As a child, I noticed that the proofs of mathematical theorems were esoteric knowledge, known only to a few adults. I struggled to follow even the simplest proofs, and hoped that one day I might learn to create a proof or two of my own. This was not only a high aspiration, but a dangerous one. I saw no reason why certain knowledge of a true fact would be accessible to humans via proof. Any-one who embarked on the quest to find a proof risked embarking on a doomed quest to find a non-existent proof.
For me Gödel's completeness theorem is the miracle. Every valid statement has a proof. Amazing!
Aim a little higher, every true statement, and there might not be a proof. It is no surprise to me that this is true. It is a big surprise to me that Gödel was able to prove it; ordinary proofs are hard to find, and proofs of the limits of provability presumably even more deeply hidden.
Non-standard models of arithmetic are weird. Theorems that are true of the standard model of arithmetic and false in some non-standard model must surely be convoluted and obscure. The first order version of the Peano axioms nail down the integers, not perfectly, but very well. Restricting one-self to theorems that are true in all models of them, even the weird, non-standard ones, feels like a very minor restriction. Gödel's completeness theorem raises the possibility of writing a computer program to find a proof of every theorem that isn't convoluted and obscure. Gödel completeness theorem is the really big deal.
Except it isn't. That computer program turns out to be one of those wretched tree search ones that soon bogs down. The real problem turns out to be the combinatorial explosion inherent in unstructured search through the Herbrand universe. One needs Unification and one needs a still missing ingredient to give search a sense of direction. The interesting questions are about the "sense of direction" that lets us find some of the deeply hidden proofs that do exist. Will LLM's help? The answer will be interesting, either way.
* There are axioms, there are models, and there are theorems.
* A model is a particular structure with objects and relationships. The "standard model of arithmetic" is just the natural numbers 0, 1, 2, ... with normal rules of addition and subtraction and so on. A different model could be the real numbers, or one that includes infinitesimally small numbers, or so on.
* A set of axioms are rules about how a model can work. For example, we can have an axiom for any set of objects called "numbers" with an operation called "addition" that the operation must be commutative (x+y = y+x). The standard model above is one model that satisfies this axiom.
* A theorem is a fact that can be true or false about a given model. For example, "the model has infinitely many objects." If we can prove a theorem from a set of axioms, then that theorem is true for every model that satisfies the axioms. However, there can also be theorems that are true for one model that satisfies the axioms but false for a different model.
Godel's completeness theorem says that if a theorem is true for every model that satisfies a set of axioms, then one can prove that theorem from the axioms.
Godel's first incompleteness theorem says that in any axiomatic system (sufficiently complex) there are theorems that are neither always true nor always false. In other words, there is a theorem that is true for some model of the axioms but false for some other model of the axioms.
Godel's completeness theorem can not be understood without bringing in first order logic, because it is a statement of the expressitivity of the language(relative to its semantics). Other more expressive languages, like second order logic (with its usual semantics) is not complete. Trying to explain Godel's completeness theorem without bringing in the language is a path to confusion.
And your explanation of the first incompleteness theorem is also at best confusing. I must preface this with the comment that your definition of a 'theorem' matches what is usually called a sentence or a statement, and a theorem is usually reserved for a sentence which is proven by a axiomatic system. If the axiomatic system is sound, all theorems will be true in all models. The question of completeness is whether or not all truths(aka sentences true in all models) can be proven(aka they are theorems). With this more common usage of the words, Gödel's incompleteness theorems show that every consistent theory containing the natural numbers has true statements on natural numbers that are not theorems of the theory (that is they cannot be proved inside the theory).
Your description of the first incompleteness theorem is also true for complete logics, even for propositional logic (with your definition of 'theorem' as actually meaning statement). It has statements which is true in some models and false in others. This does not make it incomplete.
My understanding is that for any system of axioms strong enough to encode arithmetic, you can have at most two of these three properties:
1. Complete (for any well formed statement, the axioms can be used to prove either it or its negation)
2. Consistent (can't arrive at contradictory statements ~ arriving at a both a statement and its negation )
3. The set of axioms is enumerable ~ you can write a program that lists them in a defined order (since the workaround for completeness can be just adding an axiom for the cases that are unproven in your original set)
If my understanding is correct, I believe your explanation is missing the third required property.
It's also important to point out that if we cant prove a statement or its negation (one of which must be true) then we know there are true statements that are unprovable. This is a much stronger of a finding than "Godel's first incompleteness theorem says that in any axiomatic system (sufficiently complex) there are theorems that are neither always true nor always false. "
That is interesting, I always thought that the incompleteness theorems says, there are theorems that are true or false in all models but cannot be proved to be so. But if it that is not the case and there always exist models where the theorem is true and false, that makes it sound to me, like the incompleteness theorem is not really about proving things. With that it sounds more like the inability of a sufficiently complex set of axioms to only admit isomorphic models, i.e. have all possible models agree on all expressible theorems. Makes the entire thing sound almost trivial, of course you can not prove what does not follow from the axioms.
> I always thought that the incompleteness theorems says, there are theorems that are true or false in all models but cannot be proved to be so.
As the GP points out, that's not what Godel's incompleteness theorem actually shows. Although it's a common misconception (one which unfortunately is propagated by many sources that should know better).
The key point of the incompleteness theorem is that it shows that (at least in first order logic, which is the logic in which the theorem holds) no set of axioms can ever pin down a single model. For example, no set of first-order axioms can ever pin down "the standard natural numbers" as the only model satisfying the axioms. There will always be other models that also satisfy them. So if you want to pin down a single model, you always have to go beyond just a set of first-order axioms.
Using the natural numbers as an example, consider a model that consists of two "chains" of numbers:
(0, 1, 2, 3, ....)
(..., -3a, -2a, -1a, 0a, 1a, 2a, 3a, ...)
The first chain is, of course, the standard natural numbers, but the second chain also satisfies the standard first-order axioms that we normally take to define natural numbers. So this model, as a whole, satisfies those axioms. And there is no way, within first-order logic, to say "I only want my model to include the first chain". That's what Godel's incompleteness theorem (or more precisely, his first incompleteness theorem combined with his completeness theorem) tells us.
> The key point of the incompleteness theorem is that it shows that (at least in first order logic, which is the logic in which the theorem holds) no set of axioms can ever pin down a single model.
No, this was known before the incompleteness theorem, ref Löwenheim–Skolem theorem.
Of all the incompleteness-style theorems, I find the Halting problem to be the most approachable and also the most interesting. Maybe it's because I'm a software dev that dabbles in math rather than the other way around. But that makes me wonder if all of Gödel's theorems can be stated if 'software form', so to speak.
Right, if you're a software engineer, the realization that the two theorems are nearly-equivalent really takes the air out of a lot of the existential philosophizing around Gödel's incompleteness.
Gödel's argument basically says that any system of mathematics powerful enough to implement basic arithmetic is a computer. This shouldn't be surprising to software engineers because the equivalency between Boolean logic and arithmetic is easy to show. And if you have a computer, you can build algorithms whose outcome can't be programmatically decided by other algorithms.
It's convenient that Henry Rice lived long before the age of language cults. I don't even think Rice wrote software, he's just a mathematician, he proved this nice property in mathematics. Stuff like FORTRAN and ALGOL happens later.
Also though, just as for the Halting Problem, we are always allowed a three-way split. Rice proves that "Has property" vs "Does not have property" can't be done, but "Has property" vs "Does not have property" vs "Shrug - I dunno, seems hard" is possible, and indeed easy if you're OK with lots of machines landing in the "Shrug" pile. You can expend as much work as you like to shrink that pile, Rice just proved it would need infinite work to empty it completely.
Another way around the Rice's theorem is the Curry-Howard correspondence. A constructive proof of existence of a program that has a property can be transformed into a program that has this property. Yet another way is to have a programming language where syntactic correctness implies a range of semantic properties.
e.g. that Godel didn't think this scrapped Hilbert's project totally:
>Gödel believed that it was possible to redefine what we mean by a formal mathematical framework, or allow for alternative frameworks. He often discussed an infinite sequence of acceptable logical systems, each more powerful than the last. Every well-formulated mathematical question might be answerable within one of them.
That part you quoted was interesting to me too. I remember once re-reading the incompleteness theorems - where it talks about a "finite set of axioms", it seemed there may be a loophole if we can imagine a theoretically infinite set of axioms, as a way to approach completeness.
Overall I really enjoyed this article, short interviews with mathematicians and philosophers on a topic I've often thought about.
Some people, when confronted with a Gödel's Incompleteness Theorems, think "I know, I’ll use a theoretically infinite set of axioms." Now they have aleph-nought problems.
As far as I can see people always radically exaggerate the effect of the incompleteness theorems. It seems interesting that any nontrivial axiomatic system has statements which are true but unprovable but to say that derails Hilbert’s project seems just obviously untrue when you can for example join math postgrad programs now which are focused on formalisation. [1] So formalisation is very much still going on, probably more so now than ever given advances in theorem provers.
Yes there are undecidable statements (eg the continuum hypothesis) but that doesn’t change the fact that the vast vast majority of statements in any axiomatic system are going to be decidable, and most undecidable statements are going to have “niche” significance like that.
This is more like the popular lay take that are more or less confused about the meaning and implications of the theorems. The fact is that the implications are real yet more nuanced than this - something the article touched on.
Hilbert’s program was to reduce math to a finite set of axioms and was indeed derailed by incompleteness theorems(Gödel) and undecidability (Church, Turing).
Formalizing math with automated theorem provers has little to do with the goal of Hilbert program. Also they aren’t related to the foundational research that it entailed.
Also, as the article mentions, the implications as well as Gödel was largely misunderstood.
> any nontrivial axiomatic system has statements which are true but unprovable
Although this is a common way of stating what Godel's incompleteness theorem tells us, it's actually not correct. As was posted upthread, when you combine Godel's first incompleteness theorem with Godel's completeness theorem (all this in first-order logic), you find that, for any sentence that is not provable in a system of first-order logic (such as the Godel sentence for that system), there must be a semantic model of that first-order logic in which the sentence is false. (I gave an example of such a model for the first-order axioms of arithmetic upthread.)
At the same time if you imagine a machine that can associate different maths. Would said machine encounter undecidable statements more frequently?
Would the rules of said machine have statements they themselves cannot prove by parameters set in their ‘programmed(by humans, machines, or other machines)’ assumptions?
> As far as I can see people always radically exaggerate the effect of the incompleteness theorems
Like people saying Godel theorems "prove" LLMs could never invent new mathematics because being a software system Godel applies to their operation, but not to humans which are not axiomatic systems, and thus humans can go beyond them and beyond the limits of Godel, humans can "know" a result to be true even if Godel says you can't prove it.
> the vast vast majority of statements in any axiomatic system are going to be decidable
This is just flatly untrue, in the strictest possible sense, and even in the generous definition of "statements that mathematicians care about", that is going to be very heavily biased towards questions that are decidable, because the questions that are likely to be decidable are the ones that they can even reason about to begin with.
If you're a computer programmer, the CS shadow of Godel comes up _all the time_ in practice, and undecidable situations are quite common and don't really need to be carefully constructed.
This is just the drunk-in-the-streetlight situation. Things seem to be decidable everywhere we look because we look where things are likely to be decidable.
> “incompleteness theorems” established that no formal system of mathematics — no finite set of rules, or axioms, from which everything is supposed to follow — can ever be complete.'
Natalie mentions the Newman & Nagel's text "Gödel's Proof," a
(//the//?) 1958 classic on the subject. [[1]] Having left IBM in
December 1990, I spent a month with the text, dipping into mild
insanity, taking to strange wines to relieve myself of the fear
that my previous years long study of Whitehead & Russell's
"Principia Mathematica" [[2]] was not useless.
I really appreciate the inclusion of Alvir's statement on
whether or not Gödel thought he proved all logical systems
undecidable and incomplete. About 80% into the article is her
quote:
>> Often people will speak as if the CH is the smoking gun that
>> shows sometimes mathematical questions have no answer. But
>> in my opinion, this situation provides very little evidence
>> that there are “absolutely undecidable” mathematical
>> problems, relative to any given permissible framework.
Though I would have added a reference to Infinitary Logic [[3]]
after dropping the reference to L-omega-1-omega. I suspect most
readers would find discussion of higher-order and modern logic a
bit confusing without a pause for further study. But a guide
post pointing in the appropriate direction would be good.
That this is the only critique I have of the article speaks to
Wolchover's skill in communicating complex ideas for a lay
audience. I really liked this article, so thank you @baruchel
for posting the reference to it.
:: References
1. https://search.worldcat.org/title/1543160023
2. https://search.worldcat.org/title/933122838
3. https://en.wikipedia.org/wiki/Infinitary_logic
As a child, I noticed that the proofs of mathematical theorems were esoteric knowledge, known only to a few adults. I struggled to follow even the simplest proofs, and hoped that one day I might learn to create a proof or two of my own. This was not only a high aspiration, but a dangerous one. I saw no reason why certain knowledge of a true fact would be accessible to humans via proof. Any-one who embarked on the quest to find a proof risked embarking on a doomed quest to find a non-existent proof.
For me Gödel's completeness theorem is the miracle. Every valid statement has a proof. Amazing!
Aim a little higher, every true statement, and there might not be a proof. It is no surprise to me that this is true. It is a big surprise to me that Gödel was able to prove it; ordinary proofs are hard to find, and proofs of the limits of provability presumably even more deeply hidden.
Non-standard models of arithmetic are weird. Theorems that are true of the standard model of arithmetic and false in some non-standard model must surely be convoluted and obscure. The first order version of the Peano axioms nail down the integers, not perfectly, but very well. Restricting one-self to theorems that are true in all models of them, even the weird, non-standard ones, feels like a very minor restriction. Gödel's completeness theorem raises the possibility of writing a computer program to find a proof of every theorem that isn't convoluted and obscure. Gödel completeness theorem is the really big deal.
Except it isn't. That computer program turns out to be one of those wretched tree search ones that soon bogs down. The real problem turns out to be the combinatorial explosion inherent in unstructured search through the Herbrand universe. One needs Unification and one needs a still missing ingredient to give search a sense of direction. The interesting questions are about the "sense of direction" that lets us find some of the deeply hidden proofs that do exist. Will LLM's help? The answer will be interesting, either way.
It's much easier than it seems.
* There are axioms, there are models, and there are theorems.
* A model is a particular structure with objects and relationships. The "standard model of arithmetic" is just the natural numbers 0, 1, 2, ... with normal rules of addition and subtraction and so on. A different model could be the real numbers, or one that includes infinitesimally small numbers, or so on.
* A set of axioms are rules about how a model can work. For example, we can have an axiom for any set of objects called "numbers" with an operation called "addition" that the operation must be commutative (x+y = y+x). The standard model above is one model that satisfies this axiom.
* A theorem is a fact that can be true or false about a given model. For example, "the model has infinitely many objects." If we can prove a theorem from a set of axioms, then that theorem is true for every model that satisfies the axioms. However, there can also be theorems that are true for one model that satisfies the axioms but false for a different model.
Godel's completeness theorem says that if a theorem is true for every model that satisfies a set of axioms, then one can prove that theorem from the axioms.
Godel's first incompleteness theorem says that in any axiomatic system (sufficiently complex) there are theorems that are neither always true nor always false. In other words, there is a theorem that is true for some model of the axioms but false for some other model of the axioms.
No.
Godel's completeness theorem can not be understood without bringing in first order logic, because it is a statement of the expressitivity of the language(relative to its semantics). Other more expressive languages, like second order logic (with its usual semantics) is not complete. Trying to explain Godel's completeness theorem without bringing in the language is a path to confusion.
And your explanation of the first incompleteness theorem is also at best confusing. I must preface this with the comment that your definition of a 'theorem' matches what is usually called a sentence or a statement, and a theorem is usually reserved for a sentence which is proven by a axiomatic system. If the axiomatic system is sound, all theorems will be true in all models. The question of completeness is whether or not all truths(aka sentences true in all models) can be proven(aka they are theorems). With this more common usage of the words, Gödel's incompleteness theorems show that every consistent theory containing the natural numbers has true statements on natural numbers that are not theorems of the theory (that is they cannot be proved inside the theory).
Your description of the first incompleteness theorem is also true for complete logics, even for propositional logic (with your definition of 'theorem' as actually meaning statement). It has statements which is true in some models and false in others. This does not make it incomplete.
My understanding is that for any system of axioms strong enough to encode arithmetic, you can have at most two of these three properties:
1. Complete (for any well formed statement, the axioms can be used to prove either it or its negation)
2. Consistent (can't arrive at contradictory statements ~ arriving at a both a statement and its negation )
3. The set of axioms is enumerable ~ you can write a program that lists them in a defined order (since the workaround for completeness can be just adding an axiom for the cases that are unproven in your original set)
If my understanding is correct, I believe your explanation is missing the third required property.
It's also important to point out that if we cant prove a statement or its negation (one of which must be true) then we know there are true statements that are unprovable. This is a much stronger of a finding than "Godel's first incompleteness theorem says that in any axiomatic system (sufficiently complex) there are theorems that are neither always true nor always false. "
That is interesting, I always thought that the incompleteness theorems says, there are theorems that are true or false in all models but cannot be proved to be so. But if it that is not the case and there always exist models where the theorem is true and false, that makes it sound to me, like the incompleteness theorem is not really about proving things. With that it sounds more like the inability of a sufficiently complex set of axioms to only admit isomorphic models, i.e. have all possible models agree on all expressible theorems. Makes the entire thing sound almost trivial, of course you can not prove what does not follow from the axioms.
> I always thought that the incompleteness theorems says, there are theorems that are true or false in all models but cannot be proved to be so.
As the GP points out, that's not what Godel's incompleteness theorem actually shows. Although it's a common misconception (one which unfortunately is propagated by many sources that should know better).
The key point of the incompleteness theorem is that it shows that (at least in first order logic, which is the logic in which the theorem holds) no set of axioms can ever pin down a single model. For example, no set of first-order axioms can ever pin down "the standard natural numbers" as the only model satisfying the axioms. There will always be other models that also satisfy them. So if you want to pin down a single model, you always have to go beyond just a set of first-order axioms.
Using the natural numbers as an example, consider a model that consists of two "chains" of numbers:
(0, 1, 2, 3, ....)
(..., -3a, -2a, -1a, 0a, 1a, 2a, 3a, ...)
The first chain is, of course, the standard natural numbers, but the second chain also satisfies the standard first-order axioms that we normally take to define natural numbers. So this model, as a whole, satisfies those axioms. And there is no way, within first-order logic, to say "I only want my model to include the first chain". That's what Godel's incompleteness theorem (or more precisely, his first incompleteness theorem combined with his completeness theorem) tells us.
> The key point of the incompleteness theorem is that it shows that (at least in first order logic, which is the logic in which the theorem holds) no set of axioms can ever pin down a single model.
No, this was known before the incompleteness theorem, ref Löwenheim–Skolem theorem.
Of all the incompleteness-style theorems, I find the Halting problem to be the most approachable and also the most interesting. Maybe it's because I'm a software dev that dabbles in math rather than the other way around. But that makes me wonder if all of Gödel's theorems can be stated if 'software form', so to speak.
Right, if you're a software engineer, the realization that the two theorems are nearly-equivalent really takes the air out of a lot of the existential philosophizing around Gödel's incompleteness.
Gödel's argument basically says that any system of mathematics powerful enough to implement basic arithmetic is a computer. This shouldn't be surprising to software engineers because the equivalency between Boolean logic and arithmetic is easy to show. And if you have a computer, you can build algorithms whose outcome can't be programmatically decided by other algorithms.
Halting problem concerns decidability, not completeness
Also check out https://en.wikipedia.org/wiki/Rice%27s_theorem
basically generalized the halting problem to arbitrary semantic properties.
It's convenient that Henry Rice lived long before the age of language cults. I don't even think Rice wrote software, he's just a mathematician, he proved this nice property in mathematics. Stuff like FORTRAN and ALGOL happens later.
Also though, just as for the Halting Problem, we are always allowed a three-way split. Rice proves that "Has property" vs "Does not have property" can't be done, but "Has property" vs "Does not have property" vs "Shrug - I dunno, seems hard" is possible, and indeed easy if you're OK with lots of machines landing in the "Shrug" pile. You can expend as much work as you like to shrink that pile, Rice just proved it would need infinite work to empty it completely.
Another way around the Rice's theorem is the Curry-Howard correspondence. A constructive proof of existence of a program that has a property can be transformed into a program that has this property. Yet another way is to have a programming language where syntactic correctness implies a range of semantic properties.
Interesting points in here.
e.g. that Godel didn't think this scrapped Hilbert's project totally:
>Gödel believed that it was possible to redefine what we mean by a formal mathematical framework, or allow for alternative frameworks. He often discussed an infinite sequence of acceptable logical systems, each more powerful than the last. Every well-formulated mathematical question might be answerable within one of them.
That part you quoted was interesting to me too. I remember once re-reading the incompleteness theorems - where it talks about a "finite set of axioms", it seemed there may be a loophole if we can imagine a theoretically infinite set of axioms, as a way to approach completeness.
Overall I really enjoyed this article, short interviews with mathematicians and philosophers on a topic I've often thought about.
Some people, when confronted with a Gödel's Incompleteness Theorems, think "I know, I’ll use a theoretically infinite set of axioms." Now they have aleph-nought problems.
As far as I can see people always radically exaggerate the effect of the incompleteness theorems. It seems interesting that any nontrivial axiomatic system has statements which are true but unprovable but to say that derails Hilbert’s project seems just obviously untrue when you can for example join math postgrad programs now which are focused on formalisation. [1] So formalisation is very much still going on, probably more so now than ever given advances in theorem provers.
Yes there are undecidable statements (eg the continuum hypothesis) but that doesn’t change the fact that the vast vast majority of statements in any axiomatic system are going to be decidable, and most undecidable statements are going to have “niche” significance like that.
[1] eg https://www.imperial.ac.uk/study/courses/postgraduate-taught...
This is more like the popular lay take that are more or less confused about the meaning and implications of the theorems. The fact is that the implications are real yet more nuanced than this - something the article touched on.
Hilbert’s program was to reduce math to a finite set of axioms and was indeed derailed by incompleteness theorems(Gödel) and undecidability (Church, Turing).
Formalizing math with automated theorem provers has little to do with the goal of Hilbert program. Also they aren’t related to the foundational research that it entailed.
Also, as the article mentions, the implications as well as Gödel was largely misunderstood.
> any nontrivial axiomatic system has statements which are true but unprovable
Although this is a common way of stating what Godel's incompleteness theorem tells us, it's actually not correct. As was posted upthread, when you combine Godel's first incompleteness theorem with Godel's completeness theorem (all this in first-order logic), you find that, for any sentence that is not provable in a system of first-order logic (such as the Godel sentence for that system), there must be a semantic model of that first-order logic in which the sentence is false. (I gave an example of such a model for the first-order axioms of arithmetic upthread.)
Proof by contradiction requires finding a single contradiction. Practical utility just requires that contradiction be infrequent enough.
At the same time if you imagine a machine that can associate different maths. Would said machine encounter undecidable statements more frequently?
Would the rules of said machine have statements they themselves cannot prove by parameters set in their ‘programmed(by humans, machines, or other machines)’ assumptions?
We really want to believe that we can understand everything, yet we know we cannot.
> As far as I can see people always radically exaggerate the effect of the incompleteness theorems
Like people saying Godel theorems "prove" LLMs could never invent new mathematics because being a software system Godel applies to their operation, but not to humans which are not axiomatic systems, and thus humans can go beyond them and beyond the limits of Godel, humans can "know" a result to be true even if Godel says you can't prove it.
> the vast vast majority of statements in any axiomatic system are going to be decidable
This is just flatly untrue, in the strictest possible sense, and even in the generous definition of "statements that mathematicians care about", that is going to be very heavily biased towards questions that are decidable, because the questions that are likely to be decidable are the ones that they can even reason about to begin with.
If you're a computer programmer, the CS shadow of Godel comes up _all the time_ in practice, and undecidable situations are quite common and don't really need to be carefully constructed.
This is just the drunk-in-the-streetlight situation. Things seem to be decidable everywhere we look because we look where things are likely to be decidable.
> “incompleteness theorems” established that no formal system of mathematics — no finite set of rules, or axioms, from which everything is supposed to follow — can ever be complete.'
There is usually a 'not sufficiently complex' clause in that definition. Presburger arithmetic is complete: https://en.wikipedia.org/wiki/Presburger_arithmetic
Right, you need to be able to construct numbers for Gödel's proof to apply.
Hilbert's incidence geometry, for instance, is consistent and complete. It's just rather small.
Is it releteble to Logic? I have heard that Economy is a subset of Logic, so is this theorem relatable to Economy?
Only in the sense that Economists speak entirely in statements that are unprovable.
It hints at something fundamental to how the universe works, in that there is always an adjacent possible.
I don’t think we’ll ever entirely know what they mean.
It may mean our brains are not currently equipped to understand the universe.