Improper Integrals: Integrating to Infinity

When the ordinary rules run out

Every definite integral you have computed so far lived on a tidy finite interval [a, b], with a function that stayed finite the whole way across. The Fundamental Theorem then handed you the answer as F(b) - F(a). But two things can break that tidiness. The interval can stretch to infinity — integral from 1 to infinity of something. Or the function can blow up to infinity somewhere inside the interval — like 1/sqrt(x) right at x = 0. In either case the symbol F(b) - F(a) no longer literally makes sense, because there is no honest endpoint to plug in.

An integral with either of these features is called an [[improper-integral|improper integral]]. The word "improper" is not an insult — it just flags that the definition you learned does not apply directly. So we do the same move that rescued everything else in calculus: we sneak up on the problem with a [[limit-at-infinity|limit]].

The definition: integrate to a wall, then push the wall away

Here is the trick made precise. To integrate from 1 all the way to infinity, you do not magically reach infinity. Instead you integrate up to a movable wall at x = b — that is an ordinary, well-behaved definite integral you already know how to do. Then you ask what happens to that number as you slide the wall b out toward infinity. If the number settles down to a single finite value L, we say the improper integral converges to L. If it grows without bound or never settles, we say it diverges.

integral from a to infinity of f(x) dx  =  lim b->infinity  ( integral from a to b of f(x) dx )

  converges  ->  the limit is a finite number L
  diverges   ->  the limit is infinite, or does not exist

The unbounded-function case works exactly the same way, just from the other side. If f blows up at x = a, you start the wall a tiny bit past the trouble — integrate from a + t to b — and then let t -> 0 to creep up on the bad point. Same philosophy: never touch the dangerous spot directly; approach it through a limit and see whether the area stays finite.

The shock: 1/x^2 converges, 1/x diverges

Now for the genuinely surprising part. Picture the curves y = 1/x and y = 1/x^2 to the right of x = 1. Both drop toward zero forever; both trap an infinitely long sliver of area. Your gut probably says: an infinitely long region must hold infinite area. Your gut is half wrong. The two integrals behave completely differently — and the only way to know is to take the limit.

1. Convergent case: integral from 1 to b of 1/x^2 dx = [-1/x] from 1 to b = -1/b + 1 = 1 - 1/b. As b -> infinity, the term 1/b vanishes, so the limit is 1. The total area under 1/x^2 out to infinity is exactly 1 — finite! It converges.
2. Divergent case: integral from 1 to b of 1/x dx = [ln x] from 1 to b = ln(b) - ln(1) = ln(b). As b -> infinity, ln(b) keeps climbing forever — slowly, but without any ceiling. The limit is infinite, so the area under 1/x is unbounded. It diverges.

So two curves that look almost identical — both vanishing toward zero — split into opposite fates. The deciding factor is how *fast* the function shrinks. 1/x^2 dies off quickly enough that its tail areas add up to a finite total; 1/x crawls toward zero just a hair too slowly, and its endless tail accumulates without limit. The whole drama of convergence and divergence lives in that razor's edge.

A bridge to summing infinitely many things

Look again at what just happened. You added up an *infinite* amount of stuff — area stretching out forever — and asked whether the running total approaches a finite number. That is precisely the question waiting for you in the next track. Replace the smooth area under a curve with a list of discrete numbers added one after another, and you have an [[infinite-series|infinite series]]: a sum like 1 + 1/4 + 1/9 + 1/16 + ... that never ends.

The logic transfers exactly. A series converges when its partial sums (1, then 1 + 1/4, then 1 + 1/4 + 1/9, ...) approach a finite limit, and diverges otherwise — the very same converge/diverge language, the very same limit idea. There is even a theorem (the integral test) that uses an improper integral of 1/x^2 to prove the sum 1 + 1/4 + 1/9 + ... converges, and the divergence of integral 1/x to prove the famous harmonic sum 1 + 1/2 + 1/3 + ... diverges. The two stories are the same story.

import math

# integrate 1/x^2 up to a moving wall b, watch it settle toward 1
for b in [10, 100, 10000, 10**8]:
    print(b, 1 - 1/b)        # -> 0.9, 0.99, 0.9999, ~1.0   (converges to 1)

# integrate 1/x up to the same walls, watch it climb forever
for b in [10, 100, 10000, 10**8]:
    print(b, math.log(b))    # -> 2.3, 4.6, 9.2, 18.4 ...   (no ceiling -> diverges)