Documentation

PnP2023.Lec_01_06.NatEq

Examples of Proofs #

We see our first proofs, most of which involve the relation on natural numbers.

We will see that the natural numbers are "defined" in terms of the zero and succ constructors.

Analogous to this, (modulo renaming) the relation is defined in terms of the le_refl and le_step constructors.

However, we will first focus on a simpler relation.

Example: a proof by reflexivity of equality #

Our first example is simpler: just a proof of equality. Note that a proof of a theorem in lean is very similar, both conceptually and syntactically, to a definition.

In a definition, we give a name, a type and a value (though sometimes the type is inferred so can be omitted). In a theorem, we give a name, a statement and a proof. The statement is essentially a type, and the proof is a value of that type.

Technically a statement is a proposition which is a special sort of type as we see below, but in these examples it is best to think of it as just a type.

theorem one_plus_one_is_two :
1 + 1 = 2

The theorem that 1 + 1 = 2

We look more closely at the right hand side, i.e., the proof. Below is the description in Lean of rfl, which corresponds to reflexivity of equality, i.e., the fact that any term is equal to itself:

rfl : a = a is the unique constructor of the equality type. This is the same as Eq.refl except that it takes a implicitly instead of explicitly.

This is a more powerful theorem than it may appear at first, because although the statement of the theorem is a = a, lean will allow anything that is definitionally equal to that type. So, for instance, 2 + 2 = 4 is proven in lean by rfl, because both sides are the same up to definitional equality.

To clarify what is meant by implicit and explicit parameters, we look at the types of rfl and Eq.refl. Both these represent the statement that if α is a type and a : α is a term of type α then a = a.

#check rfl  -- rfl.{u} {α : Sort u} {a : α} : a = a

#check Eq.refl -- Eq.refl.{u_1} {α : Sort u_1} (a : α) : a = a

Notice that in rfl both the arguments α and a are in braces while in Eq.refl the argument α is in braces while a in in parenthesis. When calling a function, the arguments in parenthesis are to be specified, while those in braces are to be inferred (if possible).

If we prepend @ to a function name then all arguments become explicit. Conversely we can ask Lean to infer arguments by using _. The following examples of similar proofs should clarify these.

/- argument `a` given explicitly -/
example : 2 = 2 := Eq.refl 2

/- notation `@` used and both `α` and `a` are given explicitly-/
example : 3 = 3 := @Eq.refl ℕ 3

/- the argument `a` is inferred-/
example : 1 + 2 = 3 := Eq.refl _

/- notation `@` used but `α` is inferred-/
example : 3  = 3 := @Eq.refl _ 3

Incorrect proofs #

Of course the right hand side may not be a proof of the statement in the left hand side. As mentioned above, this is equivalent to the term on the right-hand side having the type specified in the left hand side.

For example, consdier the following (incorrect) proof.

theorem wrong : 1 + 2 = 2 := rfl

We get the following error message:

type mismatch
  rfl
has type
  1 + 2 = 1 + 2 : Prop
but is expected to have type
  1 + 2 = 2 : Prop

This is because Lean inferred that a is 1 + 2 (and α is ) and deduced that the type of the right hand side is 1 + 2 = 1 + 2. This is not equal to the type specified in the left hand side, 1 + 2 = 2.

We get a different type of error if the statement we are trying to prove is not an equality. For example, consider the following (incorrect) proof.

example : 1 ≤ 2 := rfl

Here Lean is not able to infer a (or even α) and so we get the following error message:

type mismatch
  rfl
has type
  ?m.690 = ?m.690 : Prop
but is expected to have type
  1 ≤ 2 : Prop

All that Lean knows about rfl is that it is a proof that a = a, but the statement is not an equality. So Lean assign a metavariable ?m.690 to a and cannot solve for the metavariable, giving an error as above.

Proofs versus Definitions; Sorts and Types #

To clarify the similarities and differences between a definition and a theorem, as well as sorts and types, we look at the following example.

def two :

Definition of two for comparison

Instances For

    As we have remarked, this is formally very similar to the proof of a theorem. To see the differences, we see some associated types:

    #check 1 = 1 -- 1 = 1 : Prop
    #check ℕ -- ℕ : Type
    
    /- The universe of propositions. Prop ≡ Sort 0-/
    #check Prop -- Prop : Type
    
    /- A type universe. Type ≡ Type 0, Type u ≡ Sort (u + 1)-/
    #check Type -- Type : Type 1
    

    The main difference between the two cases is that the type of a theorem is a proposition, which is a special type (i.e., a type in Prop). The type of Prop is Type, which is a type in the next universe up and is the type of . The universes form so called Sorts, with all but the first being Types.

    In other closely related foundational systems (such as Homotopy Type Theory), this distinction is not present and all uninverses are Types. In Lean propositions, i.e., elements of Prop (such as 1 = 1) have a special property that two terms (i.e., proofs) of a proposition are equal by definition.

    Propositions are taken to be at "Types at level -1", i.e., "Sorts at level 0". This makes it convenient to exclude propositions in some cases where it is necessary to do so (as we will see later).