Extending a Function: Analytic Continuation

A function that knows its own future

Here is a small miracle you have already half-met. In the real world, a function defined on a little interval can do absolutely anything once you leave that interval — you can extend a smooth real function past its edge in infinitely many genuinely different ways, all still smooth. The complex world is nothing like this. A holomorphic function defined on even a tiny disk, or along a tiny arc, has almost no freedom about how to continue: there is at most one holomorphic extension to any larger connected region. The function, in a real sense, already knows its own future. This phenomenon is called analytic continuation.

Make it concrete. The geometric series f(z) = sum z^n = 1 + z + z^2 + z^3 + ... converges only on the unit disk |z| < 1 — outside it, the terms blow up and the sum is meaningless. So as a series, f lives only on that disk. But you know what it adds up to inside: 1 / (1 - z). And that formula, g(z) = 1 / (1 - z), is perfectly holomorphic everywhere except the single point z = 1. The function g agrees with f on the whole disk where f makes sense, yet g is defined on almost the entire plane. We say g is the analytic continuation of f: the same function, finally allowed to occupy all the room it was always entitled to.

Why the extension is unique

Why can there be only one extension? The engine is the identity theorem, the deepest rigidity result you carried up from the power-series rung. It says: if two holomorphic functions on a connected region agree on a set that has a limit point inside that region — even just on a tiny arc, or on a sequence of points piling up somewhere — then they agree everywhere on the whole region. A holomorphic function cannot quietly match another one on a small patch and then peel away; matching on any patch with substance locks them together forever.

Now the uniqueness follows in one line. Suppose g_1 and g_2 are both holomorphic continuations of f onto the same connected region. Both equal f on the original little patch — so g_1 and g_2 agree there, on a whole disk, which certainly has limit points. By the identity theorem they must agree on the entire connected region. There is no room for two different answers. This is exactly the uniqueness of analytic continuation: extending a holomorphic function to a given connected region can be done in at most one way.

Chaining disks: how continuation actually moves

Uniqueness tells us the extension is forced, but not how to find it. The concrete machine is the power-series chain. Recall from the power-series rung that a holomorphic function equals its Taylor series sum a_n (z - z_0)^n on a disk, and that the disk grows until the radius of convergence reaches the nearest singularity. So pick a fresh center z_1 inside your first disk but near its rim, recompute the Taylor coefficients there, and you get a new disk around z_1. If the function is healthy near z_1, this second disk can poke out beyond the first — and on the overlap the two series agree automatically, because both equal the same function there.

1. Start with a power series sum a_n (z - z_0)^n valid on its disk of convergence D_0, the function's first home.
2. Pick a new center z_1 inside D_0, close to the boundary but in a healthy direction (away from any singularity).
3. Re-expand: compute the Taylor coefficients of the SAME function at z_1, giving a new series on a new disk D_1.
4. If D_1 pokes outside D_0, you have honestly extended the function — and on the overlap D_0 ∩ D_1 the two series agree, by the identity theorem, so the extension is unambiguous.
5. Repeat, walking center by center along a path, stitching disk onto disk until you have covered as much of the plane as the function will allow.

Run this on our geometric series, centered at z_0 = 0 with its disk |z| < 1. Re-center at z_1 = -1/2: the new Taylor disk of 1/(1-z) reaches all the way to the singularity at z = 1, so its radius is the distance |1 - (-1/2)| = 3/2, and it spills well outside the original |z| < 1. Each re-expansion only ever stops at the genuine singularity z = 1, never at the artificial circle |z| = 1 that the first series happened to die on. Stepping around in this way, the chain eventually reconstructs g(z) = 1/(1-z) on the whole plane minus the point 1 — exactly the continuation uniqueness already promised must exist if anything did.

What carries across — and a warning

A beautiful bonus rides along with continuation: identities survive it. This is the permanence of functional relations. If a holomorphic function satisfies some equation on its first patch — say the exponential obeys e^(z+w) = e^z e^w, or a function squares to give z — that relation, being itself an equation between holomorphic functions, holds wherever you continue, by the same identity-theorem reasoning. This is precisely how the real exponential, sine, and the gamma function are extended to the plane while keeping the algebra you trust. The continuation is not just a function out there; it is the honest extension, the one that respects everything the original stood for.

  D_0 (around z_0=0)        D_1 (around z_1)        D_2 ...
  series in z          ->   series in (z - z_1) ->  ...
  |z| < 1                   reaches farther out      keeps going
     \______ agree on overlap (identity theorem) ______/
  the union of all reachable disks = the complete analytic function

Now the warning, and it is the heart of this whole rung. Uniqueness was guaranteed for a fixed connected region. But continuation moves along a path, and different paths to the same destination can wind around a singularity in different ways. When that happens, the disk-chain can arrive at the same final point carrying a different value — the answer depends on the road, not just the endpoint. The classic offenders are the square root and the logarithm, whose values are genuinely multivalued: continue the square root once around the origin and it comes back as its own negative. The collection of all these path-dependent pieces, stitched into one consistent object, is the complete analytic function — single-valued only after we are honest about which road we took.

Where this rung is heading

You now hold the central idea: a holomorphic function is rigid, so it extends in at most one way to a connected region, and the disk-chain is the concrete tool that performs the extension. The remaining four guides chase the consequences. The very next one makes the path-dependence precise as continuation along a path, and pins down exactly when two paths give the same answer — the monodromy theorem. Then we ask when continuation cannot proceed at all, hitting a natural boundary of singularities packed so densely that no window opens past them.

And then the grand resolution. The reason a loop can change the value of a square root or a logarithm is that we have been forcing a multivalued object to live on the flat plane, where it does not fit. The cure, due to Riemann, is to change the stage: build a new curved domain, a Riemann surface, that has just enough sheets for the function to become honestly single-valued — going once around the origin then lands you on a different sheet, not back where you started. The last two guides build these surfaces explicitly for the square root and the logarithm, turning every paradox of this guide into clean geometry. For now, hold the one idea that powers all of it: locally, a holomorphic function has no freedom; the only freedom is in the global shape of the road.