Request for support for inequality constraint equations

[Peter230655]

As I was told, opty can handle algebraic constraint equations.
Can it also handle alegebraic inequalities? What I mean is this:
Say, $a, b \in R, a < b$. Let $f (q_i)$ be a scalar function of the generalized coordinates.
Would $a < f(q_i) < b \longleftrightarrow$
$f(q_i) - b < 0$
$a - f(q_i) < 0$ be possible?

Thanks for any help!

[moorepants]

No, that is currently not supported, but opty could support anything that fits an NLP problem that IPOPT can solve. See the definition here: https://coin-or.github.io/Ipopt/

Someone would have to implement inequality algebraic constraints in opty. It only constructs equality constraints at the moment.

[Peter230655]

Clear thanks!
Should I close the issue, my question was answered!, or leave it open for possible future activity by someone?

[moorepants]

I changed the title and we can leave it as an open issue.

[moorepants]

I think this should be relatively easy to implement. We just need a flag to Problem that lets you change:

$$ 0 < g(x) < 0 $$

to

$$ lower < g(x) < upper $$

Maybe something like:

Problem(..., inequalities=((3, -1.2, 5.0), (8, 0.0, np.inf)))

Where the first number in each tuple corresponds to the index in the equations of motion that holds $g(x)$ and the next two numbers are lower and upper bounds for that equation.

[Peter230655]

I guess, you meant $0 \leq g(x) \leq 0$
The situation I wanted to model would certainly be covered by this!
Unfortunately, I would have no idea how to implement it. :-(

[moorepants]

Unfortunately, I would have no idea how to implement it. :-(

I don't believe that. :)

[Peter230655]

Easier than GitHub / PRs ? :-)

[Peter230655]

If I wanted to have a look at this, would I have to have a full understanding of Ipopt?
Where would I start to look in opty?
Presumably I would have to fix this "windows issue" first?
(This should be solvable, Timo uses windows and does not have these problems)
Thanks!

[moorepants]

If I wanted to have a look at this, would I have to have a full understanding of Ipopt?

You wouldn't have to have a full understanding, but you will have to understand the relevant parts.

Where would I start to look in opty?

The ConstraintCollocator class will need to handle extra equations. I think that it only allows # equations equal to the # of states at the moment. And the Problem class will have to be modified to allow a user to specify which equations are inequalities and whether >= or <=. We should try carefully think about the API design for specifying this, so that it isn't clunky.

[Peter230655]

This sounds beyond my capabilities.