[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Octave-bug-tracker] [bug #61732] news unclear
From: |
Markus Mützel |
Subject: |
[Octave-bug-tracker] [bug #61732] news unclear |
Date: |
Mon, 27 Dec 2021 08:43:44 -0500 (EST) |
User-agent: |
Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/96.0.4664.110 Safari/537.36 Edg/96.0.1054.62 |
Update of bug #61732 (project octave):
Category: None => Documentation
Status: None => Invalid
Open/Closed: Open => Closed
_______________________________________________________
Follow-up Comment #1:
Afaict, there is no unique solution to (nearly) singular square matrix
equations. Instead, `mldivide` returns an "arbitrary" (but hopefully
numerically close) solution `x` that minimizes `norm(A*x-b)`.
In prior version, Octave used an algorithm for which the solution `x` had a
minimum norm `norm(x)`. That's what is called a "minimum-norm solution" IIUC.
In newer versions, it'll use an LU factorization to minimize `norm(A*x-b)`.
That will return a different solution `x`. Afaict, that solution won't
(necessarily) have the minimum norm of all possible solutions. But it happens
to be more numerically stable(?). I.e., the norm of `A*x-b` will more likely
be smaller than using the solution `x` calculated in previous versions of
Octave in that equation.
I don't think that note is contradicting itself. But, it's important to
distinguish between `norm(A*x-b)` and `norm(x)` (which can be minimized at the
same time or not...).
Closing as invalid.
_______________________________________________________
Reply to this item at:
<https://savannah.gnu.org/bugs/?61732>
_______________________________________________
Message sent via Savannah
https://savannah.gnu.org/