Matlab returns different values with two equivalent functions

Question

Marcos Baños 2020-3-8

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/509676-matlab-returns-different-values-with-two-equivalent-functions

评论： Marcos Baños 2020-3-10

采纳的回答： Thiago Henrique Gomes Lobato

Let's say I have two functions:

f = @(x) (1.01.*exp(4.*x) - 4.62.*exp(3.*x) - 3.11.*exp(2.*x) + 12.2.*exp(x) - 1.99)

F = @(x) (((1.01.*exp(x) - 4.62).*exp(x)-3.11).*exp(x) + 12.2).*exp(x) - 1.99

These two functions are equivalent but when I want to execute with MATLAB it gives me different results with the same input. (for example with 0.925)

I would like to ask why cause this situation and which one has should I trust.

Thanks in advance!

2 个评论
显示无隐藏无

darova 2020-3-8

在 MATLAB Online 中打开

>> f(0.925)
ans =
  -24.2489
>> F(0.925)
ans =
  -24.2489

Works as always

Marcos Baños 2020-3-8

ans =

-24.2489252551835

ans =

-24.2489252551834

as you see it's not the same, and with 1.53 the difference is bigger

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Thiago Henrique Gomes Lobato 2020-3-8

1
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/509676-matlab-returns-different-values-with-two-equivalent-functions#answer_419088

编辑：Thiago Henrique Gomes Lobato 2020-3-8

在 MATLAB Online 中打开

The difference between then in my computer with your example is -3.197442310920451e-14 (depending of your processor may be slightely different but still extremely low). This difference is negletible and can be regarded as numerical/approximation error between taking the exp of a number and multiplying two (to four, in your case) exp's. You can reproduce it easily with an easier example:

exp(3.245*2)-exp(3.245)*exp(3.245)
ans =
    -1.136868377216160e-13

The results are then actually the same and you can trust both of them with the given accuracy, which should be enough for almost any application you may think.

6 个评论
显示 4更早的评论隐藏 4更早的评论

Thiago Henrique Gomes Lobato 2020-3-8

The problem is not with matlab, but rather with numerical computation itself. Even if you use variables with a higher precision (which, in matlab, I believe it only can be done with symbolic toolbox) you can never expect a result to be always "perfect". This truncation error is something that, unfortunately, anyone that uses computer calculations has to deal with. About your question of which one you should use, the answer is that it actually doesn't matter much. You always will have something like: Result = Realresult + numericalError. This numerical error is normally bounded and, even though some specific operations do provide less numerical errors, the difference between then is so minimal that only some very restricted application could take any advantage of it. If you really want a "black and white" answer, I would guess that the variant F can be slightely better since you have actually only one expoent computation and everything else is sum and multiplication, which should produce marginally less numerical errors.

Stephen23 2020-3-8

编辑：Stephen23 2020-3-8

"So it means that you have not enough precision in MATLAB to produce the correct answer?"

This is not specific to MATLAB, it is due to how all modern computers store fractional values as binary floating point numbers, and would occur with any numeric application that uses the very-efficient inbuilt floating point operations of your CPU. Read these to know more:

https://www.mathworks.com/matlabcentral/answers/57444-faq-why-is-0-3-0-2-0-1-not-equal-to-zero

https://www.mathworks.com/matlabcentral/answers/316889-why-two-equal-numbers-are-not-equal

http://www.mathworks.com/matlabcentral/answers/102419-how-do-i-determine-if-the-error-in-my-answer-is-the-result-of-round-off-error-or-a-bug

http://www.mathworks.com/matlabcentral/answers/140656-about-numerical-difficulty-of-equality-comparison-operation-between-two-double-real-numbers-in-matlab

This is worth reading as well:

https://docs.oracle.com/cd/E19957-01/806-3568/ncg_goldberg.html

http://faculty.tarleton.edu/agapie/documents/cs_343_arch/papers/1991_Goldberg_FloatingPoint.pdf

Marcos Baños 2020-3-10