the Phragmén-Lindelöf principle
/ FRAG-men LIN-deh-lurf /
The maximum modulus principle is a workhorse: if a holomorphic function is bounded on the boundary of a region, it is bounded by the same amount inside. But it has a loophole — it needs the region to be bounded (and the function to extend continuously to the boundary). On an unbounded region, like a strip or a sector running off to infinity, the function could in principle blow up far away even while staying small on the visible boundary. The Phragmén-Lindelöf principle is a family of theorems that plug this loophole: with a modest extra growth assumption, the maximum principle is rescued on unbounded regions too.
The shape of the argument is always the same. Take an unbounded region (say an infinite strip or a sector), suppose f is holomorphic there, bounded on the boundary, and — crucially — grows no faster than some controlled rate as you head to infinity inside the region (for a sector of angle pi/alpha the safe rate is roughly slower than exp(|z|^beta) for some beta < alpha). Then you conclude f is bounded by its boundary bound throughout the region. The trick that makes it work: multiply f by a tiny auxiliary 'damping' factor like exp(-epsilon z^beta) that decays at infinity in the region but is nearly 1 on the boundary, apply the ordinary maximum principle to the now-bounded product on a large truncated region, and let epsilon -> 0. The growth hypothesis is exactly what lets the damping factor win the race at infinity.
Phragmén-Lindelöf is indispensable in analytic number theory (bounding the Riemann zeta function and Dirichlet series in strips), in the theory of the Fourier and Laplace transforms, in interpolation theory, and anywhere you must control a holomorphic function on an unbounded domain. The three-lines and three-circles theorems are its most famous concrete incarnations. An honest and important caveat: the growth restriction is not optional decoration — it is essential. Without it the conclusion is false: the function exp(exp(z)) is bounded (in fact equals 1 in modulus) on the boundary lines of the strip |Im z| <= pi/2 but is wildly unbounded inside. Phragmén-Lindelöf only works because such monstrous growth is forbidden by hypothesis.
On the strip 0 <= Re z <= 1, suppose f is holomorphic, bounded by M on both edge lines, and grows no faster than exp(|z|^beta) for some beta < 1 inside. Phragmen-Lindelof concludes |f(z)| <= M throughout the strip. The counterexample exp(exp(pi z)) shows why the growth cap matters: it is bounded on the edges of a suitable strip but explodes inside, because its growth violates the hypothesis.
A controlled-growth hypothesis rescues the maximum principle on an unbounded strip.
The growth hypothesis cannot be dropped, and it is tied to the geometry: a wider region (larger opening angle) tolerates only slower growth. Quote both the region and its allowed growth rate together — one without the other is meaningless.