An initial-value problem specifies both the value and derivative at one endpoint. A boundary-value problem instead imposes conditions at two different endpoints, so a solution built from the left may miss the condition on the right. A Green function resolves that tension by joining one homogeneous solution adapted to each endpoint.
The join is continuous, but its derivative has a prescribed jump. That jump produces the forcing after integration. We will derive the formula, show exactly when it exists, and distinguish nonresonance from the additional sign condition that makes the kernel positive.
The regular problem
Let , with , and be real. Consider
Definition 2.4.1
(Endpoint solutions and the weighted Wronskian)
Let and solve the homogeneous equation with
The linear initial-value theorem supplies both functions throughout . Their weighted Wronskian is
The weighted Wronskian is constant: differentiating and using , gives . At the endpoints,
These two equalities include the endpoint normalizations and are useful tests of the sign in the kernel.
Theorem 2.4.2
(Nonresonant Green kernel)
The following are equivalent:
- .
- The homogeneous Dirichlet problem , has only the zero solution.
- For every continuous , the Dirichlet problem has exactly one solution.
When these conditions hold, define
Then and the unique solution is
For fixed interior , is continuous in , vanishes at , solves away from , and satisfies the flux jump
Proof
(The Wronskian detects resonance)
A homogeneous solution satisfying is a scalar multiple of : its initial derivative fixes the multiple, and uniqueness identifies the solutions. Such a multiple also vanishes at precisely when that scalar times is zero. Thus is equivalent to uniqueness of the zero homogeneous solution.
If , is a nonzero homogeneous Dirichlet solution, so even the forcing has more than one solution. Hence uniqueness for every forcing requires . The kernel construction below proves existence and uniqueness when it is nonzero.
Proof
(Join the endpoint solutions and differentiate the integral)
The two formulas for agree at . The branch vanishes at and the branch at ; away from each is a multiple of a homogeneous solution. At the join,
The negative sign is required because the operator begins with . Symmetry follows directly from the two cases in the formula.
For continuous , write the integral as
The two boundary terms from differentiating the integrals cancel, giving
Differentiating once more produces the extra term
The remaining terms satisfy the homogeneous equation, so . The formula also gives zero endpoint values. Continuity of the coefficients and gives a classical solution. If two solutions have the same forcing, their difference is a homogeneous Dirichlet solution and is zero by the preceding argument.
Proof X-ray
The sign at the join matters
Each hypothesis enters at a definite place. Regularity, with , makes the initial-value theorem supply and on all of ; each factor of is then responsible for one boundary condition, so the formula satisfies before any forcing is used. Nonresonance is used only to divide, and uniqueness comes from the homogeneous argument, not from the formula.
The non-obvious move is the second differentiation. The first derivative has no boundary terms, because is continuous at the join. The second produces exactly the flux jump, , and the minus sign in turns it into . Reversing the sign of the jump would solve the problem with forcing .
Positivity is an additional conclusion
Theorem 2.4.3
(A positive kernel when )
If on , then and
Consequently implies ; if is continuous, nonnegative and not identically zero, then for every .
Proof
(Endpoint-normalized solutions stay positive)
Near , because . As long as , the equation gives , so and is strictly increasing. It cannot have a first zero to the right of ; therefore on . Applying the same argument after reversing the interval shows on . Consequently and the kernel is positive for interior arguments.
Integrating a nonnegative forcing against this kernel gives a nonnegative solution. If is positive at some point, continuity makes it positive on a nonempty interior subinterval, so the integral is strictly positive at every interior .
Examples and resonance
Worked example 2.4.4
(The Dirichlet inverse of the second derivative)
For , , , one has , , and . Thus
For , integration gives . It is positive inside, vanishes at the endpoints and satisfies . The kernel has a corner at ; demanding a continuous derivative there would remove the very jump needed to reproduce the forcing.
Common mistake
(Asking the kernel to be smooth at the join)
If had a continuous derivative at , it would be a homogeneous solution on all of with zero boundary values, hence zero in the nonresonant case, and the integral formula would return for every . The corner is not a defect: the flux jump is what reproduces the forcing.
Worked example 2.4.5
(A negative potential can still be nonresonant)
On , let . The normalized solutions are and , giving . The kernel formula therefore applies even though . In fact these two factors are positive on the relevant interior intervals, so this particular kernel is positive as well. The condition is sufficient for positivity and nonresonance, not necessary.
Worked example 2.4.6
(A variable coefficient )
On let , so and . Since and , we get and . Similarly and give . Both endpoint tests agree: . Hence for . For ,
and the kernel formula gives
Check: , so , and , . Edge case: at the right endpoint , so the flux condition means . Normalizing instead would give and , so the test holds only for the flux normalization.
Counterexample mode
At resonance an inverse kernel cannot exist
On , consider . The nonzero function is a homogeneous Dirichlet solution, so . With forcing , no solution exists: multiplying by , integrating by parts and using both endpoint conditions gives zero on the left but on the right.
For forcing , solutions do exist but are not unique: . Thus the same resonant operator can exhibit incompatibility or nonuniqueness, depending on the forcing. Dividing the Green formula by cannot repair either issue. A modified kernel on compatible forcings with a normalization is a different construction and is outside this theorem.
Common mistake
(Expecting a positive kernel from nonresonance)
The nonresonant theorem does not claim positivity for arbitrary . For instance, replacing on by has lowest Dirichlet eigenvalue ; for , its Dirichlet solution is in the interior. It is nonresonant with , since . The sign hypothesis and invertibility answer different questions.
Exercises
Quick check
(On , use the explicit kernel for to solve with zero endpoint values. Verify the result directly.)
Solution · Answer
for and for . Integrating over and gives . Direct differentiation gives , with zero endpoint values.
Quick check
(For on , find and write for .)
Solution · Answer
, and . Therefore when ; symmetry gives the other case.
Quick check
(On , explain why has no zero-Dirichlet solution.)
Solution · Answer
Multiply the equation by the homogeneous Dirichlet function and integrate by parts using both endpoint conditions. The left side is , whereas the right side is , a contradiction.
Quick check
(Suppose is the nonresonant kernel and . Which property of gives , and what sufficient condition on was proved? Give the argument.)
Solution · Answer
The property needed is ; the positivity theorem proves it when . Since , the difference is nonnegative. The example with shows this is a sufficient hypothesis, not a necessary one.