# NSolve accuracy issues

themadmathematician shared this problem 3 years ago
Solved

NSolve seem to have accuracy issues. I have confirmed the following issue from a Swedish user. Look at the following:

A = (1, 0)

B = (9, 4)

C = (15, 0)

f(x) = RegressionPoly({A, B, C}, 2)

g: y = 3.5

D & E created from command Intersect(f, g)

NSolve(f = 3.5)

Seems innocent enough but the solutions to NSolve DO NOT match the x-values for the intersection points. Worse: If the number of displayed digits is changed, then the NSolve solutions jump about and though the accuracy slowly increases, it is about two decimal placs worse than the shown accuracy. For 0 and 1 decimals shown even the number of solutions change.

Also, the slope, is not that low at the intersection. It is not as if it is critically near 1 or anything.

I suggest looking over the accuracy conditions for NSolve, and investigating why they fail in this case. 2

Thank you, I've looped your report to the Support Team 1

Thanks. This was in Classic 5, I might add. 1

doing rounding to 5 decimals the problem disapears, then I think there is a rounding problem in Nsolve command 1

Thank you for reporting this issue. I will have a look at it. 1