0% found this document useful (0 votes)
30 views8 pages

Existence of Solutions in Gradient Flows

The document discusses the existence theory for gradient flows within the framework of Gelfand triples and semicoercive functionals. It outlines the definitions and propositions necessary for establishing the existence of solutions to the gradient flow problem, employing methods such as implicit time discretization and a priori estimates. The main results demonstrate the conditions under which minimizers exist and converge to a solution of the gradient flow equation.

Uploaded by

miru park
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
30 views8 pages

Existence of Solutions in Gradient Flows

The document discusses the existence theory for gradient flows within the framework of Gelfand triples and semicoercive functionals. It outlines the definitions and propositions necessary for establishing the existence of solutions to the gradient flow problem, employing methods such as implicit time discretization and a priori estimates. The main results demonstrate the conditions under which minimizers exist and converge to a solution of the gradient flow equation.

Uploaded by

miru park
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Existence theory for gradient flows

we start off byrecalling some preliminary definitions


Definition I Let c X Hux 0 be a Gelfand triple and let us fix a moEW

Furthermore let us consider a functional I X IR Freshet differentiable

with Freichetdifferential DI We say that a Efc Coot X is a solution to the

H gradient flow for I if I e I e coots x and

detour t estructour 0 for all N E X and almostevery feat


x x x
x
nd ee Coo No

Definition 11 semicoercive functional

Let C X H X be a Gelfandtriple Afunctional I X IR is said to besemicoercive if

there exists c or and czEIR such that

I Cro 7 Citrus cullviper for all u EX


To construct a solution to the gradient flow problem wefollow Rothe's method

In the first place we consider an implicit time discretization of the trinederivative


and we establish existence of approximations Namely we have the following proposition

Proposition CJmplicit Euler scheme

Let ex H x be a Gelfand triple and let X IR be lowersemicontinuous and

semicoercive functional N f 7
Then for
every T so such that 44oz L and K l

NK i 112N
the functionals a Ik cus n Icu with no no

haveminimers that satisfy dguk v DI reeks cus O HN E X


µ
and dguk e cute nie l
Ip
The ez appearing in 4TCz 1 refers for the ez in thedefinitionof semicoercivity of I

Proof Fromtheassumptions made abovewededuce that I Kcmo is a coercivefunctional Intact


nuns culinity
Ikea I Hu nk l
2T
t I cud z
2
Hee ar Iff te
if c EO CHURN 7,0 Itecud ogc unify callWfp 3 c IIhis

It is boundedfrombelow
and it is coercive

ref is so Ik cus I 2
Hee ar Iff te nails annuity

g 2oz Ya uh 1112N to 11h45 callNfu

scufinifutoute Itu 2 Lee uh to null CzilulRu


I f
2 Cwlink Iff t 26211WRN yousee uh I µ to Hull Czkmiffy
7,2oz link 112µ t 2ozHulky 4ozHullµflute fly to Nulty eztuff

920211we ifµ 4callallµlink IIµ to 11m11s t callmiff

72oz link iffy calmsnullg him iink ily to Nultytwinifu


72callute lift 2µ 46211we Iff t e nUNITczµ2w

Sy
c 11h11 2 Cz link I 112N

From this we obtain that when ou 0

Ik r w g c Nulty 20211we Ifw


Hence Ike us too and Ike us is boundedfrombelow and moreover Iki's
Hull too
weakly lowersemicontinuous Therefore fromthe directmethodof calculusof variations

we obtain the
existence
of a minimiser uh for IR

Since I and n un we 1112N are F r echef differentiable wededuce that

Uk satisfies
e l
t BI euro ero
O Fr e X
k w µ Oma

From the previous proposition we have obtained a sequence of minimisersLuteKelo N


which are a solution for the discretisedcalls'oretised intime gradientflow

Gnorder to conclude we need to prove two further results


i we need an a prioriestimates whichholdsuniformlyin T o so that we can

extract a convergentsubsequence ons T trO

the limit point of thesubsequence satisfies the gradientflowequation


Fororder to prove an a priori bound we needstrongerassumptionson the functional I

Definition ill Let r x N X be a Gelfandtriple Le f I X SIR be Freichefdifferent

functional we saythat SI X 0X is semicoercive andbounded if I peel too 4 0 ke


and e such that I
SIC u v z dyn riff Cz wtf tf r E X
and
HDICroll Idg itHull I Hu E X

Once this is in place wecanstate and prove the main result whichallows us to establish an
priori estimate for thesequence we eeeLo N
Proposition l A prioriestimates o Lef c x H x o be a Gelfandtriple left x IR be on

Fri chefdifferentiablefunctional and let us suppose that SI x x is semicoercive and bounded

If 4T e e l where c is the one appearing in Definitionill then wehavethat


N N N
we ex t T T Dermot Eco
III J H T
E E nd well
t
E
Where Co depends on p T no 4 CLandCj
Proof We knowthatluteSrego N are minimisersof the corresponding funotrinds

Ik cus
If u n un Iff t
I Iue respectively

Thus uk solves

If uk DI cute Cvo O H N E X
µ t w

Bysetting v nie andby using thesemicoercivity of SI weobtain


dguk we wt stink rules t
IldenteIlf t I dyIlunify t oflinklip ednullify E
Iz
IslidtukifutIdflinkIfn Is dear dear µ t funk112W link ily
nie r nee i
e nie un l we we I
µ
1 4 we we µ µ

lily uh we I t E linklifer
ET vwhiff toute z
2T
I nullify cuteuh I µ cute nk i we µ Ifweuh µ

Once this is in place we multiply the vis of b for T and sum over ie l P wit h
i e e eN As a result we obtain
I e
Indentelf tei f e nwerP Tch interfere 0 ee Pe N
t nullify truth Tag w X
K4
wherewehaveusedthe following fact abouttelescopic sums
Harith ruleinn fineIif nwo
delinkifu
If
Byrearranging the addends we obtain

1 linen
2
t t Indemnify to it finnienf stringenttea purify
P
since od e I Tod Tc's
junkif
we
lift tea purify
Therefore term 1411UPIfµ can be absorbed on left handside obtaining
PI
e InnoirutTc's EIntern
f rueihr t Tef Etdealer t ofT req unn
f
For P o N we now set

I onether t dT linkup
ye
Fromthis wededuce that
PI
P off linkif I Innov2ut Tc Eyukky
y I neither t
ei
Filedif t 41e's E Irwelfy
bI

e Ennohut414 y
k

Therefore we havethat

ye e Inmolten tote's ik i see N

Byapplying the discrete Gronwalllemma we see that


Nl
Mat P e mother expC ago44cL
P lo N
y

mas
P t N
y
P
pm
b N
X
f'sonehut dt
E we

Max I line It off well


f
P I N

Therefore wefinally have


N
N
[Link]
I line It off turf statuary exe jotted
l N
L noir exp4Th N
t IN N
kit
n t 4T EN well s Inneifu e grunge
e nl
we reranthat N fNq therefore TIN 10 IT while TN DT
N
T and T dentelle remain tobe proven
Theboundfor E dem'll

Ynorderforddthat theboundednessof SI comes in ourhelp In fact

I troop citrate 0
T
E strut on e T E G's citrarip DP e

werecallthefollowingfact poop1 1 andthat off linkIlf I Hunkyet


ftip I o

e d Teej P I t linkup I w P TN t d top i


est E Nuteil f
l OP T f zP I r iz P I
I Il ruoff e44T
2
Therefore wehave obtained

strut on E d P 1
T tog T t
ruoffet
As for the term T nd well we use the estimate previouslyproven

we recall that cute is a solutionfor the discretised gradient flow namely

If we er 7 SI cute0 c u 0 0 Kv EX
H
Therefore Il d f URYya HDI cute011 8
Hence
N
of 11 DI cute011 I lost T t P's T
T dfuk11
if 2
runtye4
toes

In order to identify the limits of theapproximations wedefineinterpolants of


the approximations
Theinterpdants of LukSaefo canbe definedinthreealternativeways
n

Let set N and fee KT Keo N


Ig
us

l
n f Mk f E Cfee yofK K l N
p
mtg t Uk u f C fee tie Kel N

If it o
tI uk t t
T
we 1 Rsi N
r fer
as
n ut n
ng

I l l S l l l S l l l s
T 2T 3T T 2T3T T 2T3T

Ourgoalis totranslatethe a priori estimates for


ate
go ngf0
NT I and IT
Thisis thepurpose of the next Lemmon

Lemmon Discreteevolutionequation

If we let Uf et andthey be the interpolants defined above them we have

It Cts 7µF SI cuteto vs D Hr et and almostall fearTO

For all To the following estimate holds

htt ut pic T int wiretaps


t
[Link] cuttle're x 5 Coo
Proof
the derivative of fur is trifled MI MI delete of El tr r tr es N
T
letCfs Up f E eteegotµ test N

Therefore since ed we er secure cus O Fr E X Hk I ON


is
it follows that r µ DIR over D A NET andalmostallofECUT
oilyego

From the following estimate previouslyproven


N N N
Dermot Eco
max
P oI
Irwin
N
t T
E wellf t T
E nd well t T
ee

we deducethat

fTnutotonfdx et EIturtle eco

11upIly Co
sup
CCCo T
Nt
11 Holly e max
P outs N
Thus theuniformboundsforthe firsttwo addends are thus proven

As for the fourthterm we use the boundednessof DI


T
of horntail dt got4 it outgeton t'dt
T
e Cobo
I fit nut roll f dt

e se coff IT too
n
Since ilmyHolly II DI interfully we obtainthat

q
Itu TigreIly df 111strutrusty dt s d
coff f T t co
and
firing iton It
III ingreen dt

wenotethat IT can be alternativelywritten as I geto mtg f ofee f figCfd therefor


we obtain

el f
fell at Ho ite fo fig roll If
te i

pl I I
e s
af St natron it t
E Yen et improv at

se if int oI dttrEi I niiheooI dt

d i
f nutpetolp If t 2P Jatin still It
e I I t
supNuffHildy D i gotnutriton df
feast
Thus wehaveobtained

firing it on dfs d p tooth D Sotnutritonefadt


ts
T 11Mtfp I'tIlplc out XA
yzP
I n
soft N
And this completes the ants
a priori estimates for the interpol

To sum
up so far we haveproved that thelinterporlonts ut ng Ip satisfy
the following estrindJ uniformly in T o

htt [Link] at pre it int wiretaps ht re x 5 Coe


ABB

This bound for the Interpolants allow us for extract a convergent subsequence

which willbe the solution to a sort of gradientflow problem


The next proposition the proof forwhich is omitted gives a precise result
onwhat sense the interpolants convergetowards a limit point a and whatkind of
Gradient flow equation is satisfied
by u itself
Proposition selectionof a limit Let Ix H X I be a Gelfandtriple andlet us

further assume that X is compactly embeddediin It


Thenthere exists me RicoTJ X P
n w eco x and 3 ELP e Coo x suchthat

t N u LN
Up y in out w
To
a
u in LP Con x
T ut y o
Uf w in w Pecoots X
Tko
in ee e cops x
stint pyo
Moreover we havethat u e cccoots H and a is a solutionto

detourf t 3K0gr o Tve X foralmost all te Coo


2
are me

Gnaddition if z stew then u is a solution for the Gradient flowproblem for I

To sum up this scheme of proving the existence of a solutionforthe gradientflowproblem

for I can be described in four steps


discretised version
Id Yn the first place weprove the existence of a solution for a

of the gradient flow problem

I we provide on a prioribound satisfied by these solutions Such bound allows

us to extract a convergent subsequence as T to a

II we prove the limit point satisfies a generalised Gradient flow equationsuchasL2

IT Underfurther assumptions on thefunctional I it can be proven that the actual

Gradient flow equation for I is satisfied

References
e I havebased these notes on the file gradientflows available on the SMSTC web page

of the course

You might also like