Unit 9, lesson 119

First-Order Substitutions

How a well chosen change of variable turns Bernoulli equations into linear ones and equations of the form y' = F(y/x) into separable ones, and which solutions and domains need extra care along the way.

  • Theory part 237
  • Practice part 238
  • Solution Methods

Many equations that are neither separable nor linear become one of those after a change of variable, the same way a hard integral becomes easy after the right substitution. The skill is to recognize the shape of the equation, pick the new unknown, and then translate the answer back without losing solutions that the substitution could not see.

Key terms

Bernoulli equation

Definition A Bernoulli equation is an equation of the form

y′+P(x) y=Q(x) yn,y' + P(x)\,y = Q(x)\,y^n,

where nn is a real number and PP and QQ are continuous on an open interval II. For n=0n = 0 and n=1n = 1 it is already linear, so the interesting cases are n≠0,1n \neq 0, 1. When nn is not an integer, yny^n is taken for y>0y > 0.

In plain words It is a linear equation spoiled by a single power of yy on the right.

Example The equation y′+y=y2y' + y = y^2 is a Bernoulli equation with P=1P = 1, Q=1Q = 1 and n=2n = 2. So is y′−1xy=xyy' - \frac{1}{x}y = x\sqrt{y} with n=12n = \frac{1}{2}, on x>0x > 0 and y>0y > 0.

Substitution to a linear equation

Definition Let n≠0,1n \neq 0, 1 and let yy be a solution of the Bernoulli equation on an interval where y≠0y \neq 0 (and y>0y > 0 when nn is not an integer). Then v=y1−nv = y^{1-n} satisfies the linear equation

v′+(1−n)P(x) v=(1−n)Q(x).v' + (1 - n)P(x)\,v = (1 - n)Q(x).

Conversely, a solution vv of this linear equation with v>0v > 0 gives the solution y=v1/(1−n)y = v^{1/(1-n)} of the Bernoulli equation, and when 1−n1 - n is an odd integer any nonzero vv will do. When nn is an odd integer, 1−n1 - n is even and y=−v1/(1−n)y = -v^{1/(1-n)} is a solution as well. The proof is one line, since v′=(1−n)y−ny′v' = (1 - n)y^{-n}y', and multiplying the Bernoulli equation by (1−n)y−n(1 - n)y^{-n} gives the equation for vv.

In plain words Dividing by yny^n and renaming y1−ny^{1-n} as a new unknown removes the nonlinear part. Then you use an integrating factor as usual.

Example Solve y′+y=y2y' + y = y^2. With n=2n = 2, v=y−1v = y^{-1} satisfies v′−v=−1v' - v = -1. The integrating factor e−xe^{-x} gives (e−xv)′=−e−x\left(e^{-x}v\right)' = -e^{-x}, so e−xv=e−x+Ce^{-x}v = e^{-x} + C and v=1+Cexv = 1 + Ce^{x}. Hence

y=11+Cex.y = \frac{1}{1 + Ce^{x}}.

As a check,

y′=−Cex(1+Cex)2andy2−y=1−(1+Cex)(1+Cex)2=−Cex(1+Cex)2.y' = -\frac{Ce^x}{(1 + Ce^x)^2} \qquad\text{and}\qquad y^2 - y = \frac{1 - (1 + Ce^x)}{(1 + Ce^x)^2} = -\frac{Ce^x}{(1 + Ce^x)^2}.

The substitution cannot see the solution y=0y = 0, which has to be added to the family.

Homogeneous equations y’ = F(y/x)

Definition A first order equation is called homogeneous (in the sense of degree zero) if it can be written as

y′=F(yx)y' = F\left(\frac{y}{x}\right)

with FF continuous on an interval. For x≠0x \neq 0, an equation y′=f(x,y)y' = f(x, y) has this form exactly when f(tx,ty)=f(x,y)f(tx, ty) = f(x, y) for all t≠0t \neq 0, and then F(v)=f(1,v)F(v) = f(1, v). On an interval where x≠0x \neq 0, the substitution v=yxv = \frac{y}{x}, that is y=vxy = vx and y′=v+xv′y' = v + xv', turns it into the separable equation

x dvdx=F(v)−v.x\,\frac{dv}{dx} = F(v) - v.

In plain words If the right side only cares about the ratio yx\frac{y}{x}, make that ratio the new unknown.

Example Solve y′=x2+y2xyy' = \frac{x^2 + y^2}{xy} for x>0x > 0, y>0y > 0. Here F(v)=1v+vF(v) = \frac{1}{v} + v, so xv′=1vxv' = \frac{1}{v} and

∫v dv=∫dxx⟹v22=ln⁡x+C⟹y=x2ln⁡x+2C,\int v\,dv = \int \frac{dx}{x} \quad\Longrightarrow\quad \frac{v^2}{2} = \ln x + C \quad\Longrightarrow\quad y = x\sqrt{2\ln x + 2C},

valid where 2ln⁡x+2C>02\ln x + 2C > 0.

Not the same as a homogeneous linear equation

Definition The word homogeneous has two unrelated meanings. A linear equation y′+P(x)y=Q(x)y' + P(x)y = Q(x) is homogeneous when Q=0Q = 0. An equation y′=f(x,y)y' = f(x, y) is homogeneous of degree zero when f(tx,ty)=f(x,y)f(tx, ty) = f(x, y). Neither property implies the other.

In plain words Check which meaning a problem intends. One is about a missing forcing term, the other is about depending only on yx\frac{y}{x}.

Example The equation y′+xy=0y' + xy = 0 is a homogeneous linear equation, but f(x,y)=−xyf(x, y) = -xy gives f(tx,ty)=−t2xyf(tx, ty) = -t^2xy, so it is not of the form F(yx)F\left(\frac{y}{x}\right). The equation y′=1+yxy' = 1 + \frac{y}{x} is of the form F(yx)F\left(\frac{y}{x}\right), yet as a linear equation y′−1xy=1y' - \frac{1}{x}y = 1 it is nonhomogeneous.

Scaling substitutions

Definition A scaling substitution replaces the variables by constant multiples, x=αsx = \alpha s and y=βuy = \beta u with α,β≠0\alpha, \beta \neq 0. Then dydx=βαduds\frac{dy}{dx} = \frac{\beta}{\alpha}\frac{du}{ds}, so y′=f(x,y)y' = f(x, y) becomes

duds=αβ f(αs,βu).\frac{du}{ds} = \frac{\alpha}{\beta}\,f(\alpha s, \beta u).

Choosing α\alpha and β\beta well removes constants from the equation. Degree zero homogeneity is exactly the statement that y′=f(x,y)y' = f(x, y) does not change under the scaling x=tsx = ts, y=tuy = tu, which is why v=yxv = \frac{y}{x} works. The same idea goes further. Suppose the equation does not change under x=tsx = ts, y=tmuy = t^m u for some fixed power mm and every t>0t > 0, which means f(tx,tmy)=tm−1f(x,y)f(tx, t^m y) = t^{m-1}f(x, y). Then on x>0x > 0 the substitution y=vxmy = vx^m turns it into the separable equation

x dvdx=f(1,v)−mv.x\,\frac{dv}{dx} = f(1, v) - mv.

In plain words Measure xx and yy in better units and the equation loses its clutter. Fewer constants make the behavior easier to see.

Example For y′=3y−6y2y' = 3y - 6y^2, put y=u2y = \frac{u}{2} and x=s3x = \frac{s}{3}, so α=13\alpha = \frac{1}{3} and β=12\beta = \frac{1}{2}. Then

duds=23(3u2−6u24)=u−u2,\frac{du}{ds} = \frac{2}{3}\left(\frac{3u}{2} - \frac{6u^2}{4}\right) = u - u^2,

an equation with no parameters at all.

Domains and equilibrium solutions

Definition A substitution is valid only where it is defined and can be inverted. The substitution v=y1−nv = y^{1-n} needs y≠0y \neq 0 (and y>0y > 0 for noninteger nn), and v=yxv = \frac{y}{x} needs x≠0x \neq 0. Solutions outside these sets must be found separately. In particular, for n>0n > 0 the function y=0y = 0 solves every Bernoulli equation, and for y′=F(yx)y' = F\left(\frac{y}{x}\right) every root v0v_0 of F(v0)=v0F(v_0) = v_0 gives the line y=v0xy = v_0 x, which separation of xv′=F(v)−vxv' = F(v) - v discards.

In plain words Each step that divides or takes a power has a blind spot. List the blind spots before you start and check them at the end.

Example Solve y′=xy+y2x2y' = \frac{xy + y^2}{x^2} for x>0x > 0. Here F(v)=v+v2F(v) = v + v^2, so xv′=v2xv' = v^2. The root v=0v = 0 gives the solution y=0y = 0. For v≠0v \neq 0,

−1v=ln⁡x+C⟹y=−xln⁡x+C,-\frac{1}{v} = \ln x + C \quad\Longrightarrow\quad y = -\frac{x}{\ln x + C},

valid on each interval of x>0x > 0 where ln⁡x≠−C\ln x \neq -C. The answer is this family together with y=0y = 0, which no value of CC produces.

Check yourself

  1. Which substitution turns y′+2xy=x2y3y' + \frac{2}{x}y = x^2y^3 into a linear equation, and what is that equation?
Show the answer

Here n=3n = 3, so v=y−2v = y^{-2}. The linear equation is v′−4xv=−2x2v' - \frac{4}{x}v = -2x^2.

  1. Solve y′=yx+1y' = \frac{y}{x} + 1 for x>0x > 0 with the substitution v=yxv = \frac{y}{x}.
Show the answer

Substituting y=vxy = vx gives v+xv′=v+1v + xv' = v + 1, so v′=1xv' = \frac{1}{x} and v=ln⁡x+Cv = \ln x + C. Hence y=xln⁡x+Cxy = x\ln x + Cx.

  1. Is y′=x+yx−yy' = \frac{x + y}{x - y} homogeneous of degree zero, and what substitution does it suggest?
Show the answer

Yes, f(tx,ty)=t(x+y)t(x−y)=f(x,y)f(tx, ty) = \frac{t(x + y)}{t(x - y)} = f(x, y), and dividing top and bottom by xx gives F(v)=1+v1−vF(v) = \frac{1 + v}{1 - v}. Use y=vxy = vx on an interval with x≠0x \neq 0.

How this lesson connects

It also uses substitutions.