simple gradient optimzation method

[Martin Larocca]

Hi, dear QuTip Team!

I was wondering if there is any implementation in QuTip of a simple gradient optimization algorithm (with no second order derivatives). I need to be able to choose the step size in order to build "continuos" optimization trayectories through control space. GRAPE's quasi-newtonian methods are no good for this.

We are really greatful to you for the development of this amazing package and have acknowledge your work in our recent pre-print (https://arxiv.org/pdf/1802.05683.pdf) where we study the quantum control landscape of a simple two level system,

Thanks,

Martin Larocca - Universidad de Buenos Aires

[Alex Pitchford]

Dear Martin,

Very happy to hear are making good use of the qutip library. Thanks for citing us.

There are two control optimisation implementations in qutip. The main one uses scipy optimize, and mostly the quasi second order methods you mention. It is described in.

http://nbviewer.jupyter.org/github/qutip/qutip-notebooks/blob/master/examples/optimal-control-overview.ipynb

Also mentioned is the first order method. There are four example notebooks for this, including:

http://nbviewer.jupyter.org/github/qutip/qutip-notebooks/blob/master/examples/control-grape-single-qubit-rotation.ipynb

I don't think this first order method is widely used.

There are many methods available scipy optimize.

https://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.minimize.html#scipy.optimize.minimize

In theory, any of them can be used by the control.pulseoptim functions.

I don't know whether any of them would suit your requirement.

Alex

 

--
You received this message because you are subscribed to the Google Groups "QuTiP: Quantum Toolbox in Python" group.
To unsubscribe from this group and stop receiving emails from it, send an email to qutip+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.