Re: [Maxima-discuss] bug in ic1

[Robert D. <rob...@gm...>]

On Sun, Feb 21, 2021 at 11:34 AM Dimiter Prodanov <dim...@gm...> wrote:

> ic1(z(x)=%c1-'integrate((z(u607)^2*(x-u607))/u607,u607,x,M),x=M,z(x)=0);
> crashes

> however
> ic1(z(x)=%c-'integrate((z(u607)^2*(x-u607))/u607,u607,x,M),x=M,z(x)=0);
>
> returns
> z(x)=z(M)-integrate((z(u607)^2*x-u607*z(u607)^2)/u607,u607,x,M)
>
> So in my view, the symbol %c should not be hardcoded in ic1.

Hi Dimiter, by the way, from where did %c1 appear?

I've looked at this problem for a bit, and what I'm seeing is that the
ode2 code assumes throughout that constants are only named %c, %k1, or
%k2. It wouldn't be too hard to change that to numbered constants;
there is already a mechanism (integration_constant and
integration_constant_counter) to generate numbered constants in the
integration code.

There are at least two problems introduced by that. The smaller one is
that the presence of numbered constants would probably make it harder
to check results automatically. OK, that can be fixed up without much
trouble.

The larger one is that the results might seem messy to users. After a
while you'll get stuff like %k67*sin(x) + %k68*cos(x) and if you want
"nice" constants you'll have to reset integration_constant_counter.
That sounds painful.

In some sense creating new constants is the right thing to do. Is it
sufficiently right to overcome the clumsiness of it? What does anyone
think about any of this stuff?

best,

Robert Dodier