search

mforets / carlin

Carleman linearization of polynomial differential equations in SageMath.
GNU General Public License v3.0
2 stars 3 forks source link

Contradictory license information #5

Closed freemin7 closed 3 years ago

freemin7 commented 3 years ago

The label in readme says MIT but the side bar says GLP-3.0

mforets commented 3 years ago

The licence is https://github.com/mforets/carlin/blob/master/LICENSE, GPL-3.0. I don't know why github displays the MIT in the icon, i can try to fix it later.

By the way, please note that I no longer maintain this repository. You may be interested to know that I recently added new methods for Carleman linearization in the project JuliaReach, more specifically in https://github.com/JuliaReach/ReachabilityAnalysis.jl , see https://github.com/JuliaReach/ReachabilityAnalysis.jl/tree/master/src/Algorithms/CARLIN

If you have ideas to apply these methods I'm interested to discuss, drop me a message!

freemin7 commented 3 years ago

Awesome, i was thinking about porting those methods to Julia! Great if they are already there.

mforets commented 3 years ago

i've updated the license badge in the README, using the service https://shields.io/category/license

Awesome, i was thinking about porting those methods to Julia! Great if they are already there.

not all methods are ported to Julia -- but the new Julia implementation is more powerful in some aspects, for example that i have successfully built matrices with millions of dimensions efficiently, because it does so by only operating with sparse matrices. the part that i didn't yet implement in Julia is the automatic translation from the polynomial ODE to the quadratic form x' = F1 * x + F2 * (x \otimes x).

freemin7 commented 3 years ago

If you ever want to implement this transform, please talk to @ChrisRackauckas and what benefits result from including it somewhere under the DiffEq eco system (probably modeling toolkit) since multiple things in the eco system could profit on the transform. If you don't implement the transform in 2021, i might look into it or something.

ChrisRackauckas commented 3 years ago

I think doing a purely symbolic version via ModelingToolkit can allow for some extra optimizations when mixed with CSE and simplifications, but yes the numerical way is more flexible and powerful so both have certain advantages and disadvantages. 2021 will be an exciting time for these kinds of problems.

mforets commented 3 years ago

@freemin7 Sure I'll ping you if I have updates on that transform! (Don't hesitate to ping me if you want to discuss sth ;) )

@ChrisRackauckas Yes a fully symbolic approach or a combined symbolic/numeric approach would be very interesting. Then we need to take the matrix exponential of that beast / compute its action on vectors obtained by successive Kronecker powers. Maybe the symbolic arrays infrastructure that you are developing can help.

My renewed interest on this matter is related to applying (in the context of set propagation for nonlinear ODEs), the new error bounds from https://arxiv.org/pdf/2011.03185.pdf (see https://www.quantamagazine.org/new-quantum-algorithms-finally-crack-nonlinear-equations-20210105/ for a less technical introduction). Maybe the folks that are developing QC algorithms (https://julialang.org/blog/2019/12/yao-differentiable-quantum-programming/ ) become interested in this transform too.