zlatko-minev
commented
4 years ago It’s a good idea. Try and see if it helps. Qutip uses some sparse matrices here and there but not sure if that will help us, actually probably not since H isn’t sparse.
We can then remove Qutip and/or add an option to export to it, by converting np.ndarray to QObj in a function
But probably not even much needed
Sent from my iPhone
On May 6, 2020, at 11:17 AM, Daniel Cohen Hillel notifications@github.com wrote:
Qutip is a great package and it provides many, very useful functionalities for quantum mechanics. One of the main things it can do is represent a quantum system with the qobj class. A qobj object is not much more than a numpy ndarray but it has some problems. Doing operations on qobjs is much slower than simply working with a numpy array (I would say around ~15 times slower on average, obviously depending on what you're doing and how much), also it manages memory worse and uses system resources less effectively. To be fair, numpy is one of the most used python packages in the world so obviously it's super optimized.
Anyway, for pyEPR, qutip is used on around ~10 lines and everything it is used for can be done with a numpy array without really changing the code much. I got this idea when I tried analyzing 10 modes and qutip returned and overflow error, and I think switching to numpy might help solve this problem. And as a bonus there won't be a dependency on the qutip package :)
All of the uses of qutip are one of these:
qutip.qeye ---> np.qeye qutip.tensor ---> np.kron qutip.destroy/create/num/basis ---> can be done in 3 lines with numpy, we can add this as a function in toolbox Replacing matrix multiplication * in qutip with @ in numpy I might do it myself later today, tell me what you think.
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or unsubscribe.
DanielCohenHillel
Qutip is a great package and it provides many, very useful functionalities for quantum mechanics. One of the main things it can do is represent a quantum system with the
qobjclass. Aqobjobject is not much more than anumpy ndarraybut it has some problems. Doing operations onqobjs is much slower than simply working with a numpy array (I would say around ~15 times slower on average, obviously depending on what you're doing and how much), also it manages memory worse and uses system resources less effectively. To be fair, numpy is one of the most used python packages in the world so obviously it's super optimized.Anyway, for pyEPR, qutip is used on around ~10 lines and everything it is used for can be done with a numpy array without really changing the code much. I got this idea when I tried analyzing 10 modes and qutip returned and overflow error, and I think switching to numpy might help solve this problem. And as a bonus there won't be a dependency on the qutip package :)
All of the uses of qutip are one of these:
qutip.qeye--->np.qeyequtip.tensor--->np.kronqutip.destroy/create/num/basis---> can be done in 3 lines with numpy, we can add this as a function in toolbox*in qutip with@in numpyI might do it myself later today, tell me what you think.