Let's take an example as below:

　employee: ｛employee-NO, employee-name, company-name,...｝ [ R ]

　The "company-name" is the name of company for which an "mid-career recruiting" employee worked, "null (or nil)" value must be, therefore, observed in the "company-name" of the records for "hiring of new graduates" employees.

　You can create "subsets", "heterotypic" (or "atypical") subsets (of different formats), with the null values as below:

　　employee [ R ]
　　　｜
　　　｜
　　　× by null (company-name)
　　　｜
　　　｜＿＿＿＿ hiring of new graduates {employee-NO, employee-name,...}
　　　｜
　　　｜
　　　｜＿＿＿＿ mid-career recruting {employee-NO, employee-name, company-name,...}

　
　Subsets can be created through the following two rules:
　　(1) An entity has a "partition-code"
　　(2) An entity has "null value" in its attribute

　The subsets created under the declaration of "partition-code", are said to be "substantial subsets", and the subsets created owing to "null value", are to be "procedural subsets."

　Just ask if the "company-name" must belong to (or an attribute of) "employee", the answer is that it's not.
　It's, therefore, strongly recommended that you should create VE for "company-name" as below:

　employee:
　 ｛employee-NO, employee-name,...｝ [ R ]

　employee. company:
　 ｛employee-NO (R), company-name,...｝ [ VE ]