Source

Adapted from my own handwritten and hand-typed notes from Year 1 to Year 3 of my EEE degree

To decide whether a critical point is a local maximum, local minimum or neither, use the first or second derivative test. Optimization problems then turn a real situation into a function and find its best value. On a closed interval, the Closed Interval Method is the simplest route.

Local vs Global Extrema

Definitions:

  • local maximum at : for all in some open interval around
  • local minimum at : for all in some open interval around
  • global extrema compare with the whole domain; a global max is also a local max

Fermat's theorem: if has a local max or min at and exists, then . So local extrema only occur at critical points (where or does not exist).

First Derivative Test

Let be a critical point where is continuous, with existing on both sides of :

  • changes from to : local minimum at
  • changes from to : local maximum at
  • does not change sign: neither

This works even when does not exist.

Examples:

  • : is undefined at , negative for and positive for , so is a local minimum
  • : on both sides of , so is neither

Second Derivative Test

If and is continuous near :

  • : local minimum (cup shape)
  • : local maximum (cap shape)
  • : no conclusion, go back to the first derivative test

Examples:

  • : at , and ; gives a local max, gives a local min
  • : , test fails; goes from to , so is a local min

Global Extremum From One Critical Point

If and on the whole interval , then is the global minimum on . With throughout, it is the global maximum.

Optimization Steps

Work through it in this order:

  1. Draw a diagram and name the variables
  2. Write the quantity to optimise as a formula
  3. Use the constraint to reduce it to one variable, and state the domain
  4. Find the critical points
  5. Justify max or min (closed interval method, first derivative test or second derivative test)
  6. Answer the question asked, with units

Worked Example: Cheapest Can

A closed cylindrical can must hold volume . Minimise the surface area:

  • and
  • constraint gives , so for
  • gives , so
  • for all , so this is the global minimum

Then and : the best can has height equal to its diameter.