Earlier this month, the news that Alibaba Quantum Lab successfully developed the world’s most powerful quantum circuit simulator, “Taizhang”, caused a lot of discussion, and there was a lot of talk about the battle for quantum hegemony. Based on the online clustering power of Alibaba Group’s computing platform, “Tai Zhang” is the first in the world to successfully simulate the benchmark Google Random quantum circuit at the 81 (9×9) bit 40 layer, according to the official information, while previous simulators could only handle 49 bits.
As for the “Tai Zhang” itself, there have been many media reports before, so we will not repeat them. So what is the strongest quantum circuit simulator? Cloud Community interviewed Dr. Shi Yaoyun and Dr. Chen Jianxin of Alibaba Quantum Laboratory team to make popular science interpretation for everyone.
Chen Jianxin, a core member of the research achievement, said that the final form of quantum computing should be based on quantum hardware, but the circuit simulation has its important significance, when the hardware is not enough scale and quality, can be used as a tool to verify the quantum algorithm, auxiliary algorithm and hardware design.
In the field of quantum computing, there are two mainstream simulation schemes in the industry at present: one is to store all the amplitudes of the quantum state; the other is to quickly calculate the results for any amplitudes. Alibaba quantum laboratory USES is the second category of simulation program, namely in the alibaba online cluster computing platform, through the calculation of arbitrary amplitude, quickly and effectively after task subtasks can be very balanced allocation to different nodes, rarely communication overhead makes the simulator adapter now widely available services of cloud computing platform.
No research team around the world has successfully simulated Google’s first-generation random test circuit with more than 50 bits and 40 layers. The biggest breakthrough of “Taizhang” is that alibaba Quantum Lab team parallelizes several layers at the same time. “We used the second type of simulation. Usually splitting circuit model requires parallelization of each layer, but this will lead to the exponential increase of the number of parallel tasks when the whole circuit has a large number of layers. By parallelizing several layers at the same time, we can effectively control the increase of parallel tasks.” Chen jianxin said.
In addition, he also pointed out that “Tai Zhang” also explained to a certain extent that Google proposed in March this year 72 bits 40 layers of the random circuit sampling task can not exceed classical computer capabilities, in other words, Google Bristlecone chip is currently unable to achieve quantum hegemony.
The following is a transcript of an interview with Chen Jianxin, a scientist at Alibaba Quantum Laboratory, and Shi Yaoyun, chief scientist of quantum computing.
Note: Image source network
Cloud community: can you introduce the quantum circuit simulator in a popular way? What is its role in full stack quantum computing? Chen Jianxin: Among the current models of quantum computing, there is a quantum circuit model, which stores information in quantum bits and performs calculations through quantum gates similar to classical logic gates. The final realization form of quantum computing should be based on quantum hardware, but circuit simulation has its important significance. When the hardware cannot reach enough scale and quality, it can be used as a tool to verify the quantum algorithm, auxiliary algorithm and hardware design. After hardware breakthroughs can be achieved on a sufficient scale, quantum hardware resources can be expected to remain expensive, and it is also necessary to check whether classical computer simulations meet expectations before implementing quantum programs. Cloud community: Why is the newly developed 81 bit quantum circuit simulator named Taizhang? Chapter Chen Jianxin: “too” from: “huai south son” delivery form training “, “yu is the step too chapter from east as to the west, seventy-five steps in two hundred million and thirty-three thousand five hundred; the vertical step hai from the North Pole as to the South Pole, two hundred million and thirty-three thousand five hundred in seventy-five. All that old hong water collection since three hundred towering above two hundred million and thirty-three thousand five hundred and fifty, nine miles deep yu, is filled with interest the floods that famous mountains.” The taizu simulator aims to understand quantum motion in a classical, understandable way, similar to how Taizu measured the distance from the east pole to the west pole on foot. Cloud community: 81 bits 40 layers, what do these two data mean? Why 81 bits and 40 layers? Chen Jianxin: The purpose of this simulator is to study the application layer in the future. The 81 bits and 40 layers are only proposed here for comparison with the 72 bits proposed by Google in March this year. The random circuit sampling scheme proposed by Google in 2016 is difficult to break through 40 layers of hardware when it is larger than 50 bits, so simulating 81 bits 40 layers means that the sampling task of 72 bits 40 layers of random circuit proposed by Google in March this year cannot exceed classical computer capabilities.
The corresponding relation between the execution time of each amplitude output of the random circuit and the circuit depth is calculated on the NXN two-dimensional grid
Cloud community: in addition, the laboratory has successfully simulated the 100 bit 35 layer, 121 bit 31 layer, 144 layer 27 layer? What is the logical relationship between bits and layers? Chen Jianxin: The more bits you have, the less layers you can simulate. The number of layers that can be simulated is approximately inversely proportional to the number of bits under the same computation force. Cloud community: According to media reports, the previous 40 layer emulator can only handle 49 bits? What breakthroughs did “Tai Zhang” make? Chen Jianxin: At present, there are two types of mainstream simulation schemes in the industry. One is to store all the amplitudes of the quantum state, and the other is to quickly calculate the results for any amplitudes. The first type of simulation scheme is basically implemented on supercomputers, because it requires Petabyte of memory to store 45 bits of quantum state, and it requires constant exchange of data between different computing nodes to operate and calculate the quantum state while storing so much data. Such communication costs are prohibitively high for normal cloud services.
The scale of the random quantum circuit simulated by “Tai Zhang” (black line) is compared to the scale that Google quantum hardware can achieve (red line) (based on Google’s estimate of 7×7 in [Characterizing Quantum Supremacy in Near-term Devices] *
We use the second type of simulation scheme. Usually splitting circuit model requires parallelization of each layer, but this will lead to the exponential increase of the number of parallel tasks when the whole circuit has a large number of layers. We parallelize several layers at the same time, which can effectively control the increase of the number of parallel tasks. Cloud community: The team only used 14% of the computing resources of the online cluster of Alibaba computing platform to achieve tasks that could not be completed by supercomputers before. What is the difficulty of this? How did you overcome it? Chen Jianxin: Computing resources span several machine rooms, which means that inter-process overhead across machine rooms is extremely expensive. Therefore, we focused on minimizing communication overhead and adapting the emulator to the cloud computing platform that is now widely available. Cloud community: What is the significance of this achievement in quantum circuit simulator for Ali’s research on quantum computing? Chen Jianxin: How to carry out the preliminary layout without hardware is the original intention of making this quantum circuit simulator. Based on this circuit simulator, we can further assist algorithm design, application development and testing. Cloud Community: What is the next plan of Alibaba Quantum Laboratory? ShiYaoYun: A central mission has been to develop quantum algorithms that are of significant value to our business. To identify such Problems, we hope very much our technology community can help (core task is to research and develop quantum algorithms of value to Alibaba business). The Cloud community: How close is it to quantum hegemony? Chen Jianxi: The quantum hegemony scheme proposed by Google in 2016 is a good idea, but our results show that the previous random-circuit sampling scheme, no matter how large the number of bits, will not be able to surpass classical computers. But we also see that classical simulations get harder as they get bigger. After all, we use a lot of classical computing resources, in other words, quantum hardware is already better at this particular task, if only to compare it with a normal desktop computer. With the development of hardware technology, I believe that quantum hegemony will not be too far away.
The original link