Hacker News Reader

Comment by Asraelite | Hacker News Reader

I feel like the algorithm is maybe poorly named. At first I thought this was a tool to help with computational quantum physics.

View on HN

Wave Function Collapse is actually the same (or essentially identical) as Model Synthesis[1]. IMO that is a better name.

[1]https://paulmerrell.org/model-synthesis/

Taking a look at the comparison paper [0] provided on that site, there does seem to be a significant difference. The images generated with MS are more prone to long horizontal lines while WFC produces larger blocks of texture. This is hard to spot with more homogeneous textures like pipes, but it's more apparent with something like the grey cat faces, where WFC generates a variety of shapes and MS mostly makes very long ones. Going back up to section 2 I found the explanation: "Model synthesis sweeps through the grid in scanline order. WFC chooses the lowest entropy cell." The results from WFC look more natural and varied to me so I'd call this a significant advance.

But it does seem appropriate to say they're variations of the same algorithm. As for the name, I don't like either: Wave Function Collapse is based on a silly pop-science interpretation of quantum mechanics, and Model Synthesis is extremely generic and forgettable. And it's not even a synthesis, more of a filtering process. I guess I'd stick with Wave Function Collapse.

[0] https://paulmerrell.org/wp-content/uploads/2021/07/compariso...

I thought the same - though it actually gets its name from wave function collapse in quantum physics, as each pixel has multiple potential positions it can collapse to, leading to the procedurally generated map

As someone who has actually studied physics (including quantum mechanics) I find this analogy extremely poor. Keeping multiple potential candidates which you gradually reduce into a single solution is a very common approach for solving a variety of problems (e.g. sudoku solver). This is not at all what makes the wave function interesting or useful in quantum mechanics. And the whole concept that you iteratively collapse each "wave element" isn't a thing which is present in quantum mechanics at all.

Oh well, I guess people like the fancy name.

> Oh well, I guess people like the fancy name.

The way "quantum leap" is used by laypeople springs to mind too; literally the opposite of what they mean (not only tiny but random).

is it though? I've heard the argument, but to me it still makes sense as it is used colloquially. It is a discrete jump forward, thus if you are not talking about electrons and just the latin definition can be an arbitrary sized quanta if defined to an an appropriate set. a quantum leap between scientific epochs is quite large for example

also just reread your comment. I don't agree with the definition "small and random", it is smallest unit of measurement an arbitrary system magnitude can change

Here's my interpretation: a classical leap is when you are somewhere, you expend a bit of energy you have, and you end up somewhere else as a result. Basically, you do something and something happens.

A quantum leap, on the other hand, happens when you are somewhere just sitting, and boom, now you're somewhere else. It just happened randomly, without any energy input from you. So to bring this back to the idiom, an example of a quantum leap would be this: "I had a dream yesterday that gave me the insight I needed to finish my proof of XYZ theorem!"

The main example of a quantum leap that comes to mind is between energy levels of an electron of an atom, and that does involve emitting or absorbing a photon, carrying the energy that needs to be accounted for.

I don’t think what you’ve said is correct.

Quantum does not mean "small and discrete", it only means "discrete", as in "quantisation" - fitting values into a discrete defined set. For instance in mathematics the operation of "rounding" quantises fractions into the set of integers.

Yes, but the quantum leaps in physics are small.

I think it is (in this usage) meant as the gain in human understanding of the photoelectric effect (Einstein 1905), specific heat, diamagnetism, and so forth once quantum mechanics began development. The leap from classical mechanical theories being fundamental to quantum mechanical theories being fundamental (with classical physics emerging from that), in other words, was a big shift even though subatomic particles themselves, and their bound states, are generally very small.

Random doesn't really enter into it. One can get that in a firmly non-quantized theory, as in one of Einstein's other 1905 papers, https://en.wikipedia.org/wiki/Brownian_motion#Einstein's_the... https://history.aip.org/exhibits/einstein/essay-brownian.htm https://physicsworld.com/a/einsteins-random-walk/

It's also not clear what this algorithm has to do with quantum physics. Maybe that's how it was discovered or it has applications to physics, but the algorithm itself doesn't require any advanced math or have anything to do with the wave function. It can be described using elementary combinatorics.

It's reasonable to be surprised when a computer science term uses a metaphor, but I don't think it's fair to call it a poor name unless you also object to almost every other name in programming.

Strings have nothing to do with textiles. And, very confusingly, they aren't even related to fibers or threads. Zippers aren't related to textiles either.

Trees and bloom filters are not related to botany. Heaps and garbage collection are not about municipal waste management. Graph coloring, despite the term, has nothing to do with either pictures or pigments. Red-black trees aren't colored or botanical.

Bubble sort does not involve fluid dynamics. Neither does bucket or cocktail shaker sort for that matter.

Circular buffers are not round. You can't eat a spaghetti stack or poke your finger on a cactus stack.

Very naive question but is it even possible to do quantum experiments with code on a regular computer? Anyone knows if it's been tried?

You can simulate quantum dynamics but the number of bits required is exponential with respect to the number of qubits simulated. So basically you can simulate really simple quantum systems but it becomes pretty much intractable when the system you are simulating has many parts.

Yes. It's just an equation. The complexity of quantum system that can be simulate is limited.

https://marl0ny.github.io/QM-Simulator-2D/index.html

From the above link you can look single slit, double slit, triple slit, step, spike, or load an arbitrary image as a potential and do experiments in 2d box.

>This WebGL program simulates the quantum mechanics of a single particle confined in a 2D box, where inside this box the user can create new potential barriers and scatter Gaussian wavepackets off them. The full instructions are found here.

https://github.com/marl0ny/QM-Simulator-2D

btw. Wave function collapse in quantum physics is completely speculative phenomenon. There is only apparent wave function collapse.

> btw. Wave function collapse in quantum physics is completely speculative phenomenon. There is only apparent wave function collapse.

To be more precise, the Born rule non-linear adjustment of the wave function to a single real value after a measurement is strictly necessary for QM to match experiments. Whether this should be interpreted as a physical phenomenon of wave function collapse, or as entanglement with the environment (MWI), or as an update of probabilities for hidden variables (Pilot wave) or some other phenomenon is speculative, but the wave function must be "collapsed" to a single real value after a measurement to correctly predict experimental results.

Adjustment is just cutting interference terms from equation. Mathematically apparent wave function collapse is caused by quantum decoherence.

Wave-function collapse as a priori process is just speculation. Finding that it actually happens would be new physics.

In which case, where does the Born rule come from?

https://en.wikipedia.org/wiki/Gleason%27s_theorem

Huh. I can't believe I missed this. Thanks.

In quantum computing the consequence of the theorem (as I understand it) is that once you have asked every yes/no question there is to ask about the state, the state has collapsed.

I didn't think you needed the theorem for that; quantum entanglement behaves very similarly to information theory (specifically, information provenance).

As far as I can tell, that is just explaining the numeric value of the Born rule, not the wave function collapse/update. I have not found any claims that Gleason's theorem (which I was unaware of, so thank you for pointing it out!) solves the measurement problem.

Quantum decoherence may give you a diagonal density matrix but you still need to "collapse" that to a single outcome in some way or another.

However you put it, [to a classical observer] the wave function still "collapses" after a measurement. This is most famously seen by adding a detector inside one of the slits for the double-slit experiment: the original wave function is not consistent with the experiment, you have to update the wave function after the interaction (or lack of interaction) with the detector.

Sure, in MWI the wave function of the universe never collapses, but something similar still happens for "parts" of the universal wave function.

If you like that simulation you might also like my own webGL Gross-Pitaevskii Equation (GPE) solver. It uses RK4 to simulate a 2D box of ultracold atoms undergoing Bose-Einstein Condensation.

The GPE models a condensate as a single-particle quantum wavefunction with a non-linear form of the Schrödinger Equation, so you get some interesting behaviour from the non-linearity while the simulation remains computationally feasible.

You can interact with the potential term by clicking and dragging inside the 2D box.

https://georgestagg.github.io/webgl_gpe/

Sure. We’ve been doing quantum theory with pencil and paper for over a century!

This gets brought up every time its mentioned. And I think its interesting. It's only tangentially related to actual physics. But parallelizing the WFC for large inputs might be exactly the kind of problem QC assists.

Reply on news.ycombinator.com