gate_rot_freq configuration parameter default value not consistent between C++ and Python interfaces

[rachelwaldon]

@srkohn

I'm currently using this config file with quandary v4.0 on Ubuntu

nlevels = 2
nessential= 2
ntime = 1061
dt = 0.0942507068803016
transfreq = 4.10595
rotfreq= 4.10595
selfkerr = 0.2198
crosskerr= 0.0
Jkl= 0.0
collapse_type = none
initialcondition = basis
control_segments0 = spline, 150
control_initialization0 = random, 0.0070710678118654745
control_bounds0 = 0.05
carrier_frequency0 = 0.0, 
control_enforceBC = False
optim_target = gate, xgate
optim_objective = Jfrobenius
optim_weights= 1.0
optim_atol= 1e-4
optim_rtol= 1e-4
optim_ftol= 1e-08
optim_inftol= 1e-09
optim_maxiter= 200
optim_regul= 0.0001
optim_regul_interpolate=False
optim_penalty= 0.1
optim_penalty_param= 0.0
optim_penalty_dpdm= 0.01
optim_penalty_energy= 0.1
datadir= ./c_plus_plus
output0=expectedEnergy, population, fullstate
output_frequency = 1
optim_monitor_frequency = 1
runtype = optimization
usematfree = False
linearsolver_type = gmres
linearsolver_maxiter = 20
timestepper = IMR
rand_seed = 1234

when I run with the C++ executable it says it is using the default gate_rot_freq=1e+20 in the warning.

With the default value gate_rot_freq=1e+20 the optimization converges as expected and matches optimization results from quandary v3.0

If I reproduce this configuration with the python interface, I would assume it chooses the same default value for gate_rot_freq but it uses 0.0 and the result is completely different.

from quandary import Quandary

quandary_opt = Quandary(
    Ne=[2],
    freq01=[4.10595],
    rotfreq=[4.10595],
    optim_target="gate, xgate",
    tol_infidelity=1e-9,
    costfunction="Jfrobenius",
    tol_costfunc=1e-8,
    usematfree=False,
    carrier_frequency=[[0.0]],
    rand_seed=1234,
    verbose=True,
    maxctrl_MHz=50.,
    nsplines=150
)

quandary_opt.optimize(datadir="/home/waldon1/quandary/py_interface")

The fidelities are different and the optimizer doesn't converge.

It seems from experimentation that gate_rot_freq is an important parameter for the optimization and is affecting whether the optimizer converges. I would like the python interface to expose this to be configurable by the user so I can have more control over the optimization and match the C++ results if I need to for backward compatibility reasons. It is currently hardcoded in the quandary.py file and I am not able to change it or configure it.

Let me know if this is possible thanks!

[steffi7574]

Fixed in commit 9aa8cc663067b. The default gate rotation in the python interface is now set to the rotational frame frequencies (same as in the C++ version). I also removed the warning of missing option, because it reported the wrong default. The default can be overwritten with the option "gate_rot_freq" (list, one float per oscillator), but it is not recommended to do so. Optimization model and target gate should typically be rotated with the same frequencies, so it is advised to stick to the default.

Thanks for finding this bug!