OTI Lumionics is raising the bar…
...and challenging assumptions about the limits.
You may have read “OTI Lumionics Establishes New Computational Chemistry Benchmark, Outperforming Traditional Quantum Models” and wondered what the big deal is. After all, it’s about running high-precision molecular simulations with the quantum-inspired Iterative Qubit Coupled Cluster (iQCC) algorithm on a single NVIDIA Blackwell gaming GPU, not on a quantum computer. It is therefore demonstrating improvements in efficiency, speed, and accuracy over supercomputing clusters without quantum, which is not exactly what any of us have come to the table for. As the case in point, this ain’t The Quantum-Inspired Dragon, folks.
Challenging Assumptions
We assume that quantum computers will be useful someday, and OTI Lumionics is challenging that assumption. The approach reminds me of DARPA’s Quantum Benchmarking (QB) program, which sought to determine whether or not there are actual use cases for quantum computers. Every approach has tradeoffs, in this case RAM for computer time, using memory compression, and the goal is to determine what the output would be on a real quantum computer if we could actually do so. The results can be examined, and the approaches can be improved.
The approach in question just raised the bar from 72 qubits to 200. In other words, we would need a 200-qubit fault-tolerant quantum computer to do what they did. We just checked the back of The Quantum Dragon’s cave, and we apparently don’t have one of those yet. The RAM requirement is N^5, so quantum computers, in principle, should surpass this approach at some point, but we’re not there yet. And in the meantime, OTI Lumionics looks poised to raise the bar further.
Steroids
The peer-reviewed approach took the Python training wheels off and translated the code to Julia. This version was translated again into C++ for supercomputing clusters. Unlike tensor networks, this approach doesn’t ignore phase, look only at nearby qubits, or take any other shortcuts; it is a full simulation. With 1TB RAM, it can simulate 200 qubits and requires only one pass with no pre-processing. And it still needs to be fully optimized for NVIDIA GPUs, which is how we suspect that the bar can still be raised further. What we’ve got here is chemists and software engineers working together to squeeze out practical value.
Limitations
With N^5 RAM requirements, we’ll eventually run out of memory to run the iQCC algorithm. However, at that point, OTI Lumionics will have left behind a gift. Once we have large-scale, fault-tolerant quantum computers, we’ll know what we’ll be able to extract from them and, therefore, we’ll know what we’ll want to do with them.
iQCC lacks some of the limitations of its variational alternatives. For example, its ansatz scales linearly with respect to the Hamiltonian. There’s no simplification, increased sampling ratios, exponential iterations, or exponential pre-preparation. There’s no sampling at all, actually; it computes directly.
Conclusion
OTI Lumionics is actually a chemical company, not a quantum company. Its interest is that this stuff needs to work, otherwise there’s no justification for it. Playtime is over; we need to be using realistic approaches and migrating away from Python.
While iQCC can be run on a quantum computer, the point is that can be simulated with up to 200 qubits on classical computers. It is raising the bar for quantum computers and, in doing so, it is making possible calculations that chemists and dragons alike actually care about.
