Source
Adapted from my own handwritten and hand-typed notes from Year 1 to Year 3 of my EEE degree
To find the highest and lowest points of a surface , we first find candidate points where the surface is "flat" or not smooth, then classify them.
Global and local extrema
- Global (absolute) maximum at : for all in the domain (minimum: )
- Local maximum at : for all in some small disk around (minimum: )
- Every global extremum is also a local extremum
Critical points
If has a local extremum at an interior point and the partial derivatives exist there, then .
Names for these points:
- Stationary point: and
- Singular point: or does not exist there
- Critical point: stationary or singular
- Saddle point: a stationary point that is neither a local max nor a local min (the surface goes up in one direction and down in another, like a horse saddle)
Not every stationary point is an extremum, so we must classify.
Second derivative test
Suppose the second partials are continuous near a stationary point . Let be the determinant of the Hessian matrix:
| Condition | Conclusion |
|---|---|
| and | local minimum |
| and | local maximum |
| saddle point | |
| no conclusion (check by hand, e.g. along paths or by completing the square) |
This test is only for two variables.
Method
- Compute , . Points where either does not exist are singular points
- Solve and together for the stationary points (factorise, do not divide by something that could be zero)
- Compute , , and , then classify each stationary point with the table
- The local max or min value is at that point
Worked example
Find the local extrema of .
Working:
- , so and
- Then , i.e. , so
- Stationary points: , ,
- , , , so
| Point | Type | ||
|---|---|---|---|
| saddle point | |||
| local min, | |||
| local min, |
Second example
has , so the stationary points are and . With , , :
- , saddle point
- and , local maxima with value
Singular point example
For , we get . The partial does not exist when , so every point is a singular point. For , forces and then forces , a contradiction, so there is no stationary point.
For extrema on a closed region, see Global Extrema on Closed Bounded Regions.