Composition

Let $f: B \rightarrow C$ be a continuous function. This means that $\lim_{z \rightarrow b} f(z) = f(b)$ for all $b \in B$. Therefore

\[\forall \epsilon_f > 0 \quad \exists \delta_f > 0 \quad \text{s.t.} \quad 0 < \lvert z - b \rvert < \delta_f \implies \lvert f(z) - f(b) \rvert < \epsilon_f\]

Let $g: A \rightarrow B$ be any function (not necessarily continuous) whose limit exists at $a \in A$. Let $M \in B$ such that $\lim_{x \rightarrow a} g(x) = M$. Therefore

\[\forall \epsilon_g > 0 \quad \exists \delta_g > 0 \quad \text{s.t.} \quad 0 < \lvert x - a \rvert < \delta_g \implies \lvert g(x) - M \rvert < \epsilon_g\]


We wish to show that

\[\lim_{x \rightarrow a} f(g(x)) = f \left ( M \right )\]


Fix any $\epsilon > 0$. In the limit definition of $f(z)$ we fix a particular $\epsilon_f = \epsilon$. Also, choose the particular $b = M$. Therefore, we have

\[\exists \delta_f > 0 \quad \text{s.t.} \quad 0 < \lvert z - M \rvert < \delta_f \implies \lvert f(z) - f(M) \rvert < \epsilon\]

Now, in the limit definition of $g(x)$ we fix a particular $\epsilon_g = \delta_f$. Therefore, we have

\[\exists \delta_g > 0 \quad \text{s.t.} \quad 0 < \lvert x - a \rvert < \delta_g \implies \lvert g(x) - M \rvert < \delta_f\]

Now, let $z = g(x)$ and let $\delta = \delta_g$. Therefore

\[\exists \delta > 0 \quad \text{s.t.} \quad 0 < \lvert x - a \rvert < \delta \implies \lvert g(x) - M \rvert < \delta_f \implies \lvert z - M \rvert < \delta_f \implies \lvert f(z) - f(M) \rvert < \epsilon\]

which proves the result.


Written in a more general notation, we have the composition law of limits

\[\lim_{x \rightarrow a} f(g(x)) = f \left ( \lim_{x \rightarrow a} g(x) \right )\]


Counter-Example

The composition law of limits does not hold if $f$ is not continuous. To show this, we provide a counter-example.


Suppose   \(f(x) = \begin{cases} 0 &\quad\text{if } x \leq 0 \\ 1 &\quad\text{if } x > 0 \end{cases}\)    is the unit step function, and $g(x) = x$. Then, $f(g(x)) = f(x)$


The composition law would suggest that $\displaystyle \lim_{x \rightarrow 0} f(x) = \lim_{x \rightarrow 0} f(g(x)) = f \left ( \lim_{x \rightarrow 0} g(x) \right ) = f \left ( \lim_{x \rightarrow 0} x \right ) = f(0) = 0$. However, using the definition of $\lim_{x \rightarrow 0} f(x) = 0$, we can show this leads to a contradiction.

\[\begin{align} &\forall \epsilon > 0 \quad \exists \delta > 0 \quad \text{s.t.} \quad 0 < \lvert x - 0 \rvert < \delta \implies \lvert f(x) - 0 \rvert < \epsilon \\[10pt] &\forall \epsilon > 0 \quad \exists \delta > 0 \quad \text{s.t.} \quad 0 < \lvert x \rvert < \delta \implies \lvert f(x) \rvert < \epsilon \end{align}\]

If $\epsilon = 1$ and $0 < x < \delta$, then $\lvert f(x) \rvert = \lvert 1 \rvert = 1 \not< 1 = \epsilon$.


The issue here is that the left-handed limit and the right-handed limit do not equal. In particular

\[\lim_{x \rightarrow 0^{-}} f(x) = 0 \quad\qquad\text{and}\qquad\quad \lim_{x \rightarrow 0^{+}} f(x) = 1\]

This issue does not happen if the function is continuous.

Limits and Continuity Series