solving symbolic inverse of big matrix takes long

9 次查看（过去 30 天）
Martin Baumann
Martin Baumann 2021-5-6
回答： Aditya Patil 2021-5-21
I would like to solve a symbolic linear system looking like this: (A1*s+A2)*x=(B1*s+B2)*u
where: x...states, u.... input, A1,A2,B1,B2.... numeric matrizes, s...laplace variable
I tried multiple approaches such as:
linsolve(A1*s+A2,B1*s+B2)
(A1*s+A2)\(B1*s+B2)
inv(A1*s*A2)*(B1*s+B2)
Minimalexample:
s=sym('s');
A1=rand(15);
A2=rand(15);
B1=ones(15,1);
B2=ones(15,1);
tic; (A1*s+A2)\(B1*s+B2); toc
Elapsed time is 35.918775 seconds.
The most expensive computation is taken by the inverse of (A1*s+A2).
Is there a way to further reduce the computation time? Goal is to calculate only one component of x. I actually don't need the solution for all x but since the system is coupled I don't see a way to further decrease the complexity.

请先登录，再回答此问题。

回答（1 个）

Aditya Patil
Aditya Patil 2021-5-21
Ran in:
In this case, using inverse is much faster than mldivide.
N = 15;
syms s
A1=rand(N);
A2=rand(N);
B1=rand(N,1);
B2=rand(N,1);
x = (A1*s+A2);
y = (B1*s+B2);
tic; result1 = x\y; toc;
Elapsed time is 48.751415 seconds.
tic; result2 = inv(x) * y; toc;
Elapsed time is 1.878644 seconds.
sum(abs(subs(result1, s, 1) - subs(result2, s, 1)))
ans = 
0
I tried the code for various sizes of N, and inverse is faster for this problem.
Regarding avoiding inverse, MATLAB doesn't provide any inbuilt functionality for this. However, you can implement custom code which calculates only the required components of the inverse, and uses them in the equation. This might not give major performance gain though.

类别

在 Help Center 和 File Exchange 中查找有关 Linear Algebra 的更多信息

标签

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by