Water trickles in at a perfectly steady pour — but the vase is narrow here and wide there, so the level races up through the narrows and crawls through the belly. Over the whole fill there’s one average rise-rate, and MVT swears the level hit exactly that rate at some instant. Pour it, and watch the needle land on the line.
It feels like it could fail — surely you could dodge that exact average? You can’t, and the reason is beautifully simple. It’s Rolle’s Theorem wearing a tilt.
Two hypotheses, one guarantee. Both hypotheses matter — drop either and the promise can break.
Continuous on the closed interval [a, b] — no breaks, jumps, or holes.
Differentiable on the open interval (a, b) — a smooth tangent at every interior point (no corners or cusps).
There is at least one point c in (a, b) where the instantaneous rate equals the average rate:
It promises such a c exists — not where, and not how many. There can be several.
The other guarantee. It’s the reason “find the absolute maximum” is even a fair question — something has to be the biggest.
Both words in “continuous on a closed interval” are load-bearing. Remove either and the max can slip through your fingers:
On the open interval (0, 1), f(x) = x gets closer and closer to 1 but never arrives. There is no single highest value — the max escaped through the open end.
Punch a jump into the graph and the peak can vanish into the gap — the function approaches a value it never actually takes. No continuity, no guarantee.
The special case where the endpoints sit at the same height. If f is continuous on [a,b], differentiable on (a,b), and f(a) = f(b), then somewhere the tangent is flat:
Interior points where f′(x) = 0 or f′(x) does not exist. Every local extremum hides at one — but not every critical point is an extremum.
A local (relative) extremum is the biggest/smallest nearby. An absolute (global) extremum is the biggest/smallest on the whole interval. EVT is about the absolute ones.
EVT guarantees the absolute max/min exist. To locate them: list every critical point and both endpoints, evaluate f at each, and pick the largest and smallest values.
MVT and Rolle need continuity on [a,b] and differentiability on (a,b). On \(|x|\) over \([-1,1]\) the corner at 0 kills differentiability — no guaranteed c.
The guaranteed point c lives in the open interval \((a,b)\) — strictly between the ends, never at a or b.
The candidates test checks critical points and both endpoints. An absolute max often sits at an endpoint, where \(f'\ne 0\).
A critical point is where \(f'=0\) or \(f'\) is undefined. Cusps and corners are critical points too — don’t miss them.