ProofComplete

Distributions and Sobolev Spaces - Key Proof

We present a proof of the Sobolev Embedding Theorem in one dimension, illustrating the key ideas that extend to higher dimensions.

TheoremSobolev Embedding in One Dimension

For u∈W1,p(a,b)u \in W^{1,p}(a,b) with 1≀pβ‰€βˆž1 \leq p \leq \infty:

  1. If p=1p = 1, then uu is absolutely continuous and βˆ₯uβˆ₯βˆžβ‰€Cβˆ₯uβˆ₯W1,1\|u\|_\infty \leq C\|u\|_{W^{1,1}}
  2. If p>1p > 1, then u∈C0,1βˆ’1/p([a,b])u \in C^{0,1-1/p}([a,b]) (HΓΆlder continuous)
Proof

Case p=1p = 1: For u∈W1,1(a,b)u \in W^{1,1}(a,b), the weak derivative uβ€²u' is in L1L^1. By the fundamental theorem of calculus for Sobolev functions, u(x)=u(c)+∫cxuβ€²(t) dtu(x) = u(c) + \int_c^x u'(t) \, dt for some c∈(a,b)c \in (a,b) and almost every x∈(a,b)x \in (a,b).

For any x,y∈[a,b]x, y \in [a,b]: ∣u(x)βˆ’u(y)βˆ£β‰€βˆ«yx∣uβ€²(t)βˆ£β€‰dt≀βˆ₯uβ€²βˆ₯L1|u(x) - u(y)| \leq \int_y^x |u'(t)| \, dt \leq \|u'\|_{L^1}

Taking yy such that ∣u(y)βˆ£β‰€βˆ₯uβˆ₯L1/(bβˆ’a)|u(y)| \leq \|u\|_{L^1}/(b-a) (mean value), we get ∣u(x)βˆ£β‰€βˆ£u(y)∣+∣u(x)βˆ’u(y)βˆ£β‰€βˆ₯uβˆ₯L1bβˆ’a+βˆ₯uβ€²βˆ₯L1≀Cβˆ₯uβˆ₯W1,1|u(x)| \leq |u(y)| + |u(x) - u(y)| \leq \frac{\|u\|_{L^1}}{b-a} + \|u'\|_{L^1} \leq C\|u\|_{W^{1,1}}

Case p>1p > 1: For x,y∈[a,b]x, y \in [a,b] with x<yx < y, HΓΆlder's inequality gives: ∣u(x)βˆ’u(y)∣=∣∫xyuβ€²(t) dtβˆ£β‰€βˆ₯uβ€²βˆ₯Lp∣yβˆ’x∣1/pβ€²|u(x) - u(y)| = \left|\int_x^y u'(t) \, dt\right| \leq \|u'\|_{L^p} |y - x|^{1/p'}

where 1/p+1/pβ€²=11/p + 1/p' = 1. Thus uu is HΓΆlder continuous with exponent Ξ±=1βˆ’1/p=1/pβ€²\alpha = 1 - 1/p = 1/p'.

For the embedding constant: βˆ₯uβˆ₯C0,Ξ±=sup⁑x∣u(x)∣+sup⁑xβ‰ y∣u(x)βˆ’u(y)∣∣xβˆ’yβˆ£Ξ±β‰€Cβˆ₯uβˆ₯W1,p\|u\|_{C^{0,\alpha}} = \sup_x |u(x)| + \sup_{x \neq y} \frac{|u(x) - u(y)|}{|x - y|^\alpha} \leq C\|u\|_{W^{1,p}}

β– 
RemarkHigher Dimensions

In nn dimensions, the proof is more involved and uses:

  1. Gagliardo-Nirenberg-Sobolev Inequality: For u∈Cc1(Rn)u \in C_c^1(\mathbb{R}^n) and 1≀p<n1 \leq p < n: βˆ₯uβˆ₯Lpβˆ—β‰€Cnβˆ₯βˆ‡uβˆ₯Lp\|u\|_{L^{p^*}} \leq C_n \|\nabla u\|_{L^p} where pβˆ—=npnβˆ’pp^* = \frac{np}{n-p}

  2. Approximation: Extend from Cc1C_c^1 to W1,pW^{1,p} by density

  3. Iteration: For higher order Sobolev spaces, iterate the result

ProofSketch for $W^{1,p}(\mathbb{R}^n) \hookrightarrow L^{p^*}(\mathbb{R}^n)$

Step 1: For u∈Cc1(Rn)u \in C_c^1(\mathbb{R}^n), write ∣u(x)∣pβˆ—=pβˆ—βˆ«βˆ’βˆžx1βˆ‚βˆ‚t∣u(t,x2,…,xn)∣pβˆ—β€‰dt|u(x)|^{p^*} = p^* \int_{-\infty}^{x_1} \frac{\partial}{\partial t} |u(t, x_2, \ldots, x_n)|^{p^*} \, dt

Step 2: Apply HΓΆlder's inequality with careful exponents: ∣u(x)∣pβˆ—β‰€Cβˆ«βˆ’βˆžx1βˆ£βˆ‡u∣∣u∣pβˆ—βˆ’1 dt|u(x)|^{p^*} \leq C \int_{-\infty}^{x_1} |\nabla u| |u|^{p^*-1} \, dt

Step 3: Integrate over Rn\mathbb{R}^n and use HΓΆlder again to separate uu and βˆ‡u\nabla u terms

Step 4: Solve the resulting inequality to get βˆ₯uβˆ₯Lpβˆ—pβˆ—β‰€Cβˆ₯βˆ‡uβˆ₯Lppβˆ₯uβˆ₯Lpβˆ—pβˆ—βˆ’p\|u\|_{L^{p^*}}^{p^*} \leq C \|\nabla u\|_{L^p}^p \|u\|_{L^{p^*}}^{p^*-p}

Step 5: Rearrange to obtain βˆ₯uβˆ₯Lpβˆ—β‰€Cβˆ₯βˆ‡uβˆ₯Lp\|u\|_{L^{p^*}} \leq C \|\nabla u\|_{L^p}

Step 6: Extend to W1,pW^{1,p} by density of Cc1C_c^1

β– 
Remark

The Sobolev embedding theorem is sharp: the exponent pβˆ—=npnβˆ’pp^* = \frac{np}{n-p} is optimal. Functions in W1,pW^{1,p} may not be in LqL^q for q>pβˆ—q > p^*, as shown by explicit counterexamples.

This theorem and its proof technique are fundamental to understanding regularity of solutions to PDEs.