Re: ublas : double * matrix

From: jhrwalter (walter_at_[hidden])
Date: 2002-01-18 17:42:04

--- In boost_at_y..., Toon Knapen <toon.knapen_at_s...> wrote:
> jhrwalter wrote:
> 
> > --- In boost_at_y..., Toon Knapen <toon.knapen_at_s...> wrote:
> > 
> >>I noticed that in ublas 'a = 2 * b' works but 'a = b * 2' does 
> >>
> > not ?!
> > 
> > Meanwhile we've found two reasons for this restriction. First, as 
far 
> > as I know multiplication with a scalar is mathematically always 
> > defined as left multiplication for vectors and matrices. 
> 
> 
> theoretically scalar-matrix multiplication is commutative and thus 
a * M 
> == M * a.

Hm. I'm not sure, if the rhs is defined at all.

> > Next, 
> > multiplication is left associative in C++ (as Mathias told me ;-
), so 
> > if we compare a = 2 * 2 * b to a = 2 * b * 2, the former is 
always 
> > guaranteed to need fewer operations.
> 
> 
> I can understand your implementation issues 

I believe, this is more an interface than an implementation issue.

> and it's good to know now 
> that Matrix * scalar can not be done but one needs to do scalar * 
matrix 
> instead. I just took me fifteen minutes to find out and maybe it's 
a 
> good idea to put it in the docs.

Certainly. Sorry for the inconvenience.

> e.g. in the mean time again I had the same trouble with division, I 
> tried M /a but to make it compile I redefined it as (1/a) * M : one 
just 
> needs to know.

Division is another topic. We once discussed to introduce such a 
division operator for accuracy reasons, although it's usually not 
defined mathematically, too.

> Although I think in the long term, all mathematical correct 
expressions 
> should be supported but it is not highest priority.

Regards

Joerg