Source : Free On-Line Dictionary of Computing

Weak Head Normal Form
     
         (WHNF) A {lambda expression} is
        in weak head normal form (WHNF) if it is a {head normal form}
        (HNF) or any {lambda abstraction}.  I.e. the top level is not
        a {redex}.
     
        The term was coined by {Simon Peyton Jones} to make explicit
        the difference between {head normal form} (HNF) and what
        {graph reduction} systems produce in practice.  A lambda
        abstraction with a reducible body, e.g.
     
        	\ x . ((\ y . y+x) 2)
     
        is in WHNF but not HNF.  To reduce this expression to HNF
        would require reduction of the lambda body:
     
        	(\ y . y+x) 2  -->  2+x
     
        Reduction to WHNF avoids the {name capture} problem with its
        need for {alpha conversion} of an inner lambda abstraction and
        so is preferred in practical {graph reduction} systems.
     
        The same principle is often used in {strict} languages such as
        {Scheme} to provide {call-by-name} evaluation by wrapping an
        expression in a lambda abstraction with no arguments:
     
        	D = delay E = \ () . E
     
        The value of the expression is obtained by applying it to the
        empty argument list:
     
        	force D = apply D ()
        		= apply (\ () . E) ()
        		= E
     
        (1994-10-31)