WP43S linked to computer?

[Paul Dale]

(04-23-2014 09:37 AM)RMollov Wrote:  I have only one problem with the WP34s - the solver, SLV command is not as good as the HP one we know.

In what way is the 34S solver worse than HP's? The 34S solver is in some ways more sophisticated than HP's solver. It is different however. In particular, it won't give as tight error bounds most of the time, this is due to the methods being used.

HP's solver is essentially a slightly modified secant method. The 34S incorporates several methods. The primary one, Brent's method, chooses between bisection, secant and an inverse quadratic interpolation depending on the function evaluations -- in theory it will converge faster than a secant based method. The 34S will also attempt Ridder's steps after bisections which seemed to improve convergence when I did my testing.

Of course, as Walter wrote, you are free to implement your own solver -- it is a keystroke program whose source is in trunk/xrom/solve.wp34s. If you do implement a better solver, feel free to submit it -- we'll definitely consider its inclusion. Writing a solver is not a simple task however -- the 34S solver took several iterations and rewrites to get it to its current state. Still, this is an interesting area of numerical mathematics.

Personally, I'd like to eventually improve the integration code to incorporate a Gauss-Kronrod quadrature as the sub-step in the Romberg method instead of the naïve method used.

- Pauli