webusers.imj-prg.frwebusers.imj-prg.fr/~eric.toubiana/Barbaresco-Final_Public_Release.…webusers.imj-prg.fr...

188
Thales Air Systems Date

Transcript of webusers.imj-prg.frwebusers.imj-prg.fr/~eric.toubiana/Barbaresco-Final_Public_Release.…webusers.imj-prg.fr...

����������� �����

Thal

es A

ir S

yste

ms

Dat

e��������

��� ��

����� �

������ �

���������

��� �

����� ��

����

2/

2/

Thal

es A

ir S

yste

ms

Dat

e

���

��� ��

������

���������� ������

Take

shi S

AS

AK

IW

. BLA

SC

HK

EC

alya

mpu

di R

. RA

ON

ikol

ai N

. CH

ENTS

OV

Hiro

hiko

SH

IMA

Jean

-Lou

is K

OS

ZUL

Von

Thom

as F

RIE

DR

ICH

Y. S

HIS

HID

O

Hom

ogen

eous

Con

vex

Con

esG

.E

.B. V

INB

ER

GJe

an-L

ouis

KO

SZU

L

Hom

ogen

eous

Sym

met

ricB

ound

edD

omai

nsG

.

Elie

CA

RTA

NC

arl L

udw

ig S

IEG

EL

Prob

abili

tyin

M

etric

Spac

eM

auric

e R

. FR

EC

HE

T

Info

rmat

ion

Theo

ryN

icol

as .L

. BR

ILLO

UIN

Cla

ude.

SH

AN

NO

N

Prob

abili

tyon

R

iem

anni

anM

anifo

ldM

iche

l EM

ER

YM

arc

ARN

AUD

ON

Geo

met

ricSc

ienc

e of

In

form

atio

n

KO

SZU

L-VI

NB

ERG

M

ETR

IC(K

OSZ

UL-

VIN

BER

G

CH

AR

AC

TER

ISTI

C

FUN

CTI

ON

)

FISH

ER M

ETR

IC

Prob

abili

ty/G

. on

str

uctu

res

Y. O

LLIV

IER

M. G

RO

MO

V

3/

3/

Thal

es A

ir S

yste

ms

Dat

e

� ��

������������� ���� �����

���

�!��� �� �"� #���

�����

� �$

�� ��

Y. O

llivi

er, «

Prob

abili

tés

sur l

es e

spac

es d

e co

nfig

urat

ion

d'or

igin

e gé

omét

rique

», P

hDw

ithad

vise

rsM

. Gro

mov

& P

. Pa

nsu

http

://w

ww

.yan

n-ol

livie

r.org

/rech

/pub

ls/th

ese.

pdf

Y. O

llivi

er, «

Le H

asar

d et

la C

ourb

ure

(Ran

dom

ness

and

Cur

vatu

re)»

, hab

ilita

tion:

http

://w

ww

.yan

n-ol

livie

r.org

/rech

/pub

ls/h

dr_i

ntro

.pdf

Y. O

llivi

er &

You

heiA

kim

oto,

«O

bjec

tive

impr

ovem

enti

n in

form

atio

n-ge

omet

ricop

timiz

atio

n»,

FO

GA

2013

, pre

prin

t20

12ht

tp://

ww

w.y

ann-

olliv

ier.o

rg/re

ch/p

ubls

/dee

ptra

in.p

dfYa

nn O

llivi

er, L

udov

ic A

rnol

d, A

nne

Auge

r, an

d N

ikol

aus

Han

sen,

«In

form

atio

n-ge

omet

ricop

timiz

atio

n: A

uni

fyin

gpi

ctur

evi

a in

varia

nce

prin

cipl

es»,

arX

iv:1

106.

3708

v1, 2

011

http

://w

ww

.yan

n-ol

livie

r.org

/rech

/pub

ls/q

uant

ile_i

go.p

dfVi

deo

«se

min

aire

Bril

loui

n»:

http

://ar

chip

rod-

exte

rne.

ircam

.fr/v

ideo

/VI0

2023

900-

226.

mp4

Yann

Olli

vier

, Par

is-S

ud U

nive

rsity

, LR

I Dep

t.C

NR

S B

ronz

e M

edal

2011

Intr

oduc

tion

of p

roba

bilis

tic m

odel

s on

str

uctu

red

obje

cts

Inte

rpla

y be

twee

n pr

obab

ility

and

diff

eren

tial g

eom

etry

Nat

ural

Gra

dien

t by

Fish

er In

form

atio

n M

atrix

(IG

O)

Mis

ha G

rom

ov, I

HES

Abe

l Priz

e20

09M

athe

mat

ics

abou

t "in

tere

stin

g st

ruct

ures

“C

ateg

ory-

theo

retic

appr

oach

of F

ishe

r Met

ricM

. Gro

mov

, « In

a S

earc

h fo

r a S

truc

ture

, Par

t 1: O

n En

trop

y»,

pre

prin

t, Ju

ly 2

012

http

://w

ww

.ihes

.fr/~

grom

ov/P

DF/

stru

ctre

-ser

ch-

entr

opy-

july

5-20

12.p

dfC

hap

2. «

Fish

er M

etric

and

Von

Neu

man

n En

trop

y»Su

ch a

resc

alin

g, b

eing

a n

on-tr

ivia

l sym

met

ry, i

s a

sign

ifica

nt s

truc

ture

in it

s ow

n rig

ht; f

or e

xam

ple,

th

e gr

oup

of fa

mili

es o

f suc

h "r

esca

lings

" le

ads

the

amaz

ing

orth

ogon

al s

ymm

etry

of t

he F

ishe

r met

ric

M. G

rom

ov, «

Con

vex

sets

and

Käh

ler m

anifo

lds»

,in

Adv

ance

s in

J. D

iffer

entia

l Geo

m.,

F. T

ricer

ri ed

., W

orld

Sci

., Si

ngap

ore,

pp.

1-3

8, 1

990

http

://w

ww

.ihes

.fr/~

grom

ov/P

DF/

%5B

68%

5D.p

df

[]�

=+

=

�� ��

�� ��

∂∂

∂∂

=�� ��

�� ��

∂∂

∂−

=

ji

ji

ij

ji

ji

ij

dd

gd

xp

xp

Kds

xp

xp

Ex

pE

g

,

2

2

),

(),

,(

),

(lo

g)

,(

log

),

(lo

g

θθ

θθ

θ

θθ

θθ

θθ

θ

4/

4/

Thal

es A

ir S

yste

ms

Dat

e

��%�����

��&����

���� ��������

�����

������� ��

��������

���� ���� �

H. S

him

a, “

The

Geo

met

ry o

f Hes

sian

Str

uctu

res”

, Wor

ld

Scie

ntifi

c, 2

007

http

://w

ww

.wor

ldsc

ient

ific.

com

/wor

ldsc

iboo

ks/1

0.11

42/6

241

dedi

cate

d to

Pro

f. Je

an-L

ouis

Kos

zul (

«th

e co

nten

t of t

he

pres

ent b

ook

finds

thei

r orig

in in

his

stu

dies

»)H

. Shi

ma,

Sym

met

ric s

pace

s w

ith in

varia

nt lo

cally

Hes

sian

st

ruct

ures

, J. M

ath.

Soc

. Jap

an,,

pp. 5

81-5

89.,

1977

H. S

him

a, «

Hom

ogen

eous

Hes

sian

man

ifold

s»,

Ann

. Ins

t. Fo

urie

r, G

reno

ble,

pp.

91-

128.

, 198

0H

. Shi

ma,

«Va

nish

ing

theo

rem

s fo

r com

pact

Hes

sian

m

anifo

lds

», A

nn. I

nst.

Four

ier,

Gre

nobl

e, p

p.18

3-20

5., 1

986

H. S

him

a, «

Har

mon

icity

of g

radi

ent m

appi

ngs

of le

vel s

urfa

ces

in a

real

a� ���

ne s

pace

», G

eom

etria

e D

edic

ata,

pp.

177

-184

., 19

95H

. Shi

ma,

«H

essi

an m

anifo

lds

of c

onst

ant H

essi

an s

ectio

nal

curv

atur

e»,

J. M

ath.

Soc

. Jap

an, p

p. 7

35-7

53.,

1995

H. S

him

a, «

Hom

ogen

eous

spa

ces

with

inva

riant

pro

ject

ivel

y fla

t a� ���

ne c

onne

ctio

ns»,

Tra

ns. A

mer

. Mat

h. S

oc.,

pp. 4

713-

4726

, 19

99

Hiro

hiko

Shim

a, E

mer

itus

Prof

esso

r of Y

amag

uchi

Uni

v.Ph

D fr

om O

saka

Uni

vers

ityIn

terp

lay

betw

een

the

Geo

met

ry o

f Hes

sian

Str

uctu

res

and

Info

rmat

ion

Geo

met

ry

Jean

-Lou

is K

oszu

l, Fr

ench

Sci

ence

s Ac

adem

yPh

Dst

uden

tof H

enri

Car

tan,

Bou

rbak

i mem

ber

Intr

oduc

tion

of K

oszu

lfor

ms,

Kos

zul-V

inbe

rgch

arac

teris

ticfu

nctio

n&

met

ricJ.

L. K

oszu

l, «

Sur l

a fo

rme

herm

itien

ne c

anon

ique

des

es

pace

s ho

mog

ènes

», c

ompl

exes

, Can

ad. J

. Mat

h. 7

, pp

. 562

-576

., 19

55J.

L. K

oszu

l, «

Dom

aine

s bo

rnée

s ho

mog

ènes

et

orbi

tes

de g

roup

es d

e tr

ansf

orm

atio

ns a

ffine

s»,

Bul

l. So

c. M

ath.

Fra

nce

89, p

p. 5

15-5

33.,

1961

J.L.

Kos

zul,

«O

uver

ts c

onve

xes

hom

ogèn

es d

es

espa

ces

affin

es»,

Mat

h. Z

. 79,

pp.

254

-259

., 19

62J.

L. K

oszu

l, «

Varié

tés

loca

lem

ent p

late

s et

co

nvex

ité»,

Osa

ka J

. Mah

t. 2,

pp.

285

-290

., 19

65J.

L. K

oszu

l, «

Déf

orm

atio

ns d

es v

arié

tés

loca

lem

ent

plat

es»,

.Ann

Inst

Fou

rier,

18 ,

103-

114.

, 196

8Se

e:M

. N. B

oyom

, «C

onve

xité

loca

le d

ans

l’esp

ace

des

conn

exio

ns s

ymét

rique

s. C

ritèr

e de

com

para

ison

de

s m

odèl

es s

tatis

tique

s»,

, Mar

ch 2

012,

IHP,

Par

isht

tp://

ww

w.c

erem

ade.

daup

hine

.fr/~

peyr

e/m

spc/

msp

c-th

ales

-12/

«Le

s co

nnex

ions

sym

étriq

ues

est u

n so

us-e

nsem

ble

conv

exe

cont

enan

t le

sous

-ens

embl

e de

s co

nnex

ions

loca

lem

ent p

late

()

()

ψψ

ψ

ψψ

log

det

log

log

log

0lo

g,

)(

conn

ectio

nfla

t co

nes,

dual

an

d ,

*

**

,

*

ds

ds

d

Dd

gdx

ex

Dx

xx

=−

=

>=

=

ΩΩ

Ω∈

Ω

Kos

zul-V

inbe

rgC

hara

cter

istic

Func

tion

&

met

ricof

regu

lar

conv

exco

neΩ ΩΩΩ

5/

5/

'������

��� ������

�����

���� ���� ��

��

����

��

��

� ���

��

����

���

���

���

�H

irohi

ko S

him

a B

ook,

«G

eom

etry

of H

essi

an

Stru

ctur

es»,

wor

ld S

cien

tific

Pub

lishi

ng 2

007,

ded

icat

ed

to J

ean-

Loui

s K

oszu

lJ.

L. K

oszu

l

6/

6/

Thal

es A

ir S

yste

ms

Dat

e

'������

��� ������

�����

���� ���� ��&��

� ���#�

������

H. S

him

a, “

The

Geo

met

ry o

f Hes

sian

Str

uctu

res”

, Wor

ld S

cien

tific,

200

7ht

tp://

ww

w.w

orld

scie

ntifi

c.co

m/w

orld

scib

ooks

/10.

1142

/624

1de

dica

ted

to P

rof.

Jean

-Lou

is K

oszu

l («

the

cont

ent o

f the

pre

sent

boo

k fin

ds th

eir o

rigin

in h

is s

tudi

es»)

A pa

ir (D

; g) o

f a f

lat c

onne

ctio

n D

and

a H

essi

an m

etric

gis

cal

led

a H

essi

an s

truc

ture

.

J.

L. K

oszu

lstu

died

a f

lat m

anifo

ld e

ndow

ed w

ith a

clo

sed

1-fo

rm α ααα

such

that

Dα ααα

is p

ositi

ve d

efin

ite,

whe

reup

on D

α αααis

a H

essi

an m

etric

. Thi

s is

the

ultim

ate

orig

in o

f the

not

ion

of H

essi

an s

truc

ture

sA

Hes

sian

str

uctu

re (D

; g) i

s sa

id to

be

of K

oszu

ltyp

e, if

ther

e ex

ists

a c

lose

d 1-

form

α α α α su

ch th

at g

= D

α ααα

The

seco

nd K

oszu

lfor

m β βββ

play

s an

impo

rtan

t rol

e si

mila

r to

the

Ric

ci te

nsor

for a

Käh

leria

nm

etric

Let

vbe

the

volu

me

elem

ent

ofg.

We

defin

ea

clos

ed1-

form

α αααan

dβ βββ

asy

mm

etric

bilin

ear

form

by:

The

form

sα ααα

and

β βββar

eca

lled

the

first

Kos

zul

form

and

the

seco

ndK

oszu

lfor

mfo

raH

essi

anst

ruct

ure

(D;g

)res

pect

ivel

y

αβ

αD

vX

vD

X=

=

and

)

(

[]

()

[]

()

[]

� �

∂∂

∂=

∂∂=∂∂

=�

∧=

ji

klji

ij

iji

in

/ij

xx

gx

vg

x�

..dx

dxg

vde

tlo

g21de

tlo

gde

t2

21

12

1

αβ

LetV

beth

eve

ctor

spac

eof

allr

eals

ymm

etric

mat

rices

ofde

gree

n,an

dle

tΩ ΩΩΩ

beth

ese

tof

all

posi

tive

defin

item

atric

esin

V,a

regu

larc

onve

xco

nein

V.

xD

dn

Dd

ge

xdx

ex

n

xyTr

yx

xx

det

log

21

log

)(

det

)(

21

dual

-se

lf

)(

,

*,

**

*+

−=

=�

==

+−

Ω=Ω

ψ

ψψ

Leve

lsur

face

ofψ ψψψ

=M

inim

alSu

rfac

eof

(Ω ΩΩΩ,g

)

Flat

Con

nect

ion

D:

tors

ion

tens

or T

and

cur

vatu

re te

nsor

R v

anis

h id

entic

ally

7/

7/

Thal

es A

ir S

yste

ms

Dat

e

� ��

�����������(���

������

�$��

����)

M. A

rnau

don,

L. M

iclo

, ”M

eans

in c

ompl

ete

man

ifold

s:

uniq

uene

ss a

nd a

ppro

xim

atio

n”, h

ttp://

arxi

v.or

g/ab

s/12

07.3

232

M. A

rnau

don,

C. D

ombr

y, A

. Pha

n, L

e Ya

ng, «

Stoc

hast

ic

algo

rithm

s fo

r com

putin

g m

eans

of p

roba

bilit

y m

easu

res.

»,

Stoc

hast

ic P

roce

sses

and

thei

r App

licat

ions

122,

pp.

143

7-14

55,

2012

M. A

rnau

don,

A. T

halm

aier

, “B

row

nian

mot

ion

and

nega

tive

curv

atur

e”, B

ound

arie

s an

d Sp

ectr

a of

Ran

dom

Wal

ks, P

rogr

ess

in P

roba

bilit

y, V

ol. 6

4, 1

45--1

63, S

prin

ger B

asel

,201

1M

. Arn

audo

n, F

. Bar

bare

sco,

Le

Yang

, ”M

edia

ns a

nd m

eans

in

Rie

man

nian

geo

met

ry: e

xist

ence

, uni

quen

ess

and

com

puta

tion”

Mat

rix In

form

atio

n G

eom

etry

, Nie

lsen

, Fra

nk; B

hatia

, Raj

endr

a (E

ds.),

Spr

inge

r, ht

tp://

arxi

v.or

g/pd

f/111

1.31

20v1

Le Y

ang,

«M

édia

nes

de m

esur

es d

e pr

obab

ilité

dan

s le

s va

riété

s rie

man

nien

nes

et a

pplic

atio

ns à

la d

étec

tion

de c

ible

s ra

dar»

, PhD

with

adv

isor

s M

. arn

audo

n &

F. B

arba

resc

oht

tp://

tel.a

rchi

ves-

ouve

rtes

.fr/d

ocs/

00/6

6/41

/88/

PDF/

Dis

sert

atio

n-Le

_YAN

G.p

df, T

HAL

ES P

hD A

war

d 20

12

Mar

c Ar

naud

onPh

D w

ith M

. Em

ery,

Bor

deau

x U

nive

rsity

P-M

eans

Com

puta

tion

on R

iem

anni

an M

anifo

ldSt

ocha

stic

Flo

w o

n R

iem

anni

an M

anifo

ld

Mic

hel E

mer

yIR

MA

Lab,

Str

asbo

urg

Uni

vers

ityPr

obab

ility

on R

iem

anni

anM

anifo

ldSt

ocha

stc

Cal

culu

son

Man

ifold

sM

. Ém

ery,

G. M

okob

odzk

i, «

Sur l

e ba

ryce

ntre

d'u

ne

prob

abili

té d

ans

une

varié

té»,

Sém

inai

re d

e pr

obab

ilité

s de

Str

asbo

urg,

25

, p. 2

20-2

33, 1

991

http

://ar

chiv

e.nu

mda

m.o

rg/a

rtic

le/S

PS_1

991_

_25_

_220

_0.p

dfM

. Ém

ery,

P.A

. Mey

er, «

Stoc

hast

ic c

alcu

lus

in

man

ifold

s»,

Spr

inge

r 198

9M

. Ém

ery,

W. Z

heng

, «Fo

nctio

ns c

onve

xes

et

sem

imar

tinga

les

dans

une

var

iété

», S

émin

aire

de

Prob

abili

tés

XVIII

, Lec

ture

Not

es in

Mat

hem

atic

s 10

59,

Sprin

ger 1

984

M. F

réch

et, «

L’in

tégr

ale

abst

raite

d’u

ne fo

nctio

n ab

stra

ite e

t son

app

licat

ion

à la

moy

enne

d’u

n él

émen

t alé

atoi

re d

e na

ture

que

lcon

que

», R

evue

Sc

ient

ifiqu

e, 4

83-5

12, 1

944

M. F

réch

et, «

Les

élém

ents

alé

atoi

res

de n

atur

e qu

elco

nque

dan

s un

esp

ace

dist

anci

é».

Ann

ales

IHP,

10

no.

4, p

. 215

-310

, 194

8 ht

tp://

arch

ive.

num

dam

.org

/art

icle

/AIH

P_19

48__

10_4

_215

_0.p

df

[]

func

tion

conv

ex

cont

inuo

us , )

()

ϕϕ

xE

x≤

[]

()

()

0)

(ex

p)

(1

=�

=

−dx

Px

gx

gE

bM

b

8/

8/

���*+,����

��� ��

������

�����

������ ������&�-��

���)���$

�����)����

��

��

��

������

���

��

�C

ompu

tatio

nal I

nfor

mat

ion

Geo

met

ry

�H

essi

an In

form

atio

n G

eom

etry

Sym

plec

tic In

form

atio

n G

eom

etry

�O

ptim

izat

ion

on M

atrix

Man

ifold

s�

Prob

abili

ty o

n M

anifo

lds

�D

iver

genc

e G

eom

etry

& A

ncill

arity

�M

achi

ne/M

anifo

ld/T

opol

ogy

Lear

ning

Opt

imal

Tra

nspo

rt G

eom

etry

Mat

hem

atic

al M

orph

olog

y fo

r Ten

sor/M

atrix

-Val

ued

Imag

es�

Geo

met

ry o

f Aud

io P

roce

ssin

g�

Shap

e Sp

aces

: Geo

met

ry a

nd S

tatis

tic�

Geo

met

ry o

f Sha

pe V

aria

bilit

y�

Geo

met

ric In

vers

e Pr

oble

ms:

Filt

erin

g/Es

timat

ion

on M

anifo

lds

�D

iffer

entia

l Geo

met

ry in

Sig

nal P

roce

ssin

g�

Rel

atio

nal M

etric

�Fi

nite

Met

ric S

pace

s

http

://w

ww.

gsi2

013.

com

28-3

0th

Augu

st

2013

Pa

ris

9/

9/

���*+,����

��� ��

������

�����

������ ������&�-��

���)���$

�����)����

��

�Ya

nn O

LLIV

IER

(Par

is-S

ud U

nive

rsity

, Fra

nce)

: «

Info

rmat

ion-

geom

etric

opt

imiz

atio

n: T

he

inte

rest

of i

nfor

mat

ion

theo

ry fo

r dis

cret

e an

d co

ntin

uous

opt

imiz

atio

�H

irohi

ko S

HIM

A (Y

amag

uchi

Uni

vers

ity, J

apan

): «

Geo

met

ry o

f Hes

sian

Str

uctu

res

»

�G

iova

nni P

ISTO

NE

(Col

legi

o C

arlo

Alb

erto

, Ita

ly):

«N

on-P

aram

etric

Info

rmat

ion

Geo

met

ry»

10/

10/

�.����� ��/.�

��0�������� �������

��

���

��

��

���

��

� �!

��

���

��

�"�

��

����

���

��#��$

����

��

���

��

�"�

��

����

���

��#��$

����

��

���

��

�"�

��

����

���

��#��$

����

��

���

��

�"�

��

����

���

��#��$

����

%

�La

bora

toire

d’a

ccue

il :

�IR

CA

M, S

alle

Igor

Stra

vins

ky

�P

roje

t IN

RIA

/CN

RS

/Irca

mM

uSyn

c(A

rshi

aC

ont)

�A

nim

atio

n : A

rshi

aC

ont(

IRC

AM

& IN

RIA

), Fr

ank

Nie

lsen

(Son

y R

esea

rch)

, F.

Bar

bare

sco

(Tha

les)

�Le

s la

bora

toire

s�

IRC

AM

(Ars

hia

Con

t, G

erar

d A

ssay

ag, A

rnau

d D

esse

in)

�P

olyt

echn

ique

(Oliv

ier S

chw

ange

r,F.N

iels

en)

�M

ines

Par

isTe

ch(S

ilver

eB

onna

bel,

Jesu

sA

ngul

o)�

UTT

Tro

yes

(Hic

hem

Sno

ussi

) �

Uni

v. P

oitie

rs (M

arc

Arn

audo

n, L

e Ya

ng)

�U

niv.

Mon

tpel

lier (

Mic

hel B

oyom

, Pau

l Bry

and)

�IN

RIA

(Xav

ier P

enne

c, A

rshi

aco

nt)

�Th

ales

(Fré

déric

Bar

bare

sco,

Ale

xis

Dec

urni

nge)

�S

ony

Res

arch

(Fra

nk N

iels

en)

�Si

te w

eb: h

ttp://

repm

us.ir

cam

.fr/b

rillo

uin/

past

-eve

nts

http

://re

pmus

.irca

m.fr

/bril

loui

n/ho

me

11/

11/

� ��

��1��

)���

�2-���(3

�4$�� �5���

�� �

���������

��� �6�

'������(����

�����'���

����7�

���)

�-��

����������

���8��

With

Pro

f. R

ajen

dra

Bha

tia (N

ew D

ehli

Indi

an In

stitu

t of S

tatis

tics)

12/

12/

Air

Sys

tem

s D

ivis

ion

«La

Mes

se e

st d

ite»

by M

arce

l Ber

ger

•Pre

mie

r Mira

cle

:La

théo

rie d

es e

spac

es s

ymét

rique

s pe

ut ê

tre

cons

idér

ée c

omm

e le

pre

mie

r mira

cle

de la

géo

mét

rie

riem

anni

enne

, en

fait

com

me

un n

œud

de

fort

e de

nsité

dan

s l’a

rbre

de

tout

es le

s m

athé

mat

ique

s …

O

n do

it à

Elie

Car

tan

dans

les

anné

es 1

926

d’av

oir d

écou

vert

que

ces

géo

mét

ries

sont

, da

ns u

ne

dim

ensi

on d

onné

e, e

n no

mbr

e fin

i, et

en

outr

e to

utes

cla

ssée

s.

•Sec

ond

Mira

cle

:En

tre

les

varié

tés

loca

lem

ent s

ymét

rique

s et

les

varié

tés

riem

anni

enne

s gé

néra

les,

il e

xist

e un

e ca

tégo

rie in

term

édia

ire, c

elle

des

var

iété

s kä

hlér

ienn

es. …

On

a al

ors

affa

ire p

our d

écrir

e le

pan

oram

a de

s m

étriq

ues

kähl

érie

nnes

sur

not

re v

arié

té, n

on p

as à

un

espa

ce d

e fo

rmes

diff

éren

tielle

s qu

adra

tique

s, tr

ès lo

urd,

mai

s à

un e

spac

e ve

ctor

iel d

e fo

nctio

ns n

umér

ique

s [le

pot

entie

l de

Käh

ler]

. …

La

riche

sse

Käh

lérie

nne

fait

dire

à c

erta

ins

que

la g

éom

étrie

käh

lérie

nne

est p

lus

impo

rtan

te q

ue la

omét

rie ri

eman

nien

ne.

•Pas

d’e

spoi

r d’a

utre

mira

cle

:N

e ch

erch

ez p

as d

’aut

res

mira

cles

du

genr

e de

s es

pace

s (lo

cale

men

t) sy

mét

rique

s et

des

var

iété

s kä

hlér

ienn

es. E

n ef

fet,

c’es

t un

fait

depu

is 1

953

que

les

seul

es v

arié

tés

riem

anni

enne

s irr

éduc

tible

s qu

i adm

ette

nt u

n in

varia

nt p

ar tr

ansp

ort p

aral

lèle

aut

re q

ue g

elle

-mêm

e (e

t sa

form

e vo

lum

e) s

ont l

es

espa

ces

loca

lem

ent s

ymét

rique

s, le

s va

riété

s kä

hlér

ienn

es, l

es v

arié

tés

kähl

érie

nne

de C

alab

i-Yau

, et

les

varié

tés

hype

rkäh

lérie

nnes

.

Mar

cel B

erge

r (IH

ES),

«15

0 an

s de

Géo

mét

rie R

iem

anni

enne

»,G

éom

étrie

au

20iè

me

sièc

le, H

isto

ire e

t hor

izon

s, H

erm

ann

Édite

ur, 2

005

Rea

d M

ore

: Mar

cel B

erge

r, «

A Pa

nora

mic

Vie

w o

f Rie

man

nian

Geo

met

ry»,

Spr

inge

r 200

3

M. B

erge

r

Ber

ger,

Mar

cel,

«Le

s es

pace

s sy

mét

rique

s no

ncom

pact

s».

Ann

ales

sci

entif

ique

s de

l'Éc

ole

Nor

mal

e Su

périe

ure,

Sér

. 3, 7

4 no

. 2 (1

957)

, p. 8

5-17

7

����������� �����

Thal

es A

ir S

yste

ms

Dat

e

!�������

� .)

. ����3(�

3(-�2

!

14/

14/

!�������

�Pr

obab

ility

in M

etric

Spa

ces

by F

réch

et�

Met

ric G

eom

etry

of H

erm

itian

Pos

itive

Def

inite

s M

atric

es (H

PD)

�V

iew

poi

nt o

f geo

met

ers:

Car

tan-

Sie

gel S

ymm

etric

Bou

nded

Dom

ains

�V

iew

poi

nt o

f pro

babi

lists

: Inf

orm

atio

n G

eom

etry

and

Fis

her I

nfor

mat

ion

met

ric

�K

arch

er F

low

s fo

r HPD

mat

rices

�B

asic

Kar

cher

flow

to c

ompu

te «

Mea

n »

Bar

ycen

ter o

f N H

PD

mat

rices

�W

eisz

feld

-Kar

cher

flow

to c

ompu

te «

Med

ian

» B

aryc

ente

r of N

HP

D m

atric

es

�K

arch

er F

low

s fo

r Toe

plitz

HPD

mat

rices

(TH

PD)

�Tr

ench

/Ver

blun

sky

Theo

rem

s: P

artia

l Iw

asaw

a D

ecom

posi

tion

and

CA

R m

odel

�K

arch

er F

low

for H

essi

an m

etric

/Info

rmat

ion

Geo

met

ry m

etric

of C

ompl

ex A

utor

egre

ssiv

e M

odel

�N

ew «

Ord

ered

Sta

tistic

Hig

h D

oppl

er R

esol

utio

n C

FAR

» (O

S-H

DR

-CFA

R)

�O

S-H

DR

-CFA

R P

roce

ssin

g C

hain

& re

sults

on

real

reco

rder

dat

a

�K

arch

er F

low

s fo

r Toe

plitz

-Blo

ck-T

oepl

itz H

PD m

atric

es (T

BTH

PD)

�M

atrix

Ext

ensi

on o

f Tre

nch/

Ver

blun

sky

Theo

rem

: Mul

tivar

iate

CA

R p

aram

etriz

atio

n

�K

arch

er F

low

in S

iege

l Dis

k w

ith M

osto

w D

ecom

posi

tion/

Ber

ger F

ibra

tion

�N

ew «

Ord

ered

Sta

tistic

Spa

ce-T

ime

Adap

tive

Proc

essi

ng»

(OS-

STAP

)�

Qua

ntiz

atio

n of

Com

plex

Sym

met

ric S

pace

s by

Ber

ezin

�An

alys

is o

n Sy

mm

etric

Con

e by

Far

aut

�H

omog

eneo

us H

yper

bolic

Affi

ne H

yper

sphe

res

by S

asak

i

����������� �����

Thal

es A

ir S

yste

ms

Dat

e

� ��

����������

�$�� ��

��9�

����

���� .��

��

16/

16/

$��

����(�

�.�� .��

�����

���

����������$

�� ��

��9�

���

Mau

rice

Ren

é Fr

éche

t

&'

()

�*+

��

���

���

���#,�

-.

��

���

���

��

����

����

� �

���

���

���

���

��

��

��

��

������

���

�������

����

����

/0

���

��

��

$�-

1

�1

����

�2�

����

���

����

��1

��

��

�23

��

����

��

��

&'4)�5

��$

�6 �

�����

-�7

+1��

����

������

,�8�

���

��

$�3

��

����

����

1928

Jacq

ues

Had

amar

dPhD

Sup

ervi

sor

17/

17/ 4$��

�6������ �

��� �)

�)������ .��

����� ��

���� ����$�� ��

��9�

��

Rig

ht T

riang

le {a

,b,h

}

h2=a

2 +b2

a

b

h

{}

22

21

with

,

,i

ii

N ii

ii

ba

hh

ba

+=

=

{}

{}

{}{

}(

)H

BA

hb

ad

Min

HB

Ai

ii

N i

p geod

esiq

ueH

BA

,,

,,

,ar

g,

,1

,,� =

=

a

b

h>9

0�

22

2b

ah

+>

Insi

de th

e co

ne(b

ound

eddo

mai

n)

Am

blig

one

Tria

ngle

p=2

: Mea

n, p

=1 M

edia

n

18/

18/

Mod

ify o

r Hid

e in

the

head

er /

foot

er p

rope

rties

: D

ate

$��

����� .�

������(��

)���-�����

������3��� �

����9�

���

3��

6 �

��

��9

�:��

���

���

��

����

���

��

����

���

���

���

��

��

��

��

���

����

���

��

��

��

���

���

���

;9�+

��

���

���

���

#,�

-9�

��&

<9�

��(

9��

�4&

=>?

&<

9�&

'(

)*�

8

� ��

� �

@�

��

��

.

�Il

cons

titue

la m

ise

au p

oint

de

cert

ains

des

cou

rs q

ue n

ous

avon

s fa

it à

la S

orbo

nne

en

1945

et e

n 19

46�

Form

idab

le e

xten

sion

du

Cal

cul [

des

prob

abili

tés

qui]

résu

lte d

e la

sim

ple

intr

oduc

tion

de la

not

ion

de d

ista

nce

de d

eux

élém

ents

alé

atoi

res

�La

nat

ure,

la s

cien

ce e

t la

tech

niqu

e of

fren

t de

nom

breu

x ex

empl

es d

’élé

men

ts a

léat

oire

s qu

i ne

sont

, ni d

es n

ombr

es, n

i des

sér

ies,

ni d

es v

ecte

urs,

ni d

es fo

nctio

ns. T

elle

s so

nt

par e

xem

ple,

la fo

rme

d’un

fil j

eté

au h

asar

d su

r une

tabl

e, la

form

e d’

un o

euf p

ris a

u ha

sard

dan

s un

pan

ier d

’oeu

fs. O

n a

ains

i une

cou

rbe

aléa

toire

, une

sur

face

alé

atoi

re. O

n pe

ut a

ussi

con

sidé

rer d

’aut

res

élém

ents

mat

hém

atiq

ues

aléa

toire

s : p

ar e

xem

ple

des

tran

sfor

mat

ions

alé

atoi

res

de c

ourb

e en

cou

rbe.

�Il

para

ît ce

rtai

n qu

e l’u

rban

ism

e co

ndui

ra à

étu

dier

des

élé

men

ts a

léat

oire

s te

ls q

ue la

fo

rme

d’un

e vi

lle p

rise

au h

asar

d, v

érifi

er p

ar e

xem

ple,

d’u

ne m

aniè

re s

cien

tifiq

ue

l’hyp

othè

se d

e la

tend

ance

au

déve

lopp

emen

t des

vill

es v

ers

l’Oue

st.

�Pa

rmi l

es p

rem

ière

s no

tions

à g

énér

alis

er fi

gure

nt c

elle

s de

loi d

e pr

obab

ilité

, de

vale

urs

typi

ques

(moy

enne

, équ

ipro

babl

e, e

tc.),

de

disp

ersi

on, d

e co

nver

genc

e st

ocha

stiq

ue,

que

nous

allo

ns d

’abo

rd e

xam

iner

.�

Nou

s in

diqu

eron

s pl

us lo

in e

n N

ote

addi

tionn

elle

, un

moy

en tr

ès g

énér

al d

’éte

ndre

la

notio

n de

fonc

tion

de ré

part

ition

à d

es é

lém

ents

alé

atoi

res

de n

atur

e qu

elco

nque

.

Appa

rtem

ent o

u vé

cu F

réch

et12

Squ

are

Des

noue

ttes,

15

ème,

Par

is

19/

19/

Mod

ify o

r Hid

e in

the

head

er /

foot

er p

rope

rties

: D

ate

$��

����� .�

������(��

)���-�����

������3��� �

����9�

��

�N

ous

avon

s en

trep

ris u

ne é

tude

exp

érim

enta

le d

e qu

elqu

es e

xem

ples

par

ticul

iers

. Ceu

x-ci

son

t act

uelle

men

t : la

form

e al

éato

ire d

’un

fil je

té s

ur u

ne ta

ble,

la fo

rme

aléa

toire

d’

une

sect

ion

horiz

onta

le (p

rise

à un

niv

eau

anat

omiq

ue d

éter

min

é) d

’un

crân

e hu

mai

nch

oisi

au

hasa

rd.

�Po

ur c

e se

cond

exe

mpl

e, n

otre

cho

ix s

’est

por

té s

ur d

eux

colle

ctio

ns d

e co

ntou

rs

obte

nus

au “

conf

orm

ateu

r” d

’une

par

t sur

des

per

sonn

es v

ivan

tes

(mai

s an

onym

es),

cont

ours

four

nis

grac

ieus

emen

t par

la g

rand

e m

aiso

n pa

risie

nne

de c

hape

llerie

Soo

ls,

d’au

tre

part

sur

une

col

lect

ion

ethn

ogra

phiq

ue d

e cr

ânes

d’in

divi

dus

mor

ts, c

olle

ctio

n m

ise

aim

able

men

t à n

otre

dis

posi

tion

par M

. Les

ter,

Dire

cteu

r du

labo

rato

ire

d’Et

hnog

raph

ie.

�L’

étud

e st

atis

tique

d’u

ne c

olle

ctio

n de

con

tour

s cr

ânie

ns p

réhi

stor

ique

s.�

Les

fonc

tions

repr

ésen

tativ

es s

ont d

ével

oppé

es e

n sé

rie d

e fo

urie

r et l

’on

proc

éder

a à

une

anal

yse

stat

istiq

ue d

e ce

s sé

ries.

On

déte

rmin

era

les

lois

de

fréq

uenc

es d

es

prem

iers

coe

ffici

ents

pris

isol

émen

t et l

’on

étud

iera

d’a

utre

par

t leu

r cor

réla

tion

au

moy

en d

e di

vers

indi

ces

de c

orré

latio

n et

en

part

icul

ier l

’indi

ce d

iago

nal q

ue n

ous

avon

s dé

fini r

écem

men

t.�

Le la

bora

toire

de

Cal

cul d

e l’I

nstit

ut H

enri

Poin

caré

opé

rant

sur

deu

x an

alys

eurs

ha

rmon

ique

s.�

Pour

les

fils,

env

iron

120

coef

ficie

nts

(nou

s di

spos

ons

de 1

00 d

e ce

s fo

rmes

). Po

ur le

s co

ntou

rs c

râni

ens

dont

les

form

es s

ont i

nfin

imen

t moi

ns v

arié

es, n

ous

nous

som

mes

co

nten

tés

de 1

2 ha

rmon

ique

s so

ient

25

coef

ficie

nts

pour

cha

que

crân

e.�

Dév

elop

emen

t en

série

de

Four

ier d

e l’é

quat

ion

en c

oord

onné

es p

olai

res

d’un

con

tour

cr

ânie

n

20/

20/

Mod

ify o

r Hid

e in

the

head

er /

foot

er p

rope

rties

: D

ate���)

�� .��

������ ��

��������

���3��

)����3 ��� �

3��

6 �

��

���

���

1

���

��8�

��3

��A

����

�M

. Ozi

l a d

ès 1

938

émis

l’id

ée q

u’on

pou

rrai

t aid

er à

la c

lass

ifica

tion

des

race

s au

moy

en d

e l’a

naly

se h

arm

oniq

ue d

es p

rofil

s hu

mai

ns. L

es ta

blea

ux d

e M

. O

zil n

e co

rres

pond

ent p

our d

es fo

rmes

en

fait

asse

z co

mpl

iqué

s qu

’à 9

ha

rmon

ique

s et

repr

ésen

tent

seu

lem

ent l

a m

oitié

exp

ress

ive

du p

rofil

, c’e

st à

di

re u

ne fo

nctio

n no

n pé

riodi

que.

Doc

umen

ts n

umér

isés

à

part

ir du

Fon

d Fr

éche

t de

s A

rchi

ves

de

l’Aca

dém

ie d

es

Scie

nces

:C

orre

spon

danc

e en

tre

M.

Fréc

het e

t M. O

zil

����������� �����

Thal

es A

ir S

yste

ms

Dat

e

$�� ��

����

��� �����

�� �

�����

������� �

�:�������

��$�� ��

���;�

�:<�

=��

���������7�

�)��

�� ��

�>

22/

22/

Mod

ify o

r Hid

e in

the

head

er /

foot

er p

rope

rties

: D

ate

� ��

��������$

�� ��

���:�����

������"��

��?� ��)

��� � ������

B�

����

���

�-

1�

1����

���

��

����

�����

� ��

���C

���

�$�

��

����

������1

��C

����4

� ����

�8� �

�1��

��D

A �

��

��8

���

���

�9��

���C

���

�$�

����

����

����1

��C

����4

��

1

�1

����

���

��

���

����

$���

���

�4�

��

��8

� �

�1

����D

������ ��

������ ��

=

− )1()1(1)1( 2

)1( 1

)1(

nn zzzz

Z�

������ ��

������ ��

=

− )2(

)2(1)2( 2

)2( 1

)2(

nn zzzz

Z�

()

???

,)2(

)1(=

ZZ

d

())1(

)1(/θ

Zp

())2(

)2(/θ

Zp

()

???

,)2(

)1(=

θθ

d

23/

23/

�������

�� ��"������ ���

������������� ����������

��� �����(

��

,�$

��

��

��

��

� � �

�C

ram

er-R

ao-F

réch

et-D

arm

ois

Bou

ndan

d Fi

sher

Info

rmat

ion

Mat

rix

�K

ulba

ck-L

eibl

erD

iver

genc

e (v

aria

tiona

ldef

initi

onby

Don

sker

/Var

adha

n) :

�R

ao-C

hent

sov

Met

ric(in

varia

nce

by n

on-s

ingu

larc

hang

e of

par

amet

eriz

atio

n)

�In

varia

nce:

1945

1943

( IH

P Le

ctur

e 19

39)

()(

)(

)1ˆ

ˆ−

+≥

� ��� ��

−−

θθ

θθ

θI

E

CR

FD B

ound

[]

() �� ��

�� ��

∂∂

∂−

=*

2

, /

ln)

(j

ij

iX

pE

θθ

θ

Fish

er In

form

atio

n M

atrix

R.A

. Fis

her

()

()

[]

�=

=+

=+

ji

ji

ji

dd

gd

Id

dX

pX

pK

ds,

*,

2)

(/

,/

θθ

θθ

θθ

θθ

()

()

[]

()

()

()

dxx

qx

px

pe

EE

Sup

qp

Kq

p�� ��

�� ��=

−=

θθ

θφ

φ

φ//

ln/

ln)

,(

N. N

. Che

ntso

v )

()

()

(2

θds

wds

Ww

=�

=

24/

24/

2��

����

�� ��

��@� �����

�������

�)���������������

#�:� �

7��

���

�1

���

� �

���6

��

��

��

��

$��

���1

��

��7

�8�

"�

��

�K

ullb

ack

Div

erge

nce

can

be n

atur

ally

intro

duce

d by

com

bina

toria

l ele

men

ts a

nd

stirl

ing

form

ula

:

()

[] )

(lo

g)

,(

φ

φφ

eE

ESu

pq

pK

qp

−=

{} in

()

∏ =

=M i

in iM

MM

nqN

qq

nn

nP

i

11

21

!!

,...,

/,..

.,,

iq� =

=M i

iN

n1

Nnp

ii

=

+∞→

≈−

n

whe

n

..2

..

!n

en

nn

[]

),

(lo

g.

log

11

qp

Kqp

pP

NLi

mM i

iii

MN

=� ��

� ��=� =

+∞→

Let

mul

tinom

ial L

aw o

f N e

lem

ents

spr

ead

on M

leve

ls

with

prio

rs ,

an

d

Sirli

ng fo

rmul

a gi

ves

:

�E

� �

���

���6

��

����

�$�

���

��1

���

�7�8

� "

��

��

�D

onsk

er a

nd V

arad

han

have

pro

pose

d a

varia

tiona

l def

initi

on o

f Kul

lbac

k di

verg

ence

:

25/

25/

This

pro

ves

that

the

supr

emum

ove

r all

is n

o sm

alle

r tha

n th

e di

verg

ence

����

��#�:� �

7��

�����@3(3

:�30*��@� �����

����3

99 ���

��

()

[] )

(lo

g)

,(

φ

φφ

eE

ESu

pq

pK

qp

−=

�D

onsk

er a

nd V

arad

han

have

pro

pose

d a

varia

tiona

l def

initi

on o

f Kul

lbac

k di

verg

ence

:

()

),

()1

ln(

),

()

()

()

(ln

)(

)(

ln)(

)(

ln)

(

)(

)(

ln)

( :

Con

side

r

qp

Kq

pK

qpq

qpp

eE

E

qp

qp

=−

=� ��

� ��− �� ��

�� ��=

−�

�� ���� ��

=

��

ωω

φ

ωωω

ωωω

φ

ωωω

φ

()

()

[]

0)

()

(ln

)(

)(

ln)

()

,(

)()

()

(w

ith

)(

)(

ln)

()

(ln

)(

ln)

(

)()

(

≥� ��

� ���� ��

�� ��=

−−

�=

�� ���� ��

� ��� ��

=�� ��

�� ���� ��

�� ��=

��

ωφ

φ

θ

θφω

φφ

ω

φ

φ

φφ

ωωω

φθω

ω

ωωω

φ

qpp

eE

Eq

pK

eq

eq

q

qqp

eE

eE

eE

E

qp

qp

qp

φ

Usi

ng th

e di

verg

ence

ineq

ualit

y,

�Li

nk w

ith «

Larg

e D

evia

tion

Theo

ry»

& F

ench

el-L

egen

dre

Tran

sfor

mw

hich

gi

ves

that

loga

rithm

of g

ener

atin

g fu

nctio

n ar

e du

al to

Kul

lbac

k D

iver

genc

e :

[]

[]

[]

[]

[]

[]

)(

(.)

)(

(.)

)(

log

)(

),

(

)(

log

)(

)(

),

(

),

()

()

()

(lo

g

xV

qp

V

xV

V

p

xV

eE

VE

Sup

qp

K

dxx

qe

dxx

px

VSu

pq

pK

qp

Kdx

xp

xV

Sup

dxx

qe

−=

−=

−=

Vara

dhan

Abel

Priz

e 20

07

26/

26/

:������

�� �

���������

��� ���$

���� � ������

��������/�"

7�

����

���

���

���

����

���F

�-

���

���

��$�

��

��

��

�P

oten

tial F

unct

ions

are

Dua

l and

rela

ted

by L

egen

dre

tran

sfor

mat

ion

:

()

()

()

()

()(

)(

)(

)(

)(

)(

) � �

−−

−+

−=

−Θ

=�

� �

+Σ−

==

==

−−

−−

−−

−−

−−

en

�)

(�

��

��

nTr

��

mm

m�,�

� )m

,(�

,�

TT

T

ππ

θθ2

log

2de

tlo

g2

1lo

g2

~~

)lo

g(2

det

log

22

~~

,~

2~

sco

ordi

nate

Dua

l

11

11

11

12

11

� �

=∂∂

=∂∂

� �

=∂∂

=∂∂

���

��

���

��

~~

and

~~

()

TT��

TrH,

�H,

��

+=

−≡

θη~

~w

ith

~~

~~

()

[] p

E�

�lo

g~

~=

Entr

opy

ji

* ijj

iij

���

g�

��

g∂

∂∂

≡∂

∂∂

=~

an

d

~2

2

�H

essi

ans

are

conv

exe

and

defin

e R

iem

anni

an m

etric

s :

27/

27/

Air

Sys

tem

s D

ivis

ion

:����:���� ��

������

����� �

���

��C

���

���

��1

��

��7

�8�

"�

��

�Fo

r eac

h D

ual G

eom

etry

, we

can

built

a d

iver

genc

e th

at is

dire

ctly

rela

ted

to K

ulba

ck-L

eibl

er D

iver

genc

e :

()

()

[]

()

()

()

()

[]

()

()

()

()

[]

()

()

()dx

mx

pm

xp

mx

p,�

m,N

,�m

ND

iv

H,�

��

��

,�m

,N,�

mN

Div

H,�

��

��

,�m

,N,�

mN

Div

ΣΣ

Σ=

≥−

+≡

≥−

+≡

22

11

11

11

22

21

21

11

22

*

12

12

11

22

,/

,/

log

,/

0~

~~

~~

~0

~~

~~

~~

Rie

man

nian

Met

ric�

As

Pot

entia

l are

con

vexe

, the

ir H

essi

ans

defin

e R

iem

anni

an M

etric

s :

()

()

()

()

()

()

()

()(

)(

)�� ��

�� ��

−−

Σ+

Σ−

ΣΣ

Σ−

=−

−−

Tm

mm

mTr

Tr, �

m,N

,�m

ND

iv2

12

11 2

1 11 2

11 2

11

12

2

det

log

21

()

()

33

2

2121

ij

iij

ij

iij

d�O

d�d�

gd�

Od�

d�g

ds+

=+

=

28/

28/

Air

Sys

tem

s D

ivis

ion���� �

���������

��� �����$

���� � ������

��������:�

�����

3�

���8

� �

���

��

�����

��7

���

�1���

��

$���

��

���

�R

iem

anni

an In

form

atio

n m

etric

is g

iven

by

Rao

Met

ric

()

[]

()

TAB

TrB

AA

AA

dd

Trds

==

ΣΣΣ

Σ=

−−

,et

,w

ith

2121

2

22/1

2/12

12

Rao

Met

ric

Mul

tivar

iate

Gau

ssia

n of

non

zer

o m

ean

�Is

omet

ries

are

give

n by

follo

win

g ho

meo

mor

phis

ms

:

()

[]

21

12

21Σ

Σ+

Σ=

−−

dTr

dmdm

dsT

()

()

()

()

),

(x

,

with

'

isom

etry

,'

,',

22

2R

nG

LR

Aa

dsds

dsA

Aa

mA

mm

n

TT

∈=

Σ+

→Σ �2

2

11

1

1th

en

A

s

−Σ

Σ

−−

Σ−=

ΣΣ

�=

ΣΣ

dsds

dd

IIn

varia

nce

by in

vers

ion

()

()

1 21 1

21

,,

−−

ΣΣ

ΣD

D

29/

29/

2��

�������-5�

�9����$

���

� ������

��������/�"

����>�

����

����

�C

�Fi

sher

Info

rmat

ion

Mat

rix fo

r Gau

ss-L

apla

ce G

auss

ian

Law

:

()(

)�� ��

�� ��=

≥� ��

� ��−

−� ��

� ��=

−−

σθ

θθ

θθ

θσ

θm

IE

IT

an

d

)(

ˆˆ

w

ith

20

01 .

)(

12 (

)(

) �� ��

�� ��+

� ��� ��

=+

==

−2

22

22

222

2.2

.2.

σσσ

σθ

θθ

ddm

ddm

dI

dds

T

�R

ao-C

hent

sov

Met

ric o

f Inf

orm

atio

n G

eom

etry

H. P

oinc

aré

.2

σim

z+

=(

) 1

<+−

ωi

zi

z(

)222

2

1.8ωω

−=

�d

ds�

Fish

er M

etric

is e

qual

to P

oinc

aré

Met

ric fo

r Gau

ssia

n La

ws:

{}{

}(

)

)*2()1(

)2()1(

)2()1(

2

)2()1(

)2()1(

22

11

2

1)

,(

with

),

(1

),

(1

log

.2,

,,

ωω

ωω

ωω

δ

ωω

δω

ωδ

σσ

−−

=

�� ���� ��

−+=

mm

d

30/

30/

�������

��@��"�����

������ ���2� ���1���7

���:��

�����$

�� ��

Hen

ri Po

inca

ré(u

pper

-hal

f pla

ne

mod

el o

f hyp

erbo

lic

geom

etry

)n=

1

Elie

Car

tan

(cla

ssifi

catio

n in

6

type

s of

sym

met

ric

hom

ogen

eous

bo

unde

d do

mai

ns )

n<=3

Erne

st V

inbe

rg(li

nk w

ith h

omog

eneo

us c

onve

xe c

ones

,

S

iege

l do

mai

ns o

f 2nd

kin

d)

Fine

stru

ctur

e of

In

form

atio

n G

eom

etry

(Hes

sian

Geo

met

ry,

Käh

leria

n G

eom

etry

)

« Il

est c

lair

que

si l’

on p

arve

nait

à dé

mon

trer

que

tous

le

s do

mai

nes

hom

ogèn

es d

ont l

a fo

rme

est d

éfin

ie p

ositi

ve s

ont s

ymét

rique

s, to

ute

la th

éorie

de

s do

mai

nes

born

és h

omog

ènes

ser

ait é

luci

dée.

C

’est

là u

n pr

oblè

me

de g

éom

étrie

her

miti

enne

ce

rtai

nem

ent t

rès

inté

ress

ant »

Der

nièr

e ph

rase

de

Elie

Car

tan,

dan

s «

Sur l

es

dom

aine

s bo

rnés

de

l'esp

ace

de n

var

iabl

es c

ompl

exes

»,

Abh

. Mat

h. S

emin

ar H

ambu

rg, 1

935

()

ji

ji

ji

zddz

zz

zz

K�

∂∂

∂=

,

2,

log

Φ

Jean

-Lou

is K

oszu

l(c

anon

ical

her

miti

an fo

rm o

f co

mpl

ex h

omog

eneo

us s

pace

s,

a co

mpl

ex h

omog

eneo

us s

pace

w

ith p

ositi

ve d

efin

ite c

anon

ical

he

rmiti

an fo

rm is

isom

orph

ic to

a

boun

ded

dom

ain,

,Stu

dy o

f A

ffine

Tra

nsfo

rm G

roup

s o

f lo

cally

flat

man

ifold

s)

Car

l Lud

wig

Sie

gel

(Sie

gel d

omai

ns in

fram

ewor

k of

Sym

plec

tic G

eom

etry

)

Look

eng

Hua

(Ber

gman

, Cau

chy

and

Poi

sson

K

erne

ls in

Sie

gel d

omai

ns)

31/

31/ 2� ���1���7

������

��� ��

����

�7��

��������

�)�)

�:��

����

{}

()

()2

*

**

11

111

11

,,

1/

zww

zK

zzC

z

IV

IIIII

I ,−

=<

∈=

==

=

Hen

ri Po

inca

ré(n

=1)

()

()

()

()

()(

)(

)do

mai

n. th

eof

volu

me

eucl

idea

n is

w

here

, :

IV Ty

pefo

r 2

ZZ1

1,

1 ,

: II

I

Type

1 ,

: II Ty

pe

, : I

Type

for

det

1,

-*

*t

**

Ωμ

νΩ

μ

ννν

Ωμ

ν

ν

n

ZWW

WW

ZK

p

p

qp

ZWI

WZ

K

IV n

III

pII pI p,q

=−

+=

� �

−=

+=

+=

−=

+−+

Look

eng

Hua

02

ZZ1,1

ZZ

:lin

e 1

and

row

sn

with

m

atric

esco

mpl

ex

: IV

Type

por

der

of m

atric

essy

mm

etric

skew

com

plex

:

III

Type

por

der

of m

atric

essy

mm

etric

com

plex

: II

Type

row

s q

and

lines

p w

ith

mat

rices

com

plex

: I

Type

):

(

Mat

rixr

Rec

tang

ula

Com

plex

:

2t

t>

−+

<

−<

+

++

ZZ

co

njug

ate

tran

spos

edI

ZZZ

IV nIII

pII pI p,q

Car

l Lud

wig

Sie

gel

Elie

Car

tan

(n<=

3)

32/

32/

Air

Sys

tem

s D

ivis

ion

��9�

�����������

� .��9

���

+�

���

�$�

��

+�

���

�$�

��

+�

���

�$�

��

+�

���

�$�

��

4 444G

��

��

����

� 1

����-

��

�� �

�-��

�G

��

��

����

� 1

����-

��

�� �

�-��

�G

��

��

����

� 1

����-

��

�� �

�-��

�G

��

��

����

� 1

����-

��

�� �

�-��

�M

öbiu

s Tr

ansf

orm

is a

tran

sitiv

e ac

tion

that

tran

sfor

ms

uppe

r hal

f-pl

ane

to it

self

(hom

ogen

eous

spa

ce) :

�M

and

–Mha

ve s

ame

actio

n th

en w

e co

nsid

er th

e qu

otie

nt G

roup

:

�C

ompl

ex u

nit d

isk

is li

nk to

upp

er h

alf p

lane

by

Cay

ley

trans

form

:

{} 0

)Im

(:

an

d

)(

,

2>

∈=

∈++

=∈ �� ��

�� ��=

zC

zH

zd

czb

azz

MR

SLd

cb

aM

22

2/

IR

SLR

PSL

±= {

}

11et

1:

zzi

z

HD

iz

iz

z

DH

zC

zD

−+→

+−→

<∈

=

��

1-1

i -i

22

2

22

2

ydzy

dydx

ds=

+=

x

y

H

D (

)222

2

14

zdzds

−=

Car

l Lud

wig

Sie

gel C

ontr

ibut

ion

in h

is s

emin

al b

ook

«Sy

mpl

ectic

Geo

met

ry»

: a

gene

raliz

atio

n of

Poi

ncar

é Sp

ace

33/

33/

�Si

egel

Met

ric fo

r the

Sie

gel U

pper

-Hal

f Pla

ne:

�U

pper

-Hal

f Pla

ne :

�Is

omet

ries

of

a

re g

iven

by

the

quot

ient

gro

up:

with

the

Sym

plec

tic G

roup

:

�U

niqu

e M

etric

inva

riant

e by

:

�� �

����

����

�3����

� 9�

���������7�

��$�� ��

������ �

nSH {

} nI

Rn

SpR

nPS

p2

/),

()

,(

±≡

),

(F

nSp

()(

)1)

(−

++

=� �� ��

�� ��=

DC

ZB

AZZ

MD

CB

AM

� �

=−

⇔∈ �� ��

�� ��=

nT

T

TT

IB

CD

AD

BC

AF

nSp

DC

BA

Msy

mm

etric

et

),

(

{}

),

2(0

0 ,

/),

2()

,(

Rn

SLI

IJ

JJM

MF

nG

LM

Fn

Spn

nT

∈ �� ���� �� −

==

∈≡

)(Z

M(

)(

)(

)Zd

YdZ

YTr

dsSi

egel

11

2−

−=

�� ==

nRYX

0(

)(

)(

)2

12

nn

dRR

Trds

−=

iYX

Z+

=

{} 0

Im/),

(>

=∈

+=

=Y

(Z)

Cn

Sym

iYX

ZSH

n

34/

34/

���7

���A99

��1������

����

����

��� �

{} 0

Im/),

(>

=∈

+=

=Y

(Z)

Cn

Sym

iYX

ZSH

n

Sieg

el U

pper

Hal

f Spa

ce

X

0>

Y

()

()

()

ZdY

dZY

Trds

11

2−

−=

()

()(

)(

)� =

−+

=n i

ii

ZZ

d1

22

12

1/1

log

λ

kk

kYi

XZ

.+

=1=

k

2=

k

()

[]

21

2dR

RTr

ace

ds−

=

()

kk

kk

RN

WRi

Z,0

if

.≡

=

1=

k2=

k

()

()

� =

=n k

kR

Rd

1

22

12

log

()

()(

)()(

)12

12

11

21

21

21,

−−

−−

−−

=Z

ZZ

ZZ

ZZ

ZZ

ZR

()

0.

..

det

2/11

22/1

1=

−−

−I

RR

()

()

0.

,de

t2

1=

−I

ZZ

35/

35/

BA3

0'�C

3'�!

0��0

�2!$�/-D

��E$

$-'(�2���3

2-�

F. B

erez

in

{}

()(

)(

)(

)(

)(

)(

)(

)(

)B

BI

trac

eB

BI

gF

AB

g

ZF

Zg

FZ

BA

trac

eZ

FZ

gF

ZZ

Itr

ace

ZZ

Iz

F

AZ

BB

AZZ

gB

AA

BI

BB

AA

II

JJ

Jgg

BB

BA

g

IZZ

ZSD

tt

t

n

++

++

++

+

+=

+=

�=

∂∂=

∂∂�

++

=−

−=

−−

=

++

= � �

=−

=−

� ��� �� −

==

� ��� ��

=

<=

log

det

log

))0((

)0(

)(

)(

log

Re

2)

())

((

log

det

log

)(

: po

tent

ial

Käh

ler

)(

w

ith

0

whe

re

00

with

an

d

with

/

1*

**

**

1*

***

**

ZgZZ

gj

ZZ

KZ

gj

zg

jgZ

gZK

ZZ

dK

ZZ

Kh

c

ZZ

dK

ZZ

KZ

gZ

fh

cg

f

h

h

∂∂=

=� ��

� ��=

� ��� ��

=

),

(, )

,(

),

()

,(

),

(

with

)

,(

)0,0()

,(

)(

),

()0,0()

,(

)(

)(

)(

,

**

**

/1*

1

*/1

*

μ

μ

()

()

()

()

()ν

βα

αββ

α

βα

βα

πμ

μ

−+

+=

∂∂

∂−

==

=

�W

WI

WW

FW

WW

WF

gdW

dWg

ds

WW

dW

WF

WW

dn

L

det

),

(

w

here

,lo

g w

ith ,

,,

**

*2

,

*,

2

**

*

36/

36/

�In

form

atio

n G

eom

etry

for M

ultiv

aria

te G

auss

ian

Law

of z

ero

Mea

n an

d in

trin

sec

Geo

met

ry o

f Her

miti

an P

ositi

ve D

efin

ite M

atric

es

(par

ticul

ar c

ase

of S

iege

l Upp

er-H

alf P

lane

) pro

vide

the

sam

e m

etric

�In

form

atio

n G

eom

etry

:

�G

eom

etry

of S

iege

l Upp

er-H

alf P

lane

:

()

()

()

21

2n

ndR

RTr

ds−

=

[]

()(

)[

]n

n

nn

nn

n

RR

Trn

nn

n

RR

E

mZ

mZ

R

eR

RZ

pn

n

=

−−

==+

−−

−−

ˆ

and

with

..

)(

)/

(1

�� ��

�� ��−

=*

2

.)

/(

ln)

(j

i

nn

ijZ

pE

g∂θ

∂θθ

∂θ

0=

nm

with

���7

�����:

�$�� �5�$

�� ��

�F���

�$�� ��

����$

���� ���

��������/�"

{} 0

Im/),

(>

=∈

+=

=Y

(Z)

Cn

Sym

iYX

ZSH

n

()

()

()

iYX

ZZd

YdZ

YTr

dsSi

egel

+=

=−

with

1

12

�� ==

nRYX

0(

)(

)(

)2

12

nn

dRR

Trds

−=

37/

37/

�Si

egel

Dis

tanc

e:�

Par

ticul

ar C

ase

(X=0

) and

Gen

eral

Cas

e:

�Pa

rticu

lar C

ase

(pur

e im

agin

ary

axis

) :

�G

ener

al C

ase

of S

iege

l Upp

er-H

alf P

lane

Dis

tanc

e:

:�����

������"��

����

:�$�� ��

����9

� ���

��� ������������7�

()

()

()

� =

−−

==

n kk

RR

RR

Rd

1

22

2/11

22/1

12

12

log

..

log

()

0de

t1

2=

−R

with

0

with

∈+

=X

SHiY

XZ

n

()

n

n kkk

Sieg

elSH

ZZ

ZZ

d∈

�� ��

�� ���� ��

�� ��

−+=� =

21

1

22

12

, w

ith

11lo

g,

λλ

with

()

0.

),

(de

t2

1=

−I

ZZ

()

()(

)()(

)12

12

11

21

21

21,

−−

−−

−−

=Z

ZZ

ZZ

ZZ

ZZ

ZR

0

avec

>

=R

iRZ

38/

38/

Air

Sys

tem

s D

ivis

ion

Sie

gel h

as d

educ

ed a

n ot

her d

ista

nce

from

:

3��

��� �)�����

���)�)

���)

�� ��

����7�

��=� #

()

[]

21

2.

ds− Σ

Σ=

dTr

()(

)12

122 12

11 2

1 11 2

1 112

.T

=R

an

d

.T

TT

+−

−−

−−

Σ−

Σ=

R is

her

miti

an p

ositi

ve d

efin

ite m

atrix

with

eig

en-v

alue

s : )

..

(

and

)(

r

with

11

2/1)1(1

)2(2/1)1(

kk

2

nn

nkk

kR

RR

Rr

−=

=�� ��

�� ��+−

λσ

λλ

We

dedu

ce th

at :

()

� =�� ��

�� ��−+

= � ��� ��

�� ���� ��

−+=

ΣΣ

n kkk

nn

rrR

IR

ITr

d1

2/1

2/12

2/1

2/12

21

2

11ln

ln,

beca

use

:

[]

[]�

�=

∞ =

−=

�� ���� ��

+=

= �� ���� ��

−+n k

j kk

k

nnr

akR

RR

RI

RI

1

j2

0

22/1

12/1

2/12

RTr

nd

1.2

..4

tanh

.4ln

39/

39/

��:�$�� ��

����

����� �

��

��

$��

� �

����

��-

���

�8�

�7�

$��

���

�3�

� ��

���"

�8�

��

1�

�G

eode

sic

:

-

��

���

��

$���

����

��

��

�Sy

met

ric S

pace

as s

tudi

ed b

y El

ie C

arta

n : E

xist

ence

of b

iject

ive

geod

esic

isom

etry

�B

ruha

t-Tits

Spa

ce: s

emi-p

aral

lelo

gram

ineq

ualit

y

�C

arta

n-H

adam

ard

Spac

e(C

ompl

ete,

sim

ply

conn

ecte

d w

ith

nega

tive

sect

iona

l cur

vatu

re M

anifo

ld)

[] 10

with

)

,(

.))

(,

(,

tR

Rdt

tR

dY

XX

∈=

γ ()

()

YX

YX

Xt

XY

XX

XR

RR

tX

RR

RR

RR

RR

RR

eR

tX

YX

�=

==

==

−−

−−

)2/1(

and

)1(

,

)0()(2/1

2/12/1

2/12/1

log

2/12/1

2/1

γγ

γγ

()

()

2/12/1

2/12/1

2/11-

),

(

avec

)(

XA

BAA

AB

AB

AB

AX

GB

A−

−=

=�

��

Xx

)d(

x,x

)d(

x,x

d(x,

z))

,xd(

xx

zx

x∈

∀+

≤+

∀∃

2

24

que

tel

,

22

21

22

21

21

40/

40/

x1x

2x

z

U

VU

+V

U−

V

2� ���*���

����� ��

��9�

��

�Sy

met

ric S

pace

as s

tudi

ed b

y El

ie C

arta

n : E

xist

ence

bije

ctiv

e ge

odes

ic

isom

etry

�B

ruha

t-Tits

Spa

ce: s

emi-p

aral

lelo

gram

ineq

ualit

y

22

22

22

VU

VU

VU

+=

++

),

(A

GB

BA

=

BG

AB

A)

,(

=(

))

(X

-1)

,(

BA

BA

XG

BA

��

=

X

()

2/1/21

2/12/1

2/1A

BAA

AB

A−

−=

()

[] 1,0)(

2/12/1

2/12/1

∈=

−−

tA

BAA

At

A= )0(γ

B= )1(γ

22

21

22

21

22

4)

d(x,

x)

d(x,

xd(

x,z)

),x

d(x

+≤

+

E. J

. Car

tan

M. B

erge

r

J. T

its

41/

41/

Air

Sys

tem

s D

ivis

ion

�Th

is is

omet

ry fo

r met

ric s

pace

:

�Is

an

exte

nsio

n of

this

one

:

�To

be

com

pare

d w

ith E

uclid

ean

«sy

mm

etric

» sp

ace

������ ��

��9�

��

()

()

()

()

� �

=

=•

••

=−

−−

22/1

2/12

2/12/1

2/12/1

2/11-

),

(lo

g,

w

ith

)(

XBA

AB

A

ABA

AA

BA

BA

BA

XG

BA

δ

()

()

()

()

� �

=

==

−2

12

1

log

,

with

ab

ba

abb

ab

ax

ba

xG

-(a

,b)

δ

��

()

� �

−=+

=� ��

� ��+

+� ��

� ��+

=2

2,

2

with

2

2b

ab

a

ba

ba

ba

-xb

ax

G(a

,b)

δ

42/

42/

Air

Sys

tem

s D

ivis

ion

�is

the

only

fixed

poin

t bec

ause

:

�du

e to

trac

e pr

oper

tyof

:

A������

������5�)

�9����

()

2/1

2/1

2/1

2/1

2/1

ABA

AA

−−

()(

)C

XI

CX

XI

CX

XC

XX

XC

CX

=�

=�

=�

=−

−−

−−

2/12/1

root

squa

reof

unic

ity

2/12/1

2/12/1

1

()

()

()

()

() )

(,2

,)

(X

),(

X1

-1-1

BA

Xd

IX

BA

BA

dX

BA

BA

d•

=•

•=

••

{}

11

1

2

)(

of sei

genv

alue

w

ith

log

)(

−=

=

� ��� ��

=•

�X

BA

)B

Ad(

X,

n ii

n ki

λ

λ

43/

43/

Air

Sys

tem

s D

ivis

ion

�Fo

r spa

ceof

Sym

met

ricPo

sitiv

e D

efin

item

atric

es, t

he a

nalo

gue

of

is

:

�It

cons

titut

esa

natu

ralf

ram

ewor

kfo

r a g

ener

aliz

edth

eory

of c

onve

xity

, w

here

the

role

of a

rithm

etic

mea

nis

play

edby

a m

idpo

intp

airin

g:

isca

lled

-con

vex

if �

For s

pace

of s

ymm

etric

posi

tive

defin

item

atric

es, a

n af

fine

func

tion

isgi

ven

by :

�an

d is

clos

ely

rela

ted

to e

ntro

py:

2��

�5���

ba

)1(

λλ

−+

()

2/1

2/1

2/1

2/1

ABA

AA

BA

λλ

−−

=•

BA

λ•

BA

A0•

=B

AA

1•=

f(

)2

1,•• ()

()

()

Yf

Xf

YX

f2

1•

≤•

det

log

=f

()

cste

REn

trop

y+

−=

det

log

44/

44/

Air

Sys

tem

s D

ivis

ion

� ��

��1'�����9�

��������

�19� ������7 ��

�/�"

��

Bru

hat-T

itsSp

ace

:�

Spa

ce(S

ym++

,δ) i

sca

lled

Bru

hat-T

itsS

pace

, whe

re, a

ccor

ding

to δ

met

ric, p

oint

ate

qual

dist

ance

of 2

poi

nts

isun

ique

and

giv

enby

ge

omet

ricm

ean:

()

()

),

()

,(

,)

,(

2)

,(

2)

,(

0de

t w

ith

log

),

(

defin

itepo

sitiv

e)

,(

,

11

1

1

2

−−

=

==

==

=−

� ��� ��

=

∈∀

BA

BA

BCC

ACC

BA

BB

AA

BA

IAB

BA

Rn

Sym

BA

TT

n ii

δδ

δδ

δδ

λλ

δ

��

�B

ruha

t-Tits

Spa

ce is

a c

ompl

ete

met

ric s

pace

that

ver

ifies

sem

i-par

alle

logr

am la

w :

Xx

)�(

x,x

)�(

x,x

�(x,

z))

,x�(

xz

xx

∈+

≤+

∃∈

all

for

2

24

:su

ch th

at

X,

22

21

22

21

21

�B

ruha

t-Tits

Spa

ce a

re p

artic

ular

cas

es o

f Car

tan-

Had

amar

d M

anifo

ld [1] F

. Bru

hat &

J. T

its, «

Gro

upes

rédu

ctifs

sur

un

corp

s lo

cal»

, IH

ES, n�4

1, p

p.5-

251,

197

2

«B

ruha

t-Tits

Spac

e»(M

etric

Spa

ce)

45/

45/

Air

Sys

tem

s D

ivis

ion

� ��

��1'�����9�

��������

�1�� ������7 ��

�/�"

�B

ruha

t-Tits

Spa

ce :

�S

emi-p

aral

lelo

gram

Law

:

Xx

)�(

x,x

)�(

x,x

�(x,

z))

,x�(

xz

xx

∈+

≤+

∃∈

all

for

2

24

:su

ch th

at

X,

22

21

22

21

21

x1x

2x

z

U

VU

+V

U−

22

22

22

VU

VU

VU

+=

++

−�

Ded

uced

from

par

alle

logr

am la

w :

V

22

22

12

23

22

21

22

2)

(x,x

d)

(x,x

d)

(x,x

d)

,x(x

d+

=+

22

22

12

22

22

12

32

22

24

.2)

(x,x

d)

(x,x

d(x

,z)d

),x

(xd

)(x

,xd

(x,z)

d+

=+

�=

3x

46/

46/

�G

eode

sic

betw

een

mat

rix P

and

Geo

desi

c be

twee

nQ

& R

:

�Th

e ge

odes

ic p

roje

ctio

n is

a c

ontr

actio

n:

���

)����

�� �G������

Q

R

PP

Q

()

()

[]

2/12/1

2/12/1

2/12/1

2/12/1

,)

(P

PQ

RQQ

QP

Pt

st

ts

s−

−−

−=

σ

()

()

QP

dist

QP

dist

MM

,)

(),

(≤

ΠΠ

M: g

eode

sica

lly

conv

ex c

lose

d su

bman

ifold

of

Had

amar

d M

anifo

ld(

) SP

dist

PM

SM

,m

inar

g)

(∈

Π ΠΠΠM

(P)

Π ΠΠΠM

(Q)

P

Q

47/

47/

Air

Sys

tem

s D

ivis

ion

��� ������

������

��� 7�

���� ��

�W

e ca

n fin

d B

ergm

an K

erne

l of K

ähle

r geo

met

ry fo

r Sie

gel U

pper

-ha

lf sp

ace

by m

ean

of S

iege

l Uni

t Dis

k :

�A

nalo

gy w

ith B

ergm

an K

erne

l for

Poi

ncar

é un

it di

sk :

),

(ln

ZZ

K�

B−

=(

){

} 0 /

on

)

,(

>−

=Z

I-Z

ZWZ

IW

ZK

B

{}

()

() (

)(

)

()2

222

2

22

22

22

22

2

1

1

1

1

)(

1

11

ln

1/

on

1

),

(

with

)

,(

ln

zdzz

dzd

zz

zgd

zdds

zz

zz

zz

zg

zzz

z

zz

wzw

zK

zz

KB

B

−=

∂∂

Φ∂

==

−=

−−

−=

∂∂

Φ∂

=�

−=

∂Φ∂�

−−

<−

=−

()(

)1−+

→Ψ

nn

nn

iIZ

iIZ

Z

SD:S

H

�{

}ZZ

ZI

ZSy

m(n

,C)/

ZSD

n=

<∈

=2

2 w

ith

With

Ber

gman

Ker

nel (

)(

)(

)Zd

ZZI

dZZZ

ITr

ZdZ

dZ

Zds

11

22

−−

−−

=∂

∂Φ

∂=

48/

48/

A99�

1��

�������

�����:��#

������

�������

� .���

)�2�/�����7�

()

()

()

*1

*1

*2

222

2

111

dww

wdw

ww

ds

wdwds

−−

−−

=

−=

iz

iz

w+−

=

()(

)1−+

−=

iIZ

iIZ

W

Upp

er-H

alf P

lane

of H

. Poi

ncar

éU

nit D

isk

of P

oinc

aré

Sieg

el U

pper

-Hal

f Pla

neU

nit S

iege

l Dis

k(

)

),

(

and

),

(w

ith

11

2

Cn

HPD

YC

nH

erm

XiY

XZ

ZddZ

YY

Trds

∈∈

+=

=−

0an

d

with

*1

12

22

2

22

2

>+

==

=+

=

−−

yiy

xz

dzdz

yy

dsydz

ydy

dxds

()

()

[]

+−

+−

+−

−=

dWW

WI

dWW

WI

Trds

11

2

49/

49/

�K

oszu

l For

ms

�1s

t Kos

zul F

orm

:

�2n

d K

oszu

l For

m:

�A

pplic

atio

n fo

r Poi

ncar

é U

pper

-Hal

Pla

ne:

�W

ith

a

nd

�W

e ca

n de

duce

that

H���

1/��

����

�%����� ��

()

()

[]

gX

XJa

dJX

adTr

Xb

g∈

∀−

)(

/

()

XdΨ

−=

41α

αβ

D=

{} 0

/ >

+=

=y

iyx

zV

dxdy

X=

dydy

Y=

()

[]

[]

� �

=−

==

�Y

JXY

YX

ZY

ZY

ad,

,.

()

()

[]

()

()

[]

�� =

−=

−02

YJa

dJY

adTr

XJa

dJX

adTr

dydx

y∧

21

()

2

22

2

2

2

2141

2

ydydx

ds

ydy

dxd

ydxX

+=

∧−

=−

=�

αΨ

50/

50/

�K

oszu

l for

m fo

r Sie

gel U

pper

-Hal

f Pla

ne:

�S

ympl

ectic

Gro

up :

�A

ssoc

iate

d Li

e A

lgeb

ra:

H���

1/��

����

�%����� ����� ����7�

��A99

� 1������

����

{} 0

/ >

+=

=Y

iYX

ZV

()

�� ���� �� −

=�� ��

�� ��=

� �

==

+=

00

an

d

0

with

,

1

II

JDB

AS

BD

DB

ID

AD

BAZ

SZT

TT

-

�� ���� ��

+=

�� ���� ��

−=

� �

−==

�� ���� ��

000

, 0

0

base

and

with

0

jiij

ijji

ijij

T

Te

ee

e

ad

bb

dba

βα

()

()

()

()

()

()

()

()

� �

+=

∧+

−=

Ω

+=

� � �

+=

Ψ

−−

−−

− ZddZ

YY

Trp

ds

ZdY

dZY

TrP

d

dXY

Trp

idY

dX

pij

ijij

11

2

11

1

81

38

13

412

13

13

0 δβα

����������� �����

Thal

es A

ir S

yste

ms

Dat

e

���

��� �����!

9���

���' ��

�9� �

��)�I�

�4�� 7�����6�)�����

������

$��

����� .�

����

52/

52/

!9���

���' ��

�9� ��'��

� ���=

���� ������)�����

��

��

���

���

���7

����

��

�C

lass

ical

form

of W

asse

rste

in d

ista

nce

in O

ptim

al tr

ansp

ort

-�

���

���

��

���

�$�

��

���

���

����

����

��

�C

ase

n =

1: M

onge

-Kan

toro

vich

-Rub

inst

ein

Dis

tanc

e

�C

ase

n=2

: dis

tanc

e de

Lev

y-Fr

éche

t Dis

tanc

e

()

()

()

[]

()

{} ν

μν

μ

πν

μν

μΠ

π

==

=

�� ���� ��

= ℘

)(

,)

(,

,,

),

()

,(

,

/1

/1

),

(

Yla

wX

law

YX

dE

Inf

W

yx

dy

xd

Inf

W

nn

n

n

nn

()

[]

YX

EIn

fQ

Pd

WY

X−

==

,1

,)

,(

νμ

()

[]

2

,2

,)

,(

YX

EIn

fQ

Pd

WY

X−

==

νμ

53/

53/

!9���

���' ��

�9� ���� .�

���1=���� ������:�����

��

6 �

��

��>

��

���

���

���7

����

��

��$

3�

���8

� �

���

��

����

���

+C

�Fr

éche

t-Was

sers

tein

Dis

tanc

e

�Pr

oof:

�Le

t

�So

lutio

n is

giv

en b

y:

[]

()(

)(

)[

][

][

][

][

]Y

XY

XR

TrR

TrR

TrY

EX

EY

XE

YX

YX

TrE

YX

E

,2

22

2)

()

(−

++

−=

−−

=−

+

0≥

� ��� ��

=� � ��

� ��=

+Y

XW

RR

RYX

Ψ

()

()

2/12/1

2/12/1

2/1

0.

.1

−+

≥−

=�

++

−X

XY

XX

RR

RR

RR

RTr

Sup

YX

ΨΨ

ΨΨ

Ψ

()

()

[]

2/12/1

2/1.

..2

XY

XR

RR

TrTr

=+

Ψ

()

XR

RR

RR

YX

XY

XX

..

.2/1

2/12/1

2/12/1

−=

54/

54/

!9���

���' ��

�9� ���� .�

���1=���� ������:�����

��

��

���

���

���7

����

��

��$

3�

���8

� �

���

��

�����

���

�C

�C

�����

�>�

�����

�Fr

éche

t-Was

sers

tein

dis

tanc

e

�G

eode

sic:

�If

we

let:

�O

ptim

al T

rans

port

:�

trans

port

from

to

()(

)(

)[

][

](

)[

]2/1

2/12/1

22

.2,

,,

XY

XY

XY

XY

YX

XR

RR

TrR

TrR

Trm

mR

mR

md

−+

+−

=

()

12/1

2/12/1

2/12/1

,−

−=

=Y

XX

XY

XX

YX

RR

RR

RR

RD

()

()

� �

+−

+−

=

+−

=

YX

kY

YX

kt

XY

t

DtI

tR

DtI

tR

mtm

tm

,,

)(

)(

.)

1(.

)1(

.)

1(

()(

)[

](

)()

[]

)1()1(

)0()0(

2)

()

()

()

(2

,,

,).

(,

,,

Rm

Rm

Ws

tR

mR

mW

tt

ss

−≤

()

()

YY

kR

mN

I,

,#

ψ∇

()

YY

Rm

N,

()

XX

Rm

N,

()

XY

YX

mm

yD

yx

+−

=∇

=,

)(

ψ(

)(

)(

ψ−

+−

−=

+X

YY

XY

mm

vD

mv

v,

21)

(

()

()

()

()

YY

YX

Xx

my

Rm

ym

xR

mx

−−

=−

−−

+−

+1

1

[]

[]

()

[]

[]

()

[]

2/12/1

2/12/1

2/12/1

2/12/1

2/1 w

ith

.2Y

XX

YX

XX

YX

YX

RI

RR

RR

RM

MM

TrR

RR

TrR

TrR

Tr−

==

−+

−+

55/

55/

!9���

���' ��

�9� ���� .�

���1=���� ������:�����

��

��� �

���

���

����

$���

����

�����*

�����

H<.

�W

asse

rste

in M

etric

:

�Ta

ngen

t Spa

ce &

Exp

onen

tial M

ap:

-

��

���

��

$���

����

��

��

�A

lexa

ndro

v Sp

ace

�Sp

ace

of P

ositi

ve s

ectio

nnal

cur

vatu

re (g

eode

sica

lly c

onve

xe a

nd

sim

ply

conn

ecte

d). S

ectio

nal C

urva

ture

is g

iven

by:

Whe

re m

atrix

S is

a s

ymet

ric m

atrix

suc

h th

at:

is a

nti-s

ymm

etric

.

()

YR

XTr

YX

gY

RN

Y.

.)

,( )

,0(=

()

() )

(,,0

,0)

.(ex

pt

RR

NY

YR

NXt

=

()

()

XtI

tR

XtI

tR

kY

kt

R Y.

)1(

.)

1()

(,+

−+

−=

()V

TYR

N,0

()

).(

exp

,0Xt

YRN

()

YRRN

,0

()

()

()

()2

22

2)1(

),0().

1()

(,

.)0(

,).

1()

(,

γγ

γα

γα

γα

dt

tt

dtd

tt

d−

−+

−≥

()

[]

()

[]

()

()T

VN

SX

YV

SX

YTr

YX

K−

−=

,,

.4/3

),

( )(

[]

() VS

XY

−,

56/

56/

!9���

���' ��

�9� ���� .�

���1=���� ������:�����

��

��

���

���

���5

� �

��

���

�$�

��

���8

� �

���

�"�

�����

���

�C

��

*��

���

H<

.

�W

asse

rste

in B

aryc

ente

r

�So

lutio

n of

Was

sers

tein

Bar

ycen

ter

for N

Mul

itvar

iate

gau

ssia

n la

ws

of z

ero

mea

n :

�C

onst

rain

ed fo

r the

bar

ycen

ter:

�Ite

rativ

e so

lutio

n (p

aris

-dau

phin

e un

iver

sity

)�

Fixe

d po

int (

conv

erge

nce

for d

=2, d

>3 c

ondi

tiona

l con

verg

ence

to in

itiat

ion)

[]

2

,2

1

2 2)

,(

w

ith

),

(Y

XE

Inf

WW

Inf

YX

N kk

−=

� =

νμ

νμ

μ

()

{}N k

kRN

1,0

=

()

RR

RR

N kk

=� =1

2/12/1

2/1

()

with

2/1

2/1

1

2/1)

(2

)(

)1(

ii

N k

nk

nn

RK

KK

KK

=� ��

� ��=� =

+

57/

57/

!9���

���' ��

�9� ���� .�

���1=���� ������:�����

��

��8�>6

������

7��

�����

�In

195

7, M

auric

e R

ené

Fréc

het i

ntro

duce

d th

is d

ista

nce,

bas

ed o

n a

prev

ious

Pau

l Lev

y’s

pape

r fro

m 1

950

()

[]

−=

−=

),

()

(,

22

,

2y

xH

dd

yx

YX

EIn

fG

Fd

yx

YX

58/

58/

!9���

���' ��

�9� ���� .�

���1=���� ������:�����

��

��

8�>6

��

��

��7

����

��

�19

57 F

réch

et p

aper

in C

RA

S :

�Le

tter f

rom

Pau

l Lev

y to

Mau

rice

Fréc

het (

2 A

pril

1958

)�

… J

’ai a

insi

pu

appr

écie

r ce

que

vous

avi

ez fa

it, e

n pr

enan

t com

me

poin

t de

dépa

rt de

vot

re

mém

oire

ce

que

vous

app

elez

ma

prem

ière

déf

initi

on d

e la

dis

tanc

e de

deu

x lo

is d

e pr

obab

ilité

(en

fait

ce n

’éta

it pa

s la

pre

miè

re).

Vou

s l’a

vez

d’ai

lleur

s gé

néra

lisée

, en

ce s

ens

que

je n

e l’a

vais

as

soci

ée q

u’à

une

de v

os d

éfin

ition

s de

deu

x va

riabl

es a

léat

oire

s. E

t j’a

i bea

ucou

p ad

miré

co

mm

ent a

vec

votre

qua

trièm

e dé

finiti

on, v

ous

arriv

ez à

faire

que

lque

cho

se d

e m

ania

ble

d’un

e id

ée q

ui p

our m

oi é

tait

surto

ut th

éoriq

ue, v

u la

diff

icul

té d

e dé

term

iner

le m

inim

um d

e la

dis

tanc

e de

deu

x va

riabl

es a

léat

oire

s ay

ant l

es ré

parti

tions

mar

gina

les

donn

ées.

59/

59/I���4��

7�����6�)�����

������� .��

������ ��

����-5� ����2�9

����

(��

����

������

$�6 �

����

�E

xtre

me

Cop

ulas

of F

réch

et

�4t

h di

stan

ce o

f Fré

chet

()

() [

])(

),(

),

(w

ith

),

(,

1

12

2

yG

xF

Min

yx

H

yx

Hd

dy

xG

Fd

yx

=−=

[]

[] )

(),

()

,(

0,1)

()

()

,(

with

),

()

,(

),

( 10

01

yG

xF

Min

yx

Hy

Gx

FM

axy

xH

yx

Hy

xH

yx

H

=−

+=

≤≤

Extr

eme

Fréc

het

Cop

ulas

),

( 1y

xH

),

(0

yx

H

),

(y

xH

≥ ≥≥≥≥ ≥≥≥

60/

60/

�G

. Dal

l’Agl

io (U

nive

rsité

La

Sapi

enza

à R

ome)

, «Fr

éche

t Cla

sses

: th

e B

egin

ning

s»,

Adv

ance

d in

Pro

babi

lity

Dis

trib

utio

ns w

ith

Giv

en M

argi

nals

Bey

ond

the

Cop

ulas

, Rom

e 19

91, K

luw

er�

With

hel

p of

Giu

sepp

e P

ompi

lj, G

. Dal

l’Agl

io rm

et M

auric

e Fr

éche

t, 80

yea

rs

old,

in 1

956

in R

oma,

dur

ing

visi

t to

l’Ist

ituto

di C

alco

lo d

elle

Pro

babi

lita

�D

all’A

glio

met

a 2

ndtim

e Fr

éche

t in

Par

is a

nd is

in c

onta

ct w

ith P

aul L

evy

�D

all’A

glio

writ

e : «

Levy

als

o sh

owed

som

e in

tere

st in

dis

trib

utio

ns w

ith

give

n m

argi

nals

. In

a no

te in

196

0 he

refe

rs to

a fo

rmul

a by

Poi

ncar

é fo

r ut

iliza

tion

in n

-dim

ensi

onal

dis

trib

utio

ns w

ith g

iven

s m

argi

nals

, with

out

deve

lopp

ing

the

idea

»

�Pa

ul L

evy,

«Su

r les

con

ditio

ns d

e co

mpa

tibili

té d

es d

onné

es m

argi

nale

s re

lativ

es a

ux lo

is d

e pr

obab

ilité

» C

RAS

, t. 2

50 p

p.25

07-2

509,

196

0, n

ote

prés

enté

e pa

r M. J

acqu

es H

adam

ard

I���)�����

������� .��

�����(��

������

�����������������

��;���

9��G&�/���&�(�%%�&�:�

��*37���&��� �

����&�����J

<

61/

61/

�G

. Dal

l’Agl

io (U

nive

rsité

La

Sapi

enza

à R

ome)

, «Fr

éche

t Cla

sses

: th

e B

egin

ning

s»,

Adv

ance

d in

Pro

babi

lity

Dis

trib

utio

ns w

ith

Giv

en M

argi

nals

Bey

ond

the

Cop

ulas

, Rom

e 19

91, K

luw

er�

By

inte

grat

ion

by p

arts

:

�as

sign

s to

all

sub-

sets

of d

iago

nal y

=x, m

axim

um o

f pro

babi

lity

com

plia

nt w

ith m

argi

nals

: S. B

ertin

o, «

Su

di u

na s

otto

clas

se d

ella

cla

sse

di

Fréc

het»

, Sta

tistic

a 25

, pp.

511

-542

, 196

8

�G

ini W

ork

: in

1914

, Gin

i int

rodu

ced

«lin

ear d

issi

mila

rity

para

met

er»

(sol

utio

n in

dis

cret

cas

e fo

r α=

1, 2

)

�To

mm

aso

Sal

vem

ini (

+ Le

ti) W

ork

: con

stru

ctio

n of

«ta

bella

di c

ogra

duaz

ione

» an

d «

di c

ontro

grad

uazi

one

» (e

quiv

alen

t to

Fréc

het E

xtre

me

Rep

artit

ion

Func

tions

)

�G

ener

aliz

atio

n to

mul

tivar

iate

by

Riz

zi in

195

7, D

all’A

glio

(196

0)

I���)�����

������� .��

�����(��

������

�����������������

��;���

9��G&�/���&�(�%%�&�:�

��*37���&��� �

����&�����J

<

[]

()

()

()

� �

≥���

�� −

≤���

�� −

−=

−+

=−

≤≤

≤≤

yx

zF

zG

Inf

yG

yx

zG

zF

Inf

xF

yx

H

dzz

zH

zG

zF

YX

E

xz

y

yz

x

R

if

0,)(

)(

max

)(

if

0,)(

)(

max

)(

),

(

:fu

nctio

n n

Rep

artit

io g

Min

imliz

in

),

()

()

(

*

[]

()

()

()

{} )

(),

()

,(

:Fu

nctio

n n

Rep

artit

ioM

inim

izin

g

)()

,(

)(

)1(

)()

,(

)(

1

:1Fo

r 1

22

yG

xF

Min

yx

H

dudv

uv

vu

Fu

F

dudv

vu

vu

Hv

GY

XE

vuv

u

=

−−

+−

−−

=−

>

<

>

− ααα

αα

αα

α

),

(*y

xH

62/

62/

2��

9� ����

����� �

���������

��� �� � ����!9���

���' ��

�9� �

For t

rans

port

opt

imal

:A.

Tak

atsu

. On

Was

sers

tein

Geo

met

ry o

f the

Spa

ce o

f Gau

ssia

n M

easu

res,

to a

ppea

r in

Osa

ka J

. M

ath.

, 201

1, h

ttp://

arxi

v.or

g/ab

s/08

01.2

250

����������� �����

Thal

es A

ir S

yste

ms

Dat

e

� ���

����"���� ��

�:�

��� ��

���

$��

����$�)

����

2��

9������

64/

64/

Air

Sys

tem

s D

ivis

ion���

��� ��

�$��

��-5���������� ����

��� ��

������� �

�:�������

���� ��

��

��

���3

��

��/

0��

���

��$

���

��

� ��

�-

���

�8�

�7�

$��

���

�3�

� ��

���

��

���3

��

��/

0��

���

��$

���

��

� ��

�-

���

�8�

�7�

$��

���

�3�

� ��

���

��

���3

��

��/

0��

���

��$

���

��

� ��

�-

���

�8�

�7�

$��

���

�3�

� ��

���

��

���3

��

��/

0��

���

��$

���

��

� ��

�-

���

�8�

�7�

$��

���

�3�

� ��

���

�C

auch

y se

ries

give

n by

har

mon

ic a

nd a

rithm

etic

mea

n :

()

() 2/

an

d

2

, w

ith 1

11

11

00

nn

nn

nn

nn

nn

BA

BB

AA

BB

AA

BLi

mA

Lim

BA

+=

+=

==

==

+−

−−

+

∞→

∞→

�S

olut

ion

of th

e fo

llow

ing

OD

E :

� �

−=

=∈

∀++

∞→

++

Sym

X

XXA

BdtdX

tX

Lim

BA

Sym

BA

t)0(

with

)

(,

,1

�S

olut

ion

to th

e in

equa

lity

:

0

s.t.

(Loe

wne

r)m

atrix

hi

gher

,

,> �� ��

�� ��∈

∀++

BX

XA

BA

Sym

BA

�S

olut

ion

of th

e eq

uatio

n :

)de

t(lo

g)

( w

ith

))(

(''1

1X

XF

BAX

XA

XF

−=

==

−−

65/

65/

Air

Sys

tem

s D

ivis

ion

���

��� ��

�$��

����

�(�������-8

�����

bx

xaab

xab

x=

⇔=

⇔=

−12

�In

sca

lar c

ase,

geo

met

ric m

ean

of a

,b>0

sat

isfie

s fo

r x>=

0 :

�Th

e la

st «

sym

met

rized

» ve

rsio

n is

sui

tabl

e fo

r gen

eral

izat

ion

to n

onco

mm

utat

ive

setti

ngs

: the

geo

met

ric m

ean

is th

e un

ique

so

lutio

n of

,

if th

e un

ique

sol

utio

n ex

ists

.�

We

defin

e a

grou

p eq

uipp

ed w

ith th

e op

erat

ion

:

bx

xa=

−1

xxy

yx

1−=

•b

ax

ba

xb

xxa

�=

⇔=

•⇔

=−1

�To

sol

ve

, we

need

to

hav

e a

uniq

ue

solu

tion.

Eve

ry e

lem

ent m

ust h

ave

a un

ique

squ

are

root

:a

ex

=•

ex

xex

==

2

2/1 aa

e=

�E

xten

sion

for M

atrix

Geo

met

ric M

ean

:

()

()

BA

ABA

AA

X

BAA

XAA

BAA

XAA

XAA

BAA

AX

AXA

AB

XXA

�=

=�

=�

=�

=�

=

−−

−−

−−

−−

−−

−−

−−

−−

−−

2/12/1

2/12/1

2/1

2/12/1

2/12/1

2/1

2/12/1

2/12/1

2/12/1

2/12/1

2/12/1

2/12/1

1

))(

(

)(

66/

66/

Air

Sys

tem

s D

ivis

ion

,�� �

��8��3

���

��$

���� �

�G��

��

����

���

�D

enm

an-B

eabe

rs It

erat

ion

:

�Sc

hultz

Iter

atio

n (w

ithou

t mat

rix in

vers

ion)

:

�B

jörk

Met

hod

by S

chur

Dec

ompo

sitio

n :

���-�

���� ����$�� �5

��8�

� ��(�

��

�� ���� ��

� ��� ��

+ � ��� ��

=� ��

� ��=

+

++

00

00

210

01

1

1

11

k

k

k

k

k

kk

YZ

ZY

ZY

X

� ��� ��

=0

00

IA

Xw

ithth

en� ��

� ��=

−∞

→0

02/

1

2/1

AA

XLi

mk

k

)3(

210

02

1

11

kk

k

kk

XI

XZ

YX

−=

� ��� ��

=+

++

TAQ

Q=

+w

ith T

Upp

er tr

iang

ular

mat

rixB

y re

curr

ence

, we

com

pute

: +

=�

=Q

UQ

AT

U2/1

2

67/

67/

Air

Sys

tem

s D

ivis

ion

'���� ���

�����17��

��� ��

����

����2

�����

����� �

����

�G

auss

(20

year

s ol

d) &

Lag

rang

e in

depe

nden

tly p

rove

d th

at

�A

rithm

etic

-geo

met

ric m

ean

of a

and

b :

�Is

rela

ted

to e

lliptic

inte

gral

:

�La

nden

tran

sfor

mat

ion

give

s al

so :

()

()

aa

aa

bb

bb

n

ba

ba

ba

ba

ba

b

ba

ab

ab

ab

a

nn

nn

nn

nn

nn

nn

nn

n

nn

n

=≤

≤≤

≤≤

≤≤

=≥

∀�

≥� ��

� ��−

=−

� ��� ��

+=

� �

=

+=

=>

≥∀

++

++

++

01

10

22

21

21

110

0

......

0

02

2

2

and

,

, ,

0

()

� �

+=

==

=

∞→

∞→ 2/ 0

22

22

sin

cos

),

(

),

(

w

ith

),

(2/

π

tb

ta

dtb

aI

bLi

ma

Lim

ba

M

ba

Mb

aI

nn

nn

()

()

()

()

()

()

� ��� ��

+=

+=

� �

+=

= � ��� ��

+=

tab

Arc

tu

dttb

ta

ba

Jb

aab

Ib

aJ

ba

J

ba

Iab

ba

Ib

aI

tan

tan

: ns

f.La

nden

tra

sin

cos

,

with

,

,,

2

,,

2,

2/ 0

22

22

11

11

π

4.J(

a,b)

: P

erim

eter

of E

llips

eof

hal

f-axi

s a

& b

68/

68/

,��G

�9�

���

���

���

$��

��������

�$�

��

��$

$��

���

���

��

$��

��

������

���

�A

rithm

etic

mea

n:

�Th

is p

oint

min

imiz

es th

e fu

nctio

n of

dis

tanc

es:

�Th

e m

edia

n (F

erm

at-W

eber

Poi

nt) m

inim

izes

:

� =

=M i

ix

cent

erx

xd

Min

x1

2)

,(

arg

1x

2x

3x

4x

5x

� =

→=

M ii

xce

nter

xxM

inx

1

arg

� =

=M i

ice

nter

xM

x1

1

{}

Mi

ix,..

.,1=

� =

=M i

ice

nter

xM

x1

1

� =

=M i

ix

cent

erx

xd

Min

x1

2)

,(

arg

2��

�� ��

��$������3 ����

�����$

������

)�$�)

���������

� =

=M i

ix

med

ian

xx

dM

inx

1)

,(

arg

1x

2x

3x

4x

5x

� =→→

=M i

ii

xm

edia

n

xxxxM

inx

1

arg

[]

mx

EM

inm

mm

edia

n−

=[

]2

mx

EM

inm

mm

ean

−=

� =

=M i

ix

med

ian

xx

dM

inx

1)

,(

arg

� =

→=

M ii

cent

erx

x1

0� =

→→

=M i

ice

nter

ice

nter

xx

xx

10

69/

69/

2� ����2��

�� ��

��$�����

�)�� ��

� ����"

��

��

���

��

�� �

$�

3�

��

�El

ie C

arta

n ha

s pr

oved

that

the

follo

win

g fu

nctio

nal :

is s

tric

tly c

onve

xe a

nd h

as o

nly

one

min

imum

(ce

nter

of

mas

s of

Afo

r dis

trib

utio

n da

) for

a m

anifo

ld o

f neg

ativ

e cu

rvat

ure

��

��

� �

6�

C

�H

erm

ann

Kar

cher

has

pro

ved

the

conv

erge

nce

of th

e fo

llow

ing

flow

to th

e C

ente

r of M

ass

:

E. J

. Car

tan

H. K

arch

er

A

daa

md

mf

),

(:

2�

Μ

()

)(

)0(

avec

)(

.ex

p)

(1

nn

nn

mn

nn

mf

mf

tt

mn

−∇=

∇−

==

γ�

)(

exp

1

−−

=∇

Am

daa

f

70/

70/

9 ��

����������������

����)���-�

� �*���

59��

�����

���� ��

���� �

�M

auric

e R

ené

Fréc

het,

inve

ntor

of C

ram

er-R

ao b

ound

in 1

939,

has

als

o in

trod

uced

the

entir

e co

ncep

t of M

etric

Spa

ces

Geo

met

ryan

d fu

nctio

nal

theo

ry o

n th

is s

pace

(any

nor

med

vec

tor s

pace

is a

met

ric s

pace

by

defin

ing

bu

t not

the

cont

rary

). O

n th

is b

ase,

Fré

chet

has

then

ext

ende

d pr

obab

ility

in

abst

ract

spa

ces.

�In

this

fram

ewor

k, e

xpec

tatio

n

of a

n ab

stra

ct p

roba

bilis

tic

varia

ble

w

here

lies

on

a m

anifo

ld is

intr

oduc

ed b

y Em

ery

as a

n ex

pone

ntia

l bar

ycen

ter:

�In

Cla

ssic

al E

uclid

ean

spac

e, w

e re

cove

r cla

ssic

al d

efin

ition

of E

xpec

tatio

n E[

.]:

xy

yx

d−

=),

(M

. R. F

réch

et, “

Les

élém

ents

alé

atoi

res

de n

atur

e qu

elco

nque

dan

s un

esp

ace

dist

anci

é”, A

nnal

es d

e l’I

nstit

ut H

enri

Poi

ncar

é, n�1

0, p

p.21

5-31

0, 1

948

()[

]x

gE

b=

)(xg

x()

()

0)

(ex

p1

=

−dx

Px

gM

b []

==

�−

=�

∈−

nn

RX

Rp

ndx

xp

xg

dxP

xg

xg

Ep

qq

Rq

p)

()

()

()

()

()

(ex

p,

1

M. E

mer

y &

G. M

okob

odzk

i, “S

ur le

bar

ycen

tre d

’une

pro

babi

lité

sur u

ne v

arié

té”,

Sém

inai

re d

e Pr

oba.

XXV

, Lec

ture

s no

te in

Mat

h. 1

485,

pp.

220-

233,

Spr

inge

r, 19

91

71/

71/

$��

����$�)

�������0��

�� ��

�����:

;�<

3�

��

�$�

@��

� �

����

��-

���

�8�

�7�

$��

���

�3�

� ��

����

-7

*�.

�So

lutio

n gi

ven

by K

arch

er F

low

with

Info

rmat

ion

Geo

met

ry m

etric

3�

���

��*

6�

��

�>�

�1

� �

-

���.

�$�

@��

��

���

���

-7

*�.

�Ph

D Y

ang

Le s

uper

vise

d by

Mar

c A

rnau

don

(Uni

v. P

oitie

rs/T

hale

s)

()

()

��

=

−−

=

==

N kk

N kk

Xm

oyen

neX

BX

BX

dM

inAr

gX

1

22/1

2/1

1

2.

.lo

g,

()

2/1lo

g2/1

11

2/12/1

n

XB

X

nn

Xe

XX

N kn

kn

�=

=

−−

+

ε ()

()

��

=

−−

=

==

N kk

N kk

Xm

edia

neX

BX

BX

dM

inAr

gX

1

2/12/1

1.

.lo

g,

()

()

{} k

nn

nX

BX

XB

X

nn

BX

kS

Xe

XX

nSk

nk

n

nk

n

≠=

�=

∈−

−−

+/

with

2/1

log

log

2/11

2/12/1

2/12/1

ε

72/

72/

$�)

������

���$

�����

�)��

� ��

� 12

� ���1� .�

���

�G

radi

ent f

low

on

Surf

ace/

Man

ifold

Gra

dien

t Flo

w :P

ushe

d by

Sum

of

Nor

mal

ized

Tan

gent

vec

tors

of

Geo

desi

cs

Fréc

het-K

arch

er B

aryc

ente

r :

S

um o

f Nor

mal

ized

Tan

gent

vec

tors

of

Geo

desi

cs is

equ

al to

zer

o

−−

−=

∇�

Μ∈

Amm

Min

A

daaa

hda

am

dm

h)

(ex

p)

(ex

p)

,(

21:

11

Sum

of N

orm

aliz

ed

Tang

ent V

ecto

rs(

) )(

exp

exp

.ex

p1

11

1�� ��

�� ��=

� =−−

+

M kk

m

km

mn

xxt

mnn

n

Com

pute

poi

nt o

n th

e su

rface

In th

e di

rect

ion

of s

um o

f Nor

mal

ized

Ta

ngen

t Vec

tor

73/

73/

$��

����

� ��

� ��� ��

����

()

()

2/1lo

g2/1

2/12/1

2/12/1

2/12/1

)(

Xe

XX

XB

XX

tX

BX

tt

kk

k−

==

−−

γ

()

2/12/1

2/12/1

0lo

g)

(X

XB

XX

dtt

dk

tk

−−

==

γ

()

��

==

−−

==

� ��� ��

=N k

N kk

tk

XX

BX

Xdt

td

1

2/1

1

2/12/1

2/10

0lo

g)

1B2B

3B

4B

5B

6B

X

74/

74/

Air

Sys

tem

s D

ivis

ion

� �)���

�����"���'�

�9���%��� ���� �

�T

hefo

llow

ing

stru

ctur

eof

aM

atri

xM

isin

vari

antb

yth

isflo

w:

�w

hereJ

isan

tidia

gona

lmat

rix.

We

can

prov

eth

at:

�is

acl

ose

subm

anifo

ldof

(Her

miti

anPo

sitiv

eD

efin

iteM

atri

x)an

dsu

b-gr

oup

of

�Fo

rSi

egel

met

ric,

this

set

isco

nvex

ina

“Geo

desi

c”m

eani

ng(a

llge

odes

ics

betw

een

two

elem

ents

ofE

are

inE)

.The

nba

ryce

nter

ofN

elem

ents

ofE

isinE.

()

2/1lo

g2/1

11

2/12/1

t

AB

A

tt

Ae

AA

N kt

kt

�−

+=

−−

MJM

J=

{}

MJM

JC

HPD

ME

n=

∈=

/)( )

(CH

PDn

)(C

GL

n

MM

JMJ

JMJ

eJ

Jee

e=

==

−−

11

����������� �����

Thal

es A

ir S

yste

ms

Dat

e� ���

����"���� �'��

9���%�

��:���� ��

���;'��

:<�

' ��

���@

� ����

�#��'���

��

76/

76/

�C

ovar

ianc

es m

atric

es a

re s

truct

ured

mat

rices

that

ver

ify th

e fo

llow

ing

cons

trai

nts

: �

Toep

litz

Stru

ctur

e (fo

r sta

tiona

ry s

igna

l) :

�H

erm

itian

stru

ctur

e :

�P

ositi

ve D

efin

ite S

truct

ure

:

�H

ow to

bui

lt a

flow

that

pre

serv

es th

e To

eplit

z st

ruct

ure

?

������ ��

������ ��

=

������ ��

������ ��

=

01

21

* 10

* 21

2

* 10

1

*1

* 2* 1

0

01

21

* 10

* 21

2

* 10

1

*1

* 2* 1

0

rr

rr

rr

rr

rr

rr

rr

rr

rr

rr

rr

rr

rr

rr

rr

rr

R

n

n

n

n

n

��

��

���

��

��

���

1−nR conj

ugua

te

and

d tr

ansp

ose

:

with

+

=+

nn

RR

()

00

det

su

ch th

at

,0,

>�

=−

∀>

∈∀

λn

nn

n�.

IR

ZR

ZC

Z

3))

�����

���2��

�� �������'�

�9���%��� ���� �

[]

kk

nn

rz

zE

n=

∀−*

,

If yo

u re

mem

ber f

irst e

xam

ple

on «

Rig

ht T

riang

le»,

in th

is c

ase

Pyt

hago

re c

onst

rain

t is

repl

aced

by

HP

D &

Toe

plitz

con

stra

ints

77/

77/

Air

Sys

tem

s D

ivis

ion

2��

9��5

�2� �

��� �$

���� � ������

��������$�)

��

G�

��

��

�"�

���3

���

�C

ompl

ex C

ircul

ar M

ultiv

aria

te M

odel

:

�R

adar

Mod

el :

�C

ompl

ex a

utor

egre

ssiv

e m

odel

:

�C

lose

Lin

k w

ith Is

sai S

chur

’s a

lgor

ithm

(187

5-19

41)

�[A

lpay

] D. A

lpay

, « A

lgor

ithm

e de

Sch

ur, e

spac

es à

noy

au re

prod

uisa

nt e

t th

éorie

des

sys

tèm

es »

, Pan

oram

as e

t syn

thès

e, n�6

, Soc

iété

Mat

hém

atiq

ue d

e Fr

ance

, 199

8

[]

()(

)[

]n

nn

nn

nn

RR

Trn

nn

n

RR

Em

Ym

YR

eR

RY

pn

n

=−

−==

+

−−

−−

ˆet

with

..

)(

)/

(1

[]

ki

kk

kk

n

n

nn

nn

eib

ay

yyY

YY

ERm

φρ

=+

=��� ��

��� ��===

+

w

ith

Posi

tive

Def

inite

Her

miti

an

Toep

litz

proc

ess

mea

n ze

ro 0

1 �

[]

[]T

N NN

Nk

kn

n

N kn

kn

N kn

aa

Ab

bE

by

ay

)(

)( 1

20,

*

1

)(

...an

d w

ith

==

+−

=−

=−

�σ

δ

78/

78/

Air

Sys

tem

s D

ivis

ion

����

#���� �5

��� �

��� �����23(���)

��

�W

e ca

n ex

ploi

t Blo

ck s

truc

ture

of c

ovar

ianc

e m

atrix

to

com

pute

Rao

met

ric fo

r CA

R m

odel

:�

Rao

met

ric fo

r Com

plex

Mul

tivar

iate

Gau

ssia

n m

odel

of z

ero

mea

n :

�B

lock

stru

ctur

e of

cov

aria

nce

mat

rix fo

r CA

R m

odel

:

()

()

()

TAB

TrB

AA

AA

dRR

RdR

RTr

ds=

==

=−

−−

,et

,

with

2

22/1

2/12

12 (

)(

)(

)(

)0

det

et

lo

glo

g,

2/11

22/1

11

22

2/11

22/1

12

12

=−

==

−−

=

−−

�I

RR

RR

RR

RR

Dn k

λ

� ��� ��

+=

+ −−

−− −

−−

+ −−

−−

11

11 1

11

11

11

..

..

nn

nn

nn

nn

nn

AA

RA

AR

αα

αα

� ��� ��

−−

+=

−−

−+ −

−−

+ −− −

11

1

11

11

11 1

..

..

nn

n

nn

nn

nn

nR

AR

RA

AR

AR

α

[]

*)

()

(1

11 1

21

.0

01

00

10

0

w

here

1.

0

and

.

1w

ith

VV

AA

An

nn

nn

nn

��� ��

��� ��=

� ��� ��

+ � ��� ��

=−

=−

− −−

− −−

�μ

αμ

α

79/

79/

Air

Sys

tem

s D

ivis

ion

����

#���� �5

��� �

��� �����5

���)

�)���7��

1 ����

�M

atrix

or

that

is

used

for R

ao M

etric

com

puta

tion

has

sam

e bl

ock

stru

ctur

e�

Inva

rianc

e of

Blo

ck s

truct

ure

:

�Th

is b

lock

stru

ctur

e to

com

pute

ext

ende

d ei

gen-

valu

es :

2/1)1(1

)2(2/1)1(

..

nn

nR

RR

−+

2/1)1(

)2(2/1

)1(.

.−

+−

nn

nR

RR

[]

)1(1)2(1

1-n)1(1

)2(1

2/1)1(1

)1(1

1

11

11

11

11

12/1)1(

1)2(

2/1)1(

an

d

.

.w

ith

..

..

..

−−−

−+

−−

+ −−

−−

−−

+ −−

−−

+

=−

=

� ��� ��

==

Ω

nnn

nn

nn-

nn

nn

nn

nn

nn

nn

n

AA

RW

WW

WW

RR

R

ααβ

α

ββ

ββ

()

()

()

� �

� ��� ��

−Λ

−=

=−

+−

=

−+ −

−−

−−

− =−

−+ −

−−

11

11

)(

11

)(

)( 1,)(

1 1)

()1

(

2)1

(1

)(

11

)(

)(

)(

..

..

.1

0.

..

nn

nn k

nn

n kn kn k

n in k

n i

n in

n kn

nn k

n kn

WU

IU

XX

XW

F

λλ

λλ

λβ

βλ

λ

80/

80/

Air

Sys

tem

s D

ivis

ion

�W

e ca

n de

duce

an

algo

rithm

that

cou

ld b

e pa

ralle

lized

acc

ordi

ng to

C

AR

mod

el o

rder

for e

ach

exte

nded

eig

en-v

alue

s :

()

()

()

� �

� ��� ��

−Λ

−=

=−

+−

=

−+ −

−−

−−

− =−

−+ −

−−

11

11

)(

11

)(

)( 1,)(

1 1)

()1

(

2)1

(1

)(

11

)(

)(

)(

..

..

.1

0.

..

nn

nn k

nn

n kn kn k

n in k

n i

n in

n kn

nn k

n kn

WU

IU

XX

XW

F

λλ

λλ

λβ

βλ

λ

()

..

.1

)( 1,)(

2/11 1

2)

()1

(

2)1

(1

2)(

)( 1,)(

n kn kn i

n kn i

n in

n kn kn k

XXX

WXX

−− =

−+ −

��� ��

��� ��

−+

→�

λλ

λ

()(

)�− =

−+ −

−−

>−

+=

=1 1

2)1

(

2)1

(1

)1(

12

)( 1,

)(

1)

(

1.

..

1.

n kn k

n kn

n kn

n kn k

nn

XW

X

η

ηβ

ηβ

∂ηη

[]

nn

nn

n nn n

n nTr

Ω<

<<

<<

<<

<−

−− −

)( 1

)1( 1

)( 2

)(

1)1

(1

)(

...0

λλ

λλ

λλ

[]

[]

� ��� ��

� ��� ��

+

−−

−1

11

1

11

.n

nn

nn

WW

TrTr

β

)()4(λ

F

stric

ly in

crea

sing

cur

ve o

n ea

ch in

terv

al :

)3( 1λ)3( 2λ

)3( 3λ

[]

Tr

)4( 1λ

Ren

orm

alis

atio

n at

eac

h it

erat

ion

:

avec

)4( 3λ)4( 2λ

)4( 4λ

()

()

.A1

1

n 1=k)

( 1,)(

1)

()

(

1-n

1

1)

()

(

� �

�� ���� ��

= � ��� ��

=�� ��

�� ��

�−

=

n kn kn k

n

n k

n kn

XXF

F

∂ηλ

∂ηλ

Inte

rpre

tatio

n in

term

of p

roje

ctio

n of

CAR

vec

tor :

0

����

#���� �5

��� �

��� �����5

���)

�)���7��

1 ����

81/

81/

Air

Sys

tem

s D

ivis

ion

(���

�� ��-59 �����

���� �(

���$

�� ��

�If

we

use

bloc

k st

ruct

ure

mat

rix, w

e ca

n co

mpu

te

Rao

met

ric a

t mod

el o

rder

n fr

om m

odel

ord

er a

t n-1

�R

ecur

sive

Exp

ress

ion

of R

ao M

etric

:

�W

e ca

n ob

serv

e th

at :

�Th

is c

ould

be

inte

rpre

ted

with

Info

rmat

ion

Geo

met

ry &

Fis

her m

atrix

()

[]

11

11

2

112

12

12

..

..

−−

+ −−

−−−

−+

�� ���� �� +

==

nn

nn

nnn

nn

ndA

RdA

dds

dRR

Trds

ααα

� ��� ��

� ��� ��

� ��� ��

=+

�� ���� ��

−−

+

−−

−+ −

−−−

1

1

11

1

11

11

1

2

11lo

g0

01 .

log

..

.n

n

nn

n

nn

nn

nnn

Ad

RA

ddA

RdA

αα

ααα

()� ��

� ��+

� ��� �� � ��

� ��=

−−

+

−1

11

1

1

1

1

00

1lo

glo

g

nn

n

n

n

n

Ri

AA

αα

[]

� �

==

−−

−−

−−

11

11

11

logˆ

nn

nn

n A

RA

IR

nA

α

α

82/

82/

Air

Sys

tem

s D

ivis

ion

0��

1���

��� ��

��8�

� ��(�

���������7�

��� ��9

�If

we

cons

ider

Cho

lesk

y de

com

posi

tion

of c

ovar

ianc

e m

atrix

:�

Cho

lesk

y de

com

posi

tion

(Gol

dber

g in

vers

ion

algo

rithm

) :

�A

ll di

strib

utio

n of

n-d

imen

sion

nal v

aria

ble

is a

ssoc

iate

d w

ith A

ffine

Gro

up. I

t is

the

elem

ent s

uch

that

its

actio

n on

vec

tor

Is tr

ansf

orm

ed to

rand

om v

ecto

r :

�Th

is re

pres

enta

tion

of A

ffine

Gro

up e

lem

ents

cou

ld b

e co

nsid

ered

as

non

sym

met

ric s

quar

e ro

ot o

f Sie

gel G

roup

ele

men

t :

()

()

+−

−−

−−

+ −−

−−

+ −+

ΩΩ

� ��� ��

Ω−

=

� ��� ��

−=

==

Ω

2/11

2/11

12/11

1

2n

11

11

12

1

.

and

0

11

with

.

1.

1.

.

nn

nn

nn

nn

nn

nn

nn

nn

n

AW

AA

AA

WW

R

μ

μα

),0(

~n

nI

NZ

),

(~

nn

nA

NX

Ω

� ��� ��

= � ��� ��

= � ��� �� � ��

� ��Ω

−−

−−

XA

ZZ

An

nn

n

1.1

1.

01

12/11

2/11

1

� ��� ��

+ −−

−−

+ −

11

11

1

.1

nn

nn

n

AA

AA

83/

83/

Air

Sys

tem

s D

ivis

ion

(���

�� ��-59 �����

������

������

#�:� �

7��

��

�W

e co

nsid

er e

xpre

ssio

n of

Kul

lbac

k D

iver

genc

e fo

r Mul

tivar

iant

e G

auss

ian

Law

of z

ero

mea

n :

�K

ullb

ack

dive

rgen

ce is

giv

en b

y :

�If

we

cons

ider

recu

rsiv

e re

latio

n :

�K

ullb

ack

Div

erge

nce

is g

iven

recu

rsiv

ely

by :

()

[]

[] n

TrR

RK

nn

nn

−Σ

−Σ

=ln

21,

)2()1(

[]

� ��� �� � ��

� ��−

−=

==

−=

� ��� ��

==

Σ

−−

+−

+

−−−

−+

+ −−

−−

−−

+ −−

−−

+

)1(1

1)1(1

2)1(

)1(

)1()1(

)1()1(

)1(1)2(1

1-n)1(1

)2(1

)1(1

1-n

11

11

11

11

12/1)1(

1)2(

2/1) 1(

00

101

1

1

..

with

and

.

V w

here

..

..

..

nnn

n

n

nn

nn

nnn

nn

nn

nn

nn

nn

nn

nn

n

FIA

F

FF

RA

AF

VV

VV

RR

R

μ

ααα

β

ββ

ββ

11.

−−

Σ=

Σn

nn

β[

][

][

] 11

11

.1.

−+ −

−−

++

Σ=

Σn

nn

nn

VV

TrTr

β

()

()

()

()

[]

11

11

1)2(1

)1(1

)2()1(

ln.

.1

21,

,−

−+ −

−−

−−

−+

−+

=n

nn

nn

nn

nn

VV

RR

KR

RK

ββ

β

84/

84/

Air

Sys

tem

s D

ivis

ion

' ��

���@

� ����

�#��'���

������� ���

���"���"

����

����� �%�

����

�A

ll To

eplit

zH

erm

itian

Posi

tive

Def

inite

Mat

rix c

anbe

para

met

eriz

edby

Ref

lect

ion/

Verb

luns

kyC

oeffi

cien

ts:

�B

lock

stru

ctur

e of

cov

aria

nce

mat

rix &

Ver

blun

sky

Par

amet

eriz

atio

n:

�V

erbl

unsk

y/Tr

ench

The

orem

: Exi

tenc

eof

diff

eom

orph

ism

ϕ

� ��� ��

+=

+ −−

−− −

−−

+ −−

−−

11

11 1

11

11

11

..

..

nn

nn

nn

nn

nn

AA

RA

AR

αα

αα

� ��� ��

−−

+=

−−

−+ −

−−

+ −− −

11

1

11

11

11 1

..

..

nn

n

nn

nn

nn

nR

AR

RA

AR

AR

α

[]

*)

()

(1

1

1- 00

1 12

1

.0

01

00

10

0

w

here

1.

0

an

d

.1

with

VV

AA

A

P

nn

nn

nn

n

��� ��

��� ��=

� ��� ��

+ � ��� ��

=

=−

=

−− −

− −−

�μ

αα

μα

()

{} 1

/w

ith

,...,

,

:

11

0

1*

<∈

=

×→

−+

zC

zD

PR

DR

THPD

nn

nn

μμ

ϕ�

S. V

erbl

unsk

y(P

hD st

uden

t of

Litt

lew

ood)

Iwas

awa

(Lie

Gro

up

The

ory)

85/

85/

� ��� �

��������'�

�9���%��� ���� ����

�@� ����

�#��2��

�������

��

�C

onfo

rmal

Info

rmat

ion

Geo

met

ry m

etric

(met

ric =

Hes

sian

of

Entr

opy)

:

�En

trop

y Φ ΦΦΦ

as K

ähle

r pot

entia

l:

�C

onfo

rmal

met

ric o

n Ve

rblu

nsky

par

amet

eriz

atio

n:

[]T

nn

P1

10

)(

−=

μμ

θ�

)*(

)(

2~

n jn i

ijgθ

θ∂

∂Φ

∂≡

()

()

[]

[]

..ln.

1ln).

(.

log

det

log

,~

0

1 1

21

0Pe

nk

ne

R)

P(R

n kk

nn

πμ

π−

−−

−=

−=

Φ�− =

with

[]

()

�− =

+

−−

+�� ��

�� ��=

=1 1

222

2

00)

()

(2

1)(

.n i

iin

ijn

n

di

nPdP

nd

gd

dsμμ

θθ

E. K

ähle

r

1t w

ith

coef

ficie

nV

erbl

unsk

y :

<k

μ

86/

86/

Air

Sys

tem

s D

ivis

ionEric

h K

ähle

r…

Ent

ropy

U=-

logd

et[R

] , f

or C

ompl

ex A

utor

egre

ssiv

e m

odel

, can

be

cons

ider

ed a

s a

Käh

ler P

oten

tial t

hat

depe

nds

on re

flect

ion

coef

ficie

nt. T

his

expr

essi

on is

exa

ctly

the

sam

e th

an th

e fir

st p

oten

tial p

ropo

sed

by

E. K

ähle

r

« K

ähle

r Eric

h, M

athe

mat

ical

Wor

ks »

, Edi

ted

by R

. Ber

ndt a

nd O

. Rie

men

schn

eide

r, B

erlin

, Wal

ter d

e G

ruyt

er, i

x,

2003

��������- ��

��

��� ���9

� &�+

K,?&����

�� 7

87/

87/

(�7�

�� �%�)

�@� ����

�#��2��

�������

����(�7

��� �%�

)���

7�3�7� ���

•R

egul

ariz

ed B

urg

Algo

rithm

(TH

ALE

S Pa

tent

)

�� +

−=

−+

=

� �

=

+==

−=

+−

+−

+−

−−

=

===

==

−−

−−

− −−

+=

− =

−−

+=

− =

− −−

−−

=

��

��

)(

.)1

()

(

)1

(.

)(

)(

11

,

.

1

).(

)2.(

w

ith

..2

)1(

)(

1

..

.2)1

().

(2

t

o1

For

: (n

)te

p .

1

)(

.1

ech.

)nb

. :

(N 1

, )

()

()

(f

:tio

n In

itial

isa

.

1*

1

11

)(

)*1(

)1(

)(

)( 0

22

1)

(

1

1 0

2)1

()

(2

12

1

1

1 1

)1(

)1(

)(

*1

1

)0( 0

1

20

00

kf

kb

kb

kb

kf

kfa

,...,n

-k=

aa

aa

nk

ak

bk

fn

N

aa

kb

kf

nN

Mn

Sa

kz

NP

,...,N

k=

kz

kb

k

nn

nn

nn

nn

nn n

nk

nn

n kn kn

n kN n

k

n k

n kn k

nn

N nk

n k

nk

nn k

n kn

n

n

N k

μμ

μμ

πγ

ββ

βμ

Reg

ular

izat

ion

88/

88/

$�)

������

�$�)

����(�

���������@�

�����#�

�2��

�������

���μ μμμ

k

n,3μ

n,1μn,2μ

nm

edia

n,μ

Cla

ssic

al K

arch

er F

low

in U

nit D

isk

1,3

+nμ

1,1

+nμ

Dua

le K

arch

er F

low

{}

�l/

av

ec

���

wl,n

m

lkk

k,n

k,n

nn

<=� ≠=1

nn

med

ian

w=

+1,

μ

* nk,

n

nk,

nk,

n.w

�w

��

−−

=+

11

1,2

+nμ

* nm

edia

n,n

nm

edia

n,n

med

ian,

nw

�w

��

++

=+

11

89/

89/

$�)

������

�$�)

����(�

���������@�

�����#�

�2��

�������

���μ μμμ

#

90/

90/

Air

Sys

tem

s D

ivis

ion

��)�

��� ��$��

��;�'����� �

<

�A

nat

ural

way

to d

efin

e th

e «

expe

ctat

ion

» E(

Y) o

f a ra

ndom

var

iabl

e Y

is to

use

ge

nera

lizat

ions

of t

he la

w o

f lar

ge n

umbe

rs.

�G

iven

any

seq

uenc

e

o

f poi

nts,

we

defin

e a

new

seq

uenc

e

o

f po

ints

by

indu

ctio

n on

n a

s fo

llow

(ind

uctiv

e m

ean)

:

�st

rong

ly d

epen

d on

per

mut

atio

ns b

ut

�Ex

tens

ion

of s

cala

r cas

e :

()

2/11

/12/1 1

2/1 12/11

/11

11

an

d

−− −

− −−

−=

•=

=n

nn

nn

nn

nn

nS

SB

SS

BS

SB

S

{}n k

kB1=

{}n k

kS1=

nn

nn

nn

bn

sn

bs

sb

s1

11

an

d

1/1

11

1+

� ��� ��

−=

•=

=−

()

()

()

11

21

),(

BV n

SB

EE

n≤

δ

91/

91/

����

�������399

���

���3 ���

)������"

Ran

dom

sel

ectio

n at

eac

h ite

ratio

n of

1 p

oint

in th

e di

sk +

driv

en o

f the

flow

al

ong

the

geod

esic n

nra

nd

nn

rand

nn�

w),

(

),(

μμ=

() �� ��

�� ��=

−−

+)

(ex

p

exp

.ex

p)

(1

)(

1

1n

rand

m

nra

ndm

nm

nxx

tm

nn

n

M. A

rnau

don,

C. D

ombr

y, A

.Pha

n, L

.Yan

g, ”

Stoc

hast

ic a

lgor

ithm

s fo

r com

putin

g

mea

ns o

f pro

babi

lity

mea

sure

s”, h

ttp://

hal.a

rchi

ves-

ouve

rtes

.fr/h

al-0

0540

623

92/

92/

nn

rand

nn

rand

nn�

w),

(

),(

μμ=

() �� ��

�� ��=

−−

+)

(ex

p

exp

.ex

p)

(1

)(

1

1n

rand

m

nra

ndm

nm

nxx

tm

nn

n

3 ���

)�������

����������"

93/

93/

3 ���

)������"

��5���

)�)���

����

9�����9

���

New

Arn

audo

n St

ocha

stic

Flo

w o

n th

e sp

here

M. A

rnau

don,

L. M

iclo

, ”M

eans

in c

ompl

ete

man

ifold

s:

uniq

uene

ss a

nd a

ppro

xim

atio

n”,

http

://ar

xiv.

org/

abs/

1207

.323

2

94/

94/

�C

as d

u di

sque

de

Poin

caré

:�

G+

le s

ous-

grou

pe d

’aut

omor

phis

mes

pré

serv

ant l

’orie

ntat

ion

avec

e

t

�G

opè

re s

ur

de m

esur

e de

pro

babi

lité

de

�P

rinci

pe :

ass

igne

r à to

ute

mes

ure

de p

roba

bilit

é

sur

un p

oint

tel q

ue

e

st c

onfo

rme

et v

érifi

e:

�Il

y a

une

seul

e m

aniè

re c

onfo

rme

d’as

sign

er à

cha

que

mes

ure

de p

roba

bilit

é

sur

u

n ch

amp

de v

ecte

urs

s

ur D

tel q

ue:

avec

�� ���

�� ��2��

�� �

��)�

������

����;:��

�)�1-�

��&J

<

{} 1

/ <

∈=

zC

zD

za

az

z*

1−−

λ�

1=

z1

<a

()1 S

P{

} 1 /

1=

∈=

zC

zS μ

1 S(

)D

B∈

μ(

μB

0)

(.

0)

(1

=⇔

= S

dB

ζμ

ζμ

μ1 S

μξ0

)(

.)0(

1

== S

μζ

ξ μ

()

()

()

0)

(*

11

)0()

('1)

(1

2)

(*=

�� ���� ��

−−

−=

= S

gw

dw

ww

wg

ww

ζμ

ζζ

ξξ

μμ

zw

wz

zg w

*1

)(

−−

=

95/

95/

�C

as d

u di

sque

de

Poin

caré

�pe

ut ê

tre é

crit

en fo

nctio

n de

qui e

st le

vec

teur

tang

ent

de la

odés

ique

en

poin

tant

ver

s

:

�E

n gé

omét

rie d

e P

oinc

aré

du d

isqu

e un

ité, l

e ch

amp

de v

ecte

ur

e

st le

gra

dien

t d’

une

fonc

tion

dont

les

ligne

s de

niv

eau

sont

les

horo

cycl

es ta

ngen

ts à

en

:

�� ���

�� ��)��� .�

�����

.)����F���

���

�� ��)����

�����

)(z

μξ)

(zζξ

Dz∈

1 S∈

ζ

=1

)(

)(

)(

S

dz

μξ

ξζ

μ

ζξζh

1 S1 S

∈ζ

[]

−=

�� ��

�� ��

−−==∇

=

−→

11

1

)(

),

()

,0()

(1

log

21)

(

)(

)(

:

avec

122

Sr

S

S

dr

zd

rd

Lim

dz

zz

h

dz

hz

hh

ζμ

ζζ

μζ

ζμ

ξ μ

ζμ

μμ

96/

96/

�� ���

�� ��)��� .�

�����

.)����F���

���

�� ��)����

�����

97/

97/

�� ���

�� ��)��� .�

�����

.)����F���

���

�� ��)����

�����

98/

98/

�D

u ce

ntre

de

mas

se d

e C

arta

n au

bar

ycen

tre

de B

usem

ann

�E

lie C

arta

n a

mon

tré p

our l

es v

arié

tés

riem

anni

enne

s si

mpl

emen

t con

nexe

s à

cour

bure

nég

ativ

e ou

nul

le, l

’exi

sten

ce e

t l’u

nici

té d

u m

inim

um d

e :

�La

fonc

tion

suiv

ante

est

pro

pre,

sim

plem

ent c

onne

xe e

t atte

int s

on m

inim

um :

�Lo

rsqu

e

tend

ver

s l’i

nfin

i en

conv

erge

ant v

ers

un p

oint

d

e

, la

fonc

tion

dist

ance

à

con

vena

blem

ent n

orm

alis

ée c

onve

rge

: fon

ctio

n de

Bus

eman

n :

�Fi

xons

une

orig

ine

.

Soi

t

une

mes

ure

finie

sur

d

ont l

e su

ppor

t a

plus

de

2 po

ints

. La

fonc

tion

stric

tem

ent c

onve

xe s

uiva

nte

atte

int s

on m

inim

um

en u

n un

ique

poi

nt, i

ndép

enda

nt d

u ch

oix

de O

et a

ppel

é ba

ryce

ntre

de

Bus

eman

n :

�� ���

�� ��)����

�����

���� �

� �.�.�)*��)

��� )

()

nH

zd

zx

dx

)(

,2

ν�

()

nH

zd

zx

dx

)(

nH∂

z)

,(

),

(lim

)(

zO

dz

xd

xB

z−

=→

θθ

nH

O∈

νn

H∂

=n

H

dB

x)

()

νβ

θν

99/

99/

2��

9� ����

����$

�������$

�)����:�

99�� ��

9��� ��

Raw

Dop

pler

S

pect

rum

Mea

n D

oppl

er

Spe

ctru

mM

edia

n D

oppl

er

Spe

ctru

m

Ran

ge a

xis

Dop

pler

axi

s

No

pres

erva

tion

of d

isco

ntin

uitie

s

Per

turb

atio

n by

ou

tlier

100

/10

0/

�D

iffus

ion

Four

ier E

quat

ion

on 1

D g

raph

(sca

lar c

ase)

�A

ppro

xim

atio

n pa

r un

Lapl

acie

n di

scre

t :

with

arit

hmet

ic m

ean

of a

djac

ent p

oint

s :

�D

icre

te F

ourie

r Hea

t Equ

atio

n fo

r Sca

lar V

alue

s in

1D

:

�B

y A

nalo

gy, w

e ca

n de

fine

Four

ier H

eat E

quat

ion

in 1

D g

rapf

of

HD

P(n)

mat

rices

:

()

nn

nn

nn

uu

xxu

uxu

ux

tuxu

tu−

∇=

� ��� ��

∇−−

∇−∇

=∂∂

�∂∂

=∂∂

−+

ˆ2

12

11

2

2

() 2/

ˆ1

1−

++

=n

nn

uu

uJ.

Fou

rier!��� ��99

���

���������

����9�

�� �������

� �� ��

����-8�

�����

2,

,1

,.2

w

ith

ˆ .).

1(xt

uu

ut

nt

nt

n∇∇

=+

−=

ρρ

()

()

()

tn

tn

tn

tn

tn

tn

tn

tn

tn

tn

tn

tn

tn

tn

XX

Xt

nt

n

XX

XX

XX

XX

XX

XX

XX

eX

Xt

nt

nt

n

,12/1

,12/1

,12/1

2/1,1

,12/1,1

2/1,1

,

2/1 ,2/1 ,

,2/1 ,

2/1 ,2/1 ,

ˆlo

g2/1 ,

1,

ˆw

ith

ˆ2/1 ,

,2/1 ,

−+

+− +

−− +

+

−−

+

==

==

−−

ρρ

101

/10

1/

���� �9���:����

��������:�9

9�� ��9�

�� ��

102

/10

2/

3����� �

9���:����

��������:�9

9�� ��9�

�� ��

103

/10

3/

2��

�������-5�

�9����$

���

� ������

��������/�"

����>�

����

����

�C

�Fi

sher

Info

rmat

ion

Mat

rix fo

r Gau

ss-L

apla

ce G

auss

ian

Law

:

()(

)�� ��

�� ��=

≥� ��

� ��−

−� ��

� ��=

−−

σθ

θθ

θθ

θσ

θm

IE

IT

an

d

)(

ˆˆ

w

ith

20

01 .

)(

12 (

)(

) �� ��

�� ��+

� ��� ��

=+

==

−2

22

22

222

2.2

.2.

σσσ

σθ

θθ

ddm

ddm

dI

dds

T

�R

ao-C

hent

sov

Met

ric o

f Inf

orm

atio

n G

eom

etry

H. P

oinc

aré

.2

σim

z+

=(

) 1

<+−

ωi

zi

z(

)222

2

1.8ωω

−=

�d

ds�

Fish

er M

etric

is e

qual

to P

oinc

aré

Met

ric fo

r Gau

ssia

n La

ws:

{}{

}(

)

)*2()1(

)2()1(

)2()1(

2

)2()1(

)2()1(

22

11

2

1)

,(

with

),

(1

),

(1

log

.2,

,,

ωω

ωω

ωω

δ

ωω

δω

ωδ

σσ

−−

=

�� ���� ��

−+=

mm

d

104

/10

4/

$�)

����������

��������"�

105

/10

5/

�Tr

avau

x de

Jac

ob B

urbe

a (1

984)

�O

n co

nsid

ère

une

dens

ité d

e pr

obab

ilité

pou

r une

var

iabl

e de

Cn

�Id

entif

icat

ion

avec

le c

arré

du

mod

ule

d’un

e fo

nctio

n no

rmal

isé:

�O

n co

nsid

ère

la p

seud

o-di

stan

ce d

e S

kwar

czyn

ski :

�O

n en

déd

uit l

’équ

ival

ence

ent

re m

étriq

ue d

e Fi

sher

et m

étriq

ue H

essi

enne

:

/������

� ��7.�

�.� ��

�������

�������)���*���� ������

()

()

()

()

()

()

∂∂

∂∂

=∂

∂∂

∂=

−)

(/

log

/lo

g/

)(

//

/)

(*

*1

td

zz

tp

zz

tp

zt

pt

dz

zt

pz

zt

pz

tp

zg

ji

ji

jiμ

μ 22/

1*

1

2lo

g)

pp

dzdz

gz

dsj

i

N iji

∂=

=� =

()

1)

/(.

av

ec

)/

(/

2

2 =

=

μψ

ψ

z

zt

zt

p

()

()

)/

(.

,/

*z

tz

zK

zt

=(

)/

(),

/(

,*

wt

gz

tg

wz

K=

()

()

μψ

ψμ

ψψ

λ)

/(

,/

1)

()

/(

)/

(1

,w

tz

tt

dw

tz

tw

z−

=−

=

[]

()

*

1,

*2

22

22

2,

log

log

,)

,(

ki

N ji

ki

zw

Berg

man

dzdz

zz

zz

KK

KK

KK

dd

wz

dds

� =

=∂

∂∂

=∂∂

=∂

−∂∂

=−

==

μμ

ψψ

ψλ

()

()

),

(lo

g)

/(

log

)/

(lo

g)

/(

log

)/

(.

*,

/

)/

()

/(

**

2

zz

Kz

tg

zt

gz

tp

zt

zz

Kz

tg

zt

zt

p−

+=

� � �

==

ψ

ψ

*

*2

*

2

*

*2

*

2)

,(

log

)/

(lo

g)

,(

log

)/

(lo

g

ji

ji

jij

ij

iz

zz

zK

zz

zt

pE

gz

zz

zK

zz

zt

p∂

∂∂

=�� ��

�� ��

∂∂

∂−

=�

∂∂

∂−

=∂

∂∂

����������� �����

Thal

es A

ir S

yste

ms

Dat

e

0�"

�4! )� �)

����������

���7��

:�99

�� �(

����������2�3

(6�

;!�1�:

(12�3

(<

107

/10

7/

Air

Sys

tem

s D

ivis

ion

(�)�

�:����$��

�� ����

�G

��

� �

����

��

����/

���

��

"�

���

��$

���

���

��

���

$�

�R

efle

ctiv

ity :

Am

plitu

de o

f EM

Wav

e re

flect

ed e

nerg

y

�Ph

ases

var

iatio

ns d

ue to

Dop

pler

-Fiz

eau

�Po

lariz

atio

n of

EM

Wav

e

108

/10

8/

Air

Sys

tem

s D

ivis

ion

G�

��

��

���

��

����/

���

��

"�

���

��$

���

���

��

���

$�

�R

efle

ctiv

ity :

Am

plitu

de o

f EM

Wav

e re

flect

ed e

nerg

y

�Ph

ases

var

iatio

ns d

ue to

Dop

pler

-Fiz

eau

�Po

lariz

atio

n of

EM

Wav

e

(�)�

�:����$��

�� ����

2

arct

an12�� ��

�� ��

=ss

φ

2

arct

an2 2

2 1

3

�� ��

�� ��

+=

sss

τ

()

()

���� ��

���� ��

=

����� ��

����� ��

−+

=

���� ��

���� ��

=

τφ

τφ

τ 2si

n2

sin

2co

s2

cos

2co

s

Im2Re

20000

* 12

* 12

22

21

22

21

3210

ssss

zz

zzz

zz

z

ssss

S

� =

=� ��� ��

��� ��=

n kk

tére

flect

ivi

n

zn

Rzz

Z0

21

1�

()

ϕπ

π

π

λ

it

ff

i

tf

rc

Vr

rD

oppl

er

ee

cRt

uA

tz

et

ut

s

Vf

cV

cV

f

Dop

pler

r

22

2

1/

0

0

0 2.

)(

: R

ecei

ve

).(

)(

:Tr

ansm

it

.2/

1/

2

+

<<

� ��� ��

−≈=

≈−

=

Chr

istia

nD

oppl

er

Toep

litz

Her

miti

an P

ositi

ve D

efin

iteC

ovar

ianc

e m

atrix

���� ��

���� ��

= ���� ��

���� ��

��� ��

��� �� ��� ��

��� ��=

−+

01

1

1

01

11

01

1

cc

cc

cc

cc

c

zz

zzE

R

n

n

nn

��

���

��

109

/10

9/

(�)�

�� ���

����7���

��)���

�2�

� ����

��$�� �5

Dop

pler

Pro

cess

ing

(Tim

e C

ovar

ianc

e M

atri

x)

Ant

enna

Pro

cess

ing

(Spa

ce C

ovar

ianc

e M

atri

x)

STA

P Pr

oces

sing

(Spa

ce-T

ime

Cov

aria

nce

Mat

rix)

Pola

rim

etri

c R

adar

Pro

cess

ing

(Pol

arim

etry

Cov

aria

nce

Mat

rix)

110

/11

0/

2���������2�3

(�2��

��� � ����!�1�:

(12�3

(

I&Q

(z

1,…z n

)D

oppl

erFi

lters

Log|

.|

CA-

CFA

R

OS

-CFA

R>S

euil

OU

OU

Cas

es d

ista

nces

>SO

S-H

DR

-CFA

R(D

oppl

er S

pect

rum

Med

ian)

Cla

ssic

alC

FAR

Cha

in

Adva

nced

OS-

HD

R-C

FAR

Cha

in

Cas

ek-

1

Cas

ek

Cas

ek+

1Lo

g|.|

CA-

CFA

R

OS

-CFA

R>S

euil

OU

>S

I&Q

(z

1,…z n

)

Hig

h D

oppl

er

Res

olut

ion

Coe

ffici

ents

(P

0,k,�

1,k,…

�m

,k)

Cas

ek-

2

Cas

ek-

1

Cas

ek

Cas

ek+

1

Cas

ek-

2

� =−

−−

+=

m nk

n

med

ian

nk

n

k

med

ian

med

ian

mm

edia

nm

edia

nk

mk

km

edia

nn

thn

mP

Pm

PP

dist

1

2*

,

,,

2

,0

,02

,,1

,0,

,1,0

)]1(

arg

)[(

)ln

())

,...,

,(

),,..

.,,

((,

μμ

μμ

μμ

μμ

111

/11

1/

!�1�:

(12�3

(���

� ����

5���

����

�������

��

Méd

iane

Cas

es d

ista

nce

Cel

lule

sou

s te

st

Cel

lule

s ig

noré

es

Cel

lule

s am

bian

ce

� 1� 2

Cal

cul d

e la

di

stan

ce>S

OS

-HD

R-C

FAR

Vale

ur m

édia

ne

(OS

) pou

r am

bian

ce

112

/11

2/

(����

��������

�����

��������������� ���

)�2�����

� 1� 2

� 3

����������� �����

Thal

es A

ir S

yste

ms

Dat

e� ���

����"���� �'��

9���%1

����

#1'��9

���%���

:���� ��

���

;'�'��

:<

114

/11

4/

-5�����������'�

�9���%�����

#�'��9

���%���

������

��:�$�� ��

��

�Pr

evio

us r

esul

ts c

an b

e ex

tent

ed to

Blo

ck-T

oepl

itz M

atric

es :

� ��� ��

=

���� ��

���� ��

=+

+++

+0

,

01

1

01

10

1,

~~ R

RR

R

RR

RR

RR

RR

R

Rn

nn

p

n

n

np

��

���

*1

~

��� ��

��� ��=

n

n

RRV

R�

����� ��

����� ��

=

00

00

00

���

��

p

p

p

JJ

J

Vw

ith

115

/11

5/

-5�����������' ��

���@

� ����

�#��'���

�����

�'�'��

:�$�� ��

��

�Ev

ery

Toep

litz-

Blo

ck-T

oepl

itz H

PD m

atrix

can

be

para

met

rized

by

Mat

rix V

erbl

unsk

y C

oeffi

cien

ts:

�Ex

tens

ion

of T

renc

h/Ve

rblu

nsky

The

orem

: Exi

sten

ce o

f D

iffeo

mor

phis

m ϕ ϕϕϕ

�� ��

�� ��

+=

+−

+−

+n

nn

np

nn

nn

nn

pA

AR

AA

R

..

..

1 ,

11

αα

� ��

�� ��

−−

+=

++

+n

pn

np

np

nn

np

nn

np

RA

RR

AA

RA

R,

,

,,

1

1,

..

..

α

[]

����� ��

����� ��

+ �� ��

�� ��=

��� ��

��� ��

=

=−

=

−− −

− −+

p

pn

p

pn n

p

n npn

n n

n

-n

n nn n

n

IJ

AJ

JA

J

AA

AAA

R�

AA

*11

*1 1

1

1 1

01 0

1 11

.0

an

d

,.

1w

ith

αα

()

{} n

n n

nn

nn

IZZ

nH

erm

ZD

AA

RR

SDTH

PDTB

THPD

<∈

=

×→

+

− −

−×

/)(

Sw

ith

,...,

,

:1 1

1 10

1

ϕ

116

/11

6/

���� �

���������

��� ��$

�� ��

���-�

� �9���

��� ������

����

�K

ähle

r pot

entia

l def

ined

by

Hes

sian

of m

ulti-

chan

nel/M

ulti-

varia

te

entr

opy

:

()

()

()

()

[]

np

ji

np

np

np

RH

ess

gcs

teµ

RTr

cste

RR

,

,,

,lo

gde

tlo

g~

φ=

+−

=+

−=

()

[]

[]

0

1 1,

det

..lo

g.

det

log

).(

~R

en

AA

Ik

nR

n k

k kk k

nn

Φ+

−−

=�− =

+

()

[]

()

()

[]

�− =

+−

+−

+−

−−

−+

=1 1

11

20

10

2)

(.

n k

k kk k

k kn

k kk k

k kn

dAA

AI

dAA

AI

Trk

ndR

RTrn

ds

We

reco

ver

the

prev

ious

met

ric fo

r TH

PD m

atrix

!!

117

/11

7/

$����12��

������� 7�3�7� ���

���$

!�A

2

�M

ulti-

Cha

nnel

Bur

g A

lgor

ithm

:[

][

][

]

[]

[][

]

[]

[]

[]

� �

==

−=

� ��� ��

+� ��

� ��−

−=

++

−=

�+

==

� �

+−

=

−+

=

=

���� ��

���� ��

=

� �

+−

=

−=

+=

++

+

−+ =

++ =

++ =

++ +

−+ ++ +

+

+ ++

==

−− −

− −− −

−−

��

���

+ +

)(

)(

)(

)(

)1(

)(

with

)(

)(

)(

)(

)1(

)(

2

2

)(

)(

)(

w

ith

)(

)1(

)(

)1(

)(

)(

an

d

00

100

11

00

with

)

()

()

(

)(

)(

)(

,,..

.,,

0,,..

.,,

,...,

,

1

11

1

1 1

1*

1 1*

00

*1 11

1 11

0

0

*

0

*11

*1 2*1 1

1 11

21

12

1 1 1

kk

EP

kk

EP

kk

EP

kk

kk

kk

A

JJP

PJ

JPP

AJ

JPP

TrM

in

kZ

kk

kJ

JAk

kk

Ak

k

IA

Jl

nk

JZk

JAk

lk

Zk

Ak

IJ

JAJ

JAJ

JAA

AA

AA

AA

b nb n

b n

f nf n

f n

b nf n

fb n

nN k

b nb n

nN k

f nf n

nN k

b nf n

n n

f nb n

fbT

nfb n

n nb n

f nA

bf

f nn n

b nf n

b nn n

f nf n

nn l

n lb n

n l

n lf n

nn n

n nn n

n nn

nn n

nn n n

εε

εε

εε

εε

εε

εε

εε

εε

εε

εεεε

��

���

()

Sieg

eln n

nn n

n nD

isk

AI

AA

∈�

<+ +

++

+ ++ +

1 11

1 11 1.

118

/11

8/

$����"�:��

��9�

�������;���� 7�

���� ����

�<

�M

osto

w T

heor

em :

Ever

y m

atrix

of

ca

n be

dec

ompo

sed

:

whe

re is u

nita

ryis

real

ant

isym

met

ricis

real

sym

met

ricC

an b

e de

duce

from

�Le

mm

a :L

et

and

t

wo

posi

tive

defin

ite h

erm

itian

mat

rices

, th

ere

exis

t a u

niqu

e po

sitiv

e de

finite

her

miti

an m

atrix

such

that

:

�C

orol

lary

:if

i

s H

erm

itian

Pos

itive

Def

inite

, the

re e

xist

a u

niqu

e re

al s

ymm

etric

mat

rix

suc

h th

at :

M(

)C

nG

L,

SiA

eU

eM

=

U A S

AB

XB

XAX

=M

SS

Se

Me

M1

*−

=

119

/11

9/

()

()

()

MM

PP

PP

PP

S

PP

PP

Pe

eP

eP

SS

S

+−

−−

==

=�

=+

w

ith

log

. 21:

corr

olar

y Le

mm

a

2/12/1

2/1*

2/12/1

2/12/1

2/1*

2/12/1

22

12

*

$����"�:��

��9�

������

3

��

C�I

��

��

3

��

C�I

��

��

3

��

C�I

��

��

3

��

C�I

��

��

All

mat

rix

of

can

be d

ecom

pose

d in

:

is u

nita

ry,

is

real

ant

isym

met

ric r

éelle

and

i

s re

al s

ymm

etric

-

$�

M(

)C

nG

L,

SiA

eU

eM

=U

AS

()

SS

SS

iAS

SS

iAS

SiA

SS

iA

eP

eP

ee

ee

ee

ee

Pe

ee

MM

Pe

Ue

M

21

2*

22

22

*

2

−−

−−

+

=�

==

==

�= (

)M

MP

Pee

iA

Pee

e

SS

SS

iA

+−

−− =

=�

=

w

ith

log

21 :y

inje

ctiv

itlle

expo

nent

ie2

iAS e

Me

U−

−=

120

/12

0/

���7

���:���

�3����

� 9�

���

�A

utom

orph

ism

of S

iege

l Dis

c

g

iven

by

:

�A

ll au

tom

orph

ism

s gi

ven

by :

�D

ista

nce

give

n by

:

�In

vers

e au

tom

orph

ism

giv

en b

y :n

SD

()

tZ

nU

ZU

ZC

nU

USD

Aut

)(

)(

/,

),(

∈∃

∈Ψ

()

�� ���� ��

−+=

∈∀

)(

�1

)(

�1

log

21,

,,

WWW

Zd

SDW

ZZZ

n

()

()(

)()2/1

00

10

02/1

00

)(

0Z

ZI

ZZ

IZ

ZZ

ZI

ZZ

+−

+−

+−

−−

−=

Σ

()

()

()(

)(

)(

)(

) � �

−−

=

++

==

−−

=−

−=

−+

+

+−

+−

−+

−+

+

2/10

02/1

00

01

01

10

02/1

00

2/10

0

with

)(

)(

0

ZZ

IZ

ZI

G

ZG

IG

ZZ

ZZ

IZ

ZZ

ZI

ZZ

IG

Z

Σ

ΣΦ

Σ

121

/12

1/

��� ���)�2��

9������

�����$

�)����������7�

��)��#

{}

{}

()

()

{}

()

()

() (

)()

()(

)(

)(

)2/1

,2/1

11

2/11

2/11

1

,1

,

,,

,2/1 ,

2/12/1 ,

* ,2/1 ,

2/1 ,,

2,

2,

,,

,

10

01

0

w

ith

then

1Fo

r

with

w

ith

log

.2/1

: w

until

on

Itera

te

0

,,

,,

,,

−+

+−

++

+−

+−

++

+

≠=

+−

−−

−−

=+

+=

−−

−−

=

==

<=

==

==

=�

=

<

==

nn

nm

edia

nn

nn

nm

edia

n,n

nn

k,n

nn

k,n

nn

k,n

k,n

Gk,

n

Fn

k

m

lkk

nk

nn

nk

nk

nk

nk

nk

nk

nk

nk

nk

S

nk

SS

nk

iAn

kk,

nS

iAn

kn

k

Fn

mm

,,

med

ian,

GG

IW

GG

IG

GG

IG

GW

GG

IW

GI

GW

GG

IW

W W

,...,m

k

H

l/

H

�G

WW

PP

PP

PP

S

ith

eW

ee

We

UH

ee

UW

Gn

,...,W

W,..

.,WW

an

d

Wtio

n :

Initi

alis

a

n

nk

nk

nk

nk

nk

nk

Φ

ε

����������� �����

Thal

es A

ir S

yste

ms

Dat

e

0�"

�4! )� �)

����������

��9

���1'��

��3)�

9�� ��

� ��

�����

76�;!

�1�'3�<

123

/12

3/

����

�)� ��:�

���2�

� ����

��$�� �5

�-����

�����

()

()

�− =

−−

��� ��

��� ��

−++

=1 1

,,2

22/1

0,0

2/10

)(

�1

)(

�1

log

log

,n k

km

edia

nk

A

km

edia

nk

Am

edia

nm

edia

nAA

RR

RR

Rd

k kk k

()

()(

)()2/1

12/1

)(

ZZ

IW

ZI

ZW

ZZI

WZ

+−

+−

+−

−−

−=

Φ {}

{}

1 ,11 ,1

,0/

11 ,1

1 ,1,0

,...,

,,..

.,,

− −=

− −→

nm

edia

nn

med

ian

med

ian

Berg

erM

osto

w

M pn

pn

pp

AA

RA

AR

����������� �����

Thal

es A

ir S

yste

ms

Dat

e

-5�����������0��

1������

�� ��

'���1:�

99�� ��

�7��

125

/12

5/

0��

1������

�� ��'��

�1:�

99�� ��

�7��

�����(�

)�

Tim

e-D

oppl

er S

igna

ture

of B

EL20

6 he

lico

on o

ne ra

nge

cell

star

ing

ante

nna

Dop

pler

sig

natu

re o

f Sea

Clu

tter

Tim

e-D

oppl

er s

igna

ture

of w

ind

farm

turb

ine

blad

es

Tim

e-D

oppl

er s

igna

ture

of H

elco

Bla

des

Wak

e-Vo

rtex

Dop

pler

sig

natu

re o

f airc

raft

wak

e-vo

rtex

5����

J�

6�

K

6�

���

(?=�3

��

?�&

=�

��

&<�<

<�

:J�

5;

≥)L�3

��

≥M?<�

3�

�≥

4�

��

G��"��

����

���

≤&�L

4��

≤4?�)

���

≤L�=

���

PRI

N.

Targ

et tr

ajec

tory

PRI

Nr

Vci

ble

rad

.Δ>

)/

36(

/

10 25

300

:)

/ 2

12(

24.

:

hkm

sm

Vcm

rkm

Port

ées

mm

sPR

INEx

empl

e

cibl

era

d>

×=

126

/12

6/

$�)

����)

�� .��

���)�����

������"��

��9�

������

�$��

����)

()

()

()

� =

−−

==

n kk

RR

RR

Rd

1

22

2/11

22/1

12

12

log

..

log

()

0de

t1

2=

−R

()

()

YX

YX

Xt

XY

XX

XR

RR

tX

RR

RR

RR

RR

RR

eR

tX

YX

�=

==

==

−−

−−

)2/1(et

)1( ,

)0()

(2/1

2/12/1

2/12/1

log

2/12/1

2/1

γγ

γγ

()

[]

()

()

()

{}

[]

[]

tedé

croi

ssan

non

surje

ctio

n 1,0

1,0:

et

)(

,)(

,1,0

,

→=∈

βα

βα

βα

tQ

tP

dM

axIn

fQ

Pd

tFr

éche

t

Stat

istiq

ue d

e si

gnat

ure

te

mps

-Dop

pler

: sta

tistiq

ues

de

chem

ins

géod

ésiq

ues

sur u

ne v

arié

ou u

n es

pace

mét

rique

()

()

()

()

{}

[]

[]

tedé

croi

ssan

non

surje

ctio

n 1,0

1,0:

et

)(

,)(

,1 0

,

→=

βα

βα

βα

dtt

Qt

Pd

Inf

QP

d THAL

ES

Util

isat

ion

de te

chni

ques

de

«Fa

st M

arch

ing

»

)(t

α

)(tβ

)(t

α

)(tβ

)( 1t

R)

( 2tR

)( 3t

R)

( 4tR

����������� �����

Thal

es A

ir S

yste

ms

Dat

e

������

���

128

/12

8/

������

���

Cov

aria

nce

Mat

rix

Spa

ceor

Tim

e C

ovar

ianc

eM

atrix

Spac

e-Ti

me

Cov

aria

nce

Mat

rix

Ent

ropy

H

essi

an m

etric

Kar

cher

/Fré

chet

bary

cent

er/M

edia

nP

artia

lIw

asaw

a D

ecom

posi

tion

Isob

aryc

entri

c &

Four

ier F

low

Med

ian

inP

oinc

aré’

sdi

sk

Her

miti

anPo

sitiv

eD

efin

ite

Mat

rixG

roup

: H

PD(n

){

} nI

Cn

SpC

nPS

p2

/),

()

,(

±≡

Sym

plec

ticQ

uotie

nt G

roup

:

Toep

litz

stru

ctur

e

Mul

ti-va

riate

Ent

ropy

H

essi

an m

etric Pa

rtial

Iwas

awa

Dec

ompo

sitio

n

Med

ian

inS

iege

l’s d

isk

Blo

ck-T

oepl

itzst

ruct

ure

Pola

r Fib

ratio

n in

Po

inca

ré’s

disk

Mos

tow

Fib

ratio

n of

Sieg

el’s

disk

INFO

RM

ATI

ON

GEO

MET

RY

Mas

lov-

Lera

y in

dex

in S

iege

l D

isk

Mas

lov-

Lera

y In

dex

in

Poi

ncar

é D

isk

()

()

() e

nR

R�

πlo

gde

tlo

g~

−−

=

()

ji

jiR

�g

θθ∂∂

=~

2

Rob

ust S

pace

-Ti

me

Proc

essi

ng

Rob

ust D

oppl

er

Proc

essi

ng

()

[]

()

()

[]

�− =

+−

+−

+−

−−

−+

=1 1

11

20

1 02

)(

.n k

k kk k

k kn

k kk k

k kn

dAA

AI

dAA

AI

Trk

ndR

RTrn

ds(

)�− =

−−

+�� ��

�� ��=

1 12

222

002

1)(

.n i

iin

di

nPdP

nds

μμ

CA

RTA

N-S

IEG

EL S

YMM

ETR

IC G

EOM

ETR

Y

TRAN

SPO

RT

GEO

MET

RY

Ber

ezin

Q

uant

ifica

tion

Dou

ady-

Ear

le

Con

f.Ba

ryce

nter

129

/12

9/

���

� ���:�����

�� 9�

������

�'�'��

:�$�� ��

��

()

{}

()

()

{} 1

/

,...,

,lo

g/

,...,

,

*

11

10

0

11 1

1 10

<=

×∈

<=

×∈

−−

+

−− −

zzz

DD

RP

RI

ZZZ

SDSD

THPD

AA

R

mm

m

nm

n n

μμ

11

:by

code

dSt

ate

Dop

pler

-

Spat

io−

−×

×∈

nm

SDD

R

()

[]

()

()

[]

�− =

+−

+−

+−

−−

−+

=1 1

11

20

10

2)

(.

n k

k kk k

k kn

k kk k

k kn

dAA

AI

dAA

AI

Trk

ndR

RTrn

ds

[]

()

()

()

�− =

+

−−

+=

=1 1

222

20

)(

)(

2

1)(

log

.m i

iim

ijm

m

di

mP

dm

dg

dds

μμθ

θ

130

/13

0/

$����'���

�������

����

.�- �

�"��

()

()

()

()

()

()

()

()

{}τ

φ

τφ

τφ

τφ

τ

,,

:M

odel

Poin

caré

y Po

larim

etr

2/ar

ctan

an

d

2/

arct

anw

ith

)2

sin(

2si

n.

2co

s2

cos

.2

cos

1

.

Im.2R

e.2

0

2 22 1

31

2

0

**

22

22

3210

s

ss

ss

s

s

zz

zz

zz

zz

ssss

Szz

E

HV

HV

VH

VH

VH

+=

=

���� ��

���� ��

=

����� ��

����� ��

−+

=

���� ��

���� ��

=� � ��

� ��=

()

{}

()

()

{} 1

/

,...,

,lo

g/

,...,

,

*

11

10

0

11 1

1 10

<=

×∈

<=

×∈

−−

+

−− −

zzz

DD

RP

RI

ZZZ

SDSD

THPD

AA

R

mm

m

nm

n n

μμ

Orie

nted

Elli

pse

Com

pact

Had

amar

d Sp

ace:

R x

Pol

y-Po

inca

ré D

isk

x Po

ly-S

iege

l Dis

k

Pola

rimet

ry P

oinc

aré

Mod

el: R

+* xS1

11

:by

code

dSt

ate

Dop

pler

-

Spat

io−

−×

×∈

nm

SDD

R

1 :

byco

ded

Stat

ePo

lar

SR

×∈

11

1 :

byco

ded

Stat

eEM

−−

××

×∈

nm

SDD

SR

131

/13

1/

��)�

�� )�2��

9��������

����

From

the

cylin

dric

al re

pres

enta

tion

ofw

e co

ntra

ct th

e ca

ps a

nd th

e si

de,

and

blow

-up

the

two

circ

ular

cor

ners

Géo

mét

rie d

es b

ords

:co

mpa

ctifi

catio

ns d

iffér

entia

bles

et re

mpl

issa

ges

holo

mor

phes

Ben

oît K

loec

kner

UM

PA, É

cole

nor

mal

e su

périe

ure

de L

yon

132

/13

2/

'��9

���%���

������

��:�$�� ��

�������

9�����

����F?

Def

inite

Posi

tive

Her

miti

an

Toep

litz0

det

0 22

2b

ah

hib

aib

ah

+>

⇔>

Ω

>� ��

� ��+

−=

Ω

Had

amar

d C

ompa

ctifi

catio

n

{}

Dis

k U

nit

Poin

caré

:

1/0

1*1

. *

D

zz

Dhib

aRh

h

<=

∈+

=∈

>� ��

� ��=

Ω

+

μ

μμ

{} {

} 1/

)lo

g(

w

ith

),lo

g(

<=

∈∈

+=

zz

DR

hhib

ah

μ

μμ

Scal

e Pa

ram

eter

Shap

e Pa

ram

eter

a

b

h>9

0�Am

blig

one

Tria

ngle

h2>a

2 +b2

()

()

22

2

22

22

22 0

02

0de

t

ba

hb

ah

ba

hh

hib

aib

ah

+>

⇔>

=�

=+

−+

−�

=� ��

� ��−

+−

λλ

λλ

λλ

133

/13

3/

Thal

es A

ir S

yste

ms

Dat

e

��� �����

�� �������

���9

9 ��

��

mat

rixH

PDTo

eplit

z

� ��� ��

+−

hib

aib

ah

{}

� ��� ��

�)

log(

*)

log(

..

,)

log(

mea

n

mea

n

hm

ean

mea

n

mea

nm

ean

h

Mea

n

mea

nM

ean

eh

he

h

μμ

μ

{}

DR

∈μ

),lo

g(:

catio

nC

ompa

ctifi

Had

amar

d

[]

[]

��� �� ��� ��

��� ��

−� ��

� ��=

−+

� ��� ��

=+

μμ

μμμ

dh

dd

hd

dhdh

ds)

log(

1

10

01

)lo

g(

12

22

222

2

{}

{} 1

/

01

1.

,za

tion

para

met

eri

New

*

*

<∈

=∈

+=

>� ��

� ��=

Ω

×∈

+

zC

zD

hiba

h

DR

h

μ

μμ

μ

{}

{}

{}{

}(

μ

μ

μ),

log(

,),

log(

arg

,)

log(

1),

log(

hh

dM

in

h

ii

N i

p geod

esiq

ueh

mea

nM

ean

� =

=

FREC

HET

Bar

ycen

ter

in M

etric

Spa

ce

����������� �����

Thal

es A

ir S

yste

ms

Dat

e

B��

���%�

��������2��

9��5

������� ��

��9�

�����

����

�%��

135

/13

5/

2������������

��� ��

����

�)�)

�:��

����

���

��

� ��

�1

��

��

���

���

���

���

��

���

� �

���

�� �

��

��

��

$����

��

� ��

���

��

���

$��

��

��

����

��

/��

���

� �

��

���

���

8�

���

��

����

� �

����

�4

type

s of

cla

ssic

al s

ymm

etric

bou

nded

dom

ains

�2

exce

ptio

nal t

ypes

(gro

up o

f mot

ion

E6 a

nd E

7)

���

����

���

���

��

� ��

�5

��

��

��7

���

��*

�0

���

��

��

$�-

��

��

��7

���

.

� �

>−

+

<

−<

+

++

02

1

1

ZZ

:su

ch th

at

row

1 an

dco

lum

nsn

with

m

atric

esC

ompl

ex

: IV

Type

por

der

ofm

atric

essy

mm

etric

-sk

ewC

ompl

ex

: II

ITy

pe

por

der

ofm

atric

essy

mm

etric

Com

plex

: II

Type

colu

mns

qan

dro

ws

p w

ith

mat

rices

Com

plex

: I

Type

):

(

Mat

rixC

ompl

ex

r re

ctan

gula

:

2

t

ZZZZ

co

njug

ate

tran

spos

edI

ZZZ

t

IV nIII

pII pI p,q

136

/13

6/

� ������

�����

���������

��� ��

����

�)�)

�:��

����

�Lu

o-G

eng

Hua

has

com

pute

d th

e ke

rnel

func

tions

for a

ll cl

assi

cal d

omai

ns :

�Pa

rtic

ular

cas

e (p

=q=n

=1) :

Poi

ncar

é U

nit D

isk

()

()

()

()

()(

)(

)do

mai

n th

eof

volu

me

eucl

idea

n th

eis

w

here

, :

IVTy

pefo

r 2

ZZ1

1,

1 ,

: II

ITy

pe

1 ,

: II Ty

pe

, : I

Type

for

de

t1

,

-*

*t

**

Ωμ

νΩ

μ

ννν

Ωμ

ν

ν

n

ZWW

WW

ZK

p

p

qp

ZWI

WZ

K

IV n

III

pII pI p,q

=−

+=

� �

−=

+=

+=

−=

+−+

{}

()

()2

*

*

*1

11

11

11

,

1/

zww

zK

zzC

z

IV

III

III ,

−=

<∈

==

==

137

/13

7/

3��

�����

������

� 9�

����������

���)

�)�:��

����

�G

roup

s of

ana

lytic

aut

omor

phis

ms

of th

ese

dom

ains

are

loca

lly

isom

orph

ic to

the

grou

p of

mat

rices

whi

ch p

rese

rve

follo

win

g fo

rms:

�Th

e gr

oup

cons

its o

f blo

ck m

atric

es A

(gen

eral

izat

ion

of th

e fr

actio

nal

linea

r tra

nsfo

rmat

ion)

. For

the

first

3 ty

pes

:

�� ���� ��−

==

=

�� ���� ��

=�� ��

�� ��−

==

=

�� ���� �� −

=�� ��

�� ��−

==

=

=�� ��

�� ��−

==

n

tIV n

p

p

p

pt

III

p

p

p

p

pt

II p

p

pI p,

q

II

HH

AHA

HAH

A

II

LI

IH

LAL

AH

AHA

II

KI

IH

KAK

AH

AHA

AI

IH

HAH

A

00

,,

, :

IVTy

pe

00

,0

0,

, ,

: II

ITy

pe

00

,0

0,

, ,

: II Ty

pe

1de

t,

00

, ,

: I Ty

pe

2**

**

()(

)122

2112

1122

21

1211

−+

+=

� �� ���� ��

=A

ZA

AZ

AgZ

AA

AA

A

138

/13

8/

3��

�����

������

� 9�

����������

���)

�)�:��

����

�A

ll cl

assi

cal d

omai

ns a

re c

ircul

ar fo

llow

ing

from

Car

tan’

s ge

nera

l the

ory,

an

d th

e po

int 0

is d

istin

guis

hed

for t

he p

oten

tial :

�B

erez

in q

uant

izat

ion

is b

ased

on

the

cons

truc

tion

of th

e H

ilber

t Spa

ce o

f fu

nctio

ns a

naly

tic in

Ω ΩΩΩ:

()

()

()

()ν

Φ−

+−

=� ��

� ��=

ZZI

KZ

ZK

ZZ

det

log

0,0,ln

,*

*

ZgZZ

gj

ZZ

KZ

gj

zg

jgZ

gZK

ZZ

dK

ZZ

Kh

c

ZZ

dK

ZZ

KZ

gZ

fh

cg

f

h

h

∂∂=

=

� ��� ��

=

� ��� ��

=

),

( w

ith

),

()

,(

),

()

,(

),

()0,0()

,(

)(

),

()0,0()

,(

)(

)(

)(

,

**

*

*/1

*1

*/1

*

μ

μ

139

/13

9/

�� �%��

�B��

�������

�������������

� .�A�

���:��#

�Th

e m

ost e

lem

enta

ry e

xam

ple

of B

erez

ian

quan

tific

atio

n is

, in

the

case

of

com

plex

dim

ensi

on 1

, giv

en b

y th

e Po

inca

ré u

nit D

isk

with

vol

ume

elem

ent :

�M

ap fr

om p

ath

on D

to a

utom

orph

y fa

ctor

:

*2

2)

1.(2/1

dzdz

zi

∧−

{}

()

()

*

2

*

2

**

2

22

**

1

)(

)(

)(

lnR

e2

)(

1ln

)(

: po

tent

ial

Käh

ler

1

whe

re w

ith

)1,1(

/)1,1(1

/

zz

zF

zz

gzF

zF

az

bgz

Fz

zF

ba

ab

ba

gSU

g

SSU

zC

zD

∂∂∂

=∂

∂∂�

++

=�

−−

=

=−

� ��� ��

=∈

=<

∈=

()

()

()

)(

)(

1ln

1ln

))0((

)0(

1**

1

2

1

21

*1

*2

2

gF

gF

ab

ba

g

ba

bg

Fa

bg

ba

=� �� ��

�� �� −−

=

+=

� ��� ��

−−

=�

=

−−

=−

−−

140

/14

0/

�Ex

tens

ion

for S

iege

l Uni

t Dis

k :

�Th

e or

bit o

f the

mat

rix Z

=0 is

the

spac

e of

mat

rices

of t

he fo

rm :

{}

()(

)(

)(

)(

)(

))

()

(ln

Re

2)

())

((

lnde

tlo

g)

( :

pote

ntia

lK

ähle

r )

(

0

whe

re

00

with

an

d

with

/

**

**

1*

***

**

ZF

Zg

FZ

BA

trac

eZ

FZ

gF

ZZ

Itr

ace

ZZ

Iz

FA

ZB

BAZ

Zg

BA

AB

IB

BA

A

II

JJ

Jgg

BB

BA

g

IZZ

ZSD

tt

t

n

∂∂=

∂∂

++

=−

−=

−−

=++

=

� �

=−

=−

� ��� �� −

==

� ��� ��

=

<=

++

++

+

�� �%��

�B��

�������

�����������7�

��A����:

��#

()

()

()

BB

Itr

ace

BB

Ig

FA

Bg

++

−+

=+

=�

=ln

det

ln))0(

()0(

1*

141

/14

1/

�9��

������ �%���2��

)������

�Fo

r eve

ry s

ymm

etric

Rie

man

nian

spa

ce, t

here

exi

st a

dua

l spa

ce

bein

g co

mpa

ct. T

he is

omet

ry g

roup

s of

all

the

com

pact

sym

met

ric

spac

es a

re d

escr

ibed

by

bloc

k m

atric

es (t

he a

ctio

n of

the

grou

p in

te

rms

of s

peci

al c

oord

inat

es is

des

crib

ed b

y th

e sa

me

form

ula

as th

e ac

tion

of th

e gr

oup

of m

otio

ns o

f the

dua

l dom

ain)

.

�B

erez

in c

oord

inat

es fo

r Sie

gel d

omai

n :

�� ���� ��

==

=

�� ���� ��

=�� ��

�� ��=

+

++−

00

with

,

:ly

eq

uiva

lent

or

, 1

**

II

LL

LI

AB

BA

AB

BA

t

tt ΓΓ

ΓΓΓΓ

()

()

()

++

−+

=+

=�

=BB

Itr

ace

BBI

gF

AB

gln

det

ln))0(

()0(

1*

()(

)

�� ���� ��

==

++

=� �� ��

�� ��=

IiI

iII

CC

C

AW

AA

WA

WA

AA

A

21 w

ith

:Is

omet

ry

)( 1

122

2112

1122

21

1211

ΓΓ

ΓΓ

142

/14

2/

�� �

������

�� ��

�����9�

������� �%���2��

)������

�Le

t M b

e a

clas

sica

l com

plex

com

pact

sym

met

ric s

pace

. The

inva

riant

vo

lum

e an

d in

varia

nt m

etric

in te

rms

of s

peci

al B

erez

in c

oord

inat

es

have

the

form

:

�Li

nk w

ith :

For a

rbitr

ary

Käh

leria

n ho

mog

eneo

us s

pace

, the

loga

rithm

of

the

dens

ity fo

r the

inva

riant

mea

sure

is th

e po

tent

ial o

f the

met

ric

()

()

()

()

()ν

βα

αββ

α

βα

βα

πμ

μ

−+

+=

∂∂

∂−

==

=

WW

IW

WF

WW

WW

Fg

dWdW

gds

WW

dW

WF

WW

dn

L

det

),

(

whe

re

,ln

with

,,

,

*

*

*2

,

*,

2

**

*

����������� �����

Thal

es A

ir S

yste

ms

Dat

e3��

��������������� ��

�2��

���

��� ��

144

/14

4/

Air

Sys

tem

s D

ivis

ion

/��#�"���

�3��

��������������� ��

�2��

���;H

���� ��

�<�

Jacq

ues

Fara

ut h

as p

ublis

hed

2 bo

oks

on A

naly

sis

on S

ymm

etric

C

ones

& o

n Li

e G

roup

:�

[1] J

. Far

aut &

A. K

oran

yi, «

Ana

lysi

s on

Sym

met

ric c

ones

», C

lare

ndon

Pre

ss, O

xfor

d,

1994

�[2

] J. F

arau

t, «

Ana

lyse

sur

les

grou

pes

de L

ie»,

Cal

vage

& M

oune

t, P

aris

, 200

6

�H

arm

onic

Ana

lysi

s in

the

spec

ial c

ase

of th

e co

ne o

f pos

itive

def

inite

m

atric

es in

the

vect

or s

pace

of a

ll re

al s

ymm

etric

mat

rices

pla

ys a

fu

ndam

enta

l rol

e in

:�

Num

ber T

heor

y (M

inko

wsk

i, si

egel

, Maa

s,…

)�

Sta

tistic

s (W

isha

rt, C

onst

antin

e, J

ames

, Mui

rhea

d)�

Phy

sic

(stu

dy o

f Lor

entz

con

e)

�G

ener

al C

ase

has

been

stu

died

by

:�

Gin

diki

n�

Vin

berg

�Sy

mm

etric

Con

es &

Tub

es (o

ver

them

) are

exa

mpl

e of

Rie

man

nian

Sy

mm

etric

Spa

ces

145

/14

5/

Air

Sys

tem

s D

ivis

ion

/��#�"���

�3��

��������������� ��

�2��

���;H

���� ��

�<

Con

vex

cone

s :

�Le

t Vbe

a fi

nite

dim

ensi

onal

real

Euc

lidea

n Sp

ace.

A s

ubse

t Cof

Vis

sa

id to

be

a co

ne if

:

�Th

e cl

osed

dua

l con

e of

any

con

e C

is d

efin

ed b

y :

�Th

e au

tom

orph

ism

gro

up

of a

n op

en c

onve

x co

ne

is d

efin

ed

by :

�is

a c

lose

d su

bgro

up o

f

,

and

henc

e is

a L

ie G

roup

. The

op

en c

one

i

s sa

id to

be

hom

ogen

eous

if

act

s on

it tr

ansi

tivel

y.

The

open

con

e is

sai

d to

be

sym

met

ric if

it is

hom

ogen

eous

and

sel

f-du

al :

Cx

Cx

∈�

���>∈

λλ

0

()

{}

()

CC

Cx

yx

Vy

C

=

∈∀

≥∈

= ###

,0/

()

()

{}

Ω=

Ω∈

gV

GL

gG

/(

()

ΩG

()

ΩG

()

ΩG

()

()

()

Ω∈

=

GL

gg

yg

xy

gx

ofel

emen

t an

of

adjo

int

the

:*

*

146

/14

6/

Air

Sys

tem

s D

ivis

ion

/��#�"���

�3��

��������������� ��

�2��

���;H

���� ��

�<

Con

vex

cone

s :

�Fo

r any

pro

per o

pen

conv

ex c

one

:

�ch

arac

teriz

es th

e sy

mm

etric

con

es

Cha

ract

eris

tic F

unct

ion

of a

Con

e :

�Le

t

be

a pr

oper

ope

n co

nvex

con

e, it

s ch

arac

teris

tic fu

nctio

n is

:

dyEu

clid

ean

Mea

sure

on

V

The

seco

nd d

eriv

ativ

e

is

pos

itive

def

inite

at e

ach

poin

t

Ω(

)(

)

()

()

Ω∈

���

Ω∈

Ω=

Ω

Ω=

Ω

Gg

Gg

GG

**

**

if

()

()

Ω=

ΩG

G* Ω

()

Ω

−=

*

)(

dye

xy

()

)(

det

)(

, 1

xg

gxG

ϕ−

∈∀

)(

)(

, 0 ,

xx

�xgx

n ϕλ

λϕ

λ−

=>

=

0)

,(

)(

)(

)(

log

),

(

0

>�

� �

+==

=

uu

Gtu

xdtd

xD

xD

Dv

uG

x

tu

vu

x

φφ

ϕ)(

log

2x

147

/14

7/

Air

Sys

tem

s D

ivis

ion

/��#�"���

�3��

��������������� ��

�2��

���;H

���� ��

�<

�C

hara

cter

istic

Fun

ctio

n of

a C

one

:�

Rie

man

nian

stru

ctur

e g

is g

iven

by

Ω

[]

()

log

log

21lo

g

log

)(

log

w

ith)

(lo

g

2u

2

22

2

+==

=

dudv

dudv

dd

du

dud

dud

xd

xd

g

vu

vu

v

u

uu

u

ϕϕ

ϕϕ

ϕϕ

ϕ

ϕϕ

ϕϕ

ϕ

148

/14

8/

Air

Sys

tem

s D

ivis

ion

/��#�"���

�3��

��������������� ��

�2��

���;H

���� ��

�<

Cha

ract

eris

tic F

unct

ion

of a

Con

e :

�Th

e ad

join

t :

�Th

e m

ap

i

s a

bije

ctio

n :

a

nd h

as u

niqu

e fix

ed p

oint

Sym

met

ric C

one

as R

iem

anni

an S

ymm

etric

Spa

ce :

�Th

e bi

linea

r for

m

is p

ositi

ve d

efin

ite,

ther

efor

e it

defin

es a

Rie

man

nian

met

ric o

n

�Th

e co

ne

e

quip

ed w

ith th

is m

etric

is a

Rie

man

nian

Man

ifold

�Si

nce

the

cone

is s

ymm

etric

, the

map

i

s a

bije

ctio

n an

d an

isom

etry

(the

man

ifold

is a

Rie

man

nian

Sym

met

ric

Spac

e gi

ven

by th

is is

omet

ry)

()

)(

)(

with

)

(lo

g*

xf

Du

xf

xx

u=

∇−∇

**

Ω∈

Ω∈

xx

�(

)n

xx

=*

)(

log

),

(x

DD

vu

Gv

ux

ϕ=

ΩΩ

Ω)

(lo

g*

xx

−∇=

()

cons

tx

xx

x=

=

)(

)(

*

**

ϕϕ

149

/14

9/

Air

Sys

tem

s D

ivis

ion

/��#�"���

�3��

��������������� ��

�2��

���;H

���� ��

�<

The

cone

of P

ositi

ve D

efin

ite S

ymm

etric

mat

rices

:

�In

ner P

rodu

ct &

qua

drat

ic fo

rm:

�Le

t

be

the

set o

f pos

itive

def

inite

sym

met

ric m

atric

es.

The

set i

s an

ope

n co

nex

cone

and

is s

elf-d

ual :

is h

omeg

eneo

us :

�C

hara

cter

istic

func

tion

:

()

()

()

()

0)

( , 0

,

,

,,

>=

≠∈

=∈

Tn

xQ

R

xyTr

yx

Rn

Sym

yx

ξξξ

ξξ

)(R

Ω=

Ω*

)(R

� �

===

�Ω

∈T

nT

gxg

xg

Ig

ggx

x)

(

)(

ρρ

()

()

nn

Ig

xI

gx

ϕρ

ϕρ

1)

(de

t)

()

(−

=�

=(

)(

)[

](

)n

n

nI

xx

gg

gx

ϕϕ

ρ21

1

2

)de

t()

()

det(

)(

det

)de

t()

det(

+−

+=

� � �

=

=

()

()

nIx

nx

ϕϕ

log

det

log

)1(

21)

(lo

g+

+−

=

1*

1)1

(21

)de

t(lo

g−

−+

=�

=∇

xn

xx

x

150

/15

0/

Air

Sys

tem

s D

ivis

ion

/��#�"���

�3��

��������������� ��

�2��

���;H

���� ��

�<

Sym

met

ric C

one

& E

xpon

entia

l Fam

illy

of p

roba

bilit

y m

easu

re

�Le

t μ μμμbe

a p

ositi

ve B

orel

Mea

sure

on

eucl

idea

n sp

ace

V. A

ssum

e th

at

the

follo

win

g in

tegr

al is

fini

te fo

r all

xin

an

open

set

:

�Fo

r

, c

onsi

der t

he p

roba

bilit

y m

easu

re (e

xpon

entia

l fam

illy)

:

Then

V⊂

Ω(

)

−=

)(

)(

yd

ex

yx

μϕ

Ω∈x

()

()

)(

)(1

,y

de

xdy

xp

yx

μϕ

−=

)(

log

),

()

(x

dyx

ypx

−∇=

=

()

()(

))

(lo

g)

,(

)(

)(

)(

xD

Ddy

xp

vx

my

ux

my

vu

xV

vu

ϕ=

−−

=

����������� �����

Thal

es A

ir S

yste

ms

Dat

e

����7

����

�����9

� ��

����

3�����

���9

� �9��

��

�������#

������ ��

����

152

/15

2/

Thal

es A

ir S

yste

ms

Dat

e

3�����

���9

� ��

������9

� �9��

+$$

���

����

� 1

���

���

� �

��

� �

�A

loca

lly s

tron

gly

conv

ex h

yper

surf

ace

in th

e af

fine

spac

e R

n+1

is

calle

d an

affi

ne h

yper

bolic

hyp

ersp

here

if th

e af

fine

norm

als

thro

ugh

each

poi

nt o

f the

hyp

ersu

rfac

eei

ther

all

inte

rsec

t at o

ne

poin

t, ca

lled

its c

ente

r, th

at is

on

the

conv

ex s

ide.

�Th

is c

lass

of h

yper

surf

aces

was

firs

t stu

died

sys

tem

atic

ally

by

W.

Bla

schk

ein

the

fram

e of

affi

ne g

eom

etry

. �

E. C

alab

irai

sed

a co

njec

ture

that

:�

thes

e hy

pers

urfa

ces

are

asym

ptot

ic to

the

boun

dary

of a

con

vex

cone

�ev

ery

non-

dege

nera

te c

one

V d

eter

min

es a

hyp

erbo

lic a

ffine

hyp

ersp

here

, as

ympt

otic

to th

e bo

unda

ry o

f V, u

niqu

ely

by th

e va

lue

of it

s m

ean

curv

atur

e.

He

prov

ed th

is c

onje

ctur

e fo

r hom

ogen

eous

con

vex

cone

s un

der s

ome

cond

ition

s on

the

actio

n of

the

auto

mor

phis

mgr

oup

of th

e co

ne.

153

/15

3/

Thal

es A

ir S

yste

ms

Dat

e

����7

����

���3�����

���9

� ��

������9

� �9��

��

"�����

+$$

�����

��� 1

���

���� �

��� �

�Th

e th

eory

of h

omog

eneo

us c

onve

x co

nes

play

s a

cent

ral r

ole.

Let V

be

a no

ndeg

ener

ate

open

con

vex

cone

in R

n+1 (

x)an

d V’

be

its

dual

. A(V

) mea

ns th

e gr

oup

of a

ll lin

ear t

rans

form

atio

ns w

hich

le

ave

V in

varia

nt. T

he c

hara

cter

istic

func

tion

of V

, is

give

n by

the

equa

tion:

with

the

val

ue o

f the

line

ar fu

nctio

nal �

at x

�W

e de

note

by

S cth

e le

vel s

urfa

ce o

f

whi

ch is

a

nonc

ompa

ctsu

bman

ifold

in V

, and

by �

the

indu

ced

met

ric o

n S c

�Th

e H

essi

an

defin

esth

e m

etric

on

V.

�A

ssum

ing

the

cone

V is

hom

ogen

eous

und

er A

(V),

Sasa

ki p

rove

d th

at S

cis

a h

omog

eneo

us h

yper

bolic

affi

ne h

yper

sphe

rean

d ev

ery

such

hyp

ersp

here

sca

n be

obt

aine

d in

this

way

Sasa

ki re

mar

ks th

at �

is id

entif

ied

with

the

affin

e m

etric

and

Sc

is

a gl

obal

Rie

man

nian

sym

met

ric s

pace

whe

n V

is a

sel

f-dua

l con

e.

Vx

de

xV

xV

∈=

− ,

)

('

φξ

ξ,x{

} cx

VV

=)(

φ

Vd

φlo

g2

KO

SZU

L-VI

NB

ERG

C

HAR

AC

TER

ISTI

C

FUN

CTI

ON

154

/15

4/

Thal

es A

ir S

yste

ms

Dat

e

� �9

� �3�����

���9

� ��

������9

� �9��

-

��

�+

$$��

���

��

� 1

���

����

� �

��

� �

�Le

t S b

e a

hype

rsur

face

in R

n+1

and

be th

e im

bedd

ing

of S

. The

imbe

ddin

g

def

ines

a v

olum

e bu

ndle

val

ued

quad

ratic

fo

rm G

on

S by

the

equa

tion:

�in

term

s of

loca

l coo

rdin

ates

of S

. Thi

s is

inva

riant

un

der u

nim

odul

ar a

ffine

tran

sfor

mat

ions

in R

n+1 .

If th

is q

uadr

atic

fo

rm is

sup

pose

d to

be

non-

dege

nera

te, i

t def

ines

a

pseu

dorie

man

nian

str

uctu

re te

nsor

gw

ith c

orre

spon

ding

vol

ume

elem

ent d

v(g)

, uni

quel

y de

fined

by

the

equa

tion:

�∧

∧⊗

�� ���� ��

∂∂∂∂

∂∂∂

=j

i

nj

in

ji

dydy

dydy

yfyf

yy

fG

,

11

2

...,..

.,,

det

1:

+→

nR

Sf

f

()

n yy

,...,

1

)(g

dvg

G⊗

=

155

/15

5/

Thal

es A

ir S

yste

ms

Dat

e

� �9

� �3�����

���9

� ��

������9

� �9��

-

��

+

$$��

���

��

� 1

���

���

� �

��

� �

�w

e as

sum

e th

at th

e se

t S is

loca

lly s

tron

gly

conv

ex. T

hen

the

tens

or g

can

be c

hose

n to

be

posi

tive

defin

ite c

hoos

ing

the

orie

ntat

ion

of S

so

that

G is

pos

itive

val

ued.

With

this

Rie

man

nian

met

ric, c

alle

d th

e af

fine

met

ric, t

he a

ffine

no

rmal

is d

efin

ed to

be

the

vect

or

w

here

�is

the

Lapl

ace-

Bel

tram

i ope

rato

r with

resp

ect t

o g.

For a

n af

fine

hype

rbol

ic h

yper

sphe

rew

ith th

e ce

nter

at t

he o

rigin

, n

satis

fies

the

equa

tion:

whe

re H

, cal

led

the

affin

e m

ean

curv

atur

e, is

a n

onze

ro c

onst

ant.

�C

alab

ipro

ved

that

the

hype

rsur

face

Sis

a p

rope

r affi

ne

hype

rsph

ere

with

its

cent

er a

t the

orig

in a

nd th

e af

fine

mea

n cu

rvat

ure

Hif

usa

tisfie

s th

e eq

uatio

n:

fn

=)

/1(

Hf

n−

=

()

22

)(

det

−−

= �� ���� ��

∂∂∂

n

ji

Hu

ξξ

156

/15

6/

Thal

es A

ir S

yste

ms

Dat

e

� �9

� �3�����

���9

� ��

������9

� �9��

@

�>�

� �

��

� ��

���

� �

���

���

����

$�����

$$��

���

��� �

��� �

�Le

t

b

e a

linea

r coo

rdin

ate

syst

em o

f Rn+

1an

d

be

the

repr

esen

tatio

n of

S a

s th

e gr

aph

of a

lo

cally

str

ongl

y co

nvex

func

tion

f

or

rang

ing

in a

dom

ain

Let

be

the

imag

e of

Dun

der t

he lo

cally

in

vert

ible

map

ping

We

defin

e th

e fu

nctio

n

by

the

equa

tion

whe

re

is th

e pa

iring

giv

ing

the

cano

nica

l dua

lity.

uis

the

Lege

ndre

tran

sfor

m o

f

and

als

o th

e do

mai

n �

the

Lege

ndre

tran

sfor

m o

f Sw

ith re

spec

t to

the

coor

dina

tes

()

11 ,..

.,+n x

x(

){

}n

nx

xf

x,..

.,1

1=

+

f(

)n x

xx

,...,

1=

nR

D⊂

()

nn

ξξ

,...,

,2

1⊂

Ω(

)i

in

xff

ff

grad

f∂∂

==

=

whe

re,..

.,1

ξ(

∈n

ξ,..

.,1

()

)(

,)

()

(x

grad

fx

xf

xgr

adf

u+

−=

.,.f

()

11 ,..

.,+n x

x

157

/15

7/

Thal

es A

ir S

yste

ms

Dat

e

2��

�� ������������

�9� ��

����������

���9

� �9��

�Th

eore

m[W

u an

d Sa

ckst

eder

]�

Let S

be a

clo

sed

hype

rbol

ic a

ffine

hyp

ersp

here

with

cen

ter a

t the

orig

in a

nd th

e af

fine

mea

n cu

rvat

ure

H. t

he h

yper

surfa

ceS

is c

ompl

ete

(with

resp

ect t

o th

e af

fine

met

ric a

nd w

ith re

spec

t to

the

indu

ced

met

ric o

f the

Rie

man

nian

met

ric o

f R

n+1 )

, non

com

pact

, orie

ntab

le, s

moo

th a

nd lo

cally

stro

ngly

con

vex.

In th

is

situ

atio

n w

e ha

ve:

�S

uch

a su

rface

is th

e fu

ll bo

unda

ry o

f som

e cl

osed

con

vex

body

and

is th

e gr

aph

of a

non

-neg

ativ

e sm

ooth

stri

ctly

con

vex

func

tion

defin

ed in

som

e hy

perp

lane

.

�B

y th

is th

eore

m th

e hy

pers

urfa

ceS

can

be w

ritte

n gl

obal

ly a

s th

e se

t

, w

here

is a

pos

itive

sm

ooth

str

ictly

co

nvex

func

tion

on

�Th

e ta

ngen

t pla

ne a

t any

poi

nt o

f Sca

nnot

con

tain

the

orig

in. I

n ot

her w

ords

the

affin

e no

rmal

is n

ot ta

ngen

t to

S:�

the

norm

al v

ecto

r in

Euc

lidea

n se

nse

at o

ne p

oint

in S

is p

ropo

rtion

al to

:

with

the

coo

rdin

ate

of th

e Le

gend

re tr

ansf

orm

atio

n

�Th

e af

fine

norm

al a

t tha

t poi

nt is

with

�H

ence

:

()

{}

nn

xx

fx

,...,

11

=+

f{

} 01

=+nx

() 1

,,..

.,1

−=

nEn

ξξ

�� ���� ��

∂∂+

∂∂∂∂

=� =n i

ii

n

n1

11

2,

,...,

ξρρ

ξρξρ

ρ

()

2/1

det

+−

=n

ijfρ

0/1

,≠

−=

ρEn

n

158

/15

8/

Thal

es A

ir S

yste

ms

Dat

e

3���9��������

����

�)� ��������

�5���

��

�S

is a

sym

ptot

ic to

the

boun

dary

of a

con

vex

cone

whe

n th

e bo

unda

ry is

equ

al to

the

set o

f all

asym

ptot

ic li

nes

of S

:�

Let

then

an

d

is a

n op

en n

on-d

egen

erat

e co

nvex

con

e.

�S

is a

sym

ptot

ic to

the

boun

dary

of V

�Th

eore

m:

�E

very

clo

sed

hype

rbol

ic a

ff�ne

hype

rsph

ere

is a

sym

ptot

ic to

the

boun

dary

of a

co

nvex

con

e. C

onve

rsel

y, e

very

non

-deg

ener

ate

cone

V d

eter

min

es a

hyp

erbo

lic

affin

e hy

pers

pher

eas

ympt

otic

to th

e bo

unda

ry o

f V, a

nd u

niqu

ely

dete

rmin

ed b

y th

e va

lue

of it

s m

ean

curv

atur

e.

�Le

gend

re tr

ansf

orm

atio

n is

an

isom

etry

with

resp

ect t

o th

e LN

-m

etric

τ τττ(in

trod

uced

by

Loew

ner&

Nire

nber

g) a

nd th

e af

fine

met

ric

g: �

For a

neg

ativ

e co

nvex

sol

utio

n u

of

Loew

ner&

Nire

nber

g d

efin

ed th

e m

etric

on

a bo

unde

d co

nvex

dom

ain �

in

{}S

pR

kpS

nk

∈∈

=+

/1

'fo

r 0

'k

kS

Sk

k≠

=�

� 0>

=k

kSV

()

22

)(

det

−−

= �� ���� ��

∂∂∂

n

ji

Hu

ξξ

ud

Hu

21

)(ξ

nR

159

/15

9/

Thal

es A

ir S

yste

ms

Dat

e

/��"

�� 10� �

��� 7�$�� ��

�LN

-met

ric τ τττ

(intr

oduc

ed b

y Lo

ewne

r& N

irenb

erg)

�Th

is m

etric

has

the

proj

ectiv

e in

varia

nce

in th

e fo

llow

ing

sens

e:

�Le

t

be

a pr

ojec

tive

trans

form

atio

n w

hich

sen

ds �

onto

A�

. Th

en A

is a

n is

omet

ryw

ith re

spec

t to

LN-m

etric

s of

�an

d A�

.

�Le

gend

re tr

ansf

orm

atio

n is

an

isom

etry

with

resp

ect t

o th

e LN

-met

ric

τ τ τ τ an

d th

e af

fine

met

ric g

�Le

t be

an e

quat

ion

of a

hyp

erbo

licaf

fine

hype

rsph

ere.

Sin

ceth

e af

fine

met

ricis

writ

ten

as

but

atth

e co

rres

pond

ing

poin

ts x

and

� b

y th

e Le

gend

re tr

ansf

orm

atio

n. H

ence

:

()

22

)(

det

−−

= �� ���� ��

∂∂∂

n

ji

Hu

ξξ

ud

Hu

21

with

()

Rn

SLA

,1+∈

()

nn

xx

fx

,...,

11

=+

()

ijn

ji

ff

ff

=,..

.,,

det

1,

()

()

fd

fg

nij

22

/1de

t+

−=

)(

)(

22

ξu

dx

fd

=

gu

du

dH

u=

==

22

τ(

)2

/1de

t+

−=

nijf

ρw

ith

160

/16

0/

Thal

es A

ir S

yste

ms

Dat

e

����7

����

�����9

� ��

����������

���9

� �9��

��

�Le

t Vbe

a n

on-d

egen

erat

e co

nvex

con

e in

Rn+

1 . Fi

rst w

e re

call

som

e pr

oper

ties

of th

e ch

arac

teris

tic fu

nctio

n:

�te

nds

to in

finity

whe

n x

appr

oach

es to

any

poi

nt o

f the

bou

ndar

y of

V

�Th

e m

easu

re

is in

varia

nt u

nder

:

�is

con

vex

on V

. Hen

ce

def

ines

a m

etric

on

V

�Th

e le

vel s

urfa

ce o

f

is

a n

onco

mpa

ctsu

bman

ifold

in V

calle

d th

e ch

arac

teris

tic s

urfa

ce o

f V.

�W

e de

note

by

the

indu

ced

met

ric o

f

o

n

Vx

de

xV

xV

∈=

− ,

)

('

φξ

)(x

Vφdx

xV

)(

φ)

(VA)

(fo

r

)de

t(/)

()

(V

AA

Ax

AxV

V∈

φVφ

log

Vd

φlo

g2

{} c

xS

Vc

V=

=)

(:

φφ

Vd

φlo

g2

cSKO

SZU

L-VI

NB

ERG

C

HAR

AC

TER

ISTI

C

FUN

CTI

ON

161

/16

1/

Thal

es A

ir S

yste

ms

Dat

e

����7

����

�����9

� ��

����������

���9

� �9��

��

�A

ssum

ing

Vis

affi

nely

hom

ogen

eous

. The

cha

ract

eris

tic s

urfa

ce

S cis

obv

ious

ly h

omog

eneo

us w

ith re

spec

t to

unim

odul

arel

emen

ts

of A

(V):

�TH

EOR

EM (S

ASA

KI):

Ever

y ch

arac

teris

tic s

urfa

ce S

cis

a c

ompl

ete

hype

rbol

ic a

ffine

hyp

ersp

here

with

mea

n cu

rvat

ure

ac2 /n

+2w

here

a

is a

neg

ativ

e co

nsta

nt d

epen

ding

onl

y on

V.

�Pr

oof:

Let

. .

c

anbe

writ

ten

loca

llyas

by a

sm

ooth

func

tion

, the

co

ordi

nate

bein

gch

osen

such

that

. Le

t

the

Lege

ndre

tran

sfor

mof

.

Sin

ceis

cons

tant

on

we

have

on

.

By

the

defin

ition

:

B

ut

Hen

ce

Ther

efor

e

)(

log

)(

xx

Vφψ

={

} cS c

log

==

ψ)

,...,

(1

1n

nx

xf

x=

+f

()

11 ,..

.,+n x

x0

1≠

+nψ

uf

ψcS

1/

+=

ni

ifψ

ψcS

11

1/

+=

+� ��

� ��+

−=

�n

n ii

in

xf

ψψ

0,

, lo

g)1

()

()

(>

∀∈

∀+

−=

kV

xk

nx

kxψ

ψ

�+ =

+−

=1 1

)1(

)(

n

nx

αα ψ

1/)1

(+

+=

nn

162

/16

2/

Thal

es A

ir S

yste

ms

Dat

e

����7

����

�����9

� ��

����������

���9

� �9��

��

�C

onsi

der

�Se

t

�W

e ha

ve�

Hen

ce

by

the

hom

ogen

eity

for s

ome

cons

tant

bw

hich

dep

ends

onl

y on

Vits

elf.

This

mea

ns

�Si

nce

, th

en

()

()

0

1de

t

,1 ,

det

det

1

11

11

1

21

1

21

11

1

11

1

+

++

++

+

++

+

++

+

++

+

=−

≤≤

�� ���� ��

++

−=

nj

nn

nj

n

iin

ij

nij

n

n

nn

ji

n

jni

jin

ijij

n

f

nj

if

ψψ

ψψ

ψψ

ψψ

ψψ

ψψψ

ψψ

ψψ

ψψ

ψψ

1,

1 , 0

)(

+≤

≤=

Φn

αψ

ψψ

β

ααβ

()

)(

, )(

det

)(

2V

AA

AxA

x∈

Φ=

Φ ()

12

det

det

− �� ���� ��

∂∂∂

=j

iij

uf

ξξ)

()

(2

xb

=Φ (

)c

ijn

nS

bcx

fon

)(

det

21

21

=−

++

ψψ

2

22

11

det

bcun

un

ji

+

� ��� ��

+−

= �� ���� ��

∂∂∂

ξξ

11 ++=

n

nu

ψR

emar

k:

163

/16

3/

Thal

es A

ir S

yste

ms

Dat

e

����7

����

�����9

� ��

����������

���9

� �9��

��

�TH

EOR

EM:

Let S

a co

mpl

ete

hype

rbol

ic a

ffine

hyp

ersp

here

with

its

cen

ter a

t the

orig

in w

hich

is h

omog

eneo

us u

nder

the

subg

roup

G

of th

e un

imod

ular

grou

p. L

et V

, the

con

vex

cone

to w

hose

bo

unda

ry th

e hy

pers

urfa

ceS

is a

sym

ptot

ic:

.

Let

, T

he e

lem

ent

acts

on

Vby

Vis

hom

ogen

eous

und

er

.

Then

Sis

a c

hara

cter

istic

sur

face

of V

.Pr

oof:

Let f

unct

ion

on V

by

the

equa

tion

Sinc

e

,

is

wel

l def

ined

. Th

en

for

A e

G.

Ther

efor

e, b

y th

e ho

mog

enei

ty,

for s

ome

nonz

ero

cons

tant

b.

�0>

=k

kSV

=R

GG~

()

Gt

gg

~,

∈=

)(

.x

gtg

=G~

kn

Sx

kx

∈=

−−

for

)(

γ'

for

'

kk

SS

kk

≠=

∩φ

γ)

det(

/)(

)(

Ax

Axγ

γ=

Vbφ

γ=

164

/16

4/

Thal

es A

ir S

yste

ms

Dat

e

����7

����

�����9

� ��

����������

���9

� �9��

��

�2

prev

ious

Sas

aki ‘

s Th

eore

ms

prov

e th

at th

e cl

assi

ficat

ion

of

hom

ogen

eous

hyp

erbo

lic a

ffine

hyp

ersp

here

sis

redu

ced

to th

e cl

assi

ficat

ion

of h

omog

eneo

us c

onve

x co

nes:

�R

otha

us, O

. S.,

The

cons

truct

ion

of h

omog

eneo

us c

onve

x co

nes,

Ann

. of M

ath.

, 83

, pp

. 358

-376

., 19

66

�V

inbe

rg, E

. B.,

The

theo

ry o

f con

vex

hom

ogen

eous

con

es, T

rans

. Mos

cow

Mat

h.

Soc

, 12

(196

3), 3

40-4

03

�E.

B. V

inbe

rgha

s de

fined

an

indu

ctiv

e m

etho

d pr

oduc

ing

all

hom

ogen

eous

con

vex

cone

s :

�Fr

om a

giv

en c

onve

x co

ne V

iin

Rn+

1 (x)

, one

can

con

stru

ct a

noth

er

hom

ogen

eous

con

e V

in R

(t) X

Rm

(y) X

Rn+

1 (x)

by

the

equa

tion

:

whe

re h

is a

line

ar m

appi

ng o

n R

n+1

who

se v

alue

s ar

e re

al s

ymm

etric

pos

itive

-de

finite

mat

rices

of o

rder

m a

nd, c

orre

spon

ding

to e

ach

elem

ent

B o

f som

e tra

nsiti

ve s

ubgr

oup

of A

{V1)

, the

re e

xist

s a

mat

rix

such

that

. Thi

s m

etho

d ca

n be

tran

spos

ed to

obt

ain

all p

roje

ctiv

ely

hom

ogen

eous

bou

nded

con

vex

dom

ains

or a

ll ho

mog

eneo

us h

yper

bolic

a�fi

nehy

pers

pher

es

{} y

xh

yt

xy

tV

t)

(/)

,,

(1−

>=

),

(R

mG

LA

∈(

)Bx

hA

xh

At=

)(

165

/16

5/

Thal

es A

ir S

yste

ms

Dat

e

����7

����

�����9

� ��

����������

���9

� �9��

��

�TH

EOR

EM: S

uppo

se V

is h

omog

eneo

us. T

hen

the

met

ric �

cis

id

entif

ied

with

the

affin

e m

etric

gup

to a

con

stan

t fac

tor.

Proo

f :

Let

.

c

an b

e w

ritte

n lo

cally

as

by a

sm

ooth

func

tion

,

the

coor

dina

te

be

ing

chos

en s

uch

that

Si

nce

we

have

But

. Th

eref

ore

The

assu

mpt

ion

need

ed is

not

the

hom

ogen

eity

of V

but

the

cond

ition

that

the

leve

l sur

face

is a

n af

fine

hype

rsph

ere.

)(

log

)(

xx

Vφψ

={

} cS c

log

==

ψ)

,...,

(1

1n

nx

xf

x=

+(

)1

1 ,...,

+n xx

f0

1≠

+nψ

� =+

+−

=n i

i

nin

dxdx

11

1

ψψ

� ≤≤

+−

==

nj

i

ji

ijn

Sc

dxdx

fd

c,

11

ψω

un

n/)1

(1

+=

Hg

nc

)1(

+−

166

/16

6/

Thal

es A

ir S

yste

ms

Dat

e

����7

����

�����9

� ��

����������

���9

� �9��

��

�R

emar

k: T

he s

olut

ion

u of

the

equa

tion

is g

iven

as

a fir

st o

rder

loga

rithm

ic d

eriv

ativ

e of

the

char

acte

ristic

func

tion,

i.e.

Sinc

e th

e ch

arac

teris

tic fu

nctio

n

of t

he p

rodu

ct o

f con

vex

cone

s V

and

W is

equ

al to

, t

he d

eriv

ativ

e of

is w

ritte

n us

ing

the

deriv

ativ

es o

f an

d

.

This

giv

es th

e co

mpo

sitio

n fo

rmul

a fo

r re

duci

ble

cone

s.

WVφ

φ

()

22

)(

det

−−

= �� ���� ��

∂∂∂

n

ji

Hu

ξξ

1/)1

(+

+=

nn

φφ

VφWφ

167

/16

7/

Thal

es A

ir S

yste

ms

Dat

e

����7

����

�����9

� ��

����������

���9

� �9��

��

�C

onsi

derin

g th

e se

lf-du

al c

one,

to

form

ulat

e an

othe

r des

crip

tion

of th

e Le

gend

re tr

ansf

orm

atio

n w

e in

trod

uce

the

*-m

appi

ng d

ue to

K

oech

er ,

It is

a m

appi

ng fr

om V

to it

s du

al V

def

ined

by

the

equa

tion:

�Th

is m

appi

ng *

has

the

follo

win

g pr

oper

ties:

�*

sets

up

a on

e to

one

cor

resp

onde

nce

betw

een

Van

d V

and

hold

s fo

r eve

ry

�If

Vis

hom

ogen

eous

, the

n V

’ is

also

hom

ogen

eous

and

is c

onst

ant

for a

ll x.

We

deno

te th

is c

onst

ant b

y

�In

hom

ogen

eous

cas

e, th

e *-

imag

e of

Sc

is a

lso

a ch

arac

teris

tic

surf

ace

of V

whi

ch w

e de

note

by

S'c.

Taki

ng a

hyp

erpl

ane

Hsu

ch

that

is a

non

-em

pty

boun

ded

conv

ex d

omai

n, w

e co

rres

pond

, for

eve

ry p

oint

th

e in

ters

ectio

n po

int o

f the

lin

e th

roug

h th

e or

igin

and

x*w

ith U

. Thu

s w

e ha

ve a

map

ping

from

S'

cto

U.

�LE

MM

A: T

he m

appi

ng *

follo

wed

by

this

map

ping

is d

efin

ed o

n S c

and

proj

ectiv

ely

equi

vale

nt to

eve

ry L

egen

dre

tran

sfor

mat

ion

of S

c.

()

Vx

xx

xn

∈−

==

+fo

r )

(),.

..,(

11

ψξ

()

*1

* )(

xA

Axt

−=

)(VA

A∈

)(

)(

*'

xx

VV

φφ

'V

HU

∩=

'*

cSx

168

/16

8/

Thal

es A

ir S

yste

ms

Dat

e

����7

����

�����9

� ��

����������

���9

� �9��

��

�PR

OPO

SITI

ON

: Sup

pose

Vis

hom

ogen

eous

. The

n * i

s an

isom

etry

w

ith re

spec

t to

the

met

rics

a

nd

�Pr

oof:

Vd

φlo

g2

'2

log

Vd

φ

()(

)()

V

lk

lk

ji

lj

Vk

iV

ji

V

ji

ji

ji

VV

k ij

kj

Vj

iV

d

kk

kj

Vj

j

jj

iV

i

Vx

VV

VV

d

dxdx

xx

xx

xx

x

dd

d

xx

xx

I

ddx

dxx

xx

xd

grad

x

ψ

ψψ

ξξψ

ξξ

ξξψ

ψ

δξ

ξψψ

ξξ

ξξψ

ξψ

ξ

ψφ

ψφ

ψ

2

,,

,

22

*'

2

,

'2

**

'2

**

*'

22

**

'2

*2

*

*'

'

)(

)(

)(

)(

).(

)(

)(

,)

()

(

,lo

g,

log

=

∂∂∂

∂∂∂

∂∂∂

=

�� ���� ��

∂∂∂

=

=∂

∂∂∂

∂∂�

=

∂∂∂

−=

∂∂∂

−=

−=

==

��

169

/16

9/

Thal

es A

ir S

yste

ms

Dat

e

����7

����

�����9

� ��

����������

���9

� �9��

��

�N

ow a

ssum

e V

is a

sel

f-dua

l con

e. T

hen

is a

n au

tom

orph

ism

of V

and,

mor

eove

r,

�Le

t

, * /

Sis

an

invo

lutiv

eau

tom

orph

ism

of S

.

Sinc

e

, we

have

.The

refo

re K

(x)i

s an

aut

omor

phis

mof

S

�PR

OPO

SITI

ON

: If

the

hom

ogen

eous

con

e V

is s

elf-d

ual,

then

the

char

acte

ristic

sur

face

Sc

is a

glo

bally

sym

met

ric s

pace

.�

Proo

f: Fo

r one

poi

nt

d

efin

e an

aut

omor

phis

ms

of S

by

. S

ince

K(x

)is

sym

met

ric, s

is a

n in

volu

tion

by

By

For a

gen

eral

Sc,

it is

eno

ugh

to tr

ansl

ate

the

sym

met

ry o

f Sto

Sc

by th

e m

appi

ng

xx

Kx

)(

*=

xx

xK

∂−∂

=/

)(

*

*KS

S=

()

Kx

xx

Kx

VV

Vde

t)

()

()

(*

φφ

φ=

=S

xx

K∈

=fo

r 1

))(

det(

Sx

∈0

*1

0)

()

(x

xK

xs

−=

dIx

xxx

Kx

xx

s−

=∂∂

=∂∂

=−

)(

)(

)(

xs

and

)

(0

*1

00

00

Sx

cS

xx

∈→

∈κ

170

/17

0/

Thal

es A

ir S

yste

ms

Dat

e

����7

����

�����9

� ��

����������

���9

� �9��

��

�Le

t H+ (

n, K

) be

the

cone

of p

ositi

ve-d

efin

ite h

erm

itian

sym

met

ric

mat

rices

ove

r K

= fi

elds

R, C

, H (q

uate

rnio

ns) o

r the

Cay

ley

alge

bra

Ca.

The

n th

e fo

llow

ings

are

the

list o

f all

irred

ucib

le s

elf-d

ual

cone

s V

and

the

corr

espo

ndin

g gl

obal

ly s

ymm

etric

spa

ces

S

EIV

type

ofsp

ace

the

),2(

)1(

/)1,1(

)(

)(

/)2(

),

()

(/)

,(

),

()

(/)

,(

),

( *

*

=�

=

−−

=�

=

=�

=

=�

==

�=

+++

SC

aH

Vn

SOn

SOS

nC

Vn

Sn

SUS

Hn

HV

nSU

Cn

SLS

Rn

HV

nSO

Rn

SLS

Rn

HV

p

171

/17

1/

Thal

es A

ir S

yste

ms

Dat

e

����7

����

�����9

� ��

����������

���9

� �9��

��

�C

orol

lary

: Le

t be

a re

gula

r con

vex

cone

and

let

be

the

cano

nica

l Hes

sian

met

ric. T

hen

each

leve

l sur

face

of t

he

char

acte

ristic

func

tion

i

s a

min

imal

sur

face

of t

he R

iem

anni

an

man

ifold

Exam

ple:

Let

be

a re

gula

r con

vex

cone

con

sist

ing

of a

ll po

sitiv

e de

finite

sym

met

ric m

atric

es o

f deg

ree

n. T

hen

is a

Hes

sian

str

uctu

re o

n

, a

nd e

ach

leve

l sur

face

of

i

s a

min

imal

sur

face

of t

he R

iem

anni

an m

anifo

ld:

ψlo

gD

dg

=

ψ(

) g,Ω

() x

Dd

gD

det

log

,−

xde

t(

) xD

dg

det

log

,−

����������� �����

Thal

es A

ir S

yste

ms

Dat

e

��� 1(��������"

173

/17

3/

Air

Sys

tem

s D

ivis

ion

2��

9��5

�3��� �7

���� ��$�)

������

�� ��

����� �

�Er

ich

Käh

ler G

eom

etry

is g

iven

by

:�

com

plex

Man

ifold

of n

dim

ensi

ons,

com

pact

or n

ot, w

ith K

ähle

rian

met

ric, t

hat

coul

d be

loca

lly g

iven

by

posi

tive

defin

ite R

iem

anni

an F

orm

:

�K

ähle

r con

ditio

n :

Loca

l Exi

sten

ce o

f Käh

ler p

oten

tial f

unct

ion

, (a

nd P

luri-

harm

onic

equ

ival

ent)

such

that

:

�R

icci

Ten

sor i

s gi

ven

by re

mar

kabl

e E

xpre

ssio

n [E

rich

Käh

ler]

:

�A

nd s

cala

r cur

vatu

re :

� =

=n j

i

ji

jidz

dzg

ds1

,

2.

2

ji

jiz

zg

∂∂

Φ∂

=2

()

ji

lkji

zz

gR

∂∂

∂−

=de

tlo

g2 � =

=n l

klk

lkR

gR

1,

.

Φ

174

/17

4/

Air

Sys

tem

s D

ivis

ion

��� ��

�� ��

��� �2

��9��5

�3��� �7

���� ��$�)

���

Inth

e fr

amew

ork

of A

ffine

Info

rmat

ion

Geo

met

ry, m

etric

is g

iven

by

Hes

sian

of E

ntro

py :

�E

ntro

py fo

r Mul

tivar

iate

Gau

ssia

n m

odel

of z

ero

mea

n:

�In

cas

of C

ompe

x A

utor

egre

ssiv

e M

odel

of o

rder

n, E

ntro

py c

ould

be

expr

esse

d by

re

flect

ion

coef

ficie

nts

:

()

()

() e

nR

R�

πlo

gde

tlo

g~

−−

=-R

H�

��

gj

iij

=∂

∂∂

and

~

2

[]

[]1 0

1 1

2..

ln.1

ln).(

~−

− =

+−

−=�

απ

μe

nk

n)

(R�

n kk

n

[]

� =

− −−

==

−=

n kk

nn

n

xn

P1

20

1 0

1 12

1

1w

ith

.1 α

αμ

α(

)[

]∏

∏− =

−−

− =

−−

==

1 1

20

1 0

11

det

n k

kn

kn

n kk

nRμ

αα

175

/17

5/

Air

Sys

tem

s D

ivis

ion

�W

e de

fine

«D

oppl

er»

met

ric in

cas

e of

Com

plex

Aut

oreg

ress

ive

Mod

el b

y H

essi

an o

f Käh

ler P

oten

tial,

whe

re P

oten

tial i

s gi

ven

as in

In

form

atio

n A

ffine

Geo

met

ry b

y En

trop

y :

�K

ähle

r Pot

entia

l is

give

n by

Ent

ropy

par

amet

rized

by

refle

ctio

n co

effic

ient

s :

�M

etric

can

be

expl

icitl

y co

mpu

ted

:[]

[]1 0

1 1

2..

ln.1

ln).(

~−

− =

+−

−=�

απ

μe

nk

n)

(R�

n kk

n []

[]T

n nn

Tn

nP

)(

)( 1

11

0)

θμ

μθ

��

==

20

2 011

−=

=nP

ng

α(

)22

1

).(

i

ijij

in

δ

−−=

()

�− =−

−+

�� ���� ��

=1 1

222

2

002

1)(

.n i

iin

di

nPdP

nds

μμ

��� ��

�� ��

��� �2

��9��5

�3��� �7

���� ��$�)

��

176

/17

6/

Air

Sys

tem

s D

ivis

ion

����� �2� �

�� �����2

��9��5

�3��� �7

���� ��$�)

���

We

use

Ric

ci T

enso

r exp

ress

ion

give

n by

Eric

h K

ähle

r in

fram

ewor

k of

Käh

ler G

eom

etry

�In

Käh

ler G

eom

etry

, Ric

ci T

enso

r is

give

n by

:

�W

e ca

n co

mpu

te R

icci

tens

or fo

r Com

plex

Aut

oreg

ress

ive

Mod

el :

�Its

neg

ativ

e sc

alar

cur

vatu

re is

giv

en b

y :(

)j

i

lkji

zz

gR

∂∂

∂−

=de

tlo

g2

()

� �

−=

−−

=

−=

1,..

.,2

for

1

2

12

22

2 011

nk

R

PR

kkllk

μδ

�=

lk

lklk

Rg

R,

.∞

−→

� ��� ��

−−

=∞

− =�n

n jj

nR

1 0)

(1

.2

177

/17

7/

Air

Sys

tem

s D

ivis

ion

3��� �7

���� ��$�)

���L��

�� 1-��������$

�� ��

�Pr

evio

usm

etric

isno

t a K

ähle

r-Ei

nste

in m

etric

, but

a c

lose

mat

rix

stru

ctur

e�

A m

etric

isca

lled

Käh

ler-

Ein

stei

nm

etric

if its

Ric

ci te

nsor

ispr

opor

tiona

lto

the

met

ric:

�In

cas

e of

Käh

ler-

Ein

stei

n M

etric

, Käh

lerP

oten

tiali

sso

lutio

n M

onge

-Am

père

E

quat

ion

:

�Fo

r Com

plex

Aut

oreg

ress

ive

Mod

el, w

eha

ve :

[]

[]

[]

()

{}

,....,

2

whe

rean

d

)(

1.2

with

1)

(

1 0

)(

)(

− =

−−

=

� ��� ��

−−

==

=�

in

diag

B

jn

BTr

Rg

BR

n

n j

nij

nij

()

ji

ji

lklk

jiz

zk

zz

gk

gk

R∂

∂Φ

∂=

∂∂

∂−

�=

2

0

2

00

.de

tlo

gco

nsta

nt :

w

ith

.

�� =

Φ−

func

tion

holo

mor

phe

Pote

ntia

lK

ähle

r

with

)

det(

02

� :

� :

eg

klk

ψ

178

/17

8/

Air

Sys

tem

s D

ivis

ion

���)

�����2����

�����

�� 1(���

�����"���� �2

3(�$�)

���

Intr

insi

c G

eom

etric

Flo

w a

re o

f fun

dam

enta

l int

eres

t in

Mat

hem

atic

&

Phy

sics

bas

ed o

n va

riatio

nal A

ppro

ach

:�

Käh

ler-

Ric

ci F

low

:

�C

alab

i Flo

w :

�In

egal

ity b

etw

een

both

func

tiona

ls

Hilb

ert A

ctio

nK

ähle

r-R

icci

Flo

w

=

M

M

gdV

gdV

gR

R)

(

)(

)(

M

gdV

gR

)(

)(

jiji

jig

nRR

tg+

−=

∂∂

Cal

abi A

ctio

nC

alab

i Flo

w(

)*

*2

, ji

jiz

zz

zg

∂∂∂

Cal

abi F

low

=

M

gdV

gR

gS

)(

)(

)(

2

*

2

ji

ji

zz

Rtg

∂∂∂

=∂∂

RR

t−

=∂∂φ

�� ���� ��

≥M

M

gdV

gdV

gR

gS

)(

/)

()

()

(2

179

/17

9/

Air

Sys

tem

s D

ivis

ion

In 2

dim

ensi

on o

f the

com

plex

var

iabl

e, C

alab

i & K

ähle

r-Ric

ci

flow

s ar

e cl

osel

y re

late

d :

�C

alab

i & K

ähle

r-R

icci

Flo

ws

can

be re

late

d by

:

�If

seco

nd d

eriv

ativ

e ac

cord

ing

to K

ähle

r-R

icci

flow

tim

e is

id

entif

icat

ed w

ith o

ppos

ite o

f sec

ond

deriv

ativ

e ac

cord

ing

to C

alab

i flo

w ti

me

, bot

h flo

ws

are

equi

vale

nt:

�����

#�����(

����������

�"��

��2����

�����

�� 1(���

�����"�

()

()

()

� �

−=

−=

∂∂

� ��� ��

∂∂

∂∂

∂=

∂∂

� �

∂∂

∂−

=

∂∂−

=∂∂

�−

=∂∂

�R

Rg

tg

tg

zz

tg

zz

gR

tRtg

Rtg

ji

jiji

ji

ji

ji

ji

jiji

jiji

,

*

2

2

2

*

2

2

2

)de

t(lo

g

)de

t(lo

g

)de

t(lo

g

CR

ttt

∂∂−

=∂∂

22

*

2

2

2

ji

ji

zz

Rtg

∂∂∂

−=

∂∂�

180

/18

0/

Air

Sys

tem

s D

ivis

ion

Giv

en K

ähle

r met

ric :

����

�����#

��������"

����

�������

��

�� ��

����

����

*)

,,

(2

.2

*

dzdz

eds

tz

=

*

2

*z

zR zz

∂∂

Φ∂

−=

()

an

d

)de

t(lo

g*

**

=∂

∂∂

−=

eg

zz

gR

zzzz

*

2

*

2

*

2

..

w

ith

zz

et

zz

te

zz

teR

tgji

ji

∂∂∂

ΔΦ=

∂Φ∂�

∂∂

Φ∂

=∂Φ∂

�∂

∂Φ

∂=

∂∂�

−=

∂∂

Φ−

ΦΦ

Käh

ler-

Ric

ci F

low

ΔΔΦ

−=

∂Φ∂�

∂∂

ΔΦ∂

−=

∂∂�

ΔΦ−=

∂∂

Φ∂

−=

=∂

∂∂=

∂∂

Φ

Φ−�

t

zz

te

zz

eR

gR

zz

Rtg

ji

jiji

ji

*

2

,*

2

*

2

an

d

Cal

abi F

low

181

/18

1/

Air

Sys

tem

s D

ivis

ion

Bas

ed o

n pr

evio

us R

ao m

etric

com

puta

tion

for

Aut

oreg

ress

ive

Mod

el :

�K

ähle

r-R

icci

Flo

w o

n R

efle

ctio

n C

oeffi

cien

ts :

�C

alab

i Flo

w o

n R

efle

ctio

n C

oeffi

cien

ts :

��� 1(�������)

�2����

�����"���� �2

3(�$�)

��

()

22

2

00

)(

22

2)0(

)(

an

d

.)0(

1)

(1

nt

in

t

ii

eP

tP

et

−−

=−

=−

μμ

()

�− =

−−

+�� ��

�� ��=

1 1

)(

4

22

24

2

0

02

22

)0(1)

()0(

)(n i

in

t

i

int

ne

di

ne

PdPn

tds

μμ

()

nti

nt

ii

eP

tP

et

−−

−=

−=

−)0(

)(

an

d

.)0(

1)

(1

00

)(

22

μμ

()

�− =

−−

+�� ��

�� ��=

1 1

)(

2

22

22

2

0

02

)0(1)

()0(

)(n i

in

t

i

int

ne

di

ne

PdPn

tds

μμ

����������� �����

Thal

es A

ir S

yste

ms

Dat

e

$���

������

���

183

/18

3/

��(�0�-(��!!��4$3'(�D��0

�!($

3'�!

0��-!

$-'(E

6�;IIM

�9�7

��<

http

://w

ww.

info

rmat

iong

eom

etry

.org

/MIG

/MIG

BO

OKW

EB

/

184

/18

4/

/������

����

J��9�

������

���

�)��� �

����

� ��

«Le

s an

imau

x se

div

isen

t en

: a) a

ppar

tena

nt à

l’Em

pere

ur,

b) e

mba

umés

, c)a

ppriv

oisé

s, d

) coc

hon

de la

it, e

) sirè

nes,

f)

fabu

leux

, g) c

hien

s en

libe

rtés

, h)in

clus

dan

s la

pré

sent

e cl

assi

ficat

ion,

i) q

ui s

’agi

tent

com

me

des

fous

, j)

inno

mbr

able

s, k

) des

siné

s av

ec u

n pi

ncea

u tr

ès fi

n en

poi

l de

cham

eau,

l) e

t cet

era,

m) q

ui v

ienn

ent d

e ca

sser

la c

ruch

e, n

) qu

i de

loin

sem

blen

t des

mou

ches

»En

cycl

opéd

ie c

hino

ise

tiré

d’un

text

e de

J.L

. Bor

ges

(En

préf

ace

de l’

ouvr

age

«Le

s m

ots

et le

s ch

oses

» de

M

iche

l Fou

caul

t)«

La m

onst

ruos

ité q

ue B

orge

s fa

it ci

rcul

er d

ans

son

énum

érat

ion

cons

iste

en

ceci

que

l’es

pace

co

mm

un d

es re

ncon

tres

s’y

trou

ve ru

iné.

Ce

qui e

st im

poss

ible

, ce

n’es

t pas

le v

oisi

nage

des

ch

oses

, c’e

st le

site

lui-m

ême

où e

lles

pour

raie

nt v

oisi

ner.

… L

es c

hose

s y

sont

‘cou

chée

s’,

‘pos

ées’

, ‘di

spos

ées’

dan

s de

s si

tes

à ce

poi

nt d

iffér

ents

qu’

il es

t im

poss

ible

de

trou

ver p

our e

ux

un e

spac

e d’

accu

eil ,

de

défin

ir au

des

sous

des

uns

et d

es a

utre

s un

lieu

com

mun

. … S

ur q

uelle

‘ta

ble’

, se

lon

quel

esp

ace

d’id

entit

és, d

e si

mili

tude

s, d

’ana

logi

e, a

vons

nou

s pr

is l’

habi

tude

de

dist

ribue

r tan

t de

chos

es d

iffér

ente

s et

par

eille

s ?

»M

iche

l Fou

caul

t «Le

s m

ots

et le

s ch

oses

»

Jorg

e Lu

is B

orge

s

185

/18

5/

' ��

���.���

��8��

�)����� �

����

����

��

«La

tram

e sé

man

tique

de

la re

ssem

blan

ce a

u XV

Ièm

e si

ècle

, est

fort

rich

e : A

mic

itia,

Ae

qual

itas

(con

trac

tus,

con

sens

us, m

atrim

oniu

m, s

ocie

tas,

pax

et s

imili

a), C

onso

nanc

ia,

Con

cert

us, C

ontin

uum

, Par

itas,

Pro

port

io, S

imili

tudo

, Con

junc

tio, C

opul

a.

Mai

s il

y en

a q

uatr

e qu

i son

t, à

coup

sûr

ess

entie

lles

:•L

a C

ON

VEN

IEN

TIA

: est

une

ress

embl

ance

liée

à l’

espa

ce d

ans

la fo

rme

du ‘p

roch

e en

pr

oche

’. El

le e

st d

e l’o

rdre

de

la c

onjo

nctio

n et

de

l’aju

stem

ent.

Elle

app

artie

nt m

oins

au

x ch

oses

qu’

au m

onde

dan

s le

quel

elle

s se

trou

vent

.•L

’AEM

ULA

TIO

: sor

te d

e co

nven

ance

affr

anch

ie d

e la

loi d

u lie

u, q

ui jo

uera

it im

mob

ile

dans

la d

ista

nce.

Les

ann

eaux

ne

form

ent p

as u

ne c

haîn

e co

mm

e le

s él

émen

ts d

e la

co

nven

ance

, mai

s pl

utôt

des

cer

cles

con

cent

rique

s, r

éflé

chis

et r

ivau

x.•L

’AN

ALO

GIE

: affr

onte

men

t des

ress

embl

ance

s à

trav

ers

l’esp

ace.

•le

JEU

X D

ES S

YMPA

THIE

S: n

ul c

hem

in n

’est

dét

erm

iné

à l’a

vanc

e, n

ulle

dis

tanc

e n’

est

supp

osée

, nul

enc

haîn

emen

t pre

scrit

. Elle

par

cour

t en

un in

stan

t les

esp

aces

les

plus

va

stes

. Sa

figur

e ju

mel

le, l

’ant

ipat

hie

mai

ntie

nt le

s ch

oses

en

leur

isol

emen

t et e

mpê

che

les

assi

mila

tions

.En

fin, l

l n’y

a p

as d

e re

ssem

blan

ce s

ans

sign

atur

e. L

e sa

voir

des

sim

ilitu

des

se fo

nde

sur l

e re

levé

des

sig

natu

res

et s

ur le

ur d

échi

ffrem

ent.

»«

Les

mot

s et

les

chos

es»

de M

iche

l Fou

caul

t

186

/18

6/

���� �

� �

Col

loqu

e de

rent

rée

Col

lège

de

Fran

ce 2

011

: La

vie

des

form

esA

nne

Fago

t-Lar

geau

lt-L

a fo

rme

chez

Pla

ton

et A

risto

teht

tp://

ww

w.co

llege

-de-

franc

e.fr/

docu

men

ts/a

udio

/Col

rent

ree1

1/co

l-fag

ot-

larg

eaul

t-201

1101

3.m

p3ht

tp://

ww

w.co

llege

-de-

franc

e.fr/

docu

men

ts/v

ideo

/col

loqu

e11/

Fran

cais

/col

-fa

got-l

arge

ault-

2011

1013

.mp4

Mot

«Fo

rme

» en

Gre

c: M

orph

è, E

idos

& Id

eaPo

ur P

lato

n, il

s’a

git d

e «

Met

tre

en é

vide

nce

des

Rel

atio

ns e

ntre

El

émen

ts: d

es R

elat

ions

app

réhe

ndée

s pa

r l’in

telli

genc

e

et

non

des

Elé

men

ts to

uché

s pa

r les

sen

D’a

bord

écr

it «

enfo

rmer

», l

e m

ot «

info

rmer

» a

ppar

aît e

n fra

nçai

s en

128

6,

empr

eint

au

latin

« in

form

are

», li

ttéra

lem

ent «

don

ner u

ne fo

rme

».

Le

mot

« in

form

atio

n »

appa

raît

conj

oint

emen

t au

XIII

ème

sièc

le.

De

l’éty

mol

ogie

gre

cque

, ����

, mor

phê

(« fo

rme

»), n

ous

est p

arve

nu le

se

ns d

e m

orph

olog

ie, l

a sc

ienc

e de

s fo

rmes

. Le

con

cept

fond

amen

tal s

embl

e do

nc ê

tre c

elui

de

« fo

rmes

» e

t de

cite

r C

icér

on: «

mat

eria

, qua

m fi

ngit

et fo

rmat

effe

ctio

»

(la

mat

ière

que

mou

le e

t met

en

form

e la

forc

e m

otric

e).

187

/18

7/

/*.

��������2 .�� ��

����/��)�

��� ������

��� �

��1�

� 7���

Lede

veni

retl

afo

rme

…O

r,la

vie

est

une

évol

utio

n.N

ous

conc

entr

ons

une

pério

dede

cette

évol

utio

nen

une

vue

stab

lequ

eno

usap

pelo

nsun

efo

rme,

et,

quan

dle

chan

gem

ent

est

deve

nuas

sez

cons

idér

able

pour

vain

cre

l'heu

reus

ein

ertie

deno

tre

perc

eptio

n,no

usdi

sons

que

leco

rps

ach

angé

defo

rme.

Mai

s,en

réal

ité,l

eco

rps

chan

gede

form

tout

inst

ant.

Ou

plut

ôtil

n'y

apa

sde

form

e,pu

isqu

ela

form

ees

tde

l'im

mob

ileet

que

laré

alité

est

mou

vem

ent.

Ce

quie

stré

el,c

'est

lech

ange

men

tcon

tinue

lde

form

e:l

afo

rme

n'es

tqu'

unin

stan

tané

pris

suru

netr

ansi

tion.

…Q

u'il

s'ag

isse

dem

ouve

men

tqu

alita

tifou

dem

ouve

men

tév

olut

ifou

dem

ouve

men

tex

tens

if,l'e

sprit

s'ar

rang

epo

urpr

endr

ede

svu

esst

able

ssu

rl'i

nsta

bilit

é.Et

ilab

outit

ains

i,co

mm

eno

usve

nons

dele

mon

trer

,àtr

ois

espè

ces

dere

prés

enta

tions

:1º

les

qual

ités,

2ºle

sfo

rmes

oues

senc

es,3

ºles

acte

s.…

Ace

str

ois

man

ière

sde

voir

corr

espo

nden

ttro

isca

tégo

ries

dem

ots

.Les

adje

ctifs

,les

subs

tant

ifset

les

verb

es,

qui

sont

les

élém

ents

prim

ordi

aux

dula

ngag

e.A

djec

tifs

etsu

bsta

ntifs

sym

bolis

ent

donc

des

état

s.M

ais

leve

rbe

lui

mêm

e,si

l'on

s'en

tient

àla

part

ieéc

lairé

ede

lare

prés

enta

tion

qu'il

évoq

ue,n

'exp

rime

guèr

eau

tre

chos

e.«

L’év

olut

ion

Cré

atric

Hen

riB

ergs

on,1

907

188

/18

8/

/*.

��������2 .�� ��

����/��)�

��� ������

��� �

��1�

� 7���

Laph

iloso

phie

des

form

eset

saco

ncep

tion

dude

veni

r.Pl

aton

etA

risto

te.

Pent

ena

ture

llede

l'int

ellig

ence

Lem

otei

dos,

que

nous

trad

uiso

nsic

ipar

Idée

,aen

effe

tce

trip

lese

ns.I

ldés

igne

:1�

laqu

alité

,2�

lafo

rme

oues

senc

e,3�

lebu

tou

dess

ein

del'a

cte

s'ac

com

plis

sant

,c'e

st-à

-di

re,a

ufo

nd,l

ede

ssin

del'a

cte

supp

osé

acco

mpl

i.C

estr

ois

poin

tsde

vue

sont

ceux

del'a

djec

tif,

dusu

bsta

ntif

etdu

verb

e,et

corr

espo

nden

tau

xtr

ois

caté

gorie

ses

sent

ielle

sdu

lang

age.

…ei

dos

estl

avu

est

able

pris

esu

rl'i

nsta

bilit

éde

sch

oses

:la

qual

itéqu

iest

unm

omen

tdu

deve

nir,

lafo

rme

qui

est

unm

omen

tde

l'évo

lutio

n,l'e

ssen

cequ

ies

tla

form

em

oyen

neau

-des

sus

etau

-des

sous

dela

quel

lele

sau

tres

form

ess'

éche

lonn

ent

com

me

des

alté

ratio

nsde

celle

-là,e

nfin

lede

ssei

nin

spira

teur

del'a

cte

s'ac

com

plis

sant

,leq

uel

n'es

tpoi

ntau

tre

chos

e,di

sion

s-no

us,q

uele

dess

inan

ticip

éde

l'act

ion

acco

mpl

ie.

…N

ous

disi

ons

qu'il

ya

plus

dans

unm

ouve

men

tqu

eda

nsle

spo

sitio

nssu

cces

sive

sat

trib

uées

aum

obile

,plu

sda

nsun

deve

nir

que

dans

les

form

estr

aver

sées

tour

àto

ur,

plus

dans

l'évo

lutio

nde

lafo

rme

que

les

form

esré

alis

ées

l'une

aprè

sl'a

utre

.…

Iltr

ouve

raqu

ela

form

ees

tes

sent

ielle

men

tét

endu

e,in

sépa

rabl

equ

'elle

est

dude

veni

rex

tens

ifqu

il'a

mat

éria

lisée

auco

urs

deso

néc

oule

men

t.To

ute

form

eoc

cupe

ains

ide

l'esp

ace

com

me

elle

occu

pedu

tem

ps.

…Le

sfo

rmes

sens

ible

sso

ntde

vant

nous

,to

ujou

rspr

êtes

àre

ssai

sir

leur

idéa

lité,

touj

ours

empê

chée

spa

rla

mat

ière

qu'e

lles

port

ent

enel

les,

c'es

t-à-d

irepa

rle

urvi

dein

térie

ur,p

arl'i

nter

valle

qu'e

lles

lais

sent

entr

ece

qu'e

lles

sont

etce

qu'e

lles

devr

aien

têt

re.

«L’

évol

utio

nC

réat

rice

»H

enri

Ber

gson

,190

7