Chapter X · The general theory of the logarithmic, exponential and circular functions
The logarithmic series
log(1 + z) = z − z²/2 + z³/3 − … for |z| ≤ 1 except z = −1, with the principal value; on the unit circle it gives the cosine and sine series of a logarithm and a saw-tooth.
Why this matters
For real with , Chapter IX proved (§220). The series converges for complex with too, but its sum must be some logarithm of , and there are infinitely many. §244 proves that it is always the principal value, and that the formula even survives on the circle , except at . §245 then splits it into real and imaginary parts and finds, on the circle, two famous trigonometric series: one sums to a logarithm, the other to a saw-tooth that jumps.
The idea
Two things have to be explained. First, why the principal value? Because for the point lies in the disc of radius about , which sits in the right half-plane, so the straight segment from to never comes near or crosses the negative real axis. Integrating along that segment gives the principal logarithm (§231), and integrating the geometric series for term by term gives the power series.
Second, what happens on the circle ? The terms no longer shrink geometrically, only like , so convergence is slow and depends on cancellation between terms that point in different directions.
Try this. Add terms with inside the unit circle and watch the polygon close in on the hollow dot, the principal value . Then switch on “keep on the circle” and drag round it, towards , with many terms (try and more).
- z (drag it)
- partial sums sₖ, one side per term
- principal value log(1 + z)
What to notice
Inside the circle the polygon converges quickly. On the circle the sides shrink only like , so the polygon creeps round a slowly tightening spiral; it still converges, except at , where the sum is and diverges. Near the target itself runs off to infinity.
Precise statements
For all complex with , except ,
where is the principal value. The series converges absolutely for and conditionally for , . Replacing by : for , .
Write . Then , and when its real part is positive, so its principal amplitude is with the inverse tangent between and . Equating real and imaginary parts:
For and all (and for , not an odd multiple of ),
For ,
Both sums are periodic with period . So the cosine series equals for all except odd multiples of (where it diverges), and the sine series equals for and at odd multiples of : a discontinuous saw-tooth.
Try this. Increase the number of terms from to and watch the blue partial sum approach the dashed saw-tooth. Put close to and compare partial sum and limit. Then switch to the real part (cosines).
- partial sum with N = 5 terms
- the sum of the series
- your θ
What to notice
Every partial sum is a continuous function of , yet the limit jumps by at each odd multiple of , where . Near the jump the partial sums converge slowly and overshoot (the Gibbs phenomenon); exactly at every term is , so the sum is , the midpoint of the jump. The cosine series dives to at , matching .
Proofs
Since the real case is known (§220), take not real. Integrate along the straight segment from to , which gives the principal value. On the segment, expand as a finite geometric sum plus an exact remainder, integrate, and show the remainder tends to . Working with a finite sum and an exact remainder avoids any question about integrating an infinite series term by term.
Parametrise the segment
Write with and . The segment is , , with . It does not pass through (that would need , i.e. real and negative), and by §231 it gives the principal value: .
Finite geometric sum
For any , , an algebraic identity (sum of a geometric progression).
Integrate
Integrating from to and using : , with .
Bound the remainder
Let
, so the partial sums converge to . This works for too; only is excluded (it is real, and there the segment ends at ). Absolute convergence for follows by comparison with ; for , is the harmonic series, so the convergence is only conditional.
Put and simplify with half-angle formulae; then read off real and imaginary parts.
Half angles
.
Principal logarithm
For , and , so .
Equate parts
The real and imaginary parts of with are the cosine and sine series. Each term has period in , so the sums do too; at every vanishes, so the sine series sums to there.
Finally, writing and for and subtracting gives for ; for real this is Gregory’s series for (§221), now explained by §241.
Worked examples and exercises
Put in the logarithmic series and equate real and imaginary parts.
Hint
Solution
. The theorem applies (, ) and gives . Real parts: , which is half the real series at . Imaginary parts: , Leibniz’s series. (Splitting a convergent complex series into real and imaginary parts is always allowed.)
In a triangle with sides and opposite the angle , prove .
Hint
Solution
Since , we have , so . As , the theorem (with ) gives , and . Hence .
Let and be real. Prove that
Hint
Solution
The corollary with angle gives , and , so the right side is . (For negative the corollary still holds: replace by and by .)
On the left, cycles with period in : it equals , , , for , and so on. With the signs the terms are , , , , …, as stated. In other words, the series is the real part of .
Using periodicity, find the sum of at , and .
Hint
Solution
lies in : sum . lies in , so : sum . is an odd multiple of : every term is , so the sum is . Note the jump between and : the sum goes from about down to as passes .
Check yourself
For , the sum of is
The series is a continuous function of on the disc and equals the principal value; the proof integrates along the straight segment from to , which stays in the right half-plane, and that path gives in §231. Its imaginary part is the amplitude of , which lies between and .
The logarithmic series converges at every point of the circle except .
True (Hardy proves slightly more than the real case: the remainder bound needs only ). At the series is , which diverges to .
What is the sum of at ?
Every term is , so the sum is , although the limits from the two sides are and . The sum function is discontinuous there.
In (for ), which value of the inverse tangent is meant?
It is the principal amplitude of , and since , that amplitude lies strictly between and . The series has a single sum, so “any value” is wrong.
Hardy’s language
The function is discontinuous for θ = (2k + 1)π.
Summary
- for , , with the principal logarithm.
- Proof: integrate along the segment from to , using a finite geometric sum with an exact remainder bounded by .
- Real and imaginary parts give and the matching arctan formula.
- On : the sine series sums to the saw-tooth on , discontinuous at odd multiples of , where .