Quantum Computing, Capabilities and Limits: An Interview with Scott Aaronson

Google promised quantum supremacy um... well... last year. Made for some great headlines, with the press breathlessly announcing that the quantum era has arrived. Only, it didn't. Internally, there are two quantum computing groups who are competing for resources and don't seem to collaborate.

Rigetti plans a 120 qubit system, but has yet to calibrate it, so I'm not holding my breath. And their qubits still have astonishingly low connectivity and the fidelity of their gates is so low that a larger qubit count won't get them anywhere. In other news, they laid off half of their software team because their chip isn't actually useful, so a full-fledged software stack was somewhat premature (if quite elegant). Ex-employees have some great opinions about the culture; check glassdoor.

Microsoft's QC efforts are fascinating, but AFAIK they don't even have a plan to build a computer. Word on the street is that huge egos and a lack of production experience are conspiring to prevent progress.

And the funny thing about these quantum computing efforts: nobody even has a plan for error correction. For all Scott Aaronson rails against DWave... he doesn't dig deep into gate-model efforts, to expose their flaws. And strangely, he's still focused on coherence time of DWave's qubits.

There's a fundamental difference between gate-model and adiabatic qubits: gate-model qubits in the "0" state are at a ground level, and "1" is an excited state -- whereas DWave's qubits have "+1" corresponding to clockwise current, "-1" corresponding to counterclockwise current; and both are ground states of the qubit. Coherence time is roughly the amount of time it takes an excited state to decay into a ground state. Adiabatic transitions begin, remain, and end in ground states, barring external perturbations -- and if such an event occurs, a low coherence time is actually preferable. Higher coherence times are still beneficial to adiabatic quantum computers like DWave's and Google's G-mons, since it boosts long-range interactions between qubits, but it isn't the showstopper that Aaronson hypes it to be.

So gate-model quantum computing is promising, in theory. But there are some very real, very difficult roadblocks to clear before we get there. And for all that DWave gets omitted from such discussion... well, the theorists say that adiabatic and gate-model quantum computing are actually equivalent[1]. Wake me up when somebody else can factor arbitrary 10-bit integers[2] -- and even so... 10-bit factoring problems don't really scream "quantum computing may be here soon" -- more like "1970 called, and says quantum computing is crushing it."

[1] https://physicsworld.com/a/quantum-adiabatic-and-quantum-cir... [2] https://www.dwavesys.com/sites/default/files/14-1002A_B_tr_B...

Assuming we double the number of physical qubits we can use in a quantum computer every year starting now, we might be able to build one capable of running interesting algorithms sometime around 2040.

The general consensus in the field is that we need to have 100,000 times more physical qubits in a machine to achieve scalable error-corrected quantum computation. That represents tens of millions of qubits; our current state of the art is floating somewhere in the low to high tens of qubits. That’s quite sobering compared to breathless tech reporting, and it doesn’t address the decoherence reductions we’ll need either.

Quantum computing is not coming soon, barring an unlikely and unpredictable series of fundamental breakthroughs. The field is still very much in the, “we have no evidence it’s impossible” stage of research.

In fact, when you see large mainframe-like pictures of quantum computers, the big box mostly contains cooling equipment. The actual circuitry is pretty small.

[1] https://www.dwavesys.com/tutorials/background-reading-series...

The relationship between the two is essentially Energy = k_B * temp = h * freq, where k_B is the Boltzmann constant and h is Planck's constant.

If the temperature was higher than 10 milliKelvin, then thermal fluctuations could have comparable energy to your control signals and thus cause unintended electronic state transitions.

I found it easy to comprehend. It starts explaining to a child & works its way up to a doctorate level.