(HP15C)(HP67)(HP41C) Bernoulli Polynomials

[John Keith]

For B(16) using my program listed above in approximate mode (all floating-point numbers) I get exactly 7.1. The largest term in the 15th row of A163626 is ~10^14 so some rounding is inevitable, and would certainly be worse for 10-digit calculators.

The largest row with all numbers < 10^10 is row 11 which should allow accurate computation of B(12). The value of B(12) to 12 digits is -0.253113553114, while the program returns -0.25312, which has only four correct digits. The exact value of B(12) is -691/2730 so we should expect more accurate results.

Your idea of scaling the terms by their LCM is interesting but I can't see how one could compute the LCM on calculators limited to 10-digit floats.