Sobolev Spaces

In this post I would like to introduce Sobolev spaces, which is one of the most important in PDE theory.

Hölder Spaces

Definition.

Assume Ω⊂Rd\Omega \subset \R^d is open and γ∈(0,1]\gamma \in (0, 1]. Then the function u:Ω→Ru : \Omega \to \R is Hölder continuous if

∣u(x)−u(y)∣≤C∣x−y∣γfor x,y∈Ω|u(x) - u(y)| \leq C | x - y |^\gamma \quad \text{for } x, y \in \Omega

for some constant CC.

Definition.

First we define the norm of continuous function

∥u∥C(Ω‾)=sup⁡x∈Ω∣u(x)∣.\| u \|_{\C (\overline{\Omega})} = \sup_{x \in \Omega} | u(x) |.

Then the γth\gamma^{th}-Hölder seminorm of u:Ω→Ru : \Omega \to \R is defined by

[u]C0,γ(Ω‾)=sup⁡x,y∈Ωx≠y{∣u(x)−u(y)∣∣x−y∣γ}[u]_{\C^{0, \gamma}}(\overline{\Omega}) = \sup_{\substack{x, y \in \Omega \\ x \neq y}} \left\{ \frac{|u(x) - u(y)|}{|x-y|^\gamma}\right\}

and the γth\gamma^{th}-Hölder norm is defined by

∥u∥C0,γ(Ω‾)=∥u∥C(Ω‾)+[u]C0,γ(Ω‾).\| u \|_{\C^{0, \gamma}(\overline{\Omega})} = \|u\|_{\C(\overline{\Omega})} + [u]_{\C^{0, \gamma}}(\overline{\Omega}).

For k≥1k \geq 1, we define

∥u∥Ck,γ(Ω‾)=∑∣α∣≤k∥Dαu∥C(Ω‾)+∑∣α∣=k[Dαu]C0,γ(Ω‾)\| u \|_{\C^{k, \gamma}(\overline{\Omega})} = \sum_{|\alpha| \leq k}\|D^\alpha u\|_{\C(\overline{\Omega})} + \sum_{|\alpha| = k} [D^\alpha u]_{\C^{0, \gamma}}(\overline{\Omega})

for u∈Ck(Ω‾)u \in \C^k(\overline{\Omega}).

Definition.

The Hölder space is defined by

Ck,r(Ω‾)={u∈Ck(Ω‾) ∣ ∥u∥Ck,γ(Ω‾)<∞},\C^{k, r}(\overline{\Omega}) = \left\{ u \in \C^k (\overline{\Omega}) \, \big | \, \| u \|_{\C^{k, \gamma}(\overline{\Omega})} < \infty \right\},

which is the set of functions whose Hölder norm is finite.

Theorem.

Ck,r(Ω‾)\C^{k, r}(\overline{\Omega}) is a Banach space.

Proof. Obvious.

Weak Derivatives

Definition.

Suppose u,v∈Lloc1(Ω)u, v \in L^1_{\loc}(\Omega) and α\alpha is a multiindex. We say that vv is the αth\alpha^{th}-weak partial derivative of uu, denoted by Dαu=vD^\alpha u = v if

∫ΩuDαϕdx=(−1)∣α∣∫Ωvϕdx\int_\Omega u D^\alpha \phi dx = (-1)^{|\alpha|} \int_\Omega v \phi dx

for all ϕ∈Cc∞(Ω)\phi \in \C^\infty_c (\Omega).

Lemma.

If DαuD^\alpha u exists, then it is uniquely defined up to a set of measure zero.

Proof. Assume that v,v~∈Lloc1(Ω)v, \tilde{v} \in L^1_{\loc}(\Omega) satisfy

∫ΩuDαϕdx=(−1)∣α∣∫Ωvϕdx=(−1)∣α∣∫Ωv~ϕdx\int_\Omega u D^\alpha \phi dx = (-1)^{|\alpha|} \int_\Omega v \phi dx = (-1)^{|\alpha|} \int_\Omega \tilde{v} \phi dx

for all ϕ∈Cc∞(Ω)\phi \in \C^\infty_c(\Omega) therefore v=v~v = \tilde{v} almost everywhere.

Sobolev Spaces

Definition.

The Sobolev space is defined as

Wk,p(Ω)={u:Ω→R ∣ u∈Lloc1, Dαu∈Lp(Ω) in the weak sense for each multiindex ∣α∣≤k}\begin{split} W^{k, p}(\Omega) & = \bigg\{ u : \Omega \to \R \, \big | \, u \in L^1_{\loc}, \, D^\alpha u \in L^p (\Omega)\\ & \qquad \text{ in the weak sense for each multiindex }|\alpha| \leq k \bigg\} \end{split}

and the Sobolev norm is defined by

∥u∥Wk,p(Ω)={(∑∣α∣≤k∫Ω∣Dαu∣pdx)1p(1≤p<∞)∑∣α∣≤kess⁡sup⁡Ω∣Dαu∣(p=∞).\| u \|_{W^{k, p}(\Omega)} = \begin{cases} \left(\sum_{|\alpha| \leq k} \int_\Omega |D^\alpha u|^p dx \right)^{\frac{1}{p}} \quad (1 \leq p < \infty)\\ \sum_{|\alpha| \leq k} \ess \sup_\Omega |D^\alpha u| (p = \infty). \end{cases}

Definition.

The sequence {um}\{ u_m\} converges to u∈Wk,p(Ω)u \in W^{k, p}(\Omega) in Wk,p(Ω)W^{k, p}(\Omega) denoted by

um→uin Wk,p(Ω)u_m \to u \quad \text{in }W^{k, p}(\Omega)

whenever

lim⁡m→∞∥um−u∥Wk,p(Ω)=0.\lim_{m \to \infty} \|u_m - u\|_{W^{k, p}(\Omega)} = 0.

Definition.

We denote

um→uin Wlock,p(Ω)u_m \to u \quad \text{in }W_\loc^{k, p}(\Omega)

to mean

um→uin Wk,p(V)u_m \to u \quad \text{in } W^{k, p}(V)

for each V⊂⊂ΩV \subset \subset \Omega.

Definition.

We denote

W0k,p(Ω)=Cc∞(Ω)‾⊂Wk,p(Ω).W_0^{k, p}(\Omega) = \overline{\C_c^\infty(\Omega)} \subset W^{k, p}(\Omega).

Properties of Sobolev Spaces

Theorem.

Assume Ω⊂Rd\Omega \subset \R^d be an open, bounded set. If u,v∈Wk,p(Ω)u, v \in W^{k, p}(\Omega) and ∣α∣≤k|\alpha| \leq k, then the followings are satisfied.

  • Dαu∈Wk−∣α∣,p(Ω)D^\alpha u \in W^{k - |\alpha|,p}(\Omega)

    and

    Dβ(Dαu)=Dα(Dβu)=Dα+βuD^\beta(D^\alpha u) = D^\alpha(D^\beta u) = D^{\alpha + \beta} u

    for all multiindices α\alpha, β\beta with ∣α∣+∣β∣≤k|\alpha| + |\beta| \leq k.

  • For each λ,μ∈R\lambda, \mu \in \R,

    λu+μv∈Wk,p(Ω)\lambda u + \mu v \in W^{k, p}(\Omega)

    and

    Dα(λu+μv)=λDαu+μDαv.D^\alpha(\lambda u + \mu v) = \lambda D^\alpha u + \mu D^\alpha v.
  • If V⊂ΩV \subset \Omega is open, then

    u∈Wk,p(V).u \in W^{k, p}(V).
  • If ζ∈Cc∞(Ω)\zeta \in \C_c^\infty(\Omega), then ζu∈Wk,p(Ω)\zeta u \in W^{k, p}(\Omega) and

    Dα(ζu)=∑β≤α(αβ)DβζDα−βu.D^\alpha (\zeta u) = \sum_{\beta \leq \alpha} \binom{\alpha}{\beta} D^\beta \zeta D^{\alpha - \beta} u.

Proof. The proof is so tedious that it is left as an exercise for the readers.

Theorem.

Assume Ω⊂Rd\Omega \subset \R^d be an open, bounded set. Then the Sobolev space Wk,p(Ω)W^{k, p} (\Omega) is a Banach space.

Proof. Obvious.

Approximation

Interior Approximation by Smooth Functions

Theorem.

Assume Ω⊂Rd\Omega \subset \R^d is an open, bounded set with C1\C^1 boundary and denote Ωε={x∈Ω∣dist⁡{x,∂Ω}>ε}.\Omega_\varepsilon = \{ x \in \Omega \mid \dist\{x, \partial \Omega\} > \varepsilon\}. If 1≤p<∞1 \leq p < \infty and u∈Wk,p(Ω)u \in W^{k, p}(\Omega), then

uε∈C∞(Ωε)u^\varepsilon \in \C^\infty (\Omega_\varepsilon)

and

uε→uu^\varepsilon \to u

in Wlock,p(Ω)W_\loc^{k, p}(\Omega) as ε→0\varepsilon \to 0 where uε=ηε∗uu^\varepsilon = \eta_\varepsilon * u in Ωε\Omega_\varepsilon.

Proof. It suffices to show that uε→uu^\varepsilon \to u \quadin Wlock,p(Ω)W_\loc^{k, p}(\Omega) as ε→0\varepsilon \to 0.

We claim that if ∣α∣≤k|\alpha| \leq k, then Dαuε=ηε∗DαuD^\alpha u^\varepsilon = \eta_\varepsilon * D^\alpha u in Ωε\Omega_\varepsilon.

The definition of the nthn^{th}-weak partial derivative implies

Dαuε(x)=Dα∫Ωηε(x−y)u(y)dy=∫ΩDxαηε(x−y)u(y)dy=(−1)∣α∣∫ΩDyαηε(x−y)u(y)dy=∫Ωηε(x−y)Dαu(y)dy=[ηε∗Dαu](x).D^{\alpha} u^{\varepsilon}(x) =D^{\alpha} \int_{\Omega} \eta_{\varepsilon}(x-y) u(y) d y \\ = \int_{\Omega} D_{x}^{\alpha} \eta_{\varepsilon}(x-y) u(y) d y \\ = (-1)^{|\alpha|} \int_{\Omega} D_{y}^{\alpha} \eta_{\varepsilon}(x-y) u(y) d y\\ = \int_\Omega \eta_\varepsilon (x-y) D^\alpha u(y) dy\\ = [\eta_\varepsilon * D^\alpha u] (x).

Now for an open set V⊂⊂ΩV \subset \subset \Omega,

Dαuε→DαuD^\alpha u^\varepsilon \to D^\alpha u

in Lp(V)L^p(V) as ε→0\varepsilon \to 0, which implies

∥uε−u∥Wlock,p(Ω)p=∑∣α∣≤k∫Ω∥Dαuε∥Lp(V)pdx→0\|u_\varepsilon - u\|_{W_\loc^{k, p}(\Omega)}^p = \sum_{|\alpha| \leq k}\int_\Omega \|D^\alpha u_\varepsilon\|_{L^p(V)}^p dx \to 0

as ε→0\varepsilon \to 0.

Global Approximation by Smooth Functions

Theorem.

Assume Ω⊂Rd\Omega \subset \R^d is an open, bounded set with C1\C^1 boundary. If 1≤p<∞1 \leq p < \infty and u∈Wk,p(Ω)u \in W_{k, p}(\Omega), then there exists a sequence of functions um∈C∞(Ω)∩Wk,p(Ω)u_m \in \C^\infty(\Omega) \cap W^{k, p}(\Omega) such that

um→uu_m \to u

in Wk,p(Ω)W^{k, p} (\Omega).

Proof. We denote

Ωi={x∈Ω ∣ dist⁡{x,∂Ω}>1i}.\Omega_i = \left\{ x \in \Omega \,\bigg|\, \dist \{x, \partial \Omega\} > \frac{1}{i} \right\}.

Write Vi=Ωi+3−Ωi+1‾V_i = \Omega_{i+3} - \overline{\Omega_{i+1}} (i≥1)(i \geq 1) and choose an open set V0⊂⊂ΩV_0 \subset \subset \Omega so that Ω=∑i=0∞Vi\Omega = \sum_{i=0}^\infty V_i.

Now let ζi∈Cc∞(Vi)\zeta_i \in \C^\infty_c (V_i) be a sequence of function such that 0≤ζi≤10 \leq \zeta_i \leq 1 and

∑i=0∞ζi(x)=1\sum_{i=0}^\infty \zeta_i(x) = 1

for all x∈Ωx \in \Omega.

Choose any u∈Wk,p(Ω)u \in W^{k, p}(\Omega) then by (\Omega)andand\supp(\zeta_i u) \subset V_i$.

Let δ>0\delta > 0 be given and choose εi>0\varepsilon_i > 0 so that ui=ηεi∗(ζiu)u^i = \eta_{\varepsilon_i} * (\zeta_i u) satisfies

∥ui−ζiu∥Wk,p(Ω)≤δ2i+1(i≥0)\| u^i - \zeta_i u \|_{W^{k, p}(\Omega)} \leq \frac{\delta}{2^{i+1}} \quad (i \geq 0)

and

supp⁡ui⊂Wi(i≥1)\supp u^i \subset W_i \quad (i \geq 1)

where Wi=Ωi+4−Ω‾i⊃ViW_i = \Omega_{i+4} - \overline{\Omega}_i \supset V_i.

Let v=∑i=0∞ui∈C∞(Ω)v = \sum_{i=0}^\infty u^i \in \C^\infty (\Omega), by noticing that u=∑i=1∞ζiuu = \sum_{i=1}^\infty \zeta_i u we have for each V⊂⊂ΩV \subset \subset \Omega

∥v−u∥Wk,p(V)=∥∑i=0∞(ui−ζiu)∥Wk,p(V)≤∑i=0∞∥ui−ζiu∥Wk,p(Ω)∥≤δ∑i=0∞12i+1=δ.\|v - u\|_{W^{k, p}(V)} = \left \| \sum_{i=0}^\infty (u_i - \zeta_i u) \right\|_{W^{k, p}(V)}\\ \leq \sum_{i=0}^\infty \| u_i - \zeta_i u\|_{W^{k, p}(\Omega)} \| \\ \leq \delta \sum_{i = 0}^\infty \frac{1}{2^{i + 1}} \\ = \delta.

Taking the supremum over sets V⊂⊂ΩV \subset \subset \Omega, we conclude ∥v−u∥Wk,p(V)≤δ\|v - u\|_{W^{k, p}(V)} \leq \delta.

Theorem.

Assume Ω⊂Rd\Omega \subset \R^d is an open, bounded set with C1\C^1 boundary. If 1≤p<∞1 \leq p < \infty and u∈Wk,p(Ω)u \in W_{k, p}(\Omega), then there exists a sequence of functions um∈C∞(Ω‾)u_m \in \C^\infty(\overline{\Omega}) such that

um→uu_m \to u

in Wk,p(Ω).W^{k, p}(\Omega).

Extensions and Traces

Extensions

Theorem.

Assume Ω⊂Rd\Omega \subset \R^d is an open, bounded set with C1\C^1 boundary. Select a bounded open set VV such that Ω⊂⊂V\Omega \subset \subset V. Then there exists a bounded linear operator

E:W1,p(Ω)→W1,p(Rn)E: W^{1, p}(\Omega) \rightarrow W^{1, p}\left(\mathbb{R}^{n}\right)

satisfying the following properties for each u∈W1,p(Ω)u \in W^{\mathbf{1}, p}(\Omega).

  • Eu=uE u = u almost everywhere in Ω\Omega.
  • supp⁡(Eu)⊂V\supp(Eu) \subset V.
  • ∥Eu∥W1,p(Rd)≤C∥u∥W1,p(Ω)\| Eu \|_{W^{1, p}(\R^d)} \leq C \| u \|_{W^{1, p}(\Omega)} for fixed pp, uu, and VV.

Proof. First we construct the operator EE by assuming that u∈C∞(Ω‾)u \in \C^\infty (\overline{\Omega}). Fix x0∈∂Ωx^0 \in \partial \Omega and let Φ:x↦y\Phi : x \mapsto y be a C1\C^1 mapping with Φ(x0)=0\Phi(x_0) = 0 and its inverse Ψ\Psi such that

Φ(∂Ω∩Br1(x0))=Br(0)∩{yn=0}andΦ(Ω∩Br1(x0))=Br(0)∩{yn>0}\Phi(\partial \Omega \cap B_{r_1}(x_0)) = B_r(0) \cap \{y_n = 0\} \quad \text{and} \quad \Phi(\Omega \cap B_{r_1}(x_0)) = B_r(0) \cap \{y_n > 0\}

for some r,r1>0r, r_1 > 0. We write y=Φ(x)y = \Phi (x), u′(y)=u(Ψ(y))u'(y) = u(\Psi(y)) and denote

{B+=Br(0)∩{yn≥0}⊂Φ(Ω‾)B−=Br(0)∩{yn≤0}⊂Φ(Rd∖Ω).\begin{cases} B^+ = B_r(0) \cap \{ y_n \geq 0 \} \subset \Phi(\overline{\Omega})\\ B^- = B_r(0) \cap \{ y_n \leq 0 \} \subset \Phi(\R^d \setminus \Omega). \end{cases}

Then we define

uˉ′(y)={u′(y)if y∈B+−3u′(y1,⋯ ,yn−1,−yn)+4u′(y1,⋯ ,yn−1,−yn2)if y∈B−.\bar{u}'(y) = \begin{cases} u'(y) \quad \text{if }y \in B^+\\ -3 u'(y_1, \cdots, y_{n-1}, -y_n) + 4u'\left(y_1, \cdots, y_{n-1}, -\frac{y_n}{2}\right) \quad \text{if }y \in B^-. \end{cases}

\begin{claim} uˉ′∈C1(Br(0))\bar{u}' \in \C^1 (B_r(0)). \end{claim} Denote u−=uˉ′∣B−u^-=\bar{u}'|_{B^{-}} and u+=uˉ′∣B+u^+=\bar{u}'|_{B^{+}} then we observe

∂u−∂xn(x)=3∂u∂xn(x1,…,xn−1,−xn)−2∂u∂xn(x1,…,xn−1,−xn2)\frac{\partial u^{-}}{\partial x_{n}}(x)=3 \frac{\partial u}{\partial x_{n}}\left(x_{1}, \ldots, x_{n-1},-x_{n}\right)-2 \frac{\partial u}{\partial x_{n}}\left(x_{1}, \ldots, x_{n-1},-\frac{x_{n}}{2}\right)

hence uxn−∣xn=0=uxn+∣xn=0u_{x_n}^-|_{x_n = 0} = u_{x_n}^+|_{x_n = 0}. Furthermore

uxi−∣xn=0=uxi+∣xn=0(1≤i≤n−1)u_{x_i}^-|_{x_n = 0} = u_{x_i}^+|_{x_n = 0} \quad (1 \leq i \leq n-1)

since u−=u+u^- = u^+ on {xn=0}\{x_n = 0\}. These imply

Dαu−∣xn=0=Dαu+∣xn=0D^\alpha u^- |_{x_n = 0} = D^\alpha u^+ |_{x_n = 0}

for each ∣α∣≤1|\alpha| \leq 1 so prove the claim.

\Omegasing these calculations we obtain

∥uˉ′∥W1,p(Br(0))≤C∥u′∥W1,p(B+)\|\bar{u}'\|_{W^{1, p}(B_r(0))} \leq C\|u'\|_{W^{1, p}\left(B^{+}\right)}

for some constant CC which does not depend on uu.

Let W=Ψ(Br(0))W = \Psi(B_r(0)) then we have the estimate

∥uˉ∥W1,p(W)≤C∥u∥W1,p(Ω).\|\bar{u}\|_{W^{1, p}(W)} \leq C\|u\|_{W^{1, p}(\Omega)}.

Since ∂Ω\partial \Omega is compact, there exist finitely many points xi0∈∂Ωx_i^0 \in \partial \Omega (1≤i≤N)(1 \leq i \leq N) so that each open set WiW_i constructed from xi0x_i^0 and extensions uˉ\bar{u} of uu to WiW_i as above such that

∂Ω⊂⋃i=1NWi.\partial \Omega \subset \bigcup_{i=1}^N W_i.

Take W0⊂⊂ΩW_0 \subset \subset \Omega so that

Ω⊂⋃i=0NWi\Omega \subset \bigcup_{i=0}^N W_i

and let ζi∈Cc∞(Vi),(1≤i≤N,0≤ζi≤1)\zeta_i \in \C^\infty_c (V_i), (1 \leq i \leq N, 0 \leq \zeta_i \leq 1) be an associated partition of unity such that

∑i=0Nζi(x)=1\sum_{i=0}^N \zeta_i(x) = 1

for all x∈Ωx \in \Omega.

Let uˉ=∑i=0Nζiuˉi\bar{u} = \sum_{i=0}^N \zeta_i \bar{u}_i where uˉ0=u\bar{u}_0 = u then we obtain the bound

∥uˉ∥W1,p(Rd)≤C∥u∥W1,p(Ω).\|\bar{u}\|_{W^{1, p}(\R^d)} \leq C\|u\|_{W^{1, p}(\Omega)}.

for some constant CC which does not depend on uu. Furthermore we can arrange supp⁡uˉ⊂V\supp \bar{u} \subset V where V⊃⊃ΩV \supset \supset \Omega. Note that EE is indeed linear.

So we have constructed the operator E:C∞(Ω)→W1,p(Rd)E : \C^\infty(\Omega) \to W^{1, p}(\R^d). Hereafter we are going to extend its domain to W1,p(Ω)W^{1, p}(\Omega), which is our desired result. Let um∈C∞(Ω‾)u_m \in \C^\infty(\overline{\Omega}) such that um→u∈W1,p(Ω)u_m \to u \in W^{1, p}(\Omega), then the previous estimate and the linearity of EE imply

∥Eum−Eul∥W1,p(Rd)≤C∥um−ul∥W1,p(Ω)\|E u_m - E u_l \|_{W^{1, p}(\R^d)} \leq C\|u_m - u_l \|_{W^{1, p}(\Omega)}

thus EumE u_m is a Cauchy sequence converging to EuEu.

Traces

Definition.

Assume Ω⊂Rd\Omega \subset \R^d is an open, bounded set with C1\C^1 boundary. The trace of uu on ∂Ω\partial \Omega is a bounded linear operator T:W1,p(Ω)→Lp(∂Ω)T : W^{1, p}(\Omega) \to L^p (\partial \Omega) such that

Tu=u∣∂Ωif u∈W1,p(Ω)∩C(Ω‾),T u = u |_{\partial \Omega} \quad \text{if }u \in W^{1, p}(\Omega) \cap \C(\overline{\Omega}),

and

∥Tu∥Lp(∂Ω)≤C∥u∥W1,p(Ω)for all u∈W1,p(Ω)\|T u \|_{L^p (\partial \Omega)} \leq C \| u\|_{W^{1, p}(\Omega)} \quad \text{for all }u \in W^{1, p} (\Omega)

where CC is a constant independent to uu.

Theorem (trace theorem).

There exists the trace TuT u of uu on ∂Ω\partial \Omega.

Proof. Assume first u∈C1(Ω‾)u \in \C^1 (\overline{\Omega}) and repeat what we have done at the proof of previous theorem as follows. Fix x0∈∂Ωx^0 \in \partial \Omega and let Φ:x↦y\Phi : x \mapsto y be a C1\C^1 mapping with Φ(x0)=0\Phi(x_0) = 0 and its inverse Ψ\Psi such that

Φ(∂Ω∩Br1(x0))=Br(0)∩{yn=0}andΦ(Ω∩Br1(x0))=Br(0)∩{yn>0}\Phi(\partial \Omega \cap B_{r_1}(x_0)) = B_r(0) \cap \{y_n = 0\} \quad \text{and} \quad \Phi(\Omega \cap B_{r_1}(x_0)) = B_r(0) \cap \{y_n > 0\}

for some r,r1>0r, r_1 > 0. We write y=Φ(x)y = \Phi (x), u′(y)=u(Ψ(y))u'(y) = u(\Psi(y)) and denote

{B+=Br(0)∩{yn≥0}⊂Φ(Ω‾)B−=Br(0)∩{yn≤0}⊂Φ(Rd∖Ω).\begin{cases} B^+ = B_r(0) \cap \{ y_n \geq 0 \} \subset \Phi(\overline{\Omega})\\ B^- = B_r(0) \cap \{ y_n \leq 0 \} \subset \Phi(\R^d \setminus \Omega). \end{cases}

Select ζ∈Cc∞(Br(0))\zeta \in \C^\infty_c (B_r(0)) with ζ≥0\zeta \geq 0 such that ζ≡1\zeta \equiv 1 in Br2(x)B_{\frac{r}{2}}(x) and let Γ=Φ(∂Ω)∩Br2(0)\Gamma = \Phi(\partial \Omega) \cap B_{\frac{r}{2}}(0). Setting y′=(y1,⋯ ,yn−1)∈Rn−1y' = (y_1, \cdots, y_{n-1}) \in \R^{n-1}, we obtain

∫Γ∣u∣pdy′≤∫{yn=0}ζ∣u∣pdy′=−∫B+(ζ∣u∣p)yndy=−∫B+∣u∣pζyn+p∣u∣p−1(sgn⁡u)uynζdy≤C∫B+∣u∣p+∣Du∣pdy\int_\Gamma |u|^p dy' \leq \int_{\{y_n = 0\}} \zeta |u|^p dy' \\ = - \int_{B^+} (\zeta |u|^p)_{y_n} dy \\ = - \int_{B^+} |u|^p \zeta_{y_n} + p |u|^{p-1} (\sgn u) u_{y_n} \zeta dy \\ \leq C \int_{B^+} |u|^p + |D u|^p dy

by the Divergence theorem and Young’s inequality.

Since ∂Ω\partial \Omega is compact, there exist finitely many points xi0∈∂Ωx_i^0 \in \partial \Omega (1≤i≤N)(1 \leq i \leq N) so that each open set Γi\Gamma_i constructed from xi0x_i^0 such that

∂Ω⊂⋃i=1NWiand∥u∥Lp(Wi)≤C∥u∥W1,p(Ω)\partial \Omega \subset \bigcup_{i=1}^N W_i \quad \text{and} \quad \| u \|_{L^p(W_i)} \leq C \|u\|_{W^{1, p}(\Omega)}

where Wi=Ψ(Γi)W_i = \Psi (\Gamma_i).

Consequently if we define Tu=u∣∂ΩT u = u |_{\partial \Omega} then we obtain

∥Tu∥Lp(∂Ω)≤C∥u∥W1,p(Ω)\| Tu \|_{L^p(\partial \Omega)} \leq C \|u\|_{W^{1, p}(\Omega)}

for some constant CC independent to uu.

Now we have the trace operator T:C1(Ω‾)→Lp(∂Ω)T : \C^1(\overline{\Omega}) \to L^p (\partial \Omega). Henceforth we are going to extend its domain to W1,p(Ω)W^{1, p}(\Omega), which is analogous to what we have done in the previous theorem. Let um∈C∞(Ω‾)u_m \in \C^\infty(\overline{\Omega}) such that um→u∈W1,p(Ω)u_m \to u \in W^{1, p}(\Omega), then

∥Tum−Tul∥Lp(∂Ω)≤C∥um−ul∥W1,p(Ω)\| Tu_m - Tu_l \|_{L^p(\partial \Omega)} \leq C \|u_m - u_l \|_{W^{1, p}(\Omega)}

thus TumT u_m is a Cauchy sequence and we define Tu=lim⁡m→∞TumTu = \lim_{m \to \infty} T u_m.

Whereas the boundary values of C1\C^1 functions are well defined, those of W1,pW^{1, p} functions are not in general. Instead, the trace operator could assign well-defined boundary values of the W1,pW^{1, p} function, which allows us to solve the boundary-value problem in Sobolev spaces.

Theorem (trace-zero functions).

Assume Ω⊂Rd\Omega \subset \R^d is an open, bounded set with C1\C^1 boundary. If u∈W1,p(Ω)u \in W^{1, p}(\Omega), then u∈W01,p(Ω)u \in W_0^{1,p}(\Omega) if and only if Tu=0T u = 0 on ∂Ω\partial \Omega.

Proof. We omit this proof for the first reading.

Sobolev Inequalities and Embedding

Gagliardo-Nirenberg-Sobolev Inequality

Definition.

If 1≤p<n1 \leq p < n, we define The Sobolev conjugate of pp as

p∗=npn−p.p^* = \frac{np}{n - p}.

Theorem (Gagliardo-Nirenberg-Sobolev).

If 1≤p<n1 \leq p < n, then there exists a constant CC independent to uu such that

∥u∥Lp∗(Rd)≤C∥Du∥Lp(Rd)\|u \|_{L^{p^*}(\R^d)} \leq C \| Du \|_{L^p (\R^d)}

for all u∈Cc1(Rd)u \in \C_c^1(\R^d).

Proof. Since uu has compact support, for 1≤i≤n1 \leq i \leq n and x∈Rdx \in \R^d we have

u(x)=∫−∞xiuxi(x1,⋯ ,xi−1,yi,xi+1,⋯ ,xn)dyiu(x) = \int_{-\infty}^{x_i} u_{x_i}(x_1, \cdots, x_{i-1}, y_i, x_{i+1}, \cdots, x_n)dy_i

and so

∣u(x)∣≤∫−∞∞∣Du(x1,⋯ ,yi,⋯ ,xn)∣dyi|u(x)| \leq \int_{-\infty}^\infty |Du(x_1, \cdots, y_i, \cdots, x_n)| d y_i

hence

∣u(x)∣nn−1≤∏i=1n(∫−∞∞∣Du(x1,⋯ ,yi,⋯ ,xn)∣dyi)1n−1.|u(x)|^{\frac{n}{n-1}} \leq \prod_{i=1}^{n} \left( \int_{-\infty}^\infty |Du(x_1, \cdots, y_i, \cdots, x_n)| d y_i \right)^\frac{1}{n-1}.

Integrating above with respect to x1x_1, we obtain

∫−∞∞∣u∣nn−1dx1≤∫−∞∞∏i=1n(∫−∞∞∣Du∣dyi)1n−1dx1=(∫−∞∞∣Du∣dy1)1n−1∫−∞∞∏i=2n(∫−∞∞∣Du∣dyi)1n−1dx1≤(∫−∞∞∣Du∣dy1)1n−1(∏i=2n∫−∞∞∫−∞∞∣Du∣dx1dyi)1n−1\int_{-\infty}^{\infty} |u|^\frac{n}{n-1} dx_1 \leq \int_{-\infty}^{\infty} \prod_{i=1}^{n} \left( \int_{-\infty}^\infty |Du| d y_i \right)^\frac{1}{n-1} dx_1 \\ = \left( \int_{-\infty}^\infty |Du| d y_1 \right)^\frac{1}{n-1} \int_{-\infty}^{\infty} \prod_{i=2}^{n} \left( \int_{-\infty}^\infty |Du| d y_i \right)^\frac{1}{n-1} dx_1 \\ \leq \left( \int_{-\infty}^\infty |Du| d y_1 \right)^\frac{1}{n-1} \left( \prod_{i=2}^{n} \int_{-\infty}^\infty \int_{-\infty}^{\infty} |Du| dx_1 dy_i \right)^\frac{1}{n-1}

by the general Hölder inequality. Continuing the integration with respect to x2,⋯ ,xnx_2, \cdots, x_n we conclude

∫Rd∣u∣nn−1dx≤∏i=1n(∫−∞∞⋯∫−∞∞∣Du∣dx1⋯dxn)1n−1=(∫Rd∣Du∣dx)nn−1.\int_{\R^d} |u|^{\frac{n}{n-1}} dx \leq \prod_{i=1}^n \left(\int_{-\infty}^{\infty} \cdots \int_{-\infty}^{\infty} |Du| dx_1 \cdots dx_n \right)^{\frac{1}{n-1}} \\ = \left(\int_{\R^d} |Du| dx\right)^\frac{n}{n-1}.

This is the very estimate we desired for p=1p = 1.

Now we consider the case of 1<p<n1 < p < n. From the previous estimate for p=1p=1, we observe that where γ>1\gamma > 1

(∫Rd∣u∣γnn−1dx)n−1n≤∫Rd∣D∣u∣γ∣dx=γ∫Rd∣u∣γ−1∣Du∣dx≤γ(∫Rd∣u∣(γ−1)pp−1dx)p−1p(∫Rd∣Du∣pdx)1p\left(\int_{\R^d} |u|^{\frac{\gamma n}{n-1}} dx\right)^\frac{n-1}{n} \leq \int_{\R^d} |D |u|^\gamma | dx \\ = \gamma \int_{\R^d} |u|^{\gamma - 1} |D u | dx \\ \leq \gamma \left(\int_{\R^d} |u|^{(\gamma - 1) \frac{p}{p-1}}dx\right)^{\frac{p-1}{p}} \left(\int_{\R^d} |D u |^p dx \right)^{\frac{1}{p}}

Here we choose γ=p(n−1)n−p>1\gamma = \frac{p(n-1)}{n-p} > 1 so that γnn−1=(γ−1)pp−1=npn−p=p∗\frac{\gamma n}{n-1} = (\gamma - 1)\frac{p}{p-1} = \frac{np}{n-p} = p^*. Then it becomes

∥u∥Lp∗(Rd)≤C∥Du∥Lp(Rd)\|u \|_{L^{p^*}(\R^d)} \leq C \| Du \|_{L^p (\R^d)}

which is what we wanted to show.

Theorem.

Assume Ω⊂Rd\Omega \subset \R^d is an open, bounded set with C1\C^1 boundary. If 1≤p<n1 \leq p < n and u∈W1,p(Ω)u \in W^{1, p}(\Omega), then

u∈Lp∗(Ω)u \in L^{p^*}(\Omega)

with the estimate

∥u∥Lp∗(Ω)≤C∥u∥W1,p(Ω)\|u\|_{L^{p^*}(\Omega)} \leq C \|u\|_{W^{1,p}(\Omega)}

where CC is a constant independent to uu.

Proof. Since ∂Ω\partial \Omega is C1\C^1 is compact, and

∥uˉ∥W1,p(Rd)≤C∥u∥W1,p(Ω),\| \bar{u} \|_{W^{1, p}(\R^d)} \leq C \|u\|_{W^{1, p}(\Omega)},

we have the inequalities

∥um−ul∥Lp∗(Rd)≤C∥Dum−Dul∥Lp(Rd)and∣um−u∥Lp∗(Rd)≤C∥Dum∥Lp(Rd)\| u_m - u_l \|_{L^{p^*}(\R^d)} \leq C \| Du_m - Du_l \|_{L^p (\R^d)} \quad \text{and} \quad | u_m - u\|_{L^{p^*}(\R^d)} \leq C \| Du_m \|_{L^p (\R^d)}

for l,m≥1l, m \geq 1 which yield

∥uˉ∥Lp∗(Rd)≤C∥Duˉ∥Lp(Rd).\| \bar{u} \|_{L^{p^*}(\R^d)} \leq C \| D \bar{u} \|_{L^p (\R^d)}.

Theorem.

Assume Ω⊂Rd\Omega \subset \R^d be an open, bounded set. If 1≤p<n1 \leq p < n and u∈W01,p(Ω)u \in W_0^{1, p}(\Omega), then

∥u∥Lq(Rd)≤C∥Du∥Lp(Ω)\|u\|_{L_{q}(\R^d)} \leq C \| D u \|_{L^p(\Omega)}

for each q∈[1,p∗]q \in [1, p*] where CC is a constant independent to uu.

\begin{note} This theorem does not require the boundary to be C1\C^1. \end{note}

Proof. Since u∈W0k,p(Ω)u \in W_0^{k, p}(\Omega), there exists a sequence of function um∈Cc∞(Ω)u_m \in \C_c^\infty(\Omega) such that

um→uin W1,p(Ω).u_m \to u \quad \text{in }W^{1, p}(\Omega).

We extend each umu_m to be 00 in \R^d \setminus \overline{\Omega}} then by we have our desired result.

Morrey’s Inequality

Theorem (Morrey).

Assume n<p≤∞n < p \leq \infty, then

∥u∥C0,γ(Rd)≤C∥u∥W1,p(Rd)\|u\|_{C^{0, \gamma}}(\R^d) \leq C \|u\|_{W^{1, p}(\R^d)}

for all u∈C1(Rd)u \in \C^1(\R^d), where γ=1−np\gamma = 1 - \frac{n}{p} and CC is a constant independent to uu.

Proof. Choose any ball B(x,r)⊂RdB(x, r) \subset \R^d. \begin{claim}

\dashintB(x,r)∣u(y)−u(x)∣dy≤C∫B(x,r)∣Du(y)∣∣y−x∣n−1dy\dashint_{B(x, r)} |u(y) - u(x)| dy \leq C \int_{B(x, r)} \frac{|Du(y)|}{|y - x|^{n-1}} dy

where CC is a constant depending only on nn. \end{claim}

Fix w∈∂B(0,1)w \in \partial B(0, 1), then for 0<s<r0 < s < r we have

∣u(x+sw)−u(x)∣=∣∫0sddtu(x+tw)dt∣=∣∫0sDu(x+tw)⋅wdt∣≤∫0s∣Du(x+tw)∣dt.|u(x + sw) - u(x)| = \left|\int_0^s \frac{d}{dt} u(x + tw) dt \right|\\ = \left|\int_0^s Du(x + tw) \cdot w dt \right|\\ \leq \int_0^s |Du(x + tw)| dt.

By letting y=x+twy = x + tw,

∫∂B(0,1)∣u(x+sw)−u(x)∣dS≤∫0s∫∂B(0,1)∣Du(x+tw)∣dSdt=∫Bs(x)∣Du(y)∣∣x−y∣n−1dy≤∫B(x,r)∣Du(y)∣∣x−y∣n−1dy\int_{\partial B(0, 1)} |u(x + sw) - u(x)| dS \leq \int_0^s \int_{\partial B(0, 1)} |Du(x + tw)| dS dt\\ = \int_{B_s(x)} \frac{|Du(y)|}{|x-y|^{n-1}}dy \\ \leq \int_{B(x, r)} \frac{|Du(y)|}{|x-y|^{n-1}}dy

which follows

∫B(x,r)∣u(y)−u(x)∣dy≤rnb∫B(x,r)∣Du(y)∣∣x−y∣n−1dy.\int_{B(x, r)} |u(y) - u(x)| dy \leq \frac{r^n}{b} \int_{B(x, r)} \frac{|Du(y)|}{|x-y|^{n-1}}dy.

Definition.

We say u∗u^* is a version of uu if u=u∗u = u^* almost everywhere.

Theorem.

Assume Ω⊂Rd\Omega \subset \R^d is an open, bounded set with C1\C^1 boundary. If n<p≤∞n < p \leq \infty and u∈W1,p(Ω)u \in W^{1, p}(\Omega), then uu has a version

u∗∈C0,γ(Ω‾)u^* \in \C^{0, \gamma}(\overline{\Omega})

such that

∥u∗∥C0,γ(Ω‾)≤C∥u∥W1,p(Ω)\|u^*\|_{\C^{0, \gamma}(\overline{\Omega})} \leq C \|u\|_{W^{1, p}(\Omega)}

where γ=1−np\gamma = 1 - \frac{n}{p} and CC is a constant independent to uu.

Proof. Choose any ball B(x,r)⊂RdB(x, r) \subset \R^d. \begin{claim}

\dashintB(x,r)∣u(y)−u(x)∣dy≤C∫B(x,r)∣Du(y)∣∣y−x∣n−1dy\dashint_{B(x, r)} |u(y) - u(x)| dy \leq C \int_{B(x, r)} \frac{|Du(y)|}{|y - x|^{n-1}} dy

where CC is a constant depending only on nn. \end{claim}

General Sobolev Inequalities

Theorem.

Assume Ω⊂Rd\Omega \subset \R^d be bounded and open with ∂Ω\partial \Omega is C1\C^1. Let u∈Wk,p(Ω)u \in W^{k, p}(\Omega).

  • If k<npk < \frac{n}{p}, then

    u∈Lq(Ω)u \in L^q (\Omega)

    and

    ∥u∥Lq(Ω)≤C∥u∥Wk,p(Ω)\|u\|_{L^q(\Omega)} \leq C \|u \|_{W^{k,p}(\Omega)}

    where 1q=1p−kn\frac{1}{q} = \frac{1}{p} - \frac{k}{n}.

  • If k>npk > \frac{n}{p}, then

    u∈Ck−⌊np⌋−1,γ(Ω‾)u \in \C^{k - \left\lfloor\frac{n}{p}\right\rfloor -1, \gamma}(\overline{\Omega})

    and

    ∥u∥Ck−⌊np⌋−1,γ(Ω‾)≤C∥u∥Wk,p(Ω)\|u\|_{\C^{k - \left\lfloor\frac{n}{p}\right\rfloor -1, \gamma}}(\overline{\Omega}) \leq C \|u\|_{W^{k, p}(\Omega)}

    where

    γ={⌊np⌋+1−npif np∉Zany positive number<1if np∈Z.\gamma = \begin{cases} \left\lfloor\frac{n}{p}\right\rfloor + 1 - \frac{n}{p} \quad \text{if }\frac{n}{p} \not\in \Z\\ \text{any positive number} < 1 \quad \text{if }\frac{n}{p} \in \Z. \end{cases}

Here the constant CC is independent to uu.

Proof. We skip the proof for now.

Summary

Suppose Ω⊂Rd\Omega \subset \R^d is bounded, ∂Ω\partial \Omega is C1\C^1, u∈W1,p(Ω)u \in W^{1, p}(\Omega) and V⊃⊃ΩV \supset \supset \Omega.

  • (\R^d)$

    {Eu=ualmost everywhere in Ωsupp⁡(Eu)⊂V∥Eu∥W1,p(Rd)≤C∥u∥W1,p(Ω)\begin{cases} Eu = u \quad \text{almost everywhere in }\Omega\\ \supp(Eu) \subset V\\ \|Eu\|_{W^{1,p}(\R^d)} \leq C \|u\|_{W^{1,p}(\Omega)} \end{cases}
  • $$

$$Tu \in L^p(\partial \Omega)$

{Tu=u∣∂Ωif u∈W1,p(Ω)∩C(Ω‾)∥Tu∥Lp(∂Ω)≤C∥u∥W1,p(Ω)\begin{cases} Tu = u |_{\partial \Omega} \quad \text{if }u \in W^{1, p}(\Omega) \cap \C(\overline{\Omega})\\ \|T u \|_{L^p (\partial \Omega)} \leq C \| u\|_{W^{1, p}(\Omega)} \end{cases}
$$ u \in W_0^{1,p}(\Omega) \iff T u = 0 \quad \text{on }\partial \Omega \quad \text{( $$
  • [label=(\roman*)]
  • 1≤p<n1 \leq p < n:

    →G-N-Su∈Lp∗(Ω),∥u∥Lp∗(Ω)≤C∥u∥W1,p(Ω)\xrightarrow{\text{G-N-S}} u \in L^{p*}(\Omega), \quad \|u\|_{L^{p^*}(\Omega)} \leq C \|u\|_{W^{1,p}(\Omega)} \|u \|_{L^{p^*}(\R^d)} \leq C \| Du \|_{L^p (\R^d)} \quad \text{if }u \in \C_c^1(\R^d) \quad \text{(
  • n<p≤∞n < p \leq \infty:

    →Morreyu∗∈C0,γ(Ω‾),∥u∗∥C0,γ(Ω‾)≤C∥u∥W1,p(Ω)\xrightarrow{\text{Morrey}} u^* \in C^{0, \gamma}(\overline{\Omega}), \quad \|u^*\|_{\C^{0, \gamma}(\overline{\Omega})} \leq C \|u\|_{W^{1, p}(\Omega)} \|u\|_{C^{0, \gamma}}(\R^d) \leq C \|u\|_{W^{1, p}(\R^d)} \quad \text{if }u \in \C^1(\R^d) \quad \text{(
  • n=pn = p: →Poincare\xrightarrow{\text{Poincare}} ?

Compactness

Now we are going to show that W1,pW^{1, p} is indeed compactly embedded in Lq(Ω)L^q(\Omega).

Definition.

Let XX and YY be Banach spaces with X⊂YX \subset Y. XX is compactly embedded in YY, denoted by

X⊂⊂YX \subset \subset Y

if

∥x∥Y≤C∥x∥X\|x\|_Y \leq C \|x\|_X

and each bounded sequence in XX is precompact in YY, where CC is a constant.

Theorem (Rellich-Kondrachov).

Assume Ω⊂Rd\Omega \subset \R^d is open, bounded and ∂Ω\partial \Omega is C1\C^1. If 1≤p<n1 \leq p < n then

W1,p(Ω)⊂⊂Lq(Ω)W^{1, p}(\Omega) \subset \subset L^q(\Omega)

for each q∈[1,p∗)q \in [1, p^*).

Proof. Fix q∈[1,p∗)q \in [1, p^*) then since Ω\Omega is bounded we have the estimate

W1,p(Ω)⊂Lq(Ω)and∥u∥Lq(Ω)≤C∥u∥W1,p(Ω)W^{1, p}(\Omega) \subset L^q(\Omega) \quad \text{and} \quad \|u\|_{L^q (\Omega)} \leq C \|u\|_{W^{1,p}(\Omega)}

by .

Now it suffices to show that every bounded sequence of functions {um}∈W1,p(Ω)\{ u_m \} \in W^{1,p}(\Omega) has a convergent subsequence which converges in Lq(Ω)L^q (\Omega). Our strategy is to use Arzela-Ascolli

In the sense of < \infty$.

Define the smoothed functions umε=ηε∗umu_m^\varepsilon = \eta_\varepsilon * u_m which have compact support on VV.

\begin{claim}

umε→umin Lq(V) uniformlyu_m^\varepsilon \to u_m \quad \text{in }L^q(V)\text{ uniformly}

as ε→0\varepsilon \to 0. \end{claim}

Note that

umε(x)−um(x)=∫B(0,1)η(y)(um(x−εy)−um(x))dy=∫B(0,1)η(y)∫01ddt(um(x−εty))dtdy=−ε∫B(0,1)η(y)∫01Dum(x−εty)⋅ydtdyu_{m}^{\varepsilon}(x)-u_{m}(x) =\int_{B(0, 1)} \eta(y)\left(u_{m}(x-\varepsilon y)-u_{m}(x)\right) d y \\ =\int_{B(0, 1)} \eta(y) \int_{0}^{1} \frac{d}{d t}\left(u_{m}(x-\varepsilon t y)\right) d t d y \\ =-\varepsilon \int_{B(0, 1)} \eta(y) \int_{0}^{1} D u_{m}(x-\varepsilon t y) \cdot y d t d y

Hence

∫V∣umε(x)−um(x)∣dx≤ε∫B(0,1)η(y)∫01∫V∣Dum(x−εty)∣dxdtdy≤ε∫V∣Dum(z)∣dz\int_{V}\left|u_{m}^{\varepsilon}(x)-u_{m}(x)\right| d x \leq \varepsilon \int_{B(0, 1)} \eta(y) \int_{0}^{1} \int_{V}\left|D u_{m}(x-\varepsilon t y)\right| d x d t d y \\ \leq \varepsilon \int_{V}\left|D u_{m}(z)\right| d z

By approximation this estimate holds if um∈W1,p(V)u_{m} \in W^{1, p}(V). Hence

∥umε−um∥L1(V)≤ε∥Dum∥L1(V)≤εC∥Dum∥Lp(V),\left\|u_{m}^{\varepsilon}-u_{m}\right\|_{L^{1}(V)} \leq \varepsilon\left\|D u_{m}\right\|_{L^{1}(V)} \leq \varepsilon C\left\|D u_{m}\right\|_{L^{p}(V)},

we thereby discover

umε→umin L1(V) uniformly.u_{m}^{\varepsilon} \rightarrow u_{m} \quad \text{in }L^1(V)\text{ uniformly}.

But then since 1≤q<p∗1 \leq q<p^{*}, using the interpolation inequality for LpL^p-norms we obtain

∥umε−um∥Lq(V)≤∥umε−um∥L1(V)θ∥umε−um∥Lp∗(V)1−θ\left\|u_{m}^{\varepsilon}-u_{m}\right\|_{L^{q}(V)} \leq\left\|u_{m}^{\varepsilon}-u_{m}\right\|_{L^{1}(V)}^{\theta}\left\|u_{m}^{\varepsilon}-u_{m}\right\|_{L^{p^{*}}(V)}^{1-\theta}

where 1q=θ+(1−θ)p∗,0<θ<1\frac{1}{q}=\theta+\frac{(1-\theta)}{p^{*}}, 0<\theta<1. Consequently our previous estimate and imply

∥umε−um∥Lq(V)≤C∥umε−um∥L1(V)θ\left\|u_{m}^{\varepsilon}-u_{m}\right\|_{L^{q}(V)} \leq C\left\|u_{m}^{\varepsilon}-u_{m}\right\|_{L^{1}(V)}^{\theta}

which follows our claim.

\begin{claim} For a given ε>0\varepsilon > 0, the sequence umεu_m^\varepsilon is uniformly bounded and equicontinuous. \end{claim}

Indeed, if x∈Rnx \in \mathbb{R}^{n}, then

∣umε(x)∣≤∫B(x,ε)ηε(x−y)∣um(y)∣dy≤∥ηε∥L∞(Rn)∥um∥L1(V)≤Cεn<∞\begin{aligned} \left|u_{m}^{\varepsilon}(x)\right| \leq \int_{B(x, \varepsilon)} \eta_{\varepsilon}(x-y)\left|u_{m}(y)\right| d y \\ \leq\left\|\eta_{\varepsilon}\right\|_{L^{\infty}\left(\mathbb{R}^{n}\right)}\left\|u_{m}\right\|_{L^{1}(V)} \leq \frac{C}{\varepsilon^{n}}<\infty \end{aligned}

for m=1,2,…m=1,2, \ldots. Similarly

∣Dumε(x)∣≤∫B(x,ε)∣Dηε(x−y)∥um(y)∣dy≤∥Dηε∥L∞(Rn)∥um∥L1(V)≤Cεn+1<∞\begin{aligned} \left|D u_{m}^{\varepsilon}(x)\right| \leq \int_{B(x, \varepsilon)}\left|D \eta_{\varepsilon}(x-y) \| u_{m}(y)\right| d y \\ \leq\left\|D \eta_{\varepsilon}\right\|_{L^{\infty}\left(\mathbb{R}^{n}\right)}\left\|u_{m}\right\|_{L^{1}(V)} \leq \frac{C}{\varepsilon^{n+1}}<\infty \end{aligned}