Running QAOA on X8?

Hey together!

i wanted to ask if it is possible to run QAOA on the X8 oder any other chip?

Thanks for your help!

Best Regards
Peter

Hey @PetMai and welcome to the forum!

QAOA requires the implementation of a Hamiltonian evolution, and it is not immediately clear (although not impossible?) to match this formalism to the setup of a GBS device - especially if it is a restricted architecture like the X8. Do you have something in mind, or are you just asking in general?

Checkout the apps layer for combinatorial optimisation problems and how to encode them into GBS!

Thank you very much for your answer!
I just want to know in general if it is possible because i read something about QAOA and the MaxCut-Problem on Pennylane, so i was wondering, if it would run on a quanten computer from Xanadu.
According to your answer it would not be possible, right?

Ah, I wonder if you are referring to this demo? If so, then this is unfortunately a qubit-based implementation, while the X8 is a special-purpose continuous-variable architecture.

But hopefully a future generation of chips will be able to do QAOA :slight_smile:

Yes, I did indeed mean this demo.
Alright, thanks for the clarification :slight_smile:

Hi,

Thanks for the great discussion!

I’m trying to implement CV-QAOA for a cost function like f(x)=(x−a)2f(x) = (x - a)^2f(x)=(x−a)2 on the Xanadu X8 chip as per this notebook. From what I understand, X8 supports only passive linear optics operations — beam splitters, phase shifters, and interferometers — acting identically on signal and idler modes initialized as two-mode squeezed states.

My question is:

  • Since CV-QAOA requires displacement gates (e.g., eipx^/ℏe^{i p \hat{x} / \hbar}eipx^/ℏ) and quadratic phase gates, is it correct that displacements cannot be implemented on X8 at all?
  • Are there any known workarounds or approximations for implementing such active Gaussian gates on X8?
  • Has Xanadu planned or implemented displacement operations on any of their hardware platforms beyond X8, like Borealis or lab testbeds?

Thanks in advance for any guidance!
Mas

Hi @mhakimi ,

Unfortunately displacements cannot be implemented on our current hardware and there’s no plan to add it. Our new hardware architecture is based on measurement-based quantum computing. These use CV gates for the creation of our GKP qubits but not for the actual computation. The computation is done via measurements, which apply DV gates (i.e. the normal gates you see everywhere in the literature).

While our latest hardware, Aurora, is not available online, you can read about it in our blog post. You can also read our demo on measurement-based quantum computing, which can be a great insight onto how our newest machines work!

Let us know if you have any additional questions.

I hope this helps!

Hi @CatalinaAlbornoz,

Thanks again for the helpful clarification!
I also noticed that the Pgate isn’t supported on the simulon_gaussian backend. However, I couldn’t find official documentation listing the supported operations on this backend. Would you mind pointing me to any resources or documents that outline its capabilities and limitations?

Additionally, as we’re working on implementing actual CV algorithms using gates like Dgate, Sgate, and Pgate, do you have any suggestions or advice on how such algorithms could be practically implemented on existing devices?

We’d really appreciate any guidance you can offer!

Thanks,
Mas

Hi @mhakimi ,

The simulon_gaussian simulator runs on the gaussian backend (see Simulating your program) and thus only supports Gaussian operations, including homodyne and heterodyne measurements, as well terminal Fock measurements. Note that there are limits to how many measurements a circuit can have depending on the type of measurement. These can be retrieved by calling engine.device_spec.modes with engine = sf.RemoteEngine("simulon_gaussian") .

You can check the info for the Gaussian backend to get more details about the gates that are supported. The primitives and decompositions in the compilers section of the docs can help.

Unfortunately the only device we have online is X8 and it’s not going to be enough for your needs. The new devices we’re building will not even give access to the CV side of things, they will require using discrete gates such as Hadamard, Pauli gates, etc. There may be other technologies that give you access hardware where you can do CV algorithms but I’m not aware of any at the moment unfortunately.

So for now I’d suggest sticking to simulators if you really want to do CV algorithms.

I’m sorry I don’t have better news.

Thank you @CatalinaAlbornoz for your support!
I appreciate it!

1 Like