News for hackers. Stuff that matters.
Re:Mathematica on Raspberry Pi for free by lutusp on Thursday November 21, 2013 @02:44PM (#6776940)
> You don't need source to audit the results.

In fact, you do. Modern computer mathematical results require full exposure, nothing can be left to the imagination. This is true and understandable.

Going back to the first important computer mathematical result (the solution to the four-color map conjecture¹ in 1976), to be taken seriously, computer math results have been required to be fully transparent.

Imagine auditing a cancer cure without knowing anything about the methods or tools. Same idea.

1. http://en.wikipedia.org/wiki/Four_color_theorem

Re:Mathematica on Raspberry Pi for free by taliesinb on Thursday November 21, 2013 @03:00PM (#6777063)
I actually somewhat sympathize with your view. But I think you overstate the problem: if someone discovers something with Mathematica, they can then formalize it with an open proof assistant like Coq before they publish, and modern journals in fact encourage this sort of thing.

Really, it's not a zero-sum-game between open-source and commercial tools.

Re:Mathematica on Raspberry Pi for free by lutusp on Thursday November 21, 2013 @08:39PM (#6778735)
> Really, it's not a zero-sum-game between open-source and commercial tools.

Yes, there are mathematical results that can be validated independently. But the problem is not so much pure mathematics as numerical results. In that arena, the method by which the results were acquired must be accessible in detail. And much of modern applied mathematics relies on numerical results. Imagine producing a result of some value and being unable to explain in detail how it was arrived at, because it emanated from a closed-source program.

The four-color map problem I mentioned earlier was the first important case where this problem had to be dealt with, and it was resolved to everyone's satisfaction only because the full source was made available.

Re:Mathematica on Raspberry Pi for free by taliesinb on Thursday November 21, 2013 @10:53PM (#6779183)
You're jumping around a bit..

Did the four-color map problem use numerical results? I imagine it was just a pure combinatorial enumeration problem (I have not read the original paper, I may be wrong about that).

If you're doing a finite element method to validate a bridge design, you're hardly publishing a paper based on the result (though of course you will do plenty of validation and then times everything by 2 for good measure!).

What kinds of actual fields of mathematics are you thinking about, here?

I know that error analysis is a deep and arcane subject, and personally I have no idea how well or badly we do in that department, but I'm struggling to think of where the details of how we get it wrong or right is crucial for the development of published academic mathematics.

"Most" maths is totally abstract these days, categories and sheaves and topi and what not.

Re:Mathematica on Raspberry Pi for free by lutusp on Friday November 22, 2013 @03:48AM (#6779818)
> You're jumping around a bit.

In that case, anyone who uses more than one example to make his point is jumping around. If allowed to prevail, this non-argument would cripple debate.

> Did the four-color map problem use numerical results?

Yes, emphatically so. There were a certain finite number of possible maps (1936 to be precise), and each of them had to be tested explicitly. The reason for the requirement for full disclosure was that the tested set needed to be examined to assure that nothing germane was excluded, nothing included on specious grounds, and the manner of testing had to be closely examined. None of this could be reduced to a simple equation, and none of it yielded to a formal, concise mathematical proof.

> I imagine it was just a pure combinatorial enumeration problem (I have not read the original paper, I may be wrong about that).

If that were true, there would have been no role for a computer program -- it would have been published as a pure and formal result, like Fermat's last theorem. No one would have cared about the role played by computers, because a formal and pure proof could be examined by anyone sufficiently versed in mathematics.

> What kinds of actual fields of mathematics are you thinking about, here?

Any field that would benefit from use of a computer. By recent count, that's nearly all mathematical fields for one reason or another, including areas we once might have thought would have no use for a computer. Computer mathematics has gotten to the point where computers, laboring independently, come up with new theorems that humans have overlooked.

> I know that error analysis is a deep and arcane subject, and personally I have no idea how well or badly we do in that department, but I'm struggling to think of where the details of how we get it wrong or right is crucial for the development of published academic mathematics.

Most struggle with that, and part of the struggle is being able to see the detail of the algorithms used to arrive at a result. No detail, no trust, no science.

> "Most" maths is totally abstract these days ...

Not relevant. Relevant would be whether that abstract mathematics is or is not fully revealed in its details. For science, it must be.

This article --

http://www.newscientist.com/article/dn7286-computer-generate...

-- discusses the role of theorem-proving software like Coq, but includes a misleading statement by someone at Microsoft: "Microsoft hopes to develop a similar system for checking the logic used in computer programs, which could pre-empt some unforeseen bugs that cause programs to crash."

I hope this is an error on the part of the journalist who wrote the article, because it implies that Turing's halting problem, and Gödel's incompleteness theorems, which lie behind it, are potentially soluble. They aren't. All the more reason for full disclosure.

Re:Mathematica on Raspberry Pi for free by jonsterling on Friday November 22, 2013 @01:38PM (#6782761)
The statement regarding checking logic in computer programs is not misleading (though of course, your characterization of it is); there are many systems which are not subject to the halting problem in fact (at least in any way that affects practice ---> You can simply limit constructible terms to those which provably terminate, if you please). Statements like yours about Turing & Gödel are a common mistake.

You don't have to _decide_ whether a program halts in order to prove it. In many cases, termination is not so much even a proof obligation as it is a _fact_ by construction; in the harder case where termination is not obvious structurally, you can simply rewrite your general recursion as a call-graph, and then termination is a proof obligation which you must satisfy if you wish to execute it.

Decidability is much a much stronger thing than "provability in most relevant cases", and it turns out that the latter suffices. By Gödel, there are indeed "true" sentences which you cannot prove in a particular system, but I am willing to give you a hundred thousand dollars if you can actually find one and print it out. That is, Gödel's result was indeed sort of constructive, in that it did provide a scheme to construct such a sentence, but nobody does so because the term is so unfeasibly large.

There are many things which we should like very much to be decidable, some of which _can_ in fact be decidable (by Turing, termination is not one of those). For instance, in some systems (such as Intensional Type Theory), type checking is decidable, which is really nice. In others, type checking is undecidable (such as in Extensional Type Theory); does this mean that type checking cannot be done? Of course not. It simply means that type checking becomes a proof obligation, since the typing derivations may not be entirely synthesized by the type checker. (A simple demonstration of this is equality in a NuPRL-like system: propositional equality is reflected into definitional equality; there is not decidable equality for functions; therefore, there are equations which might hold definitionally by equality reflection, but for which further proof would be required; therefore, definitional equality is undecidable; therefore type checking is undecidable; but we can still prove that a term is well-typed, even if type checking is undecidable.)

Re:Mathematica on Raspberry Pi for free by jordigh on Thursday November 21, 2013 @03:21PM (#6777203)
"Open source" and "commercial" are not antonyms, but people working on proprietary software think they are.
Re:Mathematica on Raspberry Pi for free by comex on Thursday November 21, 2013 @07:14PM (#6778486)
Only if you use an overly literal interpretation of "commercial" yet don't use the literal interpretation of "open source" (since it's also possible to make source code visible to the public without using a license that would be considered "open source").
Re:Mathematica on Raspberry Pi for free by jlgreco on Thursday November 21, 2013 @07:26PM (#6778521)
You have just rediscovered the reason that some people talk about "free software", and some people talk about "open source".

The sort of access to source code that is being expected here is really no more than the sort of access that you can get to Windows code. It really is not an unreasonable expectation.

Re:Mathematica on Raspberry Pi for free by comex on Thursday November 21, 2013 @09:22PM (#6778861)
No - when most people use the term "open source", they mean something along the lines of the Open Source Definition (http://opensource.org/definition), rather than the literal meaning; this is independent from the distinctions between open source and free software, which are mostly about philosophy rather than the set of applicable licenses (http://www.gnu.org/philosophy/free-software-for-freedom.html)

I suppose that you, however, do not, which makes what I was trying to say irrelevant.

Re:Mathematica on Raspberry Pi for free by Someone on Thursday November 21, 2013 @07:11PM (#6778477)
Some people may think that, but I doubt that that applies here. Wolfram certainly knows that "proprietary" and "free" aren't antonyms.

I do think there is a commercial side to this, too, however. This is a kind of a hook to catch future customers of the commercial package.

Re:Mathematica on Raspberry Pi for free by throwawaykf on Friday November 22, 2013 @09:15AM (#6780766)
And a lot of people using open source products think so too :-)
Re:Mathematica on Raspberry Pi for free by thinkpad20 on Thursday November 21, 2013 @03:16PM (#6777175)
If what one is publishing is an algorithm or formula or what-have-you, it's only required that the algorithm itself be fully exposed. It particular implementation, much less so the implementation of the particular software that implementation is written in, is only incidental. When you're giving a driving test, you don't need to know how a fuel-injection system works.
Re:Mathematica on Raspberry Pi for free by lutusp on Friday November 22, 2013 @03:56AM (#6779837)
> It particular implementation, much less so the implementation of the particular software that implementation is written in, is only incidental.

Not at all. In computer programming, examples abound in which the gap between the formal algorithm and its practical embodiment led to catastrophe. One can scarcely count the number of times in which the specific manner by which a simply stated algorithm was implemented needed to be examined more closely:

http://en.wikipedia.org/wiki/Ariane_5

"Ariane 5's first test flight (Ariane 5 Flight 501) on 4 June 1996 failed, with the rocket self-destructing 37 seconds after launch because of a malfunction in the control software.[20] A data conversion from 64-bit floating point value to 16-bit signed integer value to be stored in a variable representing horizontal bias caused a processor trap (operand error) ..."

> When you're giving a driving test, you don't need to know how a fuel-injection system works.

No, but someone does. If the software is inaccessible, then no one can check it except its owners, who have an incentive not to check it thoroughly, or to conceal what they have discovered.

Re:Mathematica on Raspberry Pi for free by blhack on Thursday November 21, 2013 @06:30PM (#6778314)
Out of curiosity: how deep does that need go? Low level hardware analysis? Chip manufacturing facility audits?

How would you even approach this?

Re:Mathematica on Raspberry Pi for free by lutusp on Thursday November 21, 2013 @08:32PM (#6778717)
> Out of curiosity: how deep does that need go?

Deep enough that the meaning of the result cannot reasonably be called into question, just like in science.

> How would you even approach this?

As scientists. Someone once described science as ... not so much knowing as knowing that we know. Same idea.

Many results that seemed certain when published, evaporated once the mathematics were examined closely. This happens even when the entire process -- data, equations, conclusions -- are all included in the publication. Given that history, imagine how much more cautious people want to be when part of the process of arriving at a critical conclusion is concealed perpetually.

Re:Mathematica on Raspberry Pi for free by mkingston on Friday November 22, 2013 @04:53AM (#6779980)
"Deep enough that the meaning of the result cannot reasonably be called into question, just like in science."

I think what is really being asked is: "What level of confidence is required?".

I would assume you trust the hardware and only scrutinise the software? In contrast: are results tested on a plurality of machines with intentionally, measurably different hardware and the absence of discrepancies implies that the hardware is not a factor with regard to the problem being solved? Presumedly you'd need some sort of standardised test suite to pass on your hardware to really confidently declare the hardware to not be a factor.

I doubt this is the case, but like the previous poster, I'm quite curious as to the specific level of scrutiny applied here.

Re:Mathematica on Raspberry Pi for free by lutusp on Friday November 22, 2013 @05:50AM (#6780146)
> I think what is really being asked is: "What level of confidence is required?".

That's something that varies from field to field. In physics, a new discovery requires a five-sigma result, meaning a result that has a one in 3.5 million probability of arising by chance rather than the legitimate detection of a new particle or phenomenon.

http://www.select-statistics.co.uk/blog/webmaster/one-sigma-...

In psychology, by contrast, a p-value of 0.05 (i.e. 2 sigma) is accepted as statistically significant, which by itself may explain the sad state of the field.

http://en.wikipedia.org/wiki/P-value

> I would assume you trust the hardware and only scrutinise the software?

No, a scientist doesn't trust either hardware or software, but hardware can be checked and validated. In the case of Mathematica, the software cannot be checked and validated, it must be trusted. And for a scientist, trust cannot be distinguished from faith.

Re:Mathematica on Raspberry Pi for free by mkingston on Saturday November 23, 2013 @05:19PM (#6787666)
Thanks for your reply.

"No, a scientist doesn't trust either hardware or software, but hardware can be checked and validated." Do you know if or how this is done?

Re:Mathematica on Raspberry Pi for free by noonespecial on Thursday November 21, 2013 @08:04PM (#6778637)
Recalling the The Pentium FDIV bug, I'd say yes. Chip level.