Chapter IX · The logarithmic, exponential and circular functions of a real variable
The series for the inverse tangent
The same integration trick gives for , hence ; and small arguments turn it into a practical way of computing .
Why this matters
In §163 the inverse tangent was identified with an integral, , taking the value between and . The integrand is again a geometric series, now in , so the method of §220 applies almost word for word. Hardy leaves the details to the reader in §221; this lesson supplies them. The payoff is the most famous series in mathematics, , and with it the first serious method of calculating .
The idea
For every real and every ,
as you can check by multiplying by . This time the denominator never vanishes and is never less than , which makes the remainder even easier to control than for the logarithm.
Try this. Set and increase ; then try and ; finally move beyond .
- the first terms of the geometric series
- area between them over = error
What to notice
The picture is symmetric: is even, so the area from to is minus the area from to , and is odd. Unlike the logarithm, both endpoints and work. Beyond the polynomial runs off to and the error grows with .
Precise statements
If then
where is the value between and . For the series diverges. In particular (Leibniz),
Proofs
Integrate the finite identity from to and bound the remainder by integrating . By oddness it is enough to treat .
The exact identity
Integrating from to (valid for every real , since ):
The bound for
Since , we get , which tends to .
Negative
Both sides of the identity are odd functions of : the integral of an even function from to is minus the integral from to , and only odd powers occur on the right. So the case follows from the case . In particular the endpoint is covered, which was not so for the logarithm.
Divergence for
The terms tend to when , so the series cannot converge.
Which value of the arctangent?
The integral is a continuous odd function that increases from and stays between and (see §163), so the series gives the principal value. At this is .
Leibniz’s series is beautiful but, as a method of computing , hopeless: the bound for the error of means two million terms for six decimals. The cure is to use the series only at small , where the error falls like . Identities such as Euler’s and Machin’s (proved from the addition formula of §225) do exactly that.
Try this. Step from to and read off how many decimals each formula gives. How many terms does Machin's formula need for six decimals?
What to notice
On the log scale the Euler and Machin errors fall along straight lines: each term gains a fixed number of digits (about for Euler, since , and for Machin, since ). Leibniz's error barely moves: after terms it still has only one correct decimal. Machin passes six correct decimals at .
Worked examples and exercises
Show that for ,
Why does this series, unlike the one for , fail at ?
Hint
Hint
Solution
For both and hold. Subtracting, the even powers cancel and the odd ones double: . At the series becomes , which diverges like the harmonic series; correspondingly there. The arctan series has the same terms with alternating signs, and alternation is what rescues it at .
How many terms of does the bound require before is guaranteed to be within of ?
Hint
Solution
means , that is, . The bound is essentially sharp: the true error after terms is close to , so about a million terms are genuinely needed. Machin’s formula needs five.
Assuming (proved in the lesson on §225), show how to compute to six decimals, and say how many terms of each series are enough.
Hint
Solution
. Using terms of the series for , the error in is at most : for this is , not quite enough; for it is . For two terms give an error at most (even one term gives about ). With five and two terms:
which is with a total error below . So to six decimals. The widget above reproduces these numbers.
Show that for , and deduce that
Hint
Hint
Solution
The series. . Integrating, the remainder is at most for .
The value. At the left side is . Partial fractions give an antiderivative (check by differentiating)
At everything vanishes (the two arctangents are ). At , , so the logarithm is ; and because the two arguments are reciprocals of each other (both positive). Hence .
Check yourself
What is the smallest for which the bound on the error of is less than ?
means , so (at the bound is ). Two hundred terms for two decimals.
Putting in the series gives .
False. For the terms grow without bound, so the series diverges, even though exists. To compute use and the series at .
Why does the arctan series converge at , while the logarithmic series diverges there?
The log integrand blows up at , and the series at is minus the harmonic series. The arctan integrand is at most everywhere, so the bound holds on both sides, which is the oddness of seen in the series. At the series is . It converges only conditionally there (the absolute values give , which diverges), and the fact that is defined everywhere is no help: the series still diverges for .
Hardy’s language
The value of arc tan x which is represented by the series is of course that which lies between −½π and ½π.
Summary
- for (principal value); at , .
- Proof: integrate the finite geometric sum in ; the remainder satisfies .
- Both endpoints work because and the function is odd; for the series diverges.
- The error falls like , so the series should be used at small : Machin’s gives six decimals with five terms.