The Riemann zeros as energy levels of a Dirac fermion

in a potential built from the prime numbers in Rindler spacetime

Germán Sierra
Instituto de Fı́sica Teórica, UAM-CSIC, Madrid, Spain
(Dated: August 4, 2014)
We construct a Hamiltonian HR whose discrete spectrum contains, in a certain limit, the Riemann
zeros. HR is derived from the action of a massless Dirac fermion living in a domain of Rindler space-
time, in 1+1 dimensions, that has a boundary given by the world line of a uniformly accelerated
observer. The action contains a sum of delta function potentials that can be viewed as partially
reflecting moving mirrors. An appropriate choice of the accelerations of the mirrors, provide primi-
tive periodic orbits associated to the prime numbers p, whose periods, measured by the observer’s
clock, are log p. Acting on the chiral components of the fermion χ∓ , HR becomes the Berry-Keating
Hamiltonian ±(xp̂ + p̂x)/2, where x is identified with the Rindler spatial coordinate and p̂ with the
arXiv:1404.4252v3 [math-ph] 1 Aug 2014

conjugate momentum. The delta function potentials give the matching conditions of the fermion
wave functions on both sides of the mirrors. There is also a phase shift eiϑ for the reflection of
the fermions at the boundary where the observer sits. The eigenvalue problem is solved by transfer
matrix methods in the limit where the reflection amplitudes become infinitesimally small. We find
that for generic values of ϑ the spectrum is a continuum, where the Riemann zeros are missing, as
in the adelic Connes model. However, for some values of ϑ, related to the phase of the zeta function,
the Riemann zeros appear as discrete eigenvalues immersed in the continuum. We generalize this
result to the zeros of Dirichlet L-functions, associated to primitive characters, that are encoded in
the reflection coefficients of the mirrors. Finally, we show that the Hamiltonian associated to the
Riemann zeros belongs to class AIII, or chiral GUE, of Random Matrix Theory.

I. INTRODUCTION

A century ago, Pólya and Hilbert suggested that the imaginary part of the non trivial zeros of the Riemann zeta
function, would be the oscillation frequencies of a physical system. The reality of these frequencies will provide a
proof of the celebrated Riemann hypothesis (RH) [1] which has deep consequences for the distribution of the prime
numbers [2]-[7]. There are evidences that the Riemann zeros are the eigenvalues of a quantum Hamiltonian: i) the
Montgomery-Odlykzo law according to which the statistical distribution of the zeros is given, locally, by the Gaussian
Unitary Ensemble (GUE) of Random Matrix Theory (RMT) [8]-[12]; ii) analogies between trace formulas relating
periods of classical trajectories and spectra in Quantum Chaos Theory and explicit formulas relating prime numbers
and Riemann zeros [13]-[18], and iii) Selberg’s trace formula relating the lengths of the geodesics on a compact Riemann
surface with negative curvature and the eigenvalues of the Laplace-Beltrami operator [19, 20]. The picture proposed
by Berry in 1986, is that the binomium primes/zeros is similar to the binomium classical/quantum for a dynamical
chaotic system [13]. Furthermore, it was conjectured that the classical Hamiltonian underlying the Riemann zeros
should be quasi-one dimensional, breaking time reversal symmetry and with isolated periodic orbits whose periods
are the logarithm of the prime numbers (see [16, 18] for reviews of this approach, [21, 22] for general references and
[23]-[26] for introductions and historical background).
In 1999 Berry and Keating (BK) [27, 28], and Connes [29], suggested that a spectral realization of the Riemann
zeros could be achieved from the quantization of the simple classical Hamiltonian Hcl = xp, where x and p are
the position and momentum of a particle moving in the real line. The xp Hamiltonian is one dimensional, breaks
the time reversal symmetry, is integrable, not chaotic, with unbounded classical trajectories and its quantization
yields a continuum spectrum [32, 33]. The connection found by Berry, Keating and Connes between xp and the
Riemann zeros, was semiclassical, and relied on two different regularizations schemes. Berry and Keating introduced
a Planck cell regularization of the phase space imposing the constraints |x| ≥ `x and |p| ≥ `p with `x `p = 2π~,
obtaining semiclassical energies that agree asymptotically with the average location of the Riemann zeros [27]. In
Connes’s work, there is a cutoff Λ, and the constraints |x| ≤ Λ and |p| ≤ Λ, such that in the limit Λ → ∞, the
semiclassical spectrum becomes a continuum with missing spectral lines associated to the average Riemann zeros
[29]. The interpretation of the zeros as missing spectral lines, would also explain a mysterious sign problem in the
fluctuation term of the number of zeros [29]. The possible connection between xp and the Riemann zeros motivated
several works in the past two decades, some of them will be discussed in more detail below [30]-[53].
The previous semiclassical versions of xp were formulated as consistent quantum mechanical models in references
[38, 43, 44, 47]. Connes’s version was realized in terms of a charge particle moving in a box of size Λ × Λ, and subject
to the action of a uniform perpendicular magnetic field and an electrostatic potential xy [38]. For strong magnetic
fields the dynamics is restricted to the lowest Landau level, where the xy potential acts effectively as the quantum xp
 2

Hamiltonian. In this realization, the smooth part of the counting formula of the Riemann zeros appears as a shift of
the energy levels (that become a continuum in the limit Λ → ∞) and not as an indication of missing spectral lines.
The Landau model with xy potential, has been used in an analogue model of Hawking radiation in a quantum Hall
fluid [54].
On the other hand, the Berry-Keating version of xp was revisited recently using the classical Hamiltonian Hx/p =
x(p + `2p /p) defined in the half line x ≥ `x [43, 44] (hereafter denoted as the x/p model). The role of the term
`2p x/p is to bound the classical trajectories which become periodic, unlike the trajectories of xp that are unbounded.
√ √
The x/p Hamiltonian can be quantized in terms of the operator Ĥx/p = x(p̂ + `2p p̂−1 ) x, where p̂ = −i~d/dx
∞
and (p̂−1 ψ)(x) = −i/~ x dy ψ(y), and its spectrum agrees asymptotically with the average Riemann zeros provided
R

`x `p = 2π~. A similar result was obtained by Berry and Keating using the Hamiltonian HBK = (x + `2x /x)(p + `2p /p)
that is invariant under the exchange x ↔ p [47]. These two works provided an spectral realization of the average
Riemann zeros, but not of the actual zeros. From the Quantum Chaos perspective, the reason of this failure lies in the
fact that these variants of xp are non chaotic and does not contain periodic orbits related to the prime numbers [47].
More generally, any one dimensional classical and conservative Hamiltonian is integrable and therefore non chaotic,
which seems to lead to nowhere.
In this work we propose a solution of the puzzle that leads to a spectral realization of the Riemann zeros. The main
ideas can be explained as follows. Let us consider a chaotic billiard in two spatial dimensions, such as the Sinai’s
billiard [55]-[58]. A ball thrown with some energy follows chaotic trajectories that in most cases cover the entire table,
except for a discrete set of periodic trajectories, whose periods, that are independent of the energy, dominate the path
sum that gives rise to the Gutzwiller formula for the fluctuations of the energy levels. Unlike this, a one dimensional
billiard, made of two walls, will be boring since the ball will go back and forth periodically between the walls.
Let us now take semitransparent walls, so that with a certain probability the ball passes through or bounces off. In
such a billiard the particle may follow several trajectories depending on the outcome at each wall. One may say this
is a quasi-one dimensional billiard. In Sinai’s billiard, or in the motion on compact Riemann surfaces, the particle
follows geodesics, which implies that the periods of the closed orbits are independent of the energy. In the quasi-1D
billiard, one can achieved the same property by choosing massless particles, say photons or massless fermions, whose
trajectories lie on the light cone in Minkowsky space-time. The soft walls should then be viewed as semitransparent
mirrors, or beam splitters.
The last ingredient one has to incorporate into the quasi-1D billiard, or rather the array of mirrors, is chaos. In table
billiards, chaos is generated by a border that defocuses the trajectories, and in compact billiards, chaos is produced
by the negative curvature of the space that separates nearby trajectories exponentially fast. If the 1D mirrors stay
at fix positions, nearby light-ray trajectories will stay close in space-time. However, if the mirrors are accelerated,
then slightly delayed light rays, will generally have their reflected rays departing exponentially fast from one another
after several reflections. Hence in this model, the source of chaos is acceleration. The simplest situation is when the
mirrors are uniformly accelerated, in which case they are called moving mirrors in the literature of Quantum Field
Theory in curved spacetimes [59]. We shall then consider an infinite array of moving mirrors whose accelerations,
and reflection properties, will be used to encode number theoretical information. In particular, we shall choose the
accelerations inversely proportional to a power of integers, that leads to the appearance of primitive periodic orbits
whose periods are the logarithms of the prime numbers. These periods are measured by a moving observer whose
acceleration sets the units of this magnitude.
This model realizes in a relativistic framework, Berry’s suggestion of associating primitive periodic orbits to prime
numbers [13]. Quantum mechanically, the waves propagating in the array generate an interference pattern that encodes
the accelerations and reflection coefficients of the mirrors. Here we find several situations: destructive interference
where the Riemann zeros are missing spectral lines as in the adelic Connes’s model and constructive interference where
the Riemann zeros are point like spectrum embedded in a continuum. In both cases, the connection between the
spectrum of the model and the Riemann zeros involves a limit where the reflection amplitudes vanish asymptotically.
This limit is analogue to the semiclassical limit ~ → 0, that leads to the Gutzwiller formula, that was the starting
point of the analogies between Quantum Chaos and Number Theory.
The paper is organized as follows. In section II we review the basic definitions of Rindler spacetime that describes
the geometry of the model. We formulate the massive Dirac fermion in a domain of the Rindler space-time, we find
the Hamiltonian, study its relation with xp and recover the spectrum of the x/p model, obtaining the interpretation
of the parameters `x and `p as inverse acceleration and fermion mass. In section III we construct an ideal array
of moving mirrors with accelerations c2 /`n and study the reflections of the light rays emitted and absorbed by an
observer with acceleration c2 /`1 . Using special√relativity, we show that the proper time of the observer’s clock is
proportional to log `n , and that the choice `n ∝ n singularizes the trajectories associated to the prime numbers. We
also propose an array where `n ∝ en/2 , which has regular trajectories, and whose discrete spectrum is proportional
to the integers (this model will be denoted harmonic). In section IV we construct the Hamiltonian of a massless
 3

x0 x0

S
Sc
x1 {x x1
{x

FIG. 1: Regions S and Sc defined in eq. (8). The union of S and Sc is the right Rindler wedge (4).

Dirac fermion with delta function potentials associated to the moving mirrors of section III. We derive the matching
conditions for the wave functions and show that the corresponding Hamiltonian is self-adjoint for generic values of
the accelerations and the reflection coefficients. The eigenvalue problem for the Hamiltonian is formulated using
transfer matrix methods, and in a semiclassical limit we find the conditions for the existence of discrete eigenvalues.
In so far the construction is general, but then we analyze several examples and make contact with the Riemann zeta
function and the Dirichlet L−functions. In section V, the identify the symmetry class of the Hamiltonian under time
reversal (T), charge conjugation (C) and parity (P). In the conclusions we summarize our results and discuss future
developments. In Appendix A we derive the spectrum of the harmonic model.

II. THE MASSIVE DIRAC FERMION IN RINDLER SPACETIME

A. Rindler spacetime

We start with some basic definitions and set up our conventions. The 1+1 dimensional Minkowski space-time is
defined by a pair of coordinates (x0 , x1 ) ∈ IR2 , a flat metric ηµν , with signature (−1, 1), and line element

ds2 = ηµν dxµ dxν , −η00 = η11 = 1, η01 = η10 = 0, (1)

that is invariant under translations and Lorentz transformations (we set the units of the speed of light c = 1). The
change of coordinates,

x0 = ρ sinh φ, x1 = ρ cosh φ, (2)

brings (1) to the form

ds2 = dρ2 − ρ2 dφ2 . (3)

ρ and φ are the Rindler space and time coordinates respectively [60, 61]. The Rindler metric (3) is invariant under
shifts of φ, generated by the Killing vector ∂/∂φ. Restricting ρ to positive values, defines the region called right
Rindler wedge [60]-[62],

R+ = {(ρ, φ) | ρ > 0, −∞ < φ < ∞} , (4)

that in Minkowski coordinates is described by the quadrant x1 > |x0 |. The value ρ = 0 represents a horizon of the
metric (3), that is similar to the horizon of a black hole, the reason being that the Schwarzschild metric, near the
black hole horizon, is approximately the Rindler metric [61].
Rindler spacetime is the natural arena to study the physical phenomena associated to accelerated observers [60],
such as the Unruh effect [62, 63]. Let us consider an observer whose world line is given by (2), with ρ = ` > 0 a
 4

constant value. The proper time, τ , measured by the observer is defined by ds2 = −dτ 2 , and its relation to φ follows
from (3)
τ = ` φ. (5)
plus an additive constant that is set to 0 in (5). The observer’s trajectory (2), written in terms of its proper time,
reads
x0 = ` sinh(τ /`), x1 = ` cosh(τ /`), (6)
µ
and has a constant proper acceleration, a, defined as the Minkowski norm of the vector a
d2 xµ 1
aµ = , aµ aµ = −a2 → a = . (7)
dτ 2 `
Restating the speed of light, a = c2 /`. This quantity plays an central role in the Unruh effect according to which an
observer, with proper acceleration a, detects a thermal bath with temperature TU = a/(2π) (in units c = ~ = TB = 1)
[63]. Replacing a by the surface gravity κ of an observer near the horizon of a black hole, yields the Hawking
temperature TH = κ/(2π) [64, 65]. The similarity between the Hawking and Unruh formulas lies in the equivalence
principle of General Relativity.
In our approach to a spectral realization of the Riemann zeros, we shall introduce an observer with acceleration,
a1 = 1/`1 . The observer’s world line, ρ = `1 , divides R+ into the regions S and Sc located to her right and left,
S = {(ρ, φ) | ρ ≥ `1 , −∞ < φ < ∞, } , (8)
Sc = {(ρ, φ) | 0 < ρ ≤ `1 , −∞ < φ < ∞, } ,
such that R+ = S ∪ Sc . Let us next study the dynamics of a Dirac fermion in S.

B. Dirac fermion

A representation of the Dirac’s gamma matrices in 1+1 dimensions is given by [66]

γ 0 = σx , γ 1 = −iσ y , γ 5 ≡ γ 0 γ 1 = σz , (9)
a µ
where σ (a = x, y, z) are the Pauli matrices. γ satisfy the Clifford algebra
{γ µ , γ ν } = −2η µν ,
 µ 5
γ , γ = 0, (µ = 0, 1). (10)
In an abuse of notation, γ 5 denotes the 1+1 analogue of the 1+3 gamma matrix γ 5 = γ 0 γ 1 γ 2 γ 3 , and it defines the
chirality of the fermions. A Dirac fermion ψ is a two component spinor
 
ψ− † †
ψ= , ψ̄ = ψ † γ 0 = (ψ+ , ψ− ), (11)
ψ+
†
where ψ± are the conjugate of ψ± . The fields ψ∓ are the chiral components of ψ, namely γ 5 ψ∓ = ±ψ∓ . Let us
introduce the light cone coordinates x± , and the derivatives ∂± = ∂/∂x±
x± = x0 ± x1 = ±ρ e±φ , (12)
1 1
∂± = (∂0 ± ∂1 ) = ± e∓φ (∂ρ ± ρ−1 ∂φ ),
2 2
where the Minkowski metric (1) becomes ds2 = −dx+ dx− . Under a Lorentz transformation with boost parameter λ,
that is velocity v = tanh λ, the light cone coordinates and the Dirac spinors transform as
x± → e∓λ x± , ∂± → e±λ ∂± , ψ± → e±λ/2 ψ± , (13)
and the Rindler coordinates as
φ → φ − λ, ρ → ρ. (14)
Hence the spinors χ±
χ± = e±φ/2 ψ± , (15)
remain invariant under (14). The spaces S, Sc and R+ are mapped into themselves under Lorenzt transformations.
 5

C. Dirac action of a massive fermion

The Dirac action of a fermion with mass m in the space-time domain S is (units ~ = c = 1)
Z
i
S = d2 x ψ̄(∂/ + im)ψ, (16)
2 S
Z  
i † † im  † †
= dx+ dx− ψ− ∂+ ψ− + ψ+ ∂− ψ+ + ψ− ψ+ + ψ+ ψ−
2 S 2
i ∞
Z Z ∞ 
1 1
= dφ d ρ χ†− (∂φ + ρ∂ρ + )χ− + χ†+ (∂φ − ρ∂ρ − )χ+
2 −∞ `1 2 2
 i
† †
+ imρ χ− χ+ + χ+ χ− .

S has a boundary ∂S corresponding to the worldline ρ = `1 > 0. The variational principle applied to (16) gives the
Dirac equation
(∂/ + im)ψ(x) = 0, x ∈ S, (17)
and the boundary condition
εµν ẋµ ψ̄(x)γ ν δψ(x) = 0, x ∈ ∂S, (18)
µ
where δψ is an infinitesimal variation of ψ, εµν is the Levi-Civita tensor (ε01 = −ε10 = 1, εµµ = 0), and ẋ =
dxµ (`1 , φ)/dφ is the tangent to ∂S in the Rindler coordinates (2). The Dirac equation (17) reads in components
im
(∂0 ∓ ∂1 )ψ± + imψ∓ = 0 −→ ∂∓ ψ± + ψ∓ = 0. (19)
2
In the massless case, the fields ψ± decouple and describe a right moving fermion, ψ+ (x+ ), and a left moving fermion,
ψ− (x− ), in terms of which one can construct a Conformal Field Theory with central charge c = 1 [66]. The mass term
couples the two modes and therefore conformal invariance is lost. The action principle applied to the last expression
of eq.(16) gives
1
(∂φ ± ρ∂ρ ± )χ∓ + imρχ± = 0, (20)
2
and the boundary condition
χ†− (`1 , φ) δχ− (`1 , φ) = χ†+ (`1 , φ) δχ+ (`1 , φ), ∀φ. (21)
Eqs.(20) and (21) are of course equivalent to (19) and (18) respectively. The infinitesimal generator of translations of
the Rindler time φ, acting on the fermion wave functions, is the Rindler Hamiltonian HR , that can be read off from
(20)
 
χ−
i∂φ χ = HR χ, χ= , (22)
χ+

−i(ρ ∂ρ + 21 )
 
mρ √ √
HR = = ρ p̂ρ ρ σ z + mρ σ x , (23)
mρ i(ρ ∂ρ + 12 )
where p̂ρ = −i∂/∂ρ, is the momentum operator associated to the radial coordinate ρ. The operator

1 1 √ √
Hxp = −i(ρ ∂ρ + ) = (ρ p̂ρ + p̂ρ ρ) = ρ p̂ρ ρ, (24)
2 2
coincides with the quantization of the classical xp Hamiltonian proposed by Berry and Keating [27], where x is the
radial Rindler coordinate. The eigenfunctions of (24) are

1
Hxp ψE = E ψE , ψE = √ ρ−1/2+iE , (25)
2π
with eigenvalues E in the real numbers IR if ρ > 0 [32, 33]. Thus HR consists of two copies of xp, with different
signs corresponding to opposite chiralities that are coupled by the mass term. In CFT the operator HR , with m = 0,
corresponds to the sum of Virasoro operators L0 + L̄0 that generate the dilation transformations.
 6

D. Self-adjointess of HR

The action (16) is invariant under the U (1) transformation ψ → eiα ψ. The corresponding Noether current is
J µ = ψ̄γ µ ψ, and is conserved, i.e. ∂µ J µ = 0. The charge associated to J µ can be integrated along the space-like line
(2) with constant φ. Using (15) one finds

Z Z
µ ν
dρ χ∗− χ− + χ∗+ χ+ ,


dx νµ ψ̄γ ψ = (26)
φ=cte

where χ∗± are the complex conjugate of χ± . Hence the scalar product of two wave functions, in the domain S, can be
defined as

Z ∞
dρ χ∗1,− χ2,− + χ∗1,+ χ2,+ .


hχ1 |χ2 i = (27)
`1

The Hamiltonian HR is hermitean (or symmetric) respect to this scalar product if

hχ1 |HR χ2 i = hHR χ1 |χ2 i, (28)
when acting on a subspace of the total Hilbert space. Partial integration gives
Z ∞
dρ∂ρ ρ χ∗1,+ χ2,− − χ∗1,− χ2,+ .


hχ1 |HR χ2 i − hHR χ1 |χ2 i = (29)
`1

Hence HR is hermitean provided

lim ρ1/2 χ± (ρ) = 0, χ∗1,− χ2,− − χ∗1,+ χ2,+ = 0, at ρ = `1 . (30)

ρ→∞

The latter condition is equivalent to eq.(21). The solution of (30) is

− ieiϑ χ− = χ+ at ρ = `1 , (31)
where ϑ ∈ [0, 2π). The quantity −ieiϑ has the physical meaning of the phase shift produced by the reflection of the
fermion with the boundary. It will play a very important role in what follows.
HR is not only hermitean but also self-adjoint. According to a theorem due to von Neumann, an operator H
is self-adjoint if the deficiency indices n± are equal [67, 68]. These indices are the number of linearly independent
eigenfunctions of H with positive and negative imaginary eigenvalues,
n± = dim {ψ± | H ψ± = ±zψ± , Im z > 0}. (32)
If n+ = n− = 0, the operator H is essentially self-adjoint, while if n+ = n− > 0, then H admits self-adjoint extensions.
In the case of HR , one finds n+ = n− = 1, and the self-adjoint extensions are parameterized by the phase eiϑ in
eq.(31).

E. Spectrum of HR

The eigenvalues and eigenvectors of the Hamiltonian (23), are given by the solutions of the Schroedinger equation
HR χ = E χ, χ± (ρ, φ) = e−iEφ f± (ρ), ρ ≥ `1 , (33)
that satisfy the boundary condition (31). E is the Rindler energy. The equations for f± (ρ) that follows from (23) are
 
1
ρ∂ρ + ± iE f± ∓ imρf∓ = 0, (34)
2
that lead to the second order differential eqs.
" 2 #
2

2 ∂ ∂ 1 2 2
ρ +ρ − ± iE − m ρ f± = 0, (35)
∂ρ2 ∂ρ 2
 7

whose general solution is a linear combination of the modified Bessel functions [69]

f± (ρ) = C1 e∓iπ/4 K 21 ±iE (mρ) ± C2 e∓iπ/4 I 12 ±iE (mρ). (36)

The phases e±iπ/4 follows from (34). Let us compute the deficiency indices of HR , for which it is enough to take
E = ±i/2,

f+ (ρ) = C1 e−iπ/4 K0 (mρ) + C2 e−iπ/4 I0 (mρ),


i
E= → (37)
2 f− (ρ) = C1 eiπ/4 K1 (mρ) − C2 eiπ/4 I1 (mρ),
f+ (ρ) = C1 e−iπ/4 K1 (mρ) + C2 e−iπ/4 I1 (mρ),

i
E=− → (38)
2 f− (ρ) = C1 eiπ/4 K0 (mρ) − C2 eiπ/4 I0 (mρ).

The functions I0,1 (mρ) diverge exponentially as ρ → ∞, that forces C2 = 0, while K0,1 (mρ) decreases exponentially
and are normalizable provided `1 > 0. The deficiency indices are therefore equal, n+ = n− = 1 and the eigenfunctions
are (setting C1 = 1)

f± (ρ) = e∓iπ/4 K 21 ±iE (mρ), ρ ≥ `1 , (39)

which yields
χ± (ρ, φ) = e−iEφ ∓iπ/4 K 21 ±iE (mρ), (40)

up to a common normalization constant. Plugging (40) into (31) yields the equation for the eigenenergies
eiϑ K 21 −iE (m`1 ) − K 12 +iE (m`1 ) = 0. (41)

This equation has positive and negative solutions, but only if ϑ = 0 or π, they come in pairs {En , −En }. Moreover,
if ϑ = 0, then E0 = 0 is also an eigenvalue. The imaginary part of the Riemann zeros also form pairs {tn , −tn } (i.e.
ζ(1/2 ± itn ) = 0), and t = 0 is not a zero since ζ(1/2) 6= 0. This situation corresponds to the choice ϑ = π.
The number of eigenvalues, n(E), in the interval [0, E] (with E > 0) is given in the asymptotic limit E  m`1 by
 
E 2E ϑ
n(E) ' log −1 − + O(E −1 ), E  m`1 . (42)
π m`1 2π
For E < 0 there is a similar formula with ϑ → −ϑ. Let us compare this expression with the Riemann-Mangoldt formula
that counts the number of zeros, N (t), of the zeta function ζ(s) that lie in the rectangle 0 < Re s < 1, 0 < Im s < t
[2]
N (t) = hN (t)i + Nfl (t) (43)
 
θ(t) t→∞ t t 7
hN (t)i = + 1 −→ log −1 +
π 2π 2π 8
 
1 1
Nfl (t) = Im log ζ + it = O(log t)
π 2
where hN (t)i is the average term and Nfl (t) the oscillation term. This expression agrees with (42), to order t log t and
t, with the identifications
t
E= , m`1 = 2π. (44)
2
However, the constant term 7/8 is not reproduced by eq.(42) for ϑ = π, which is the choice for the absence of the
zero mode E0 = 0 (note that neither ϑ = 0 gives the 7/8).

Comments:
• In the limit m`1 → 0, the spectrum (42) becomes a continuum. This situation arises in three cases: 1) if m > 0
is kept constant and `1 → 0, the domain S becomes R+ , and one recovers the spectrum of a massive Dirac
equation in R+ ; 2) if `1 > 0 is kept constant, and m → 0, the particles become massless and the effect of the
boundary ∂S, is to exchange left and right moving fermions, that constitutes a boundary CFT [66], and 3) m
and `1 → 0 that corresponds to a massless fermion in R+ .
 8

• Equation (41) coincides with the spectrum of the quantum x/p Hamiltonian [43, 44]
√
√
Hx/p = x p̂ + `2p p̂−1 x, x ≥ `x , (45)
with the identifications
Ex/p
E= , m = `p , `1 = `x . (46)
2
The origin of this coincidence lies in the fact that a classical Hamiltonian of the form H = U (x)p + V (x)/p [44]
can be formulated as a massive Dirac model in a space-time metric built from the potentials U (x) and V (x)
[49]. In the case of the x/p model this metric is flat, which allows us to formulate this model in terms of the
Dirac equation in the domain S.
• Gupta et al. proposed recently the Hamiltonian (xp /+p/x)/2 as a Dirac variant of the xp Hamiltonian [48]. The
former Hamiltonian is defined in 2 spatial dimensions and after compactification of one coordinate becomes 1D,
with an spectrum that depends on a regularization parameter and which is similar to the one found from the
Landau theory with electrostatic potential xy [38].
• Burnol has studied the causal propagation of a massive boson and a massive Dirac fermion in the Rindler right
edge R+ relating the scattering from the past light cone to the future light cone to the Hankel transform of zero
order and suggesting a possible relation to the zeta function [70].

F. General Dirac action

For later purposes we shall introduce the general relativistic invariant Dirac action in 1+1 dimensions
Z
i
S = / + im − m0 γ 5 )ψ,
d2 x ψ̄(∂/ + iA (47)
2 S

where, in addition to the mass term, there is a chiral interaction ψ̄γ 5 ψ and a minimal coupling ψ̄ Aψ / to a vector
potential Aµ (A / = γ µ Aµ ). Moreover m, m0 can be functions of the space-time position, in which case m becomes a
scalar potential and m0 a pseudo scalar potential. We saw above that for a constant mass m the action (16) can be
restricted to the domain S, preserving the invariance under shifts of the Rindler time φ. It is clear that if m depends
in ρ, but not in φ, the action remains invariant under translations of φ, and that the Hamiltonian is equal to (23),
with m replaced by m(ρ). The same happens with the term ψ̄γ 5 ψ if m0 = m0 (ρ). Concerning the vector potential,
the action (47) is invariant provided

A0 ± A1 = a± (ρ) e∓φ , (48)

in which case (47) becomes

Z ∞ Z ∞
i 1
S = dφ d ρ χ†− (∂φ + ρ∂ρ + + ia+ ρ)χ− (49)
2 −∞ `1 2
1 i
+ χ†+ (∂φ − ρ∂ρ − + ia− ρ)χ+ +(im + m0 )ρχ†− χ+ + (im − m0 )ρχ†+ χ− .
2
Using gauge transformations one can reduce the number of fields in (49). This issue will be consider below in a
discrete realization of this model. The equations of motion derived from (49) are

1
(∂φ ± ρ∂ρ ± + ia± ρ)χ∓ + (im ± m0 )ρχ± = 0, (50)
2
and can be written as the Schroedinger equation (22) with Hamiltonian

−i(ρ ∂ρ + 21 ) + a+ ρ (m − im0 )ρ
 
HR = (51)
(m + im0 )ρ i(ρ ∂ρ + 21 ) + a− ρ)
√ √ a+ + a− a+ − a− z
= ρ p̂ρ ρ σ z + ρ1 + ρ σ + mρ σ x + m0 ρ σ y ,
2 2
 9

x0
x0

4
x1
1 2 3

x1

FIG. 2: Left: worldlines of the mirrors with accelerations an = 1/n that corresponds to the choice `n = n (n = 1, 2, . . . ). The
case of eq.(52), is a simple rescaling of lengths. S is the grey region. Right: A light ray (dotted line) is sent at x0 = 0 by
the observer at x1 = `1 . The ray either passes through the 2nd mirror or is reflected back to the 1st mirror, where is reflected
totally.

that acts on the wave functions that satisfy the boundary condition (31).
In summary, we have shown in this section, that the spectrum of the Rindler Hamiltonian, in the domain S agrees
asymptotically with the average Riemann zeros, under the identifications (44). However, there is no trace of their
fluctuations that depend on the zeta function on the critical line. This observation is no surprising since the prime
numbers should be included into the model. In the next section we shall take a first step in that direction.

III. MOVING MIRRORS AND PRIME NUMBERS

Here we construct an ideal optical system, in Rindler space-time, that allows to distinguish prime numbers from
composite. In the next section we shall give a concrete realization of this optical system in the Dirac model. The
system consists of an infinite array of mirrors, labelled by the integers n = 1, 2, . . . , ∞, that have the following
properties:

• The first mirror, n = 1, is perfect, while the remaining ones, n > 1, are one-way mirrors (beam splitters) that
reflect and transmit the light rays partially. The light rays can be replaced by massless fermions.
• The mirrors move in Minkowski space-time with uniform accelerations an , with a1 > a2 > . . . , such that
limn→∞ an = 0.
• At time x0 = 0, the mirrors are placed at the positions `n = 1/an (units c = 1).
• The worldlines of the mirrors are contained in the domain S, whose boundary corresponds to the first mirror,
n = 1, such that ρ = `1 .
• An observer carries the first mirror, and sends and receives light rays whose departure and arrival times she
measures with a clock.
• The lengths `n are given by

`n = `1 n1/2 , n = 1, 2, . . . , ∞. (52)

This ansatz can be replaced by nα (α > 0), but the parameter α can be set to 1/2 by scaling the clock’s ticks.
Figure 2 depicts the mirror’s worldlines satisfying these conditions with α = 1.
 10

We shall next study the propagation of light rays in this optical array using the laws of special relativity. Let us
consider a light ray emanating from the point (ρ1 , φ1 ) and reaching the point (ρ2 , φ2 ), where (ρi , φi ) are the Rindler
coordinates. Along this trajectory the line element (3) vanishes,
ρ2
dρ = ±ρ dφ → ρ2 = ρ1 e±(φ2 −φ1 ) → φ2 − φ1 = | log |, (53)
ρ1
corresponding to right moving (ρ2 > ρ1 ) or left moving (ρ1 > ρ2 ) rays. Suppose that the ray is emitted at the first
mirror at time φin = 0, i.e. (ρin , φin ) = (`1 , 0), moves rightwards, reflects on the nth mirror and returns to the first
mirror, (ρend , φend ) = (`1 , φn ). The value of φn follows from eq.(53) and (52)
`n
φn = 2 log = log n, n > 1. (54)
`1
which is twice the change in φ from ρ = `1 to ρ = `n . φn can be measured by the observer’s clock traveling with the
(1)
perfect mirror where the ray was emitted and received. The change in the clock’s proper time is given by τn = `1 φn
(see eq.(5)), which in units of `1 = 1, reads
τn = log n, n > 1. (55)
Hence, measuring τn , the observer can find the value of n. If n = 2, the ray travels forth and back between the mirrors
n = 1 and n = 2. However if n > 2, the ray must pass through the intermediate mirrors 2, . . . , n − 1. Let us next
analyze the case of two reflections. Now the ray is emitted at (ρin , φin ) = (`1 , 0), reaches the mirror n1 , returns to
the perfect mirror, reflects again, reaches the mirror n2 and comes back finally to the perfect mirror where the clock
records the proper time, τn1 ,n2 that is given by the sum of the intermediate times (55)
τn1 ,n2 = τn1 + τn2 = log(n1 n2 ), n1,2 > 1. (56)
Hence the measurement of τn1 ,n2 allows the observer to compute the product n1 n2 . There are more complicated cases
as the one illustrated by the following sequence
1 → n1 → n2 → n3 → 1, (57)
1 < n1 > n2 < n3 > 1,
where the ray emitted by the observer is reflected by the mirror n1 , back to the mirror n2 < n1 , that reflects the ray
forwards to the mirror n3 > n2 , that reflects the ray back to the first mirror. The proper time recorded by the clock
is
n1 n3
τn1 ,n2 ,n3 = τn1 − τn2 + τn3 = log . (58)
n2
Notice that the case n2 = 1 reproduces eq.(56). It is easy to derive a general formula for the proper time elapsed for
a trajectory involving 2k intervals, that starts and ends at the first mirror,
1 → n1 → n2 → . . . → n2k−1 → n2k = 1, (59)
1 < n1 > n2 < . . . < n2k−1 > n2k = 1,
and is given by

n1 n3 . . . n2k−1
τ{ni } = log . (60)
n2 n4 . . . n2k
The numerator of (60) corresponds to reflections: right mover → left mover, while the denominator corresponds to
reflections: left mover → right mover. The argument of log in (60) will be in general a rational number. Let us
suppose it is the prime p. It is clear, from (55), that after one reflection, i.e. k = 1, the observer will detect one ray
arriving at τp = log p. Suppose now that the argument of the log is a composite number n, say 4. At time τ4 = log 4,
the observer will detect the ray reflected from the forth mirror, but also one ray from two reflections on the second
mirror τ2,2 = 2τ2 , which is of course the same as τ4 . In a real experiment the two rays arriving at the observer will
interfere. The study of this interference is left to the next section.
The previous example suggests that prime numbers correspond to unique paths characterized by observer proper
times equal to log p. Let us prove this statement in the case k = 2. Equation (58) becomes
n1 n3
= p, (61)
n2
 11

where n1,2,3 satisfy the constraints (57). According to (61), p divides the product n1 n3 , so it is a prime factor of n1
or n3 . In the former case one has
n01 n3
n1 = p n01 (n01 ≥ 1) → = 1, (62)
n2
which is a contradiction because n2 < n3 by eq.(57). The same result holds if p divides n3 . The generalization to any
k ≥ 2 goes as follows. Suppose that τ{ni } = log p, then from eq.(60)
n1 n3 . . . n2k−1
= p. (63)
n2 n4 . . . n2k−2

If p divides, say n1 , one gets

n01 n3 . . . n2k−1
n1 = p n01 (n01 ≥ 1) → = 1, (64)
n2 n4 . . . n2k−2

which cannot be satisfied because n2 < n3 , . . . , n2k−2 < n2k−1 by the conditions (59). This proves that the observer
detects a single ray only when it comes from the reflection on a prime mirror, while it detects more than one ray
when the rays comes from composite mirrors. This interpretation is purely classical because it presuposes that the
rays can be distinguished, and disregards the interference effects. Both effects have of course to be taken into account
in a realistic implementation using identical particles, such as photons or masless fermions. The interference pattern
emerging for fermions in this array will be purpose of the next section. In any case, one can easily show that this
classical model can be used to implement the classical Eratosthenes’s sieve to construct prime numbers.

Comments:

• We mentioned in the introduction, that similarities between counting formulas in Number Theory and Quantum
Chaos led Berry to conjecture the existence of a classical chaotic Hamiltonian whose primitive periodic orbits
are labelled by the primes p, with periods log p, and whose quantization will give the Riemann zeros as energy
levels. A classical Hamiltonian with this property has not yet been constructed, but the mirror system presented
above, displays some of its properties. In particular, the rays associated to prime numbers behave as primitive
orbits with a period log p. Moreover, the trajectories and periods of these primitive rays are independent of
their energy, that is the frequency of the light.
• One can construct an array of moving mirrors in the domain Sc (recall (8)) with the same properties as the
array in S. The parameters `n that characterize the array can be labeled with the negative integers,

`n = `1 |n|−1/2 , n = −1, −2, · · · − ∞, (65)

where n = −1 is the perfect mirror located at the boundary ∂Sc = ∂S. The proper time elapsed is now given by
τn = log |n|, and is identical to (55) . The relation between the arrays in S and Sc corresponds to the inversion
transformation ρ → `21 /ρ.
• The array of mirrors defined in S is an analogue computer for the multiplication operation or rather the addition
of log’s. To implement the addition operation we shall define an array of mirror with positions, i.e. inverse
accelerations,

`n = `0 en/2 , n = 0, 1, 2, . . . ∞. (66)

Where n = 0 labels the perfect mirror. The analogue of eqs.(55) and (56) is (with `0 = 1)

τn = n, (67)
τn1 ,n2 = n1 + n2 .

This is an harmonic array in the sense that the proper times (67), as well as the discrete spectrum (see Appendix
A) are given by integer numbers. The harmonic array shows that it is not enough to have accelerated mirrors
to generate chaos. The location/accelerations of the mirrors is essential. In the harmonic case, the exponential
separation of the mirrors (66) compensates the exponential time dependence of the ray trajectories (53), that
give rise to a tessellation of Rindler space-time. In the case of (52), the whole set of ray trajectories does not
tessellate the Rindler spacetime which is a manifestation of the arithmetic chaos.
 12

IV. MASSLESS DIRAC FERMION WITH DELTA FUNCTION POTENTIALS

In this section we present a mathematical realization of the moving mirrors in the Dirac theory. We start from
the massless Dirac action and represent the mirrors by delta function potentials that are obtained discretizing the
interacting terms in the general action (49). The mass term becomes a contact interaction that turns left moving
fermions into right moving ones, and viceversa. The new action remains invariant under translations in Rindler time,
so that the Rindler Hamiltonian is a conserved quantity and we look for its spectrum using transfer matrix methods.
The Dirac equation with delta function potentials requires a special treatment that we describe in detail (see [71]-[76]
for the Dirac equation with delta interactions in Minkowski coordinates).

A. Discretization in Rindler variables

Let us consider the integral

Z ∞ Z ∞
1
dρ ρ f (ρ) = du f (u), u = ρ2 , u1 = `21 , (68)
`1 2 u1

where f (ρ) is a generic function, and dρ ρ is the radial part of the measure d2 x = dφ dρ ρ. The Rindler time φ can be
easily incorporated into the equations. We shall discretize (68), using the positions `n of the mirrors given in eq.(52),
which amounts to a partition of the half-line u ≥ u1 into segments of equal width `21 , separated by the points

uk = `2k = `21 k, k = 1, 2, . . . . (69)

Let us discretize (68) as

∞ ∞
`21 X `2 ∞
Z Z
1 X
du f (u) → f (uk ) → 1 du f (u) δ(u − uk ). (70)
2 u1 2 2 u1
k=2 k≥2

If f is a continuous function the last expression of (70) coincides with the middle one, but if f is discontinuous one
gets
∞ ∞ −
`21 X f (u+
Z
1 k ) + f (uk )
du f (u) → , (71)
2 u1 2 2
k=2

where we used
Z v+
1
du f (u)δ(u − v) = (f (v + ) + f (v − )), (72)
v− 2
±
f (v ) = lim f (v ± ).
>0→0

In the limit `1 → 0, the last expression in eq.(70) converges towards the integral (68) for well behaved functions.
Replacing (69) into (70) yields

Z ∞ X `2
1
dρ ρ f (ρ) δ(ρ − `k ). (73)
`1 2`k
k≥2

One could use this formula to discretize the mass term in the Dirac action (16). Taking f = mψ̄ψ this would yield

i
Z X m`2
1
dφ dρ ρ ψ̄ψ δ(ρ − `k ). (74)
2 S 2`k
k≥2

However, the corresponding Dirac equation is problematic, because the matching conditions are not consistent with the
equations of motion, as shown in references [71]-[76]. This problem is solved by replacing the local delta interactions
by separable delta function potentials, that amounts to a point splitting. More concretely, the integral
 13

Z ∞ Z ∞
1
dρ ρ f (ρ)g(ρ) = du f (u) g(u), (75)
`1 2 u1

should be replaced by

`21 R∞
R∞
du u1 du0 f (u)g(u0 ) k≥2 δ(u − uk )δ(u0 − uk ),
P
2 u1
(76)
R∞ R∞ `2
= `1 dρρ `1 dρ0 ρ0 f (ρ)g(ρ0 ) k≥2 2`12 δ(ρ − `k )δ(ρ0 − `k ).
P
k

If f and g are continuous functions at `k , this expression is equivalent to (73).

B. The Dirac action with delta function potentials

Applying the discretization formula (76) to the general action (49) one obtains

SDM = S0 + Sint (77)

where S0 is the massless action

i ∞
Z Z ∞  
† 1 † 1
S0 = dφ d ρ χ− (∂φ + ρ∂ρ + )χ− + χ+ (∂φ − ρ∂ρ − )χ+ , (78)
2 −∞ `1 2 2

and Sint is the discretization of the mass terms and vector potential of (49) in the gauge a+ = a−

Z
i X 2
Sint = dφ dρ ρ dρ0 ρ0 δ(ρ − `k )δ(ρ0 − `k ) (79)
2 S `k
k≥2
h
× irk χ− (ρ, φ) χ− (ρ0 , φ) + irk00 χ†+ (ρ, φ) χ+ (ρ0 , φ)
00 †

i
+(irk + rk0 )χ†− (ρ, φ) χ+ (ρ0 , φ) + (irk − rk0 )χ†+ (ρ, φ) χ− (ρ0 , φ) ,

where
m(`k )`21 m0 (`k )`21 a± (`k )`21
rk = , rk0 = , rk00 = , k ≥ 2, (80)
4`k 4`k 4`k

are dimensionless parameters that are equal to the values taken by the functions m, m0 , a± in (49) multiplied by
`21 /(4`k ). One can verify that the (77) is invariant under the scale transformation

ρ → λρ, χ± → λ−1/2 χ± , `k → λ`k , (81)

so that the physical observables only depend on the parameters rk , rk0 , rk00 . In the Dirac model considered in the
previous section, the mass is constant, which after discretization implies that rn = m`1 /(4n1/2 ). Later on we shall
generalize this eq. to rn ∝ n−σ , and make contact with the Riemann zeta function ζ(σ + it).

C. The Hamiltonian

The equations of motion that follows from (77) are

1 X
− −
`k δ(ρ − `k ) irk00 (χ∓ (`+ 0 +
 
(∂φ ± ρ∂ρ ± )χ∓ + k ) + χ∓ (`k )) + (irk ± rk ) (χ± (`k ) + χ± (`k )) = 0, (82)
2
k≥2
 14

where (72) has been used to integrate around ρ = `k (to simplify the notation the variable φ has been suppressed).
This equation implies that the left and right moving modes propagate freely and independently between the positions
`n ,

(∂φ ± ρ∂ρ ± 12 )χ∓ = 0, ρ 6= `k . (83)

Hence the Hamiltonian is (recall (23)),

√ √ 
ρ p̂ρ ρ 0
HR = √ √ , p̂ρ = −i∂ρ , ρ 6= `n , ∀n. (84)
0 − ρ p̂ ρ

The delta function terms yield the matching conditions between the wave functions on both sides of `k ,
− −
(±1 + irk00 )(χ∓ (`+ 0 +
k ) − χ∓ (`k )) + (irk ± rk )(χ± (`k ) + χ± (`k )) = 0, k ≥ 2, (85)

that can be written in matrix form as

−
Nk,+ χ(`+
k ) = Nk,− χ(`k ), k ≥ 2, (86)

where χ is the two component vector (22) and

1 ± irk00 ±(irk + rk0 )

 
Nk,± = 0 . (87)
±(−irk + rk ) 1 ∓ irk00

These matrices are invertible provided

2 2
det Nk,± = 1 − rk2 − r0 k + r00 k 6= 0. (88)

Otherwise, eqs.(85) become a decoupled set of equations, hence from hereafter we shall assume that condition (88) is
satisfied and leads to

χ(`− +
k ) = Lk χ(`k ), k ≥ 2, (89)

where
−1
Lk = Nk,− Nk,+ (90)
2 2
1 + rk2 + r0 k − r00 k + 2irk00 2(irk + rk0 )
 
1
= 0 2 2 .
1 − rk2 − r0 2k + r00 2k 2(−irk + rk ) 1 + rk + r0 k − r00 k − 2irk00
2

Finally, the variation of the action at the boundary ρ = `1 gives the condition (31)
 
iϑ 1
− ie χ− (`1 ) = χ+ (`1 ) −→ χ(`1 ) ∝ , ϑ ∈ [0, 2π). (91)
−ieiϑ

D. Self-adjointness of the Hamiltonian

One should expect the Hamiltonian (84), acting on wave functions subject to the BC’s (89) and (91), to be self-
adjoint. We shall next show that this is indeed the case. The scalar product, given by eq. (27), will be written
as
∞ Z `n+1
X
dρ χ∗1,− χ2,− + χ∗1,+ χ2,+ .


hχ1 |χ2 i = (92)
n=1 `n

To show that HR is a self-adjoint operator we follow the approach of Asorey et al. based on the consideration of the
boundary conditions that turns out to be equivalent to the von Neumann theorem [77]. The starting point is the
bilinear

hχ1 |HR χ2 i − hHR χ1 |χ2 i ≡ iΣ(χ1 , χ2 ), (93)

 15

where |χ1,2 i are two wave functions. This quantity measures the net flux or probability flowing across the boundary
of the system, which for a unitary time evolution generated by e−iφHR must vanish. The self-adjoint extensions of
HR select subspaces of the total Hilbert space where Σ(χ1 , χ2 ) = 0. In the case of the Hamiltonian (84) one finds

Σ(χ1 , χ2 ) = `1 χ∗1,− (`1 ) χ2,− (`1 ) − χ∗1,+ (`1 ) χ2,+ (`1 )

(94)
X 
∗ + + ∗ + +
+ `n χ1,− (`n ) χ2,− (`n ) − χ1,+ (`n ) χ2,+ (`n )
n≥2

−χ∗1,− (`− − ∗ − −

n ) χ2,− (`n ) + χ1,+ (`n ) χ2,+ (`n ) .

The term proportional to `1 already cancels out by eq.(91). Imposing the independent cancellation of the terms
proportional to `n (n ≥ 2) yields

χ∗1,− (`+ + ∗ + + ∗ − − ∗ − −
n ) χ2,− (`n ) − χ1,+ (`n ) χ2,+ (`n ) = χ1,− (`n ) χ2,− (`n ) − χ1,+ (`n ) χ2,+ (`n ) (95)

which we write as
− −
χ1 (`+ z + z



n )|σ |χ2 (`n ) = χ1 (`n )|σ |χ2 (`n ) , (96)

where |χ) = (χ− , χ+ )t and (χ| = (χ∗− , χ∗+ ). The general solution of (96) is obtained if χ1,2 (`− +
n ) and χ1,2 (`n ) are
related by a transformation

χ(`− +
n , φ) = Un χ(`n , φ), (97)

where Un satisfies

Un† σ z Un = σ z , (98)

which implies that Un belongs to the Lie group U (1) ⊗ SU (1, 1). The non compact character of SU (1, 1) arises in this
problem from to the relative minus sign of the xp terms in the Hamiltonian (84). The U (1) factor can be eliminated
by a phase transformation of the field χ(ρ) in the interval ρ ∈ (`n , `n+1 ), which reduces the group to SU (1, 1), that is
  
a b 2 2
SU (1, 1) = , a, b ∈ C , |a| − |b| = 1 . (99)
b∗ a∗

The matrices Ln (90) are of this form, that ensures that HR is self-adjoint acting on the wave functions satisfy-
ing the conditions (89) and (91). One can further reduce the number of parameters in Ln applying another U (1)
transformation. Let us label the wave functions with an integer n

χn (ρ) = χ(ρ), ρ ∈ (`n , `n+1 ), n = 1, 2, . . . , ∞ (100)

and make the transformation

z
χn (ρ) → eiαn σ χn (ρ), (101)

that induces the following changes in the boundary values

z z
χ(`+
n) → e
iαn σ
χ(`+
n ), χ(`−
n) → e
iαn−1 σ
χ(`−
n ), (102)

and in the Ln matrices (recall (89))

z z
Ln → eiαn−1 σ Ln e−iαn σ . (103)

The parameters αn can then be used to bring (90) into the form
2
1 + rn2 + r0 n 2(irn + rn0 )
 
1
Ln = 2 , n ≥ 2, (104)
1 − rn2 − r0 2n 2(−irn + rn0 ) 1 + rn2 + r0 n

corresponding to an element in the coset SU (1, 1)/U (1). We use the same notation for the transformed parameters
rn , rn0 as in (90).
In summary, we have shown that the Hamiltonian HR , defined in the domain S, is self-adjoint acting on wave
functions that satisfy the BC’s (89) and (91); and is characterized by the set {`n , rn , rn0 }∞
n=2 and ϑ (by scale invariance
(81) we set `1 = 1).
 16

E. Eigenvalue problem

We shall now consider the eigenvalue problem of the Hamiltonian (84), acting on normalizable wave functions
subject to the boundary conditions (89) and (91) The eigenfunctions of HR are given by

 
1
HR χ = E χ → −i ρ ∂ρ + χ∓ = ±E χ∓ → χ∓ ∝ ρ−1/2±iE . (105)
2

so for the nth interval (recall (100))

A∓,n
χ∓,n (ρ) = e±iπ/4 , ρ ∈ (`n , `n+1 ), (106)
ρ1/2∓iE

where A∓,n will depend, in general, on the eigenenergy E and ρ is measured in units of `1 (in what follows we take
`1 = 1). The phases e±iπ/4 have been introduced by analogy with the eigenfunctions (40). The values of χ on both
sides of ρ = `n are given by

± 4 iπ A∓,n
χ∓ (`+
n ) = lim χ∓ (`n + ) = χ∓,n (`n ) = e 1/2∓iE
, (107)
→0+ `n
iπ A∓,n−1
χ∓ (`−
n ) = lim χ∓ (`n − ) = χ∓,n−1 (`n ) = e
± 4
1/2∓iE
.
→0+ `n
Let us write these equations in matrix form in terms of the vector
 
A−,n
|An i = , (108)
A+,n

and the matrix

 
eiπ/4 `iE
n 0
Kn,± = , (109)
0 e−iπ/4 `−iE
n

such that (107) read

−1/2
|χ(`+
n )i = `n Kn |An i, (110)
|χ(`−
n )i = `−1/2
n Kn |An−1 i.

Plugging these eqs. into the matching condition (89) yields

Kn |An−1 i = Ln Kn |An i, (111)

so

|An−1 i = Tn |An i, Tn = Kn−1 Ln Kn , (112)

and substituting (104)

2
1 + rn2 + r0 n 2(rn − irn0 ) `−2iE
 
1 n
Tn = 2 . (113)
1 − rn2 − r0 2n 2(rn + irn0 )`2iE
n 1 + rn2 + r0 n

To simplify the notations let us define

%n = rn − ir0 n , (114)

so that
1 + |%n |2 2%n `−2iE
 
1 n
Tn = T (E, %n , `n ) = . (115)
1 − |%n |2 2%∗n `2iE
n 1 + |%n |2
 17

The parameters %n have the meaning of reflections amplitudes associated to the nth mirror. The absence of a mirror
at the position `n is expressed by the condition %n = 0, so that An−1 = An . Concerning the condition (91), equation
(106) implies

eiϑ A−,1 = A+,1 . (116)

that can be written as

 
1
|A1 i = |A1 (ϑ)i = . (117)
eiϑ

Therefore, the eigenvalue problem has been reduced to find the energies E for which the amplitudes An , satisfying
(112) and (117), yield wave functions (106) that are normalized in the discrete sense (Kronecker delta function) or in
the continuous sense (Dirac delta function), corresponding to the discrete or continuum spectrum of the Hamiltonian.
To this aim, we shall also need the norm and scalar product of the wave functions written in terms of the amplitudes.
Using eq.(92), the norm of (106) is given by
∞
X `n+1
hχ|χi = log hAn |An i, hAn |An i = |A−,n |2 + |A+,n |2 , (118)
n=1
`n

and the scalar product of two eigenfunctions with energies E1 and E2 by

∞
1 X nh iE12 i
(1) ∗ (2)
h i
(1) ∗ (2)
o
hχ1 |χ2 i = `n+1 − `iE
n
12
A−,n A−,n − `−iE 12 −iE12
n+1 − `n A+,n A+,n (119)
iE12 n=1

where E12 = E1 − E2 6= 0 and A±,n ∗ is the complex conjugate of A±,n . Since the Hamiltonian is self-adjoint this
product will vanish.

F. Semiclassical approximation

The recursion relation (112), together with the initial condition (117), gives all the vectors Ak in terms of A1 (ϑ)

|Ak i = Tk−1 Tk−1

−1
· · · T2−1 |A1 (ϑ)i, k ≥ 2. (120)

Except for some simple cases, as the harmonic model (see Appendix A), it will impossible to find close analytic
expressions of the product of matrices appearing in (120). The only hope to made progress is to evaluate (120) in the
limit where the coefficients %n are infinitesimally small. We shall then assume that %n are proportional to a parameter
ε that will be taken to zero at the end of the computation. This parameter plays the role of Planck’s constant, so the
limit ε → 0, will be interpreted as semiclassical. This interpretation is supported by the discretization of the massive
Dirac equation, that led to eq.(80), according to which ρn = m`1 /(4n1/2 ). The connection with the average Riemann
zeros was achieved for m`1 = 2π, that corresponds in the semiclassical xp Berry-Keating model to the Planck constant
`x `p = 2π~.
Taking %n = O(ε), the matrix Tn given in eq.(115) can be replaced in the limit ε → 0, by

2%n `−2iE
 
0
Tn ≈ 1 + τn ≈ eτn + O(ε2 ), τn = n . (121)
2%∗n `2iE
n 0

Tn can be expressed exactly as the exponential of a matrix of the form of τn , but in the limit ε → 0, it will converge
towards the expression given in (121) up to order ε2 . Plugging (121) into (120) yields

|Ak i = e−τk e−τk−1 . . . e−τ2 |A1 (ϑ)i + O(ε2 ). (122)

It is convenient to define a matrix τ1 , of order ε, corresponding to the choice n = 1 of τn , which does not depend on
E because `1 = 1. The vector |A1 (ϑ)i can be replaced by e−τ1 |A1 (ϑ0 )i, where ϑ0 is equal to ϑ up to terms of order
ε. Eq.(122) can then be written as

|Ak i = e−τk e−τk−1 . . . e−τ1 |A1 (ϑ)i + O(ε2 ). (123)

 18

where ϑ0 has been replaced by ϑ since they become the same quantity in the limit ε → 0. The product of exponentials
of matrices can be approximated by the Baker-Campbell-Haussdorf formula [68]

eεA eεB = eε(A+B) + O(ε2 ), (124)

that yields
Pk
|Ak i = |ATk i + O(ε2 ), |ATk i ≡ e− n=1 τk
|A1 (ϑ)i. (125)
Let us next define
k
X k
X
−iΦk
e Rk = %n `−2iE
n , eiΦk
Rk = %∗n `2iE
n , (126)
n=1 n=1

where Rk and Φk are real for real values of E. The factor 2 multiplying %n in eq.(121) can be absorbed into the
parameter ε, so without loss of generality we can write
x
+sin Φk σ y )
|ATk i = e−Rk (cos Φk σ |A1 (ϑ)i, (127)
whose norm is
hATk |ATk i = e2Rk (1 − cos(Φk − ϑ)) + e−2Rk (1 + cos(Φk − ϑ)). (128)
The discrete eigenvalues of the Hamiltonian are those for which the norm (118) is finite, which is ensured if (128)
vanishes sufficiently fast when k → ∞. The approximation we have performed above is a sort of inverse Trotter-Suzuki
decomposition, e.g. limn→∞ (eA/n eB/n )n = eA+B , where a product of exponentials of non commuting operators is
replaced by the exponential of their sum [78, 79]. Let us consider some examples.

1. Harmonic model

This model is defined by the parameters (see eq.(66) with `0 = 1)

`n = en/2 , %n = ε, n = 0, . . . , ∞. (129)
Let us remind that in this case the mirrors (delta functions) are labelled by n = 1, 2, . . . , while the boundary is located
at `0 = 1. ε can be positive or negative and is the semiclassical parameter. This model has an exact solution given
in Appendix A, that allow us to verify the semiclassical approximation done below. From (126) one finds
k
(
−iΦk
X
−inE
ε(k + 1) E ∈ 2π ZZ
e Rk = ε e = sin((k+1)E/2) −ikE/2 . (130)
ε sin(E/2) e E∈ / 2π ZZ
n=0

In the case E ∈ 2π ZZ , one can choose Rk = ε(k + 1) and Φk = 0. This identification is not unique but simplifies the
calculation. The norm (128) is given by

E ∈ 2π ZZ → hATk |ATk i = e2ε(k+1) (1 − cos ϑ) + e−2ε(k+1) (1 + cos ϑ), (131)

while the norm of the wave function χT , whose amplitudes are ATk , reads (see eq. (118))
∞ ∞
1X T T 1 X h 2ε(n+1) i
hχT |χT i = hAn |An i = e (1 − cos ϑ) + e−2ε(n+1) (1 + cos ϑ) . (132)
2 n=0 2 n=0

This series diverges for all values of ϑ different from 0 and π. This means that the energies E = 2πn, do not belong
to the spectrum. On the other hand, if
ϑ = 0 for ε > 0, or ϑ = π for ε < 0 (133)
the state has the norm
1 1
hχT |χT i = → , ε → 0, (134)
e2|ε| − 1 2|ε|
 19

The parameter ε can be absorbed into the normalization constant of the state. Hence, in the two cases (133) there is
an infinite number of normalized states with energies Em = 2πm (m ∈ ZZ ).
In the case E ∈
/ 2π ZZ , one can choose
sin((k + 1)E/2) kE
Rk = ε , Φk = . (135)
sin(E/2) 2

The limit E → 2πm, yields Rk → ε(−1)mk (k + 1), eiΦk → (−1)mk and one recovers the previous result. If E ∈ / 2π ZZ ,
eq.(135) shows that |Rk | < |ε/ sin(E/2)| ∀k, hence the norm of ATk will be bounded and the corresponding states will
be normalizable in terms of Dirac delta functions, e.g. they belong to the continuum spectrum.
In summary, the spectrum of Hamiltonian of the harmonic model, in the limit ε → 0, is given by the union of a
continuum part, Specc , and a discrete part, Specd , where

ϑ = 0, ε → 0+ or ϑ = π, ε → 0− : Specc → IR − 2π ZZ , Specd = 2π ZZ (136)

ϑ 6= 0 and π, ε → 0± : Specc → IR − 2π ZZ , Specd = ∅ (137)

Note that in the case ϑ 6= 0, π, the energies E = 2πn are missing in the continuum. For finite values of ε the spectrum
is given by (see Appendix A)

ϑ = 0, ε > 0 or ϑ = π, ε < 0 : Specc = ∪∞

n=−∞ 2π[n + δ, n + 1 − δ], Specd = 2π Z
Z, (138)

where sin(πδ) = 2ε/(1 + ε2 ) (see eq.(A16)), so if ε → 0 the gap between the intervals 2π[n + δ, n + 1 − δ] closes and
the continuum spectrum becomes IR − 2π ZZ , and the discrete spectrum 2π ZZ as in (136). When ϑ 6= 0, π, one also
recovers the spectrum in (137).
Let us consider another example with the same values of `n but exponential decaying reflection amplitudes

`n = en/2 , %n = ε e−λn (λ > 0), n = 0, . . . , ∞, (139)

that yields
sinh((λ + iE)(k + 1)/2)
e−iΦk Rk = ε e−(λ+iE)k/2 . (140)
sinh((λ + iE)/2)

If λ = 0, one recovers eq.(130). Let us see if the discrete spectrum of the previous model survives when λ > 0. Taking
E ∈ 2π ZZ yields

1 − e−λ(k+1)
E ∈ 2π ZZ → e−iΦk Rk = ε . (141)
1 − e−λ
Hence for any finite value of λ, the quantities Rk are bounded, and therefore the state belongs to the continuum, e.g.
the discrete spectrum disappears, but it can be recovered in the limit λ → 0, where Rk → ε(k + 1).

2. Polylogarithm model

This model is defined by the parameters (see eq.(52) with `1 = 1 )

e−λn
`n = n1/2 , %n = ε (σ > 0, λ > 0), n = 1, . . . , ∞. (142)
nσ
The positivity conditions on σ and λ ensures that %n → 0 as n → ∞, including the case where λ = 0. The norm of
the wave function χT is given by (see eq.(118))
∞  
1X 1
hχT |χT i = log 1 + hATn |ATn i, (143)
2 n=1 n

so its convergence depends on the asymptotic behavior of hATn |ATn i/n. The ansatz (142) yields
k
X e−λn
e−iΦk Rk = ε , (144)
n=1
nσ+iE
 20

that in limit k → ∞ becomes

e−iΦ∞ R∞ = lim e−iΦk Rk = εLiσ+iE (e−λ ), (145)
k→∞

where Lis (z) is the polylogarithm function [80]. In the limit λ → 0+ there is the expansion
e−iΦ∞ R∞ = ε λσ−1+iE Γ(1 − σ − iE) + ζ(σ + iE) + O(λ) .
 
(146)

If σ > 1, the term proportional to λσ−1+iE drops out and one is left with
σ > 1, λ → 0 → e−iΦ∞ R∞ = εζ(σ + iE), (147)
which implies that the norm of ATk converges towards a constant, for any value of ϑ, and therefore the series (143)
diverges logarithmically signaling a continuum of states normalizable in the Dirac delta sense. If σ < 1, the term
proportional to λσ−1+iE dominates hence
σ < 1, λ → 0 → e−iΦ∞ R∞ ' ελσ−1+iE Γ(1 − σ − iE), (148)
leading to
 1/2
Γ(1 − σ − iE)
R∞ ' |ελσ−1 Γ(1 − σ − iE)|, e−iΦ∞ ' sign (ε)λiE , (149)
Γ(1 − σ + iE)
If λ is kept fixed, the norm of ATk is bounded and one gets again a continuum spectrum. Let us see what happens in
the limit λ → 0. Here R∞ → +∞, and therefore the norm of ATk does not blow up, and actually converges to zero,
if and only if
 1/2
Γ(1 − σ + iE)
cos(Φ∞ − ϑ) = 1 =⇒ sign (ε)λ−iE = eiϑ , (150)
Γ(1 − σ − iE)
This result suggests that a regularized version of this model, involving the limit λ → 0+ , should contain eigenstates
satisfying (150). For E >> 1, the Stirling formula provides the asymptotic behavior
 
E 1 E E 1 3 ϑ
n(E) ' log + log + − σ − sign ε − , E  1, (151)
2π λ 2π 2πe 4 2 2π
that in the limit λ → 0, becomes a continuum. The parameter λ regularizes the model and in some respects is
analogue to the cutoff Λ in Connes’s xp model, with Λ ∝ 1/λ. However, in eq.(151) the term E/(2π) log E/(2πe),
appears with an opposite sign as compared with Connes model, and it does not have the meaning of missing spectral
lines, but rather of a finite energy correction. Notice also that if σ = 1/2, the spectrum has the symmetry E ↔ −E,
provided ϑ = 0 if ε > 0 or ϑ = π if ε < 0, like in the harmonic model analyzed above.
In summary, for σ > 1 the spectrum is a continuum related to the Riemann zeta function ζ(σ + iE). This result is
consistent with the studies carried by several authors in the past where the zeta function appears in connection to the
scattering states of some physical system [81]-[89]. However, for 0 < σ < 1 the connection with the zeta function is lost
and only the smoothed zeros appear as a finite size correction to the level counting formula, in analogy with Connes’s
version on the xp model, but with opposite sign. We are then forced to look for another ansatz for the reflection
coefficients %n , if the zeros are to be realized as discrete eigenvalues of the Hamiltonian HR in the semiclassical limit.
A hint is provided by the harmonic model, where the eigenvalues En = 2πn, arises from the blow up of Rk (En ) as
k → ∞. This property leads us to the next example.

3. Riemann model

The model is defined by the parameters

µ(n)
`n = n1/2 , %n = ε (σ > 0), n = 1, . . . , ∞, (152)
nσ
where µ(n) is the Moebius function that vanishes if n contains as divisor the square of a prime, and it is equal to 1
(−1) if n is the product of an even (odd) number of distinct primes, that is [90]

(−1)r If n = p1 . . . pr
µ(n) = . (153)
0 If ∃ p, p2 |n
 21

In this expression p1 , . . . , pr are different prime numbers. Integer numbers for which µ(n) 6= 0 are called square free.
The Moebius function has been used to construct an ideal gas of primons with fermionic statistics [91, 92]. In our
model, µ(n) appears in order to amplify the interference between waves so that the Riemann zeros are not swept out
in the semiclassical limit and become visible. In the limit k → ∞ one finds
∞ 
X µ(n) ε σ≥1
e−iΦ∞ R∞ = ε = , if 1 . (154)
n=1
nσ+iE ζ(σ + iE) 2 < σ < 1, RH : True

The case σ = 1 is equivalent to the Prime Number Theory, that was proved by Hadamard and de la Vallée-Poussin
by showing that ζ(1 + it) 6= 0, ∀t ∈ IR [2]. Repeating the analysis performed in the previous models, we find that
for σ ≥ 1, the spectrum is given by IR. Indeed, for all values of E, the norm of Ak approaches a constant value in
the limit k → ∞, and consequently the norm (143) diverges logarithmically. The same thing occurs for 1/2 < σ < 1,
under the RH according to which this region will be free of zeros.
We are then left with the model with σ = 1/2 to provide a spectral realization of the zeros. Let us first approach
this case in the limit σ → 1/2 when E = En is a zero, that is ζ(1/2 + iEn ) = 0. Expressing the zeta function on the
critical line as ζ(1/2 + iE) = Z(E)e−iθ(E) , where Z(E) and θ(E) are the Riemann Siegel functions [2] one finds

1+
ζ(σ + iEn ) = −i(σ − 1/2)Z 0 (En )e−iθ(En ) + O((σ − 1/2)2 ), σ→ , (155)
2
where Z 0 (E) = dZ(E)/dE. For the sake of the argument we have assumed that En is a simple zero of Z(E), a fact
that is unknown to hold for all zeros. If that is the case, the sign of the derivative of Z(E) at the zeros satisfy the
rule

1
sign Z 0 (En ) = (−1)n+ 2 (1+sign(n)) (156)

where the positive zeros are labelled as n = 1, 2, . . . , and the negative zeros as n = −1, −2, . . . . Equation (156) can
be derived from the continuity of Z(E) and the fact that Z(1/2) < 0. Plugging (155) and (156) into (154) gives
π 1
εei(θ(En )+ 2 ) εei(θ(En )−π(n+ 2 sign(n))) 1+
e−iΦ∞ R∞ ' 0
= , σ→ , (157)
(σ − 1/2)Z (En ) (σ − 1/2)|Z 0 (En )| 2
that leads to the choice
ε 1
R∞ ∼ > 0, e−iΦ∞ ∼ ei(θ(En )−π(n+ 2 sign(n))) . (158)
(σ − 1/2)|Z 0 (En )|

Hence the limit σ → 1/2, implies R∞ → +∞, whereby the amplitude AT∞ blows up unless the parameter ϑ satisfies
1
cos(Φ∞ − ϑ) = 1 → cos(θ(En ) − π(n + sign(n)) + ϑ) = 1, (159)
2
in which case AT∞ = 0, corresponding to a normalizable state. This is an important result that we shall derive more
rigorously a bit later, but now let us discuss its implications.

• The value of ϑ(En ) satisfying eq.(159) can be written as (with En > 0)

ϑ(En ) 1 θ(En ) 3
=n+ − = n − hN (En )i + , (160)
π 2 π 2
where hN (E)i is the average number of zeros in the range [0, E] (see eq.(43)). In the absence of fluctuations the
average would be exact, that is hN (En )i = n, whereby ϑ(En ) = 3π/2, ∀n > 0. Hence a single value of ϑ would
work for all the zeros. But the existence of fluctuations make things more interesting. To hear a given zero [87]
one has to fine tune ϑ according to eq.(160), which pass to depend on the phase of the zeta function, θ(En ).
One thus obtains that the Riemann zeros and the phase of the zeta function both acquire a physical meaning
in the model.
• Berry [13], and Badhuri et al. [88] have argued that a better approximation to the average zeros is obtained if

θ(En ) 1
cos θ(En ) ' 0 → 'n+ , n = 1, 2, . . . . (161)
π 2
 22

JHEn L Frequency

0.12
2

0.10
1
0.08

n 0.06
200 400 600 800 1000
0.04
-1
0.02

-2 JHEL
-2 -1 0 1 2

FIG. 3: Left: plot of ϑ(En ) ∈ (−π, π] given in eq.(160) for the 103 first positive Riemann zeros. Right: histogram of ϑ(En ).

This result can be visualized by plotting the real and imaginary parts of ζ(1/2 + it) in the complex plane for a
large interval of t. One obtains a collection of loops that cut the real axis at the Gram points where sin θ(t) = 0,
and that just before crossing the origin at t = En , the loop cuts the imaginary axis, where cos θ(t) = 0. Badhuri
et al. also show that θ(t) gives roughly the scattering phase shift of a non relativistic particle in an inverted
harmonic potential (V (x) ∝ −x2 ) with a hard wall at the origin. Now, replacing (161) into (160) gives that
in average ϑ(En ) ' 0 mod(2π). This result is confirmed in Fig. 3 which shows ϑ(En ) for the first 103 zeros
together with an histogram.

So far we have considered the value of e−iΦ∞ R∞ , using eq.(154). To show the existence of discrete states for
σ = 1/2, we need the finite sum giving e−iΦk Rk , that can be computed using the Perron’s formula [90, 93]
c+iT
xs
Z
X ∗ µ(n) ds 1
= lim , c > 0, c > 1 − σ, σ = Re z (162)
nz T →∞ c−iT 2πi ζ(s + z) s
1≤n≤x

P∗
where means that the last term in the sum must be multiplied by 1/2 when x is an integer. The integral (162)
can be done by residue calculus [93]
c+iT
xs
Z
ds X 1
lim F (s) = Ressj F (s), F (s) = , (163)
T →∞ c−iT 2πi ζ(s + z) s
Re sj <c

where the sum runs over the poles sj of F (s) located to the left Pof the line of integration, that is Re sj < c. For
x
z = 0, the sum (162) is basically the Mertens functions M (x) = n=1 µ(n) that plays an important role in Number
Theory [2]-[4]. Here we are interested in the values z = 21 + iE, with E a real number. Eq.(162) imposes the condition
c > 1 − Re z = 1/2. Hence the poles of F (s) that contribute to (163) have their real part smaller than c, that is
Re sj ≤ 1/2. The origin s = 0 is a simple pole if ζ(z) 6= 0 and a multiple pole if ζ(z) = 0. In the latter case we shall
assume that z is a simple zero so that the pole is double. The corresponding residues are given by

1/ζ(z) if ζ(z) 6= 0,
Ress=0 F (s) = (164)
log(x)/ζ 0 (z) if ζ(z) = 0, ζ 0 (z) = 0.

The remaining poles of F (s) come from the zeros of ζ(s + z), except the case s = 0, that is included in (164). The
trivial zeros of ζ contribute with the poles sn = −2n − z (n = 1, 2, . . . ), with residue

x−2n−z
Ress=−2n−z F (s) = , n = 1, 2, . . . , ∞. (165)
−(2n + z)ζ 0 (−2n)

The non trivial zeros of ζ, denoted as ρm , contribute with the poles sm = ρm − z (note that Re sm < 1/2), with
residue
xρm −z
Ress=ρm −z F (s) = , ρm 6= z (166)
(ρm − z)ζ 0 (ρm )
 23

Collecting results we find

∞
X ∗ µ(n) 1 X xρm −z X x−2n−z
= + + , if ζ(z) 6= 0, (167)
n z ζ(z) ρ
(ρm − z)ζ (ρm ) n=1 −(2n + z)ζ 0 (−2n)
0
1≤n≤x m
∞
X ∗ µ(n) log x X xρm −z X x−2n−z
= + + , if ζ(z) = 0, ζ 0 (z) 6= 0,
nz ζ 0 (z) (ρm − z)ζ 0 (ρm ) n=1 −(2n + z)ζ 0 (−2n)
1≤n≤x ρm 6=z

These equations are formally exact, so it would be interesting to prove them and find their range of validity. In what
follows, we shall derive their consequences. Let us recall that the LHS of (167) gives essentially e−iΦk Rk , where k is
to be identified with x. Neglecting for a while the summands in these expressions one finds

ε eiθ(E)
e−iΦk (E) Rk (E) ∼ , Z(E) 6= 0, (168)
Z(E)
i(θ(En )+ π
2)
ε log k e
e−iΦk (En ) Rk (En ) ∼ , Z(En ) = 0, Z 0 (En ) 6= 0.
Z 0 (En )

In the first case Rk (E) remains bounded in the limit k → ∞, that would yield a state in the continuum. In the second
case we choose

ε log k 1
Rk (En ) ∼ > 0, e−iΦk (En ) ∼ ei(θ(En )−π(n+ 2 sign(n))) , k→∞ (169)
|Z 0 (En )|

that can be compared with (158). Then imposing eq.(159), yields the asymptotic behavior of the norm of ATk (see
eq.(128))
0
hATk |ATk i → 2e−2Rk = k −2ε/|Z (En )|
, k → ∞, (170)

and a wave function χT whose norm given by (143)

∞ ∞  
1X1 T T X 1 2ε
hχT |χT i ∼ hAn |An i ∼ = ζ 1 + , (171)
2 n=1 n n=1
n1+2ε/|Z 0 (En )| |Z 0 (En )|

which is finite for any ε > 0 corresponding to a discrete eigenstate with energy En .
Fig. 4 shows |Ak |2 in the case where E corresponds to a state in the continuum, and for the first zero E1 = 14.13..,
corresponding to a discrete state. In the latter case we took ϑ = ϑ(E1 ), that guarantees that the norm converges to
zero except for some jumps. For other values like ϑ = π the norm increases with k. The same pattern is observed for
other zeros. The values of ϑ(En ) are mostly concentrated around 0 (see Fig. 3). Finally, we have computed the norm
of the vector χT , using eq.(143) observing that, for the zeros, it converges to a finite value with the choice (160), and
diverges in the remaining cases. These results are in agreement with eq.(171).
Let us now return to the summands in eq.(167) that we neglected in the previous computation. The last term
corresponding to the trivial zeros quickly converges to 0 as x = k → ∞. The term associated to non trivial zeros
on the critical line, ρm = 1/2 + iEm , Em ∈ IR, oscillates as xρm −z = ei(Em −En ) log x , and we expect that it gives
subleading contributions to the main term that goes with log x. Finally,0 a zero off the critical line, say ρm =
0 0
1/2 + Em + iEm , Em , Em > 0 ∈ IR, would give a contribution xρm −z = xEm ei(Em −E) log x , that dominates over the
remaining terms, for all values of E, leading to
0
Rk (En ) ∼ εC(ρm , E) k Em > 0, e−iΦk (En ) ∼ ei(Em −E) log k+α(ρm ,E)) , k→∞ (172)

where C(ρm , E) and α(ρm , E) do not depend on k. The expression of Rk (E) diverges when k → ∞, but unlike the
case of (169), the phase Φk (E) cannot be fixed to a value that cancels the divergent term e2Rk in the norm (128).
0
Hence |ATk |, would grow typically as exp(εCk Em ), so that the wave function χT will not be normalizable even in the
continuum sense. This occurs for any value of E, so we arrive at the paradoxical conclusion that a zero off the critical
line implies that the Hamiltonian HR does not have eigenstates !! That’s not certainly the case because HR is a well
defined self-adjoint operator, so we must conclude that off critical zeros do not exist. This result seems to provide a
proof of the Riemann hypothesis, but one must be very cautious since it relies on several unproven assumptions.
Comments:
 24

ÈAk 2 ÈAk 2
6
0.14

5 0.12

4 0.10

0.08
3
0.06
2
0.04
1
0.02

k k
500 1000 1500 2000 500 1000 1500 2000

FIG. 4: Left: Plot of |Ak |2 given in eq.(128) for E = 24, with k = 10, . . . , 2000, ε = 0.5, ϑ = π. Right: E1 = 14.13..,
k = 10, . . . , 2000, ε = 0.25, ϑ = ϑ(En )

• In the spectral realization proposed above there cannot exist zeros outside the critical line in the form of
resonances. As explained above, their presence leads to the non existence of eigenvectors of the Hamiltonian
HR .
• von Neumann and Wigner showed that in ordinary Quantum Mechanics it is possible to have a bound state
immersed in the continuum [68, 94]. They used a potential that decays as 1/r with oscillations that trap the
particle thanks to interference effects. There is a general class of models with this property [95]-[97], and they
all require a fine tuning of couplings. That this phenomena may happen for the Riemann zeros was suggested
in [32] using a xp model with non local interactions.
• The delta function potential needed to reproduce the zeros depends on the Moebius function µ(n) that exhibits
an almost random behavior. This result is reminiscent of the fractal structure of the quantum mechanical
potentials built to reproduce the lowest values of prime numbers and the Riemann zeros. The latter potentials
were built from the smooth ones obtained by Mussardo, for the primes [98], and Wu and Sprung, for the zeros
[99], and have a fractal dimension around 2 and 1.5 respectively [100]-[102].

4. Dirichlet model

We shall briefly describe how the previous results can be generalized to Dirichlet Lχ −functions. These are analytic
functions constructed from a Dirichlet character χ as [4]
∞
X χ(n) Y 1
Lχ (s) = s
= −s
< s > 1, (173)
n=1
n p
1 − χ(p)p

where the product runs over the prime numbers (Euler formula). A character χ of modulus q, is an arithmetic function
satisfying χ(mn) = χ(m)χ(n), χ(1) = 1, χ(m + qn) = χ(m), χ(n) = 0 if q divides n [4]. One has that |χ(n)| = 1 or
0, and if χ(n) = ±1, 0, the character is called real. The number of characters with modulus q is given by the Euler
totient function φ(q), that counts the number of coprime divisors of q. For q = 1, the character is χ1 (n) = 1, ∀n, that
corresponds to the Riemann zeta function ζ(s). Primitive characters are those that cannot be written as products of
characters of smaller modulus q, and can be classified into even or odd if χ(−1) = 1, −1 respectively.
The L-functions associated to primitive characters satisfy the functional relation
 −(s+aχ )/2  
π s + aχ Gχ
Λχ (s) ≡ Γ Lχ (s) = eiχ Λχ̄ (1 − s), eiχ = i−aχ √ , (174)
q 2 q

where aχ = 0 (1) for an even (odd) character, χ̄ is the character conjugate to χ, e.g. χ̄(n) = χ∗ (n), and Gχ is the
Gauss sum
q
X √
Gχ = χ(n)e2πin/q , |Gχ | = q. (175)
n=1
 25

On the critical line s = 1/2 + it, eq.(174) reads

 −it/2      it/2    
π 1 + 2aχ it 1 iχ π 1 + 2aχ it 1
Γ + Lχ + it = e Γ − Lχ̄ − it . (176)
q 4 2 2 q 4 2 2

For t real it is useful to parameterize the L functions as

   
1 1
t ∈ IR → L∗χ̄ − it = Lχ + it = Zχ (t) e−iθχ (t) , (177)
2 2

where Zχ (t) is real and eiθχ (t) is a phase that can be found from eq.(176)
 
 −it Γ 1+2aχ + it
π 4 2
e2iθχ (t) = e−iχ   , e2i(θχ (t)+θχ (−t)) = e−iχ . (178)
q Γ
1+2aχ
− it
4 2

θχ (t) coincides with θ(t) when χ is the identity character. For real characters, the zeros of Lχ , appear symmetrically,
Lχ (1/2 + itn ) = Lχ (1/2 − itn ) = 0, that is reflected in the properties, Zχ (−t) = Z(t) and θχ (−t) = −θχ (t). However
for complex characters this symmetry is broken.
The Dirac model associated to Lχ will be defined by the parameters

µ(n)χ(n)
`n = n1/2 , %n = ε (σ > 0), n = 1, . . . , ∞. (179)
nσ
Note that one can deal with complex characters thanks to the existence of two types of mass terms, ψ̄ψ and ψ̄γ 5 ψ.
The characters χ(n) acquire a physical meaning related to the reflection coefficient of the nth mirror. This provides
a unified framework to deal with the whole family of Dirichlet L-functions.
We can repeat the analysis done for the zeta function to find the discrete eigenenergies in the spectrum. For
example, the asymptotic limit of e−iΦk Rk is given by
∞
X µ(n)χ(n) ε
e−iΦ∞ R∞ = ε σ+iE
= , (180)
n=1
n Lχ (σ + iE)

which shows that the zeros of Lχ (1/2 + iE) appear as poles of R∞ , that lead eventually to discrete states, provided
ϑ is fine tuned appropriately. Assuming that the zeros of Lχ (1/2 + iE) are simple, the value of ϑ(En ) for which En
is a discrete state is given by

ϑ(En ) 1 θ(En )
= n + (1 + bχ + sign(n)) − , bχ = sign Zχ (1/2). (181)
π 2 π
For the zeta function, bχ = −1, eq.(160) is recovered from (181). We expect the zeros of the Dirichlet L functions,
associated to primitive characters, to form the discrete spectrum of the corresponding Hamiltonians. That would
amount to a proof of the Generalized Riemann hypothesis.

V. CPT SYMMETRIES AND AZ CLASSES

The statistical properties of the Riemann zeros, conjectured by Montgomery and confirmed numerically by Odlyzko,
have been one of the main motivations to search for a spectral origin of these numbers. The conjecture is that the
zeros satisfy, locally, the GUE law, which implies that the Riemann dynamics breaks the time reversal symmetry.
Deviations from the GUE law were later on identified by Berry and collaborators, as a trace of the semiclassical origin
of the zeros and a breakdown of universality. It is thus of great interest to study the discrete CPT symmetries and
RMT universality classes of the Hamiltonians discussed in previous sections.
The action of the time reversal symmetry (T ), charge conjugation or particle-hole symmetry (C), and parity or
chirality (P) on a vector ψ are defined as [103, 104]

ψT = T ψ∗ , ψC = C ψ∗ , ψ P = P ψ, (182)
 26

where T, C, P are unitary matrices, i.e. T T † = CC † = P P † = 1 and ψ ∗ is the complex conjugate of ψ. Here ψ is
the column vector formed by the coefficients of a pure state of a Hilbert space in an orthonormal basis. T and C are
antiunitary transformations and P is unitary. A Hamiltonian H has T , C or P symmetry if it satisfies the conditions

H T = T H ∗ T † = H, H C = C H t C † = −H, H P = P H P † = −H. (183)

Since H is hermitean, H ∗ = H t , then the T symmetry becomes T H t T † = H. Notice that P is the product of T and
C, and one can choose P = CT † . There is a basis where T and C are real, and symmetric or antisymmetric matrices,
i.e. T t = ±T and C t = ±C, in which case unitarity implies T 2 = ±1 and C 2 = ±1. If a symmetry is broken, say
T , one writes T 2 = 0. Counting all the possibilities one arrives to ten symmetry classes, as found by Altland and
Zirnbauer (AZ), that include the classical Wigner-Dyson gaussian ensembles: GOE, GUE, and GSE as well as their
chiral versions chGOE, chGUE, chGSE [105]. Among the 10 AZ classes there are 4 that break time reversal symmetry,
which are the candidates to describe the zeros of ζ and other L-functions (see Table 1).

AZ T2 C2 P 2 Top
A 0 0 0 0
AIII 0 0 1 ZZ
D 0 1 0 ZZ 2
C 0 -1 0 0
Table 1.- The AZ classes where the time reversal symmetry is broken. Column ”Top” denotes the topological
invariants of the class in one spatial dimension.
Let us review briefly the main properties of the classes of Table 1 and their relations with the problem at hand.
Class A characterizes Hamiltonians of the form H = A, where A is an hermitean matrix, with no further conditions
placed on it. The statistical properties of random matrices of this form are described by GUE. As shown in Table 1,
all the CPT symmetries are broken and there is no topological invariant in 1D.
Class D characterizes Hamiltonians of the form H = A, where A is imaginary and antisymmetric. The eigenvalues
of A appear in pairs {En , −En }, and if the dimension of the matrix A is odd there is a zero eigenvalue. The Berry-
t
Keating
√ Hamiltonian
2 −1
√ H = (xp̂+ p̂x)/2 belongs to this class, since here T = C = 1 and H = −H [46]. The Hamiltonian
H = x(p̂ + `p p̂ ) x also belongs to class D provided the parameter ϑ, that characterizes its self-adjoint extensions,
is 0 or π. The latter choices ensure that the eigenvalues of H come in pairs {En , −En }, and that if ϑ = 0, then E0 = 0
is eigenvalue [43, 44]. We can thus identify ϑ/π = 0, 1 as the ZZ 2 = {0, 1} topological invariant of class D in 1D.
Class C characterizes Hamiltonians of the form H = A + ~σ · S, ~ where At = A∗ = −A, S ~t = S~∗ = S ~ and ~σ are
the Pauli matrices that act in an additional 2 dimensional Hilbert space, that can be seen as a spin 1/2 [46]. Here
C = iσ y , so T 2 = −1. The eigenvalues of H come in pairs {En , −En }, as in class D. Srednicki proposed recently
that class C is associated to the zeros of the Dirichlet L-functions whose characters are real and even, that includes
the zeta function [46]. He was led to this proposal by a conjecture due to Katz and Sarnak [106] according to which
these L-functions form a family related by a sort of symplectic symmetry, and by the fact that the spacings of their
zeros agree asymptotically with the GUE distribution.
Class AIII, is a chiral version of GUE (chGUE) and characterizes Hamiltonians with the block structure
! !
0 A 1 0
H= , P = , P HP † = −H, (184)
A† 0 0 −1

where A is a complex matrix and P the chiral operator. The eigenvalues of H come in pairs {En , −En }. If A is a
matrix of dimension N+ × N− , then the number of zeros eigenvalues is |N+ − N− | that explains the ZZ topological
invariant of this class. Class chGUE, together with its relatives chGOE and chGSE, describe massless Dirac fermions
and have been applied to study the QCD Dirac operator, partition functions, etc [107]-[110].
Let us next study the symmetry classes of the Rindler Hamiltonians constructed in previous sections. First of all,
the massless Hamiltonian
√ √
HR = ρ p̂ρ ρ σ z , (185)

admits different representations of the CP T symmetries [104]. Indeed, T and P can be realized as σ x or iσ y , and C
as σ z or 1. Adding a mass term to (185) selects a particular realization, e.g.
√ √
HR = ρ p̂ρ ρσ z + mρσ x → T = σ x , C = σ z , P = iσ y . (186)
 27

This Hamiltonian acts on the wave functions satisfying the BC (31). One can verify that this BC preserves T , for
all values of ϑ, however the symmetry C is preserved only if ϑ = 0, π. Hence in these two cases, the Hamiltonian
2 2
(186) √belongs to class√ BDI (chGOE) since T = C = 1. We mentioned in section II.E that the Hamiltonian
2 −1
H = x(p̂ + `p p̂ ) x( see eq. (45)) has the same spectrum as (186) with the identifications (46). We show above
that the former Hamiltonian belongs to class D, while we have found now that (186) belongs to class chGOE. There
is no contradiction between these results. In both cases the C symmetry explains the pairing of energies {En , −En },
while the T symmetry appears from the doubling of degrees of freedom in the Dirac model.
Finally, let us show that the Dirac model associated to the Riemann zeros belongs to class AIII, with chiral operator
P = iσ y . First of all notice that the matching conditions (89) are preserved by the action of P , because in this model
rn0 = rn00 = 0, and then

χ(`− + P − P +
n ) = Ln χ(`n ) → χ (`n ) = Ln χ (`n ), Ln = P Ln P † . (187)

On the other hand, if χ is the eigenfunction with energy En it satisfies the BC (91)

− ieiϑ(En ) χ− (`1 ) = χ+ (`1 ) → −ie−iϑ(En ) χP P

− (`1 ) = χ+ (`1 ), (188)

where χP∓ = ±χ± . The function ϑ(En ), given in eq.(160) for n > 0, satisfies that −ϑ(En ) = ϑ(−En ) which, together
with the equation P HR P † = −HR , implies that if χ is an eigenstate with energy En , then χP is an eigenstate with
energy −En . Hence the pairing of energies is explained by the chiral symmetry and not by the charge conjugation
symmetry as the Hamiltonian (186). One can perform a change of basis that brings the chiral operator P and
√ √
the Hamiltonian (185) into the form (184), with A = i ρ p̂ρ ρ that is a real and antisymmetric matrix in some
orthonormal basis. The Hamiltonian we are dealing with is therefore a subclass within the class chGUE.

VI. CONCLUSIONS

We have proposed in this paper that a combination of Quantum Mechanics and Relativity Theory is the key to the
spectral realization of the Riemann zeros. The old message of Pólya and Hilbert was that Quantum Mechanics should
play a central role in this realization, an idea that is behind the successful applications of RMT, and Quantum Chaos,
to Number Theory. What is new is that Einstein’s theory of Relativity must also be present. The reason is that the
properties of accelerated objects, in Minkowski space-time, can be used to encode and process arithmetic information.
In some sense, spacetime becomes an analogue computer, or simulator, that allows an accelerated observer to multiply
numbers, and distinguish primes from composite, by measuring the proper times of events involving massless particles.
In this way the prime numbers acquire a classical relativistic meaning associated to primitive trajectories of particles,
along the lines suggested by the periodic orbit theory in Quantum Chaos.
The combination of Quantum Mechanics and Relativity is of course a Relativistic Quantum Field Theory, that in
our case is the Dirac theory of fermions in Rindler space-time, the latter being the geometry associated to moving
observers. The time evolution in this space is generated by the Rindler Hamiltonian that coincides with the Berry-
Keating Hamiltonian ±(xp̂ + p̂x)/2, each sign corresponding to the chirality of the fermion. This operator is also the
generator of dilations of the Rindler radial coordinate x, but in the presence of delta function potentials, associated
to the prime numbers, scale invariance is broken. This has the dramatic effect that the spectrum is a continuum,
where the Riemann zeros are missing, in analogy with the result found by Connes in the adelic approach to the RH.
However, the moving observer can fine tune the phase factor of the reflection of the fermion at the boundary, in such
a way that a bound state appears with an energy given by a Riemann zero. The phase factor at the boundary is
given essentially by the phase of the zeta function at the corresponding zero. This result is obtained in a limit where
the perturbation of the massless Dirac action by the delta potential is infinitesimally small, so that the bound state
is immersed in a continuum of states. This weak coupling limit is reminiscent to the semiclassical limit that leads to
the Gutzwiller formula for the fluctuations of the energy levels in chaotic systems. The previous scenario leads to a
proof of the Riemann hypothesis but more work is required to fully support this claim.
The present work suggests the possibility of an experimental observation of the Riemann zeros as energy levels.
The system proposed here does not look realistic at the moment, but there are theoretical proposals to realize the
Rindler geometry using cold atoms and optical lattices [111, 112]. Another possible route is to use the Quantum Hall
effect where xp arises as an effective low-energy Hamiltonian [38, 54]. On a more theoretical perspective, this work
points towards a relation between Riemann zeros and black holes, whose near horizon geometry is in fact Rindler
[113]. The role played by the privileged observer in our construction reminds the brick wall model introduced by ’t
Hooft to regularize the horizon of a black hole [114]. Throughout this paper the prime numbers have been treated as
classical objects. However, they turn into quantum objects in the Prime state formed by the superposition of primes
 28

in the computational basis of a quantum computer [115, 116]. This poses the question whether the Prime state could
be created with the same tools as the Riemann zeros become energy levels.

Acknowledgements.- I am very grateful for suggestions, and conversations to José Ignacio Latorre, Belén Paredes,
Paul Townsend, Michael Berry, Jon Keating, Javier Rodrı́guez-Laguna, Manuel Asorey, Jose Luis Fernández Barbón,
Giuseppe Mussardo, André LeClair, Mark Srednicki, Javier Molina, Luis Joaquı́n Boya and Miguel Angel Martı́n-
Delgado. This work has been financed by the Ministerio de Ciencia e Innovación, Spain (grant FIS2012- 33642),
Comunidad de Madrid (grant QUITEMAD) and the Severo Ochoa Program.

Appendix A: The harmonic model

We first discuss some general properties of the transfer matrix Tk that will be used later on to derive the exact
spectrum of the harmonic model. Let us consider the transfer matrix (115), with rk0 = 0, and the condition |rk | < 1.
It is easy to see that Tk can be written as
!
1 rk2 + 1 2rk `−2iE z x z
Tk = 2 2iE 2
k = e−iϕk σ egk σ eiϕk σ , (A1)
1 − rk 2rk `k rk + 1

where e±gk are the eigenvalues of Tk (positive from the condition |rk | < 1)
1 + rk 1 − rk
egk = , e−gk = , (A2)
1 − rk 1 + rk
sign gk = sign rk , |rk | < 1,

and

ϕk = E log `k , k = 1, . . . , ∞. (A3)

Replacing (A1) into (112) gives the recursion relation

z x z
|Ak−1 i = e−iϕk σ egk σ eiϕk σ |Ak i, k ≥ 2. (A4)

We now define the vector

z
|Âk i = eiϕk σ |Ak i, k ≥ 1, (A5)

that satisfies
z x
|Âk−1 i = e−i∆k σ egk σ |Âk i, k ≥ 2, (A6)

where
`k
∆k = ϕk − ϕk−1 = E log , k ≥ 2. (A7)
`k−1
Notice that

|Â1 i = |A1 (ϑ)i. (A8)

Eq.(A6), gives an alternative way to solve the eigenvalue problem that admits an interestingz physical interpretation
as the evolution of a kicked rotator with spin 1/2 [58]. In this interpretation the term e−i∆k σ represents the rotation
of the spin around the z-axis, with an energy E during a time elapse log(`k /`k−1 ), after which the spin is kicked
with an imaginary magnetic field with strength gk along the x-direction. Kicked rotators of this sort are currently
employed to analyze quantum chaos in simple situations [58].
Let us next consider the harmonic model defined by eq.(129), that according to eqs.(A2, A7) correspond to constant
values of gk = g and ∆k = ∆
1+ε E
`n = en/2 , %n = rn = ε −→ eg = , ∆= . (A9)
1−ε 2
 29

g
3

2
4Π∆
1

E
2 4 6 8 10 12

-1

-2

-3

FIG. 5: The region in shadow represents the values of E and g where the matrix S is elliptic, that corresponds to states in the
continuum (see eq.(A15)). In the white region the matrix S is hyperbolic, and contains the discrete spectrum whose location
depend on g and ϑ by eq.(A14). For ϑ = 0, π, the eigenstates are at En = 2πn for all values of g (see eq.(A13)). The horizontal
line corresponds to a value of the coupling constant g. Its intersection with the shadow regions give the continuum spectrum
that form the bands (A16) separated by a gap 4πδ.

In this case the recursion relation (A6) can be iterated yielding

|Âk+1 i = S k |A1 (ϑ)i, k ≥ 0, (A10)

where the matrix S is given by

!
−gσ x iEσ z /2 cosh g eiE/2 − sinh g e−iE/2
S=e e = . (A11)
− sinh g eiE/2 cosh g e−iE/2

Making the replacement E → E +2π, the matrix S changes by an overall sign, which in turn implies that the spectrum
has a 2π periodicity. Let us consider the case where E = 2πn, that yields
!
n −gσ x nk −gkσ x 1
S = (−1) e −→ |Âk+1 i = (−1) e , (A12)
eiϑ

which is convergent only in the following two cases

!
nk −gk 1
ϑ = 0, g > 0 −→ |Âk+1 i = (−1) e , (A13)
1
!
nk −|g|k 1
ϑ = π, g < 0 −→ |Âk+1 i = (−1) e .
−1

These are normalizable eigenstates for all values of the coupling constant g, even in the limit g ' 2ε → 0. Normalizable
states for ϑ 6= 0, π are also possible. The relation between the energy, E, g and ϑ is given by
sin ϑ
tan(E/2) = , 1 + cos ϑ coth g > 0, (A14)
cos ϑ + coth g

that in the limit g → 0 gives the discrete eigenvalues E → 2πn. Plugging ϑ = 0, π in (A14) we recover the cases
(A13). The continuum spectrum of the model can be derived from the condition that S is an elliptic matrix (note
that det S = 1)

|Tr S = 2 cosh g cos E/2| < 2 ←→ | sin E/2| > tanh g, (A15)
 30

and consists of energy bands separated by a gap 4πδ that are shown in Fig. 5
2ε
Specc = ∪∞
n=−∞ 2π[n + δ, n + 1 − δ], sin(πδ) = tanh g = . (A16)
1 + ε2
In the limit g → 0, the gap closes, and the discrete energy states En = 2πn become immersed in the continuum IR.

[1] B. Riemann, ”On the number of primes less than a given magnitude”, translated by H. M. Edwards from ”Ueber die
Anzahl der Primzahlen unter einer gegebenen Grösse” (1859) in [2].
[2] H.M. Edwards, Riemann’s Zeta Function, Academic Press, New York, 1974.
[3] E.C. Titchmarsh, The Theory of the Riemann Zeta Function, Oxford Univ. Press, Oxford, 1986.
[4] H. Davenport, Multiplicative Number Theory, Grad. Texts in Math., vol. 74, Springer-Verlag, New York, 2000.
[5] E. Bombieri, Problems of the Millennium: The Riemann Hypothesis, Clay Mathematics Institute, (2000).
[6] P. Sarnak, Problems of the Millennium: The Riemann Hypothesis, Clay Mathematics Institute, (2004).
[7] B. Conrey, ”The Riemann Hypothesis, Notices of AMS, March, 341 (2003).
[8] H.L. Montgomery, ”Distribution of the zeros of the Riemann zeta function”, in Proceedings Int. Cong. Math. Vancouver
1974, Vol. I, Canad. Math. Congress, Montreal, 379 (1975).
[9] A.M. Odlyzko, ”Supercomputers and the Riemann zeta function”, in Supercomputing 89: Su- percomputing Structures
& Computations, Proc. 4-th Intern. Conf. on Supercomputing, In- ternational Supercomputing Institute, St. Petersburg,
FL, 348 (1989).
[10] A. M. Odlyzko, The 1020-th zero of the Riemann Zeta Function and 70 Million of its Neighbors,
www.dtc.umn.edu/?odlyzko/unpublished.
[11] Bohigas and M. J. Gianonni, in Mathematical and Computational Methods in Nuclear Physics, edited H. Araki, J. Ehlers,
K. Hepp, R. Kippenhahn, H. Weidenmuller and J. Zittart (Springer Verlag, Berlin, 1984).
[12] M. L. Mehta, Random Matrices, Elsevier Ltd. The Netherlands, 2004.
[13] M. V. Berry, “Riemann’s zeta function: a model for quantum chaos?”, in Quantum Chaos and Statistical Nuclear Physics,
edited by T. H. Seligman and H. Nishioka, Springer Lecture Notes in Physics Vol. 263 , p. 1, Springer, New York (1986).
[14] E. Bombieri, “Prime territory: exploring the infinite landscape at the base of the number system”, The Sciences 32, 30
(1992).
[15] M. C. Gutzwiller, Chaos in Classical and Quantum Mechanics, SpringerVerlag, New York, 1990.
[16] J. P. Keating, ”Periodic orbits, spectral statistics and the Riemann zeros”, in Supersymmetry and Trace Formulae: Chaos
and Disorder, eds. I.V. Lerner, J.P. Keating & D.E Khmelnitskii (Plenum Press), 1-15 (1999).
[17] P. Leboeuf, A. G. Monastra, O. Bohigas, ”The Riemannium”, Reg. Chaot. Dyn. 6, 205 (2001); arXiv:nlin/0101014.
[18] E. Bogomolny, ”Quantum and Arithmetical Chaos”, in Frontiers in Number Theory, Physics and Geometry, Les Houches,
2003; arXiv:nlin/0312061
[19] A. Selberg, ”Harmonic analysis and discontinuous groups in weakly symmetric Riemannian spaces with applications to
Dirichlet series”, Journal of the Indian Mathematical Society 20, 47 (1956).
[20] D. A. Hejhal, ”The Selberg trace formula for P SL(2, R)”., vol. 548 of Lecture Notes in Mathematics, Springer- Verlag
(1976).
[21] D. Schumayer, D. A. W. Hutchinson, ”Physics of the Riemann Hypothesis”, Rev. Mod. Phys. 83, 307 (2011);
arXiv:1101.3116.
[22] M. Watkins, Number Theory and Physics arxiv: http://empslocal.ex.ac.uk/people/staff/mrwatkin//zeta/physics.htm
[23] M. du Satoy, The Music of the Primes, Harper Collins, NewYork (2003).
[24] K. Sabbagh, Dr. Riemann’s zeros, Atlantic books, London (2003).
[25] J. Derbyshire, Prime obsession, A plume book, New York (2003).
[26] D. Rockmore, Stalking the Riemann hypothesis, Jonathan Cape, London (2005).
[27] M.V. Berry, J.P. Keating, “H = xp and the Riemann zeros”, in Supersymmetry and Trace Formulae: Chaos and Disorder,
ed. J.P. Keating, D.E. Khmelnitskii and I. V. Lerner, Kluwer, 1999.
[28] M. V. Berry, J. P. Keating, “The Riemann zeros and eigenvalue asymptotics”, SIAM Review 41, 236, 1999.
[29] A. Connes, “Trace formula in noncommutative geometry and the zeros of the Riemann zeta function”, Selecta Mathe-
matica New Series 5 29, (1999); math.NT/9811068.
[30] B. Aneva, ”Symmetry of the Riemann operator”, Phys. Lett. B 450, 388 (1999).
[31] G. Sierra, ”The Riemann zeros and the cyclic renormalization group”, J. Stat. Mech.: Theor. Exp. (2005) P12006;
math.NT/0510572.
[32] G. Sierra, ”H = xp with interaction and the Riemann zeros”, Nucl. Phys. B 776, 327 (2007); math-ph/0702034.
[33] J. Twamley and G. J. Milburn, ”The quantum Mellin transform”, New J. Phys. 8, 328 (2006); quant-ph/0702107.
[34] G. Sierra, ”Quantum reconstruction of the Riemann zeta function”, J. Phys. A: Math. Theor. 40 (2007) 1; math-
ph/0711.1063.
[35] G. Sierra, ”A quantum mechanical model of the Riemann zeros”, New J. Phys. 10, 033016 (2008); arXiv:0712.0705.
[36] J. C. Lagarias, ”The Schroëdinger operator with Morse potential on the right half line”, Communications in Number
Theory and Physics 3 (2009), 323; arXiv:0712.3238.
 31

[37] J-F. Burnol, ”On some bound and scattering states associated with the cosine kernel”, arXiv:0801.0530.
[38] G. Sierra and P.K. Townsend, ”The Landau model and the Riemann zeros”, Phys. Rev. Lett. 101, 110201 (2008);
arXiv:0805.4079.
[39] S. Endres and F. Steiner, ”The Berry-Keating operator on L2 (R> , dx) and on compact quantum graphs with general
self-adjoint realizations”, J. Phys. A: Math. Theor. 43, 095204 (2010); arXiv:0912.3183.
[40] G. Regniers and J. Van der Jeugt, ”The Hamiltonian H = xp and classification of osp(1|2) representations”, Workshop
Lie Theory and Its Applications in Physics VIII (Varna, 2009); arXiv:1001.1285.
[41] G. Sierra, ”A Physics pathway to the Riemann hypothesis”, Eds M. Asorey, et al. J. Vicente Garcı́a Esteve, M. F. Rañada,
J. Sesma, Prensas Univ. Zaragoza, Spain (2009); arXiv:1012.4264.
[42] E. de Rafael, “Large-Nc QCD, Harmonic Sums and the Riemann Zeros”, Nucl. Phys. Proc. Suppl. 207, 290 (2010);
arXiv:1010.4657.
[43] G. Sierra and J. Rodrı́guez-Laguna, ”The H = xp model revisited and the Riemann zeros”. Phys. Rev. Lett. 106, 200201
(2011); arXiv:1102.5356
[44] G. Sierra, ”General covariant xp models and the Riemann zeros”, J. Phys. A: Math. Theor. 45 055209 (2012);
arXiv:1110.3203.
[45] M. Srednicki, ”The Berry-Keating Hamiltonian and the Local Riemann Hypothesis”, J. Phys. A: Math. Theor. 44 305202
(2011); arXiv:1104.1850.
[46] M. Srednicki, ”Nonclasssical Degrees of Freedom in the Riemann Hamiltonian”, Phys. Rev. Lett. 107, 100201 (2011);
arXiv:1105.2342.
[47] M. V. Berry and J. P. Keating, ”A compact hamiltonian with the same asymptotic mean spectral density as the Riemann
zeros”, J. Phys. A: Math. Theor. 44, 285203 (2011).
[48] K. S. Gupta, E. Harikumar and A. R. de Queiroz, ”A Dirac type xp-Model and the Riemann Zeros”, Eur. Phys. Lett.
102 10006 (2013); arXiv:1205.6755.
[49] J. Molina-Vilaplana and G. Sierra, ”An xp model on AdS2 spacetime”, Nucl. Phys. B 877, 107 (2013); arXiv:1212.2436.
[50] R. V. Ramos and F. V. Mendes, “Riemannian Quantum Circuit”, Phys. Lett. A 378, 1346 (2013) arXiv:1305.3759.
[51] J. C. Andrade, “Hilbert-Pólya conjecture, zeta-functions and bosonic quantum field theories”, Int. J. Mod.Phys. A28,
1350072 (2013); arXiv:1305.3342.
[52] J. Kuipers, Q. Hummel and K. Richter, ”Quantum graphs whose spectra mimic the zeros of the Riemann zeta function”,
Phys. Rev. Lett 112, 070406 (2014); arXiv:1307.6055.
[53] M. C. Nucci, ”Spectral realization of the Riemann zeros by quantizing H = w(x)(p + `2p /p): the Lie-Noether symmetry
approach”, Journal of Physics: Conference Series 482, 012032 (2014).
[54] M. Stone, ”An analogue of Hawking radiation in the quantum Hall effect”, Class. Quantum Grav. 30, 085003 (2013);
arXiv:1209.2317.
[55] Ya. G. Sinai, ”Dynamical systems with elastic reflections”, Russ. Math. Surv. 25, 137, (1970).
[56] M. V. Berry, ”Quantizing a Clasically Ergodic system: Sinai’s billiard and the KKR method”, Annals of Physics, 131,
163 (1981).
[57] O. Bohigas, M. J. Giannoni and C. Schmit, ”Characterization of Chaotic Quantum Spectra and Universality of Level
Fluctuation Laws”, Phys. Rev. Lett. 52, 1 (1984).
[58] H.J. Stöckmann, Quantum Chaos: an introduction, Cambridge University Press, Cambridge, 1999.
[59] C. Holzhey, F. Larsen and F. Wilczek, ”Geometric and Renormalized Entropy in Conformal Field Theory”, Nucl. Phys.
B 424 , 443 (1994); arXiv:hep-th/9403108.
[60] W. Rindler, ”Kruskal space and the uniformly accelerated frame”, Am. J. Phys. 34, 1174 (1966).
[61] C. W. Misner, K. S. Thorne, and J. A. Wheeler, Gravitation, Ed. W. H. Freeman, (1973).
[62] L. C. B. Crispino, A. Higuchi, G. E. A. Matsas, ”The Unruh effect and its applications”, Rev. Mod. Phys. 80 787 (2008);
arXiv:0710.5373.
[63] W. G. Unruh, ”Notes on black-hole evaporation”, Phys. Rev. D 14, 870 (1976).
[64] S. W. Hawking, ”Black-hole explosions”, Nature (London) 248, 30 (1974).
[65] S. W. Hawking, ”Particle creation by black-holes”, Comm. Math. Phys. 43, 199 (1975).
[66] P. Di Francesco, P. Mathieu and D. Sénéchal, Conformal Field Theory, Springer-Verlag, London, 1996.
[67] J. von Neumann. Math. Ann., 102, 49 (1929).
[68] A. Galindo and P. Pascual, ”Quantum Mechanics I”, Springer-Verlag, Berlin, 1991.
[69] M. Abramowitz and I. A. Stegun, Handbook of mathematical functions, Dover, New York 1974. (1972).
[70] J-F. Burnol, ”Spacetime causality in the study of the Hankel transform”, Ann. Henri Poincar 7, 1013 (2006);
arXiv:math/0509619.
[71] B. Sutherland and D. C. Mattis, ”Ambiguities with the relativistic delta-function potential”, Phys. Rev. A 24, 1194
(1981).
[72] B.H.J. McKellar and G.J.. Stephenson Jr., ”Relativistic quarks in one- dimensional periodic structures”, Phys. Rev. C
35, 2262 (1987).
[73] M. G. Calkin, D. Kiang and Y. Nogami, ”Nonlocal separable potential in the one-dimensional Dirac equation”, Phys.
Rev. B 38, 1076 (1988).
[74] C. L. Roy, ”Boundary conditions across a d-function potential in the one-dimensional Dirac equation”, Phys. Rev. A 47,
3417 (1993).
[75] F. Domı́nguez-Adame, ”The Dirac equation with surface delta interactions”, J. Phys. A 23, 1993 (1990).
[76] F. A. B. Coutinho, Y. Nogami, and J. F. Perez, ”Generalized point interactions in one-dimensional quantum mechanics”,
 32

J. Phys. A 30, 3937 (1997).

[77] M. Asorey, A. Ibort, G. Marmo, ”Global Theory of Quantum Boundary Conditions and Topology Change”, Int. J. Mod.
Phys. A20, 1001 (2005);hep-th/0403048.
[78] H. F. Trotter, ”On the product of semi-groups of operators”, Proc. Amer. Math. Soc. 10, 545 (1959).
[79] M. Suzuki, ”Generalized Trotter’s Formula and Systematic Approximants of Exponential Operators and Inner Derivations
with Applications to Many-Body Problems”, Commun. Math. Phys. 51, 183 (1976).
[80] http://en.wikipedia.org/wiki/Polylogarithm
[81] B. S. Pavlov and L. D. Fadeev, ”Scattering theory and authomorphic functions”, Sov. Math. 3, 522 (1975). Plenum
Publishing Corp. translation, N.Y.
[82] P. D. Lax and R.S. Phillips, Scattering Theory for Automorphic Functions, Princeton University Press (1976).
[83] M. C. Gutzwiller, ”Classical Quantization of a Hamiltonian with Ergodic Behavior”, Phys. Rev. Lett. 45, 150 (1980).
[84] M. C. Gutzwiller, ”Stochastic behaviour in quantum scattering”, Physica D 7, 341 (1983).
[85] N. L. Balazs, and A. Voros, ”Chaos on the pseudo sphere”, Phys. Rep. 143, 109 (1986).
[86] M. V. Berry, ”Quantum chaology” (The Bakerian lecture), R. Soc. London, Ser. A 413, 183 (1987).
[87] M. V. Berry, “Hearing the music of the primes: auditory complementarity and the siren song of zeta”, J. Phys. A: Math.
Theor. 45 382001 (2012).
[88] R.K. Bhaduri, A. Khare, and J. Law, ”Phase of the Riemann zeta function and the inverted harmonic oscillator”, Phys.
Rev. E 52, 486 (1995); chao-dyn/9406006.
[89] S. Joffily, ”Jost function, prime numbers and Riemann zeta function”, math-ph/0303014
[90] T. M. Apostol, Introduction to Analytic Number Theory, Springer, New-York, 1976.
[91] B. Julia, “Statistical theory of numbers, in Number Theory and Physics”, eds. J. M. Luck, P. Moussa, and M. Waldschmidt,
Springer Proceedings in Physics, Vol. 47, 276, Springer-Verlag, Berlin, 1990.
[92] D. Spector, “Supersymmetry and the Moebius Inversion Function”, Commun. Math. Phys. 127 , 239 (1990).
[93] R. Baillie, ”Experiments with Zeta zeros and Perron’s formula”, arXiv:1103.6226.
[94] J. von Neumann and E. P. Wigner, ”Über Merkwürdige Diskrete Eigenwerte”, Z. Phys. 30, 465 (1929).
[95] B. Simon, ”On positive eigenvalues of one body Schrödinger operators”, Comm. Pure. Appl. Math. 12, 531 (1969).
[96] M. Arai and J. Uchiyama, ”On the von Neumann and Wigner Potentials”, J. Diff. Eq. 157, 348 (1999).
[97] J. Cruz-Sampedro, I. Herbst, R. Martı́nez-Avendaño, ”Perturbations of the Wigner-Von Neumann potential leaving the
embedded eigenvalue fixed”, Annales Henri Poincaré 3, 331 (2002).
[98] G. Mussardo, “The quantum mechanical potential for the prime numbers”; cond-mat.9712010.
[99] H. Wu and D. W. L. Sprung, “Riemann zeros and a fractal potential”, Phys. Rev. E 48, 2595 (1993).
[100] A. Ramani, B. Grammaticos, and E. Caurier, “Fractal potentials from energy levels”, Phys, Rev. E 51, 6323 (1995).
[101] B. P. van Zyl and D. A. W. Hutchinson, “Riemann zeros, prime numbers, and fractal potentials”, Physical Review E, 67,
066211 (2003).
[102] D. Schumayer, B. P. van Zyl, P. Brandon and D. A. W. Hutchinson, “Quantum mechanical potentials related to the
prime numbers and Riemann zeros”, Phys. Rev. E 78, 056215 (2008).
[103] D. Bernard and A. LeClair, ”A Classification of random Dirac fermions”, J. Phys A 35, 2555 (2002); arXiv:cond-
mat/0109552
[104] D. Bernard, E.A. Kim, and A. LeClair, ”Edge states for topological insulators in two dimensions and their Luttinger-like
liquids”, Phys. Rev. B86 205116 (2012); arXiv:1202.5040.
[105] A. Altland and M. Zirnbauer, ”Nonstandard symmetry classes in mesoscopic normal-superconducting hybrid structures”,
Phys. Rev. B55, 1142 (1997).
[106] N.M. Katz and P. Sarnak, ”Random matrices, Frobenius eigenvalues and monodromy”, Bull. Am. Math. Soc. 36, 1 (1999)
(Am. Math. Soc., Providence, 1999).
[107] J. J. Verbaarschot and T. Wettig, ”Random Matrix Theory and Chiral Symmetry in QCD”, Annu. Rev. Nucl. Part. Sci.
50, 343 (2000); arXiv:hep-ph/0003017.
[108] M.A. Stephanov, J.J.M. Verbaarschot and T. Wettig, ”Random Matrices”, Iin the Wiley Encyclopedia of Electrical and
Electronics Engineering, Supplement 1 (2001); arXiv:hep-ph/0509286
[109] J.J.M. Verbaarschot, ”Handbook Article on Applications of Random Matrix Theory to QCD”, Oxford Handbook of
Random Matrix Theory, arXiv:0910.4134.
[110] Y. V. Fyodorov, E. Strahov, ”On correlation functions of characteristic polynomials for chiral Gaussian Unitary Ensem-
ble”, Nucl. Phys. B 647, 581 (2002); arXiv:hep-th/0205215.
[111] O. Boada, A. Celi, J. I. Latorre and M. Lewenstein, ”Dirac Equation For Cold Atoms In Artificial Curved Spacetimes”,
New J.Phys.13:035002 (2011); arXiv:1010.1716.
[112] J. Rodrı́guez-Laguna, A. Celi, L. Tarruell and M. Lewenstein, ”Synthetic Unruh effect in cold atoms”. In preparation.
[113] P. R. Giri, ”Physical realization for Riemann zeros from black hole physics”, arXiv:0905.4939.
[114] G. t Hooft, ”On the quantum structure of a black hole”, Nucl. Phys. B 256, 727 (1985).
[115] J. I. Latorre and G. Sierra, “Quantum Computation of Prime Number Functions”, Quant. Inf. and Comp., 14, 0577
(2014); arXiv:1302.6245.
[116] J. I. Latorre and G. Sierra, ”There is entanglement in the primes”, arXiv:1403.4765.