| 5 Jun 2022 |
 @tmonz:matrix.org | In reply to @tmonz:matrix.org
btw - some of you interested more on the progress and status on the academic quantum side may like the videos here: https://www.youtube.com/channel/UCYfq48NHj6zbudywnLW3aYw
usually the next speaker is presenting live on thursday afternoon, and takes questions via the chat or email Andreas Walraff, Prof on Experimental Physics at ETH Zuerich, and likely one of the best heads in Europe on superconducting quantum computing, will give a talk on the coming Thursday, June 9th. The paper above on repetitive error correction, published in nature, was from his team. Feel free to attend | 20:23:19 |
| 7 Jun 2022 |
|  Daniel changed their display name from Daniel Peterson to RECLAIM THE RAINBOW. | 17:36:18 |
|  @holl:matrix.org joined the room. | 20:48:35 |
| @holl:matrix.org set a profile picture. | 21:02:59 |
| 8 Jun 2022 |
 Hendrik Weimer | What is a scalable Shor algorithm? | 15:34:54 |
| @tmonz:matrix.org | the argument back then was that all previous implementations of shor's algorithm very based on precompilation of the circuits knowing one of the factors - which is by definition of finding factors not scalable. the methods often presented in those days (with precompilation based on known factors) was pushed to the extreme by john smolin. he 'demonstrated' factoring large numbers using a coin-toss ... fully in line with methods used back then. he not only managed to write a harsh paper about it, but also get it published in nature: https://www.nature.com/articles/nature12290
subsequently the focus was to implement shor's algorithm without precompilation, but building purely on top of modular multipliers without prior knowledge | 16:01:30 |
| @tmonz:matrix.org | i think john did the 'demonstration' at a damop conference or such, if i remember correctly | 16:03:26 |
| @tmonz:matrix.org | Hendrik Weimer: is that what you askde/ | 16:53:31 |
| @tmonz:matrix.org | * Hendrik Weimer: is that what you asked? | 16:53:37 |
| @tmonz:matrix.org | * Hendrik Weimer: does that cover your question? | 16:54:30 |
| Hendrik Weimer | I am aware of this criticism and that "scalable Shor" is somehow meant to counter this. I also remember the old modular exponentiation circuits by Beckman et al., where you need something like 3,000 Toffoli gates to factorize 15. I haven't followed the recent developments in this area, so I assume there has been some improvement on these numbers? | 17:24:22 |
| @tmonz:matrix.org | well. you can have a look at the circuits in the science paper | 17:30:56 |
| @tmonz:matrix.org | it's question of: (a) are we talking about physical or logical qubits? (b) which basis are we looking at? for some basis it's relatively easy, and other basis are very much unforgiving | 17:31:32 |
| @tmonz:matrix.org | seeing the numbers you have in mind, i assume you talk about logical qubits | 17:31:42 |
| @tmonz:matrix.org | there i'm not sure how the toffolis scale ... but for me it feels as if the bigger challenge with fault tolerant circuits are the t-gates | 17:33:14 |
| Hendrik Weimer | Are you referring to https://www.science.org/doi/10.1126/science.aad9480 ? Sure, the circuits are short, but that was also true for the old non-scalable ones (no offense). With "basis" you mean the a in a^x mod N, I assume? So the scalable strategy would be to generate circuits for different choices of a and execute the shallowest ones that you can find? | 17:42:02 |
| @tmonz:matrix.org | t-gate injection is a pain in the neck | 17:42:04 |
| @tmonz:matrix.org | the problem is that you don't know 'the shallowest' one | 17:42:44 |
| @tmonz:matrix.org | because you ideally would have a 'block' that corresponds to a^1 mod N, a^2 mod N, a^4 mod N ... and you would run as many as you can, because you don't know the order in advance | 17:43:28 |
| @tmonz:matrix.org | so even if you would have a 'shallow' instance in terms of circuits, it could be that the order is long ... and thus you don't (necessarily) win | 17:43:55 |
| Hendrik Weimer | Why does the order have to be short? I'd even guess that asymptotically, the order should scale with N for almost all bases. | 17:50:49 |
| @tmonz:matrix.org | i like to get the answer fast ;) | 18:06:33 |
| @tmonz:matrix.org | what i meant is: you can't optimize for circuit length, because it has no influence on the order. and the order essentially tells you how many blocks you need at the very least. as you said, on average, you just need 'a lot' | 18:07:50 |
| Hendrik Weimer | So how is this scalable then? Again, no offense, I simply would like to understand what "scalable Shor" means because I might want to implement it at some point. | 18:12:47 |
| @tmonz:matrix.org | we argued about 'scalable' simply by saying 'no compilation without knowing the answer already' | 18:13:19 |
| @tmonz:matrix.org | properly using modular multipliers | 18:13:42 |
| @tmonz:matrix.org | and, one thing that helped a lot, was to do the semi-classical qft from kitaev | 18:13:58 |
| @tmonz:matrix.org | that reduced the resourced by a factor of (about) 3 | 18:14:06 |
| @tmonz:matrix.org | these days ... not sure what i'd do differently. gates are different. we didn't have compilers back then. likely i'd try to do a circuit design that takes error propagation into account ... maybe add error mitigation. | 18:15:43 |
| @tmonz:matrix.org | depneds a bit on the dominant error sources ... | 18:15:58 |
