Photo by Brian Kostiuk – @BriKost on Unsplash, this article is from WeChat public account:Smart Things (ID:zhidxcom)< /span>, Reporter: Hearts

October 23 news, remember a month ago Google claimed the first to achieve “quantum hegemony (Quantum Supremacy)“? It is known as a major advance in the development of quantum computing, and it takes 3 minutes and 20 seconds to complete the calculation that the world’s first supercomputer takes 10,000 years to complete.

Recently, another core player of quantum computers, IBM, has not seen it. The direct public criticism of Google is misleading. There are bugs in the calculation method. Modern supercomputer does not need to take such a long time.

“We have sufficient evidence that the term ‘quantum hegemony’ has been widely misunderstood and has caused more and more confusion. We recommend that you stop using the word and hope that the community will give up using the word as soon as possible.” Written in the IBM blog.

IBM researchers say that quantum computers areIt is not the “hegemony” of traditional computers, because the experiments done in the laboratory are aimed at implementing a very specific quantum sampling procedure without practical application.

The blogger pointed out that Google has a problem in estimating the estimate that the classic supercomputer needs to calculate in 10,000 years, and IBM’s method allows the supercomputer to complete the same computing task with higher fidelity within 2.5 days. This is still a “conservative, worst-case estimate,” and other studies can further reduce time.

And Google’s 10,000-year estimate is based on the false assumption that the RAM required to run a problem in a classic computer will have a very high RAM. Google used time to make up for the lack of space, only to estimate 10,000 years.

The five IBM researchers also published a paper “Leveraging Secondary Storage to Simulate Deep 54-qubit Sycamore Circuits” to support their views.

Google has not commented on this yet.

IBM paper address: chromeextension://cbnaodkpfinfiipjblikofhlhlcickei/src/pdfviewer/web/viewer.html?file=https://arxiv.org/pdf/ 1910.09534.pdf

The following is the full text of the IBM research blog:

Quantum computers have begun to approach the limits of classical simulations. Importantly, we continue to benchmark our progress and explore their simulation difficulties. This is an interesting scientific question.

The latest advances in quantum computing have produced two 53-qubit processors: one is from usThe IBM Group’s processor, and the other is the device described in Google’s leaked paper preprint.

In preprints, some say their devices have reached “quantum hegemony” and “a state-of-the-art supercomputer will take about 10,000 years to perform the same task.”

We believe that (supercomputer) system within 2.5 days. =”Remarks”> (and Google Quantum Computer) The ideal simulation of the same task, and the fidelity is higher. In fact, this is a conservative, worst-case estimate, and we expect further improvements to further reduce the cost of the simulation.

Because the original meaning of the term “mass hegemony” proposed by John Preskill in 2012 is to describe the quantum computer can do things that traditional computers can’t do, so this threshold has not been reached.

This special concept of “quantum hegemony” is based on the implementation of a random quantum circuit whose size cannot be simulated with any available classical computer.

In particular, Google’s preprint shows a computational experiment with a 53-qubit quantum processor that implements a very large two-qubit (2-qubit) quantum circuit with 430 two qubits and 1113 single qubit gates with an estimated total fidelity of 0.2%.

The 10,000-year classic simulation estimate is based on the observation that RAM memory requirements for storing full state vectors in Schrödinger-type simulations are prohibited, so Schrödinger-Feynman simulation is needed to weigh space and time.

The concept of “quantum hegemony” shows the unique resources of quantum computers, such as directly reaching entanglement and superposition. However, classic computers have their own resources, such as storage hierarchies and high-precision calculations in hardware, various software assets, and a broad database of algorithmic knowledge, in the quantum (computer) with classic (Supercomputer) It is important to take advantage of all of these features when making comparisons.

When compared to classic computers, they rely on advanced simulations that take advantage of parallelism, fast and error-free computing, and massive aggregate RAM, but fail to adequately consider large amounts of disk storage.

In contrast, our Schrödinger-style classic simulation method uses both RAM and hard disk space to store and manipulate state vectors. The performance enhancement techniques used in our simulation methods include circuit partitioning, tensor and decrement, gate aggregation and batch processing, careful coordination of aggregated communications, and well-known optimization methods (eg cache block and double buffering) to allow communication traversal to overlap between the CPU and GPU components of the hybrid node for computation. Further details can be found in the paper “Leveraging Secondary Storage to Simulate Deep 54-qubit Sycamore Circuits”.

▲Explain the classic calculation run time and the analysis of the circuit depth of the “Google Sycamore Circuit”. The blue line estimates the classic run time of a 53 qubit processor (2.5 days for a circuit depth of 20), while the orange line estimates the runtime of a 54 qubit processor.

Our simulation methods have many nice features that don’t shift directly from the classic world to the quantum world. For example, once the classic calculations are made, the entire state vector can be accessed as many times as desired.

The running time of our simulation method is roughly proportional to the circuit depth (see above), there is no limit due to limited coherence time. New and better classic hardware for more efficient use of classic hardwareCode optimization, not to mention the use of GPU-direct communication to run some of the strongest simulations of interest, can greatly speed up our simulation speed.

Building quantum systems is a feat of science and engineering, and benchmarking them is a huge challenge. Google’s experiments showcase the progress of quantum computing based on superconductivity, demonstrating the latest door fidelity on 53-qubit devices, but it should not be seen as evidence of quantum computer “hegemony” relative to classical computers. .

As we all know in the quantum community, we are worried about the directionality of the term “quantum hegemony”. John Preskill recently discussed the origins of the term in a thoughtful article published in Quantum, including reasonable defenses and frank reflections on certain controversial dimensions.

Professor Preskill summed up two main objections to the term, saying that “this term exacerbates the coverage of the state of quantum technology that has been exaggerated,” and “through the connection with white hegemony has caused offensive political stance.” .”

The two are wise opposition. And we have to add that the term “hegemony” is almost exclusively by everyone (outside the scarce world of quantum computing experts who can put them in the appropriate context) ) misunderstood. The appearance of some variants containing “quantum hegemony has been achieved” will inevitably mislead the public.

First, because as mentioned above, according to its strictest definition, this goal has not been achieved. But fundamentally, quantum computers will never replace traditional computers, but only work with them, because each computer has its own unique advantages.

For the above reasons, and because we have enough evidence to show that the term “quantum hegemony” is widely misunderstood and causes more confusion, we urge the community to deal with quantum computers for the first time. Advocate. Because a proper metric for benchmarking is complex, a classic computer can’t question it.

In order for quantum to have a positive impact on society, the current task is to continue to build and make the more widely accessible, more powerful programmable quantum computing system. The system is capable of reproducibly and reliably implementing a wide variety of quantum presentations, algorithms, and programs. This is the only way to implement a practical solution in a quantum computer.

The last thought. The concept of quantum computing is spurring a new generation of physicists, engineers, and computer scientistsChange the pattern of information technology. If you are already at the forefront of quantum computing, stay tuned. If you are new to the field, join the community. Go ahead and run your first program on today’s real quantum computer.

The best is still behind.

——Dmitri Maslov, chief architect of IBM Q, also contributed to this article.

This article is from WeChat public account:Smart Things (ID: zhidxcom), Reporter: Hearts