Application related to nonlinear optimization

[sandeep026]

Dear developers,

I want to solve constrained nonlinear optimization problems using symforce for small to medium sized problems. Can I formulate objective , constraints and compute their derivatives for use in a solver like IPOPT? If there are some existing works, references will be greatly helpful.

Additional information

1. derivatives are usually sparse
2. objectives and constraints can be evaluated in parallel

[ryan-brott-skydio]

Can I formulate objective , constraints and compute their derivatives for use in a solver like IPOPT?

1. derivatives are usually sparse

Yes, SymForce can be used for these problems. We don't have any direct examples of this, but I'll sketch out how I would approach it.

• Break the constraints into dense fragments and generate code for each of them.
• Wrap these functions in sym::Factor objects and pass them to a sym::Linearizer. I would follow the sparse linearizer test.
  You can do something similar for the objective (perhaps with the dense linearizer instead of the sparse one).

2. objectives and constraints can be evaluated in parallel

The SymForce linearizers do not exploit thread-level parallelism, though user code may strategically split factors between multiple linearizers, run linearization in parallel, and combine the linearizations together (one can imagine a few variants of this strategy).