Partial Differentiation

14 Partial Differentiation 14.1 Fun tions of Several Variables In single-variable calculus we were concerned with functions that map the real numbers R to R, sometimes called “real functions of one variable”, meaning the “input” is a single real number and the “output” is likewise a single real number. In the last chapter we considered functions taking a real number to a vector, which may also be viewed as functions f : R → R3 , that is, for each input value we get a position in space. Now we turn to functions of several variables, meaning several input variables, functions f : Rn → R. We will deal primarily with n = 2 and to a lesser extent n = 3; in fact many of the techniques we discuss can be applied to larger values of n as well. A function f : R2 → R maps a pair of values (x, y) to a single real number. The threedimensional coordinate system we have already used is a convenient way to visualize such functions: above each point (x, y) in the x-y plane we graph the point (x, y, z), where of course z = f (x, y). EXAMPLE 14.1.1 Consider f (x, y) = 3x + 4y − 5. Writing this as z = 3x + 4y − 5 and then 3x +4y −z = 5 we recognize the equation of a plane. In the form f (x, y) = 3x +4y −5 the emphasis has shifted: we now think of x and y as independent variables and z as a variable dependent on them, but the geometry is unchanged. EXAMPLE 14.1.2 We have seen that x2 + y 2 + z 2 = 4 represents a sphere of radius 2. We cannot write this in the form f (x, y), since for each x and y in the disk x2 +y 2 < 4 there are two corresponding points on the sphere. As with the equation of a circle, we can resolve 349 350 Chapter 14 Partial Differentiation p p this equation into two functions, f (x, y) = 4 − x2 − y 2 and f (x, y) = − 4 − x2 − y 2 , representing the upper and lower hemispheres. Each of these is an example of a function with a restricted domain: only certain values of x and y make sense (namely, those for which x2 + y 2 ≤ 4) and the graphs of these functions are limited to a small region of the plane. √ √ EXAMPLE 14.1.3 Consider f = x + y. This function is defined only when both √ x and y are non-negative. When y = 0 we get f (x, y) = x, the familiar square root function in the x-z plane, and when x = 0 we get the same curve in the y-z plane. Generally speaking, we see that starting from f (0, 0) = 0 this function gets larger in every direction in roughly the same way that the square root function gets larger. For example, √ if we restrict attention to the line x = y, we get f (x, y) = 2 x and along the line y = 2x √ √ √ √ we have f (x, y) = x + 2x = (1 + 2) x. 6 5 4 3 10.0 2 7.5 1 0 0.0 5.0 y 2.5 2.5 5.0 x 7.5 10.0 Figure 14.1.1 0.0 f (x, y) = √ x+ √ y (AP) A computer program that plots such surfaces can be very useful, as it is often difficult to get a good idea of what they look like. Still, it is valuable to be able to visualize relatively simple surfaces without such aids. As in the previous example, it is often a good idea to examine the function on restricted subsets of the plane, especially lines. It can also be useful to identify those points (x, y) that share a common z-value. EXAMPLE 14.1.4 Consider f (x, y) = x2 + y 2 . When x = 0 this becomes f = y 2 , a parabola in the y-z plane; when y = 0 we get the “same” parabola f = x2 in the x-z plane. Now consider the line y = kx. If we simply replace y by kx we get f (x, y) = (1 + k 2 )x2 which is a parabola, but it does not really “represent” the cross-section along y = kx, because the cross-section has the line y = kx where the horizontal axis should be. In 14.1 Functions of Several Variables 351 order to pretend that this line is the horizontal axis, we need to write the function in p p terms of the distance from the origin, which is x2 + y 2 = x2 + k 2 x2 . Now f (x, y) = p x2 + k 2 x2 = ( x2 + k 2 x2 )2 . So the cross-section is the “same” parabola as in the x-z and y-z planes, namely, the height is always the distance from the origin squared. This means that f (x, y) = x2 + y 2 can be formed by starting with z = x2 and rotating this curve around the z axis. Finally, picking a value z = k, at what points does f (x, y) = k? This means x2 +y 2 = k, √ which we recognize as the equation of a circle of radius k. So the graph of f (x, y) has parabolic cross-sections, and the same height everywhere on concentric circles with center at the origin. This fits with what we have already discovered. 8 6 4 2 −3 0 −3 −2 −2 −1 −1 0 0 1 1 2 2 3 3 Figure 14.1.2 f (x, y) = x2 + y2 (AP) As in this example, the points (x, y) such that f (x, y) = k usually form a curve, called a level curve of the function. A graph of some level curves can give a good idea of the shape of the surface; it looks much like a topographic map of the surface. In figure 14.1.2 both the surface and its associated level curves are shown. Note that, as with a topographic map, the heights corresponding to the level curves are evenly spaced, so that where curves are closer together the surface is steeper. Functions f : Rn → R behave much like functions of two variables; we will on occasion discuss functions of three variables. The principal difficulty with such functions is visualizing them, as they do not “fit” in the three dimensions we are familiar with. For three variables there are various ways to interpret functions that make them easier to understand. For example, f (x, y, z) could represent the temperature at the point (x, y, z), or the pressure, or the strength of a magnetic field. It remains useful to consider those points at which f (x, y, z) = k, where k is some constant value. If f (x, y, z) is temperature, the set of points (x, y, z) such that f (x, y, z) = k is the collection of points in space with temperature 352 Chapter 14 Partial Differentiation k; in general this is called a level set; for three variables, a level set is typically a surface, called a level surface. 2 2 2 EXAMPLE 14.1.5 Suppose the temperature at (x, y, z) is T (x, y, z) = e−(x +y +z ) . This function has a maximum value of 1 at the origin, and tends to 0 in all directions. If k is positive and at most 1, the set of points for which T (x, y, z) = k is those points satisfying x2 + y 2 + z 2 = − ln k, a sphere centered at the origin. The level surfaces are the concentric spheres centered at the origin. Exercises 14.1. 1. Let f (x, y) = (x − y)2 . Determine the equations and shapes of the cross-sections when x = 0, y = 0, x = y, and describe the level curves. Use a three-dimensional graphing tool to graph the surface. ⇒ 2. Let f (x, y) = |x| + |y|. Determine the equations and shapes of the cross-sections when x = 0, y = 0, x = y, and describe the level curves. Use a three-dimensional graphing tool to graph the surface. ⇒ 2 2 3. Let f (x, y) = e−(x +y ) sin(x2 +y2 ). Determine the equations and shapes of the cross-sections when x = 0, y = 0, x = y, and describe the level curves. Use a three-dimensional graphing tool to graph the surface. ⇒ 4. Let f (x, y) = sin(x − y). Determine the equations and shapes of the cross-sections when x = 0, y = 0, x = y, and describe the level curves. Use a three-dimensional graphing tool to graph the surface. ⇒ 5. Let f (x, y) = (x2 − y2 )2 . Determine the equations and shapes of the cross-sections when x = 0, y = 0, x = y, and describe the level curves. Use a three-dimensional graphing tool to graph the surface. ⇒ 6. Find the domain of each of the following functions of two variables: p p a. 9 − x2 + y2 − 4 b. arcsin(x2 + y2 − 2) p 16 − x2 − 4y2 c. ⇒ 7. Below are two sets of level curves. One is for a cone, one is for a paraboloid. Which is which? Explain. 14.2 14.2 Limits and Continuity 353 Limits and Continuity To develop calculus for functions of one variable, we needed to make sense of the concept of a limit, which we needed to understand continuous functions and to define the derivative. Limits involving functions of two variables can be considerably more difficult to deal with; fortunately, most of the functions we encounter are fairly easy to understand. The potential difficulty is largely due to the fact that there are many ways to “approach” a point in the x-y plane. If we want to say that lim f (x, y) = L, we need to (x,y)→(a,b) capture the idea that as (x, y) gets close to (a, b) then f (x, y) gets close to L. For functions of one variable, f (x), there are only two ways that x can approach a: from the left or right. But there are an infinite number of ways to approach (a, b): along any one of an infinite number of lines, or an infinite number of parabolas, or an infinite number of sine curves, and so on. We might hope that it’s really not so bad—suppose, for example, that along every possible line through (a, b) the value of f (x, y) gets close to L; surely this means that “f (x, y) approaches L as (x, y) approaches (a, b)”. Sadly, no. EXAMPLE 14.2.1 Consider f (x, y) = xy 2 /(x2 + y 4 ). When x = 0 or y = 0, f (x, y) is 0, so the limit of f (x, y) approaching the origin along either the x or y axis is 0. Moreover, along the line y = mx, f (x, y) = m2 x3 /(x2 + m4 x4 ). As x approaches 0 this expression approaches 0 as well. So along every line through the origin f (x, y) approaches 0. Now suppose we approach the origin along x = y 2 . Then f (x, y) = y4 1 y2y2 = = , 4 4 4 y +y 2y 2 so the limit is 1/2. Looking at figure 14.2.1, it is apparent that there is a ridge above x = y 2 . Approaching the origin along a straight line, we go over the ridge and then drop down toward 0, but approaching along the ridge the height is a constant 1/2. Thus, there is no limit at (0, 0). Fortunately, we can define the concept of limit without needing to specify how a particular point is approached—indeed, in definition 2.3.2, we didn’t need the concept of “approach.” Roughly, that definition says that when x is close to a then f (x) is close to L; there is no mention of “how” we get close to a. We can adapt that definition to two variables quite easily: DEFINITION 14.2.2 Limit Suppose f (x, y) is a function. We say that lim (x,y)→(a,b) f (x, y) = L if for every ǫ > 0 there is a δ > 0 so that whenever 0 < |f (x, y) − L| < ǫ. p (x − a)2 + (y − b)2 < δ, 354 Chapter 14 Partial Differentiation Figure 14.2.1 f (x, y) = xy2 (AP) x2 + y4 This says that we can make |f (x, y) − L| < ǫ, no matter how small ǫ is, by making the distance from (x, y) to (a, b) “small enough”. EXAMPLE 14.2.3 We show that 3x2 y = 0. Suppose ǫ > 0. Then (x,y)→(0,0) x2 + y 2 lim x2 3x2 y = 3|y|. x2 + y 2 x2 + y 2 p p Note that x2 /(x2 + y 2 ) ≤ 1 and |y| = y 2 ≤ x2 + y 2 < δ. So x2 3|y| < 1 · 3 · δ. x2 + y 2 14.2 Limits and Continuity 355 We want to force this to be less than ǫ by picking δ “small enough.” If we choose δ = ǫ/3 then ǫ 3x2 y < 1 · 3 · = ǫ. 2 2 x +y 3 Recall that a function f (x) is continuous at x = a if lim f (x) = f (a); roughly this x→a says that there is no “hole” or “jump” at x = a. We can say exactly the same thing about a function of two variables. DEFINITION 14.2.4 f (x, y) is continuous at (a, b) if lim (x,y)→(a,b) f (x, y) = f (a, b). EXAMPLE 14.2.5 The function f (x, y) = 3x2 y/(x2 + y 2 ) is not continuous at (0, 0), because f (0, 0) is not defined. However, we know that lim f (x, y) = 0, so we can (x,y)→(0,0) easily “fix” the problem, by extending the definition of f so that f (0, 0) = 0. This surface is shown in figure 14.2.2. -3 -1 1.8 1 0.8 3 3 -0.2 1 -1.2 -2.2 Figure 14.2.2 -1 -3 f (x, y) = 3x2 y (AP) x2 + y2 356 Chapter 14 Partial Differentiation Note that in contrast to this example we cannot fix example 14.2.1 at (0, 0) because the limit does not exist. No matter what value we try to assign to f at (0, 0) the surface will have a “jump” there. Fortunately, the functions we will examine will typically be continuous almost everywhere. Usually this follows easily from the fact that closely related functions of one variable are continuous. As with single variable functions, two classes of common functions are particularly useful and easy to describe. A polynomial in two variables is a sum of terms of the form axm y n , where a is a real number and m and n are non-negative integers. A rational function is a quotient of polynomials. THEOREM 14.2.6 Polynomials are continuous everywhere. Rational functions are continuous everywhere they are defined. Exercises 14.2. Determine whether each limit exists. If it does, find the limit and prove that it is the limit; if it does not, explain how you know. x2 ⇒ 1. lim (x,y)→(0,0) x2 + y 2 xy ⇒ 2. lim (x,y)→(0,0) x2 + y 2 xy 3. lim ⇒ 2 (x,y)→(0,0) 2x + y 2 4. x4 − y4 ⇒ (x,y)→(0,0) x2 + y 2 lim sin(x2 + y2 ) ⇒ (x,y)→(0,0) x2 + y2 xy p 6. lim ⇒ (x,y)→(0,0) 2x2 + y2 5. lim 2 2 7. e−x −y − 1 ⇒ (x,y)→(0,0) x2 + y2 8. x3 + y3 ⇒ (x,y)→(0,0) x2 + y 2 9. x2 + sin2 y ⇒ (x,y)→(0,0) 2x2 + y 2 lim lim lim (x − 1)2 ln x ⇒ (x,y)→(1,0) (x − 1)2 + y 2 11. lim 3x + 4y ⇒ 10. lim (x,y)→(1,−1) 12. 4x2 y ⇒ (x,y)→(0,0) x2 + y 2 lim 14.3 13. Does the function f (x, y) = x−y ? Explain. 1 + x2 + y2 14.3 Partial Differentiation 357 x−y have any discontinuities? What about f (x, y) = 1+x+y Partial Differentiation When we first considered what the derivative of a vector function might mean, there was really not much difficulty in understanding either how such a thing might be computed or what it might measure. In the case of functions of two variables, things are a bit harder to understand. If we think of a function of two variables in terms of its graph, a surface, there is a more-or-less obvious derivative-like question we might ask, namely, how “steep” is the surface. But it’s not clear that this has a simple answer, nor how we might proceed. We will start with what seem to be very small steps toward the goal; surprisingly, it turns out that these simple ideas hold the keys to a more general understanding. 8 6 4 2 −3 0 −3 Figure 14.3.1 −2 −1 −2 0 −1 1 0 1 x 2 2 3 f (x, y) = x2 + y2 , cut by the plane x + y = 1 (AP) Imagine a particular point on a surface; what might we be able to say about how steep it is? We can limit the question to make it more familiar: how steep is the surface in a particular direction? What does this even mean? Here’s one way to think of it: Suppose we’re interested in the point (a, b, c). Pick a straight line in the x-y plane through the point (a, b, 0), then extend the line vertically into a plane. Look at the intersection of the 358 Chapter 14 Partial Differentiation plane with the surface. If we pay attention to just the plane, we see the chosen straight line where the x-axis would normally be, and the intersection with the surface shows up as a curve in the plane. Figure 14.3.1 shows the parabolic surface from figure 14.1.2, exposing its cross-section above the line x + y = 1. In principle, this is a problem we know how to solve: find the slope of a curve in a plane. Let’s start by looking at some particularly easy lines: those parallel to the x or y axis. Suppose we are interested in the cross-section of f (x, y) above the line y = b. If we substitute b for y in f (x, y), we get a function in one variable, describing the height of the cross-section as a function of x. Because y = b is parallel to the x-axis, if we view it from a vantage point on the negative y-axis, we will see what appears to be simply an ordinary curve in the x-z plane. 8.8 8.0 7.2 6.4 8 5.6 4.8 4.0 6 3.2 2.4 z −3 4 1.6 −2 0.8 −1 y 0 2 0.0 −3 −2 −1 0 1 2 3 1 2 0 3 2 1 0 −1 −2 −3 x Figure 14.3.2 f (x, y) = x2 + y2 , cut by the plane y = 2 (AP) Consider again the parabolic surface f (x, y) = x2 +y 2 . The cross-section above the line y = 2 consists of all points (x, 2, x2 + 4). Looking at this cross-section from somewhere on the negative y axis, we see what appears to be just the curve f (x) = x2 + 4. At any point on the cross-section, (a, 2, a2 + 4), the steepness of the surface in the direction of the line y = 2 is simply the slope of the curve f (x) = x2 + 4 at x = a, namely 2a. Figure 14.3.2 shows the same parabolic surface as before, but now cut by the plane y = 2. The left graph shows the cut-off surface, the right shows just the cross-section, looking up from the negative y-axis toward the origin. 14.3 Partial Differentiation 359 If, say, we’re interested in the point (−1, 2, 5) on the surface, then the slope in the direction of the line y = 2 is 2x = 2(−1) = −2. This means that starting at (−1, 2, 5) and moving on the surface, above the line y = 2, in the direction of increasing x values, the surface goes down; of course moving in the opposite direction, toward decreasing x values, the surface will rise. If we’re interested in some other line y = k, there is really no change in the computation. The equation of the cross-section above y = k is x2 + k 2 with derivative 2x. We can save ourselves the effort, small as it is, of substituting k for y: all we are in effect doing is temporarily assuming that y is some constant. With this assumption, the derivative d 2 2 dx (x + y ) = 2x. To emphasize that we are only temporarily assuming y is constant, we ∂ use a slightly different notation: ∂x (x2 + y 2 ) = 2x; the “∂” reminds us that there are more variables than x, but that only x is being treated as a variable. We read the equation as “the partial derivative of (x2 + y 2 ) with respect to x is 2x.” A convenient alternate notation for the partial derivative of f (x, y) with respect to x is fx (x, y). EXAMPLE 14.3.1 The partial derivative with respect to x of x3 + 3xy is 3x2 + 3y. Note that the partial derivative includes the variable y, unlike the example x2 + y 2 . It is somewhat unusual for the partial derivative to depend on a single variable; this example is more typical. Of course, we can do the same sort of calculation for lines parallel to the y-axis. We temporarily hold x constant, which gives us the equation of the cross-section above a line x = k. We can then compute the derivative with respect to y; this will measure the steepness of the curve in the y direction. EXAMPLE 14.3.2 is fy (x, y) = The partial derivative with respect to y of f (x, y) = sin(xy) + 3xy ∂ ∂ sin(xy) + 3xy = cos(xy) (xy) + 3x = x cos(xy) + 3x. ∂y ∂y So far, using no new techniques, we have succeeded in measuring the slope of a surface in two quite special directions. For functions of one variable, the derivative is closely linked to the notion of tangent line. For surfaces, the analogous idea is the tangent plane—a plane that just touches a surface at a point, and has the same “steepness” as the surface in all directions. Even though we haven’t yet figured out how to compute the slope in all directions, we have enough information to find tangent planes. Suppose we want the plane tangent to a surface at a particular point (a, b, c). If we compute the two partial derivatives of the function for that point, we get enough information to determine two lines tangent to the surface, both through (a, b, c) and both tangent to the surface in their 360 Chapter 14 Partial Differentiation Figure 14.3.3 Tangent vectors and tangent plane. (AP) respective directions. These two lines determine a plane, that is, there is exactly one plane containing the two lines: the tangent plane. Figure 14.3.3 shows (part of) two tangent lines at a point, and the tangent plane containing them. How can we discover an equation for this tangent plane? We know a point on the plane, (a, b, c); we need a vector normal to the plane. If we can find two vectors, one parallel to each of the tangent lines we know how to find, then the cross product of these vectors will give the desired normal vector. z 3 2 ... ..... ..... ..... . . . . .. ..... ..... ..... ..... . . . . ...... ...... ...... .......................... . . . . . . . ... .................. x ......... ........... ........... ................. ........... . . . . . . . . . . . . . . . . ................................... f (2, b) 1 x 0 0 1 Figure 14.3.4 2 3 4 A tangent vector. How can we find vectors parallel to the tangent lines? Consider first the line tangent to the surface above the line y = b. A vector hu, v, wi parallel to this tangent line must have y component v = 0, and we may as well take the x component to be u = 1. The ratio 14.3 Partial Differentiation 361 of the z component to the x component is the slope of the tangent line, precisely what we know how to compute. The slope of the tangent line is fx (a, b), so fx (a, b) = w w = = w. u 1 In other words, a vector parallel to this tangent line is h1, 0, fx(a, b)i, as shown in figure 14.3.4. If we repeat the reasoning for the tangent line above x = a, we get the vector h0, 1, fy (a, b)i. Now to find the desired normal vector we compute the cross product, h0, 1, fy i × h1, 0, fxi = hfx , fy , −1i. From our earlier discussion of planes, we can write down the equation we seek: fx (a, b)x + fy (a, b)y − z = k, and k as usual can be computed by substituting a known point: fx (a, b)(a) + fy (a, b)(b) − c = k. There are various more-orless nice ways to write the result: fx (a, b)x + fy (a, b)y − z = fx (a, b)a + fy (a, b)b − c fx (a, b)x + fy (a, b)y − fx (a, b)a − fy (a, b)b + c = z fx (a, b)(x − a) + fy (a, b)(y − b) + c = z fx (a, b)(x − a) + fy (a, b)(y − b) + f (a, b) = z √ Find the plane tangent to x2 + y 2 + z 2 = 4 at (1, 1, 2). This p point is on the upper hemisphere, so we use f (x, y) = 4 − x2 − y 2 . Then fx (x, y) = √ −x(4 − x2 − y 2 )−1/2 and fy (x, y) = −y(4 − x2 − y 2 )−1/2 , so fx (1, 1) = fy (1, 1) = −1/ 2 and the equation of the plane is EXAMPLE 14.3.3 √ 1 1 z = − √ (x − 1) − √ (y − 1) + 2. 2 2 The hemisphere and this tangent plane are pictured in figure 14.3.3. So it appears that to find a tangent plane, we need only find two quite simple ordinary derivatives, namely fx and fy . This is true if the tangent plane exists. It is, unfortunately, not always the case that if fx and fy exist there is a tangent plane. Consider the function xy 2 /(x2 + y 4 ) pictured in figure 14.2.1. This function has value 0 when x = 0 or y = 0, and we can “plug the hole” by agreeing that f (0, 0) = 0. Now it’s clear that fx (0, 0) = fy (0, 0) = 0, because in the x and y directions the surface is simply a horizontal line. But it’s also clear from the picture that this surface does not have anything that deserves to be called a “tangent plane” at the origin, certainly not the x-y plane containing these two tangent lines. 362 Chapter 14 Partial Differentiation When does a surface have a tangent plane at a particular point? What we really want from a tangent plane, as from a tangent line, is that the plane be a “good” approximation of the surface near the point. Here is how we can make this precise: DEFINITION 14.3.4 Let ∆x = x − x0 , ∆y = y − y0 , and ∆z = z − z0 where z0 = f (x0 , y0 ). The function z = f (x, y) is differentiable at (x0 , y0 ) if ∆z = fx (x0 , y0 )∆x + fy (x0 , y0 )∆y + ǫ1 ∆x + ǫ2 ∆y, and both ǫ1 and ǫ2 approach 0 as (x, y) approaches (x0 , y0 ). This definition takes a bit of absorbing. Let’s rewrite the central equation a bit: z = fx (x0 , y0 )(x − x0 ) + fy (x0 , y0 )(y − y0 ) + f (x0 , y0 ) + ǫ1 ∆x + ǫ2 ∆y. (14.3.1) The first three terms on the right give the value of z on the tangent plane, that is, fx (x0 , y0 )(x − x0 ) + fy (x0 , y0 )(y − y0 ) + f (x0 , y0 ) is the z-value of the point on the plane above (x, y). Equation 14.3.1 says that the z-value of a point on the surface is equal to the z-value of a point on the plane plus a “little bit,” namely ǫ1 ∆x + ǫ2 ∆y. As (x, y) approaches (x0 , y0 ), both ∆x and ∆y approach 0, so this little bit ǫ1 ∆x + ǫ2 ∆y also approaches 0, and the z-values on the surface and the plane get close to each other. But that by itself is not very interesting: since the surface and the plane both contain the point (x0 , y0 , z0 ), the z values will approach z0 and hence get close to each other whether the tangent plane is “tangent” to the surface or not. The extra condition in the definition says that as (x, y) approaches (x0 , y0 ), the ǫ values approach 0—this means that ǫ1 ∆x + ǫ2 ∆y approaches 0 much, much faster, because ǫ1 ∆x is much smaller than either ǫ1 or ∆x. It is this extra condition that makes the plane a tangent plane. We can see that the extra condition on ǫ1 and ǫ2 fits neatly with the definition of partial derivatives. Suppose we temporarily fix y = y0 , so ∆y = 0. Then the equation from the definition becomes ∆z = fx (x0 , y0 )∆x + ǫ1 ∆x or ∆z = fx (x0 , y0 ) + ǫ1 . ∆x Now taking the limit of the two sides as ∆x approaches 0, the left side turns into the partial derivative of z with respect to x at (x0 , y0 ), or in other words fx (x0 , y0 ), and the 14.3 Partial Differentiation 363 right side does the same, because as (x, y) approaches (x0 , y0 ), ǫ1 approaches 0. Essentially the same calculation works for fy . Almost all of the functions we will encounter are differentiable at points we will be interested in, and often at all points. This is usually because the functions satisfy the hypotheses of this theorem. THEOREM 14.3.5 If f (x, y) and its partial derivatives are continuous at a point (x0 , y0 ), then f is differentiable there. Exercises 14.3. 1. Find fx and fy where f (x, y) = cos(x2 y) + y3 . ⇒ xy 2. Find fx and fy where f (x, y) = 2 . ⇒ x +y 3. Find fx and fy where f (x, y) = ex 2 +y 2 . ⇒ 4. Find fx and fy where f (x, y) = xy ln(xy). ⇒ p 5. Find fx and fy where f (x, y) = 1 − x2 − y2 . ⇒ 6. Find fx and fy where f (x, y) = x tan(y). ⇒ 1 . ⇒ 7. Find fx and fy where f (x, y) = xy 8. Find an equation for the plane tangent to 2x2 + 3y2 − z 2 = 4 at (1, 1, −1). ⇒ 9. Find an equation for the plane tangent to f (x, y) = sin(xy) at (π, 1/2, 1). ⇒ 10. Find an equation for the plane tangent to f (x, y) = x2 + y3 at (3, 1, 10). ⇒ 11. Find an equation for the plane tangent to f (x, y) = x ln(xy) at (2, 1/2, 0). ⇒ 12. Find an equation for the line normal to x2 + 4y2 = 2z at (2, 1, 4). ⇒ 13. Explain in your own words why, when taking a partial derivative of a function of multiple variables, we can treat the variables not being differentiated as constants. 14. Consider a differentiable function, f (x, y). Give physical interpretations of the meanings of fx (a, b) and fy (a, b) as they relate to the graph of f . 15. In much the same way that we used the tangent line to approximate the value of a function from single variable calculus, we can use the tangent plane to approximate a function from multivariable calculus. Consider the tangent plane found in Exercise 11. Use this plane to approximate f (1.98, 0.4). ⇒ 16. Suppose that one of your colleagues has calculated the partial derivatives of a given function, and reported to you that fx (x, y) = 2x + 3y and that fy (x, y) = 4x + 6y. Do you believe them? Why or why not? If not, what answer might you have accepted for fy ? 17. Suppose f (t) and g(t) are single variable differentiable functions. Find ∂z/∂x and ∂z/∂y for each of the following two variable functions. a. z = f (x)g(y) b. z = f (xy) c. z = f (x/y) 364 Chapter 14 Partial Differentiation ⇒ 14.4 The Chain Rule Consider the surface z = x2 y + xy 2 , and suppose that x = 2 + t4 and y = 1 − t3 . We can think of the latter two equations as describing how x and y change relative to, say, time. Then z = x2 y + xy 2 = (2 + t4 )2 (1 − t3 ) + (2 + t4 )(1 − t3 )2 tells us explicitly how the z coordinate of the corresponding point on the surface depends on t. If we want to know dz/dt we can compute it more or less directly—it’s actually a bit simpler to use the chain rule: dz = x2 y ′ + 2xx′ y + x2yy ′ + x′ y 2 dt = (2xy + y 2 )x′ + (x2 + 2xy)y ′ = (2(2 + t4 )(1 − t3 ) + (1 − t3 )2 )(4t3 ) + ((2 + t4 )2 + 2(2 + t4 )(1 − t3 ))(−3t2 ) If we look carefully at the middle step, dz/dt = (2xy + y 2 )x′ + (x2 + 2xy)y ′ , we notice that 2xy + y 2 is ∂z/∂x, and x2 + 2xy is ∂z/∂y. This turns out to be true in general, and gives us a new chain rule: THEOREM 14.4.1 Suppose that z = f (x, y), f is differentiable, x = g(t), and y = h(t). Assuming that the relevant derivatives exist, ∂z dx ∂z dy dz = + . dt ∂x dt ∂y dt Proof. If f is differentiable, then ∆z = fx (x0 , y0 )∆x + fy (x0 , y0 )∆y + ǫ1 ∆x + ǫ2 ∆y, where ǫ1 and ǫ2 approach 0 as (x, y) approaches (x0 , y0 ). Then ∆x ∆y ∆x ∆y ∆z = fx + fy + ǫ1 + ǫ2 . ∆t ∆t ∆t ∆t ∆t (14.4.1) 14.4 The Chain Rule 365 As ∆t approaches 0, (x, y) approaches (x0 , y0 ) and so dz ∆z = ∆t→0 ∆t dt ∆x dx lim ǫ1 =0· ∆t→0 ∆t dt ∆y dy lim ǫ2 =0· ∆t→0 ∆t dt lim and so taking the limit of (14.4.1) as ∆t goes to 0 gives dx dy dz = fx + fy , dt dt dt as desired. We can write the chain rule in way that is somewhat closer to the single variable chain rule: df = hfx , fy i · hx′ , y ′ i, dt or (roughly) the derivatives of the outside function “times” the derivatives of the inside functions. Not surprisingly, essentially the same chain rule works for functions of more than two variables, for example, given a function of three variables f (x, y, z), where each of x, y and z is a function of t, df = hfx , fy , fz i · hx′ , y ′ , z ′ i. dt We can even extend the idea further. Suppose that f (x, y) is a function and x = g(s, t) and y = h(s, t) are functions of two variables s and t. Then f is “really” a function of s and t as well, and ∂f ∂f = fx gs + fy hs = fx gt + fy ht . ∂s ∂t The natural extension of this to f (x, y, z) works as well. Recall that we used the ordinary chain rule to do implicit differentiation. We can do the same with the new chain rule. EXAMPLE 14.4.2 x2 + y 2 + z 2 = 4 defines a sphere, which is not a function of x and y, though it can be thought of as two functions, the top and bottom hemispheres. We can think of z as one of these two functions, so really z = z(x, y), and we can think of x 366 Chapter 14 Partial Differentiation and y as particularly simple functions of x and y, and let f (x, y, z) = x2 + y 2 + z 2 . Since f (x, y, z) = 4, ∂f /∂x = 0, but using the chain rule: 0= ∂x ∂y ∂z ∂f = fx + fy + fz ∂x ∂x ∂x ∂x = (2x)(1) + (2y)(0) + (2z) ∂z , ∂x noting that since y is temporarily held constant its derivative ∂y/∂x = 0. Now we can solve for ∂z/∂x: ∂z 2x x =− =− . ∂x 2z z In a similar manner we can compute ∂z/∂y. Exercises 14.4. 1. Use the chain rule to compute dz/dt for z = sin(x2 + y2 ), x = t2 + 3, y = t3 . ⇒ 2. Use the chain rule to compute dz/dt for z = x2 y, x = sin(t), y = t2 + 1. ⇒ 3. Use the chain rule to compute ∂z/∂s and ∂z/∂t for z = x2 y, x = sin(st), y = t2 + s2 . ⇒ 4. Use the chain rule to compute ∂z/∂s and ∂z/∂t for z = x2 y2 , x = st, y = t2 − s2 . ⇒ 5. Use the chain rule to compute ∂z/∂x and ∂z/∂y for 2x2 + 3y2 − 2z 2 = 9. ⇒ 6. Use the chain rule to compute ∂z/∂x and ∂z/∂y for 2x2 + y2 + z 2 = 9. ⇒ 7. Use the chain rule to compute ∂z/∂x and ∂z/∂y for xy2 + z 2 = 5. ⇒ 8. Use the chain rule to compute ∂z/∂x and ∂z/∂y for 2 sin(xyz) = 1. ⇒ 9. Chemistry students will recognize the ideal gas law , given by P V = nRT which relates the Pressure, Volume, and Temperature of n moles of gas. (R is the ideal gas constant). Thus, we can view pressure, volume, and temperature as variables, each one dependent on the other two. a. If pressure of a gas is increasing at a rate of 0.2P a/min and temperature is increasing at a rate of 1K/min, how fast is the volume changing? b. If the volume of a gas is decreasing at a rate of 0.3m3 /min and temperature is increasing at a rate of .5K/min, how fast is the pressure changing? c. If the pressure of a gas is decreasing at a rate of 0.4P a/min and the volume is increasing at a rate of 3L/min, how fast is the temperature changing? ⇒ 10. Verify the following identity in the case of the ideal gas law: ∂P ∂V ∂T = −1 ∂V ∂T ∂P 11. The previous exercise was a special case of the following fact, which you are to verify here: If F (x, y, z) is a function of 3 variables, and the relation F (x, y, z) = 0 defines each of the 14.5 Directional Derivatives 367 variables in terms of the other two, namely x = f (y, z), y = g(x, z) and z = h(x, y), then ∂x ∂y ∂z = −1 ∂y ∂z ∂x 14.5 Dire tional Derivatives We still have not answered one of our first questions about the steepness of a surface: starting at a point on a surface given by f (x, y), and walking in a particular direction, how steep is the surface? We are now ready to answer the question. We already know roughly what has to be done: as shown in figure 14.3.1, we extend a line in the x-y plane to a vertical plane, and we then compute the slope of the curve that is the cross-section of the surface in that plane. The major stumbling block is that what appears in this plane to be the horizontal axis, namely the line in the x-y plane, is not an actual axis—we know nothing about the “units” along the axis. Our goal is to make this line into a t axis; then we need formulas to write x and y in terms of this new variable t; then we can write z in terms of t since we know z in terms of x and y; and finally we can simply take the derivative. So we need to somehow “mark off” units on the line, and we need a convenient way to refer to the line in calculations. It turns out that we can accomplish both by using the vector form of a line. Suppose that u is a unit vector hu1 , u2 i in the direction of interest. A vector equation for the line through (x0 , y0 ) in this direction is v(t) = hu1 t + x0 , u2 t + y0 i. The height of the surface above the point (u1 t + x0 , u2 t + y0 ) is g(t) = f (u1 t + x0 , u2 t + y0 ). Because u is a unit vector, the value of t is precisely the distance along the line from (x0 , y0 ) to (u1 t + x0 , u2 t + y0 ); this means that the line is effectively a t axis, with origin at the point (x0 , y0 ), so the slope we seek is g ′ (0) = hfx (x0 , y0 ), fy (x0 , y0 )i · hu1 , u2 i = hfx , fy i · u = ∇f · u d d (u1 t+x0 ) = u1 and dt (u2 t+y0 ) = u2 . Here we have used the chain rule and the derivatives dt The vector hfx , fy i is very useful, so it has its own symbol, ∇f , pronounced “del f”; it is also called the gradient of f . EXAMPLE 14.5.1 Find the slope of z = x2 + y 2 at (1, 2) in the direction of the vector h3, 4i. We first compute the gradient at (1, 2): ∇f = h2x, 2yi, which is h2, 4i at (1, 2). A unit vector in the desired direction is h3/5, 4/5i, and the desired slope is then h2, 4i·h3/5, 4/5i = 6/5 + 16/5 = 22/5. 368 Chapter 14 Partial Differentiation When doing such problems, it is easy to forget that we require a unit vector in the calculation ∇f · u. You may prefer to remember that this can always be written as ∇f · v/|v|. In the previous example, we might then have computed h2, 4i · h3, 4i/|h3, 4i|, rather than remembering to first compute u = h3, 4i/|h3, 4i|. EXAMPLE 14.5.2 Find a tangent vector to z = x2 + y 2 at (1, 2) in the direction of the vector h3, 4i and show that it is parallel to the tangent plane at that point. Since h3/5, 4/5i is a unit vector in the desired direction, we can easily expand it to a tangent vector simply by adding the third coordinate computed in the previous example: h3/5, 4/5, 22/5i. To see that this vector is parallel to the tangent plane, we can compute its dot product with a normal to the plane. We know that a normal to the tangent plane is hfx (1, 2), fy (1, 2), −1i = h2, 4, −1i, and the dot product is h2, 4, −1i · h3/5, 4/5, 22/5i = 6/5 + 16/5 − 22/5 = 0, so the two vectors are perpendicular. (Note that the vector normal to the surface, namely hfx , fy , −1i, is simply the gradient with a −1 tacked on as the third component.) The slope of a surface given by z = f (x, y) in the direction of a (two-dimensional) unit vector u is called the directional derivative of f , written Du f . The directional derivative immediately provides us with some additional information. We know that Du f = ∇f · u = |∇f ||u| cos θ = |∇f | cos θ if u is a unit vector; θ is the angle between ∇f and u. This tells us immediately that the largest value of Du f occurs when cos θ = 1, namely, when θ = 0, so ∇f is parallel to u. In other words, the gradient ∇f points in the direction of steepest ascent of the surface, and |∇f | is the slope in that direction. Likewise, the smallest value of Du f occurs when cos θ = −1, namely, when θ = π, so ∇f is anti-parallel to u. In other words, −∇f points in the direction of steepest descent of the surface, and −|∇f | is the slope in that direction. EXAMPLE 14.5.3 Investigate the direction of steepest ascent and descent for z = x2 + y 2 . The gradient is h2x, 2yi = 2hx, yi; this is a vector parallel to the vector hx, yi, so the direction of steepest ascent is directly away from the origin, starting at the point (x, y). The direction of steepest descent is thus directly toward the origin from (x, y). Note that at (0, 0) the gradient vector is h0, 0i, which has no direction, and it is clear from the plot of this surface that there is a minimum point at the origin, and tangent vectors in all directions are parallel to the x-y plane. If ∇f is perpendicular to u, Du f = |∇f | cos(π/2) = 0, since cos(π/2) = 0. This means that in either of the two directions perpendicular to ∇f , the slope of the surface is 0; this 14.5 Directional Derivatives 369 implies that a vector in either of these directions is tangent to the level curve at that point. Starting with ∇f = hfx , fy i, it is easy to find a vector perpendicular to it: either hfy , −fx i or h−fy , fx i will work. If f (x, y, z) is a function of three variables, all the calculations proceed in essentially the same way. The rate at which f changes in a particular direction is ∇f · u, where now ∇f = hfx , fy , fz i and u = hu1 , u2 , u3 i is a unit vector. Again ∇f points in the direction of maximum rate of increase, −∇f points in the direction of maximum rate of decrease, and any vector perpendicular to ∇f is tangent to the level surface f (x, y, z) = k at the point in question. Of course there are no longer just two such vectors; the vectors perpendicular to ∇f describe the tangent plane to the level surface, or in other words ∇f is a normal to the tangent plane. EXAMPLE 14.5.4 Suppose the temperature at a point in space is given by T (x, y, z) = T0 /(1 + x2 + y 2 + z 2 ); at the origin the temperature in Kelvin is T0 > 0, and it decreases in every direction from there. It might be, for example, that there is a source of heat at the origin, and as we get farther from the source, the temperature decreases. The gradient is ∇T = h = −2T0 y −2T0 z −2T0 x , , i 2 2 2 2 2 2 2 2 (1 + + y + z ) (1 + x + y + z ) (1 + x + y 2 + z 2 )2 (1 + x2 −2T0 hx, y, zi. + y 2 + z 2 )2 x2 The gradient points directly at the origin from the point (x, y, z)—by moving directly toward the heat source, we increase the temperature as quickly as possible. EXAMPLE 14.5.5 Find the points on the surface defined by x2 + 2y 2 + 3z 2 = 1 where the tangent plane is parallel to the plane defined by 3x − y + 3z = 1. Two planes are parallel if their normals are parallel or anti-parallel, so we want to find the points on the surface with normal parallel or anti-parallel to h3, −1, 3i. Let f = x2 + 2y 2 + 3z 2 ; the gradient of f is normal to the level surface at every point, so we are looking for a gradient parallel or anti-parallel to h3, −1, 3i. The gradient is h2x, 4y, 6zi; if it is parallel or anti-parallel to h3, −1, 3i, then h2x, 4y, 6zi = kh3, −1, 3i for some k. This means we need a solution to the equations 2x = 3k 4y = −k 6z = 3k but this is three equations in four unknowns—we need another equation. What we haven’t used so far is that the points we seek are on the surface x2 + 2y 2 + 3z 2 = 1; this is the 370 Chapter 14 Partial Differentiation fourth equation. If we solve the first three equations for x, y, and z and substitute into the fourth equation we get 2  2 −k 3k +2 +3 1= 4 6   2 3 9 k2 + + = 4 16 4  = 3k 2 2  25 2 k 8 √ 2 2 so k = ± . The desired points are 5 √ √ ! √ 2 2 3 2 and ,− , 5 10 5 √ √ √ ! 2 2 3 2 − . The , ,− 5 10 5 ellipsoid and the three planes are shown in figure 14.5.1. Figure 14.5.1 Ellipsoid with two tangent planes parallel to a given plane. (AP) Exercises 14.5. 1. Find Du f for f = x2 + xy + y2 in the direction of v = h2, 1i at the point (1, 1). ⇒ 2. Find Du f for f = sin(xy) in the direction of v = h−1, 1i at the point (3, 1). ⇒ 3. Find Du f for f = ex cos(y) in the direction 30 degrees from the positive x axis at the point (1, π/4). ⇒ 14.5 Directional Derivatives 371 4. The temperature of a thin plate in the x-y plane is T = x2 + y2 . How fast does temperature change at the point (1, 5) moving in a direction 30 degrees from the positive x axis? ⇒ p 5. Suppose the density of a thin plate at (x, y) is 1/ x2 + y2 + 1. Find the rate of change of the density at (2, 1) in a direction π/3 radians from the positive x axis. ⇒ p 6. Suppose the electric potential at (x, y) is ln x2 + y2 . Find the rate of change of the potential at (3, 4) toward the origin and also in a direction at a right angle to the direction toward the origin. ⇒ 7. A plane perpendicular to the x-y plane contains the point (2, 1, 8) on the paraboloid z = x2 + 4y2 . The cross-section of the paraboloid created by this plane has slope 0 at this point. Find an equation of the plane. ⇒ 8. A plane perpendicular to the x-y plane contains the point (3, 2, 2) on the paraboloid 36z = 4x2 + 9y2 . The cross-section of the paraboloid created by this plane has slope 0 at this point. Find an equation of the plane. ⇒ 9. Suppose the temperature at (x, y, z) is given by T = xy + sin(yz). In what direction should you go from the point (1, 1, 1) to decrease the temperature as quickly as possible? What is the rate of change of temperature in this direction? ⇒ 10. Suppose the temperature at (x, y, z) is given by T = xyz. In what direction can you go from the point (1, 1, 1) to maintain the same temperature? ⇒ 11. Find an equation for the plane tangent to x2 − 3y2 + z 2 = 7 at (1, 1, 3). ⇒ 12. Find an equation for the plane tangent to xyz = 6 at (1, 2, 3). ⇒ 13. Find a vector function for the line normal to x2 + 2y2 + 4z 2 = 26 at (2, −3, −1). ⇒ 14. Find a vector function for the line normal to x2 + y2 + 9z 2 = 56 at (4, 2, −2). ⇒ 15. Find a vector function for the line normal to x2 + 5y2 − z 2 = 0 at (4, 2, 6). ⇒ 16. Find the directions in which the directional derivative of f (x, y) = x2 + sin(xy) at the point (1, 0) has the value 1. ⇒ 17. Show that the curve r(t) = hln(t), t ln(t), ti is tangent to the surface xz 2 − yz + cos(xy) = 1 at the point (0, 0, 1). 18. A bug is crawling on the surface of a hot plate, the temperature of which at the point x units to the right of the lower left corner and y units up from the lower left corner is given by T (x, y) = 100 − x2 − 3y3 . a. If the bug is at the point (2, 1), in what direction should it move to cool off the fastest? How fast will the temperature drop in this direction? b. If the bug is at the point (1, 3), in what direction should it move in order to maintain its temperature? ⇒ 19. The elevation on a portion of a hill is given by f (x, y) = 100 − 4x2 − 2y. From the location above (2, 1), in which direction will water run? ⇒ 20. The contour map here shows wind speed in knots during Hurricane Andrew on August 24, 1992. Use it to estimate the value of the directional derivative of the wind speed at Homestead, FL, in the direction of the eye of the hurricane. Explain the meaning of your answer to a lay person. ⇒ 372 Chapter 14 Partial Differentiation 21. Suppose that g(x, y) = y − x2 . Find the gradient at the point (−1, 3). Sketch the level curve to the graph of g when g(x, y) = 2, and plot both the tangent line and the gradient vector at the point (−1, 3). (Make your sketch large). What do you notice, geometrically? ⇒ 22. The gradient ∇f is a vector valued function of two variables. Prove the following gradient rules. Assume f (x, y) and g(x, y) are differentiable functions. a. ∇(f g) = f ∇(g) + g∇(f ) b. ∇(f /g) = (g∇f − f ∇g)/g 2 c. ∇((f (x, y))n ) = nf (x, y)n−1 ∇f 14.6 Higher order derivatives In single variable calculus we saw that the second derivative is often useful: in appropriate circumstances it measures acceleration; it can be used to identify maximum and minimum points; it tells us something about how sharply curved a graph is. Not surprisingly, second derivatives are also useful in the multi-variable case, but again not surprisingly, things are a bit more complicated. It’s easy to see where some complication is going to come from: with two variables there are four possible second derivatives. To take a “derivative,” we must take a partial derivative with respect to x or y, and there are four ways to do it: x then x, x then y, y then x, y then y. EXAMPLE 14.6.1 Compute all four second derivatives of f (x, y) = x2 y 2 . Using an obvious notation, we get: fxx = 2y 2 fxy = 4xy fyx = 4xy fyy = 2x2 . 14.7 Maxima and minima 373 You will have noticed that two of these are the same, the “mixed partials” computed by taking partial derivatives with respect to both variables in the two possible orders. This is not an accident—as long as the function is reasonably nice, this will always be true. THEOREM 14.6.2 Clairaut’s Theorem tinuous, they are equal. EXAMPLE 14.6.3 If the mixed partial derivatives are con- Compute the mixed partials of f = xy/(x2 + y 2 ). fx = y 3 − x2 y (x2 + y 2 )2 fxy = − x4 − 6x2 y 2 + y 4 (x2 + y 2 )3 We leave fyx as an exercise. Exercises 14.6. 1. Find all first and second partial derivatives of f = xy/(x2 + y2 ). ⇒ 2. Find all first and second partial derivatives of x3 y2 + y5 . ⇒ 3. Find all first and second partial derivatives of 4x3 + xy2 + 10. ⇒ 4. Find all first and second partial derivatives of x sin y. ⇒ 5. Find all first and second partial derivatives of sin(3x) cos(2y). ⇒ 2 6. Find all first and second partial derivatives of ex+y . ⇒ p 7. Find all first and second partial derivatives of ln x3 + y4 . ⇒ 8. Find all first and second partial derivatives of z with respect to x and y if x2 +4y2 +16z 2 −64 = 0. ⇒ 9. Find all first and second partial derivatives of z with respect to x and y if xy + yz + xz = 1. ⇒ 2 2 10. Let α and k be constants. Prove that the function u(x, t) = e−α the heat equation ut = α2 uxx k t sin(kx) is a solution to 11. Let a be a constant. Prove that u = sin(x − at) + ln(x + at) is a solution to the wave equation utt = a2 uxx . 12. How many third-order derivatives does a function of 2 variables have? How many of these are distinct? 13. How many nth order derivatives does a function of 2 variables have? How many of these are distinct? 14.7 Maxima and minima Suppose a surface given by f (x, y) has a local maximum at (x0 , y0 , z0 ); geometrically, this point on the surface looks like the top of a hill. If we look at the cross-section in the plane y = y0 , we will see a local maximum on the curve at (x0 , z0 ), and we know from ∂z ∂z = 0 at this point. Likewise, in the plane x = x0 , ∂y = 0. single-variable calculus that ∂x 374 Chapter 14 Partial Differentiation So if there is a local maximum at (x0 , y0 , z0 ), both partial derivatives at the point must be zero, and likewise for a local minimum. Thus, to find local maximum and minimum points, we need only consider those points at which both partial derivatives are 0. As in the single-variable case, it is possible for the derivatives to be 0 at a point that is neither a maximum or a minimum, so we need to test these points further. You will recall that in the single variable case, we examined three methods to identify maximum and minimum points; the most useful is the second derivative test, though it does not always work. For functions of two variables there is also a second derivative test; again it is by far the most useful test, though it doesn’t always work. THEOREM 14.7.1 Suppose that the second partial derivatives of f (x, y) are continuous near (x0 , y0 ), and fx (x0 , y0 ) = fy (x0 , y0 ) = 0. We denote by D the discriminant: D(x0 , y0 ) = fxx (x0 , y0 )fyy (x0 , y0 ) − fxy (x0 , y0 )2 . If D > 0: if fxx (x0 , y0 ) < 0: there is a local maximum at (x0 , y0 ); if fxx (x0 , y0 ) > 0: there is a local minimum at (x0 , y0 ); if D < 0: there is neither a maximum nor a minimum at (x0 , y0 ); if D = 0: the test fails. EXAMPLE 14.7.2 Verify that f (x, y) = x2 + y 2 has a minimum at (0, 0). First, we compute all the needed derivatives: fx = 2x fy = 2y fxx = 2 fyy = 2 fxy = 0. The derivatives fx and fy are zero only at (0, 0). Applying the second derivative test there: D(0, 0) = fxx (0, 0)fyy (0, 0) − fxy (0, 0)2 = 2 · 2 − 0 = 4 > 0 and fxx (0, 0) = 2 > 0, so there is a local minimum at (0, 0), and there are no other possibilities. EXAMPLE 14.7.3 The derivatives: fx = 2x Find all local maxima and minima for f (x, y) = x2 − y 2 . fy = −2y fxx = 2 fyy = −2 fxy = 0. Again there is a single critical point, at (0, 0), and D(0, 0) = fxx (0, 0)fyy (0, 0) − fxy (0, 0)2 = 2 · −2 − 0 = −4 < 0, so there is neither a maximum nor minimum there, and so there are no local maxima or minima. The surface is shown in figure 14.7.1. 14.7 Maxima and minima 375 7.5 5.0 2.5 0.0 −5.0 −2.5 −2.5 0.0 −5.0 y 2.5 5.0 2.5 0.0 x Figure 14.7.1 EXAMPLE 14.7.4 The derivatives: fx = 4x3 5.0 −2.5 −5.0 A saddle point, neither a maximum nor a minimum. (AP) Find all local maxima and minima for f (x, y) = x4 + y 4 . fy = 4y 3 fxx = 12x2 fyy = 12y 2 fxy = 0. Again there is a single critical point, at (0, 0), and D(0, 0) = fxx (0, 0)fyy (0, 0) − fxy (0, 0)2 = 0 · 0 − 0 = 0, so we get no information. However, in this case it is easy to see that there is a minimum at (0, 0), because f (0, 0) = 0 and at all other points f (x, y) > 0. EXAMPLE 14.7.5 The derivatives: fx = 3x2 Find all local maxima and minima for f (x, y) = x3 + y 3 . fy = 3y 2 fxx = 6x2 fyy = 6y 2 fxy = 0. Again there is a single critical point, at (0, 0), and D(0, 0) = fxx (0, 0)fyy (0, 0) − fxy (0, 0)2 = 0 · 0 − 0 = 0, so we get no information. In this case, a little thought shows there is neither a maximum nor a minimum at (0, 0): when x and y are both positive, f (x, y) > 0, and when x and 376 Chapter 14 Partial Differentiation y are both negative, f (x, y) < 0, and there are points of both kinds arbitrarily close to (0, 0). Alternately, if we look at the cross-section when y = 0, we get f (x, 0) = x3 , which does not have either a maximum or minimum at x = 0. EXAMPLE 14.7.6 Suppose a box with no top is to hold a certain volume V . Find the dimensions for the box that result in the minimum surface area. The area of the box is A = 2hw + 2hl + lw, and the volume is V = lwh, so we can write the area as a function of two variables, A(l, w) = Then Al = − 2V +w l2 2V 2V + + lw. l w and Aw = − 2V + l. w2 If we set these equal to zero and solve, we find w = (2V )1/3 and l = (2V )1/3 , and the corresponding height is h = V /(2V )2/3 . The second derivatives are All = 4V l3 so the discriminant is D= Aww = 4V w3 Alw = 1, 4V 4V − 1 = 4 − 1 = 3 > 0. l3 w3 Since All is 2, there is a local minimum at the critical point. Is this a global minimum? It is, but it is difficult to see this analytically; physically and graphically it is clear that there is a minimum, in which case it must be at the single critical point. This applet shows an example of such a graph. Note that we must choose a value for V in order to graph it. Recall that when we did single variable global maximum and minimum problems, the easiest cases were those for which the variable could be limited to a finite closed interval, for then we simply had to check all critical values and the endpoints. The previous example is difficult because there is no finite boundary to the domain of the problem—both w and l can be in (0, ∞). As in the single variable case, the problem is often simpler when there is a finite boundary. THEOREM 14.7.7 If f (x, y) is continuous on a closed and bounded subset of R2 , then it has both a maximum and minimum value. 14.7 Maxima and minima 377 As in the case of single variable functions, this means that the maximum and minimum values must occur at a critical point or on the boundary; in the two variable case, however, the boundary is a curve, not merely two endpoints. EXAMPLE 14.7.8 The length of the diagonal of a box is to be 1 meter; find the maximum possible volume. If the box is placed with one corner at the origin, and sides along the axes, the length p of the diagonal is x2 + y 2 + z 2 , and the volume is V = xyz = xy p 1 − x2 − y 2 . Clearly, x2 + y 2 ≤ 1, so the domain we are interested in is the quarter of the unit disk in the first quadrant. Computing derivatives: y − 2yx2 − y 3 Vx = p 1 − x2 − y 2 x − 2xy 2 − x3 Vy = p 1 − x2 − y 2 √ If these are both 0, then x = 0 or y = 0, or x = y = 1/ 3. The boundary of the domain is composed of three curves: x = 0 for y ∈ [0, 1]; y = 0 for x ∈ [0, 1]; and x2 + y 2 = 1, where p x ≥ 0 and y ≥ 0. In all three cases, the volume xy 1 − x2 − y 2 is 0, so the maximum √ √ √ occurs at the only critical point (1/ 3, 1/ 3, 1/ 3). See figure 14.7.2. Exercises 14.7. 1. Find all local maximum and minimum points of f = x2 + 4y2 − 2x + 8y − 1. ⇒ 2. Find all local maximum and minimum points of f = x2 − y2 + 6x − 10y + 2. ⇒ 3. Find all local maximum and minimum points of f = xy. ⇒ 4. Find all local maximum and minimum points of f = 9 + 4x − y − 2x2 − 3y2 . ⇒ 5. Find all local maximum and minimum points of f = x2 + 4xy + y2 − 6y + 1. ⇒ 6. Find all local maximum and minimum points of f = x2 − xy + 2y2 − 5x + 6y − 9. ⇒ 7. Find the absolute maximum and minimum points of f = x2 + 3y − 3xy over the region bounded by y = x, y = 0, and x = 2. ⇒ 8. A six-sided rectangular box is to hold 1/2 cubic meter; what shape should the box be to minimize surface area? ⇒ 9. The post office will accept packages whose combined length and girth is at most 130 inches. (Girth is the maximum distance around the package perpendicular to the length; for a rectangular box, the length is the largest of the three dimensions.) What is the largest volume that can be sent in a rectangular box? ⇒ 378 Chapter 14 Partial Differentiation 0.0 0.25 0.5 0.75 0.15 1.0 0.1 1.0 0.75 0.05 0.5 0.25 0.0 Figure 14.7.2 0.0 The volume of a box with fixed length diagonal. 10. The bottom of a rectangular box costs twice as much per unit area as the sides and top. Find the shape for a given volume that will minimize cost. ⇒ 11. Using the methods of this section, find the shortest distance from the origin to the plane x + y + z = 10. ⇒ 12. Using the methods of this section, find the shortest distance from the point (x0 , y0 , z0 ) to the plane ax + by + cz = d. You may assume that c 6= 0; use of Sage or similar software is recommended. ⇒ 13. A trough is to be formed by bending up two sides of a long metal rectangle so that the cross-section of the trough is an isosceles trapezoid, as in figure 6.2.6. If the width of the metal sheet is 2 meters, how should it be bent to maximize the volume of the trough? ⇒ 14. Given the three points (1, 4), (5, 2), and (3, −2), (x − 1)2 + (y − 4)2 + (x − 5)2 + (y − 2)2 + (x − 3)2 + (y + 2)2 is the sum of the squares of the distances from point (x, y) to the three points. Find x and y so that this quantity is minimized. ⇒ 15. Suppose that f (x, y) = x2 + y2 + kxy. Find and classify the critical points, and discuss how they change when k takes on different values. 16. Find the shortest distance from the point (0, b) to the parabola y = x2 . ⇒ 17. Find the shortest distance from the point (0, 0, b) to the paraboloid z = x2 + y2 . ⇒ 18. Consider the function f (x, y) = x3 − 3x2 y + y3 . a. Show that (0, 0) is the only critical point of f . b. Show that the discriminant test is inconclusive for f . c. Determine the cross-sections of f obtained by setting y = kx for various values of k. 14.8 Lagrange Multipliers 379 d. What kind of critical point is (0, 0)? 19. Find the volume of the largest rectangular box with edges parallel to the axes that can be inscribed in the ellipsoid 2x2 + 72y2 + 18z 2 = 288. ⇒ 14.8 Lagrange Multipliers Many applied max/min problems take the form of the last two examples: we want to find an extreme value of a function, like V = xyz, subject to a constraint, like 1 = p x2 + y 2 + z 2 . Often this can be done, as we have, by explicitly combining the equations and then finding critical points. There is another approach that is often convenient, the method of Lagrange multipliers. It is somewhat easier to understand two variable problems, so we begin with one as an example. Suppose the perimeter of a rectangle is to be 100 units. Find the rectangle with largest area. This is a fairly straightforward problem from single variable calculus. We write down the two equations: A = xy, P = 100 = 2x + 2y, solve the second of these for y (or x), substitute into the first, and end up with a one-variable maximization problem. Let’s now think of it differently: the equation A = xy defines a surface, and the equation 100 = 2x + 2y defines a curve (a line, in this case) in the x-y plane. If we graph both of these in the three-dimensional coordinate system, we can phrase the problem like this: what is the highest point on the surface above the line? The solution we already understand effectively produces the equation of the cross-section of the surface above the line and then treats it as a single variable problem. Instead, imagine that we draw the level curves (the contour lines) for the surface in the x-y plane, along with the line. 50 40 30 y 20 10 0 0 10 20 30 40 50 x Figure 14.8.1 Constraint line with contour plot of the surface xy. 380 Chapter 14 Partial Differentiation Imagine that the line represents a hiking trail and the contour lines are, as on a topographic map, the lines of constant altitude. How could you estimate, based on the graph, the high (or low) points on the path? As the path crosses contour lines, you know the path must be increasing or decreasing in elevation. At some point you will see the path just touch a contour line (tangent to it), and then begin to cross contours in the opposite order—that point of tangency must be a maximum or minimum point. If we can identify all such points, we can then check them to see which gives the maximum and which the minimum value. As usual, we also need to check boundary points; in this problem, we know that x and y are positive, so we are interested in just the portion of the line in the first quadrant, as shown. The endpoints of the path, the two points on the axes, are not points of tangency, but they are the two places that the function xy is a minimum in the first quadrant. How can we actually make use of this? At the points of tangency that we seek, the constraint curve (in this case the line) and the level curve have the same slope—their tangent lines are parallel. This also means that the constraint curve is perpendicular to the gradient vector of the function; going a bit further, if we can express the constraint curve itself as a level curve, then we seek the points at which the two level curves have parallel gradients. The curve 100 = 2x + 2y can be thought of as a level curve of the function 2x + 2y; figure 14.8.2 shows both sets of level curves on a single graph. We are interested in those points where two level curves are tangent—but there are many such points, in fact an infinite number, as we’ve only shown a few of the level curves. All along the line y = x are points at which two level curves are tangent. While this might seem to be a show-stopper, it is not. 50 40 30 y 20 10 0 0 10 20 30 40 50 x Figure 14.8.2 Contour plots for 2x + 2y and xy. 14.8 Lagrange Multipliers 381 The gradient of 2x + 2y is h2, 2i, and the gradient of xy is hy, xi. They are parallel when h2, 2i = λhy, xi, that is, when 2 = λy and 2 = λx. We have two equations in three unknowns, which typically results in many solutions (as we expected). A third equation will reduce the number of solutions; the third equation is the original constraint, 100 = 2x+2y. So we have the following system to solve: 2 = λy 2 = λx 100 = 2x + 2y. In the first two equations, λ can’t be 0, so we may divide by it to get x = y = 2/λ. Substituting into the third equation we get 2 2 + 2 = 100 λ λ 8 =λ 100 so x = y = 25. Note that we are not really interested in the value of λ—it is a clever tool, the Lagrange multiplier, introduced to solve the problem. In many cases, as here, it is easier to find λ than to find everything else without using λ. The same method works for functions of three variables, except of course everything is one dimension higher: the function to be optimized is a function of three variables and the constraint represents a surface—for example, the function may represent temperature, and we may be interested in the maximum temperature on some surface, like a sphere. The points we seek are those at which the constraint surface is tangent to a level surface of the function. Once again, we consider the constraint surface to be a level surface of some function, and we look for points at which the two gradients are parallel, giving us three equations in four unknowns. The constraint provides a fourth equation. 2 Recall example 14.7.8: the diagonal of a box is 1, we seek to p maximize the volume. The constraint is 1 = x2 + y 2 + z 2 , which is the same as 1 = x2 + y 2 + z 2 . The function to maximize is xyz. The two gradient vectors are h2x, 2y, 2zi and hyz, xz, xyi, so the equations to be solved are EXAMPLE 14.8.1 yz = 2xλ xz = 2yλ xy = 2zλ 1 = x2 + y 2 + z 2 If λ = 0 then at least two of x, y, z must be 0, giving a volume of 0, which will not be the maximum. If we multiply the first two equations by x and y respectively, we get xyz = 2x2 λ xyz = 2y 2 λ 382 Chapter 14 Partial Differentiation so 2x2 λ = 2y 2 λ or x2 = y 2 ; in the same way we can show x2 = z 2 . Hence the fourth √ √ equation becomes 1 = x2 + x2 + x2 or x = 1/ 3, and so x = y = z = 1/ 3 gives the maximum volume. This is of course the same answer we obtained previously. Another possibility is that we have a function of three variables, and we want to find a maximum or minimum value not on a surface but on a curve; often the curve is the intersection of two surfaces, so that we really have two constraint equations, say g(x, y, z) = c1 and h(x, y, z) = c2 . It turns out that at points on the intersection of the surfaces where f has a maximum or minimum value, ∇f = λ∇g + µ∇h. As before, this gives us three equations, one for each component of the vectors, but now in five unknowns, x, y, z, λ, and µ. Since there are two constraint functions, we have a total of five equations in five unknowns, and so can usually find the solutions we need. EXAMPLE 14.8.2 The plane x + y − z = 1 intersects the cylinder x2 + y 2 = 1 in an ellipse. Find the points on the ellipse closest to and farthest from the origin. p We want the extreme values of f = x2 + y 2 + z 2 subject to the constraints g = x2 + y 2 = 1 and h = x + y − z = 1. To simplify the algebra, we may use instead f = x2 + y 2 + z 2 , since this has a maximum or minimum value at exactly the points at p which x2 + y 2 + z 2 does. The gradients are ∇f = h2x, 2y, 2zi ∇g = h2x, 2y, 0i ∇h = h1, 1, −1i, so the equations we need to solve are 2x = λ2x + µ 2y = λ2y + µ 2z = 0 − µ 1 = x2 + y 2 1 = x + y − z. Subtracting the first two we get 2y − 2x = λ(2y − 2x), so either λ = 1 or x = y. If λ = 1 then µ = 0, so z = 0 and the last two equations are 1 = x2 + y 2 and 1 = x + y. Solving these gives x = 1, y = 0, or x = 0, y = 1, so the points of interest are (1, 0, 0) and (0, 1, 0), which are both distance 1 from the origin. If x = y, the fourth equation is 2x2 = 1, 14.8 Lagrange Multipliers 383 √ √ giving x = y = ±1/ 2, and from the fifth equation we get z = −1 ± 2. The distance p √ √ √ √ from the origin to (1/ 2, 1/ 2, −1 + 2) is 4 − 2 2 ≈ 1.08 and the distance from the p √ √ √ √ origin to (−1/ 2, −1/ 2, −1 − 2) is 4 + 2 2 ≈ 2.6. Thus, the points (1, 0, 0) and √ √ √ (0, 1, 0) are closest to the origin and (−1/ 2, −1/ 2, −1 − 2) is farthest from the origin. This applet shows the cylinder, the plane, the four points of interest, and the origin. Exercises 14.8. 1. A six-sided rectangular box is to hold 1/2 cubic meter; what shape should the box be to minimize surface area? ⇒ 2. The post office will accept packages whose combined length and girth are at most 130 inches (girth is the maximum distance around the package perpendicular to the length). What is the largest volume that can be sent in a rectangular box? ⇒ 3. The bottom of a rectangular box costs twice as much per unit area as the sides and top. Find the shape for a given volume that will minimize cost. ⇒ 4. Using Lagrange multipliers, find the shortest distance from the point (x0 , y0 , z0 ) to the plane ax + by + cz = d. ⇒ 5. Find all points on the surface xy − z 2 + 1 = 0 that are closest to the origin. ⇒ 6. The material for the bottom of an aquarium costs half as much as the high strength glass for the four sides. Find the shape of the cheapest aquarium that holds a given volume V . ⇒ 7. The plane x − y + z = 2 intersects the cylinder x2 + y2 = 4 in an ellipse. Find the points on the ellipse closest to and farthest from the origin. ⇒ 8. Find three positive numbers whose sum is 48 and whose product is as large as possible. ⇒ 9. Find all points on the plane x + y + z = 5 in the first octant at which f (x, y, z) = xy2 z 2 has a maximum value. ⇒ 10. Find the points on the surface x2 − yz = 5 that are closest to the origin. ⇒ 11. A manufacturer makes two models of an item, standard and deluxe. It costs $40 to manufacture the standard model and $60 for the deluxe. A market research firm estimates that if the standard model is priced at x dollars and the deluxe at y dollars, then the manufacturer will sell 500(y − x) of the standard items and 45, 000 + 500(x − 2y) of the deluxe each year. How should the items be priced to maximize profit? ⇒ 12. A length of sheet metal is to be made into a water trough by bending up two sides as shown in figure 14.8.3. Find x and φ so that the trapezoid–shaped cross section has maximum area, when the width of the metal sheet is 2 meters (that is, 2x + y = 2). ⇒ ..... .... ..... ..... ..... ..... . ..... . . . ..... ... . .... ........ ....... ... ........ ..... ... ................................................................................................ .. ... φ x x φ y Figure 14.8.3 Cross-section of a trough. 13. Find the maximum and minimum values of f (x, y, z) = 6x + 3y + 2z subject to the constraint g(x, y, z) = 4x2 + 2y2 + z 2 − 70 = 0. ⇒ 384 Chapter 14 Partial Differentiation 14. Find the maximum and minimum values of f (x, y) = exy subject to the constraint g(x, y) = x3 + y3 − 16 = 0. ⇒ p 15. Find the maximum and minimum values of f (x, y) = xy + 9 − x2 − y2 when x2 + y2 ≤ 9. ⇒ 16. Find three real numbers whose sum is 9 and the sum of whose squares is a small as possible. ⇒ 17. Find the dimensions of the closed rectangular box with maximum volume that can be inscribed in the unit sphere. ⇒ 18. Find the isoceles triangle with perimeter 12 and maximum area. ⇒