Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock...

24
Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock University of Stellenbosch South Africa Zimányi 2009 Winter School on Heavy Ion Physics

Transcript of Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock...

Page 1: Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock University of Stellenbosch South Africa Zimányi 2009.

Gram-Charlier and Edgeworth expansions for nongaussian correlations in

femtoscopy

Michiel de Kock University of Stellenbosch

South Africa

Zimányi 2009 Winter School on Heavy Ion Physics

Page 2: Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock University of Stellenbosch South Africa Zimányi 2009.

Experimental Femtoscopy

][ 1),( )( 1)(23 rqrrq SdC

1)( qC

),( rq

)(rS

r

Fireball Detector

21 ppq

Relative distance distribution

Wave function

Correlation function

Position Momentum

r q

2p

1p

Fourier Transform

Identical,non-interacting particles

Page 3: Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock University of Stellenbosch South Africa Zimányi 2009.

2longlongout

2side

longout2out

12

0

00

0

qqq

q

qqq

Rij

First Approximation: Gaussian

ijjiji qRqC 2exp1)( q

• Assume Gaussian shape for correlator:

• Out, long and side

• Measuring Gaussian Radii through fitting

),,( sidelongout qqqq

Page 4: Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock University of Stellenbosch South Africa Zimányi 2009.

• Measured 3D Correlation function are not Gaussian.• The traditional approach: fitting of non-Gaussian functions.• Systematic descriptions beyond Gaussian:

Harmonics (Pratt & Danielewicz, http://arxiv.org/abs/nucl-th/0612076v1)Edgeworth and Gram-Charlier series Reference: T. Csörgő and S. Hegyi, Phys. Lett. B 489, 15 (2000).

High-Statistics Experimental Correlation functions: Not Gaussian!

)GeV/c(outq)(1 qC

Data: http://drupl.star.bnl.gov/STAR/files/starpublications/50/data.htm

STAR Au+Au 200 GeV

Page 5: Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock University of Stellenbosch South Africa Zimányi 2009.

Derivation of Gram-Charlier series

2221 ]E[][][ qqq

)(1)( qgC q• Assume one dimension,

with

• Moments:

• Cumulants:

We want to use cumulants to go beyond the Gaussian.

1)( dqqg

rrrr qdqqgqq )(][

2longlongout

2side

longout2out

12

0

00

0

qqq

q

qqq

RK ijij

Page 6: Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock University of Stellenbosch South Africa Zimányi 2009.

First four CumulantsMean Variance

Skewness Kurtosis

1 2

3 4

Page 7: Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock University of Stellenbosch South Africa Zimányi 2009.

Why Cumulants?

036

03

44224

4

323

3

22

222

11

• Cumulants are invariant under translation • Cumulants are simpler than moments

One-dimensional Gaussian:

Moments of a Gaussian Cumulants

2

)(exp

2

1)(

2qqg

cqq

Page 8: Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock University of Stellenbosch South Africa Zimányi 2009.

0

)()(

x

r

r xGixd

d

0

)(log)(

x

r

r xGixd

d222

21

413144

312133

2122

11

364

3

Generating function

dqeqgj

ixixixxG

iqx

jj

j

)(!

)()()(1)(

02

2!2

11

Moment generating function (Fourier Transform).

Cumulant generating function (Log of Fourier Transform).

Moments:

Cumulants:

1

33

!31

22

!21

1 !

)()()()()](log[

jj

j

j

ixixixixxG

Moments to Cumulants:

)(xG

Page 9: Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock University of Stellenbosch South Africa Zimányi 2009.

Reference function

)(

)(

xF

xG 33221

!3!2!11 x

cx

cx

c

Measured correlation function• Want to approximate g in terms of a reference functionGenerating functions of g and f:

Start with a Taylor expansion in the Fourier Space

)(qf

)(qg

1

*

0

*

!

)(exp

!

)()()(

jj

j

jj

jixq

j

ix

j

ixdqeqfxF

10 !

)(exp

!

)()()(

jj

j

jj

jixq

j

ix

j

ixdqeqgxG

Page 10: Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock University of Stellenbosch South Africa Zimányi 2009.

Gram-Charlier Series

)(!3

)(''!2

)('!1

)()( )3(321 qfc

qfc

qfc

qfqg

Coefficients are determined by the moments/cumulants

Useful property of Fourier transforms

Expansion in the derivatives of a reference function

)()()()()()(''

)()()(')()(3)3(2 xFixqfxFixqf

xFixqfxFxf

33221

!3!2!11

)(

)(x

cx

cx

c

xF

xG

Page 11: Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock University of Stellenbosch South Africa Zimányi 2009.

222

21

413144

312133

2122

11

364

3

c

c

c

c

Determining the Coefficients

0 !

)(log)(logj

j

j

j

xxGxF

*jjjjc

33221

!3!2!11

)(

)(x

cx

cx

c

xF

xG

Taking logs on both sides and expanding

Coefficients in terms of Cumulant Differences:

Cumulant differences to Coefficients

Page 12: Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock University of Stellenbosch South Africa Zimányi 2009.

Infinite Formal Series

Truncate series to form a partial sum, from infinity to k

How good is this approximation in practice?

)(!

)1()(0

qfdq

d

j

cqg

j

j

jj

Partial Sums

)(

)(!

)1()(0

qf

qfdq

d

j

cqg

k

jk

j

jj

Truncate to k terms

Page 13: Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock University of Stellenbosch South Africa Zimányi 2009.

We will now use analytical functions for the correlator to test the Gram-Charlier expansion.

Kurtosis

Negative Kurtosis Zero Kurtosis Positive Kurtosis

Beta Distribution Gaussian Hypersecant

Student’s t

Normal Inverse Gaussian

Gaussian

Negative kurtosis Positive kurtosisZero kurtosis

qqf vs.)(log

Page 14: Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock University of Stellenbosch South Africa Zimányi 2009.

Gram-Charlier Type A Series:Gaussian reference function

Gaussian gives Orthogonal Polynomials;Rodrigues formula for Hermite polynomials.Gram-Charlier Series is not necessarily orthogonal!

154515)(

36)(

1)(

2466

244

22

qqqqH

qqqH

qqH

)]([)()( )(1 qfqH r

dqd

qfr

])(!3

)(!2

)(!1

1)[()( 33

22

11 qH

cqH

cqH

cqfqg

2

)(exp

2

1)(

2qqf

Page 15: Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock University of Stellenbosch South Africa Zimányi 2009.

)()(!

)1()(0

qfqHj

cqf j

k

j

jjk

Negative-Kurtosis g(q)

Gaussian

q

Negative probabilitiesq

q

q

)(qf

Beta)(qg

)(6 qf

)()(6 qgqf

BetaGram-Charlier (6th order) )(qg

Page 16: Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock University of Stellenbosch South Africa Zimányi 2009.

)()(!

)1()(0

qfqHj

cqf j

k

j

jjk

Positive-kurtosis g(q)

4th Gram-Charlier

6th Gram-Charlier is worse 8th Gram-Charlier

q q

q q

Gaussian)(qf Hypersecant

)(qg

)(4 qf

)(6 qf )(8 qf

Hypersecant

)(qgHypersecant Hypersecant

)(qg

)(qg

Page 17: Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock University of Stellenbosch South Africa Zimányi 2009.

)()10(

)(

)(

)(

6236!6

1

55!51

44!41

33!31

qH

xH

qH

qH

Edgeworth Expansion

• Same series; different truncation• Assume that unknown correlator g(q) is the sum of n

variables.

1 !exp)|(

j

jjn

jn

ixn

n

xGnxG

Truncate according to order in n instead of a number of terms (Reordering of terms).

...)(3556)(

)(280)(35)(

)(10)(

)(

82453!8

16!6

16

9!913

37!71

435!51

5

6!612

34!41

4

33!31

qHqH

qHqHqH

qHqH

qH

Gram-Charlier Edgeworth

Page 18: Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock University of Stellenbosch South Africa Zimányi 2009.

Edgeworth does better

Gram-Charlier (6 terms) Edgeworth (6th order in n)

4th order arethe same

Hypersecant)(qg

Gaussian)(qf

q q

q q

)(6 qf )(6 qf

Hypersecant)(qg

Hypersecant)(qg

Hypersecant)(qg

Page 19: Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock University of Stellenbosch South Africa Zimányi 2009.

Interim Summary• Asymptotic Series• Edgeworth and Gram-Charlier have the same convergence

• Gaussian reference will not converge for positive kurtosis.• Negative kurtosis will converge, but will have negative tails.

Different reference function for different measured kurtosis

• Negative kurtosis g(q): use Beta Distribution for f(q)1. Solves negative probabilities.2. Great convergence .

• Small positive kurtosis g(q): use Edgeworth Expansion for f(q)• Large positive kurtosis g(q): use Student’s t Distribution for

f(q) and Hildebrandt polynomials, investigate further...

Page 20: Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock University of Stellenbosch South Africa Zimányi 2009.

Hildebrandt Polynomials2

1

2

2

21

1),(

1)(

m

a

q

maBqf

Orthogonal polynomials:

)4(61

)4(32

)6(454

6

212

43

24

22

246

)(

)1()2)(1(2

)3)(2)(1(

)1(2

mq

mq

mqmS

mmqmmm

qmmmmS

mqmmS

• Student’s t distribution has limited number of moments (2m-1).• Hildebrandt polynomials don’t exist for higher orders.

Student’s t distribtion:

Page 21: Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock University of Stellenbosch South Africa Zimányi 2009.

Orthogonality vs. Gram-Charlier

• Pearson family: Orthogonal and Gram-Charlier• Choose: Either Gram-Charlier(derivatives of reference)or Orthogonal Polynomials Gram-Charlier

Orthogonal Polynomials

Pearson Family 22

221)(

q

qaKexf

Normal Inverse Gaussian• Finite moments and simple cumulants• Construct polynomials or take derivatives

Page 22: Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock University of Stellenbosch South Africa Zimányi 2009.

Strategies for Positive kurtosis: ComparisonGauss-Edgeworth Hildebrandt

NIG Gram-Charlier NIG Polynomials

q

)(6 qf

q

q

Hypersecant)(qg

Hypersecant

q

Hypersecant Hypersecant

)(qg

)(qg)(qg

)(6 qf

)(6 qf)(6 qf

Page 23: Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock University of Stellenbosch South Africa Zimányi 2009.

Strategies for Positive kurtosis: Difference

)()(6 qgqf

02.0Gauss-Edgeworth Hildebrandt 004.0

001.0

Partial Sum-Hypersecant

005.0 NIG PolynomialsNIG Gram-Charlier

q

q

q

q

Page 24: Gram-Charlier and Edgeworth expansions for nongaussian correlations in femtoscopy Michiel de Kock University of Stellenbosch South Africa Zimányi 2009.

Conclusions

• The expansions are not based on fitting; this might be an advantage in higher dimensions.

• For measured distributions g(q) close to Gaussian, the Edgeworth expansion performs better than Gram-Charlier.

• For highly nongaussian distributions g(q), both series expansions fail.

• Choosing nongaussian reference functions f(q) can significantly improve description.

– Negative kurtosis g(q): use Beta distribution for f(q)

– Positive kurtosis g(q): choose reference f(q) to closely resemble g(q)

• Cumulants and Moments are only a good idea if the shape is nearly Gaussian.