Multivariable calculus

Partial Derivatives of x^4 + y^4 - 4xy

For f(x, y) = x^4 + y^4 - 4xy the partial with respect to x is 4x^3 - 4y, and the partial with respect to y is 4y^3 - 4x. Setting both to zero gives y = x^3 and x = y^3, so the critical points are (0, 0), (1, 1), and (-1, -1). The value at both non-zero critical points is -2.

f(x,y)=x4+y4−4xyf(x,y) = x^4 + y^4 - 4xy
∂f∂x=4x3−4y\frac{\partial f}{\partial x} = 4x^3 - 4y
∂f∂y=4y3−4x\frac{\partial f}{\partial y} = 4y^3 - 4x

Differentiate, one variable at a time

With yy frozen, y4y^4 is a constant and contributes nothing, while −4xy-4xy contributes −4y-4y.

∂f∂x=4x3−4y\frac{\partial f}{\partial x} = 4x^3 - 4y

The function is symmetric under swapping the variables, so the yy partial is the mirror image.

∂f∂y=4y3−4x\frac{\partial f}{\partial y} = 4y^3 - 4x

At (2,1)(2,1) these are 32−4=2832 - 4 = 28 and 4−8=−44 - 8 = -4. The steep positive value in xx reflects how quickly a fourth power grows once you leave the origin.

The mistake: solving the critical point system carelessly

Setting 4x3−4y=04x^3 - 4y = 0 and 4y3−4x=04y^3 - 4x = 0 gives y=x3y = x^3 and x=y3x = y^3. Substituting the first into the second gives x=x9x = x^9, and the temptation is to divide by xx and report only x=±1x = \pm 1.

Dividing by xx throws away the solution x=0x = 0. Factor instead.

x9−x=x(x8−1)=x(x4−1)(x4+1)=0x^9 - x = x\left(x^8 - 1\right) = x\left(x^4 - 1\right)\left(x^4 + 1\right) = 0
  • x=0x = 0 gives y=0y = 0, the critical point (0,0)(0,0).
  • x=1x = 1 gives y=1y = 1, and x=−1x = -1 gives y=−1y = -1.
  • x4+1x^4 + 1 has no real roots, so there are exactly three critical points.

Never divide an equation by a variable that could be zero. Factoring keeps every root in view.

Classifying the three points

The second partials are fxx=12x2f_{xx} = 12x^2, fyy=12y2f_{yy} = 12y^2, and fxy=−4f_{xy} = -4, so the discriminant is 144x2y2−16144x^2y^2 - 16.

At (0,0)(0,0) the discriminant is −16<0-16 < 0, so the origin is a saddle point. At (1,1)(1,1) and (−1,−1)(-1,-1) the discriminant is 144−16=128>0144 - 16 = 128 > 0 with fxx=12>0f_{xx} = 12 > 0, so both are local minima.

f(1,1)=1+1−4=−2,f(−1,−1)=1+1−4=−2f(1,1) = 1 + 1 - 4 = -2, \qquad f(-1,-1) = 1 + 1 - 4 = -2

Since x4+y4x^4 + y^4 eventually overwhelms 4xy4xy far from the origin, these two local minima are in fact the global minimum value of the function.

Frequently asked questions

Why are there two separate minimum points?

The cross term −4xy-4xy is most negative when xx and yy have the same sign and similar size. That happens on both the first-quadrant diagonal and the third-quadrant diagonal, so the function dips to the same depth of −2-2 in each.

What makes this a standard textbook example?

It puts a saddle and two separate minima in one problem, so the second derivative test has to be run three times with two different verdicts, and the critical point system has a hidden root at the origin that punishes dividing by xx. That combination is why it appears in so many Calc 3 courses.