By Brian Siegelwax ithunyelwe ngomhlaka-22 Mar 2024
A few short years ago, quantum computer simulators were quite limited. On a laptop, maybe you could simulate around 10 qubits. Via the cloud, maybe you could simulate around 20. Depending on what you were running, algorithms with these low qubit counts could already take hours to process. In fact, I discovered a cloud simulator’s 10,000-second runtime limit while using only about 20 qubits. I waited 2.75 hours just to get an error message at the end.
In the years since, quantum computers have greatly improved, but so have their simulators. I haven’t tested them all, but it’s common to find claims of 30-40 qubit simulation capabilities. We have also seen the rise of emulators, which are simulators that have noise models that mimic specific types of quantum computers, or even specific quantum computers.
More recently, we have seen growth in the use of tensor networks. These classical solvers can claim to simulate more than 100 qubits. Now, here comes QPerfect, which claims their MIMIQ-Circ umndeni of simulators can handle many hundreds of qubits, perhaps up to a couple of thousand qubits. I was briefly granted access, and I used this time to test their claims.
MIMIQ-Circ, by QPerfect
The challenge of classically simulating quantum computers is that each entangled qubit we add doubles the amount of memory we need to represent the quantum system. One way to reduce the overall memory requirement is to not fully describe the system. The memory requirement still grows exponentially, but smaller numbers are being doubled. Another way to simulate more qubits is to restrict the operations that can be implemented, as is the case with a Clifford simulator, which can simulate several thousand qubits.
MIMIQ-Circ follows the first approach, using a partial state space with a full set of operations. The qubit count isn’t as high as a Clifford simulator, yet it’s much higher than other simulators.
MIMIQ-Circ is actually a small family of simulators: a statevector simulator and an MPS simulator.
Statevector Simulation
During the current trial period, QPerfect is limiting its statevector simulator to only 32 qubits and a shot limit of 216. It doesn’t actually return the statevector, which represents the state of the qubits before measurement, but that’s in the pipeline and there is a way to get it in the meantime. For now, it returns a sampling as counts, as if you’re using a QASM simulator.
What’s interesting is that I compared local installations of simulators to a cloud-hosted MIMIQ-Circ simulator. This placed MIMIQ-Circ at a distinct disadvantage because the data had to make a roundtrip through the Internet.
I tested the simulators against QPE and HHL circuits, which are some of the deepest quantum circuits you’ll find. At the smallest scales, the local implementations were faster. But as I increased the qubit counts, MIMIQ-Circ became faster even with the Internet issue.
To show you how quickly this happens with QPE, I used molecular hydrogen, which is the smallest possible molecule we can use. To make a precise calculation, we need nine total qubits. And with nine total qubits, MIMIQ-Circ over the cloud was already faster than the local simulators. With HHL, MIMIQ-Circ tied a local simulator at 15 qubits and surpassed it at 16 qubits.
MIMIQ-Circ is efficient enough that even with network latency it overtakes local simulators. Importantly, the results of MIMIQ-Cirq qualitatively match the local simulators, building confidence that it actually works.
MPS Simulation
This is the tensor network simulator that can supposedly simulate hundreds of qubits. But you can’t do that anywhere else, so I don’t have quantum circuits that large just lying around. Fortunately, it’s easy to build a massive circuit using a subroutine called the SWAP Test. So, I built a large circuit, ran it, scaled it up, and ran it again until MIMIQ-Circ finally broke.
MIMIQ-Circ processed a 1401-qubit circuit in just under 6 minutes.
Somewhere between 1401 and 1421 qubits with somewhere between 700 and 710 controlled-SWAP gates, MIMIQ-Circ finally starts returning runtime errors. That’s almost 1400 qubits more than your average quantum computer simulator can handle.
Importantly, at small scales, the results of MIMIQ-Circ qualitatively match the local simulators. Unfortunately, other simulators don’t scale up very far. However, the SWAP Test is easy to verify, and MIMIQ-Circ seems to hold up much better at large scales than other simulators do at small scales.
Local Simulation vs Network Latency
To solve the network latency problem, which is where you have to send data round trip across the Internet, QPerfect said they are working on batch jobs, variational algorithm support, and a local 20-qubit statevector simulator. From what I’ve seen, a local simulator ought to comfortably outperform other local alternatives. As a bonus, you won’t have to send your data across the Internet, which not everyone wants to do, anyway.
Isiphetho
MIMIQ-Circ ought to be able to simulate every quantum circuit that we can possibly run on every quantum computer in existence today, including the two 1000+ processors that are not publicly available. In fact, MIMIQ-Circ has two major advantages over these processors:
- There is no noise. In the absence of quantum error correction, which we don’t have in production, MIMIQ-Circ ought to be qualitatively better than the 1000+ processors.
- MIMIQ-Circ has all-to-all qubit connectivity. Although one of the 1000+ processors has the potential for all-to-all connectivity, that hasn’t been confirmed, and the other one definitely does not.
Although I focused on stress testing MIMIQ-Circ, it’s important to reiterate that its results matched up qualitatively with the results of the local simulators. At the smallest scales where other simulators can operate, it’s easy to confirm that MIMIQ-Circ works. And at a large scale, the results of the SWAP Test are promising. MIMIQ-Circ seems to be fast, accurate, and in a league of its own.
Brian N. Siegelwax ingumklami ozimele we-Quantum Algorithm kanye nombhali ozimele we Ngaphakathi kweQuantum Technology. Waziwa ngamagalelo akhe emkhakheni we-quantum computing, ikakhulukazi ekwakhiweni kwe-quantum algorithms. Uhlole izinhlaka eziningi zekhompiyutha ye-quantum, amapulatifomu, nezinsiza futhi wabelane ngemininingwane yakhe nokutholakele ngemibhalo yakhe. USiegelwax naye ungumbhali futhi ubhale izincwadi ezinjengethi “Dungeons & Qubits” kanye nethi “Khetha Owakho Uhambo LweQuantum”. Uhlala ebhala ku-Medium ngezihloko ezahlukahlukene ezihlobene ne-quantum computing. Umsebenzi wakhe uhlanganisa ukusetshenziswa okusebenzayo kwe-quantum computing, ukubuyekezwa kwemikhiqizo ye-quantum computing, nezingxoxo zemiqondo ye-quantum computing.
- I-SEO Powered Content & PR Distribution. Khuliswa Namuhla.
- I-PlatoData.Network Vertical Generative Ai. Zinike Amandla. Finyelela Lapha.
- I-PlatoAiStream. I-Web3 Intelligence. Ulwazi Lukhulisiwe. Finyelela Lapha.
- I-PlatoESG. Ikhabhoni, I-CleanTech, Amandla, Environment, Ilanga, Ukuphathwa Kwemfucuza. Finyelela Lapha.
- I-PlatoHealth. I-Biotech kanye ne-Clinical Trials Intelligence. Finyelela Lapha.
- Source: https://www.insidequantumtechnology.com/news-archive/quantum-simulator-leap-looking-at-mimiq-circ-by-qperfect/
