Opens profile photo
Follow
Click to Follow CraigGidney
Craig Gidney
@CraigGidney
Programmer turned research scientist on Google's quantum computing team. Maker of Quirk, a fun drag-and-drop quantum circuit simulator ( algassert.com/quirk ).
Santa Barbara, CAalgassert.comJoined July 2013
122 Following
2,786 Followers

Craig Gidney’s Tweets

Craig Gidney
@CraigGidney
Suppose as a hypothetical that quantum mechanics is wrong, in a way that breaks quantum computers. In that world, how do you expect quantum computers to break *first*? The simplest task that couldn't be made to work sufficiently well, even with error correction, would be...
When you make a selection it cannot be changed
145 votes11 hours left
8
11
Craig Gidney
@CraigGidney
On a plane, the blossom algorithm dual variables form visual proofs the matching is optimal. A better solution would correspond to applying an augmenting cycle or path. But the augment crosses each non-overlapping circle or blossom-layer *twice*. So it can only move or add costs.
Image
11
Craig Gidney
@CraigGidney
In other words, prove that entanglement distillation stop working at 50% depolarization, instead of only at 75% depolarization, by showing the state you get after applying 50% depolarization to |00>+|11> can be created while using only non-entangling operations.
1
2
Show this thread
Craig Gidney
@CraigGidney
Quantum circuit puzzle: Alice and Bob share a Bell pair corrupted by 50% depolarization on Alice's qubit (max mixing is 75%). Show this error destroyed all entanglement by finding a circuit that makes the same mixed state using local operations and classical communication.
Image
1
13
Show this thread
Craig Gidney
@CraigGidney
The color code decoding problem is similar to a game of LightsOut on a hex grid, with the lights on faces and buttons on nodes. The goal is to turn the lights out in as few moves as possible. AFAIK the QEC community doesn't know an efficient way to solve it exactly. Does anyone?
Image
4
50
Craig Gidney
@CraigGidney
Although, now that I think about it more, maybe the visual diff can be run the other way. If turning the square prism into a triangle prism worked turned [[7,1,3]] into [[5,1,3]], maybe turning the five sided pyramid into a seven sided pyramid will turn [[5,1,3]] into [[7,1,3]]?
1
Show this thread
Craig Gidney
@CraigGidney
Thanks to this flawless poll with no issues whatsoever, we finally know the nature of reality: amplitudes are continuous. Unless you account for vote splitting, in which case they're discretized, most likely as double precision floating point numbers. Phew. Glad that's resolved.
4
Show this thread
Craig Gidney
@CraigGidney
It also included another view of the sphere. Did you know that, from the correct perspective, every sphere has a triforce? Neither did I! (On the other hand, it's amazing it can make any text diagram at all in the first place.)
Image
8
Show this thread
Craig Gidney
@CraigGidney
It's funny how ChatGPT fails when it fails. I asked it for a text diagram of a sphere and it made an upside down ice cream cone shape. But don't worry; it helpfully labelled it "sphere" in case that was confusing.
Image
13
146
Show this thread
Craig Gidney
@CraigGidney
arxiv.org/abs/2212.12372 sure does spend a lot of space not mentioning the expected number of circuit shots it requires. It's critical to the entire premise of the paper to have a small bound on this number, and as far as I can tell they just don't talk about it. Very bad sign.
arxiv.org
Factoring integers with sublinear resources on a superconducting...
Shor's algorithm has seriously challenged information security based on public key cryptosystems. However, to break the widely used RSA-2048 scheme, one needs millions of physical qubits, which is...
5
69
Craig Gidney Retweeted
Patrick McKenzie
@patio11
Last year, a group of geeks convened on Discord in response to a Twitter thread about California’s woes with the vaccine rollout. They very quickly found themselves running a critical piece of national public health infrastructure. Here is our story:
worksinprogress.co
The story of VaccinateCA - Works in Progress
Nobody had a plan to get vaccines out of freezers and into Americans’ arms–except VaccinateCA. Its CEO tells the story of how a small team brought order to a chaotic rollout.
18
661
Show this thread
Craig Gidney
@CraigGidney
Advent of code 2022 starts tonight! A bunch of us playing along has become a minor yearly tradition on google's quantum team. See y'all on the puzzlefield.
1
14
Craig Gidney
@CraigGidney
Rule of thumb: avoid big stabilizers. By default big stabilizers hurt detection fractions cubically: more qubits, more ops per measurement, and spreadier hook errors. It's REAL BAD to use big stabilizers. It's a common mistake to pump code distance by using bigger stabilizers.
2
12
Show this thread
Craig Gidney
@CraigGidney
Quantum circuit challenge: take as input one S state |0> + i|1> and output two such states, using the gate set {ResetX,ResetZ,MeasureX,MeasureZ,CNOT}. Or prove that it's not possible.
4
23
Craig Gidney
@CraigGidney
Most likely this is because of the stabilizer generator corresponding to the observable being perturbed by the intermediate measurements. In a way that requires a lot of iterations before the specific choice of basis being used from iteration to iteration starts repeating.
Show this thread
Craig Gidney
@CraigGidney
Something that surprised me: the tableau representation of the honeycomb code's state varies from round to round. I tried optimizing stim's measurement sampling by watching for states recurring during loops. It works, technically... but not on the case I wanted it to improve!
Image
1
9
Show this thread
Craig Gidney
@CraigGidney
Fun fact: this is also basically the qec equivalent of three interlocked gears unable to turn with the words "working together" written next to it😂. See if you can fix it making as few changes as possible.
2
Show this thread
Craig Gidney
@CraigGidney
Office hours complete! Ended up seeing two people. I think they each went well, in that they conveyed random important details that aren't well explained in papers but are absolutely crucial to making something that works.
1
1
Show this thread
Craig Gidney
@CraigGidney
The underlying reason is that everything only matters up to Paulis, so there's not really a distinction between period 2 and period 4. Also, I sort of lied when I said it's 4:2 and 2:4; really there are also examples of 4:4 and 2:2 it's just the cheapest ones are 4:2 and 2:4.
Show this thread
Craig Gidney
@CraigGidney
You'd expect the period of single qubit Cliffords to correspond to the periods of surface code corner permutations implementing them, like S has period 4 so it would do a period 4 permutation. But actually the period correspondence is 4:2, 3:3, 2:4.
Image
1
Show this thread
Craig Gidney
@CraigGidney
I'm doing another Stim office hours, where I help y'all with QEC/coding problems. 30 minute slots wednesday November 9th from 1pm to 5pm PST. If you want a slot, find my email from a paper and email me with subject STIM OFFICE HOURS.
2
25
Show this thread
Craig Gidney
@CraigGidney
Probably the main way stim gains convenience functions is someone asks a question on stack exchange, I write demo code for it, and I'm not happy at how long the demo code is.
1
11
Craig Gidney
@CraigGidney
Here are quantum solutions in quantum flytrap and in quirk. Here's the saved JSON of the flytrap setup gist.github.com/Strilanc/3a6d5 and here's the link to the Quirk circuit algassert.com/quirk#circuit=
Image
Image
2
Show this thread