I have a few questions on photonics quantum computing.
Since photonics quantum computing are continuous variable, are they also known as analog quantum computing?
Am I right to say that a single qumode can in theory encode an infinite (large) dimensional data, e.g: 1000-dimensional data, but in practice, is truncated to a lower number of dimensions per qumode? If yes, what is the normal number of dimension per qumode that works with current hardware?
If I entangle 4 qumodes with each encoding a 10-dimension data, the state space will be 10^4?
I looked at the all the operations that are supported in Strawberry Fields through the class inheritance diagram at https://strawberryfields.readthedocs.io/en/stable/code/sf_ops.html#class-inheritance-diagram.
I’m not sure when should I apply which operations, e.g. which preparation routine will work for which problem. Are there any resources where I could learn the functionality of each operation? Are these operations in Strawberry Fields supported in MrMustard too?
Thanks for your questions. I’ll try to answer as best as I can.
I don’t know if analog computing has additional or different implications. Xanadu’s old approach to quantum computing (with our X series devices and Borealis) used discrete gates applied on quantum modes, which have a continuum of coefficients. You can read more here and check if this aligns with your definition for analog quantum computing or not. Important note: Xanadu’s newest devices and developments use GKP qubits, which are discrete, not continuous.
Again, I don’t think the framing for this question is correct, so reading the tutorials above can help clarify this.
It depends on what you want to do. From your questions it sounds like learning PennyLane might be an easier way to start, and getting familiar with concepts like measurement-based quantum computing. MrMustard is specialized in a specific type of simulations, more about the photonic system itself and the information side of things, not really making large computations.
Strawberry Fields operations are not supported in MrMustard per-se. There are some equivalent gates such as a Displacement gate (Dgate) but they’re built from scratch in MrMustard and there’s no “converter” that will turn a Strawberry Fields program into a MrMustard program.
For more context:
Strawberry Fields was designed for people to run programs on Xanadu’s old devices and simulators. It’s still valuable to many people doing research in photonics, but not really for computing. MrMustard was designed for doing specific simulations of the newer generation of Xanadu devices and architectures, so for people doing research in photonics (not so much computing) this is a very useful library.
For anyone doing quantum computing, especially if you’ve come from a background in gate-based quantum computing, PennyLane is the way to go. As we further develop our hardware, we’re also developing PennyLane so that in the future you can run PennyLane programs on any kind of hardware, not only gate-based devices which is the case today.
I hope this helps you explore new paths towards quantum computing!
Final thought: It’s a steep learning curve going from gate-based to continuous-variable quantum computing. Feel free to ask other questions here if you have them, and hopefully the resources I’ve shared will clarify many of your questions.
Thank you for your detailed answered. I read through the resources that you gave and I cleared much of the misconceptions I had.
I have some questions:
Do the latest GKP qubits only work as qubits and can’t be extended to qudits? I saw a post that support for qudits might be added in the future at Qudits in Pennylane. I have some knowledge on qudits and I am thinking qudits can be useful for encoding high-dimensional data in quantum machine learning.
Also on the same vein as the first question, since the latest devices are GKP qubits, can the systems still be used for continuous variable experiments using MrMustard for example?
Measurement-based quantum computing is interesting! Will the bottleneck be the speed of measurement because there will be a lot of measurement in the computation?
Our GKP qubits in fact only act as qubits, not qudits. Qudits are not in the PennyLane roadmap unfortunately.
I don’t think the new GKP qubits can be used for continuous-variable experiments but I’m not 100% sure. Also, please note that these devices are not available on the cloud, so you wouldn’t really be able to use them in the near future.
At the moment our bottleneck is loss. What this means is that if our devices don’t have the quality we need, then the light bounces off in the wrong places and gets lost. This is the main challenge our team is tackling at the moment!
Overall it might be interesting to study qudits at a theoretical level, but so far there aren’t many tools I’m aware of to simulate them. This may respond to the fact that quantum hardware companies like Xanadu are building qubits instead of qudits.
I know this isn’t great news but I hope this helps!
I have some questions about measurement-based quantum computing. From Measurement-based quantum computation | PennyLane Demos, I understand that there are a lot of measurements and operations conditioned on the measurements that are involved in MBQC.
What is the real advantage of measurements based quantum computing (MBQC) over normal gate based model? For instance, can it do some algorithm that is beyond normal gate based model?
You can see some of the advantages on the Xanadu website.
A lot of the advantages relate to scaling. For some of the other technologies it might be easier to get to hundreds of qubits but fundamentally harder to get to thousands or millions of them. With MBQC it might be harder to make a few qubits but once we have achieved certain specifications it might be comparatively easier to scale to thousands or millions of qubits.