A positive harmonic function cannot change arbitrarily fast inside a disk. Knowing its value at the center controls it above and below throughout every smaller disk. This is Harnack's inequality. The constant is sharp, and the equality cases are explicit positive harmonic functions.
The estimate also connects local information to global behavior. A chain of overlapping disks compares values on any compact subset of a connected domain. Applied to differences of an increasing sequence of harmonic functions, that comparison yields a useful convergence theorem without assuming derivative bounds in advance.
A sharp disk inequality
Theorem 3.11.1
(Harnack inequality and equality)
Let be real harmonic on . For , ,
If , then . If and equality holds in either bound at a nonzero , then for some ,
For the upper bound ; for the lower bound . These functions attain the respective equalities.
Proof
(Map a positive half-plane to the disk)
The strong minimum principle gives if ; otherwise throughout the connected disk. Since the disk is simply connected, has a harmonic conjugate. Choose a holomorphic with and subtract an imaginary constant so that .
Define
Because , and . Schwarz's lemma gives . Solving for yields
The bounds give
and monotonicity in gives the stated inequality.
If equality holds at nonzero , equality must hold in Schwarz's lemma, so with . Equality in the upper Harnack bound requires ; for the lower bound it requires . The resulting functions attain those bounds by direct substitution.
Proof X-ray
Why positivity is structural
The right half-plane is mapped into the unit disk by a fractional-linear transformation. Positivity ensures the transformed function stays inside the disk, so Schwarz's lemma controls its size. Equality requires both a rotation in Schwarz's lemma and the correct phase in the elementary modulus inequality.
For a disk , apply the theorem to . For ,
In particular, on , values lie between and .
Harnack chains and convergence
Theorem 3.11.2
(Comparison on a connected domain)
Let be connected and open, , and compact. There is such that every nonnegative harmonic on satisfies
The constant depends on the domain, and , not on . In particular, applying both inequalities to shows that pointwise boundedness of a family at is equivalent to boundedness at .
Proof
(Connect a compact set by finitely many disk steps)
Any two points in an open connected subset of the plane can be joined by a path in it. The path image is compact, so there is such that for every point of the path. Uniform continuity of permits a finite subdivision so consecutive points are less than apart; each is then in the preceding point's half-radius disk. Each application of (1) changes the comparison factor by at most three. A finite chain therefore compares the values at the endpoints in both directions.
For each choose with . Compactness gives a finite cover of by the half-radius disks . Connect their centers to by such finite chains and take the largest resulting factor, including one final factor of three. This supplies . If , the strong minimum principle gives , so the inequality remains valid.
Theorem 3.11.3
(Harnack's convergence theorem)
Let be a pointwise increasing sequence of nonnegative harmonic functions on a connected domain . Exactly one of the following occurs:
- converges locally uniformly to a finite harmonic function on .
- uniformly on every compact subset of .
The first case occurs exactly when is bounded at one, and hence every, point .
Proof
(Apply the chain to differences)
Fix . If is bounded, monotonicity makes it a convergent real sequence. For , the difference is nonnegative and harmonic. Harnack's chain gives
Thus is uniformly Cauchy on every compact and converges locally uniformly to a finite continuous function. The mean-value property passes to the limit on each closed disk contained in ; the mean-value characterization of harmonic functions then makes the limit harmonic.
If is unbounded, monotonicity gives . The lower Harnack-chain estimate gives for every compact . These cases are mutually exclusive and exhaustive.
Proof X-ray
Why monotonicity matters twice
It makes the values at one point either converge or tend to infinity. More subtly, it makes nonnegative, so Harnack's inequality applies to the differences and upgrades pointwise control to uniform Cauchy control on compact sets.
Applications and examples
Worked example 3.11.4
(A nonnegative entire harmonic function is constant)
Let be harmonic on all of . For fixed , apply (1) on and let . Both factors tend to one, forcing . Thus is constant. More generally, a harmonic function on that is bounded below is constant: subtract a lower bound and apply the nonnegative result. The same follows for a function bounded above by applying it to its negative.
Worked example 3.11.5
(Sharpness and a holomorphic derivative estimate)
The positive harmonic function
has . At it equals ; at it equals . Both constants in Harnack's inequality are therefore sharp.
If is holomorphic on with , the same half-plane normalization gives
Indeed compose with the disk automorphism and subtract . The associated map , , fixes zero. Writing gives . Schwarz's derivative estimate, and give the formula. The function gives equality at .
Worked example 3.11.6
(A Harnack chain along a strip)
Suppose is harmonic on the strip and . No disk in centered at reaches the point , since such disks have radius at most . Instead use the disks of radius one centered at ; each lies in because its points have . Each next center, and finally the point , lies at distance from the previous center, so by (1) each step changes the value by a factor between and . After four steps,
The constant depends only on the chain, not on , which is exactly the content of the comparison theorem.
The convergence theorem uses monotonicity in an essential way, as the next sequence shows.
Counterexample mode
Boundedness at one point is not enough without monotonicity
On , let . Every is positive and harmonic, and for all . Yet at any point with nonzero real part the values alternate, so the sequence does not converge. The Harnack convergence theorem needs monotonicity; boundedness at one point alone gives neither its conclusion nor a preferred limit.
Common mistake
(Applying Harnack's inequality to a signed function)
The inequality also needs nonnegativity. The harmonic functions all vanish at zero but are nonzero elsewhere, so no analogous Harnack-type bound can hold for arbitrary signed harmonic functions. For a function bounded below by on the disk, apply the inequality to , not to .
Exercises
Quick check
(If is harmonic on and , bound using Harnack's inequality.)
Use (1) with and .
Solution · Answer
With and , the factors are and . Thus .
Quick check
(Verify directly that attains both disk bounds at and .)
Write the real part as .
Solution · Answer
The displayed expression is . At it is ; at it is . Since , these are exactly the upper and lower bounds.
Quick check
(Suppose are harmonic and increasing on a connected domain, and for every . What type of convergence follows on compact subsets? Why are differences relevant?)
Decide which alternative of the convergence theorem occurs.
Solution · Answer
The values at converge because they are increasing and bounded. The Harnack convergence theorem gives locally uniform convergence to a finite harmonic function. For , the difference is nonnegative harmonic, allowing the chain inequality to prove uniform Cauchy estimates.
Quick check
(If is holomorphic on , , and , bound and give an example attaining the bound.)
Use the derivative estimate at .
Solution · Answer
The derivative estimate gives . Equality holds for : here and , so .