#quantum #AI

If you want to learn more I will be at APS March meeting talking about this work!

#quantum #physics #quantumcomputing

@bbrunetto.bsky.social

This behavior also appears to be stable: shifting the model away from the DU point, we still observe faster state design preparation, but the story is more complicated than being characterized only by entangling power.

The rate of approaching Haar statistics is governed by the entangling power of the static 2-local gates. Even modest entangling power is enough to be faster than circuits made up of 2-local Haar random gates.

We do this with two strategies: random driving only on the boundary, or random driving each qubit for each layer of the curcuit. Remarkably, at the dual unitary point this prepares state designs significantly faster than circuits built entirely from random unitaries.

In the past, studies have focused on the case where circuits are made up entirely of local random unitaries. Instead we approach the problem with fixed two-local gates, and restrict ourselves to inserting driving with one qubit random gates.

#quantum #physics #science

I've always been passionate about this, and I recently had the opportunity to do another podcast: youtu.be/xtnFRWCvo9E?...

AND of course signatures of more generic dynamics than traditional integrable models. The answer is yes! But with lots of open questions to push this truly to a spatially extended locally interacting Hamiltonian. #quantum #physics

With dual unitary models, and random circuit models showing us that it is possible to write down minimal models for quantum chaos and generic dynamics we asked the question: is it possible to construct a model for continuous (energy preserving) dynamics, which has analytically tractable properties?

Agreed I feel like we are saying the same thing. I'm advocating / wishing more scientists would pick up c/c++.

Agreed! The problem is numpy and other libraries fairly often don't fit the workload you want or need for a given problem.

I agree python is a lovely front end when the tool you need exists.

Of course this is more nuanced than "python slow, c++ fast". In a lot of cases folks are building complex solutions in python with tools that aren't designed for their workload or problem, this is the edge case I care about. If a compiled library in python meets what you need, I'm all for it!

Oo I'm tempted! How do you get involved?