Profile avatar
craiggidney.bsky.social
Research scientist on Google's quantum team, working on reducing the cost of quantum error correction. Useful tools I've made: - Quirk: https://algassert.com/quirk - Stim: https://github.com/quantumlib/stim - Crumble: https://algassert.com/crumble
84 posts 662 followers 19 following
Regular Contributor
Active Commenter
Posts 19 Comments 31

A misleading slide shown at QIP2025. The surface code number is 10x too high (should be 0.1% not 1%; see fig 1c of arxiv.org/abs/2408.13687) and the tesseract number omits the worst 18% of shots (see table III of arxiv.org/abs/2409.04628).

submitted 3 days ago • 0 comments

Slides from my QIP talk on magic state cultivation: docs.google.com/presentation...

submitted 4 days ago • 1 comment

If qubits only talk to 1D neighbors, is constant rate fault tolerance possible? Last year I referee'd a paper claiming it wasn't. My review was "this is wrong but the constructive disproof is too large for this review". Clearly a reviewer 2 move. Sorry. But I was right! scirate.com/arxiv/2502.1...

submitted 4 days ago • 3 comments

If you approximate a phase gradient state with stabilizers it kinda... just... works? Mostly? The approximation has a fidelity of ~80%, so there needs to be 80% overlap with the correct result even when doing things like QFTs that want fine grained angles. Weird! algassert.com/quirk#circui...

submitted 11 days ago • 0 comments

Me? Nerd sniped? I needed phase gradient qubits, and was using sequences from gridsynth. That has 33% waste because gridsynth makes rotations (3DOF) not states (2DOF). But this isn't a key part of the paper. And it's fine. It doesn't matter. There's more important... Hm. Maybe mitm would work...

submitted 11 days ago • 1 comment

Here is the Simons institute talk I never got to give, due to a flight cancellation. Given to whoever I could round up, instead: youtu.be/SyW1LkbFv6k It definitely shows that this was intended for a technical audience. But there's some good audience interaction.

submitted 12 days ago • 0 comments

With two qubits of storage you can measure a series of shingled operators like AB, BC, CD, DE, etc while touching each data qubit once. The trick is to mask measurements of A,B,C,D,E, etc with a GHZ state. ...kinda equivalent to storing X til you get Y, to compute XY, now drop X but store Y, etc.

submitted 15 days ago • 0 comments

Who hurt you

submitted 23 days ago • 2 comments

Tried my hand at vectorizing sampling of a geometric distribution (it's useful for qec sims). Got it down to 1.5 nanoseconds per sample, but with sorta low quality (~100k samples for a perfect distinguisher to become 2:1 confident it's not perfect). gist.github.com/Strilanc/4f0...

submitted 26 days ago • 1 comment

It's weird. Often when I answer a stack exchange question, they ask me wtf I did to make the circuit. Isn't it obvious? /sarc

submitted 37 days ago • 1 comment

God github actions are so brittle. Every month there's some random thing that breaks despite no change to the config file. Currently my javascript tests are failing because puppeteer can't start chromium because [???]. Separately, clion decided it doesn't understand my cmake files now. Ughhhhhhhh

submitted 38 days ago • 1 comment

Superposition masking (arxiv.org/abs/2008.04577) gives a cheap way to compute |x⟩ → |x⁻¹,x⟩ using a few quantum*=classical multiplications. But the trick doesn't work for uncomputing |x⁻¹,x⟩ → |x⁻¹⟩, so afaik ~all the cost of inverting is clearing the original register. Easy to grow, hard to squish.

submitted 39 days ago • 0 comments

How to write a paper. (1) Come up with a clever idea. (2) Write a lot of paragraphs describing it. (3) Delete those paragraphs because there's a simpler idea that makes all that irrelevant. (4) Repeat.

submitted 40 days ago • 1 comment

After spending a day trying to solve a thing ... wait... is this just matrix multiplication ...? 🤦 okay well that fixes that now onto the next thing that's secretly just matrix multiplication

submitted 53 days ago • 2 comments

A tell for theory-vs-practice in quantum magic states is mentioning γ. In theory, γ dominates by determining how costs compound as stages are added. In practice we'll do MAYBE 1 stage of distillation past cultivation, enough to be more reliable than transistors, and then stop. Making γ irrelevant.

submitted 53 days ago • 2 comments

Stim is a tool for simulating quantum error correction. I made it because previous tools were limited to operating on the abstractions used to make QEC circuits, instead of the circuits themselves. It took a few years to catch on, but is now common. Used by 198 papers in 2024.

submitted 57 days ago • 1 comment

Interesting paper implementing magic state distillation with neutral atoms: arxiv.org/abs/2412.15165 They're in a weird regime: below the distillation threshold but over the color code threshold. They see improvements!... But hit a ceiling imposed by their d=3 memory error (because d=5 does worse).

submitted 71 days ago • 1 comment

Over the past year I played a supporting role on experiments showing we have a flexibility in how we do QEC on a planar chip. We do a color code (arxiv.org/abs/2412.14256), and three interesting variations of the surface code circuit (arxiv.org/abs/2412.14360), all below threshold.

submitted 71 days ago • 0 comments

The "hide the exponential" game Scott Aaronson has complained about for decades is alive and well: arxiv.org/abs/2412.13164 Yes, Shor's algorithm with 3 registers only needs very basic operations. No, you can't pack a 2^2000 level quantum state into 1 oscillator and then operate accurately on it.

submitted 72 days ago • 2 comments