types of errors part 5
TRANSCRIPT
-
7/28/2019 types of errors part 5
1/21
6/12/2013 http://numericalmethods.eng.usf.edu 1
Floating Point Representation
Major: All Engineering Majors
Authors: Autar Kaw, Matthew Emmons
http://numericalmethods.eng.usf.edu Transforming Numerical Methods Education for STEM
Undergraduates
http://numericalmethods.eng.usf.edu/http://numericalmethods.eng.usf.edu/ -
7/28/2019 types of errors part 5
2/21
Floating Point Representation
http://numericalmethods.eng.usf.edu
http://numericalmethods.eng.usf.edu/http://numericalmethods.eng.usf.edu/ -
7/28/2019 types of errors part 5
3/21
http://numericalmethods.eng.usf.edu3
Floating Decimal Point : Scientific Form
2
3
2
10.56782aswrittenis78.256
10678.3aswrittenis003678.0105678.2aswrittenis78.256
-
7/28/2019 types of errors part 5
4/21
http://numericalmethods.eng.usf.edu4
Example
exponent10mantissasignThe form is
or
Example: For
em 10
2105678.2
2
5678.2
1
e
m
-
7/28/2019 types of errors part 5
5/21
http://numericalmethods.eng.usf.edu5
Floating Point Format for BinaryNumbers
mm
m y e
22 101mantissave-for 1ve,for 0number of sign
2
1 is not stored as it is always given to be 1.
exponentinteger e
-
7/28/2019 types of errors part 5
6/21
http://numericalmethods.eng.usf.edu6
Example
22
52210
1011011.1
21011011.111.11011075.54
9 bit-hypothetical wordthe first bit is used for the sign of the number,the second bit for the sign of the exponent,the next four bits for the mantissa, and
the next three bits for the exponent
We have the representation as
0 0 1 0 1 1 1 0 1
Sign of thenumber
mantissaSign of theexponent
exponent
-
7/28/2019 types of errors part 5
7/21
http://numericalmethods.eng.usf.edu7
Machine EpsilonDefined as the measure of accuracy and foundby difference between 1 and the next number
that can be represented
-
7/28/2019 types of errors part 5
8/21
-
7/28/2019 types of errors part 5
9/21
http://numericalmethods.eng.usf.edu9
Relative Error and MachineEpsilon
The absolute relative true error in representinga number will be less then the machine epsilon
Example
201102
5210
21100.1
21100.102832.0
10 bit word (sign, sign of exponent, 4 for exponent, 4 for mantissa)
0 1 0 1 1 0 1 1 0 0
0625.02034472.0
02832.00274375.002832.0
0274375.021100.1
4
01102
2
a
Sign of thenumber mantissaSign of theexponent
exponent
-
7/28/2019 types of errors part 5
10/21
IEEE 754 Standards for SinglePrecision Representation
http://numericalmethods.eng.usf.edu
-
7/28/2019 types of errors part 5
11/21
IEEE-754 Floating PointStandard
Standardizes representation of floating point numbers on
different computers in single anddouble precision.
Standardizes representation of floating point operations ondifferent computers.
-
7/28/2019 types of errors part 5
12/21
One Great Reference
What every computer scientist (and even if you are not) should know about floating point
arithmetic!
http://www.validlab.com/goldberg/paper.pdf
http://www.validlab.com/goldberg/paper.pdfhttp://www.validlab.com/goldberg/paper.pdf -
7/28/2019 types of errors part 5
13/2113
IEEE-754 Format Single
Precision32 bits for single precision
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
Sign(s)
BiasedExponent (e)
Mantissa (m)
127'2 21)1(Value . e s m
-
7/28/2019 types of errors part 5
14/2114
Example#1
127'2 2.11Value e s m
127)10100010(21 2210100000.11
127162
2625.11 1035 105834.52625.11
1 1 0 1 0 0 0 1 0 1 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
Sign(s)
BiasedExponent (e)
Mantissa (m)
-
7/28/2019 types of errors part 5
15/2115
Example#2
? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?
Sign(s)
BiasedExponent (e)
Mantissa (m)
Represent -5.5834x10 10 as a singleprecision floating point number.
?110
2?.11105834.5
-
7/28/2019 types of errors part 5
16/2116
Exponent for 32 Bit IEEE-754
8 bits would represent2550 e
Bias is 127; so subtract 127 fromrepresentation
128127 e
-
7/28/2019 types of errors part 5
17/21
Exponent for Special Casese Actual range of
2541 e
0e and 255e are reserved for special numbers
Actual range of
127126 e
e
-
7/28/2019 types of errors part 5
18/21
Special Exponents and Numbers
0e all zeros255e all ones
s m Represents0 all zeros all zeros 01 all zeros all zeros -00 all ones all zeros
1 all ones all zeros0 or 1 all ones non-zero NaN
e
-
7/28/2019 types of errors part 5
19/2119
IEEE-754 Format
The largest number by magnitude
The smallest number by magnitude
Machine epsilon
381272 1040.321........1.1
381262 1018.220......00.1
723 1019.12mach
-
7/28/2019 types of errors part 5
20/21
Additional ResourcesFor all resources on this topic such as digital audiovisuallectures, primers, textbook chapters, multiple-choice tests,worksheets in MATLAB, MATHEMATICA, MathCad andMAPLE, blogs, related physical problems, please visit
http://numericalmethods.eng.usf.edu/topics/floatingpoint_representation.html
http://numericalmethods.eng.usf.edu/topics/floatingpoint_representation.htmlhttp://numericalmethods.eng.usf.edu/topics/floatingpoint_representation.htmlhttp://numericalmethods.eng.usf.edu/topics/floatingpoint_representation.htmlhttp://numericalmethods.eng.usf.edu/topics/floatingpoint_representation.htmlhttp://numericalmethods.eng.usf.edu/topics/floatingpoint_representation.html -
7/28/2019 types of errors part 5
21/21
THE END
http://numericalmethods.eng.usf.edu
http://numericalmethods.eng.usf.edu/http://numericalmethods.eng.usf.edu/