Newton’s cosh iteration
Newton's rule is N(z) = z − f(z) / f′(z). Here f(z) = cosh z − 1 and f′(z) = sinh z. Each step subtracts the correction (cosh z − 1) / sinh z, which is tanh(z/2) away from the zeros of sinh z.
The equation cosh z = 1 has infinitely many solutions: z = 2πik for every integer k. Each is a double root, so nearby Newton errors are approximately halved on each step: convergence is linear, not quadratic. The denominator vanishes at every πik. At even multiples the quotient has a removable value because those points are roots; the genuine poles are the odd multiples (2k+1)πi.
Try z₀ = 0.2i. The correction is tanh(0.1i) = i tan(0.1) ≈ 0.1003i, so z₁ ≈ 0.0997i: the error is visibly about halved. Near an odd multiple such as πi, the correction can become very large. In the implementation, the real component is clamped for numerical stability, and the step is left unchanged when |sinh(z)|² < 1e-10; those are safeguards, not claims about the unsafeguarded map.
The function repeats with period 2πi in the imaginary direction, and the Newton map does too: N(z + 2πi) = N(z) + 2πi. That repeated rule is the source of the horizontal banding.


