frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

Open in hackernews

Solving LinkedIn Queens with SMT

https://buttondown.com/hillelwayne/archive/solving-linkedin-queens-with-smt/
125•azhenley•23h ago

Comments

TheBozzCL•22h ago
Hah, about a month ago I wrote a DLX solver for exact cover problems and LiQueens was one of my first implementations.

Next I want to try to solve the Tango and Zip games.

sevensor•22h ago
SMT is so much fun. The Z3 Python api lets you write your problem very directly and then gives you fast answers, even for quite large problems.
doctorpangloss•16h ago
This post is the programming joke about Python, "import solution; solution()".
sevensor•13h ago
Barely a joke, this is literally what using the Python Z3 bindings feels like.
ndr•6h ago
I did write a shockingly similar solution few months ago:

https://gist.github.com/enigma/98ea0392471fa70211251daa16ce8...

Recursing•4h ago
Note that CVC5 has basically the same Python API ( https://cvc5.github.io/docs/cvc5-1.1.2/api/python/python.htm... ) and is often much faster
OutOfHere•22h ago
The article fails to even say what SMT is. It also fails to describe and explain it. This article should help:

https://en.wikipedia.org/wiki/Satisfiability_modulo_theories

Jtsummers•22h ago
In the article:

> "Satisfiability Modulo Theories"

rook37•22h ago
Huh, where? I didn't see it going through and opening it in reader mode and ctrl+f-ing any of those words turns up nothing for me still.
Jtsummers•21h ago
He embeds the footnotes in the web version (it's a proper footnote in the email newsletter version). Find the "..." in this paragraph:

> Ryan solved this by writing Queens as a SAT problem, expressing properties like "there is exactly one queen in row 3" as a large number of boolean clauses. Go read his post, it's pretty cool. What leapt out to me was that he used CVC5, an SMT solver. (...) SMT solvers are "higher-level" than SAT, capable of handling more data types than just boolean variables. It's a lot easier to solve the problem at the SMT level than at the SAT level. To show this, I whipped up a short demo of solving the same problem in Z3 (via the Python API).

GZGavinZhao•21h ago
Formal Methods in general are underrated in the industry. Pretty much no large companies except AWS (thank you Byron Cook!) use them at a large scale.

Edit: maybe there are large companies that use them behind the curtains, but AWS is the only place I know of where they publicly acknowledge how much they appreciate and use formal methods. If you know any of them, please comment and I'd be curious to learn about how they're using it!

steamrolled•21h ago
> Formal Methods in general are underrated in the industry. Pretty much no large companies except AWS (thank you Byron Cook!) use them at a large scale.

At least Microsoft and Google poured a lot of money into this by funding well-staffed multi-year research projects. There's plenty of public trail in terms of research papers. It's just that not a whole lot came out of it otherwise.

The problem isn't that the methods are underrated, it's that they aren't compatible with the approach to software engineering in these places (huge monolithic codebases, a variety of evolving languages and frameworks, no rigid constraints on design principles).

fakedang•11m ago
Can you ELI5 what formal methods are and how not the industry standard apparently? As a complete noob, from what I'm reading online, they're pretty much how you should approach software engineering, or really any sort of programming right?
IshKebab•21h ago
I don't think they are underrated. They are heavily used where they work really well and bugs have a very high cost (e.g. hardware design).

For the vast majority of software though they don't really make much sense because formally verifying the software is 10-100x more work than writing it and testing it with normal methods. And formal verification of software generally requires faaaaar more expertise than most people have. (The situation is much better for hardware design because it tends to be way simpler.)

It's a very powerful tool but also extremely difficult to use.

tgma•17h ago
> Pretty much no large companies except AWS (thank you Byron Cook!) use them at a large scale.

I don't think that's true at all. I suppose that depends on what you mean by formal methods and in what context you're concerned about those. Off the top of my head this comes to mind from Microsoft: https://learn.microsoft.com/en-us/windows-hardware/drivers/d...

Twirrim•16h ago
Microsoft, Amazon, Oracle, Google, all sorts of large companies use them, and talk about it, publicly. They've all published whitepapers and resources about them. Microsoft even employs Dr. Leslie Lamport who created and maintains TLA+ (among other things).

Just for some quick examples:

Microsoft: https://github.com/Azure/azure-cosmos-tla, https://www.youtube.com/watch?v=kYX6UrY_ooA

Google: https://research.google/pubs/specifying-bgp-using-tla/, https://www.researchgate.net/publication/267120559_Formal_Mo...

Oracle: https://blogs.oracle.com/cloud-infrastructure/post/sleeping-... (note the author is a "Formal Verification Engineer", it's literally his job at Oracle to do this stuff)

Intel: https://dl.acm.org/doi/10.1145/1391469.1391675, https://link.springer.com/chapter/10.1007/978-3-540-69850-0_...

Jetbrains: https://lp.jetbrains.com/research/hott-and-dependent-types/

Arm: https://ieeexplore.ieee.org/document/9974354

dfc•13h ago
Lamport+Microsoft was the first thing that I thought of when I read the comment. FWIW he retired at the beginning of this year.
nhatcher•21h ago
SAT solvers and the algorithms surrounding them are so much fun. I agree they are very unappreciated.

Shameless plug: I wrote a (admittedly very deriative) introduction with some examples I thought at the time were cool.

https://www.nhatcher.com/post/on-hats-and-sats/

cubancigar11•9h ago
Thanks that was quite informative, perfect for me.
cpatuzzo•20h ago
I tried to write a programming language that compiles to SAT many years ago: https://sentient-lang.org/
hwayne•20h ago
I remember you showing me this! Wow that was a long time ago.
superlopuh•6h ago
I love that language and frequently show it to people. I'm sad to see that my local install doesn't work any more. I actually used it to solve a puzzle in Evoland 2 that I'm relatively sure was added as a joke, and is not solvable in a reasonable time without a solver. I'm actually doing a PhD in compilers right now, and would love to chat about sentient if you have the time. My email is sasha@lopoukhine.com.
robinhouston•20h ago
If you want a language for expressing constraint satisfaction problems that's higher-level than SAT, I think MiniZinc is pretty interesting. https://www.minizinc.org/
osmarks•20h ago
I was briefly looking into using SMT for Minecraft autocrafting, but it turns out you can do integer linear programming and the mapping is easier.
b0a04gl•20h ago
you mentioned SMT is slower than SAT and left it there, but that feels incomplete. in problems like this, solve time barely matters unless you’re generating at scale. the real weight is in how fast you can write, refactor, and trust the constraints. SAT might give faster outputs, but SMT usually gets you to correct models quicker with less friction. wondering if you actually benchmarked both and stuck with SAT on numbers, or if it was more of a default comfort pick. felt like a missed moment to shift the lens from solver speed to model dev loop speed
stong1•19h ago
Reminds me of a small project I did back in undergrad: Minesweeper using a SMT solver. https://github.com/stong/smt-minesweeper
jononor•19h ago
How good are current LLMs at translating problems given as text into something SMT solvers can operate on? Be it MiniZinc, Z3, Smtlib, Python bindings, etc. Anyone tried it out?
Jaxan•19h ago
I tried it many months ago and it was garbage. But this was trying smtlib directly. Maybe via the python bindings it works better?
hwayne•18h ago
Apparently they're getting very good: https://emschwartz.me/new-life-hack-using-llms-to-generate-c...

I try not to use them too much because I want to build the skill of using SMTs directly for now.

Twirrim•16h ago
I've found them to be bad, for the most part. There aren't enough blog posts and examples of code out there for them to leach from.

Besides which, I would argue the process of writing proof in the language is integral to building the understanding you need to deal with the results. You'll spot bugs as you're creating the model.

naet•19h ago
I actually wrote a backtracking solution to the LinkedIn queens game a while ago (and the tango game).

I know nothing about SMT or SAT and I imagine they might be faster, but the backtracking appears to solve just as instantaneously when you push the button.

Might be cool to learn a bit about SMT or SAT. Not sure how broadly applicable they are but I've seen people use them to solve some difficult advent of code problems if nothing else.

gbacon•15h ago
Definite broad applicability.

NP-complete are the hardest problems in NP. Cook in 1971 proved SAT to be in NP-complete. In the worst case for any other problem in NP, we can quickly (i.e., in polynomial time) convert instances of that problem into instances of SAT. In other words, we can use SAT to solve any problem in NP.

It turns out there are many problems in NP-complete. The fast conversion applies among them too, so in some sense, problems in NP-complete are all the same because we can use them all to solve instances of each other. However, for some of those problem instances the best known algorithm is to try all possible inputs, which requires exponential time (very, very slow for even modestly large inputs).

Lots of research has been and continues to be poured into SAT because any gains automatically yield improvements to everything else in NP-complete and the rest of NP. Using a SAT solver allows you to hitch a ride more or less for free on the results of all that research. Each incremental improvement to SAT solvers benefits programs that use them.

As the author noted, forming SAT instances by hand can be a pain. SMT or SAT Modulo Theories is sort of a high-level language that “compiles down” to SAT. Expressing problems with SMT is more natural and reduces the burden of converting your problem to SMT and SMT solutions back to your problem domain.

spencerflem•19h ago
Some additional context: Outside of Microsoft, this puzzle is often known as Star Battle.

Terrific little puzzle, highly recommend it!

https://www.puzzles.wiki/wiki/Star_Battle

https://www.puzzle-star-battle.com/?size=5

refulgentis•17h ago
> Which is the correct solution to the queens puzzle. I didn't benchmark the solution times, but I imagine it's considerably slower than a raw SAT solver. Glucose is really, really fast.

I'm new to this area, neither the original article nor the link to Glucose have enough info to tell me order of magnitude here: milliseconds? hours?

zero_k•16h ago
Haha, Marijn Heule who is pushing a lot of limits of SAT solving would love this. If you manage to get him excited, he might spend a few years on this problem :) He's kinda famous for solving the Boolean Pythagorean Triples problem using SAT [1]. He loves puzzles. He also got Knuth excited about a bunch of fun puzzles.

BTW, these puzzles also tend to have a lot of symmetries, which SAT solvers are pretty bad at handling. You can break them, though, using a variety of techniques, e.g. static symmetry breaking [2], or symmetric learning.

[1] https://www.cs.utexas.edu/~marijn/ptn/ [2] https://github.com/markusa4/satsuma

anArbitraryOne•15h ago
What about a CP solver?

IBM Has a Roadmap to a 'Fault-Tolerant' Quantum Computer by 2029

https://www.wsj.com/articles/ibm-has-a-roadmap-to-a-fault-tolerant-quantum-computer-by-2029-91645d73
1•gmays•38s ago•0 comments

Webb telescope spots infant planets in different stages of development

https://www.dawn.com/news/1916812
1•Brajeshwar•3m ago•0 comments

I built an app that turns any topic into a 1-minute video

https://apps.apple.com/us/app/fenne-learn-anything-fast/id6746964450
1•prvn_•8m ago•1 comments

Fucoidans are senotherapeutics that enhance SIRT6-dependent DNA repair

https://www.researchsquare.com/article/rs-6613032/v1
2•bilsbie•8m ago•0 comments

The Hat, the Spectre and SAT Solvers (2024)

https://www.nhatcher.com/post/on-hats-and-sats/
7•todsacerdoti•10m ago•0 comments

Transfer – use Android as a local server

https://github.com/matan-h/Transfer
1•matan-h•11m ago•0 comments

Luxe Game Engine

https://luxeengine.com/
11•garrypettet•12m ago•2 comments

Tech Giants' New AI Ad Tools Threaten Big Agencies

https://www.wsj.com/articles/tech-giants-new-ai-ad-tools-threaten-big-agencies-75d54a8a
1•thm•13m ago•0 comments

A Simple Showcase for the Sea-of-Nodes Compiler IR

https://github.com/SeaOfNodes/Simple
2•surprisetalk•14m ago•0 comments

'We're done with Teams': German state hits uninstall on Microsoft

https://www.france24.com/en/live-news/20250613-we-re-done-with-teams-german-state-hits-uninstall-on-microsoft
8•taubek•15m ago•0 comments

Israel "had clear U.S. green light for Iran strike"

https://www.axios.com/2025/06/13/how-israel-executed-strike-iran-nuclear
6•cempaka•16m ago•0 comments

Parquet on Iceberg Outperforms MergeTree

https://altinity.com/blog/the-future-has-arrived-parquet-on-iceberg-finally-outperforms-mergetree
1•higeorge13•16m ago•0 comments

Ask HN: Why does EU institutions and member gov. websites have cookie banners?

3•karel-3d•16m ago•1 comments

Ask HN: Is ageism in tech still a problem?

4•leonagano•16m ago•1 comments

Show HN: 10k Lines of Basic

http://10klob.com/
1•tuveson•17m ago•0 comments

Show HN: Algochat – Real-Time AI Chatbots for Streamers

https://www.algochat.io
2•khanwave•18m ago•1 comments

NASA scientist tasked with identifying asteroids on collision course with Earth

https://www.abc.net.au/news/2025-05-31/nasa-officer-tasked-with-protecting-earth-from-asteroids/105354010
1•Tomte•19m ago•0 comments

Stuck? Build Your Language Backwards

https://jimmyhmiller.com/build-your-language-backwards
1•surprisetalk•19m ago•0 comments

What I talk about when I talk about IRs

https://bernsteinbear.com/blog/irs/
2•surprisetalk•20m ago•0 comments

RFC: type safe search params defined in routes.ts

https://github.com/remix-run/react-router/discussions/13800
2•atroxone•20m ago•0 comments

US Streetlights Are Turning Purple

https://www.scientificamerican.com/article/streetlights-are-mysteriously-turning-purple-heres-why/
1•surprisetalk•20m ago•0 comments

30 Minutes with a Stranger

https://pudding.cool/2025/06/hello-stranger/
2•surprisetalk•21m ago•0 comments

Honored to accept a direct commission as a Lt. Colonel

https://twitter.com/boztank/status/1933512877140316628
4•tiahura•22m ago•0 comments

Compiling C with Zig

https://www.mitchellhanberg.com/compiling-c-with-zig/
3•mhanberg•24m ago•0 comments

Delay, Interfere, Undermine

https://www.propublica.org/article/bukele-trump-el-salvador-ms13-gang-vulcan-corruption-investigation
3•JumpCrisscross•25m ago•0 comments

Ask HN: Casual Math Book Suggestions

2•vriendspookstem•25m ago•1 comments

Launch your WebSocket within seconds

https://cloudfully.io
2•firasabb•26m ago•0 comments

Show HN: Garlic – Java/Android decompiler written in C

1•neocanable•28m ago•3 comments

Why did Apple build Linux Container support?

https://micahhausler.com/posts/why-did-apple-build-linux-container-support/
2•micah_chatt•28m ago•0 comments

Just 2% of tidal and offshore solar energy could make a dent in CO₂ emissions

https://techxplore.com/news/2025-06-tidal-offshore-solar-energy-dent.html
3•PaulHoule•28m ago•0 comments