The Untyped Lambda Calculus

Transcription

1 CS738: Advanced Compiler Optimizations The Untyped Lambda Calculus Amey Karkare Department of CSE, IIT Kanpur

2 Reference Book Types and Programming Languages by Benjamin C. Pierce

3 The Abstract Syntax t := x Variable

4 The Abstract Syntax t := x Variable λx.t Abstraction

5 The Abstract Syntax t := x Variable λx.t Abstraction t t Application

6 The Abstract Syntax t := x Variable λx.t Abstraction t t Application Parenthesis, (... ), can be used for grouping and scoping.

7 Conventions λx.t 1 t 2 t 3 is an abbreviation for λx.(t 1 t 2 t 3 ), i.e., the scope of x is as far to the right as possible until it is

8 Conventions λx.t 1 t 2 t 3 is an abbreviation for λx.(t 1 t 2 t 3 ), i.e., the scope of x is as far to the right as possible until it is terminated by a ) whose matching ( occurs to the left pf λ, OR

9 Conventions λx.t 1 t 2 t 3 is an abbreviation for λx.(t 1 t 2 t 3 ), i.e., the scope of x is as far to the right as possible until it is terminated by a ) whose matching ( occurs to the left pf λ, OR terminated by the end of the term.

10 Conventions λx.t 1 t 2 t 3 is an abbreviation for λx.(t 1 t 2 t 3 ), i.e., the scope of x is as far to the right as possible until it is terminated by a ) whose matching ( occurs to the left pf λ, OR terminated by the end of the term. Applications associate to the left: t 1 t 2 t 3 to be read as (t 1 t 2 )t 3 and not as t 1 (t 2 t 3 )

11 Conventions λx.t 1 t 2 t 3 is an abbreviation for λx.(t 1 t 2 t 3 ), i.e., the scope of x is as far to the right as possible until it is terminated by a ) whose matching ( occurs to the left pf λ, OR terminated by the end of the term. Applications associate to the left: t 1 t 2 t 3 to be read as (t 1 t 2 )t 3 and not as t 1 (t 2 t 3 ) λxyz.t is an abbreviation for λxλyλz.t which in turn is abbreviation for λx.(λy.(λz.t)).

12 α-renaming The name of a bound variable has no meaning except for its use to identify the bounding λ.

13 α-renaming The name of a bound variable has no meaning except for its use to identify the bounding λ. Renaming a λ variable, including all its bound occurrences, does not change the meaning of an expression. For example, λx.x x y is equivalent to λu.u u y

14 α-renaming The name of a bound variable has no meaning except for its use to identify the bounding λ. Renaming a λ variable, including all its bound occurrences, does not change the meaning of an expression. For example, λx.x x y is equivalent to λu.u u y But it is not same as λx.x x w

15 α-renaming The name of a bound variable has no meaning except for its use to identify the bounding λ. Renaming a λ variable, including all its bound occurrences, does not change the meaning of an expression. For example, λx.x x y is equivalent to λu.u u y But it is not same as λx.x x w Can not change free variables!

16 β-reduction (Execution Semantics) if an abstraction λx.t 1 is applied to a term t 2 then the result of the application is

17 β-reduction (Execution Semantics) if an abstraction λx.t 1 is applied to a term t 2 then the result of the application is the body of the abstraction t 1 with all free occurrences of the formal parameter x replaced with t 2.

18 β-reduction (Execution Semantics) if an abstraction λx.t 1 is applied to a term t 2 then the result of the application is the body of the abstraction t 1 with all free occurrences of the formal parameter x replaced with t 2. For example, (λfλx.f (f x)) g β λx.g (g x)

19 Caution During β-reduction, make sure a free variable is not captured inadvertently.

20 Caution During β-reduction, make sure a free variable is not captured inadvertently. The following reduction is WRONG (λxλy.x)(λx.y) β λy.λx.y

21 Caution During β-reduction, make sure a free variable is not captured inadvertently. The following reduction is WRONG (λxλy.x)(λx.y) β λy.λx.y Use α-renaming to avoid variable capture (λxλy.x)(λx.y) α β (λuλv.u)(λx.y) λv.λx.y

22 Exercise Apply β-reduction as far as possible 1. (λx y z. x z (y z)) (λx y. x) (λy.y) 2. (λx. x x)(λx. x x) 3. (λx y z. x z (y z)) (λx y. x) ((λx. x x)(λx. x x))

23 Church-Rosser Theorem Multiple ways to apply β-reduction

24 Church-Rosser Theorem Multiple ways to apply β-reduction Some may not terminate

25 Church-Rosser Theorem Multiple ways to apply β-reduction Some may not terminate However, if two different reduction sequences terminate then they always terminate in the same term

26 Church-Rosser Theorem Multiple ways to apply β-reduction Some may not terminate However, if two different reduction sequences terminate then they always terminate in the same term Also called the Diamond Property

27 Church-Rosser Theorem Multiple ways to apply β-reduction Some may not terminate However, if two different reduction sequences terminate then they always terminate in the same term Also called the Diamond Property Leftmost, outermost reduction will find the normal form if it exists

28 Programming in λ Calculus Where is the other stuff?

29 Programming in λ Calculus Where is the other stuff? Constants?

30 Programming in λ Calculus Where is the other stuff? Constants? Numbers

31 Programming in λ Calculus Where is the other stuff? Constants? Numbers Booleans

32 Programming in λ Calculus Where is the other stuff? Constants? Numbers Booleans Complex Types?

33 Programming in λ Calculus Where is the other stuff? Constants? Numbers Booleans Complex Types? Lists

34 Programming in λ Calculus Where is the other stuff? Constants? Numbers Booleans Complex Types? Lists Arrays

35 Programming in λ Calculus Where is the other stuff? Constants? Numbers Booleans Complex Types? Lists Arrays Don t we need data?

36 Programming in λ Calculus Where is the other stuff? Constants? Numbers Booleans Complex Types? Lists Arrays Don t we need data?

37 Programming in λ Calculus Where is the other stuff? Constants? Numbers Booleans Complex Types? Lists Arrays Don t we need data? Abstractions act as functions as well as data!

38 Numbers: Church Numerals We need a Zero

39 Numbers: Church Numerals We need a Zero Absence of item

40 Numbers: Church Numerals We need a Zero Absence of item And something to count

41 Numbers: Church Numerals We need a Zero Absence of item And something to count Presence of item

42 Numbers: Church Numerals We need a Zero Absence of item And something to count Presence of item Intuition: Whiteboard and Marker

43 Numbers: Church Numerals We need a Zero Absence of item And something to count Presence of item Intuition: Whiteboard and Marker Blank board represents Zero

44 Numbers: Church Numerals We need a Zero Absence of item And something to count Presence of item Intuition: Whiteboard and Marker Blank board represents Zero Each mark by marker represents a count.

45 Numbers: Church Numerals We need a Zero Absence of item And something to count Presence of item Intuition: Whiteboard and Marker Blank board represents Zero Each mark by marker represents a count. However, other pairs of objects will work as well

46 Numbers: Church Numerals We need a Zero Absence of item And something to count Presence of item Intuition: Whiteboard and Marker Blank board represents Zero Each mark by marker represents a count. However, other pairs of objects will work as well Lets translate this intuition into λ-expressions

47 Numbers Zero = λm w. w

48 Numbers Zero = λm w. w No mark on the whiteboard

49 Numbers Zero = λm w. w No mark on the whiteboard One = λm w. m w

50 Numbers Zero = λm w. w No mark on the whiteboard One = λm w. m w One mark on the whiteboard

51 Numbers Zero = λm w. w No mark on the whiteboard One = λm w. m w One mark on the whiteboard Two = λm w. m (m w)

52 Numbers Zero = λm w. w No mark on the whiteboard One = λm w. m w One mark on the whiteboard Two = λm w. m (m w)...

53 Numbers Zero = λm w. w No mark on the whiteboard One = λm w. m w One mark on the whiteboard Two = λm w. m (m w)... What about operations?

54 Numbers Zero = λm w. w No mark on the whiteboard One = λm w. m w One mark on the whiteboard Two = λm w. m (m w)... What about operations? add, multiply, subtract, divide,...?

55 Operations on Numbers succ = λx m w. m (x m w)

56 Operations on Numbers succ = λx m w. m (x m w) Verify: succ N = N + 1

57 Operations on Numbers succ = λx m w. m (x m w) Verify: succ N = N + 1 add = λx y m w. x m (y m w)

58 Operations on Numbers succ = λx m w. m (x m w) Verify: succ N = N + 1 add = λx y m w. x m (y m w) Verify: add M N = M + N

59 Operations on Numbers succ = λx m w. m (x m w) Verify: succ N = N + 1 add = λx y m w. x m (y m w) Verify: add M N = M + N mult = λx y m w. x (y m) w

60 Operations on Numbers succ = λx m w. m (x m w) Verify: succ N = N + 1 add = λx y m w. x m (y m w) Verify: add M N = M + N mult = λx y m w. x (y m) w Verify: mult M N = M * N

61 More Operations pred = λx m w. x (λg h. h (g m))(λu. w)(λu. u)

62 More Operations pred = λx m w. x (λg h. h (g m))(λu. w)(λu. u) Verify: pred N = N - 1

63 More Operations pred = λx m w. x (λg h. h (g m))(λu. w)(λu. u) Verify: pred N = N - 1 nminus = λx y. y pred x

64 More Operations pred = λx m w. x (λg h. h (g m))(λu. w)(λu. u) Verify: pred N = N - 1 nminus = λx y. y pred x Verify: nminus M N = max(0, M - N) natural subtraction

65 Church Booleans True and False

66 Church Booleans True and False Intuition: Selection of one out of two (complementary) choices

67 Church Booleans True and False Intuition: Selection of one out of two (complementary) choices True = λx y. x

68 Church Booleans True and False Intuition: Selection of one out of two (complementary) choices True = λx y. x False = λx y. y

69 Church Booleans True and False Intuition: Selection of one out of two (complementary) choices True = λx y. x False = λx y. y Predicate:

70 Church Booleans True and False Intuition: Selection of one out of two (complementary) choices True = λx y. x False = λx y. y Predicate: iszero = λx. x (λu.false) True

71 Operations on Booleans Logical operations and or not = λp q. p q p = λp q. p p q = λp t f.p f t

72 Operations on Booleans Logical operations and or not = λp q. p q p = λp q. p p q = λp t f.p f t The conditional operator if if = λc e t e f. (c e t e f )

73 Operations on Booleans Logical operations and or not = λp q. p q p = λp q. p p q = λp t f.p f t The conditional operator if if c e t e f reduces to e t if c is True, and to e f if c is False if = λc e t e f. (c e t e f )

74 More... More such types can be found at

75 More... More such types can be found at It is fun to come up with your own definitions for constants and operations over different types

76 More... More such types can be found at It is fun to come up with your own definitions for constants and operations over different types or to develop understanding for existing definitions.

77 We are missing something!! The machinery described so far does not allow us to define Recursive functions Factorial, Fibonacci,... There is no concept of named functions So no way to refer to a function recursively! Fix-point computation comes to rescue

78 Fix-point and Y -combinator A fix-point of a function f is a value p such that f p = p

79 Fix-point and Y -combinator A fix-point of a function f is a value p such that f p = p Assume existence of a magic expression, called Y-combinator, that when applied to a λ-expression, gives its fixed point Y f = f (Y f)

80 Fix-point and Y -combinator A fix-point of a function f is a value p such that f p = p Assume existence of a magic expression, called Y-combinator, that when applied to a λ-expression, gives its fixed point Y f = f (Y f) Y-combinator gives us a way to apply a function recursively

81 Recursion Example: Factorial fact = λn. if (iszero n) One (mult n (fact (pred n))) = (λf n. if (iszero n) One (mult n (f (pred n)))) fact

82 Recursion Example: Factorial fact = λn. if (iszero n) One (mult n (fact (pred n))) = (λf n. if (iszero n) One (mult n (f (pred n)))) fact fact = g fact fact is a fixed point of the function g = (λf n. if (iszero n)one (mult n (f (pred n))))

83 Recursion Example: Factorial fact = λn. if (iszero n) One (mult n (fact (pred n))) = (λf n. if (iszero n) One (mult n (f (pred n)))) fact fact = g fact fact is a fixed point of the function g = (λf n. if (iszero n)one (mult n (f (pred n)))) Using Y-combinator, fact = Y g

84 Factorial: Verify fact 2 = (Y g) 2

85 Factorial: Verify fact 2 = (Y g) 2 = g (Y g) 2 by definition of Y-combinator

86 Factorial: Verify fact 2 = (Y g) 2 = g (Y g) 2 by definition of Y-combinator = (λfn. if (iszero n) 1 (mult n (f (pred n)))) (Y g) 2

87 Factorial: Verify fact 2 = (Y g) 2 = g (Y g) 2 by definition of Y-combinator = (λfn. if (iszero n) 1 (mult n (f (pred n)))) (Y g) 2 = (λn. if (iszero n) 1 (mult n ((Y g) (pred n)))) 2

88 Factorial: Verify fact 2 = (Y g) 2 = g (Y g) 2 by definition of Y-combinator = (λfn. if (iszero n) 1 (mult n (f (pred n)))) (Y g) 2 = (λn. if (iszero n) 1 (mult n ((Y g) (pred n)))) 2 = if (iszero 2) 1 (mult 2 ((Y g)(pred2)))

89 Factorial: Verify fact 2 = (Y g) 2 = g (Y g) 2 by definition of Y-combinator = (λfn. if (iszero n) 1 (mult n (f (pred n)))) (Y g) 2 = (λn. if (iszero n) 1 (mult n ((Y g) (pred n)))) 2 = if (iszero 2) 1 (mult 2 ((Y g)(pred2))) = (mult 2 ((Y g) 1))

90 Factorial: Verify fact 2 = (Y g) 2 = g (Y g) 2 by definition of Y-combinator = (λfn. if (iszero n) 1 (mult n (f (pred n)))) (Y g) 2 = (λn. if (iszero n) 1 (mult n ((Y g) (pred n)))) 2 = if (iszero 2) 1 (mult 2 ((Y g)(pred2))) = (mult 2 ((Y g) 1))... = (mult 2 (mult 1 (if (iszero 0) 1 (...))))

91 Factorial: Verify fact 2 = (Y g) 2 = g (Y g) 2 by definition of Y-combinator = (λfn. if (iszero n) 1 (mult n (f (pred n)))) (Y g) 2 = (λn. if (iszero n) 1 (mult n ((Y g) (pred n)))) 2 = if (iszero 2) 1 (mult 2 ((Y g)(pred2))) = (mult 2 ((Y g) 1))... = (mult 2 (mult 1 (if (iszero 0) 1 (...)))) = (mult 2 (mult 1 1))

92 Factorial: Verify fact 2 = (Y g) 2 = g (Y g) 2 by definition of Y-combinator = (λfn. if (iszero n) 1 (mult n (f (pred n)))) (Y g) 2 = (λn. if (iszero n) 1 (mult n ((Y g) (pred n)))) 2 = if (iszero 2) 1 (mult 2 ((Y g)(pred2))) = (mult 2 ((Y g) 1))... = (mult 2 (mult 1 (if (iszero 0) 1 (...)))) = (mult 2 (mult 1 1)) = 2

93 Recursion and Y -combinator Y-combinator allows to unroll the body of loop once similar to one unfolding of recursive call

94 Recursion and Y -combinator Y-combinator allows to unroll the body of loop once similar to one unfolding of recursive call Sequence of Y-combinator applications allow complete unfolding of recursive calls

95 Recursion and Y -combinator Y-combinator allows to unroll the body of loop once similar to one unfolding of recursive call Sequence of Y-combinator applications allow complete unfolding of recursive calls

96 Recursion and Y -combinator Y-combinator allows to unroll the body of loop once similar to one unfolding of recursive call Sequence of Y-combinator applications allow complete unfolding of recursive calls BUT, what about the existence of Y-combinator?

97 Y -combinators Many candidates exist Y 1 = λf. (λx. f (x x)) (λx. f (x x))

98 Y -combinators Many candidates exist Y 1 = λf. (λx. f (x x)) (λx. f (x x)) Y = λabcdefghijklmnopqstuvwxwzr.r(thisisafixedpointcombinator)

99 Y -combinators Many candidates exist Y 1 = λf. (λx. f (x x)) (λx. f (x x)) Y = λabcdefghijklmnopqstuvwxwzr.r(thisisafixedpointcombinator) Y funny = TTTTT TTTTT TTTTT TTTTT TTTTT T

100 Y -combinators Many candidates exist Y 1 = λf. (λx. f (x x)) (λx. f (x x)) Y = λabcdefghijklmnopqstuvwxwzr.r(thisisafixedpointcombinator) Y funny = TTTTT TTTTT TTTTT TTTTT TTTTT T Verify that (Y f) = f (Y f) for each

101 Summary A cursory look at λ-calculus

102 Summary A cursory look at λ-calculus Functions are data, and Data are functions!

103 Summary A cursory look at λ-calculus Functions are data, and Data are functions! Not covered but important to know: The power of λ calculus is equivalent to that of Turing Machine ( Church Turing Thesis )