Lecture 33 - Supplemental Material, Weirstrass Function

We talked about:

L'Hopital's Rule: 00 case

Let f,g be continuous on an interval containing a and suppose f,g are differentiable on this interval with the possible exception of the point a. If limxaf(x)=limxag(x)=0 and g(x)0 for all xa, then:
limxaf(x)g(x)=LRlimxaf(x)g(x)=L

Proof

Apply the Generalized Mean Value Theorem via the following method. For argument let (b,c) be our interval.

  1. Consider the case that a=b. Let ε>0. Supposing limxaf(x)g(x)=L, choose δ>0 such that 0<|xb|<δ implies that |f(x)g(x)L|<ε2. Let x(b,b+δ). Let y(b,x). By the Generalized Mean Value Theorem, then t(y,x) such that:
    f(t)(g(x)g(y))=g(t)(f(x)f(y))
    and because g(x)g(y) and g(t)0:
    f(t)g(t)=f(x)f(y)g(x)g(y)
    Now since b<t<b+δ then our implication we started with gives:
    Lε2<f(x)f(y)g(x)g(y)<L+ε2
    for all y(a,x). Let yb+ then by the Algebraic Limit Theorem for Functional Limits and Limits and Order (Order Limit Theorem) then taking a limit and using limxaf(x)=0=limxag(x) then:
    Lε<f(x)g(x)<L+ε
  2. Doing the cases for a=c and ca,b are done very similarly.

L'Hopital's Rule: case

Suppose f,g are differentiable on (a,b) and that g(x)0 for all x(a,b). If limxag(x)= (defined here) (or ), then:
limxaf(x)g(x)=Llimxaf(x)g(x)=L

Notice that we only need the denominator to go to in this case. This is different than the 00 case.

Proof

Let ε>0. Because limxaf(x)g(x)=L then δ1>0 such that:
|f(x)g(x)L|<ε2
for all a<x<a+δ1. For convenience of notation, let t=a+δ1 and note that t is fixed for the remainder of the argument.

Our functions are not defined at a, but for any x(a,t) we can apply the Generalized Mean Value Theorem on the interval [x,t] to get:
f(x)f(t)g(x)g(t)=f(c)g(c)
for some c(x,t). Our choice of t then implies that:
Lε2<f(x)f(t)g(x)g(t)<L+ε2
for all x(a,t).

In an effort to isolate the fraction f(x)g(x), the strategy is to multiply the inequality by g(x)g(t)g(x). We need to be sure, however, that this quantity is positive, which amounts to insisting that 1g(t)g(x). Now, because t is fixed and limxag(x)= then we can choose δ2>0 so that g(x)g(t) for all a<x<a+δ2 (because t is fixed, so eventually we pass g(t) as g(x)). Carrying out the desired multiplication results in:
(Lε2)(1g(t)g(x))<f(x)f(t)g(x)<(L+ε2)(1g(t)g(x))
which after some manipulation becomes:
Lε2+Lg(t)+ε2g(t)+f(t)g(x)<f(x)g(x)<L+ε2+Lg(t)ε2g(t)+f(t)g(x)
Notice the $g(x)

Weirstrass Function

Let's construct the function that is everywhere continuous but nowhere differentiable.

One function (we don't actually cover this) defined the function:
n=0ancos(bnx)
where a,bR are constants. Specifically there were more restrictions on a,b not discussed here. This is one function that would have the property of being everywhere continuous but nowhere differentiable, but the cos here is making such high oscillations that it makes the derivative not exist.

A different function that (only starts to) take on this role is:
ϕ(x)={|x|1x1
Where we extend this function to all of R by requiring that ϕ(x+2)=ϕ(x).

Now this function is differentiable at any xN while xR, but the idea is to try to fill in the gaps. We'll consider the function of the form:
g(x)=n=0anϕ(bnx)
where a,b have the similar restraints (ab>1 and a<1). We'll specifically want to use:
f(x)=n=0(34)nϕ(4nx)
We'll show that f is continuous, and then next that f is differentiable nowhere.

Continuity

f(x) above is defined for all of xR. Namely we could compare our series defined via f converges since it's less than the geometric series:
n=0(34)n
which itself is a convergent geometric series.

For continuity at every xR, notice that NN that:
f(x)=SN(x)+RN(x)
where:
SN(x)=n=0N(34)nϕ(4nx)
is clearly continuous, and:
RN(x)=n=N+1(34)nϕ(4nx)n=N+1(34)n=(34)N+1134=3(34)N
So thus 0RN(x)3(34)N.

So the argument is the following:

Proof

Let xR and let ε>0. Choose some NN such that 3(34)N<ε2. Using the argument above, then:
f(x)=SN(x)+RN(x)
where SN(x) is continuous. Thus then by definition then choose δ>0 such that |xy|<δ implies |SN(x)SN(y)|<ε2. Now let y be such that |xy|<δ. Then:
|f(x)f(y)|=|SN(x)+RN(x)SN(y)RN(y)||SN(x)SN(y)|+|RN(x)RN(y)|<ε2+ε2=ε
Thus f is uniformly continuous.

Not Differentiable

Notice that for any x,yR then:
|ϕ(x)ϕ(y)||xy|

with equality only when an integer between x,y. Let's prove the failure for differentiability at any xR.

Proof

Let xR. and fix mN.

What may seem weird, let δm=±1214m, where:

  • Use + if an integer between 4mx and 4mx+12
  • Use if an integer between 4mx12 and 4mx.
    Thus using our δm then 4m(x+δm) and 4mx have no integers between them.

Now we'll show that f(x+δm)f(x)δm has no limit as m (which makes δm0). This is the h-definition of a derivative. Doing this:
f(x+δm)f(x)δm=1δm(n=0(34)nϕ(4n(x+δm))n=0(34)nϕ(4nx))=n=0(34)nϕ(4n(x+δm))ϕ(4nx)δm
Now notice that the term in the box is some γn so making that replacement:
=n=0(34)nγn
We have the following cases:

  1. Say n>m. Then into ϕ we get:
    4n(x+δm)=4nx±124nm
    now since n>m then the 4nm has at least one factor of 4, and the 4n also has a factor of 4, so the whole term is an even integer. Since ϕ has period 2, then:
    ϕ(4n(x+δm))ϕ(4nx)=0
    Thus for n>m then γn=0. Thus the infinite series becomes a finite series.
  2. Say n=m. Then notice:
    4n(x+δm)=4m(x+δm)4nx=4mx
    Now there cannot be an integer between these values:
    ϕ(4m(x+δm))ϕ(4mx)=±4mδmγn=±4m
  3. Say 0n<m. We won't get the exact value but we will approximate γn. Notice that using what we found at the start of the differentiability section:
    |ϕ(4n(x+δm))ϕ(4nx)||4n(x+δm)4nx|4n|δm|
    so then |γn|4n.

Now:
f(x+δm)f(x)δm=n=0(34)nγn=n=0m(34)nγn(γn=0n>m)=(34)mγm+n=0m1(34)nγn
Let the left value be A and for the right denote it Bn (the n is chosen to say that it is a sum for a sequence, not that it depends on n). Notice:
|A|=|(34mγm)|=|(34)m(±4m)|=3m
|B|=|n=0m1(34)nγn|n=0m1(34)n|γn|n=0m13n=3m131123m
Therefore:
|f(x+δm)f(x)δm|=|A+Bn|||A||Bn||3m123m=123m
so the limit doesn't exist.

s in the denominator. Again, because $t$ is fixed then $\lim_{ x \to a } g(x) = \infty$. Thus, we can choose a $\delta_{3}$ such that $x \in (a,a + \delta_{3})$ implies that $g(x)$ is large enough to ensure that the terms: $ \frac{-Lg(t) + \frac{\varepsilon}{2}g(t) + f(t)}{g(x)}, \frac{-Lg(t) - \frac{\varepsilon}{2}g(t) + f(t)}{g(x)} $ are less than $\frac{\varepsilon}{2}$ in absolute value (because the numerators are fixed, and $g(x) \to \infty$), so using that in our original inequality and choosing $\delta = \min\{ \delta_{1}, \delta_{2}, \delta_{3} \}$ guarantees that: $ L - \frac{\varepsilon}{2} + \frac{\varepsilon}{2} \lt \frac{f(x)}{g(x)} \lt L + \frac{\varepsilon}{2} + \frac{\varepsilon}{2} $ Thus: $ \left| \frac{f(x)}{g(x)} - L \right| \lt \varepsilon $ for all $a \lt x \lt a + \delta$.

Weirstrass Function

Let's construct the function that is everywhere continuous but nowhere differentiable.

!Wierestrass Function