Differential Equations › Section B · First order differential equations › Topic B.6

Existence, uniqueness, and intervals of validity

Before solving an equation it pays to know whether a solution exists, whether it is the only one, and how far it reaches. Two theorems answer these questions, and they reveal a deep difference between linear and nonlinear equations.

Questions to ask before solving

Every method so far has produced a formula, but three questions sit underneath any formula. Does the initial value problem have a solution at all? Is it the only one, so that the formula you found is the answer rather than an answer? And on what interval does it live? These are not idle worries. A model that predicts two different futures from the same present cannot be trusted, and a numerical method will happily march past a point where the true solution has blown up. Existence and uniqueness theorems settle these questions, often without solving anything.

The linear theorem

For a linear initial value problem

y' + P(t)y = Q(t), \qquad y(t_0) = y_0

the answer is as good as it could be. If \(P\) and \(Q\) are continuous on an open interval \(\alpha < t < \beta\) containing \(t_0\), then the problem has exactly one solution, and that solution exists on the whole interval. The integrating-factor formula from "Linear equations and integrating factors" proves it: every step makes sense wherever \(P\) and \(Q\) are continuous. So for linear equations the interval of validity is the largest open interval containing \(t_0\) on which \(P\) and \(Q\) are both continuous, and the value \(y_0\) plays no part.

Worked example: an interval without solving

Find the interval of validity of \((t - 6)y' + ty = \sqrt{t + 4}\) with \(y(1) = 2\), without solving. 1. Standard form: divide by \(t - 6\), giving \(P = \frac{t}{t - 6}\) and \(Q = \frac{\sqrt{t + 4}}{t - 6}\). 2. \(P\) fails to be continuous at \(t = 6\). 3. \(Q\) fails there too, and the root requires \(t \ge -4\), so on an open interval \(t > -4\). 4. The candidate intervals are \(-4 < t < 6\) and \(t > 6\). 5. The one containing \(t_0 = 1\) is \(-4 < t < 6\). Changing \(y(1) = 2\) to any other value leaves this answer unchanged; changing the starting time to \(t_0 = 10\) would give \(t > 6\).

The nonlinear theorem

For a general initial value problem \(y' = f(t, y)\), \(y(t_0) = y_0\), the guarantee is weaker. If \(f\) and \(\frac{\partial f}{\partial y}\) are continuous on some rectangle \(\alpha < t < \beta\), \(\gamma < y < \delta\) containing \((t_0, y_0)\), then the problem has exactly one solution on some interval \(t_0 - h < t < t_0 + h\) inside \(\alpha < t < \beta\). The theorem promises a solution near \(t_0\) but says nothing about how large \(h\) is. To find the actual interval of validity of a nonlinear problem, you must solve it, or at least study it closely. When the hypotheses fail, a solution may still exist, but uniqueness can be lost.

Worked example: losing uniqueness

Find solutions of \(y' = 2\sqrt{y}\) with \(y(0) = 0\). 1. The constant \(y = 0\) works, since both sides are 0. 2. Try \(y = t^2\) for \(t \ge 0\): \(y' = 2t\), and \(2\sqrt{t^2} = 2t\), so it works too. Joined to \(y = 0\) for \(t < 0\), it is differentiable at 0, with slope 0 from both sides. 3. In fact, for every \(c \ge 0\), the function equal to 0 up to \(t = c\) and \((t - c)^2\) afterwards is a solution: infinitely many. 4. Why the theorem allows this: \(\frac{\partial f}{\partial y} = \frac{1}{\sqrt{y}}\) is not continuous at \(y = 0\), so no rectangle around \((0, 0)\) satisfies the hypotheses. The same failure lets \(y' = 3y^{2/3}\), \(y(0) = 0\), have both \(y = 0\) and \(y = t^3\) as solutions.

Worked example: an interval that depends on y(0)

Solve \(y' = y^3\) with \(y(0) = y_0 \ne 0\), and find the interval of validity. 1. Separate: \(y^{-3}\,dy = dt\), so \(-\frac{1}{2y^2} = t + C\), with \(C = -\frac{1}{2y_0^2}\). 2. Solve: \(y^2 = \frac{y_0^2}{1 - 2y_0^2t}\), and taking the root with the sign of \(y_0\),

y = \frac{y_0}{\sqrt{1 - 2y_0^2\,t}}

3. The root needs \(1 - 2y_0^2t > 0\), so the interval is \(-\infty < t < \frac{1}{2y_0^2}\). 4. With \(y_0 = 1\) the solution blows up at \(t = \frac{1}{2}\); with \(y_0 = 0.1\), at \(t = 50\). Nothing in the equation is discontinuous, yet the interval depends on where the solution starts.

Linear and nonlinear compared

The two theorems summarise how differently the two families behave. For linear equations, the interval of validity can be read from \(P\) and \(Q\) before solving, it does not depend on \(y_0\), and the general solution from the integrating factor contains every solution. For nonlinear equations, solutions can blow up at places no coefficient predicts, the interval can depend on \(y_0\), uniqueness can fail where \(\frac{\partial f}{\partial y}\) is discontinuous, solutions may be available only implicitly, and the general solution can miss solutions, such as the constants lost when separating. Linear problems are tame; nonlinear ones need checking case by case.

Uniqueness keeps solutions apart

Uniqueness has a geometric meaning. Where the nonlinear theorem applies everywhere, two different integral curves can never meet: if they shared a point, two solutions would satisfy the same initial condition there. This is why the curves of a direction field never cross. It also turns constant solutions into barriers. For \(y' = y(3 - y)\), the lines \(y = 0\) and \(y = 3\) are solutions, so a solution starting between them stays between them forever. It approaches 3 but can never reach it in finite time, since reaching it would mean meeting the constant solution. The topic "Equilibrium solutions and stability" builds on this fact.

Common mistakes to avoid

First, reading a linear interval before putting the equation in standard form, and so missing the points where the coefficient of \(y'\) is zero. Second, choosing an interval that does not contain \(t_0\), or giving all the candidate intervals at once. Third, closing an interval at a point where \(P\) or \(Q\) fails. Fourth, applying the linear theorem's conclusion to a nonlinear equation, for instance claiming \(y' = y^3\) has solutions on all of \(\mathbb{R}\) because nothing is discontinuous. Fifth, assuming a formula you found is the only solution when \(\frac{\partial f}{\partial y}\) is not continuous at the initial point.

Key takeaways

For \(y' + P(t)y = Q(t)\), \(y(t_0) = y_0\), continuity of \(P\) and \(Q\) on an open interval containing \(t_0\) guarantees exactly one solution on that whole interval, so the interval of validity is found without solving and does not depend on \(y_0\). For \(y' = f(t, y)\), continuity of \(f\) and \(\frac{\partial f}{\partial y}\) near \((t_0, y_0)\) guarantees exactly one solution on some, possibly small, interval. Nonlinear intervals must be found from the solution and can depend on \(y_0\). Where uniqueness holds, integral curves never cross, and constant solutions act as barriers.

Flashcards

Flashcard

Existence

The initial value problem has at least one solution

Guaranteed by the theorems' hypotheses.

Test your understanding

Question 1 of 5

Without solving, what is the interval of validity of (t − 5)y′ + y = 1/(t + 2) with y(0) = 3?

Question 2 of 5

For a linear initial value problem, how does the interval of validity depend on the initial value y₀?

Question 3 of 5

What is the interval of validity of y′ = y³ with y(0) = 2?

Question 4 of 5

For which IVP does the nonlinear theorem fail to guarantee a unique solution?

Question 5 of 5

For y′ = y(3 − y), why can a solution with 0 < y(0) < 3 never reach y = 3 in finite time?

← Previous topicNext topic →