24

Say we have some finite set, and some theory about a set, say "All elements of the finite set $X$ satisfy condition $Y$".

If we let a computer check every single member of $X$ and conclude that the condition $Y$ holds for all of them, can we call this a proof? Or is it possibly something else?

JMCF125
  • 2,738
  • 1
  • 20
  • 34
Igglyboo
  • 353
  • 3
  • 5
  • 49
    Yes.​​​​​​​​​​​ – user2345215 Mar 18 '14 at 23:33
  • 1
    The older generation of pure mathematicians would prefer a concise, elegant proof. But yes, if you find the condition is satisfied for all elements then it is a proof. –  Mar 18 '14 at 23:42
  • 19
    @MitchKnight Most all mathematicians prefer elegant proofs to brute-force proofs (no matter the source). This preference has nothing to do with age. – Bill Dubuque Mar 18 '14 at 23:46
  • 2
    That's true. I was just suggesting that the use of computers to solve proofs is relatively new with regard to solve mathematical problems. –  Mar 18 '14 at 23:47
  • I was told that there is some controversy behind computer assisted brute force 'proofs' because it's not entirely rigid, you're using a machine that isn't garantueed to never fail. – camel Mar 19 '14 at 00:32
  • 5
    @camel: Human brains are certainly not failproof either. Computer proofs are verified exactly the same way any proof is: by having enough eyes on it until it is "pretty certain" to be right. Mathematical proofs have gone decades with flaws in them, only to be found later. It doesn't invalidate the endeavor of mathematics. – ex0du5 Mar 19 '14 at 00:35
  • 1
    Humans are not failproof, but you can rigorously check the logical assumptions they made (although it can be hard). If the computer is making the mistake (maybe there's a hidden software bug), it'd be impossible to check because some bugs can be due to the computer/compiler instead of the source code (you can't prove rigorously that it's getting executed correctly)

    See wikipedia for the four color theorem as an example, there is-was controversy

     – camel Mar 19 '14 at 00:39
  • 8
    The four color theorem was proved in that way (http://en.wikipedia.org/wiki/Four_color_theorem) – Quimey Mar 19 '14 at 03:42
  • You must be more clear in the question. If I have convinced myself that a statement is true, by whatever method, then this is not in itself a proof. A proof is a (written) trace of my reasoning that could lead somebody else to also be convinced of the truth of the statement. If an exhaustive computer search produces such a trace, then that may be considered a proof. The actual program text itself, if it can be easily checked and run, may also be quite convincing; I'm not sure I would call it a proof though. – Marc van Leeuwen Mar 19 '14 at 15:57
  • 8
    @camel brute force proofs, in general, are frowned upon not just because the chance of errors is much greater (which is one reason), but also because it tends to imply that there is no fundamental mathematical "reason" why the proof works, it "just so happens to work" for every possible case. IOW, there is a vague notion that a brute force proof is "a valid proof of a now provably-boring statement". – KutuluMike Mar 19 '14 at 16:45
  • @MarcvanLeeuwen I could write source code and then write a formal proof that says: "if the output of this program is X, then the assertion is true". Proving correctness is a big thing in Computer Science. – Cruncher Mar 19 '14 at 17:10
  • 3
    One common example of this kind of proofs is in boolean logic. E.g. prove that (a & b) | c equals (a | c) & (b | c). You simply write the truth table, enumerate all cases and assert the equation. – Shahbaz Mar 19 '14 at 17:51
  • lots of examples/refs on recent empirical/experimental math/TCS research – vzn Mar 19 '14 at 20:27
  • 1
    Related: Why are mathematical proofs that rely on computers controversial? – doppelgreener Mar 19 '14 at 23:22
  • Complete induction is in a way a brute-force method without doing all the work, isn't it? ;-) And it works on infinite sets. – Peter - Reinstate Monica Mar 20 '14 at 10:55

5 Answers5

43

Yes, you can. This method is known as proof by exhaustion.

Also, see computer-assisted proof.

Edit: As others have noted, this of course works only for finite sets.

foobar1209
  • 1,047
  • 10
    To nitpick, it does work for infinite sets (brute forcing, not proof by exhaustion) if you only n0eed to prove existence. – TC1 Mar 19 '14 at 11:23
  • 2
    @TC1 True, but I was really thinking more of what the asker was talking about. – foobar1209 Mar 19 '14 at 16:36
  • 4
    @TC1 note that proving existence is recognizable, but not decidable using brute force. (That is, if it's true, you will eventually find it, but if it's false it will run forever, and at no point can you determine "it definitely doesn't exist" as it may find it on the next iteration) http://math.stackexchange.com/questions/25802/recognizable-vs-decidable – Cruncher Mar 19 '14 at 17:12
  • 2
    @Cruncher Of course, that's why I said "brute forcing, not proof by exhaustion". Obviously you can't exhaust an infinite set. :) – TC1 Mar 19 '14 at 17:17
  • 6
    @TC1 Chuck Norris can :D – Spook Mar 20 '14 at 07:33
15

The more general question is, when is anything a proof.? There are a least two answers, but the one I think is relevant here is that something is a proof if it is a persuasive argument that someone skilled in the art will find convincing. In this sense, your finite enumeration is a proof, if:

  1. There is a persuasive argument that your enumeration of the cases is in fact exhaustive
  2. The argument for each case is persuasive

There was a lot of controversy in the 1980s about the Haken-Appel proof of the four-color map theorem, which states that that every map in the plane can be colored with only four colors so that no adjoining regions are the same color. Haken and Appel had an argument that showed that every map could be reduced to one of a few thousand cases, an argument that showed that if a case satisfied certain conditions then the corresponding maps could be four-colored, a computer enumeration of the several thousand cases, and a computer-generated demonstration that each case had the desired property.

The arguments were checked and now everyone agrees that this was a proof. But for a while it wasn't clear.

MJD
  • 65,394
  • 39
  • 298
  • 580
9

Yes, but some mathematicians think of it as an inferior way of proving things. They think that it's not "pure" and it doesn't really explain(on a deeper level) about what's going on.

One of the most famous computer-assisted proofs is the proof for the four color theorem.

EDIT:

Also note that this technique does not work for infinite sets(I personally think this is one of the biggest reasons why some prefer a more traditional proof). There are conjectures in number theory that are true for a huge huge huge chunk of numbers but are ultimately false.

  • 4
    Funny that you say "more direct". I'd say exhaustion is the most direct way possible, though it's certainly not always the most efficient / concise / elegant way. – leftaroundabout Mar 19 '14 at 01:53
  • @leftaroundabout You're certainly right. I'll change the wording a bit. –  Mar 19 '14 at 03:26
6

Using a computer to brute-force can be the first step to a proof. The next step is to prove that the program is correct!

A few ways you might do this are:

  1. Have the program output a proof for each member of the set. We can then check these proofs without having to trust the program at all. We could even send them all through an automated proof checker, which of course would also need to be proven correct! This may be worth doing, since proof checkers are generally simpler (easier to prove correct) and more general than proof finders; you might output proofs in a format understood by an existing proof checker.
  2. Prove that the program is correct for each member of the set. This might defeat the point of using a program in the first place!
  3. Prove that the program is correct for all possible inputs. This can be a good strategy, since the program only needs to simplify the problem, it doesn't need to solve it. For example, we might prove that our program returns "TRUE" if our property holds and "FALSE" if it doesn't; we specifically don't have to prove the more difficult part about which one it will actually return. To do that, we just run it.
Warbo
  • 257
  • 5
    What a spoil-sport! Prove the program correct. Then, prove the compiler correct. Then prove the design of the CPU correct. And so on... – GEdgar Mar 19 '14 at 12:40
  • Then prove the proof is correct. – Lazar Ljubenović Mar 19 '14 at 17:25
  • @GEdgar So of course this is going to rest on some assumptions. Hell, there's no guarantee that the transistors work... These happen to be assumptions that we base on every day. If we run the same code, on many platforms, using many compilers and/or interpreters, different processors and architectures and produce the same result every time... I'd say we're looking pretty good. – Cruncher Mar 19 '14 at 17:27
  • 1
    @GEdgar FYI, the compiler step can be skipped, and that's probably the most likely place for this to fail. – Cruncher Mar 19 '14 at 17:28
  • We have a huge amount of empirical data for the behaviour of compilers, CPUs and transistors; they're not perfect, but many of their bugs are known and can be avoided. We have exactly no data about the brand-new program I just wrote to help with a proof, so we have no reason to trust it. In fact, all of the empirical data from software engineering says that it will be buggy, hence we have specific reason not to trust it! – Warbo Mar 20 '14 at 10:02
  • @GEdgar Really, the chances of software/hardware errors conspiring to make an incorrect theorem seem correct by incorrectly running a provably correct program seem-- small. A lot smaller than the chance of having a bug in a traditional proof, in fact. Especially if there are multiple interpreters/compilers for the language, and it can be run on multiple types of CPU and operating system, and it is published in a way to be repeatable by others, then I see no realistic need for deeper proofs. – MarnixKlooster ReinstateMonica Apr 05 '14 at 07:01
3

Exhaustive checking is certainly a valid method proof, that is, everyone agrees that it is a proof.

It is usually considered a not very satisfying proof, though, because it usually fails to produce any insight why such-and-such is true. It just observes that it is true, but mathematicians are generally interested in understanding rather than mere facts, and that is something you don't get from an exhaustive search.

In contrast an "elegant" proof will generally tell you not only that such-and-such is true but will also give you an understanding why it has to be true.

hmakholm left over Monica
  • 286,031