Quantum Algorithms vs. Quantum-Inspired Algorithms
Quantum computing, as a field, is very inspirational: the promise of solving hard computational problems beyond the reach of regular computers feeds our hope of deploying energy-efficient solutions to logistics problems, speeding up and saving costs in material and drug discovery with more realistic ab-initio simulations, and better predicting the behavior of complex dynamical systems that affects everyone’s lives, including weather and financial markets.
This inspirational view drew a lot of attention and investment to the field, and, as it is usual, the fertile ground provides a general opportunity for growth, both for the expected and the unexpected. Among the unexpected topics that have been growing together with the quantum computing industry is the field of quantum-inspired solutions.
But what even are these? What is their relation to quantum computing? Since quantum-inspired solutions are growing in tandem with the quantum computing industry, at times competing for resources, it should do well to revisit the topic and bring some clarity to the questions above and provide perspective on expectations for end-users for now and the future.
Quantum-inspired algorithms refer usually to either of the two: (i) classical algorithms based on linear algebra methods — often methods known as tensor networks — that were developed in the recent past, or (ii) methods that attempt to use a classical computer to simulate the behavior of a quantum computer, thus making the classical machine operate algorithms that benefit from the laws of quantum mechanics that benefit real quantum computers.
On (i), while the physics community has leveraged these methods to address problems in quantum mechanics since the 70s [Penrose], tensor networks have an independent origin as far back as the 80s in neuroscience as well, as there is nothing really quantum behind them; it really is just linear algebra.
For (ii), the process of emulating a quantum system falls back on the limitations of classical hardware. It is very hard to emulate classically the full dynamics of a large quantum system for the exact same reasons that one wants to actually build a real one!
So, does this mean that quantum-inspired algorithms are bogus? Not really. These are quite novel classical algorithms, and running state-of-the-art algorithms made for state-of-the-art hardware means that real situations arise where one can get better performance for problem-solving today — in other words, running decades-old software in freshly-bought classical machines is bound to not achieve optimal execution.
This performance improvement also helps drive the friendly competition between classical and quantum methods, guaranteeing those full quantum solutions really are doing their job of challenging — and beating — classical solutions.
So, organizations should ask themselves: what do I see as my journey through quantum? If their focus is to leverage the most recent developments in computational problem solving, perhaps driven by quantum computing, to address the company’s problems today, then certainly pushing for the adoption of quantum-inspired solutions can be a fair approach.
But if one’s goal is to really discover what are the limits of classical computation in their business, and get ready for quantum disruption, then one should not settle for quantum-inspired approaches.
And even if a company’s goal is to search for solutions for today, it does not need to settle for quantum-inspired solutions. The advance of new application-specific quantum hardware — at times referred to as analog quantum computers — has released a new wave of machines in the market with real quantum coherence and scale at hundreds of qubits that have already demonstrated value in applications at their native playfield of physics.
The goal of this type of hardware is to limit the breadth of applicability, favoring performant use of quantum resources for particular use cases. While gate-based machines are universal, they are limited to a few tens of qubits that can be emulated on a laptop.
The new analog quantum computers cannot be easily emulated with classical hardware, and the jury is out there on whether the success they have demonstrated so far in areas of scientific application will be transferred to more general applications.
It is finally worth noting that we mentioned two types of quantum-inspired methods above: case (i) really does not bring anyone closer to quantum readiness, but a case can perhaps be made for situation (ii), where the methods being emulated are really closer to what a quantum computer would perform, meaning that the transfer of the quantum-inspired solution to a fully quantum solution — once the quantum hardware is on-par — would probably require a lower adaptation barrier.
Furthermore, even if, at times, these methods do not really have anything to do with the real quantum in quantum computing, these methods provide a venue for companies to really improve their solutions for problems of today, with ROI today.
This means that these methods can really contribute to bridging the development and adoption of quantum technologies through valleys of pessimism or lower investment, helping upper management keep engaged with the technology as it evolves to full maturity.
Despite the advent of quantum computing, classical computing has not stagnated. Quantum-inspired methods are but one of the new developments of classical computing that participate in the productive competition that brings both fields farther ahead. But if your aspiration is to be quantum-ready, you will probably need more than inspiration. You will need to be quantum.
