Multivariable calculus
Critical Points of sin(x+y) + cos(x-y)
The critical points of f(x,y) = sin(x+y) + cos(x-y) sit where cos(x+y) = 0 and sin(x-y) = 0. The discriminant there is D = 4 sin(x+y) cos(x-y). The point (pi/4, pi/4) is a local maximum with D = 4 and value 2, (3pi/4, -pi/4) is a saddle with D = -4, and (5pi/4, pi/4) is a local minimum with D = 4.
- local maximumdiscriminant D = 4
- saddle pointdiscriminant D = -4
- local minimumdiscriminant D = 4
These are not necessarily all of them; the page says which it covers.
Add and subtract to decouple the system
The chain rule contributes an inside derivative of to in both variables, but contributes in and in . That single sign difference is what makes the system solvable.
Add the two equations and the sine cancels. Subtract them and the cosine cancels. Two combinations replace a coupled pair with two independent conditions.
Writing and , the critical points are exactly the places where and . Solving back gives and , which is an infinite grid rotated 45 degrees relative to the axes.
- , gives .
- , gives .
- , gives .
The discriminant in closed form
Differentiate once more. Write and to keep the algebra readable.
The two diagonal entries are equal, so the discriminant is a difference of two squares and it collapses.
At a critical point forces , and forces . So is always or and the test never stalls. When and have the same sign, and settles it: both equal to gives and a local maximum, both equal to gives and a local minimum. When and have opposite signs, and the point is a saddle.
That reads off the three listed points immediately. At , so and , the global maximum. At , and so , , and . At , so and , the global minimum.
The mistake students make
The inside derivative of is where most of the errors live, because it is in one variable and in the other. Write the chain rule out in full rather than pattern matching.
That opposite pair of signs is the only difference between and , and it is exactly what makes adding and subtracting productive. Get it wrong and the two partials come out identical, both combinations collapse to the same equation, and the system looks unsolvable.
There is a free check available. is symmetric: swapping and leaves alone and turns into , which is the same number because cosine is even. A symmetric function must satisfy , and the stated partials do, since swapping the variables in flips the sine and gives . If your two partials fail that test, one inside derivative has the wrong sign.
The second slip is assuming means the test is inconclusive at . The discriminant is what decides, and here , comfortably negative. A vanishing diagonal entry with a nonzero mixed partial is the signature of a saddle, not of a degenerate point.
Frequently asked questions
Why does the substitution u = x + y, v = x - y help so much?
Because is genuinely a function of and alone: , which is separable. In those coordinates the critical point conditions are just and , one variable each. The change of variables is a rotation and a rescaling, so it does not create or destroy critical points, and the classification transfers.
Is the list of three points complete?
No. The conditions and each have infinitely many solutions, so the critical points form an infinite grid. The three listed here are one representative of each behaviour. Every other critical point is a copy: shift by or by and both the discriminant and the classification come back unchanged.