Lectures on Random Matrices

Tiago Pereira
Department of Mathematics
Imperial College London, UK

1
 Preface
Lecture notes for a short course on the school for complex systems in Sao Jose,
Brazil. The goal of these lecture is to expose the student to the main concepts and
tools of random matrices. This short course consists of a few lectures to students of
various backgrounds. Therefore, I have chosen to include many elementary exam-
ples throughout the text. I tried to combine heuristic arguments and to communicate
the main ideas. There are beautiful connections between many branches of mathe-
matics, and the theory is aesthetically very pleasing. I tried to capture some of these
connections and to highlight the main points, however, often I chose to sacrifice the
precision of the statements. I also don’t present the proofs, but I will offer the ref-
erence. I’m in debt with Alexei Veneziani and Daniel Maia for his critical reading
of the text and for the exercises.

2
Contents
Motivations 4

Lecture 1: Wigner Semi-Circular Law 6

A neat example: Graph Theory . . . . . . . . . . . . . . . . . . . . 11
Stability of Species . . . . . . . . . . . . . . . . . . . . . . . . . . 12

Lecture 2: Gaussian & Unitary Ensembles 13

Orthogonal and Unitary Ensembles . . . . . . . . . . . . . . . . . . 13
General Unitary Ensembles: . . . . . . . . . . . . . . . . . . . . . 16
Joint Probability of Eigenvalues . . . . . . . . . . . . . . . . . . . 17

Lecture 3: Universality 19
Eigenvalues as a Gas of electrons . . . . . . . . . . . . . . . . . . . 19
Orthogonal Polynomials and Integral Kernel . . . . . . . . . . . . . 20
Universality – Heuristics . . . . . . . . . . . . . . . . . . . . . . . 22
Rescaled Integral Kernel and Universal Limit . . . . . . . . . . . . 22

Lecture 4: Non-Hermitian Random Matrices 23

3
 Motivations
In the mid 50s a large number of experiments with heavy nuclei was performed.
These heavy atoms absorb and emit thousands of frequencies. So an experiment of
this kind offers us a great number of differences in the energy levels, and difficult
to find the set of levels behind the given differences.
In fact, it was virtually impossible to know the levels energy exactly and label
them according to good numbers quantum. To tackle this problem one is required
to understand the eigenvalue problem

Hψi = Ei ψi

where H is the Hamiltonian of the system, and Ei is the energy levels along with
the eigenfunctions ψi . Not surprisingly, writing the Hamiltonian H is already a
hard problem, as there are hundreds of nucleons involved. This large systems are
typically non-integrable, so solving the eigenvalue problem is undoable.
Wigner and Dyson were the first to attack the problem through a statistical point
of view. Instead of searching an approximate solution for the nuclear system, they
focused on the distribution of energy levels. Dyson [1] summarizes the motivation
behind the use of statistical methods:
The statistical theory will not predict the detailed sequence of levels in any one
nucleus, but it will describe the general appearance and the degree of irregularity of
the level structure, that is expected to occur in any nucleus which is too complicated
to be understood in detail.
This view led Wigner to develop a theory based on random matrices for ex-
plaining the distribution of the energy levels [2]. Wigner assumed that the detailed
knowledge of the system would not be relevant for statistical description of the
system. Starting from these assumptions Wigner proposed the description of prop-
erties of a heavy nucleus through an ensemble random matrices, where the entries
(elements) of the matrix would be independently chosen following a distribution.
Additional system information could be obtained through the inherent symmetries,
for example, invariance under time translation invariance and rotational. Such sym-
metries would place a distinct ensembles matrices.
This approach was indeed very successful. Much of the motivation for the study
of random matrices comes from the fact that once removed the dependent part of the
model used, the correlation of levels of different systems exhibit universal features
in a variety of physical situations [3, 4].
Today, random matrices have a wide range of applications starting particle
physics elementary [5] covering quantum hydrodynamics with applications in fluid
Hele-Shaw [6] and applications detection of epilepsy [7]. Another important prob-
lem that can be addressed using the theory of random matrices is the emergence of
collective behavior in complex networks [8, 9].

4
 Zeros of the Riemann zeta function: Lets discuss this interesting example.
Recall that the Riemann zeta function
∞
X 1
ζ(s) = s
, Re s > 1,
n=1
n

has a meromorphic continuation to all C, and apart from the trivial zeros at z =
−2n, on the negative axis, all the other zeros are conjectured to lie on the critical
line Re s = 1/2, this is precisely the Riemann hypothesis: The non-trivial zeros
zeros lie on the line Res = 1/2, that is, s = 1/2 + iγ.
A surprising observation is that the random matrix theory describe the distribu-
tion of the non-trivial zeros of ζ. Assuming the Riemann hypothesis, Montgomery
rescaled the imaginary parts of the zero
γj log γj
γj → γ̃j = ,
2π
to have a mean spacing of 1

#{j ≥ 1 : γ̃j < T }

→ 1.
T
He then obtained an expression for pairs of zeros

1
R(a, b) = lim #{ pairs (j1 , j2 ) : 1 ≤ j1 , j2 ≤ n, γ̃j1 − γ̃j2 ∈ (a, b)}
n→∞ n
for any interval (a, b). Montgomery gave a talk in Princeton about his results, and
Dyson could not attend the talk. However, they sope later and Montgomery went
on explaining that he wants to obtain an expression for the pairs of zeros. Dyson
then asked whether he found
Z b  2 !
sin πu
R(a, b) = 1− du
a πu

which was precisely Montgomery’s results... Wait a bit, how come? Dyson explain
that this is what one should obtain if the zeros were to behaving as the eigenvalues
of the GUE, see Ref. [10] for details.

5
 Lecture 1: Wigner Semi-Circular Law
First lets fix some notation. Recall that a matrix H = (Hij )ni,j=1 is Hermitian if and
only if
H = H †,
where † stands for the transpose conjugate. In terms of the matrix elements the
hermitian properties reads
Hij = H̄ji ,
where ¯· stands for the complex conjugate. If we need to explicit the real and com-
plex components of the elements we denote

Hij = HijR + iHijI ,

where HijR is the real part and HijI the complex part.
A particular case is real symmetric matrices. A matrix H is real symmetric if
and only if all its entries are real and

H = HT ,

where T stands for the transpose.

Exercise 1. Let H be a Hermitian matrix. Show that all eigenvalues of H are real
The spectral theorem for Hermitian matrices states that
Theorem 1. Let H be a Hermitian matrix. Then, there exists an orthonormal basis
consisting of eigenvectors of H, and each eigenvalue is real. Moreover, H admits
the decomposition
H = U ΛU †
where U is the matrix of eigenvectors and Λ = diag (λ1 , · · · , λn ) is the matrix of
eigenvalues. And
U U † = U † U = 1,
that is, the matrix U is unitary.
Hence, Hermitian matrices can be decomposed in terms of its spectral coordi-
nates Now we are ready to define our object of study

Definition 1. A Wigner matrix ensemble is a random matrix ensemble of Hermitian

matrices H = (Hij )ni,j=1 such that
– the upper-triangular entries Hij , i > j are iid complex random variables with
mean zero and unit variance.

– the diagonal entries Hii are iid real variables, independent of the upper-
triangular entries, with bounded mean and variance.

6
Example 1 (Real-Symmetric Matrices). As we discussed real symmetric matrices
are a particular case of Hermitian matrices. Lets see how the Wigner ensemble
takes form for 2 by 2 matrices. Any two by two real symmetric matrix has the form
 
a b
H= .
b c

To have a Wigner ensemble we impose that a, b and c are independent and identi-
cally distributed. For example, they could be distributed according to a Gaussian
with zero mean and variance 1. The collection of all these matrices will form the
Wigner Ensemble.
There are many statistics of the Wigner ensemble one wishes to consider such
as the eigenvalues. Of particular interest is the operator norm

kHk := sup |Hx|

x∈Cn :|x|=1

where | · | is a vector norm. This is an interesting quantity in its own right, but also
serves as a basic upper bound on many other quantities. For example, all eigen-
values λi (H) of H have magnitude at most kHk. Because of this, it is particularly
important to get good understanding of the norm.
Theorem 2 (Strong Bai-Yin theorem, upper bound). Let h be a real random vari-
able with mean zero, variance 1, and finite fourth moment, and for all 1 ≤ i ≤ j, let
Hij be an iid sequence with distribution h, and set Hji := Hij . Let H := (Hij )ni,j=1
be the random matrix formed by the top left n × n block. Then almost surely one
has
kHk
lim sup √ ≤ 2.
n→∞ n
√
This means that operator norm of H√is typically of size O( n). So it is natural
to work with the normalised matrix H/ n.
The Semi-Circular Law: A centerpiece in random matrix theory is the Wigner
semi-circle law. It is concerned with the asymptotic distribution of the eigenvalues
   
H H
λ1 √ ≤ . . . ≤ λn √
n n

of a random Wigner matrix H in the limit n → ∞.

It appears that the histogram of eigenvalues, called the density of eigenvalues,
converges to a deterministic shape. In fact, this is true. The density of eigenvalues
of any Wigner matrix has a limiting distribution known as Wigner’s semicircle law:
1 1/2
µ0sc (x) := (4 − |x|2 )+ dx,
2π
where (x)+ = x if x > 0 and 0 otherwise

7
Example 2. Let A be a n × n random matrix with real entries. Notice that
A + AT
H=
2
is a real symmetric matrix. In matlab the command

n = 2000;
A = (randn(n)/sqrt(n))*sqrt(2);
H = (A+A’)/2;

generates a real symmetric random matrix with Gaussian entries. We plot the
distribution of eigenvalues we can use the command

d = eig(H);
f,x = hist(d,50);
bar(x,f/trapz(x,f))

The result is well described by the semi-circle law. This would this need a proper
normalization.
0.35

0.3

0.25

0.2

0.15

0.1

0.05

0
ï2 ï1.5 ï1 ï0.5 0 0.5 1 1.5 2

Figure 1: Distribution of Eigenvalues

We will state two results on this convergence.

Theorem 3 (Wigner’s Semicircle Law). Let Hn be a sequence of Wigner matrices
and I an interval. Then introduce the random variables
√
#{λj (H/ n) ∈ I}
En (I) = . (1)
n
Then En (I) → µsc (I) in probability as n → ∞.
Wigner realised that one can study the behavior of the random variables En (I)
without computing the eigenvalues directly. This is accomplished in terms of a
random measure, the empirical law of eigenvalues.

8
Definition 2. The empirical law of eigenvalues µn is the random discrete probabil-
ity measure
n
1X √ .
µn := δ
n j=1 λj (H/ n)
Clearly this implies that for any continuous function f ∈ C(R) we obtain
Z n
1X
f dµn = f (λj ) (2)
n j=1
For the matrix ensemble the corresponding function µn is now a random mea-
sure, i.e. a random variable taking values in the space of probability measures on
the real line. The semicircle law first proved by Wigner states that the eigenvalue
distribution of the normalized matrices converges in probability as n → ∞ to a non
random distribution
µn → µsc .
This last statement can be slightly confusing. The sequence of random measures
µn (in the space of probability measures in the real line) converge in probability
(resp. converge almost surely) to a deterministic limit, which is a deterministic
probability measure! The precise statement is the following:
Theorem 4 (Wigners Semicircle Law). Let Hn be a sequence of Wigner matrices.
Then the empirical law of eigenvalues µn converges in probability to µsc as n → ∞.

Precisely, for any continuous bounded function f and each ε > 0,

Z Z 
lim P f dµn − f µsc > ε = 0.
n→∞

Comments about the proof: There are two basic schemes to prove the theorem,
the so-called Moment approach and the resolvent approach. The classical Wigner
method concerns with the moments. We will brief discuss the main ideas on this
technique.
Exercise 2. Without loss of generality (why?), let f ∈ C(R) be a polynomial. Show
that Z
1
f dµn = Tr f (H)
n
Hint: Use that
Z
1X
f dµn = f (Λjj )
n j
1
= Tr f (Λ)
n
1
= Tr U † f (Λ)U
n
1
= Tr f (H) (3)
n
9
 In this formulation, we can use the spectral theorem to eliminate the explicit
appearance of eigenvalues in the law. This is done using our last exercise. Consider
the moments Z
Mj = E λj dσn (λ)
n
R
Notice that
1
Mjn = E Tr Aj .
n
After a long labor, one derives the relations

n tk if j = 2k
lim Mj = mj =
n→∞ 0 if j = 2k + 1

where k ∈ N, and tk ’s are given by the recurrence relation

k−1
X
t0 = 1 and tk = tk−1−j tj
j=0

Actually, one can obtain

 
1 2k
t2k = Ck :=
k+1 k

The numbers Ck are the Catalan numbers. These are precisely the moments of the
semicircle law
Exercise 3. Let µsc be the semicircle law defined above. Let
Z
mk = xk µsc (dx)

By symmetry, m2k+1 = 0 for all k. Use a trigonometric substitution to show that

2(2k − 1)
m0 = 1 and m2k = m2(k−1) .
k+2
This recursion completely determines the even moments; show that, in fact,

m2k = Ck

Hence, the moments are the same, and one can show that this is equivalent to
the semi-circle law (by the problems of moments).

10
 A neat example: Graph Theory
Consider random graphs of n nodes of labelled undirected graphs of n nodes. We
will use a random graph model and terminology from references [11, 12]. This
model is an extension of the Erdös-Rényi model for random graphs with a general
degree distribution. The model consists in to prescribing the expected values of the
node degrees. For convenience, any given sequence of expected degrees

wn = (w1 , w2 , · · · , wn ).
We consider thus an ensemble of random graphs G(wn ) in which an edge be-
tween nodes i and j is independently assigned with success probability
wi wj
pij = Pn .
k=1 wk

In order to ensure that pij ≤ 1, we assume that wn is chosen so that

2 P
max1≤k≤n wk ≤ nk=1 wk .
A realisation of a graph in the ensemble G(wn ) is encoded in the adjacency
matrix A = (Aij ) with (0, 1)-entries determining the connections among nodes of
the graph. The degree κi of the ith node is the number of connections that it receives:
n
X
ki = Aij .
j=1

Notice that κi is a random variable whose expected value is exactly the prescribed
quantity wi . In particular, w1 = max1≤i≤n wi is the largest expected value of a
degree.
Now consider the the combinatorial Laplacian

L=D−A

where D = diag(k1 , · · · , kn ) is the matrix of degrees. This matrix is important for

collective dynamics in networks [13, 14] and for counting the number of spanning
trees. Now consider the normalised Laplacian

L = 1 − D−1/2 AD−1/2

this controls the isoperimetrical properties and the mixing rates of a random walk
on a graph [12]. For graphs with uneven degrees, the above three matrices can have
very different distributions.

The eigenvalues of the normalised Laplacian L satisfy the semicircle law under
the condition that the minimum expected degree is relatively large (much larger than
the square root of the expected average degree).
1
max |λi − 1| ≤ [1 + o(1)] p + O(1/wmin )
i6=0 hwi

11
 Stability of Species
An important question in ecology is the stability of a given food web. In particular, it
is important to understand how the stability of the food web depends on the number
of species and the structure of interaction.
In early 70’s, Gardner and Ashby suggested that large complex systems which
are randomly connected may become unstable when the number of interacting
species increase [15]. In a follow up, Robert May proposed an explanation using
random matrices [16]. Lets discuss May’s ideas here.
In an ecological system these are populations of n interacting species, which in
general obey non-linear differential equations. However, the stability of the stable
configurations is determined by the linearised dynamics around such configurations.
Performing a linearisation one obtains the linear equation
dx
= Ax,
dt
where x is a vector of disturbed populations xj , and A a n × n interaction ma-
trix with elements ajk which characterise the effect of species k on species j near
equilibrium.
Assume that in the absence of interaction disturbances are damped. That is, the
system is stable. We can choose then aii = −1. This sets a time scales in which the
disturbances decay. Then, the interactions are switched on. We also assume that the
strength of interaction from species k to j is the same as j to k. That is, the matrix
A is symmetric 1 . Moreover, the independent elements of A are is equally likely
to be positive and negative and independent. The matrix elements is assigned from
a distribution of random numbers, and this distribution has mean zero and mean
square value α.
Here α is thought of as expressing the average interaction strength. May as-
sumed no symmetry in the matrix interaction. Hence,

A = αH − I

where H is a Wigner matrix. System is stable if and only if all the eigenvalues of A
have negative real parts. The eigenvalues are λ(A) = αλ(H) − 1. In particular we
know that the largest eigenvalue of A must be
√
λmax (A) ≤ 2α n − 1

Hence we obtain
2
α< √
n
Roughly speaking this suggests that within a web species which interact with
many other should do this weakly.

1
We make this assumption for simplicity as it is not quite realistic to think of a symmetric ecosys-
tem.

12
 Lecture 2: Gaussian & Unitary Ensembles
Lets for a moment consider the Wigner ensemble of orthogonal matrices with iid
elements satisfying E(Hij ) = 0 and E(Hij2 ) = 1.
Exercise 4. Show that independence implies that

E(Hij Hpq ) = δij δjq + δip δjp .

Lets assume for a moment that all elements has a Gaussian distribution From
these assumptions on the matrix elements it is possible to obtain a probability den-
sity in the space of real-symmetric matrices. That is, the probability P (H)dH that
an symmetric matrix H lies in a small parallelopiped dH is Gaussian. The volume
of the parallelopiped is given in terms in the independent coordinates
Y
dH = dHkj .
k≤j

Then, using the independence we can compute

1 Y −Hij2 /2 Y −Hii2 /2
P (H) = e e
Z j<i i
( !)
1 1 X 2 X 2
= exp − H + Hii
Z 2 j<i ij i

where Z is the normalization constant. Recall that Hij = Hji , hence

( n
)
1 1X 2
P (H) = exp − H
Z 4 i,j=1 ij

or one can write this as a trace

Exercise 5. Show that the above P(H) can be written as
 
1 1 2
P (H) = exp − Tr H (4)
Z 4
This means that Gaussianity in the matrix elements leads to a probability density
represented a the trace of the matrix squared. Next, we will see that is this the case
regardless the form of the distribution of the elements.

Orthogonal and Unitary Ensembles

Definition 3. The Gaussian unitary ensemble GOE(n) is described by the Gaussian
measure
1 − n Tr H 2
P (H)dH = e 2 dH (5)
Zo
on the space of n × n real symmetric matrices H = (Hij )ni,j=1 .

13
 Here, Zo is a normalization constant, chosen so that the integral against the mea-
sure is equal to one. This ensemble can be redefined in terms of the matrix elements.
The GOE(n) defined on the space of real-symmetric matrices is characterized by
the following properties
1. The ensemble is invariant under every transformation

H 7→ W t HW

where W is any orthogonal matrix, that is W T W = W W T = 1.

2. The elements Hkj with k ≤ j are statistically independent. This means that
the density P (H) will be decomposed into a product form
Y
P (H) = fkj (Hkj )
k≤j

Theorem 5. The conditions 1. and 2. on the matrix elements uniquely determine

the Gaussian measure P (H)dH, up to shifts H → H + const and a rescaling. This
is regardless the form of the distribution (with finite variance).
The proof is rather lengthy and can be found in Ref. [17] Therefore, invariance
under orthogonal transformations and the statistical independence of matrix ele-
ments completely determine the Gaussian orthogonal ensemble. The motivation for
the name of the ensemble becomes from the computation we just performed.
Exercise 6 (Why this Gaussian Ensemble?). Let H = U ΛU T , show that

P (H) = P (Λ)

which implies that that P is a Gaussian density in the spectral coordinates

( )
1 nX 2
P (H) = exp − λ
Z 2 i i

Hint: Use that Tr ABC = Tr BCA.

Remark 1. Notice the factor n difference between the deduced Eq. (4) and the
model Eq. (5). There are a few ways to see the rescaling
√
– Rescaling of the matrices H 7→ H/ n. So we rescale the eigenvalues
 
λ(H) H
√ = λ̃ √
n n

for any eigenvalue λ.

14
 – Rescaling the standard deviation
σ
σ 7→ √
n

Notice this rescaling leads to the factor n in front of the trace, and since
EHij = 0, this variance rescaling corresponds precisely to the matrix rescal-
ing.
This means that our definition of the ensemble already accounts for the normal-
isation. It is possible to show that that the volume element is invariant under real
orthogonal transformation [17, 18]. Moreover, the invariance of P (H) imposes that

P (H 0 )dH 0 = P (H)dH

where H 0 = W t HW , and W is any orthogonal matrix. This equation shows that

this probability view by another observer (via W ) the matrices in H realizing
the same operators fill a parallelopiped dH 0 . The second observer computes the
probability using the same laws as the first observer and obtains the same result.

Lets now introduce the unitary ensemble. Roughly speaking, the only differ-
ences are: i) we deal with Hermitian matrices which are complex as opposed to
real symmetric, ii) ensemble invariance is with respect to unitary transformations,
which can be think of as the generalisation of orthogonal transformations to Hermi-
tian matrices.

Definition 4. The Gaussian unitary ensemble GUE(n) is described by the Gaussian

measure
1 − 1 Tr H 2
P (H) = e 2
Zu
on the space of n × n Hermitian matrices H = (Hij )ni,j=1 .
Here, Zu is a normalization constant, chosen so that the integral against the
measure P is equal to one. The Gaussian unitary ensemble can also be redefined in
terms of properties of the matrix elements by the following properties

U1 the ensemble is invariant under unitary transformations

H 7→ U t HU,

where U is a unitary matrix U † U = U U † = 1.

U2 linearly independent elements of H are statistically independent. This means

that is written as products of functions which depend only on the independent
matrix elements.

15
 Notice that the volume element is also given in terms of the independent entries.
Y Y
R I
dH = dHkj dHkj .
k≤j k<j

The conditions on the matrix elements U1 and U2 also completely charaterize

the GU E. Similar to the GOE case, these conditions on the matrix elements yield
the Guassian measure uniquely P , up to the the shifting and rescaling. This has a
nice consequence: regardless the specific details of the distribution of the matrix
elements the statistical properties of the ensemble is the same.

General Unitary Ensembles:

We will discuss some properties of the unitary ensembles. However, it makes more
sense and yet it yields more interesting results to consider slightly more general
classes of ensembles than just the GUE. Lets just rephrase our observations. Con-
sider the polynomial
X k
V (x) = aj x j
j=2

and define the extension of this polynomial as a matrix function

k
X
V (H) = aj H j .
j=2

Then, we can definite the unitary random matrix unitary ensemble

Definition 5. The unitary ensemble is described by the measure
1 −n Tr V (H)
P (H)d(H) = e dH.
Z
in the space of Hermitian matrices.
This ensemble is indeed invariant under unitary transformation. For any unitary
matrix W we have that
dH 0 = dH
where H † = W HW † . Moreover, because the trace is cyclic Tr(ABC) =
Tr(BCA), for any matrices A, B and C, we obtain that
P (H) = P (H 0 ).
In particular, P (H) = P (Λ), where Λ is the matrix of eigenvalues.

Our discussion up to now revealed statistical independence of the matrix ele-

ments leads uniquely to V (H) = H 2 . For this observations we are guarantee that
polynomials of degrees higher than 2 (or real-analytic functions) V will correspond
the matrix ensembles with correlations in the matrix elements. We will take a close
look in these models.

16
 Joint Probability of Eigenvalues
In the last lecture we saw how the independence of the matrix entries and invariance
leads to the Guassian measure. Our goal in this lecture will be to obtain a equation
for the joint probability of eigenvalues from this probability measure.
The idea here is to write the matrix in its spectral form

H = U ΛU † (6)

where U is the matrix of eigenvectors and Λ the matrix of eigenvalues. Notice that
the probability P (H)dH is given in terms of the matrix elements. One could try to
express P (H)dH in terms of the pair (U, Λ) and integrate over the variables U . At
the end (of this laborious process) one is left with a probability depending only on
the eigenvalues Λ. As we discussed in the previous lecture P (H) = P (Λ), that is,
this density does not depend on the eigenvector variables. On the other hand, we
will discuss that the volume transforms as
Y
dH = |λj − λi |2 dΛdU
i<j

Combining these two results we obtain that in the spectral coordinates (Λ, U ) the
unitary ensemble reads

P (H)dH = P (U ΛU † )d(U ΛU † )
= P (Λ)JdΛdU
1 −n Pi V (λi ) Y
= e |λi − λj |2 dΛdU.
Z i<j

Integrating over the variables U we are left with a joint density of eigenvalues.
The main result of the section is the following
Claim 1. The joint distribution of eigenvalues is given by
1 −n Pi V (λi ) Y
P (λ1 , . . . , λn ) = e |λi − λj |2
Z

The contribution |λi − λj |2 came from the volume element dH. This terms
Q
will play a major role in the analysis. And it is the main responsible for the beautiful
behaviour of the ensemble such as universality in the eigenvalue statistics. To be
more clear I should have change the letter for the joint probability of eigenvalues
as it may be confusing since P used to denote the Gaussian density in the space of
Hermitian matrices. However, this is the standard notation.

Claim 2. For the GOE(n) the joint distribution of eigenvalues

1 −n Pi V (λi ) Y
P (λ1 , . . . , λn ) = e |λi − λj |
Z

17
We will discuss the main ideas of this construction. In mathematical language we
wish to parametrize the probability in terms of the spectral coordinates. Consider
the mapping from the spectral variables to the matrix elements
(Λ, U ) 7→ H = U ΛU †
This spectral parametrisation must be a bijective change of variables. There are
two obstacles to this: a) if there is a multiple eigenvalue then there is no canonical
way to select orthonormal eigenvectors from the corresponding eigenspace. And b)
eigenvectors are is determined only up to an arbitrary sign or phase.
For the eigenvectors, we could simply say: we don’t care about the phase (or
sign), that is, we only consider eigenvectors up to an arbitrary change in the phase.
It turns out, that this can be actually done in a rigorous way in terms of cosets in
the unitary group [10, 17]. So I won’t develop this ideas here. It is possible to
overcome the challenge a) if we take a measure theoretic perspective. It is possible
to show that almost every hermitian matrix one picks at random will have distinct
eigenvalues. This means that for all what matters the coordinate change can be
done. I wish to be more precise here.
Recall that the map must be injective and surjective. By the spectral theorem,
we can always write a Hermitian matrix in its spectral representation 6. This means
that the map is surjective. It remains to show that the map is
Let H(n) be the space of hermitian matrices
H(n) := {H ∈ Cn×n : H † = H}
Moreover, let H(n)
b ⊂ H(n) be the subset of hermitian matrices with distinct
eigenvalues. The first major observation is the following
Lemma 1. H(n)
b ⊂ H(n) is open, dense and has full measure.
I won’t present the proof of this claim here, see [10, 17, 18]. Hence, consid-
ering the eigenvectors coordinates (up to phases) and the eigenvalues is an honest
parametrisation in H(n).
b Moreover, since one performs an integration to obtain
Claim 1 the zero measure set of non simple eigenvalues plays no role.
Performing the change of variables, one obtains the Jacobian of the transforma-
tion:
dH = J(H)dΛdU
As I discussed before the volume element is invariant under unitary similarity trans-
formations. This implies that
dH = d(QHQ∗ ) = J(QHQ∗ )dΛd(QU ) = J(QHQ∗ )dΛd(U )
Therefore J(H) = J(QHQ∗ ), hence, J(H) is a symmetric function of the eigen-
values of H only. Thus we can write J(H) = J(Λ).
Proposition 1. Y
J(λ1 , . . . , λn ) = |λj − λi |2
i<j

18
 Lecture 3: Universality
Physical systems in equilibrium obey the laws of thermodynamics. So, whatever
the precise nature of the interaction between the particles at a detailed level, at
the macroscopic level, physical systems exhibit universal behavior – they are all
governed by the same thermodynamical laws. The random matrix theory share the
same similarities.

Eigenvalues as a Gas of electrons

Note that we can rewrite the joint distribution of eigenvalues as
Y Pn
P (λ1 , · · · , λn ) = Zn−1 |λi − λj |2 e−n i=1 V (λi ) (7)
1≤i<j≤n
( n
!)
X −1
X
= Zn−1 exp − log |λi − λj | +n V (λi )
1≤i6=j≤n i=1

Those educated in statistical mechanics may be overjoyed. The normalisation

constant Z is the same as the partition function of a gas of electrons in a plane
when the electrons are confined to a one-dimensional straight wire under a quadratic
potential. Lets explore this analogy for two reasons, first it is very nice, and second
it will give us an nice intuition about the eigenvalues.
Just recall that the Couloumb potential in the plane is logarithmic (to see this you
either solve the poisson equation in two dimensions or solve the Maxwell equations
with a infinitely long wire passing through the origin). Hence that eigenvalues are
charges of the same sign on a 2-d Coulomb potential. Therefore, the interaction
between eigenvalues is repulsive. This is know as repulsion of eigenvalues.
If one consider the coulomb term alone the eigenvalues would then scape to
infinity. Here is where the function V plays a role of a confining potential. And
hence, the name. This gives a good reasoning for having the size n multiplying the
potential, it provide a balance between coulomb and electrostatic. Otherwise, the
equilibrium configuration could be found.

Variational Problem and Equilibrium Configurations: Lets just explore a bit

more the relation the similarities between our joint distribution of electrons and the
Coulomb gas. Notice that we could rewrite the joint distribution as

2IV
P (λ1 , · · · , λn ) = Zn−1 e−n (µn )
(8)
where n
1X
µn (λ) = δ(λ − λi ) (9)
n i=1
is the empirical law of eigenvalues and
Z
I (µ) ≡ (V (z) + U µ (z)) dµ (z)
V
(10)

19
is the “electrostatic energy ” associated with µn with
Z
U (z) = log |z − w|−1 dµ (w)
µ
(11)

being the logarithmic potential.

Remark 2. We see this equation as the energy functional of n charges distributed
according to the empirical law µn under the influence of an external potential V .
Thus, the first contribution of (10) is the interaction energy of the n charges with the
potential V and the second contribution comes from the of the Coulomb interaction
energy between charges.
Note that the scale n in the the potential V in (15) has an important effect, and
balances the interaction energy with the potential of the chargers and the energy
of interaction between the charges appear the same order n : this is the physical
argument which justifies the scale of the eigenvalues of H for the support empirical
law has a compact support as n → ∞ .
We can then turn this problem to a variational problem. The leading contribution
to the integrals with respect to (8) in the thermodynamic limit n → ∞ is given by
the following variational problem:

EV ≡ inf I V (µ) , E < ∞ (12)

µ∈M(R)

where the infimum is taken over the set M(R) of all probability measure on R. If
such probability measure exists µV for the potential V

E V = I(µV ),

we say that µV is the equilibrium measure associated with the potential V .

Notice that µV is the equilibrium distribution of the charges in the thermo-
dynamic limit n → ∞) and at the zero temperature limit (T −1 → ∞) of a bi-
dimensional Coulomb gas at a temperature T = N −1 .

Orthogonal Polynomials and Integral Kernel

There is a beautiful and extremely useful connection between the theory of orthog-
onal polynomials and random matrices. We can write the joint probability of eigen-
values as determinant
1
Pn (λ1 , · · · , λn ) = det (Kn (λi , λj ))ni,j=1 (13)
n!
of the integral kernel
n
n n
X
Kn (x, y) = e− 2 V (x) e− 2 V (y) φnj (x) φnj (y) (14)
j=1

20
  n
where φnj j=1
is a set of orthogonal polynomials with respect to the weights

dν (x) = e−nV (x) dx.

The demonstration is somewhat laborious and we won’t present it here. Please,

see Ref. [17].

Correlation Functions: We can introduce the n-point correlation function as-

sociated with Pn in a similar manner as in statistical mechanics
Z n
n! Y
Rkn (λ1 , · · · , λk ) = Pn (λ1 , · · · , λn ) d2 λi .
(n − k)! i=k+1

A remarkable result reveals that the n−point correlation function can also be
represented as a determinant of the Integral kernel

Rkn (λ1 , · · · , λk ) = det (Kn (λi , λj ))ki,j=1 .

Therefore, the asymptotic behavior of of the integral kernel KN will dictate the
statistics of the eigenvalues of the unitary ensemble as n → ∞.

21
 Universality – Heuristics
The universality in the eigenvalue statistics can be heuristically understood in terms
of the Dyson interpretation of the eigenvalues as a Coulomb gas [17]. For a hermi-
tian matrix the joint probability of the eigenvalues reads as
n X n
X o
Pn (λ1 , · · · , λn ) ∝ exp ln |λi − λj |−1 − n V (λi ) . (15)
1≤i6=j≤n i=1

As we explored last section, this joint distribution of eigenvalues has the same
form as a 2 dimensional Coulomb gas restricted to one dimension. The eigenvalue
interaction is given by a Coulomb term ln |λi − λj |−1 under a potential V (λ) that
confines the eigenvalues in a bounded set of the real line.
Since the eigenvalues are restricted to one dimension, if an eigenvalue is located
between two eigenvalues the Coulomb potential does not allow it to switch its posi-
tion with the surrounding eigenvalues. As the number of eigenvalues increase (the
size of the matrix) the eigenvalues become closer and the effect of the Coulomb
potential overtakes the potential V (λ) (due to the logarithm behavior near zero).
If we then rescale the mean distance between eigenvalues to the unity, heuris-
tically this is equivalence to switch off the external potential. Thus no matter the
external potential, after the rescaling only the Coulomb potential affects the eigen-
values. Notice that the ”Coulomb” interaction comes front he volume element and
therefore, is intrinsic of the matrix ensemble. Hence, the universality. Recently, all
this reasoning has been rigorously proven (see e.g. [10] and references therein). We
will discuss a sketch of these ideas here.

Rescaled Integral Kernel and Universal Limit

Lets discuss in slightly more detail the meaning of universality for Hermitian en-
sembles. The quantity playing a major role in the analysis of random ensembles is
the integral kernel
n
n
X
Kn (x, y) = e− 2 (V (x)+V (y)) φj (x) φj (y) (16)
j=1

where V (x) is the potential [see Eq. (15)] and {φj }nj=1 denotes the set of poly-
nomials up to order n − 1, orthogonal with respect to the weight exp{−nV (x)}.
The important statistical quantities
R associated with the matrix ensemble such as
the eigenvalue density ρ(λ1 ) = Pn (λ1 , · · · , λn )dλ2 · · · λn can be obtained by the
Kn [10]. For large n the relation reads

Kn (λ, λ) = nρ(λ) (1 + o (1)) ,

where o(1) converges to 0 as n → ∞.

To proceed the universality analysis, we must rescale the integral kernel. The
scale is chosen such that the fraction of eigenvalues in an interval of length s close
to a point λ equals s, in other words, the average spacing between eigenvalues is

22
unity. It can be shown that the correct scale is Kn (λ, λ). Indeed, since the fraction
of eigenvalues in the interval A ⊂ R is given by
n
1X
fn (A) = χA (λi ), (17)
n i=1

it is possible to show that [19] for large n we obtain

  
s
nfn λ, λ + = s[1 + o(1)].
Kn (λ, λ)

Once we have the proper scale Kn (λ, λ), we proceed the analysis by rescaling the
integral kernel
 
1 x y
K
e n (x, y) = Kn λ + ,λ + . (18)
Kn (λ, λ) Kn (λ, λ) Kn (λ, λ)

The astonishing result [19] is then

e n (x, y) = sin π(x − y)

lim K (19)
n→∞ π(x − y)
exists pointwise for Freud–type or real analytic potentials V . Since it does not
depend on V the Hermitian ensembles in this sense are universal. The natural ques-
tions is whether normal ensembles also display universal eigenvalue statistics.

Lecture 4: Non-Hermitian Random Matrices

The case of non-Hermitian ensembles are significantly less developed than the Her-
mitian. If the matrix is normal2 some deeper understanding is possible. For exam-
ple, Hermitian ensembles and the Wigner semi-circle law can be obtain as a limit
of the normal case [20, 21]. Moreover, these ensemble have physical interpretations
to quantum Hall effects and pattern formation [24–26]. Universality questions for
normal ensemble have been addressed recently [6, 27, 28].
Lets is discuss a model introduced in Ref. [29]. Consider an ensemble of random
matrices such that
hHij Hji i = τ
where τ ∈ [−1, 1]. This means that for τ → −1 this mimics a totally anti-symmetric
ensemble, whereas for τ → 1, we would have a symmetric ensemble. Moreover,
assume that the distribution of the independent elements is Guassian. Then, it is
possible to show that
 
1 n T 2
P (H)dH = exp − Tr(HH − τ H ) dH
Z 2(1 − τ 2 )
2
A matrix is normal if and only if it commutes with its adjoint

23
In this case, the eigenvalues are complex, but tt can be shown that the eigenval-
ues are uniformly distributed inside ellipsoids determined by τ . Indeed, density of
eigenvalues reads as
 1
0 πab
, if(x/a)2 + (y/b)2 ≤ 1
µ (x + iy) =
0, otherwise

where a = 1 + τ and b = 1 − τ . In the case of totaly assymetric ensembles τ = 0,

we recover (almost surely) the Girko ensemble, with eigenvalues being uniformly
distributed inside the circle.

References
[1] F.J. Dyson, J. Math. Phys. 3, 140 (1962).

[2] E.P. Wigner, Conference on Neutron Physics, Tenessee, Lab, Rept. ORNL-
2309 p.67 (1959). Reimpresso em [3].

[3] C.E. Porter, Statistical Theories of Spectra: Fluctuations, (Academic Press,

New York, 1965).

[4] P. Deift, Orthogonal Polynomials and Random Matrices: A Riemann-Hilbert

Approach, (New York University 1999).

[5] C. Fernández-Ramı́rez and A. Relaño, Phys. Rev. Lett. 98, 062001 (2007).

[6] H. Hedenmalm e N. Makarov, Quantum Hele-Shaw flow,

arXiv:[Link]/0411437 (2004).

[7] Y. Lai, M. G. Frei, e I. Osorio, Phys. Rev. Lett. 98, 108102 (2007).

[8] M. Timme, F. Wolf, and T. Geisel, Phys. Rev. Lett 92, 07410 (2004).

[9] T Pereira, D Eroglu, G Bagci, U Tirnakli, and HJ Jensen, Phys. Rev. Lett. 110,
234103 (2013).

[10] P. Deift. “Orthogonal Polynomials and Random Matrices: A Riemann-Hilbert

Approach”. American Mathematical Society. New York University, New York
(2000).

[11] B. Bollobás, Random Graphs, Cambridge University Press; 2 edition (2001).

[12] F. R. K. Chung and L. Lu, Complex Graphs and Networks, American Mathe-
matical Society (2006).

[13] L. M. Pecora and T. Carrol, Phys. Rev. Lett. 80, 2109 (1998), H. Fujisika and
T. Yamada, Progr. Theoret. Phys. 69, 32 (1983).

24
[14] T. Pereira, J. Eldering, M. Rasmussen, A. Veneziani, Nonlinearity 27, 501
(2014).

[15] M.R Gardner, and W.R. Ashby, Nature 228, 784 (1970).

[16] R. May, Nature 238, 413 (1972).

[17] D. L. Mehta. “Random Matrices - Revised and Enlarged”. 2nd Edition. Aca-
demic Press. 1991.

[18] A Venezianni, T Pereira, A note on the volume form in normal matrix space,
(2013).

[19] P. Deift, T. Kriecherbauer, K. T.-R. McLaughlin, S. Venekides, and X. Zhou,

J. Comput. Appl. Math. 133, 47-63 (2001).

[20] AM Veneziani, T Pereira, and DHU Marchetti, Journal of Physics A: Math.

Theor. 44, 075202 (2011).

[21] P. Elbau e G. Felder, arXiv:[Link]/0406604v2 (2005).

[22] P.B. Wiegmann, A. Zabrodin, Comm. Math. Phys. 213, 523 (2000).

[23] Ellison R, Gardner V, Lepak J, et ali., Inter. Jour. Bif. Chaos 15 (2005) 2283;
W.J. Freeman and J.M. Barrie, Jour. Neurophys. 84 (2000) 1266.

[24] L.L. Chau and Y. Yu, Phys. Lett. A 167, 452 (1992).

[25] A. Marshakov, P. Wiegmann, A. Zabrodin, Commun. Math. Phys. 227, 131

(2002).

[26] A. Zabrodin, arxiv:0210331 (2002).

[27] AM Veneziani, T Pereira, and DHU Marchetti, Journal of Mathematical

Physics 53, 023303 (2012).

[28] L.L. Chau and O. Zaboronsky, Comm. Math. Phys. 196, 203 (1998).

[29] H.J. Sommers, A. Crisanti, H. Sompolinsky, and Y. Stein, Phys. Rev. Letts.
60, 1895 (1998).

25