Exact equations

An equation such as (2xy - 3) + (x^2 + 4y)y’ = 0 is neither separable nor linear. When does such an equation say that some function of x and y stays constant, how can that be tested from the equation alone, and what can be done when the test fails?

Where you are 6 in First-order equations

Why this exists

Separable equations in general and Linear equations and integrating factors solve two shapes of first-order equation. The equation

(2xy - 3) + (x^2 + 4y)\,y' = 0

has neither: y' = \dfrac{3 - 2xy}{x^2 + 4y} is not a product g(x)\,h(y) and not affine in y. Yet its left side is the derivative of x^2y - 3x + 2y^2 along the graph of y, so its solutions are the curves on which that function is constant, as energy is along a motion. This lesson recognizes such equations through Partial derivatives, builds the conserved quantity, and multiplies other equations into this form. Throughout, y is a function of x, except where a curve is better traced with x as a function of y.

Definition

In plain words: an equation is exact when its left side is the total change of one function of x and y, so that a solution is a curve along which that function does not change.

Let R be an open set in the plane and M, N continuous on R. A solution of

M(x, y)\,dx + N(x, y)\,dy = 0

as a function of x is a differentiable \varphi on an open interval I with (x, \varphi(x)) \in R and M + N\,\varphi' = 0 at (x, \varphi(x)) for every x \in I. A solution as a function of y is a differentiable \psi with (\psi(y), y) \in R and M\,\psi' + N = 0 at (\psi(y), y). The equation y' = f(x, y) is the case M = -f, N = 1. An initial value problem adds a condition y(x_0) = y_0.

F is of class C^1 on R when its partial derivatives F_x and F_y exist and are continuous there. We write M_y for \dfrac{\partial M}{\partial y} and N_x for \dfrac{\partial N}{\partial x}. The equation is exact on R if some F of class C^1 on R has

\frac{\partial F}{\partial x} = M \quad\text{and}\quad \frac{\partial F}{\partial y} = N \quad\text{on } R.

Such an F is a potential function. An integrating factor on R is a function \mu of class C^1 on R, never 0, for which \mu M\,dx + \mu N\,dy = 0 is exact on R. An open rectangle is a set (a, b) \times (c, d) with -\infty \le a < b \le \infty and -\infty \le c < d \le \infty, such as the plane or the half-plane x > 0.

Intuition

Read F(x, y) as the height of the ground above (x, y). A small step (dx, dy) changes the height by about F_x\,dx + F_y\,dy, so M\,dx + N\,dy = 0 says every step is level: the solutions are the contour lines of F (Figure 1), and (M, N) points straight uphill.

If M and N are the slopes of one surface toward the east and the north, then M_y and N_x are both the mixed partial of the height, so slopes with M_y \ne N_x describe no surface. When M_y = N_x, the height is rebuilt by walking: add up M along a horizontal segment, then N along a vertical one. The walk must stay in the region, which is why the converse needs a region without holes (Example 4).

2 4 −2 2 4 x y
Figure 1Contour lines of the arboretum hill (Example 5), 1, 3 and 6 meters below the summit (dot at (2, 1)). They are solutions of M\,dx + N\,dy = 0 with M = 5 - 2x - y and N = 4 - x - 2y. The thick arc is the solution y(x) through the post at (1, 0); it ends at (0, 2) and (4, 0), where the contour turns vertical.

Results

1. Solutions lie on level curves. Let F be a potential function for M\,dx + N\,dy = 0 on an open set R.

(a) A differentiable \varphi on an open interval I, with graph in R, is a solution if and only if F(x, \varphi(x)) is constant on I. The same holds for solutions x = \psi(y).

(b) If (x_0, y_0) \in R and N(x_0, y_0) \ne 0, the initial value problem M + N\,y' = 0, y(x_0) = y_0, has a solution on an open interval containing x_0, and any two of its solutions agree on some open interval containing x_0.

2. The test for exactness. Let M and N be of class C^1 on an open set R.

(a) If the equation is exact on R, then M_y = N_x on R.

(b) If R is an open rectangle, M_y = N_x on R and (x_0, y_0) \in R, the equation is exact on R and

\begin{aligned} F(x, y) &= \int_{x_0}^{x} M(s, y_0)\,ds \\ &\quad + \int_{y_0}^{y} N(x, t)\,dt \end{aligned}

is a potential function. Every potential function on R is F + K for a constant K.

3. Integrating factors. Let M and N be of class C^1 on an open rectangle R = I \times J.

(a) If \mu is continuous and never 0 on R, then M\,dx + N\,dy = 0 and \mu M\,dx + \mu N\,dy = 0 have the same solutions in R.

(b) A function \mu(x) of class C^1 on I, never 0, is an integrating factor if and only if M_y - N_x = \dfrac{\mu'(x)}{\mu(x)}\,N on R. So if M_y - N_x = P(x)\,N with P continuous on I, then \mu(x) = e^{\int P(x)\,dx} is an integrating factor.

(c) Likewise \mu(y) is one if and only if N_x - M_y = \dfrac{\mu'(y)}{\mu(y)}\,M, and if N_x - M_y = Q(y)\,M with Q continuous on J, then \mu(y) = e^{\int Q(y)\,dy} is one.

(d) For p and q continuous on I, the linear equation y' + p(x)\,y = q(x), written with M = p(x)\,y - q(x) and N = 1, has M_y - N_x = p(x)\,N, so (b) gives \mu(x) = e^{\int p(x)\,dx}, the integrating factor of Linear equations and integrating factors.

Why it is true

A chain rule. If F is of class C^1 on R and u, v are differentiable with (u(t), v(t)) \in R, then \dfrac{d}{dt}F(u, v) = F_x\,u' + F_y\,v' (a case of the The multivariable chain rule). Split the change of F from t to t + h into changes along a horizontal and a vertical segment in R; the mean value theorem writes them as \Delta u\,F_x(\xi, v(t + h)) and \Delta v\,F_y(u(t), \eta); divide by h and let h \to 0, using the continuity of F_x and F_y.

Result 1. (a) By the chain rule with u(x) = x and v = \varphi, the derivative of F(x, \varphi(x)) is M + N\,\varphi' at (x, \varphi(x)). It is 0 on I exactly when \varphi is a solution, and a function on an interval has derivative 0 exactly when it is constant. For \psi, exchange x and y.

(b) Since F_y(x_0, y_0) = N(x_0, y_0) \ne 0, the implicit function theorem, proved in advanced calculus and used here without proof, gives open intervals I \ni x_0 and J \ni y_0 such that the points of I \times J on the level curve F = F(x_0, y_0) form the graph of a differentiable \varphi : I \to J. By (a), \varphi is a solution with \varphi(x_0) = y_0. Another such solution stays in J near x_0, by continuity, and lies on the same level curve by (a), so it equals \varphi there. \square

Result 2. (a) If F_x = M and F_y = N with M, N of class C^1, then F is of class C^2, and Clairaut’s theorem (Partial derivatives, Result 3) gives M_y = F_{xy} = F_{yx} = N_x.

(b) First, differentiation under the integral sign: for (x, y) \in R,

\frac{\partial}{\partial x}\int_{y_0}^{y} N(x, t)\,dt = \int_{y_0}^{y} N_x(x, t)\,dt.

Let K \subseteq R be a closed rectangle with x-side [x - \delta, x + \delta] and y-side between y_0 and y. A continuous function on a closed bounded set is uniformly continuous (a theorem of analysis, used without proof), so given \varepsilon > 0, |N_x(p) - N_x(q)| < \varepsilon for points of K closer than some \delta_1 < \delta. For 0 < |h| < \delta_1 the mean value theorem gives N(x + h, t) - N(x, t) = h\,N_x(x + \theta h, t) with 0 < \theta < 1, so the difference quotient of the integral is within \varepsilon\,|y - y_0| of \int_{y_0}^{y} N_x(x, t)\,dt.

The segments from (x_0, y_0) to (x, y_0) and on to (x, y) lie in the rectangle, so F is defined. Its first integral does not involve y, so the fundamental theorem gives F_y = N(x, y). By the lemma and N_x = M_y,

\begin{aligned} F_x &= M(x, y_0) + \int_{y_0}^{y} M_y(x, t)\,dt \\ &= M(x, y). \end{aligned}

Both partials are continuous, so F is of class C^1. If G is another potential, D = G - F has D_x = D_y = 0, and the mean value theorem along the same two segments gives D(x, y) = D(x_0, y_0). \square

Result 3. (a) As \mu \ne 0, M + N\varphi' = 0 at a point exactly when \mu M + \mu N\varphi' = 0 there; likewise for \psi. (b) By Result 2 on the rectangle, \mu M\,dx + \mu N\,dy = 0 is exact if and only if (\mu M)_y = (\mu N)_x, which for \mu = \mu(x) reads \mu M_y = \mu' N + \mu N_x; divide by \mu. If M_y - N_x = P\,N, then \mu = e^{\int P} is never 0 and \mu'/\mu = P. (c) The same, with (\mu M)_y = \mu' M + \mu M_y and (\mu N)_x = \mu N_x. (d) Here M_y - N_x = p(x) - 0 = p(x)\,N. \square

Worked examples

Example 1

Testing, then solving

Solve explicitly, on the largest open interval:

(2xy - 3)\,dx + (x^2 + 4y)\,dy = 0, \qquad y(1) = 1.
Solution

M = 2xy - 3 and N = x^2 + 4y are of class C^1 on the plane, an open rectangle, and M_y = 2x = N_x, so by Result 2(b) the equation is exact. A potential has F_x = 2xy - 3, so integrating in x with y held fixed,

F(x, y) = x^2y - 3x + A(y),

where the constant of integration may depend on the variable held fixed. Then F_y = x^2 + A'(y) must equal x^2 + 4y, so A(y) = 2y^2. By Result 1(a) the solution satisfies x^2y - 3x + 2y^2 = F(1, 1) = 0, a quadratic in y with roots \displaystyle \frac{1}{4}\left(-x^2 \pm \sqrt{x^4 + 24x}\right). At x = 1 these are \displaystyle \frac{-1 \pm 5}{4}, so the plus sign holds. Where x^4 + 24x > 0 the roots differ and a continuous solution cannot jump between them, so

y = \frac{-x^2 + \sqrt{x^4 + 24x}}{4}, \quad 0 < x < \infty.

The interval is maximal: x^4 + 24x = x(x^3 + 24) is negative for \displaystyle -\sqrt[3]{24} < x < 0, so the level curve has no point with such an x. Along the solution \displaystyle N = \sqrt{x^4 + 24x} > 0.

Example 2

An integrating factor in x

Solve

(y^2 + xy + y)\,dx + (x + 2y)\,dy = 0, \qquad y(0) = 1.
Solution

M_y = 2y + x + 1 and N_x = 1 differ, so by Result 2(a) the equation is not exact. But M_y - N_x = x + 2y = 1 \cdot N, so Result 3(b) with P(x) = 1 gives \mu = e^{x}. Integrating e^{x}(x + 2y) in y,

F(x, y) = e^{x}(xy + y^2) + B(x),

and F_x = e^{x}(xy + y^2 + y) + B'(x) equals e^{x}M when B' = 0. By Results 3(a) and 1(a), e^{x}(y^2 + xy) = F(0, 1) = 1, that is y^2 + xy - e^{-x} = 0, and the root with y(0) = 1 is

y = \frac{-x + \sqrt{x^2 + 4e^{-x}}}{2}, \quad x \in \mathbb{R}.

The radicand is positive for every x, and along the solution \displaystyle N = x + 2y = \sqrt{x^2 + 4e^{-x}} > 0, so the solution lives on the whole line.

Example 3

An integrating factor in y, and a lost solution

Find the curves that carry the solutions of the equation below, and the solution with y(2) = 1 along its circle, on its largest open interval:

2xy\,dx + (y^2 - x^2)\,dy = 0.
Solution

M_y = 2x and N_x = -2x, so the equation is exact on no open set. But N_x - M_y = -4x = -\dfrac{2}{y} \cdot 2xy, so on each half-plane y > 0 and y < 0, Result 3(c) with Q(y) = -\dfrac{2}{y} gives \mu = e^{-2\ln|y|} = \dfrac{1}{y^2}. Then \mu M = \dfrac{2x}{y} and \mu N = 1 - \dfrac{x^2}{y^2}. Integrating in x, F = \dfrac{x^2}{y} + B(y), and F_y = \mu N gives B(y) = y. The level curves \dfrac{x^2}{y} + y = C are

x^2 + y^2 = Cy, \quad\text{or}\quad x^2 + \left(y - \tfrac{C}{2}\right)^2 = \tfrac{C^2}{4},

circles tangent to the x-axis at the origin, which is removed. On y = 0, where \mu is undefined, M = 0, so y = 0 is a solution, lost when we multiplied (Figure 2).

Through (2, 1), C = 5. Of the roots of y^2 - 5y + x^2 = 0, the one equal to 1 at x = 2 is

y = \frac{5 - \sqrt{25 - 4x^2}}{2}, \quad -\tfrac{5}{2} < x < \tfrac{5}{2}.

It touches the origin at x = 0, where M = N = 0 and the equation holds trivially. At \displaystyle x = \pm\frac{5}{2}, y = |x|, so N = 0: the circle turns vertical, and the arc ends. Other solutions with y(2) = 1 leave the circle at the origin (Boundaries).

−2 2 −2 2 4 x y
Figure 2Solutions of 2xy\,dx + (y^2 - x^2)\,dy = 0 lie on the circles x^2 + y^2 = Cy and on the line y = 0 (Example 3). The thick arc is the solution with y(2) = 1 along its circle, on \displaystyle -\frac{5}{2} < x < \frac{5}{2}. Every circle and the line meet at the origin, where M = N = 0.
Example 4

A short proof: a region with a hole

Let M = \dfrac{-y}{x^2 + 4y^2} and N = \dfrac{x}{x^2 + 4y^2} on the plane with the origin removed. Prove that M_y = N_x there but M\,dx + N\,dy = 0 is not exact there, and solve it on the half-plane x > 0.

Solution

By the quotient rule,

M_y = \frac{4y^2 - x^2}{(x^2 + 4y^2)^2} = N_x.

Suppose F were a potential on the punctured plane, and let g(\theta) = F(2\cos\theta, \sin\theta). On this ellipse x^2 + 4y^2 = 4, so by the chain rule

\begin{aligned} g'(\theta) &= M \cdot (-2\sin\theta) + N\cos\theta \\ &= \frac{2\sin^2\theta + 2\cos^2\theta}{4} = \frac{1}{2}. \end{aligned}

Then g(2\pi) - g(0) = \pi, while both values are F(2, 0), a contradiction. The punctured plane is not a rectangle, so Result 2(b) does not apply.

The half-plane x > 0 is an open rectangle, and \displaystyle F = \frac{1}{2}\arctan\dfrac{2y}{x} is a potential there:

F_x = \frac{1}{2} \cdot \frac{x^2}{x^2 + 4y^2} \cdot \frac{-2y}{x^2} = M,

and similarly F_y = N. Its level curves are \dfrac{2y}{x} = \text{const}, so the solutions in x > 0 are the half-lines y = kx, x > 0, for every real k. \square

Example 5

Naomi’s arboretum hill

Naomi lays out paths on an arboretum hill, with x and y in hundreds of meters east and north of the visitor center. Hiro’s survey gives the grade, in meters of rise per hundred meters, as M = 5 - 2x - y toward the east and N = 4 - x - 2y toward the north. (a) Show that some height function has these grades, and find the summit and how far below it the trail post at (1, 0) stands. (b) Find the level path through the post, its westernmost and easternmost points, and the solution y(x) through the post.

Solution

(a) M_y = -1 = N_x on the plane, so Result 2(b) gives a height h with h_x = M and h_y = N, unique up to a constant. Integrating M in x, then matching h_y = N,

h(x, y) = h(0, 0) + 5x + 4y - x^2 - xy - y^2.

Both grades vanish where 2x + y = 5 and x + 2y = 4, at (2, 1). With u = x - 2 and v = y - 1,

\begin{aligned} h(2, 1) - h(x, y) &= u^2 + uv + v^2 \\ &= \left(u + \tfrac{v}{2}\right)^2 + \tfrac{3}{4}v^2, \end{aligned}

positive except at (2, 1), so (2, 1) is the summit. The post, with u = v = -1, stands 1 + 1 + 1 = 3 meters below it.

(b) Along a path (x, y(x)) the height changes at the rate M + N\,y', so a level path solves M\,dx + N\,dy = 0, and by Result 1(a) the one through the post lies on the ellipse u^2 + uv + v^2 = 3 (Figure 1). At its westernmost and easternmost points the tangent is vertical, so N = 0, that is u = -2v; then 3v^2 = 3, giving (0, 2) and (4, 0). As a quadratic in v, the ellipse gives \displaystyle v = \frac{1}{2}\left(-u \pm \sqrt{12 - 3u^2}\right), and at the post the minus sign holds:

y = \frac{4 - x - \sqrt{12 - 3(x - 2)^2}}{2}, \quad 0 < x < 4.

The northern half of the loop is a second solution.

Boundaries

  • Result 2(b) needs a region without holes. Example 4 passes the test on the punctured plane and has no potential. The test does suffice on every simply connected region, one where each closed curve can be shrunk to a point, by Line integrals and Green's theorem.
  • Where M = N = 0 the equation says nothing, and uniqueness can fail. In Example 3 the function equal to 0 for x \le 0 and to \displaystyle \frac{1}{2}\left(5 - \sqrt{25 - 4x^2}\right) for \displaystyle 0 < x < \frac{5}{2} also solves it with y(2) = 1, on \displaystyle \left(-\infty, \frac{5}{2}\right). No solution passes through the summit of Example 5, whose level set is one point.
  • Where N = 0 a solution y(x) can end while its level curve goes on (Examples 3 and 5), as x = \psi(y) (Exercise 14).
  • One-variable integrating factors are the exception (Exercise 9). In general \mu solves the partial differential equation \mu_x N - \mu_y M = \mu\,(M_y - N_x), and it is never unique (Exercise 17).

Where this goes next

For y'' = -g(y) and v = y', the equation v\,dv + g(y)\,dy = 0 is exact, with the energy \displaystyle \frac{1}{2}v^2 + G(y), G' = g, as potential; its level curves are orbits in Springs, damping and resonance and Nonlinear systems and linearization. Existence and uniqueness gives uniqueness for y' = f(x, y) with no potential at all. In vector calculus, exactness says the field (M, N) is a gradient (Conservative fields and potential functions).

Exercises

Testing and solving

  1. Show that the equation with M = 3x^2y + e^{y} and N = x^3 + xe^{y} - 2y is exact, and find its general solution in implicit form.
  2. Decide which equation is exact, and solve that one: (a) M = y^2 + 2x, N = 2xy + 3; (b) M = x^2 + y, N = x^2 - y.
  3. Solve the equation with M = e^{x}y + 2x and N = e^{x} + 3y^2, y(0) = 1, in implicit form, and find y'(0).
  4. Solve the equation with M = y + 3x^2 and N = x + 2y, y(1) = 1, explicitly, and give the largest open interval of the solution, its right end to four decimal places.
  5. On y > 0, solve the equation with M = 2x\ln y and N = \dfrac{x^2}{y} + 2y, y(1) = 1, by integrating N in y first, and find y'(1).
  6. For M = ye^{xy} + 2x, find every N of class C^1 on the plane that makes the equation exact.

Integrating factors

  1. Find an integrating factor depending on x alone for (3y^2 + 4x)\,dx + 2xy\,dy = 0, and solve the equation.
  2. Find an integrating factor depending on y alone for 2x\,dx + (x^2 + y + 1)\,dy = 0, and solve the equation.
  3. Show that the equation of Exercise 2(b) has no integrating factor depending on x alone and none depending on y alone, on any open rectangle.
  4. Write y' - 3y = e^{x} as M\,dx + N\,dy = 0, find the integrating factor Result 3(b) gives, and solve the equation as an exact equation.
  5. Find exponents a and b for which x^a y^b is an integrating factor on x > 0, y > 0 of the equation with M = 2y^2 + 3xy and N = 3xy + 2x^2, and solve it.

Level paths and level curves

  1. On another slope of the arboretum, a survey reports grades M = 2 - y east and N = 1 + x north. Show that no ground surface has these grades.
  2. In a hollow of the arboretum the grades are M = 2x - y east and N = 4y - x - 7 north. Show that a height function exists, find the lowest point and how far a post at (3, 2) stands above it, and find where the level path through the post crosses the line y = 2.
  3. For (2x - y)\,dx + (2y - x)\,dy = 0, show that no solution y = \varphi(x) with \varphi(2) = 1 exists on an open interval containing 2, and find the solution x = \psi(y) with \psi(1) = 2.

Proofs

  1. Let g be continuous on an open interval I and h continuous and never 0 on an open interval J. Prove that g(x)\,dx - \dfrac{dy}{h(y)} = 0 is exact on I \times J, and that its solutions as functions of x are exactly the solutions of y' = g(x)\,h(y) with graph in I \times J.
  2. Under the hypotheses of Result 2(b), let G be built the other way round: integrate N(x_0, t) from y_0 to y, then add the integral of M(s, y) from x_0 to x. Prove that G is a potential function and G = F.
  3. Let F be a potential function for M\,dx + N\,dy = 0 on R, and let g be of class C^1 on \mathbb{R} and never 0. Prove that g(F(x, y)) is an integrating factor on R.
Answers
  1. x^3y + xe^{y} - y^2 = C. Both M_y and N_x equal 3x^2 + e^{y}; integrating M in x gives x^3y + xe^{y} + A(y), and F_y = N gives A' = -2y.
  2. (a) is exact, M_y = 2y = N_x, with solution xy^2 + x^2 + 3y = C. (b) is not: M_y = 1 and N_x = 2x differ off the line \displaystyle x = \frac{1}{2}, so the test fails on every open set.
  3. e^{x}y + x^2 + y^3 = 2, from the potential e^{x}y + x^2 + y^3, whose value at (0, 1) is 2. Then y'(0) = -\dfrac{M(0, 1)}{N(0, 1)} = -\dfrac{1}{4}.
  4. \displaystyle y = \dfrac{-x + \sqrt{x^2 - 4x^3 + 12}}{2} on (-\infty, 1.5306). The potential xy + x^3 + y^2 equals 3 at (1, 1). The radicand decreases for x < 0 and for \displaystyle x > \frac{1}{6}, so it has one real zero, near 1.5306, where N = x + 2y reaches 0.
  5. x^2\ln y + y^2 = 1 and y'(1) = 0. Here M_y = \dfrac{2x}{y} = N_x; integrating N in y gives x^2\ln y + y^2 + B(x) with B' = 0. At (1, 1), M = 0.
  6. N = xe^{xy} + g(y) with g any function of class C^1. By Result 2, exactness on the plane means N_x = M_y = e^{xy} + xye^{xy}, the x-partial of xe^{xy}. So N - xe^{xy} has x-partial 0 and depends on y alone.
  7. \mu = x^2; solution x^3y^2 + x^4 = C. Here M_y - N_x = 4y = \dfrac{2}{x} \cdot 2xy, so P = \dfrac{2}{x} on x > 0 or x < 0, and x^2M, x^2N are the partials of x^3y^2 + x^4. The line x = 0 is also a solution as x = \psi(y), since N(0, y) = 0.
  8. \mu = e^{y}; solution e^{y}(x^2 + y) = C. Here N_x - M_y = 2x = 1 \cdot M, so Q = 1, and e^{y}(x^2 + y) has partials 2xe^{y} and e^{y}(x^2 + y + 1).
  9. A factor \mu(x) would give 1 - 2x = P(x)\,(x^2 - y) by Result 3(b). For fixed \displaystyle x \ne \frac{1}{2} the left side does not change with y, so P(x) = 0 and 1 - 2x = 0, a contradiction. A factor \mu(y) would give 2x - 1 = Q(y)\,(x^2 + y); for fixed y the right side has degree 2 in x, or is 0, never degree 1.
  10. M = -3y - e^{x} and N = 1 give M_y - N_x = -3N, so \mu = e^{-3x}, with potential \displaystyle e^{-3x}y + \frac{1}{2}e^{-2x}. Setting it equal to C gives \displaystyle y = Ce^{3x} - \frac{1}{2}e^{x}.
  11. a = b = 1; solution x^2y^3 + x^3y^2 = C. Dividing (x^a y^b M)_y - (x^a y^b N)_x by x^a y^b leaves a linear combination of y and x that vanishes exactly when 2(b + 2) = 3(a + 1) and 3(b + 1) = 2(a + 2), whose only solution is a = b = 1. Then xyM and xyN are the partials of x^2y^3 + x^3y^2.
  12. M_y = -1 and N_x = 1. A height with these grades would be a potential, and Result 2(a) would give -1 = 1.
  13. M_y = -1 = N_x, so a height exists: h = h(0, 0) + x^2 - xy + 2y^2 - 7y. With u = x - 1, v = y - 2, \displaystyle h - h(1, 2) = u^2 - uv + 2v^2 = \left(u - \frac{v}{2}\right)^2 + \frac{7}{4}v^2, so (1, 2) is the lowest point, and the post stands 4 meters above it. The level path u^2 - uv + 2v^2 = 4 meets v = 0 where u^2 = 4: at x = -1 and x = 3.
  14. The potential x^2 - xy + y^2 equals 3 at (2, 1), where N = 0. On that level curve \displaystyle 3 = \left(y - \frac{x}{2}\right)^2 + \frac{3}{4}x^2, so x \le 2, with equality only at (2, 1). A solution \varphi with \varphi(2) = 1 would lie on the curve by Result 1(a) and need points with x > 2; there are none. As a quadratic in x, the curve gives \displaystyle \psi(y) = \dfrac{y + \sqrt{12 - 3y^2}}{2} on -2 < y < 2, with \psi(1) = 2.
  15. Let G and H be antiderivatives of g and of \displaystyle \frac{1}{h}. Then F(x, y) = G(x) - H(y) has continuous partials F_x = g(x) and \displaystyle F_y = -\frac{1}{h(y)}, so the equation is exact on I \times J. A differentiable \varphi with graph in I \times J satisfies g(x) - \dfrac{\varphi'}{h(\varphi)} = 0 exactly when \varphi' = g(x)\,h(\varphi), since h(\varphi) \ne 0. \square
  16. The segments from (x_0, y_0) to (x_0, y) and on to (x, y) lie in R. By the fundamental theorem, G_x = M(x, y). The lemma of Result 2(b), with x and y exchanged, and M_y = N_x give G_y = N(x_0, y) + \int_{x_0}^{x} N_x(s, y)\,ds = N(x, y). So G = F + K by Result 2(b), and G(x_0, y_0) = 0 = F(x_0, y_0) gives K = 0. \square
  17. Let \Gamma be an antiderivative of g. Holding y fixed, the one-variable chain rule gives \big(\Gamma(F)\big)_x = g(F)\,M, and likewise \big(\Gamma(F)\big)_y = g(F)\,N, both continuous. So \Gamma(F) is a potential for g(F)M\,dx + g(F)N\,dy = 0, and g(F) is never 0, with continuous partials g'(F)\,M and g'(F)\,N. \square
math.wiki · CC-BY-SA-4.0