Investors In Quantum Computing Should Be Very Cautious (NASDAQ:QUBT)

Quantum Computing, Inc. (NASDAQ:QUBT), hereafter presented as QCI, bills itself as “full-stack quantum software and hardware company”. Their marquee products are QAmplify and Dirac 1. QAmplify is a collection of software tools designed to simplify the development of quantum computing software. Dirac 1 is an “Entropy Quantum Computing” device developed by QCI’s recently acquired subsidiary, Qphoton. QCI uses a lot of superlatives when they discuss their products. But neither their website nor their SEC filings contain the level of detail needed to justify their claims. This is especially true for QAmplify. For Dirac 1 they do provide some interesting details. However, the level of information provided falls short of what other companies in this industry generally provide. The information that they do provide lacks the type of detail that is needed for making investment decisions.

Quantum annealing and quantum gate computing

Before discussing specifics about QCI, I want to draw a distinction between Quantum annealing computers and quantum gate computers.

Annealing algorithms are named after the annealing process in metallurgy. In metallurgy, annealing is a complex, carefully controlled heating and cooling process used to produce metals and alloys with desirable properties. Annealing algorithms are mathematical analogs of this process; they are useful for finding the minimum value of a function that has a lot of variables. Quantum annealers refer to a class of computers that use an annealing like process to solve minimization problems. They can only be used on problems that are solvable with annealing algorithms. Based on what QCI has and has not said, I suspect that they are focusing on the annealing problem.

When you hear predictions of how quantum computers will revolutionize the world, they are NOT referring to quantum annealers, they are referring to quantum gate machines (note 1). While functioning quantum annealers, like the ones that D-Wave builds, are currently available, true quantum gate machines don’t exist. But, if and when they hit the market, they will replace devices that are specifically designed for annealing (D-Wave, no doubt, would disagree with that statement).

The QCI website conflates the annealing market with the gate computing market. However, from an investor’s point of view, these are completely separate markets. Quantum annealers capable of solving real-world problems currently exist. They compete in a market that is currently dominated by massively parallel (digital) computing systems. Quantum gate computers capable of solving real-world problems don’t currently exist. But their potential for changing the world would be difficult to overstate.

QAmplify and Entropy computing

QCI likes to make grandiose claims that, in my opinion, are either not backed up with details or just fall apart upon scrutiny. Some examples follow.

• “For example, gate models can typically process from 10-120 data variables, and annealing machines can process approximately 400 variables in a simple problem set. These small problem sets restrict the size of the problems that can be solved by today’s QPUs, limiting businesses’ ability to explore the value of quantum computing. QCI’s patent-pending [I assume that they are referring to this patent, see note 2] QAmplify suite of powerful QPU-expansion software technologies overcomes these challenges, dramatically increasing the problem set size that each can process. The QAmplify gate model expansion’s demonstrated capabilities have been benchmarked at a 500% (5x) increase and the annealing expansion has been benchmarked at up to a 2,000% (20x) increase.” (Accessed on 09/14/22.)

The above statement sounds very impressive, but it is short on important details. I’ve attempted to find additional details on these claims but came up blank. What I have been able to find is that QAmplify/Qatalyst is an Application Programming Interfaces (API); in other words, it’s a set of tools and pre-programmed algorithms designed to make life easier for programmers. Software developed with the aid of the QAmplify/Qatalyst package was used to compete in a competition sponsored by BMW and AWS. For the contest, QCI used an algorithm that they have named the Variational Analog Quantum Oracle. Search of their website and on Google Scholar (searched on 09/13/22) did not provide any details on the algorithm.

They did not win the competition. But competitions like these are not good ways to test software, so normally, the fact that they did not win isn’t terribly meaningful. But given the complete lack of information on the algorithms that are contained in QAmplify, the fact that they did not win does not speak well for the claims being made.

For another example.

• “The Company achieved this landmark by applying a new quantum hardware technology called Entropy Quantum Computing (EQC) to the BMW Vehicle Sensor Placement challenge, a complex problem consisting of 3,854 variables and over 500 constraints. In comparison, today’s Noisy Intermediate Scale Quantum (NISQ) computers can process approximately 127 variables for a problem of similar complexity.”

NISQ machines are gate-based quantum computers. The excitement surrounding them is not because of what they are currently able to do, it is because of what they theoretically will be able to do in the future. At the present time, these machines are not even able to outperform a mediocre laptop. Outperforming a NISQ processor is not something to brag about.

When it comes to providing details about what they are doing, their article on the BMW challenge problem is about as good as it gets. So that’s the one that I’m going to pick on.

In the article, they refer to something called “Entropy Quantum Computing” (EQC). Entropy Quantum Computing is a term that they invented, which is fine. But when you invent a term you need to explain it – they don’t.

QCI does not provide a lot of details on their EQC device. They do however state that “the EQC carefully engineers the environment and its backaction through the Hamiltonian to relax the system to the ground state“, which is basically what a quantum annealer does. Combining that with the fact that quantum annealers reduce entropy, leads me to suspect that the EQC is some sort of quantum annealer (note 3).

QCI uses their EQC prototype to solve one of the BMW challenge problems. The EQC solution is compared to other solutions:

	EQC	QCI’s D-Wave Solution	QCI’s Digital Solution
Coverage	96%	99.6%	62.8%
Number of sensors	15	373	15
Runtime (sec)	363	26373 (≈ 7.3hours)	197

Click to enlarge

Bold means the numbers are bad.

Note that the EQC solution is compared to their D-Wave solution, and not the actual solution that won the BMW challenge. As for QCI’s digital solution, did they use a state-of-the-art algorithm running on a high-performance computing system, or did they use a crude brute force algorithm running on an ancient “computer cluster” in their comparison? These issues, however, are not my main objections to this press release.

The sole reason for pursuing quantum computing is for their theoretical ability to “scale”. By scale, I mean what happens when you increase the number of variables. So for example, in the above benchmark, the simulation used 3,854 variables. If you increase the number of variables to 7,700 will the EQC processor runtime grow linearly to around 800 seconds or will it grow exponentially to 800 hours.

Now let’s assume that the runtime for the EQC processor grows exponentially to 800 hours, while the D-Wave solution grows linearly (in real life, D-Wave scales better than a digital computer, but not linearly) to 15 hours. In this case, since the D-Wave processor has superior scaling, it would be considered to be the superior computer. For problems that are difficult to solve, scaling is the most important factor in determining which processor/algorithm is the superior one.

Bottom line, without scaling information, there is no way to determine how EQC fares against the competition.

Bending the truth to make your product appear to be revolutionary is easy to do. That is why high-quality companies in this industry (and in other high-tech industries) will submit their work to peer-reviewed journals. The peer review process means that a neutral expert on the subject reviews the work to determine if the claims being made are reasonable. The importance of this cannot be overstated. Without peer review of some form or another, there is absolutely no way for a non-expert in the field to determine if there is any validity to the claims being made.

As an example of what, in my opinion, legitimate quantum computing companies look like, take a look at the peer-reviewed and waiting to be peer-reviewed publications of Cambridge Quantum, IonQ, Rigetti, and D-Wave. QCI appears to have completely avoided the peer review process.

Wrap Up

Qamplify and EQC processors may ultimately turn out to be revolutionary products, but for now, the company has not presented any evidence that their exuberance is justified. What they have presented are web pages, press releases, and Youtube videos that are very professionally done and extol the virtues of their products, but don’t provide sufficient information to make an investment decision. This is a stock to be completely avoided.

Note 1. Companies such as IBM, Honeywell, IONQ, Rigetti, Google, etc. have developed primitive quantum gate computers. IBM, Honeywell, IONQ, and Rigetti are integrating their quantum gate machines with high-performance digital computers. The goal is to produce a synergy between the two technologies.

Note 2. The patent talks about breaking up large problems into smaller problems that are readily solvable and then using those solutions to find the solution to the larger problem. This is a very basic computing concept that has long been in use in massively parallel high-performance (digital) computing systems and is also in use in hybrid quantum-digital systems. Methods for achieving these types of decomposition are a major area of research. The patent doesn’t say anything that is even remotely original, and it completely ignores real-world issues that are encountered when performing such decompositions. What this patent is trying to do is claim ownership of any decomposition that is done on a quantum computing system.

Based on the discussion in the patent, I am assuming that QAmplify has some sort of algorithm for decomposing a large problem into several smaller problems. These types of decompositions are not trivial. In order to determine how competitive QAmplify is in this area, some specific knowledge of QAmplify’s algorithm would be needed. That information is not provided in the patent.

Note 3. The QCI corporate website does not contain any useful information on their EQC computer, and the Qphoton website is password protected (accessed on 09/14/22). They do however provide enough information to justify the use of the label “quantum” for their device