Differentiability

Elementary calculus books normally state that a function $f$ is differentiable at $a$ if $f’(a)$ exists, which means the limit

\[f'(a) = \lim_{h \to 0} {f(a + h) - f(a) \over h}\]

exists, and $f’(a)$ is called the derivative of $f$ at $a$.

There is another notion of derivative which is very useful and leads to proper notion of differentiability for functions of several variables.

Differentiability in One Variable

A function $f: \mathbb{R} \to \mathbb{R}$ is differentiable at $x = a$ if it is approximately linear near $x = a$. Geometrically, this means that the graph of $f$ has a tangent line at $x = a$. Analytically, this means there is a linear function $l(x) = mx + b$ satisfying the following two conditions:

\[l(x) = f(a) + m(x - a)\] \[{f(x) - l(x) \over x - a} \to 0 \text{ as } x \to a\]

Denoting the increment $x - a$ by $h$, we have

\[f(x) - l(x) = f(a + h) - (f(a) + mh)\]

We think of this difference as a function of $h$ and denote it by $E(h)$, which represents the error when we approximate $f(a + h)$ by the linear function $f(a) + mh$.

Definition. Suppose $f$ is a real-valued function defined on some open interval in $\mathbb{R}$ containing the point $a$. It is differentiable at $a$ if these is a number $m$ such that

\[\begin{equation} \tag{1} \label{eq1} f(a + h) = f(a) + mh + E(h), \quad \text{where } \quad \lim_{h \to 0} {E(h) \over h} = 0 \end{equation}\]

In other words, $f$ is differentiable at $a$ if we can represent $f(a + h)$ as the sum of the linear function $f(a) + mh$ and an error term $E(h)$ that tends to zero more rapidly than $h$ as $h \to 0$. In this case, we have

\[m = {f(a + h) - f(a) \over h} - {E(h) \over h}\]

When $h \to 0$, the last term vanishes, so

\[m = \lim_{h \to 0} {f(a + h) - f(a) \over h}\]

Thus, the number $m$ is uniquely determinted and it is called the derivative of $f$ at $a$, denoted by $f’(a)$.

Coversely, if the limit $m$ exists, we set $E(h) = f(a + h) - f(a) - mh$ and

\[\lim_{h \to 0} {E(h) \over h} = \lim_{h \to 0} {f(a + h) - f(a) \over h} - m = m - m = 0\]

so $\eqref{eq1}$ holds and it is like the usual definition.

Also, as $E(h) / h$ vanishes as $h \to 0$, so as $E(h)$ itself and hence so does $f(a + h) - f(a)$. It means differentiability at $a$ implies continuity at $a$.

Little-o Notation

Definition. $f(h) = o(g(h))$ as $h \to 0$ if

\[\lim_{h \to 0} {f(h) \over g(h)} = 0\]

Property.

  • If $f(h) = o(h^a)$ and $g(h) = o(h^a)$ then $f(h) + g(h) = o(h^a)$

  • If $f(h) = o(h^a)$ and $g(h) = o(h^b)$ then $f(h)g(h) = o(h^{a+b})$

  • If $f(h) = o(h)$ then $cf(h) = o(h)$

  • If $f_1(h) = o(f_2(h))$ and $f_2(h) = o(f_3(h))$ then $f_1(h) = o(f_3(h))$

  • If $f(h) = o(h^a)$ where $a \in \mathbb{N}$ then $f(h) = o(h)$

Base on the definition, we can see that $E(h) = o(h)$ and we have the following theorem:

Theorem. If $f$ is differentiable at $x = a$, then

\[f(a + h) = f(a) + f'(a)h + o(h)\]

Conversely, if there exist constants $A$ and $B$ such that

\[f(a + h) = A + Bh + o(h)\]

then $f$ is differentiable at $a$, with $A = f(a)$ and $B = f’(a)$.

This theorem means that the tangent line is a good linear approximation to $f$ and the tangent line is the only line that gives such a good approximation.

This notation is handy for deriving basic laws of differentiation.

Theorem. [Product Rule] If $f$ and $g$ are differentiable at $x = a$, then $fg$ is differentiable and the derivative is $f(a)g’(a) + f’(a)g(a)$.

Proof.

Given

\[f(a + h) = f(a) + f'(a)h + o(h) \quad \text{and} \quad g(a + h) = g(a) + g'(a)h + o(h)\]

Therefore,

\[\begin{align*} f(a + h)g(a + h) &= [f(a) + f'(a)h + o(h)][g(a) + g'(a)h + o(h)] \\ &= f(a)g(a) + [f(a)g'(a) + f'(a)g(a)]h \\ &\quad \quad + \left\{ [f(a) + f'(a)h]o(h) + [g(a) + g'(a)h]o(h) + o(h^2) \right\} \\ &= f(a)g(a) + [f(a)g'(a) + f'(a)g(a)]h + o(h) \end{align*}\]

Hence, $fg$ is differentiable at $x = a$ and its derivative is $f(a)g’(a) + f’(a)g(a)$.

Theorem. [Chain Rule] If $f$ is differentiable at $x = a$ and $g$ is differentiable at $x = f(a)$, then $g \circ f$ is differentiable at $x = a$ and its derivative is $g’(f(a))f’(a)$

Proof.

Given

\[f(a + h) = f(a) + f'(a)h + o(h) \quad \text{and} \quad g(f(a) + k) = g(f(a)) + g'(f(a))k + o(k)\]

Substitute $k = f(a + h) - f(a)$ in the second formula,

\[g(f(a + h)) = g(f(a)) + g'(f(a))[f(a + h) - f(a)] + o(f(a + h) - f(a))\]

Then substitute $f(a + h) - f(a) = f’(a)h + o(h)$,

\[\begin{align*} g(f(a + h)) &= g(f(a)) + g'(f(a))[f'(a)h + o(h)] + o(f'(a)h + o(h)) \\ &= g(f(a)) + [g'(f(a))f'(a)]h + o(h) \\ \end{align*}\]

Hence, $g \circ f$ is differentiable at $x = a$ and its derivative is $g’(f(a))f’(a)$.

Leibniz’s Rule

Theorem. [Leibniz’s Rule] If $f = uv$, then

\[f^{(n)}(x) = \sum_{r=0}^n {n \choose r} u^{(r)} v^{(n-r)}\]

Proof.

When $n = 2$, it is true by the Product Rule.

Assume it is true when $n = k$. When $n = k+1$,

\[\begin{align*} f^{(k+1)}(x) &= \sum_{r=0}^k {k \choose r} (u^{(r)}v^{(k-r)})' \\ &= \sum_{r=0}^k {k \choose r} (u^{(r+1)}v^{(k-r)} + u^{(r)}v^{(k-r+1)}) \\ &= u^{(0)}v^{(k+1)} + \sum_{r=1}^{k} \left({k \choose r-1} + {k \choose r}\right) u^{(r)}v^{(k+1-r)} + u^{(k+1)}v^{(0)} \\ &= \sum_{r=0}^{k+1} {k+1 \choose r} u^{(r)}v^{(k+1-r)} \end{align*}\]

By induction, Leibniz’s Rule is true for all positive integers $n$.

References