Source
Adapted from my own handwritten and hand-typed notes from Year 1 to Year 3 of my EEE degree
Lagrange multipliers find the max and min of when must lie on a constraint curve .
Idea
At a constrained max or min, the level curve of just touches the constraint curve. Touching curves share a tangent line, so their normals are parallel. Since the gradient is the normal to a level curve:
The number is the Lagrange multiplier. If the point is a stationary point of itself.
Method (one constraint)
Assumption: the global max and min on the constraint exist (true when the constraint set is closed and bounded and is continuous).
The system to solve for :
Steps:
- Solve the system above for all solutions
- Evaluate at every point found
- Largest value is the max, smallest value is the min
For 3 variables, add and use .
Tips for solving: factor equations like into cases, divide only when you know the divisor is non-zero, and try multiplying equations by , , to make them look alike.
If the constraint is simple (like ), you can instead substitute it into and do a 1-variable max/min problem.
Lagrangian form
Equivalently, find stationary points of
where the constraint is written as . Setting gives back the constraint.
Two constraints
To optimise subject to and , solve
then compare the values of as before.
Worked example
Extreme values of on the circle (closed and bounded, so extrema exist).
Cases:
- From the first, or
- gives
- gives , so and
- and
Maximum at , minimum at .
Worked example: open box
A lidless box uses m of card: . Maximise .
Multiply by , , respectively. All three left sides become , so . This gives and . Substituting: , so , , and m.
Worked example: two constraints
Maximise subject to and .
So and . Then , so .
Two candidates:
- : , , ,
- :
The maximum value is .