5 Equivalent Versions of Fourier Restriction

Searching for (p,q) for which

∥f^∥Lq(Sn−1)≤Cp,q,n∥f∥Lp(ℝn).

Reminder: RSn−1(p→q) means

∥f^∥Lq(Sn−1)≤Cp,q,n∥f∥Lq(ℝn)∀⁡f∈S(ℝn).

Conjecture (Restriction Conjecture). RSn−1(p→q) if and only if n+1−n−1q≤n+1p and p<2nn+1.

First proved for S1⊆ℝ2 by Fefferman (1970) and Zygmund (1974).

Special things happen in ℝ2, classical harmonic analysis techniques apply.

Same conjecture for RPn−1(p→q), where

Pn−1={(ξ,|ξ|2)∈ℝn:|ξ|<1}.

∥f^∥Lq(Pn−1)q=∫|ξ|<1|f^(ξ,|ξ|2)|qdξ.

PIC

For n≥3, open and active!

Restriction theory can be used to deduce continuum incidence geometry estimates.

Surprisingly, we can go the other way too (very recent progress, whereas the above direction has been well-known since at least the 90s).

Equivalent formulations of RPn−1(p→q)

Dual version is called “Fourier extension”:

∥f^∥Lq(Pn−1)=sup⁡g∈Lq′(Pn−1)∥g∥Lq′(Pn−1)=1|∫|ξ|<1ξ∈ℝn−1f^(ξ,|ξ|2)g(ξ)dξ|.

(1q+1q′=1). The integral equals:

∫|ξ|<1∫ℝne−2πix⋅(ξ,|ξ|2)f(x)dxg(ξ)dξ=∫ℝnf(x)∫|ξ|<1e−2πix⋅(ξ,|ξ|2)g(ξ)dξ⏟Eg(x)dx∥Eg∥Lp′(ℝn)≤Cp,q,n∥g∥Lq′(Pn−1)∀⁡S(Pn−1)

Call the last inequality RPn−1∗(q′→p′).

Local, dual version: allows us to work with functions, F.T.

For any R≥1, any BR⊆ℝn, we have

(∫ℝn|f(x)|p′)1p′≲R1q(∫|f^(ξ)|q′dξ)1q′

for all f∈S(ℝn) with supp⁡f^⊆NR−1(Pn−1).

Call this RPn−1∗,loc(q′→p′).

RPn−1∗(q′→p′)⟹Rpn−1∗,loc(q′→p′). First thing bounds Eg, while secound thing bounds f when we have supp⁡f^⊆NR−1(Pn−1).

Let f∈S(ℝn), supp⁡f^⊆NR−1(Pn−1). Try to express f in trems of ext. op.

f(x)=Fourier inversion∫ℝne2πix⋅ξf^(ξ)dξ=∫ℝn−1∫ℝe2πix⋅(ξ′,ξn)f^(ξ′,ξn)dξndξ′=∫ℝe2πxnξn∫ℝn−1e2πix⋅(ξ′,|ξ′|2)f^(ξ′,|ξ′|2+ξn)⏟=:gξn(ξ′)dξ′dξnξn↦ξn+|ξ′|2=∫[−R−1,R−1]e2πxnξn∫|ξ′|<1e2πix⋅(ξ′,|ξ′|2)f^(ξ′,|ξ′|2+ξn)⏟=:gξn(ξ′)dξ′⏟Egξn(x)dξn∫ℝn|f(x)|p′dx=∫ℝn|∫IR−1e2πixnξnEgξn(x)dξn|p′dxR∗ bounds ∫|Eg|p′≤Hölders’s∫ℝn|IR−1|p′−1∫IR−1|Egξn(x)|p′dξndx≲(R−1)p′−1∫IR−1(∫|ξ′|<1|gξn(ξ′)|q′dξ′)p′q′dξn⏟(∗)using RPn−1∗(q′→p′) Goal is to bound this last expression by
R−p′q∫IR−1∫|ξ′|<1|gξn(ξ′)|q′dξ′dξn⏟∥f^∥Lq′(NR−1(Pn−1))

Lucy case: p′q′<1, i.e. 1<q′p′. Then

(∗)≤R−p′+1∫IR−1
∫Ah≤Hölder’s|A|1−1s(∫Ahs)1s,

s≥1 (actually ≈ since h approximately constant on A)

RPn−1∗(q′→p′)⟹RPn−1∗,loc(q′→p′) continued.

Spatial: x,f(x),Eg(x)

∫ℝn|f(x)|p′dx
∫ℝn|Eg(x)|p′dx

PIC

Frequency ξ:

PIC

g(ξ′,|ξ′|2)=g(ξ′), supp⁡f^⊆NR−1(Pn−1),

∫NR−1(Pn−1)|f^(ξ)|q′dξ
∫|ξ′|<1|g(ξ′)|q′dξ′.
∫BR|f(x)|p′dx≤⋯≤(∗)R−(p′−1)∫IR−1(∫|ξ′|<1|gξn(ξ′)|q′dξ′)p′q′dξn.

Aiming for

(∗)≤R−p′q(∫IR−1∫|ξ′|<1|gξn(ξ′)|q′dξ′)p′q′.

Lucky case: p′q′≤1.

(∗)≤Hölder’s inequality in ξnR−(p′−1)|IR−1|1−p′q′(∫IR−1∫|ξ′|<1|gξn(ξ′)|q′dξ′dξn)p′q′

R−(p′−1)(R−1)1−p′q′=R−p′q (1=1q+1q′)

Unlucky case: p′q′>1. Hölder’s inequality goes in the wrong direction:

∫Ah≤|A|1−1s(∫Ahs)1s

for all s≥1, h≥0.

This is equality if h is constant on A.

PAUSE THIS.

Useful Harmonic Analysis tool

The locally constant property.

Convolution: Let f,g∈S(ℝn). Define f∗g∈S(ℝn) by

f∗g(x)=∫ℝnf(x−y)g(y)dy=∫ℝnf(Y)g(x−y)dy.

See Young’s convolution:

∥f∗g∥Lr≤∥f∥p∥g∥q

when 1+1r=1p+1q.

Example. g=χB (B is the unit ball in ℝn),

f∗χB(x)=∫y∈Bf(x−y)dy

RHS is “average value of f on B(x)”.

“f∗χB is approximately constant on balls of radius ∼1”.

Support property: supp⁡f=A, supp⁡g=B, supp⁡f∗g⊆A+B={a+b:a∈A,b∈B}.

Convolution and Fourier Transform: fg^=f^∗g^.

Locally constant property: Let f∈S(ℝn), supp⁡f^⊆B1(0). Then for any unit ball B′⊆ℝn and any x∈B′,

∥f∥L∞(B′)≲m∫ℝn|f|(x−y)1(1+|y|2)mdy.

1(1+|y|2)m is ∼1 on |y|≲1.

PIC

Digesting f≥0.

For any unit interval I′⊆ℝn, x∈I′,

∥f∥L∞(I′)≤∫(−12,12)f(x−y)dy.

Suppose this:

PIC

LHS has to be constant.

∥f∥L∞(I′)≲∫(−12,12)f(x−y)dy.

PIC

∥f∥L∞(I′)≲(∗)∫(−12,12)f(x−y)dy

Lemma (Locally constant property). Assuming that:

  • f∈S(ℝn)

  • supp⁡f^⊆B1(0)

Then for any unit ball B1⊆ℝn and any x∈B′,
∥f∥L∞(B′)≲m∫ℝn|f|(x−y)1(1+|y|2)mdy.

The proof of this fact is more important than the statement – we will be using the strategy in future.

Proof of locally constant property. Let f∈S(ℝn), supp⁡f^⊆B1(0). Let φ∈Cc∞(ℝn) such that φ≡1 on B1(0), supp⁡φ⊆B2(0).

By Fourier inversion:

f=ˇ(f^)=ˇ(f^φ)=ˇ(f^)∗ˇφ=f∗ˇφ.

Let x0,x∈B1 (unit ball in ℝn).

|f(x0)|=|∫f(x0−y)ˇφ(y)dy|≤∫|f|(x0−y)|ˇφ(y)|dy=∫|f|(x−y)|ˇφ|(y−(y−x0))dy≲m∫|f|(x−y)1(1+|y−(x−x0)⏟|x−x0|<1|2)mdy≲m∫|f|(x−y)1(1+|y|2)m□
Returning to RPn−1∗(q′→p′)⟹RPn−1∗,loc(q′→p′)
∫BR|f(x)|p′dx≲∫BR|fφBR|p′dx

where φBR∈S(ℝn) satisfies φBR∼1 on BR, supp⁡φBR^⊆BR−1(0).

fφBR^=f^⏟R−1 neighbourhood of Pn−1∗φBR^⏟R−1 ball(ξ)

supported in 2R−1-neighbourhood of Pn−1.

Repeat steps of proof:

∫BR|f(x)|p′dx≲R−(p′−1)∫IR−1(∫|ξ′|<1|hξn(ξ′)|q′dξ′)p′q′dξn

PIC

h(ξn)=f^∗φBR^(ξ′,|ξ′|2+ξn)