+ All Categories
Home > Documents > Gradient bounds for nonlinear strictly elliptic equations ...acesar/slides MB60/ley.pdf · strictly...

Gradient bounds for nonlinear strictly elliptic equations ...acesar/slides MB60/ley.pdf · strictly...

Date post: 02-Mar-2018
Category:
Upload: phungminh
View: 216 times
Download: 2 times
Share this document with a friend
13
Gradient bounds for nonlinear strictly elliptic equations with coercive Hamiltonians Olivier Ley Feb 2016 Gradient bounds for nonlinear strictly elliptic equations with coercive Hamiltonians Olivier Ley IRMAR, INSA de Rennes, France http://ley.perso.math.cnrs.fr Collaboration with Vinh Duc Nguyen (Cardiff) Nonlinear PDEs : Optimal Control, Asymptotic Problems and Mean Field Games On the occasion of Martino Bardi’s 60th birthday Padova, February 25-26, 2016 1/13
Transcript

Gradientbounds fornonlinear

strictly ellipticequations

with coerciveHamiltonians

Olivier Ley

Feb 2016

Gradient bounds for nonlinear strictlyelliptic equations with coerciveHamiltonians

Olivier Ley

IRMAR, INSA de Rennes, Francehttp://ley.perso.math.cnrs.fr

Collaboration with Vinh Duc Nguyen (Cardiff)

Nonlinear PDEs : Optimal Control, Asymptotic Problemsand Mean Field GamesOn the occasion of Martino Bardi’s 60th birthdayPadova, February 25-26, 2016

1/13

Gradientbounds fornonlinear

strictly ellipticequations

with coerciveHamiltonians

Olivier Ley

Feb 2016

Statement of the problem

(HJ) εφε − Tr(A(x)D2φε) + H(x ,Dφε) = 0

• x ∈ TN periodic setting

• ε > 0

• A(x) ≥ νI , ν > 0 strict ellipticity

• A(x) = σ(x)σ(x)T , σ ∈W 1,∞(TN)

• Assume there exists a continuous viscosity solution φε

ê Goal : To obtain gradient bounds |Dφε|∞ ≤ K

with K independent of ε

2/13

Gradientbounds fornonlinear

strictly ellipticequations

with coerciveHamiltonians

Olivier Ley

Feb 2016

Motivation for ε-independent bounds

It allows to solve the associated ergodic problem :

εφε → −c (ergodic constant), φε − φε(0)→ v as ε→ 0

and (c, v) ∈ R×W 1,∞(TN) is solution to

(HJerg) − Tr(A(x)D2v) + H(x ,Dv) = c .

[Lions-Papanicolau-Varadhan 86, Evans 89, Arisawa-Lions 98,Alvarez-Bardi 10, etc.]

Remark : |φε|∞ ∼ 1ε in general

3/13

Gradientbounds fornonlinear

strictly ellipticequations

with coerciveHamiltonians

Olivier Ley

Feb 2016

Motivation for Lipschitz bounds

ε-independent bounds for (HJ) leads in general totime-independent gradient bounds |Du(·, t)|∞ for the solutionof (HJevol)

∂tu − Tr(A(x)D2u) + H(x ,Du) = 0 (x , t) ∈ TN×(0,+∞)

u(x , 0) = u0(x) x ∈ TN

allowing a linearization procedure which permits to use theStrong Maximum Principle for viscosity solutions [Bardi-DaLio 99] to prove the large time behavior [Barles-Souganidis 01]

u(x , t) + ct → v(x) uniformly as t → +∞,

where (c , v) are solution to (HJerg).

4/13

Gradientbounds fornonlinear

strictly ellipticequations

with coerciveHamiltonians

Olivier Ley

Feb 2016

Existing results for gradient bounds for (HJ)

• Elliptic regularity for classical solutions[Gilbarg-Trudinger] : |H(x , p)| ≤ C (1 + |p|2) (subquadratic)

• Ishii-Lions’ method :

[Ishii-Lions 90]

|H(x , p)− H(y , p)| ≤ C + ω(|x − y |)|x − y |τ |p|τ+2 τ ∈ [0, 1]

C , τ = 0, ω(|x − y |) = |x − y | ⇒ |DxH| ≤ |p|2

typical case : a(x)|p|2 (subquadratic)

[Barles 91]

|H(x , p)− H(y , p)| ≤ C + C |x − y ||p|3 + C |p|2little more than quadratic but restriction on the growth of H in p

These bounds depend on |φε|∞

5/13

Gradientbounds fornonlinear

strictly ellipticequations

with coerciveHamiltonians

Olivier Ley

Feb 2016

Gradient bounds for (HJ) with arbitrary growth

Bernstein method [Bernstein 1910]

• classical solutions [Gilbarg-Trudinger]

• viscosity solutions, weak Bernstein method : [Barles 91]Need of structural assumptions of “convexity type”[Barles-Souganidis 01] (A(x) ≡ I )

∃L > 0 : ∀|p| ≥ L, x ∈ TN , L(Hp · p − H − |H(·, 0)|∞) ≥ |Hx |

typical case : H(x , p) = a(x)|p|1+β + f (x),

β>0, a,f ∈W 1,∞(TN), a(x)>0

6/13

Gradientbounds fornonlinear

strictly ellipticequations

with coerciveHamiltonians

Olivier Ley

Feb 2016

Remarks

• 1st order equations : εφε + H(x ,Dφε) = 0 with coerciveHamiltonian (of arbitrary growth) : immediate gradients boundsindependent of ε for subsolutions

• Holder bounds for possibly degenerate (HJ)[Capuzzo Dolcetta-Leoni-Porretta 10]Very general result for subsolutions :

if H(x , p) ≥ 1C |p|

k − C , k > 2 then φε ∈ C 0, k−2k−1 (TN)

with a Holder bound independent of ε

7/13

Gradientbounds fornonlinear

strictly ellipticequations

with coerciveHamiltonians

Olivier Ley

Feb 2016

Our slight extensions : “subquadratic case”

Theorem 1. |Dφε|∞ ≤ K for (HJ) when

(H1) ∃L > 1 : ∀x , y ∈ TN , if |p| = L then

H(x , p) ≥ |p|(H(y , p

|p|) + |H(·, 0)|∞ + N|x − y ||σx |2∞)

(H2) ∃α,C > 0 : ∀x , y ∈ TN ,|H(x , p)− H(y , p)| ≤ C |x − y |α|p|α+2 + C (1 + |p|2)

Comments :(H1) holds when H(x ,p)

|p| → +∞ as |p| → +∞ (superlinearity)

(H2) reduces to [Barles 91] when α = 1

Example : εφε−Tr(A(x)D2φε) + |Σ(x)Dφε|m + K (x ,Dφε) = 0

Σ∈C 0,γ(TN), m≤2+γ,K (x , p)

|p|→

|p|→+∞+∞, |K (x , p)|≤C (1+|p|2)

8/13

Gradientbounds fornonlinear

strictly ellipticequations

with coerciveHamiltonians

Olivier Ley

Feb 2016

Our slight extensions : “arbitrary growth”

Theorem 2. |Dφε|∞ ≤ K for (HJ) when

(H3) H(x , p) ≥ 1C |p|

k − C , k > 2

(H4) ∃α,β>0,B(p) such that β<(k−1)α+k andB(p)

|p|k→

|p|→+∞0

|H(x , p)− H(y , p)| ≤ C |x − y |α|p|β + B(p)

Comments :By (H3), φε ∈ C 0, k−2

k−1 (TN) [Capuzzo Dolcetta-Leoni-Porretta 10]

(H4) : no convexity-type assumptions, k , α can be big

Example : εφε − Tr(A(x)D2φε) + |Σ(x)Dφε|m + K (x ,Dφε) = 0

0 < Σ∈C 0,γ(TN), k ≤m≤ (k − 1)α + k ,K(x , p)

|p|k→

|p|→+∞0

(nonconvex) εφε − Tr(A(x)D2φε) + a(x)G (Dφε) + K (x ,Dφε) = 0

0 < a∈C 0,γ(TN), |p|k

C ≤ G (p) ≤ C |p|β , K(x , p)

|p|k→

|p|→+∞0

9/13

Gradientbounds fornonlinear

strictly ellipticequations

with coerciveHamiltonians

Olivier Ley

Feb 2016

Ideas of proof (1) : Oscillation bound

Lemma 1. [L-Nguyen 15] Under (H1) (“superlinearity”),osc(φε) := maxTNφε −minTNφε ≤ O independent of ε.

Proof very simple, OK for degenerate and nonlocal equations.Will allow some “localization arguments” in the proof of regularity

Proof. Choose L >> 1 so that (H1) holds for |p| = L. Consider

M = maxx,y∈TN

φε(x)− Lφε(y) + (L− 1)minφε − L|x − y |

If M ≤ 0 we are done ; otherwise M > 0 and at maximum, x 6= y .Writing the viscosity inequalities with p = L x−y

|x−y | leads to

ε(φε(x)− Lφε(y))︸ ︷︷ ︸|εφε| ≤ |H(·, 0)|∞maximum principle

−Tr(A(x)X−A(y)Y )︸ ︷︷ ︸≤N|σx |2∞L|x−y |

+ H(x , p)− LH(y ,p

L)︸ ︷︷ ︸

|p| = L >> 1 sofrom (H1) bigger than

L(|H(·, 0)|∞ + N|σx |2∞|x − y |)

≤ 0

Contradiction 210/13

Gradientbounds fornonlinear

strictly ellipticequations

with coerciveHamiltonians

Olivier Ley

Feb 2016

Ideas of proof (2) : Ishii-Lions’ method

Lemma 2. Let ψ a concave smooth function with ψ(0) = 0.If maxx ,y∈TN{φε(x)− φε(y)− ψ(|x − y |)} > 0

is achieved at x , y with x 6= y and q = x−y|x−y | , then

−4νψ′′(|x − y |)︸ ︷︷ ︸>0 if ψ strict.concave

−C |x − y |ψ′(|x − y |)+H(x , ψ′(|x − y |)q)− H(y , ψ′(|x − y |)q) < 0

Idea : use ψ strictly concave s.t. −ψ′′(r)− Crψ′(r) >> 1

1) Holder bounds : ψ(r) = Krγ , γ ∈ (0, 1)2) Lipschitz bounds : ψ(r) = r − Kr1+γ , γ ∈ (0, 1)

Need to have r = |x − y | ≤ r0 small.For instance, in 2), r0 ≤ ((1 + γ)K )−γ , K very big

ê crucial use of Lemma 1 (oscillation bound)

11/13

Gradientbounds fornonlinear

strictly ellipticequations

with coerciveHamiltonians

Olivier Ley

Feb 2016

Ideas of proof (3) : Theorem 1

Step 1. (Holder estimate) |φε|C 0,γ ≤ K for some γ ∈ (0, 1).M = max{φε(x)− φε(y)− ψ(|x − y |}, ψ(r) = Lrγ .(H1)+Lemma 1 ê osc(φε) ≤ O independent of ε.Choosing K , r0 s.t. Krγ0 = O + 1 we have r = |x − y | < r0.If M > 0, then |x − y | 6= 0 and (H2)+Lemma 2 imply

−4νψ′′(r)− Crψ′(r)− Crαψ′(r)α+2 − Cψ′(r)2 − C < 0

...Contradiction for γ small enough.

Step 2. (Lipschitz estimate)M =max{φε(x)−φε(y)−ψ(|x − y |)}, ψ(r)=A1(A2r−(A2r)1+γ).We earn something, “Revenge of the ellipticity” : nonlinearity can bestronger than 2nd order terms but Holder estimate weaken thenonlinear terms :

rψ′(r) ≤︸︷︷︸concavity

ψ(r) <︸︷︷︸max.point

φε(x)− φε(y) ≤︸︷︷︸Step 1

Krγ

ê Better estimate of ψ′ : ψ′(r) ≤ Krγ−1

12/13

Gradientbounds fornonlinear

strictly ellipticequations

with coerciveHamiltonians

Olivier Ley

Feb 2016

Ideas of proof (4) : Theorem 2

Step 1. |φε|C0,γ ≤ K for γ = k−2k−1

[Capuzzo Dolcetta-Leoni-Porretta 10]

Give a first improvement of the first derivative.

Step 2. |φε|C0,γ ≤ K for every γ ∈ (0, 1).Use ψ(r) = Krγ with the above improvement.

Step 3. Improvement to Lipschitz continuity as in Step 2 ofTheorem 1.

13/13


Recommended