MATLAB returns a wrong value of determinant.

Question

Jem 2021-11-10

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1583544-matlab-returns-a-wrong-value-of-determinant

回答： Paul 2021-11-10

采纳的回答： Paul

在 MATLAB Online 中打开

Consider a 2x2 Matrix X whose entries are

5 25

7 35

By hand calculations, we know that the determinant of the Matrix is 0. How come when MATLAB computes this, it gives a value of 3.8858e-15? Here's the Code snippet.

Using also MATLAB R2021b

X =[5 25; 7 35];
D = det(X)

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Paul 2021-11-10

1
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1583544-matlab-returns-a-wrong-value-of-determinant#answer_828594

在 MATLAB Online 中打开

The reason det() doesn't return zero for this matrix is because the floating point errors in the computation of the determinant. If det() computed the determinant using the textbook formula, the result would, of course, be zero.

X =[5 25; 7 35];
D0 = X(1,1)*X(2,2) - X(2,1)*X(1,2)
D0 = 0

But det() uses a more general algorithm as shown on its doc page and repeated here. This algorithm like any other is subject to numerical error

D = det(X)
D = 3.8858e-15
[L,U]=lu(X);
s = det(L);
D1 = s*prod(diag(U));
[D D1]
ans = 1×2
	1.0e+-14 *

    0.3886    0.3886
isequal(D,D1)
ans = logical
   1

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

Answer 2

Jon 2021-11-10

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1583544-matlab-returns-a-wrong-value-of-determinant#answer_828539

There will always be some tolerance in numerical methods compared with exact solutions. 3.88858e-15 is considered very small, and effectively can be treated as zero. It is important to realize this whenever working with numerical methods particularly when testing for certain conditions. You should always define a threshold for what is considered close enough to zero. One useful value you can get to help you scale this is the machine precision, which you can get using MATLAB's eps function. You can then define zero as being some multiple of the machine precision.

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Rik 2021-11-10

在 MATLAB Online 中打开

The reason behind this is that Matlab stores data with finite memory, which might not be enough to yield the mathematically correct answer.

The reason you should consider a value of 1e-15 as 0 is precisely this.

det([5 25;7 35]*1e-15)
ans = 3.7397e-44

Here you see the exact same thing. I personally treat every value 15 orders of magnitude smaller than the input numbers as 0. If you need more precision, you shouldn't be using the standard tools.

请先登录，再进行评论。