Why it is so hard

The phase is where the whole advantage lives, and the phase is the first thing to go. This is why quantum computers sit at ten thousandths of a degree above absolute zero, and why the field has spent thirty years on error correction rather than on algorithms.

1970s – today18 min

By the end you should be able to

Distinguish T₁ from T₂ and say why T₂ is the number that matters
Explain why superconducting qubits require millikelvin temperatures
Assess honestly where quantum computing currently stands

We met decoherence in Act 6 as an explanation: why you never see a cat in superposition, why the world looks classical, why the measurement cut can be placed anywhere sensible. Nothing is ever isolated; correlation with an environment you do not track destroys interference; and it happens extraordinarily fast. Here it is the adversary. Building a quantum computer means holding a superposition together long enough to do something with it, against a universe that is constantly measuring it. The same phenomenon, viewed from the other side. And the numbers from Act 6 apply: a dust grain in air is struck 10¹⁸ times a second. A qubit is far more delicate than a dust grain.

A dilution refrigerator with quantum processors enclosed, layered copper and gold stages. — A dilution refrigerator, and an honest picture of what decoherence costs to fight. Every stage exists to remove one more channel by which the outside world could learn something about the state — thermal photons, vibration, stray fields. You cannot stop the environment measuring a quantum system; you can only make it slower at it.

Start on the equator and let time run. Watch the vector leave the surface.

Loading the sphere…

A pure state lies on the surface of the sphere. Decoherence pulls the vector inwards, and a point inside the sphere is not a superposition at all — it is a statistical mixture, a state of ignorance about which of two definite things the qubit is.

Watch the coherence readout as time passes. Note that purity falls and then recovers towards 1 — because the qubit ends in the pure ground state, having lost the superposition entirely. Purity returning to 1 is not the qubit recovering. Coherence is the number that tells the truth here.

Loading the sphere…

The sideways component is the phase; the vertical is the energy. With T₂ shorter than T₁, the sideways part decays first. The parameter that no measurement can see is also the parameter that decoheres soonest.

The invisible thing is the fragile thing

Two lessons ago: the phase is the parameter no measurement can detect, and it is where every quantum algorithm does its work — arranging phases so that amplitudes for wrong answers cancel. This lesson: the phase is the first thing decoherence destroys, because dephasing needs no energy exchange and therefore happens more readily than relaxation. So the valuable thing and the fragile thing are the same thing. That is not a coincidence. What makes the phase invisible is that it does not show up in the probabilities — and what makes it fragile is that anything in the environment which learns even a little about the qubit's state randomises it. Being invisible to you does not mean being invisible to the environment. The environment is measuring constantly, in every basis, without recording the result anywhere you can reach.

The experiment

Watching a superposition decohere in real time

Michel Brune, Serge Haroche, Jean-Michel Raimond and colleagues · 1996 · École Normale Supérieure, Paris

The question: Decoherence is invoked to explain why superpositions vanish. Can the process itself be observed as it happens — and does the decay rate depend on the "size" of the superposition, as theory requires?
The apparatus: A microwave cavity made of superconducting niobium mirrors, so good that a photon survives inside it for a substantial fraction of a millisecond. Rubidium atoms in highly excited Rydberg states were passed through one at a time; the first atom placed the cavity field into a superposition of two states with different phases — a small "Schrödinger cat" of a few photons — and a second atom sent through after a controlled delay probed whether the coherence between them had survived.

Theory predicted

If decoherence is a real dynamical process rather than a bookkeeping device, the coherence should decay measurably, and faster for superpositions of more distinguishable states — the decoherence time should fall as the separation between the two components grows.

They measured

The coherence decayed on a timescale much shorter than the cavity’s own photon lifetime, and the rate increased with the separation between the two field components, exactly as predicted. Decoherence was observed as a continuous process with a measurable rate.

How sure could they be? Sufficient to trace the decay curve and confirm the scaling with the size of the superposition, which is the theoretically distinctive prediction. Haroche shared the 2012 Nobel Prize with David Wineland for this line of work.

Why it mattered

It turns decoherence from an explanation into a measured phenomenon with a rate you can compute and check. That matters both ways: it supports the account of why the classical world looks classical, and it establishes the quantitative adversary that anyone building a quantum computer has to defeat. Note also what it does not do — it says nothing about why a single outcome occurs, which remains exactly where Act 6 left it.

Correcting errors you are not allowed to look at

From the last lesson: you cannot copy a qubit, and you cannot inspect one without destroying it. Shor's 1995 scheme evades both — spread the information across several entangled qubits, then measure relationships between them rather than the qubits themselves. Parity checks reveal that an error occurred, and where, while revealing nothing about the protected state. It works, subject to two brutal conditions. A threshold. The physical error rate must fall below roughly 1%, depending on the scheme. Above it, adding more qubits makes things worse — you introduce errors faster than you correct them. Below it, adding more makes things better, without limit. There is no gentle slope: it is a phase transition. Overhead. Current estimates run to around a thousand physical qubits per logical qubit you can rely on. A machine that could factor a 2048-bit number needs millions of physical qubits. Present devices have hundreds.

You might think

Quantum computers have already achieved supremacy over classical computers, and are close to breaking encryption.

Actually

Neither, and it is worth being precise because this is where most of the exaggeration lives. Google’s 2019 Sycamore result demonstrated a sampling task that was chosen to be hard classically and natural for the hardware — it computes nothing anyone wanted computed, and improved classical algorithms subsequently narrowed the claimed gap by many orders of magnitude. As of now, no quantum computer has performed a useful calculation faster than a classical machine could. Not one. On encryption: Shor’s algorithm has factored numbers like 15 and 21, and factoring anything cryptographically relevant requires millions of physical qubits against the hundreds that exist. None of this casts doubt on the physics, which is not in question. It is a statement about engineering, and about the distance between a demonstration and a machine.

That said, the progress has been real, and on precisely the thing that matters. The threshold is the whole game. Below it, more qubits make things worse; above it, more make things better. Everything else is detail. In 2023 and 2024, several groups demonstrated logical qubits with lower error rates than their constituent physical qubits — Google's surface-code work showing error suppression that improves as the code grows, and neutral-atom and trapped-ion groups showing logical operations outperforming physical ones. That is the line between a laboratory curiosity and a technology, and it appears to have been crossed. What remains is scale, and scale is a matter of a great deal of engineering rather than of new physics. Which is a much better problem to have than the alternative.

1970Zeh proposes that environmental interaction, not measurement, destroys superposition.
1981–82Zurek develops decoherence quantitatively; "pointer states" explain which quantities look classical.
1996Haroche’s group watches a superposition decohere in real time.
1995Shor shows quantum error correction is possible; Steane independently.
1999First superconducting qubit: coherence 2 ns, enough for two operations.
2019Sycamore samples a contrived distribution; classical algorithms then close much of the gap.
2023–24Logical qubits outperform their physical constituents. The threshold is crossed.

Where this act leaves us

Act 7 ends roughly where it began, which is a pleasing shape for it. The strangeness Einstein found intolerable in 1935 turns out to be a resource — and the reason it is difficult to exploit is the very same reason the world looks classical in the first place. The environment is always measuring, and the thing it destroys first is the thing you most needed. But something has been assumed throughout, without comment, and it is time to notice it. Everything in this act has treated particles as the fundamental objects. Electrons, photons, qubits — things that exist, carry properties, and get measured. That picture does not survive contact with three facts. Every electron in the universe is exactly identical, to the limit of measurement, in a way that manufactured objects never are. Particles are created and destroyed routinely — collide two protons hard enough and hundreds of particles come out that were not inside them. And the vacuum, with nothing in it at all, has measurable physical effects. None of that makes sense if particles are the basic furniture of the world. The last act is about what happens when you stop assuming they are.