A point source is too concentrated to be an ordinary function, yet it is a useful idealization: once we know its response, we can add responses from a distributed source. The singularity is not an inconvenience to discard. Its flux carries exactly the unit mass that makes the construction work.
Finding the radial kernel
Let and let denote the surface area of the unit sphere in ; thus and . We use the convention . Away from the origin a radial harmonic function satisfies
so is constant. The desired radial derivative must also give the flux , which fixes that constant as . Integrating once more, and discarding the additive constant, gives the kernel below.
Definition 5.5.1
(The fundamental solution of the Laplacian)
For set
In both cases . A distribution with is called a fundamental solution of ; the theorem below shows that is one.
Each kernel is locally integrable: in polar coordinates, the singular radial integrals behave like for and for . Hence defines a distribution, and the question is what its distributional Laplacian is.
Theorem 5.5.2
(The radial kernel is a fundamental solution)
The function in (1) belongs to , is harmonic on , and satisfies in the sense of distributions: for every ,
Proof
(Excising a small ball around the source)
For choose a large ball containing its support and remove . Green's identity and on the resulting annulus yield
The outward normal of the punctured domain is , which explains the signs; the outer sphere contributes nothing because vanishes there. The first term is bounded by , which is for and for . In the second term , so it equals the average of on the sphere and tends to by continuity. Local integrability of lets the left side tend to the whole-space integral by dominated convergence.
Proof X-ray
Where the delta mass comes from
The proof never differentiates at the origin. It moves both derivatives onto , then cuts out so that the classical identity can be used on what remains. Three facts are used, each at one place. Local integrability passes the limit on the left. The size estimate kills the term containing . The flux normalization turns the remaining term into a spherical average, and this is the only step that produces . A kernel would pass the first two checks and give , so the constant in (1) is forced by the flux and nothing else. The fact that away from the origin does not erase the source.
The last sentence is worth isolating, because the pointwise computation is easy to over-read.
Counterexample mode
Harmonic away from the source is not harmonic everywhere
Claim: a locally integrable function that is harmonic on is harmonic as a distribution on all of . In three dimensions is locally integrable and has zero classical Laplacian at every , but , so its outward flux through every centered sphere is , not . By the theorem, , so cannot be extended as a harmonic function across the origin. The correct statement needs a size condition: an isolated singularity of a harmonic function is removable when the function stays bounded near it.
Superposition for a smooth source
Let and set
This function is called the Newtonian potential of (the logarithmic potential when ).
Theorem 5.5.3
(The Newtonian potential solves Poisson's equation)
Let and . Then is and
If , then as , and is the only classical solution of (3) that tends to zero at infinity.
Proof
(Differentiate the smooth factor, then convolve with the delta)
The integral is finite by local integrability. On every compact set in , the relevant values of lie in a fixed bounded set. After substituting , derivatives can be passed to the smooth compactly supported factor under the integral. Thus and . Distributional differentiation and the previous theorem give
and this is a classical equality because both sides are smooth. For , if is supported in then for the bound on the support gives , so . Any other classical solution tending to zero at infinity differs from by a harmonic function tending to zero. On the maximum principle gives , and letting gives .
In two dimensions the situation is different, because itself grows logarithmically. Put .
Proposition 5.5.4
(Logarithmic growth in two dimensions)
Let , supported in , and . Then for
If , no classical solution of tends to zero at infinity.
Proof
(Expand the logarithm and follow the circular average)
For and , and , so uniformly on the support. Integrating against gives (4).
Let be any classical solution and its average over the circle of radius . For , integrating over and using the divergence theorem gives , so
Hence for , which does not tend to zero when ; so cannot tend to zero either.
If , the particular potential (4) does tend to zero. A bounded entire harmonic function may still be added to a solution, so boundedness alone does not determine its additive constant.
Quick check
(For and , can the potential be bounded on ?)
Look at the leading term in (4).
Solution · Answer
No. By (4), is bounded for large , so like . The sign of only decides whether (for ) or (for ).
Worked example: a uniform ball
Worked example 5.5.5
(Uniform ball in three dimensions)
For , let with . Although this source is discontinuous at the sphere and lies outside the smooth-source theorem, radial symmetry and matching give a solution that is and piecewise :
Inside, gives ; outside, is harmonic, so . At both pieces equal and both radial derivatives equal , so no surface delta is created. The total source is , and the exterior field is exactly . This is a special consequence of radial symmetry; a general source has higher-order spatial information.
The same matching works in the plane, where it shows the logarithmic growth of the proposition in closed form.
Worked example 5.5.6
(Uniform disk in two dimensions)
Let and with , so . For a radial source, the mean value property of the harmonic function on the disk shows that the potential outside the support is exactly :
Inside, a radial function with has the form , since in the plane. The radial derivatives already agree at : on both sides. Matching the values gives , so
This is an edge case for the uniqueness statement: the source has positive mass, and the potential tends to at the rate instead of decaying.
The flux condition also fixes the constant in dimensions where is less familiar.
Worked example 5.5.7
(The kernel in four dimensions)
For the unit sphere has area , so (1) gives
Check: , and the sphere of radius has area , so the outward flux equals for every . Local integrability holds because near the origin.
Common mistakes
Common mistake
(Mixing the sign conventions)
Some texts normalize instead of . Then : for it is , and for it is . Using the kernel from one convention in the solution formula of the other produces with the wrong sign. A quick test is the flux: with the convention here, must be .
Common mistake
(Matching values but not normal derivatives)
When pieces of a radial solution are glued at an interface, continuity of is not enough. A jump across the sphere adds the surface distribution on that sphere to . In the uniform-ball example both and must match for to hold with no extra layer.
Practice with solutions
Quick check
(Verify that the kernel (1) has flux through the sphere of radius .)
Multiply the radial derivative by the area of the sphere.
Solution · Answer
and , so , independent of and of .
Quick check
(For and compactly supported smooth , give the large- expansion of (2) through order .)
Expand for bounded.
Solution · Answer
with , by the first-order Taylor estimate , uniform for in the support. If , the displayed term vanishes and .
Quick check
(For , can a source of nonzero total mass have a solution tending to zero at infinity?)
Use the circular-average argument.
Solution · Answer
No. The circular average of any solution satisfies outside the support, so , which does not tend to zero; (4) shows the same logarithmic growth for the potential itself.
Quick check
(Find the value at the center in (5), and explain why matching only but not would be insufficient.)
Set in the inner formula.
Solution · Answer
. A jump of the normal derivative at would produce an additional distribution supported on the interface, so the glued function would solve Poisson's equation with an extra surface source.