Unit 2, lesson 12

Bernoulli Equations

How the power substitution reduces a Bernoulli equation to a linear equation, why the exponents 0 and 1 are exceptional, and how to restore the zero solution and respect the domain of the power.

  • Video 12
  • Substitution Methods

The logistic equation y′=ry(1−y/K)y' = ry(1 - y/K), with time tt as the independent variable and y′=dy/dty' = dy/dt, would be linear without its y2y^2 term. Equations that are linear except for one power of the unknown on the right side are Bernoulli equations. Dividing by that power and taking a suitable power of yy as the new unknown turns them into linear equations, which an integrating factor solves. This lesson derives the substitution, works an example through to a checked answer, and shows which solutions the substitution cannot see.

Key terms

Bernoulli equation

Definition With tt as the independent variable and y′=dy/dty' = dy/dt, a Bernoulli equation has the form

y′+p(t) y=q(t) yn,y' + p(t)\,y = q(t)\,y^n,

where pp and qq are continuous on an open interval II and nn is a real constant. The coefficient of y′y' must be 11, so an equation a(t)y′+b(t)y=g(t)yna(t)y' + b(t)y = g(t)y^n is first divided by a(t)a(t) on an interval where a(t)≠0a(t) \ne 0. For n=0n = 0 and n=1n = 1 the equation is linear, and for every other nn it is nonlinear. The power must make sense for the values of yy involved. A positive integer nn allows every real yy, a negative integer nn excludes y=0y = 0, and for a non-integer nn the usual real power requires y≥0y \ge 0, or y>0y > 0 when n<0n < 0.

In plain words A Bernoulli equation looks like a linear equation, except that its right side is multiplied by a power of the unknown. That power is the only nonlinear part.

Example The logistic equation y′=ry(1−y/K)y' = ry(1 - y/K) with constants r>0r > 0 and K>0K > 0 can be written as

y′−ry=−rK y2,y' - ry = -\frac{r}{K}\,y^2,

so it is a Bernoulli equation with n=2n = 2, p(t)=−rp(t) = -r and q(t)=−r/Kq(t) = -r/K. By contrast, y′=1+y2y' = 1 + y^2 is not a Bernoulli equation, because its right side mixes two different powers of yy and only a first power can be moved to the left side.

Power substitution

Definition For n≠0,1n \ne 0, 1, the power substitution replaces the unknown yy by

z=y1−n.z = y^{1-n}.

On an interval where y≠0y \ne 0, and y>0y > 0 if nn is not an integer, the chain rule gives

z′=(1−n) y−n y′.z' = (1 - n)\,y^{-n}\,y'.

The exponent 1−n1 - n is chosen because dividing the Bernoulli equation by yny^n gives y−ny′+p(t) y1−n=q(t)y^{-n}y' + p(t)\,y^{1-n} = q(t). In this equation y1−ny^{1-n} appears, and y−ny′y^{-n}y' is its derivative up to the constant factor 1−n1 - n.

In plain words Divide by the power that causes the trouble and look at what remains. The term p(t) y1−np(t)\,y^{1-n} suggests the new unknown, and the chain rule shows that its derivative is already in the equation.

Example For y′−y=−y3y' - y = -y^3 we have n=3n = 3, so z=y−2z = y^{-2} and z′=−2y−3y′z' = -2y^{-3}y'. Multiplying the equation by −2y−3-2y^{-3} gives

−2y−3y′+2y−2=2,-2y^{-3}y' + 2y^{-2} = 2,

that is z′+2z=2z' + 2z = 2. Its solutions are z=1+Ce−2tz = 1 + Ce^{-2t}, so y2=1/(1+Ce−2t)y^2 = 1/(1 + Ce^{-2t}). The new unknown z=y−2z = y^{-2} forgets the sign of yy, so the sign has to come from the data. With y(0)=−12y(0) = -\tfrac{1}{2} we get z(0)=4z(0) = 4, so C=3C = 3 and

y=−11+3e−2t,y = -\frac{1}{\sqrt{1 + 3e^{-2t}}},

which is defined for every real tt.

Reduction to a linear equation

Definition Let n≠0,1n \ne 0, 1 and let pp and qq be continuous on II. On a subinterval where y≠0y \ne 0, and y>0y > 0 if nn is not an integer, a differentiable function yy solves the Bernoulli equation exactly when z=y1−nz = y^{1-n} solves the linear equation

z′+(1−n) p z=(1−n) q.z' + (1 - n)\,p\,z = (1 - n)\,q.

The reason is that the two equations differ by the factor (1−n)y−n(1 - n)y^{-n}, which is never zero there. The linear equation has continuous coefficients, so it is solved with the integrating factor e(1−n)∫p(t) dte^{(1-n)\int p(t)\,dt}, and through any initial value its solution exists on all of II. The solution yy is then recovered from y1−n=zy^{1-n} = z, choosing the root or the sign that agrees with the initial value. It is valid only while zz stays nonzero, and positive when an even root or a non-integer power is needed to recover yy, so its interval can be shorter than II. For n>1n > 1 a zero of zz is a point where yy blows up.

In plain words Once the power has a new name, the equation is linear, and the integrating factor for linear equations solves it. The only new work is turning zz back into yy and checking where that is allowed.

Example Solve y′+y=y2y' + y = y^2 with y(0)=12y(0) = \tfrac{1}{2}. Here n=2n = 2 and z=1/yz = 1/y, so z′=−y′/y2z' = -y'/y^2. Dividing the equation by y2y^2 and multiplying by −1-1 gives z′−z=−1z' - z = -1. With the integrating factor e−te^{-t},

(e−tz)′=−e−t,e−tz=e−t+C,\begin{gathered} \left(e^{-t}z\right)' = -e^{-t}, \\[0.4em] e^{-t}z = e^{-t} + C, \end{gathered}

so z=1+Cetz = 1 + Ce^t. The initial value z(0)=2z(0) = 2 gives C=1C = 1, and

y=11+et.y = \frac{1}{1 + e^t}.

To check, y′=−et/(1+et)2y' = -e^t/(1 + e^t)^2, and over the same denominator

y2−y=1−(1+et)(1+et)2=−et(1+et)2=y′.\begin{aligned} y^2 - y &= \frac{1 - (1 + e^t)}{(1 + e^t)^2} \\[0.4em] &= -\frac{e^t}{(1 + e^t)^2} = y'. \end{aligned}

The denominator 1+et1 + e^t is positive for every tt, so this solution exists on the whole line, with 0<y<10 < y < 1, and it decreases toward 00 as t→∞t \to \infty. Other members of the family y=1/(1+Cet)y = 1/(1 + Ce^t) behave differently. The value C=0C = 0 gives the equilibrium y=1y = 1, and any C<0C < 0 gives a pole where 1+Cet=01 + Ce^t = 0.

Exceptional exponents

Definition The exponents n=0n = 0 and n=1n = 1 are exceptional, because for them the Bernoulli equation is already linear and no substitution is needed. For n=0n = 0 it reads y′+p(t)y=q(t)y' + p(t)y = q(t). For n=1n = 1 the two terms in yy combine into the homogeneous linear equation

y′+(p(t)−q(t))y=0.y' + \bigl(p(t) - q(t)\bigr)y = 0.

For n=1n = 1 the power substitution would even fail, since z=y0=1z = y^0 = 1 is constant and the factor 1−n1 - n is 00. The other exponents are not exceptional, but some need care with the domain, as a negative nn excludes y=0y = 0 and a non-integer nn restricts yy to nonnegative values.

In plain words When the power is 00 or 11, there is nothing nonlinear to remove, so the equation is solved directly as a linear one.

Example The equation y′+y=tyy' + y = ty has n=1n = 1, p(t)=1p(t) = 1 and q(t)=tq(t) = t. Moving the right side over gives y′+(1−t)y=0y' + (1 - t)y = 0, and the integrating factor et−t2/2e^{t - t^2/2} gives the solutions

y=Cet2/2−ty = Ce^{t^2/2 - t}

for every real constant CC, on the whole line. The zero solution is the case C=0C = 0, because nothing was divided by yy. With n=0n = 0, as in y′+y=ty' + y = t, the equation is linear from the start, with solutions y=t−1+Ce−ty = t - 1 + Ce^{-t}.

Zero solution

Definition For every n>0n > 0 the zero function y=0y = 0 solves the Bernoulli equation on all of II, since both sides vanish. The power substitution assumes y≠0y \ne 0, so this solution never comes out of the formula for yy and must be restored by checking the original equation. For n=0n = 0 the zero function is a solution only when qq is identically zero, and for n<0n < 0 it is not allowed at all. For an integer n≥2n \ge 2 the right side q(t)yn−p(t)yq(t)y^n - p(t)y of y′=q(t)yn−p(t)yy' = q(t)y^n - p(t)y has a continuous partial derivative in yy, so solutions are unique, and a solution that is nonzero at one time is nonzero on its whole interval. For 0<n<10 < n < 1 that partial derivative is unbounded near y=0y = 0 wherever q≠0q \ne 0, uniqueness can fail there, and a solution can stay at zero for a while and then leave it.

In plain words Dividing by yny^n quietly assumes y≠0y \ne 0. Before dividing, check whether y=0y = 0 solves the equation, and add it to the answer if it does.

Example For y′+y=y2y' + y = y^2, substituting y=0y = 0 gives 0=00 = 0, so the zero function is a solution, but no constant CC in y=1/(1+Cet)y = 1/(1 + Ce^t) produces it. The complete set of solutions is that family together with y=0y = 0. For y′=yy' = \sqrt{y}, a Bernoulli equation with p=0p = 0, q=1q = 1 and n=12n = \tfrac{1}{2}, the initial value y(0)=0y(0) = 0 has the zero solution and also, for every a≥0a \ge 0, the solution

y(t)={0,t≤a,14(t−a)2,t>a.y(t) = \begin{cases} 0, & t \le a, \\ \tfrac{1}{4}(t - a)^2, & t > a. \end{cases}

The substitution z=yz = \sqrt{y} gives z′=12z' = \tfrac{1}{2} and only the pieces y=14(t−c)2y = \tfrac{1}{4}(t - c)^2 with t>ct > c, so it cannot detect where a solution leaves zero.

Check yourself

  1. Write ty′+y=t2y3ty' + y = t^2y^3, for t>0t > 0, in Bernoulli form, and find the linear equation for the new unknown.
Show the answer

Dividing by tt gives the normalized form y′+1t y=t y3y' + \frac{1}{t}\,y = t\,y^3, so n=3n = 3, p(t)=1/tp(t) = 1/t and q(t)=tq(t) = t. The new unknown is z=y−2z = y^{-2}, with z′=−2y−3y′z' = -2y^{-3}y'. Multiplying the normalized equation by −2y−3-2y^{-3} gives

z′−2t z=−2t,z' - \frac{2}{t}\,z = -2t,

valid on intervals with t>0t > 0 and y≠0y \ne 0. With the integrating factor t−2t^{-2} its solutions are z=t2(C−2ln⁡t)z = t^2(C - 2\ln t), so y2=1/(t2(C−2ln⁡t))y^2 = 1/\bigl(t^2(C - 2\ln t)\bigr) wherever C−2ln⁡t>0C - 2\ln t > 0. The zero function is also a solution and has to be kept separately. Skipping the division by tt would put the wrong coefficients into pp and qq.

  1. Solve y′=2y−y2y' = 2y - y^2 with y(0)=1y(0) = 1.
Show the answer

Written as y′−2y=−y2y' - 2y = -y^2, it is a Bernoulli equation with n=2n = 2. With z=1/yz = 1/y we get z′=−y′/y2=−2/y+1z' = -y'/y^2 = -2/y + 1, so z′+2z=1z' + 2z = 1. The integrating factor e2te^{2t} gives (e2tz)′=e2t(e^{2t}z)' = e^{2t}, so z=12+Ce−2tz = \tfrac{1}{2} + Ce^{-2t}. From z(0)=1z(0) = 1 we get C=12C = \tfrac{1}{2}, and

y=21+e−2t.y = \frac{2}{1 + e^{-2t}}.

The denominator is positive for every tt, so the solution exists on the whole line, and it rises from 00 toward the equilibrium 22. A direct check works, since y′=4e−2t/(1+e−2t)2y' = 4e^{-2t}/(1 + e^{-2t})^2 and 2y−y22y - y^2 simplifies to the same expression.

  1. A classmate solves y′+y=y2y' + y = y^2 with z=1/yz = 1/y and reports that every solution has the form y=1/(1+Cet)y = 1/(1 + Ce^t) and exists for all tt. Find the solution with y(0)=2y(0) = 2 and its maximal interval, and name the solution that the formula leaves out.
Show the answer

From z(0)=12=1+Cz(0) = \tfrac{1}{2} = 1 + C we get C=−12C = -\tfrac{1}{2}, so

y=11−12et=22−et.y = \frac{1}{1 - \tfrac{1}{2}e^t} = \frac{2}{2 - e^t}.

The denominator vanishes at t=ln⁡2≈0.693t = \ln 2 \approx 0.693, and y→+∞y \to +\infty as tt approaches ln⁡2\ln 2 from the left, so the maximal interval is (−∞,ln⁡2)(-\infty, \ln 2), not the whole line. The formula also leaves out the zero solution y=0y = 0, which satisfies the original equation but corresponds to no value of CC, because z=1/yz = 1/y is undefined there.

How this lesson connects