Random Field Theories

Prepared for submission to JHEP
Random Field Theories in The Mirror Quintic Moduli

Space
arXiv:1608.05189v1 [hep-th] 18 Aug 2016
Kate Eckerlea and Brian Greeneb

a
Center for Theoretical Physics, Columbia University,
New York, New York 10027, USA
b
Center for Theoretical Physics and Department of Mathematics, Columbia University,
New York, New York 10027, USA
E-mail: [email protected], [email protected]
Abstract: We investigate the distribution of field theories that arise from the low energy
limit of flux vacua built on type IIB string theory compactified on the mirror quintic. For
a large collection of these models, we numerically determine the distribution of Taylor
coefficients in a polynomial expansion of each model’s scalar potential to fourth order, and
show that they differ significantly from potentials generated by random choices of such
coefficients over a flat measure.
Contents
1 Introduction 1
2 Background 3
2.1 Review of Flux Compactifications 3
2.2 Period Integrals 9
2.3 Period Expansions for the Mirror Quintic 14
2.4 Structure of the Hessian 20
3 Calculational Approach 22
3.1 Generating a Random Sample of Vacua 22
3.2 Computing Coefficients 25
4 Results 29
4.1 Masses 30
4.2 Couplings 42
5 Discussion 50
A Near Conifold Period Expansion Coefficients 52
1 Introduction
The enormous number of vacua scattered throughout the string landscape poses one of
the most significant challenges for making contact between string theory and observable
physics. Generally speaking, three approaches for tackling this challenge have been ad-
vanced. The first focuses on the distribution of string vacua within the moduli space of
geometrical compactificaitons, seeking mathematical structure that might entail patterns
in physical observables emerging from the low energy sector of the theory. Seminal pa-
pers of this sort are [1–4]. A second approach has been to model the collection of low
energy models arising from string compactifications as a space of random quantum field
theories, with coefficients of all renormalizable terms drawn from a random distribution of
perturbatively sensible values over a flat measure. The papers [5–10] illustrate the types of
conclusions that can be drawn using this approach. The third approach combines aspects
of the the first two by investigating the space of low energy field theories arising from
string compactifications, and determining the degree to which this space is well modeled
by random field theories drawn from a flat measure. That is, the third approach seeks
nontrivial structure in the space of low energy string dynamics which can then be used to
sharpen conclusions drawn from the first two approaches noted above. Examples of this
approach include [11–14]. In this paper, we push forward on this third approach.
–1–
The third approach involves two competing influences. The starting point for low en-
ergy string dynamics, for definiteness type IIB supergravity in ten dimensions, is surely
a theory which enjoys a great deal of structure. As we compactify this theory, preserv-
ing various amounts of low energy supersymmetry, we inject additional structure such as
the rich topology and geometry of Calabi-Yau compactifications. At the same time, the
distribution of low energy string dynamics arising from such compactifications becomes
substantially broader as masses and couplings depend sensitively on the detailed topolog-
ical and geometrical data of the compact manifold. This becomes all the more apparent
when we adorn our compactifications with branes and fluxes, which introduce yet more
degrees of freedom on which low energy properties depend. Collectively, then, one might
imagine that notwithstanding the structure of supergravity and string geometry, the im-
pact of varying the choices of fluxes in conjunction with the distinct locations of the vacua
in moduli space associated with each such flux choice, would result in the low energy field
theories arising from string theory being essentially random. As above, various authors,
(those following “approach two”) have indeed relied on this perspective.
Nevertheless, papers in approach three have noted that even with the statistical ten-
dency toward flattened distributions, residual structure in the the low energy dynamics can
persist. For example, [3, 10] have shown that the distribution of mass eigenvalues fill out
a non-random, and by now well-understood, mathematical pattern. The purpose of the
current paper is to push this perspective further by considering higher order coefficients
beyond the Hessian, and to determine the degree to which these terms are well-modeled
– or not – by a random distribution over a flat measure. We will argue that, much as
was found for mass eigenvalues, the third and fourth order terms in the low energy scalar
potential retain non-random structure.
The analysis leading to this result is conceptually straightforward, albeit computa-
tionally technical. We begin in section 2 by providing background on the various elements
required for understanding low energy string dynamics arising from flux compactifications,
focusing for definiteness on the mirror of the quintic hypersurface in CP4 . We include this
material for completeness and to set up notation; the reader familiar with the geometrical
machinery of flux compactifications can skip the first three subsections of this discussion.
The fourth and final background subsection, 2.4, reviews how flux compactifications give
rise to a particular pattern of mass eignenvalues, illustrating in the simplest setting the
kind of mathematical features relevant for the third approach. In section 3 we lay out
our calculational approach for generating a sample set of flux vacua, for computing the
form of the low energy Lagrangian describing small fluctuations about such vacua up to
fourth order, and for comparing these Lagrangians to those emerging from a random set of
theories drawn from a flat measure. In section 4 we provide our results and, in particular,
reveal nontrivial structure in the third and fourth order coefficients. Finally in section 5
we summarize our results and suggest further directions for study.
–2–
2 Background
2.1 Review of Flux Compactifications

The low energy dynamics of the type IIB string is governed by the type IIB supergravity
action, which provides our starting point [15],
2
"Z #
dτ ∧ ∗dτ̄ G(3) ∧ ∗Ḡ(3) F̃(5)
Z
2π 10
p
10 10 1
SIIB = 8 d x −g R − + + + C(4) ∧ H(3) ∧ F(3) +Sloc ,
`s 2 (Im(τ ))2 Im(τ ) 2
(2.1)
10
where R is the 10d Ricci Scalar in the Einstein frame, G(3) is the combined 3-form flux,
G(3) = F(3) − τ H(3) , (2.2)
τ is the axio-dilaton related to the dilaton, φ, by
τ = C(0) + ie−φ (2.3)
and F(p) and H(3) are obtained from potentials C(p−1) and B(2) ,
F(p) = dC(p−1) (2.4)

H(3) = dB(2) . (2.5)
This theory can be compactified on a Calabi-Yau 3-fold to yield an effective action for
the moduli fields, which describe how the compact manifold M varies from one spacetime
location to another in the four large dimensions. Such parameters are complex valued and
change continuously across the given family of Calabi-Yau, so they enter the 4d theory
as complex scalar fields. It is instructive to sketch the derivation of the effective action,
and give a very brief review of the geometry of Calabi-Yau moduli spaces. In the process
we introduce notation and summarize our strategy for generating an ensemble of random
effective field theories. Experienced readers may wish to skip this section.
Calabi-Yau moduli come in two different types: those associated with deformations
of the manifold’s complex structure, and those associated with deformations of its Kähler
form, J. The former are in one-to-one correspondence with elements of the (2, 1)-de Rham
cohomology group, H (2,1) (M), and the latter with H (1,1) (M). We denote the dimension
of these vector spaces by their Hodge numbers, h2,1 and h1,1 , respectively.
Complexifying the Kähler form, we deal with a moduli space of complex dimension
h (2,1) + h(1,1) , itself a Kähler manifold that factors locally into the direct product of two
separate Kähler manifolds: one spanned by the complex structure moduli and the other
spanned by the complexified Kähler moduli, with Kähler potential of the form,
2,1 1,1
Kcs (z 1 , z 2 , ..., z h ) + Kka (w1 , w2 , ..., wh ). (2.6)
Lower case indices (a, b, c, ...) will refer to Calabi-Yau moduli. They are ordered from 1 to
h2,1 + h1,1 running through all the complex structures first, followed by those of Kähler
type. However, their range in certain expressions may be restricted to moduli of one of
–3–
the two types. Most often this will be obvious from the context, but where there is the
possibility for ambiguity we will state which if any moduli are excluded.
The Kähler potential for the complex structure moduli is,
Z
cs 1 h2,1
K (z , ..., z ) = − log −i Ω ∧ Ω̄ (2.7)
M
where Ω is the holomorphic three-form of the Calabi-Yau manifold. It can be shown that
differentiating Ω with respect to any of the complex structure moduli yields a component
proportional to Ω, and a remaining closed (2, 1)-form. That is,
∂Ω
= ka Ω + χa (2.8)
∂z a
with χa ∈ H (2,1) (M). In particular the proportionality constant ka turns out to be,
ka = −Ka = −∂a K. (2.9)
This allows us to construct a basis for H (2,1) by acting on the holomorphic 3-form with
a Kähler covariant derivative, Da , whose action on Ω is defined by,
Da Ω ≡ χa = ∂a Ω + Kacs Ω. (2.10)
Furthermore, since Z
∂Ω
Ω∧ = 0, (2.11)
M ∂z a
such (2, 1)-forms are orthogonal to Ω.
We can now compute the term proportional to G(3) ∧ ∗Ḡ(3) in the supergravity action
upon compactification. This process amounts to taking the 10d spacetime to be the direct
product of a four dimensional (noncompact) Lorentzian manifold, M4 , and a compact
Riemannian one, M, which for us is a Calabi-Yau 3-fold. We write,
2,1 1,1
M10 = M4 × M(z 1 , ..., z h , w1 , ..., wh ), (2.12)
and perform the integration over M in the action. Since M is parameterized by the
aforementioned moduli, and since the Calabi-Yau are allowed to vary across locations in
M4 , performing the integral over M will yield an effective field theory involving moduli
fields, φa (xµ ).
Requiring Poincaré invariance in M4 implies only G(3) ’s components with all indices
in the compact dimensions may be nontrivial, and so
Z Z
G(3) ∧ ∗Ḡ(3) = G(3) · Ḡ(3) . (2.13)
M M
This is essentially a norm of a (for now general) closed 3-form on M. We may expand G(3)
and Ḡ(3) in an orthogonal basis for
H (3) (M) = H (3,0) (M) ⊕ H (2,1) (M) ⊕ H (1,2) (M) ⊕ H (0,3) (M), (2.14)
–4–
namely,
2,1 2,1
Ω , {χa }ha=1 , {χ̄a }ha=1 and Ω̄ (2.15)
allowing us to write,
Z Z Z Z Z
i ab̄
G(3) · Ḡ(3) = R G(3) ∧ Ω̄ Ḡ(3) ∧ Ω + K G(3) ∧ χ̄a Ḡ(3) ∧ χb .
M M Ω ∧ Ω̄ M M M M
(2.16)
Each of these can be expressed in terms of covariant derivatives of the Gukov-Vafa-
Witten superpotential, W , defined in terms of the (3)-form flux as,
Z
W (z, τ ) = Ω ∧ G(3) . (2.17)
M
The second term on the right hand side of eq. 2.16 involves Kähler covariant derivatives
of the superpotential with respect to the complex structure moduli (because it is built out
of (2, 1)-forms). It is proportional to,
K ab̄ Da W D̄b̄ W̄ . (2.18)
As is standard, we can define a “Kähler potential” for the axio-dilaton such that the
first term in on the right hand side of eq. 2.16 has the same form as eq. 2.18, i.e. so it is
∼ |Dτ W |2 . Specifically, we choose
Kax = − log(−i(τ − τ̄ )), (2.19)
and
Kaxτ τ̄ = (Kτaxτ̄ )−1 = (∂τ ∂τ̄ Kax )−1 . (2.20)
One makes this choice because
Z
1 i
G(3) ∧ Ω̄ = ∂τ − W = (∂τ + ∂τ Kax ) W ≡ Dτ W, (2.21)
(τ̄ − τ ) τ − τ̄
and so first term in 2.16 is proportional to
|Dτ W |2 = Kaxτ τ̄ Dτ W D̄τ̄ W̄ , (2.22)
which parallels the form arising for the other complex structure moduli, reflecting the rela-
tionship of type IIB string theory to F-theory in which the axio-dilaton explicitly becomes
another complex structure modulus.
Notationally, to include the axio-dilaton as as additional modulus we use new indices
– capital letters – that begin from zero, the index value reserved for the axio-dilaton. We
denote the full Kähler potential by K. It is the sum of all three pieces, Kcs , Kax and Kka .
The result, then, of dimensionally reducing the 3-form flux term is
2π i ¯ 2π cs ax ¯
8
R KI J DI W D̄J¯W̄ = 8 eK +K KI J DI W D̄J¯W̄ (2.23)
`s 2Im(τ ) M Ω ∧ Ω̄ `s
where the Kähler moduli are excluded.
–5–
The kinetic terms for all the Calabi-Yau moduli come from the Einstein Hilbert term in
the 10d action. They are noncanonical. We identify where they come from as well compute
the total relative factor between the kinetic and potential terms which will involve one
remaining expression given in terms of the volume of the compactification manifold. This
is meant as a qualitative description. First we decompose the 10d curvature scalar into the
trace of the noncompact component of the Ricci tensor, that of the compact component
(which is zero because Calabi-Yaus are Ricci flat), and the remaining terms which will
involve products of the metric and its derivatives with indices in both the compact manifold
and large dimensions, which we label Rmix ,
R10 = R4 + R6 + Rmix . (2.24)
Since R4 is a constant over the Calabi-Yau, integration of it over M yields a factor of

the volume of the Calabi-Yau. The factor then in front of the resulting 4d Einstein-Hilbert
term is
2π 2π Mp2
Vol(M) = V0 = 4πV0 (2.25)
`8s `2s 2
where we’ve defined the dimensionless constant V0 , the volume of the Calabi-Yau manifold
in string units. Since the string length is the fundamental length scale at which one will
see string modes, the volume of the Calabi-Yau must be large compared to `6s for the direct
compactification procedure we are employing to be valid.
In order to have a canonical Einstein-Hilbert term in the effective action one must
1
rescale the 4d metric so that the curvature rescales precisely with a factor of 4πV 0
. The new
curvature term also comes with kinetic terms for the volume modulus because the volume,
and thus the factor by which the 4d spacetime metric is rescaled, may be expressed in
terms of the volume modulus, ρ. 1
Incidentally, the term in eq. 2.1 which clearly yields kinetic terms for the axio-dilaton,
namely,
dτ ∧ ∗dτ̄
Z
∼ , (2.26)
(Im(τ ))2
arises in precisely the same manner; specifically from transforming from the string metric
to the Einstein metric by rescaling the string metric by eφ/2 . The resulting kinetic terms
for ρ and τ are noncanonical, specifically given by,
Mp2
Z
3
d4 x ∂µ ρ∂ µ ρ̄ + Kτaxτ̄ ∂µ τ ∂ µ τ̄ (2.27)
2 (ρ − ρ̄)2
ρ is not itself one of the Calabi-Yau moduli denoted by our indices a, b, ... Rather it is a
specific function of all the Kähler moduli. We shall shortly see that their noncanonical
kinetic terms (involving the contraction with their Kähler metric) reside in that for ρ in
eq. 2.27.
The kinetic terms for the complex structure moduli come from integration of Rmix
over M, and so are also generally noncanonical involving contraction with their respective
1
The imaginary part of ρ cubed is proportional to the volume squared.
–6–
Kähler metric as follows,
Mp2
Z
d4 xKab̄ ∂µ φa ∂ µ φ̄b (2.28)
2
We can thus write the effective action describing the moduli,
cs ax
Mp2 Mp2 eK +K I J¯
Z
4 I µ J
Sef f = d xKI J¯∂µ φ ∂ φ̄ − K DI W D̄ ¯
J W̄ . (2.29)
2 4π V02
Any consistent flux compactification of type II string theories on Calabi-Yau manifolds
requires the addition of negative tension localized objects. This is necessary in order to
satisfy F̃(5) ’s equation of motion which, when integrated over the Calabi-Yau, yields the
following tadpole condition,
Z
1
F ∧ H(3) + Qloc
3 = 0. (2.30)
`4s M (3)
This is effectively a statement of the consistency of the configuration of field lines wrap-
ping the Calabi-Yau 3-cycles (i.e. field lines in the small dimensions curl and close onto
themselves, while those in large dimensions end on mathematically valid sources).
It can be shown that the term in eq. 2.30 involving the R-R and NS-NS fluxes is
positive definite. Of the allowed localized objects that preserve Poincaré invariance in the
four large dimensions and can act as sources for the fluxes, only the O3-planes contribute a
negative charge to the total Qloc
3 , thus they must be included in the compactifation in order
to cancel all the remaining positive definite terms in eq. 2.30. Though the O3-planes are
not dynamical, they do in general impact the moduli space geometry. We assume a model
in which such effects are negligible. O3-planes also reduce the N = 2 supersymmetry we
began with to N = 1.
Generic theories with N = 1 supersymmetry involve complex scalars described by a
potential of the form,
V = eK Kab̄ Da W D̄b̄ W̄ − 3|W |2 , (2.31)
where the superpotential, W , is a holomorphic function of the complex scalars {φa }. Notice
that the −3|W |2 term is absent in our effective action. This ‘no-scale’ form arises from a
simple but general cancellation inherent to Calabi-Yau compactifications at the classical
level. Namely, because the classical superpotential is independent of the Kähler moduli, the
Kähler dependence of the scalar potential arises solely from the contribution to ∼ |DW |2
from
Kab̄ Ka W Kb̄ W̄ = Ka Ka |W |2 (2.32)
with indices running over the Kähler moduli only. The classical expression for their Kähler
potential (i.e. that which comes from the special geometry) is,
Z
ka
K = −2 log J ∧J ∧J (2.33)
M
Z
1
= −2 log 6 dV = − log(V02 ) (2.34)
`s M
= −3 log (−i(ρ − ρ̄)) . (2.35)
–7–
Figure 1. The qualitative dependence of the scalar potential on the volume of a Calabi-Yau
manifold.
So,
1
Ka Kb̄ Kab̄ = ∂a ρ∂b̄ ρ̄Kρ Kρ̄ × (∂ρ φa ∂ρ̄ φ̄b ) (2.36)
∂ρ ∂ρ̄ K
= Kρ Kρ̄ Kρρ̄ (2.37)
which gives,
−3 +3 −(ρ − ρ̄)2
Ka K a = = +3. (2.38)
ρ − ρ̄ ρ − ρ̄ 3
This then yields the cancellation
Kab̄ Ka W Kb̄ W̄ − 3|W |2 = 0. (2.39)
The resulting scalar potential is positive semi-definite, and so its zeros are its global
minima. Solutions of the SUSY condition, DI W = 0 for all I = 0, 1, ..., h2,1 , are the only
zeros because the metric and eK are positive definite. In general, when V 6= 0 the scalar
potential depends on the Kähler moduli through its overall dimensionful factor, Mp2 /4πV02 ,
but when V = 0, all such dependence drops out. The flattening of the potential at a zero
in the volume direction of parameter space is shown schematically in Figure 1.
We see too that the ρ-dependent factor in the kinetic term for the volume modulus
in eq. 2.27 is indeed its Kähler metric, Kρρ̄ . We may identify this term as the net kinetic
term for the Kähler moduli in eq. 2.29, similarly by the chain rule. Finally, we recognize
ka
1/V02 in eq. 2.29 as eK , and thus, the effective action we obtain upon compactification as
that of a theory with 1 + h2,1 + h1,1 complex scalars and N = 1 supersymmetry with an
additional/non-generic feature. Namely, the cancellation of the −3|W |2 in the potential
by the contribution from a subset of the scalars, specifically h1,1 of them. The feature is
entirely due to the fact we’ve compactified on a Calabi-Yau manifold and used only classical
expressions.
–8–
These are, generally speaking, subject to both α0 and gs corrections. In type IIB, the
complex structure Kähler potential is protected from both types, but the Kähler moduli
are not shielded from either. However, as is well known, such corrections are suppressed
in the large volume limit, 2 as are the instanton corrections the superpotential receives.
This setting also ensures backreaction of the fluxes on the geometry of the manifold is
subdominant. We will work in this regime and so now use the formulae we’ve reviewed to
set up explicit calculations on the mirror quintic.
2.2 Period Integrals

To search for local minima of the effective potential and compute its Taylor coefficients in
the expansion about these minima one must express the quantities in eq. 2.29 as explicit
functions of the complex structure(s) and axio-dilaton. To accomplish this we need only
express W and Kcs in this fashion, as all terms in eq. 2.29 are obtained from them.
Generally, the integrals over the compactification manifold need not be computed directly.
Rather they can be expressed in terms of a basis of systematically calculable functions, the
period integrals of the Calabi-Yau manifold, which are solutions to differential equations
(specific to the compactification manifold) known as the Picard-Fuchs equations.
By the Poincaré dulaity H (3) (M) is isomorphic to H(3) (M), the space of nontrivial
3-cycles. Thus, for any two closed 3-forms α(3) and β(3) there exist two 3-cycles A and B
such that, Z Z Z
α(3) ∧ β(3) = β(3) = α(3) . (2.40)
M A B
If {Ci } are a basis of 3-cycles, the right hand sides of eq. 2.40 are a linear combination of
the integrals of the relevant 3-form over the basis cycles.
Z h3 Z
X
i
β(3) = A β(3) (2.41)
A i=1 Ci
Z h3 Z
X
i
α(3) = B α(3) (2.42)
B i=1 Ci
where the Ai and B i are real numbers (the components of A and B in the Ci basis). Thus,
W can be expressed as a linear combination of the integrals of the holomorphic 3-form over
the basis cycles for H(3) (M). These are known as the period integrals, or period functions.
They are functions of the complex structure moduli only.
We note the existence of an integral and symplectic basis. The first of these properties
means Ci is a geometrical cycle (that is, a submanifold, not merely a formal object defined
as the dual to a 3-form). The second means each basis cycle intersects only one other basis
cycle, and does so exactly one time. We denote the period functions in this basis as follows,
Z
1 h2,1
Πi (z , ...z ) = Ω (2.43)
Ci
2
This limit is one in which not only the 6-volume but all subvolumes are large compared to the natural
sizes (involving the dimensionful constants).
–9–
where the index i ranges from zero to 2h2,1 + 1 for a total of h3 different periods. The
symplectic basis is the one most natural for us because the period functions have well
defined expansions about special points in the moduli space where vacua accumulate, as
we shall discuss at greater length shortly.
The intersection form allows us to express the effective action in terms of these natural
period functions. Two 3-cycles intersect at points in a 6-dimensional manifold. Since
cycles are oriented such points will have multiplicity ±1. The intersection form, in the
context where the (3)-homology groups are the domain, takes in two 3-cycles and sums the
intersection multiplicities. In light of the Poincaré duality this is equally viewed as a map
from two copies of the (3)-cohomology groups. That is, we write
Qij = Q(Ci , Cj ) = hCi ^ Cj [M]i (2.44)

Z
= Q̃ij = Q̃(αi , αj ) = αi ∧ αj . (2.45)
M
In an integral and symplectic basis Qij are the entries of a symplectic h3 × h3 -matrix.
The superpotential, W, can now be expressed as follows,
4
X
W = Gi Πi (z) (2.46)
i=0
= (F − τ H) · Π(z) (2.47)
where F and H are row vectors whose four entries indicate the quantity of R-R and NS-
NS flux wrapping the basis cycles, and Π(z) is a column vector containing the h3 period
functions. It can be shown that the 3-form fluxes wrapping the integral and symplectic
basis cycles are integrally quantized in units of 4π 2 α0 ,
Z
1
F ∈ 2πZ (2.48)
2πα0 Ci (3)
and similarly for H(3) . Since the overall dimensionful factor has been pulled outside the
potential, this amounts to requiring the entries of the F and H vectors in eq. 2.47 be
integers.
The Kähler potential for the complex structure modulus is expressed in terms of the
period functions as follows,
Kcs (z, z̄) = − log(iΠ† (z̄)Q−1 Π(z)) (2.49)
In evaluating these functions, one can avoid performing an integration over the compacti-
fication manifold because the periods are solutions to particular differential equations, the
Picard-Fuchs equations (associated with the given Calabi-Yau). Given the above expres-
sions for the superpotential and Kähler potential, one need only find the solutions to these
differential equations to write down an explicit effective action for the moduli.
The complexity of the Picard-Fuchs equations quickly mounts as the number of moduli
increase. We consider the simplest case, where h2,1 = 1, and so there are a total of four
– 10 –
period functions. There is a complete list 14 such compactifications, the most well known
being the mirror quintic. For these 14 models the Picard-Fuchs equation takes the following
form,
4
δ − z(δ + α1 )(δ + α2 )(δ + α3 )(δ + α4 ) u(z) = 0 (2.50)
d
where δ ≡ z dz , and the αj are rational numbers specific to the compactification (the mirror
quintic has αj = j/5).
A convenient basis for expressing solutions to this ODE, which we shall label {Ui (z)}3i=0 ,
is as follows [16]
U0 (z) = c G1,3
4,0 (−z; {1 − α1 , 1 − α2 , 1 − α3 , 1 − α4 }, {0, 0, 0, 0}) (2.51)
c 2,2
U1 (z) = G (z; {1 − α1 , 1 − α2 , 1 − α3 , 1 − α4 }, {0, 0, 0, 0}) (2.52)
2πi 4,0
c
U2− (z) = G3,1 (−z; {1 − α1 , 1 − α2 , 1 − α3 , 1 − α4 }, {0, 0, 0, 0})
(2πi) 4,0
2
c
U3 (z) = G4,0 (z; {1 − α1 , 1 − α2 , 1 − α3 , 1 − α4 }, {0, 0, 0, 0}) (2.53)
(2πi)3 4,0
(
U2− (z) Im(z) ≤ 0
U2 (z) = −
(2.54)
U2 (z) − U1 (z) Im(z) > 0
The Gm,n
p,q are Meijer-G functions defined in terms of contour integrals in the complex, say,
s-plane,
Πm n
j=1 Γ(bj − s)Πj=1 Γ(1 − aj + s)
Z
1
Gm,n (z; {a1 , ...ap }, {b1 , ..., b q }) = ds zs
p,q
2πi L Πqj=m+1 Γ(1 − bj + s)Πpj=n+1 Γ(aj − s)
(2.55)
where c is a constant specific to the given Calabi-Yau (one of the 14 models),
1
c= . (2.56)
Γ(α1 )Γ(α2 )Γ(α3 )Γ(α4 )
The particular linear combinations of the Ui (z) that yield the periods in the integral
symplectic basis, the Πi (z)’s, are fixed by the calculable monodromy transformations of
the homology cycles when transported about certain special points in the moduli space.
For the case of h2,1 = 1 there are three such special points: the large complex structure
point (which corresponds to z = 0 in our coordinates), the conifold point (z = 1) and
the Landau-Ginsburg point (z = ∞). These nontrivial monodromy transformations of the
3-cycles in turn yield nontrivial transformations for the corresponding period functions.
For instance, if we donate the shrinking sphere as the conifold is approached by C3 ,
and the cycle it intersects by C0 , then
Q03 = hC0 ^ C3 , [M]i →hC0 + nC3 ^ C3 , [M]i (2.57)

= hC0 ^ C3 , [M]i + nhC3 ^ C3 , [M]i (2.58)
= Q03 + n ∗ 0 = Q03 . (2.59)
The integer n is specified by requiring mutual consistency between all the monodromy
transformations in the mirror quintic’s moduli space, and as is well-known, this requires
– 11 –
n = 1. The monodromy transformations imply that the linear combinations of the afore-
mentioned solutions to the Picard-Fuchs equation, {Ui }, that correspond to the period
integrals in a symplectic basis for the mirror quintic are given by,
Π0 (z) = U0 (z) (2.60)

Π1 (z) = −U1 (z) (2.61)
Π2 (z) = 3U1 (z) − 5U2 (z) (2.62)
Π3 (z) = 5U1 (z) + 5U3 (z) (2.63)
where Π3 is the (analytic) period that vanishes at the conifold point, it’s partner, Π0 , picks
up a copy of Π3 for each revolution about the conifold point, and the remaining periods
are analytic and nonvanishing. For a detailed derivation including the general form for any
of the 14 one parameter models see, for instance, Appendix A of [16].
The transformations of the periods upon circling a given special point in the moduli
space fix their expansions in the neighborhood of the special point. In the case of the
conifold point the transformation,
Π0 (z) → Π0 (z) + Π3 (z) (2.64)
for each revolution z → (z − 1)e2πi + 1, implies
log(z − 1)
Π0 (z) = Π3 (z) + f (z) (2.65)
2πi
where f (z) is analytic and nonvanishing at the conifold point. The expansions of the
period functions are discussed in detail in the following section. For now we remark that
the branch cut for Π0 introduces a branch point singularity in first derivative of Π0 which
in turn results in a singularity in the Kähler metric at the conifold point.
We also adopt the standard convention (see, e.g., [16] for details) where the entries of
the period vector, Π(z), are given in descending order,
 
Π3 (z)
 
 Π (z) 
 2
Π(z) =  (2.66)


 Π1 (z) 
 
Π0 (z)
while those in the flux vectors are labeled in ascending order,

F = F0 , F1 , F2 , F3 (2.67)

H = H0 , H1 , H2 , H3 . (2.68)
With this review of notation and conventions, all functions in the effective action have
now been specified. The only free parameters are the fluxes, which for us consist of eight
integers (four R-R and four NS-NS fluxes). So, by randomly selecting a set of eight integers,
constructing the corresponding scalar potential, searching for local minima (in the z − τ
– 12 –
field space), and evaluating the potential’s Taylor coefficients about the local minima so
identified, one obtains the masses and couplings of a sample of effective field theories in
the landscape of the mirror quintic. Since this model’s only dynamical fields are the axio-
dilaton and the mirror quintic’s sole complex structure modulus there are a total of four
real degrees of freedom.
For the masses and couplings to have physical significance one must trade the {z, τ, z̄, τ̄ }
basis for one that simultaneously diagonalizes the Hessian of the scalar potential, and yields
kinetic terms that are canonical (i.e. the Kähler metric evaluated at the vacuum is the
identity). This transformation and several other technicalities are discussed in detail in
section 3, but here we finish outlining the strategy in broad strokes.
To minimize a function numerically we must begin by providing a guess for the vacuum
location. Vacua are known to accumulate near the aforementioned special points in the
moduli space, especially near the conifold point. We focus our search there. Moreover, we
look specifically for zeros of the scalar potential, which are solutions to the SUSY condition
Dz W = Dτ W = 0. This restriction both dramatically reduces the computational expense
of searching for vacua by decreasing the (real) dimension of the space over which the
function needs to be minimized from four to two, as well as enables us to compute a guess
location given a choice of fluxes (which is essential for numerical minimization). These are
not SUSY vacua in the traditional sense because we do not require that the superpotential
itself vanish at the vacua.
The first of these simplifications is due the fact that the two SUSY conditions imply
that the vacuum value of the axio-dilaton τSU SY for a given choice of fluxes is an explicit
function of the complex structure vacuum location. In particular,
F · Π̄(z̄SU SY )
τSU SY = (2.69)
H · Π̄(z̄SU SY )
So, we evaluate the axio-dilaton in the function we seek to minimize, |Dz W |2 , at τ =

τSU SY (z). We then need only minimize over the variation of two real fields (the real and
imaginary parts of z). The guess location, zguess , for the given set of fluxes can be computed
straightforwardly by using the near conifold period expansions in the period vectors and
the Kähler potential below,

F · Π̄(z̄)
Dz W (z, τSU SY (z))= F − H · (Π0 (z) + Kz Π(z)) = 0, (2.70)
H · Π̄(z̄)
We compute the leading order solution to the above, which amounts to retaining the
log(z − 1) and constant terms, and dropping everything O(z − 1). The resulting zguess is
given in terms of the flux integers and period expansion coefficients in section 3.
To proceed further, it is essential to have high accuracy approximations to the period
functions near the conifold point. The Meijer-G functions, with respect to which the periods
and their derivatives can be expressed, are generally slow to evaluate numerically. As the
singularities of the Meijer-G’s are approached (for example the branch point singularity
for Π00 (z), and terms ∼ (z−1)1
k−1 for its k
th order derivative) this becomes a significant
obstacle. We not only evaluate such expressions multiple times while searching for a single
– 13 –
vacuum, but we then must compute the Taylor coefficients at the near conifold vacuum
found. This will involve many additional evaluations of increasingly divergent (due to the
derivatives taken) Meijer-G’s near their singularities. The tremendous number of times
the search algorithm needs to be run to find a sufficiently large random sample of vacua,
and the subsequent computation of the Taylor coefficients makes it essential to have high
accuracy fast approximations to the period functions near the conifold point.
Additionally, we note that such approximations are also needed near the large complex
structure point (z = 0). The minima we are searching for typically have basins of attraction
that narrow sharply near the minimum. Though a particular set of fluxes may yield a guess
in the neighborhood of the conifold point, and so be worthy of pursuing, the guess may
lie far up the minimum’s basin outside the basin’s thin throat. Iterative minimization
procedures work by taking a steps in the direction of the gradient of the function being
minimized. A narrow basin that then flattens out can result in significant overshooting of
the minimum during the first steps. The search region needs to be large enough to contain
these initial sweeps as it ping-pongs around the minimum, and eventually spiral into it.
The surrounding buffer area we need includes the large complex structure point. The
behavior of the periods here is well known, Πi → (z log(z))i . Due to the branch cuts in
the periods, many of the Meijer-G’s in the expressions we seek to minimize are singular.
So, when searching for vacua we use “patched period functions”– piecewise defined fast
approximations to the exact expressions in terms of the Meijer-G’s. Outside the neighbor-
hoods of both the large complex structure point and conifold point (where the expansions
are used), we build interpolating functions by evaluating the exact periods on a grid. The
entire search region showing the neighborhoods where each of the three type of approxima-
tions to the period functions are used is found in Figure 2. We postpone further discussion
of the search algorithm until the Calculational Approach section, and now turn to the
computation of the fast approximations to the period functions.
2.3 Period Expansions for the Mirror Quintic

Three of the mirror quintic’s four period integrals (in the integral and symplectic basis) are
analytic in the neighborhood of the conifold point. These are the intersecting pair Π1 (z)
and Π2 (z) which are nontrivial at the conifold point, and Π3 (z) which vanishes because it
is an integral over the collapsing three cycle. These can be approximated straightforwardly
by truncating their Taylor series. We write,
q
X
Π1 (z) = bn (z − 1)n (2.71)
n=0
Xq
Π2 (z) = cn (z − 1)n (2.72)
n=0
Xq
Π3 (z) = dn (z − 1)n . (2.73)
n=1
The periods and their first derivatives enter the scalar potential. Since we seek to
collect up to fourth order Taylor coefficients of the scalar potential at the vacua located,
– 14 –
Figure 2. We search for no scale vacua in the square portion of the complex plane for z depicted
above. The three regions where we use different fast approximations to the period functions are
shown using different colors. The near-conifold patch consists of the disk of radius 0.5 centered at
the conifold, z = 1. The portion of the disk of radius 0.8 centered at the LCS point, z = 0, that
is not contained within the near-conifold region is shown in purple. Here we use the 12th order
expansions about z = 0 obtained directly from Mathematica. Lastly, an interpolating function built
from discrete Meijer-G data is used in the remaining portion of the square search region, shown in
light green. Branch cuts are indicated by the red zigzag lines, with the one emanating from the
conifold point along the positive real axis applying to Π0 , and those emanating along the negative
real axis from the LCS point of relevance to all periods excluding Π0 .
we will be evaluating fifth order derivatives of the periods near z = 1. For the sake of
accuracy we take q = 8. The expansion coefficients can be found in Table 3 of Appendix
A.
The remaining period, Π0 , picks up one factor of its intersecting partner, Π3 , for each
loop about the conifold point. This transformation property of Π0 restricts its form to
log(z − 1)
Π0 (z) = Π3 (z) + f (z) (2.74)
2πi
for some function f (z) that is analytic at the conifold point. Before proceeding we note that
we shall henceforth include an overall minus sign in front of the argument in the logarithm
in the expansion of Π0 so that all explicit values of the expansion coefficients correspond to
a consistent choice of branch cuts in Mathematica. Specifically, the expressions given for
the periods in terms of the Meijer-G’s use the convention of branch cuts emanating from
the conifold point along the positive real axis, and from the large complex structure point
along the negative real axis. Since Mathematica’s logarithm function places the branch cut
– 15 –
along the argument’s negative real axis, it is necessary to include a minus sign in front of
the log’s argument in the expansion, eq. 2.74, to flip it from (−∞, 1] to [1, +∞).
Note that the argument of the logarithm in Π0 ’s expansion may be rescaled freely
because this amounts to a relabeling of analytic terms. The righthand side of eq. 2.74 is
equivalently written as
log(−(z − 1)) Π3 (z) log(−(z − 1))

Π3 (z) + f (z) − = Π3 (z) + f˜(z). (2.75)
2πi 2 2πi
The shifted function, f˜(z), is still analytic because Π3 is. Relabeling f˜(z) by f (z) we have
the same expression as eq. 2.74, only with a negative sign in front of the (z − 1). We choose
however to keep the “extra” analytic term, −Π3 /2, separate and take the form,

log(−(z − 1)) 1
Π0 (z) = Π3 (z) − + f (z). (2.76)
2πi 2
It is a convenient choice for performing checks of the accuracy of the Π0 approximation

because the factor multiplying Π3 does not change sign (the range of the imaginary part
of the logarithm function in Mathematica is [−π, π]).
Since we have a polynomial expansion for Π3 , the task of obtaining a fast approximation
for Π0 amounts to finding one for the unknown f (z). Since we have no special restrictions
to this function’s properties aside from analyticity, the simplest approximation is a Taylor
series about z = 1. We write
q
X
f (z) = an (z − 1)n . (2.77)
n=0
The zeroth coefficient is the value of Π0 at the conifold point, which is trivial to compute.
The higher order coefficients are more difficult.
Although each
1 dn f 1 dn Π0 1 dn

1
an = n
= n
− n
Π3 (z) log(−(z − 1)) − (2.78)
n! dz z=1 n! dz z=1 2πi dz
2
z=1
is finite, the fact that the divergences between the two terms on the righthand side cancel
exactly at each order is lost if one attempts to evaluate (numerically) the righthand side
exactly the conifold point. Mathematica’s “Limit” function cannot be used to remedy this.
However, the next coefficient, a1 , is nonetheless easily obtained numerically by exploiting
the weakness of the divergences that cancel in the first derivative, which are logarithmic.
In particular, to leading order in s ≡ (z − 1), a1 is given by,

dΠ0 log(−s) id1 d1
a1 = − d1 + + + O(s log(s)). (2.79)
dz z=1
2πi 2π 2
We obtain an approximate value for a1 by dropping the O(s) terms which involve higher
order (unknown as of now) ai ’s and evaluating the remaining known expressions on the
righthand side sufficiently close to the conifold point that errors due to the truncation are
– 16 –
negligible. Taking the form s = e−t , the negligibility of such errors at a finite t = t∗ is
ensured if the value of ã1 (t) defined by,
id1 d1 log(−e−t )
ã1 (t) = Π00 (1 + e−t ) ++ − d1 (2.80)
2π 2 2πi
converges within the relevant precision one is using for t → t∗ . Such convergence is exhib-
ited in Table 1.
t ã1 (t)
2 0.0082657465 − 0.1488734062i
3 0.0143193561 − 0.1662234075i
4 0.0193211265 − 0.1734738528i
5 0.0221685901 − 0.1762752502i
6 0.0235728366 − 0.1773249249i
7 0.0242171435 − 0.1777137131i
8 0.0245004660 − 0.1778570993i
9 0.0246216047 − 0.1779098968i
10 0.0246723700 − 0.1779293266i
12 0.0247018694 − 0.1779391052i
14 0.0247067045 − 0.1779404287i
16 0.0247074729 − 0.1779406078i
20 0.0247076106 − 0.1779406353i
25 0.0247076138 − 0.1779406359i
Table 1. Depiction of the the convergence of the expansion coefficient a1 computed numerically.
The third coefficient in f ’s expansion, a2 , can be obtained in a similar fashion. We

write
3id2 id1 et d2 log(−e−t )

1 00 −t
ã2 (t) ≡ Π0 (1 + e ) + d2 + + − (2.81)
2 2π 2π π
However, here it is essential to use high-digit accuracy computations when evaluating the
righthand side for a given value of t. This is because we’re extracting a small number by
taking the difference of two large numbers, Π000 (1 + e−t ) and the term proportional to et in
eq. 2.81. Table 2 displays the convergence of a2 .
Clearly this strategy is limited to the lowest expansion coefficients. At each higher order
the righthand side will involve evaluating increasingly divergent terms near the conifold
point and extracting an ever (comparatively) smaller difference. To obtain the higher order
coefficients we instead derive a recursion relation for the an ’s by using the fact that both
Π0 and Π3 are solutions to the Picard-Fuchs equation. Specifically, since the Picard-Fuchs
equation is linear, f (z) must satisfy

log(−(z − 1)) 1
ÔP F [f (z)] = −ÔP F Π3 (z) − . (2.82)
2πi 2
– 17 –
t ã2 (t)
2 0.00213219053 + 0.09247047120i
3 −0.00142768524 + 0.11114853496i
4 −0.00594277215 + 0.11937168288i
5 −0.00880198201 + 0.12261391639i
6 −0.01027763172 + 0.12383805148i
7 −0.01097135473 + 0.12429273911i
8 −0.01128102108 + 0.12446060393i
10 −0.01147126928 + 0.12454519737i
12 −0.01150432390 + 0.12455665210i
16 −0.01151065730 + 0.12455841226i
20 −0.01151081435 + 0.12455844450i
24 −0.01151081798 + 0.12455844509i
28 −0.01151081806 + 0.12455844510i
Table 2. Depiction of the convergence of the expansion coefficient a2 computed numerically.
The righthand side is a known, albeit messy, analytic function due to the fact that
Π3 ’s near conifold expansion coefficients are known. Note that because OP F is a fourth
order differential operator the righthand side will contain terms that are individually di-
vergent (from derivatives acting on the log times the lower order terms in Π3 so as to yield
contributions ∼ s−1 and s−2 ). The divergences, though, exactly cancel due to the strict
relationship among Π3 ’s coefficients, owing to the fact that it satisfies
ÔP F [Π3 (z)] = 0. (2.83)
Thus, we need only express the lefthand side of eq. 2.82 as a power series in s (whose
coefficient at a given order is a linear combination of a subset of the {ai }) and ensure we
have enough of the lowest order coefficients to generate the rest. It will turn out that the
zeroth order term on the lefthand side involves f ’s four lowest order coefficients, and all
those of order n > 0 involve the n − 1th and subsequent four coefficients: {an−1 , .., an+3 }.
Hence, the a0 , a1 and a2 obtained numerically as described above will be sufficient to start
off the recursive procedure, and provide us as many coefficients as we need.
To that end, we rewrite the Picard-Fuchs differential operator as follows,
ÔP F = −sδ 4 − s(k1 δ 3 − k2 δ 2 − k3 δ − k4 ) − (k1 δ 3 − k2 δ 2 − k3 δ − k4 ) (2.84)

4
X 4
X
= −s ki δ 4−i − ki δ 4−i
i=0 i=1
– 18 –
where the ki are the constants,
k0 = 1 (2.85)
4
X
k1 = αi (2.86)
i=1
X4 4
X
k2 = αi αj (2.87)
i=1 j=i+1
k3 = α1 α2 α3 + α1 α2 α4 + α2 α3 α4 (2.88)
k4 = α1 α2 α3 α4 . (2.89)
Next note that δ acts on sn as,

d
δsn = (s + 1) (z − 1)n
dz
= (s + 1)nsn−1
= n(sn + sn−1 ).
By repeatedly applying this rule each of the δ j sn terms can be computed. For instance,
δ 2 sn = δ[n(sn + sn−1 )]
= n(n(sn + sn−1 ) + (n − 1)(sn−1 + sn−2 ))
= n2 sn + n(2n − 1)sn−1 + n(n − 1)sn−2 .
After similarly obtaining δ 3 and δ 4 on sn , collecting terms and shifting indices of summa-
tion, the Picard-Fuchs operator’s action on f (z) can be expressed in the form,
∞
X
ÔP F [f (z)] = C−1 (n)an−1 + C0 (n)an + C1 (n)an+1 + C2 (n)an+2 + C3 (n)an+3 sn ,
n=0
(2.90)
where a−1 ≡ 0 and the constants Cj (n) are also functions of the ki . Though a tedious
exercise, the Cj (n) can be obtained straightforwardly with the aid of Mathematica. A
similar procedure yields the expansion of the righthand side of eq. 2.82 thus completing
the recursion relation. The resulting values for the an are given in Table 4 located in
Appendix A.
The analogous expansions about the large complex structure point are far easier to
obtain numerically, despite the fact that three as opposed to one of the cycles transform
nontrivially upon circling it. Being finite but multiple-valued, the form of the corresponding
three periods involve linear combinations of powers of z log z. As mentioned at the end of
subsection 2.3, the leading order behavior of the jth period in our notation (that is the
term that contributes the most divergent term to the period’s derivative) is (z log z)j . This
form includes the behavior of the analytic period, Π0 , which is u 1.
Since each of the other three periods has its own residual analytic term (analogous to
f (z) in eq. 2.74) as well as an additional subleading logarithmic term at each period index
– 19 –
j higher, the near large complex structure expressions are more complicated. Nonetheless,
the approximations can be obtained using Mathematica’s “Series” function, due to the
special properties of Meijer-Gs. Essentially, one can expand the Gm,n p,q in eq. 2.55 about
z = 0 to yield a series of integrals whose individual terms are easy to evaluate.
2.4 Structure of the Hessian

The masses of the moduli in the effective field theory associated with a given vacuum are
contained in the Hessian of the scalar potential specified by the particular flux configuration.
When evaluated at the vacuum location in the moduli space, the eigenvalues of the Hessian
in canonically normalized field coordinates are the squares of the masses in the effective
theory. No scale vacua have additional structure built in from the outset as compared to
ordinary general N = 1 supersymmetric theories.
We begin by expressing the general N = 1 scalar potential – that which includes
the Kähler moduli and does not assume cancellation of the 3|W |2 term – and its partial
derivatives in terms of the appropriately invariant quantities. It is convenient to adopt the
standard notation for the Kähler and geometrically covariant derivatives of the superpo-
tential, up to third order,
FI ≡ DI W ; ZIJ ≡ DI DJ W ; UIJK ≡ DI DJ DK W (2.91)
Note that FI is not to be confused with the amount of R-R flux wrapping a particular
3-cycle of the compact manifold. We express general N = 1 scalar potential then as,
V = eK (FI F̄ I − 3|W |2 ). (2.92)
where indices run over all moduli.

Due to the Kähler invariance of eq. 2.92 we may trade partial derivatives for covariant
ones and obtain the following covariant expressions [3].
∂I V = eK (DI DJ W )F̄ J − 2FI W̄ = eK ZIJ F̄ J − 2FI W̄

(2.93)
∂I ∂J V = eK (DI DJ DK W )F̄ K − DI DJ W̄ = eK UIJK F̄ K − ZIJ W̄

(2.94)

∂I ∂J¯V = eK −RI JK K L̄
¯ L̄ F̄ F + KI J¯FK F̄
K
− FI F̄J¯ + (DI DK W )(D̄J¯D̄K W̄ ) − 2KI J¯|W |2
(2.95)

= eK −RI JK K L̄
¯ L̄ F̄ F + KI J¯FK F̄
K
− FI F̄J¯ + Z Z̄I J¯ − 2KI J¯|W |2 . (2.96)
When the no scale cancellation takes place, and the SUSY condition for the remaining
dynamical moduli in the theory is imposed the nontrivial components of the Hessian reduce
to,
∂I ∂J V = 2eK W̄ ZIJ (2.97)

K 2

∂I ∂J¯V = e Z Z̄I J¯ + KI J¯|W | . (2.98)
where Z Z̄ is defined with the contraction of one holomorphic and one anti-holomorphic
index using the (inverse) Kähler metric, and indices now run over only the axio-dilaton
and complex structure moduli.
– 20 –
Next choose a basis for the complex moduli fields that is orthonormal with respect to
the vacuum Kähler metric,
KI J¯|vac = δI J¯. (2.99)
Such a basis is only unique up to unitary transformations. An arbitrary choice will not
in general simultaneously diagonalize Z can Z̄ can . Note that while Z and Z̄ are complex
and symmetric, Z Z̄ is Hermitian and positive definite, and is thus related to the diagonal
matrix containing N = 1 + h2,1 eigenvalues by a unitary transformation. In canonical
coordinates we write,
Z can Z̄ can = U Σ2 U † (2.100)
and express the eigenvalues as the squares of real positive numbers λi . The columns of
the unitary matrix, U , are of course the corresponding eigenvectors of the canonical Z Z̄.
(An excellent resource for understanding no scale structure and its implications is [10].
We’ve adopted their notation in our abridged calculation here in order to facilitate its use
to readers seeking greater detail).
Thus, the Hessian in canonical coordinates,
!
(Z Z̄) can + 1 |W | 2 2Z̄ can W
I J¯ 2×2 I¯J¯
Hcan = eK U † U (2.101)
can can 2
2ZIJ W̄ ¯ + 12×2 |W |
(Z Z̄)IJ
can be diagonalized by a unitary transformation defined in terms of the 2N×2N-matrix U,

!
U 0
U= . (2.102)
0 U†
In particular, one can rewrite eq. 2.101 as,

!
Σ2 + 12×2 |W |2 2ΣW
Hcan = eK U † U (2.103)
2ΣW̄ Σ2 + 12×2 |W |2
A permutation of the rows and columns of matrix between U and U † in eq. 2.103 casts
it as block diagonal, with each of the N 2×2 blocks having the form,
!
λ2i + |W |2 2λi W
. (2.104)
2λi W̄ λ2i + |W |2
The eigenvalues of the Hessian then come in pairs, namely those of each block times the
overall factor of eK ,
m2i± = eK (λi ± |W |)2 . (2.105)
The fact that the scalar masses in no scale supergravity take the form of eq. 2.105 does
not ensure a discernible pattern among the masses of an ensemble of vacua will emerge.
Which pattern is present, if any, depends on the relative scales of the λi as well as how they
compare to the magnitude of the superpotential at vacua. We shall see that a pronounced
– 21 –
hierarchy and splitting of the field space consistent across the ensemble arises due to the
special features of the conifold point, where our vacua accumulate. This is discussed at
length in section 4. It is worth remarking that no association of one particular kind of
moduli field (or a particular linear combination) with a heavy or light mass pair, nor the
existence of separated mass pairs, is imposed by eq. 2.105.
3 Calculational Approach
There are two components to our procedure for generating a random sample of effective
field theories in the mirror quintic’s moduli space. In this section we discuss each of these
in turn.
3.1 Generating a Random Sample of Vacua

Recall that we make the assumptions that the effect of O3-planes on the compact geometry
is negligible at the level of the 4d action for the moduli, that all Kähler moduli are stabilized,
and that the backreaction from fluxes (warping) can be ignored thus preserving the no scale
structure given by compactifying type IIB supergravity on a Calabi-Yau.
The effective action for the two remaining complex scalars takes the form,
Mp2
Z
Sef f = d4 x Kzcsz̄ ∂µ z∂ µ z̄ + Kτaxτ̄ ∂µ τ ∂ µ τ̄ − V (z, τ, z̄, τ̄ ), (3.1)
2
where both components of the field space metric are obtained by taking one holomorphic
and one antiholomorphic derivative of the relevant Kähler potential. The Kähler potential
for the complex structure is given by eq. 2.49, which is known explicitly in terms of Meijer-
G functions via eqs. 2.60–2.63 and 2.51–2.54. That for the axio-dilaton is obtained from
eq. 2.19.
The scalar potential, V , is not a holomorphic function of z and τ . It is however defined
in terms of the holomorphic superpotential, W (z, τ ),
Mp2 Kcs (z,z̄)+Kax (τ,τ̄ ) z z̄

K Dz W D̄z̄ W̄ + Kτ τ̄ Dτ W D̄τ̄ W̄ .

V = 2 e (3.2)
4πV0
The superpotential is parameterized by eight integers indicating the amount of R-R and
NS-NS fluxes wrapping/piercing each of the mirror quintic’s four 3-cycles. In particular,
we write the superpotential as in eq. 2.47; a linear combination of the mirror quintic’s
period integrals in a symplectic basis (the Πi ’s).
Solutions to the SUSY condition, Dz W = Dτ W = 0, are zeros of the scalar potential
and thus are global minima of the theory. We search specifically for such solutions by
randomly scanning through models defined by eq. 3.1, that is by randomly drawing eight
flux integers. For simplicity, we assume a flat measure for the fluxes, and draw from the
interval, [−20, 20]. Once the set of fluxes is drawn, the corresponding superpotential can
be built, and the zeros of DI W can be searched for numerically.
The SUSY condition for the axio-dilaton implies it is an explicit function of the com-
plex structure vacuum location, specifically that in eq. 2.69. By evaluating the SUSY
– 22 –
condition for the complex structure, Dz W = 0, at τ = τSU SY (z) we accomplish an impor-
tant reduction in the computational expense of finding vacua numerically – we need only
minimize a single (semipositive definite) function of two real variables, namely,
2
u(x, y; {Fi , Hi }) = (F − τSU SY (x + iy)H) · Π0 (x + iy) + Kz Π(x + iy) .

(3.3)
where Kz is of course also a function of the real and imaginary parts of the complex
structure, x and y. Specifically,
Π† (x − iy)Q−1 Π0 (x + iy)
Kz = − . (3.4)
Π† (x − iy)Q−1 Π(x + iy)
For a given set of fluxes the function, u, defined in eq. 3.3 can be assembled and
minimized directly in Mathematica using its FindMinimum function, provided that an
initial starting point (for x and y) is specified. Vacua are known to accumulate near the
conifold point, z = 1, so it is reasonable to target our search here. Since we have simple
expansions for the period functions here, namely eqs. 2.71–2.74, we may expand the Kähler
covariant derivative of the superpotential with respect to the complex structure in z − 1.
The term involving Π00 in Dz W will yield a logarithm of z − 1. This is the most divergent
term. By retaining only the logarithmic and O(1) terms in Dz W = 0 we can solve for z in
terms of the fluxes. The result is,
zguess = 1 − eϕ (3.5)
!
d1
βa0 − a1 − 2 F2 − tH2 βb0 − b1 F1 − tH1 βc0 − c1 F0 − tH0
ϕ = −1 + 2πi + + −
d1 F3 − tH3 d1 F3 − tH3 d1 F3 − tH3
(3.6)
a¯0 d1 − c¯0 b1 + b0 c¯1
β=− (3.7)
b¯0 c0 − c¯0 b0
F3 a¯0 + F2 b¯0 + F1 c¯0
t= (3.8)
H3 a¯0 + H2 b¯0 + H1 c¯0
Note that the axio-dilaton has been evaluated at τSU SY (z; F, H) and expanded as well. It
is the the O(1) constant, t, above.
There is no guarantee that a random choice of fluxes will have a near conifold minimum
that satisfies the SUSY condition. In fact, the vast majority do not. Whether this is the
case can be determined from the initial guess. If zguess − 1 is so large that O(z − 1) terms
dominate log(z − 1), then the expansion that lead to zguess was not valid to begin with,
and so the eight fluxes are redrawn.
To summarize, then, the steps of our search algorithm are:
1. Randomly draw eight integers independently from the interval [−20, 20].
2. Compute the guess via eq. 3.6. If it is more than 0.5 away from conifold redraw
the fluxes. Otherwise construct u(x, y; F, H) using the patched period functions.
3. Minimize u using FindMinimum, with the real and imaginary parts of the guess as
the starting point. (We further invoke the option that limits the search region to avoid
– 23 –
Mathematica searching far away from the region of interest when there is no near conifold
minimum, and/or extrapolating the periods so as to give artificial minima).
4. If a minimum is found, and that minimum is sufficiently close to the conifold, we
collect a variety of useful information including the list of fluxes, the minimum’s location,
the magnitude of u, and the axio-dilaton’s location τSU SY (zmin ).
5. Repeat.
Lastly, it is necessary to filter these local minima of u. Since the imaginary part of the
axio-dilaton is proportional to the inverse of the string coupling, any minima found during
the search that have negative Im(τSU SY (zmin )) are unphysical. Removing these, the list
of potential vacua is approximately cut in half. Next, only the zeros of u should be kept,
as the evaluation of the axio-dilaton at 2.69 assumes vacua solve DI W = 0. We eliminate
minima whose u is above a threshold of 10−6 .
To compare vacua properly, particularly with regard to their locations in the the τ
field space, we need to perform an SL(2, Z) transformation that maps each vacua’s τ value
into the fundamental domain. A set of four integers {a, b, c, d} is found such that the
transformation,
aτ + b
τ→ (3.9)
cτ + d
1
maps the axio-dilaton into the strip in the upper half-plane with both Re(τ ) < 2 and
|τ | > 12 . The vacuum’s fluxes are then mapped as follows,
Fi → aFi + bHi
Hi → cFi + dHi .
This is the four dimensional incarnation of the original SL(2, Z) symmetry enjoyed
by the 10d type IIB supergravity action. In fact, one reason for formulating the theory
in terms of the axio-dilaton is so this symmetry is made manifest. The transformation is
stated for the total 3-form flux as,
G(3) → cG(3) + d. (3.10)
Performing the transformation also enables us to check for duplicate vacua. The transfor-
mation is not 1-to-1, so vacua with different fluxes prior to mapping may actually corre-
spond to the same vacuum in the fundamental domain. Though possible, no instances of
double counting were found among the vacua identified with our algorithm.
Finally, we impose the type IIB tadpole condition on the fluxes, eq. 2.30. This integral
expression can be restated conveniently in terms the number of orientifold planes and D3-
branes, and the dimensionless fluxes wrapping the mirror quintic’s 3-cycles as
1
F · Q · H = NO3 − ND3 , (3.11)
4
where Q is the intersection matrix, and F and H are the vectors containing the four R-R
and four NS-NS flux integers. The condition is often stated as an inequality by defining
– 24 –
Figure 3. Complex structure and axio-dilaton vacuum locations for the random sample of 1358
vacua.
the maximum of the righthand side as the positive number Lmax ,
F · Q · H ≤ Lmax . (3.12)
We make an admittedly arbitrary choice of Lmax = 300, and dispose of vacua whose
fluxes combine via Q to violate this threshold. We are interested in the statistical features
of flux vacua in the mirror quintic’s moduli space as a probe of the landscape more broadly.
So long as our results are not sensitive to the particular choice of Lmax , we believe it is
reasonable to relax the condition. We find this is the case and so proceed without concern
for the actual maximum number of orientifold planes the mirror quintic can support.
The results for the masses and couplings given in the following section are for the
largest random sample of vacua we found using this search algorithm and filtering. It
consists of 1358 near conifold vacua whose complex structure and axio-dilaton’s locations
are shown in Figure 3.
3.2 Computing Coefficients

Now that we have a random sample of flux vacua, we turn to our second computational
task – obtaining the masses and couplings to quartic order in the corresponding ensemble of
effective field theories. These are the data whose statistics we want to analyze. For a given
model, the nth order couplings are the nth order Taylor coefficients of the corresponding
scalar potential expanded about the model’s vacuum and transformed appropriately so
that the kinetic terms in all the effective field theories are canonical.
Though the vacuum axio-dilaton coordinate location in each model is fixed in terms
of the vacuum’s complex structure location, the axio-dilaton is a full degree of freedom.
We fixed it as an explicit function of z in our search algorithm as a short-cut to finding
– 25 –
the location of minima of the scalar potential. Here we leave τ as a variable in the scalar
potential, and so have two complex degrees of freedom. The Hessian then is a 4 × 4-matrix,
and the four masses come in two pairs due to the special structure of the mass matrix in
no scale models, as we reviewed in section 2.4.
This structure is relevant to the higher order couplings because we need to report
them in a basis which not only yields canonical kinetic terms but also diagonalizes the
mass matrix. In this subsection we first discuss the field redefinitions,
{z, τ, z̄, τ̄ } → {y1 , y2 , y3 , y4 } (3.13)

(z, τ ) ∈ C2 , y ∈ R4 (3.14)
that accomplish this, and we then the numerical algorithm for evaluating the Taylor co-
efficients. Though the original complex coordinates are the simplest in which to evaluate
the Taylor coefficients because we have expansions for the periods in z − 1, it will still be
necessary to design an efficient algorithm. This somewhat tedious exercise is discussed in
the second half of this section after defining the specific transformation that is applied to
each tensor of Taylor coefficients calculated with the algorithm.
To that end, we define the column vector of fields in our original complex basis,
 
z
 
τ 
Φ ≡  . (3.15)
 
 z̄ 
 
τ̄
The effective field theory is obtained by expanding about a homogeneous background,

Φvac . We begin by writing, Φ = Φvac + BΨ, where the matrix B will serve to canonically
normalize the kinetic terms. We have,
L(Φ) = L(Φvac ) + ∂µ Ψ† B † (Gvac + O(Ψ))B∂ µ Ψ − V (BΨ) (3.16)

1
= Lvac + ∂µ Ψ† B † Gvac B∂ µ Ψ − Ψ† B † M BΨ + O(Ψ3 ) (3.17)
2
where G(Φ) is the Hermitian matrix containing the components of the Kähler metrics.
Specifically,  
Kz z̄ 0 0 0
 
 0 K
τ τ̄ 0 0 
G(Φ) =  . (3.18)
 
 0 0 Kz̄z 0 
 
0 0 0 Kτ̄ τ
and Gvac is that evaluated at the vacuum.
The effective field theory will have canonical kinetic terms provided
B † Gvac B = 1. (3.19)
This is easily accomplished by rescaling the fields. A normalized complex basis, which we
denote {ξ, σ}, will be useful in discussing our results so we define one during this otherwise
– 26 –
intermediate step. It will be convenient, in addition, to shift the normalized complex
structure field so that it is zero at the conifold. Thus we write,
z = 1 + C1 ξ , τ = C2 σ (3.20)
p p
where C1 and C2 are the constants 1/ Kz z̄ |vac and 1/ Kτ τ̄ |vac , respectively (we drop the
superscripts on the Kähler potentials indicating the complex structure and axio-dilaton as
its diagonal form in z and τ renders them superfluous in the metric). Then,
 
ξ − ξvac
 
σ −σ 
vac 
Ψ≡ . (3.21)

 ξ¯ − ξ¯vac 
 
σ̄ − σ̄vac
The matrix M contains the partial derivatives of the scalar potential (in the original
coordinates) evaluated at the vacuum, but with the appropriate ordering so that it is
Hermitian. We take the ordering (z, τ, z̄, τ̄ ) for the columns, so our rows have the barred
ordering (z̄, τ̄ , z, τ ). Using the like orderings for columns and rows will not yield a Hermitian
matrix, only a symmetric one. We included the subscripts in the definition of G in part to
emphasize this.
The “rescaled” mass matrix, B † M B, is not in general diagonal. This is because the
scalar potential nontrivially mixes the complex structure with the axio-dilaton so that
mixed partials, like ∂z ∂τ V , do not vanish at the vacuum. We choose first to transform to
a set of four real fields, and then to diagonalize the resulting real and symmetric matrix
by an orthogonal transformation (for reasons that will become clear momentarily.)
We express Ψ as
 
Re(ξ − ξvac )
 
1  Im(ξ − ξvac ) 

Ψ = T X, where X = √  . (3.22)
2 Re(σ − σvac ) 

Im(σ − σvac )
Note that T it is unitary. Lastly, we take
X = OY (3.23)
where O is the orthogonal matrix containing the (normalized) eigenvectors of T † B † M BT

as columns. That is,
4
X
T T † † T
Y O T B M BT OY = Y DY = m2i yi2 (3.24)
i=1
where D is diagonal.
– 27 –
The full transformation to the real basis that simultaneously diagonalizes the Hessian
and canonically normalizes the kinetic terms (locally) is
J˜ : (Φ − Φvac ) → Y
J˜ = (BT O)−1
So, if we compute the rank three and four symmetric tensors of partial derivatives of the
scalar potential at the vacuum in the original complex coordinates,

∂ ∂ ∂
Aijk = V , (3.25)
∂Φi ∂Φj ∂Φk Φvac
we need to transform according to the standard tensor transformation law,
∂Φi ∂Φj ∂Φk

Ai0 j 0 k 0 = Aijk (3.26)
∂Y i0 ∂Y j 0 ∂Y k0
= Jii0 Jjj0 Jkk0 Aijk , (3.27)
with J defined as the inverse of J,˜ and similarly for the fourth order coefficients, Ai0 j 0 k0 l0 as
well. By choosing the particular unitary transformation that diagonalizes the rescaled mass
matrix and yields a basis of real fields, yi , we avoid concerning ourselves about reordering of
the entries of the complex symmetric rank three and four tensors we compute numerically.
This final task, evaluating the Taylor coefficients in the original complex field coor-
dinates, may seem trivial. After all, the vacua reside near the conifold point where the
period functions are polynomial in (z − 1) and/or (z − 1)n log(z − 1), so we never need
evaluate the divergent Meijer-G’s in the scalar potential resulting from derivatives of Π0
(and Π̄0 ). However, the seemingly mundane exercise of symbolically simplifying the near
conifold scalar potential resulting from plugging in the period expansions and its partial
derivatives proves prohibitive.
This is mainly due to the cumbersome way the scalar potential mixes up the periods
and the relative factors of
∂ m Π† Q−1 ∂ n Π
∼ (3.28)
(Π† Q−1 Π)m+n
between its summands. Since we are plugging in eighth order expressions for the periods,
the number of terms that need to be expanded, collected and organized is large. Rather
than try to undo the natural packaging of the period functions, we make use of it.
Our strategy is to express each entry in the tensors we wish to evaluate, the Aijk and
Aijkl , in terms of simple combinations of a small number of blocks. Each block is built out
of smaller elements, which include the periods, their derivatives and combinations thereof
(the Kähler potential for the complex structure and its partial derivatives). For each model
in the ensemble, we evaluate the periods and their derivatives up to fifth order once. The
values in this array are then combined appropriately to obtain the rest of the elements
needed to construct the blocks. The blocks are then assembled into each entry required by
the tensors. Essentially, we are exploiting the fact that many expressions appear repeatedly
within a given entry and across entries, and so we need not evaluate them repeatedly.
– 28 –
The blocks consist of all the Kähler covariant derivatives of the superpotential, FI , and
their partial derivatives up to third order taken with respect to any of the complex fields.
Some of these are zero, for instance ∂τ̄ Fz , but note ∂z̄ Fz is in general non-vanishing. Since
the scalar potential is
V = eK Kz z̄ Fz F̄z̄ + Kτ τ̄ Fτ F̄τ̄

(3.29)
its third and fourth order partial derivatives involve several terms linear in FI . All such
contributions vanish however because the SUSY condition, FI = 0, is satisfied at all the
vacua. By only retaining those terms in Aijk and Aijkl that have at least one partial
derivative on FI and at least one on its conjugate we have more manageable expressions
for each Taylor coefficient that need to be combined.
The individual blocks are compact when expressed in terms of the elements. For
example,
∂z ∂z̄ ∂τ Fz = ∂z ∂z̄ ∂τ (F − τ H) · (Π0 + Kz Π) (3.30)

= −H · (Π00 + Kz z̄ Π0 + Kzz z̄ Π). (3.31)
This approach enables us to compute both rank 3 and 4 tensors for the entire sample of
1358 vacua in time of order tens of minutes.
Lastly, we note that the form of the Hessian outlined in subsection 2.4 is confirmed
by comparing that obtained by direct differentiation with that built from the metric and
(separately constructed) Z and Z̄ matrices in the original noncanonical basis. The percent
deviation between the eigenvalues of the two are on the order of 10−13 .
4 Results
In this section we analyze the distributions of masses and couplings for a collection of 1358
vacua, found using the vacuum hunting algorithm described in subsection 3.1. There is a
great deal of structure built in from the get-go. The task is to untangle the randomness
that is present from that structure. As indicated in subsection 2.4, the no-scale structure
for a theory with N complex moduli is responsible for pairing the 2N scalar masses of the
effective field theory.
The association of each mass pair with a single one of the complex scalars (for us,
either z or τ ) is not expected, a priori, because of the mixing between the axio-dilaton and
the complex structure in the scalar potential.
However, for near conifold flux vacua in the mirror quintic’s moduli space that satisfy
the SUSY condition, DI W = 0, the two scalar fields do approximately separate; ever more
so as the vacuum-to-conifold distance is diminished. Tied to this cleaving of the field
space is also a scale separation between the associated axio-dilaton and complex structure
mass pairs. We observe that such a hierarchy percolates through third and fourth order
couplings. All this structure, nearly universal across our ensemble, can be traced back to
a single quantity: the mirror quintic’s Yukawa coupling.
We will show that the larger the Yukawa coupling, the more exaggerated this structure
becomes. Its singularity, located precisely where vacua accumulate – at the conifold point
– 29 –
– is responsible for the expected pattern of near conifold mass and couplings dominating
the ensemble. We use the qualifier “expected” because there is one random ingredient: the
vev of the superpotential. Depending on your point of view, it muddies otherwise sharply
defined features, or provides the possibility of freedom from rigidity (albeit a vanishingly
small possibility as the conifold is approached).
In this section we first establish the distance of a vacuum from the conifold as the key
quantity controlling the degree to which structure is amplified or diluted. Next, we build
intuition for the mass pairs and their distributions. Finally we present the hierarchies
and correlations observed in the data for cubic and quartic couplings, which similarly
is attributable to the singular dependence on the vacuum-to-conifold distance. For ease
of discussion, we will loosely refer to the magnitude of a vacuum’s canonical complex
structure coordinate, |ξvac |, as its distance to the conifold in moduli space. More precisely,
this distance is a monotonically increasing fucntion of |ξ|, but is not identically equal
to it. During our investigation we developed a Random Matrix Model that accurately
captures this particular combination of both regularity and randomness. We comment on
the possible generalizations of these results to models with more complex structure moduli
in the Discussion section.
4.1 Masses
Since the SUSY condition is satisfied at our vacua, the complex 2×2-matrix ZIJ ≡ DI DJ W
is the matrix of vacuum values of the partial derivatives of FI , specifically,
!
∂τ Fτ ∂z Fτ
Z= . (4.1)
∂τ Fz ∂z Fz
Not all entries in this matrix are independent. It’s form is restricted because there is
no mixing between the complex structure and the axio-dilaton at the level of the Kähler
potential (Kzτ = 0) and also because Kτ τ̄ = −Kτ 2 . These two, together with the SUSY
condition in τ , imply Z has the form,
!
0 Z01
Z= , (4.2)
Z10 Z11
where the entries are complex valued (and Z01 = Z10 ).

The two nontrivial entries, it turns out, are related by a known analytic function when
the canonical basis is used (the fields we labeled ξ and σ, whose corresponding Z matrix
is Z can ). For compactifications of type IIB on general Calabi-Yau the following equation
is valid at solutions to DI W = 0,
ZIJ = FIJK Z̄ 0K (4.3)
in a basis where the fields in the effective action are canonically normalized. The FIJK
in eq. 4.3 are the Yukawa couplings between the Calabi-Yau’s complex structure moduli
– 30 –
and their fermionic counterpart in the effective field theory. Since the mirror quintic has a
single complex structure modulus we have the direct proportionality,
Z11 = F111 Z̄ 01 . (4.4)
where we’ve suppressed the “can” superscripts.

The Yukawa coupling is singular at the conifold point. Its exact analytic form was
found by Candelas and de la Ossa in their seminal papers on the manifold (see for e.g. [17]).
Stated in terms of their complex structure field coordinate before canonical normalization,
ψ, which is related to ours by z = ψ −5 the Yukawa coupling takes the form
2πi 3 5ψ 2

κψψψ = . (4.5)
5 1 − ψ5
Note that while zero and infinity switch under the coordinate transformation between ψ
and z, the conifold point is fixed. The conifold singularity in eq. 4.5 persists through
the coordinate transformation to our F111 , and manifests as one naively expects (as a 1/ξ
divergence) with minor modification. This is because the prepotential from which κψψψ
derives is the same as ours.
It is useful to briefly sketch the calculation of Candelas and de la Ossa in order to
understand the origin of the divergence, as well as its leading order form in our coordinates.
They define a set of functions, “Wronskians”, in terms of derivatives of the prepotential.
The kth Wronskian is given by
dk i d
k
Wk = Z i G i − G Zi (4.6)
dψ k dψ k
where Z i and Gi are an intersecting pair of periods (in an integral and symplectic basis).
Since there are four nontrivial cycles, i ranges from 1 to 2. The prepotential is
1
G = Z i Gi . (4.7)
2
A crucial next step is to identify the Yukawa coupling,
d3 Ω
Z
κψψψ = Ω ∧ , (4.8)
dψ 3
with the third Wronskian. Together with the properties of Calabi-Yau, particularly the
fact that the periods solve the Picard-Fuchs equation, they obtain an ordinary differential
equation for W3 , whose solution is given in eq. 4.5.
When the fields in the effective action are canonically normalized, the Yukawa coupling
receives a total factor of the inverse vacuum Kähler metric for the complex structure raised
to the three halves; one half power from the rescaling of the scalar field and one full power
from the transformation of the fermion (one half power for each of the two factors of the
fermion in the original interaction term ∼ κφχχ).
We’ve already accounted for one half of the total three halves by canonically normal-
izing our z coordinate. This implies the ratio of our Z11can to Z can will have a leading order
01
– 31 –
Figure 4. Plotted in light purple is the ratio of the magnitudes of the entries of the Z matrix for
our vacua (the 11 entry over the 01 entry) on the vertical, against |ξvac | which is a measure of the
vacuum to conifold distance. Special geometry implies that this ratio ought to be the magnitude of
the Calabi-Yau Yukawa coupling. The function plotted in red is a numerical fit of the data to the
leading order form of the mirror quintic’s |F|2 . Note the extremely good agreement between the
two, and the divergence at the conifold point, precisely where vacua accumulate.
behavior of 1/(ξ log ξ), since the Kähler metric goes like log(ξ) near the conifold point,
ξ = 0. In Figure 4 we display the actual vacuum data for the magnitude of this ratio
against the conifold distance. A numerical fit to the leading order form is overlaid in red.
Note the exceedingly tight agreement between the two. Essentially, the ∼ 1ξ dependence
d 3
comes from a contribution ∼ Π3 dξ 3 Π0 , since
d3 d3
Π3 Π 0 = O(ξ) (O(ξ) log ξ + analytic) (4.9)
dξ 3 dξ 3
d3 d2

= O(ξ) O(ξ) 3 log ξ + O(1) 2 log ξ + O(1) (4.10)
dξ dξ
= O(ξ)O(1/ξ 2 ) (4.11)
= O(1/ξ) (4.12)
Our vacua live in a region where |F| >> 1, so Z11 always dominates Z01 . This
is consequential for the mass spectra and coordinate transformation that enters into the
computation of the subsequent higher order couplings. Expressed in terms of the magnitude
of the Yukawa coupling (where we’ve suppressed the “111” indices), the Z Z̄ matrix takes
– 32 –
the form, !
2 0 |F|e−iδ
Z Z̄ = |Z01 | , (4.13)
|F|e+iδ |F|2 + 1
whose eigenvalues come in the pair,
|Z01 |2 2 p
Λ2± = |F| + 2 ± |F | |F|2 + 4 . (4.14)
2
The larger of these is always Λ2+ , so, in our labeling convention for the λi ’s we identify
λ21 = Λ2+ , and λ22 = Λ2− .
Note that in either limit, |F| >> 1, or the reverse, we have Λ2+ >> Λ2− . If |F| >> 1
the eigenvector associated with the larger eigenvalue is almost entirely contained within
the span of the complex structure field, and in the opposite limit within that of axio-
dilaton field. Since we always have the former case, the largest eigenvalue of Z Z̄, λ21 , is
associated always with the complex structure, and the smaller, λ22 , with the axio-dilaton.
An immediate consequence of this is the cleaving of the eigenspace of the Hessian in two.
One subspace is spanned almost entirely by the complex structure and is associated
with the mass pair m21± , while the other is spanned by the axio-dilaton and is associated
with m22± . This is because the 2×2 blocks entering into the diagonalization of the Hessian,
which would otherwise mix these two fields, are approximately equal to the identity. These
2×2 blocks in the complex field coordinates of section 2.4 are U and its Hermitian conjugate.
In the real field coordinates of section 3.2 they form two by two blocks in O, upon a
reordering of rows and columns.
Whether or not the hierarchy in the λi leads to a hierarchy between the two mass pairs
– a heavy complex structure pair and a light axio-dilaton pair – depends on the relative
sizes of |W |, λ1 and λ2 . More precisely, we begin by noting there is no ambiguity about
the heaviest mass. It is always m21+ , which for us is always associated with z. It’s partner
(still associated with z) need not be second heaviest, however. To see why note that
(λ1 − |W |)2 < (λ2 ± |W |)2

λ21 − λ22 < 2|W |(λ1 ± λ2 )
λ1 ∓ λ2 < 2|W |.
So, if half of the gap between the λi is less than the magnitude of the vacuum superpotential
the second heaviest of the four masses is the larger of the axio-dilaton masses, m2+ . If
the average of the λi is also less than the magnitude of the superpotential then the third
heaviest mass is the lighter of the axio-dilaton pair, and the lightest of the four masses
is the lighter of the complex structure pair. To summarize, the naive/expected ordering
among the masses,
m21+ > m21− > m22+ > m22− (4.15)
is realized if the difference condition is not met (large discrepancy between the λi ’s).
The middle two masses swap places,
m21+ > m22+ > m21− > m22− (4.16)
– 33 –
Figure 5. The difference between the λi divided by twice the vev of the superpotential vs. |ξvac |,
illustrating the key quantity in the hierarchy condition. Data points that fall below the dashed
pink line do not satisfy the condition and have a mass hierarchy that differs from the expected one
by at least one swap. Note that as the conifold distance decreases the data points float upwards,
confirming the expectation (based on the divergence of the Yukawa coupling) that the condition
becomes ever more difficult to satisfy as the location of the vacuua approaches the conifold.
if the gap condition is met but the average condition is not. Lastly, if both conditions are
met the lighter complex structure mass shuffles all the way to the bottom of the mass scale,
m21+ > m22+ > m22− > m21− . (4.17)
As we’ve seen, the difference in scale between the two distinct nontrivial entries of Z
diverges as the conifold point is approached. Since the larger the scale difference the larger
the gap between the λi will be, we expect the likelihood of the gap condition being met to
diminish as the conifold point is approached. This is precisely what we find. In Figure 5
we plot λ1 − λ2 divided by twice the magnitude of the superpotential against the conifold
distance for each vacuum. A horizontal line at 1 is indicated by the dashed line, so points
above this line fail the gap condition and the naive order exists, while those below have at
least one rightward shift of m21− down the hierarchy in 4.15.
There are two important observations. First, the vacua migrate upward as the coni-
fold is approached making the condition ever more unlikely to be satisfied, verifying our
expectation. Second, there are nonetheless a few vacua for whom the condition is met.
Specifically, we find 33 out of 1358 such instances, or 2.4%. The image toward the upper-
right of Figure 5 shows the portion of the plot focused near the bottom (with exactly 33
points below the dashed line). The random ingredient that allows for vacua to dip below
– 34 –
Figure 6. A scatter plot of the magnitude of the vev of the superpotential vs. ξvac . Note that
the two bear no significant dependence on one another. Data points become more clustered as one
moves toward the peak or either quantity’s distribution independently.
the threshold for mass swaps is the magnitude of the superpotential. The vev shows no
dependence on the distance of the vacuum from the conifold. This is shown in Figure 6.
We reiterate that in all cases, including these nonconformist 33, the Hessian’s eigenspace
enjoys an approximate separation between the complex structure field space, and axio-
dilaton field space. The angle between the subspace spanned by one of the moduli – ξ or
σ – and that spanned by the two eigenvectors associated with one of the mass pairs – m21±
or m22± – can be computed. In Figure 7 we display the histogram of angles between the
complex structure subspace and the i = 1 mass pair for all vacua. The mean angle is 5.85
degrees, indicating that the subspaces are approximately parallel. The identical statement
holds for the axio-dilaton subspace and the second mass pair. A visual depiction of the
subspaces is included to the right of the histogram, and uses the mean angle.
Now that we have established that this separation between z and τ lines up with the
half-way marker between the masses in virtually all cases, we turn to developing intuition
for each pair. We display the distributions of λ1 and λ2 in Figure 8, and of |W | in Figure
9. We have absorbed a factor of the vev of eK/2 into the definitions of each of these three,
as they are the correct Kähler invariant quantities, i.e. the physically relevant values to
consider.
As expected, the λ1 distribution’s scale is significantly larger than λ2 ’s due to the
accumulation of vacua where the Yukawa coupling diverges. Specifically, we find a difference
of two to three orders of magnitude. The characteristics of the corresponding mass pairs
will depend on the relative sizes of the λi to |W | individually. We find a superpotential
that is approximately one order of magnitude larger than λ2 , but one order smaller than
λ1 (several orders smaller for vacua in the tail).
The resulting two mass pairs are displayed in Figures 10 and 11, with the larger mass
of each couple plotted on the horizontal. We immediately notice that the complex structure
– 35 –
Figure 7. A distribution of relative angle, θ, between the complex structure subspace of the moduli
space and the m21± eigenspace, which is identical to that between the axio-dilaton subspace and the
m22± eigenspace. The fact that the angles for all vacua are small indicates that the former pair are
approximately parallel to each other, and likewise for the latter. A visual aid depicting this split of
the eigenspace is shown to the right using the mean value of this angle, which is 5.85 degrees for
our vacua.
Figure 8. Histograms of the Käahler independent λi for our ensemble of vacua. Estimated distri-
butions obtained numerically are plotted over each histogram in blue.
mass pair looks more tightly correlated than the axio-dilaton pair. This, as we’ll analyze
more precisely later, is entirely an artifact of the difference in scale between the two field’s
pairs; an effect that is exaggerated by the particularly wide range needed to include all of
the z mass data points in Figure 10. The distribution (for both members of the z pair)
peaks at much lower values, around 100. A fairer comparison with the τ masses, which are
more widely/evenly distributed, would come from zooming in and excluding the z masses’
– 36 –
Figure 9. The histogram of the Käahler independent vevs of the superpotential for our ensemble.
The estimated distribution obtained numerically is overlayed.
long tails. A partial zoom is shown in the ellipsoidal window on the right in Figure 10.
A more refined analysis will nonetheless reveal that the two fields have virtually identical
levels of relative degeneracy between the members of their respective pairs.
Turning to the τ data points shown in Figure 11, note firstly that they fill in more
of the triangular half below the diagonal including the region immediately beneath the
diagonal. This indicates that there is a larger variety among the dimensionful mass gaps
for the axio-dilaton, than for the complex structure. There are more instances of near
equality between the masses – in an absolute/dimensionful sense – as compared to those
in the lower range of z’s distribution (there are far more data points along the diagonal
boundary in the axio-dilaton’s scatter plot than in the zoomed in complex structure’s).
There are also more instances of large differences for the τ pairs than the z’s. Clearly, the
latter statement remains true when z’s tail is considered, but the former may not. These
distinctions make sense given the distributions for the λi and |W |. Essentially, the τ masses
are dominated by the superpotential, which has a rather large spread and is not skewed
(roughly Gaussian). This leads to a more uniform distribution horizontally throughout the
triangle.
The lack of space between data points and the diagonal is due to the fact that λ2
peaks very near zero, and decays quickly before its decline steadies around ∼ 0.5. This
increases the frequency of λ2 ’s that are completely negligible compared to |W |, and thus
very nearly equal masses among the given pair. The axio-dilaton pairs’ greater vertical
extent throughout the triangle is due to the combination of the larger spread in λ2 and
|W |, and the fact that the intervals where they are supported partially overlap. Thus, more
instances of close competition between λ2 and |W | occur than for λ1 and W .
These observations are helpful for building intuition, but a comparison of the degree
of degeneracy in the mass pairs of the two fields should involve dimensionless mass gaps,
namely those scaled by the mean of the masses in each pair. Starting with the difference
in the squared masses of the two members in the ith pair, 4λi |W |, one finds the limiting
– 37 –
Figure 10. A scatter plot for the mass pair associated with the complex structure modulus, with
the heavier of the two, m1+ , on the horizontal and the lighter, m1− on the vertical. A dashed line
with slope one is plotted in purple. The portion of the plot focused where the masses distributions
peak (i.e. where the data points cluster) is shown to the right.
Figure 11. The analogous scatter plot for the mass pair associated with the axio-dilaton as that
in Figure 10.
form,
s
∆mi /2 λi |W |
= (4.18)
mi,avg λ2i + |W |2
∆mi /2 λi
λi << |W | : → << 1 (4.19)
mi,avg |W |
∆mi /2 |W |
λi >> |W | : → << 1 (4.20)
mi,avg λi
That the result is the same for both limits simply reflects the fact that one can equally
well view λi as the degeneracy breaking term as one can |W |. A small λi compared to |W |
– 38 –
Figure 12. The cumulative relative mass gaps for each pair of masses, mi± . The pair associated
with the axio-dilaton is shown in on top in the darker shade of blue, while that associated with
the complex structure is shown behind in the lighter turquoise color. The two are nearly identical
meaning that the relative degree of non-degeneracy between the masses in a given pair is distributed
in the same manner across the vacua of our ensemble, regardless of with which of the moduli the
mass pair is associated.
yields a mass pair mi± ≈ |W | ± , and the reverse yields a mass pair ≈ λi ± .
Now, we may consider a probability density for each modulus as a function of the
rescaled half mass gaps. For a given one of the moduli its value integrated over an interval
[a, b] would yield the probability of finding a vacuum for whom that modulus’ associated
masses each lie within (b−a)mi,avg of their mean, mi,avg . We can then consider a cumulative
density function obtained by integrating the probability density from a = 0. In Figure 12
we display the histograms for our data corresponding to the cumulative density functions
(their discrete analogs) for our sample of vacua.
Notice that the complex structure and the axio-dilaton’s histograms are virtually in-
distinguishable. They both achieve 50% within a threshold of 0.26, and continue to rise
together in step with 75% of both fields having scaled gaps within the threshold of 0.40.
This assessment of relative degeneracy, or relative spread, is not evident from looking at
the scatter plots alone.
The structure and patterns we’ve encountered in the masses clearly won’t be replicated
with an ordinary Random Matrix Model where the Hessian for each vacuum is taken to
be Wirshart – a Hermitian random matrix that is positive definite by construction. One
essentially “squares” a random (Wigner) matrix, A, which is not in general Hermitian, by
– 39 –
Figure 13. The left panel shows a scatter plot of the (Kähler independent) magnitude of the 01
entry of the Z matrix in canonical coordinates against the vev of the superpotential. The sharp
quarter-circle boundary√is a manifestation of the tadpole condition, with the radius of the arc being
√
Lmax which for us is 300. Within the region allowed by the tadpole condition the data points
exhibit no correlation. The panel on the right displays a plot of the Z matrix entry against the
remaining control parameter, the conifold distance. No correlation is evident between these either.
multiplying it with its complex conjugate. The entries of the Wigner matrix are taken
to be independent and identically distributed, that is drawn from an O(N 2 ) dimensional
Gaussian.
In light of the analytic form of the Hessian with which we begin, and the limiting
behavior of one of its essential building blocks, Z, for near conifold vacua, we design a
different Random Matrix Model. We’ve seen three (Kähler independent) parameters are
ultimately in control. These are (1) the proximity of a given vacuum to the conifold point,
(2) its value of eK/2 |Z01 | and (3) its value of eK/2 |W |. We should be able to mimic the
actual mass data with a random sample of these triples. The simplest case would be to
treat each parameter independently.
We saw earlier that control parameters (1) and (3) do not appear to depend on one
another, as indicated by Figure 6. A similar scatter plot for (1) and (2) is displayed in the
right panel of Figure 13, demonstrating their lack of correlation. The plot for (2) and (3)
shows a sharp cutoff because the tadpole condition forbids these data points from leaving
the quarter circle. It can be shown that the tadpole condition implies,
|Z01 |2 + |W |2 ≤ Lmax (4.21)
in Gaussian normal coordinates.

√ √
The radius of the arc plotted in Figure 13 is indeed Lmax = 300. Within this region
however the data points vary independently. The empty bands along both axes are simply
– 40 –
Figure 14. The random matrix model data for the artificial complex structure mass pair are
plotted in light blue over that of the actual vacuum data, shown in light pink. The panel to the
right magnifies the portion of the plot where the mass distributions peak.
a reflection of the fact that the two distributions are peaked away from zero, with relatively
little of their support coming from the interval ≈ [0, 5]. Just as with the other two scatter
plots, the density of data points increases as either parameter is pushed towards the value
where its distribution peaks while the other is held fixed. The lack of correlation within
the region suggests we do the following.
First obtain estimated probability densities for the Kähler invariant magnitudes in the
canonically normalized fields, namely, eK/2 |W | and eK/2 |Z01 |, as well as for the conifold
distance. Draw a value from each distribution independently. If it has parameters (2) and
(3) that violate inequality 4.21 dispose of it and redraw the triple until it is satisfied. Then
compute the random eigenvalues, λ21 and λ22 , by evaluating the Yukawa coupling at the
randomly drawn conifold distance, and using it with the random |Z01 | in eq. 4.14.
In Figure 14 we display the resulting scatter plot for the i = 1 random mass pair –
the artificial complex structure pair – atop that from our sample of actual flux vacua for
the full range of masses. The image in the ellipsoidal window zooms in on the range where
complex structure mass distributions peak. The RMM does a good job in reproducing the
data’s features in both regimes: the peak and the tail of the mass distributions. The same
is true for the RMM’s performance with the axio-dilaton mass pair. The two are virually
indistinguishable in their superposed scatter plots, shown in Figure 15.
The scaled mass gap CDFs for the RMM also agree rather well with the actual data,
and is shown in Figure 16. It is worth noting that without the step in the RMM procedure
that eliminates draws that live outside the quarter circle allowed by the tadpole condition
there is a noticeable overdensity of RMM axio-dilaton data points away from the diagonal
in the analogous version of Figure 15. The effect on the complex structure’s scatter plot of
removing RMM tadpole condition is not perceptible.
– 41 –
Figure 15. The random matrix model data for the artificial axio-dilaton mass pair are plotted in
light blue over that of the actual vacuum data, shown in light pink.
Figure 16. The analogous histograms to those in Figure 12 for the random matrix model.
4.2 Couplings
The hierarchy present in the masses, due to the fact that our vacua accumulate where the
Yukawa coupling is singular, persists through the third and fourth order couplings. Since
the basis in which couplings ought to be reported factors into one half associated almost
entirely with the complex structure and the other with the axio-dilaton, we naively expect
each additional index associated with the former at the expense of the latter to involve the
– 42 –
evaluation of increasingly more divergent terms near they singularity.
In particular, one expects four distinct scales to emerge among the distributions of
third order coefficients, and five scales for the fourth order couplings. These correspond to
the 3-choose-2 and 4-choose-2 ways one can differentiate the scalar potential. For instance,
at third order we expect the Ai0 j 0 k0 with all indices related to the complex structure (that
is, equal to 1 or 2 in our convention for the basis of real canonically normalized fields) to
be dominated by the term involving a derivative of the Yukawa coupling. The next highest
scale expected would then be that with two complex structure and one axio-dilaton indices
(3 or 4 in our convention), followed by one complex structure and two axio-dilaton, and
lastly that with all three axio-dilaton.
First we establish that such a hierarchy of scales is realized, and then confirm the ex-
planation in the preceding paragraph is valid by showing that the scale separation becomes
ever more prominent as the vacuum-to-conifold distance diminishes. We then qualitatively
investigate the correlations between the couplings at a given order, and indicate that they
are not the result of the coordinate transformation alone. We accomplish this with the
use of another random construction. Specifically, we generate a set of random rank three
symmetric tensors and transform each by the orthogonal matrices from the actual set of
vacuum data. Though one might expect correlations to be built in by the special structure
of the Hessian’s eigenspace, the fact that none of the correlations present in the mirror quin-
tic data is replicated by the random procedure indicates that they are not the consequence
of diagonalization.
Turning to the hierarchy among the magnitudes of the couplings, the scale separation
can be shown visually by first imagining each of the entries in the third order couplings
for a given vacuum, Ai0 j 0 k0 , as living in one of 64 cells of a 4-by-4-by-4 celled cube. We
have one cube for each vacuum, and its entries take on positive or negative values (with
equal likelihood, as indicated by the roughly Gaussian distributions centered at zero found
for all coefficients. A representative sample of the histograms and estimated distributions
can be found in Figure 17). Taking the labeling convention for the real and canonically
normalized field coordinates defined in subsection 3.2, one 2-by-2-by-2 subblock in, say,
the front-bottom corner of this cube will involve all complex structure related indices. The
subblock diagonally opposite it in the top-far corner will involve all axio-dilaton indices,
and the two types of mixed index subblocks will live interspersed throughout the remaining
6 off-diagonal subblocks.
Next, consider taking the magnitude of the value in each cell and then computing the
median for each entry across the ensemble of cubes. The median is the more appropriate
quantity because the distributions of the magnitudes are heavily skewed, just as the masses
were. We may then represent the cube containing the ensemble’s median values visually
by coloring each cell according to a continuous scheme. The resulting hierarchy is, not
surprisingly, best illustrated using a logarithmic scale. Two views of the resulting cube are
shown in Figure 18, with a color gradient of green to white to pink indicating smallest to
largest.
The cube arranges itself into the four 8-celled subcubes of different scale, which we’ve
described. This is indicated by the green quadrant, which is flanked by much paler green
– 43 –
Figure 17. A representative pair of distributions of the higher order couplings.
Figure 18. The median across the ensemble of the magnitude of the transformed third order
couplings, Ai0 j 0 k0 . Each cell represents one choice for the three indices. i0 = 1 or 2 corresponds to
the complex structure associated eigenvectors, y1 and y2 , whereas i0 = 3 or 4 corresponds to the
axio-dilaton associated eigenvectors, y3 and y4 . The scale is logarithmic with green representing
the smallest median magnitude and pink representing the largest. The “origin,” so to speak, is
located in the bottom right corner of the back face in the view of the cube in the left panel. The
pink 2-by-2-by-2 subcube in this corner contains the all-complex-structure subset of couplings since
the indices are all either 1 or 2. Similarly the green corner diagonally opposite contains the all-
axio-dilaton couplings. The four expected hierarchies based on the leading order behavior of the
Yukawa coupling near the conifold can be seen by the partitioning of the cube into four types of
subblocks each with cell colors in a different regime of the scale: pink, light pink/pale green, light
green, and green. A view of the cube rotated about the vertical axis is shown on the right.
(identical by symmetry) subcubes adjacent to it, the vibrant pink quadrant diagonally
across from the green corner and lastly the (identical) subcubes with pale pink and green
cells that share an edge with the pink corner.
The smallest couplings (green) do indeed reside in the all-axio-dilaton subblock, which
is located at the top left of the front face of the cube in the first view. The pink corner is
in fact the all-complex structure subblock. Its neighboring subcubes – those that share an
– 44 –
edge with it (for instance those directly above and directly to the left of the pink corner in
the front face of the second view) – still have two complex structure indices because they
are in the same 2-cell thick “slice” of the cube, but have only one axio-dilaton index. The
fact that the pale colors in these neighboring subcubes are pinker/less green than the pale
subcubes that neighbor the green axio-dilaton corner means the ξ-ξ-σ couplings are larger
than the other mixed index cubic couplings, ξ-σ-σ.
The hierarchy among the quartic couplings can be visualized in much the same way,
only with a stack of four 64 celled cubes instead of a single one. We show two views of
this hypercube in Figure 19. The blocks are arranged top to bottom according to the first
index, i0 , in Ai0 j 0 k0 l0 ; the top having i0 = 1 and bottom having i0 = 4. The color scheme
here is CMYK with cyan/blue representing the smallest magnitude, followed by purple,
magenta, orange, yellow, gray and finally black indicating the largest. Notice that each
cube in the stack partitions itself into quadrants of four distinct scales (just like the single
cube of third order couplings).
The top pair of blocks then has one additional index associated with the canonical
complex structure coordinate, relative to the bottom pair with the canonical axio-dilaton.
The largest magnitudes (the blackest cells) do in fact fill in the ξ-ξ-ξ subcube of the top pair
of blocks. These are the all-complex-structure quartic couplings. These corner subcubes
each share an edge with (identical) yellow subcubes. Since neighboring subcubes differ by
one index type these neighbors contain couplings with three complex structure indices and
one axio-dilaton. The fact that it is yellow means ξ-ξ-ξ-σ couplings rank second largest.
Across from the black corner subcubes but within the same 2-cell thick slice we have the
couplings that involve one more σ in place of ξ, the ξ-ξ-σ-σ couplings. The fact that they
are orange indicates they are the third largest scale.
The remaining two scales in the hierarchy are displayed by the purple corners of the top
pair of blocks in the stack and the blue corner cubes that are only present in the bottom
two blocks in the stack. The purple corners of the top pair of blocks blocks do in fact
lie diagonally opposite the black corners, making them cells containing ξ-σ-σ-σ couplings.
The subcube located in this same top back corner position in the bottom pair of blocks
in the stack differs from the purple ones of the preceding top pair in the stack by the first
index, making them the all-axio-dilaton couplings. The cells are indeed cyan/blue, making
these couplings the smallest in scale.
Now that we’ve confirmed the existence of the naively expected hierarchies we turn to
their source – the proximity to the conifold point. A priori it is possible that the divergent
contributions due to the Yukawa coupling and its derivatives could have been tempered
by some other mechanism as the conifold is approached. It is also important to assess the
degree of variation in the expected scale separation. Just as the vev of the superpotential
played the role of a random element complicating an otherwise clean analytic dependence
on the conifold distance, here too we will have a layer of noise atop the signal. The
significance of this noise, and importantly the degree to which it changes as the conifold is
approached, is not obvious at the outset.
We show the conifold distance dependence of the cubic couplings’ scale separation
visually as well. For each vacuum’s four independent 8-cell subcubes we first compute each
– 45 –
Figure 19. The analogous data as that displayed in Figure 18 for the quartic couplings Ai0 j 0 k0 l0
with a logarithmic scale for the magnitudes represented with a CMYK color scheme (black being
the largest, followed by yellow, magenta then cyan). The five scales expected due to the behavior of
the Yukawa coupling near the conifold manifest themselves as the five different types of subcubes
– those with cells in the black, yellow, orange/pink, pink/purple, and blue. The origin of each of
the four cubes in the stack is at the bottom left of the front face in the view on the left, making
the all-complex-structure-couplings contained in the black corners of the top pair of cubes. The
panel on the right shows a view of the hypercube rotated about the vertical axis, with the all-axio-
dilaton couplings in the top front corner of the bottom pair of cubes in the stack, which are blue
as expected.
subcube’s mean magnitude. The mean is the appropriate measure here since the entries in
a single subcube for an individual vacuum are comparable. Each vacuum then has a list of
four positive values– the average magnitude of each of the four type of cubic couplings. We
take the logarithm of each value in the list, as well that of the magnitude of the canonical
– 46 –
Figure 20. A log-log scatter plot of the vacuum coupling subcube average magnitude versus
conifold distance. Each subcube contains couplings of one of the four types: all-ξ (shown in pink), ξ-
ξ-σ (green), ξ-ξ-σ (purple), and all-σ (navy). The fact that the data points organize themselves into
approximately linear bands with increasingly negative slope for each ξ at the expense of a σ confirms
the Yukawa coupling (through its successively more singular partial derivatives) is responsible for
the hierarchy observed. The width of individual bands signals the presence/role of the random
element, the vev of the superpotentail.
vacuum coordinate, |ξvac |. The ensemble data for all four types of cubic couplings are
displayed in the single log-log scatter plot in Figure 20, with different colors used for each
of the four types.
Notice first that the colors separate into four approximately linear bands with negative
slope. This indicates that each type of coupling has an inverse power law dependence on
the canonical vacuum coordinate, |ξvac |. The data points with most negative slope, the
pink band, are the ensemble of all-complex-structure cubic couplings. Each pink point is a
different vacuum’s mean ξ-ξ-ξ–type coupling magnitude. Below this band lies the second
largest scale in the cubic couplings involving two ξ and one σ, shown in teal, followed by
purple and navy blue for the σ-σ-ξ and the all-axio-dilaton couplings, respectively. The
fact that the bands are approximately linear reflects the domination of the leading order
term in the Yukawa coupling and derivatives thereof over other terms in the expressions
for the cubic couplings.
The same analysis can be performed for the quartic couplings. We show the resulting
log-log scatter plot for the five types of couplings in Figure 21, with the same coloring
scheme descending from the largest in pink (ξ-ξ-ξ-ξ–type), and the addition of a fifth
– 47 –
Figure 21. The analogous plot as that in Figure 20 for the quartic couplings. The colors are
ordered in the same manner according to the number of ξ’s in the coupling-type, descending from
all-ξ (pink), with the addition of light-blue for the last of the five types, all-σ. The self organization
of the vacuum data into the approximately linear bands of increasingly negative slope for every ξ at
the expense of a σ confirms the validity of our explanation of the hierarchies based on the behavior
of the Yukawa coupling near the conifold.
color, light-blue, for the smallest (all-axio-dilaton type). The same reasoning indicates
that the source of the hierarchy among the quartic couplings are the terms involving the
most ξ derivatives of Yukawa coupling evaluated near the conifold point. For both the
cubic and the quartic scatter plots we may view the statistical variation within a given
band as being supplied by the random element, the vev of the superpotential.
We conclude with a qualitative discussion of the remaining aspect of the structure
among the couplings that is not captured by a Random Matrix Model, for example that of
[7]. These are the pattern of nontrivial correlations we find between couplings. That is, the
ensemble of Ai0 j 0 k0 and Ai0 j 0 k0 l0 are not accurately modeled by totally symmetric tensors
whose entries are drawn separately from independent distributions. We’ve seen that the
distribution of a particular cubic or quartic coupling is roughly Gaussian and is centered at
zero. The hierarchies discussed mean that the spread of these distributions differ in scale,
according to index type. For instance the A112 distribution is comparable to A222 in this
regard, but not to, say, A113 , whose spread is smaller by comparison.
The hierarchies and the non-flat distributions themselves need not have come with
correlations between couplings. The fact that we find approximately linear scatter plots
between particular pairs of couplings renders a random approach involving independent
– 48 –
Figure 22. A representative sample of the scatter plots of pairs of cubic and quartic couplings from
the vacuum data (pink, green, and purple), as well as from the random matrix model couplings
(blue) designed as a diagnostic. Note that whereas the vacuum data exhibits sharply defined
correlations between certain pairs of couplings, all the random matrix model pairs do not. This
indicates that correlations are not merely built in by the diagnoalization of the Hessian in canonical
coordinates.
distributions – uniform or otherwise – a poor approximation to the actual coefficients. A

representative sample of the nontrivial correlations for the cubic coupling data sets are
shown eight of the nine panels in Figure 22, excluding that in the bottom right corner (in
– 49 –
blue).
The pink plots on the top row show that while the pairs A111 with A122 , and A112
with A222 have an approximately constant ratio across the ensemble of vacua, there is no
relationship between A111 and A222 . We also find correlations in the couplings of medium
scale, namely those that mix moduli type. For instance, A144 and A133 are approximately
equal in magnitude, but opposite in sign, across models. This is shown in the teal plots in
the middle row.
A reasonable hypothesis for the source of these correlations is the transformation per-
formed to the field coordinates that simultaneously diagonalize the Hessian and canonically
normalizes the kinetic terms. This seemingly mundane step in the processing of the raw
coupling data might be suspected as being nontrivial at the level of correlations because
of the special structure of the Hessian’s eigenspace. We test this hypothesis by comparing
the results of a modified Random Matrix Model designed entirely as a diagnostic for this
purpose.
If it is the case that the transformation from the original noncanonical complex co-
ordinates builds in the patterns of correlations we observe, then an ensemble of real and
totally symmetric tensors with i.i.d. entries acted upon by the orthogonal transformation
O (defined in subsection 3) ought to exhibit correlations. Since we have 1358 O matrices,
we build the same number of random rank-3 tensors and perform the transformation,
i j
Arand k rand
ijk → Oi0 Oj 0 Ok0 Aijk . (4.22)
The result is that the transformed random couplings are uncorrelated. We’ve included a
single scatter plot of these RMM couplings as a representative example. This is the ninth
panel in Figure 22.
5 Discussion
The initial expectation that string theory would result in a unique, or nearly unique, vac-
uum state whose low energy excitations would explain the familiar properties of particle
physics has not been borne out by developments over the past few decades. Instead, a
wealth of discoveries have revealed an ever greater abundance of mathematically consistent
vacua, without any allied developments that single out one (or perhaps a few) such vacua
as physically relevant. Because of this, significant attention has shifted to statistical prop-
erties of these vacua and, more generally, to statistical properties of the easier to analyze
surrogate, random field theories in high dimensional moduli spaces.
In this paper, we have investigated the degree to which this latter surrogate faithfully
models the space of low energy field theories arising from string compactifications. We
reviewed arguments which suggest the relevance of random field theories–namely, the ran-
domizing effects of arbitrary fluxes coupled with the broad spectrum of vacuum locations
in moduli space associated with each such flux choice. We then tested this argument by
focusing our attention on one particular compactification of the type IIB string, the famous
mirror to the quintic hypersurface. We identified a class of 1358 low energy flux models
– 50 –
built on this compactification, computed the scalar potential for the canonically normal-
ized scalar fields in each such model, and considered the statistical distributions of the
renormalizable coefficients in the Taylor expansions of the potentials. We confirmed previ-
ously known results for the second order coefficients – mass terms – and went on to study
the third and fourth order terms. Our main result is that we found significant deviations
from a random collection of coefficients, as illustrated in Figures 18, 19 and 22, showing
that some of the rich structure inherent in type IIB supergravity survives the randomizing
influence of flux compactifications.
The lesson, then, is that one must exercise care when invoking random field theories as
a model for the space of low energy compactified string dynamics. More particularly, our
results, and generalizations thereof to higher dimensional moduli spaces, provide a sharper
ensemble for accurate statistical modeling of the features of low energy string theory.
Going forward, these results suggest a number of research directions. For ease of
computation we have focused on a Calabi-Yau compactification with a single complex
structure modulus. One would like to acquire an understanding of the distributions we
have studied in more generic cases with higher dimensional moduli spaces. Explicit analysis
of the sort we’ve undertaken here would be difficult. However, in the vicinity of a conifold
locus – where vacua generally accumulate – we’ve reduced the statistical dependence to
the three dominant control parameters introduced earlier. These each have natural higher
dimensional generalizations and so it would be of interest to see if we can gain insight into
more general Calabi-Yau compactificaitons guided by the results we found here, and thus
avoiding direct calculation. We hope to return to this shortly. It would also be interesting
to revisit the works [7] which have investigated the quantum stability of vacua in random
high dimensional scalar field theories as a surrogate for the stability of the string landscape.
Are those results modified by studying a collection of field theories whose distribution more
closely aligns with that of low energy dynamics of string theory? We intend to return to
this question as well.
Acknowledgments
It is a pleasure to thank Thomas Bachlechner, David Kagan, Ruben Monten, and especially
Frederik Denef, for numerous insightful conversations and observations, which greatly as-
sisted the completion of the research reported herein. We also gratefully acknowledge the
support of the Department of Energy through grant DE-FG02-92-ER40699.
References
[1] M. R. Douglas, JHEP 0305, 046 (2003).
[2] F. Denef and M. R. Douglas, JHEP 0405, 072 (2004).
[3] F. Denef, and M. R. Douglas, JHEP 0503, 061 (2005).
[4] F. Denef, M. R. Douglas, and B. Florea, JHEP 0406, 034 (2004).
[5] N. Arkani-Hamed, S. Dimopoulos, and S. Kachru, SLAC-PUB 10928 (2005)
arXiv:hep-th/0501082 [hep-th].
– 51 –
[6] J. Distler, and U. Varadarajan, (2005) arXiv:hep-th/0507090 [hep-th].
[7] B. Greene, D. Kagan, A. Masoumi, D. Mehta, E. J. Weinberg and X. Xiao, Phys. Rev. D 88,
no. 2, 026005 (2013) arXiv:1303.4428 [hep-th].
[8] X. Chen, G. Shiu, Y. Sumitomo, and S. H. H. Tye, JHEP 1204, 026 (2012).
[9] D. Marsh, L. McAllister, and T. Wrase, JHEP 1203, 102 (2012) arXiv:1112.3034 [hep-th].
[10] D. Marsh, B. Vercnocke, and T. Wrase, JHEP 1505, 081 (2015) arXiv:1411.6625 [hep-th].
[11] U. H. Danielsson, N. Johansson, and M. Larfors, JHEP 0703, 080 (2007).
[12] D. I. Podolsky, J. Majumder, and N. Jokela, JCAP 0805, 024 (2008) arXiv:0804.2263
[hep-th].
[13] M. Dine, and S. Paban, (2015) arXiv:1506.06428 [hep-th].
[14] M. Dine, (2015) arXiv:1512.08125 [hep-th].
[15] S. B. Giddings, S. Kachru, and J. Polchinski, Phys. Rev. D 66, 106006 (2002).
[16] P. Ahlqvist, B. R. Greene, D. Kagan, E. A. Lim, S. Sarangi, and I-S. Yang, JHEP 1103, 119
(2011).
[17] P. Candelas, X. C. de la Ossa, P. S. Green, and L. Parkes, Nucl. Phys. B 359 (1991), 21-74.
A Near Conifold Period Expansion Coefficients
Recall that our integral and symplectic basis for mirror quintic’s period functions are
denoted Πi , with i = 0 and 3 an intersecting pair, and i = 1 and 2 the other. Π0 is the
only non-analytic period at the conifold point. It’s partner, Π3 , is the period obtained by
integrating the holomorphic 3-form over the cycle that collapses. This period is nevertheless
well-behaved (it simply vanishes at the conifold). The two periods associated with the
other intersecting pair of cycles are also analytic. These are nonvanishing. The following
expansions about the conifold point at z = 1 hold,
q
X
Π1 (z) = bn (z − 1)n (A.1)
n=0
Xq
Π2 (z) = cn (z − 1)n (A.2)
n=0
Xq
Π3 (z) = dn (z − 1)n . (A.3)
n=1
The values for the coefficients were computed in Mathematica by evaluating derivatives of
the expressions for the Πi in terms of the Meijer-G functions (the Ui ) at the conifold. We
used 40 digit accuracy in these computations. The values are listed in Table 3.
The remaining period, Π0 , is multiple-valued at the conifold point. The cycle it is
associated with picks up one copy of its vanishing partner for each revolution around the
conifold in moduli space. This fixes the form of Π0 to 2.74. To match the branch cuts of
– 52 –
n bn cn dn
0 +1.293574i 6.19502 − 7.11466i 0
1 −0.150767i −1.016605 + 0.829217i −0.355881i
2 +0.0777445i 0.570733 − 0.427595i 0.249117i
3 −0.0522815i −0.401804 + 0.287548i −0.194548i
4 +0.0393684i 0.312044 − 0.216526i 0.161285i
5 −0.0315669i −0.256050 + 0.173618i −0.138686i
6 +0.0263447i 0.217649 − 0.144896i 0.122217i
7 −0.0226046i −0.189607 + 0.124325i −0.109620i
8 0.0197941i 0.168193 − 0.108868i 0.0996353i
Table 3. Expansion coefficients for period functions Π1 , Π2 and Π3 .
the logarithm in the expansion with that of the relevant Mejer-G (U0 ) in Mathematica we
must negate the argument of the logarithm. Ultimately the expression we write for Π0 is,

log(−(z − 1)) 1
Π0 (z) = Π3 (z) − + f (z) (A.4)
2πi 2
with f (z) analytic.

Its expansion coefficients, an , were computed using a recursion relation based on the
fact Π0 satisfies the Picard-Fuchs equation. This is discussed at length in section 2.3. Here
we simply tabulate the resulting values. The first three an are needed by the recursion to
generate the rest. They are obtained numerically, as discussed in section 2.3. All coefficients
were calculated using 30–40 digit accuracy computations and are listed in Table 4.
n an
0 1.07073
1 0.024708 − 0.177941i
2 −0.0115108 + 0.1245584i
3 0.0065650 − 0.0972742i
4 −0.0042768 + 0.0806427i
5 0.0030290 − 0.0693428i
6 −0.0022701 + 0.0611087i
7 0.0017719 − 0.0548102i
8 −0.0014261 + 0.0498177i
Table 4. Expansion coefficients for analytic contribution to Π0
– 53 –

Random Field Theories

Uploaded by

Copyright:

Available Formats

Random Field Theories

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Random Field Theories

Uploaded by

Copyright:

Available Formats

Prepared for submission to JHEP

Random Field Theories in The Mirror Quintic Moduli

Kate Eckerlea and Brian Greeneb

E-mail: [email protected], [email protected]

A Near Conifold Period Expansion Coefficients 52

2.1 Review of Flux Compactifications

G(3) = F(3) − τ H(3) , (2.2)

τ is the axio-dilaton related to the dilaton, φ, by

τ = C(0) + ie−φ (2.3)

F(p) = dC(p−1) (2.4)

ka = −Ka = −∂a K. (2.9)

K ab̄ Da W D̄b̄ W̄ . (2.18)

Kax = − log(−i(τ − τ̄ )), (2.19)

and so first term in 2.16 is proportional to

|Dτ W |2 = Kaxτ τ̄ Dτ W D̄τ̄ W̄ , (2.22)

where the Kähler moduli are excluded.

R10 = R4 + R6 + Rmix . (2.24)

Since R4 is a constant over the Calabi-Yau, integration of it over M yields a factor of

Kab̄ Ka W Kb̄ W̄ − 3|W |2 = 0. (2.39)

2.2 Period Integrals

Qij = Q(Ci , Cj ) = hCi ^ Cj [M]i (2.44)

Kcs (z, z̄) = − log(iΠ† (z̄)Q−1 Π(z)) (2.49)

Q03 = hC0 ^ C3 , [M]i →hC0 + nC3 ^ C3 , [M]i (2.57)

Π0 (z) = U0 (z) (2.60)

Π0 (z) → Π0 (z) + Π3 (z) (2.64)

for each revolution z → (z − 1)e2πi + 1, implies

while those in the flux vectors are labeled in ascending order,

So, we evaluate the axio-dilaton in the function we seek to minimize, |Dz W |2 , at τ =

2.3 Period Expansions for the Mirror Quintic

log(−(z − 1)) Π3 (z) log(−(z − 1))

It is a convenient choice for performing checks of the accuracy of the Π0 approximation

The third coefficient in f ’s expansion, a2 , can be obtained in a similar fashion. We

Table 2. Depiction of the convergence of the expansion coefficient a2 computed numerically.

ÔP F [Π3 (z)] = 0. (2.83)

ÔP F = −sδ 4 − s(k1 δ 3 − k2 δ 2 − k3 δ − k4 ) − (k1 δ 3 − k2 δ 2 − k3 δ − k4 ) (2.84)

Next note that δ acts on sn as,

2.4 Structure of the Hessian

FI ≡ DI W ; ZIJ ≡ DI DJ W ; UIJK ≡ DI DJ DK W (2.91)

V = eK (FI F̄ I − 3|W |2 ). (2.92)

where indices run over all moduli.

∂I V = eK (DI DJ W )F̄ J − 2FI W̄ = eK ZIJ F̄ J − 2FI W̄

∂I ∂J V = 2eK W̄ ZIJ (2.97)

Z can Z̄ can = U Σ2 U † (2.100)

can be diagonalized by a unitary transformation defined in terms of the 2N×2N-matrix U,

In particular, one can rewrite eq. 2.101 as,

3.1 Generating a Random Sample of Vacua

Mp2 Kcs (z,z̄)+Kax (τ,τ̄ ) z z̄

G(3) → cG(3) + d. (3.10)

the maximum of the righthand side as the positive number Lmax ,

3.2 Computing Coefficients

{z, τ, z̄, τ̄ } → {y1 , y2 , y3 , y4 } (3.13)

The effective field theory is obtained by expanding about a homogeneous background,

L(Φ) = L(Φvac ) + ∂µ Ψ† B † (Gvac + O(Ψ))B∂ µ Ψ − V (BΨ) (3.16)

Note that T it is unitary. Lastly, we take

where O is the orthogonal matrix containing the (normalized) eigenvectors of T † B † M BT

we need to transform according to the standard tensor transformation law,

∂Φi ∂Φj ∂Φk

∂z ∂z̄ ∂τ Fz = ∂z ∂z̄ ∂τ (F − τ H) · (Π0 + Kz Π) (3.30)

where the entries are complex valued (and Z01 = Z10 ).

ZIJ = FIJK Z̄ 0K (4.3)