LOSS OF SIGNIFICANCE IN NUMERICAL COMPUTING AND POLYNOMIAL FORM

Abstract: One of the most common and often avoidable peculiarity of using normalized floating point representation of numbers in arithmetic operations is Loss of Significance. Use of normalized floating point representation of numbers in arithmetic operations lead to some unexpected and unfortunate consequences. There is a loss of significance due to subtraction of a number from nearly equal number. In this paper, the loss of significance in various numerical computations including its effect in polynomial form is discussed. Further, the alternatives to minimize its effect on final result are also discussed.  

Keywords: normalized floating point, arithmetic operations, loss of significance, polynomial form.