The Mean Value Theorem: At Some Point, the Slope Must Match the Average

You drive from one city to another, two hundred kilometres apart, in two hours flat. Your average speed for the journey is one hundred kilometres per hour. Now consider the instantaneous speed your speedometer reads at each moment. It might have read $80$ at one point, $130$ at another, $0$ when you were stopped at a red light. But at some instant during the trip, your speedometer must have read exactly $100$. Not “around $100$”, not ”$100$ on average”, but precisely $100$ at one specific moment of clock time.

That this is true should feel inevitable, even obvious. Your speed varied continuously, the average is $100$, so by continuity the instantaneous speed must hit $100$ on its way through the various values it takes. The remarkable thing is that this everyday physical intuition is exactly the content of one of the most-used theorems in all of calculus — the Mean Value Theorem, which says, in mathematical form, that the instantaneous rate of change of a smooth function must equal its average rate of change at some point in any interval. This article is about that theorem, its short proof, and why it sits at the foundation of so much of analysis.

The statement

Let $f$ be a continuous function on a closed interval $[a, b]$, and differentiable on the open interval $(a, b)$. Then there exists at least one point $c$ in the interior of the interval such that

$f'(c) = \frac{f(b) - f(a)}{b - a}.$

The right-hand side is the average slope of the function between $a$ and $b$ — the slope of the straight line (the secant) connecting the endpoints. The left-hand side is the instantaneous slope of the function at the single point $c$. The theorem says: at some point in the interval, the tangent line to the curve is parallel to the secant line that joins the endpoints.

A picture

The geometric content is best seen all at once. The curve in the picture below climbs from $(a, f(a))$ to a peak and then descends to $(b, f(b))$; the secant line is drawn in green, and at one specific point $c$ a tangent line of the same slope is shown in red, parallel to the secant.

The point $c$ is where the curve, having climbed and then started to descend, is steepest in the same direction as the overall trend from start to finish. Visually it is the place where the tangent line would slide along the curve in perfect parallel to the green secant. The MVT is the statement that such a $c$ must exist — and indeed it usually does in more than one place on a curve that wiggles.

Why it is true: Rolle’s theorem

The proof of the mean value theorem is short and goes through a special case, Rolle’s theorem: if $f(a) = f(b)$, then there exists $c \in (a, b)$ with $f'(c) = 0$.

This is intuitively obvious. If the curve starts and ends at the same height, it must rise and then come back down (or vice versa), and at the highest or lowest point in between it has a horizontal tangent. A rigorous proof uses the extreme value theorem — a continuous function on a closed interval attains its maximum and minimum — and Fermat’s theorem that the derivative vanishes at an interior extremum. So Rolle’s theorem follows almost from definitions.

To deduce the general mean value theorem, define an auxiliary function

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

This is $f$ minus the linear function whose slope is the average slope. By direct check, $g(a) = f(a)$ and $g(b) = f(a)$ as well, so $g$ takes the same value at both endpoints. Rolle’s theorem applies to $g$ and yields a point $c$ where $g'(c) = 0$. Computing $g'(c) = f'(c) - (f(b) - f(a))/(b - a) = 0$ rearranges into exactly the statement of the mean value theorem. Three lines of algebra and the job is done.

What it really gives us

The reason the mean value theorem is so important is not the picture but its consequences. It is the bridge from derivative information to function information.

Monotonicity. Suppose $f'(x) > 0$ for all $x$ in some interval. For any two points $a < b$ in that interval, the mean value theorem says $f(b) - f(a) = f'(c)(b - a)$ for some $c$ between them. Since $f'(c) > 0$ and $b - a > 0$, the right side is positive, so $f(b) > f(a)$. The function is increasing. This is the standard calculus-textbook fact that “positive derivative means increasing,” and it is proved by the mean value theorem. Without it, the connection between the sign of $f'$ and the behaviour of $f$ would be a heuristic, not a theorem.

Constant functions. If $f'(x) = 0$ everywhere on an interval, then for any two points $a < b$, the mean value theorem gives $f(b) - f(a) = f'(c) \cdot (b - a) = 0$. So $f(b) = f(a)$ for any two points — meaning $f$ is constant. The MVT is what justifies the leap from “the derivative is zero” to “the function does not change.” This in turn underwrites the indefinite integral: two antiderivatives of the same function differ by a constant precisely because their difference has derivative zero everywhere.

Lipschitz bounds. If $|f'(x)| \leq M$ on an interval, the MVT immediately gives $|f(b) - f(a)| \leq M|b - a|$, a uniform bound on how fast the function can change. This is the foundation of the mean value inequality, and it is the reason bounding the derivative bounds the function — a fact used implicitly in almost every error estimate in numerical analysis.

Taylor’s remainder. When you approximate a function by its Taylor polynomial of degree $n$, the error has a clean form called the Lagrange remainder:

$R_n(x) = \frac{f^{(n+1)}(c)}{(n+1)!}(x - a)^{n+1},$

for some $c$ between $a$ and $x$. This is essentially the mean value theorem applied iteratively. It is what lets you bound Taylor-polynomial errors by bounding higher derivatives, and it is the basis of nearly every quantitative error analysis in classical analysis.

Newton’s method. The convergence proof of Newton’s method for root-finding relies on the mean value theorem to control how quickly successive iterates approach the root. Many iteration-based numerical algorithms in fact prove their convergence by applying MVT-style inequalities to the error.

A speedometer between cities

The opening physical picture is in fact a perfect summary of what the theorem really says. The position-versus-time graph of a car making a journey is a continuous, differentiable function (assuming smooth driving). The total displacement divided by total time is the average velocity. The mean value theorem says: at some specific instant during the trip, the instantaneous velocity — what the speedometer reads — equals the average. This is a real, observable, physical fact, and it falls directly out of the theorem.

What makes the result mathematically interesting is not the existence of the moment — that is intuitively unavoidable — but how generally the same logic works. Replace position by any continuously differentiable quantity, replace time by any continuous parameter, and the same conclusion holds: somewhere along the way, the rate of change matched the average rate. It is the kind of statement whose physical content seems almost too obvious to need proving, until you try to prove it without the theorem and find that all your proofs implicitly need it. That is the situation classical calculus was in for nearly a century before Lagrange and Cauchy wrote down the statement and proof properly.

A small theorem with a long reach

The mean value theorem is a small result. It says, in essence, that derivatives cannot lie about averages. A continuously differentiable function’s instantaneous slope is forced to coincide, somewhere, with its average slope across any interval. But from that tiny seed grows the entire structural connection between local and global behaviour that makes the calculus of real-valued functions a coherent subject rather than a collection of formulas. Without it, the relationship between $f$ and $f'$ would be a matter of intuition rather than logic, and most of single-variable analysis would be built on sand.

It is among the cleanest examples in mathematics of how a single physically obvious fact, once stated precisely and proved, becomes the foundation on which an entire theory can be built. The car between two cities had to hit exactly $100$ kilometres per hour at some moment — and somehow that single sentence is the engine that drives Taylor’s theorem, the proof that the derivative determines whether a function is increasing, the convergence of numerical methods, and a host of inequalities and bounds that working analysts use every day. Lagrange and Cauchy gave us, in just a few lines, one of the most quietly powerful theorems in all of mathematics.

Frequently asked

Who proved the mean value theorem first?

Augustin-Louis Cauchy gave the first modern statement and rigorous proof in his 1823 lectures at the École Polytechnique, building on a near-version published by Joseph-Louis Lagrange in 1797. The special case where f(a) = f(b), now called Rolle's theorem, was proved by Michel Rolle in 1691 for polynomials and extended to general smooth functions later. The general MVT is therefore sometimes called the 'Lagrange–Cauchy theorem' in older textbooks.

What is Rolle's theorem?

It is the special case of the mean value theorem where f(a) and f(b) happen to be equal. The statement is then that f'(c) = 0 for some c between a and b — there is a point where the curve has a horizontal tangent, which is intuitively obvious if you imagine a smooth curve that starts and ends at the same height: it must have a peak or a valley somewhere in between. Rolle's theorem is the essential lemma from which the full mean value theorem is derived in one short algebraic step.

Why is the mean value theorem useful?

Because it converts a global statement about a function into a local statement about its derivative. Want to know whether f is increasing on an interval? Check whether f' is positive there — MVT guarantees that the sign of the derivative controls the change of the function. Want to bound the error in a Taylor polynomial? MVT gives you the Lagrange remainder. Want to prove a function is constant? Show its derivative is zero everywhere; MVT does the rest. The theorem is the workhorse that makes elementary calculus proofs work.

Does MVT have a multidimensional version?

Yes, but it is more subtle. For a function f from R^n to R, the gradient version of MVT holds: f(b) − f(a) = ∇f(c) · (b − a) for some c on the segment between a and b. For maps from R^n to R^m with n ≥ 1 and m > 1, the literal statement fails — there may be no single point c at which the derivative matrix takes the average value. Instead one uses an integral version, sometimes called the fundamental theorem of calculus for line integrals, which substitutes an average derivative along the segment for a single derivative at a point.

Continue reading

Blog

Lagrange Multipliers: How to Optimise When You Are Not Free to Move Anywhere

At a maximum of a function on a curve, the contour lines must run parallel to the curve. That geometric fact turns one of the hardest problems in calculus into a tidy system of equations.

Read →Blog

The Fundamental Theorem of Calculus: Why Areas and Slopes Are Two Sides of the Same Idea

Adding up infinitely many thin strips and finding the steepness of a curve look like utterly different problems. One short theorem reveals that they are exact inverses of each other.

Read →Blog

Taylor Series: How a Function Becomes an Infinite Polynomial

Pick a smooth function, write down its value, slope, curvature, and every higher derivative at a single point — and you have rebuilt the whole function from that one spot, as an infinite polynomial.

Read →