Grid of atoms is both a quantum computer and an optimization solver | Hacker News Reader

Grid of atoms is both a quantum computer and an optimization solver

arstechnica.com

58 pointsisaacfrond15 comments

If we want to simulate molecular systems that have quantum effects, with a quantum computer, is it a good fit?

Say we want to create designer molecules for flow batteries. A few benzene rings and some "arms" with a few carbons and hydroxyl group at the end. The potential design space is absolutely humongous. Can a quantum computer simulate something like that relatively directly, and then see how that reacts in various situations, oxidizing and reducing? If you have a system with 20 atoms, how many qubits would you need, and so on?

Right now researchers make a best guess, then synthesize the material. That can take months. Then they make experiments, and usually there are some surprises. Machine learning has been proposed to help but probably has limitations...

Little known fact: Simulating quantum physical systems was the original motivation for building quantum computers, because those simulations are infeasible without one, and someone figured out it would be, in principle, feasible to run those simulations using a computer based on quantum effects.

There's a neat symmetry to it. We can't be certain quantum computers work at useful scales, until we've built one, because there's no feasible way to simulate the physics of a useful size quantum computer without using one.

> there's no feasible way to simulate the physics of a useful size quantum computer without using one.

What do the diverging growth curves look like? We probably could brute force simulate the exponential states within a small-n using conventional computing, just by paying for more parallelism and time. At some point that doesn't scale (more computers needed than atoms on earth, for longer than the heat-death of the universe or whatever) but surely there are interesting problems we can model before whatever inflection point exists? If so - what kind of scale is that?

In theory yes, but we can't get one with enough qbits (you would need more qbits than your molecule because you need to encode positions and interactions of electrons at the very least), enough configurability (current q-computers are somewhat limited to which qbit can interact with which through which mean) and that is able to keep decoherence away (which limits how much it can scale).

It seems that we are really hitting a wall every time we want to scale them to a useful size and keep the number of possible interactions between qbits high and noise/decoherence level low.

So for now we don't have the technology (if ever), it is not just a problem of "just scaling by adding more qbits" we need a new approach.

So you can make a grid with less than 40 quantum atom pairs, each of which is coupled in some way to another atom, and since measuring one atom guarantees the other atom is in the opposite state, you can bypass Heisenberg uncertainty.

But wait. To solve an NP problem of size n, you need square(n) atoms. But 40 atoms is barely more than 36, so you can solve a NP problem of size 6, which is easy and possible.

I think brute force on an NP problem is factorial(n) So factorial(6) == 6 ! == 6 * 5 * 4 * 3 * 2 * 1 == 720.

Am I misunderstanding something here ?

The other atom is not in the 'opposite state' for all properties, rather it has one property which is entangled so whatever accuracy you just measured, you know to the same accuracy of the other in the entangled property only.

Also 40 bits in superposition can represent 2^40 possibilities.

> 40 bits in superposition can represent 2^40 possibilities.

In classical QC yes, is this the case any longer in the Rydberg blockade configuration? From what I understood from the article, it's what GP said: the atom configuration implements the independent set problem using entanglement. So the number of qubits is the order of the graph, which is the size of the problem. I'm not sure where the quadratic bound is derived from, and I didn't understand where the solution to the maximum independent set comes from so maybe I didn't understand the article...

There are so many other gray areas in the article that reading it feels a bit like riding on the hype train.

> I think brute force on an NP problem is factorial(n)

No, NP ⊆ EXPTIME, which means that exponential time should be enough.

Still find it nuts you need to operate the system at 0 kelvin. I assume if we had super efficient cooling equipment (not LN2) we could also do superconductors on a small scale?

LN2 is not enough you need LHe for most superconductors. It is usually cooled by LHe that is itself shieled by LN2 to limit losses.

LN2 is relatively cheap not LHe.

That's why there are recylcing systems to recover the boiling off He in MRIs, NMRs and other supraconducting systems.

There are systems that allow cooling without wasting LHe/LN2 but they are pretty limited in size or costly:

https://hexus.net/tech/news/cooling/131414-worlds-first-abso...

https://www.wired.co.uk/article/mri-magnet-cooling

https://en.wikipedia.org/wiki/Dilution_refrigerator

edit: s/LH2/LHe

Liquid He right? Liquid hydrogen is relatively cheap

Thanks will edit

You use lasers to optically trap and cool the atoms, you can get them insanely cold very reliably with methods like this:

https://en.wikipedia.org/wiki/Optical_molasses

https://en.wikipedia.org/wiki/Polarization_gradient_cooling

https://en.wikipedia.org/wiki/Doppler_cooling

The components in a system like this are probably operating at room temperature.

> Still find it nuts you need to operate the system at 0 kelvin. I assume if we had super efficient cooling equipment (not LN2) we could also do superconductors on a small scale?

Neutral atom quantum computers (the type outlined in the article) work at room temperature.

Superconducting qubit based quantum computers do indeed require cooling to these levels, though.

Smells like d-wave nonsense.

‹finishes reading›

Yeah, it's the same nonsense.

Reminder: quantum computing is not expected to be able to improve the solutions of NP-complete problems by more than a quadratic factor, unless some commonly accepted conjectures can be proven false. If that happens, you'll be hearing about it from more than just a analog computer company's marketing materials.

Reply on news.ycombinator.com