next up previous
Next: Bibliography

Second assignment with solutions
was due September 18 in class
18.155 Fall 2001

Richard Melrose


Department of Mathematics, Massachusetts Institute of Technology

rbm@math.mit.edu

In the main these questions form theorems in Hörmander's book [1], so the proofs are available there. I suggest that you try to work them out on your own and in any case I expect written proofs, even if you need to get the idea from the book. Of course, at the very least, you will have to translate the notation.

Problem 1   [Hörmander, Theorem 3.1.4] Let $ I\subset\mathbb{R}$ be an open, non-empty interval.

  1. [i)] Show (you may use results from class) that there exists $ \psi\in\mathcal{C}_c(I)$ with $ \int_{\mathbb{R}}\psi(x)ds=1.$

    Proof. [Solution] We showed in class that there exists $ \phi\in\mathcal{C}_c(\mathbb{R})$ which is strictly positive on $ (0,1)$ and vanishes on $ \mathbb{R}\setminus(0,1).$ If $ I$ is a non-empty open interval it contains an interval $ [\bar x-\epsilon,\bar x+\epsilon]$ for some $ \bar x\in\mathbb{R}$ and $ \epsilon>0.$ Consider $ \psi(x)=C\phi((x-\bar x)/\epsilon)$ where $ C>0$ is to be chosen. The properties of $ \phi$ imply that $ \psi\in\mathcal{C}_c(I)$ and that $ \int_{\mathbb{R}}\psi=1/C,$ $ C>0.$ This value of $ C$ make the integral of $ \psi$ equal $ 1$ as desired. $ \qedsymbol$

  2. [ii)] Show that any $ \phi \in\mathcal{C}_c(I)$ may be written in the form

    $\displaystyle \phi=\tilde\phi+c\psi,\ c\in\mathbb{C},\ \tilde\phi \in\mathcal{C}_c(I)$ with $\displaystyle \int_{\mathbb{R}}\tilde\phi =0.$    

    Proof. [Solution] Set $ c=\int_{I}\phi$ and observe that $ \tilde\phi=\phi-c\psi$ has vanishing integral. $ \qedsymbol$

  3. [iii)] Show that if $ \tilde\phi\in\mathcal{C}_c(I)$ and $ \int_{\mathbb{R}}\tilde\phi =0$ then there exists $ \mu \in\mathcal{C}_c(I)$ such that $ \frac{d\mu }{dx}=\tilde\phi$ in $ I.$

    Proof. [Solution] If we consider $ \mu(x)=\int_{-\infty}^x\tilde\phi(s)ds$ then $ \mu$ is infinitely differentiable since $ d\mu(x)/dx=\tilde\phi(x)$ by the fundamental theorem of calculus. Certainly $ \mu(x)=0$ if $ x<\inf I',$ where $ I'$ is a compact interval outside which $ \tilde\phi$ vanishes. The assumption that $ \int_{I}\tilde\phi(x)dx=\int_{I'}\tilde\phi(x)dx=0$ means that $ \mu(x)=0$ for $ x>\sup I'$ so $ \mu\in\mathcal{C}_c(I).$ $ \qedsymbol$

  4. [iv)] Suppose $ u\in\mathcal{C}^{-\infty}(I)$ satisfies $ \frac{du}{dx}=0,$ i.e.

    $\displaystyle u(-\frac{d\phi}{dx})=0 \forall \phi \in\mathcal{C}_c(I),$ (1)

    show that $ u=c$ for some constant $ c.$

    Proof. [Solution] Using ii) and iii) we may write each element $ \phi \in\mathcal{C}_c(I)$ in the form $ \phi=c\psi+d\mu(x)/dx$ where $ \psi\in\mathcal{C}_c(I)$ is fixed and $ \mu \in\mathcal{C}_c(I)$ depends on $ \phi.$ Then the definition of the vanishing of the derivative in 1 shows that

    $\displaystyle u(\phi)=u(c\psi+d\mu/dx)=cu(\psi)-u(-d\mu/dx)=cu(\psi)=\int_IC\phi(x)dx$    

    where we have used the value of $ c$ and set $ C=u(\psi).$ The last integral is the definition of the (constant) function $ C$ as a distribution on $ I.$ Thus $ u=C$ is an equality between distributions. $ \qedsymbol$

  5. [v)] Suppose that $ u\in\mathcal{C}^{-\infty}(I)$ satisfies $ \frac{du}{dx}=c,$ for some constant $ c,$ show that $ u=cx+d$ for some $ d\in\mathbb{C}.$

    Proof. [Solution] Consider the continuous function $ cx.$ This defines a distribution, which we just denote the same way. The definition of the derivative means that $ d(cx)/dx=c$ (as you would expect). We proved this in class, it is the integration by parts formula

    $\displaystyle \frac{d(cx)}{dx}(\phi)=cx(-\frac{d\phi}{dx})=-\int_{\mathbb{R}}cx\frac{d\phi}{dx}dx =c(\phi).$    

    Now, if we consider the difference $ v=u-cx$ it is a distribution on $ I$ which satisfies $ dv/dx=0,$ so by iv) it is equal to a constant, $ d.$ That is $ u-cx=d,$ or $ u=cx+d.$ $ \qedsymbol$

Problem 2   [Hörmander Theorem 3.1.16]
  1. [i)] Use Taylor's formula to show that there is a fixed $ \psi
\in\mathcal{C}_c(\mathbb{R}^n)$ such that any $ \phi \in\mathcal{C}_c(\mathbb{R}^n)$ can be written in the form

    $\displaystyle \phi =c\psi +\sum\limits_{j=1}^nx_j\psi _j$ (2)

    where $ c\in\mathbb{C}$ and the $ \psi _j\in\mathcal{C}_c(\mathbb{R}^n)$ depend on $ \phi.$

    Proof. [Solution] By Taylor's formula, or theorem, any infinitely differentiable function which vanishes at 0 can be written in the form

    $\displaystyle \phi=\sum\limits_{j=1}^nx_ju_j$ (3)

    where the $ u_j$ are also infinitely differeniable. Choosing $ \psi\in\mathcal{C}_c(\mathbb{R}^n),$ as in class, with $ \psi(0)=1$ we can apply this to $ \phi-c\psi$ where $ c=\phi(0).$ The problem is that the $ u_j$ need not have compact support. However, we can choose another function (depending on $ \phi),$ $ \psi'\in\mathcal{C}_c(\mathbb{R}^n)$ which is equal to $ 1$ on the support of $ \phi$ and the support of $ \psi.$ Thus $ \phi\psi'=\phi$ and $ \psi\psi'=\psi.$ Then 3 gives the desired 2 where $ \psi_j-u_j\psi'\in\mathcal{C}_c(\mathbb{R}^n).$ $ \qedsymbol$

  2. [ii)]Recall that $ \delta _0$ is the distribution defined by

    $\displaystyle \delta _0(\phi)=\phi(0)\ \forall\ \phi \in\mathcal{C}_c(\mathbb{R}^n);$    

    explain why $ \delta _0\in\mathcal{C}^{-\infty}(\mathbb{R}^n).$

    Proof. [Solution] Certainly $ \delta _0$ so defined is a linear functional and it is continuous since

    $\displaystyle \vert\delta _0(\phi)\vert\le \sup_{\mathbb{R}^n}\vert\phi\vert=\Vert\phi\Vert _0.$    

    $ \qedsymbol$

  3. [iii)] Show that if $ u\in\mathcal{C}^{-\infty}(\mathbb{R}^n)$ and $ u(x_j\phi )=0$ for all $ \phi \in\mathcal{C}_c(\mathbb{R}^n)$ and $ j=1,\dots,n$ then $ u=c\delta _0$ for some $ c\in\mathbb{C}.$

    Proof. [Solution] We can use 2 to evaluate

    $\displaystyle u(\phi)=u(c\psi)+\sum\limits_{j=1}^nu(x_j\psi_j)=C\phi(0)$    

    where the given identity means that the sum is zero and we write $ C=u(\psi)$ for a fixed constant and note that $ c=\phi(0).$ This just means $ u=C\delta_0.$ $ \qedsymbol$

  4. [iv)] Define the `Heaviside function'

    $\displaystyle H(\phi )=\int_0^\infty\phi (x)dx\ \forall\ \phi \in\mathcal{C}_c(\mathbb{R});$    

    show that $ H\in\mathcal{C}^{-\infty}(\mathbb{R}).$

    Proof. [Solution] The linearity of the integral shows that $ H$ is a linear functional on $ \mathcal{C}_c(\mathbb{R}).$ The basic integral estimate

    $\displaystyle \vert H(\phi)\vert\le L\sup\vert\phi\vert$ if $\displaystyle \phi(x)=0$ in $\displaystyle \vert x\vert>L$    

    shows that it is continuous. $ \qedsymbol$

  5. [v)] Compute $ \frac{d}{dx}H\in\mathcal{C}^{-\infty}(\mathbb{R}).$

    Proof. [Solution] By definition

    $\displaystyle \frac{dH}{dx}(\phi)=-H(\frac{d\phi}{dx})=-\int_0^\infty\frac{d\phi}{dx}dx=\phi(0)$    

    where we have used the fundamental theorem of calculus. Thus

    $\displaystyle \frac{dH}{dx}=\delta_0.$ (4)

    $ \qedsymbol$

Problem 3   Using Problems 1 and 2, find all $ u\in\mathcal{C}^{-\infty}(\mathbb{R})$ satisfying the differential equation

$\displaystyle x\frac{du}{dx}=0$ in $\displaystyle \mathbb{R}.$    

Proof. [Solution] By definition $ v=\frac{du}{dx}$ is a distribution which satisfies $ xv=0.$ Problem 2, part iii), in one dimension shows that $ v=c\delta_0.$ Thus $ \frac{du}{dx}=c\delta_0.$ By Problem 2, part v), $ dH/dx=\delta.$ Thus $ \tilde u=u-cH$ satisfies $ d\tilde u/dx=0.$ By Problem 1, part iv), we conclude that $ v=d$ is constant. Thus the only distributional solutions of $ x\frac{du}{dx}=0$ are

$\displaystyle u=cH+d,\ c$ and $\displaystyle d$ constants.    

This means that $ u$ is constant in $ x<0$ and in $ x>0,$ with possibly different values, but it means more than this as regards its `behaviour at 0'. $ \qedsymbol$




next up previous
Next: Bibliography
Richard B. Melrose 2001-09-17