If we know how large a holomorphic function can be on two circles, how large can it be on a circle between them? The maximum-modulus principle gives the larger of the two boundary bounds, but it does not use the position of the intermediate circle. Hadamard's three-circle theorem gives a sharper multiplicative interpolation: the relevant coordinate is , not .
A related three-lines theorem works on a vertical strip. Its domain is unbounded, so the usual maximum principle on a compact rectangle needs an additional argument. Comparing the two proofs shows both the power of harmonic barriers and the importance of growth hypotheses at infinity.
The three-circle theorem
Let , and let be holomorphic on a neighborhood of the closed annulus . Write
Work on the connected component of that neighborhood containing the annulus. If is not identically zero there, on every circle: otherwise the identity theorem would force on that component.
Theorem 1.6.1
(Hadamard's three-circle theorem)
Assume is not identically zero on that component. For put
Then
Equivalently, is convex on . If equality holds at any one intermediate radius, then on the annulus for a nonzero constant and an integer . Conversely these functions give equality at every intermediate radius.
Proof
(A harmonic barrier on an annulus)
Let be chosen so that the harmonic function equals on , . We claim
throughout the annulus, interpreting at zeros.
For , consider
On the boundary its value is at most . It is upper semicontinuous on the compact closed annulus, including at zeros, so it attains a maximum. An interior point where the maximum exceeds the boundary bound cannot be a zero of . Near that point is harmonic, because has a local holomorphic logarithm. Hence there, contradicting the second-derivative test at an interior maximum. Thus . Let .
Taking the maximum on and substituting the affine function of gives the displayed inequality. This argument permits zeros of ; it does not take a global logarithm of or of .
Proof
(When equality forces a monomial)
Suppose equality holds at and choose on that circle with . In a small disk about , choose a branch of . The holomorphic function has modulus at most there and attains that bound at . By the maximum-modulus principle it is locally constant. Differentiating gives locally, and the identity theorem extends this equation to the connected annulus.
A nonzero solution of this first-order equation cannot have a zero in the annulus: local uniqueness for the analytic equation would otherwise force it to vanish identically. Thus globally. Integrating around a circle gives
the winding number about zero of the image of the positively oriented source circle. Write ; then , so . Conversely , making affine and giving equality.
Proof X-ray
The correct variable is logarithmic radius
The boundary interpolation is linear in because is harmonic away from zero. Exponentiating converts this linear bound into a geometric mean. Equality at one interior radius forces the local maximum principle to become rigid, and the winding number then forces the exponent to be an integer.
Using the circle inequality
Worked example 1.6.2
(A geometric-mean bound)
Suppose is holomorphic on a neighborhood of , with and . At , the logarithmic position is , so
This is stronger than the maximum principle's immediate bound . The bound is sharp for : it has , and .
Worked example 1.6.3
(Strict convexity without zeros)
For , and , which is strictly convex. Thus all nontrivial three-circle inequalities are strict, even though has no zeros. By contrast, is holomorphic on every annulus avoiding zero and attains equality; negative exponents are allowed because the domain need not contain the origin.
Common mistake
(Interpolating in instead of )
The exponent is , not . For on , and . The correct bound at is . The linear exponent would give , which is smaller than the true value , so it is not a valid bound.
Counterexample mode
A bound on maximum modulus is not a pointwise lower bound
For on an annulus containing , the circle maximum is , but . The three-circle theorem controls the largest modulus on each circle. It does not claim that every point has modulus comparable to or that has no zeros.
For a nonzero entire function, the same theorem applies on any two positive radii. It therefore adds a convexity constraint to its maximum-modulus growth. The separate growth-order chapter studies the asymptotic ratio ; convexity alone does not determine that order or prove a factorization theorem.
Three lines on an unbounded strip
Theorem 1.6.4
(Hadamard's three-lines theorem, bounded form)
Let be bounded and continuous on the closed strip , and holomorphic in its interior. Set
For every and ,
when . If either boundary supremum is zero, then in the strip.
Proof
(Damp the ends of a large rectangle)
First assume and put . For define
On both vertical sides, : balances the two boundary bounds and . Since and are bounded on the strip, on the horizontal sides the factor tends uniformly to zero as . For large enough , those sides also satisfy the same bound. The maximum-modulus principle on the rectangle yields inside it.
For fixed , take a sufficiently large rectangle and then let . This gives .
If a boundary supremum is zero, replace in the argument by , use those as upper bounds and let . Every interior value must then vanish; continuity gives the same on the boundary.
Counterexample mode
Boundary bounds alone do not control an unbounded strip
The entire function satisfies for every real , because the sine is purely imaginary on both boundary lines. But , and is unbounded. Omitting the boundedness hypothesis would make the three-lines conclusion false.
The map sends a vertical strip to an annular region, with a periodic identification in the imaginary direction. In fact, apply the three-lines theorem after an affine rescaling to on : it is bounded by continuity and -periodicity, and its supremum on each vertical line is the corresponding circle maximum. This recovers the three-circle inequality. The strip theorem itself also covers nonperiodic functions.
Worked example 1.6.5
(A Gaussian on the strip)
Let on . Since , is bounded, and , . The theorem gives . The true supremum on the line is , attained at , and on . So the bound holds, with equality only on the two boundary lines: is convex in and lies below the chord , which is exactly the three-lines inequality.
Exercises
Quick check
(If and , bound for a holomorphic function on a neighborhood of .)
Compute first.
Solution · Answer
Here , so .
Quick check
(For on , verify equality in the three-circle theorem at .)
Compute exactly.
Solution · Answer
. Hence , , and .
Quick check
(Let satisfy the bounded strip theorem with and . Bound uniformly in .)
Here .
Solution · Answer
The theorem gives for every . If either boundary supremum is zero, the function is zero and the bound remains valid.
Quick check
(Suppose a nonzero entire function satisfies equality in a three-circle inequality at an intermediate radius. Show it must be a nonnegative integer power of times a constant.)
Use the equality case, then the behavior at .
Solution · Answer
Equality forces on the annulus with . The identity theorem extends this equality to . Since is entire, a negative would give a pole at zero, impossible. Thus .