Scaling functions for finite-size corrections in EVS Zoltán Rácz

20
Scaling functions for finite-size corrections in EVS Zoltán Rácz Institute for Theoretical Physics Eötvös University E-mail: [email protected] Homepage: cgl.elte.hu/~racz Collaborat ors: G. Gyorgyi N. Moloney K. Ozogany I. Janosi I. Bartos EVS looks like a finite-size scaling problem of critical phenomena – try to use the methods learned there. Finite size corrections to limiting distributions (i.i.d. variable Numerics for the EVS of signals ( Improved convergence by using the right scaling variables. Distribution of yearly maximum temperatures. ion: ion: Do witches exist if there were 2 very large hurricanes in a century? uction uction : : Extreme value statistics (EVS) for physicists in 10 minutes. Slow convergence to limiting distributions. Not much is known about the EVS of correlated variables. f / 1 0

description

Institute for Theoretical Physics Eötvös University E-mail: [email protected] Homepage: c gl.elte.hu/~racz. Scaling functions for finite-size corrections in EVS Zoltán Rácz. Collaborators: G. Gyorgyi N. Moloney K. Ozogany I. Janosi I. Bartos. - PowerPoint PPT Presentation

Transcript of Scaling functions for finite-size corrections in EVS Zoltán Rácz

Page 1: Scaling functions for finite-size corrections in EVS Zoltán Rácz

Scaling functions for finite-size corrections in EVS

Zoltán Rácz

Institute for Theoretical PhysicsEötvös UniversityE-mail: [email protected]: cgl.elte.hu/~racz

Collaborators: G. Gyorgyi N. Moloney K. Ozogany I. Janosi I. Bartos

IdeaIdea: : EVS looks like a finite-size scaling problem of critical phenomena – try to use the methods learned there.

Results: Results: Finite size corrections to limiting distributions (i.i.d. variables). Numerics for the EVS of signals ( ). Improved convergence by using the right scaling variables. Distribution of yearly maximum temperatures.

MotivaMotivation: tion: Do witches exist if there were 2 very large hurricanes in a century?

IIntroductionntroduction: : Extreme value statistics (EVS) for physicists in 10 minutes.

ProblemsProblems: : Slow convergence to limiting distributions. Not much is known about the EVS of correlated variables.

f/1 0

Page 2: Scaling functions for finite-size corrections in EVS Zoltán Rácz

Extreme value statistics

is measured: Question: Question: What is the distribution of the largest number?

)( NzP

Nz

)(0 yP

y

NN yyyz ,...,,max 21

Y Nyyy ,...,, 21

LogicsLogics::

Assume something about iy

Use limit argument: )( N

E.g. independent, identically distributed

Family of limit distributions (models) is obtained

Calibrate the family of models by the measured values of Nx

)( ii thy

AimAim:: Trying to extrapolate to values where no data exist.

Page 3: Scaling functions for finite-size corrections in EVS Zoltán Rácz

Extreme value statistics: i.i.d. variables

is measured: )(0 yP

y

)( NzP

Nz

NN yyyz ,...,,max 21

Y Nyyy ,...,, 21

)(zGN zzN probability of

)()( 0 yPdyzFz

z

limN

Question:Question: Is there a limit distribution for ?

NN zFzG )]([)(

NN bxaz

)()]([ xGbxaF NNN

)( NNN bxaG

limN

N

ResultResult:: Three possible limit distributions depending on the tail of the parent distribution, .)(0 yP

z

Page 4: Scaling functions for finite-size corrections in EVS Zoltán Rácz

Extreme value limit distributions: i.i.d. variables

)(0 yP

y0y

aye

1 y

10 )( yy

Fisher & Tippet (1928)Gnedenko (1941)

Fisher-Tippet-Gumbel (exponential tail)

))exp(exp()( xxGFTG

Fisher-Tippet-Frechet (power law tail)

0 0 0 )exp()(

xxxxGFTF

Weibull (finite cutoff)

0 1 0 ))(exp()(

xxxxGW

Characteristic shapes of probability densities:

FTGFTF W

dxxdGx II /)()(

/)( xx /)( xx/)( xx

12/5

Page 5: Scaling functions for finite-size corrections in EVS Zoltán Rácz

Gaussian signalsf/1

)(xh

0 L

x

Edwards-Wilkinson

Random walk

Random acceleration

Mullins-Herring

f/1noise

White noise

Single mode,random phase

2

~)(

k kL hk

k eh

Independent, nonidentically distributed Fourier modes

khk ~2

with singular fluctuations 12122 ~)( LhLhhwk k

425.00 1

Berman, 1964)

Majumdar-Comtet, 2004

EVS

h

Page 6: Scaling functions for finite-size corrections in EVS Zoltán Rácz

Slow convergence to the limit distribution (i.i.d., FTG class)

)(0 yP

y0y

2ye The Gaussian results are characteristic for the whole FTG class

yeyP ~)(0

0

except for

425.00 1

Page 7: Scaling functions for finite-size corrections in EVS Zoltán Rácz

Finite-size correction to the limit distribution de Haan & Resnick, 1996Gomes & de Haan, 1999

)(0 yP

y

)()( 0 yPdyzFz

NNN

N bxaFdx

d

dx

xdGNxP )]([

)(),(

Fix the position and the scale of by ),( NxP

0x 12 x Na Nb

z

, is determined.Nexpand in

substitute

...)(ln

1)(),( 1 x

NxNxP

FTG

2ye

3 1 )( 2

1 )( ) ( 2

02001

00 xaxa eexaaxaxxFTG

/60 a

...577.0

0

Page 8: Scaling functions for finite-size corrections in EVS Zoltán Rácz

Finite-size correction to the limit distribution

3 1 )( 2

1 )( ) ( 2

02001

00 xaxa eexaaxaxxFTG

For Gaussian )(0 yP

Comparison with simulations:

How universal is ? )(1 x Signature of corrections? 2)(ln N

0

Page 9: Scaling functions for finite-size corrections in EVS Zoltán Rácz

Finite-size correction: How universal is ? )(1 x

Determines universality

2~)( zzf

different (known) function

Gauss class

0

)(0 yP )(0 1~)()( zf

z

eyPdyzF

z

2ye

)(p

1-p )( 11 xx pzzf ~)( )(1 x

zzf ~)( )(1 x Exponential class

Exponential class is unstable

Szzzf ~)(

Gauss class eves for 10 s

)1ln(~)( zazzf

)exp(~)( azzzf

1a

1a Gauss class

Exponential class

Weibull, Fisher-Tippet-Frechet?!

Page 10: Scaling functions for finite-size corrections in EVS Zoltán Rácz

Maximum relative height distribution ( ) Majumdar & Comtet, 2004

)(xh

0 L

x

mhh

2

)(),( 00 xLhPh mm

Lwh Lm ~120

Connection to the PDF of the area under Brownian excursion over the unit interval

26~ xe

2/5~ xaex

mhmaximum height measured from the average height

?),( LhP m

Result: Airy distribution

Choice of scaling

0/ mm hhx

425.00 1

Page 11: Scaling functions for finite-size corrections in EVS Zoltán Rácz

Finite-size scaling :

2 Schehr & Majumdar (2005)

Solid-on-solid models:pL

ii hhKH 1 1

...)(2

1)(),( 000 x

LxLhPh mm

0/ mm hhx

Lwh Lm ~120

)(0 x

)(0 x

...

2

1100

Lhch mm

x

Page 12: Scaling functions for finite-size corrections in EVS Zoltán Rácz

Finite-size scaling : Derivation of …

2

Assumption: carries all the first order finite size correction.

)(2

1)(),( 000 x

LxLhPh mm

Lhm 0

212 !

)()(exp)(

mm

mdk

L cm

ikcyikyP

)()( 00 xhPhx mLmL

Lmh

...1)1(1

)0(11 LcLchc Lm

Cumulant generating function

...)0(2

mmm LccScaling with xhy m 0

0/ mhqk

2

)0()1(1

)0(12 !

)()(exp)( 1

mm

mdq

L cm

iqLccxiqx

)()()( 0

)1(1

0 1x

L

cxxL

Expanding in :1L

Shape relaxes faster than the position0 mh

Page 13: Scaling functions for finite-size corrections in EVS Zoltán Rácz

Finite-size scaling : Scaling with the average1

Assumption: carries all the first order finite size correction(shape relaxes faster than the position).

mhc1

212 !

)()(exp)(

mm

mdk

L cm

ikcyikyP

)()( xhPhx mLmL

Lmh

...1)1(1

)0(11 LcLchc Lm

Cumulant generating function

...)0(2

mmm Lcc

Scaling with xhy m

mhqk /

2)1(

1)0(

1

)0(

2 ][!

)()1(exp)(

1m

mm

mdq

L Lcc

c

m

iqxiqx

)]()()1[()()( 00)0(1

)1(1

0 1xxx

Lc

cxxL

Expanding in :1L

Page 14: Scaling functions for finite-size corrections in EVS Zoltán Rácz

Finite-size scaling : Scaling with the fluctuations1

Assumption: relaxes faster thanany other .

2c

212 !

)()(exp)(

mm

mdk

L cm

ikcyikyP

)()( 1 xcPx LL

2c

...)( 22)1(2

2)0(2

22 LcLchhc Lmm

Cumulant generating function

...)0(3

mmm Lcc

Scaling with xcy

1 /qk

32/)1(

2)0(

2

)0(2

2 ][!

)(

2)1(exp)(

2m

mm

mdq

L Lcc

c

m

iqqxiqx

)]()()([2

)()( 000)0(2

)1(2

0 2xxxx

Lc

cxxL

Expanding in :2L

2mc

Faster convergence

Page 15: Scaling functions for finite-size corrections in EVS Zoltán Rácz

Finite-size scaling: Comparison of scaling with and .

Much faster convergence

mh

mh

scaling

scaling

Page 16: Scaling functions for finite-size corrections in EVS Zoltán Rácz

Possible reason for the fast convergence for ( )

)(xh

0 L

x

mhh

1

12122 ~)( LhLhhwk k

mhw ~2 xwPw L )( 22

Width distributions Antal et al. (2001, 2002)

22 / wwx

Cumulants of 2w

...~1

~1

)1(

1

)1()( 2

mm

mm

L

nm

mwm L

baL

nLc

...1~12

2)1()(2

2

L

bLc w

...1~1

1)1()(1

2

L

bLc w

4 , 2 31 , LL 73 , LL

Page 17: Scaling functions for finite-size corrections in EVS Zoltán Rácz

Extreme statistics of Mullins-Herring interfaces ( )4and of random-acceleration generated paths

mm hhx / mm hhx /

)(0 x

)(0 x

/)( mm hhx/)( mm hhx

425.00 1

Page 18: Scaling functions for finite-size corrections in EVS Zoltán Rácz

Extreme statistics for large .

425.00 1

mm hhx /

)(0 x

2

~)(

k kL hk

k eh

2

1

2

k

k

1kOnly the mode remains

0 mh1km hh

||||~)(11

2

1

max kk

hhdheh kL

2

0 4exp

2)( xxx

Page 19: Scaling functions for finite-size corrections in EVS Zoltán Rácz

Skewness, kurtosis

)( maxTP

Distribution of the daily maximal temperature

0max T

12max

2max TT

Scale for comparability

Calculate skewness and kurtosis

Put it on the map

skewness s

curtosis ...1.1FTGs 4.2FTG

Reference values:

Page 20: Scaling functions for finite-size corrections in EVS Zoltán Rácz

Yearly maximum temperatures

Corrections to scalingDistribution in scaling