Can We Approximate the Matrix Square Root of A+B Under Certain Conditions?

  • Context: Graduate 
  • Thread starter Thread starter jostpuur
  • Start date Start date
  • Tags Tags
    Approximate
Click For Summary
SUMMARY

The discussion focuses on approximating the matrix square root of A + B, where A and B are non-commuting matrices, and B is smaller than A. It explores the potential of using infinite series, specifically the Taylor series for square roots, to derive an approximation for the square root of A + B. The conversation highlights that while the series for (A + B)^{-1} can be effectively utilized, a direct application for the square root is complicated due to the lack of a straightforward relationship between the square roots of individual matrices. The challenge remains that eigendecomposition is only applicable if A + B is non-singular.

PREREQUISITES
  • Understanding of matrix operations and properties, specifically non-commuting matrices.
  • Familiarity with Taylor series and their applications in matrix analysis.
  • Knowledge of eigendecomposition and its conditions for application.
  • Proficiency in linear algebra concepts, particularly related to matrix inverses and square roots.
NEXT STEPS
  • Research the application of Taylor series in matrix approximations, focusing on square roots.
  • Study the conditions under which eigendecomposition can be applied to matrix sums.
  • Explore advanced topics in matrix analysis, including perturbation theory for non-commuting matrices.
  • Investigate numerical methods for approximating matrix functions, particularly for non-singular matrices.
USEFUL FOR

Mathematicians, researchers in linear algebra, and professionals working with matrix computations who are interested in advanced matrix approximations and their applications in various fields.

jostpuur
Messages
2,112
Reaction score
19
Assume that [itex]A,B\in\mathbb{R}^{N\times N}[/itex] are some matrices which do not commute. Is there any way to approximate the matrix

[tex] \sqrt{A + B}[/tex]

under the assumption that [itex]B[/itex] is smaller than [itex]A[/itex]? Could it be possible to write the square root precisely with some infinite series under some conditions?

Here's a related problem, which I know already. Suppose we were interested in the matrix

[tex] (A + B)^{-1}[/tex]

We know for complex numbers [itex]|z|<1[/itex] the series

[tex] \frac{1}{1 + z} = 1 - z + \sum_{n=2}^{\infty}(-1)^n z^n[/tex]

The same series will work for bounded operators with norm less than one, so by using the formula [itex](XY)^{-1}=Y^{-1}X^{-1}[/itex] we get

[tex] (A + B)^{-1} = ((\textrm{id} + BA^{-1})A)^{-1} = A^{-1}(\textrm{id} + BA^{-1})^{-1}[/tex]
[tex] = A^{-1}\Big(\textrm{id} - BA^{-1}+ BA^{-1}BA^{-1} + \sum_{n=3}^{\infty}(-1)^n(BA^{-1})^n\Big)[/tex]
[tex] = A^{-1} - A^{-1}BA^{-1}+ A^{-1}BA^{-1}BA^{-1} - \cdots[/tex]

We also have Taylor series for square root:

[tex] \sqrt{1 + z} = 1+ \frac{1}{2}z- \frac{1}{4}z^2 + \frac{3}{8}z^3<br /> + \sum_{n=4}^{\infty}\frac{(-1)^{n+1}3\cdot 5\cdot\ldots\cdot (2n-3)}{2^n}z^n[/tex]

Could it be, that this series could be used get something for the [itex]\sqrt{A+B}[/itex]? Actually the previous trick doesn't work now. The problem is that [itex]\sqrt{XY}[/itex] is not related to [itex]\sqrt{X}\sqrt{Y}[/itex] or [itex]\sqrt{Y}\sqrt{X}[/itex] in any obvious way. So I was left without further ideas here.
 
Last edited:
Physics news on Phys.org
Eigendecomposition would work if ##A+B## is non-singular.
 

Similar threads

  • · Replies 8 ·
Replies
8
Views
3K
  • · Replies 21 ·
Replies
21
Views
2K
  • · Replies 26 ·
Replies
26
Views
2K
  • · Replies 4 ·
Replies
4
Views
3K
  • · Replies 3 ·
Replies
3
Views
2K
  • · Replies 52 ·
2
Replies
52
Views
4K
  • · Replies 1 ·
Replies
1
Views
2K
  • · Replies 7 ·
Replies
7
Views
3K
  • · Replies 14 ·
Replies
14
Views
4K
  • · Replies 2 ·
Replies
2
Views
2K