An expression denotes a value which may depend on some variables. The basic example of an expression is simply a variable x. Of course, x must be defined somewhere in order for such an expression to make sense. It can be either a parameter (of a definition, or a lambda expression, or a pi expression, or a sigma expression), a variable defined in a let expression, or a variable defined in a pattern.

If e, e_1, … e_n are expressions and x_1, … x_n are variables, then we will write e[e_1/x_1, … e_n/x_n] for the substitution operation. This is a meta-operation, namely, it is a function on the set of expressions of the language and not an expression itself. The expression e[e_1/x_1, … e_n/x_n] is simply e in which every occurrence of each of the variables x_i is replaced with the expression e_i.


There is a binary relation => on the set of expressions called the reduction relation. If e_1 =>=> e_n, we will say that e_1 reduces to e_n. If there is no e’ such that e => e’, we will say that e is a normal form. If e reduces to e’ and e’ is a normal form, we will say that e’ is a normal form of e and that e evaluates to e’. Every expression has a unique normal form.

The relation => is a meta-relation on the set of expressions of the language, that is you cannot refer to it explicitly in the language. This relation is used by the typechecker to compare expressions. The typechecker never compares expressions directly. To compare expressions e_1 and e_2, it first evaluates their normal forms and then compares them. Since normal forms always exist, the comparison algorithm always terminates, but it is easy to write an expression that does not evaluate in any reasonable time.

The reflexive, symmetric, and transitive closure of => is denoted by == and called the computational equality. We will often call terms t_1 and t_2 such that t_1 == t_2 simply equivalent.


Every expression has a type. The notation e : E is used for the judgement that an expression e has type E.

A type is an expression which has type \Type. The expression \Type is discussed in Universes.

Every variable has a type which is the one specified when the variable is defined or, if it is not specified, the one that can be inferred. An expression of the form x has the type of the variable x.

The type of an expression can usually be inferred automatically, but in rare cases, when it cannot be inferred, or just for the sake of readability it can also be specified explicitly. An expression of the form (e : E) (parentheses are necessary) is equivalent to e, but also has an explicit type annotation. In this expression, e must have type E and the type of the whole expression is also E (since it is equivalent to e).


A defcall is an expression of the form f a_1 … a_n, where f is the name of a definition with n parameters and a_1,, a_n are expressions. Note that classes and records do not have parameters and any defcall in this case is of the form f. Expression of the form f a_1 … a_n, where f is the name of a class, are class extensions, see Class extensions for details.

Defcall expressions have the following properties:

  • If f is a definition with parameters x_1, … x_n and the result type R, then the type of a defcall f a_1 … a_n is R[a_1/x_1, … a_n/x_n].

  • If f is either a class, a record, a data type, a constructor without conditions, an instance, or a function defined by copattern matching, then f a_1 … a_n is a normal form whenever a_1, … a_n are.

  • If f is a function defined as \func f (x_1 : A_1) … (x_n : A_n) => e, then f a_1 … a_n reduces to e[a_1/x_1, … a_n/x_n]. If f is a function defined by pattern matching or a constructor with conditions, then the evaluation of defcalls f a_1 … a_n is described in Functions. If f is an instance or a function defined by copattern matching, then the evaluation of defcalls f a_1 … a_n is described in Classes.

  • If f has n parameters and k < n, an expression of the form f a_1 … a_k is also valid and is equivalent to \lam a_{k+1} … a_n => f a_1 … a_n.