Zuber Talk Randommatrices

Free Probability and Large N Limit
Introduction to Random Matrices

from a physicists perspective
Jean-Bernard Zuber
Berkeley March 2007
Random matrices: a major theme in theoretical physics since Wigner (1951);

a major theme in mathematical physics and mathematics for the last 20 years.
Plan of talk
A short history of the subject
Random systems, from nuclear Hamiltonians to mesoscopic systems to
financial markets. . .
Large N limit of U(N) gauge theory Topological expansion and the
counting of maps and other planar objects Stat mech models on
random lattices
Double scaling limit and 2D quantum gravity
QCD, Dijkgraaf-Vafa, etc
Feynman diagrams and large N limit
Counting of maps or triangulations (cf Edouard Maurel-Segalas talk)
Computational methods: saddle point; [loop equations (cf EMS)];
[orthogonal polynomials (cf Paul Zinn-Justin)]
A short history of the subject

1. Random systems [Wigner 1951]
Study of spectrum of large size Hamiltonians of big nuclei, regarded as
Gaussian random matrices, subject to some symmetry or reality property
(GOE, GUE, GSE, . . .)
Eugene P. Wigner
1902 - 1995
Freeman Dyson
1923 -
Madan Lal Mehta

1932 - 2006
Statistics of energy levels of random Hamiltonians
Level spacing histogram

for a large set of nuclear
levels
(Nuclear
Data
Ensemble,
[O. Bohigas et al])
Other random systems and random matrices : from transport properties (for
ex. universal fluctuations of electric conductivity) in disordered mesoscopic
systems to financial markets. . .
2. Large N limit of U(N) gauge theories

Gauge theories : quantum field theories based on a principal fibre bundle with a compact
group G. Gauge field A (connection) lives in
Lie algebra of G. For G = U(N), A is a N N
(anti-Hermitian) matrix.
In the search of a non-trivial approximation, it
is natural to look at large N limit, expansion
parameter 1/N 2 , [t Hooft 1974], see below.
Indeed, major simplification in the large N limit of Feynman expansions of
matrix field theories. . .
Consider toy field theories : integrals over a finite number of (large size)
matrices.
Basics of Feynman diagrams

Consider a Gaussian integral over n real variables xi , A = AT > 0 def. matrix
Z
12 xi Ai j x j
d xe
(2)n/2
1
det 2 A
n/2 1
1
1
(2)
b
A
x
A
x
+
b
x
i
i
j
j
i
i
ij bj
2
e
=
d xe 2
1
2
det A
Differentiate w.r.t. bi R n
1
d x xk1 xk2 xk e 2 x.A.x
1 b.A1 .b
e2
hxk1 xk2 xk i :=
=

R
1
2 x.A.x
n
b=0
b
b
k1
k
d xe
1
A1
A
kP1 kP2
kP1 kP
all distinct
y
X
XXX
pairings P of the k
X
...
Wick theorem
xk x k
2
1
xk
propagators
HH
j
H

...
P kP kP
1
kP
Wick theorem also applies to monomials (n = 1 variable for simplicity):

p vertices propagator A1
h(x4 ) p i =
graphs
Non Gaussian integrals (g < 0): power series perturbative expansions

Z
=
=
log Z
1
2 + g x4
Ax
4!
dx e 2
2 12
A
2 21
=
A
p=0 graphs G with 2p lines

and p 4valent vertices
gp
p!
p=0
gp
A2p
|Aut G |
= connected Feynman diagrams

g
1
g2 1
=
+ 4 +
+ 4
2
8A
A 2.4! 2
=
x4 p 1 Ax2
e 2
dx
4!
Matrix Integrals: Feynman Rules

N N Hermitean matrices M, dM = i dMii i< j deMi j dmMi j
Z
=: eF
1
2 + g tr M 4 ]
N[
tr
M
2
4
= dM e
Feynman rules: propagator ij

q
nm
4-valent vertex :
i
l
1
k = N i jk
= gN jk m np qi
For each connected diagram contributing

to log Z: fill each closed index loop with a
disk discretized closed 2-surface
[t Hooft]
10
Power of N in a connected diagram

each vertex N;
each double line N 1 ;
each loop N.
Thus N #vert. #lines + #loops = N Euler ()
[t Hooft (1974)]. For example, compare
gN
A topological expansion :
gN
= log Z =
conn. surf.
22h (h)
N
F (g).
h=0
22genus()
g#vert.()
symm. factor
11
Thus large N limit of matrix integral DMeNtr (M + 4 M ) = generating

function of planar 4-valent graphs. . .(cf census of planar maps by Tutte)
R
[Br
ezin, Itzykson, Parisi, Z. 1978]
limN N12 log Z =
gn
planar diagrams symm.factor
with n 4vertices
or in a dual way, of quadrangulations of 2D surfaces of genus 0

[Kazakov; David; Kazakov-Kostov-Migdal; Ambjrn-Durhuus-Fr
ohlich 85]
12
Thus large N limit of matrix integral DMeNtr (M + 3 M ) = generating

function of planar 3-valent graphs. . .[Brezin, Itzykson, Parisi, Z. 1978]
R
or in a dual way, of triangulations of 2D surfaces of genus 0

[Kazakov; David; Kazakov-Kostov-Migdal; Ambjrn-Durhuus-Fr
ohlich 85]
Triangulated surfaces and discrete 2D gravity
Thus : Large N limit of matrix integrals Counting of planar objects :

maps, triangulations, alternating knots and links [P Z-J & J-B Z], etc, or
of objects of higher topology . . .
13
Large N limit of matrix integral DA DB eNtr (A +cAB+B +gA

generating function of bicolored planar 4-valent graphs. . .
R
4 +gB4 )
14
Large N limit of matrix integral DA DB eNtr (A +cAB+B +gA

generating function of bicolored planar 4-valent graphs. . .
R
4 +gB4 )
i.e. describes Ising model on a random quadrangulated sphere

[Boulatov-Kazakov 1985]
+
+
+ +
+ +
+
+
+
+
+
+
+ ++
++
etc etc, many variations on that theme
Statistical mechanics models on a random lattice
15
Two remarks useful in connection with free probabilities. . .

1
1
1
1
Factorization property h N tr P1 N tr P2 i = h tr P1 ih tr P2 i +O( N12 )
|N
{z N
}
disconnected diagrams
Compare (in Gaussian theory

h N1 tr M12 M22 i
M1
2 +M 2 )
tr
(M
1
2 )
e
and
M2
2
gN
h N1 tr M1 M2 M1 M2 )i
gN
Similar behavior in non Gaussian theory etr (V1 (M1 )+V2 (M2 )) provided
hM1 i = 0, hM2 i = 0.
16
If hM1 i =
6 0, hM2 i =
6 0,
1
0
0
1
tr M1 M2 M1 M2
1
0
0
1
11
00
00
11
11
00
00
11
1
0
0
1
1
0
0
1
0
1
0
1
1
0
0
1
0
1
0
1
1
0
0
1
1
0
0
1
0
1
0
1
1
0
0
1
0
1
0
1
1
0
0
1
11
00
00
11
1
0
0
1
1
0
0
1
1
0
0
1
1
0
0
1
1
0
0
1
Compare (a1 a2 a1 a2 ) = (a21 )2 (a2 ) + 2 (a1 )(a22 ) 2 (a1 )2 (a2 ) for two
free variables
17
3. Double scaling limit

[Br
ezin-Kazakov; Douglas-Shenker; Gross-Migdal 1989, . . .]
F (0) (g) and more generally F (h) (g) have a singularity at g = gc ,

F (h) (g) (gc g)(2str )(1h)
with str , the string susceptibility, typically equal to 1/2 (for the
simplest models M 3 or M 4 above), see below.
As g gc , h# trianglesi =
log F
g
diverges. Expect to make contact with
continuum 2D gravity. Keep all genera h in h=0 N 22h F (h) (g) by letting
gc g 0 as N in such a way that (gc g)(2str )/2 N = fixed.
Very interesting limit : appearance of integrable equations (KdV . . .),
solutions to Painleve equations . . .
Thus, double scaling limit models of 2D quantum gravity
18
4. Other physical applications
Cell decomposition of moduli space of Riemann surfaces, intersection

numbers . . .
[Witten, Kontsevich, 1991 . . .]
QCD, the Dirac operator D

/ = / + A
/ in the presence of a gauge field and
RMT [Verbaarschot et al.]
Dijkgraaf-Vafa 2002 : computing the effective action of supersymmetric
gauge theories in terms of matrix integrals
etc etc
19
Computational techniques
Consider integral over N N Hermitian matrices
Z=
dMeNtr V (M) ,
V (M) a polynomial of degree d + 1. For ex. V3 (M) = ( 21 M 2 + g3 M 3 ) and

V4 (M) = ( 21 M 2 + 4g M 4 ). Note that multi-traces are excluded, for example
(tr M 2 )2 .
Integrand and measure are invariant under U(N) transformations
M UMU . Express both in terms of eigenvalues 1 , , N of M :
Z=
Z N
2 N i=1 V (i )
,
(
)
e
d
i
j
i

i=1
i< j
Several ways to treat this integral: saddle point approximation ; orthogonal

polynomials; loop equation. . .
20
1. Saddle point approximation

Rewrite
Z=
Z N
di exp
i=1
2 log |i j | N V (i )
i< j
i=1
In the large N limit, if O(1), both terms in exponential are of order N 2 .

Look for the stationary point, i.e. the solution of
2
N
1
i j = V (i ) .
j6=i
To solve this problem, introduce the resolvent

1 E D1 N 1 E
1D
tr
=
G(x) =
x i .
N
xM
N i=1
()
21
Computing its square leads after some algebra to

E
1 D
1
1
2
G (x) = 2
=
G
(x)
+V
(x)G(x) P(x)
(x i )(x j )
N i, j=1,,N
N
E
D
(x)V ( )
V
i
a polynomial in x of degree d 1, i.e.
with P(x) := N1 Ni=1
xi
G2 (x) V (x)G(x) +
1
G (x) + P(x) = 0 .
N
(Beware ! Not exact for N finite!) For N large, neglect the 1/N term
quadratic equation for G(x), with yet unknown polynomial P, hence

q
1
G(x) =
V (x) V (x)2 4P(x)
2
(minus sign in front of

dictated by the requirement that for large |x|,
G(x) 1/x.)
In that large N limit, the s form a continuous distribution with density
22
() on a support S ,
S d() = 1,
and G(x) =
R 2a d ()
2a
For a purely Gaussian potential V () = 12 2 , Wigners semi-circle law:
1
() = 2 4 2 on the segment [2, 2].
For more general potentials, assume first S to be still a finite segment
[2a , 2a ], in such a way that (*) becomes
2 P.P.
Z 2a
d ()
2a
= V ()
if [2a , 2a ] .
(P.P.= principal part), expressing that, along its cut,

1
G(x i) = V () i(x)
x [2a , 2a ] .
Thus
2
p
1
G(x) = V (x) Q(x) (x + 2a )(x 2a )
2
where the coefficients of the polynomial Q(x) and a , a are determined by
the condition that G(x) 1/x for large |x|. Q is of degree d 1. The
solution is unique (under the one-cut assumption).
23
Example For the quartic potential V () = 21 2 + g4 4 , by symmetry

a = a =: a,
p
1 g 2
1
3
2
G(x) = (x + gx ) ( + x + ga ) x2 4a2
2
2 2
with a2 the solution of
3ga4 + a2 1 = 0
(EQa2 )
which goes to 1 as g 0 (a limit where we recover Wigners semi-circle

law). From this we extract
p
1 1 g 2
2
() = ( + + ga ) 4a2 2
2 2
and we may compute all invariant quantities like the free energy or the
moments
E Z
D1
tr M 2p = d 2p () .
G2p :=
N
For example G2 = (4 a2 )a2 /3, G4 = (3 a2 )a4 , etc. All these functions
24
1
of a2 are singular as functions of g at the point gc = 12
where the two
roots of (EQa2 ) coalesce. For example the genus 0 free energy

1 2
Z(g)
1
2
(0)
2
= log a (a 1)(9 a2 )
F (g) : = lim (1/N ) log
N
Z(0)
2
24
3g (2p 1)!
= ( 2 )p
[Tutte 62, BIPZ 78]
p!(p + 2)!
p=1 t
...
has a power-law singularity
F (0) (g) |g gc |5/2
ggc
which reflects on its series expansion

F
(0)
(g) =
fn gn
n=0
fn const|gc |n n7/2 .
n
Comments
i) Nature of the 1/N 2 and of the g expansions, algebraic singularity at
finite gc
ii) Universal singular behavior at gc
iii) Extension to several cuts, the role of the algebraic curve (cf Eynard).
iv) Connected correlation functions and free (or non crossing) cumulants
v) Factorization property, localization of the matrix integral and the
master field [. . .]
25
26
2. Orthogonal polynomials
Z
d Pm ()Pn ()eNV () = hn mn
[Mehta, Bessis, . . .]
cf Paul Z-J . . .
3. Loop (or Schwinger-Dyson) equations
Z
dM
{ eNtrV (M) } = 0
Mi j
and make use of factorization property . . .

cf Edouard M-S . . .
27
Comments on Free or non-crossing cumulants fn

[BIPZ 78, Cvitanovic 81, Voiculescu 85, Speicher 94, Biane, PZ-J]
Generating function of moments mn = N1 tr M n
1
1
i = jn mn
Z( j) = h tr
N 1 jM
n=0
n1 m . The generating function of free

or G(u) = u1 Z(u1 ) =
n
n=0 u
cumulants
e (z)
W (z) = 1 + zn fn = 1 + W
n=1
or P(z) = z1W (z), is defined by the relations

W (z) = Z( j(z))
with
j(z) = z/W (z)
or equivalently
Z( j) = W (z( j))
with z( j) = jZ( j)
28
Graphically [Cvitanovic]
Z( j) =
Z(j)
1 +
111111111
000000000
~
000000000
111111111
W(j)
000000000
111111111
000000000
111111111
000000000
111111111
Z
= W ( jZ( j))
...
...
These relations amount to saying that P et G are functional inverses of one

another P G(u) = u. Indeed
P(G(u)) = G1 (u)W (G(u)) = uZ 1 (u1 )W (u1 Z(u1 )) = u, since
Z(u1 ) = W (u1 Z(u1 )).
Using Lagrange formula, one computes
f1 1 f2 2
k!
mk = k (k+1 q )! 1 ! 2 !
or conversely
(k2+ q )! (m1 )1 (m2 )2

fk = k
k!
1 !
2 !
29
End of Act I
30

Zuber Talk Randommatrices

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Zuber Talk Randommatrices

Hochgeladen von

Copyright:

Verfügbare Formate

Free Probability and Large N Limit

Free Probability and Large N Limit

Introduction to Random Matrices

Random matrices: a major theme in theoretical physics since Wigner (1951);

Free Probability and Large N Limit

Free Probability and Large N Limit

A short history of the subject

Madan Lal Mehta

Free Probability and Large N Limit

Statistics of energy levels of random Hamiltonians

Level spacing histogram

[O. Bohigas et al])

Free Probability and Large N Limit

2. Large N limit of U(N) gauge theories

Free Probability and Large N Limit

Basics of Feynman diagrams

Free Probability and Large N Limit

Wick theorem also applies to monomials (n = 1 variable for simplicity):

Non Gaussian integrals (g < 0): power series perturbative expansions

p=0 graphs G with 2p lines

= connected Feynman diagrams

Free Probability and Large N Limit

Matrix Integrals: Feynman Rules

Feynman rules: propagator ij

For each connected diagram contributing

Free Probability and Large N Limit

Power of N in a connected diagram

Free Probability and Large N Limit

Thus large N limit of matrix integral DMeNtr (M + 4 M ) = generating

limN N12 log Z =

or in a dual way, of quadrangulations of 2D surfaces of genus 0

Free Probability and Large N Limit

Thus large N limit of matrix integral DMeNtr (M + 3 M ) = generating

or in a dual way, of triangulations of 2D surfaces of genus 0

Triangulated surfaces and discrete 2D gravity

Thus : Large N limit of matrix integrals Counting of planar objects :

Free Probability and Large N Limit

Large N limit of matrix integral DA DB eNtr (A +cAB+B +gA

Free Probability and Large N Limit

Large N limit of matrix integral DA DB eNtr (A +cAB+B +gA

i.e. describes Ising model on a random quadrangulated sphere

Free Probability and Large N Limit

Two remarks useful in connection with free probabilities. . .

Compare (in Gaussian theory

Free Probability and Large N Limit

Free Probability and Large N Limit

3. Double scaling limit

F (0) (g) and more generally F (h) (g) have a singularity at g = gc ,

diverges. Expect to make contact with

Free Probability and Large N Limit

4. Other physical applications

Cell decomposition of moduli space of Riemann surfaces, intersection

QCD, the Dirac operator D

Free Probability and Large N Limit

V (M) a polynomial of degree d + 1. For ex. V3 (M) = ( 21 M 2 + g3 M 3 ) and

Several ways to treat this integral: saddle point approximation ; orthogonal

Free Probability and Large N Limit

1. Saddle point approximation

In the large N limit, if O(1), both terms in exponential are of order N 2 .

To solve this problem, introduce the resolvent

Free Probability and Large N Limit

Computing its square leads after some algebra to

(minus sign in front of

Free Probability and Large N Limit