Code for determinant.

Question

eden hazard 2018-11-14

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/429944-code-for-determinant

评论： Walter Roberson 2021-2-26

I would like to built a code to find the determinant of a 24x24 matrix.I dont want to use the det(X) function,but a fuction that would be quick.

The one i created,whick is too slow for that.

2 个评论
显示无隐藏无

Stephen23 2018-11-15

"I dont want to use the det(X) function,but a function that would be quick."

Ah, the sweet, blind optimism of youth.

Have you actually looked at the definition of the determinant, and figured out how many calculations it would require for a 24x24 matrix? When is your estimated completion?

eden hazard 2018-11-15

Haha it is the optimism of my teacher. The fuction should calculate the determinant in 5-30 minutes

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

John D'Errico 2018-11-15

3
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/429944-code-for-determinant#answer_347009

在 MATLAB Online 中打开

NO NO NO!!!!!

First, you should not be using the determinant here. I don't know why you think you need to use it. But if you don't understand how to efficiently compute a determinant, the probability is 1 that you should not be using the determinant in the first place. There are far better tools to determine is amatrix issingular, which is why you are surely wanting to compute the determinant of a 24x24 matrix.

Ok, why is the code you wrote a an obscenely bad one? Do you understand that the code you wrote will require O(factorial(24)) computations to execute?

factorial(24)
ans =
      6.20448401733239e+23

Even the largest supercomputer the NSA owns would grind to a miserable halt.

So, CAN you compute the determinant more efficiently? Well, yes. USE DET! (In fact there are ways to compute the determinant efficiently. But since you should not be using it in the first place, I won't go into details.)

So better yet, don't compute the determianant at all. So instead, you need to learn how to use tools like rank, cond or svd to do what you are trying to do. Of course, we don't really yet know what you are doing.

6 个评论
显示 4更早的评论隐藏 4更早的评论

James Tursa 2018-11-15

编辑：James Tursa 2018-11-15

My teacher in college once described the minors/cofactor method to me this way:

"Well, it may be slow, but it's innacurate!"

Steven Lord 2018-11-16

"innaccurate" as in "start it computing, go to bed, and maybe you'll have something close to right when you wake up?" grin

请先登录，再进行评论。

Answer 2

Reza Lashani 2021-2-3

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/429944-code-for-determinant#answer_614042

编辑：Reza Lashani 2021-2-3

One of the fastest ways to determine the determinant of a matrix is doing row operation. For an invertible matrix we know that row operations finally reach to identity matrix which has determinant equal to 1. For calculating determinant we can write an efficient and of course fast code to do row operation and it is not necessary to make the first non-zero entry 1 by dividing it by itself. It cause we finally reach to a diagonal matrix (stead of identity matrix) and the determinant is equal to the product of pivots of the final matrix. Doing this algorithm on my ordinary system for a 24 × 24 matrix lates less than 1 second.

4 个评论
显示 2更早的评论隐藏 2更早的评论

John D'Errico 2021-2-3

编辑：John D'Errico 2021-2-3

在 MATLAB Online 中打开

No. The best way to compute the determinant, If your matrix is non-symbolic is to use det. (det does use LU, so it is close to what you have suggested.) But even there, computing the determinant of a 24x24 matrix is just a silly thing to do. It is highly unstable.

Anyway, while you may THINK that your time of "less than a second" is a good thing. So it might be a smart idea to actually see how fast det is.

A = rand(24);
timeit(@() det(A))
ans =
   4.8397e-06

So less than a second is correct. But in this case, MATLAB actually takes only a few MILLIONTHS of a second.

Writing your own code to do matrix operations is a BAD thing to do, especially if you don't really understand what is happening. And even if you do, it is STILL a bad thing. Never write your own code to perform computations when you already have high quality code available for the same purpose, written by people who are seriously good at what they do.

Consider that random matrix.

>> det(A)
ans =
     0.030614
     
>> det(10*A)
ans =
   3.0614e+22
   
>> det(A/10)
ans =
   3.0614e-26

Is computing the determinant useful for ANYTHING? Not really, if you consider that simply scaling the matrix produces either a result that is huge, or incredibly tiny for the determinant. And that is almost always the reason people have to want to compute a determinant. Is the determinant zero?

Is that matrix singular? NO. Of course not. But if you use the determinant to decide if A/10 is singular, you might easily decide it is.

rank(A)
ans =
    24

So while the best way to compute the determinant of a matrix is to use det, an even better way is to not compute the determinant at all. There is literally no good reason to compute the determinant of a 24x24 matrix.