A locally unique solution can still end
For , , separation gives on . The right-hand side is smooth and hence locally Lipschitz, so local existence and uniqueness are not in doubt. Nevertheless the solution cannot remain finite at . A local theorem says how to start; a continuation argument decides whether the solution can be extended.
Let be open. Assume is continuous and locally Lipschitz in , locally uniformly in : around every there is a product neighborhood and a constant for which .
Definition 1.9.1
(Solution and maximal solution)
A solution of is a differentiable function on an open interval such that and for every . It is maximal if no solution on a strictly larger open interval agrees with it on . A maximal interval is written with .
The local existence–uniqueness theorem then supplies a unique maximal solution through each initial point in .
Theorem 1.9.2
(Continuation theorem)
If , the graph eventually leaves every compact subset of : for each compact there is such that for all .
This says more than merely leaving a particular bounded region once. It rules out repeatedly returning to any compact set strictly inside the domain at times tending to the endpoint.
Proof by a restart near the endpoint
Proof
(Restart near the endpoint)
Suppose instead that for some compact there are with . A subsequence converges to . Choose a closed product neighborhood of contained in on which is bounded by and is Lipschitz in . Write the neighborhood as . For starting points in its half-sized product, choose with , , and ; omit a restriction if its coefficient is zero. The Picard construction uses the same bounds and Lipschitz constant throughout this rectangle, so this single works for every such starting point.
For large , lies in that smaller neighborhood and . Restart the equation at this point. Uniqueness throughout the common interval follows from local uniqueness in by an open–closed argument, even if the original graph leaves the chosen rectangle. Thus the restarted solution agrees with wherever both are defined. Glue before to the restarted solution afterward; this extends beyond , contradicting maximality. Therefore no such sequence exists. If the graph did not eventually leave , such a sequence could be chosen, proving the stated form.
Proof X-ray
Why compactness matters
The proof has one key move and two supports. The key move is the uniform restart time . Because the subsequence limit lies in , one closed rectangle around it carries one bound and one Lipschitz constant , so the Picard construction gives the same for every nearby starting point, and once is close to . Compactness of is used exactly once, to place that limit inside rather than on its boundary. Uniqueness is used once, to glue the restarted solution to . In the restricted-domain example below, the graph tends to , a boundary point of ; no rectangle around it lies in , and no restart is possible.
The full-space blow-up alternative
Corollary 1.9.3
(Full-space blow-up alternative)
If and , then
Proof
(A bounded sequence would stay in a compact set)
If (1) failed, some and times would satisfy . For large the graph points lie in the compact set , contradicting the continuation theorem. Here is any fixed time in the solution interval. The conclusion is a norm limit, not just an unbounded supremum.
Worked example 1.9.4
(Blow-up of )
The equation , has on . Direct differentiation verifies the equation; the formula shows divergence as . By uniqueness, any extension would agree with this formula before and hence could not be continuous there. The finite maximal time is consistent with (1).
Counterexample mode
Boundedness does not suffice inside a restricted domain
Take , and . The maximal solution is on . It stays bounded as , but approaches the boundary of . The full-space conclusion (1) does not apply to this restricted problem.
A growth bound that prevents finite-time blow-up
Theorem 1.9.5
(A linear growth bound excludes finite-time blow-up)
Suppose now and, for every finite , there is such that
Then every maximal solution is defined on all of , that is, and .
Proof
(Grönwall on every bounded time interval)
For a solution starting at , the integral equation gives . Grönwall's inequality on a finite forward interval contained in yields
If the maximal forward time were finite, choose to contain it. This bound would keep bounded up to that time, contradicting (1). Thus . Reversing time gives under the same assumptions. The condition is sufficient, not necessary: a particular equation may be globally solvable despite faster growth of its vector field.
Worked example 1.9.6
(Forward global existence without a linear growth bound)
For , , the vector field is not globally linearly bounded. Nevertheless every solution exists for all forward time: whenever , , and uniqueness prevents crossing an equilibrium at . Thus for forward time, so the continuation theorem excludes finite forward blow-up. This argument only establishes forward global existence; backward behavior requires separate analysis.
Worked example 1.9.7
(Backward blow-up for the same equation)
Take in . The solution stays positive, since it cannot cross the equilibrium . Put ; then
so
Forward in time decreases to the equilibrium . Backward, the bracket vanishes at , and as , as the blow-up alternative in reversed time requires at a finite endpoint. The growth-bound theorem would have given ; it does not apply, because admits no bound . The forward trapping argument has no backward counterpart: reversing time turns the equation into , whose large states grow.
Common mistakes
Common mistake
(Smoothness does not give global existence)
A smooth or polynomial right-hand side guarantees only local solutions. For the solution with is , which ends at ; larger data end sooner. Global existence needs an additional estimate, such as the growth bound (2) or a trapping argument as for , that keeps the graph inside a compact set on every bounded time interval.
Common mistake
(Checking the growth bound only on bounded sets)
On each ball one has , so satisfies a linear bound there with constant . That is not condition (2), which asks for one constant for all states once the time range is fixed. Every continuous satisfies a bound of the local kind, so such a bound can never exclude blow-up; the constant must not depend on the size of the state.
Practice and solutions
Quick check
(Find the maximal forward time for , .)
Solution · Answer
, so .
Quick check
(Why does local Lipschitz continuity alone not guarantee a solution for all time?)
Solution · Answer
The solution of , blows up at even though is smooth. Local Lipschitz continuity controls the solution only on a short interval around each point.
Quick check
(For on , does (2) hold?)
Solution · Answer
Yes: , so one may take .
Quick check
(In the restricted-domain example, which hypothesis of the full-space blow-up alternative fails?)
Solution · Answer
The domain is , not all of ; the solution approaches its boundary while remaining bounded.