247A Notes on Lorentz Spaces Definition 1. for 1 ≤ P < ∞ and F : R D → C We Define (1) Weak(Rd) := Sup Λ∣∣{X

247A Notes on Lorentz Spaces Definition 1. for 1 ≤ P < ∞ and F : R D → C We Define (1) Weak(Rd) := Sup Λ∣∣{X

247A Notes on Lorentz spaces Definition 1. For 1 ≤ p < 1 and f : Rd ! C we define ∗ 1=p (1) kfkLp ( d) := sup λ fx : jf(x)j > λg weak R λ>0 and the weak Lp space p d ∗ L ( ) := f : kfk p d < 1 : weak R Lweak(R ) p −p Equivalently, f 2 Lweak if and only if jfx : jf(x)j > λgj . λ . Warning. The quantity in (1) does not define a norm. This is the reason we append the asterisk to the usual norm notation. To make a side-by-side comparison with the usual Lp norm, we note that ZZ 1=p p−1 kfkLp = pλ dλ dx 0≤λ<jf(x)j Z 1 dλ1=p = jfjfj > λgj pλp 0 λ 1=p 1=p = p λjfjfj > λgj p dλ L ((0;1); λ ) and, with the convention that p1=1 = 1, ∗ 1=1 1=p kfkLp = p λjfjfj > λgj 1 dλ : weak L ((0;1); λ ) This suggests the following definition. Definition 2. For 1 ≤ p < 1 and 1 ≤ q ≤ 1 we define the Lorentz space Lp;q(Rd) as the space of measurable functions f for which ∗ 1=q 1=p (2) kfkLp;q := p λjfjfj > λgj q dλ < 1: L ( λ ) p;p p p;1 p From the discussion above, we see that L = L and L = Lweak. Again ∗ k · kLp;q is not a norm in general. Nevertheless, it is positively homogeneous: for all a 2 C, ∗ −1 1=p ∗ (3) kafkLp;q = λ fjfj > jaj λg Lq (dλ/λ) = jaj · kfkLp;q (strictly the case a = 0 should receive separate treatment). In lieu of the triangle inequality, we have the following: ∗ 1=p kf + gkLp;q = λ fjf + gj > λg Lq (dλ/λ) 1 1 1=p ≤ λ fjfj > 2 λg + fjgj > 2 λg Lq (dλ/λ) 1 1=p 1 1=p ≤ λ fjfj > 2 λg Lq (dλ/λ) + λ fjgj > 2 λg Lq (dλ/λ) by the subadditivity of fractional powers and the triangle inequality in Lq(dλ/λ). Thus ∗ ∗ ∗ (4) kf + gkLp;q ≤ 2kfkLp;q + 2kgkLp;q : Combining (3), (4), and the fact that kfkLp;q = 0 implies f ≡ 0 almost every- ∗ where, we see that k · kLp;q obeys the axioms of a quasi-norm. When p > 1, this quasi-norm is equivalent to an actual norm (see below). When p = 1 and q 6= 1, there cannot be a norm that is equivalent to our quasi-norm. However there is a metric that generates the same topology. In either case, we obtain a complete metric space. ∗ ∗ Notice that (i) if jfj ≥ jgj then kfkLp;q ≥ kgkLp;q and (ii) The quasi-norms are ∗ ∗ rearrangement invariant, which is to say that kfkLp;q = kf ◦φkLp;q for any measure preserving bijection φ : Rd ! Rd. 1 2 p;q P Proposition 3. Given f 2 L , we write f = fm where fm(x) := f(x)χfx:2m≤|f(x)j<2m+1g: Then ∗ kfk p;q ≈p;q kfmk p d L Lx(R ) q `m(Z) p;q1 p;q2 In particular, L ⊆ L whenever q1 ≤ q2. P m Proof. It suffices to consider f of the form f = 2 χEm with disjoint sets Em (cf. m m+1 Em = f2 ≤ jfj < 2 g). Now Z 1 ∗ q q q=p dλ kfkLp;q = p λ jfjfj > λgj 0 λ Z 2m q=p X q X dλ = p λ jEnj m−1 λ m 2 n≥m 1=p q X m X ≈ 2 jEnj : m n≥m To obtain a lower bound, we keep only the summand n = m; for an upper bound, we use the subadditivity of fractional powers. This yields m 1=p ∗ X m 1=p (5) 2 jE j q kfk 2 jE j : m ` . Lp;q . n m q m≤n `m m m 1=p p As k2 χEm kL = 2 jEmj , we have our desired lower bound. To obtain the upper bound, we use the triangle inequality in `q(Z): 1 1 X −k m+k X −k m RHS(5) = 2 k2 χE kLp ≤ 2 k2 χE kLp m+k m q q `m k=0 `m k=0 This completes the proof of the upper bound. Lemma 4. Given 1 ≤ q < 1 and a finite set A ⊂ 2Z, X X q q X Aq ≤ A ≤ 2 max A ≤ 2q Aq A2A where all sums are over A 2 A. More generally, for any subset A of a geometric series and any 0 < q < 1, q X q X A ≈ A where the implicit constants depend on q and the step size of the geometric series. Proposition 5. For 1 < p < 1 and 1 ≤ q ≤ 1, R ∗ ∗ (6) sup j fgj : kgkLp0;q0 ≤ 1 ≈ kfkLp;q : Indeed, LHS(6) defines a norm on Lp;q. Note that by (6), this norm is equivalent to our quasi-norm. Moreover, under this norm, Lp;q is a Banach space and when 0 0 q 6= 1, the dual Banach space is Lp ;q , under the natural pairing. R Remark. When p = 1 (and q 6= 1), the LHS(6) is typically infinite; indeed, E jfj may well be infinite even for some set E of finite measure. In fact, there there cannot be a norm on Lp;q equivalent to our quasi-norm. For example, the impossibility of finding an equivalent norm for L1;1(R) can be deduced by computing N ∗ N X −1 X −1 ∗ jx − nj ≈ N log(N) and jx − nj L1;1 ≈ N: 1;1 n=0 L n=0 3 Proof. Because the quasi-norm is positively homogeneous, we need only verify (6) in the case that f and g have quasi-norm comparable to one. We may also assume P n P m that f = 2 χFn and g = 2 χEm . By the normalization just mentioned, 0 X n 1=pq X m 1=p0 q (7) 2 jFnj ≈ 1 ≈ 2 jEmj n m Combining the above with Lemma 4, we obtain q X X n 1=p X X n 1=pq (8) 2 A ≈ 2 jFnj ≈ 1: A22Z n:jFn|≈A A22Z n:jFn|≈A and similarly for g. Now we compute: Z X n m jfgj dx = 2 2 jFn \ Emj n;m X X n X m ≤ 2 · min(A; B) · 2 A;B22Z n:jFn|∼A m:jEm|∼B 1 1 0 X X n 1=p A 0 B p X m 1=p ≤ 2 A · min p ; · 2 B : B A A;B22Z n:jFn|∼A m:jEm|∼B Notice that this has the structure of a bilinear form: two vectors (indexed over 2Z) with a matrix sitting between them. Moreover, by Schur's test, the matrix is a bounded operator on `q(2Z). Thus, Z X n 1=p X m 1=p0 jfgj dx . 2 A · 2 B ≈ 1 0 `q (A22Z) `q (B22Z) n:jFn|∼A m:jEm|∼B by (8) and the corresponding statement for g. This completes proof of the . part P n p;q of (6). We turn now to the opposite inequality. Given f = 2 χFn 2 L , we choose q−1 1 − 1 q−p X n p p0 X n(q−1) p g = 2 jFnj jFnj χFn = 2 jFnj χFn : n n Then Z q−1 q q 1 1− 1 1 X n p n p0 X n p ∗ fg = 2 jFnj 2 jFnj = 2 jFnj ≈ kfkLp;q ≈ 1: n n ∗ It remains to show that kgk 0 0 1. By Proposition 3, Lp ;q . q0=p0 q0 0 q−p ∗ X q X n(q−1) p kgk p0;q0 ≈ A jFnj where n 2 N(A) , 2 jFnj ≈ A: L A22Z n2N(A) Notice that for each A, the sum in n is over part of a geometric series; indeed, p −n p(q−1) n 2 N(A) () jFnj ≈ A q−p 2 q−p : Thus Lemma 4 applies and yields 0 ∗ q X q0 X q0=p0 X nq q=p kgkLp0;q0 ≈ A jFnj ≈ 2 jFnj ≈ 1: A22Z n2N(A) n This provides the needed bound on g and so completes the proof of (6). The fact that LHS(6) is indeed a norm is a purely abstract statement about vector spaces and (separating) linear functionals. The proof that Lp;q is complete in this norm differs little from the usual Riesz{Fischer argument. p;q 1=p Let ` be a continuous linear functional on L . By definition, j`(χE)j . jEj and so the measure E 7! `(χE) is absolutely continuous with respect to Lebesgue measure and so is represented by some locally L1 function g. This is the Radon{ Nikodym Theorem. By linearity this representation of the functional extends to 4 simple functions. Boundedness when tested against simple functions suffices to 0 0 show that g 2 Lp ;q . When q 6= 1, the simple functions are dense in Lp;q and so our linear functional admits the desired representation. When q = 1 the simple functions are not dense. For example, one cannot approximate jxj−d=p 2 Lp;1(Rd) by simple functions. Indeed, inspired by the Banach limit linear functionals on `1(Z) we can construct a non-trivial linear functional on Lp;1 that vanishes on simple functions. Let L denote the vector space of f 2 Lp;1 such that `(f) := lim jxjd=pf(x) exists. x!0 Notice that L contains the simple functions and that ` vanishes on these.

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    5 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us