My answer to this question is negative.
In order to create a pure star schema of (1:m) relationships between Dim tables and a fact table, at least 1 Foreign Key must correspond to 1 Dim table Primary Key which can then be successively merged as the Foreign Keys accumulate during the Fact table construction.
For example, using a simple model with 3 Dim tables:
- Imagine Dim2 holds FK of Dim1 PK but not Dim3 PK
- Dim3 PK is an attribute of Dim(2+1) but neither of Dim1 nor Dim2 tables separately.
Chaining or the process of merging in sequence (1) dim2 to dim1, and (2) dim(2+1) to dim(3), aggregates FKs together and avoids null fields in FK attributes of the fact table.
Without the chained FKs, even a single large fact table would contain null fields in - almost - every single FK attribute; and it would be impossible to produce measures in the fact table using values from several dim tables because the relationship is (1:0) in most or all cases. Surrogate keys are of no help in these instances.
In other words, there must be implicit relationships between dim tables for a fact table to make sense, correct?