# mpmath, ordinary measure for real k with abs(k+4)>16.
import mpmath as mp
def mahler_q_exterior(k, dps=120):
with mp.workdps(dps):
u = mp.mpf(k) + 4
if abs(u) <= 16:
raise ValueError("This series requires abs(k+4)>16")
r = 16 / abs(u)
a, b, power = 1, 4, 1/u
result = mp.log(abs(u)) - 4*power
n = 1
while r**(n+1)/((n+1)*(1-r)) > mp.eps:
n += 1
c = (2*(2*n-1)*(5*n*n-5*n+2)*b - 64*(n-1)**3*a)//n**3
a, b = b, c
power /= u
result -= c*power/n
return +result
print(mp.nstr(mahler_q_exterior(20), 100))The generator uses Bertin's eta quotient and divisor-sum expansion [1], Theorem 1.1(2), with explicit branch choices for the nome. It reconstructs the exact CM nomes where available and solves the eta equation afresh at 150 and 210 working decimal digits elsewhere. The endpoint $k=-4$ uses the factorization in (1) and Smyth's evaluation. Every entry agrees between the two precisions to at least 149.33 digits relative to the larger of one and the value magnitude; 100 significant digits are retained. The omitted divisor-sum tail is explicitly bounded, but the root solving and floating-point arithmetic are not certified, so the common rigour level is heuristic (agreement-checked).
An independent implementation starts with the Domb-moment series in its convergence region and continues its differential equation through the upper half-plane to each of the 23 nonsingular selected arguments. Its known-answer control continues from $u=32$ to $u=64$ and agrees with a direct exterior sum. At $k=0,12$, a separate real hypergeometric endpoint integral and the proved factor-four relation provide the check. These independent values agree with the generator to at least 139.32 absolute decimal digits. Arb separately evaluates the zeta and Dirichlet formulas at $k=-4,-3$, agreeing to more than 185 digits.
Exact exponent arithmetic verifies both torus substitutions; direct binomial sums verify the first 35 Domb coefficients. The exterior series is also summed directly with an explicit tail bound for every selected $|k+4|>16$. The endpoint derivative is checked against Watson's gamma product. A deliberately perturbed value is rejected. Generator verification and the same independent comparisons are applied to the stored decimal intervals. The attached checks, argument-selection note and numerical evidence record the computations and software versions.