Changing precision in rho_ss

[Perry Rice]

I am having trouble with the syntax for this, looking at a system that is very weakly driven, so small numbers. My program runs but gives nonsense below a certain pump level. Tried various solver methods, direct,eigen, svd..... and they are all consistent, so I figure it must be in the lack sig figures somewhere.

Thanks

[Éric Giguère]

Have you tried changing the tolerances?

`options={"atol": 1e-8, "rtol": 1e-6, "nsteps": 2500}` 

Lowering these tolerances will make the simulation slower, but should help with weaker drive.

[Perry Rice]

I have but have struggled with the syntax. Does your code segment go inside the rho_ss(H........) call or as a separate options = command in a separate line? Thanks

[Perry Rice]

adding that inside the rho_ss call runs, changing the tolerances does not improve performance nor take things longer to run

using 5.1

 rho_ss=steadystate(H,c_ops,options={"atol": 1e-8, "rtol": 1e-6, "nsteps": 2500})

[Perry Rice]

that is then passed to correlation_3op_2t

[Éric Giguère]

The `options` arguments should be passed to the `correlation_...` function.

For the steadystate call, the `solver` parameter control which function from numpy.linalg or scipy.sparse.linalg is called and extra keyword parameter are passed to these function.

rho_ss = steadystate(H, c_ops, solver="bicgstab", atol=1e-16, rtol=1e-5)

will call

scipy.sparse.linalg.bicgstab(L_as_scipy, zeros_vec, atol=1e-10, rtol=1e-5)

Per default scipy.sparse.linalg.spsolve is used, which does not have an explicit parameter.

[Perry Rice]

Thanks. That runs but does not solve my issue. Is there a way to specify precision in correlation_3op_2t?

Appreciate the help

[Éric Giguère]

Correlation uses mesolve, which receive the options:

`correlation_3op_2t(H, c_ops, args, options={"atol": 1e-10, "rtol":1e-8, nsteps=1e5})`

"atol"/"rtol" is applied by internal step, not total evolution, nor tlist / tau step. 

So it should be set a lot smaller than what you need.

There is a limit on how much precision we can get using double and qutip does not support any higher precision.