Entity-Relationship modeling

Entity-Relationship modeling

This is a variant (actually a predecessor) of object modeling (eg UML or CRC cards or Booch diagrams). In the latter, everything is an object. In ER modeling, we will make a distinction between entities (things) and relationships. As a simple example, students and courses are entities; but the enrolled_in table is a relationship. Sections most likely would be modeled as entities too, though there is a relationship to COURSE.

The ER process starts, like most software-engineering projects, with obtaining requirements from users. What data needs to be kept, what queries need to be asked, and what business rules do we build in? (For example, if the DEPARTMENT table has a single column for manager, then we have just committed to having a single manager for each department.)

The goal of the E-R modeling process is to create an E-R diagram, which we can then more-or-less mechanically convert to a set of tables. Both entities and relationships will correspond to tables; entity tables will often have a single-attribute primary key while the key for relationship tables will almost always involve multiple attributes.

Here is an E-R diagram for the OFFICE database.  (The figure below was Fig 3.2 in an earlier edition of E&N; it is Fig 7.2 in the 6th edition.)

E-R diagram for COMPANY database

This style of diagram was introduced by Peter Chen in 1976, and is sometimes known at the Chen format. It is well-suited to conceptual and logical modeling of a database, in that it makes a clear distinction between entities and relationships.


The first modeling step is to identify the entities. These should represent physical things, such as employees or parts or (more abstractly) departments. Note that customer_orders might be modeled as an entity at this point, but might also be modeled as a relationship.

Entities have attributes, which will later more or less become the fields. For each attribute we have the following aspects:
We also must decide which attributes can be NULL.

Attributes at this point should not be references to other tables; instead, we will create those references when we create relationships.

The book uses () to represent sub-attributes of composite attributes, and {} to surround multi-valued attributes.

Traditionally we represent the entity with a rectangular box, and the attributes are little oval tags.

An entity type is our resulting schema for the entity; the entity set is the actual set of entities.

In the diagram, we will underline the key attributes. If a key is composite, say (state,regnum), then we make a composite attribute out of those pieces.

This is a slight problem if the key can be either (state,regnum) or (state,license_plate); how could we best address this?

Note that key attributes really represent constraints.

In the early stages, we allowed entity attributes to be composite or computed or multi-valued; all of these will eventually be handled in specific ways as we translate into SQL.

Often there is more than one way to do things. In the COMPANY example, we might list dept as an attribute of EMPLOYEE, and eventually conclude that because dept represented an instance of another entity (DEPARTMENT), we would have a foreign-key constraint on EMPLOYEE.dept, referring to DEPARTMENT.dnumber.

Note, however, that we could instead list employees as a multi-valued attribute of DEPARTMENT. One reason for not doing this is that we do want to minimize the use of multi-valued attributes, but this arrangement would have been a possible option. Later, we even could implement this second approach by adding an attribute dept to the EMPLOYEE table (the table, not entity).

We actually could have both forms, but we would need to understand the constraint that if employee e is in the employees multi-valued attribute for DEPARTMENT d, then department d must be be the value of the EMPLOYEE e's dept attribute. That is, the dual attributes would have to be inverses.

As for naming entities, a common practice (used by E&N) is to name them with singular nouns. Nouns because they should represent things; singular for the individual objects. Eventually we will have a table of employees, plural, but we call it EMPLOYEE to represent what entities it contains.

Weak entities

The usual definition of a weak entity is that it is an entity that does not have key attributes of its own. The classic example is the DEPENDENT entity, with attributes name, birth_date, sex and relationship; a dependent is uniquely determined by the name and the employee to whom the dependent is associated. You might wonder why we don't add an attribute employee at the beginning, and have ⟨name, employee⟩ be the key. One problem with that approach is that employee is a reference to a different entity; such references should really be described as relationships. After all, the EMPLOYEE is really someone else, not an atomic attribute of the DEPENDENT person itself (the other attributes of DEPENDENT are atomic). We will say, instead, that there is a relationship between the DEPENDENT entity and EMPLOYEE; this relationship is the identifying relationship for DEPENDENT.

In general, during the design process, the statement that "dependents do not have a key" is subject to interpretation; we can always declare that the associated employee's SSN is part of the key. However, the point is that dependents do not have a "natural" key that is an attribute of the dependent itself. Also, using the "employee_ssn" as an attribute is suspect because it realistically is an attempt to refer to another table.

There is a total participation constraint between DEPENDENT and EMPLOYEE; every DEPENDENT must be connected to some EMPLOYEE. As the book points out, however, every DRIVERS_LICENSE is associated with some PERSON, but the DRIVERS_LICENSE entity does in fact have its own key: the drivers_license_number.

The way that DEPENDENT could become a strong entity is if we added its own key: dependent_ssn. But typically the SSNs of dependents are not known (minor dependents may not even have SSNs), so we choose not to implement the database this way.

The DEPENDENT entity does have a partial key: the attribute name, which, together with the associated EMPLOYEE object, does define a key.

We could also represent dependents as a multi-valued, composite attribute of EMPLOYEE.

Fig 7.8 (6th-edition numbering) lists all the entities:

basic entities

Entity summary

Here's a summary for the construction of entities:


Initially we arrive at Fig 7.8, with four entities: DEPARTMENT, PROJECT, EMPLOYEE, DEPENDENT. Note that Works_on here is shown as an EMPLOYEE attribute; it could also be represented as a PROJECT attribute. How are we representing department membership? Who works on what? Who is in charge of what projects?

Note some of the attributes in figure 7.8 refer to other entities. These are our first relationships; these will likely end up translated into foreign key constraints.

A relationship formally is a set of ordered tuples ⟨e1,e2,...,en⟩ where each ei is a member of entity Ei. Some entities here may simply be attributes (eg the hours attribute of the WORKS_ON relationship ⟨employee,project,hours⟩.

The tuples in a relationship must each have a clear meaning to the application. Relationship names are usually verbs, and should make sense "left to right" (and sometimes top to bottom). That is, we would prefer the relationship name supervises because it fits in with
    SUPERVISOR----- supervises ------EMPLOYEE
We could also use
    EMPLOYEE ----- reports_to ------ SUPERVISOR

Most relationships are binary (possibly with added attributes); ternary and higher-degree relationships are less common (and less tractable).

In early stages we may model a relationship as a (typically multivalued) entity attribute; consider again how we modeled WORKS_ON in Figure 7.8.

When a relationship involves multiple entities, we can assign a role name to each entity. Commonly this is just the name of the entity (eg EMPLOYEE), but in relationships between an entity and itself (so-called recursive relationships), we have to use different names. Consider the example of the SUPERVISES relationship.

Example: fig 7.11; note that the righthand SUPERVISION oval contains references to pairs of entities in the lefthand EMPLOYEE oval. (Which numeric label is used to indicate the supervisor?)

For entities, it is often the case that we elect to use synthetic keys: arbitrarily generated "ID numbers". This makes sense for departments and employees. Relationships, however, typically have a natural key consisting of one primary key from each entity; using synthetic keys (eg order numbers) should stand out. A good example of this is the GRADE_REPORT table, indexed by student_number and section_identifier (and with attribute grade).

How should we model SECTION in the school database? We did model it as an entity, but could we model it as a ternary relationship between course, semester, and instructor? No, if we allow an instructor to teach two sections of the same course in the same semester.

What about an INVOICE? This consists of a number of ITEMs, each with quantity, ordered by a single CUSTOMER. We can create a relationship ORDERS between CUSTOMER and ITEM, but an invoice is more than that. If a customer places multiple orders on the same day, the customer likely expects them to remain different. So instead we might choose to have an entity for INVOICE, with attributes invoice_number (synthetic), and date, and customer, and then create a relationship ORDERS between INVOICE and ITEM, with attributes for price and quantity:


See invoice.


Binary relationships can be classified as 1:1, 1:N, N:1, or M:N. In the WORKS_FOR relationship, between DEPARTMENT and EMPLOYEE, this is 1:N. Each employee works for 1 department, but a department can have multiple employees. (Again, the 1 here in 1:N represents a constraint; the N represents no constraint. It is not actually required that all departments have multiple employees.)

The MANAGER relationship is 1:1 (though see the note): every dept has one manager and vice-versa. This is a 1-1 relationship between EMPLOYEE and DEPARTMENT. Note that most employees are not managers; this does not change the fact that no employee manages two departments. See Fig 7.12 for a diagram representing this.

Note: that the MANAGER relationship is 1:1 expresses a business rule: no employee manages more than one department, and no department has two managers. The latter is pretty universal; the former, while common, is not.

Many relationships are 1:N (one-to-many):

    DEPARTMENT ----1---  employs ----N----- EMPLOYEE  (or employee works_for department)
    EMPLOYEE -----1----- supervises ----N------EMPLOYEE (boss is on left side)
    DEPARTMENT ----1---- controls-----N------PROJECT

Think of "1 department = N employees"; the 1 goes on the side that the other entity can have only 1 of. The 1 goes on the "larger" unit: a department is made of N employees, a boss supervises N employees, a department controls N projects.

See Fig 7.9.

The supervises relationship is "recursive" (a better word, used in the UML community, is "reflexive"). See figure 7.11 for a diagram.

The WORKS_ON relationship is M:N.
Similarly, the enroll relationship is M:N
STUDENT -----M----- enrolls ----N----SECTION
A section may have several students; each student may enroll in several sections.

See fig 7.13 for a diagram of the WORKS_ON relationship.

What do we do if, after we've gotten started, we decide that the location attribute of a DEPARTMENT should be multi-valued? We can model multi-valued attributes as relationships instead:

    DEPARTMENT ----N----is_located_at-----M----LOCATION

Clearly, we would not want this to be 1:M, which would mean that a location could be used by only one department. If we do decide that departments have single locations, we go back to an N:1 relationship:

    DEPARTMENT ----N----is_located_at-----1----LOCATION

Participation constraints on relationships

Suppose every employee must work for some department. Then the WORKS_FOR relationship involves total participation of the EMPLOYEE entity. The MANAGES relationship involves partial participation of the EMPLOYEE entity, at least as far as supervisors are concerned.

We represent total participation by a double line, and partial by a single line.

Relationships can have attributes; eg hours of WORKS_ON or grade for the GRADE_REPORT table.

As was described above, entities usually have a single (possibly composite) key; entities are often given a synthetic key (ie an employee_id or student_number). Relationships typically have a key with as many attributes as the degree of the relationship. Synthetic keys are often awkward for these.

The key to a relationship should be a composite of the keys to each entity. Otherwise the relationship is not just about the two entities involved.

Note that synthetic keys work very well for joins.

Now we should be able to go through Figure 7.2 (E&N p 204, below) in detail. The relationships are supervises, works_for, manages, controls, works_on, and dependents_of. Note that the name "supervision" is awkward; it is not clear who is supervising whom. As a result, the entity links need annotation with the role names "supervisor" and "supervisee". However, such annotation is often a good idea for clarity.

(The figure below was Fig 3.2 in an earlier edition of E&N; it is Fig 7.2 in the 6th edition.)

ER diagram for the COMPANY database

Sometimes, as we rethink things, an attribute can be changed to a relationship, or vice-versa. Sometimes an attribute may be promoted to an entity, particularly if it was used in several other entities, in which case we may also add a relationship to those other entities.

Relationship attributes can sometimes be moved to entities. For a 1:1 relationship, the relationship attribute can be moved to either entity. For a 1:N relationship, the relationship attribute can only be moved to the N side. Consider the earlier examples:

    DEPARTMENT ----1---  employs ----N----- EMPLOYEE     attribute: start_date, etc
    EMPLOYEE -----1----- supervises ----N------EMPLOYEE    attribute: review_date
    DEPARTMENT ----1---- controls-----N------PROJECT       attribute: project_budget_num

Sometimes we have entity attributes that need to be translated into relationships. See Section 7.6. We would move manager information from the DEPARTMENT entity to the MANAGES relationship. We started out with manager as an attribute of departments, but later realized that there was a relationship involved because two entities were involved: DEPARTMENT and EMPLOYEE. This suggests the need for a relationship.

We would move controlling-department information from the PROJECT entity to the CONTROLS relationship. We would remove department, supervisor, and works_on from EMPLOYEE. Note that some of these will eventually be added back. At this point, we should have eliminated most multi-valued attributes.

Entities usually have a single (possibly composite) key; entities are often given a synthetic key (ie an employee_id or student_number). Relationships typically have a key with as many attributes as the degree of the relationship. Synthetic keys are often awkward for these.

The key to a relationship should be a composite of the keys to each entity. Otherwise the relationship is not just about the two entities involved.

Note that synthetic keys work very well for joins.

ER diagram for the STUDENT database


    course----< PREREQUISITE >---- course

    section----< IS_OFFERING_OF >---- course

    student ----< REGISTERS_FOR >---- section

(min,max) annotation

Instead of labeling lines connecting a relationship to an entity with 1, M, or N, we can also use a (min,max) notation, meaning that each entity e in the entity set E must participate in at least min entries of the relationship, and at most max. If min>0, the participation is total; min=0 means partial participation. The max is denoted N when we mean it is allowed to be >1.

Note that a 1-N relationship would have the values reversed using the (min,max) notation:
    DEPARTMENT ===1===  employs ===N=== EMPLOYEE             // 1 dept = N employees
    DEPARTMENT -----(1,N)----  employs --- (1,1)----- EMPLOYEE   //  dept can have 1..N employees      

The second line, above, means that a given department can appear multiple times in the EMPLOYS relationship; ie a department can have multiple employees. An employee can appear only once; that is, can work for only a single department. Every employee must appear at least once, and every department.

(Recall that the parallel lines === in the first line above represent total participation: every department has an employee, and every employee works for some department. This is represented in the second line with the 1 as the first coordinate of each pair.)

Full example of (min,max) annotation: Fig 7.15
Why do they say
    department ----(4,N)---employs ----(1,1)----employee

Manages relationship: put into entity on (1,1) side rather than entity on (0,1) side

(0,1) doesn't say anything about how often the participation can be 0. Consider
    department --- MANAGED_BY------- (0,1)-----manager (employee)        
    supervisor  ---  SUPERVISES  --- (0,1)--- supervisee (employee)
Most employees are not managers. Almost all employees are supervised.

Crows-foot diagrams

The above Chen-style diagrams are characterized by separate symbols for entities and relationships; they are best suited for so-called logical design, before the relationships are translated into tables. For describing the physical model of a database, the so-called crows-foot notation is often useful. In this notation, there is a box for each table. The box lists the attributes of that table, identifying keys. Boxes represent entities after the relationships have been transformed into entity attributes or into new tables, as appropriate.

Lines between boxes represent relationships, and are often associated with foreign-key constraints. Dashed lines are used for ordinary relationships, and solid lines for weak-entity relationships. Relationships don't get their own boxes because at this point they have been reduced to entities (that is, tables), and thus no longer have their own attributes.

Cardinality is represented by how the ends of these lines are decorated. Here are the basics:

────┼    one

────<    many

───┼<    one or many

──o─<    zero or many

───┼┼    exactly one

For an example, see dellstore.png. Note that, in this example, some of the "crows' feet" get partially obscured by the drop-shading on the boxes.

UML diagrams

See Figure 7.16. The previous example, dellstore.png, is similar. UML diagrams have space for operations,which in the world of databases we're not much concerned about. The big boxes are for entities; relationships have been reduced to boxes that annotate links. A (min,max) notation is used, but the label goes on the opposite entity.

UML relationships (actually, ER relationships as well) may either be of association or of aggregation. As examples of the latter we have:

How do we translate this to tables?

We'll get to this next, but note that a 1:1 relationship can be represented as an attribute of either entity. A 1:N relationship can be modeled as an attribute of one of the entities (the entity on the side of the N). M:N relationships must get their own table.

ER-to-relational mapping

How do we build a database schema from an ER diagram?

Step 1: regular entities
We define a table for each non-weak entity. We use all the leaf attributes; composite attributes are represented by their ungrouped components. Keys are also declared. Attributes that were earlier pushed into relationships are not yet included.

Step 2: weak entities
We create a table for each weak entity, adding the keys for the owner entity type (or types) (this would mean employee ssn), and adding a foreign key constraint to the owner-entity table.

We are likely to use the CASCADE option for drop/updates: if an employee ssn is updated, then the dependent essn must be updated, and if an employee is deleted, then all the dependents are deleted too.

Step 3: binary 1:1 relationships
Let S and T be the participating entities to 1:1 relationship R. We pick one of the two -- say S -- and add to S a column that represents the primary key of T, and all the attributes of R.

It is better to choose as S the entity that has total (or at least closer to total) participation in R. For example, the manages relationship between departments and employees is 1:1, but is total only for DEPARTMENT, and is nowhere near total for EMPLOYEE. Thus, we add a column manager to DEPARTMENT. However, adding a column manages to EMPLOYEE would work.

We also add a foreign key constraint to S, on the new attribute, referring to the primary key of T.

One alternative is to merge S and T into a single relationship; this makes sense only if both have total participation in R. This means that S and T each have the same number of records, and each record s in S corresponds to exactly one t in T.

A third alternative is to set up a table R containing ⟨sk,tk⟩ key pairs.

Step 4: binary 1:N relationships
Let us suppose S---N---R---1---T. We now add T's key to S as an attribute with foreign-key constraint. We must add T's key to S; we cannot do it the other way around. In the relationship

    DEPARTMENT ----1---  employs ----N----- EMPLOYEE

we would have S be EMPLOYEE; we would put a dno column in EMPLOYEE (why can't we add an essn column to DEPARTMENT?)

An alternative is the ⟨Skey,Tkey⟩ keypair table. This might be more efficient if only a few s in S participate in the relationship; otherwise we would have many NULLs in the T-column of S.

Step 5: binary M:N relationships
Here we must create a table R of tuples including the key of S (sk), the key of T (tk), and any attributes of R; we can not push the data into either S or T. Call the new table also R (note that E&N call it S). The sk column of R should have a foreign key constraint referring to the key column of S, and the tk column of R should similarly have a foreign key constraint to the key column of T.

The WORKS_ON table is a canonical example; so is the GRADE_REPORT table.

Again we would likely to use the CASCADE option for deletion or update of records in the participating entities S and T.

Step 6: multivalued attributes
If we have any left, they must be moved into their own tables. For example, if employees can have several qualifications (eg degrees or certifications), we would create a table QUALIFICATION with two columns: essn and qualification. The DEPT_LOCATIONS table is similar. Again, we would have an appropriate foreign key constraint back to the original table.

Step 7: higher-degree relationships
These are handled like binary M:N relationships. Sort of.

More on Foreign Keys

Here's the seven-step ER-to-relation algorithm again, slightly simplified:
  1. create a table for each regular entity
  2. create a table for each weak entity, adding the key field from the owner entity as a foreign key for the new entity. Example: table Dependents, with a column essn referencing Employee.
  3. for binary 1:1 relationships between entities E1 and E2, pick one of them (eg E1) and add to it a field containing the key to E2. Make this a foreign key in E1. Example: the dept-manager relationship, implemented as column mgr_ssn in table Department.
  4. for binary 1:N relationships between E1 and E2, E1---1---R---N---E2, add a column to E2 containing the key of E1 (we can not implement the relationship with a column in E1!). Make this new column a foreign key in E2, referencing E1. Example: the works-for relationship, implemented as column dno in table Employee.
  5. For binary N:M relationships between E1 and E2, create a new table R consisting of ⟨E1.key, E2.key, R.attributes⟩. Make E1.key and E2.key foreign keys in R. Example: The works-on relationship, implemented as a table which has as key the pair ⟨essn,pno⟩, each of which are keys to their respective tables.
  6. For multivalued attributes of entity E, create a new relation R. One column of R will be E.key; this should be a foreign key in R.
  7. ternary and higher-degree relationships: like step 5.
Joins arise in steps 2, 3, 4, 5, 6, and 7, for recovering the original relationships (or attribute sets for 6, or entities for 2). In every case, the join field is a key of one relation and a foreign key in the other.

Not all joins are about recovering relations from an ER diagram.

Also, I said earlier that entity T should not have an attribute that was another entity of type S; instead, we should create a relationship R between T and S. If S was at all a candidate for an attribute, each T would be related to at most one S and so this would have cardinality constraint T---N---R---1---S. Then, when we did the above conversion, in step four we would add S's key to T with a foreign key constraint referring to S.

But suppose we did add S as an entity attribute to T. Then we would end up with the same situation: we would use the key of S as an attribute of T, and create the same foreign-key constraint. So in the end we get the same thing.


How shall we model invoices? An invoice is a collection of parts ordered, each with a quantity. One way is to try to model an invoice (or at least an invoice_item) as a binary relationship between CUSTOMER and PART, with attributes date and quantity. An invoice is thus all the items to the same customer with the same date.

                 /        \
              date       quantity

An invoice would be uniquely determined by the date and customer, so if Customer c ordered Part p on Date d with Quantity q we would have ⟨c,p,d,q⟩ ∈ Invoice. Given ⟨c,d⟩ we can look up all the parts p and, for each part, the quantity.

For a given c and d there might be multiple parts p that were part of the invoice. We can search the Invoice table for those ⟨c,d⟩, and find the balance of each record.

Problem: INVOICE is not actually a "relationship set" for entities Customer and Part, as defined in EN6 §7.4.1; a relationship would have to be a subset of the cross product Customer × Part; we can add attributes, but the ⟨c,p⟩ part is supposed to determine the record. However, the values of c and p do not determine an INVOICE record. The key for INVOICE is the triple ⟨c,p,d⟩; a customer c can order 100 units of d on 2005-12-01 and then 200 more units on 2006-01-27.

If we want INVOICE to be a relationship, we need to recognize that it is really a ternary relationship between Customer, Part, and a single-attribute entity Order_Date. Ternary relationships tend to be inefficient. None of the relationships in the COMPANY database were ternary; in WORKS_ON, a record was uniquely determined by the essn and the project_num; in WORKS_FOR, by the ssn and the dept_no.

Even if we do this, we have another issue: if a customer places multiple orders on the same day, the customer likely expects them to remain different.

So, instead, a much more common approach (which also allows multiple invoices on a single day) is to make Invoice an entity, with synthetic key invoice_num. That is, we declare that orders are "things" rather than relationships. This is an instance of a rather general strategy that might be called the synthetic-key trick: convert a putative relationship to an entity by assigning a "serial number" to each tuple in the relationship. In this case the synthetic key has a natural interpretation: we number each order as it is placed. For the works_on relationship of the COMPANY database we might use a synthetic key called Job_Assignment_Num; for the Works_For relationship between Employees and Departments we might use Job_Association_Num.

After we create an entity Invoice, with attributes Cust_id and Order_date and identified by invoice_num, we will create a relationship Invoice_Item, between Invoice and Item, with attributes for price and quantity. This table effectively lists what a given Invoice actually includes:

Table Invoice_Item

Relationship Invoice_Item is often called Orders: the relationship identifies all the items ordered.

    Invoice --------- Invoice_Item ---------- Part

(Actually, Invoice also has a relationship Ordered_By to Customer; that is N:1 so I have immediately implemented it by adding a Cust_id attribute to Invoice. We replaced one sort-of-binary relationship Invoice between Customer and Part with a new entity Invoice with binary relationships to each of Customer and Part. But only the relationship with Part is M:N and so needs it own table.)

We implement Invoice_Item as its own table listing invoice numbers, part numbers and quantities. The primary key is the pair ⟨invoice_num, part_num⟩; the table also has an attribute for quantity (and perhaps also for current_price, or for discount). The INVOICE table (table Orders in the dellstore database) itself might look like this:

Table Invoice

and the table Invoice_Items (table Orderlines in the dellstore database) might look like this:

Table Invoice_Item

The Invoice_Item table has a true dual-attribute key, as it represents an M:N relationship between invoices and parts. (Though note that, in the Dellstore, the primary key for Orderlines is in fact the synthetic key OrderlineID.)

Bottom line:
Is an invoice more like an entity or a relationship?

What about a course registration?

ER-to-relation mapping of ternary and other higher-degree relationships

Consider the SUPPLY relationship on a supplier s, project j, and part p. The tuple ⟨s,j,p⟩ is included if supplier s supplies part p for project j.

We might try to model this with three binary relationships, SUPPLIES(s,j), CAN_SUPPLY(s,p), and USES(j,p). It is true that if ⟨s,j,p⟩ is in SUPPLY, then ⟨s,j⟩ is in SUPPLIES, ⟨s,p⟩ is in CAN_SUPPLY, and ⟨j,p⟩ is in USES. But the converse is not true (example). If we build the three binary tables, we cannot reconstruct the ternary table.

See Fig 7.17.

As for binary relationships, a ternary relationship key is a triple of keys from each participating entity.

Ternary relationships can be problematic, and so we often include corresponding binary relationships, sometimes even if they are reundant.

One approach is to model a ternary relationship as a weak entity, with three identifying relationships (Fig 7.17(c)). This is usually done only when the underlying ER-modeling tools do not support ternary relationships. The resultant entity has the requisite three-attribute key to describe the ternary relationship accurately.

Alternatively,  we can give SUPPLY a synthetic ("surrogate") key, supply_id, and then relate it to SUPPLIER, PROJECT, AND PART by binary relationships. The synthetic key would uniquely determine a ⟨s,j,p⟩ triple; we can say this in SQL by saying that ⟨s,j,p⟩ is a secondary key. With a synthetic key we now have an entity SUPPLY, with key supply_id si, and with three relationships SUPPLIES3(si, s, j), CAN_SUPPLY3(si,s,p) and USES3(si,j,p). We may still need a ternary relationship explaining the relationship of all three, but from the entity SUPPLY(supply_id, supplier, project, part) we can now reconstruct the original ternary table.

The next example is OFFERS, for a school database; see Fig 7.18; ⟨i,s,c⟩ belongs to OFFERS if INSTRUCTOR i teaches COURSE c during SEMESTER s. Again, the binary projections fail to adequately model the ternary relationship. E&N suggest that if one of the binary relationships is 1:1 (eg if the CAN_TEACH(i,c) relationship is 1:1), then this does work, but that is seldom if ever the case.

Actually, in a real school database one would not use OFFERS, one would use SECTION. The latter would likely be an entity, complete with synthetic key section_id. However, SECTIONs are determined by instructor i, course c, semester s, and timeslot ts, or, alternatively, by semester and course and section_number or semester and class number (Loyola's mechanism).

Ternary relationships can have cardinality tags, like binary relationships, but they are not as straightforward. For the SUPPLY relationship, suppose each project,part pair (j,p) can have only one supplier. Then we might put a 1 on the SUPPLIER link. But it might also be the case that each project j uses a unique supplier,part pair (s,p) (that is, each supplier can supply only one part to each project). We now have an unrelated 1 on the PROJECT link.

The (min,max) relationship is more straightforward: E------(m,M)-----R means that for each e in E, there are at least m tuples involving e in R, and at most M. But this method cannot describe the case where each project/part combination can have only one supplier.

Translation from higher-degree relationships to SQL table definitions is done the same way as for binary M:N relationships: we create a table consisting of columns for keys for each of the participating entities, and any relationship attributes. The entity keys form the primary key for the new table, and each entity key has a foreign key constraint referring back to its defining entity table.

The Enhanced ER (EER) model

In this model, we allow for some forms of inheritance.

Figure 8.1 is a starting point. Note that there are several kinds of employee; some kinds participate in relationships and some do not. Note the symbol denoting inheritance; arrows from the parent class to the child class are more common in OOP design.

Ultimately, we may implement the diagram of Fig 8.1 with an EMPLOYEE table, and also tables for SECRETARY, TECHNICIAN, ENGINEER, MANAGER, and HOURLY_EMPLOYEE, each indexed by the ssn and having additional columns for the subclass-specific attributes.

To be in a subclass, you must also be in the superclass.

Note that some subclasses have subclass-specific attributes, and other subclasses have subclass-specific participation in relationships (eg Manager and Hourly_Employee).

The circled (d) in Fig 8.1 stands for "disjoint"; one cannot be a SECRETARY and a TECHNICIAN. However, one can simultaneously be a SECRETARY, a MANAGER, and an HOURLY_EMPLOYEE (at least as the relationship is drawn). In practice, it is likely that each of Secretary, Technician, Manager and Engineer would also belong either to Hourly_Employee or Salaried_Employee. In general, membership in multiple subclasses is to be allowed unless explicitly forbidden with the (d) notation.

The alternative to (d) is (o), for overlapping. EN's example for overlapping subclasses is in Fig 8.5: the parent class is PART and the subclasses are MANUFACTURED_PART and PURCHASED_PART. Some parts can be both here.

Fig 8.2 example (showing disjointness). Secretaries, Engineers and Technicians are all Employees, but everyone belongs to at most one category.

Generalization is the process of realizing that two existing entities, CAR and TRUCK, are really both instances of VEHICLE. See fig 8.3.

(But note there is some debate as to whether CAR and TRUCK are really disjoint classes; below is an El Camino.)
El Camino (car in front, truck in back)

Sometimes subclass membership is determined by a field value or Boolean expression involving the parent class (eg jobtype = engineer). Note that such "tag" fields are frowned upon in classic OOP in, say, Java. This arrangement is also called attribute-defined subclassing (or specialization). If the value of a single attribute determines the subclass, this necessarily leads to disjoint subclasses. This is illustrated in Fig 8.4. In other examples, subclass membership represents a form of new data; these are user-defined subclasses. As new subclass records are inserted into the database, the appropriate subclass must also be indicated.

Besides disjoint/overlapping, subclasses may be described as total or partial. Total means that every member of the base class must be in some subclass (ie that the base class is abstract in java notation). Partial means that base-class-only objects may exist. In the Fig 8.1 example, every employee is either salaried or hourly, so the right-hand subclass is total. The double line is used to denote this. Note that this has nothing to do with any of the other subclass relationships.

Multiple inheritance means that we may end up with a lattice of relationships: see Fig 8.6 and Fig 8.7. If multiple inheritance is involved, the classes will not be disjoint. A common OOP issue with multiple inheritance -- resolving method or attribute names when the same name is used in more than one parent class -- is usually handled by requiring attribute names to be unique.

Postgres supports table inheritance (see www.postgresql.org/docs/9.5/static/tutorial-inheritance.html)

create table employee (
    -- as before
create table engineer (
    degree    varchar(30),
    eng_type varchar(30)
) inherits (employee);
Note that we do not list ssn!
Now look at the result of \d engineer.

Let's add an engineer:
insert into engineer values('ralph', 'j', 'wiggum', '000000001', null, 'no fixed abode', 'm', 34000, null, 5, 'Loyola 2021', 'mechanical');
insert into manager values('ralph', 'j', 'zoggum', '000000002', null, 'no fixed abode', 'm', 43000, null, 5, 'Loyola 2020', 'MBBS');
select fname, lname from engineer;
select fname, lname from employee;
select fname, lname from employee*;
select fname, lname from only employee;
select tableoid, fname, lname from employee;      -- oid = object identifier
select tableoid::regclass, fname, lname from employee;
We see the engineers are also employees.

We can also add this:

create table manager (
    degree    varchar(30),
    projectcount  integer
) inherits (employee);

Warning: the postgres documentation contains the following note:
Note: Although inheritance is frequently useful, it has not been integrated with unique constraints or foreign keys, which limits its usefulness.
As an example:
insert into engineer values('dalph', 'k', 'ziggums', '000000001', null, 'no fixed abode', 'm', 34000, null, 5, 'Loyola 2022', 'electrical');
Oops. (Try 'select ssn from employee')
create unique index engindex on engineer(ssn);
alter table engineer add constraint engineer_index primary key using index engindex;
Now try again to insert Dalph Ziggums.

We can delete the engineer and manager tables with 'drop table engineer' and 'drop table manager'.

Union types

Sometimes the best way to model a Vehicle type is simply as a union of existing types Car and Truck. See Fig 8.8 for two examples.

Union types generally mean that the designer has not taken advantage of any common attributes. It is particularly helpful to identify a primary key that can be moved to the base class.

EER-to-Relations mapping

Given a parent class C with subclasses S1, S2, ..., Sm, here are some options for defining relations:

A. Create a table for representing C, and separate tables for each Si. Each Si will include a column representing the corresponding C data.
For example, C might be the Employees table, with key ssn; we might have tables for Secretary, Technician and Engineer also with keys ssn.

Multiple inheritance can be handled by having someone in the Employees, Technician and Engineer tables.

The Postgres inheritance example above follows this approach, where the Si will inherit from C. There is no need to explicitly include a column from C in the Si.

B. Create a separate table for each Si, including in each table all the common attributes. This only works if subclassing is total; that is, if every member of the parent class is in some subclass (why?). It becomes inefficient of the Si are not disjoint.

C. Create a single table including all attributes of C and all the Si, and an additional type attribute indicating to which Si the record belongs. For example, we might have fields
fname, lname, ssn, address, type (secy, tech, eng), typing_speed, Tgrade, Eng_type, Eng_degree, year
The value of the type attribute determines which of the remaining attributes are actually used. Disjoint subclasses are necessary here, and space may not be used efficiently. Multiple inheritance is not supported.

D. Like C, but instead use m Boolean attributes to indicate membership in each Si:
fname, lname, ssn, address, is_secy, typing_speed, is_tech, Tgrade, is_eng, Eng_type, Eng_degree, year

This mechanism can handle multiple inheritance reasonably well.