Combining relations
You can have more than one relation at the same time and combine them. For example if you also have relations
X ::= (i == 5 and j == 1)
or (i == 2 and j == 1)
Y ::= i == 5 and k == 3
then Z ::= X and Y
becomes
Z ::= i == 5 and j == 1 and k == 3
The i == 2
alternative of X
does not occur in the new relation since Y
does not hold for that value of i
. Also note that all equalities are preserved from both relations, the j == 1
equality and the k == 3
equality are part of Z
even though only one of the input relations stated them. The reason is that if an alternative in a relation does not say anything about a variable, it is assumed it may have any value (the not mentioned variable is independent).
This also works if both relations are a disjunction. For example P ::= X and Q
with
X ::= (i == 5 and j == 1)
or (i == 2 and j == 1)
Q ::= (i == 5 and k == 5)
or (i == 5 and k == 3)
or (i == 3 and k == 1)
or (i == 2 and j == 1 and k == 4)
gives
P ::= (i == 5 and j == 1 and k == 5)
or (i == 5 and j == 1 and k == 3)
or (i == 2 and j == 1 and k == 4)
All combined alternatives of X
and Q
that are not trivially false due to conflicting equalities become part of the result.
From a logical point of view the above is simple Boolean algebra, but if you change your view on what X
and Q
express you can see the hidden power of decision diagrams.
Instead of X holds when j is equal to 1 and i must be either 2 or 5, read X
as a function from i
to j
, that is, when i equals 2 then j equals 1 or when i equals 5 then j equals 1. Similarly, relation Q
can be read as a function from i
and j
to k
, that is, when i equals 5 then k becomes 5 or 3, if i equals 3 then k becomes 1, if i equals 2 and j equals 1 then k equals 4.
If you look at what P
contains in the function view, you can see you get the conjunction of both functions. In one X and Q
step you computed the combined function for all values of all variables at the same time!
An example of this property is used below, computing the result of an assignment.