The execution plan for start join first must access the fact table. However before accessing this table a set of access primary keys must found as intersection pk-filters for conditions
imposed on conditions.
Example:
Consider a star join schema with the fact table F and 2 dimension table D1 and D2.
Let D1,D2 and F are defined as follows

create table D1 (a int primary key, c int, index d1_c(c));

create table D2 (b int primary key, c int, index d2_c(c));

create table F (pk int primary key, a int, b int, index f_a(a), index f_b(b));

Consider the query

select count(*) from F, D1, D2

where D1.a=F.a and D2.b=F.b and D1.a and D1.c in (C1,C2) and D2.c in (C3,C4);

Using the index d1_c pk-filter for the condition D1.c in (C1,C2) is built D1_F. Then using the index f_a pk-filter D1_F for the table F is built from D1_F.
Using the index d2_c pk-filter D1_F for the condition D2.c in (C3,C4) is built . Then using the index f_a pk-filter F_F2 for the table F is built from D2_F.
F_F1 is intersected with F_F2. The intersection provides the set of pk keys of F that has to be accessed.