Delete array element by index

There is no built-in function that I would know of. In Postgres 9.3+ you have the silimar function array_remove(anyarray, anyelement) to (per documentation):

remove all elements equal to the given value from the array (array must be one-dimensional)

Say, you have a 1-dimenstional array arr and want to delete the 3rd element:

If array elements are unique, you can:

SELECT array_remove(arr, arr[3]);

If uniqueness is not guaranteed the above might delete multiple elements. You can concat two array slices instead:

SELECT arr[1:2] || arr[4:2147483647] AS arr_without_3rd_elem

2147483647 is just the maximum possible array subscript to cover all possibilities. Works for any version since at least 8.3, probably more.
About array subscripts:

• Normalize array subscripts for 1-dimensional array so they start with 1

SQL Fiddle.

If you need this a lot, create a function. I suggest an IMMUTABLE polymorphic SQL function. (An implementation in C would probably be faster.)

CREATE OR REPLACE FUNCTION f_array_remove_elem(anyarray, int)
  RETURNS anyarray LANGUAGE sql IMMUTABLE AS
'SELECT $1[1:$2-1] || $1[$2+1:2147483647]';

Call:

SELECT f_array_remove_elem('{1,2,3,4,5}'::int[], 4)    AS i
     , f_array_remove_elem('{foo,bar,baz}'::text[], 1) AS t;

Result:

i         | t
----------+-----------
{1,2,3,5} | {bar,baz}

It would be more complicated:

• for multi-dimensional arrays.
  You cannot just remove a single element, dimensions have to stay in sync. So you have to define which slice to remove exactly. Then I would probably unnest the array using generate_suscripts(), remove selected subscripts and array_agg() / array_agg_mult() the result.

• for non-standard array-subscripts.
  Normalize subscripts in the source like demonstrated in the linked question or adapt subscripts in the expression with array_lower() and array_upper() ...