Section 14,55: denormalized_number

 denormalized number

 A computational floating-point result smaller than the lowest value
 in the normal range of a data type (the smallest representable
 normalized number) that gradually underflows and loses precision
 until it reaches the smallest representable subnormal value (.GT.
 or .EQ.  0) for that data type.  You cannot write a constant for a
 denormalized number.