HNHacker News
TopNewBestAskShowJobs

fastneutron

455 karma · joined May 10, 2020

Principal R&D engineer - Deeptech, nuclear energy, scientific computing, quantum technologies, artificial intelligence.

fastneutron235 -at- gmail -dot- com

submissionscomments
fastneutron··on Ask HN: Freelancer? Seeking freelancer? (November 2023)
SEEKING WORK | Remote | US

I am an experienced applied physicist with 10 years of experience in implementing the full hardware and software stack of emerging technologies. I am available for part-time consulting, including informal discussion and brainstorming, in the following areas:

- Design, validation and due diligence for deeptech concepts, especially quantum technologies, energy technologies, and AI-augmented engineering tools,

- Design and prototyping of sensor and instrumentation hardware/software,

- Architecture and prototyping of scientific & numerical software, including generative modeling.

Email in profile, CV available on request.

fastneutron··on Large deposit of "white" hydrogen found underground in France
I had to look up what “white hydrogen” actually was: https://en.wikipedia.org/wiki/Natural_hydrogen
fastneutron··on NY bill would require a criminal history check for the purchase of a 3D printer
Are there any documented cases of people actually committing crimes with a 3D printed firearm?
fastneutron··on Most demographers now predict that human population will plateau
Care to elaborate? Are you envisioning some decoupling of energy and economic growth?
fastneutron··on Leaked screenshot shows Amazon now tracking individual employee attendance
At places I’ve worked, they were logged, but only ever audited in cases of suspected time fraud or other legal infractions. Anonymized statistics were also sometimes used for covid tracking. There was never somebody actively sifting through the data and taking attendance.
fastneutron··on Show HN: TG – Fast geometry library for C
You might consider Shewchuk’s methods for robust geometric computations: https://www.cs.cmu.edu/~quake/robust.html

There are more modern implementations available on GitHub, but I’ve found it to be fairly easy to integrate into C-family projects without much modification.

fastneutron··on “Most notorious” illegal shadow library (LibGen) sued by textbook publishers
It’s impossible to even get a sense of a textbook’s content from the so-called “previews” that you find on Amazon, Google Books, and the publisher’s’ websites. I prefer print, so I’ll happily pay if the content is good, but just letting me see the preface, contents, index and maybe a few pages of chapter 1 isn’t enough for me to discern that. Libgen at least lets me see what I might be getting.

Libgen is also great for books that are long out of print, and the authors long dead. It’s a crime against humanity that these works are locked up by our archaic copyright laws, for things that have long since ceased to be remotely profitable for the publisher.

fastneutron··on Wireless electrical–molecular quantum signalling for cancer cell apoptosis
Reported on in a popular article here: https://interestingengineering.com/health/first-quantum-canc...

This work purports to be the first kind of medical treatment whose mechanism makes direct use of an engineered quantum-mechanical effect.

fastneutron··on Sucking carbon dioxide out of the sky is moving from science fiction to reality
CO2 removal and sequestration will be a net energy sink no matter what you do, since it amounts to “unburning” some portion of fossil fuel combustion that’s already been committed. The question is what methods are the least bad in terms of thermodynamics, and how much renewable and nuclear energy capacity are we willing to divert?
fastneutron··on How to make a synthetic diamond (2009)
Element Six [1] makes all sorts of synthetic diamonds for industrial and analytical applications. Some are actually quite cheap, although there’s some irony perhaps in that they’re a DeBeers subsidiary.

We use these diamonds in my lab for experiments with nitrogen vacancy centers, which are an interesting platform for solid state qubits.

1. https://e6cvd.com/

fastneutron··on Ask HN: In person industry tours and site visits?
Local and university chapters of the American Nuclear Society sometimes organize tours of nuclear power plants in their respective regions. I went on a few of these tours in college and it was truly a unique experience. It's easy to talk about a gigawatt of power, but to actually feel it screaming past you in the turbine hall really opens your eyes to the immense scale of the electrical infrastructure we've built to power our civilization.
fastneutron··on BoTorch – Bayesian Optimization in PyTorch
These two references are both really good, and freely available:

Bayesian Optimization - https://bayesoptbook.com/

Surrogates - https://bobby.gramacy.com/surrogates/

fastneutron··on BoTorch – Bayesian Optimization in PyTorch
It depends on the needs of the specific application. What typically happens is that you’d use BO to globally converge to within some tolerance and use the resulting surrogate to get a map of where the interesting regions are. You can then more densely sample these candidate regions or switch to a gradient-based method (via finite difference). For uncertainty information, we usually add this as a noise parameter, either on the input samples, or as part of the GP kernel.
fastneutron··on BoTorch – Bayesian Optimization in PyTorch
It wasn’t terribly difficult once we got a feel for the underlying API. GP surrogates work out of the box, but you can plug in other kinds as well. Same goes for acquisition functions.

As far as the objective, usually we’re calling an external physics-based solver (e.g. finite elements) and post-processing the solution to get the quantity we’re trying to optimize. There’s almost never any gradient information, so Bayesian optimization winds up being the method of choice.

fastneutron··on BoTorch – Bayesian Optimization in PyTorch
If you’re not from a traditional engineering background, some of my terminology may have been confusing. Here’s a less jargon-heavy version:

- We use Bayesian optimization to find the optimum (or worst-case) configuration of real manufactured objects and systems. - Bayesian optimization lets us arrive at that design configuration faster than explicit, physics-based simulation of many samples within the space of all possible configurations.

- We built the framework to do that using Botorch.

- It’s not an uncommon practice by any means, but the availability of tools like Botorch now makes it a lot easier to implement Bayesian optimization in-house, vs relying on a vendor-based engineering tool.

fastneutron··on BoTorch – Bayesian Optimization in PyTorch
We built (internally) an engineering optimization framework on top of Botorch. Bayesian optimization is a great tool to have in the toolbox for mapping a design space and finding the limiting cases with fewer calls to the expensive physics-based solvers. The computational budget often winds up being 5-10x smaller than it would have been using traditional design-of-experiments sampling methods.
fastneutron··on Declining fertility ideals? Young people more often prefer having no children
If you’re in a developed nation, and unless you expect the earth to go through an irreversible Venus syndrome in the next century, I don’t see how this argument is anything more than a fig leaf for more fundamental social ailments. What would you say to someone in 700 AD, for whom the next 1000 years were more likely to be objectively worse in terms of quality of life than the next 1000, even accounting for what climate change is expected, intervention notwithstanding.

The costs of childcare, housing and education have exploded, and even establishing the relationship needed to do the deed in the first place is arguably harder than it s been. That sucks and there’s little any individual can do about it, despite the rhetoric blaming it on the generations most likely to be impacted by it. Climate is a convenient scapegoat to deflect blame, but we were never to blame in the first place.

fastneutron··on Ask HN: How long until quantum computers break 512 bit RSA/ECDSA? 1024 bit?
On what basis do you make that statement? The hardware is literally manipulating coherent quantum states of trapped ions or transmon junctions in the cases of IonQ and IBM, respectively. Noisy and unreliable, sure, but most definitely not classical.
fastneutron··on Peter Shor's MIT Fall 2022 course lecture notes on quantum computing
This video, “Quantum Computing for Computer Scientists” is quite approachable: https://m.youtube.com/watch?v=F_Riqjdh2oM&pp=ygUeTWljcm9zb2Z...
fastneutron··on Introduction to the Conjugate Gradient Method Without Agonizing Pain (1994) [pdf]
Shewchuk’s work on mesh generation is nothing short of a masterpiece. I will always direct people to the source of his Triangle code as an example of what good, literate C code should look like. His Berkeley page is here: https://people.eecs.berkeley.edu/~jrs/
fastneutron··on Cargo Cult Science (1974) [pdf]
The big "if" in quantum computing is in our ability to engineer error correction and fault tolerance. Implementing fault tolerance requires a physical qubit overhead of 10-10,000 qubits per logical qubit, depending on the technology, base error rate and whose analysis you look at.

It's not entirely accurate to say nobody knows how to do this. The theory of quantum error-correction has been rigorously developed since the 1990s, and experimental implementations have shown that improved fidelity is possible with known error-correction schemes. I think we'll know definitively within the next 3-5 years how hard it will be to engineer fault tolerance, and workable solutions will emerge from those findings.

fastneutron··on Cargo Cult Science (1974) [pdf]
I work adjacent to this field, and I largely agree when it comes to the software side of things. With some notable exceptions, a lot of pure-play "quantum software" companies are premature optimizations to a field who's hardware can best be described as physics experiments with a Python API. Many of these places are pivoting to "AI" applications, which is simply them putting their mouths where the money is. If a "quantum winter" happens, many of these places will be the first to go.

I think this is partially a symptom of cargo culting the SV model, like you suggested, along with the much more fundamental reason that software has a very low barrier to entry. Hardware is hard, but it's really the most impactful investment to make in quantum tech at this early stage because of the spin-off applications in sensing, timing and networking.

fastneutron··on The US government is taking a step toward space-based nuclear propulsion
> Once it reaches a safe orbit, the reactor will be turned on.

Fission reactors aren't appreciably more radioactive than rocks you can dig out of the ground until you turn them on. This is something that wouldn't be brought online until the riskiest part of the launch has past.

fastneutron··on Wendelstein 7-X: Gigajoule energy turnover generated for eight minutes
Nitpick: You’re always going to get some parasitic D+D reactions because it’s reaction cross section is appreciably higher than D + 3He below 50 keV, and non-negligible even after the crossover point.
fastneutron··on A Critical Problem (2017)
Nuclear criticality safety was one of my favorite courses back during my nuclear engineering degree. The failure modes, like the article mentions, can be incredibly non-intuitive. Swirling vortices, firefighting foam, and even the mere presence of a human body can lead to a fatal power excursion. A lot of complex computational modeling and simulation goes into determining safety limits for fissile systems, as well as a lot of sheer imagination.

The most important thing I took away from that experience is the concept of engineered compliance. People will break administrative rules and regulations all the time if they get int the way of work, so you need to engineer the physical system to account for that (e.g. flat or skinny geometry, or lots of neutron absorbers). This can apply to any compliance activity related to safety and security, so the lessons are universal.

Los Alamos published a good survey on various worldwide criticality accidents (PDF): https://www.nrc.gov/docs/ML0037/ML003731912.pdf. Worth a read for all the various ways things can go wrong.

fastneutron··on Replace peer review with “peer replication” (2021)
As much as I agree with the sentiment, we have to admit it isn't always practical. There's only one LIGO, LHC or JWST, for example. Similarly, not every lab has the resources or know-how to host multi-TB datasets for the general public to pick through, even if they wanted to. I sure didn't when I was a grad student.

That said, it infuriates me to no end when I read a Phys. Rev. paper that consists of a computational study of a particular physical system, and the only replicability information provided is the governing equation and a vague description of the numerical technique. No discretized example, no algorithm, and sure as hell no code repository. I'm sure other fields have this too. The only motivation I see for this behavior is the desire for a monopoly on the research topic on the part of authors, or embarrassment by poor code quality (real or perceived).

fastneutron··on Using fluids to model inaccessible realms of the cosmos
For a highly in-depth look at the theory behind this, I can recommend “The Universe in a Helium Droplet” by Volovik [1].

1. https://academic.oup.com/book/11557

fastneutron··on Quantinuum H-Series quantum computer
Depends on the time of day. Sometimes quantum topics generate good discussion here, and other times it’s a peanut gallery of minimally-informed dismissals. I’ve had a heck of a time trying to find a decent online community to discuss this stuff in a properly nuanced way.
fastneutron··on Quantinuum H-Series quantum computer
In principle, the advantages of quantum computers are delivered not in speed or throughput, but in algorithmic scaling. For example, the well-known Grover's search algorithm scales as O(sqrt(N)), as opposed to O(n) for classical linear search. Similarly, Shor's factoring algorithm comes with an exponential scaling speedup, taking a nominally O(2^n) problem and bringing it to O(n^k). Problems related to physics and chemistry simulations are falling in various places along that quadratic-to-exponential speedup spectrum as well.

In practice, there's expected to be a larger constant overhead than with classical computing hardware, since error correction would need to be performed continuously during operation, and the physical operations of addressing qubits are inherently slower than classical CPU or GPU operations. There have been a few publications [1,2] asserting that this overhead essentially negates any quadratic scaling benefits, and that only cubic or better speedups would deliver any practical advantage. This assumes no further improvement in operation times or error correction overhead.

As far as timeframe, it's anyone's guess. Error correction will be needed to get beyond these research prototypes, and I think we'll have a better sense over the next 2-5 years as to how hard that will be to engineer.

1. https://arxiv.org/abs/2011.04149

2. https://arxiv.org/abs/2307.00523

fastneutron··on Quantinuum H-Series quantum computer
The migration to post-quantum cryptosystems is a different issue than standardizing one in the first place, and I'm not going to pretend to know anything about the organizational cat-herding needed to achieve such a migration. The US government is beating the drum to have the migration of critical systems completed by 2030, and there are plenty of IT consulting firms who will happily sell post-quantum migration services to get ahead of that deadline.

If you're worried about pre-harvested ciphertext that nation-states are sitting on, waiting for the right hardware to come along to decrypt it, I personally have no solution. The genie's probably out of the bottle on that one. Hopefully the affected parties took, or are taking mitigating actions.

← PreviousPage 3 of 5Next →