Shulou(Shulou.com)11/24 Report--

Today, IBM verifies 100 + qubits for the first time, and without error correction, it can still achieve accurate results, even surpassing classical computers. The latest research is on the cover of Nature.

Today's Nature cover belongs to IBM.

New research from IBM and the University of California, Berkeley shows "a path to useful quantum computing".

It is proved for the first time that a 100 + qubit quantum processor can achieve accurate results and surpass the leading classical methods.

Most importantly, classical computers can be surpassed without error correction.

The CEO of IBM said that this is proof of a milestone and marks the practical application of quantum computing!

Four years ago, Google claimed that its own quantum computers had achieved "quantum hegemony", but only made a breakthrough in niche computing without practical application.

According to the latest Nature paper, quantum computers will soon beat ordinary computers in useful tasks.

In https://www.nature.com/ articles / s41586,023-06096-3, the researchers simulated the behavior of magnetic materials on an IBM 127qubit eagle (Eagle) quantum processor.

Crucially, they managed to bypass the "quantum noise" and achieved reliable results. You know, quantum noise will introduce computational errors, which is the main obstacle to this technology.

Knocking on the door of "Quantum advantage" has always been a key milestone in quantum computing. IBM defines quantum advantage as a significant improvement in the running time of quantum algorithms in real cases.

So far, there are no useful applications to prove the quantum advantage, and the reason is simple:

Quantum computers are noisy, error-prone, and too small to solve big problems in the real world.

However, most papers on quantum advantage are usually based on random circuit sampling or Gaussian boson sampling, both of which are not useful applications.

In IBM's view, quantum computers must solve three main problems in order to perform useful tasks:

A method of dealing with quantum noise is needed.

Qubits must be scalable to a large number of numbers.

-Quantum processors must have sufficient speed (measured in circuit layer operations per second or CLOPS).

Among them, there is a direct relationship between quantum computing noise and the scale of the problem that can be solved.

Loud noise will lead to errors, and uncorrected errors will limit the number of qubits added to the circuit, which in turn limits the complexity of the algorithm.

Obviously, the "error control" of quantum computing is very important, that is, quantum error correction (QEC) is needed.

On the other hand, IBM goes in the opposite direction, surpasses the classical computer without error correction, and even achieves the unique usefulness of 100 + quantum ratio.

In this paper, the researchers turned to a method to intentionally amplify the noise and then measure different levels of processor noise.

In response, physicists accurately measured the noise in each qubit.

The researchers simulated 127 interacting spin states using a 127qubit Eagle R3 processor.

In the simulation, each qubit acts as a spin, using two qubits gates with a depth of 60.

Experiments show that they can run 127 qubits and up to 60 processing steps involving all Eagles, more than any other reported quantum computing experiment.

These results validate IBM's short-term strategy to provide useful calculations by mitigating errors rather than correcting them.

The researchers used "error mitigation" technology to enable the team to perform quantum computing on a "scale that cannot be achieved by classical computers."

Santa Barbara, a physicist at the University of California, Santa Barbara, who led the Google team to a 2019 milestone, said

Although the problem they overcome uses a simplified and unrealistic material model, it makes people more optimistic about the future that it will apply to other systems and more complex algorithms.

Abhinav Kandala, manager of the quantum capabilities and demonstration department of IBM, who is not afraid of noise, says the key part is to be able to control noise outside pulse broadening (pulse stretching).

"once we start working, we can make more complex inferences to suppress the deviation caused by the noise. This could not have been achieved before. "

This noise amplification is the last piece of jigsaw puzzle needed by IBM.

With a representative noise model, people can control and amplify the noise more accurately. Then, the classical post-processing method can be used to infer the calculation results without noise, which is called "zero noise inference" (Zero Noise Extrapolation,ZNE).

At the same time, error mitigation (Error mitigation) requires high performance hardware. IBM must continue to advance in terms of scale, quality and speed.

With the 127bit IBM Quantum Eagle processor, IBM finally has a system that can run circuits large enough.

Now it's time to use ZNE to test IBM SOTA-level processors.

Of course, "error mitigation" technologies like ZNE are not a panacea.

To realize the full potential of quantum computing, IBM needs to establish redundancy in the system and allow multiple qubits to work together to correct each other, that is, error correction.

However, through error mitigation, IBM realized that before the era of full error correction, it was possible to find a way to generate some kinds of accurate calculations, even for noisy quantum computers.

And these calculations may come in handy.

IBM just needs to test whether their error mitigation techniques are really effective.

First, the researchers tried to run increasingly complex quantum computing on IMB's cloud service, and then compared it with traditional computers.

At the same time, IBM also needs some external experts to verify the correctness of these calculations. So they used the help of researchers Sajant Anand, Yantao Wu and Michael Zaletel, an associate professor at the University of California, Berkeley.

There are several ways to run quantum circuits with classical computers.

The first is to rely on brute force to calculate the expected value, similar to the method in which physics students calculate the expected value by hand. This requires first writing all the information about the wave function into a list, and then creating a matrix to calculate.

With each additional qubit, the difficulty of these methods doubles, so researchers end up unable to capture the complexity of circuits large enough.

But for a subset of quantum circuits, there are techniques that allow researchers to use rough calculations to get accurate answers, even if the circuit uses all 127qubits of the IBM Quantum Eagle.

Starting from these circuits and methods, IBM conducts corresponding benchmark tests for classical and quantum methods respectively.

To deal with more complex circuits, the Berkeley team used two different tensor network state (TNS) methods to approximate wave functions with fewer numbers.

This classical approximation method attempts to express the quantum state of many qubits as a network of tensors. TNS comes with a set of instructions on how to use this data for calculations and how to use it and recover specific information about the quantum state, such as expected values, and so on.

This method is a bit like image compression, in the case of limited computing power and space, in order to retain only the information needed to get an accurate answer, get rid of less important information.

The experiment will be carried out as follows: IBM will use all 127qubits of the Quantum Eagle processor to simulate the changing behavior of a system, which naturally maps to a quantum computer, called the Quantum Ising Model (Ising model).

The Ising model is a simplification of nature, which represents the interacting atoms as the lattice of a quantum system in an energy field.

IBM will use ZNE to try and accurately calculate an attribute of the system, that is, average magnetization. The expected value is basically the weighted average of the possible results of the circuit.

Meanwhile, the team at the University of California, Berkeley, with the help of the National Center for Scientific Computing for Energy Research (NERSC) at the Lawrence Berkeley National Laboratory and the advanced supercomputer at Purdue University, will try to simulate the same system using tensor network methods.

Specifically, IBM's calculations will run partly on NERSC's "Cori" supercomputer, partly on Lawrence Berkeley National Laboratory's internal "Lawrencium" cluster, and partly on Purdue University's "Anvil" supercomputer funded by the National Science Foundation.

IBM will then compare the two with precise methods to see how they perform.

From the results, the quantum method is consistent with the exact method. However, with the increase of difficulty, the classical approximation method began to go wrong.

Finally, IBM asked the two computers to run more calculations than could be calculated accurately, and they were confident about the results.

Bring useful quantum computing to the world recently, there is a debate about whether quantum computers can provide computational advantages for useful tasks before full error correction.

Fault tolerance is the ultimate goal, and "error mitigation" is the way to make quantum computing useful.

IBM's latest paper shows that noisy quantum computers can also provide value.

The key to this work is that all 127 qubits of Eagle can now be used to run a fairly deep circuit, and the numbers are correct.

This paper is a node, indicating that IBM is entering the era of quantum dominance. They have said before that quantum advantage is a continuous path that requires two things:

First of all, we must prove that quantum computers can surpass classical computers. Second, we must find the useful problems of acceleration and figure out how to map them to qubits.

The first point has been achieved in the latest paper. This is an important moment for the quantum field.

The IBM step is just the starting point, and netizens say they will achieve accurate results at 1000 + qubits by the end of this year.

Reference:

Https://www.nature.com/articles/d41586-023-01965-3

Https://research.ibm.com/blog/utility-toward-useful-quantum

Https://www.nytimes.com/2023/06/14/science/ibm-quantum-computing.html

This article comes from the official account of Wechat: Xin Zhiyuan (ID:AI_era)

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.