Comparing Methods of First Derivative Approximation

Forward, Backward and Central Divided Difference

Ana Catalina Torres, Autar Kaw

University of South Florida
United States of America
[email protected]

Introduction

This worksheet demonstrates the use of Mathematica to to compare the approximation of first order derivatives using three
different methods. Each method uses a point h ahead, behind or both of the given value of x at which the first derivative of f(x) is
to be found.

f ' HxL @
f Hx+hL- f HxL
Forward Difference Approximation (FDD)
h
Backward Difference Approximation (BDD)

f ' HxL @
f HxL- f Hx-hL
h

f ' HxL @
Central Difference Approximation (CDD)
f Hx+hL- f Hx-hL
2h

Section 1: Input

The following simulation approximates the first derivative of a function using different methods of approximation (FDD,BDD,-
CDD). The user inputs are
a) function, f(x)
b) point at which the derivative is to be found, xv
c) starting step size, h
d) number of times user wants to halve the starting step size, n

The outputs include

a) approximate values of the first derivative at the point and initial step size given using different types of approximation
b) exact value
c) true error, absolute relative true error, approximate error and absolute relative approximate error, number of at least correct
significant digits in the solution as a function of step size.

Function f(x):

f@x_D := x ∗ Exp@2 ∗ xD;

Value of x at which f '(x) is desired, xv

xv = 4.0;
2 nbm_dif_sim_comparedif.nb

Starting step size, h

h = 0.2;

Number of times step size is halved

n = 12.0;

This is the end of the user section. All the information must be entered before proceeding to the next section.

Section 2: Procedure

The following procedure estimates the solution of first derivate of an equation at a point xv using different methods of approxima-
tion.
f (x) = function
xv = value at which the solution is desired
h = step size value
n = number of times step size is halved

Forward Divided Difference Procedure

deriv = Hf@xv + hD − f@xvDL ê HhL;

FDD@f_, xv_, h_D := Module@8deriv<,

derivD

Backward Divided Difference Procedure

deriv = Hf@xvD − f@xv − hDL ê HhL;

BDD@f_, xv_, h_D := Module@8deriv<,

derivD

Central Divided Difference Procedure

deriv = Hf@xv + hD − f@xv − hDL ê H2 ∗ hL;

CDD@f_, xv_, h_D := Module@8deriv<,

derivD

Section 3: Calculation

The exact value Ev of the first derivative of the equation:

First, using the diff command the solution is found. In a second step, the exact value of the derivative is shown.
f@x_D
2 x_
x_

f '@x_D
2 x_
+2 2 x_
x_

Ev = N@f '@xvDD

26 828.6
 nbm_dif_sim_comparedif.nb 3

The next loop calculates the following:

Av: Approximate value of the first derivative using various first derivative approximation methods by calling the procedures
"FDD", "BDD", and "CDD"
Ev: Exact value of the second derivative
et: Absolute relative true percentage error
ea: Absolute relative approximate percentage error
Sig: Least number of correct significant digits in an approximation
Do@
Nn@iD = 2 ^ i;
H@iD = h ê Nn@iD;
AVFDD@iD = FDD@f, xv, H@iDD;
AVBDD@iD = BDD@f, xv, H@iDD;
AVCDD@iD = CDD@f, xv, H@iDD;
etFDD@iD = Abs@HEv − AVFDD@iDL ê EvD ∗ 100.0;
etBDD@iD = Abs@HEv − AVBDD@iDL ê EvD ∗ 100.0;
etCDD@iD = Abs@HEv − AVCDD@iDL ê EvD ∗ 100.0;
If@i > 0,
eaFDD@iD = Abs@HAVFDD@iD − AVFDD@i − 1DL ê AVFDD@iDD ∗ 100.0;
eaBDD@iD = Abs@HAVBDD@iD − AVBDD@i − 1DL ê AVBDD@iDD ∗ 100.0;
eaCDD@iD = Abs@HAVCDD@iD − AVCDD@i − 1DL ê AVCDD@iDD ∗ 100.0;
SigFDD@iD = Floor@H2 − Log@10, eaFDD@iD ê 0.5DLD;
SigBDD@iD = Floor@H2 − Log@10, eaBDD@iD ê 0.5DLD;
SigCDD@iD = Floor@H2 − Log@10, eaCDD@iD ê 0.5DLD;
If@SigFDD@iD < 0, SigFDD@iD = 0D;
If@SigBDD@iD < 0, SigBDD@iD = 0D;
If@SigCDD@iD < 0, SigCDD@iD = 0D;
D
, 8i, 0, n − 1, 1<D

The loop halves the value of the step size n times. Each time, the approximate values of the first derivative are calculated and
saved in different vectors depending on the method of approximation. The approximate error is calculated after at least two
approximate values of the first derivative have been saved. The number of significant digits is calculated and written as the
lowest real number for each method. If the number of significant digits calculated is less than zero, then is shown as zero.

Section 4: Spreadsheet

The next table shows the step size value, exact value, approximate value, and the absolute relative true error as a function of the
step size value for each method of approximation.
Print@" ", "h", " ", "AvFDD", " ", "etFDD",
" ", "AvBDD", " ", "etBDD", " ", "AvCDD", " ", "etCDD"D;

Grid@Table@8H@iD, AVFDD@iD, etFDD@iD, AVBDD@iD, etBDD@iD, AVCDD@iD, etCDD@iD<, 8i, 0, n − 1<DD

Print@" "D
4 nbm_dif_sim_comparedif.nb

h AvFDD etFDD AvBDD etBDD AvCDD etCDD

0.2 33 769.2 25.8702 21 653.4 19.2898 27 711.3 3.29021

0.1 30 040.6 11.9724 24 054.8 10.3389 27 047.7 0.816743
0.05 28 375.3 5.76494 25 391.3 5.35729 26 883.3 0.203824
0.025 27 587.7 2.82941 26 096.9 2.72754 26 842.3 0.0509334
0.0125 27 204.7 1.40171 26 459.4 1.37624 26 832. 0.012732
0.00625 27 015.8 0.697638 26 643.2 0.691272 26 829.5 0.0031829
0.003125 26 922. 0.348019 26 735.7 0.346428 26 828.8 0.000795719
0.0015625 26 875.3 0.17381 26 782.1 0.173412 26 828.7 0.00019893
0.00078125 26 851.9 0.0868553 26 805.3 0.0867558 26 828.6 0.0000497324
0.000390625 26 840.3 0.0434152 26 817. 0.0433903 26 828.6 0.0000124331
0.000195313 26 834.4 0.0217045 26 822.8 0.0216983 26 828.6 3.10818 × 10−6
0.0000976563 26 831.5 0.0108515 26 825.7 0.0108499 26 828.6 7.77234 × 10−7

Section 5: Graphs

The following graphs show the approximate solution, absolute relative true error and absolute relative approximate error as a
function of step size.

data = Table@8H@iD, AVFDD@iD<, 8i, 0, n − 1<D;

Needs@"PlotLegends`"D

data2 = Table@8H@iD, AVBDD@iD<, 8i, 0, n − 1<D;

data3 = Table@8H@iD, AVCDD@iD<, 8i, 0, n − 1<D;
plot1 = ListPlot@8data<,
PlotJoined → True,
PlotStyle → 8Green<,
DisplayFunction → Identity,
PlotRange → Full D;
plot2 = ListPlot@8data2<,
PlotJoined → True,
PlotStyle → 88Blue, Thick<<,
DisplayFunction → Identity,
PlotRange → Full D;
plot3 = ListPlot@8data3<,
PlotJoined → True,
PlotStyle → 88Red, Dashed<<,
DisplayFunction → Identity,
PlotRange → Full D;
alltogetherplot = ShowAplot1, plot3, plot2, PlotRange → Automatic,
PlotLabel → "Approximate Solution of the First Derivative of a Function

AxesLabel → 9"Step Size", "Approximate Value"=E;

using\ndifferent Methods of Approximation as a Function of Step Size",

ShowLegend@alltogetherplot, 888Graphics@8Green, Line@880, 0<, 81, 0<<D<D, "FDD"<,

8Graphics@88Blue, Line@880, 0<, 81, 0<<D<<D, "BDD"<,
8Graphics@8Red, Line@880, 0<, 81, 0<<D<D, "CDD"<<,
LegendPosition → 8.5, −.6<, LegendSize → 80.3, 0.4<<D

data = Table@8H@iD, etFDD@iD<, 8i, 0, n − 1<D;

data2 = Table@8H@iD, etBDD@iD<, 8i, 0, n − 1<D;
 nbm_dif_sim_comparedif.nb 5

data3 = Table@8H@iD, etCDD@iD<, 8i, 0, n − 1<D;

plot1 = ListPlot@8data<,
PlotJoined → True,
PlotStyle → 8Green<,
DisplayFunction → Identity,
PlotRange → Full D;
plot2 = ListPlot@8data2<,
PlotJoined → True,
PlotStyle → 88Blue, Thick<<,
DisplayFunction → Identity,
PlotRange → Full D;
plot3 = ListPlot@8data3<,
PlotJoined → True,
PlotStyle → 88Red, Dashed<<,
DisplayFunction → Identity,
PlotRange → Full D;
alltogetherplot = ShowAplot1, plot3, plot2, PlotRange → Automatic,
PlotLabel → "Absolute Relative True Percentage\nError as a Function of Step Size",
AxesLabel → 9"Step Size", "Absolute Relative
True Error"=E;
ShowLegend@alltogetherplot, 888Graphics@8Green, Line@880, 0<, 81, 0<<D<D, "FDD"<,
8Graphics@88Blue, Line@880, 0<, 81, 0<<D<<D, "BDD"<,
8Graphics@8Red, Line@880, 0<, 81, 0<<D<D, "CDD"<<,
LegendPosition → 8.5, .1<, LegendSize → 80.3, 0.4<<D

data = Table@8H@iD, eaFDD@iD<, 8i, 0, n − 1<D;

data2 = Table@8H@iD, eaBDD@iD<, 8i, 0, n − 1<D;
data3 = Table@8H@iD, eaCDD@iD<, 8i, 0, n − 1<D;
plot1 = ListPlot@8data<,
PlotJoined → True,
PlotStyle → 8Green<,
DisplayFunction → Identity,
PlotRange → Full D;
plot2 = ListPlot@8data2<,
PlotJoined → True,
PlotStyle → 88Blue, Thick<<,
DisplayFunction → Identity,
PlotRange → Full D;
plot3 = ListPlot@8data3<,
PlotJoined → True,
PlotStyle → 88Red, Dashed<<,
DisplayFunction → Identity,
PlotRange → Full D;
alltogetherplot = Show@plot1, plot3, plot2, PlotRange → Automatic,
PlotLabel → "Absolute Relative Approximate Percentage\n Error as a Function of Step Size",
AxesLabel → 8"Step Size", "Absolute Relative

ShowLegend@alltogetherplot, 888Graphics@8Green, Line@880, 0<, 81, 0<<D<D, "FDD"<,

Approximate Error"<D;

8Graphics@88Blue, Line@880, 0<, 81, 0<<D<<D, "BDD"<,

8Graphics@8Red, Line@880, 0<, 81, 0<<D<D, "CDD"<<,
LegendPosition → 8.5, .1<, LegendSize → 80.3, 0.4<<D
6 nbm_dif_sim_comparedif.nb

Approximate Solution of the First Derivative of a Function using

different Methods of Approximation as a Function of Step Size
Approximate Value

27 400

27 200

Step Size
0.02 0.04 0.06 0.08 0.10 0.12

26 800
FDD
26 600
BDD
26 400
CDD

Absolute Relative True Percentage

Error as a Function of Step Size
Absolute Relative FDD
True Error
BDD
8
CDD

Step Size
0.02 0.04 0.06 0.08 0.10 0.12

Absolute Relative Approximate Percentage

Error as a Function of Step Size
Absolute Relative FDD
Approximate Error
BDD

6 CDD

Step Size
0.02 0.04 0.06 0.08 0.10

References

Numerical Differentiation of Continuous Functions.

See http://numericalmethods.eng.usf.edu/mws/gen/02def/
 nbm_dif_sim_comparedif.nb 7

Questions

1. The velocity of a rocket is given by

vHtL = 2000 ln 140 000

140 000- 2100 t
- 9.8 t
Use three different methods with a step size of 0.25 to find the acceleration at t=5s. Compare with the exact answer and study the
effect of the step size.

2. Look at the true error vs. step size data for problem # 1. Do you see a relationship between the value of the true error and step
size ? Is this concidential? Is it similar for Forward and Backward Divided Difference? Is it different for Central Divided Differ-
ence method?

Conclusions

The worksheet shows the nature of accuracy of the three different methods of finding the first derivative of a continuous function.
Forward and Backward Divided Difference methods exhibit similar accuraciees as they are first order accurate, while central
divided difference shows more accuracy as it is second order accurate.

Legal Notice: The copyright for this application is owned by the author(s). Neither Maplesoft nor the author are responsible for
any errors contained within and are not liable for any damages resulting from the use of this material. This application is
intended for non-commercial, non-profit use only. Contact the author for permission if you wish to use this application for-profit
activities.