2 let plt b. m density fzgteproblab.ca/louigi/courses/20172018/math547/bbm_kpp.pdf · 2018. 4....
TRANSCRIPT
Brownian motion and the heat equation .
�1�
Prep : Fix 9 :# R"
nice"
,let vk.tl = E
,.[ g ( BH ) ) ]
.
Then o¥v( x. f) =ov=:tzf÷zv( x. t)(
[ 2,
bounded
Lemmas : Let Plt , x. g) =
fzgte" " " " "
be B. M.
transition density . Then
ftp.x.ykoplt.x.y ) .
Proof pH , x. y ) =P H,
x -
y , 0 ) so can assume y- 0 .§
PH ," ,d=- ztfeiiht + 1- e-
" 42?
¥,= pit , x. o ) (2¥ - zt )
Fhtt
§=
plt.x.ot-PH.x.dz#o2z.pH.x.d=pH.x.dfz2g+4zI)=2fpH.x.o:
Proof of Proposition
¥ vk.ti.gg/ngcy)pH.x.y)dy=/ngcy)dqPH.x.y)dy=f.gcyzg,2gpH.x.y)dydt
'
= £ Ifngcyipct, x. g) dy=0VC' '
.t )
⇐-
°
0 Exlg CBHID
Branching Brownian Motion
yEx →
starting position x ( default ; so )�2�
- Time 0 : Particle at
xeR ( Rdl '
- Time t :
Dt=# particles at time t .
Particles at positions (
X.HI, ...
,
Xottl)
. Particles branch at rate 1 C i. e. Time for a particle to branch is EXPN - distributed ;
or,
PC fixed particle branches eftttdt ) ) - dt )Tt
- Particles more as Brownian Motion
- Particles move and branch indep . from one - another.
wvmar4NB When
k particles ,time to branch is m ;n( Ei
,
ieiek ) where E, ,
. .Et are independent Expa ) .
So,
( Dt ,t 30 ) is a pure birth process with rates 7 ,.=k
Filtration of BBM is Feokxicstie Ds ) ,set ) = all info
. aboutBBM up to time t
Pap ( Fisher - KPP ) Fix fR→lR nice,
write ZHI :
II. FCXICH) �3�
ulx.tl = IEXZH ).
Then osfucxtl= On ( x. t ) - ucxt ) ( I - uc at )).
proof" Backward equation
"- Think about what happens in time ( ahl
,h
smallu( x.
tth) = IE
,fZ(t.tt/=lE,c/7(t+h)/
no branching in 6. h) ) .IN No branching in (
ah) ]
+ E×[ Z ( tth ) / branching in (ah )|P[ branching inCah )) (#
Hold t fixed,
let gcx ) = ul x. t ),
v( g. s ) - IEY( g ( Bcs ) )).
Then ffv ( g. s ) =
ovcy,s )
Exffltth)|no branching in 6.
41= InEYIZLHIPH,x.g) dy =),nuly.t)p(hx,y)dy=v(x,
h)
# =v ( x. at hDv( x. o ) + 01h ) =ubstltHOUCx. t ) + 01h )
P( No branching in ( ah) ] = I - h + och )
E ,dZ(tth ) / branching in ( ah) ] = uln , f) 2+0 ( h) P[ branching in ( ah))= htolh )
: ( x. tth ) = ( i - htolh ) ) @( x , D thou( x. Htoch ) ) + ( htolh ) ) ( U ( x. f)2+0 ( h ) )
= ( 1- buthow + hit + och )
sodycx.tl = DUGST ) - ucx ,
f) ( tucx ,t )) Dat
�4�
Important special case : fcx )= 1 no
Then
Egftp.fkiltl ) =
Pg( min ( Xict )
,ie Dt ) >
OlyPo ( max ( XICH ,
is
DDEY)
X. (
H=y- Xia ) gives BBM from 0 ,
and min ( Xilt ),
is Dt ) > 0 < ⇒ max ( Xi ( t ),
isDDEYThe ( Neveu ) tiny t'
'm. IF
I
NB.
This implies that from Heaviside initial condition o
Ul x. 0 ) = Tfxzo x=o
f t > o, ulx ,t ) is increasing in x ( and CT in fact )
,
and too, tianya.nu#=otlIId?k+atu"
' t' - l'
Prop :
IEDC= et for all t �5�
Proof :
EDt+h=µ§MDt=klE[Dt+hIDEK]
=
§,PlDt=k) . ( Ktkhtockh ) ) =( lthtoch )
)ED£so ddqEDT =
EDT,
and
EDo=Do=l. D
Corollary : Let Mt =
imax,
Xilt ) = maximal position of a particle at timetD
Thenflimsy My gfz almost surely
Proof If lineup Ptt > Fate,
eecon ) then can find times ( tn )na ,tn+ ,
? tntl,
St . My > F. + £ for all n.
IfMtzfktfttn
then there is i st.
Xiltn ) 3152+ E) tn .
For a Bmstarted fromany
ice R,
P,c( BH ) > x ) =L.
so
PolMrt .it#EHnl2+n.Mtn3Cr+EItn) ?± .
3 ( if + E)ftn7 - i ) > ( f+§ ) # for n large
�6�
So if lizards, if + ethen can find
anincreasing seq
. ( Sk ) # 1 of integer times
s.t.MS, ¢ + f) V. k
Sk
Each time - n particle has positionBut for any
of > 0,
NEN , with dist Ncanl ± if Z,
Z~N( al )
P( Mn 's (52+8) n ) =
§P( Mink 'S)n/Dn=k)P(Dn=k)K
s E H P ( AZ > ( Eton ) p(Dn=k ).
+ E I . P(Dn=k )KerenKnier
s menextpc
. (( rzts)rn)% ) + PC Dn ? n' en )
s nze' on
+ IE
Dnlrien= n2 @-
on+ n
- 2
Son,§P( Mn ? fztrh ) < a
0%
Lenya For Z ~ Ncol ), @
PCZ = x ) >
rffexe' '%
for xs ,016 )=¥e -
%> ¥Fe e-
542for se t.tt
proof PCZ . x ) ± IPIZEK ,xt±H=k" Finds ¥f⇒e×e⇒2o
Proof of Neveu 's Thm : Already showed lifeaupmtlttif. Remains toprove life,sgpMtA¥ik .
f. i. recoil) For ( E - d) t ? I,
e-( rzrttk
,et ECFEH
.
-
p( tkz > ( if . d) t ) = P( Z 3 ( E - d) th ) 7 rate . ( rent't 14 eoet
' '
Since ED+= et and or - E > %
lE[ # { is Dt : XIH ) > ( rz . dlt } ]= et . P( tkz > ( F . of t )
-ITt->
, @
onYruttot' "
Fix T = TW ) big enough that
eottyrttert"z > I
. Icp ICT )
Then IEIIOI> I.
For each jetolet
Ij = { it DKT ) : Xi(2T ) a descendant ofXJCT)
,X:( ZT ) -Xj ( T) 3 ( if - d) T }
.
Then ( Ij , je Io ) are conditionally indep . given Io ,and all have dist
.
of I¢ .
�8�
Continuing this way gives a brandingprocess :
Ij , , . → jµ ,
= { is D*+, )t:X :( HADa descendant of Xjw ( ttl , Xi CKTDTI - X ;kCkT) > af . d) T }
Any particle in gen . g of this process has X ; ( 9T ) > ( rz . on )gT .
So if the B. P.
survives then fiminaf M¥ > ligninaf lgng¥> rz - or
.
And the process survives with positive probability since supercritical !
This gives PC timid Mitt > R - o ) > e > o . But
P ( timid Mitt < R - o ) =P ( timid Mitt < R - or )2 ( consider first branch time ) .
so =o
so PC lining Mitt > R . o ) = I as