Instructions

This is a MIDTERM. You cannot ask for any external help and your work is strictly individual.

You may consult the the course textbooks (chapters relative to the contents covered), your personal notes, and HW solutions posted on the course website.

You may assume all statements we have covered in class and in the HW but cite their statement clearly when you use them.

All problems require proofs.

You may use integrals but they are not necessary.

You may use L'Hopital but be very careful with the hypotheses.

You hereby agree to follow the UVA Code of Conduct and abide by the Honor code.

You have 120 hours from the moment of download to do this Midterm, in any case you cannot turn in past <2020-11-26 Thu 11:00>. You can decide when download the midterm. Download and submission is through Gradescope. You MUST send a private Piazza post at the moment of download.

Please post any questions or concerns on Piazza in a private post.

1

Consider the function

\begin{equation*} f(x)=\left\{ \begin{aligned} &e^{-\frac{1}{x}} &&\text{if } x\gt 0\\ &0 &&\text{if } x\leq 0\\ \end{aligned}\right. \end{equation*}

Find all of its derivatives at \(0\). (Proof required)

1.1 Solution

We claim that \(f^{(n)}(0)=0\) for all \(n\in\N\).

The derivative at \(x_0\) is given by

\begin{equation*} \lim_{t\to 0} \frac{f(x_0+t)-f(x_0)}{t}. \end{equation*}

It is local (by that we mean that if \(f\) coincides with \(g\) on an open ball and \(g\) is differentiable at a point of that ball, then \(f\) is too and \(f'=g'\) at that point).

Let us compute the derivatives \(f^{n}(x)\) for \(x\gt 0\). Using the chain rule:

\begin{equation*} \begin{aligned}[t] & f(x)= e^{-1/x} \\ & f'(x)= x^{-2} e^{-1/x} \\ & f''(x)= (x^{-4}-2 x^{-3}) e^{-1/x} \end{aligned} \end{equation*}

We claim that for each \(n\in\N\) the \(n\)th derivative \(f^{(n)}(x)\) has the form

\begin{equation*} f^{n}(x)= P_n(1/x) e^{-1/x} \end{equation*}

where \(P_n\) is some polynomial. Let us show this by induction. It is true for \(n=0\) with \(P_0(t)=1\). Supposing that it is true for \(n\) we have

\begin{equation*} \begin{aligned}[t] f^{(n+1)}(x)&=\frac{d }{dx}f^{(n)}(x)=\frac{d }{dx}\Big( P_{n}(1/x)e^{-1/x}\Big) = \\ &= -(1/x^2) P'_n(1/x)e^{-1/x} + P_n(1/x) (1/x^2) e^{-1/x} =(P_n(1/x) - P'_n(1/x))(1/x^2) e^{-1/x} = P_{n+1}(1/x) e^{-1/x} \end{aligned} \end{equation*}

where \(P_{n+1}(t)= t^2(P_n- P_n'(t))\).

Let us show by induction that \(f^{(n)}(0)=0\). This is true for \(n=0\) and \(n=1\) (by HW).

Supposing that \(f^{(n)}(0)=0\) let us compute \(f^{(n+1)}(0)\):

\begin{equation*} \begin{aligned}[t] f^{(n+1)}(0)=\lim_{t\to 0} \frac{f^{(n)}(t)-f^{(n)}(0)}{t}=\lim_{t\to 0} \frac{f^{(n)}(t)}{t} \end{aligned} \end{equation*}

Let us split the limit into \(t\to 0^-\) and \(t\to 0^+\). For \(t\to 0^-\) this is \(0\) since \(f^{(n)}(t)=0\) if \(t\lt 0\) (by locality of derivatives and by the fact that for \(t\lt 0\) one has \(f(t)=0\)).

For \(t\gt 0\) we use the inductive claim about the form of \(f^{(n)}(t)\) to obtain

\begin{equation*} \begin{aligned}[t] \lim_{t\to 0^+} \frac{f^{(n)}(t)}{t} = \lim_{t\to 0^+} \frac{P_n(1/t) e^{-1/t}}{t} \end{aligned} \end{equation*}

Changing variables like in HW06P1.5 we compute

\begin{equation*} \begin{aligned}[t] \lim_{t\to 0^+} \frac{P_n(1/t) e^{-1/t}}{t}= \lim_{s\to +\infty} sP_n(s) e^{-s} =0 \end{aligned} \end{equation*}

The last identity follows by the fact that \(\lim_{s\to +\infty} \frac{sP_n(s)}{s^d}=0\) for any \(d\gt \mathrm{deg}(P_n)\) and \(\lim_{s\to +\infty}s^de^{-s}=0\) as per solution to the HW06P1.5.

2

Show that the sequence of functions

\begin{equation*} f_n(x)=2^{-n^2} x^n e^{-x^2} \end{equation*}

converge uniformly to \(0\).

2.1 Solution

We need to show that \(\lim_n \|f_n\|_\sup =0\). Since \(f\) is either even or odd (depending on \(n\)) we can only study it for \(x\geq 0\)

Let us show that \(f_n\) is bounded and find bounds for \(|f_n(x)|\).

We have (by solution of HW06P1.5 or L'Hopital) that for any fixed \(n\) we have. L

\begin{equation*} \lim_{x\to +\infty} f_n(x)=0 \end{equation*}

So by midterm01 the function above is non-negative. Also, being continuous on \([0,+\infty)\), \(f_n\) is bounded and attains max (since \(0\) cannot be its \(\sup\)). Since \(f_n\) is differentiable (being a composition and product of differentiable functions). The point of maximum \(x\) must satisfy

\begin{equation*} f'_n(x)=0 \end{equation*}

i.e.

\begin{equation*} n\,2^{-n^2}x^{n-1} e^{-x^2} - 2\,2^{-n^2} x^{n+1}e^{-x^2}=0 \end{equation*}

i.e.

\begin{equation*} x^{n-1}(n - 2 x^2) =0. \end{equation*}

The maximum is not at \(x=0\) where \(f(x)=0\) so it is at \(x=\sqrt{n/2}\). There the value of the function is

\begin{equation*} f_n(\sqrt{n/2}) = 2^{-n^2}(n/2)^{n/2} e^{-n/2}\leq e^{-\log(2)n^2 + (\log(n/2)-1) n/2} \end{equation*}

For the numerator we have

\begin{equation*} \lim_{n\to \infty}\Big(-\log(2)n^2 + (\log(n/2)-1) n/2\Big)=\lim_{n\to \infty}n^2\Big(-\log(2) + (\log(n/2)-1)/( 2n)\Big)=-\infty \end{equation*}

since \((\log(n/2)-1)/( 2n)\to 0\). To see this last part apply either change variables \(n= e^t\) to obtain:

\begin{equation*} \lim_{n\to +\infty}(\log(n/2)-1)/( 2n)=-\lim_{t\to +\infty} \frac{t-\log(2)-1}{e^t}=0 \end{equation*}

or apply L'Hopital to obtain

\begin{equation*} \lim_{n\to +\infty}(\log(n/2)-1)/( 2n)=\lim_{n\to +\infty}(1/n)/(2) = 0. \end{equation*}

Thus we have shown that

\begin{equation*} 0\leq \|f_n\|_{\sup}\leq f_n(\sqrt{n/2})\leq e^{-\log(2)n^2 + (\log(n/2)-1) n/2}\to 0 \end{equation*}

concluding our claim.

3

Let \(F\colon \R \to \R\) be \(C^2\) (continuous, \(F'\) continuous, and \(F''\) continuous). Consider the map \(T\) defined on \(C^0\big([0,1];\R\big)\) with values in \(C^0\big([0,1];\R\big)\) given by

\begin{equation*} \Big(T(f)\Big):=\Big(x \to F(f(x))\Big) \end{equation*}

3.1

Show that \(F\) has bounded first and second derivative on any set of the form \([-M;M]\).

3.1.1 Solution

\(F'\) and \(F''\) are continuous \([-M,M]\) is bounded and closed thus compact by Heine-Borel. Continuous functions on cpt sets are bounded.

3.2

Show that \(T\) is continuous as a map from \(\Big(C^0\big([0,1];\R\big),\, \|\cdot\|_{\sup}\Big)\) to \(\Big(C^0\big([0,1];\R\big),\, \|\cdot\|_{\sup}\Big)\).

3.2.1 Solution

First we argue that for any fixed \(f\) the function \(T(f)\) is continuous. But this is just because \(F\) is continuous and a composition of continuous functions is continuous.

Now we need to show that if \(T\colon C^0\big([0,1];\R\big)\to C^0\big([0,1];\R\big)\) is continuous. This means we must show that for any fixed \(f\in C^0\big([0,1];\R\big)\) and any sequence \(f_n\in C^0\big([0,1];\R\big)\) such that \(\|f_n-f\|_\sup \to 0\) we have \(\|T(f_n)-T(f)\|_\sup\to 0\). Let us write out \(T(f_n)-T(f)\):

\begin{equation*} \begin{aligned}[t] \Big|\Big(T(f_n)-T(f)\Big)(x)\Big| = |F(f_n(x))-F(f(x))| \leq\sup_{c\in\big[f(x),f_n(x)\big]} |F'(c)| |f_n(x)-f(x)| \end{aligned} \end{equation*}

where we used Lagrange. (the segment \(\big(f(x),f_n(x)\big)\) denotes the points between \(f(x)\) and \(f_n(x)\); it might be that \(f_n(x)\leq f(x)\)).

Eventually \(\|f_n-f\|_{\sup}\leq 1\) so \(|f_n(x)|\leq \|f\|_{\sup}+1\) and obviously \(|f(x)|\leq \|f\|_{\sup}+1\) so \(|c|\|f\|_{\sup}+1\). Let \(L=\sup_{[-\|f\|_{\sup}-1,+\|f\|_{\sup}+1]}|F'(c)|\lt \infty\) by the previous point so

\begin{equation*} \Big|\Big(T(f_n)-T(f)\Big)(x)\Big| \leq L |f_n(x)-f(x)| \end{equation*}

so by taking \(\sup\) on both sides

\begin{equation*} \Big\|T(f_n)-T(f)\Big\|_\sup \leq L \|f_n-f\|_\sup \end{equation*}

showing that if \(\|f_n-f\|_\sup\to 0\) then \(\|T(f_n)-T(f)\|_\sup\to 0\).

3.3

Show that \(T\) is continuous as a map from \(\Big(C^1\big([0,1];\R\big),\, \|\cdot\|_{C^1}\Big)\) to \(\Big(C^1\big([0,1];\R\big),\, \|\cdot\|_{C^1}\Big)\).

3.3.1 Solution

First we argue that for any fixed \(f\in C^1\) the function \(T(f)\) is \(C^1\). But this is just because \(F\) is and differentiable and a composition of differentiable functions is differentiable (chain rule).

Now we need to show that if \(T\colon C^1\big([0,1];\R\big)\to C^1\big([0,1];\R\big)\) is continuous. This means we must show that for any fixed \(f\in C^1\big([0,1];\R\big)\) and any sequence \(f_n\in C^1\big([0,1];\R\big)\) such that \(\|f_n-f\|_\sup+\|f_n'-f'\|_\sup \to 0\) we have \(\|T(f_n)-T(f)\|_\sup+\|T(f_n)'-T(f)'\|_\sup\to 0\). We have already shown that \(\|T(f_n)-T(f)\|_\sup\to 0\). It remains to show that \(\|T(f_n)'-T(f)'\|_\sup\to 0\). Let us write out \(T(f_n)'-T(f)'\):

\begin{equation*} \begin{aligned}[t] \Big|\Big(T(f_n)-T(f)\Big)'(x)\Big| &= |F'(f_n(x))f'_n(x)-F'(f(x))f'(x)| \leq \\ &|F'(f_n(x))\big(f'_n(x)-f'(x)\big)+ \big(F'(f_n(x))-F'(f(x))\big)f(x)| \\ &\leq|F'(f_n(x))\big(f'_n(x)-f'(x)\big)|+ |\big(F'(f_n(x))-F'(f(x))\big)f'(x)| \\ &\leq |F'(f_n(x))| |f_n'(x)-f'(x)| + \sup_{c\in\big[f(x),f_n(x)\big]} |F''(c)| |f_n(x)-f(x)| |f'(x)| \end{aligned} \end{equation*}

where we used Lagrange. (the segment \(\big(f(x),f_n(x)\big)\) denotes the points between \(f(x)\) and \(f_n(x)\); it might be that \(f_n(x)\leq f(x)\)).

Eventually \(\|f_n-f\|_{\sup}+\|f_n'-f'\|_{\sup}\leq 1\) so \(|f_n(x)|+|f_n'(x)|\leq \|f\|_{C^1}+1\). Let \(L=\sup_{[-\|f\|_{C^1}-1,+\|f\|_{C^1}+1]}\big(|F''(c)|+|F'(c)|\big)\lt \infty\) by the previous point so

\begin{equation*} \Big|\Big(T(f_n)-T(f)\Big)'(x)\Big| \leq L |f_n'(x)-f'(x)|+ L |f'(x)| |f_n(x)-f(x)| \leq L\|f_n-f\|_{C^1} + L \|f'\|_{C^1}\|f_n-f\|_{C^1} \end{equation*}

showing that if \(\|f_n-f\|_{C^1}\to 0\) then \(\|T(f_n)'-T(f)'\|_\sup\to 0\) as required.

3.4 Hints:

To solve the last two points you may assume initially that \(F'\) \(F''\) are bounded.

Then use the first point to remove that assumption.

4

Let \(f\colon(-1,1)\to \R\).

4.1

Suppose that \(f\) is differentiable at \(x_0\). Show that

\begin{equation*} \lim_{t\to 0} \frac{f(x_{0}+t)-f(x_{0}-t)}{2t} = f'(x_{0}) \end{equation*}

Careful, no information about the behavior of \(f\) outside of \(x_0\) is given, except for what follows from differentiability at \(x_0\).

4.1.1 Solution

Write

\begin{equation*} \frac{f(x_{0}+t)-f(x_{0}-t)}{2t}= \frac{f(x_{0}+t)-f(x_0)+f(x_0)-f(x_{0}-t)}{2t} = \frac{1}{2}\frac{f(x_{0}+t)-f(x_0)}{t} + \frac{1}{2}\frac{f(x_{0}-t)-f(x_0)}{-t} \end{equation*}

By definition of derivative we have

\begin{equation*} \begin{aligned}[t] & \lim_{t\to 0}\frac{f(x_{0}+t)-f(x_0)}{t} = f'(x_0) \\ & \lim_{t\to 0}\frac{f(x_{0}-t)-f(x_0)}{-t} = f'(x_0) \end{aligned} \end{equation*}

The claim follows by algebra of limits.

4.2

Suppose that \(f\) is twice differentiable at \(x_0\). Show that

\begin{equation*} \lim_{t\to 0} \frac{f(x_{0}+t)+f(x_{0}-t)-2f(x_{0})}{t^{2}} = f''(x_{0}) \end{equation*}

Careful, no information about the behavior of \(f\) outside of \(x_0\) is given, except for what follows from differentiability at \(x_0\).

4.2.1 Solution

Using Taylor to second order we have

\begin{equation*} f(x_0+t)= f(x_0)+ tf'(x_0) + \frac{t^2}{2} f''(x_0) + \mathrm{err}(t) \end{equation*}

with \(\lim_{t\to0}\frac{\mathrm{err}(t)}{t^2}=0\). So we have

\begin{equation*} \frac{f(x_{0}+t)+f(x_{0}-t)-2f(x_{0})}{t^{2}} = \frac{f(x_{0})+tf'(x_0)+\frac{t^2}{2}f''(x_0) + \mathbf{err}(t)+f(x_{0})-t f'(x_0)+\frac{t^2}{2}f''(x_0) + \mathbf{err}(-t)-2f(x_{0})}{t^{2}} \end{equation*}

Simplyfying and taking the limit

\begin{equation*} \lim_{t\to 0}\frac{f(x_{0}+t)+f(x_{0}-t)-2f(x_{0})}{t^{2}} =\lim_{t\to 0} \frac{t^2 f''(x_0) + \mathbf{err}(t)+ \mathbf{err}(-t)}{t^{2}} = f''(x_0) \end{equation*}

as required.

4.3

Let \(p_{1},p_{2},p_{3},p_{4}\) be four numbers in \(\R\). Show that a sufficient and necessary condition for

\begin{equation*} \lim_{t\to 0} \frac{\sum_{i=1}^{4}a_{i}f(x_{0}+p_{i}t)}{t^{3}}= f'''(x_{0}) \end{equation*}

for any \(f\colon(-1,1)\to \R\) three times differentiable at \(x_{0}\in(-1,1)\) is that

\begin{equation*} \begin{aligned} &\sum_{i=1}^{4}a_{i}p_{i}^{k}=0 \qquad \forall k\in\{0,1,2\} \\ & \sum_{i=1}^{4}a_{i}p_{i}^{3}=3! \end{aligned} \end{equation*}

Careful, no information about the behavior of \(f\) outside of \(x_0\) is given, except for what follows from differentiability at \(x_0\).

Hint: To check necessity test against \(f\) that are polynomials. To show that it is sufficient, Taylor expand.

4.3.1 Solution

Sufficiency

Taylor expand \(f\) to \(3\)-rd order:

\begin{equation*} f(x_0+p t) = f(x_0) + pt f'(x_0) + p^2\frac{t^2}{2} f''(x_0) + p^3 \frac{t^3}{3!} f'''(x_0) + \mathbf{err}(pt) \end{equation*}

with \(\lim_{s\to 0}\frac{\mathbf{err}(s)}{s^3} =0\). Summing up we have

\begin{equation*} \sum_{i=1}^{4} a_i f_i(x_0+p_i t) =\sum_{i=1}^{4}\Big( a_if(x_0) + a_ip_it f'(x_0) + a_ip_i^2\frac{t^2}{2} f''(x_0) + a_ip_i^3 \frac{t^3}{3!} f'''(x_0) + a_i\mathbf{err}(p_it)\Big) \end{equation*}

and using the assumptions we get

\begin{equation*} \sum_{i=1}^{4} a_i f_i(x_0+p_i t)=t^3 f'''(x_0) + \sum_{i=1}^4 a_i\mathbf{err}(p_it) \end{equation*}

Plugging this into the limit we need to find we get

\begin{equation*} \lim_{t\to 0}\Big(\frac{\sum_{i=1}^{4} a_i f_i(x_0+p_i t)}{t^3}\Big) = f'''(x_0) + \lim_{t\to 0}\Big( \sum_{i=1}^4 a_i\frac{\mathbf{err}(p_it)}{t^3}\Big) = f'''(x_0) \end{equation*}

where the last limit follows from the error estimate.

Necessity

Let us take \(f(x)=x^k\) for \(k\in\{0,1,2\}\). We get that \(f'''(0)=0\) so we must have

\begin{equation*} \lim_{t\to 0}\frac{\sum_{i=1}^{4} a_ip_i^k t^k}{t^3}=\lim_{t\to 0}t^{k-3}\sum_{i=1}^{4} a_ip_i^k =0 \end{equation*}

and since \(k-3\leq -1\) we have that \(\lim_{t\to 0}|t^{k-3}|=+\infty\) so the above can hold only if

\begin{equation*} \sum_{i=1}^{4} a_ip_i^k=0 \end{equation*}

Finally. Testing against \(f(x)=x^3\) we get \(f'''(0)=3!\) and

\begin{equation*} \lim_{t\to 0}\frac{\sum_{i=1}^{4} a_ip_i^k t^3}{t^3}=\lim_{t\to 0}\sum_{i=1}^{4} a_ip_i^k =\sum_{i=1}^{4} a_ip_i^k \end{equation*}

obtaining the last condition.

5

Consider the function \(F\colon l^\infty_c(\N;\R)\to l^\infty(\N;\R)\) given by

\begin{equation*} F(\vec{x})_k= e^{-x_k^2}. \end{equation*}

Find a continuous extension of \(F\) to \(l^\infty_0(\N;\R)\) and show that it is unique.

Hint: Show that the derivative of \(e^{-x^2}\) is bounded.