Scientists have developed algorithms for quantum computer simulations that can run efficiently on distributed systems. With classical computers they can simulate more qubits than originally thought.
The Google researchers ran their simulation on virtual machines from Google’s Cloud Compute platform. “Our use of modest computer hardware is democratizing quantum simulations,” they argue in their study.
Simulations of quantum computers usually take place on supercomputers, because of the complex calculations. This is mainly because of the fast interconnects that connect the systems in a compute cluster of supercomputers. The scientists’ algorithms do not require such interprocess communication, so they can be deployed on distributed clusters, such as those of cloud platforms.
One of the benefits of this is that the huge amounts of memory required for quantum simulations on supercomputers are not needed. In 2016, researchers, also from Google, predicted that simulating circuits of more than 48 high-depth qubits, of some 40 layers, would be impossible with the supercomputers of that time. Running the algorithms on a 6 × 8 qubit circuit would require 2.252 petabytes of RAM, they argued. Because of this practical limit, Google spoke of ‘quantum supremacy’: the point where classical computers are no longer sufficient in terms of simulations and quantum computers surpass their classical counterparts.
However, in the current study, the scientists ran a simulation of a grid of 7 × 8 qubits and depth of 40 layers on the Cloud Compute platform. That required ‘only’ 16TB of ram and they estimate the cost at $35,184. Their technique is scalable to more complex circuits: they estimate that a quantum circuit of 7 × 7 qubits with a depth of 48 layers costs about a million dollars.
The depth refers to the number of layers into which a circuit can be divided without overlapping gates for the qubits, i.e. how many logic operations can be implemented. The calculation that quantum circuits can perform is determined by grids of qubits in combination with that depth. Simulating large numbers of qubits with a shallow depth on classical computer architecture has been possible for some time.
The results of the simulations as proposed by the researchers are approximate, but the same applies to real quantum circuits. Unlike regular bits, qubits do not represent either the value 0 or a 1, but can take on a 0 and 1 at the same time. The number of different values increases exponentially with increasing the number of qubits. In this way, a quantum computer can perform parallel calculations in one go and immediately arrive at a definitive result, for example by checking all the results at once, leaving only the correct ones. Scientists worldwide are experimenting with this fact to develop a quantum computer that can be used for actual calculations.
Space-time volume of a quantum circuit calculation. Source: Google