Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
[
Applied Mathematics
Statistics of Real Eigenvalues
in GinOE Spectra
]
Statistics of Real Eigenvalues
in GinOE Spectra
Eugene Kanzieper
Gernot Akemann
Department of Applied Mathematics
H.I.T. - Holon Institute of Technology
Holon 58102, Israel
(Brunel)
Phys. Rev. Lett. 95, 230501 (2005)
arXiv: math-ph/0703019 (J. Stat. Phys.)
Alexei Borodin
(Caltech)
in preparation
Snowbird Conference on Random Matrix Theory & Integrable Systems, June 25, 2007
42
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» The Problem
2005
2007
What is the probabilityStatistics
that an n × of
n random real matrix
with Gaussian i.i.d. entries has exactly k real eigenvalues?
Spectra
A.Complex
Edelman (mid-nineties)
41
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Outline
Ginibre’s random matrices
• Definitions & physics applications
Ginibre’s real random matrices (GinOE)
• Overview of major developments since 1965
• Real vs complex eigenvalues: What is (un)known ?
• Probability to find exactly k real eigenvalues and
inapplicability of the Dyson integration theorem
• Pfaffian integration theorem
Conclusions & What is next ?
40
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
39
» Ginibre’s random matrices: also physics
P
(H)DH
N / 2
exp tr HH dHij( q )
GinOE : H R N N ( 1) GOE
GinSE : H QN N ( 4) GSE
GinUE : H CN N ( 2) GUE
N
q 1 i , j 1
1965
success
complexity
( )
N
2
Dropped Hermiticity…
Statistics of
Is
there any
Complex
Spectra
physics
?
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Ginibre’s random matrices: also physics
Is there any
physics
?
38
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
» Ginibre’s random matrices: also physics
• Dissipative quantum chaos (Grobe and Haake 1989)
• Dynamics of neural networks (Sompolinsky et al 1988, Timme et al 2002, 2004)
• Disordered systems with a direction (Efetov 1997)
• QCD at a nonzero chemical potential (Stephanov 1996)
• Integrable structure of conformal maps (Mineev-Weinstein et al 2000)
• Interface dynamics at classical and quantum scales (Agam et al 2002)
• Time series analysis of the brain auditory response (Kwapien et al 2000)
• More to come: Financial correlations in stock markets (Kwapien et al 2006)
]
37
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Ginibre’s random matrices: also physics
directed chaos
<< 1
?
Is there any
HL HS h g H A
physics
GinOE model
~1
36
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
35
» Ginibre’s random matrices: also physics
Asymmetric L-R
Cross-Correlation
Matrices
Universal noise dressing
is still there !
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Ginibre’s random matrices: also physics
• Dissipative quantum chaos (Grobe and Haake 1989)
• Dynamics of neural networks (Sompolinsky et al 1988, Timme et al 2002, 2004)
• Disordered systems with a direction (Efetov 1997)
• QCD at a nonzero chemical potential (Stephanov 1996)
• Integrable structure of conformal maps (Mineev-Weinstein et al 2000)
• Interface dynamics at classical and quantum scales (Agam et al 2002)
• Time series analysis of the brain auditory response (Kwapien et al 2000)
• More to come: Financial correlations in stock markets (Kwapien et al 2006)
Back to
1965 and Ginibre’s maths curiosity…
34
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Outline reminder
Ginibre’s random matrices
• Definitions & physics applications
Ginibre’s real random matrices (GinOE)
• Overview of major developments since 1965
• Real vs complex eigenvalues: What is (un)known ?
• Probability to find exactly k real eigenvalues and
inapplicability of the Dyson integration theorem
• Pfaffian integration theorem
Conclusions & What is next ?
33
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Spectra of Ginibre’s random matrices
( )
N
P
(H)DH
(almost) uniform
distribution
N / 2
2
exp tr HH dHij( q )
N
q 1 i , j 1
depletion from
real axis
1965
accumulation along
real axis
32
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Spectra of Ginibre’s random matrices
1965
(almost) uniform
distribution
depletion from
real axis
accumulation along
real axis
31
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Spectra of Ginibre’s random matrices
N
PN z1 ,..., z N C2 ( N ) zk1 zk2
k1 k2
2 N
zk zk
e
1965
k 1
GinUE : jpdf + correlations
(almost) uniform
distribution
30
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Spectra of Ginibre’s random matrices
N
PN z1 ,..., zN C4 ( N ) zk1 zk2 zk1 zk2
k1 k2
2
2 N
zk zk
z
z
e
k k
k 1
2
1965
Mehta, Srivastava 1966
GinUE : jpdf + correlations
GinSE : jpdf + correlations
depletion from
real axis
29
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Spectra of Ginibre’s random matrices
N
PH T ( N / N ) ( w 1 ,, w N ) j1 j2
j1 j2
N
e
j 1
j2 / 2
1965
Mehta, Srivastava 1966
GinUE : jpdf + correlations
GinSE : jpdf + correlations
accumulation along
real axis
28
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Spectra of Ginibre’s random matrices
NN
N
k j10
1 j2
j 1
PHHTT((NN/)N()w( w
, , w )
PH Tj1( N/ k) (j2w 1
,e, w N )
1 ,1 , w N N)
j2 / 2
?
1965
Key Feature H T (N )
T (N )
…
N
accumulation along
T ( Nreal
) axis T ( N
k 10
/ k)
T ( N / 0) ... T ( N / k ) ... T ( N / N )
number of real eigenvalues
27
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Outline reminder
Ginibre’s random matrices
• Definitions & physics applications
Ginibre’s real random matrices (GinOE)
• Overview of major developments since 1965
• Real vs complex eigenvalues: What is (un)known ?
• Probability to find exactly k real eigenvalues and
inapplicability of the Dyson integration theorem
• Pfaffian integration theorem
Conclusions & What is next ?
26
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
25
]
» Overview of major developments since 1965
Ginibre
1965
Lehmann & Sommers
Edelman, Kostlan & Shub
Edelman
1994
1997
1991
quarter of a century !!
NN
2n
1
1
E
[
k
]
k
p
1
O
) , wPNH,kT{( N /,
PHT ( N / N ) ({wP})
H T ( N ) ( w 1 , , w N
k ) ( w 1 , n, w N )}
w w 1k,
0
N
1 , k , z 1 , z 1 ,, z l , z l2
k 0
1
k real eigenvalues l pairs of c.c. eigenvalues
Correlation
Functions ?!
N
PH T ( N / k ) ({w}) C N , k wi w j
i j
N w 2
e j erfc w j w j
2
j 1
1/ 2
[
Applied Mathematics
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Overview of major developments since 1965
Ginibre
Lehmann & Sommers
1965
Edelman, Kostlan & Shub
Edelman
1994
1997
1991
quarter of a century !!
N
PH T ( N ) ( w 1 ,, w N ) PH T ( N / k ) ( w 1 ,, w N )
k 0
1
Correlation
Functions ?!
Borodin & Sinclair, arXiv: 0706.2670
Forrester & Nagao, arXiv: 0706.2020
Sommers, arXiv: 0706.1671
detailed k-th partial correlation functions are not available…
24
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Outline reminder
Ginibre’s random matrices
• Definitions & physics applications
Ginibre’s real random matrices (GinOE)
• Overview of major developments since 1965
• Real vs complex eigenvalues: What is (un)known ?
• Probability to find exactly k real eigenvalues and
inapplicability of the Dyson integration theorem
• Pfaffian integration theorem
Conclusions & What is next ?
23
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Real vs complex eigenvalues
Edelman
N
p N ,k dw j PH T ( N / k ) ({w})
1997
j 1
Probability to have all eigenvalues real
pN ,N 2 N ( N 1) / 4
(the smallest one)
Theorem
pN ,k rN ,k sN ,k 2
( rN ,k
& s N ,k
rational)
22
Applied Mathematics
[
]
Statistics of Real Eigenvalues
in GinOE Spectra
» Real vs complex eigenvalues
p N ,k
l
2
2
cN ,k k
zp zp
z p z p ( zEdelman
N
j2 / 2
2
p zp ) / 2
d Z p
erfc
e
d j e
k!l ! pj 1
2
i
i
2
1
({w})
dw
Im{
Z }0 pP
N ,k
k
j 1
i j
i j
N k 2l
j
l
z
pq
p
H T ( N / k )
zq
2
z p zq
2
k
1997
l
j 1 p 1
j
z p j z p
wp w 1,
,2
wNN(N{1) 1/ ,4, k , z 1, z 1,, z l , z l }
pN,k
rN ,k
N ,N
k real eigenvalues l pairs of c.c. eigenvalues
Solved ?..
s N ,k 2
+
21
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Real vs complex eigenvalues
MATHEMATICA
code up to
N 9
No Closed
Formula for
p N ,k
pN , N 2
N ( N 1) / 4
20
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Outline reminder
Ginibre’s random matrices
• Definitions & physics applications
Ginibre’s real random matrices (GinOE)
• Overview of major developments since 1965
• Real vs complex eigenvalues: What is (un)known ?
• Probability to find exactly k real eigenvalues and
inapplicability of the Dyson integration theorem
• Pfaffian integration theorem
Conclusions & What is next ?
19
[
Applied Mathematics
Statistics of Real Eigenvalues
in GinOE Spectra
]
18
» Probability to find exactly k real eigenvalues
The Answer
a probability to have all eigenvalues real
pN ,k pN , N Fl ( x1 ,, xl )
N k 2l
universal multivariate polynomials
j
1 l 2xl j z p z p x j z ptr( 0z,p[ N / 2](z1) zˆ) / 2j
c N ,k
l / 2
Fl p(Nx,k1 ,, xl ) ( d1)j e d Z p
erfc
e
! 1 l j 2i
k!l ! j 1
{l } jIm{
1 Z
i 2
}0j p
k
p (l )
g
2
j
2
p
2
p
a nonuniversal ingredient
k
integer partitions
i j
i j
l
z
pq
p
zq
Even Better
l l 1 1 , l 2 2 ,, l gGg N ( z )
Starting point
2
z p zq
[ N / 2]
2
1
Fl l ( x1 ,...,
x
)
l zZ(1l ) ( x1 ,..., xl )
z
j
p
j l !p
j 1 p 1
k
l
z
p N , N det
1̂ z ˆ
polynomials
pN , N 2l zonal
l 0
Jack polynomials at α=2
Applied Mathematics
[
17
]
Statistics of Real Eigenvalues
in GinOE Spectra
» Probability to find exactly k real eigenvalues
0.01
0.0001
1.
10 6
1.
10 8
1.
10 10
0
2
4
6
8
10
12
8
10
12
0.4
0.3
0.2
0.1
0
0
2
4
6
No visible discrepancies with numeric simulations over 10 orders of magnitude !!
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
16
» Sketch of derivation: I. Integrating out j ‘ s
cancellation
Reduced
integral representation
({z, z})
2l
1
l
pf K N ({z, z})
z p z p ( z p2 z p2 ) / 2
2l 22l
( k kK
) characteristic
GOE pf
polynomial
2l ({pzN,,kz})
e
A N ,l d Z p erfc
N ({ z , z }) 2l 2l
i 2
Im{ Z } 0 p 1
l
det z
Nagao-Nishigaki (2001), Borodin-Strahov (2005)
j 1
Starting point
p N ,k
j
Oˆ det z j Oˆ
l
cN ,k k
zp zp
z p z p ( z p2 z p2 ) / 2
j2 / 2
2
d
e
d
Z
erfc
e
j
p
k!l ! j 1
2i
i 2
Im{ Z }0 p 1
k
i j
i j
l
z
pq
p
zq
2
z p zq
2
k
l
j 1 p 1
j
z p j z p
Oˆ GOE( k k )
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
15
» Sketch of derivation: I. Integrating out j ‘ s
Reduced integral representation
pN ,k A N ,l
S I
K
T
D S
z p z p ( z p2 z p2 ) / 2
e
d Z p erfc
pf K N ({z, z}) 2l 2l
i 2
Im{ Z } 0 p 1
l
2
D –part of a GOE 2 2 matrix kernel
GOE skew-orthogonal polynomials
not a projection
operator !
Dyson
Integration
Theorem
Inapplicable !!
How do we calculate the integral ?..
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Outline reminder
Ginibre’s random matrices
• Definitions & physics applications
Ginibre’s real random matrices (GinOE)
• Overview of major developments since 1965
• Real vs complex eigenvalues: What is (un)known ?
• Probability to find exactly k real eigenvalues and
inapplicability of the Dyson integration theorem
• Pfaffian integration theorem
Conclusions & What is next ?
14
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Sketch of derivation: II. Pfaffian integration theorem
Two fairly
compact proofs
13
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Sketch of derivation: II. Pfaffian integration theorem
p N ,k A N ,l
z p z p ( z p2 z p2 ) / 2
e
d Z p erfc
pf K N ({z, z}) 2l 2l
i 2
Im{ Z } 0 p 1
l
2
12
[
Applied Mathematics
Statistics of Real Eigenvalues
in GinOE Spectra
11
]
» Sketch of derivation: II. Pfaffian integration theorem
pN ,k
pN , N
l!
a probability to have all eigenvalues real
Z (1l ) ( x1 ,, xl )
N k 2l
pN , N 2 N ( N 1) / 4
Zonal polynomials
GN ( z )
[ N / 2]
z
l 0
l
pN , N 2l pN , N det 1̂ z ˆ
x j tr (0, [ N / 2]1) ˆ
j
a nonuniversal ingredient
Solved !!
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Sketch of derivation: II. Pfaffian integration theorem
10
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
09
» Sketch of derivation: II. Pfaffian integration theorem
Fredholm Pfaffian (Rains 2000)
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Sketch of derivation: II. Pfaffian integration theorem
08
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Sketch of derivation: II. Pfaffian integration theorem
07
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Sketch of derivation: II. Pfaffian integration theorem
06
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Sketch of derivation: II. Pfaffian integration theorem
05
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Outline reminder
Ginibre’s random matrices
• Definitions & physics applications
Ginibre’s real random matrices (GinOE)
• Overview of major developments since 1965
• Real vs complex eigenvalues: What is (un)known ?
• Probability to find exactly k real eigenvalues and
inapplicability of the Dyson integration theorem
• Pfaffian integration theorem
Conclusions & What is next ?
04
[
Applied Mathematics
Statistics of Real Eigenvalues
in GinOE Spectra
]
» Conclusions
Statistics of real eigenvalues in GinOE
Exact formula for the distribution of the number k of
real eigenvalues in the spectrum of n × n random
Gaussian real (asymmetric) matrix
Solution highlights a link between integrable structure
of GinOE and the theory of symmetric functions
Even simpler solution is found for the entire generating
function of the distribution of k
0.01
0.0001
Pfaffian Integration Theorem as an extension of the
1.
10 6
1.
10 8
Dyson Theorem (far beyond the present context)
1.
10 10
0
2
4
6
8
10
12
03
Applied Mathematics
[
Statistics of Real Eigenvalues
in GinOE Spectra
]
» What is next ?
Looking for specific
physical applications
(weak non-Hermiticity)
!
directed chaos
<< 1
HL HS h g H A
GinOE model
~1
?
Asymptotic analysis of the distribution of k (matrix size n
taken to infinity)
work in progress
Asymptotic analysis of the distribution
of k (when k scales with E[k] and the matrix size n that is
taken to infinity)
Further extension of the Pfaffian integration theorem to
determine all partial correlation functions
02
[
Applied Mathematics
Statistics of Real Eigenvalues
in GinOE Spectra
]
Statistics of Real Eigenvalues
in GinOE Spectra
Eugene Kanzieper
Gernot Akemann
Department of Applied Mathematics
H.I.T. - Holon Institute of Technology
Holon 58102, Israel
(Brunel)
Phys. Rev. Lett. 95, 230501 (2005)
arXiv: math-ph/0703019 (J. Stat. Phys.)
Alexei Borodin
(Caltech)
in preparation
Snowbird Conference on Random Matrix Theory & Integrable Systems, June 25, 2007
01