6 ms·
> I don't claim to have solved P=NP > nor do I claim to have an algorithm. But you know someone who claims to be able to solve subset-sum problems quickly.
by ColinWright 11y ago
> I don't claim to have solved P=NP
> nor do I claim to have an algorithm.
But you know someone who claims to be able to solve subset-sum problems quickly. I have a graph to 3-color, I wonder if I should convert it into a subset-sum problem for them to solve. That would take some time, and I'm pretty sure it won't be worth it. I'll have a think.
> The word "theoretical" was only used
> to distinguish between problems and
> instances.
OK, noted.
> The question is,
> "Solving which instances, in which order,
> would lead to the most positive outcomes?",
> where "positive outcomes" is intentionally
> ill-defined so that the answerer can inject
> their own perspective as well.
Now I'm starting to get some idea of what you're asking. Give me some time (I'm about to go into a string of meetings) and I'll see if I can phrase it differently to see if I really do know what you're asking.
> I think the question is also interesting
> because people often talk about how great
> the world would be if P=NP and we could
> easily solve NP problems.
Actually, people mostly talk about how the world would be much better if NP != P.
> However, this talk is typically only
> "theoretical" (about problems and not
> instances).
Of course - people are researching how to solve the problems, specific instances are specific instances, whereas the research challenge is algorithms for the problem.
> If someone can claim the world would
> be better off with faster solutions,
> they should be able to provide an
> instance whose solution would lead
> to a better world.
Why? Often we don't know the true benefits of an advance in science or technology until well after it's done, dusted, commercialized, and in the hands of ordinary people. Given that there currently are no ways to solve big, hard instances of NPC problems, why should people bother to find specific instances?
- jsprogrammer 11y ago>Actually, people mostly talk about how the world would be much better if NP != P. I know I've seen some arguments stating as such. If you know of a well-founded one, I'd be interested in reading it. Personally, I don't know or have a strong opinion. Seems like either could be viewed as "better" in some way. >But you know someone who claims to be able to solve subset-sum problems quickly. I have a graph to 3-color, I wonder if I should convert it into a subset-sum problem for them to solve. That would take some time, and I'm pretty sure it won't be worth it. I'll have a think. Well, I actually wrote up some tests for him and he ran his algorithm against it. I also gave him a ~10MB text file containing a very large list of integers (I forget the range at the moment. I could find it if you'd like, but I'm not sure how meaningful it is.) with a randomly selected target sum. He provided the solution (which I have not verified) in ~6-9 months. He also ran through many thousands of problem instances (subset sum...in JSON arrays) with varying parameters (number of integers; range of integers) over the course over several days. I made a rudimentary plot of the parameters and timings and in my very limited analysis, it did appear that there was a sub-exponential trend (at least for the instances that he provided solutions to [:)]). I've been searching for the results (all instances, solutions, and timings) that I had collected from the test machine, but I haven't been able to find them yet. However, all of the test code is on GitHub and packaged into Docker containers for easy deployment of the test server and basic visualization of the results. I'd guess he (the person who claims a P=NP proof and sub-exponential-time subset sum solver) would be willing to run his algorithm against the test server again.
- ColinWright 11y ago>> Actually, people mostly talk about how >> the world would be much better if NP != P. > Seems like either could be viewed as > "better" in some way. Indeed, it depends on your internal axioms about "better". I don't have strong feelings either way either, and my feelings don't affect what's true and what isn't. The problem is open, remains open, and we deal with what's in front of us. >> But you know someone who claims to be >> able to solve subset-sum problems quickly. > ... He also ran through many thousands > of problem instances ... with varying > parameters ... it did appear that there > was a sub-exponential trend ... Unfortunately that means exactly nothing. The whole point is that for many (perhaps most) problems, the proportion of hard instances gets vanishingly small, and yet those are the ones that we often need to solve. Showing a sub-exponential trend for randomly chosen instances, or worse, instances he's chosen to show you, is no evidence at all. >> I have a graph to 3-color, I wonder if >> I should convert it into a subset-sum >> problem for them to solve. > I'd guess he (the person who claims a > P=NP proof and sub-exponential-time > subset sum solver) would be willing > to run his algorithm ... Perhaps that would be interesting. I'll look into how easy the conversion might be.
- jsprogrammer 11y ago>Unfortunately that means exactly nothing. The whole point is that for many (perhaps most) problems, the proportion of hard instances gets vanishingly small, and yet those are the ones that we often need to solve. Showing a sub-exponential trend for randomly chosen instances, or worse, instances he's chosen to show you, is no evidence at all. I wouldn't say exactly nothing. If no one can present an algorithm than can do at least as good on the same instances...that's interesting. He did not choose the instances. They were generated using multiple calls to a built-in pseudo-random function. The instances were generated when he requested an instance and the timer started when the instances were transmitted to him. The timer was stopped when a response was received and the response was then verified. He never submitted an incorrect answer. [Aside: There is a particular web page of a university professor that lists a smallish subset sum problem and claims that the naive algorithm would take some 10^large_number years to find the answer. His algorithm solves that particular instance in some very short period of time (seconds or less I believe). Is that particular instance "hard"? I don't know. Apparently not. He likes to use that instance as an example of his algorithm solving a "hard" problem. My problem is that I don't know how to verify if an instance is hard (or if such an analysis is even possible/valid). My instincts say that it very much depends on the "density of solutions" in the particular instance. Maybe that instance has lots of solutions and so a random walk will easily find one? On the other hand, maybe there is a limit to the number of solutions that can exist in a well-formed instance?] While I do not understand his entire proof (and I do believe he has intentionally withheld a portion of the proof [and certainly a key piece of the algorithm (if it exists)]), he does point to some interesting things (such as an apparent fractal pattern in the sum of all sets in the powerset of a set of integers). He claims a function that provides an ordering over the sums of the sets in the powerset, which can apparently be computed on-demand in polynomial time of the original set. A binary search is then applied over the ordering. In my own theory, it seems like it could be possible; after all, there isn't an exponential number of items, only combinations. >The whole point Sorry, the whole point of what? NP-Completeness?