Motivation
A nonlinear equation can become linear after changing the dependent variable. That is useful only if we know how to return. A real power may require a positive argument, an even power may erase a sign, and division by the solution may remove its zero branch. The linear equation can be globally defined while its nonlinear reconstruction is not.
Bernoulli's method is a clean place to learn this discipline. We will solve the transformed equation explicitly and then let its admissible sign determine the original interval. The same zero of the transformed function means infinite state for one exponent and zero state for another. These different endpoints require different continuation arguments.
Definitions
Let the coefficient interval be open and contain the initial point, and let the scalar coefficients be continuous. For a real exponent different from zero and one, first consider positive solutions:
For arbitrary real exponents the power is defined on positive states by the real logarithm. A positive initial-value solution is continuously differentiable on a nonempty open subinterval containing the initial point and stays strictly positive. A maximal positive solution cannot be extended within this positive-state class. That is not the same as maximality after enlarging the problem to allow zero or negative states. All nonlinear integer exponents receive a separate real-state treatment below. For noninteger exponents this chapter keeps the positive-state convention, even when a rational power could also be defined for negative states.
Definition 1.7.1
(The reversible power coordinate)
Set
For either sign of the exponent in this coordinate change, it is a smooth bijection from positive states onto positive states. Both directions are therefore legitimate there. Nothing in this definition supplies an inverse for negative transformed values or justifies differentiating at an original zero.
The omitted exponents are already linear: exponent zero means that the nonlinear term is replaced by the forcing itself, and exponent one combines with the linear state coefficient. These cases are solved directly in the practice below. In particular the zero-exponent problem is defined as a linear equation, without any need to assign a value to zero raised to zero.
Theorem/Proposition
Theorem 1.7.2
(Positive Bernoulli solutions and their full interval)
Define the positive integrating factor, bracket and linear solution by
Let the admissible interval be the connected component of the positive bracket set containing the initial point. Then the unique maximal positive solution of (1) is
Only this component is the solution interval through the datum, even if the bracket is positive again elsewhere.
Proof
(Linearize, then invert on the positive component)
On positive states, the chain rule and the original equation give
Multiplication by the factor in (3) makes the left side a product derivative. Integration from the initial point gives exactly (3). The integrals exist at every point of the coefficient interval, so the transformed linear solution exists on that entire interval and is unique.
Since the factor is positive, the transformed solution is positive exactly where the bracket is. On the component in (4), the inverse power is differentiable and direct differentiation reverses (5), proving the original equation. Every other positive solution transforms to the same linear initial-value solution, so it must coincide with (4). It cannot continue positively through a zero bracket; it can continue through any interior point where the bracket remains positive. This proves the precise maximal positive interval, without assuming global boundedness of the original solution.
Theorem 1.7.3
(What an interior endpoint of the positive branch means)
Suppose a finite endpoint of the maximal positive interval lies inside the coefficient interval. Then the transformed bracket and transformed solution are zero there. Approaching that endpoint from within the positive branch,
For an exponent strictly between zero and one, enlarge the equation to nonnegative states using the continuous power value zero at zero. The positive branch then has derivative tending to zero and can be joined to the zero solution on the other side of the endpoint, within the coefficient interval, as a continuously differentiable nonnegative solution. This is an extension in the enlarged class, not a positive continuation. At a left endpoint it is a departure from zero as the independent variable increases; at a right endpoint it is an arrival.
For a negative exponent the original power is undefined at zero. A finite zero-state limit alone does not give a classical continuation through it. Endpoints of the coefficient interval require a separate analysis; (6) concerns endpoints strictly inside that interval.
Proof
(The linear solution is continuous at the endpoint)
Continuity of the globally defined linear solution gives a finite value at the interior endpoint. That value cannot be positive, or (4) would extend there, and it cannot be negative, since the one-sided values are positive. It is zero. The sign of the reciprocal exponent in (2) gives (6).
For an exponent between zero and one, continuity of the coefficients and the original equation imply that the derivative tends to zero as the state tends to zero. The zero function solves the enlarged equation. Joining the two at their common zero value produces matching one-sided derivatives, hence a continuously differentiable solution satisfying the equation also at the join. This proves existence of that particular continuation, not uniqueness of all nonnegative continuations. For a negative exponent, the right-hand side has not been defined at zero, so no such argument is available without specifying a different extended problem.
For an exponent greater than one, the continuous nonnegative extension is locally Lipschitz in the state even at zero, since the derivative of the power is bounded on every bounded nonnegative interval. Gronwall applied to differences in either direction gives uniqueness through zero. A positive solution therefore cannot reach zero at a finite interior point, consistently with (6).
Proof X-ray
Solve linearly, reconstruct conditionally
The endpoint proof uses four facts in a fixed order. First, and are continuous on all of , so the linear solution is defined and continuous at an interior endpoint; at a coefficient boundary, such as in the first worked example, this step is unavailable and the theorem says nothing. Second, a connectedness squeeze fixes the endpoint value: a positive value would contradict maximality of the component, a negative one the positive one-sided values, so there. Third, only the sign of translates into a state limit, which is why the same linear zero means blow-up for and arrival at zero for . Fourth, continuation is decided in the original equation, not in the linear one. For the joined function is because ; for the Lipschitz property at zero forbids reaching zero at all; for the right-hand side is undefined at zero and the argument stops.
Theorem 1.7.4
(All real branches for nonlinear integer powers)
For an integer exponent other than zero and one, consider
For positive integer exponents the state domain is the whole real line and the zero function is a solution. For negative integer exponents the state domain excludes zero. A solution with nonzero initial value has fixed sign and is obtained by solving
then using the integer coordinate exponent and the appropriate inverse:
The linear formula is still (3), with exponent and initial transformed value replaced as in (8). Since the integrating factor is positive, the maximal interval is explicitly
This lists every real solution through an admissible datum, with the zero solution added in the positive-integer case. Different inverse signs cannot be switched freely along one solution.
Proof
(Fixed signs and one-to-one powers on each half-line)
For positive integer exponents the polynomial field is continuous in the independent variable and locally Lipschitz in every real state. Thus a nonzero solution cannot meet the zero solution at an interior point: local uniqueness through that point in both directions would force equality. For negative integer exponents the field is locally Lipschitz on each nonzero state half-line, and zero is outside the prescribed state domain. In either case the sign of a nonzero solution is fixed. On either half-line the integer power coordinate is smooth with nonzero derivative, and (5) remains valid.
An odd integer coordinate power has the same sign as the state; an even integer coordinate power is positive on both state half-lines and forgets the sign. This is true also for negative integer coordinate powers. On each fixed half-line the map is one-to-one. These facts give (9) and the bracket signs in (10). The inverse formulas solve the original equation by the chain rule; their intervals end at a coefficient boundary or a zero of their transformed value. At an interior zero the state diverges if the original exponent is positive, and tends to zero if it is negative. In the latter case the endpoint state is outside the equation's domain, so it still has no classical continuation within that domain. This proves the classification without claiming a removable-state extension.
Concept lensStructural
The sign belongs to the inverse problem
The transformed linear coordinate propagates one number. The original state also needs an admissible range and, for some integer powers, a remembered sign. A linear zero can mean infinite amplitude, a zero-state boundary, or a forbidden state. The formula becomes a solution only after those inverse requirements and its connected interval are fixed.
Worked examples
Worked example 1.7.5
(A nonconstant-coefficient equation with different signed branches)
On the positive independent-variable half-axis, consider
The transformed equation and its general solution are
For initial value one at first coordinate one, the constant is two and the maximal solution interval is from zero to two. The first endpoint is a coefficient-domain boundary, where the coefficient is singular and this explicit graph also diverges; it is outside the interior-endpoint theorem. The other endpoint is an interior transformed zero and produces blow-up. For initial value minus one at the same first coordinate, the constant is zero, giving
The negative solution exists throughout the positive coefficient interval. Here , so the coordinate exponent is odd. The transformed value is negative, as required for the odd-power negative branch. The zero solution must be included separately because the reciprocal coordinate cannot represent it.
Worked example 1.7.6
(A square root, a zero endpoint and a valid zero extension)
Take the nonnegative-state equation and positive initial datum
On a positive branch the linear coordinate is the positive square root, so
The branch tends to zero at its left endpoint when the independent variable decreases. In the increasing direction it departs from zero; there is no blow-up there. Extending the state by zero on the left gives a continuously differentiable nonnegative solution on the whole real line. Squaring the negative transformed values instead would give a different positive formula that fails the original square-root equation. The legitimate continuation and the inadmissible algebraic reconstruction are distinct operations.
Worked example 1.7.7
(Integer powers and the remembered initial sign)
For the cubic equation, equal positive and negative initial magnitudes produce the same transformed value:
The sign is fixed by the original initial datum; both signs give genuine solutions, but switching between them does not. At the transformed zero both magnitudes diverge. There is also the original zero solution, which never entered this calculation.
For comparison, a negative integer exponent can carry a negative state to a forbidden zero endpoint:
Its transformed bracket is negative on the selected interval. The state tends to zero from below at the finite right endpoint, while its derivative diverges positively. There is no zero solution of this prescribed negative-power equation.
Common mistakes / subtle points
Common mistake
(Squaring a negative linear value back)
For the positive square-root equation, the formal square below appears to be real everywhere:
When the transformed value is negative, the derivative of the candidate is negative while the original square-root right-hand side is positive. In particular, at first coordinate minus three the derivative is minus one half and the right-hand side is one half. The inverse formula is valid only when the transformed value is positive. A real-valued expression is not sufficient evidence of equivalence.
Common mistake
(Losing the zero solution in the substitution)
Dividing by to obtain (5) assumes . For an integer exponent the zero function is a separate solution that no value of the transformed constant represents: for the family never vanishes. For the loss is more serious, because nonuniqueness sits exactly at the zero solution, as the waiting solutions below show. Whenever zero is an admissible state, a complete answer lists the zero solution next to the transformed family.
Counterexample mode
Linear uniqueness does not decide the original zero initial value
For an exponent strictly between zero and one, the nonnegative equation has waiting solutions:
The derivative of the positive piece tends to zero at departure, since its power is greater than one. Thus all departure times at or after a zero initial point give solutions through that datum, in addition to the permanently zero solution. The transformed waiting function is zero before departure and linear after it; it is not differentiable at departure and does not solve the transformed inhomogeneous equation on the zero segment. The derivation of (5) divided by a state power and was never valid there. No uniqueness theorem for that linear equation can remove the original waiting family.
The waiting solutions leave a finite zero limit that can be continued. For a negative power, a finite zero limit can instead be a genuine end of the solution.
Counterexample mode
A finite zero limit can be a singular endpoint
For a negative power, take
Here the positive state tends to zero at its finite endpoint and its derivative diverges negatively. Initial value minus one instead gives the negative square-root branch, with positively diverging derivative at the same endpoint. Setting either state to zero afterward is not a classical continuation: the original right-hand side is undefined there. Derivative divergence is a fact of this example, not an assertion about every negative-power coefficient choice. The state limit alone does not determine whether the equation can be continued.
Practice and solutions
Quick check
(Solve the equation below for initial value one and initial value minus one, each at first coordinate zero. Give the maximal real interval for each and identify the zero solution.)
Use the reciprocal coordinate and keep its sign when selecting the interval.
Solution
The transformed equation is
For the two data this gives
Each interval is the correct-sign component containing zero. At its finite endpoint the denominator vanishes and the solution diverges, so it cannot be extended there. The zero solution exists on the entire real coefficient interval and was excluded by the reciprocal substitution.
Quick check
(Treat exponents zero and one directly. Derive their initial-value formulas for continuous coefficients, and state when the zero function is a solution. Interpret exponent zero as the linear forcing equation.)
At exponent one, do not divide by one minus the exponent. Exponent zero is the usual linear forcing equation.
Solution
For exponent zero, the equation and formula are
The product rule proves the formula and its uniqueness, with no positivity restriction. The zero function solves it exactly when the forcing is identically zero on the interval. For exponent one,
Direct differentiation verifies it; multiplying by the reciprocal exponential proves uniqueness. The zero function is always a solution, and a nonzero solution retains its initial sign. The attempted power coordinate at exponent one is constant and carries no initial-state information.
Quick check
(Solve the fourth-power equation below from initial values one and minus one at zero. Compare their forward behavior and maximal intervals.)
The reciprocal cube has the same sign as the state.
Solution
Use the signed linear coordinate:
The positive solution blows up in finite forward time. The negative one increases toward zero as the independent variable tends to positive infinity and has no finite forward blow-up; its blow-up is backward at the other endpoint. Both statements follow from the signed inverse and agree with the positive derivative of the original equation. The two signs have different future behavior for an even power. The zero function is the remaining solution through zero initial state and exists on the whole real line.
Quick check
(For the nonnegative-state equation below, find the positive branch through the datum. Show that its positive interval ending at zero does not determine a unique nonnegative past through that endpoint.)
Use the square-root coordinate where positive, then verify two continuations in the original equation.
Solution
The linear coordinate and positive branch are
The maximal positive component through the datum excludes zero. Two nonnegative continuations on the full real line are
For the first candidate, its state derivative and the original right-hand side agree explicitly:
For the second, both value and derivative match at zero, and the original equation holds on each side and at the join. They agree at and after zero but have different pasts. These are two examples, not an exhaustive list of nonnegative continuations. The square-root transform of the second is constant on its zero segment, so it does not satisfy the transformed linear equation there. Positive-branch uniqueness and uniqueness in the enlarged nonnegative class are different questions.