One coefficient out of infinitely many
On the last rung you learned to expand a function around an isolated singularity as a Laurent series, sum a_n (z - z_0)^n, where the index n now runs over negative values too. That two-way series carries an infinite ledger of coefficients: a_0, a_1, a_2, ... reaching out into the analytic part, and a_(-1), a_(-2), ... reaching down into the principal part that records the blow-up. The residue of f at z_0 is, almost shockingly, just one entry from that ledger: the coefficient a_(-1), the one on the single power 1/(z - z_0).
It is fair to ask why this one coefficient earns its own name and a chapter of theory. Why a_(-1) and not a_(-2), or a_5? The reason is a single clean fact about integrating powers around a small circle centred at z_0, and it is worth feeling in your bones. Walk once counterclockwise around a tiny loop and integrate (z - z_0)^n. For every whole-number power except one, the answer is exactly zero — the function comes back to where it started and the contributions cancel perfectly. The lone exception is the power n = -1, where the integral is the magic number 2 pi i.
integral over |z - z_0| = r of (z - z_0)^n dz = 0 for every integer n != -1 integral over |z - z_0| = r of (z - z_0)^(-1) dz = 2 pi i (the one survivor) so, integrating the Laurent series term by term: integral over C of f(z) dz = 2 pi i * a_(-1) = 2 pi i * Res(f, z_0)
Now integrate the whole Laurent series around that loop, term by term. Every term collapses to zero and vanishes from the sum — every term but one. Only a_(-1) times 1/(z - z_0) survives, and it contributes a_(-1) times 2 pi i. So the entire loop integral, no matter how complicated f looks, reduces to 2 pi i times that single coefficient. The residue is precisely the part of f that you cannot integrate away. That is the whole reason it deserves a name: it is the residue as a Laurent coefficient, the one number the contour leaves behind.
A residue lives only where the function misbehaves
Before chasing recipes, get the geography right, because it saves real work. A residue is attached to an isolated singularity — a point where the function fails to be analytic but is perfectly well behaved on a small punctured disk around it. At a point where f is analytic, there is no principal part at all: the Laurent series is just an ordinary Taylor series, every negative coefficient is zero, and so a_(-1) = 0. The residue at a good point is simply zero. You only ever go hunting for residues at the bad points.
Even among the bad points, two of the three kinds of isolated singularity often hand you the answer for free. At a removable singularity there is no principal part either — the function was merely missing a value that you can fill in — so once more a_(-1) = 0 and the residue is zero. The interesting cases are a pole, where the principal part has finitely many terms and a_(-1) is the coefficient you must extract, and an essential singularity, where infinitely many negative terms swirl and you typically have to find a_(-1) by genuinely expanding the series. Most of everyday computation, though, lives at poles.
The simple-pole recipe: a one-line limit
Expanding the full Laurent series just to read off a_(-1) would be wasteful, so the subject builds shortcuts keyed to the order of the pole — and the most common case, a simple pole (order 1), is the easiest of all. Suppose z_0 is a simple pole, so f(z) = a_(-1)/(z - z_0) + (analytic stuff). Multiply through by (z - z_0): the troublesome term becomes just a_(-1), and the analytic stuff, now multiplied by (z - z_0), quietly dies as z approaches z_0. What survives the limit is exactly the residue. This is the simple-pole residue formula.
In symbols the rule reads Res(f, z_0) = lim as z -> z_0 of (z - z_0) f(z). Watch it work on f(z) = e^z / (z - 1), which has a simple pole at z = 1. Multiplying by (z - 1) cancels the denominator and leaves e^z; sending z to 1 gives e^1 = e. The whole computation is one cancellation and one substitution — no series in sight. That clean one-liner is why simple poles are the friendliest residues to meet.
There is a beautiful special case worth singling out, because it appears constantly. Suppose f = g/h, where g and h are both analytic at z_0, with g(z_0) not equal to 0, and h has a simple zero there (so h(z_0) = 0 but h'(z_0) is nonzero). Then the limit in the formula rearranges, using the definition of the derivative of h, into a clean quotient: the residue is just g(z_0) divided by h'(z_0). This is the quotient residue shortcut, and it lets you skip cancelling factors entirely — you differentiate the denominator once and divide.
Higher-order poles: differentiate, then take the limit
When the pole has order m greater than 1, multiplying by (z - z_0) is not enough — there are still negative powers left over. So multiply by the full (z - z_0)^m to clear every negative power at once. That turns f into an analytic function near z_0, whose Taylor series you now hold; but the coefficient you want, a_(-1), has been shifted up to become the coefficient of (z - z_0)^(m-1). To dig out that particular Taylor coefficient you differentiate m - 1 times, divide by the factorial (m - 1)!, and take the limit. This is the higher-order pole residue formula.
- Find the order m of the pole at z_0 — for f = g/h with g(z_0) nonzero, it is the order of the zero of h.
- Form the cleaned-up function phi(z) = (z - z_0)^m * f(z), which is analytic at z_0.
- Differentiate phi exactly m - 1 times.
- Divide by (m - 1)! and take the limit as z -> z_0; the result is Res(f, z_0).
Try it on f(z) = e^z / z^3, which has a pole of order m = 3 at the origin. The cleaned-up function is phi(z) = z^3 f(z) = e^z. We need m - 1 = 2 derivatives, and the second derivative of e^z is still e^z. Divide by (m - 1)! = 2! = 2 and let z go to 0: the residue is e^0 / 2 = 1/2. You can sanity-check it against the series e^z = 1 + z + z^2/2 + ..., so e^z/z^3 = 1/z^3 + 1/z^2 + (1/2)(1/z) + ..., and indeed the coefficient of 1/z is 1/2. The recipe and the raw series agree, as they must.
Choosing the right recipe — and when to ignore them all
Notice that the simple-pole formula is exactly the higher-order formula with m = 1: zero derivatives, divide by 0! = 1, take the limit. So you are really carrying one method with a dial for the order. The practical workflow is: locate the singularity, decide its order, then turn the dial. Get the order wrong and the recipe quietly betrays you — using m too large injects spurious extra derivatives, using m too small leaves uncancelled negative powers that send the limit to infinity. Always pin down the order first; it is the single most common source of error.
There is an honest catch, and it is worth saying plainly: these limit-and-derivative recipes only work at poles. At an essential singularity there is no finite order to plug in, the principal part runs forever, and no amount of multiplying by (z - z_0)^m will ever clear it. There the only general route back to a_(-1) is to expand the Laurent series by hand and read off the 1/(z - z_0) coefficient directly. For example, e^(1/z) at the origin expands as 1 + 1/z + 1/(2! z^2) + ..., so its residue is simply the coefficient of 1/z, namely 1 — found by inspection, not by any pole formula.
Even at poles, do not let the formulas crowd out a sharper eye. Often the fastest route is to recognise a known series and read a_(-1) off directly, exactly as we cross-checked e^z/z^3 above. The recipes are reliable machinery for when inspection is hard, not a law forbidding you to think. With residues in hand and a single small number computed at each singularity, you are perfectly poised for the next guide, where the residue theorem turns these one-point computations into a master formula for whole contour integrals — the grand generalisation of the Cauchy integral formula you already know.