This class maintains a member field of type AtomicReference. It implements the following algorithm to create and initialize an object in its get() method:

• First it is checked whether the AtomicReference variable contains already a value. If this is the case, the value is directly returned.
• Otherwise the initialize() method is called. This method must be defined in concrete subclasses to actually create the managed object.
• After the object was created by initialize() it is checked whether the AtomicReference variable is still undefined. This has to be done because in the meantime another thread may have initialized the object. If the reference is still empty, the newly created object is stored in it and returned by this method.
• Otherwise the value stored in the AtomicReference is returned.

Because atomic variables are used this class does not need any synchronization. So there is no danger of deadlock, and access to the managed object is efficient. However, if multiple threads access the AtomicInitializer object before it has been initialized almost at the same time, it can happen that initialize() is called multiple times. The algorithm outlined above guarantees that get() always returns the same object though.

Compared with the LazyInitializer class, this class can be more efficient because it does not need synchronization. The drawback is that the initialize() method can be called multiple times which may be problematic if the creation of the managed object is expensive. As a rule of thumb this initializer implementation is preferable if there are not too many threads involved and the probability that multiple threads access an uninitialized object is small. If there is high parallelism, LazyInitializer is more appropriate.