# Relationship of lambda calculus to the rest of math

I just started reading "The calculi of lambda conversion" by Church.

Church defines functions like: id x = x, and says the domain and range are understood to be as permissible as possible. Permitting even itself, id id = id

In my experience, I've always been told to specify a domain and range with the functions I've defined. And they are usually relatively limited, in contrast to id.

This is the first time I've seen functions with a domain and range this large. Are there uses for functions with wide domains and ranges in mathematical contexts other then logic or lambda calculus?

• Did the edit help? – Polymer Dec 10 '16 at 4:29
• Yes. That's much clearer. – Willie Wong Dec 10 '16 at 21:04

The question you are trying to ask is "What is a denotational semantics for the untyped lambda calculus?"

This is a difficult problem because, as Bjorn Kjos-Hanssen said in his answer, if you try and make variables range over elements of some set $D$ you find that you must have $D \times D \cong D$ and $D^D \cong D$. Unfortunately this implies that $D$ is the singleton set and all lambda terms must represent the same function.

Dana Scott solved the problem of giving a denotational semantics to the untyped calculus with the invention of domain theory.

• So if I understand this answer correctly. Untyped lambda calculus is deceptively complicated. The lambda terms are "really" continuous functions of some odd space. Is that correct? – Polymer Dec 10 '16 at 19:46
• I think you need to be more careful about the distinction between a language and a semantics for that language. Lambda terms aren't anything but strings of symbols obeying a certain grammar and rewriting rules. One may attempt to associate mathematical objects to lambda terms (give a denotational semantics) but there are many ways to this. – Justin Hilburn Dec 10 '16 at 21:46
• As you said, one way of doing this makes lambda terms into elements of a domain. – Justin Hilburn Dec 10 '16 at 21:47
• It is just like how there are many different models of the theory of groups, i.e., particular groups. It doesn't make sense to say that an element of a group is "really" a matrix even though that might be true in a particular model. – Justin Hilburn Dec 10 '16 at 21:58
• If you are just trying to connect programming languages and mathematics I would learn the typed lambda calculus and look at mathoverflow.net/questions/903/… – Justin Hilburn Dec 11 '16 at 0:05

In the standard set theoretical setup, a function cannot have itself as an input. This is because the rank of the function is strictly larger than that of its inputs and outputs.

https://en.m.wikipedia.org/wiki/Von_Neumann_universe

So when they say id(id)=id, it is meant in a more algebraic sense where composition is really just a kind of multiplication or binary operation.

• So id is weird. To verify your thought, I define an "application" function $id \times x = x$. Now $id \times id = id$, and we can use it $(id \times id) \times 2 = 2$. No more functions "calling themselves". I'm still confused why I haven't seen something like this sooner. Would studying ZFC closer clarify why functions calling themselves is strange? If I wanted to use this in some mathematical context, what problems might come up? – Polymer Dec 10 '16 at 16:36
• @Polymer well, functions can still "call themselves" in a different sense, like say $f(n+1):=2^{f(n)}$. – Bjørn Kjos-Hanssen Dec 10 '16 at 16:51
• @Polymer I guess you can try to get into Russell's paradox-like trouble like this: if you define $f(g)=1$ if $g(g)=0$, and $f(g)=0$ if $g(g)=1$, then what is $f(f)$? – Bjørn Kjos-Hanssen Dec 10 '16 at 16:55

First, these are not functions. These are lambda expressions that can be artistically interpreted as functions. If you want to model lambda theory in a theory where functions are a part of the discourse, it's a different story. As it was told above, one solution is to model lambda in Set Theory. Sets have functions defined, so lambda expressions would be represented as functions. The problem is, of course, that it's impossible to model without tricks. The solution was provided by Dana Scott in the '60s. See, e.g., http://www.users.waitrose.com/~hindley/SomePapers_PDFs/2006CarHin,HistlamRp.pdf or watch him at LambdaConf, https://www.youtube.com/watch?v=mBjhDyHFqJY&t=2s