From one known solution to a usable formula
Constant-coefficient equations admit exponential trials, but a variable coefficient can make such trials unhelpful. Two complementary questions arise. Given one homogeneous solution, can we construct another? Given two independent homogeneous solutions, can we use them to solve an equation with a forcing term?
Both answers require care about division. A nontrivial solution can vanish at individual points, so a formula that divides by it may work only on a smaller interval. A fundamental pair has a different safeguard: its Wronskian never vanishes on the coefficient interval. We derive these facts instead of treating the resulting integrals as recipes.
Let be an open interval and let be continuous. Consider
Solutions are twice continuously differentiable on . We use existence and uniqueness for the associated first-order linear initial-value system on .
The Wronskian and Abel's identity
Definition 1.11.1
(Wronskian and fundamental pair)
For two differentiable functions on , the Wronskian is
Two solutions of the homogeneous equation form a fundamental pair if every solution of that equation is for some constants .
The Wronskian of two solutions obeys its own first-order equation, which is why it can vanish only in a controlled way.
Theorem 1.11.2
(Abel's identity and the Wronskian test)
Let solve on and let . Then
Hence is either identically zero or nowhere zero on . Moreover, are linearly independent if and only if , and in that case they form a fundamental pair.
Proof
(A first-order equation for the Wronskian)
Differentiating, the terms cancel, and the equation gives
This first-order linear equation has the unique solution (2). In particular, if , it is nonzero everywhere on . The initial-value vectors and then form a basis of . Matching any homogeneous solution's initial data and using uniqueness proves that form a fundamental pair. Conversely, if , a nonzero linear combination has zero value and derivative at , so uniqueness makes that combination identically zero. Thus for solutions of this same homogeneous equation, independence and nonzero Wronskian are equivalent.
Variation of parameters
Let be a fundamental pair with Wronskian , which is nowhere zero by Abel's identity.
Theorem 1.11.3
(Variation of parameters)
For , the solution of (1) with is
Every solution of (1) is with constants .
Proof
(Two unknown coefficients and one auxiliary condition)
Seek and impose the auxiliary condition . Then , and substitution into (1) reduces the remaining condition to . Solving this two-by-two system gives
Integrating (3) from gives (4). The integrands are continuous because is nowhere zero. At both integrals vanish, and the auxiliary condition gives . Differentiating confirms (1). Adding gives every solution: the difference of any two solutions is homogeneous. The auxiliary condition is a choice that removes redundant derivatives of the two unknown coefficient functions; it is not an additional condition on the desired solution.
Proof X-ray
What permits division by the Wronskian?
The coefficient matrix in (3) has determinant . Abel's identity and independence ensure this determinant stays nonzero on the entire interval. This is different from checking that the individual functions are nonzero: either may cross zero while their Wronskian remains nonzero, whereas two nonzero functions may have zero Wronskian.
Worked example: variable coefficients and forcing
Worked example 1.11.4
(Forcing with the homogeneous pair , )
On solve
Direct substitution gives the homogeneous pair , , with . Applying (4) at gives
Indeed and gives . With , the left side is . Uniqueness identifies this as the solution on . The coefficient functions are singular at zero, so this calculation does not assert an initial-value theorem through .
Reduction of order, including its domain
Theorem 1.11.5
(Reduction of order)
Suppose a nontrivial homogeneous solution has no zeros on an interval , and let . Then
solves the homogeneous equation on , and its Wronskian with is . Thus form a fundamental pair on .
Proof
(Seeking the second solution as a multiple of the first)
To derive this, write . Substituting into the homogeneous equation leaves . Since on , the integrating-factor equation for gives . Equation (5) follows. Its Wronskian with is exactly , so the two solutions are independent on .
A zero of restricts the formula, not necessarily the second solution itself.
Worked example 1.11.6
(Reduction from near its zeros)
For , take on and . Here , so the exponential factor is , and
Its Wronskian with is , as the theorem predicts. The function extends as a solution across the zeros of , although the formula itself was only derived on .
Common mistake
(Basing the reduction integral at a zero of )
In the sine example the integral in (5) cannot be based at : diverges. Extension of across a zero of is justified by the simplified solution or by the original initial-value theorem, not by treating a divergent integral as finite.
Counterexample mode
A Wronskian test needs the equation hypothesis
The functions and are and linearly independent on , but their Wronskian is zero everywhere: on each half-line they are proportional, and at zero both functions and derivatives vanish. They cannot both solve one equation of the form (1) with and continuous coefficients on an interval containing zero, because uniqueness with zero initial data would force each to vanish identically. Abel's equivalence is a theorem about solutions of the same regular equation, not arbitrary functions.
The response kernel and a nonconstant leading coefficient
Formula (4) was written using one particular fundamental pair. Two questions remain. First, the pair is not unique: replacing by , say, changes the first integral in (4) and the factor in front of the second. Does it change the answer? Second, equations rarely arrive in the form (1). They usually arrive as
and the right-hand side is not necessarily the function in (4). Both questions are settled by rewriting (4) as a single integral against a kernel and identifying the kernel by its initial data.
Combining the two integrals in (4) under one integral sign gives
Proposition 1.11.7
(Characterization of the kernel)
Let be continuous on the open interval . For each fixed , the function is the unique solution of the homogeneous equation on satisfying
Consequently does not depend on which fundamental pair is used to build it.
Proof
(Three properties determine the kernel)
For fixed , the function is a constant linear combination of and , so it solves the homogeneous equation. Setting makes the numerator zero. Differentiating in and then setting gives . The initial-value theorem gives exactly one homogeneous solution with these two values at . Any other fundamental pair produces a function with the same three properties, hence the same function.
Proof X-ray
Which data are invariant, and which are choices?
The proof never computes from a formula; it uses three properties: solves the homogeneous equation, , and . The decisive hypothesis is uniqueness for the homogeneous initial-value problem with base point , for every ; this is why must be continuous on all of , not only near . The diagonal derivative equals only because the numerator's -derivative at is the Wronskian, and allows the division.
The fundamental pair, the base point used in Abel's identity, and the constants of integration are choices; the kernel is determined by the equation alone. Its diagonal values are the normalization for an equation whose leading coefficient is one. For (6), the kernel relevant to is , whose diagonal derivative is ; this is precisely where the division by enters.
The proposition also gives a direct proof that (7) solves (1), which does not refer back to the auxiliary condition. Because are and is continuous and nowhere zero, , and are continuous on . Differentiating under the integral with a variable upper limit gives
Adding times the first line and times to the second leaves plus the integral of , and that bracket is zero for every . The two normalizations have separate roles. removes a boundary term from . produces exactly one copy of in . The same computation is valid for , since the integrals are oriented.
This interpretation is concrete. is the homogeneous motion that is at position zero with unit velocity at time : the response of a system at rest at the origin to a unit velocity impulse delivered at time . Formula (7) superposes these motions with weights . For constant coefficients, is the translate of the normalized kernel with value and derivative at time . Then depends only on , which recovers the convolution formula for constant-coefficient operators. With variable coefficients the response generally need not depend only on , so this simple convolution formula need not apply.
Equations with a leading coefficient. In (6) assume that are continuous on an open interval and that has no zero on . Dividing by gives (1) with
The homogeneous solutions, the Wronskian and the kernel are the same for (6) and the divided equation, since a function satisfies one homogeneous equation exactly when it satisfies the other. The forcing, however, is divided. The solution of (6) with zero value and derivative at is therefore
Abel's identity (2) must likewise use . If one omits the division and uses in place of , the resulting function solves . This generally changes the forcing; the two forcings agree exactly when throughout the interval. If has zeros, division applies on each connected open interval where ; behavior of the original equation at the zeros requires separate analysis, as with the singular leading coefficient discussed in the first-order linear note.
Worked example 1.11.8
(A leading coefficient divides the forcing)
On solve
The leading coefficient has no zero on . Substitution shows that and solve the homogeneous equation: for the left side is , and for it is . Their Wronskian is , which agrees with Abel's identity for : up to a constant. The kernel is
One checks and . The divided forcing is , so (8) gives
Then , and gives . To verify the equation, write plus the homogeneous term . For any function , a short calculation gives
With the bracket is , as required. Uniqueness identifies this as the solution on . In particular, is a particular solution, and the remaining terms are homogeneous and adjust the initial data.
Common mistake
(Inserting in place of )
If the undivided forcing is inserted in place of , the integral becomes . With the same identity gives . That function solves the equation with forcing , which is times the intended forcing. The error is not a misprint in the final answer but a misidentification of .
Practice
Quick check
(For , , use , in (4), and simplify.)
Solution · Answer
Here . Integration by parts gives and , so
It has both stated initial values and .
Quick check
(If and a fundamental pair has , find on .)
Solution · Answer
Equation (2) gives , which never vanishes.
Quick check
(For , start with and . Use (5) to find and its Wronskian with .)
Solution · Answer
With and , equation (5) gives and .
Quick check
(Why does the formula (5) not directly apply on an interval containing a zero of , even when a second solution exists there?)
Solution · Answer
It divides by . At a zero the integrand is undefined and may be nonintegrable; continuation of the resulting solution requires a separate argument, as in the sine example.
Exercise 1.11.9
(On , use the pair , to compute the kernel in (7) for . Check its two normalizations, and confirm that the pair , gives the same kernel. Then use (8) to solve with .)
Core
Solution(5)
- 1
Compute the Wronskian and the kernel
, so .
- 2
Check the normalizations
, and equals at .
- 3
Change the fundamental pair
For , the Wronskian is , and the numerator is . The kernel is unchanged, as the proposition predicts.
- 4
Divide the forcing by the leading coefficient
Key step
The leading coefficient is , so , and (8) gives .
- 5
Verify
, , . Hence , and .