Week 7 Discussion Notes

Table of Contents

Partial Derivatives

Partial derivatives are like usual derivatives, but you just think of the other variables as constants. For example,

xexy=exy(xxy)=exyy.\pderiv{}{x} e^{xy} = e^{xy} \p{\pderiv{}{x} xy} = e^{xy} y.

There's also a useful theorem about mixed partial derivatives:

Theorem (Clairaut's theorem)

If the mixed partial derivatives fxyf_{xy} (i.e., you take the xx-derivative first and then the yy-derivative second) and fyxf_{yx} (yy first and xx second) exist and are continuous in an open set around (a,b)\p{a, b}, then

fxy(a,b)=fyx(a,b).f_{xy}\p{a, b} = f_{yx}\p{a, b}.

In a nutshell, this theorem tells you that for most functions you can write down, you can take the partial derivatives in any order you want. Sometimes, one order is easier than the other:

Example 1.

Calculate fyyyyxf_{yyyyx}, where f(x,y)=ey2siny+xy4f\p{x, y} = e^{y^2\sin{y}} + xy^4.

Solution.

As you can imagine, if you take four yy-derivatives, you will get something very complicated. However, ff is differentiable infinitely many times, so you can apply Clairaut's theorem here:

fyyyyx(x,y)=fxyyyy(x,y)f_{yyyyx}\p{x, y} = f_{xyyyy}\p{x, y}

Immediately, you get

fx(x,y)=y4,f_x\p{x, y} = y^4,

since there's no xx in the first term. From here, you can easily take the remaining yy-derivatives:

fxyyyy(x,y)=24.f_{xyyyy}\p{x, y} = \boxed{24}.

Linearization

In one variable, the linearization of ff is just the tangent line:

f(x)f(x0)+f(x0)(xx0)f\p{x} \approx f\p{x_0} + f'\p{x_0}\p{x - x_0}

In several variables, a similar formula holds. We don't have "normal" derivatives in several variables, but instead of ff', we have the gradient

f(x0,y0)=fx(x0,y0),fy(x0,y0).\nabla f\p{x_0, y_0} = \ang{f_x\p{x_0, y_0}, f_y\p{x_0, y_0}}.

We can't multiply vectors though, but given two vectors, we can get a scalar by using the dot product, so we get

f(x,y)f(x0,y0)+f(x0,y0)xx0,yy0.f\p{x, y} \approx f\p{x_0, y_0} + \nabla f\p{x_0, y_0} \cdot \ang{x - x_0, y - y_0}.

The right-hand side is the equation of the tangent plane for ff at (x0,y0)\p{x_0, y_0}, and you can write it as

z=f(x0,y0)+f(x0,y0)xx0,yy0=f(x0,y0)+fx(x0,y0)(xx0)+fy(x0,y0)(yy0).\begin{aligned} z &= f\p{x_0, y_0} + \nabla f\p{x_0, y_0} \cdot \ang{x - x_0, y - y_0} \\ &= f\p{x_0, y_0} + f_x\p{x_0, y_0}\p{x - x_0} + f_y\p{x_0, y_0}\p{y - y_0}. \end{aligned}

Rearranging, this becomes

fx(x0,y0)x+fy(x0,y0)yz=d.f_x\p{x_0, y_0}x + f_y\p{x_0, y_0}y - z = d.

From here, you can read off the normal vector to the tangent plane:

n=fx(x0,y0),fy(x0,y0),1.\vec{n} = \ang{f_x\p{x_0, y_0}, f_y\p{x_0, y_0}, -1}.