Mean Value Theorem

The picture, before the formula

Draw a smooth curve from the point (a, f(a)) to (b, f(b)). Now draw the straight secant line connecting those two endpoints. The Mean Value Theorem makes a striking claim: somewhere strictly between a and b, the curve has a tangent line exactly parallel to that secant.

Translated to numbers: the secant slope is the average rate of change of f across the interval, namely (f(b) − f(a))/(b − a). The tangent slope at c is the instantaneous rate of change f′(c). MVT says these two are equal at some interior point c. The average is realised, somewhere along the way, as an instantaneous reading.

The driving-speed analogy: if you drive 60 miles in 1 hour, the MVT guarantees that at some moment your speedometer read exactly 60 mph — even if you never drove at exactly that speed for any sustained stretch. The average is always achieved instantaneously, somewhere.

The formal statement

Mean Value Theorem. Let f: [a, b] → ℝ be continuous on the closed interval and differentiable on the open interval (a, b). Then there exists at least one c ∈ (a, b) such that

f′(c) = (f(b) − f(a))/(b − a).

Both hypotheses are essential. Continuity must include the endpoints — otherwise the secant slope itself is meaningless. Differentiability is required only on the open interior, because the theorem doesn't claim anything about derivatives at a or b.

Proof sketch — through Rolle's theorem

The standard proof reduces MVT to its special case, Rolle's theorem (where the secant is horizontal). Define an auxiliary function

h(x) = f(x) − [(f(b) − f(a))/(b − a)] · (x − a) − f(a).

This is f minus the secant line. It is continuous on [a, b], differentiable on (a, b), and satisfies h(a) = h(b) = 0. Rolle's theorem gives a c ∈ (a, b) with h′(c) = 0, which rearranges to f′(c) = (f(b) − f(a))/(b − a).

Rolle's theorem itself follows from the extreme value theorem (a continuous function on a closed interval attains its max and min) and Fermat's theorem (interior extrema have zero derivative).

Worked example: f(x) = x² on [1, 3]

The endpoints give f(1) = 1 and f(3) = 9, so the secant slope is (9 − 1)/(3 − 1) = 4. The derivative is f′(x) = 2x. Setting 2c = 4 gives c = 2, which lies inside (1, 3). The tangent line at x = 2 has slope 4 — parallel to the secant from (1, 1) to (3, 9). For polynomials this c is usually unique; for more oscillatory functions it may not be.

MVT vs Rolle's vs Cauchy MVT

	Rolle's theorem	Mean Value Theorem	Cauchy MVT
Hypothesis	f cts [a,b], diff (a,b), f(a) = f(b)	f cts [a,b], diff (a,b)	f, g cts [a,b], diff (a,b), g′ ≠ 0
Conclusion	∃ c: f′(c) = 0	∃ c: f′(c) = (f(b)−f(a))/(b−a)	∃ c: f′(c)/g′(c) = (f(b)−f(a))/(g(b)−g(a))
Geometric picture	Horizontal tangent	Tangent parallel to secant	Tangent ratio matches secant ratio in xy-plane
Used to prove	MVT itself	Lipschitz bounds, FTC, Taylor remainder	L'Hôpital's rule
Special case of	MVT (with f(a) = f(b))	Cauchy MVT (with g(x) = x)	—
Year proved	1691 (Rolle)	1823 (Cauchy)	1823 (Cauchy)
Counter-example needed	|x| on [−1,1] (not differentiable at 0)	step function (not continuous)	g constant (g′ = 0)

Classical applications

Error bounds. If f′(x) is bounded by M on [a, b], MVT gives |f(b) − f(a)| ≤ M(b − a). This is the Lipschitz inequality and underpins virtually every numerical analysis estimate. Newton's method, fixed-point iteration, and explicit ODE solvers all bound their errors through MVT.

Increasing/decreasing test. If f′ > 0 on an interval, then f is strictly increasing there. The proof uses MVT directly: for any a < b, f(b) − f(a) = f′(c)(b − a) > 0.

Constancy from zero derivative. If f′ ≡ 0 on an interval, MVT forces f to be constant: any two values f(a), f(b) differ by f′(c)(b − a) = 0. This is why the antiderivative of a function is unique up to a constant.

Fundamental theorem of calculus. The connection from local rate of change (the derivative) to total change (the integral) is forged through MVT applied on every sub-interval of a partition. Every formal proof of the FTC has MVT under the hood.

Taylor's theorem with Lagrange remainder. The remainder R_n(x) = f⁽ⁿ⁺¹⁾(c)/(n+1)! · (x − a)ⁿ⁺¹ is the iterated MVT applied to higher derivatives.

When the hypotheses fail

No continuity at endpoints. The function defined as f(x) = x on (0, 1) and f(0) = f(1) = 5 has secant slope 0 but interior derivative always 1. MVT fails because f isn't continuous at the endpoints.

No differentiability somewhere inside. f(x) = |x| on [−1, 1] has average slope 0 but never has f′(c) = 0 — its slopes are always −1 or +1, with no derivative at x = 0. The corner kills the conclusion.

Common mistakes

• Dropping continuity. The hypothesis is continuity on the closed interval, not just the open one. Differentiability implies continuity inside, but you still need it at the endpoints separately.
• Treating c as unique. The theorem says at least one. Many cs can satisfy the equation; you cannot solve for c uniquely without more information.
• Demanding c be computable. MVT is an existence result. For complicated f, finding c may require solving a transcendental equation. The theorem doesn't help you compute it.
• Forgetting that differentiability is on the open interval. A square-root-shaped function like f(x) = √x on [0, 4] satisfies MVT — even though f′(0) doesn't exist — because differentiability is only required on (0, 4).
• Applying MVT to vector-valued functions. The classical MVT fails for ℝ → ℝⁿ. The vector version is an inequality: |f(b) − f(a)| ≤ sup |f′| · (b − a).