Holography and local fields

Ingmar Saberi∗
Mathematisches Institut
Ruprecht-Karls-Universität Heidelberg

This paper, which is a summary (in which considerable creative license has been taken)
of the author’s talk at the sixth international conference on p-adic mathematical physics
and its applications (CINVESTAV, Mexico City, October 2017), reviews some recent work
connecting field theories defined on the p-adic numbers and ideas from the AdS/CFT corres-
pondence. Some results are included, along with general discussion of the utility and interest
arXiv:1801.04942v1 [hep-th] 15 Jan 2018

of p-adic analogues of Lagrangian field theories, at least from the author’s perspective. A
few challenges, shortcomings, and ideas for future work are also discussed.

∗
saberi@mathi.uni-heidelberg.de
 2

1. INTRODUCTION

Das also war des Pudels Kern !

Ein fahrender Skolast ? Der Kasus macht mich lachen.
Goethe, »Faust«

Over the last couple of years, several papers have appeared ([17–19, 21, 22, 24]; see also [3, 11,
39]), revisiting the topic of field theories on p-adic spacetimes in an attempt to develop an inherently
discrete model of the physics of the AdS/CFT correspondence. Rather than going through calcu-
lations in any great detail, we would like to discuss the general arc of this work, emphasizing the
analogies that are in play, the parts of the formalism that carry over straightforwardly due to these
analogies, and the places where the analogy is incomplete and further work is necessary in order
to be able to tell a fully satisfactory story. (For another perspective on the same material, see [16]
from the proceedings of Strings 2016.) The reader may not be intimately familiar with all of p-adic
techniques, quantum field theory, and the ideas behind the AdS/CFT correspondence; indeed, the
increasing dearth of generalists, even within the limited domain of mathematical physics itself, is
likely responsible for the fact that these analogies (which are striking, once one knows where to
look) went unnoticed for so long. For this reason, we will give at least a cursory and teleological
review of each, in a way that we hope will make structural similarities apparent.
p-adic techniques have a long history in physics, and we will unfortunately not have the oppor-
tunity to offer a broader review of the literature here. See [9, 10] for recent reviews, and [4] for a
classic one. We will mention only that much attention was devoted to p-adic analogues of string
theory [13, 14, 33, 41, 42, 47, 52, among others], in which the string worldsheet (which, although
it starts life as a two-dimensional smooth manifold, can be thought of due to some peculiarities of
low-dimensional physics as an algebraic curve over C) was replaced by a p-adic algebraic curve. In
this context, theories of scalar fields and analogues of conformal symmetry were studied, and objects
such as Green’s functions were first computed [34, 52] The connection between p-adic models and
hierarchical models in the style of Dyson—and therefore to renormalization group theory [12, 49]—
was also remarked on some time ago; we will mention results of Lerner and Missarov [27, 28, 35, 36],
among others.

2. WHAT IS QUANTUM FIELD THEORY?

We will begin by giving an acknowledgedly biased and incomplete definition of a quantum field
theory. Of course, many such definitions have been given over the preceding decades, by people far
 3

more qualified to do so than the current author, and we do not pretend that our definition encom-
passes a complete and precisely delineated idea. Indeed, the term definition may be a misnomer,
and perhaps a misleading one; let us instead say that we will offer a sketch of something called
quantum field theory, without claiming that no other sketches are possible or useful.
We will say that a quantum field theory is a theory of many identical local degrees of freedom,
parameterized by some geometric space X, and coupled together in a local and homogeneous way
with respect to the geometrical structure of X.
This is a definition that is broad enough to include many different things: standard Lagrangian
field theories, of course, fit into it, as well as lattice field theories and even spin systems. And these
are all systems that a physicist would think of as being described by quantum field theory, at least
in some range of their parameters.
To translate our sketch into more technical language, we might agree to say that a QFT is a
quantum theory whose degrees of freedom are functions on X. By a function, we will understand
any locally defined object, such as a connection, tensor field, or a section of some other bundle;
in the case when X is a manifold, by the Serre–Swan theorem, these are just projective modules
over the algebra C ∞ (X), or another suitable algebra of functions. However, none of these beyond
functions themselves will play a great role in the story that follows.
Such a function is understood to represent a measurement or observation that can be made
independently everywhere in X; one should imagine a quantity like the temperature or electric field
in a room. We will always assume, in keeping with everyday experience, that measurements are
made using real numbers, and therefore that fields are real- or complex-valued. More exotic number
systems will appear in a different role: namely, in the construction of different choices for X.
We’ll also imagine that the interactions of the theory are encoded in an action functional:

S : F (X) → R. (2.1)

While it is perhaps not conceptually essential, the point of this requirement is to have a clear
concept of the classical limit of the theories we are interested, and to be able to use pieces of the
standard formalism (path integrals, Feynman diagrams, and the like) wholesale.
While we are of course being intentionally vague in our sketch, it should be emphasized that,
at this point, X could be anything: a manifold, of course, but also a lattice, a graph, or simply a
set. Different such choices of X will lead to different classes of models (lattice field theories, for
example); these models will carry different amounts of structure, corresponding to the presence or
absence of various geometric structures on X itself. In the case when X is a set and no additional
 4

structures are present at all, we are just talking about many-body quantum mechanics.
One big question that has preoccupied researchers in the border regions of mathematical physics
over the last thirty years or so might be stated as follows: How is geometric and topological
information about X reflected in the behavior of theories on X? As is perhaps to be expected for
such a broad question, many of the big developments of recent years can be interpreted as offering
sketches of an answer.
To just briefly mention one example, when X is a smooth four-manifold, one theory that can
be defined on it is a “twist” of N = 2 supersymmetric SU (2) gauge theory. Considering this
theory leads one to consider the moduli space M (X) of anti-self-dual instantons on X. The work
of Donaldson and others [7, 50] showed that crude topological invariants (such as the homology)
of M (X) are sophisticated invariants of the smooth structure of X itself, which provide powerful
tools for understanding four-dimensional smooth manifolds!
The use of field theories, in particular topological field theories, to construct and organize invari-
ants of geometric and topological manifolds thus has a long and fruitful history. But it is natural
to wonder if one can carry information across this bridge in reverse. For some time, physicists have
been speculating about the vague question of how the classical geometry of X is encoded in (e.g.)
the Hilbert space of a theory, or the entanglement structure of states therein.
An idea for how to start sketching partial answers to such a question is as follows: Vary X, so
that different amounts of its structure are present in different ways, and use this to try and pick
apart the question into smaller and more manageable pieces. From the author’s viewpoint, this is
nothing more than reasoning by analogy, which is perhaps the basic form of reasoning in all of pure
math and theoretical science; we will expand a bit on the role of analogy in the section that follows.
After this, we will return to the question of where one might find interesting choices of X to think
about.

3. ANALOGY IN MATHEMATICS

In some sense, the whole story of mathematics (and of theoretical reasoning more generally)
is a story about analogies. Much of the axiomatization that has taken place is about codifying
analogies between different situations: when we say that the integers are an abelian group, we are
calling upon an analogy that says that says that the integers are exactly like the real numbers,
the integers modulo n, or the complex numbers, at least as far as properties that only have to do
with the existence of a commutative composition law with inverses are concerned. The definition
 5

of “an abelian group,” then, is really just an embodiment of this particular analogy or shared piece
of structure, so that its consequences can be explored in and of themselves, without reference to
the particularities of any example.

Reasoning by analogy is of crucial importance in theoretical science too, being visible (just for
example) in the simple but powerful idea that the same equations have the same solutions. (The
importance of this point was hammered home to the author by teachers, many years ago.) That is,
when a model of a situation leads to a particular set of equations, precisely what is being modeled
(along with what other equations might or might not be valid) becomes irrelevant, at least as far
as properties depending only on those equations are concerned.

So far, we’ve just emphasized that structural reasoning, or abstraction, is, at its heart, a form
of reasoning by analogy. But there are examples of looser or larger-scale analogies in both physics
and mathematics as well. For the former, analogies between quantum field theory and statistical
mechanics come to mind; as to the latter, perhaps the best and most fruitful example of such a
large-scale conceptual analogy is the function field analogy in number theory [40, 48, for example].
With an eye towards motivating the value of such analogies, we will recall (in brief and inexpert
fashion) some aspects of it here.

One route of approach into this analogy might proceed as follows: One is a number theorist,
and as such is interested in studying integers. One of the key features of integers is that they admit
a unique factorization into prime numbers, up to an ambiguity of sign:

Y
n=u pi , n ∈ Z, (3.1)
i

where the pi are (not necessarily distinct) prime numbers, taken to be positive, and u is a sign
(±1), or in other words a unit of Z.

Then, one starts to ask naive questions about the primes: How many primes are there of a given
size, i.e., less than or equal to some integer N ? What is the longest possible span of numbers that
contains no primes? How close together can two consecutive large primes be? As everyone knows,
many questions like these, while they are easy to state, are incredibly difficult to answer; some
remain open to this day.

One way to think about difficult problems in number theory, then, is to look for toy models
exhibiting the same kinds of structural behavior—unique factorization, for instance—in a simpler
environment. And one such model consists of polynomials; any polynomial over a field k can be
 6

factored into a product of irreducible components,

Y
f (x) = u pi (x), f ∈ k[x], (3.2)
i

where pi (x) are irreducible polynomials (taken to be monic), and u is a unit (i.e. element of the
ground field k).
One now has a model where (unlike the integers) there are some parameters that can be tweaked:
in particular, any choice of field will give a ring of polynomials with unique factorization, where
similar questions can perhaps be formulated. Depending on the choice of field, understanding
the behavior of polynomials may be more or less straightforward—and may have more or less in
common with the behavior of integers. For example, when the ground field k = C, the irreducible
polynomials are exactly the linear polynomials x − t for t ∈ C. This case is quite simple (the
irreducibles are simple to understand completely), but also unlike the integers (for example, in that
there are uncountably many irreducibles).
A big part of the utility of this analogy is the geometric picture it allows one to develop. For
example, we have said that the integers can be thought of as being like regular polynomial functions
on the complex plane (or, more generally, the affine line). The field Q is then like the field of rational
functions on the same curve. Moreover, a finite extension of Q (or an algebraic number field) should
then be analogous to a finite extension of k(t). But that is the same as a function field of some
other curve C over the ground field k (or possibly a finite extension of k; see e.g. [5, 6] for precise
structure theorems on extensions of nonarchimedean local fields). Considerations like this lead one,
for example, to the notion of the genus of a number field, which was first defined by Weil [48].
Now, each irreducible polynomial (which one can think of as closed points of the curve C, or
equivalently as Gal(k̄/k)-orbits of k̄-points of C) defines a valuation on the function field, roughly
speaking by order of vanishing at the corresponding point. This is perhaps easiest to think about
when C is a Riemann surface, and the corresponding valuation is literally the order of pole or zero
of a rational function at a point p ∈ C. Of course, there are also valuations corresponding to the
points at infinity of C, i.e., to points in its projective closure C; in the case of the affine line, the
valuation at infinity is just the valuation by the degree of the polynomial. By well-known theorems,
perhaps most familiar from complex analysis, not all of these valuations are independent: in fact,
they satisfy the single relation
X
ordp (f ) = 0, (3.3)
p∈C

for any rational function f on C.

 7

One can complete the field with respect to any of its valuations. For simplicity, let’s just think
about the origin inside of the affine line. The completion of k[t] with respect to the corresponding
valuation (which is just by powers of t) is, of course, the ring of formal power series k[[t]], and the
corresponding completed field is the field of formal Laurent series k((t)). It is fruitful to think of
this as looking at functions “locally” around a point of C,
The best analogy is obtained by letting k = Fq , the finite field with q = pn elements. In fact,
by results of Artin and Whaples [1], this gives a complete list of all “global” fields: they are either
number fields (finite extensions of Q) or function fields of curves over a finite base field k. A global
field is defined by having a set of places or valuations, satisfying a relation, and for which at least
one valuation is either Archimedean or discrete with finite residue field. “Local” fields are then
completions of global fields at a chosen place; equivalently, they are locally compact topological
fields, with respect to a non-discrete topology.
A key way in which the analogy is strengthened when k = Fq is that all the prime ideals of the
ring of integers are then of finite index, and there are finitely many primes of a given polynomial
degree, just as for Z. So it is meaningful to “count” primes and ask questions about their density.
A sharp analogue of the Riemann hypothesis for function fields can be given, in the form of a
structure theorem about the Hasse-Weil zeta function of the curve C; it was conjectured by Artin
and proved by Weil (and earlier by Hasse for elliptic curves). Similar statements for other varieties,
the famous Weil conjectures later proved by Deligne, can thus be thought of as the fruit of this
analogy; unfortunately, though, this is as far as we can go into this fascinating topic here.
As already alluded to, Q is also a global field, and so has a relation between its different valua-
tions. There is one of these corresponding to each prime p (analogous to the valuations measuring
order of vanishing at a closed point); these are simply defined by

a
x = pν ⇒ ordp (x) = ν, |x|p = p−ν , (3.4)
b

for a and b prime to p. There is also the usual notion of absolute value on Q; a theorem of Ostrowski
says that, up to equivalence, these are the only nontrivial valuations. Moreover, they satisfy the
obvious relation
Y
|x|p = 1, (3.5)
p

where p runs over all places, i.e., the primes together with infinity.
The locally compact completions of Q are therefore the real numbers and the p-adic fields; these
and their finite extensions exhaust the locally compact topological fields of characteristic zero, and
 8

they will be important characters in the rest of our story. Based on the function field analogy,
one can think of the p-adics as being like “germs of functions around the point p;” however, it is
important to note that none of these local fields are isomorphic.
We will not take the time to review many of the peculiar properties of the p-adic norm here;
we will content ourselves with giving just one piece of intuition. The p-adic norm (as well as
any valuation coming from a similar concept, such as order of vanishing or divisibility) is best
understood in terms of a hierarchical structure, like the successive division of species into smaller
and smaller taxa. The closeness of two elements is then understood in terms of the lowest “height”
of taxon at which they fall into the same group.

4. STRUCTURAL INGREDIENTS OF A LAGRANGIAN FIELD THEORY

To come up with something that fits our sketch of a field theory, it’s probably a basic requirement
to be able to make sense of the concepts of locality, homogeneity, and isotropy.
Locality suggests that X has a notion of distance, measure, or causal structure, which is respected
by the interactions in the theory. In a standard field theory on affine space (or a smooth manifold),
this often means something like
Z
S[φ] = L [φ(x)]. (4.1)
X

So we need a notion of a “local Lagrangian density” (a map L from functions to functions, densities,
or top forms, depending only on functions and their derivatives at each point), as well as a notion
of integration that assigns a number to a local Lagrangian density.
For lattice models, though, we might need a different notion of local density: it’s too strong to
require that individual lattice sites interact only with themselves. Instead, we would ask for some-
thing like “nearest-neighbor” interactions: i.e., S is a sum of terms corresponding to each site, where
the term associated to a site depends only on the degrees of freedom from some finite neighborhood
of that site. The notion of locality might even be further relaxed, to mean that interactions between
separated points decay sufficiently fast (exponentially, for example) with distance.
Now, depending on the choice of X, there may be “spacetime” symmetries (implemented by the
action of a group G on X). The theory may or may not respect the action of these symmetries on
the fields F (X).
Often (when X is a maximally symmetric space), some group acts freely transitively on X. The
theory is homogeneous when it respects the action of such a spacetime symmetry, which plays the
 9

role of translations (and in fact consists of translations, in the case when X is just affine Rn ).
But the group of translations may not be the whole story. For X = Rn , the group of affine
transformations is a semidirect product of translations and rigid rotations,

G = Rn ⋊ SO(n). (4.2)

Poincaré-invariant theories always preserve these two symmetries, as makes sense for any theory that
is formulated to depend on the metric, but not on any specific choice of coordinates. Translation
invariance represents homogeneity, and rotation invariance represents isotropy (or, in Lorentzian
signature, Lorentz invariance); in other cases, the combined notion of a theory invariant under
isometries of X will make sense, even when the splitting as a semidirect product does not.
But there are other symmetries present, which may or may not be preserved in interesting ways,
and may respect less structure than Poincaré transformations. For example, there are the discrete
symmetries of quantum field theory (in particular, P and T ), coming from the group of components
of the Lorentz group. These may or may not be respected, and are in fact broken in the standard
model of particle physics.
There are also symmetries that act on the spacetime, while failing to respect the metric structure
in prescribed ways. An important symmetry is scale invariance. While most quantum field theories
of interest fail to be scale-invariant, the study of how scale invariance is broken is the theory of the
renormalization group.
One generally expects local scale-invariant theories to be invariant under a larger group of trans-
formations: the conformal group, consisting of local (position-dependent) scale transformations.
(While general reasoning about locality, similar to the reasoning that accompanies the promotion
of global to gauge symmetries, suggests that scale-invariant theories are often conformal, there are
subtleties related to this idea; we do not go into these here. For a recent review, see [37]).
In order to be able to write down something that deserves to be called a standard field theory
Lagrangian, we will need a couple of additional structures. In particular, we would like to be able
to make sense of a standard-looking kinetic term, which (in momentum space) looks something like
this:
Z
φ(−k) |k|2 + m2 φ(k) + · · ·


S[φ] = (4.3)
X∨

Of course, making sense of this expression just means being able to assign a meaning to each of the
symbols. Let’s think about what that entails.
First off, the dual space X ∨ has to be defined. That means X has a notion of (mutually com-
muting) translation symmetries, with a complete basis of eigenfunctions φk ∈ F (X), diagonalizing
 10

those translations. Here the parameter k takes values in X ∨ , which is by definition the joint spec-
trum of the translation operators. This amounts to saying that there is a notion of mode expansion,
or equivalently, of the Fourier transform. (Of course, this requirement isn’t absolutely necessary—of
course, such a requirement fails for field theories on generic curved spacetimes—but we’ll keep it
around for now.)
In the case of Rn , we further have that X ∼
= X ∨ . But this isn’t necessarily true: in lattice
models, for example, the relevant fact is Z∨ = S 1 . Physicists would refer to this under the name of
the “Brillouin zone.”
I will also need a notion of “size” on X ∨ (generalizing the length of a vector). And, finally,
I’ll need a notion of integration on X ∨ . Luckily, one set of assumptions is enough to ensure both
a notion of translation-invariant integral and a good notion of Fourier transform. Whenever the
translations of X are a locally compact abelian group, one can integrate with respect to Haar
measure, and the Pontryagin duality theorem ensures that we have a good theory of the Fourier
transform.
Once I can do this, I can make sense of the kinetic term of a real scalar field, and so I’m really
in familiar territory. The essential point in this is to identify the structures that allow us to make
sense of such a term. The more of these structures X has, the closer one is to being able to define
a theory on it in perfect analogy to your favorite typical QFT.

5. PARALLELISM

We’ve catalogued many of the important structures on Rn by now. To reiterate, the central
idea is that many of these structures exist just because Rn is an affine space over a locally compact
field. Taking our cue from mathematics, we’ll therefore try to define field theory models over other
locally compact fields, and see how far we can get. We will hope to rely on the fact that, at least
as far as the algebraic structures are concerned, affine spaces over fields all behave similarly.
We have already recalled that the p-adics are the completions of Q at its finite places. A generic
element of Qp can be written in the form
∞
X
ai p i , (5.1)
i=−ν

where the digits ai ∈ {0, . . . , p − 1}; the corresponding sequence of partial sums is always Cauchy
with respect to the p-adic distance, and thus by definition has a limit in the completion. They
are locally compact fields (with respect to a nondiscrete topology), and therefore have additively
 11

and multiplicatively invariant Haar measures; we will denote the additive-invariant measure simply
by dx. The multiplicative Haar measure is then simply dx/|x|.
The theory of additive characters of Qp is completely understood: they are of the form

χp (kx) = e2πi{kx}p , (5.2)

where {x}p denotes the p-adic “fractional part:” it is the natural map
∞
X −1
X
i
Qp → Qp /Zp ⊂ Q, ai p 7→ ai pi . (5.3)
i=−ν i=−ν

For the sake of unity, we’ll let {x}∞ = x for real numbers. Now, additive characters span the space
of well-behaved (locally constant) functions on Qp , so that there is a good theory of the Fourier
transform. (5.2) exhibits the fact that Qp is self-dual.
Many of the classical special functions of mathematics can be constructed using these ingredients.
For example, Tate’s famous thesis [6] contains (among many other things) a recipe for constructing
the local zeta function of any locally compact, Fourier-self-dual field. Let θ(x) be the fixed point
of the Fourier transform. One should therefore think of it as an analogue of the Gaussian; in fact,

2
θ∞ (x) = e−πx , (5.4)

while θp (x) is the characteristic of the unit ball (p-adic integers).

dx
Z
θ(x)|x|s = ζ(s) (5.5)
|x|

It is then a good exercise to check that

1
ζp (s) = , ζ∞ (s) = π −s/2 Γ(s/2), (5.6)
1 − p−s

so that Tate’s definition recovers precisely the local factors of the “completed” Riemann zeta func-
tion. We refer the reader to e.g. [15] for a more thorough discussion of p-adic Lie groups, special
functions, and character theory.
There are natural pseudodifferential operators, most easily defined in momentum space as mul-
tiplication by |k|s . (Here s is a parameter controlling the “order” of derivative, which may take on
real values; these operators, which are nonlocal in position space for p-adic fields, are the Vladimirov
derivatives. For a quick review of some of their properties, see the appendix of [17]; a more complete
reference is [46].)
As all of the preceding discussion hopefully suggests, p-adic analogues of familiar field theory
models, such as the O(N ) model, can now be defined straightforwardly. The resulting models
 12

correspond, in the structural way that we have outlined, to the ordinary models; in some sense,
they can be thought of as somehow intermediate between standard continuum field theories and
lattice models. For example, the p-adic line (unlike the lattice) admits a scaling symmetry, but
that symmetry is discrete rather than continuous. There are even some novel features: for instance,
while marking an open set or interval in the real line breaks its translation symmetry, there exists
a nontrivial subgroup of the p-adic translations preserving any chosen open set.
The utility of such models has recurred across mathematics; in general, one can think of a
hierarchical (or p-adic) model as replacing the real number line by an analogue, that has even
more powerful geometric and algebraic structures—at the expense of its one-dimensional structures
(ordering, path connectedness, . . . ) [43]. See also [31] for related discussion.
Computations in theories defined at general places (i.e., over the reals or over the p-adics)
sometimes exhibit universal answers, independent of which place the theory is defined at! The
independence of the result on the place, however, has to be suitably interpreted. What seems to
happen is that the result can be expressed in terms of special functions, which admit a uniform
definition at all places. The numerical values of the results thus do not necessarily agree—but the
structural form of the results is identical.
The example of this I have in mind is a result, due to Gubser, Jepsen, Parikh, and Trundy, for
leading-order anomalous dimensions in the O(N ) model [19]. The model is defined by a standard-
looking action:
Z
dk φi (−k) |k|s + m2 φi (k)


S=
!
λ
Z X
+ dk1 · · · dk4 Ti1 i2 i3 i4 φi1 (k1 )φi2 (k2 )φi3 (k3 )φi4 (k4 ) δ ki . (5.7)
4!
i

The tensor T just contains the three possible contractions of indices,

1
Ti1 i2 i3 i4 = (δi1 i2 δi3 i4 + δi1 i3 δi2 i4 + δi1 i4 δi2 i3 ) (5.8)
3

The anomalous dimensions for the operators φ and φ2 in this model are given by

γφ,φ2 = Resδ=0 gφ,φ2 (δ) + O(1/N 2 ) (5.9)

where the functions gφ,φ2 are given by

1 B(n − s, δ − s)
gφ (δ) = ,
N B(n − s, n − s)
  (5.10)
2 B(n − s, δ − s) 1 B(δ, δ) B(n − s, n − 2s)
gφ2 (δ) = − + 2 −1 .
N B(n − s, n − s) N B(n − s, n − s) B(n − s, n − s)
 13

Note that these results apply equally well to real or to p-adic models, assuming the special functions
involved are defined uniformly! The gamma and beta functions that appear are defined (for a theory
on Rn or Qnp , which here means the unique unramified degree-n extensions of Qp ) by the relations

ζp (s) Γp (t1 )Γp (t2 )

Γp (s) = , Bp (t1 , t2 ) = , (5.11)
ζp (n − s) Γp (t1 + t2 )

p, as usual, is a place (a prime or infinity).

6. EXACTLY SOLVABLE FERMIONIC MODELS

As another example of an interesting model that can be treated uniformly across places (and
exhibits some interesting new features in the non-Archimedean case), one can consider models
of interacting fermions on the line [18], defined in analogy to the recently introduced Klebanov–
Tarnopolsky models.
To define fermionic models on the p-adic line, one needs to use Grassmann field variables. Of
course, this requires that the the symmetric quadratic form (propagator) in the kinetic term be
replaced with an antisymmetric one. Normally, this is achieved by writing an action with one time
derivative, instead of two; however, since any order of Vladimirov derivative defines a symmetric
form, a little more thought is required.
In the real case, the propagator of a free fermion is

1 sgn(k)
G(k) ∼ = . (6.1)
k |k|

So, by analogy, one can antisymmetrize the propagator by using a quadratic multiplicative character
of the field. It’s also possible to introduce more than one flavor of fermion, and to contract flavor
indices with an antisymmetric tensor.
This leads one to the following class of actions for theories of interacting fermions:
1
Z
Sfree = dω φabc (−ω) |ω|sp sgn(ω) φabc (ω),
2
Z (6.2)
′ ′ ′ ′ ′ ′
Sint = dt φabc φab c φa bc φa b c .

Here, the field is either commuting or anticommuting; pairs of flavor indices are contracted either
with δ or with a fixed antisymmetric matrix Ω; and the sign character may be either “even” or “odd,”
meaning that sgn(−1) = ±1. There is already a constraint on these choices: for the kinetic term
to be nontrivial, we must have that σψ σΩ = sgn(−1). In fact, a second constraint appears in the
infrared, so that exactly one specific collection of these choices leads to consistent behavior at each
 14

place! In addition, to ensure that the field has positive scaling dimension and that the interaction
term we write is relevant, we ask that the spectral parameter satisfy 1/2 < s ≤ 1.
Just like the ordinary Klebanov–Tarnopolsky model (and other models of SYK type), this theory
is dominated in the large-N limit by the “melon” diagrams. These melon diagrams can be resummed
into an exact Schwinger–Dyson equation, which determines the two-point function in the interacting
large-N theory. In the limit of large N and weak coupling, with g2 N 3 held fixed, this Schwinger–
Dyson equation takes the form

G = F + σΩ (g2 N 3 ) G ⋆ G3 ⋆ F. (6.3)

Here G is the interacting, and F the free, two-point function. In the infrared, this equation can be
solved, giving a universal limiting behavior:

sgn(t)
G(t) = b , |t| ≫ (g 2 N 3 )1/(2−4s) (6.4)
|t|1/2

where

1
= −σΩ Γ(π−1/2,sgn )Γ(π1/2,sgn ). (6.5)
b4 g 2 N 3

Notice that the scaling in the IR limit is completely independent of the spectral parameter of the
UV theory!
Here we are making use of a generalized gamma function, depending on a multiplicative character
of the field:

dt
Z
Γ(π) = χ(t)π(t). (6.6)
|t|

The characters appearing in (6.5) are just products of characters coming from the norm and from
the specific quadratic character appearing in the propagator, i.e.,

πs,sgn (t) = |t|s sgn(t). (6.7)

For fermionic theories with direction-dependent characters (i.e., characters that are nontrivial
when restricted to Z×
p ), one can do even better. It is possible to explicitly solve the Schwinger-

Dyson equation for behavior at all scales, interpolating between the UV and the universal IR! If we
introduce the notation that F (t) = f (|t|) sgn(t) (and similarly for G), then the Schwinger–Dyson
equation reduces to

g2 N 3 2 4
g = f + σψ |t| g f. (6.8)
p
 15

And this quartic can be explicitly solved for g when σψ = −1. However, when σψ = +1 (i.e., the
fields are bosonic), (6.8) does not have a well-defined limit in the deep infrared. As such, while
fermionic theories have a well-defined two-point function at all scales, the IR limit of theories with
bosonic fields appears to be problematic.

7. HOLOGRAPHY

In all of what we’ve said so far, we’ve just discussed p-adic analogues of Poincaré-invariant
field theory. We haven’t said much about conformal transformations, or about the AdS/CFT
correspondence, at all. But the original motivation behind recent papers revisiting p-adic field
theory was to investigate whether the discrete scale symmetry present in such models could be used
to construct analogues of AdS/CFT that naturally live on discrete bulk spaces. We will say a few
words about this now, beginning with a brief reminder of some of the central ideas of AdS/CFT.
Let Rn,m denote an (n + m)-dimensional real space, equipped with the standard metric of
signature (n, m). m will be zero for Euclidean signature, and one for Lorentzian; when it does not
appear, it is taken to be zero. The group of conformal transformations of Rn,m is SO(n + 1, m + 1).
And the anti-de Sitter space H of signature (n, m) can be constructed by taking the vectors of unit
norm inside Rn,m+1 ; SO(n, m + 1) naturally acts on this space, and it has a universal cover which
is H n,m . For m = 0, we are just constructing the standard hyperbolic space. Note also that H is a
maximally symmetric space: it can be thought of as the quotient

H n = SO(n, 1)/K, (7.1)

where K ∼
= SO(n) is a maximal compact subgroup.
So the group Conf(n) = SO(n + 1, 1) naturally appears acting on two different spaces: on H n+1
by isometries, and on Rn (or its compactification S n ) by conformal transformations. In fact, there is
a good reason for this. While a metric on H n+1 does not induce a metric on the boundary at infinity
(because the metric diverges there), the conformal class of the induced metric is well-defined. So
isometries act on the asymptotic boundary S n in a way that preserves the conformal structure.
As we’ve emphasized, symmetries of a space X can often be expected to act on a theory defined
on X. As such, there are two natural ways to make Conf(n)-invariant (Euclidean) theories: One
can take X = S n (or Rn ), and look for conformal theories: fixed points of renormalization group
flow. Or, one can choose X = H n+1 , and take any field theory!
It is then perhaps natural to wonder if a field theory on hyperbolic spacetime naturally couples
 16

to a (conformally invariant) field theory at asymptotic infinity. Another clue was given by the
Bekenstein–Hawking formula for the entropy of a black hole,

c3
S= A, (7.2)
4~G

where A is the area of the event horizon. Speaking very loosely, one normally expects the entropy
to be a count of the number of degrees of freedom, and so to be extensive: that is, for a field theory,
it should scale with the volume of a region. But Bekenstein and Hawking’s results (along with
other ideas in general relativity) were interpreted to suggest that the entropy in a theory of gravity
scales with the area of the boundary of the region. This is intuitively because of the presence of an
enormous group of gauge symmetries (diffeomorphisms).
The AdS/CFT correspondence [20, 30, 51], which grew out of observations like these, states that
(certain) conformally invariant theories on S n are equivalent to gravitational theories on H n+1 . The
original construction of a pair of theories proposed to be dual in this fashion relied on string theory;
the worldvolume theory of a large stack of coincident D3-branes (maximally supersymmetric Yang–
Mills theory on R4 , in the large-color limit) is expected to be dual to the gravitational theory defined
by type-IIB string theory on the near-horizon geometry of the branes, H 5 × S 5 .
But a precise understanding of the correspondence involves not just the prediction that two
theories are equivalent, but a set of statements about how constituents of (and computations in)
the two theories are meant to correspond to one another. The first few entries in this dictionary
were worked out in bottom-up fashion by Witten [51], who proposed the following ansatz:
 Z 
exp φ0 O = Zbulk (φ0 ) ≈ exp (−Scl (φ0 )) . (7.3)
CFT

Here φ0 is a particular boundary condition at asymptotic infinity for the bulk field φ. The statement
is thus that fields in the bulk correspond to operators in the boundary theory, with the coupling
by taking the boundary value of the field as a smearing function for insertions of the corresponding
operator in the CFT. The inverse transformation is possible because of a particular crucial prop-
erty of the Klein–Gordon (or Laplace) equation on hyperbolic space: it has precisely one solution
extending any chosen boundary condition at asymptotic infinity. This makes the correspondence
sensible in the limit where the bulk dynamics are classical.
The existence of a unique solution to the generalized Dirichlet problem for the equations of
motion can be expressed in terms of a bulk-to-boundary Green’s function, which is the unique
solution extending a delta-function boundary condition. This Green’s function is a key object in
 17

AdS/CFT. In the half-space model of H n+1 , i.e. R+ × Rn with metric

1 X 2
ds2 = dxi , (7.4)
x20 i

such a Green’s function for a massless field is explicitly given by

xn0
K(x) = . (7.5)
(x20 + x21 + · · · + x2n )n

As one might expect, both fields and operators are classified by representations of the same
group, Conf(d), and so the Casimir operators also correspond; in particular, the mass of the bulk
field corresponds to the conformal dimension of the boundary operator. The precise correspondence
between these two quantities can be extracted from the asymptotic behavior of K(x), by demanding
that (7.3) make sense; the scaling dimension κ of the operator must therefore be

1 p 
κ= d + d2 + 4m2 . (7.6)
2

This also suggests the intuitive idea that the extra, radial coordinate of H plays the role of a scale
in the boundary theory.
Now, in low dimensions, due to exceptional isomorphisms, the whole story of conformal invari-
ance can be formulated algebraically, without reference to a metric [32]:

S 2 = P 1 (C), H 3 = SL(2, C)/K,

(7.7)
where K = SU (2), Conf(2) ∼
= P GL(2, C).

Similarly,

S 1 = P 1 (R), H 2 = SL(2, R)/K,

(7.8)
where K = SO(2)), Conf(1) ∼
= P GL(2, R).

Now that everything is in purely algebraic language, depending only on a choice of field, we can
simply make another choice! For example, we can take the p-adics:

∂Tp = P 1 (Qp ), Tp = SL(2, Qp )/K

(7.9)
where K = SL(2, Zp )), Conf p ∼
= P GL(2, Qp ).

Here Tp (the p-adic maximally symmetric space) is the famous “tree” of Bruhat and Tits: it is a
uniform infinite tree of valence p + 1. As we hoped, Tp is naturally a discrete space, equipped with
a metric—the ideal setting for lattice field theory. And there is another natural “round” measure,
the Patterson–Sullivan measure dµ0 (x) on P 1 (Qp ), which stands in the same relation to the Haar
measure as the Fubini–Study metric does to the translation-invariant metric on C.
 18

Now, it’s off to the races: One can try to understand or formulate the simplest pieces of the
holographic dictionary, starting with free bulk scalar fields propagating without backreaction on Tp ,
just as in Witten’s paper [21, 24]. All of the necessary facts—compatible group actions on bulk
and boundary, a bulk Klein-Gordon equation with a well-posed Dirichlet problem at infinity—are
available; in fact, scalar fields on Tp were thoroughly studied in the context of the p-adic string
theory (see, for example, [52]). The Klein–Gordon equation is

X
φ(v ′ ) − φ(v) = m2 φ,


△φ = (7.10)
v′ ∼v

and it admits a bulk-to-boundary Green’s function solution, analogous to K(x):

K(v) = pκhv,xi . (7.11)

This allows one to reconstruct the unique solution extending a given boundary condition, according
to the formula

p
Z
φ(v) = dµ0 (x) φ0 (x)phv,xi . (7.12)
p+1 Qp

Here hv, xi is the distance from v to the boundary point x, regularized to be zero at the (arbitrary)
center vertex C of the tree. κ is a parameter; a moment’s thought shows that the pair of κ and x
is identical data to the wavevector k of an ordinary plane wave, corresponding to its magnitude
and direction respectively, and that κhv, xi is a precise analogue of k · x. The corresponding mass
eigenvalue is

m2κ = pκ + p1−κ − (p + 1). (7.13)

√
Thus, the BF bound is m2κ ≥ −( p − 1)2 . Just as in the normal case, there are solutions in the
bulk whose mass-squared is negative (but bounded from below)! Bulk fields of mass mκ couple to
boundary operators of conformal dimension κ (for a study of p-adic CFT, see e.g. Melzer [34]):

1
hOκ (x)Oκ (y)i ∼ . (7.14)
|x − y|2κ
p

Pleasingly, some intuitive features of ordinary AdS/CFT—for example, that the radial coordi-
nate represents a scale in the boundary theory—are present even more sharply in the p-adic case.
For instance, if the boundary field φ0 is a single mode (something like an additive character of Qp ,
or one of the p-adic wavelets of Kozyrev [26]), it stops contributing to the reconstruction of bulk
physics abruptly, at a height determined by its wavelength!
 19

To be precise, we can take the boundary condition to be of the form


exp 2πi{kx}, x ∈ Zp ,

φ0 (x) = (7.15)
0

x 6∈ Zp .

Let v be a vertex in the branch of the tree above Zp , at a depth ℓ such that 0 ≤ ℓ ≤ − ordp (k) − 1.
Then the bulk function φ(v), reconstructed according to the prescription (7.12), is zero. The
standard intuition that moving along the radial coordinate corresponds to a renormalization group
scale in the boundary theory is thus confirmed.

8. ANALOGUES OF GRAVITY: EDGE LENGTH DYNAMICS

Since a crucial ingredient of ordinary AdS/CFT is the fact that the bulk theory includes gravity,
it is natural to wonder what an analogue of gravity for the tree would look like. There is no
completely satisfactory answer yet, but I will briefly describe one attempt to construct a theory for
metric degrees of freedom [17].
The only obvious metric-type data in the bulk space Tp are the edge lengths. Unlike an actual
metric on a manifold, they don’t have any obvious gauge freedom (since they are geodesic lengths
and hence physical), but they do couple naturally to matter through the Laplacian. They are
naturally set to a uniform value by the construction of the tree as a maximally symmetric space;
when they are deformed to a nonconstant configuration, the group action on the tree is broken, but
of course this is analogous to the normal case.
So it is natural to try allowing the edge lengths to fluctuate dynamically. Since the simplest
possible coupling to matter is obvious, it remains only to come up with a sensible action principle
for the edge lengths, which is some kind of analogue of the Einstein–Hilbert action for the metric.
A plausible choice of action can be defined using a notion of Ricci curvature for graphs [29, 38],
which has been studied in the mathematics literature.
Since there are no loops in a tree-like graph and therefore no “plaquette”-type operators like those
studied in lattice gauge theory, it does not seem possible to define a curvature as the commutator of
covariant derivatives. Rather, the graph Ricci curvature is defined in a global fashion, by computing
the rate of change in the average (Wasserstein) distance between two separated heat kernels at t = 0.
On a tree-like graph, it reduces to the following function on edges of the graph:
bxy  X  b 
xy
X 
κxy = bxy − bxxi + bxy − byyi
dx dy
≡ κx→y + κy→x .
 20

xi x y yi

b2xx′ . As a straightforward sanity check, the

P
Here b denotes an inverse edge length, and dx =
x′ ∼x
Bruhat–Tits tree has constant negative curvature:

q−1
κxy = −2 . (8.1)
q+1

So it seems reasonable to use κ to write an analogue of the Einstein–Hilbert term. The resulting
action for a massive scalar field minimally coupled to edge length fluctuations, looks as follows:
X b2e
 X
2 1 2 2
S= (κe − 2Λ) + |δe φ| + m φv . (8.2)
e
2 v
2

Now, in light of the holographic ansatz (7.3), one is interested in the on-shell value of the classical
action. But this is actually infinite for the solution corresponding to a uniform tree! Just like in
ordinary gravity, we need to regularize by a suitable choice of the constant term Λ, and must also
include a suitable boundary term (the analogue of the Gibbons–Hawking–York term) to ensure that
the action makes sense when truncated to a finite region.
Let Γ ⊂ Tp be a finite, connected subgraph, such that all vertices of Γ have valence either p + 1
or 1. ∂Γ is then the set of univalent vertices. Then, one can consider truncating the action to Γ:
the action must be modified to
X X
SΓ = SEH + Sbdy = (κe − 2Λ) + ℓx , (8.3)
e∈Γ x∈∂Γ

where
X
ℓx = K + κx→y . (8.4)
y∼x
y ∈Γ
/

There are now two undetermined constants, Λ and K, in it; one constraint ensures that S remains
finite as Γ → Tp :

2q
K= Λ + q. (8.5)
q−1

When this constraint is satisfied, the on-shell action is

 
q+1
Scl = (2 − 2g) 1 + Λ . (8.6)
q−1
 21

Here g is the genus (the result holds for the tree of any Mumford curve/higher-genus black hole).
So the on-shell action is in fact topological!
This suggests that this model is, perhaps, more like dilaton gravity in two dimensions than
honest Einstein gravity. Further evidence for that interpretation is provided by holographically
computing correlation functions of the operator T dual to edge-length fluctuations. While the two-
point functions are as one would expect for a massless bulk field, the three-point function actually
vanishes identically, at least up to possible contact terms!
Just for the sake of completeness, here are some of those results more explicitly. The two point
function of T is
pn ζ(2n) 1
hT (z1 )T (z2 )i = . (8.7)
4 ζ(n)2 |z12 |2n
For an operator O of dimension κ,
−ζ(n)ζ(2κ) 1
hT (z1 )O(z2 )O(z3 )i = . (8.8)
ζ(2κ − n)ζ(−κ)ζ(κ − n) |z12 | |z13 |n |z23 |2κ−n
n

And finally,

hT (z1 )T (z2 )T (z3 )i = 0! (8.9)

The reader is referred to [17] for more details.

9. OUTLOOK

Throughout this story, there is a long list of excellent questions, unclear points, and other things
that remain (at least to the present author) opaque. We hope to continue to address some of these
questions in future work, and hope that others will be motivated to take up the challenge as well.
For a start, there is an enormous amount of literature having to do with statistical mechanics
models or spin systems defined on similar trees [2]. It would be interesting to find any connection
between such models and lattice field theories defined on trees, or p-adic boundary field theories.
Based on the earlier discussion of the function field analogy, the reader may have wondered if
a similar story is possible for local fields of finite characteristic (i.e., fields like Fq formal Laurent
series). It would be interesting to think about such function field analogues in more detail, as
well as about adèlic versions of the story. Furthermore, it is perhaps natural to wonder if there
is any natural notion of field theory associated to the global object. While this is somewhat wild
speculation, thinking about the function field case might be an easier and illustrative setting in
which to try and define global notions.
 22

On a somewhat more down-to-earth note, one also still lacks a clear candidate for a pair of dual
theories, analogous to the pair provided by string theory in the normal case (although some ground-
up steps in this direction have been taken [22]). One part of the problem is that no p-adic theory
nearly as complicated as maximally supersymmetric Yang–Mills seems to be readily available; one
mostly has free theories, or theories of scalar fields with simple polynomial interactions.
In the years since the AdS/CFT correspondence was first proposed, work has appeared that
proposes versions of the holographic correspondence for simpler boundary theories, which are dual
to correspondingly more complicated bulk gravitational theories; the O(N ) model, in particular, is
meant to be related to a higher-spin gravity theory, as studied by Vasiliev [25, 44, 45]. Relatedly,
it has been proposed [8] that holographic duals of (Archimedean) bilocal free field theories can be
constructed by using the exact renormalization group equations due to Polchinski. It would be very
interesting to find such a construction for p-adic theories; we hope to report progress on this soon.
Another route of attack on this problem would involve coming up with more honest analogues
of fields of nonzero spin. This would hopefully allow one to begin talking about gauge interactions,
as well as more full-fledged analogues of gravity. The difficulty in doing this can be seen in several
ways: for example, in the inability to construct any “plaquette”-type operators as discussed above,
or in the difficulty of defining a good notion of conserved current in p-adic field theory. Some
vague speculation about using multiplicative characters of the field is given in [24], but this remains
largely uncharted territory.
It would also be interesting to make a connection back to the subject of “tensor network” models,
which are the other widespread class of discrete toy systems that are meant to relate to holography.
Indeed, these models were part of the original motivation for the study of holography over the
p-adics. In such a model, the bulk path integral is typically represented by the composition of
many tensors or linear maps, acting between local (finite-dimensional) Hilbert spaces. One class of
such tensors (“perfect” tensors or quantum error-correcting codes) are naturally associated to the
algebraic data of the tree, in particular to the geometry of algebraic curves over finite fields; such
codes are reviewed in [24].
It is worth mentioning that there is, at least in a schematic sense, a natural tensor network
(albeit with infinite-dimensional local Hilbert spaces) which is the path integral of the bulk field.
Given a region Γ ⊂ Tp , the Hilbert space is supposed to be constructed as the space of wave-
functionals over the space of boundary conditions of fields; that is, for a single real scalar, it is (at
least roughly) a tensor product of copies of L2 (R), one for each vertex v ∈ ∂Γ. The wave-functional,
or time evolution matrix, is then determined by doing the path integral over the vertices in the
 23

interior of Γ with prescribed boundary conditions. Evolution “across one vertex” thus corresponds
to the linear operator from L2 (R) to L2 (R)⊗p , with integral kernel

!
X
K(x; xν ) = exp − (x − xν )2 − µx2 . (9.1)
ν

Is there a precise sense in which this tensor network exhibits quantum error-correction properties?
Does it admit natural truncations to a low-energy approximation, in which the local Hilbert spaces
are finite-dimensional? If so, is there any connection of the resulting tensors to the algebrogeo-
metric codes mentioned before? Or is there any other sort of natural path-integral model or spin
system on Tp in which algebrogeometric perfect tensors naturally appear? We plan to address these
questions in future work [23].

Acknowledgements

The author is deeply grateful to the organizing committee of the sixth international conference
on p-adic mathematical physics and its applications, for the invitation to speak and for an enor-
mously productive and enjoyable week. Special thanks are owed to W. Zúñiga, for his unsurpassed
hospitality, as well as to B. Dragovich, for the kind invitation to contribute this article to the con-
ference proceedings and for his efforts towards their publication. The author also thanks all of the
participants of the conference, for numerous good questions and insight-laden conversation; I am
grateful for everything I learned from you all.

None of this work would have been possible without the diligence and talent of my collaborators.
I am so fortunate to have had the chance to work with you all. Thanks to each of you: S. Gubser,
M. Heydeman, C. Jepsen, M. Marcolli, S. Parikh, B. Stoica, B. Trundy, and P. Witaszczyk. Grat-
itude is due also to the many friends and colleagues who have patiently listened to chatter about
this work, provided opportunities for it to be spoken about, and offered invaluable conversation
and feedback. Thanks to you all, and especially to T. Dimofte, V. Disarlo, R. Eager, M. Fluder,
V. Forini, J. Keating, M. Kim, P. Longhi, T. McKinney, B. Pozzetti, M. Salmhofer, A. Schwarz,
J. Walcher, A. Waldron, D. Xie, W. Yan, S.-T. Yau, and M. Zidikis.

The author’s work is supported in part by the Deutsche Forschungsgemeinschaft, within the
framework of the Exzellenzinitiative an der Universität Heidelberg; he wishes to also express thanks
to Roost Coffee & Market, where a large fraction of this manuscript was prepared.
 24

[1] Artin, E. and G. Whaples (1945). Axiomatic characterization of fields by the product formula for
valuations. Bulletin of the American Mathematical Society 51 (7), 469–492.
[2] Baxter, R. J. (1985). Exactly solved models in statistical mechanics. World Scientific.
[3] Bhattacharyya, A., L.-Y. Hung, Y. Lei, and W. Li (2017). Tensor network and (p-adic) AdS/CFT.
arXiv preprint arXiv:1703.05445 .
[4] Brekke, L. and P. G. O. Freund (1993). p-adic numbers in physics. Phys. Rept. 233, 1–66.
[5] Cassels, J. W. S. (1986). Local fields. Cambridge University Press.
[6] Cassels, J. W. S. and A. Fröhlich (1967). Algebraic number theory. Academic Press.
[7] Donaldson, S. K. (1983). An application of gauge theory to four dimensional topology. Journal of
Differential Geometry 18 (2), 279–315.
[8] Douglas, M. R., L. Mazzucato, and S. S. Razamat (2011). Holographic dual of free field theory. Physical
Review D 83 (7), 071701.
[9] Dragovich, B., A. Y. Khrennikov, S. V. Kozyrev, and I. V. Volovich (2009). On p-adic mathematical
physics. p-adic Numbers, Ultrametric Analysis, and Applications 1 (1), 1–17.
[10] Dragovich, B., A. Yu. Khrennikov, S. V. Kozyrev, I. V. Volovich, and E. I. Zelenov (2017). p-adic
mathematical physics: the first 30 years. p-adic Numbers, Ultrametric Analysis and Applications 9 (2),
87–121.
[11] Dutta, P., D. Ghoshal, and A. Lala (2017). Notes on exchange interactions in holographic p-adic cft.
Physics Letters B 773, 283–289.
[12] Dyson, F. J. (1969). Existence of a phase transition in a one-dimensional Ising ferromagnet. Commun.
Math. Phys. 12, 91–107.
[13] Freund, P. G. O. and M. Olson (1987). Nonarchimedean strings. Phys. Lett. B199, 186.
[14] Freund, P. G. O. and E. Witten (1987). Adelic string amplitudes. Phys. Lett. B199, 191.
[15] Gel’fand, I. M., M. I. Graev, and I. Piatetskii-Shapiro (1968). Representation theory and automorphic
functions. Saunders.
[16] Gubser, S. S. A p-adic version of AdS/CFT. arXiv:1705.00373.
[17] Gubser, S. S., M. Heydeman, C. Jepsen, M. Marcolli, S. Parikh, I. Saberi, B. Stoica, and B. Trundy
(2017). Edge length dynamics on graphs with applications to p-adic AdS/CFT. JHEP 06, 157.
[18] Gubser, S. S., M. Heydeman, C. Jepsen, S. Parikh, I. Saberi, B. Stoica, and B. Trundy (2017). Signs
of the time: Melonic theories over diverse number systems. arXiv preprint arXiv:1707.01087 .
[19] Gubser, S. S., C. Jepsen, S. Parikh, and B. Trundy (2017). O(N ) and O(N ) and O(N ).
arXiv:1703.04202.
[20] Gubser, S. S., I. R. Klebanov, and A. M. Polyakov (1998). Gauge theory correlators from noncritical
string theory. Phys. Lett. B428, 105–114.
[21] Gubser, S. S., J. Knaute, S. Parikh, A. Samberg, and P. Witaszczyk (2016). p-adic AdS/CFT. Com-
 25

munications in Mathematical Physics 352 (3), 1019–1059. arXiv:1605.01061.

[22] Gubser, S. S. and S. Parikh (2017). Geodesic bulk diagrams on the Bruhat–Tits tree. arXiv:1704.01149.
[23] Heydeman, M., M. Marcolli, S. Parikh, and I. Saberi. To appear.
[24] Heydeman, M., M. Marcolli, I. Saberi, and B. Stoica (2016). Tensor networks, p-adic fields, and
algebraic curves: arithmetic and the AdS3 /CFT2 correspondence. arXiv:1605.07639.
[25] Klebanov, I. R. and A. M. Polyakov (2002). AdS dual of the critical O(N ) vector model. Physics
Letters B 550 (3), 213–219.
[26] Kozyrev, S. V. (2002). Wavelet analysis as p-adic spectral analysis. Izvestia Academii Nauk Seria
Math. 66 (2).
[27] Lerner, E. Yu. and M. D. Missarov (1989). Scalar models of p-adic quantum field theory and hierarchical
models. Theor. Math. Phys. 78, 177–184.
[28] Lerner, E. Yu. and M. D. Missarov (1994). Fixed points of renormalization group for the hierarchical
fermionic model. Journal of statistical physics 76 (3-4), 805–817.
[29] Lin, Y., L. Lu, and S.-T. Yau (2011). Ricci curvature of graphs. Tohoku Math. J. (2) 63 (4), 605–627.
[30] Maldacena, J. M. (1999). The large N limit of superconformal field theories and supergravity. Int. J.
Theor. Phys. 38, 1113–1133.
[31] Manin, Yu. I. (1991). Three-dimensional hyperbolic geometry as ∞-adic Arakelov geometry. Inventiones
mathematicæ 104 (2), 223–244.
[32] Manin, Yu. I. and M. Marcolli (2002). Holography principle and arithmetic of algebraic curves. Adv.
Theor. Math. Phys. 5, 617–650.
[33] Marshakov, A. V. and A. V. Zabrodin (1990). New p-adic string amplitudes. Mod. Phys. Lett. A5, 265.
[34] Melzer, E. (1989). Nonarchimedean conformal field theories. International Journal of Modern Physics
A 4 (18), 4877–4908.
[35] Missarov, M. D. (1991). Renormalization group and renormalization theory in p-adic and adelic scalar
models. Adv. Soviet Math 3, 143–164.
[36] Missarov, M. D. (2012). p-adic renormalization group solutions and the euclidean renormalization
group conjectures. p-adic Numbers, Ultrametric Analysis, and Applications 4 (2), 109–114.
[37] Nakayama, Y. (2015). Scale invariance versus conformal invariance. Physics Reports 569, 1–93.
[38] Ollivier, Y. (2009). Ricci curvature of Markov chains on metric spaces. Journal of Functional Analy-
sis 256 (3), 810 – 864.
[39] Osborne, T. J. and D. E. Stiegemann (2017). Dynamics for holographic codes. arXiv preprint
arXiv:1706.08823 .
[40] Poonen, B. (2006). Lectures on rational points on curves. Available from
http://www-math.mit.edu/~poonen/papers/curves.pdf.
[41] Ruelle, P., E. Thiran, D. Verstegen, and J. Weyers (1989a). Adelic string and superstring amplitudes.
Mod. Phys. Lett. A4, 1745.
[42] Ruelle, P., E. Thiran, D. Verstegen, and J. Weyers (1989b). Quantum mechanics on p-adic fields. J.
 26

Math. Phys. 30, 2854.

[43] Tao, T. (July 27, 2007). Dyadic models. Article in What’s new
(https://terrytao.wordpress.com/2007/07/27/dyadic-models/).
[44] Vasiliev, M. A. (1992). More on equations of motion for interacting massless fields of all spins in 3 + 1
dimensions. Physics Letters B 285 (3), 225–234.
[45] Vasiliev, M. A. (2013). Holography, unfolding and higher spin theory. Journal of Physics A: Mathe-
matical and Theoretical 46 (21), 214013.
[46] Vladimirov, V. S., I. V. Volovich, and E. I. Zelenov (1994). p-adic Analysis and Mathematical Physics.
World Scientific.
[47] Volovich, I. V. (1987). p-adic string. Classical and Quantum Gravity 4 (4), L83.
[48] Weil, A. (1939). Sur l’analogie entre les corps de nombres algébriques et les corps de fonctions al-
gébriques. Revue Scient 77, 104–106.
[49] Wilson, K. G. and J. B. Kogut (1974). The renormalization group and the ǫ expansion. Phys. Rept. 12,
75–200.
[50] Witten, E. (1988). Topological quantum field theory. Communications in Mathematical Physics 117 (3),
353–386.
[51] Witten, E. (1998). Anti-de Sitter space and holography. Adv. Theor. Math. Phys. 2, 253–291.
[52] Zabrodin, A. V. (1989). Nonarchimedean strings and Bruhat–Tits trees. Commun. Math. Phys. 123,
463–483.