Proof Strategy for L'Hopital's Rule

Proof Strategy: Control the Intermediate Point

L’Hopital’s Rule connects a quotient of function values with a quotient of derivatives, but the connection is made at an intermediate point. A sound proof must explain why information about the derivative quotient applies at that point. Near a finite endpoint, the intermediate point is trapped between the endpoint and the variable. At infinity, merely being beyond a fixed anchor is not enough: the anchor itself must be chosen in response to the desired error tolerance.

The previous tutorial established the Intermediate Derivative Quotient proposition and the finite-point and infinity-over-infinity forms of L’Hopital’s Rule. We will use those results rather than re-proving them. The aim here is to make two proof techniques explicit: turning a uniform estimate on derivative quotients into an estimate on function quotients, and using an epsilon-dependent anchor to control a tail argument. These techniques also provide a practical way to audit an application of the rule.

A Quantitative Estimate Near a Finite Endpoint

The finite-point argument works because each intermediate point lies between $a$ and $x$. Thus, as $x$ approaches $a$ from the right, every such intermediate point is forced to approach $a$ too. The following estimate makes the implication precise.

Proposition (Finite-Endpoint Quotient Estimate): Suppose $f$ and $g$ are continuous on $[a,a+\delta]$, differentiable on $(a,a+\delta)$, and $f(a)=g(a)=0$. Suppose $g'(t)\ne0$ for every $t\in(a,a+\delta)$. If, for some $L\in\mathbb{R}$ and $\varepsilon>0$, $$ \left|\frac{f'(t)}{g'(t)}-L\right|\leq\varepsilon $$ for every $t\in(a,a+\delta)$, then $g(x)\ne0$ and $$ \left|\frac{f(x)}{g(x)}-L\right|\leq\varepsilon $$ for every $x\in(a,a+\delta]$.

Proof. Fix $x\in(a,a+\delta]$. Apply the Intermediate Derivative Quotient proposition on $[a,x]$. It gives $g(x)\ne0$ and a point $c_x\in(a,x)$ such that

\frac{f(x)}{g(x)}=\frac{f'(c_x)}{g'(c_x)}.

Since $c_x\in(a,a+\delta)$, the assumed derivative-quotient estimate applies at $c_x$. Substitution into the displayed equality gives

\left|\frac{f(x)}{g(x)}-L\right| = \left|\frac{f'(c_x)}{g'(c_x)}-L\right| \leq\varepsilon.

This proves both claims. $\square$

For a limit proof, one usually does not have a single fixed $\varepsilon$ that works on the whole interval. Instead, given any desired error $\varepsilon>0$, convergence of $f'/g'$ to $L$ provides an interval on which the estimate holds. The proposition then transfers that estimate to $f/g$. This is the finite-endpoint proof strategy in quantitative form: the intermediate point cannot escape the region where the derivative estimate is valid.

Worked Example: Turning a Derivative Bound into a Quotient Bound

Consider the limit as $x\to0^+$ of

\frac{x^3+2x^4}{x^2}.

Set $f(x)=x^3+2x^4$ and $g(x)=x^2$, with $f(0)=g(0)=0$. For $x>0$, their derivatives are

f'(x)=3x^2+8x^3,\qquad g'(x)=2x, \qquad \frac{f'(x)}{g'(x)}=\frac{3}{2}x+4x^2.

Here $g'(x)\ne0$ for every $x>0$, and the derivative quotient tends to zero. More explicitly, if $0<x<1/10$, then

\left|\frac{f'(x)}{g'(x)}\right| =\frac{3}{2}x+4x^2 <\frac{3}{20}+\frac{4}{100} =\frac{19}{100}.

The Finite-Endpoint Quotient Estimate with $L=0$ and $\varepsilon=19/100$ therefore bounds the original quotient by $19/100$ throughout this interval. Since the derivative quotient itself tends to zero, the same estimate can be made arbitrarily small by taking a sufficiently short interval. Hence the original quotient tends to zero. Direct simplification confirms the result:

\frac{x^3+2x^4}{x^2}=x+2x^2\longrightarrow0.

The important feature of the proof is not the simplification. It is that the derivative estimate is valid at every possible intermediate point.

The Anchor Must Depend on the Error Tolerance

At infinity, an application of the Cauchy Mean Value Theorem on $[T,x]$ gives a point $c\in(T,x)$. This tells us that $c$ lies beyond $T$, but it does not tell us that $c\to\infty$ as $x\to\infty$. The proof must not rely on that unsupported inference.

Instead, fix an error tolerance first. Choose the anchor $T$ so far out that the derivative quotient is close to its limit at every point beyond $T$. Then any intermediate point produced on $[T,x]$ inherits that closeness simply because $c>T$. The anchor is allowed to depend on the tolerance; that dependence is exactly what makes the argument work.

Proposition (Anchored Cauchy Estimate): Suppose $f$ and $g$ are differentiable on $(A,\infty)$, $g'(t)\ne0$ for every $t>A$, and $|g(x)|\to\infty$ as $x\to\infty$. Suppose also that $f'(t)/g'(t)\to L$, where $L\in\mathbb{R}$. For every $\varepsilon>0$, there is an anchor $T>A$ such that, for every $x>T$, some $c\in(T,x)$ satisfies $$ f(x)-L g(x) = f(T)-L g(T) + \left(\frac{f'(c)}{g'(c)}-L\right)\bigl(g(x)-g(T)\bigr), $$ with $$ \left|\frac{f'(c)}{g'(c)}-L\right|<\varepsilon. $$ In particular, these estimates imply $f(x)/g(x)\to L$.

Proof. Fix $\varepsilon>0$. By the definition of the limit of $f'/g'$, there is a $T>A$ such that, for every $t>T$,

\left|\frac{f'(t)}{g'(t)}-L\right|<\varepsilon.

Fix any $x>T$. Differentiability of $f$ and $g$ on $(A,\infty)$ ensures that they are continuous on $[T,x]$ and differentiable on $(T,x)$. The Cauchy Mean Value Theorem gives a $c\in(T,x)$ for which

f(x)-f(T) = \frac{f'(c)}{g'(c)}\bigl(g(x)-g(T)\bigr).

The denominator $g'(c)$ is nonzero by hypothesis. Rearranging this identity gives the claimed formula. Also $c>T$, so the choice of $T$ ensures that the coefficient in the formula has absolute value less than $\varepsilon$. Notice that no claim that $c\to\infty$ is needed.

Since $|g(x)|\to\infty$, $g(x)\ne0$ for all sufficiently large $x$. For those $x$, divide the identity by $g(x)$ and use the triangle inequality:

\left|\frac{f(x)}{g(x)}-L\right| \leq \frac{|f(T)-L g(T)|}{|g(x)|} + \varepsilon\frac{|g(x)-g(T)|}{|g(x)|}.

The first term tends to zero because $T$ is fixed and $|g(x)|\to\infty$. For the second factor, the triangle inequality gives

\frac{|g(x)-g(T)|}{|g(x)|} \leq 1+\frac{|g(T)|}{|g(x)|}.

The right-hand side tends to $1$, so the factor on the left is less than $2$ for all sufficiently large $x$. Now let $\eta>0$ be arbitrary and choose $\varepsilon=\eta/4$ at the start of the proof. For sufficiently large $x$, the first term in the estimate is less than $\eta/2$, and the second is less than $2(\eta/4)=\eta/2$. Thus

\left|\frac{f(x)}{g(x)}-L\right|<\eta.

This is the definition of $f(x)/g(x)\to L$, proving the final assertion. $\square$

This estimate supplies a proof of the finite-limit infinity-over-infinity case from the previous tutorial. Its essential safeguard is the order of choices: first choose the tolerance, then choose an anchor that works for that tolerance, and only then apply the Cauchy Mean Value Theorem. Choosing one fixed anchor and hoping that its intermediate points drift to infinity does not justify the conclusion.

Worked Example: Applying the Anchored Strategy

Evaluate

\lim_{x\to\infty}\frac{x+\ln x}{2x+\sqrt{x}}.

Both numerator and denominator tend to $+\infty$. For $x>1$, the functions are differentiable, and the denominator derivative is

\frac{d}{dx}(2x+\sqrt{x})=2+\frac{1}{2\sqrt{x}}>0.

The quotient of derivatives is

\frac{1+1/x}{2+1/(2\sqrt{x})}\longrightarrow\frac{1}{2}.

Thus the hypotheses of the infinity-over-infinity form established earlier hold, and the limit is $1/2$. In the anchored proof, given a tolerance $\varepsilon$, choose $T>1$ so that the derivative quotient differs from $1/2$ by less than $\varepsilon$ for every $t>T$. For each $x>T$, the Cauchy Mean Value Theorem supplies $c\in(T,x)$, so the coefficient at $c$ is already controlled by $\varepsilon$. Dividing the resulting anchored identity by $2x+\sqrt{x}$, whose absolute value tends to infinity, makes the fixed anchor term vanish. No assumption about $c$ tending to infinity is involved.

Two Useful Proof Audits

A proof strategy should also help detect when the rule is being used outside its hypotheses. First check the form of the original quotient. Differentiating numerator and denominator is not justified merely because the quotient looks complicated. Second, remember that L’Hopital’s Rule gives a sufficient condition for a limit: a derivative-quotient limit can establish the original limit, but the derivative quotient need not have a limit whenever the original quotient does.

Worked Example: A Derivative Quotient Does Not Force the Original Limit

As $x\to0^+$, consider $(1+x)/x$. The numerator tends to $1$, while the denominator tends to zero, so this is not a $0/0$ form. Differentiating the two functions separately gives the derivative quotient $1/1=1$, but that calculation does not establish the limit of the original quotient. In fact,

\frac{1+x}{x}=1+\frac{1}{x}\longrightarrow+\infty.

The issue is not an incorrect derivative calculation; it is the failure of the form required by the finite-point rule. Checking the limits of the numerator and denominator is part of the proof, not a preliminary formality.

Worked Example: The Rule Is Sufficient, Not Necessary

Define $f(0)=0$, $f(x)=x^2\sin(1/x^2)$ for $x>0$, and let $g(x)=x$ for $x\geq0$. For $x>0$,

\frac{f(x)}{g(x)}=x\sin(1/x^2), \qquad \left|x\sin(1/x^2)\right|\leq x.

The squeeze inequality proves that $f(x)/g(x)\to0$ as $x\to0^+$. However, the derivative quotient does not tend to a limit. For $x>0$,

f'(x)=2x\sin(1/x^2)-\frac{2}{x}\cos(1/x^2), \qquad g'(x)=1.

Take $x_n=1/\sqrt{2\pi n}$. Then $x_n\to0^+$, $\sin(1/x_n^2)=0$, and $\cos(1/x_n^2)=1$, so

\frac{f'(x_n)}{g'(x_n)}=-\frac{2}{x_n}\longrightarrow-\infty.

On the other hand, take $y_n=1/\sqrt{\pi/2+2\pi n}$. Then $y_n\to0^+$, $\sin(1/y_n^2)=1$, and $\cos(1/y_n^2)=0$, giving

\frac{f'(y_n)}{g'(y_n)}=2y_n\longrightarrow0.

These two sequences rule out a limit for the derivative quotient, even though the original quotient tends to zero. L’Hopital’s Rule is a method for proving a limit when its hypotheses hold; it is not a test that every limit must pass.

A Reliable Order of Work

The proof ideas above can be condensed into a sequence of checks. Each check prevents a different gap in reasoning:

Identify the form. At a finite point, verify that both functions tend to zero. At infinity, verify that both grow without bound in absolute value.
Check the domain and differentiability. Find an interval or tail on which the functions are differentiable and the denominator derivative is nonzero.
Establish the derivative-quotient limit. Compute it carefully and justify its limiting value.
Localize the intermediate point. Near a finite endpoint, use $a<c_x<x$. At infinity, choose an anchor after fixing the error tolerance, then use $T<c<x$.
Finish with the definition of limit. At infinity, divide the anchored identity by $g(x)$, control the fixed anchor term, and bound the remaining factor involving $g(x)-g(T)$.

The key distinction is between what is known about an intermediate point and what is merely hoped for. At a finite endpoint, its location between $a$ and $x$ forces it toward $a$. At infinity, its location beyond an epsilon-dependent anchor makes the derivative estimate valid there, whether or not the intermediate point itself tends to infinity. Keeping those arguments separate makes the proof both rigorous and reusable.

Check Your Understanding

Use the proof strategies and examples in this tutorial to answer the following questions.

Why does the finite-endpoint quotient estimate transfer a derivative-quotient bound to $f(x)/g(x)$?
In the infinity proof, why must the anchor be chosen after fixing an error tolerance?
Why does $T<c<x$ alone fail to prove that $c\to\infty$ as $x\to\infty$?
Which term in the anchored estimate vanishes because $|g(x)|\to\infty$?
Why does the example $x^2\sin(1/x^2)/x$ show that L’Hopital’s Rule is sufficient but not necessary?