7

In Adams & Fournier's book "Sobolev Spaces" page 255, there are assertions:

$$ W^{s,p} = F^{s}_{p,2} \\ B^s_{p,p} = F^s_{p,p}$$

Note: Here $s>0$, $1\leq p<\infty$, and the fractional order Sobolev spaces are defined as complex interpolation spaces $W^{s,p} = [L^p , W^{m,p} ] _{s/m}$ where $m$ is the smallest integer greater than $s$.

AFAIK, the above definition is equivalent to the following definition (at least for $0<s<1$), employing Gagliardo seminorms: $f \in W^{s,p}$ iff $f \in W^{\left \lfloor{s}\right \rfloor ,p}$ and $[D^\alpha f] := (\int \frac{|D^{\alpha}f(x)-D^{\alpha}f(y)|^p } {|x-y|^{(s-\left \lfloor{s}\right \rfloor)p + n}} dxdy)^{1/p} < \infty$ for all $|\alpha|=\left \lfloor{s}\right \rfloor$.

The Triebel-Lizorkin spaces and the Besov spaces are defined by Paley-Littlewood decompositions.

In Triebel's book "Interpolation theory, function spaces, differential operators(1978)", page 169, the fractional order Sobolev spaces are defined as $W^{s,p} = B^s_{p,p}$ for $s>0$ non-integer.

Therefore, the two monographs are coherent only when $p=2$. Could someone explain about this situation?

My background: I have little experience and knowledge of function spaces. To get some knowledge, I am starting to have a brief overview of the famous monographs. However, the problem described above makes me very confused.

Any bits of help are welcome. Thank you!!

Calvin Khor
  • 36,697
  • 6
  • 48
  • 105
J. Doe
  • 1,155

1 Answers1

6

The notations in the two books are different.

With the definition using the Gagliardo seminorms, these are the real interpolation spaces (which are now more usually denoted by $W^{s,p}$ I think) when $s$ is not an integer $$ W^{s,p} = B^s_{p,p} = F^s_{p,p} = (L^p,W^{n,p})_{s/n,p} $$ while the complex interpolation gives you $$ H^{s,p} = F^s_{p,2} = [L^p,W^{n,p}]_{s/n} $$ with the seminorm $\left|f\right|_{H^{s,p}} = \|\Delta^{s/2}f\|_{L^p}$. The notation is quite misleading since when $n$ is an integer, then the classical $W^{n,p}$ is $H^{n,p}$ and not $B^n_{p,p}$.

By the general embeddings theorems between Besov and Triebel-Lizorkin spaces, you get $$ W^{s,p} ⊂ H^{s,p} $$ when $p≤ 2$ and the reverse inclusion when $p\geq 2$. Of course, when $p=2$, then $H^{s,2} = W^{s,2} = H^s$.

LL 3.14
  • 14,421
  • 1
    Thanks for your elaborate answer. I think I got the point: I should take care of the differences between Sobolev-Slodobeckii space (Gagliardo seminorm) and the Bessel potential space (using the Fourier approach), which amounts to real interpolation spaces versus complex interpolation spaces of the classical Sobolev spaces.
    1. Did I correctly rephrase your answer?
    2. Could I ask for some references for the equivalences of spaces quoted in your answer?

    Thank you!

    – J. Doe Jun 07 '20 at 07:55
  • 1
    The introduction of the book of Triebel Theory of function spaces II is good to explain the relation between the spaces. And you will also find the complex interpolation of classical Sobolev spaces there. – LL 3.14 Jun 07 '20 at 16:39
  • 1
    You can find the real interpolation of Sobolev spaces in the book of Lunardi, Interpolation Theory, section 1.1. – LL 3.14 Jun 07 '20 at 16:42
  • I think your complex interpolation is incorrect. It should be $[L^p,W^{n,p}]_{s/n}$. There is only one index rather than two. – Liding Yao Mar 18 '24 at 00:54
  • Yes, of course, corrected, thanks for the comment. – LL 3.14 Mar 18 '24 at 16:00