An optimization problem asks for the largest or smallest value of a quantity subject to constraints. The calculus is short: differentiate and solve . The mathematical content lies in the two steps around it. Before differentiating, the problem must be reduced to a function of one variable on an explicitly stated domain. After solving, one must prove that the candidate is a global extremum, not merely a stationary point. This unit relies on the extreme value theorem from the unit on continuity, on the mean value theorem and monotonicity tests, and on the derivative rules.
Set-up and terminology
Definition 4.7.1
(Objective, constraint and feasible set)
The objective function is the quantity to be maximized or minimized, such as a volume, an area or a distance. The constraint is an equation or inequality linking the variables. The feasible set is the set of all variable values allowed by the constraint and by the physical meaning of the problem (lengths are positive, a cut cannot be longer than the sheet). A point of an interval is a global maximum point of on if for all ; a global minimum point is defined with the reverse inequality.
A constraint is used to eliminate variables. If the objective is and the constraint is , we study on the set of for which a feasible exists. The domain of is part of the problem, and it is the domain, not the formula, that decides whether endpoints must be examined.
The two certificates
Theorem 4.7.2
(Extreme values on a closed interval)
Let be continuous on . Then attains a global maximum and a global minimum on . Each is attained at an endpoint, at a point where does not exist, or at a point of where .
Proof
(Existence, then the sign of the difference quotient)
Existence is the extreme value theorem. Suppose a global maximum is attained at and is differentiable at . Then for near , so the difference quotient is for and for . Letting gives and , hence . The minimum case is symmetric.
Consequently, on a closed interval one lists the finitely many candidates, evaluates at each, and compares the numbers. No second-derivative test is needed.
Theorem 4.7.3
(A single sign change gives a global extremum)
Let be an interval (bounded or not, open or not), let be continuous on and differentiable in the interior of except possibly at , and suppose for interior and for interior . Then is the global maximum of on , and it is attained only at . The statement with the signs reversed gives the global minimum.
Proof
(Mean value theorem on each side of the turning point)
Let in . By the mean value theorem on there is with , so . Likewise for the mean value theorem on gives .
Proof X-ray
Where the single-sign-change proof uses its hypotheses
The proof never searches for ; it compares with for each other separately. The key move is to turn information about the sign of on an open interval into an inequality between two function values, using the mean value theorem. That theorem needs continuous on the closed interval and differentiable on its interior. So continuity at the endpoint is exactly where the hypothesis on is used, while the sign of is used only strictly inside. Strict signs make each inequality strict, which is why the maximum point is unique. The argument closes because every lies on one side of , so one of the two cases applies, and neither boundedness nor closedness of is needed. If continuity at fails, the mean value theorem cannot be applied up to , and the next counterexample shows the conclusion can then fail.
Note that the single-sign-change theorem needs continuity of at but does not need to exist. A corner at is allowed.
Counterexample mode
Continuity at the turning point cannot be dropped
Claim: if to the left of and to the right, then is the global maximum, even without continuity at . Let , , and define for , , and for . Then for and for . But , so is not the maximum. In fact has no maximum at all, since everywhere while as . The repair is the hypothesis that is continuous at .
A commonly used shortcut is: if and for all , then is the global minimum point, because is strictly increasing and therefore changes from negative to positive at . A positive value at one point only gives a local minimum.
A procedure
- Name the variables and write the objective in words and symbols.
- Write the constraint and use it to reduce to one variable.
- State the domain of the one-variable function, including any restrictions that come from the geometry.
- Find all critical points in the domain, and note endpoints and points of non-differentiability.
- Justify the extremum by the closed-interval theorem (compare values) or the single-sign-change theorem (sign change), and state the answer with the value of every original variable.
Worked examples
Worked example 4.7.4
(Closed domain: an open box)
From a square sheet of side cm, a square of side is cut from each corner and the sides are folded up to form an open box. Find the largest possible volume.
The base has side and the height is , so . The cut cannot exceed half the side, so the domain is . Then
The critical points are and . Since the domain is closed, compare values:
The global maximum is , attained at . The endpoints are included in the comparison but give the degenerate boxes of volume .
Worked example 4.7.5
(Open domain: a can of fixed volume)
A closed cylindrical can must hold . Find the radius and height minimizing the surface area.
Let be the radius and the height. The constraint is , so . The objective is
Then , which vanishes only when . Put . For the numerator is negative, and for it is positive, so changes from negative to positive at . By the single-sign-change theorem (with the reversed signs), is the global minimum on the open unbounded interval. The domain has no endpoints at which a minimum could hide, and that theorem rules out any behavior at the ends. At the optimum, gives
so the optimal can has height equal to its diameter, and .
Concept lensAlgebraic
The can minimum without derivatives
Write . The three terms have constant product , so the arithmetic–geometric mean inequality gives for every , with equality exactly when the three terms are equal. Equality is , the same . The inequality is a global certificate with no sign table.
Worked example 4.7.6
(Distance, with an endpoint check)
Find the point on the parabola closest to .
A point of the parabola is with and . The squared distance is . Minimizing and minimizing are equivalent, because the square root is strictly increasing on ; using avoids a radical. Thus we minimize on . Now is negative for and positive for , so by the single-sign-change theorem, gives the global minimum, with . The endpoint has , consistent with this. The closest points are and the distance is .
Common mistakes and subtle points
Common mistake
(No domain, or forgotten endpoints)
If the domain of is taken to be , the function is unbounded and there is no maximum. The restriction is what makes the problem meaningful. On a closed interval the maximum may occur at an endpoint, where need not vanish. For on the maximum is at and .
Common mistake
(Treating a stationary point or a local extremum as global)
has with no extremum. Always supply a certificate. A cubic may have a local maximum that is lower than values elsewhere in the domain. Compare with the behavior at the boundary of the domain.
Common mistake
(Wrong reduction, or the wrong quantity)
Substituting a constraint that does not hold, or leaving two variables in the objective, makes meaningless. Check that the final answer satisfies the original constraint. The question may ask for the dimensions, not the optimal value; report both.
Exercises
Quick check
(Among all rectangles of perimeter , find the dimensions of the one with the greatest area, and justify that it is a global maximum.)
Solution · Guided solution
Let the sides be and , with . The area is , and vanishes at . On the closed interval compare , , . The global maximum is , at the square with side .
Quick check
(An open-top box with a square base must have volume . Find the base side and height that minimize the amount of material (surface area including the base, excluding the lid).)
Solution · Guided solution
Let the base side be and the height . Then , so . The surface area is , . Differentiating gives , which is negative for and positive for . By the single-sign-change theorem, gives the global minimum. Then and . The optimal box has base side m and height m.
Quick check
(Find the global maximum and global minimum of on , or explain why one of them does not exist.)
Solution · Guided solution
, positive on and negative on . By the single-sign-change theorem, is the global maximum on . Since for and , the global minimum is , attained at the endpoint (where ). Note that as , but this value is never attained for because there; the minimum is attained at only.
Quick check
(Find the point of the line closest to the origin and the distance.)
Solution · Guided solution
A point of the line is . The squared distance is , defined for all real . Then is negative for and positive for , so is the global minimum point. The point is and , so the distance is .