This morning, I revisited a stackoverflow question, I answered awhile back.  I wanted to provide a more complete answer and also saw the opportunity for a new post.  The question asked, "What's the difference between JPA and Hibernate?"

The Java Persistence Architecture API (JPA) is a Java specification for accessing, persisting, and managing data between Java objects / classes and a relational database. 

Let's take a further look at this definition.  As the API portion of the name implies, JPA is a specification, meaning it provides guidelines for developing an interface that complies with a certain standard.  While JPA dictates an interface, it does not provide an implementation of that interface, meaning there is no underlying code that performs the operations to persist an object to a relational database.

It should also be noted the term Object Relational Mapping, is often used to describe the process of accessing, persisting and managing data between Java objects and a relational database.

To look at the concept of JPA from another perspective, imagine if you were provided this interface.

public interface JPA {
	
	public void insert(Object obj);
	
	public void update(Object obj);
	
	public void delete(Object obj);
	
	public Object select();
	
}

Without further development, what immediate value does this interface provide?  While this interface has potential to provide value, at this point, very little value is provided because the interface lacks an implementation.  If this interface is used within any code it will not execute because no concrete objects that implement this interface exist to perform the work.  This same concept applies to JPA just on a larger scale since the API specification defines many interfaces and annotations.

This is where the role of the JPA provider comes into play.  JPA providers develop a JPA implementation that meets the requirements of the JPA specification.  Hibernate is a JPA Provider, as well as others such as EclipseLink and TopLink.  With a JPA implementation in place Java objects can be now be persisted to a relational database, since there is underlying code to perform the work.

Returning to our interface analogy, if JPA is the interface then Hibernate represents a class that implements the interface.

public class Hibernate implements JPA {

	public void insert(Object obj) {
       //Persistence code
	}

	public void update(Object obj) {
	   //Persistence code
		
	}

	public void delete(Object obj) {
	  //Persistence code
		
	}

	public Object select() {
	    //Persistence code
	}
	
	public Object superSelect(){
		//Persistence Code
	}
}

Notice that in addition to implementing the JPA interface, the Hibernate class contains some methods superfluous to the interface.  Keep this in mind for later.  Given this JPA implementation, we could now write some code that relies upon it to persist some data to a relational database.

public class MyApplication {

	public static JPA jpa = new Hibernate();
	
	public static void main(String[] args) {
		Object object = new Object();
		jpa.insert(object);  //writes to DB
	}
}

The new application works well initially, but after a couple of months its performance degrades.  Let's assume that the Hibernate implementation behind the scenes has several deficiencies causing  the poor performance.  Remember, this is for example purposes and I judging the merit of Hibernate.

Upon encountering this issue, another provider may decide the need for another JPA implementation exists.  This provider creates their own implementation of the JPA specification and publishes the code.

public class ToThoughtJpa implements JPA {
	//Implementation
}

Having used the JPA interface in our application, we can now easily make the switch to the more reliant JPA implementation.

public class MyApplication {

	public static JPA jpa = new ToThoughtJpa();
	
	public static void main(String[] args) {
		Object object = new Object();
		jpa.insert(object);  //writes to DB
	}
}

The concept illustrated in our simple example is the main value JPA provides only on a much larger scale.  If we choose to use JPA, we can eventually switch out our chosen JPA implementation for another implementation as long as they both meet the JPA specification.  In reality, this is not always a seamless transition, since we often utilize features of the implementation that are not support by the specification and each implementation has its own little quirks.  To illustrate this point, consider if we had called the superSelect method within our application.

public class MyApplication {

	public static Hibernate jpa = new Hibernate();
	
	public static void main(String[] args) {
		Object object = new Object();
		jpa.insert(object);  //writes to DB
		jpa.superSelect();
	}
}

Notice that in order to call the method, the interface of type JPA must be replaced with the Hibernate implementation.  At this point, we cannot swap our JPA implementation to the ToThoughtJpa JPA implementation because its interface does not contain the superSelect method.  This example attempts to illustrate the restrictions that occur when a developer chooses to use the straight Hibernate implementation, which is not bound by the JPA specification.

In summary, JPA is not an implementation, it will not provide any functionality within your application.  Its purpose is to provide a set of guidelines that can be followed by JPA providers to create an ORM implementation in a standardized manner.  This allows the underlying JPA implementation to be swapped and for developers to easily transition (think knowledge wise) from one implementation to another.  Hibernate is arguably the most popular JPA provider.  Hibernate's JPA implementation is used by many developers, however some choose to use the actual Hibernate implementation itself because the implementation may contain advanced functionality not contained in the JPA implementation.