• Aucun résultat trouvé

Harmonic analysis and P-representation of generalized Gaussian fields

N/A
N/A
Protected

Academic year: 2021

Partager "Harmonic analysis and P-representation of generalized Gaussian fields"

Copied!
7
0
0

Texte intégral

(1)

HAL Id: jpa-00208142

https://hal.archives-ouvertes.fr/jpa-00208142

Submitted on 1 Jan 1974

HAL

is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire

HAL, est

destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

Harmonic analysis and P-representation of generalized Gaussian fields

M Rousseau

To cite this version:

M Rousseau. Harmonic analysis and P-representation of generalized Gaussian fields. Journal de

Physique, 1974, 35 (3), pp.193-198. �10.1051/jphys:01974003503019300�. �jpa-00208142�

(2)

LE JOURNAL DE PHYSIQUE

HARMONIC ANALYSIS AND P-REPRESENTATION OF GENERALIZED GAUSSIAN FIELDS

M. ROUSSEAU

Laboratoire d’étude des Phénomènes

Aléatoires,

Université de

Paris-Sud,

Centre

d’Orsay,

Bâtiment

210,

91405

Orsay,

France

(Reçu

le 8

février

1973, révisé le 2 octobre

1973)

Résumé. 2014 Nous établissons classiquement la distribution de probabilité (P-distribution) des amplitudes complexes de

champs

optiques multimodes et monomodes formant la classe des champs

Gaussiens généralisés. Ce résultat donne immédiatement la

représentation-P

unique pour cette classe de champs. Mais, excepté le

champ

chaotique, ils ne sont pas totalement stationnaires ; on peut néanmoins construire un champ stationnaire équivalent à chacun d’eux, qui a la même inten- sité. Nous exprimons la représentation-P de ce

champ

stationnaire équivalent dans le cas monomode.

Abstract. 2014 We establish classically the probability distribution (P-distribution) for the complex amplitudes of multimodal and unimodal

optical

fields

belonging

to the class of generalized Gaussian fields recently introduced. From this result follows immediately the unique P-representation for

this class of fields. However, all the generalized Gaussian fields, except the chaotic field, are not fully stationary. To each of them we can associate a stationary equivalent field which possesses the same

intensity.

We give the P-representation of this stationary equivalent field in the monomode case.

Classification

Physics Abstracts

1.140 - 1.650 - 2.400

1. Introduction. - The class of

generalized

Gaus-

sian fields was

recently presented [1] from

a classical

point

of view. Let us recall that these

electromagne-

tic

(e. m.)

fields are

Gaussian,

so that the chaotic

field is a

particular

element of this class.

They present

the

advantage

of

being experimentally

realizable

[2]

and possess some

interesting properties

when

they

are detected and also in information

theory [3].

In

fact,

it was established

[1] that,

among the

generalized

Gaussian

fields,

the chaotic field possesses the smallest’ °

bunching

effect

(h

=

2)

for

photoelectrons,

whereas

the real Gaussian field possesses the greatest

(h = 3) [2].

In this paper we

study

the harmonic

analysis

of

generalized gaussian

fields. We are interested in

finding ,the unique P-representation

of such fields.

So far as we

know,

this

problem

is still

unsolved,

even . for the well-known chaotic,field for which

Glauber [6]

gave a

P-representation,

each mode

being

chaotic

and

independent

of the others. But Mollow

[9] proved

later that when the

length

of the

cavity

tends to

infinity,

it is not necessary to

suppose

any

particular property

for each

mode,

the sufficient condition for the field to be chaotic is

independence

between the modes.

So,

we can ask about the

uniqueness

of the

P-representation, furthermore,

this quantum des-

cription

does not seem to be well

adapted

when the

length

of the

cavity

tends to

infinity.

To establish the

unique P-representation

of each

generalized gaussian field,

we

adopt

the artifice of

solving

the

problem classically by using

the well-

known harmonic

analysis

of

gaussian

random func- tions

[r. f.’s].

-

First of all we consider the e. m. field in free space, with a continuous

spectrum.

Such a field cannot be studied in

quantum mechanics,

unless it is considered

as the limit of a field in a

cavity ;

but what is then the

meaning

of the

P-representation ?

On the other

hand, free-space

e. m. fields can be studied classi-

cally owing

to the

theory

of r.

f., which

has been

developed

since 1944

[ 11 ].

Let us recall the most

important

result. about the

decomposition

of a

temporal

random function

E(x, t),

its harmonic

analysis.

It can be written for x =

0,

as

where

de(v)

is a differential element of a r. f.

e(v)

that

Article published online by EDP Sciences and available at http://dx.doi.org/10.1051/jphys:01974003503019300

(3)

194

can be obtained from

E(t) by

linear

filtering [4]

through

a filter that cuts out all the

frequencies greater

than v,

This new r. f.

e(v)

is very

special,

for it has no deri-

vative

(even

if

E(x, t)

is

differentiable),

unlike deter- ministic

temporal functions,

so that we can

only

consider its increments

e(v

+

Av) - e(v).

Through

linear

filtering theory,

the statistical pro-

perties

of

E(O, t)

are

equivalent

to those of

e(v).

When

E(O, t)

is

Gaussian, e(v)

is Gaussian too since any

linearly

filtered Gaussian function is still Gaus- sian. For

generalized

Gaussian fields we

specify

some’

of the

properties

of

e(v) [§§ 2 .1

and

3 . 2 .1 ].

After that we treat the discrete

spectrum

as a par- ticular case of the continuous

spectrum

for which the

frequencies

are distributed on Dirac masses.

When an e. m. field is inserted in a

cavity,

it can

be written as

where L is the

length

of the

cavity

and c the

speed

of

light.

The statistical

properties

of the classical field described

by

eq.

(3)

are all defined when we know the multidimensional distribution P

{ Cn }

of the

random variables

(r. v.) Cn ;

we call it the P-distribu- tion.

Let us show that P

{ Cn }

can be obtained from the harmonic

analysis

of the

corresponding

Gaussian

field in the free space. In

fact,

eq.

(1)

and

(3)

are

equivalent

when we set

Integrating

from

obtain

Hence the

complex amplitudes

of the different modes of an e. m. field enclosed in a

cavity

can be obtained

by

a linear

filtering

of

E(t)

since

they

are

only

linear

combinations of

e(vi).

For each

generalized

Gaussian field we set up the

unique

P-distribution

(§§ 2.2, 3.2.2, 4).

Now let us retum to thé quantum

description

of

such

fields which can be very useful for

solving

some

physical problems [3]

- for

instance,

the interac- tion of radiation with matter, the Raman

effect, multiphoton absorption

and so on... It is now well

known that every classical field can be described quantum

mechanically by taking

the

corresponding

P-distribution as its

P-representation.

Thus

by using

the harmonic

analysis,

we have established the

unique P-representation

for the set of

generalized

Gaussian fields.

In section 5 we discuss some aspects of the quantum

description

of

generalized

Gaussian

fields, especially

the chaotic fields

(§ 5.1)

and their

stationarity (§ 5.2).

Any generalized

Gaussian field other than a chaotic

one is

only quasi-stationary

and not

fully stationary,

but

by introducing

a

uniformly

distributed

phase

we

can construct a

fully stationary.

e. m. field that has the same

intensity

as the Gaussian one but which is

no

longer

Gaussian. We call it the

stationary equi-

valent

field

and we

give

its

P-representation

in the.

unimodal case.

2. The P-distribution for the chaotic field. - 2.1 CHAOTIC FIELD IN THE FREE SPACE. - We are

going

to show that the random function

e(v) appearing

in the harmonic

analysis

of an

electromagnetic

field

in free space

(eq. (2)) is,

for a chaotic

field,

a random

Gaussian function with

independent increments,

which

is

commonly

known in

probability theory

as a

Wiener-Lévy

process. In

fact,

the function

e(v)

is

obviously

Gaussian since it is filtered from

E(t)

which

is Gaussian

[4]. Therefore,

every increment

is a Gaussian r. v.

Moreover,

as the function

E(t)

is

fully stationary [7],

the second moments of

de(v) satisfy

the

equations

Let us

decompose

the

complex

increment

Ae(v)

into

its real and

imaginary parts :

From eq.

(7),

we can prove that the real increments

are

independent

and that the

complex

increments

Ae(v)

and

Ae(v’)

for two

disjoint

intervals

(Av, Av’)

are

also

independent..

.

Therefore, if

the

r. f. e(v)

is a Gaussian

r. f.

with

independent

increments, then it is a

complex

Wiener-

Lévy

process whose real and

imaginary

parts are

independent Wiener-Lévy

processes.

2.2 NECESSARY AND SUFFICIENT CONDITION FOR A FIELD ENCLOSED IN A CAVITY TO BE CHAOTIC. - Neces- sary condition. - An

electromagnetic

field enclosed in

a

cavity

can be described

by

çq.

(3),

or

by

the group

(4)

of eq.

(1), (2)

and

(5)

as well. If the function

E(t)

is

stationary ’and Gaussian,

the results

of paragraph

2 .1

are

valid,

and the

complex

r. v.

Cn

are circular inde-

pendent

Gaussian r. v.

(with uniformly

distributed

phase).

Hence their multidimensional

probability

distribution

(P-distribution)

is

where

Ck = C kx

+

iCk, and 2013

is the common

variance of the r. v.

Ckx

and

Cky.

The

sufficient

condition is evident. When we start from eq.

(9),

we can show after a brief calculation that the field has the same moments as the chaotic field.

z

3.

P-representation

of the real Gaussian field. - 3. 1 Let us recall that the real Gaussian field was

experimentally

obtained

[2] by

modulation of a laser beam

passing through

a KD* P

crystal

inserted

between two crossed

polarizers.

The

intensity

of the

emerging light,

when a

voltage x(t)

is

applied along

the

optical

axis of the

crystal,

is then

proportional

to the square of the

voltage only

for small

voltages,

Here a is a constant related to the

crystal

and

Io

is

the

intensity

of the incident laser

light.

When the

electrical source

x(t)

is a Gaussian

noise,

the

intensity

of the

light

is the square of a Gaussian r.

f.,

so that

the

electromagnetic

field

produced by

such a device

can be described

by

the

analytic signal

whose

amplitude

is a real Gaussian function. We therefore we call it a real

Gaussian field.

Let us recall

that the

analytic signal

of a chaotic field is

where

z(t)

is a

complex

Gaussian r. f. whose real and

imaginary parts

are Hilbert transforms of each other.

Let us recall too, that the conditional

probability density

of

detecting

one

photon

at time t + T, if

another one has been detected at time t is propor- tional to

which is

equal

to

3 I >

for a real Gaussian

field,

whereas it

equals only 2 I >

for a chaotic

field ;

the real Gaussian field therefore has a

bunching

effect greater than the chaotic field.

The r. f. described

by

eq.

(11)

is not

fully stationary sine x(tl) x(t 2) > =1=

0. Let us consider the

fully stationary

r. f. that has the same

intensity

as

E(t)

given by

eq.

( 11 ),

where 9 is a

uniformly

distributed

phase.

The e. m.

field

E(t)

is no

longer Gaussian ;

we shall present its P-distribution in section 5.

3.2 HARMONIC ANALYSIS OF THE REAL GAUSSIAN

FIELD. - 3 .2.1 The real Gaussian

field

in the

free

space. - Let us

decompose x(t)

into its

negative

and

positive frequency parts :

where

de(v)

is the differential increment

appearing

in the harmonic

analysis

of the real Gaussian field described

by

eq.

(11)

and 2 nvo = coo. The r. f.

e(v)

.

filtered from

E(t)

is then

Gaussian ;

moreover,

x(t)

is a real

function,

so that

Hence the differential increments are not

indepen- dent,

but

symmetrically

correlated in

pairs.

The

optical

spectrum is indeed

symmetric

since

x(t)

and

its correlation function

Fx(-r)

are real functions.

Let us suppose that

x(t)

is a

stationary

r.

f. ;

thus

From eq.

(17)

it follows that

Ae(vo

+

v) given by

eq.

(6)

satisfies

Therefore the r. v.

Ae(vo

+

v)

are circular Gaussian

r. v.

3.2.2 Field in a

cavity.

- From the last results of

paragraph 3.2,

and from eq.

(5)

it follows that the

complex amplitudes Cn

are also circular Gaussian

r. v.,

correlated, by pairs, symmetrically,

with respect

Hence their P-distribution is

given by

(5)

196

where

( n >no+m

is the variance

of Cno+m I2.

When

the real Gaussian field is

unimodal,

eq.

(20)

reduces to

where C =

Cx

+

iCy.

4. Generalized Gaussian fields. - These are defined in classical

theory [1] ] by

eq.

(12)

where

z(t)

is any

stationary

Gaussian r. f. whose correlation functions

are

rzz( 7:)

and

rzz*( 7:).

We are

going

to

give

their P-distribution in terms of these second moments.

When the

generalized

Gaussian field is enclosed in

cavity,

the

complex amplitude

of the

analytic signal E(t)

can be

expanded

as

Since

zL(t)

is a

stationary

r.

f.,

Hence,

as for the real Gaussian

field,

for any genera- lized Gaussian field

except

the chaotic

field,

the r. v.

{ Cn }

are circular and correlated

by pairs,

symme-

trically

with respect to the mean

frequency

VO.

We shall express the characteristic function of the r. v.

in terms of the Fourier transform of

TZZ(i)

and

Ti (z)

where (k, j) = (z, z*).

Identifying

the moment of

zi(t)

we obtain

where

Eq. (23)

to

(26)

lead to the

following

relations :

where we set

fn

= ocn +

if3 n.

The characteristic function of the set of Gaussian

r. v. can thus be written as

where wm is the colum vector

(um vm

u-m

v-m)

and

F.

the correlation matrix of

(xm ym X-m y -m)

The P-distribution is thus the Fourier transform of this characteristic function :

where

Vm

=

(xm ym x _ m Y - m)

and from eq.

(12)

we

conclude that

where the function on the

right

hand side is

given by

eq.

(30).

We can

verify for y

= 0

(or y

=

kx),

which cor-

responds

to the real Gaussian

field,

that eq.

(30)

and

(31)

are identical to the P-distribution of the real Gaussian field

(eq. (20)).

Moreover,

if we introduce the relations

which characterize the chaotic field

(i.

e. a,, =

/3n

=

0),

we obtain

Thus the characteristic function can be

factorized,

the modes are

independent,

and eq.

(30)

leads to the

P-distribution of the chaotic field in eq.

(9).

5. Extension to quantum mechanics. - We

explain-

ed in the introduction

why

we have established these last results from the classical

theory,

and we

have now obtained the P-distribution of the

complex

amplitudes

for

chaotic,

real

Gaussian,

and genera- lized Gaussian fields

(eq. (9), (20)

and

(30)).

If we

(6)

consider as

equivalent

two

optical

fields that have the same set of moments,

determining

the P-distri- bution in classical mechanics is the same

problem

as

finding

the

P-representation

in quantum mecha-

nics

[5], [6].

Hence eq.

(9), (20)

and

(30) give

the

P-representation

of

chaotic,

real

Gaussian,

and gene-

ralized Gaussian

fields, ( nk >

in eq.

(9) being

the

mean number of

photons

in the kth mode.

Let us now consider two further

points.

For the

chaotic field we shall

give

three

equivalent definitions,

and then we shall present a

stationary

field with the

same

intensity

as each

generalized

Gaussian field.

5. 1 THE CHAOTIC FIELD : 1 THREE EQUIVALENT DEFI- NITIONS. - We started

by defining

the chaotic field

as the

only

one that is Gaussian and

fully stationary [1] ]

and in section

2.2,

we

proved

that the modes of a

chaotic field are

necessarily independent

and Gaussian.

We shall now show that an

equivalent

definition

of this field related to the entropy concept, is the

following.

For a chaotic Gaussian field the modes are

independent

and for

given

mean numbers of

pho- tons

nk

>,

the

entropy

in each mode is minimum.

To prove that this definition is

equivalent

to the

previous

one, we can write the

density

matrix in two

différent ways, on the basis of the

photon

number

states

[6] 1 { nk } >,

or on the basis of the coherent states

[6] 1 { llk } >.

or

when the

P-representation

exists.

The coherent states are linear combinations of the

photon

number states as :

The identification of eq.

(33)

and

(35)

enables us to

obtain the matrix elements p{nk,mk} when the

P-repre-

sentation is known.

For the chaotic

field,

since the modes are

indepen- dent,

the identification is

easy and

we obtain

This

expression corresponds

to the maximum of the function

Sk

= -

Tr {

Pk

log Pk }

which characterizes the entropy in the k mode

[8].

Therefore

a

stationary

Gaussian e. m.

field

in a

cavity

is a

field

whose modes are

independent

and

for

which the entropy is maximum in each

mode ;

we can

define

in a

unique

way such a chaotic

field

in quantum

mechanics,

either

by

its

P-representation (eq. (9))

or

by

the distribution

given

in eq.

(36)

on the basis

of

the

n

photons

states.

5.2 STATIONARY EQUIVALENT FIELD. - Let us now treat the consequences of

nonstationarity

for the

generalized

Gaussian fields

3,

eq.

(11».

Any generalized

Gaussian field except the chaotic

one is

only quasi-stationary

and not

fully stationary,

i. e. second moments such

as ( E(t1) E(t2) +

are not

functions

of , (tl - t2).

We

supposed only

that the

moments E(tl) £*(t2) >

are functions of

(tl

-

t2).

We said

previously

that it is

possible

to obtain from such a

quasi-stationary

field

E(t),

a

fully stationary

field

É(t) by introducing

a uniform

phase (see

eq.

(14))

Hence every moment such

as E(t)p >

vanishes. This

new field

£(1)

has the same

intensity

as

E(t)

but is not

Gaussian

(cf. § 5.2,

eq.

(45)).

All the fields

differing

from

E(t) only by

a

phase

fàctor

belong

to the same

equivalence

class

(J)

with respect to the

intensity.

In quantum

mechanics,

the

stationarity

condition is

which

corresponds

to the classical one of invariance

of all the moments under time

translation,

i. e. the

stationarity

in the strict sense. We shall examine the

particular

case of unimodal fields.

The relation

(38)

for an unimodal e. m. field reduces

[10]

to the

following

condition on the

density

matrix

We can prove that this relation is

equivalent

to the

condition that the

P-representation

is a function

only

of the modulus r of a = r ejqJ.

The matrix elements in the basis of the

photon

states are

This

expression

vanishes for all

(n, m) only

if

P(r, ç)

is

independent

of ç because eq.

(40)

represents the Fourier coefficients of the

periodic

function

defined in

(0,

2

n).

The condition pnm = 0

V(n, m)

is

equivalent

to

Ck

= 0 for all k =1= 0 or

f«({J) = 2,

i. e.

P(a)

is

independent

of ç. Hence for an unimodal

e. m.

field,

there is an

equivalence

between the three

relations

r" -1 - a 1

.

(7)

198

Now we show that the

P-representation

of one

of the elements of this

equivalence

class

(J)

is

subject

to the condition

where

Ps(1 ri. 1)

is the

P-representation

of the statio- nary field

É(t)

in this

class,

and

P( oc 1, ç)

represents the

P-representation

in

polar

coordinates.

As a matter of

fact,

the statistical

properties

of the

intensity

of an e. m. field are described

by

the set of

moments 6’"’"

{ tb ti }

We see from eq.

(43)

that all fields whose

P-represen-

tation satisfies eq.

(42)

have the same set of moments

G n,n(j ti ti })

i. e.

they

are

equivalent

as far as the

statistical

properties

of the

intensity

are concerned.

The

only

one among them that is

fully stationary

has

a

P-representation

that is

independent

of 0 in

polar

coordinates and

depends only

of the modulus of a.

Hence each

generalized

Gaussian field that is not

fully stationary

has the same

intensity properties

as

a

fully stationary

field which we call the

stationary equivalent

field and whose

P-representation

is

given by

eq.

(42).

For

example,

the

P-representation

of a unimodal

generalized

Gaussian field is

given by

the two dimen-

sional Gaussian

probability

distribution for correlated

r. v. X = a cos ç and Y =

LJ, sin cp :

where

The

P-representation

of the

stationary equivalent field,

for (J i = Q2 = 6, can be obtained from eq.

(42) :

where dt is the modified Bessel

function of order zero.

References

[1] PICINBONO, B. and ROUSSEAU, M., Phys. Rev. A 1 (1970) 635.

[2] BENDJABALLAH, C. and PERROT, F., Opt. Commun. 3 (1971) 21.

PERROT, F., Thèse de 3e Cycle, Université de Paris XI (1971).

[3] ROUSSEAU, M., Thèse de Doctorat d’Etat, Université de Paris XI (1972).

[4] BLANC-LAPIERRE, A. et FORTET, R., Théorie des fonctions aléatoires (Masson) 1953.

[5] GLAUBER, R. J., Proceeding of the 10th session of the Scottish Universities Summer School in Physics 1969. Ed. by

S. M. Kay and A. Maitland (Academic Press).

[6] GLAUBER, R. J., Quantum Optics and Electronics (Les Houches) 1964, Ed. by C. de Witt, A. Blandin and C. Cohen-

Tannoudji (Gordon and Breach Science Publ. inc, N. Y.) 1965.

[7] We show in ref. [1] that the chaotic field is, among all the Gaussian fields, the only one which is fully stationary.

This property derives : from the fact that the moments

like E*(t1) E*(t2) > are identical to zero for all (t1, t2).

[8] LANDAU, L. D. and LIFSCHITZ, E. M., Statistical Physics (Pergamon Press), 1959, p. 25.

See also YvoN, J., Les corrélations et l’entropie en mécanique statistique classique (Dunod), 1965, p. 83.

[9] MOLLOW, B. R., Phys. Rev. 175 (1968) 1555.

[10] KANO, Y., Ann. Phys. 30 (1964) 127.

[11] RICE, S., Bell. Syst. Tech. Journal 23 (1944) 282 and 24 (1945) 46.

LOÈVE, M., C. R. Hebd. Séan. Acad. Sci. 220 (1945) 380.

Références

Documents relatifs

Utlisez la droite gradu´ ee pour calculer chaque somme ou diff´

L´evy’s construction of his “mouvement brownien fonction d’un point de la sph`ere de Riemann” in [11], we prove that to every square integrable bi-K-invariant function f : G → R

Using current implementations of the Kummer algorithm on fields whose ideal class groups are cyclic with order a power of 2, we are able to find defining polynomials for the

finite fields, digit sums, Hooley-Katz Theorem, polynomial equa- tions, Rudin-Shapiro function.1.

To use this program all one needs to know is how to write his coefficients as MAD arithmetic expressions 3 and how to operate his remote console. Quixot Dulcin

Thus, studying the statistics of the Dyck cost in a model of equispaced points is a special case of our original problem of determining the distribution of A (ω p ) (w) induced by

and their confidence intervals, Part I: The Cohen’s d family”: The degrees of freedom for paired samples designs.. The Quantitative Methods for Psychology

It is an immediate consequence of the above remarks that the Gaussian Markov fields corresponding to the hyperelliptic operators as above have local Markov property. The