
be described as a natural solution in the string landscape.
These results, on the other hand, are still far from complete in realizing the full details
of the Standard Model in string theory. Since the E7 models are inherently strongly coupled
and there is extensive use of fluxes, the machinery for computing continuous parameters, or
at least the moduli dependence of continuous parameters, is very limited. While we can more
or less fully specify the discrete data in the E7 models, our arguments are at best qualitative
when it comes to questions on Yukawa couplings, mass scales, etc. Such limitations lead to
a number of shortcomings of the E7 models, especially the unavoidable presence of (mostly
inert) vector-like exotics with masses not determined. The Higgs hierarchy problem, i.e. the
µ-problem, also remains unsolved in our models.
There are many challenges to answering these important questions. First, we need to
develop new tools beyond the ultra-local approach [88–91] to compute the moduli dependence
of various quantities. After that, we still need to understand more fundamental questions like
the realization of moduli stabilization and SUSY breaking in F-theory, which by themselves
are essential components of realistic model building. Solutions to these questions also involve
tackling some open problems such as computing the Kähler potential in F-theory. All these
tasks are particularly challenging when there is no weakly coupled type IIB limit for the E7
models. Nonetheless, some insight into aspects like gauge coupling unification may be possible
by considering the 8D world-volume theory on the E7 7-branes, where flux breaking should
have a classical (if nonperturbative) description.
There are also several extensions of the E7 models presented in this paper that should
be investigated further. Throughout the paper, we have made several strong assumptions
on the geometry, such as restricting the matter curve to be P1 , to enable more interesting
computations on the discrete data. It will be interesting, although technically challenging, to

– 42 –
relax these assumptions and explore more behavior of the E7 models. It is also important to
study the statistics of these E7 models in the F-theory landscape. By scanning through a large
set of F-theory bases and flux configurations, we can quantitatively analyze the genericity of
different features of the E7 models, which further sheds light on where our Universe sits in
the landscape.
We hope to address some of these questions in future studies.

Acknowledgments

We would like to thank Lara Anderson, Martin Bies, Mirjam Cvetic, James Gray, Daniel
Harlow, Jonathan Heckman, Patrick Jefferson, Manki Kim, Paul Oehlmann, Jesse Thaler,
and Andrew Turner for helpful discussions. This work was supported by the DOE under
contract #DE-SC00012567.

A Toric hypersurfaces

In this Appendix, we explain how to count h1,1 , or the number of divisors, of a threefold
hypersurface in an ambient toric fourfold, following the general approach of Danilov and
Khovanskii [96].16 This technique is useful in §5. To simplify the discussion, we focus on simple
cases where there is a triangulation such that both the ambient space and the hypersurface
are smooth. We also assume that the hypersurface does not have any base locus.
The geometry of the hypersurface can be understood from its stratification. First we
look at the stratification of the ambient space. A d-dimensional toric variety is given by a
disjoint union of algebraic tori (C∗ )k , where 0 ≤ k ≤ d. These algebraic tori, called strata,
are associated with the cones of a toric (polyhedral) fan. To be more precise, for a toric fan
Σ with n-dimensional cones σ (n) ∈ Σ(n) (where 0 ≤ n ≤ d), the toric variety PΣ is given by

Tσ(n) ∼
a a
PΣ = Tσ(n) , = (C∗ )d−n . (A.1)
n σ (n) ∈Σ(n)

Notice that the unique σ (0) corresponds to the prime stratum (C ∗ )d , which is the defining
feature of toric varieties. The one-dimensional cones σ (1) are also given by the toric rays ⃗v ,
associated with prime toric divisors D⃗v .
Now consider a hypersurface Z as a divisor in PΣ . We abuse notation and use Z to also
denote its divisor class:
X
Z= a⃗v D⃗v . (A.2)
⃗v

Note that the prime toric divisors are not all independent and there are multiple choices of the
coefficients a⃗v for the same Z; our final results are independent of such a choice. We assume
16
We thank Manki Kim for teaching us these techniques.

– 43 –
that all the strata of PΣ intersect Z transversely. Then Z admits the following stratification
a a
Z= Zσ(n) , Zσ(n) = Z ∩ Tσ(n) . (A.3)
n σ (n) ∈Σ(n)

Note that the dimension of the strata Zσ(n) is d − n − 1. To understand the geometry of Zσ(n) ,
it is useful to construct the Newton polytope ∆ of Z

∆ = {m
⃗ |m
⃗ · ⃗v ≥ −a⃗v , ∀⃗v ∈ Σ(1)} . (A.4)

The Newton polytope encodes information about the holomorphic sections of the line bundle
OPΣ (Z). Below we restrict to the case where this line bundle is big, i.e. ∆ is also d-dimensional.
The faces of ∆ encode the geometry of Z in the following way. From ∆ we can construct
the so-called normal fan Σ(∆), where each k-dimensional face Θ(k) is associated with a (d−k)-
dimensional cone in Σ(∆).17 The resulting toric variety PΣ(∆) is a blowdown of PΣ , which is
singular in general. The corresponding blowdown of Z is denoted by Z(∆). An important fact
is that Z(∆) is an ample divisor in PΣ(∆) . Now given the one-to-one correspondence between
faces of ∆ and cones of Σ(∆), we can write the stratification of Z(∆) as (again, assuming all
strata of PΣ(∆) intersect Z(∆) transversely)
aa
Z(∆) = ZΘ(k) . (A.5)
k Θ(k)

Note that the dimension of the strata ZΘ(k) is k − 1. Now including the blowups from Z(∆)
back to Z, the stratification of Z can be written as
a a a
Z = ZΘ(d) ZΘ(d−1) EΘ(d−k) × ZΘ(d−k) , (A.6)
Θ(d−1) k≥2 Θ(d−k)

where
k−1
a a 
EΘ(d−k) = (C ∗ )i , (A.7)
i=0

is the exceptional set associated with Θ(d−k) resulting from the blowups. The geometry of
Zσ(n) can then be seen by comparing Eqs. (A.3) and (A.6).
A great advantage of studying the stratification of Z is that the Hodge numbers of Z
can be computed using the Hodge-Deligne numbers together with the stratification [96, 39]
(see also [97] for more recent review and applications). In our case where Z is smooth, the
Hodge-Deligne numbers are just certain signed combinations of the Hodge numbers, but they
behave nicely under disjoint unions and products. One can then compute the Hodge-Deligne
numbers of Z by combining those of its strata, which are easy to get. Although the formulas
for general Hodge numbers are more complicated, it can be shown that for d ≥ 4, h1,1 (Z) is
17
The explicit construction of Σ(∆) is more complicated but is not important for our purpose.

– 44 –
simply given by counting the irreducible components of Zσ(1) . In terms of ∆, we should look
at the faces
Θi = {m
⃗ |m
⃗ · ⃗v ≥ −a⃗v , ∀⃗v ̸= ⃗vi ; m
⃗ · ⃗vi = −a⃗vi } . (A.8)

All Θi ’s are nontrivial when Z does not have any base locus, but they can have different
dimensions and contribute differently to h1,1 (Z):

• dim(Θi ) = 0: Zσ(1) is given by the components in EΣ(0) × ZΣ(0) . For generic moduli,
however, ZΣ(0) is an empty set and such Θi does not contribute to h1,1 (Z).

• dim(Θi ) = 1: Zσ(1) is given by the components in EΣ(1) × ZΣ(1) . Notice that ZΘi is a
degree n = l∗ (Θi )+1 hypersurface in C∗ , where l∗ denotes the number of interior points.
For generic moduli, this hypersurface is a collection of n points, so there are n copies of
an irreducible component in Zσ(1) , contributing n to h1,1 (Z).

• dim(Θi ) = k ≥ 2: Zσ(1) is given by the components in EΣ(k) × ZΣ(k) . We see that Zσ(1)
is irreducible, contributing 1 to h1,1 (Z).

Finally, the above procedure overcounts h1,1 (Z) by d, since there are d linear relations between
prime toric divisors in PΣ . One can check that the above procedure reproduces the famous
Batyrev formula for toric hypersurface Calabi-Yau manifolds [98].
For applications in §5, it is now clear that to obtain h1,1 (Z) = h1,1 (PΣ ), we can pick Z
such that dim(Θi ) ≥ 2 for all rays ⃗vi . It is straightforward to check this condition for the
example in §5.

B Flux quantization

In this Appendix, we compute c2 (Ŷ ) and determine the impact of flux quantization on the
vertical flux parameters in §5. The computation of c2 (Ŷ ) involves an explicit choice of reso-
lution. We consider the singular Weierstrass model in Eq. (3.1), and resolve it by performing
blowups. We denote
(x,y,s|e1 )
Y1 −→ Y, (B.1)

as the blowup from Y to Y1 by the redefinition

x → xe1 , y → ye1 , s → se1 . (B.2)

The resulting locus e1 = 0 is a divisor in the ambient space, denoted by E1 . Using the same
notation, we can then write down the resolution as the following steps [99, 100, 50]:

(e4 ,e5 |e7 ) (e2 ,e4 |e6 ) (e2 ,e3 |e5 ) (y,e3 |e4 ) (x,e2 |e3 ) (y,e1 |e2 ) (x,y,s|e1 )
Ŷ −→ Y6 −→ Y5 −→ Y4 −→ Y3 −→ Y2 −→ Y1 −→ Y. (B.3)

– 45 –
This resolution smooths out all singularities on Y up to codimension 3. The exceptional
divisors on Ŷ are given by

D1 = (E1 − E2 ) ∩ Ŷ ,
D2 = (−E1 + 2E2 − E3 − E5 − E6 ) ∩ Ŷ ,
D3 = (E1 − 2E2 + E3 + 2E5 + E6 − E7 ) ∩ Ŷ ,
D4 = E7 ∩ Ŷ ,
D5 = (E3 − E4 − E5 ) ∩ Ŷ ,
D6 = (−E3 + 2E4 + E5 − E6 − E7 ) ∩ Ŷ ,
D7 = (−E1 + 2E2 − E3 − 2E5 + E7 ) ∩ Ŷ . (B.4)

Using the above information, we can then compute c2 (Ŷ ) using the techniques in [99, 50].
The computation involves a pushforward formula from Ŷ to Y for the total Chern class, and
homology relations to relate all Di · Dj to Di · Dα . The result is

[c2 (Ŷ )] = [c2 (B)] + 11π ∗ KB

2

+ (−12D0 + 14D1 + 30D2 + 48D3 + 41D4 + 28D5 + 17D6 + 27D7 ) · π ∗ KB

+ (2D1 + 6D2 + 12D3 + 12D4 + 8D5 + 6D6 + 8D7 ) · π ∗ Σ . (B.5)

It is known that the first row of the above is even [101]. Therefore, the potentially odd terms
are Di · π ∗ KB for i = 4, 6, 7. To determine the parity of these terms, it is more convenient to
work with their pushforward π∗ (Di · π ∗ KB ) = Σ · KB . For the model in §5, we calculate

Σ · KB = KΣ + NΣ = −σA · (9Fσ · FE1 + 6Fσ · FE2 + 11Fσ · FF + 6FE1 · FF ) , (B.6)

which has odd coefficients. Notice that

Σ · (FE1 + FF )|B = B · σA · (FE1 + FF ) = σA · (Fσ · FE1 + Fσ · FF + 2FE1 · FF ) , (B.7)

has the same parity as Σ · KB , so the pullback

(D4 + D6 + D7 ) · π ∗ (FE1 + FF )|B , (B.8)

has the same parity as [c2 (Ŷ )]. We see that flux quantization as in Eq. (2.1) can be satisfied
by turning on half-integer ϕiE1 and ϕiF for i = 4, 6, 7. From Eq. (3.3), this is the same as
half-integer flux parameters nE1 , nF , ϕ6E1 , ϕ6F .

References
[1] W. Taylor and Y.-N. Wang, The F-theory geometry with most flux vacua, JHEP 12 (2015) 164
[1511.03209].
[2] M. Cvetic, J. Halverson, G. Shiu and W. Taylor, Snowmass White Paper: String Theory and
Particle Physics, 2204.01742.

– 46 –
 [3] F. Marchesano, B. Schellekens and T. Weigand, D-brane and F-theory Model Building, 12,
2022, 2212.07443.
[4] C. Vafa, Evidence for F theory, Nucl. Phys. B469 (1996) 403 [hep-th/9602022].
[5] D. R. Morrison and C. Vafa, Compactifications of F theory on Calabi–Yau threefolds — I,
Nucl. Phys. B473 (1996) 74 [hep-th/9602114].
[6] D. R. Morrison and C. Vafa, Compactifications of F theory on Calabi–Yau threefolds — II,
Nucl. Phys. B476 (1996) 437 [hep-th/9603161].
[7] T. Weigand, TASI Lectures on F-theory, [1806.01854].
[8] J. J. Heckman, Particle Physics Implications of F-theory, Ann. Rev. Nucl. Part. Sci. 60 (2010)
237 [1001.0577].
[9] S. Y. Li and W. Taylor, Natural F-theory constructions of standard model structure from E7
flux breaking, Phys. Rev. D 106 (2022) L061902 [2112.03947].
[10] S. Y. Li and W. Taylor, Gauge symmetry breaking with fluxes and natural Standard Model
structure from exceptional GUTs in F-theory, JHEP 11 (2022) 089 [2207.14319].
[11] M. Bies, M. Cvetič, R. Donagi, M. Liu and M. Ong, Root bundles and towards exact matter
spectra of F-theory MSSMs, JHEP 09 (2021) 076 [2102.10115].
[12] M. Bies, M. Cvetič and M. Liu, Statistics of limit root bundles relevant for exact matter
spectra of F-theory MSSMs, Phys. Rev. D 104 (2021) L061903 [2104.08297].
[13] M. Bies, M. Cvetič, R. Donagi and M. Ong, Brill-Noether-general limit root bundles: absence
of vector-like exotics in F-theory Standard Models, JHEP 11 (2022) 004 [2205.00008].
[14] M. Bies, Root bundles: Applications to F-theory Standard Models, 3, 2023, 2303.08144.
[15] M. Bies, M. Cvetič, R. Donagi and M. Ong, Improved statistics for F-theory standard models,
2307.02535.
[16] M. Cvetic, L. Lin, M. Liu, H. Y. Zhang and G. Zoccarato, Yukawa Hierarchies in Global
F-theory Models, JHEP 01 (2020) 037 [1906.10119].
[17] J. Halverson, C. Long and B. Sung, Algorithmic universality in F-theory compactifications,
Phys. Rev. D96 (2017) 126006 [1706.02299].
[18] W. Taylor and Y.-N. Wang, Scanning the skeleton of the 4D F-theory landscape, JHEP 01
(2018) 111 [1710.11235].
[19] W. Taylor, Y.-N. Wang and Y. Yu, work in progress, .
[20] S. Andriolo, S. Y. Li and S. H. H. Tye, String Landscape and Fermion Masses, Phys. Rev. D
101 (2020) 066005 [1902.06608].
[21] A. P. Braun and T. Watari, Distribution of the Number of Generations in Flux
Compactifications, Phys. Rev. D90 (2014) 121901 [1408.6156].
[22] W. Taylor and Y.-N. Wang, A Monte Carlo exploration of threefold base geometries for 4d
F-theory vacua, JHEP 01 (2016) 137 [1510.04978].
[23] R. Donagi and M. Wijnholt, Model Building with F-Theory, Adv. Theor. Math. Phys. 15
(2011) 1237 [0802.2969].

– 47 –
[24] C. Beasley, J. J. Heckman and C. Vafa, GUTs and Exceptional Branes in F-theory - I, JHEP
01 (2009) 058 [0802.3391].
[25] C. Beasley, J. J. Heckman and C. Vafa, GUTs and Exceptional Branes in F-theory - II:
Experimental Predictions, JHEP 01 (2009) 059 [0806.0102].
[26] R. Donagi and M. Wijnholt, Breaking GUT Groups in F-Theory, Adv. Theor. Math. Phys. 15
(2011) 1523 [0808.2223].
[27] R. Blumenhagen, T. W. Grimm, B. Jurke and T. Weigand, Global F-theory GUTs, Nucl.
Phys. B 829 (2010) 325 [0908.1784].
[28] J. Marsano, N. Saulina and S. Schafer-Nameki, Compact F-theory GUTs with U(1) (PQ),
JHEP 04 (2010) 095 [0912.0272].
[29] T. W. Grimm, S. Krause and T. Weigand, F-Theory GUT Vacua on Compact Calabi-Yau
Fourfolds, JHEP 07 (2010) 037 [0912.3524].
[30] S. Krause, C. Mayrhofer and T. Weigand, G4-flux, chiral matter and singularity resolution in
f-theory compactifications, Nuclear Physics B 858 (2012) 1 [1109.3454].
[31] V. Braun, T. W. Grimm and J. Keitel, Geometric Engineering in Toric F-Theory and GUTs
with U(1) Gauge Factors, JHEP 12 (2013) 069 [1306.0577].
[32] C.-M. Chen, J. Knapp, M. Kreuzer and C. Mayrhofer, Global SO(10) F-theory GUTs, JHEP
10 (2010) 057 [1005.5735].
[33] C.-M. Chen and Y.-C. Chung, On F-theory E6 GUTs, JHEP 03 (2011) 129 [1010.5536].
[34] J. C. Callaghan and S. F. King, E6 Models from F-theory, JHEP 04 (2013) 034 [1210.6913].
[35] J. C. Callaghan, S. F. King and G. K. Leontaris, Gauge coupling unification in E6 F-theory
GUTs with matter and bulk exotics from flux breaking, JHEP 12 (2013) 037 [1307.4593].
[36] C. Mayrhofer, E. Palti and T. Weigand, Hypercharge Flux in IIB and F-theory: Anomalies
and Gauge Coupling Unification, JHEP 09 (2013) 082 [1303.3589].
[37] A. P. Braun, A. Collinucci and R. Valandro, Hypercharge flux in F-theory and the stable Sen
limit, JHEP 07 (2014) 121 [1402.4096].
[38] M. Buican, D. Malyshev, D. R. Morrison, H. Verlinde and M. Wijnholt, D-branes at
Singularities, Compactification, and Hypercharge, JHEP 01 (2007) 107 [hep-th/0610007].
[39] A. P. Braun and T. Watari, The Vertical, the Horizontal and the Rest: anatomy of the middle
cohomology of Calabi-Yau fourfolds and F-theory applications, JHEP 01 (2015) 047
[1408.6167].
[40] M. Cvetič, J. Halverson, L. Lin, M. Liu and J. Tian, Quadrillion F -Theory Compactifications
with the Exact Chiral Spectrum of the Standard Model, Phys. Rev. Lett. 123 (2019) 101601
[1903.00009].
[41] D. Klevers, D. K. Mayorga Pena, P.-K. Oehlmann, H. Piragua and J. Reuter, F-Theory on all
Toric Hypersurface Fibrations and its Higgs Branches, JHEP 01 (2015) 142 [1408.4808].
[42] N. Raghuram, W. Taylor and A. P. Turner, General F-theory models with tuned
(SU(3) × SU(2) × U(1))/Z6 symmetry, JHEP 04 (2020) 008 [1912.10991].

– 48 –
[43] P. Jefferson, W. Taylor and A. P. Turner, Chiral spectrum of the universal tuned (SU(3) ×
SU(2) × U(1))/Z6 4D F-theory model, JHEP 02 (2023) 254 [2210.09473].
[44] D. R. Morrison and W. Taylor, Classifying bases for 6D F-theory models, Central Eur. J.
Phys. 10 (2012) 1072 [1201.1943].
[45] D. R. Morrison and W. Taylor, Non-Higgsable clusters for 4D F-theory models, JHEP 05
(2015) 080 [1412.6112].
[46] D. R. Morrison and W. Taylor, Toric bases for 6D F-theory models, Fortsch. Phys. 60 (2012)
1187 [1204.0283].
[47] P. Candelas, D.-E. Diaconescu, B. Florea, D. R. Morrison and G. Rajesh, Codimension three
bundle singularities in F theory, JHEP 06 (2002) 014 [hep-th/0009228].
[48] C. Lawrie and S. Schäfer-Nameki, The Tate Form on Steroids: Resolution and Higher
Codimension Fibers, JHEP 04 (2013) 061 [1212.2949].
[49] I. Achmed-Zade, I. n. García-Etxebarria and C. Mayrhofer, A note on non-flat points in the
SU(5) U(1)P Q F-theory model, JHEP 05 (2019) 013 [1806.05612].
[50] P. Jefferson, W. Taylor and A. P. Turner, Chiral matter multiplicities and
resolution-independent structure in 4D F-theory models, 2108.07810.
[51] P. Jefferson, S. Y. Li and W. Taylor, work in progress, .
[52] R. Donagi and M. Wijnholt, Model Building with F-Theory, Adv. Theor. Math. Phys. 15
(2011) 1237 [0802.2969].
[53] T. W. Grimm, The N=1 effective action of F-theory compactifications, Nucl. Phys. B 845
(2011) 48 [1008.4133].
[54] T. W. Grimm, M. Kerstan, E. Palti and T. Weigand, Massive Abelian Gauge Symmetries and
Fluxes in F-theory, JHEP 12 (2011) 004 [1107.3842].
[55] K. Kodaira, On compact analytic surfaces: Ii, Annals of Mathematics 77 (1963) 563.
[56] A. Néron, Modèles minimaux des variétés abéliennes sur les corps locaux et globaux,
Publications Mathématiques de l’IHÉS 21 (1964) 5.
[57] T. Shioda, On elliptic modular surfaces, J. Math. Soc. Japan 24 (1972) 20.
[58] R. Wazir, Arithmetic on elliptic threefolds, Compositio Mathematica 140 (2004) 567–580
[math/0112259].
[59] E. Witten, On flux quantization in M theory and the effective action, J. Geom. Phys. 22
(1997) 1 [hep-th/9609122].
[60] K. Becker and M. Becker, M theory on eight manifolds, Nucl. Phys. B 477 (1996) 155
[hep-th/9605053].
[61] S. Gukov, C. Vafa and E. Witten, CFT’s from Calabi-Yau four folds, Nucl. Phys. B 584
(2000) 69 [hep-th/9906070].
[62] S. Sethi, C. Vafa and E. Witten, Constraints on low dimensional string compactifications,
Nucl. Phys. B 480 (1996) 213 [hep-th/9606122].
[63] T. W. Grimm and H. Hayashi, F-theory fluxes, Chirality and Chern-Simons theories, JHEP
03 (2012) 027 [1111.1232].

– 49 –
[64] T. W. Grimm and R. Savelli, Gravitational Instantons and Fluxes from M/F-theory on
Calabi-Yau fourfolds, Phys. Rev. D 85 (2012) 026003 [1109.3191].
[65] K. Dasgupta, G. Rajesh and S. Sethi, M theory, orientifolds and G - flux, JHEP 08 (1999)
023 [hep-th/9908088].
[66] S. Y. Li and W. Taylor, Large U(1) charges from flux breaking in 4D F-theory models, JHEP
02 (2023) 186 [2211.11768].
[67] A. Braun, A. Collinucci and R. Valandro, G-flux in f-theory and algebraic cycles, Nuclear
Physics B 856 (2012) 129–179 [1107.5337].
[68] J. Marsano and S. Schäfer-Nameki, Yukawas, g-flux, and spectral covers from resolved
calabi-yau’s, Journal of High Energy Physics 2011 (2011) [1108.1794].
[69] M. Bies, C. Mayrhofer and T. Weigand, Gauge Backgrounds and Zero-Mode Counting in
F-Theory, JHEP 11 (2017) 081 [1706.04616].
[70] M. Bies, C. Mayrhofer, C. Pehle and T. Weigand, Chow groups, Deligne cohomology and
massless matter in F-theory, 1402.5144.
[71] R. Blumenhagen, V. Braun, T. W. Grimm and T. Weigand, GUTs in Type IIB Orientifold
Compactifications, Nucl. Phys. B 815 (2009) 1 [0811.2936].
[72] M. Bershadsky, K. A. Intriligator, S. Kachru, D. R. Morrison, V. Sadov and C. Vafa,
Geometric singularities and enhanced gauge symmetries, Nucl. Phys. B481 (1996) 215
[hep-th/9605200].
[73] J. J. Heckman, D. R. Morrison and C. Vafa, On the Classification of 6D SCFTs and
Generalized ADE Orbifolds, JHEP 05 (2014) 028 [1312.5746].
[74] F. Apruzzi, J. J. Heckman, D. R. Morrison and L. Tizzano, 4D Gauge Theories with
Conformal Matter, JHEP 09 (2018) 088 [1803.00582].
[75] E. Witten, Nonperturbative superpotentials in string theory, Nucl. Phys. B 474 (1996) 343
[hep-th/9604030].
[76] R. Blumenhagen, M. Cvetic and T. Weigand, Spacetime instanton corrections in 4D string
vacua: The Seesaw mechanism for D-Brane models, Nucl. Phys. B 771 (2007) 113
[hep-th/0609191].
[77] L. E. Ibanez and A. M. Uranga, Neutrino Majorana Masses from String Theory Instanton
Effects, JHEP 03 (2007) 052 [hep-th/0609213].
[78] R. Blumenhagen, M. Cvetic, D. Lust, R. Richter and T. Weigand, Non-perturbative Yukawa
Couplings from String Instantons, Phys. Rev. Lett. 100 (2008) 061602 [0707.1871].
[79] T. Banks and N. Seiberg, Symmetries and Strings in Field Theory and Gravity, Phys. Rev. D
83 (2011) 084019 [1011.5120].
[80] D. Harlow and H. Ooguri, Symmetries in quantum field theory and quantum gravity, Commun.
Math. Phys. 383 (2021) 1669 [1810.05338].
[81] R. Tatar and T. Watari, Proton decay, Yukawa couplings and underlying gauge symmetry in
string theory, Nucl. Phys. B 747 (2006) 212 [hep-th/0602238].

– 50 –
 [82] Particle Data Group collaboration, Review of Particle Physics, PTEP 2022 (2022)
083C01.
[83] L. E. Ibanez, F. Marchesano, D. Regalado and I. Valenzuela, The Intermediate Scale MSSM,
the Higgs Mass and F-theory Unification, JHEP 07 (2012) 195 [1206.2655].
[84] A. Hebecker and J. Unwin, Precision Unification and Proton Decay in F-Theory GUTs with
High Scale Supersymmetry, JHEP 09 (2014) 125 [1405.2930].
[85] E. Palti, Vector-Like Exotics in F-Theory and 750 GeV Diphotons, Nucl. Phys. B 907 (2016)
597 [1601.00285].
[86] J. J. Heckman and C. Vafa, Flavor Hierarchy From F-theory, Nucl. Phys. B 837 (2010) 137
[0811.2417].
[87] S. Cecotti, M. C. N. Cheng, J. J. Heckman and C. Vafa, Yukawa Couplings in F-theory and
Non-Commutative Geometry, 0910.0477.
[88] A. Font, L. E. Ibanez, F. Marchesano and D. Regalado, Non-perturbative effects and Yukawa
hierarchies in F-theory SU(5) Unification, JHEP 03 (2013) 140 [1211.6529].
[89] A. Font, F. Marchesano, D. Regalado and G. Zoccarato, Up-type quark masses in SU(5)
F-theory models, JHEP 11 (2013) 125 [1307.8089].
[90] F. Marchesano, D. Regalado and G. Zoccarato, Yukawa hierarchies at the point of E8 in
F-theory, JHEP 04 (2015) 179 [1503.02683].
[91] F. Carta, F. Marchesano and G. Zoccarato, Fitting fermion masses and mixings in F-theory
GUTs, JHEP 03 (2016) 126 [1512.04846].
[92] W. Taylor, D-brane field theory on compact spaces, Phys. Lett. B 394 (1997) 283
[hep-th/9611042].
[93] S. P. Martin, A Supersymmetry primer, Adv. Ser. Direct. High Energy Phys. 18 (1998) 1
[hep-ph/9709356].
[94] R. Blumenhagen, Gauge Coupling Unification in F-Theory Grand Unified Theories, Phys. Rev.
Lett. 102 (2009) 071601 [0812.0248].
[95] M. Demirtas, C. Long, L. McAllister and M. Stillman, The Kreuzer-Skarke Axiverse, JHEP
04 (2020) 138 [1808.01282].
[96] V. I. Danilov and A. G. Khovanskii, Newton polyhedra and an algorithm for computing
hodge–deligne numbers, Mathematics of the USSR-Izvestiya 29 (1987) 279.
[97] P. Jefferson and M. Kim, On the intermediate Jacobian of M5-branes, 2211.00210.
[98] V. V. Batyrev, Dual polyhedra and mirror symmetry for calabi-yau hypersurfaces in toric
varieties, alg-geom/9310003.
[99] M. Esole, P. Jefferson and M. J. Kang, Euler Characteristics of Crepant Resolutions of
Weierstrass Models, Commun. Math. Phys. 371 (2019) 99 [1703.00905].
[100] L. Bhardwaj and P. Jefferson, Classifying 5d SCFTs via 6d SCFTs: Rank one, JHEP 07
(2019) 178 [1809.01650].
[101] A. Collinucci and R. Savelli, On Flux Quantization in F-Theory, JHEP 02 (2012) 015
[1011.6388].

– 51 –