Re:Object Oriented Mathematics (1995) [pdf]
by jhanschoo on Monday December 14, 2015 @12:16AM
(#10729521)
The author seems to write from the perspective of an undergraduate whose main exposure is limited to lecture notes and handouts.
I find that most pedagogical mathematical literature actually do lay out what prerequisite knowledge they expect of the reader, and make sure to identify the exact nature of symbols used in each expression, contrary to the experience of the author.
Mathematics today is still very much prose, in which equations are shorthand for common, repetitive expressions.
Nevertheless, the ambiguity the author observes is something I commonly find in lecture notes, where the prose and context around certain theorems and equations are stripped away for brevity.
Re:Object Oriented Mathematics (1995) [pdf]
by ajarmst on Sunday December 13, 2015 @10:06PM
(#10729197)
The author could have benefitted significantly from some exposure to Abstract Algebra, as he seems completely unaware of pretty rudimentary things in Field and Group Theory. The author might also benefit from some exposure to an actual mathematician, judging by the laughable canards he levels at mathematicians and their lack of "rigour" in his abstract. Finally, the author would benefit from some exposure to actual published papers, judging by his apparent unfamiliarity with what abstracts are for.
Re:Object Oriented Mathematics (1995) [pdf]
by pervycreeper on Sunday December 13, 2015 @11:27PM
(#10729417)
His fundamental conceit, namely that mathematics sometimes suffers from ambiguous or sloppily used notation, is a valid one, and this fact has caused a great deal of confusion for a great number of people.
Re:Object Oriented Mathematics (1995) [pdf]
by ajarmst on Monday December 14, 2015 @12:00AM
(#10729492)
That's a reasonable point, but I wonder if that problem is usually encountered when mathematicians from a particular branch of study read works from a different one whose usages and assumptions differ. I've bounced of it in even very simple areas, like whether 0 is a natural number, whether is is positive or negative (or other), whether 1 is prime, etc. Within the context of a particular community, such ambiguities aren't present.
Re:Object Oriented Mathematics (1995) [pdf]
by archgoon on Monday December 14, 2015 @01:16AM
(#10729672)
"This works perfectly well for communication between researchers in the same field, but may be an obstacle
for communication between researchers from different fields or for newcomers such as students"
It appears you did not read the abstract after all.
Re:Object Oriented Mathematics (1995) [pdf]
by ajarmst on Sunday December 13, 2015 @10:49PM
(#10729332)
I usually give the benefit of the doubt and assume ignorance, not incompetence. If he was familiar with Abstract Algebra, then the omission of any reference to groups and fields in the examples he gives seems bizarre, as are the pompous declarations in the "abstract" that doesn't abstract his arguments or findings.
Re:Object Oriented Mathematics (1995) [pdf]
by thwd on Monday December 14, 2015 @12:13AM
(#10729516)
On page 4 he does make a reference to the abelian group (Z, +). Though he just calls it "a structure" (which is not wrong), I agree that this paper is somewhat unusual. It even includes some typos ("yeild," page 4), yet it is readable and exposes an interesting topic; the ambiguity of mathematical notation.
Re:Object Oriented Mathematics (1995) [pdf]
by ajarmst on Sunday December 13, 2015 @11:07PM
(#10729370)
Fair enough, and thanks for reminding me of that paper, which is a gem. I'd argue that Lamport makes a convincing case with evidence and proposes changes that would rectify the evidenced issues. This author simply asserts it as a fact and then introduces an odd notation with little concrete justification. I also think Lamport's critique is less of the abstract symbolic systems used in mathematical proofs, but a lamentable tendency to use prose.
Re:Object Oriented Mathematics (1995) [pdf]
by Animats on Monday December 14, 2015 @12:09AM
(#10729510)
Way too many years ago, I was trying to invent what should have been "object oriented constructive mathematics", but we didn't have object oriented programming yet.
There was an actual reason for this. I was working on program verification [1], and we'd put together a system which used the Oppen-Nelson prover in combination with the Boyer-Moore prover. We needed to prove that their theories were consistent.
The Oppen-Nelson prover is a complete, fast, automatic, decision procedure for expressions composed of addition, subtraction, multiplication by constants, conditionals, logical operators, and structure and array access. This subset of mathematics is completely decidable. (If you add multiplication of two variables, it becomes undecidable.) It can also accept "rules", which are identities that it accepts as true. In our system, the Boyer-Moore prover was used to prove any new rules needed, which could then be imported into the Oppen-Nelson prover. Anything complicated involving loops usually required a new rule.
The Boyer-Moore system is completely constructive, and is based on recursive functions. Numbers are defined as (add1 (add1 (add1 (zero)))), for example. One can then write recursive functions for addition and subtraction, and work up to multiplication and division. A few hundred theorems cover basic number theory.
Arrays, though, were a problem. There are four classic axioms, from McCarthy, which define basic array semantics. The Oppen-Nelson prover has those built in, but the Boyer-Moore system does not. We thus wanted to prove them in the Boyer-Moore system. If we could do that, it became safe to prove new things about arrays and import those rules into the Oppen-Nelson prover.
The axioms: arrays have two operations, SELECT and STORE. SELECT(array, index) returns the appropriate element from an array. STORE(array, index, newvalue) returns a new array where newvalue has replaced the element previously at index. We then have rules such as
SELECT(STORE(A, I, V), I) = V // what you store, you get back
or, in Boyer-Moore notation:
(implies (and (arrayp! A) (numberp I))
(equal (selecta! (storea! A I V) I)
V))
The rest can be seen at p. 129 of the manual[1] if you care.
Arrays had to be defined in the Boyer-Moore system as a list of (subscript, value) tuples. Not a set, a list. A set isn't a constructive construct, because, informally, a set is a collection of unique values, the order of which is not significant. In the Boyer-Moore world, two values are equal iff they are identical. Two sets of tuples with a different order would not be equal.
So we had to define an array in the Boyer-Moore world as a list of (subscript, value) tuples ordered by increasing subscript. This is a clunky notation, because then we have to prove that the STORE operation preserved the correctness of the ordered list. Then we had to prove that all the rules for arrays were always true for that clunky representation. This required about 50 pages of machine proofs.
Back then (this was around 1981-1982), long machine proofs were not acceptable in mathematics. I had a JACM paper rejected for that reason. The approach was just too ugly.
Years later, I realized that what was needed was a kind of object oriented version of constructive mathematics. The key concept is that two things are equal if there is no way they can be distinguished. This comes from the theory of uninterpreted functions:
forall f, x: f(x) = f(y) implies x = y
So we would like to be able to define an type with public and private functions, one which exposes a new "equal" operation for the type. Then, if we can prove that the new "equal" function obeys the rule above for all public functions, and we disallow all further access to the private functions, we can construct a consistent theory with a new, more abstract notion of "equal". Now we can write set theory in Boyer-Moore theory without adding new axioms.
Unfortunately, I figured this out about a decade too late. We didn't really need that result to get a valid verification system, but it would have cleaned up the theory and made it publishable. But anyway, there's a form of object oriented mathematics which could be potentially useful.
The verification system was never used much; it was for a dialect of Pascal for Ford engine control programs that was never used in production. We looked at extending it to Ada (too hard) and C (too ill-defined). DEC SRL did a similar system for Modula III, but that died with DEC SRL, DEC, and Modula III. Some of the ideas were reused, decades later, in Microsoft's Spec#.
Re:Object Oriented Mathematics (1995) [pdf]
by chris_wot on Friday December 18, 2015 @10:15PM
(#10762455)
You wrote that:
forall f, x: f(x) = f(y) implies x = y
What if f(x) is x^2 and f(y) is y^2? You could have x=2 and y=-2.
I'm sure there is something I'm missing, but wouldn't that make the proposition false? i.e. f(x) = f(y) is true, but x = y is false, thus true implies false is false
Re:Object Oriented Mathematics (1995) [pdf]
by Animats on Saturday December 19, 2015 @02:26AM
(#10762945)
The key is "forall f". That is, this must hold for all functions. In a constructive world, the set of all functions is finite. There's an existing set of known functions on the object. Any new functions must be created using those functions, so if the known functions have that property, so must functions built from them.
The key idea is that new functions can't look inside the object directly. Think private data and function members in C++. So two objects may have different internal representations, but if the exported functions won't let you distinguish them, you can treat them as equal.
Re:Object Oriented Mathematics (1995) [pdf]
by chris_wot on Saturday December 19, 2015 @07:16AM
(#10763377)
So if the result of all values of x of the function f and g then the functions f and g are equivalent. Is that correct? If so, then does the theory handle encapsulation only?
Re:Object Oriented Mathematics (1995) [pdf]
by kmicklas on Sunday December 13, 2015 @08:23PM
(#10728841)
This guy really needs to learn more type/category theory, or Haskell, or something. OOP may have been hip and cool in 1995 when this was written but applying it to math is an even worse idea than for programming.
I find that most pedagogical mathematical literature actually do lay out what prerequisite knowledge they expect of the reader, and make sure to identify the exact nature of symbols used in each expression, contrary to the experience of the author.
Mathematics today is still very much prose, in which equations are shorthand for common, repetitive expressions.
Nevertheless, the ambiguity the author observes is something I commonly find in lecture notes, where the prose and context around certain theorems and equations are stripped away for brevity.
It appears you did not read the abstract after all.
http://www.diku.dk/~grue/
Looking at his CV, he might even have had taken a few classes in Abstract Algebra.
Pooh-pooh.
https://en.wikipedia.org/wiki/Pooh-pooh
Also, ad hominem.
http://research.microsoft.com/en-us/um/people/lamport/pubs/p...
There was an actual reason for this. I was working on program verification [1], and we'd put together a system which used the Oppen-Nelson prover in combination with the Boyer-Moore prover. We needed to prove that their theories were consistent.
The Oppen-Nelson prover is a complete, fast, automatic, decision procedure for expressions composed of addition, subtraction, multiplication by constants, conditionals, logical operators, and structure and array access. This subset of mathematics is completely decidable. (If you add multiplication of two variables, it becomes undecidable.) It can also accept "rules", which are identities that it accepts as true. In our system, the Boyer-Moore prover was used to prove any new rules needed, which could then be imported into the Oppen-Nelson prover. Anything complicated involving loops usually required a new rule.
The Boyer-Moore system is completely constructive, and is based on recursive functions. Numbers are defined as (add1 (add1 (add1 (zero)))), for example. One can then write recursive functions for addition and subtraction, and work up to multiplication and division. A few hundred theorems cover basic number theory.
Arrays, though, were a problem. There are four classic axioms, from McCarthy, which define basic array semantics. The Oppen-Nelson prover has those built in, but the Boyer-Moore system does not. We thus wanted to prove them in the Boyer-Moore system. If we could do that, it became safe to prove new things about arrays and import those rules into the Oppen-Nelson prover.
The axioms: arrays have two operations, SELECT and STORE. SELECT(array, index) returns the appropriate element from an array. STORE(array, index, newvalue) returns a new array where newvalue has replaced the element previously at index. We then have rules such as
or, in Boyer-Moore notation: The rest can be seen at p. 129 of the manual[1] if you care.Arrays had to be defined in the Boyer-Moore system as a list of (subscript, value) tuples. Not a set, a list. A set isn't a constructive construct, because, informally, a set is a collection of unique values, the order of which is not significant. In the Boyer-Moore world, two values are equal iff they are identical. Two sets of tuples with a different order would not be equal.
So we had to define an array in the Boyer-Moore world as a list of (subscript, value) tuples ordered by increasing subscript. This is a clunky notation, because then we have to prove that the STORE operation preserved the correctness of the ordered list. Then we had to prove that all the rules for arrays were always true for that clunky representation. This required about 50 pages of machine proofs.
Back then (this was around 1981-1982), long machine proofs were not acceptable in mathematics. I had a JACM paper rejected for that reason. The approach was just too ugly.
Years later, I realized that what was needed was a kind of object oriented version of constructive mathematics. The key concept is that two things are equal if there is no way they can be distinguished. This comes from the theory of uninterpreted functions:
So we would like to be able to define an type with public and private functions, one which exposes a new "equal" operation for the type. Then, if we can prove that the new "equal" function obeys the rule above for all public functions, and we disallow all further access to the private functions, we can construct a consistent theory with a new, more abstract notion of "equal". Now we can write set theory in Boyer-Moore theory without adding new axioms.Unfortunately, I figured this out about a decade too late. We didn't really need that result to get a valid verification system, but it would have cleaned up the theory and made it publishable. But anyway, there's a form of object oriented mathematics which could be potentially useful.
The verification system was never used much; it was for a dialect of Pascal for Ford engine control programs that was never used in production. We looked at extending it to Ada (too hard) and C (too ill-defined). DEC SRL did a similar system for Modula III, but that died with DEC SRL, DEC, and Modula III. Some of the ideas were reused, decades later, in Microsoft's Spec#.
[1] http://www.animats.com/papers/verifier/verifiermanual.pdf
I'm sure there is something I'm missing, but wouldn't that make the proposition false? i.e. f(x) = f(y) is true, but x = y is false, thus true implies false is false
The key idea is that new functions can't look inside the object directly. Think private data and function members in C++. So two objects may have different internal representations, but if the exported functions won't let you distinguish them, you can treat them as equal.