In the last article (check here), we met a strange situation: running an experiment was easier than simulating it. Boson sampling gave us a glimpse of how quantum systems can overwhelm even our biggest classical supercomputers.

Now let’s take that idea one step further (please read the last episode, if you haven’t already - here).

Suppose you build a quantum computer. How do you actually test it?

How do you verify that the results it gives you are correct…
when no classical computer can tell you what the right answer should be?

Welcome to the weird world of quantum benchmarking.

With a classical computer, benchmarking is boringly simple:

Done.

But with quantum computers, especially when you get beyond 50–100 qubits or start doing complex sampling experiments, this becomes impossible. The “state space” grows so fast that even the biggest supercomputers (TOP-500) can’t compute the full answer to compare with.

This means that for a real quantum device:

That’s the paradox. And the solution, surprisingly, circles us back to last week’s hero:
random sampling problems.

Let’s revisit the boson sampling idea for a moment. You send photons into a network of beam splitters, let the whole system evolve, and measure where the photons come out. An example case is shown below, where a photon entering input 3 emerges from output 8. You repeat many times and get a distribution.

Here’s the key insight:

You can test properties of the distribution instead.

Imagine testing a dice factory. You don’t need to predict each exact roll. You only need to check whether the rolls “look” like they came from a fair die. Quantum benchmarking works in a similar way.

Because we cannot precisely simulate a full quantum device, researchers do something counterintuitive:

The Boson sampling is one such random sampling problem.

Why randomness?

We might not know every detail of the quantum output, but we do know certain features it must have: the shape, spacing, correlations, collision rates, heavy-output tails, etc.

If these statistical properties match theory, we gain confidence that:

This is the essence of quantum benchmarking.

But it’s still incredibly hard. Even with randomness, benchmarking quantum devices is not trivial. Not even close. There are several challenges:

Classical computers keep getting faster and smarter. Every time someone claims quantum advantage, another group finds a new shortcut or approximation that pushes the goalpost.

Quantum systems are notoriously error-prone. If the output distribution is “wrong,” is that because the machine is faulty? Or because the classical approximation we’re comparing to is inaccurate?

Some companies might claim they achieved quantum advantage with special hardware.
But verifying that claim requires extremely careful statistical tests — and a lot of data.

A 50-qubit random circuit might be checkable with heroic supercomputers. A 200-qubit one? Completely out of reach.

This is why quantum benchmarking is its own mini-research field.

Because boson sampling is easy to implement physically but hard to compute classically, and it is innately connected to photonic systems, it becomes a kind of stress test for quantum photonic devices.

Companies use it to:

Even if boson sampling itself isn’t “useful” in the practical sense, it is extraordinarily useful as a benchmarking tool. It forces the hardware to operate in a regime where quantum effects dominate and classical simulation fails — exactly the regime companies care about.

Quantum technologies are often expected to solve problems that classical methods cannot. But the first step for any new technology is to verify that it works as intended. Without verification, a device might appear to solve a problem, yet we would not know whether the solution is correct. This creates a paradox: if quantum computers are meant to outperform classical computers, how can classical computers be used to check them?

This is where the newest high-performance classical chips, such as those from Nvidia, become important. They make it possible to test certain special cases. Beyond that, we can use random sampling problems to gain statistical evidence.

These random sampling tasks allow us to say, with high confidence:

This is why companies are investing billions in the field. They are not only building quantum computers, but also building reliable ways to verify them.

For more updates on the science behind deep tech - stay tuned.

Some administrative details:

Some of you have mentioned issues with emails not displaying correctly or figures not loading. If that happens, you can always click “Read online” in the top-right corner of the email to view the post in your browser. You can also visit the homepage https://qubit-and-neuron.beehiiv.com/ at anytime to read all the articles.

Thank you all for your feedback and for letting me know about these issues. Please contact at: [email protected]