Postgresql order by and limit

Question

I have a table R from which want to select the 3 predecessors or successor based on the z column. I have a btree index over the z column.

SELECT key, z, x, y FROM R
    WHERE z <= some_value
    ORDER BY z DESC 
    LIMIT 3

retrieves the predecessors in 'no time' but

SELECT key, z, x, y FROM R
    WHERE z >= some_value
    ORDER BY z ASC
    LIMIT 3

is useless for retrieving the next entries because is starts to scan through all the rows that have z >= some_value. Explain gives identical explanations on how the query is satisfied. Why is this and what could I do about it?

EDIT: There are two indexes actually, one ascending and one descending. Explain for both queries above says: "Limit (cost=0.00..31.91 rows=3 width=20)" " -> Index Scan Backward using zreverse on r (cost=0.00..17727815.24 rows=1666667 width=20)" " Filter: (z >= some_value")

Answer 1

Uhm...

some_value used in the WHERE clause is calculated by a simple function which was (falsely) defined VOLATILE. Changing it to STABLE solved the issue.

Answer 2

Is the index on z an "ordinary" btree index with nothing else specified? Or was it created in "descending" order, ie as

create index <name> on R(z) desc;

?

Quite possibly the addition of the opposite order index to the one you have may help.