An absolute extremum on a closed region can occur in its interior, along an edge, or at a corner. A single stationary-point calculation does not compare all these possibilities. A reliable solution first establishes existence, then describes every part of the boundary and finally compares actual values.
What compactness does and does not provide
Definition 4.7.1
(Absolute extremum on a set)
Let and . A point is an absolute maximum point if for every , and an absolute minimum point if for every . The set is compact if it is closed and bounded.
Theorem 4.7.2
(Extreme value theorem)
If is nonempty and compact and is continuous, then attains an absolute maximum and an absolute minimum on .
The theorem does not say where the extrema are, and does not say their values occur at stationary points of the unrestricted function.
At an interior local extremum where is differentiable, . For a smooth boundary arc, one can restrict to a parametrization and solve a one-variable problem, or use Lagrange multipliers when the constraint gradient is nonzero. A corner is an endpoint of adjacent arcs and should be checked separately; its two incident edges need not share a single tangent.
Worked problem: a triangular domain
Worked example 4.7.3
(A triangular domain with three edges)
On
find the absolute extrema of . The triangle is closed and bounded, and is continuous, so both extrema exist. Completing squares gives
Equality holds exactly at , which lies inside the triangle, so this is the unique absolute minimum.
For the maximum, examine the three edges. On , for . This convex quadratic has maximum at ; its other endpoint has value zero. On , for , whose maximum is at . On , substitute to obtain
This quadratic is convex and its endpoint values are and , so its maximum is at . An interior maximum is impossible: vanishes only at the already identified minimum. Thus the absolute maximum is at .
Proposition 4.7.4
(Convex quadratic on an interval)
If is convex on , then .
Proof
(Convexity bounds the interior by the endpoints)
For with , convexity gives .
This observation cannot be reversed into the claim that the minimum occurs at an endpoint.
Worked example 4.7.5
(An extremum along a whole edge)
On the same triangle find the absolute extrema of .
Both extrema exist by compactness. Since , there is no interior stationary point, so extrema lie on the boundary. On , ranges over ; on , ranges over ; on , is constant. The minimum is attained only at the corner . The maximum is attained at every point of the sloping edge, including both of its corners. Reporting a single point would be incomplete: the extremum value is unique, the set of extremum points is a segment.
Worked example 4.7.6
(An interior saddle and extrema between corners)
On the square find the absolute extrema of .
The unique stationary point is with ; it is a saddle, so it is only a candidate. On , takes values in , largest at and zero at the corners. On , takes values in , smallest at . All four corner values are . Hence the absolute maximum is at and the absolute minimum is at . The extrema occur at edge midpoints, not at corners, and the interior stationary point is neither.
A boundary critical point is only a candidate
For functions , on an equality constraint with , an interior point of a smooth constraint arc that is a constrained extremum must satisfy . The equation finds candidates, including possible minima, maxima and non-extrema. At an endpoint of the feasible arc or a point where , the hypothesis used to obtain that equation fails; inspect those cases separately.
Concept lensGeometric
Level curves sweeping across the region
The level curves of are the ellipses around . The absolute maximum is the largest whose ellipse still meets the triangle. That ellipse, , passes through but is not tangent to either edge there, because the corner is where it leaves the region. The multiplier equation detects only tangency, so it cannot find this maximum.
For an inequality such as , the line is only one component of the boundary. The coordinate inequalities also contribute edges. Even a perfectly solved multiplier equation on the sloping edge will not inspect the two coordinate edges automatically.
Counterexample mode
Existence fails without compactness
A tempting claim is that a continuous function always has a maximum on a bounded region. Remove the boundary from the triangle: , and take . Then on , while the points with give . If some were a maximum point, then , and the point with would lie in and give the larger value . So has no maximum on , and no stationary point either since . The extreme value theorem needs the region to be closed as well as bounded.
Common mistake
(Solving only the multiplier equation on the sloping edge)
For the triangle above, with gives , so , and with we get , where . This is the minimum of on that edge, not the maximum . The maximum sits at the endpoint , where the multiplier hypothesis does not apply, and the two coordinate edges are not inspected at all.
Common mistake
(Reading the boundary as a single smooth curve)
A corner joins arcs with different tangents, so one parametrization or one multiplier equation cannot cover it. List the corner values separately: here , , .
Checklist
- State the domain and verify continuity and compactness if invoking existence.
- Find interior stationary points, remembering they are candidates rather than automatic extrema.
- Parameterize each boundary component, including endpoints and singular points. Check every permitted candidate.
- Compare values on all components, report all points attaining each value, and distinguish local from absolute conclusions.
Quick checks
Quick check
(Find the absolute maximum and minimum of on the triangle , , .)
There is no interior stationary point.
Solution · Answer
, so check the corners (and edges, where is linear). , , . The maximum is at and the minimum is at .
Quick check
(Why does the extreme value theorem not apply to on , and what actually happens?)
Which hypothesis fails?
Solution · Answer
The interval is bounded but not closed, so it is not compact. The maximum is attained at , but values come arbitrarily close to without reaching it: no minimum exists.
Quick check
(Where does attain its maximum and minimum on ?)
Compare the endpoints with the vertex.
Solution · Answer
The maximum is at an endpoint: , , so the maximum is at . The minimum is at the interior vertex , where ; it is not at an endpoint.