01.01.2015 Views

Pointwise decay estimates and asymptotics for semilinear wave ...

Pointwise decay estimates and asymptotics for semilinear wave ...

Pointwise decay estimates and asymptotics for semilinear wave ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Nikodem Szpak ESI, 2010-02-16<br />

<strong>Pointwise</strong> <strong>decay</strong> <strong>estimates</strong> <strong>and</strong> <strong>asymptotics</strong><br />

<strong>for</strong> <strong>semilinear</strong> <strong>wave</strong> equations with small <strong>and</strong> large data<br />

Nikodem Szpak<br />

University Duisburg-Essen<br />

Workshop on Quantitative Studies of Nonlinear Wave Phenomena 2010<br />

Erwin Schrödinger International Institute <strong>for</strong> Mathematical Physics (ESI), Wien<br />

16 Feb 2010<br />

• Plan of the talk:<br />

– [Review:] Optimal pointwise <strong>decay</strong> <strong>estimates</strong> <strong>for</strong> ✷u = F (u) ∼ u p , p ≥ 3(> 1 + √ 2)<br />

Same <strong>decay</strong> <strong>estimates</strong> <strong>for</strong> ✷u = F (∂u) ∼ (∂u) p with p ≥ 3<br />

– Delicate case ✷u = q ab ∂ a u ∂ b u (p = 2)<br />

→ Structure of terms: Huygensian, null, weak null <strong>and</strong> singular (blowup in finite time)<br />

– <strong>Pointwise</strong> <strong>decay</strong> <strong>for</strong> defocusing ✷φ = −|φ| p−1 φ with big initial data by con<strong>for</strong>mal method, 3 ≤ p < 5<br />

[+Roger Bieli]<br />

1


Nikodem Szpak ESI, 2010-02-16<br />

Motivation – Einstein equations<br />

• Einstein eqs. around Minkowski metric g µν = η µν + h µν in harmonic gauge ✷ g x µ = 0<br />

˜✷ g h µν ≡ g αβ ∂ α ∂ β h µν = 1 4 ∂ µh α α ∂ vh β β − 1 2 ∂ µh αβ ∂ ν h αβ + Q µν (∂h, ∂h) + G µν (h)(∂h, ∂h)<br />

• Linearized Einstein eqs. (→ gravitational <strong>wave</strong>s)<br />

✷ η h µν = 0 + terms of order O(h 2 )<br />

Solutions of linearized eqs. ≠ solutions of nonlinear eqs.<br />

• Essential <strong>for</strong> the <strong>asymptotics</strong><br />

– <strong>semilinear</strong> terms ∼ ∂h · ∂h (“self-interaction of <strong>wave</strong>s”)<br />

– quasilinear terms ∼ h · ∂ 2 h (“scattering on non-flat geometry”)<br />

• Other models: <strong>wave</strong> maps, Yang-Mills equations<br />

✷φ = F (φ)(∂φ, ∂φ)<br />

2


Nikodem Szpak ESI, 2010-02-16<br />

Overview of <strong>decay</strong> <strong>estimates</strong><br />

• ✷u = 0<br />

Huygens principle: compactly supported initial data → support on the light-cone, <strong>decay</strong> 1/t<br />

• ✷u = F (u, ∂u) ∼ u p · (∂u) q <strong>for</strong> u ≈ 0 <strong>and</strong> p + q ≥ 3<br />

– in null direction t − r = const: st<strong>and</strong>ard 1/t <strong>decay</strong><br />

– <strong>for</strong> t ≫ r nonlinear scattering: tail u(t, x) ∼ 1/t p+q−1<br />

• ✷u = q ab (u) ∂ a u ∂ b u<br />

– Blowup in finite time [John] <strong>for</strong> ✷u = (∂ t u) 2<br />

– Global existence if null condition [Ch, Kl]: q µν ξ µ ξ ν = 0 on every null vector ξ (η µν ξ µ ξ ν = 0)<br />

<strong>and</strong> <strong>decay</strong> |u(t, x)| 〈t + |x|〉 −1 〈t − |x|〉 −1/2<br />

(→ generalized to quasilinear <strong>wave</strong> equations, systems of <strong>wave</strong> equations...)<br />

– Weak null condition [Li-Ro]: asymptotic eq. (ignore light-cone derivatives <strong>and</strong> cubic terms)<br />

(∂ t + ∂ r )(∂ t − ∂ r )(ru) ∼ rA mn (∂ t − ∂ r ) m u(∂ t − ∂ r ) n u + (...)<br />

Asymptotically (light-cones) not free ↔ slower <strong>decay</strong> than 1/t<br />

Example: ✷u = u∆u [Lindblad, Alinhac] → <strong>decay</strong> only like ε/t 1−cε or log(t)/t<br />

→<br />

Nonlinear <strong>wave</strong>s may behave essentially different than linear! - depending on the structure of nonlinearities<br />

(Gravitational <strong>wave</strong>s ← structure of the Einstein equations)<br />

More structural conditions<br />

3


Nikodem Szpak ESI, 2010-02-16<br />

• ✷u = F with initial data (u, ∂ t u)| t=0 = (f, g)<br />

→ Duhamel representation <strong>for</strong>mula<br />

∫<br />

∫<br />

f(y)<br />

u(t, x) = ∂ t<br />

4πt dµ S(y) +<br />

S(t,x)<br />

Decay <strong>estimates</strong> (in weighted L ∞ spaces)<br />

S(t,x)<br />

g(y)<br />

4πt dµ S(y) + 1<br />

4π<br />

∫<br />

K(t,x)<br />

F (s, y)<br />

t − s dµ K(s, y)<br />

• Weighted-L ∞ estimate [Asakura: (f, g, F ) ∈ C 3 × C 2 × C 2 → C 2 ; NS: (˜L ∞ ) 3 → ˜L ∞ ]<br />

‖u‖˜L∞<br />

≡ ‖τ + τ q−1<br />

−<br />

u(t, x)‖ L ∞ ‖〈x〉q f‖ L ∞ + ‖〈x〉 q+1 ∇f‖ L ∞ + ‖〈x〉 q+1 g‖ L ∞<br />

+ ‖〈x〉 q τ + τ q−1<br />

−<br />

F (t, x)‖ L ∞<br />

with τ ± := 〈t ± |x|〉 <strong>and</strong> 〈x〉 := 1 + |x|. Estimates <strong>for</strong> ‖τ + τ q−1<br />

−<br />

∂α u(t, x)‖ L ∞ follow by analogy<br />

• By iteration ✷u n+1 = F (u n ) converges in ˜L ∞ <strong>for</strong><br />

|F (u)| |u| p , |f(x)| ≤ f 0<br />

〈x〉 m, |∇f(x)| ≤ f 1<br />

〈x〉 m+1, |g(x)| ≤ g 0<br />

〈x〉 m+1<br />

<strong>and</strong> p > 1 + √ 2, m ≥ 2. Then u n → u ∈ ˜L ∞ <strong>and</strong><br />

‖u‖˜L∞<br />

≤ C ⇔ |u(t, x)| ≤<br />

C<br />

〈t + |x|〉〈t − |x|〉 q−1, q := min(p − 1, m) 4


Nikodem Szpak ESI, 2010-02-16<br />

Perturbation theory → optimal <strong>decay</strong> rates<br />

• ✷u = F (u) with data (u, ˙u)| t=0 = (εf, εg) (← natural scale)<br />

u =<br />

∞∑<br />

ε n v n<br />

n=1<br />

→<br />

{ ✷v1 = 0 with data(f, g)<br />

✷v n = F n (v n−1 , ..., v 1 ) with data(0, 0)<br />

where F n (v n , ..., v 1 ) ← collection of powers of ε. [First nontrivial: F p ≠ 0]<br />

• From convergence in ˜L ∞ → pointwise control of the truncation error<br />

n∑<br />

u(t, x) − ε k v k (t, x)<br />

∣ ∣ ε n<br />

〈t + |x|〉〈t − |x|〉 q−1<br />

k=0<br />

• Optimal <strong>decay</strong> rates ← low orders solvable analytically:<br />

– First order: ✷u 0 = 0 → Huygens<br />

– Second (p-th) order: ✷u p = A(u 1 ) p → already generic <strong>decay</strong> rate: u p ∼ u !<br />

– All higher orders: too weak to change the <strong>asymptotics</strong> (← convergence of the perturbation series)<br />

– We can calculate v p (almost) explicitly<br />

5


Nikodem Szpak ESI, 2010-02-16<br />

2nd perturbation → expected <strong>decay</strong> rates<br />

• ✷u n+1 = F (u n , ∂u n ) (assume spherical symmetry <strong>for</strong> simplicity) leads to<br />

✷u 1 = 0 + data (f, g) → u 1 (t, r) =<br />

✷u 2 = F 2 (u 1 , ∂u 1 ) + data (0, 0) → u 2 (t, r) = 1 r<br />

h(t − r)<br />

r<br />

∫ t<br />

0<br />

(h depends on f, g)<br />

∫ r+(t−t<br />

′ )<br />

dt ′ dr ′ r ′ F 2 (u 1 , ∂u 1 )(t ′ , r ′ )<br />

|r−(t−t ′ )|<br />

Tail (t ≫ r) Wave (t ∼ r) Iteration<br />

Structure F 2 (u 1 , ∂u 1 ) = u 2 (t, r) ∼ = u 2 (t, r) ∼ = scheme<br />

u p 1<br />

At −p+1 Bt −1 <strong>for</strong> p ≥ 3<br />

Blog(t)/t <strong>for</strong> p = 2<br />

works<br />

fails → blowup<br />

(∂u 1 ) p At −p+1 Bt −1 <strong>for</strong> p ≥ 3 works<br />

null → ∂ + u 1 · ∂ − u 1 0 Bt −1 works<br />

weak null → ∂ + u 1 · ∂ + u 1 At −3 Bt −1 works<br />

“singular” → ∂ − u 1 · ∂ − u 1 At −1 Blog(t)/t fails→ blowup<br />

6


Nikodem Szpak ESI, 2010-02-16<br />

Modified null derivatives → classification of tails <strong>and</strong> cancellations<br />

• Modified null deriv’s: D ± := ∂ t ± 1 r ∂ r(r·) = 1 r ∂ ±(r·) → quick <strong>decay</strong> <strong>for</strong> D + u (st<strong>and</strong>ard <strong>for</strong> D − u)<br />

→ D ± u(t, r) = 1 ∫<br />

r ′ 1<br />

F dv ∓ → |D + u| <br />

r<br />

r〈t + r〉 p−1, |D 1<br />

−u| <br />

r〈t − r〉 p−1<br />

null<br />

• ✷u = D + u∂u → any <strong>decay</strong> rate! → Huygen’s principle (no scatter) – iteratively kills the rhs’s...<br />

(<br />

✷u = q σλ ∂ σ u∂ λ u = q σλ D σ u − σ ) (<br />

r u D λ u − λ )<br />

r u ≡ Q σλ D σ uD λ u + P λ u<br />

r D λu + R u2<br />

r 2<br />

Order with respect to presence of D + <strong>and</strong> ∂ − (← to see one more cancellation!)<br />

✷u = Q −− ∂ − u∂ − u<br />

+ Q ++ D + uD + u + Q +− D + u∂ − u + P + u<br />

r D +u<br />

+ P − u<br />

r ∂ −u<br />

}<br />

→ blowup in finite time<br />

}<br />

→ Huygens (no scattering)<br />

}<br />

→ Huygens (cancellation)<br />

+ R u2<br />

r 2 }<br />

→ tail ∼ 1/t<br />

3<br />

Extra cancellation from total derivative:<br />

∫<br />

r u r ∂ −u dv − = 0<br />

7


Nikodem Szpak ESI, 2010-02-16<br />

Asymptotics <strong>for</strong> big data from positive energy (defocusing nonlinearity)<br />

• Defocusing <strong>semilinear</strong> <strong>wave</strong> equations<br />

✷φ = −|φ| p−1 φ (✷ ≡ ∂ 2 t − ∆)<br />

give rise to a conserved positive energy<br />

∫ ( 1<br />

E =<br />

R 3 2 |∂ tφ| 2 + 1 2 |∇φ|2 + 1 )<br />

p + 1 |φ|p+1<br />

d 3 x ≥ 0<br />

• Existing results:<br />

– Jörgens’61: global existence of C 2 solutions in the energy subcritical case 1 < p < 5<br />

(Struwe’88: the spherically symmetric critical p = 5 case; Grillakis’90: the general critical case p = 5)<br />

– Global existence is not known <strong>for</strong> p > 5<br />

– Pecher’74: uni<strong>for</strong>m boundedness of solutions <strong>for</strong> 2 < p < 5<br />

– Strauss’68: uni<strong>for</strong>m <strong>decay</strong> 1/t 1−ɛ <strong>and</strong> scattering <strong>for</strong> 3 ≤ p < 5<br />

Bahouri <strong>and</strong> Shatah’98: finite energy solutions <strong>decay</strong> to zero <strong>for</strong> p = 5<br />

Hidano’01: scattering <strong>and</strong> <strong>decay</strong> to zero <strong>for</strong> 2.5 < p ≤ 3<br />

Ginibre <strong>and</strong> Velo’89: scattering in the energy space <strong>for</strong> 1 < p < 5, but in n = 3 only <strong>for</strong> small data<br />

• We go further <strong>and</strong> study the pointwise behavior of solutions <strong>for</strong> 3 ≤ p < 5.<br />

<strong>Pointwise</strong> <strong>estimates</strong> exist only <strong>for</strong> small data [Asakura, Strauss-Tsutaya, N.S.] ← perturbation techniques.<br />

Con<strong>for</strong>mal compactification [Choquet-Bruhat, Christodoulou et al] → adjusted to the nonlinearity.<br />

8


Nikodem Szpak ESI, 2010-02-16<br />

Asymptotics <strong>for</strong> big data – defocusing nonlinearity, spherical symmetry<br />

• For null coordinates u = t + r, v = t − r define<br />

con<strong>for</strong>mal mapping K + ∋ (u, v) → (ũ, ṽ) ∈ K −<br />

ũ := − 1<br />

1<br />

up−2, ṽ := −<br />

v p−2<br />

1<br />

t<br />

I<br />

K +<br />

• The con<strong>for</strong>mal factor Ω := ˜r/r<br />

multiplies the trans<strong>for</strong>med solution ˜φ(ũ, ṽ) := Ω −1 φ(u, v)<br />

which satisfies the trans<strong>for</strong>med <strong>wave</strong> equation in K −<br />

˜✷˜φ = − 1 [<br />

]<br />

ũ − ṽ<br />

p−1<br />

|˜φ| p−1 ˜φ<br />

(p − 2) 2 (−ṽ) 1/(p−2) − (−ũ)<br />

} {{ 1/(p−2) }<br />

α(ũ,ṽ)<br />

• We consider some ”pseudo-energy” flux (Ẽt, Ẽr) in K −<br />

such that div Ẽ ≤ 0<br />

→ boundedness of the flux along null lines ũ = const.<br />

−1<br />

J<br />

1<br />

S<br />

r<br />

• → Uni<strong>for</strong>m boundedness of |˜φ(ũ, ṽ)| ≤ ˜C<br />

H<br />

K − 9<br />

• → <strong>Pointwise</strong> boundedness of |φ(u, v)| ≤ ˜C · Ω(u, v)


Nikodem Szpak ESI, 2010-02-16<br />

Asymptotics <strong>for</strong> big data – defocusing nonlinearity, spherical symmetry<br />

• Consider equations<br />

t<br />

˜✷˜φ = α(ũ, ṽ) |˜φ| p−1 ˜φ<br />

K +<br />

with α ≥ 0 <strong>and</strong> ∂ t α ≤ 0<br />

1<br />

I<br />

• And the corresponding ”pseudo-energy” flux in K −<br />

Ẽ t := r 2 [ 1<br />

2 (∂ t ˜φ) 2 + 1 2 (∂ r ˜φ) 2 + 1<br />

p + 1 α · |˜φ| p+1 ]<br />

,<br />

Ẽ r := −r 2( ∂ t ˜φ · ∂r ˜φ) ,<br />

1<br />

r<br />

• → div Ẽ = 1<br />

p+1 r2 · ∂ t α · |˜φ| p+1 ≤ 0<br />

S<br />

• Boundedness of the null Flux(ũ) := ∫ ũ=const. (Ẽt + Ẽr) dṽ<br />

→ 1<br />

p + 1<br />

∫<br />

ũ=const.<br />

˜r 2 · α · |˜φ| p+1 dṽ ≤ Flux(ũ) ≤ E 0 ,<br />

−1<br />

H<br />

J<br />

• [Pecher’74] Hölder ineq. + integral repr. of the <strong>wave</strong> eq.<br />

→ Uni<strong>for</strong>m boundedness of |˜φ(ũ, ṽ)| ≤ ˜C<br />

K − 10


Nikodem Szpak ESI, 2010-02-16<br />

Asymptotics <strong>for</strong> big data – defocusing nonlinearity, spherical symmetry<br />

• Finally, the inverse trans<strong>for</strong>mation (ũ, ṽ) → (u, v) <strong>and</strong> φ = Ω · ˜φ provides the desired estimate:<br />

the global pointwise bound<br />

|φ(u, v)| ≤ ˜C · Ω = ˜C<br />

∣ 1<br />

r<br />

∣<br />

(t − r) − 1 ∣∣∣<br />

≤<br />

p−2 (t + r) p−2<br />

<strong>for</strong> t ≥ 1 <strong>and</strong> 0 ≤ r ≤ t<br />

• Optimal Numerical simulations done by Bizoń et al suggest (generically) yes.<br />

C<br />

(1 + t + r)(1 + t − r) p−2<br />

• The same estimate as in the small data regime where it is optimal. We have <strong>for</strong> t ≫ r<br />

φ(t, r) =<br />

C<br />

t p−1 + O(t−p )<br />

• The power p − 2 in the con<strong>for</strong>mal mapping cannot be increased<br />

→ stronger pointwise <strong>decay</strong>, but α(t, r) becomes singular<br />

• Spherical symmetry essential <strong>for</strong> the method to yield optimal <strong>decay</strong> <strong>for</strong> p > 3.<br />

On R × R 3 the obvious analogue of the map (u, v) → (ũ, ṽ) is no longer con<strong>for</strong>mal, unless p = 3<br />

(radial <strong>and</strong> angular parts of the <strong>wave</strong> operator trans<strong>for</strong>m diversely → the resulting equation is not <strong>semilinear</strong>).<br />

Apply the trans<strong>for</strong>mation with p = 3 (con<strong>for</strong>mal) to an equation with nonlinearity of power p > 3<br />

→ gives a global 1/[(1 + t + r)(1 + t − r)] <strong>decay</strong>, however not optimal. Improve...<br />

11


Nikodem Szpak ESI, 2010-02-16<br />

Asymptotics <strong>for</strong> big data – defocusing nonlinearity, no symmetry<br />

• Con<strong>for</strong>mal mapping K + ∋ (u, v, θ, ϕ) → (ũ, ṽ, ˜θ, ˜ϕ) ∈ K −<br />

ũ := −1/u, ṽ := −1/v, ˜θ := θ, ˜ϕ := ϕ<br />

• ˜φ(ũ, ṽ) := Ω −1 φ(u, v) with the con<strong>for</strong>mal factor Ω := ˜r/r = (t 2 − |x| 2 ) −1 satisfies (in K − )<br />

˜✷˜φ = −|˜φ| p−1 ˜φ<br />

• Energy flux (E t , E r ) in K − such that div Ẽ = 0 → Flux(ũ) Bounded → |˜φ(ũ, ṽ)| ≤ ˜C<br />

• The inverse trans<strong>for</strong>mation (ũ, ṽ, ˜θ, ˜ϕ) → (u, v, θ, ϕ) <strong>and</strong> φ = Ω · ˜φ give the pointwise bound<br />

|φ(u, v)| ≤ ˜C · Ω ≤<br />

C<br />

(t + r)(t − r)<br />

• Improve it by inserting into the <strong>wave</strong> equation, inverting ✷ (Duhamel <strong>for</strong>mula) <strong>and</strong> estimating<br />

|φ(t, x)| ≤ ✷ −1 |φ| p ✷ −1 1<br />

(uv) p 1<br />

(1 + t + |x|)(1 + t − |x|) p−2.<br />

<strong>for</strong> t ≥ 1 <strong>and</strong> 0 ≤ r ≤ t. (Initial data of compact support do not contribute to the tail)<br />

• Optimal (= spherical symmetry ↔ numerics & small data results)<br />

12


Nikodem Szpak ESI, 2010-02-16<br />

Overview of nonlinear <strong>asymptotics</strong><br />

Equations (initial data ∼ ε) linearized nonlinear<br />

✷u = F (u) ∼ |u p | no tail tail ∼ ε p /t p−1<br />

tail<br />

✷u + V (x)u = F (u) ∼ |u p |, |V (x)| ∼ 1/|x| k u ∼ ε/t k tail ∼ ε p /t p−1<br />

<strong>wave</strong>s<br />

✷u = (∂ − v) 2 , ✷v = 0 u, v ∼ ε/t u ∼ ε 2 log(t)/t<br />

<strong>wave</strong>s<br />

✷u = u∆u u ∼ ε/t u ()<br />

∼ ε 2 log log(t)/t<br />

ε 2 /t 1−ε<br />

asymptotic <strong>for</strong>m of the Einstein eqs.[L.-R., harmonic coord.] tail<br />

✷φ 1 = Q(∂φ, ∂φ) + (∂φ 2 ) 2 + φ 3 ∂ 2 φ 1 φ i = 0 φ 1 t −1<br />

✷φ 2 = Q(∂φ, ∂φ)<br />

<strong>wave</strong>s<br />

✷φ 3 = Q(∂φ, ∂φ) φ i ∼ ε/t φ 1 ε 2 log(t)/t<br />

→ Weyl scalars (ψ 4 ), algebraic invariants...<br />

13


Nikodem Szpak ESI, 2010-02-16<br />

Self-gravitating scalar field in spherical symmetry<br />

Parametrization of the metric<br />

ds 2 = e 2α(t,r) ( −e 2β(t,r) dt 2 + dr 2) + r 2 dΩ 2 ,<br />

Define the mass function m(t, r) = (1 − e −2α )r d−2 . Then<br />

m ′ = κ r d−1 ( ˙φ2 + φ ′ 2 ) ,<br />

ṁ = 2κ r d−1 ˙φφ ′ ,<br />

β ′ =<br />

(d − 2)m<br />

r d−1 ,<br />

✷φ = 2β ¨φ + ˙β ˙φ + β ′ φ ′ .<br />

No clear <strong>asymptotics</strong> <strong>for</strong> φ. After rearrangement of the nonlinear terms on the r.h.s.<br />

✷φ = − 1 r β∂ −φ + 2∂ + β∂ + φ + 2 r ∂ −(rβ∂ − φ) + 2 r β∂2 + (rφ) ∼ = − 1 r β∂ −φ.<br />

The leading <strong>asymptotics</strong> can be calculated exactly<br />

φ(t, r) = C(φ 0, φ 1 )<br />

t 3<br />

( ) 1<br />

+ O .<br />

t 4<br />

14


Nikodem Szpak ESI, 2010-02-16<br />

Conclusions<br />

• Properties of the linearized <strong>wave</strong>s vs. nonlinear <strong>wave</strong>s<br />

– in case of Einstein equations: gravitational <strong>wave</strong>s<br />

– <strong>decay</strong> ← linear <strong>and</strong> nonlinear terms<br />

– propagation of <strong>wave</strong>s ← null/weak null structure<br />

• Solution with a finite weighted-L ∞ norm:<br />

– Global existence<br />

– Nonlinear stability<br />

‖u‖ L ∞<br />

t,x<br />

(R 1+3 ) ≤ C<br />

initial data (εf, εg) → ‖u‖ L ∞<br />

t,x<br />

(R 1+3 ) ≤ εC<br />

– <strong>Pointwise</strong> <strong>decay</strong><br />

|(∂ n C<br />

)u(t, x)| ≤<br />

〈t + |x|〉〈t − |x|〉 q−1<br />

q describes the rate of scattering (even <strong>for</strong> comp. supp. data)<br />

Perturbation theory → optimal <strong>decay</strong> rates.<br />

15


Nikodem Szpak ESI, 2010-02-16<br />

• Null condition [Klainerman]:<br />

– <strong>for</strong> <strong>semilinear</strong> <strong>wave</strong> equations<br />

Null <strong>and</strong> weak null structures<br />

✷u = Q(u)(∂u, ∂u) = q µν (u) ∂ µ u ∂ ν u<br />

<strong>and</strong> q µν ξ µ ξ ν = 0 on every null vector ξ (→ q µν ∼ η µν )<br />

– <strong>for</strong> quasilinear <strong>wave</strong> equations<br />

✷ g(u,∂u) u ≡ g µν (u, ∂u)∂ µ ∂ ν u = 0, g µν = ĝ µνα ∂ α u + ...<br />

<strong>and</strong> ĝ µνα ξ µ ξ ν ξ α = 0 on every null vector ξ<br />

– For systems of <strong>wave</strong> equations<br />

✷u A = q A µν BC<br />

∂ µ u B ∂ ν u C<br />

<strong>and</strong> q BCµν A ξ µ ξ ν = 0 on every null vector ξ<br />

→ Global existence [Christodoulou, Klainerman] <strong>and</strong> <strong>decay</strong> 〈t + |x|〉 −1 〈t − |x|〉 −1/2 ∼ 1/t (lightcone)<br />

• Weak null condition [Lindblad & Rodnianski]: asymptotic eq.<br />

(∂ t + ∂ r )(∂ t − ∂ r )(ru) ∼ rA mn (∂ t − ∂ r ) m u(∂ t − ∂ r ) n u + (...)<br />

– ignored: derivatives tangential to the lightcone <strong>and</strong> cubic terms (Here: null condition ↔ A mn = 0)<br />

Asymptotically (light-cones) not free ↔ slower <strong>decay</strong> (than 1/t)<br />

• → Global existence + <strong>decay</strong> estimate 1/t or less.<br />

↩→ True <strong>decay</strong> rates: on the lightcone (t ∼ |x|) <strong>and</strong> inside (t ≫ |x|)<br />

16


Nikodem Szpak ESI, 2010-02-16<br />

Precise <strong>asymptotics</strong> from scaling<br />

Simplified <strong>for</strong> the talk → pure power nonlinearity & spherical symmetry<br />

✷u = u p , u(0, r) = εf(r), ∂ t u(0, r) = εg(r)<br />

The linear(ized) version of the equation with removed scale factor ε<br />

✷u 0 = 0, u 0 (0, r) = f(r), ∂ t u 0 (0, r) = g(r)<br />

has solution<br />

u 0 (t, r) =<br />

h(t − r) − h(t + r)<br />

, h(x) := − x r<br />

2 f(x) − 1 2<br />

∫ ∞<br />

x<br />

yg(y)dy<br />

Introduce the difference<br />

<strong>and</strong> scale it<br />

In the limit ε → 0<br />

w(t, r) := u(t, r) − εu 0 (t, r) → ✷w = (w + εu 0 ) p<br />

W ε (t, r) := ε −b w(ε −a t, ε −a r), b = p + a(p − 1), a > 0<br />

[ ] p<br />

ε −a(p+1) u 0 (ε −a t, ε −a δ(t − r) − δ(t + r)<br />

r) → Cp , C<br />

r p−k p :=<br />

∫<br />

h p (x)dx.<br />

17


Nikodem Szpak ESI, 2010-02-16<br />

We get a limiting equation<br />

✷W (t, r) = C p<br />

δ(t − r)<br />

r p , lim<br />

ε→0<br />

‖W ε − W ‖ L ∞<br />

1,p−1<br />

= 0.<br />

We can solve exactly<br />

W (t, r) = A p<br />

Θ(t − r)<br />

r<br />

[<br />

]<br />

1<br />

(t − r) − 1<br />

.<br />

p−2 (t + r) p−2<br />

with A p := 2 p−3 C p /(p − 2).<br />

Finally, we reconstruct the <strong>asymptotics</strong> of u <strong>for</strong> small ε. Set a = 1 <strong>for</strong> convenience.<br />

u(t, r) = εu 0 (t, r) + ε 2p−1 W ε (εt, εr)<br />

[<br />

]<br />

= εu 0 (t, r) + ε p Θ(t − r) 1<br />

A p<br />

r (t − r) − 1<br />

p−2 (t + r) p−2<br />

+ o(ε 2p−1 ).<br />

with the error term o(ε 2p−1 ) controlled in the L ∞ 1,p−1<br />

It gives (note that u 0 is localized near t = r)<br />

norm → guarantees at least the same falloff.<br />

u(t, r) ∼ 1/t p−1 <strong>for</strong> t ≫ r, u(t, r) ∼ 1/r <strong>for</strong> t, r ≫ 1, t − r = const.<br />

18


Nikodem Szpak ESI, 2010-02-16<br />

Applications (<strong>semilinear</strong> w/o derivatives)<br />

• → Semilinear <strong>wave</strong> equations without derivatives (on r.h.s.)<br />

✷u = F (u, x)<br />

• Wave map M 1+3 → S 3 , spherical symmetry, corotational<br />

∂ 2 t u − ∂2 r u − 2 r ∂ ru = − sin(2u) ∼<br />

2 4u3<br />

= −<br />

r 2 r2u +<br />

3r + 2 O(u5 )<br />

→ |u(t, r)| ≤<br />

Cr<br />

〈t + |x|〉 2 〈t − |x|〉 3<br />

• Yang-Mills M 1+3 → SU(2), spherical symmetry<br />

∂ 2 t u − ∂2 r u − 2 r ∂ ru + 2 r 2u = −u3 − 3u2<br />

r<br />

→ |u(t, r)| ≤<br />

Cr<br />

〈t + |x|〉 2 〈t − |x|〉 2 19


Nikodem Szpak ESI, 2010-02-16<br />

Problem with derivatives → spherical symmetry<br />

• In ✷u = q µν ∂ µ u∂ ν u one needs to control 1st derivatives: ✷∂u = ∂F → ‖∂u‖˜L∞ ‖∂F ‖˜L∞<br />

∂u(t, x) = 1<br />

4π<br />

∫ t<br />

0<br />

∫<br />

ds<br />

S(x,t−s)<br />

∂F (s, y)<br />

t − s<br />

d2 y<br />

Problem: ∂F contains ∂ 2 u → no control (loss) of derivatives (in the iteration)<br />

• In spherical symmetry – no problem ← gain of regularity<br />

→<br />

u(t, r) = 1 2r<br />

∫<br />

∂ ± u(t, r) = 1 r<br />

null<br />

∫ t<br />

0<br />

ds<br />

∫ r+(t−s)<br />

|r−(t−s)|<br />

r ′ F (s, r ′ )dr ′ =: 1 r<br />

∫ ∫<br />

r ′ F dv + dv −<br />

r ′ F dv ∓ ∓ 1 ∫ ∫<br />

r ′ F dv<br />

r 2 + dv − → ‖∂ ± u‖˜L∞<br />

‖F ‖˜L∞<br />

• Modified null deriv’s: D ± := ∂ t ± 1 r ∂ r(r·) = 1 r ∂ ±(r·) → quick <strong>decay</strong> <strong>for</strong> D + u (st<strong>and</strong>ard <strong>for</strong> D − u)<br />

→<br />

D ± u(t, r) = 1 r<br />

∫<br />

null<br />

r ′ F dv ∓ → |D + u| <br />

1<br />

r〈t + r〉 p−1,<br />

|D −u| <br />

1<br />

r〈t − r〉 p−1 20

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!