:point_right: Here is the fourth set of problems for you.


Problem 1. Consider the heat equation ut=uxxu_{t} = u_{xx} in R×R\mathbb{R}\times\mathbb{R}, with an initial condition

u(x,0)=11+x2,xR.\displaystyle u(x,0) = \frac 1{1+x^2}\, , \qquad x\in \mathbb{R}.

Prove that this problem has no analytic solution (i.e., the Taylor series of a solution converges only at (0,0)(0,0).)

(This example is due to Kowalewska.)



If you solve all the following problems, you will know that Δu=f(u)-\Delta u = f(u) has a nontrivial solution in W01,2(Ω)W^{1,2}_0(\Omega) for a large class of nonlinear ff’s.

Here are the assumptions and notation:

(1) ff is C1C^1 on R\mathbb{R} and satisfies two growth conditions: f(z)C(1+zp)\vert f(z) \vert \le C(1+\vert z\vert^p) and f(z)C(1+zp1)\vert f'(z) \vert \le C(1+\vert z\vert^{p-1}) for some 1<p<(n+2)/(n2)% <![CDATA[ 1 < p < (n+2)/(n-2) %]]>;

(2) FF is the primitive function of ff such that F(0)=0F(0)=0; we also assume that for some γ<1/2% <![CDATA[ \gamma < 1/2 %]]> the following holds:

0F(z)γzf(z),zR,\displaystyle 0 \le F(z) \le \gamma z f(z), \qquad z\in \mathbb{R},

and azp+1F(z)Azp+1a\vert z\vert^{p+1}\le \vert F(z)\vert \le A \vert z\vert ^{p+1} for some Aa>0A\ge a> 0;

(3) HH stands for W01,2(Ω)W^{1,2}_0(\Omega) with the scalar product

(u,v)=ΩDuDvdx,\displaystyle (u,v)=\int_{\Omega} Du\cdot Dv\, dx\, ,

and ΩRn\Omega \subset\mathbb{R}^n is bounded and smooth;

(4) Finally, we define I ⁣:HRI\colon H\to \mathbb{R} by

I(u)=12ΩDu2dxΩF(u)dx=:I1(u)I2(u).\displaystyle I(u) = \frac 12 \int_{\Omega} \vert Du\vert ^2 dx - \int_{\Omega} F(u)\, dx =: I_{1}(u) - I_{2}(u)\, .


Now, the problems to solve. (You may wish to consult the proof of Theorem 3 in Evans, Section 8.5.2, as you go through them.)


Problem 2. Prove that for uHu\in H we have f(u)Lqf(u)\in L^q for q=2n/(n+2)q=2n/(n+2); hence, II is well defined on HH. (Use Sobolev imbedding theorem).

Problem 3. Write H1(Ω)H^{-1}(\Omega) to denote the space of continuous linear functionals on HH. Prove that, for each uHu\in H, the function f(u)f(u) defines a functional ϕH1(Ω)\phi\in H^{-1}(\Omega) via

ϕ(v)=Ωf(u)vdx,vH.\displaystyle \phi (v) = \int_{\Omega} f(u)v\, dx\, , \qquad v \in H\, .

(A comment: people, including the owner of this page, often abuse notation and write things like f(u)H1(Ω)f(u)\in H^{-1}(\Omega) or L2n/(n+2)H1L^{2n/(n+2)}\subset H^{-1} etc.)

Problem 4. Prove that for each wH1(Ω)w\in H^{-1}(\Omega) the equation Δv=w-\Delta v = w has a unique weak solution vW01,2(Ω)-v\in W^{1,2}_0(\Omega). (Hint: recall Lax–Milgram; Evans, Section 6.2.1.)

Check that the map

K ⁣:H1(Ω)wvW01,2(Ω)\displaystyle K\colon H^{-1}(\Omega) \ni w \longmapsto v \in W^{1,2}_0(\Omega)

is an isometry. (Do not be afraid: this is true practically by definition, no hard computations required.)

Problem 5. Use the definition of the derivative (of a function on a Banach space) and check that I1(u)=uI_{1}'(u)=u and I2(u)=K(f(u))I_{2}'(u)=K(f(u)) for all uHu\in H. Further, use the growth condtions for ff to check that IkI_k' is Lipschitz on bounded subsets of HH, k=1,2k=1,2.

Problem 6. Use assumptions on FF and Rellich-Kondrashov to prove that II satisfies the Palais–Smale condition on HH:

(PS) if the sequence I(uk)I(u_k) is bounded and I(uk)0I'(u_k)\to 0, then a subsequence of uku_k converges to some uHu\in H.

(Hint: Rellich-Kondrashov allows to identify the limit uu of uku_k; then, one has to rely on the results of Problems 4-5.)

Problem 7. Check that I(u)I(u) is positive on small spheres S(0,r)H\mathbb{S}(0,r)\subset H (use Sobolev imbedding again) and negative for u>R\| u \|> R for large RR (for this, consider the behaviour of II on rays: analyze g(t)=I(tu)g(t)=I(tu) for t>0t>0 large).


Once you do all that, use Mountain Pass Theorem to conclude immediately that Δu=f(u)-\Delta u = f(u) has a nonzero solution with zero Dirichlet boundary conditions.


:christmas_tree: :gift: :sparkles: