| ??? 01/10/08 17:59 Read: times |
#149249 - Convergence Responding to: ???'s previous message |
I haven't tested for all possible 4290000000+ inputs, but the only convergence issues are caused by the rounding errors introduced by integer arithmetic. This is actually more of a problem with very small numbers, where the lost fractional part is a bigger percentage of the integer values. As an example, an input of 15 produces an oscillating iteration. By forcing a termination after a fixed number of iterations, a result accurate to +- 1 bit will be produced in a predictable time.
The ever present problem of rounding in integer arithmetic means you must have guard digits when working with real world signals. |
| Topic | Author | Date |
| Integer square root algorithm performance | 01/01/70 00:00 | |
| How many cycles per iteration ? | 01/01/70 00:00 | |
| The edge of chaos | 01/01/70 00:00 | |
| Yes, but he can simulate the performance... | 01/01/70 00:00 | |
| Convergence | 01/01/70 00:00 | |
| Why not ? | 01/01/70 00:00 | |
Sample testing | 01/01/70 00:00 | |
| some links | 01/01/70 00:00 |