Researchers from Jülich Supercomputing Centre (Germany), Wuhan University (China), and the University of Groningen (Netherlands) announced that they broke the world record for how many qubits can be simulated on a classical supercomputer.
The team was able to simulate 46 qubits on a supercomputer, breaking the previous record of 45 qubits. They were also able to simulate 32 qubits on a laptop with 16GB of RAM.
Solidifying Quantum Simulation Limits For Supercomputers
In order for quantum computer developers to claim that their quantum computers have achieved “quantum supremacy,” first we’ll need to actually know the maximum limits of our classical supercomputers.
That limit was previously 45 qubits that could be simulated on a supercomputer, but the researchers from Jülich, Wuhan, and the University of Gronigen were able to take that one qubit further, and did it with with much less memory needed for the simulation, too.
Although the difference between 45 and 46 qubits may seem tiny, it’s important to remember that the performance required to simulate qubits grows exponentially for every extra qubit added into the mix. Normally, you need to double the memory requirements for every additional qubit being simulated, if all else is equal.
IBM was previously able to simulate 56 qubits in a more narrow type of simulation that used low-depth quantum circuits. Meanwhile, the new record is achieved for simulations of universal quantum computing circuits of any type.
Prof. Kristel Michielsen from the Jülich Supercomputing Centre explained to Tom’s Hardware the differences between the two:
"Our quantum computer simulator can simulate ANY quantum circuit with 46 qubits," said Professor Michielsen, Group leader for Quantum Information Processing at the Jülich Supercomputing Centre.
"We compute the full state vector, meaning that we compute all (2 to the power 46) amplitudes of the quantum state vector. Hence, we simulate the universal quantum computer.
IBM calculates quantum state amplitudes for measured outcomes only. IBM has presented empirical evidence that suggests that the problem of calculating quantum state amplitudes for measured outcomes may not grow to intractable levels quite as quickly as one might expect (in practice many amplitudes may be zero).
The empirical evidence derives from simulations of two universal random quantum circuits, one with depth 27 for 49 qubits and one with depth 23 for 56 qubits. The depth of a circuit is the number of layers that the circuit can be partitioned into in such a way that the gates acting on the qubits at any given layer do no overlap. It is unknown (or not yet known) what the limitations of the IBM method are in terms of numbers of qubits and depths of circuits," she added.
How The Simulation Was Achieved
Prof. Michielsen noted that there are only a few supercomputers on which even a 45-qubit simulation could have been done, because few have the required memory, computing nodes, and fast enough network connections to perform the simulation.
She added that software is also just as important to be able to run the simulation efficiently. Michielsen said that her team has been developing software that scales almost perfectly across millions of computing nodes with almost no loss in performance. Michielsen was the first person to simulate 42 qubits in 2010 on Jülich’s previous generation supercomputer and she was able to surpass her own record with a 43-qubit simulation in 2012.
Besides using a more powerful supercomputer to simulate 46-qubits, the researchers were also able to use only 2 bytes of information to represent a qubit state, down from 16 bytes required before. This is also what now allows 32-qubits to be simulated on a laptop with only 16GB of RAM.
For comparison, Rigetti, the quantum computing startup we covered recently was able to simulate only 30 qubits initially, but it has since moved to a 36-qubit simulation. Microsoft was also initially offering 30-qubit simulation, but it can now offer 40-qubit simulation. The latest Jülich discovery may allow these companies to push their number of simulated qubits even higher.
Jülich researchers also believe that being able to simulate a (relatively) high number of qubits already means that developers will be able to experiment with algorithms that will work on quantum computers the moment they reach quantum supremacy and can prove themselves to be more useful than the most powerful supercomputers for certain applications.