Towards a large deviation theory for statistical mechanical complex systems G. Ruiz López G. Ruiz López 1,2 1,2 , C. Tsallis , C. Tsallis 1,3 1,3 1 Centro Brasileiro de Pesquisas Fisicas. Brazil Centro Brasileiro de Pesquisas Fisicas. Brazil 2 Universidad Politécnica de Madrid. Spain. Universidad Politécnica de Madrid. Spain. 3 Santa Fe Institute, USA Santa Fe Institute, USA
26
Embed
Towards a large deviation theory for statistical mechanical complex systems
Towards a large deviation theory for statistical mechanical complex systems. 1 Centro Brasileiro de Pesquisas Fisicas . Brazil 2 Universidad Politécnica de Madrid. Spain . 3 Santa Fe Institute , USA. G. Ruiz López 1,2 , C. Tsallis 1,3. - PowerPoint PPT Presentation
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Towards a large deviation theory
for statistical mechanical
complex systemsG. Ruiz LópezG. Ruiz López1,21,2, C. Tsallis, C. Tsallis1,31,3
11Centro Brasileiro de Pesquisas Fisicas. BrazilCentro Brasileiro de Pesquisas Fisicas. Brazil22Universidad Politécnica de Madrid. Spain.Universidad Politécnica de Madrid. Spain.
33Santa Fe Institute, USASanta Fe Institute, USA
Towards a large deviation theory
for statistical mechanical
complex systemsG. Ruiz LópezG. Ruiz López1,21,2, C. Tsallis, C. Tsallis1,31,3
11Centro Brasileiro de Pesquisas Fisicas. BrazilCentro Brasileiro de Pesquisas Fisicas. Brazil22Universidad Politécnica de Madrid. Spain.Universidad Politécnica de Madrid. Spain.
33Santa Fe Institute, USASanta Fe Institute, USA
Large deviation theory and Statistical Large deviation theory and Statistical MechanicsMechanics
Rare events:
Tails of probability distributions
Rates of convergence to equilibrium
BG: lies on LDT NEXT: ¿ q-LDT ?
Large deviation theory and Statistical Large deviation theory and Statistical MechanicsMechanics
G. Ruiz & C. Tsallis, Phys. Lett .A 376 (2012) 2451-2454.
G. Ruiz & C. Tsallis, Phys. Lett. A 377 (2013) 491-495.
Physical scenario of a possible LDT Physical scenario of a possible LDT generalizationgeneralization
a) Standard many-body Hamiltonian system in thermal equilibrium a) Standard many-body Hamiltonian system in thermal equilibrium ((TT))
LDT standard results: LDT standard results: NN uncorrelated coinsuncorrelated coins
a) Independent random variables2
2
( )
21
2
( )x
p x e
N ��������������Standard CLT
:
Rate function and relative entropyRate function and relative entropy
( )p n
(0)
11
lnW
ii
i i
pI p
p
11
1ln ln
W
ii i
I W pp
11 ln
B
SI W
k 1
1
lnW
B i ii
S k p p
Relative entropy:(0) 1
ipW
N uncorrelated coins (W=2, p1=x, p2=1-x): 1 1(x)= ln 2 ln (1 ) ln(1 ) (x)I x x x x r
q-Generalized relative entropy:(0) 1
ipW(0)
1
lnW
iq i q
i i
pI p
p
1 1
1ln
1
Wqi
q iq
pW W
q
1 ln qqq q
B
SI W W
k
1
1
1
Wqi
iq B
pS k
q
C. Tsallis, Phys. Rev. E 58 (1998) 1442-1445.
:q b) Strongly correlated random variables
N ��������������q-CLT 2( )( ) x
q qp x Ae ( )p n
S.Umarov, C. Tsallis, S. Steinberg, Milan J. Math. 76 (2008) 307.S. Umarov, C. Tsallis, M. Gell-Mann, S. Steinberg, J. Math. Phys. 51 (2010) 033502.
Non-BG: Non-BG: NN strongly correlated coins strongly correlated coins
2
2( )y
q
yq
ep y
e dy
,( ) N nN
NP y p
(0 1)
N
��������������
Histograms:
A. Rodriguez, V. Schwammle, C. Tsallis, J. Stat. Mech (2008)P09006.
Discretization:
Suport:
,,
,0
2 1 1
0 2( )
( )
( ) [1 ( 1) ]
Q N nN n N
Q N nn
QQ
Qp y
pp y
p z Q z
( 1) ( 0;0 1)N N
,
1,
2 2 2N N
N n N
ny
N
( 0,..., )n N
Average number of heads per toss :
:
Large deviations in Large deviations in (Q, (Q, )-)-modelmodel
,,
: : ,0
( );
( )
Q N nN n N
n nn x n x Q N nN N
n
p ynP N x p
N p y
(0 1/ 2)x
( , ), Q
Large deviations in Large deviations in (Q, (Q, )-)-modelmodelAverage number of heads per toss :
:
,,
: : ,0
( );
( )
Q N nN n N
n nn x n x Q N nN N
n
p ynP N x p
N p y
(0 1/ 2)x
( , ), Q
11 (0 1/ 2; )
(3 )
Qq x
Q
00 1/ 2
12x
x
qq
1 1:Q q
q 1
1( ) ( )
; ;uncorr
QNr x Nr x
q
n n
N NN x N xPP e e
1 1:Q q
1/( 1); 1/ qnP N x N
N
Large Deviation Principle in Large Deviation Principle in (Q, (Q, )-)-modelmodel
Average number of heads per toss :
:
,,
: : ,0
( );
( )
Q N nN n N
n nn x n x Q N nN N
n
p ynP N x p
N p y
(0 1/ 2)x
( , ), Q
( )(0 1 / 2)
( , ) :
; qN r x
q x
q q Q
nP N x e
N
Generalized q-rate function:
•
•
• What about q-generalized relative entropy?
1/
1/N
( ; ; ; ) ( ; ; ;1) ( , ), 0
q q
N
r x Q r x Q Q
N
0(0) lim ( ) q q
xr r x
( 2)1 11
ln [1 2 [ (1 ) ]]1
Wqq q q q
q qB
SI W W x x
k q
( ) ( ) q qr x I x
Large Deviation Principle in Large Deviation Principle in (Q, (Q, )-)-modelmodel
11 (0 1/ 2; )
(3 )
Qq x
Q
00 1/ 2
12x
x
qq
Average number of heads per toss :
:
,,
: : ,0
( );
( )
Q N nN n N
n nn x n x Q N nN N
n
p ynP N x p
N p y
(0 1/ 2)x
( , ), Q
( )(0 1 / 2)
( , ) :
; qN r x
q x
q q Q
nP N x e
N
;( ) ( ) 1 (3 )
1 ( ) 0; ( ) 0 , were 01 1
n
NN x
B x C x QP B x C x
N N q Q
; / ( ) 0 / lim 1 ( )
( )N
N n N xP Nx B x N C x
B x
Large Deviation Principle in Large Deviation Principle in (Q, (Q, )-)-modelmodel
max1,2,..., ( )
! ( ) ( , , )
j j j x
B x xQ
C
maxj ( 0.1) =10max
j ( 0.15) = 20
4.57660794B
Asymptotic numerical behavior
max1,..., ( )
;( )( ) 1 (3 )
1 ! ( ) 0; 0 where 01 1
jj j j x
n
NN x
C xB x QP B x C
N N q Q
Numericaly known calculation ( )( ) , ( )jj qa x r x
( )
max
( )
1,...,( )
jqr x N
j qj j
a x e
Large Deviation Principle in Large Deviation Principle in (Q, (Q, )-)-modelmodel
Asymptotic expansion of q-exponential :
( , , ) (3 / 2,1/ 2,1)Q
Bounding numerical results:
Large Deviation Principle in Large Deviation Principle in (Q, (Q, )-)-modelmodel
( ) ( )( ) ; / ( )lower upperq qr x N r x Nlower upper
q q q qa x e P N n N x a x e
( )2 ( )
1( ) ( ) 1
1( )
( 1) ( )
( ) ( ) ( 1) ( )
lower upperq lower upper
lower upper lower upper qq q
r xq C x
a x B x q r x
Bounding numerical results:
Large Deviation Principle in Large Deviation Principle in (Q, (Q, )-)-modelmodel
( ) ( )( ) ; / ( )lower upperq qr x N r x Nlower upper
q q q qa x e P N n N x a x e
( )2 ( )
1( ) ( ) 1
1( )
( 1) ( )
( ) ( ) ( 1) ( )
lower upperq lower upper
lower upper lower upper qq q
r xq C x
a x B x q r x
4.69lowerC
18.75upperC
Large Deviation Principle in Large Deviation Principle in (Q, (Q, )-)-modelmodel
For all strongly correlated systems which have Q-Gaussians (Q>1) as attractors in the sense of the central limit theorem, a model-dependent set [q>1, B(x)>0,rq
(low)(x)>0, rq(up)(x)>0] might exists such that
P(N;n/N<x) satisfies these inequalities:
1 12 21 1
( ) 1 1 1 ( ) 1 1 11 ; / 1
( 1) ( ) ( 1) ( )lower upperq qq q
B x B xo P N n N x o
q r x N N q r x N NN N
4.69lowerC
18.75upperC
( , )x N
Large Deviation Principle in Large Deviation Principle in (Q, (Q, )-)-modelmodel
For all strongly correlated systems which have Q-Gaussians (Q>1) as attractors in the sense of the central limit theorem, a model-dependent set [q>1, B(x)>0,rq
(low)(x)>0, rq(up)(x)>0] might exists such that
P(N;n/N<x) satisfies these inequalities:
1 12 21 1
( ) 1 1 1 ( ) 1 1 11 ; / 1
( 1) ( ) ( 1) ( )lower upperq qq q
B x B xo P N n N x o
q r x N N q r x N NN N
( , )x N
ConclusionsConclusions We address a family of models of strongly correlated variables of a certain class We address a family of models of strongly correlated variables of a certain class
whose attractors, in the probability space, are whose attractors, in the probability space, are QQ-Gaussians (-Gaussians (Q>1Q>1). They ). They illustrate how the classical Large Deviation Theory can be generalized. illustrate how the classical Large Deviation Theory can be generalized.
We conjecture that for all strongly correlated systems that have We conjecture that for all strongly correlated systems that have QQ-Gaussians -Gaussians ((Q>1Q>1) as attractors in the sense of the central limit theorem, a model-dependent ) as attractors in the sense of the central limit theorem, a model-dependent set [set [q>1q>1, , B(x)>0B(x)>0,,rrqq
(low)(low)(x)>0(x)>0, , rrqq(up)(up)(x)>0(x)>0] might exists such that] might exists such that P(N;n/N<x) P(N;n/N<x)
satisfies:satisfies:
The argument of the The argument of the qq-logarithmic decay of large deviations remains -logarithmic decay of large deviations remains extensiveextensive in in our model. This reinforces the fact that, according to NEXT for a wide class of our model. This reinforces the fact that, according to NEXT for a wide class of systems whose elements are strongly correlated, a value of index systems whose elements are strongly correlated, a value of index qq exists such exists such thar thar SSqq preserves extensivity. preserves extensivity.
Our models open the door to a Our models open the door to a qq-generalization of virtually many of the classical -generalization of virtually many of the classical results of the theory of large deviations.results of the theory of large deviations.
The present results do suggest the mathematical basis for the ubiquity of The present results do suggest the mathematical basis for the ubiquity of qq--exponential energy distributions in nature. exponential energy distributions in nature.
( ) ( ), ( ), ( ) ; / , ( ), ( )lower upperq qr x N r x Nlower upper
j q j qa q B x r x e P N n N x a q B x r x e
Kaniadakis’ Kaniadakis’ logarithm and logarithm and --exponentialexponential