Quantum Computers Can Do Useful Work Despite Noise

Quantum Computers Can Do Useful Work Despite Noise

The assumption was always that quantum computers would need to be pristine. Perfectly isolated. Absolutely silent. A single stray photon, a wandering atom, a whisper of heat, and the whole computation would collapse into a pile of meaningless probabilities. For years, this was the dogma: quantum computers were beautiful, fragile machines that could only work if you kept them in a state of near-perfect isolation.

But here is the thing about dogma. It is often wrong.

In 2022, a team of physicists led by Kishor Bharti, Alba Cervera-Lierta, Thi Ha Kyaw, and Tobias Haug published a comprehensive review in Reviews of Modern Physics that quietly upended this assumption (Bharti et al., 2022). Their argument, backed by years of experimental and theoretical work, is surprisingly simple: noisy quantum computers can already do useful things. Not someday. Not after we solve the error correction problem. Right now.

This is not a claim about future potential. It is a claim about the present.

The Noise Isn't Going Away

Let me be clear about what noise means in a quantum computer. It is not like static on a radio. It is more fundamental. A quantum computer stores information in qubits, which are not just 0s or 1s but exist in superpositions of both states simultaneously. This is what gives them their power. But it also makes them exquisitely sensitive. A qubit can be disrupted by a cosmic ray, a temperature fluctuation, a manufacturing defect in the chip, or even the act of measuring it.

In a classical computer, you can check your work. If a bit flips from 0 to 1, you detect it and fix it. In a quantum computer, you cannot do that without destroying the superposition you are trying to preserve. So for years, the conventional wisdom was that you needed fault tolerant quantum computing: thousands of physical qubits working together to create one logical qubit that could survive long enough to do something useful.

The catch is that fault tolerance requires extremely low noise levels. Lower than any current hardware achieves. Lower than most experts expect within the next decade.

Bharti and colleagues looked at this problem from a different angle. What if you do not need fault tolerance? What if you can work with the noise instead of fighting it?

The NISQ Era Is Not a Waiting Room

The term "noisy intermediate scale quantum" or NISQ was coined by John Preskill in 2018. It refers to quantum computers with 50 to a few hundred qubits that are too noisy for full fault tolerance but too powerful to simulate classically. Many researchers treated NISQ as a transitional phase, a necessary but unglamorous stepping stone to the real thing.

Bharti et al. argue this is a mistake. They show that NISQ devices are not just prototypes. They are a new computational paradigm with their own strengths and weaknesses. The key insight is that you can design algorithms that are inherently resistant to noise, or that even use noise as a resource.

Consider the variational quantum eigensolver, or VQE. This is a hybrid algorithm that splits the work between a quantum computer and a classical computer. The quantum part prepares a trial state and measures its energy. The classical part optimizes the parameters to find the lowest energy state. Because the quantum part only runs briefly before offloading to the classical computer, it never needs to maintain coherence for long. The noise does not have time to accumulate to destructive levels.

Bharti et al. review dozens of experiments using VQE to simulate molecular structures, from hydrogen molecules to more complex compounds (Bharti et al., 2022). The results are not perfect. They are not high precision. But they are good enough to predict chemical properties that classical computers cannot calculate exactly.

This is the first surprise: noise does not have to be fatal if you design the algorithm to work in short bursts.

What Actually Works Right Now

The authors systematically catalog the tasks where NISQ devices have already demonstrated capability. The list is more impressive than you might expect.

Simulating Quantum Systems

This is the most natural application. Quantum computers are good at simulating quantum physics because they play by the same rules. Bharti et al. describe experiments where noisy quantum processors have simulated the behavior of spin chains, lattice gauge theories, and topological phases of matter (Bharti et al., 2022). These simulations are not perfectly accurate, but they capture the essential physics. For many purposes, that is enough.

The authors note that even a noisy simulation can reveal phase transitions, critical points, and other qualitative features that are invisible to classical methods. It is like looking through a dirty window. You cannot see every detail, but you can still see the shape of the building.

Quantum Chemistry

This is where the practical payoff is most immediate. Classical computers struggle to model molecules because the number of possible electron interactions grows exponentially with the number of atoms. Quantum computers, even noisy ones, can handle this complexity more naturally.

Bharti et al. review experiments that have calculated the ground state energies of small molecules like lithium hydride and beryllium hydride (Bharti et al., 2022). The results are within a few percent of exact values. For drug discovery or materials design, this level of accuracy is already useful for screening candidates. You do not need a perfect answer. You need an answer that is better than the alternatives.

Combinatorial Optimization

This is the most counterintuitive application. Optimization problems, like finding the shortest delivery route or the optimal portfolio allocation, are hard for classical computers because the search space is enormous. Quantum computers might help, but noise seems like it would ruin the optimization.

Bharti et al. describe a different approach. Instead of trying to find the single best answer, NISQ algorithms sample from a distribution of good answers. The noise broadens the distribution, but it also prevents the algorithm from getting stuck in local optima. In some cases, the noisy quantum computer finds better solutions than a perfect quantum computer would, because the noise provides a form of exploration (Bharti et al., 2022).

This is the second surprise: noise can be a feature, not a bug.

Machine Learning

Quantum machine learning is a crowded field with inflated claims, but Bharti et al. identify a specific niche where NISQ devices excel: classifying high dimensional data. The authors review experiments where quantum kernels, which measure similarity between data points, outperform classical kernels on certain tasks (Bharti et al., 2022). The noise does not destroy the advantage because the kernel is computed as a statistical average over many measurements. More noise means more measurements, but the signal survives.

How to Program a Noisy Machine

The real innovation of the NISQ era is not in the hardware. It is in the software. Bharti et al. describe a set of programming paradigms that are fundamentally different from traditional quantum computing.

Variational Algorithms

These are the workhorses of the NISQ era. The idea is to use a parameterized quantum circuit, which is a sequence of operations whose strength can be adjusted. A classical optimizer tunes the parameters to minimize a cost function. Because the quantum circuit is short, noise does not accumulate. Because the classical optimizer handles the heavy lifting, the quantum part only needs to be good enough to provide useful information.

Bharti et al. note that variational algorithms have been successfully implemented on hardware from IBM, Google, and Rigetti (Bharti et al., 2022). The same algorithm works across different platforms, suggesting that the approach is robust to the specific noise profile of each machine.

Error Mitigation

This is different from error correction. Error correction adds redundancy to detect and fix errors. Error mitigation accepts that errors will happen but tries to subtract their effects from the final answer.

The authors describe a technique called zero noise extrapolation. You run the same computation at different noise levels, then extrapolate backward to what the answer would be with zero noise. It is not perfect. It does not eliminate all errors. But it improves accuracy by a factor of ten or more in many experiments (Bharti et al., 2022).

Circuit Compilation

This is the art of translating a quantum algorithm into the specific operations that a particular machine can execute. Bharti et al. review techniques that rearrange operations to minimize the time a qubit sits idle, because idle qubits are vulnerable to noise. Other techniques exploit the fact that certain noise patterns are predictable and can be compensated for.

The result is that a well compiled circuit on a noisy machine can outperform a poorly compiled circuit on a cleaner machine. The hardware matters, but the software matters just as much.

What This Does Not Prove

I want to be careful here. The NISQ era is real, but it is not a solution to every problem.

Bharti et al. are explicit about the limitations. NISQ devices cannot run Shor's algorithm, which would break RSA encryption. They cannot run Grover's algorithm for unstructured search at scale. They cannot factor large numbers or solve discrete logarithms. These tasks require thousands of logical qubits with error rates below 10^-15 per operation. Current hardware has error rates around 10^-3.

The authors also acknowledge that the advantage of NISQ devices over classical computers is not always clear. For some problems, a classical computer with a good heuristic algorithm will outperform a noisy quantum computer. The question is not whether quantum is better in principle. It is whether quantum is better in practice, given the noise.

This is an active area of research. Bharti et al. describe ongoing efforts to find "quantum advantage" problems that are provably hard for classical computers but feasible for NISQ devices (Bharti et al., 2022). So far, no one has found a definitive example. But the search has not been fruitless. The authors point to several candidate problems in quantum simulation and machine learning that look promising.

The open question is whether NISQ devices will ever achieve a practical advantage, or whether they will always be overtaken by classical computers that are also getting faster. The answer is not yet known. But the fact that we can even ask the question, that noisy quantum computers are competitive at all, is itself a surprise.

The New Paradigm

The most important contribution of Bharti et al. is conceptual. They argue that the NISQ era has produced new ways of thinking about computation itself.

In classical computing, you write a program, run it, and get an answer. If the answer is wrong, you debug the program. In quantum computing, the noise means you never get exactly the right answer. You get a distribution of answers. The challenge is to extract useful information from that distribution.

This is a different mindset. It is closer to statistical inference than to traditional programming. You do not ask "what is the correct answer?" You ask "what can I learn from the data that this noisy machine produces?"

Bharti et al. show that this mindset has already produced practical results. Chemists are using NISQ devices to study molecules that are too large for classical simulation. Physicists are simulating quantum systems that were previously inaccessible. Optimizers are finding good solutions to hard problems.

None of these results are perfect. But they are useful.

What This Actually Means

• ▸Short circuits beat long coherence. The most important metric for a NISQ algorithm is not how long a qubit can stay coherent. It is how quickly you can run a circuit before noise destroys the information. Algorithms that use short, repeated runs are the ones that work today.

• ▸Classical computers are part of the solution. The hybrid approach, where quantum and classical computers share the work, is not a compromise. It is the optimal strategy for noisy hardware. The classical part handles the optimization, the error mitigation, and the statistical analysis. The quantum part handles the parts that are exponentially hard.

• ▸Noise is not uniform. Different qubits on the same chip have different error rates. Different operations have different error rates. The key to making NISQ work is to characterize the noise on your specific machine and then design around it. This is tedious, but it works.

• ▸The goal is not perfection. If you need a precise answer to a mathematical problem, you should use a classical computer. If you need a good enough answer to a problem that classical computers cannot solve, a NISQ device might be your best option. The standard is not perfection. It is usefulness.

• ▸The NISQ era is not a stepping stone. It is a distinct phase of computing with its own methods, its own successes, and its own open questions. The algorithms developed for NISQ devices will influence fault tolerant quantum computing when it arrives. But they are also valuable in their own right.

The noise is not going away. But as Bharti and his colleagues have shown, that does not mean we have to wait.