a natural boundary
Analytic continuation usually lets you push a function out past the rim of its first disk. But sometimes you hit a wall you can never get through: every point on the boundary circle of convergence is a singularity, packed so densely that no continuation across the circle is possible at any point. That wall is called a natural boundary. The function lives inside the disk and absolutely nowhere outside it — the boundary is not a temporary obstacle to route around, it is the edge of the function's entire universe.
Here is why such walls form. A power series can fail to continue past a single point (a pole or branch point on the boundary), and normally you just step around that one bad point through a nearby good arc. But if the singularities are DENSE on the circle — every arc of the boundary, no matter how short, contains one — then there is no good arc to step through, and continuation is blocked everywhere along the circle simultaneously. The classic mechanism for manufacturing this is a lacunary (gap-filled) series, where the surviving exponents grow fast enough that, by a clever argument, singularities appear at a dense set of boundary points. A standard example is sum z^(2^n) = z + z^2 + z^4 + z^8 + ..., which is holomorphic on |z| < 1 but has the unit circle as a natural boundary.
Why this matters: it is the honest counterweight to the optimism of continuation. Not every holomorphic function reaches out to a bigger world; some are sealed inside their disk forever. This also sharpens what 'the complete analytic function' can be — for a function with a natural boundary, the complete analytic function is just the function on its one disk, with nothing beyond. And it has real consequences in number theory and physics: certain generating functions and modular-type series have natural boundaries, which tells you something deep about the arithmetic or dynamics they encode rather than being a mere technical annoyance.
The series f(z) = sum_{n>=0} z^(2^n) = z + z^2 + z^4 + z^8 + z^16 + ... converges for |z| < 1. As z approaches any root of unity of the form e^(2 pi i p / 2^k), the terms blow up, and such roots are dense on the unit circle. So singularities crowd the entire circle: it is a natural boundary, and f cannot be continued to even a single point outside |z| < 1.
Dense singularities on the boundary circle seal the function inside its disk — no continuation is possible anywhere across it.
A natural boundary is not a single bad point you can detour around; it is a dense wall of singularities blocking continuation everywhere along the curve at once. Having a natural boundary is the rule for 'random-looking' series, not a rare pathology.