3 1. Recall that `ω ≡ \x. x x`, and `Ω ≡ ω ω`. Is `Ω` a fixed point for `ω`? Find a fixed point for `ω`, and prove that it is a fixed point.
5 ANSWER: Most of you just stated that `Ω` a fixed point for `ω`, but didn't explain why. The mere fact that `ω Ω` doesn't immediately reduce to `Ω` doesn't show that the two terms are not convertible. However, one can argue that `Ω` syntactically has two lambdas, and that upon reduction it always yields itself. So it suffices to show that neither `ω Ω` nor anything it can reduces to can have two lambdas. And indeed it has three lambdas, and anything it reduces to will either have three or four lambdas. For the last part of the question, `Y X` is a fixed point for any `X`, as we've already demonstrated in the notes.
7 2. Consider `Ω ξ` for an arbitrary term `ξ`.
8 `Ω` is so busy reducing itself (the eternal narcissist) that it
9 never gets around to noticing whether it has an argument, let alone
10 doing anything with that argument. If so, how could `Ω` have a
11 fixed point? That is, how could there be an `ξ` such that
12 `Ω ξ <~~> ξ`? To answer this
13 question, begin by constructing `Y Ω`. Prove that
14 `Y Ω` is a fixed point for `Ω`.
16 ANSWER: Already demonstrated in the notes. We don't need `Ω ξ` to reduce to `ξ`, because `Y Ω` is a `ξ` that can do its own reducing.
18 3. Find two different terms that have the same fixed point. That is,
19 find terms `F`, `G`, and `ξ` such that `F ξ <~~> ξ` and `G ξ
20 <~~> ξ`. (If you need a hint, reread the notes on fixed
23 ANSWER: Everything is a fixed point of `I`, so in particular `I` is. `I` is also a fixed point for `K I`. There are many other examples.
25 4. Assume that `Ψ` is some fixed point combinator; we're not telling you which one. (You can just write `Psi` in your homework if you don't know how to generate the symbol `Ψ`.) Prove that `Ψ Ψ` is a fixed point of itself, that is, that `Ψ Ψ <~~> Ψ Ψ (Ψ Ψ)`.
27 ANSWER: By the definition of a fixed point operator, `Ψ Ψ` is a fixed point for `Ψ`; and `Ψ (Ψ Ψ)` is a fixed point for `Ψ Ψ`. That is: (a) `Ψ (Ψ Ψ) <~~> Ψ Ψ`; and (b) `Ψ Ψ (Ψ (Ψ Ψ)) <~~> Ψ (Ψ Ψ)`. Now a fact we did not discuss in class, but which Hankin and other readings do discuss, is that substitution of convertible subterms preserves convertibility (and as part of that, `<~~>` is transitive). Hence in the lhs of (b), we can substitute `Ψ Ψ` for the subterm `Ψ (Ψ Ψ)`, because of (a), getting (c) `Ψ Ψ (Ψ Ψ) <~~> Ψ (Ψ Ψ)`. But then by (a) and transitivity of `<~~>`, we get (d) `Ψ Ψ (Ψ Ψ) <~~> Ψ Ψ`. Which states that `Ψ Ψ` is a fixed point for itself.
30 ## Writing recursive functions ##
32 5. Helping yourself to the functions given below,
33 write a recursive function called `fact` that computes the factorial.
34 The factorial `n! = n * (n - 1) * (n - 2) * ... * 3 * 2 * 1`.
35 For instance, `fact 0 ~~> 1`, `fact 1 ~~> 1`, `fact 2 ~~> 2`, `fact 3 ~~>
36 6`, and `fact 4 ~~> 24`.
39 let false = \y n. n in
40 let pair = \a b. \v. v a b in
41 let fst = \a b. a in ; aka true
42 let snd = \a b. b in ; aka false
44 let succ = \n s z. s (n s z) in
45 let zero? = \n. n (\p. false) true in
46 let pred = \n. n (\p. p (\a b. pair (succ a) a)) (pair zero zero) snd in
47 let add = \l r. r succ l in
48 let mult = \l r. r (add l) 0 in
49 let Y = \h. (\u. h (u u)) (\u. h (u u)) in
55 ANSWER: `let fact = Y (\fact n. (zero? n) 1 (mult n (fact (pred n))))`.
57 6. For this question, we want to implement **sets** of numbers in terms of lists of numbers, where we make sure as we construct those lists that they never contain a single number more than once. (It would be even more efficient if we made sure that the lists were always sorted, but we won't try to implement that refinement here.) To enforce the idea of modularity, let's suppose you don't know the details of how the lists are implemented. You just are given the functions defined below for them (but pretend you don't see the actual definitions). These define lists in terms of [[one of the new encodings discussed last week|/topics/week3_lists#v5-lists]].
59 ; all functions from the previous question, plus
60 ; `num_cmp x y lt eq gt` returns lt when x<y, eq when x==y, gt when x>y
61 let num_cmp = (\base build consume. \l r. r consume (l build base) fst)
63 (pair (\a b c. b) (K (\a b c. a)))
65 (\p. pair (\a b c. c) p)
67 (\p. p fst p (p snd) (p snd)) in
68 let num_equal? = \x y. num_cmp x y false true false in
69 let neg = \b y n. b n y in
70 let empty = \f n. n in
71 let cons = \x xs. \f n. f x xs in
72 let empty? = \xs. xs (\y ys. false) true in
73 let head = \xs. xs (\y ys. y) err in
74 let tail = \xs. xs (\y ys. ys) empty in
75 let append = Y (\append. \xs zs. xs (\y ys. (cons y (append ys zs))) zs) in
76 let take_while = Y (\take_while. \p xs. xs (\y ys. (p y) (cons y (take_while p ys)) empty) empty) in
77 let drop_while = Y (\drop_while. \p xs. xs (\y ys. (p y) (drop_while p ys) xs) empty) in
80 The functions `take_while` and `drop_while` work as described in [[Week 1's homework|assignment1]].
82 Using those resources, define a `set_cons` and a `set_equal?` function. The first should take a number argument `x` and a set argument `xs` (implemented as a list of numbers assumed to have no repeating elements), and return a (possibly new) set argument which contains `x`. (But make sure `x` doesn't appear in the result twice!) The `set_equal?` function should take two set arguments `xs` and `ys` and say whether they represent the same set. (Be careful, the lists `[1, 2]` and `[2, 1]` are different lists but do represent the same set. Hence, you can't just use the `list_equal?` function you defined in last week's homework.)
84 Here are some tips for getting started. Use `drop_while`, `num_equal?`, and `empty?` to define a `mem?` function that returns `true` if number `x` is a member of a list of numbers `xs`, else returns `false`. Also use `take_while`, `drop_while`, `num_equal?`, `tail` and `append` to define a `without` function that returns a copy of a list of numbers `xs` that omits the first occurrence of a number `x`, if there be such. You may find these functions `mem?` and `without` useful in defining `set_cons` and `set_equal?`. Also, for `set_equal?`, you are probably going to want to define the function recursively... as now you know how to do.
86 Some comments comparing this exercise to *The Little Schemer*, and Scheme more generally:
88 * The `set_equal?` you're trying to define here is like `eqset?` in Chapter 7 of *The Little Schemer*, and `set_cons x xs` would be like `(makeset (cons x xs))`, from that same chapter.
89 * `mem?` and `without` are like the `member?` and `rember` functions defined in Chapter 2 and 3 of *The Little Schemer*, though those functions are defined for lists of symbolic atoms, and here you are instead defining them for lists of numbers. *The Little Schemer* also defines `multirember`, which removes all occurrences of a match rather than just the first; and `member*` and `rember*` in Chapter 5, that operate on lists that may contain other, embedded lists.
90 * The native Scheme function that most resembles the `mem?` you're defining is `memv`, though that is defined for more than just numbers, and when that `memv` finds a match it returns a *list* starting with the match, rather than `#t`.
95 let neq? = \n m. neg (num_eq? n m) in
96 let mem? = \n xs. neg (empty? (drop_while (neq? n) xs)) in
97 let without = \n xs. append (take_while (neq? n) xs) (tail (drop_while (neq? n) xs)) in
98 let set_cons = \x xs. (mem? x xs) xs (cons x xs) in
99 let set_equal? = Y (\set_equal?. \xs ys. (empty? xs)
101 ; else when xs aren't empty
103 (set_equal? (tail xs) (without (head xs) ys))
104 ; else when head xs not in ys
107 7. Linguists often analyze natural language expressions into **trees**. We'll need trees in future weeks, and tree structures provide good opportunities for learning how to write recursive functions. Making use of our current resources, we might approximate trees as follows. Instead of words or syntactic categories, we'll have the nodes of the tree labeled with Church numbers. We'll think of a tree as a list in which each element is itself a tree. For simplicity, we'll adopt the convention that a tree of length 1 must contain a number as its only element.
109 Then we have the following representations:
132 Some limitations of this scheme: there is no easy way to label an inner, branching node (for example with a syntactic category like VP), and there is no way to represent a tree in which a mother node has a single daughter.
134 When processing a tree, you can test for whether the tree is a leaf node (that is, contains only a single number), by testing whether the length of the list is 1. This will be your base case for your recursive definitions that work on these trees. (You'll probably want to write a function `leaf?` that encapsulates this check.)
136 Your assignment is to write a Lambda Calculus function that expects a tree, encoded in the way just described, as an argument, and returns the sum of its leaves as a result. So for all of the trees listed above, it should return `1 + 2 + 3`, namely `6`. You can use any Lambda Calculus implementation of lists you like.
141 8. The **fringe** of a leaf-labeled tree is the list of values at its leaves, ordered from left-to-right. For example, the fringe of all three trees displayed above is the same list, `[1, 2, 3]`. We are going to return to the question of how to tell whether trees have the same fringe several times this course. We'll discover more interesting and more efficient ways to do it as our conceptual toolboxes get fuller. For now, we're going to explore the straightforward strategy. Write a function that expects a tree as an argument, and returns the list which is its fringe. Next write a function that expects two trees as arguments, converts each of them into their fringes, and then determines whether the two lists so produced are equal. (Convert your `list_equal?` function from last week's homework into the Lambda Calculus for this last step.)
147 ## Arithmetic infinity? ##
149 The next few questions involve reasoning about Church arithmetic and
150 infinity. Let's choose some arithmetic functions:
152 succ = \n s z. s (n s z)
153 add = \l r. r succ l in
154 mult = \l r. r (add l) 0 in
155 exp = \base r. r (mult base) 1 in
157 There is a pleasing pattern here: addition is defined in terms of
158 the successor function, multiplication is defined in terms of
159 addition, and exponentiation is defined in terms of multiplication.
162 9. Find a fixed point `ξ` for the successor function. Prove it's a fixed
163 point, i.e., demonstrate that `succ ξ <~~> ξ`.
165 We've had surprising success embedding normal arithmetic in the Lambda
166 Calculus, modeling the natural numbers, addition, multiplication, and
167 so on. But one thing that some versions of arithmetic supply is a
168 notion of infinity, which we'll write as `inf`. This object sometimes
169 satisfies the following constraints, for any finite natural number `n`:
176 (Note, though, that with *some* notions of infinite numbers, like [[!wikipedia ordinal numbers]], operations like `+` are defined in such a way that `inf + n` is different from `n + inf`, and does exceed `inf`; similarly for `*` and `^`. With other notions of infinite numbers, like the [[!wikipedia cardinal numbers]], even less familiar arithmetic operations are employed.)
178 ANSWER: Let `H ≡ \u. succ (u u)`, and `X ≡ H H ≡ (\u. succ (u u)) H`. Note that `X ≡ H H ~~> succ (H H)`. Hence `X` is a fixed point for `succ`.
180 10. Prove that `add ξ 1 <~~> ξ`, where `ξ` is the fixed
181 point you found in (1). What about `add ξ 2 <~~> ξ`?
183 Comment: a fixed point for the successor function is an object such that it
184 is unchanged after adding 1 to it. It makes a certain amount of sense
185 to use this object to model arithmetic infinity. For instance,
186 depending on implementation details, it might happen that `leq n ξ` is
187 true for all (finite) natural numbers `n`. However, the fixed point
188 you found for `succ` and `(+n)` (recall this is shorthand for `\x. add x n`) may not be a fixed point for `(*n)`, for example.
190 ANSWER: Prove that `add X 1 <~~> X`:
192 add X 1 == (\m n. n succ m) X 1
194 == (\s z. s z) succ X
197 Which by the previous problem is convertible with `X`. (In particular, `X ~~> succ X`.) What about `add X 2`?
199 add X 2 == (\m n. n succ m) X 2
201 == (\s z. s (s z)) succ X
204 And we know the inner term will be convertible with `X`, and hence we get that the whole result is convertible with `succ X`. Which we already said is convertible with `X`. We can readily see that `add X n <~~> X` for all (finite) natural numbers `n`.
208 ## Mutually-recursive functions ##
210 11. (Challenging.) One way to define the function `even?` is to have it hand off
211 part of the work to another function `odd?`:
213 let even? = \x. (zero? x)
214 ; if x == 0 then result is
216 ; else result turns on whether x-1 is odd
219 At the same tme, though, it's natural to define `odd?` in such a way that it
220 hands off part of the work to `even?`:
222 let odd? = \x. (zero? x)
223 ; if x == 0 then result is
225 ; else result turns on whether x-1 is even
228 Such a definition of `even?` and `odd?` is called **mutually recursive**. If you
229 trace through the evaluation of some sample numerical arguments, you can see
230 that eventually we'll always reach a base step. So the recursion should be
231 perfectly well-grounded:
234 ~~> (zero? 3) true (odd? (pred 3))
236 ~~> (zero? 2) false (even? (pred 2))
238 ~~> (zero? 1) true (odd? (pred 1))
240 ~~> (zero? 0) false (even? (pred 0))
243 But we don't yet know how to implement this kind of recursion in the Lambda
246 The fixed point operators we've been working with so far worked like this:
251 Suppose we had a pair of fixed point operators, `Y1` and `Y2`, that operated on
252 a *pair* of functions `h` and `g`, as follows:
259 If we gave you such a `Y1` and `Y2`, how would you implement the above
260 definitions of `even?` and `odd?`?
265 let proto_even = \even odd. <span class="ul">\n. (zero? n) true (odd (pred n))</span> in
266 let proto_odd = \even odd. <span class="ul">\n. (zero? n) false (even (pred n))</span> in
267 let even = Y1 proto_even proto_odd in
268 let odd = Y2 proto_even proto_odd in
272 By the definitions of `Y1` and `Y2`, we know that `even <~~> proto_even even odd`, and `odd <~~> proto_odd even odd`. Hence the bound variables `even` and `odd` inside the `proto_...` functions have the values we want. `even` will be bound to (something convertible with) the underlined portion of `proto_even`, and `odd` will be bound to (something convertible with) the underlined portion of `proto_odd`.
274 12. (More challenging.) Using our derivation of `Y` from [[this week's notes|topics/week4_fixed_point_combinators#deriving-y]] as a model, construct a pair `Y1` and `Y2` that behave in the way described above.
276 Here is one hint to get you started: remember that in the notes, we constructed a fixed point for `h` by evolving it into `H` and using `H H` as `h`'s fixed point. We suggested the thought exercise, how might you instead evolve `h` into some `T` and then use `T T T` as `h`'s fixed point. Try solving this problem first. It may help give you the insights you need to define a `Y1` and `Y2`. [[Here are some hints|assignment4_hint]].
278 ANSWER: One solution is given in the hint. Here is another:
280 let Y1 = \f g . (\u v . f (u u v)(v v u))
281 (\u v . f (u u v)(v v u))
282 (\v u . g (v v u)(u u v)) in
283 let Y2 = \f g . Y1 g f in