Rolle's Theorem

Suppose a smooth trip starts and ends at the same altitude. Must the road be perfectly flat somewhere along the way? This must happen even for just an instant. Common sense says yes. You cannot go up and come back down to your starting height without pausing at some peak. Nor can you dip down and return without resting at some valley. Rolle's theorem makes this intuition rigorous. It translates a statement about heights (a function returning to the same value) into a statement about slopes (a derivative equal to zero). This small theorem is a hinge on which a great deal of calculus turns. It is the seed from which the Mean Value Theorem grows. Indeed, the Mean Value Theorem is proved by tilting the graph of a general function. This tilting continues until Rolle's theorem applies directly. From the Mean Value Theorem one derives several key results. These include the sign tests for increasing and decreasing functions. They also include the characterization of functions with zero derivative as constants. Taylor's theorem with remainder and L'Hôpital's rule follow as well. Much of the logical architecture connecting a function's derivative to its global behavior emerges too. Without Rolle's theorem, calculus would be different. It would be a collection of local facts about instantaneous rates of change. There would be no bridge to global statements about the function itself. The proof is a beautiful two-line argument. It is built entirely from two facts you already have in hand. The first is the Extreme Value Theorem. This guarantees a continuous function on a closed bounded interval attains a maximum and a minimum. The second is the interior-extremum criterion from Interior Extrema and the Vanishing Derivative. This says a differentiable function must have zero derivative at any interior point. Such a point is where the function attains a local extreme value. Rolle's theorem is essentially the observation that these two facts, placed side by side, cannot fail to produce a stationary point. This article states the theorem with all its hypotheses made explicit. It proves the theorem completely from the two prerequisite results. It examines what happens when each hypothesis is dropped. It also traces the theorem's history back to its 17th-century origin. This origin is a surprising one. Michel Rolle himself distrusted the infinitesimal calculus that now bears his name in this form.

Key takeaways

Rolle's theorem: if $f$ is continuous on $[a,b]$ , differentiable on $(a,b)$ , and $f(a) = f(b)$ , then some $c \in (a,b)$ has $f'(c) = 0$ .
All three hypotheses (continuity on the closed interval, differentiability on the open interval, equal endpoint values) are load-bearing; dropping any one breaks the conclusion.
The proof combines exactly two prior tools: the Extreme Value Theorem (existence of a max/min) and the interior-extremum criterion (a local extreme point has zero derivative there, if the derivative exists).
The theorem guarantees existence of a critical point, not uniqueness, and gives no formula for finding it.
It is the direct ancestor of the Mean Value Theorem, obtained by "tilting" an arbitrary function to make its endpoint values match.
Key idea: a continuous function that returns to its starting value must attain an interior extremum, and at a point of interior extremum a differentiable function's slope is forced to zero.

Where this lives today

Rolle's theorem is the direct logical predecessor of the Mean Value Theorem: given any differentiable $f$ on $[a,b]$ (with no assumption that $f(a)=f(b)$ ), one defines the auxiliary function

g(x) = f(x) - f(a) - \frac{f(b)-f(a)}{b-a}(x-a),

which subtracts off the secant line joining $(a,f(a))$ to $(b,f(b))$ ; one checks $g(a) = g(b) = 0$ , applies Rolle's theorem to $g$ , and translates the resulting $g'(c)=0$ back into $f'(c) = \frac{f(b)-f(a)}{b-a}$ . This single trick — reduce to Rolle by tilting — is the standard proof pattern reused throughout analysis. Beyond this direct descendant, Rolle's theorem (or its underlying Fermat-critical-point idea) recurs constantly:

Root separation in numerical analysis: between any two consecutive real roots of a differentiable function, Rolle's theorem guarantees a root of the derivative, which underlies bisection-style root-isolation algorithms and Sturm's theorem on counting real roots of polynomials.
Polynomial theory: Rolle's original algebraic statement bounds how the real roots of a polynomial and its derivative interlace, still used in establishing bounds on the number of real roots of a polynomial (Descartes' rule of signs proofs use related ideas).
Physics and mechanics: any smooth motion that returns to its starting position at time $b$ having started at time $a$ must have zero instantaneous velocity at some intermediate time — a direct physical reading of the theorem, used in kinematics and control theory to argue the existence of "turning instants."
Optimization and machine learning theory: the underlying interior-extremum principle Rolle's theorem depends on is the same principle behind first-order optimality conditions (gradient equals zero at an unconstrained optimum) used throughout convex optimization and the analysis of gradient descent.
Complex and functional analysis generalizations: analogues and failures of Rolle's theorem in higher dimensions and for complex-differentiable functions are an active topic; the theorem does not generalize verbatim to complex analytic functions, which is itself an instructive boundary of the real-variable theory.

:::quiz[Let $f(x) = |x|$ on $[-1,1]$ . Why does this NOT contradict Rolle's theorem?]

Because $f(-1) \neq f(1)$
Because $f$ is not continuous on $[-1,1]$
Because $f$ fails to be differentiable at the interior point $x=0$ , so a hypothesis of the theorem is not met ✓
Because the Extreme Value Theorem fails for $f$ on $[-1,1]$

:::quiz[For $f(x) = (x-1)^2(x-4)$ on $[1,4]$ , which point(s) are guaranteed to exist, and confirmed, by Rolle's theorem?]

$x = 1$ only, since it is a root of $f$
$x = 2.5$ , the midpoint of the interval
$x = 3$ , an interior point where $f'(3) = 0$ ✓
$x = 4$ , the right endpoint

:::faq[Is Rolle's theorem true if $[a,b]$ is replaced by an unbounded interval, like $[0,\infty)$ ?] Not in general, precisely because the Extreme Value Theorem's compactness hypothesis (closed and bounded) fails. Consider $f(x) = \frac{x}{1+x^2}$ for a bounded counterexample-adjacent case, or more simply note that a continuous function on $[0,\infty)$ with $f(0) = \lim_{x\to\infty} f(x)$ need not attain any extremum at all if the limiting behavior is only approached, never reached. The closed-and-bounded hypothesis on the interval is doing real work by way of guaranteeing compactness for the Extreme Value Theorem.

Ask the tutor: How does tilting a general function by its secant line turn Rolle's theorem into the Mean Value Theorem?

a < b

Intuition

Picture the graph of $f$ over $[a,b]$ : it starts at height $f(a)$ , wanders around, and returns to the exact same height $f(b) = f(a)$ . If $f$ were constant, every point is a stationary point and there is nothing to prove. Otherwise the graph must rise above, or dip below, its starting height at some point in between — because it is continuous, it cannot jump directly from $f(a)$ back to $f(a)$ without doing something in between (unless it does nothing at all). Wherever the graph reaches its highest point (or its lowest point) strictly inside the interval, the tangent line there must be momentarily horizontal: an instant of zero velocity, exactly like the peak of a ball thrown straight up, where the ball's height stops increasing and starts decreasing, so its velocity passes through zero. The proof strategy exploits the Extreme Value Theorem, which is only about existence of a global max and global min on the closed interval, and combines it with the purely local fact that an interior extremum forces a zero derivative. The subtlety the proof must handle is that the max and min guaranteed by the Extreme Value Theorem might both occur at the endpoints $a$ and $b$ — and endpoints are not "interior," so the local criterion does not apply there. The resolution is a case split: if both extrema sit at the endpoints, then since $f(a)=f(b)$ , the max equals the min, forcing $f$ to be constant, and any interior point works. Otherwise, at least one extremum is genuinely interior, and the criterion applies directly to that point.

Figure

A function returning to its starting height must level off somewhere in between

a,b \in \mathbb{R}

[a,b]

Worked examples

Example 1 (an asymmetric cubic). Let $f(x) = x^3 - 5x^2 + 3x + 9$ on the interval $[a,b] = [-1, 3]$ . Check the hypotheses. $f$ is a polynomial, hence continuous on $\mathbb{R}$ , so certainly continuous on $[-1,3]$ ; and differentiable everywhere, in particular on $(-1,3)$ , with

f'(x) = 3x^2 - 10x + 3.

Check the endpoint condition:

f(-1) = (-1)^3 - 5(-1)^2 + 3(-1) + 9 = -1 - 5 - 3 + 9 = 0,

f(3) = 3^3 - 5(3)^2 + 3(3) + 9 = 27 - 45 + 9 + 9 = 0.

So $f(-1) = f(3) = 0$ : all three hypotheses of Rolle's theorem hold, even though the interval $[-1,3]$ is not centered at a "nice" point and the function has no symmetry. Find the guaranteed point(s). Solve $f'(c) = 3c^2 - 10c + 3 = 0$ using the quadratic formula:

c = \frac{10 \pm \sqrt{100 - 36}}{6} = \frac{10 \pm 8}{6}.

This gives $c = 3$ or $c = \tfrac{1}{3}$ . The root $c = 3$ is an endpoint, not in the open interval, so it is not the point Rolle's theorem promises (indeed the theorem only asserts existence of a point in the open interval — the appearance of $c=3$ as an algebraic root of $f'$ is incidental here). The other root,

c = \frac{1}{3} \in (-1, 3),

lies strictly inside the interval, confirming the theorem: $f'\!\left(\tfrac13\right) = 3\cdot\tfrac19 - 10\cdot\tfrac13 + 3 = \tfrac13 - \tfrac{10}{3} + 3 = \tfrac{1 - 10 + 9}{3} = 0$ , as required.

Example 2 (a case where the naive "midpoint guess" fails, testing the proof's logic). Let $f(x) = (x-1)^2(x-4)$ on $[a,b] = [1,4]$ . Check hypotheses. $f$ is a polynomial: continuous on $[1,4]$ and differentiable on $(1,4)$ . Endpoints:

f(1) = (0)^2(1-4) = 0, \qquad f(4) = (3)^2(0) = 0.

So $f(1) = f(4) = 0$ , satisfying the theorem's hypotheses. Locate the zero-derivative point. Expand:

f(x) = (x^2 - 2x + 1)(x-4) = x^3 - 6x^2 + 9x - 4.

Then

f'(x) = 3x^2 - 12x + 9 = 3(x^2 - 4x + 3) = 3(x-1)(x-3).

The roots of $f'$ are $x = 1$ and $x = 3$ . Here $x=1$ is an endpoint (excluded from the open interval, consistent with the theorem only guaranteeing an interior point), while

c = 3 \in (1,4)

is genuinely interior, and $f'(3) = 3(2)(0) = 0$ , confirming the conclusion. Note the midpoint of $[1,4]$ is $2.5$ , not $3$ : this example is deliberately asymmetric to show the guaranteed point need not be, and generally is not, the midpoint of the interval — Rolle's theorem promises existence of some interior critical point, never where it sits or that there is only one (here there would have been two critical points of $f$ inside $[1,4]$ had $x=1$ not been an endpoint, illustrating that the theorem's " $c$ " need not be unique when it exists in the interior more than once).

f(x) = |x|

Scope, edge cases & common traps

Situation	What happens
$f$ differentiable on $[a,b]$ (closed), not just $(a,b)$	Hypotheses are still satisfied (differentiability on the open interval is implied); theorem applies as stated, with room to spare.
$f$ continuous on $[a,b]$ but only differentiable on $(a,b)\setminus\{x_0\}$ for one bad point $x_0$	Conclusion can fail — see the $\|x\|$ counterexample.
$f(a) \neq f(b)$	Rolle's theorem does not apply directly; this is precisely the setting generalized by the Mean Value Theorem.
Domain is an open interval $(a,b)$ with no continuity assumed at endpoints (or endpoints not in domain at all)	The Extreme Value Theorem no longer guarantees a maximum/minimum is attained; the conclusion can fail (e.g. $f(x) = x$ on $(0,1)$ , extended arbitrarily, has no interior critical point).
$f$ constant on $[a,b]$	Every point of $(a,b)$ satisfies $f'(c) = 0$ ; this is "Case A" of the proof and is the only situation where the guaranteed point is not unique in an essential way tied to a genuine extremum.
More than one point with $f'(c)=0$	Perfectly consistent with the theorem — it asserts existence of at least one such point, never uniqueness.

Common trapThe classic error: assuming differentiability at the endpoints, or forgetting it can fail at just one interior point

Students often state Rolle's theorem requiring $f$ differentiable on the closed interval $[a,b]$ , including one-sided derivatives at $a$ and $b$ . This is too strong: the correct and standard hypothesis only requires differentiability on the open interval $(a,b)$ ; the endpoints only need continuity. Conversely, an equally common error is thinking "differentiable except possibly at finitely many points is close enough" — the $|x|$ counterexample shows a single non-differentiable point, if it happens to coincide with where the extremum must occur, is fatal to the conclusion. The hypothesis is not "differentiable almost everywhere"; it is "differentiable at every point of $(a,b)$ , no exceptions."

The mathematician behind it

Origins · 1 minMichel Rolle and how this was discovered

Michel Rolle

1652–1719 · French Academy of Sciences, algebra and the foundations of calculus

Michel Rolle was a self-taught French mathematician who worked as a clerk and later became a respected member of the Académie Royale des Sciences. He is best remembered today for a 1691 result, stated purely in the language of polynomial algebra, concerning the location of roots of a polynomial's derivative between consecutive roots of the polynomial itself — a discrete, algebraic ancestor of the theorem that now carries his name in the language of calculus. Ironically, Rolle was for years an outspoken critic of the newly invented infinitesimal calculus of Newton and Leibniz, viewing it as insufficiently rigorous, and he had no notion of a general continuous, differentiable function in the modern sense. The theorem's generalization to arbitrary differentiable functions, and its formulation via limits, emerged only later, notably in the rigorous treatments developed in the 19th century by mathematicians building the modern theory of Continuity and differentiation, including Cauchy, who used essentially this result as a stepping stone toward the Mean Value Theorem.

History

- 1691 — Michel Rolle publishes an algebraic version of the theorem for polynomials, in the context of a method for approximating roots of equations.

- 1823 — Augustin-Louis Cauchy, in his lectures on the calculus, formalizes limits and continuity rigorously and proves a general version of the Mean Value Theorem, using a Rolle-type lemma as its key step.

- Late 1800s — the theorem is given its modern name "Rolle's Theorem" in calculus textbooks, extended fully to general differentiable real functions rather than polynomials alone.

- 20th century — the theorem becomes a standard early result in real analysis courses, typically proved (as here) directly from the Extreme Value Theorem and Fermat's interior-extremum criterion.

- Today — the theorem is a foundational lemma cited in the proof of the Mean Value Theorem in essentially every modern analysis textbook.

Does Rolle's theorem require

f(a) = f(b) = 0

, or just

f(a) = f(b)

Just $f(a) = f(b)$ — the common value need not be zero. Many textbook examples use functions vanishing at both endpoints because it makes finding the interval easy (just find two roots of $f$ ), but the theorem itself only cares that the two endpoint heights match, whatever that common height is. For instance $f(x) = \cos x$ on $[-\pi/3, \pi/3]$ has $f(-\pi/3) = f(\pi/3) = 1/2 \neq 0$ , and Rolle's theorem still guarantees a point $c \in (-\pi/3,\pi/3)$ with $f'(c) = -\sin c = 0$ , namely $c = 0$ .

Why is the Extreme Value Theorem needed — couldn't we just look at where

f' = 0

directly?

The whole difficulty is that we are not handed a formula for $f'$ to solve directly; we only know $f$ is some differentiable function. Rolle's theorem is an existence statement valid for every differentiable function satisfying the hypotheses, including ones without a closed-form derivative. The Extreme Value Theorem is what guarantees, abstractly and for any continuous function on a closed bounded interval, that a highest and lowest point actually exist to reason about — without it, a function could in principle have a supremum that is never attained, and the whole argument would collapse.

What if

f

is differentiable but

f'

is not continuous — does Rolle's theorem still apply?

Yes. Rolle's theorem only requires $f$ to be differentiable on $(a,b)$ ; it says nothing about the derivative function $f'$ being continuous. A classic example is $f(x) = x^2 \sin(1/x)$ for $x \neq 0$ and $f(0) = 0$ , whose derivative exists everywhere but is discontinuous at $0$ ; Rolle's theorem applies to such functions exactly as stated, because the interior-extremum criterion it relies on likewise never assumes continuity of the derivative.

Can Rolle's theorem produce more than one valid point

c

Absolutely, and this is not a contradiction of the theorem, which only claims existence of at least one point. For example $f(x) = \sin x$ on $[0, 2\pi]$ has $f(0) = f(2\pi) = 0$ , and $f'(x) = \cos x$ vanishes at both $c = \pi/2$ and $c = 3\pi/2$ , both interior to $(0,2\pi)$ . The theorem is silent on how many zeros of $f'$ exist; it only guarantees the set is nonempty.

How does Rolle's theorem relate to finding roots of a derivative numerically?

It gives a sufficient condition for a root of $f'$ to exist between two points where $f$ agrees, which underlies bracketing methods: if you can find two points $a<b$ with $f(a)=f(b)$ , Rolle's theorem certifies a critical point of $f$ lies strictly between them, and further subdividing (using the Intermediate Value Theorem on $f'$ , if $f'$ is continuous) can narrow down its location — this is one heuristic behind root-isolation techniques for derivatives in numerical root-finding and critical-point search algorithms.

Does an analogous theorem hold for complex-differentiable (holomorphic) functions?

No — Rolle's theorem is a distinctly real-variable phenomenon and fails for complex differentiable functions. For example $f(z) = e^{2\pi i z}$ satisfies $f(0) = f(1) = 1$ , yet $f'(z) = 2\pi i\, e^{2\pi i z}$ is never zero anywhere in $\mathbb{C}$ , including on the segment $(0,1)$ . The proof of the real theorem relies essentially on the Extreme Value Theorem for real-valued continuous functions and the real-variable interior-extremum criterion, neither of which has a direct analogue for complex-valued functions in the same form.