Building a 5-Ton Calculator From 19th-Century Plans
alphadogg writes "Starting in May, many will have the opportunity to see computing done the old-fashioned way: with lots of gears, a big crank, and some muscle. The Computer History Museum in Mountain View, CA, will unveil a new construction, the first in the US, of the 19th-century British mathematician Charles Babbage's Difference Engine No. 2, an improved version of his earlier design for a mechanical digital calculator. It weighs in at two tons more than the Difference Engine built in 1991 at London's Science Museum. Microsoft millionaire Nathan Myhrvold commissioned and paid for the US model."
As I recall from some of the mechanical calulators I used several decades ago, division was performed by repeated subtraction. I don't remember trying to divide by zero, but my guess is that you'd have to keep turning the crank forever . . .
It's far more specialized than that. It basically computes values of a polynomial from a starting set (interpolate / extrapolate). It doesn't have an explicit fexible divide operation. Exactly what sort of error you get out is going to depend on how you carry out the division, but most likely you would do exp(log(a) - log(b)), which would produce a very large negative number for log(b) (an incorrect result, obviously), and a very large number for the result. It might or might not overflow, depending on the precision of your approximating polynomials for log and exp at the values of interest.