Motivation
A harmonic function is initially defined by two derivatives and one equation. Yet averaging over every small sphere forces much more regularity than that definition requests. The key is to replace a sharp spherical average with a smooth radial average. It gives exactly the same function, while allowing arbitrarily many derivatives to fall on the averaging kernel.
Positivity adds another kind of control. A nonnegative harmonic function cannot be tiny at one interior point and arbitrarily large nearby. Comparing overlapping ball averages proves a quantitative Harnack inequality. Connecting finitely many such comparisons extends it across compact subsets of a connected domain.
These local facts have global consequences, but their hypotheses matter. An entire harmonic function bounded on just one side is constant. A bounded planar harmonic function near an isolated missing point extends through it. Neither statement says that every positive harmonic function on an arbitrary domain is constant, or that an unbounded singularity can be ignored.
Definitions
Let be open, with . All functions are real-valued. A harmonic function is with . Write for an open ball, for unit-sphere area, and require whenever a closed ball is used below.
Definition 5.4.1
(Mean values and a smooth radial average)
The sphere and ball mean-value identities are
Choose a nonnegative radial with integral one, and set . A multi-index records spatial differentiation order; write .
The smoothing kernel is chosen once. Its constants may depend on the dimension, derivative and this choice, but not on the harmonic function or the ball radius. All conclusions here concern the interior; no regularity of the domain boundary is assumed.
Theorem/Proposition
Theorem 5.4.2
(Mean values characterize harmonicity and give derivative bounds)
Every harmonic function satisfies both mean-value identities and is in the interior. Whenever and , it obeys
The constants depend only on dimension, the multi-index and the fixed averaging kernel, as defined above.
Conversely, a continuous function satisfying the sphere mean-value identity for every admissible center and radius is smooth and harmonic. Agreement of averages about only one center is not this hypothesis.
Theorem 5.4.3
(An elementary Harnack inequality)
If is harmonic and , then for ,
Consequently the supremum on that smaller ball is at most times its infimum. These constants are sufficient, not optimal. If is connected and is a nonempty compact subset of it, there is a finite constant depending on and such that
A nonnegative harmonic function on a connected domain is either everywhere positive or identically zero.
Concept lensGeometric
Harnack comparison controls scale and distance, not just values
The constant comes from interior balls and the number of comparisons needed to travel between points. A compact subset stays a positive distance from the boundary, so finitely many controlled steps suffice. As points approach a boundary singularity, that control can deteriorate. Positivity is what allows integral containment to become a value comparison.
Positivity also drives the global statements: a one-sided bound is converted into a two-sided one on large balls.
Theorem 5.4.4
(Global rigidity and bounded planar punctures)
An entire harmonic function on that is bounded below, or bounded above, is constant. Two-sided boundedness is unnecessary.
In the plane, a harmonic function on a punctured disk that is bounded in a neighborhood of its missing center has a unique harmonic extension across that center. This is a classical bounded-singularity theorem; it does not assert arbitrary distributional or boundary regularity.
Proof sketch or proof idea
Proof
(Mean values, exact smoothing and the converse)
Differentiate the sphere average with respect to radius. The divergence theorem gives
Its limit at zero radius is the center value, proving the sphere identity. Radial integration proves the ball identity. This argument only uses smooth spheres contained in the domain, not smoothness of the domain's own boundary.
For a radial averaging kernel, integrate first over concentric spheres. Each sphere average equals the center value, so its weighted average is still that value:
The same identity holds near the center with a fixed admissible radius. Locally one may multiply the function by a compact cutoff equal to one on the needed neighborhood, so convolution requires no global extension of the harmonic equation. Differentiating the smooth compact kernel then proves smoothness and
This proves the estimate. The same smoothing argument starts with a merely continuous function having all sphere means. After smoothness is known, Taylor expansion of its sphere average gives
The linear terms average to zero and the coordinate second moments are equal to one divided by the dimension. Equality to the center value for every small radius forces zero Laplacian, proving the converse.
Proof X-ray
A smooth average can equal the original function exactly
General continuous functions are only approximated by mollification. Harmonic mean values make every sufficiently local radial mollification equal to the function itself, and the proof uses each hypothesis at one place. The equation enters only through the divergence theorem on spheres inside , giving the sphere identity. Radial symmetry of is what turns the convolution into a weighted average of sphere averages, so that it reproduces . The support condition and keep every sphere inside the domain. The identity then lets all derivatives fall on the kernel, and the scaling produces the power of in the estimate. Regularity is proved before derivatives of arbitrary order are used, so the argument does not assume the smoothness it is meant to establish. For the converse, only continuity and the sphere means are needed to run the same smoothing step.
Harnack's inequality needs only the ball identity and positivity.
Proof
(Harnack by ball containment and chains)
For Harnack, fix a point in the quarter-radius ball. The geometric containments are
Nonnegativity permits comparison of the integrals over these sets. The mean-value identity on the middle ball, together with the volume ratios, gives
The ball closures remain inside the original admissible ball, so all averages are legitimate. A connected open subset of Euclidean space can be traversed by a finite chain of sufficiently small interior balls between any two fixed points: the points reachable by such chains form an open and relatively closed subset. Consecutive reference points can be placed in quarter-radius balls, and each comparison multiplies by a finite dimensional factor. Cover a compact set by finitely many smaller balls and connect their centers to one reference point. There are only finitely many factors, giving a constant independent of the particular nonnegative harmonic function.
If such a function vanishes at one point, the local inequality makes it zero in a neighborhood. Its zero set is closed by continuity and open by the same argument, so connectedness makes it the whole domain. Applied to the nonnegative difference between a global attained maximum and a harmonic function, this also proves the strong maximum conclusion for an interior maximum over a connected domain.
The two global statements combine these local estimates with a limit in which the radius grows or shrinks.
Proof
(Liouville by large balls, and a logarithmic barrier at the puncture)
For Liouville, first suppose an entire harmonic function is nonnegative. On every ball about a fixed point, Harnack bounds the supremum on its quarter-radius ball by a fixed multiple of the center value. Apply the first-derivative estimate on that smaller ball. Its right side is a constant times the center value divided by the large radius. Let the radius tend to infinity: every first derivative at the chosen point is zero. The point was arbitrary, so the function is constant. A function bounded below becomes nonnegative after subtracting a constant; one bounded above is handled by changing sign.
For planar puncture removal, translate the missing point to zero and choose a circle of radius strictly within the disk. The function is bounded on the punctured smaller closed disk: use its bound near zero and compactness away from zero. Let be the harmonic Poisson extension of its continuous values on that circle, as constructed in the disk Dirichlet note. The difference is bounded on the punctured disk and zero on its outer circle. Choose bounding its absolute value there. On an annulus with inner radius , the function
is harmonic, equals the bound on the inner circle and is zero on the outer one. The annulus maximum principle applied to both signs gives
At a fixed nonzero point, let the inner radius decrease to zero. The right side tends to zero, so the difference vanishes. The Poisson extension therefore supplies the harmonic value at the puncture. Elsewhere it already agrees with the original function, and continuity makes that value unique. This proof uses the previously constructed continuous-data disk solution and the classical annulus maximum principle, not an unproved general weak regularity theorem.
Proof X-ray
Where the one-sided bound is spent
After a constant is subtracted, nonnegativity is used once: Harnack on turns one-sided information into the two-sided bound , whose right side does not grow with . Entire harmonicity is used once: balls of every radius are admissible. The first-derivative estimate on then gives
and only the factor depends on the radius. Without positivity the supremum could grow with ; on a bounded or punctured domain the radius cannot grow. These are exactly the escapes of and below.
Worked examples
Worked example 5.4.5
(A positive harmonic quadratic on a bounded disk)
In the open unit disk of the plane, is strictly positive because there. Its Laplacian is zero. On every admissible circle centered at a point, the quadratic corrections have equal coordinate averages and cancel, giving the center value. It is a nonconstant function on a bounded domain; its polynomial extension to the whole plane is unbounded below as well as above. Neither observation contradicts Liouville, whose domain and one-sided bound must hold together.
Worked example 5.4.6
(Interior comparison cannot be uniform up to a singular boundary point)
In the unit disk consider
Direct differentiation verifies harmonicity, and the function is positive inside. Its center value is one. On the closed concentric disk of radius , its maximum and minimum are
For a fixed radius the denominator is smallest toward the boundary point on the positive horizontal axis and largest in the opposite direction; the extreme ratios grow monotonically with radius. Their quotient diverges as the smaller closed disk approaches the boundary. This shows why compact-interior dependence in Harnack cannot simply be removed.
Worked example 5.4.7
(Positive harmonicity on a punctured space is not entire harmonicity)
In three dimensions is positive and harmonic away from zero. For a radial function the Laplacian is its second radial derivative plus twice its first derivative divided by radius; substitution gives zero. It is nonconstant although bounded below. The missing point prevents use of entire Liouville, and its divergence there prevents a continuous harmonic extension. Decay at infinity does not repair a singularity inside the domain one wishes to include.
Counterexample mode
Signed harmonic functions need not obey positive Harnack bounds
The function is harmonic and zero at the origin but nonzero nearby. The upper Harnack bound in terms of its center value would incorrectly force it to vanish. The proof fails exactly where it compares integrals by set inclusion: a signed integrand has no such monotonicity. Its mean-value identities remain valid.
The puncture theorem has a hypothesis of the same kind, and dropping it fails just as concretely.
Counterexample mode
An unbounded planar puncture need not be removable
The function is harmonic on a punctured planar disk and zero on the unit circle, but diverges to negative infinity at the missing center. There is no finite continuous extension. Boundedness near the puncture was essential to choose one bound independent of the shrinking inner circle in the removal proof.
Common mistakes
Common mistake
(Checking mean values about one center only)
Mean values must hold at every admissible center and radius. A function odd in one coordinate has zero averages about the origin, even if it is not harmonic elsewhere: is the example in the exercise below on circle averages of .
Common mistake
(Treating interior constants as uniform up to the boundary)
Interior smoothness says nothing by itself about boundary values, and a compact-set Harnack constant is not a uniform constant for points arbitrarily close to the boundary; the second worked example shows the ratio diverging. In Liouville's argument, large balls must remain available inside the domain. A lower bound on a punctured or bounded domain does not replace entire harmonicity.
Common mistake
(Exchanging the two limits in the puncture proof)
In the puncture proof, the inner-circle estimate tends to zero at each fixed point, not uniformly on the varying annulus. Fix the point first, then shrink the inner radius; the annulus exercise below shows that the supremum over the open annulus stays equal to one.
Summary
Mean values yield exact radial smoothing and scale-sensitive derivative estimates. Positivity converts overlapping ball averages into Harnack comparison, and finite chains extend it across compact interior sets. Large-scale estimates force one-sided bounded entire harmonic functions to be constant. A shrinking logarithmic annulus barrier removes a bounded planar puncture, with the hypotheses and limiting operations kept explicit.
Exercises
Quick check
(Suppose an entire harmonic function satisfies , with and . Prove that it is constant.)
Apply the first-derivative estimate on increasingly large balls about a fixed point.
Solution
Fix a point . On its radius- ball the growth bound gives supremum at most . The derivative estimate is therefore bounded by a constant times this expression divided by . It tends to zero because . Every first derivative at the fixed point vanishes; varying the point proves constancy. This allows growth but requires it to be sublinear, rather than invoking the one-sided theorem without its hypothesis.
Quick check
(In the plane compute circle averages of about the origin and about a general center. Explain why the origin averages do not establish harmonicity.)
Odd angular terms vanish, but a translated quadratic angular term remains.
Solution
About a center , the circle average is
It equals the center value at the origin, but fails at centers whose first coordinate is nonzero. Indeed , not identically zero. The mean-value characterization requires all centers, not just a symmetry point.
Quick check
(In fix and . On the annulus , find the radial harmonic function with boundary values on and on . Compare its fixed-point limit with its open-annulus supremum as .)
The logarithmic radial solutions make the two limiting statements different.
Solution
The function is . Fix a point of radius and evaluate it only for . It tends to zero as the inner radius decreases. On each open annulus its supremum is one and is only approached at the inner circle; on the closed annulus with the boundary extension, the maximum one is attained there. Thus convergence is not uniform on the changing open annulus. The puncture-removal proof needs only the fixed-point limit after the same bound has been imposed on every inner circle.
Quick check
(Assume the Arzelà–Ascoli theorem. For nonnegative harmonic functions on a connected domain whose values at one fixed point are uniformly bounded, prove that a subsequence converges uniformly on compact subsets to a harmonic function.)
Combine compact-set Harnack, derivative estimates, a compact exhaustion and the mean-value converse.
Solution
Choose an increasing compact exhaustion containing the fixed reference point, with interiors covering the domain. For each compact neighborhood used in an estimate, adjoin that reference point before applying compact-set Harnack. The common pointwise bound then controls the sequence uniformly there. The first-derivative estimate on slightly larger interior balls gives local equicontinuity; a finite cover suffices for each compact set, with the short line segments used for comparison staying inside those balls. Arzelà–Ascoli on the exhaustion, followed by a diagonal subsequence, gives a continuous local uniform limit. On each fixed admissible sphere, uniform convergence passes the mean-value identity to that limit. The continuous mean-value converse proves it is harmonic. No global boundedness, boundary convergence or strictly positive limiting value is asserted; the limit may be identically zero.