Multivariable calculus
Partial Derivatives of e^(x^2 - y^2)
For f(x, y) = e^(x^2 - y^2), the partial with respect to x is 2x e^(x^2 - y^2) and the partial with respect to y is -2y e^(x^2 - y^2). The outer exponential never changes; only the inside derivative does, and it is 2x in one case and -2y in the other.
Chain rule with a saddle in the exponent
The exponent is the standard saddle. Hold fixed and ; hold fixed and . In both cases the exponential itself is copied down unchanged and multiplied by that inside derivative.
At the exponent is , so , and the slopes are and . Stepping in the positive direction raises the surface; stepping in the positive direction lowers it by the same amount.
The mistake: a dropped sign or a mishandled exponent
- Writing for the partial keeps the size but loses the sign. The exponent falls as grows, so for the slope in is negative.
- Writing differentiates the whole exponent in a single partial. Each partial sees only its own variable move.
- Writing copies the exponent down in front instead of its derivative, and writing pushes the inside derivative up into the exponent. The inside derivative is , and it multiplies the exponential; the exponential itself is copied unchanged.
Use the axes as a check. Along the function is , whose slope is , and along it is , whose slope is . Both agree with the general formulas, and a candidate answer that fails these one-variable cases is wrong.
The origin is still a saddle
The exponential is never zero, so the gradient vanishes only where and , that is, only at the origin, where .
That point is a saddle. Along the -axis the function is , a valley bottom that rises away from the origin; along the -axis it is , a ridge that falls away. Increasing in one direction and decreasing in the other at the same point is exactly what a saddle means.
This is a useful general fact: composing a function with an increasing outer function such as changes the heights but not the pattern of increase and decrease, so the saddle of survives the exponential intact.
Frequently asked questions
Where is the gradient of e^(x^2-y^2) zero?
Only at the origin. The exponential factor is never zero, so 2x and -2y must both vanish, which happens at the single point (0, 0), where the value is 1.
Does the exponential change the shape of the saddle?
It stretches the heights but keeps the pattern. Since e^u is increasing, the function rises exactly where x^2 - y^2 rises, so the origin stays a saddle point.