math.wiki
Calculus › Partial derivatives › The multivariable chain rule
Sign in

The multivariable chain rule

A quantity depends on several variables, and each of those depends on others in turn. How fast does the quantity change when one of the variables at the bottom moves?

Where you are 8 in Partial derivatives

Why this exists

The chain rule of The chain rule differentiates f(g(t)) when each function has one input. In several variables compositions are the normal case: a temperature read along a moving path, a function rewritten in Polar coordinates, two quantities tied by an equation. By Tangent planes and linear approximation, a differentiable function changes by a linear function of the step in its inputs, so if the inputs change at known rates, the output changes at that linear function of the rates: a sum of products, one for each route from the bottom variable to the top one.

Isabel and Andre run a ramen shop and pull noodles by hand. As Andre stretches a rope of dough it grows longer and thinner with its volume fixed, and the chain rule ties the two rates together (Example 1).

Definition

Let f : D \to \mathbb{R} with D \subseteq \mathbb{R}^2, and let x(t) and y(t) be defined on an open interval I with (x(t), y(t)) \in D for t \in I. The composite is z(t) = f(x(t), y(t)). The variable t is the independent variable, x and y are intermediate variables, and z is the dependent variable.

More generally, let U \subseteq \mathbb{R}^m be open, D \subseteq \mathbb{R}^n, and G = (g_1, \dots, g_n) : U \to D. Write x_i = g_i(t_1, \dots, t_m) and z = f(G(t_1, \dots, t_m)), the composite f \circ G. The Jacobian matrix of G at \mathbf{a} \in U is the n \times m matrix DG(\mathbf{a}) whose entry in row i and column j is \dfrac{\partial g_i}{\partial t_j}(\mathbf{a}); row i holds the partial derivatives of x_i. A column of numbers and the vector in angle brackets with the same components are the same vector.

A dependence diagram (Figure 1) puts the dependent variable at the top, a branch down to each intermediate variable, and from each intermediate variable a branch down to each independent variable it depends on. Each branch is labeled with the partial derivative of its upper variable with respect to its lower one. A path from z to t_j is a chain of branches leading down from z to t_j.

Intuition

Let t move by a small \Delta t. Then x moves by about x'(t)\,\Delta t and y by about y'(t)\,\Delta t, and by the linear approximation of Tangent planes and linear approximation f responds to the step \langle \Delta x, \Delta y \rangle by about f_x\,\Delta x + f_y\,\Delta y. Substituting,

\Delta z \approx \left(f_x\,x'(t) + f_y\,y'(t)\right)\Delta t.

The change in t reaches z through x and through y. Along each route the rates multiply, and the linear part of f adds the routes: multiply the labels along each path of the diagram, and add over the paths.

z x y s t s t
\displaystyle \frac{\partial z}{\partial x}
\displaystyle \frac{\partial z}{\partial y}
\displaystyle \frac{\partial x}{\partial s}
\displaystyle \frac{\partial x}{\partial t}
\displaystyle \frac{\partial y}{\partial s}
\displaystyle \frac{\partial y}{\partial t}
Figure 1The dependence diagram of z = f(x, y) with x = x(s, t) and y = y(s, t) (Example 2). Two paths lead from z to s, so z_s = z_x x_s + z_y y_s; the two paths to t give z_t = z_x x_t + z_y y_t.

Results

Result 1 (the chain rule along a curve). Let x and y be differentiable at t_0, with (x(t), y(t)) \in D for t in an open interval about t_0. Let (a, b) = (x(t_0), y(t_0)) and let f be differentiable at (a, b). Then z(t) = f(x(t), y(t)) is differentiable at t_0 and

\begin{aligned} \frac{dz}{dt}(t_0) &= f_x(a, b)\,x'(t_0) \\ &\quad + f_y(a, b)\,y'(t_0). \end{aligned}

For f of n variables and x_1(t), \dots, x_n(t) differentiable at t_0,

\frac{dz}{dt} = \sum_{i=1}^{n} \frac{\partial f}{\partial x_i}\,\frac{dx_i}{dt},

with the partials of f at (x_1(t_0), \dots, x_n(t_0)).

Result 2 (several independent variables). Let G : U \to D be as in the definition and \mathbf{a} \in U. Suppose each g_i has a partial derivative with respect to t_j at \mathbf{a}, and f is differentiable at G(\mathbf{a}). Then z = f \circ G has a partial derivative with respect to t_j at \mathbf{a}, and

\frac{\partial z}{\partial t_j} = \sum_{i=1}^{n} \frac{\partial f}{\partial x_i}\,\frac{\partial x_i}{\partial t_j},

with the partials of f at G(\mathbf{a}) and those of x_i at \mathbf{a}: one term for each path from z to t_j. If this holds for every j, the row of partial derivatives of z is the matrix product

D(f \circ G)(\mathbf{a}) = Df(G(\mathbf{a}))\,DG(\mathbf{a}).

Result 3 (implicit differentiation). (a) Let \varphi be differentiable at a, with F(x, \varphi(x)) = c for every x in an open interval about a. If F is differentiable at (a, \varphi(a)) and F_y(a, \varphi(a)) \ne 0, then

\varphi'(a) = -\frac{F_x(a, \varphi(a))}{F_y(a, \varphi(a))}.

(b) Let \varphi(x, y) have both partial derivatives at (a, b), with F(x, y, \varphi(x, y)) = c on an open disk about (a, b). If F is differentiable at P = (a, b, \varphi(a, b)) and F_z(P) \ne 0, then \varphi_x(a, b) = -\dfrac{F_x(P)}{F_z(P)} and \varphi_y(a, b) = -\dfrac{F_y(P)}{F_z(P)}.

The implicit function theorem supplies \varphi; it is stated without proof and proved in Real Analysis. If F has continuous partials near (a, b), F(a, b) = c and F_y(a, b) \ne 0, there are open intervals I \ni a and J \ni b such that for each x \in I exactly one y = \varphi(x) in J has F(x, y) = c, and \varphi is differentiable on I. Likewise F(x, y, z) = c with F_z \ne 0 gives z = \varphi(x, y) with continuous partials.

Why it is true

Result 1. Let A = f_x(a, b), B = f_y(a, b) and E(h, k) = f(a + h, b + k) - f(a, b) - Ah - Bk, as in Tangent planes and linear approximation. Set \displaystyle \varepsilon(h, k) = \dfrac{E(h, k)}{\sqrt{h^2 + k^2}} for (h, k) \ne (0, 0) and \varepsilon(0, 0) = 0. Differentiability says exactly that \varepsilon is continuous at (0, 0), and since E(0, 0) = 0,

E(h, k) = \varepsilon(h, k)\sqrt{h^2 + k^2}

for every (h, k), the origin included. For small \Delta t \ne 0 put h = x(t_0 + \Delta t) - a and k = y(t_0 + \Delta t) - b. Then z(t_0 + \Delta t) - z(t_0) = f(a + h, b + k) - f(a, b), so

\begin{aligned} \frac{z(t_0 + \Delta t) - z(t_0)}{\Delta t} &= A\,\frac{h}{\Delta t} + B\,\frac{k}{\Delta t} \\ &\quad + \varepsilon(h, k)\,\frac{\sqrt{h^2 + k^2}}{\Delta t}. \end{aligned}

As \Delta t \to 0, \displaystyle \frac{h}{\Delta t} \to x'(t_0) and \displaystyle \frac{k}{\Delta t} \to y'(t_0). The last fraction has absolute value \displaystyle \sqrt{(h/\Delta t)^2 + (k/\Delta t)^2}, which tends to \displaystyle \sqrt{x'(t_0)^2 + y'(t_0)^2} and so stays bounded. Being differentiable, x and y are continuous at t_0, so (h, k) \to (0, 0), and \varepsilon(h, k) \to 0 because \varepsilon is continuous at the origin (Limits and continuity in several variables). The last term tends to 0, and the quotient to A x'(t_0) + B y'(t_0).

The function \varepsilon is needed because h and k can both be 0 for some \Delta t \ne 0 (a curve that pauses), where E cannot be divided by \displaystyle \sqrt{h^2 + k^2}. For n variables, replace \displaystyle \sqrt{h^2 + k^2} by the length of \langle h_1, \dots, h_n \rangle; nothing else changes.

Result 2. Hold every t_l with l \ne j at its value in \mathbf{a}. What remains is a path t_j \mapsto G(t_1, \dots, t_m) of one variable. Its ith component has derivative \displaystyle \frac{\partial g_i}{\partial t_j}(\mathbf{a}) at a_j, since a partial derivative is the derivative of exactly this restriction (Partial derivatives), and the derivative of z along it is \displaystyle \frac{\partial z}{\partial t_j}(\mathbf{a}). Result 1 in its n-variable form gives the sum. Entry j of the row Df(G(\mathbf{a}))\,DG(\mathbf{a}) is \displaystyle \sum_i \frac{\partial f}{\partial x_i}\,\frac{\partial g_i}{\partial t_j}, the same sum.

Result 3. (a) The function g(x) = F(x, \varphi(x)) is constant near a, so g'(a) = 0. By Result 1 with x(t) = t and y(t) = \varphi(t), g'(a) = F_x \cdot 1 + F_y\,\varphi'(a), with F_x and F_y at (a, \varphi(a)). Set this equal to 0 and divide by F_y \ne 0. (b) Apply Result 2 to G(x, y) = (x, y, \varphi(x, y)). Differentiating F(G(x, y)) = c with respect to x gives F_x \cdot 1 + F_y \cdot 0 + F_z\,\varphi_x = 0, and with respect to y, F_y + F_z\,\varphi_y = 0.

Worked examples

Example 1

Stretching a rope of dough

Andre holds a cylindrical rope of dough 10 centimeters long with radius 2 centimeters and stretches it so that the radius falls at 0.1 centimeters per second, with the volume fixed. How fast is the rope growing longer at that moment, and how fast is its curved surface area (the ends left out) changing?

Solution

With radius r(t) and length \ell(t), the volume V = \pi r^2 \ell is a differentiable function of (r, \ell), so Result 1 applies:

\frac{dV}{dt} = 2\pi r\ell\,\frac{dr}{dt} + \pi r^2\,\frac{d\ell}{dt}.

The volume is constant, so this is 0. With r = 2, \ell = 10 and \displaystyle \frac{dr}{dt} = -0.1,

\begin{aligned} 0 &= 40\pi(-0.1) + 4\pi\,\frac{d\ell}{dt}, \\ \frac{d\ell}{dt} &= 1. \end{aligned}

The rope grows at 1 centimeter per second. The curved area is S = 2\pi r\ell, so

\begin{aligned} \frac{dS}{dt} &= 2\pi\left(\ell\,\frac{dr}{dt} + r\,\frac{d\ell}{dt}\right) \\ &= 2\pi(-1 + 2) = 2\pi, \end{aligned}

about 6.28 square centimeters per second, although the radius shrinks.

Example 2

Two independent variables

Let z = x^3 + xy^2 with x = s - 2t and y = st. Find z_s and z_t at (s, t) = (1, 1).

Solution

Figure 1 is the dependence diagram: two paths lead from z to s and two to t. The partials f_x = 3x^2 + y^2 and f_y = 2xy are continuous, so f is differentiable (Result 2 of Tangent planes and linear approximation), and Result 2 gives

\begin{aligned} z_s &= f_x\,x_s + f_y\,y_s, \\ z_t &= f_x\,x_t + f_y\,y_t. \end{aligned}

At (s, t) = (1, 1), x = -1 and y = 1, so f_x = 4 and f_y = -2, while x_s = 1, x_t = -2, y_s = t = 1 and y_t = s = 1. Hence z_s = 4 - 2 = 2 and z_t = -8 - 2 = -10, as differentiating z = (s - 2t)^3 + (s - 2t)s^2t^2 directly confirms.

Example 3

Three intermediate variables in matrix form

Let w = x^2 y + yz + zx with x = u\cos v, y = u\sin v and z = uv. Find w_u and w_v at \displaystyle (u, v) = \left(2, \frac{\pi}{2}\right) as a product of matrices.

Solution

At \displaystyle \left(2, \frac{\pi}{2}\right), (x, y, z) = (0, 2, \pi), and the partials of the polynomial f are

\begin{aligned} f_x &= 2xy + z = \pi, \\ f_y &= x^2 + z = \pi, \\ f_z &= y + x = 2. \end{aligned}

The rows of DG hold the partials of x, y and z: (\cos v, -u\sin v), (\sin v, u\cos v) and (v, u). By Result 2,

\begin{aligned} \begin{bmatrix} w_u & w_v \end{bmatrix} &= \begin{bmatrix} \pi & \pi & 2 \end{bmatrix} \begin{bmatrix} 0 & -2 \\ 1 & 0 \\ \frac{\pi}{2} & 2 \end{bmatrix} \\ &= \begin{bmatrix} 2\pi & 4 - 2\pi \end{bmatrix}. \end{aligned}
Example 4

A curve that is the graph of a function

Show that the curve y^5 + x^2 y + x = 3 is the graph of a differentiable function y = \varphi(x) defined for every real x. Find the tangent line at (1, 1) and use it to estimate \varphi(1.1).

Solution

Fix x and let g(y) = y^5 + x^2 y + x - 3. Its derivative 5y^4 + x^2 is positive except at y = 0 when x = 0, so g is strictly increasing, and g(y) \to \pm\infty as y \to \pm\infty: by the intermediate value theorem g has exactly one zero, \varphi(x). With F(x, y) = y^5 + x^2 y + x, the partials F_x = 2xy + 1 and F_y = 5y^4 + x^2 are continuous, and F_y = 0 only at (0, 0), off the curve. So the implicit function theorem applies at every point of the curve, its function is \varphi (the only solution), and Result 3 gives

\varphi'(x) = -\frac{2x\varphi(x) + 1}{5\varphi(x)^4 + x^2}.

The point (1, 1) is on the curve since 1 + 1 + 1 = 3, and \displaystyle \varphi'(1) = -\frac{3}{6} = -\frac12. The tangent line is \displaystyle y = 1 - \frac12(x - 1), so \varphi(1.1) \approx 0.95. Solving y^5 + 1.21y = 1.9 numerically gives \varphi(1.1) \approx 0.9456.

Example 5

Proof: Euler’s theorem for homogeneous functions

A function f on \mathbb{R}^2 minus the origin is homogeneous of degree p if f(\lambda x, \lambda y) = \lambda^p f(x, y) for every \lambda > 0 and every (x, y) \ne (0, 0). Prove that a differentiable homogeneous function of degree p satisfies

x f_x(x, y) + y f_y(x, y) = p f(x, y).
Solution

Fix (x, y) \ne (0, 0) and let g(\lambda) = f(\lambda x, \lambda y) for \lambda > 0. The intermediate variables \lambda x and \lambda y have derivatives x and y with respect to \lambda, so Result 1 gives

g'(\lambda) = x f_x(\lambda x, \lambda y) + y f_y(\lambda x, \lambda y).

By homogeneity g(\lambda) = \lambda^p f(x, y) as well, with derivative p\lambda^{p - 1} f(x, y). The two expressions for g' agree at \lambda = 1, which is the claim. For instance, \displaystyle f(x, y) = \sqrt{x^4 + y^4} is homogeneous of degree 2, and indeed \displaystyle x f_x + y f_y = \dfrac{2x^4 + 2y^4}{\sqrt{x^4 + y^4}} = 2f.

Example 6

Proof: second derivatives in polar coordinates

Let f have continuous second partial derivatives on an open set U, and let z(r, \theta) = f(r\cos\theta, r\sin\theta) for r > 0 with (r\cos\theta, r\sin\theta) \in U. Prove that, with the left side at (r\cos\theta, r\sin\theta),

f_{xx} + f_{yy} = z_{rr} + \frac{1}{r}z_r + \frac{1}{r^2}z_{\theta\theta}.
Solution

Write c = \cos\theta and s = \sin\theta; then x_r = c, y_r = s, x_\theta = -rs and y_\theta = rc. The functions f, f_x and f_y have continuous partials, so all three are differentiable (Result 2 of Tangent planes and linear approximation) and Result 2 applies to each. First,

\begin{aligned} z_r &= c f_x + s f_y, \\ z_\theta &= -rs f_x + rc f_y. \end{aligned}

In r, c and s are constants, while f_x changes at c f_{xx} + s f_{xy} and f_y at c f_{yx} + s f_{yy}. Clairaut’s theorem (Partial derivatives) gives f_{xy} = f_{yx}, so

z_{rr} = c^2 f_{xx} + 2cs f_{xy} + s^2 f_{yy}.

In \theta, f_x changes at -rs f_{xx} + rc f_{xy} and f_y at -rs f_{yx} + rc f_{yy}, and the product rule on z_\theta gives

\begin{aligned} z_{\theta\theta} &= r^2\left(s^2 f_{xx} - 2cs f_{xy} + c^2 f_{yy}\right) \\ &\quad - r\left(c f_x + s f_y\right). \end{aligned}

The last bracket is z_r. Dividing by r^2 and adding z_{rr}, the mixed terms cancel and c^2 + s^2 = 1 leaves

z_{rr} + \frac{1}{r^2}z_{\theta\theta} = f_{xx} + f_{yy} - \frac{1}{r}z_r,

which is the claim: the Laplacian f_{xx} + f_{yy} in polar coordinates.

Boundaries

Partial derivatives of f are not enough. Let f(x, y) = \dfrac{x^2 y}{x^2 + y^2} for (x, y) \ne (0, 0) and f(0, 0) = 0. Both partials at the origin are 0, since f(h, 0) = 0 = f(0, k). Along x = y = t, \displaystyle z(t) = \frac{t}{2} and \displaystyle z'(0) = \frac12, while the formula of Result 1 would give 0. So f is not differentiable at the origin.

A partial derivative depends on what is held fixed. Let w = x + 2y. With u = x + y, in the variables (x, u) it reads w = 2u - x. So \displaystyle \left(\frac{\partial w}{\partial x}\right)_y = 1 but \displaystyle \left(\frac{\partial w}{\partial x}\right)_u = -1.

Implicit differentiation needs a nonzero denominator. On the circle x^2 + y^2 = 1 at (1, 0), F_y = 2y = 0, and no function y = \varphi(x) describes the circle nearby: for x just below 1 there are two points, and for x > 1 none.

Where this goes next

Result 1 says \displaystyle \frac{dz}{dt} is the dot product of \langle f_x, f_y \rangle with the velocity \langle x'(t), y'(t) \rangle. In Gradient and directional derivatives that vector becomes the gradient. Applying Result 1 twice to g(t) = f(\mathbf{a} + t\mathbf{h}) gives the quadratic approximation behind Extrema and the second derivative test, and differentiating a constraint along a curve leads to Lagrange multipliers. The product Df(G)\,DG returns in Change of variables and the Jacobian, and integrating Result 1 along a curve proves the fundamental theorem for line integrals in Conservative fields and potential functions.

Exercises

Along a curve

  1. Let z = x^2 y - y^3 with x = 1 + 2t and y = t^2 - 1. Find \dfrac{dz}{dt} at t = 1.
  2. Let w = xyz with x = \cos t, y = \sin t and z = t. Find \dfrac{dw}{dt} as a function of t.
  3. Let f(x, y) = \arctan\dfrac{y}{x} for x > 0. Use Result 1 to find the rate of change of f along x = t, y = t^2 at t = 1.
  4. A function g is differentiable at (3, 1) with g(3, 1) = 5, g_x(3, 1) = -2 and g_y(3, 1) = 4. Let h(t) = g(t^2 + 2, t^3). Find h'(1) and estimate h(1.1).

Several independent variables

  1. Let z = ye^x with x = st^2 and y = s^2 t. Find z_s and z_t in terms of s and t.
  2. Let z = x^2 + xy with x = r\cos\theta and y = r\sin\theta. Find z_r and z_\theta at \displaystyle (r, \theta) = \left(2, \frac{\pi}{4}\right).
  3. Let u = x^2 y z^3 with x = p + q, y = p - q and z = pq. Find u_p and u_q at (p, q) = (2, 1) as a product of matrices.
  4. A differentiable function f(u, v) has f_u(1, 0) = 2 and f_v(1, 0) = -3. Let g(s, t) = f(s^2 - t^2, 2st). Find g_s(1, 0) and g_t(1, 0).

Implicit differentiation

  1. The curve xe^y + ye^x = 1 passes through (0, 1). Find \dfrac{dy}{dx} there.
  2. Near (1, 1, 1) the equation x^2 + 2y^2 + 3z^2 - xyz = 5 defines z as a function of (x, y). Find z_x and z_y at (1, 1).
  3. Find the points of the ellipse x^2 - xy + y^2 = 3 where the tangent line is horizontal, and those where it is vertical.

In the ramen shop

  1. Isabel rolls a rectangular sheet of dough 30 centimeters long, 20 centimeters wide and 1 centimeter thick. Its length grows at 0.5 centimeters per second and its width at 0.2 centimeters per second, with its volume fixed. How fast is its thickness changing, and how fast is the area of its top growing?
  2. Andre pulls a rope of dough of radius 0.5 centimeters and length 80 centimeters; its length grows at 20 centimeters per second with its volume fixed. Find the rate of change of its radius and of its curved surface area.

Second derivatives

  1. Let f have continuous second partials and z = f(s + t, s - t). Show that z_{st} = f_{xx} - f_{yy}.
  2. Use Example 6 to show that u = \dfrac{x}{x^2 + y^2} satisfies u_{xx} + u_{yy} = 0 at every point other than the origin.
  3. Let f have continuous second partials on \mathbb{R}^2 and z(s, t) = f(e^s\cos t, e^s\sin t). Prove that z_{ss} + z_{tt} = e^{2s}\left(f_{xx} + f_{yy}\right).

Proofs

  1. Let F(u_1, u_2, u_3) be differentiable on \mathbb{R}^3 and w = F(x - y, y - z, z - x). Prove that w_x + w_y + w_z = 0.
  2. Let f be differentiable on \mathbb{R}^2 with y f_x - x f_y = 0 at every point. Prove that f is constant on every circle centered at the origin.
  3. Let f be differentiable on \mathbb{R}^2 minus the origin with x f_x + y f_y = p f at every point there. Prove that f is homogeneous of degree p, the converse of Example 5.
  4. Let f be differentiable on an open set containing the segment from P = (p_1, p_2) to Q = (q_1, q_2). Prove the mean value theorem in two variables: for some point C of the segment other than its ends, f(Q) - f(P) equals f_x(C)(q_1 - p_1) + f_y(C)(q_2 - p_2).
Answers
  1. 18. At t = 1, (x, y) = (3, 0), so f_x = 2xy = 0 and f_y = x^2 - 3y^2 = 9, while x' = 2 and y' = 2t = 2: \displaystyle \frac{dz}{dt} = 0 \cdot 2 + 9 \cdot 2 = 18.
  2. \displaystyle \frac{dw}{dt} = t\cos 2t + \frac12\sin 2t. By Result 1, \displaystyle \frac{dw}{dt} = yz(-\sin t) + xz\cos t + xy, which is t(\cos^2 t - \sin^2 t) + \sin t\cos t on the curve.
  3. \displaystyle \frac12. \displaystyle f_x = -\frac{y}{x^2 + y^2} and \displaystyle f_y = \frac{x}{x^2 + y^2} are \displaystyle -\frac12 and \displaystyle \frac12 at (1, 1), and x' = 1, y' = 2t = 2, so the rate is \displaystyle -\frac12 + 1 = \frac12, the derivative of \arctan t at t = 1.
  4. h'(1) = 8 and h(1.1) \approx 5.8. At t = 1 the point is (3, 1), with x' = 2t = 2 and y' = 3t^2 = 3, so h'(1) = -2 \cdot 2 + 4 \cdot 3 = 8 and h(1.1) \approx 5 + 8(0.1).
  5. z_s = e^{st^2}(s^2t^3 + 2st) and z_t = e^{st^2}(2s^3t^2 + s^2). With f_x = ye^x and f_y = e^x, z_s = ye^x \cdot t^2 + e^x \cdot 2st and z_t = ye^x \cdot 2st + e^x \cdot s^2.
  6. z_r = 4 and z_\theta = -4. At \displaystyle \left(2, \frac{\pi}{4}\right), \displaystyle x = y = \sqrt2, so \displaystyle f_x = 2x + y = 3\sqrt2 and \displaystyle f_y = x = \sqrt2. With \displaystyle x_r = y_r = \frac{\sqrt2}{2}, \displaystyle x_\theta = -\sqrt2 and \displaystyle y_\theta = \sqrt2: z_r = 3 + 1 and z_\theta = -6 + 2.
  7. u_p = 228 and u_q = 192. At (2, 1), (x, y, z) = (3, 1, 2), so f_x = 2xyz^3 = 48, f_y = x^2 z^3 = 72 and f_z = 3x^2 y z^2 = 108. The Jacobian matrix has rows (1, 1), (1, -1) and (q, p) = (1, 2); the row \begin{bmatrix} 48 & 72 & 108 \end{bmatrix} times it is \begin{bmatrix} 228 & 192 \end{bmatrix}.
  8. g_s(1, 0) = 4 and g_t(1, 0) = -6. With u = s^2 - t^2 and v = 2st, at (1, 0) the point is (u, v) = (1, 0) and u_s = 2, u_t = 0, v_s = 0, v_t = 2. So g_s = 2 \cdot 2 + (-3) \cdot 0 and g_t = 2 \cdot 0 + (-3) \cdot 2.
  9. -(e + 1). With F = xe^y + ye^x, at (0, 1) F_x = e^y + ye^x = e + 1 and F_y = xe^y + e^x = 1.
  10. \displaystyle z_x = -\frac15 and \displaystyle z_y = -\frac35. At (1, 1, 1), F_x = 2x - yz = 1, F_y = 4y - xz = 3 and F_z = 6z - xy = 5.
  11. Horizontal at (1, 2) and (-1, -2); vertical at (2, 1) and (-2, -1). Here F_x = 2x - y and F_y = 2y - x. A horizontal tangent needs F_x = 0 with F_y \ne 0: y = 2x gives 3x^2 = 3, and there F_y = 3x \ne 0. Swapping the roles of x and y, \displaystyle \frac{dx}{dy} = -\frac{F_y}{F_x} = 0 where x = 2y and 3y^2 = 3.
  12. The thickness falls at \displaystyle \frac{2}{75} \approx 0.0267 centimeters per second, and the top grows at 16 square centimeters per second. With V = \ell w\tau constant, 0 = w\tau\,\ell' + \ell\tau\,w' + \ell w\,\tau', that is 0 = 10 + 6 + 600\,\tau'. The top has area A = \ell w, and A' = w\ell' + \ell w' = 10 + 6.
  13. The radius falls at \displaystyle \frac{1}{16} = 0.0625 centimeters per second, and the area grows at 10\pi \approx 31.42 square centimeters per second. From V = \pi r^2\ell constant, 2\pi r\ell\,r' + \pi r^2\,\ell' = 0, so \displaystyle r' = -\frac{r\ell'}{2\ell} = -\frac{0.5 \cdot 20}{160}. Then S' = 2\pi(\ell r' + r\ell') = 2\pi(-5 + 10).
  14. With x = s + t and y = s - t, Result 2 gives z_s = f_x + f_y. The partials f_x and f_y are differentiable, since their own partials are continuous, so Result 2 applies to them in t: z_{st} = (f_{xx} - f_{xy}) + (f_{yx} - f_{yy}). Clairaut’s theorem gives f_{xy} = f_{yx}, leaving f_{xx} - f_{yy}.
  15. u has continuous second partials away from the origin, and every such point is (r\cos\theta, r\sin\theta) with r > 0. There \displaystyle z = \frac{r\cos\theta}{r^2} = \frac{\cos\theta}{r}, so \displaystyle z_r = -\frac{\cos\theta}{r^2}, \displaystyle z_{rr} = \frac{2\cos\theta}{r^3} and \displaystyle z_{\theta\theta} = -\frac{\cos\theta}{r}, and Example 6 gives \displaystyle u_{xx} + u_{yy} = \frac{2\cos\theta}{r^3} - \frac{\cos\theta}{r^3} - \frac{\cos\theta}{r^3} = 0.
  16. Here x_s = x, y_s = y, x_t = -y and y_t = x, so Result 2 gives z_s = x f_x + y f_y and z_t = -y f_x + x f_y. Differentiate again with the product rule, Result 2 for f_x and f_y (differentiable, as their partials are continuous) and f_{xy} = f_{yx}: z_{ss} = x^2 f_{xx} + 2xy f_{xy} + y^2 f_{yy} + x f_x + y f_y and z_{tt} = y^2 f_{xx} - 2xy f_{xy} + x^2 f_{yy} - x f_x - y f_y add to (x^2 + y^2)(f_{xx} + f_{yy}), and x^2 + y^2 = e^{2s}.
  17. Write F_1, F_2, F_3 for the partials of F at (x - y, y - z, z - x). The intermediate variables have partials 1, 0, -1 with respect to x, then -1, 1, 0 with respect to y and 0, -1, 1 with respect to z. By Result 2, w_x = F_1 - F_3, w_y = -F_1 + F_2 and w_z = -F_2 + F_3, which sum to 0.
  18. Fix R > 0 and let g(t) = f(R\cos t, R\sin t). By Result 1, g'(t) = -R\sin t\,f_x + R\cos t\,f_y, which is -(y f_x - x f_y) = 0 at (x, y) = (R\cos t, R\sin t). A function with zero derivative on \mathbb{R} is constant by the mean value theorem (The mean value theorem), and every point of the circle is (R\cos t, R\sin t) for some t.
  19. Fix (x, y) \ne (0, 0) and let g(\lambda) = \lambda^{-p} f(\lambda x, \lambda y) for \lambda > 0. By the product rule and Result 1, g'(\lambda) is -p\lambda^{-p - 1} f(\lambda x, \lambda y) + \lambda^{-p}(x f_x + y f_y) with the partials at (\lambda x, \lambda y). The hypothesis there says \lambda x f_x + \lambda y f_y = p f(\lambda x, \lambda y), so g'(\lambda) = 0. Hence g is constant on (0, \infty): g(\lambda) = g(1), that is, f(\lambda x, \lambda y) = \lambda^p f(x, y).
  20. Let g(t) = f(p_1 + t(q_1 - p_1), p_2 + t(q_2 - p_2)). For t in [0, 1], and for t near such a value, the point lies in the open set, so Result 1 makes g differentiable on [0, 1] with g'(t) = f_x(q_1 - p_1) + f_y(q_2 - p_2) at that point. The mean value theorem (The mean value theorem) gives c \in (0, 1) with f(Q) - f(P) = g(1) - g(0) = g'(c). Take C = (p_1 + c(q_1 - p_1), p_2 + c(q_2 - p_2)).
math.wiki · CC-BY-SA-4.0