MATLAB Answers

Understanding Precision in Matlab

10 views (last 30 days)
Ahmad Gad
Ahmad Gad on 1 Jul 2020
Commented: Walter Roberson on 1 Jul 2020
I have a simple question. How to split the different between the two numbers into 1e6 increment (unisg linspace)?
The numbers: 5.047639467288342e+4 and 5.047639467288343e+4.
Thanks

  2 Comments

Stephen Cobeldick
Stephen Cobeldick on 1 Jul 2020
"How to split the different between the two numbers into 1e6 increment (unisg linspace)?"
You can't. There are no binary floating point numbers inbetween the two values that you gave.
We can confirm this very easily by noting that their hex representations differ only by 1:
>> A = 5.047639467288342e+4;
>> B = 5.047639467288343e+4;
>> num2hex(A)
ans = 40e8a58ca12906dd
>> num2hex(B)
ans = 40e8a58ca12906de
Ahmad Gad
Ahmad Gad on 1 Jul 2020
Thank you! Then, I have to consider changing the algorithm then.

Sign in to comment.

Answers (1)

Walter Roberson
Walter Roberson on 1 Jul 2020
You cannot do that with double precision.
If you have the symbolic toolbox then:
r = linspace(sym('5.047639467288342e+4'), sym('5.047639467288343e+4'), 1e6+1);
Note that the results will be symbolic numbers with higher precision than double(), so if you were to double() them, it would be with loss of precision

  2 Comments

Ahmad Gad
Ahmad Gad on 1 Jul 2020
The code is numeric and those values will be inserted a matrix determinant function. I think it will be time consuming.
Walter Roberson
Walter Roberson on 1 Jul 2020
Determinants can have a lot of problems with cancellation. If you were to use symbolic numbers it would be better to use
r = linspace(sym(5.047639467288342e+4,'f'), sym(5.047639467288343e+4,'f'), 1e6+1);
However, symbolic determinants are slow. To put 1e6 numbers into an array you would need an array at least 1e3 x 1e3, and a proper accurate determinent of that would involve over 4 * 10^2567 terms.

Sign in to comment.

Tags

Products


Translated by