G. Z.

Garber

Foundations of Excel VBA Programming and Numerical Methods. —

Moscow: PRINTKOM, 2013. — 528 p.

ISBN 978-5-91146-894-1

Intended for university students studying computer science,

applied mathematics and information technology, as well as for post-
graduate students, scientific workers and other readers wishing to
refine their skill in solving problems by using tabular processor
Microsoft Office Excel 2007 – 2013.
Elements of an environment for developing programs (macros)
and base constructs of programming languages Visual Basic and VBA
are considered. Modern and classical numerical methods and their
program realization are also considered. As examples, Excel macros
are developed for solving mathematical, physical, engineering and
economic problems. A compact disk containing program modules and
other text information is enclosed. No preliminary programming expe-
rience is required for grasping the material.
This book is based on the author’s two previous books approved
by the Scientific and methodical council in computer science at the
Ministry of Education and Science of the Russian Federation as
a manual on discipline “Computer science” for university students.

ISBN 978-5-91146-894-1 © G. Z. Garber, 2013

 Many things are incomprehensible to us
not because our comprehension is weak,
but because those things are not within
the frames of our comprehension.
Kozma Prutkov, 1854

About the author

Gennadiy Z. Garber graduated from the faculty of applied mathematics at the
Moscow Institute of Electronic Machinery in 1972. Professor in Computer
Science, Doctor of Science in Microelectronics, principal scientific worker of
Pulsar R&D Manufacturing Company, Moscow, participant of the International
Conference on “Computer as a Tool”, IEEE EUROCON 2005 and 2007.
Area of interests: mathematical modeling of semiconductor devices and inte-
grated circuits for radio-, micro- and nanoelectronics; development of teaching
techniques in applied mathematics and programming for Excel.

About the prototype books published in Russian

Sample programs and descriptions of their usage allow to learn programming
from scratch.
Using concrete examples with a minimum portion of theoretical introduction,
the author has managed to show the first (and, I must say, main) steps of the
object-oriented programming technology in VBA.
The author’s approach is very successful, in which a theoretical material on
each numerical method is accompanied by a description of the program realiza-
tion, by a scenario of the computing experiments for applied problems and by an
analysis of the calculation results.
The student can obtain not only mathematical knowledge from the manual,
but also acquire practical skills, which are very important in a training course on
numerical methods.
From the reviews of the expert of the Scientific and
methodical council in computer science at the
Ministry of Education and Science of the Russian
Federation on author’s books [1, 2]
Contents

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Chapter 1. Programming in Visual Basic . . . . . . . . . . . . . . . . . . . 12
1.1. Elements of Visual Basic Environment. . . . . . . . . . . . . . . . 13
1.2. Main commands of the program debugger. . . . . . . . . . . . . . 19
1.3. Variables. Data types . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.4. Two main functions for conversion of data types . . . . . . . . . . 27
1.5. Constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.6. Obtaining information . . . . . . . . . . . . . . . . . . . . . . . . 32
1.7. Assignment operator . . . . . . . . . . . . . . . . . . . . . . . . . 36
1.8. Arithmetic expression. . . . . . . . . . . . . . . . . . . . . . . . . 38
1.9. Mathematical functions. Functions of date and time . . . . . . . . . 45
1.10. Logical expression . . . . . . . . . . . . . . . . . . . . . . . . . . 48
1.11. GoTo operator. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
1.12. Decision-making constructs. . . . . . . . . . . . . . . . . . . . . . 53
1.13. Cycles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
1.14. Manifestation of the error of real numbers’ computer representation 65
1.15. Arrays. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
1.16. User-defined procedures . . . . . . . . . . . . . . . . . . . . . . . 77
1.17. Built-in procedures. Usage of standard windows. . . . . . . . . . . 85
1.18. Records . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
1.19. Work with strings. . . . . . . . . . . . . . . . . . . . . . . . . . . 94
1.20. Work with text files. . . . . . . . . . . . . . . . . . . . . . . . . . 101
1.21. Matrix terminology. Formulation of demonstration tasks . . . . . . 109
1.22. Program for transposing a matrix relative to its auxiliary diagonal . 111
1.23. User-defined forms . . . . . . . . . . . . . . . . . . . . . . . . . . 116
1.24. Digression. Developing programs with the form in Microsoft
Visual Studio. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
Chapter 2. Programming in VBA . . . . . . . . . . . . . . . . . . . . . . . 133
2.1. Loading the form from the Excel window. Running the program
executable file. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

4
 Contents

2.2. Layout of the control elements on the Excel worksheet . . . . . . . 136

2.3. User-defined functions of Excel . . . . . . . . . . . . . . . . . . . 139
2.4. Two methods for developing Excel macros . . . . . . . . . . . . . 144
2.5. Excel Macro Recorder . . . . . . . . . . . . . . . . . . . . . . . . 145
2.6. VBA code generated by Excel Macro Recorder and its editing . . . 148
2.7. Objects and events . . . . . . . . . . . . . . . . . . . . . . . . . . 151
2.8. Object Application . . . . . . . . . . . . . . . . . . . . . . . . . . 154
2.9. Objects Workbook, Workbooks and ActiveWorkbook. . . . . . . . 161
2.10. Objects Worksheet, Worksheets and ActiveSheet . . . . . . . . . . 167
2.11. Objects Range, Selection and ActiveCell. . . . . . . . . . . . . . . 171
2.12. Study of objects. . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
2.13. Using the Excel table as the user interface of programs . . . . . . . 180
2.14. Two more Excel macros. Personal Macro Workbook . . . . . . . . 182
2.15. One more user-defined function of Excel. . . . . . . . . . . . . . . 188
2.16. Digression. Change of Excel options . . . . . . . . . . . . . . . . . 193
Chapter 3. Finite Difference Method for Solving Differential Equations. . . 195
3.1. Finite difference analogs of derivatives for a uniform grid. . . . . . 196
3.2. Finite difference scheme for the linear differential equation.
The decomposition method . . . . . . . . . . . . . . . . . . . . . . 199
3.3. Sufficient stability conditions for the decomposition method . . . . 204
3.4. Simplification of the second-order linear differential equation. . . . 208
3.5. Program realization of the decomposition method . . . . . . . . . . 210
3.6. Examples of using the decomposition method . . . . . . . . . . . . 212
3.7. Examples of the computing error. Instability and loss of accuracy. . 217
3.8. Solving the system of linear algebraic equations by using Excel
functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
3.9. Solving the system of linear algebraic equations by the Gaussian
elimination method . . . . . . . . . . . . . . . . . . . . . . . . . . 225
3.10. Two subroutines for solving the system of linear algebraic
equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
3.11. Reduction of the computing error. . . . . . . . . . . . . . . . . . . 235
3.12. Solving the nonlinear differential equation by the quasilinearization
method. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
3.13. Solving the Shockley-Poisson equation. . . . . . . . . . . . . . . . 241
3.14. Finite difference analogs of derivatives for a nonuniform grid. . . . 249
3.15. The decomposition method for a nonuniform grid . . . . . . . . . . 252
3.16. Solving the Shockley-Poisson equation on a nonuniform grid. . . . 255
3.17. Use of solution symmetry. . . . . . . . . . . . . . . . . . . . . . . 261

5
 Contents

3.18. The cyclic decomposition method . . . . . . . . . . . . . . . . . . 267

3.19. Program realization of the cyclic decomposition method . . . . . . 271
3.20. Solving the oscillation equation. . . . . . . . . . . . . . . . . . . . 273
Chapter 4. Cubic Spline . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
4.1. Definition of cubic spline. Spline moments . . . . . . . . . . . . . 282
4.2. Spline interpolation . . . . . . . . . . . . . . . . . . . . . . . . . . 288
4.3. Use of cubic spline for processing transistor electrical
characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
4.4. Spline integration . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
4.5. Iterative methods for solving the nonlinear algebraic equation. . . . 303
4.6. Noniterative method for solving the nonlinear algebraic equation. . 314
4.7. Calculating the charge storage capacity . . . . . . . . . . . . . . . 321
4.8. Subroutine for automatic creation of graphs . . . . . . . . . . . . . 327
4.9. Cubic spline usage for solving the second-order linear differential
equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
4.10. Program realization of the cubic spline method for solving
the linear differential equation . . . . . . . . . . . . . . . . . . . . 334
4.11. Solving the linear differential equation by the cubic spline method . 337
4.12. Modeling of heating of a geophysical cable. Locally
one-dimensional scheme . . . . . . . . . . . . . . . . . . . . . . . 340
Chapter 5. Quadratic and Linear Splines . . . . . . . . . . . . . . . . . . . 352
5.1. Definition of quadratic spline. Spline slopes . . . . . . . . . . . . . 353
5.2. Method for solving the initial value problem for the system
of differential equations. . . . . . . . . . . . . . . . . . . . . . . . 356
5.3. Program for solving the initial value problem . . . . . . . . . . . . 359
5.4. Solving the system of nonlinear algebraic equations by the Newton
method. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362
5.5. Newton and Newton-like methods for solving the single nonlinear
algebraic equation . . . . . . . . . . . . . . . . . . . . . . . . . . 365
5.6. Modeling of the piano mechanism linking a key with hammer. . . 372
5.7. Definition of linear spline. . . . . . . . . . . . . . . . . . . . . . . 383
5.8. The least-squares method. . . . . . . . . . . . . . . . . . . . . . . 386
5.9. Program to determine the dependence of the wheat productivity
on the land quality. . . . . . . . . . . . . . . . . . . . . . . . . . . 390
5.10. The forward and backward Fourier transforms of a periodic
function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395
5.11. Subroutines for the forward and backward discrete Fourier
transforms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 400

6
 Contents

5.12. Solving the sound insulation problem . . . . . . . . . . . . . . . . 406

Chapter 6. Numerical Methods for Nonlinear Programming . . . . . . . . . 415
6.1. Minimizing linear and nonlinear functions of several variables
by the Solver add-in. . . . . . . . . . . . . . . . . . . . . . . . . . 417
6.2. Method for minimizing a nonlinear function of one variable. . . . . 428
6.3. The coordinate-descent method. . . . . . . . . . . . . . . . . . . . 432
6.4. Examples of using the minimization methods . . . . . . . . . . . . 437
6.5. The Powell minimization method. . . . . . . . . . . . . . . . . . . 446
6.6. Determining the equilibrium state of a four-spring system. . . . . . 456
6.7. Minimization with nonlinear constraints . . . . . . . . . . . . . . . 462
6.8. Minimization of the multimodal function. . . . . . . . . . . . . . . 475
6.9. Minimization of the tabular function . . . . . . . . . . . . . . . . . 483
6.10. Solving the nonlinear differential equation by the shooting method . 490
6.11. Modeling of the hammer motion in the piano mechanism . . . . . . 493
6.12. Nonlinear programming and the least-squares method. . . . . . . . 501
Instead of Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 510
Appendix 1. Data Types of Visual Basic and VBA . . . . . . . . . . . . . 513
Appendix 2. Greek and Russian Alphabets Denoted by Latin Letters. . . . 515
Appendix 3. The Main Mathematical Functions . . . . . . . . . . . . . . . 517
Appendix 4. Material for Tasks . . . . . . . . . . . . . . . . . . . . . . . . 518
Appendix 5. Analytical Method for Solving the Cubic Algebraic Equation . 520
Appendix 6. Realization of the Tangent Method by Using the Excel
Circular Reference. . . . . . . . . . . . . . . . . . . . . . . . 521
References List . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 523
Subject Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 525

7
Introduction

Because of many advantages (above all, availability), tabular processor

Excel, which is a part of Microsoft Office, is used in various areas of human
activity: in economics and finances, electrical engineering and electronics,
medicine, building construction, etc. This book is about Excel usage in applied
mathematics.
While writing this book, the author pursued the following goals:
to teach the reader to program in the modern programming language,
Visual Basic (VB), and its extension, Visual Basic for Applications (VBA);
on the basis of these programming languages, to give the reader full
enough representation about numerical methods aimed at obtaining a solution of
a task in the form of numbers (instead of formulas that are a result of using
analytical methods);
to show that Excel with programs (macros), written by the reader in VBA,
is convenient for solving applied tasks by numerical methods.
The book is intended for the reader familiar with Excel, Windows Explorer,
Windows Clipboard and text editor Notepad for Windows. Besides, the reader
should be conversant with higher mathematics and general physics. No prelimi-
nary programming experience is necessary.
Learning this book is possible only by using a computer equipped with tabu-
lar processor Excel.
According to the author’s opinion, there is no difference between terms
“program” and “macro” when these terms concern the programming for Excel.
Therefore, words “program” and “macro” are synonyms in this book.
The book contains six chapters, six appendices, a list of references and a sub-
ject index.
In the first two chapters, we consider elements of Visual Basic Environment
and main facilities of programming languages VB and VBA. The standard win-
dow of operating system Windows, text file, user-defined form and Excel table
are considered as the program user interface — the facility of dialogue between
the user and program. We consider the creation of Excel user-defined functions.
Besides, we demonstrate how to work with the program debugger, reference
systems, Excel Macro Recorder and Personal Macro Workbook.

8
 Introduction

In the third chapter, we consider the finite difference method for solving the
second-order linear differential equation with two kinds of conditions on the
solution, namely, the boundary and periodicity conditions. This is followed by
a review of two versions of the decomposition method for solving systems of
linear algebraic equations of special form called finite difference schemes. The
simplest scheme is also solved by the Gaussian elimination method. The ques-
tion of stability of the decomposition and Gaussian methods is investigated in
respect of not increasing the computing error during solving the scheme. Using
the Shockley-Poisson equation as an example, we consider the quasilinearization
method for solving the nonlinear differential equation with boundary conditions.
To demonstrate the possibilities of the finite difference method, we develop sub-
routines and programs for solving mathematical and applied problems. We use
the Excel scatter diagrams for visualization of calculation results.
Chapter 4 is devoted to the use of the third-degree (cubic) spline:
for interpolation, differentiation and integration of tabular (grid) func-
tions;
for solving the nonlinear algebraic and linear differential equations.
Besides, we consider:
two classical methods for solving the nonlinear algebraic equations,
namely, the bisection and secant methods;
the locally one-dimensional scheme for solving the heat equation with
two spatial coordinates.
We solve a series of applied problems to demonstrate the possibilities of the
cubic spline construction. In addition to the macros and user-defined procedures
(subroutines and function) realizing the numerical methods, a subroutine for
automatic creation of graphs is developed.
In Chapter 5, we review the use of the second-degree (quadratic) spline for
solving the initial value problem (of Cauchy) for the system of differential equa-
tions. The first-degree (linear) spline is used in the least-squares method intended
for determining parameters of a function. Besides, we review the following
methods:
the Newton method for solving the system of nonlinear algebraic equa-
tions;
the tangent, secant and Steffensen methods (called Newton-like methods)
for solving a single nonlinear algebraic equation;
methods for the forward and backward discrete Fourier transforms of
a periodic function.
Based on this theoretical material, we develop procedures and programs for
solving applied problems.
Chapter 6 is mainly devoted to nonlinear programming, more precisely, to
the question of finding the minimum of a nonlinear function of one or several

9
 Introduction

variables without calculating the function derivative or partial derivatives. This

chapter begins with the use of the Solver add-in for Excel to minimize concrete
linear and nonlinear functions of several variables. Further, we develop subrou-
tines for finding the local minimum of a nonlinear function of general form,
which are based on the coordinate-descent and Powell methods.
We review the following applications of the developed minimization subrou-
tines:
for optimizing the dimensions of a one-liter tin can;
for determining the equilibrium state of a four-spring mechanical system;
for minimizing a nonlinear function with nonlinear constraints and a tabu-
lar function of two variables;
for determining the local minima of a multimodal function of two varia-
bles (with several local minima);
in the shooting method intended for solving the nonlinear differential
equation with boundary conditions;
in the least-squares method.
Appendix 1 presents the data types of Visual Basic and VBA.
Appendix 2 contains the Greek alphabet with English names of the letters and
the Russian alphabet denoted by Latin letters. The inclusion of this appendix is
justified by the possible lack of Greek and Russian letters on the computer key-
board. English names of Greek letters are used in texts of program modules and
source data for programs. Russian letters in Latin are mainly used in the refer-
ences list.
Appendix 3 contains the main mathematical functions of Visual Basic. In ad-
dition, this appendix contains operators allowing the use of mathematical func-
tions not included in the programming language.
Appendix 4 contains data for tasks intended to consolidate the book material
and check up its understanding.
Appendix 5 presents an analytical method for solving the cubic algebraic
equation. We use this method in Chapter 4.
Appendix 6 demonstrates the use of circular reference in Excel for solving
the nonlinear algebraic equation by the tangent method.
The subject index contains the main terms and designations with numbers
of pages, on which their sense is uncovered. It will allow using the book as
a reference manual.
The present book is based on author’s books [1, 2] approved by the Scientific
and methodical council in computer science at the Ministry of Education and
Science of the Russian Federation as a manual on discipline “Computer science”
for university students.
In the book, we often speak about mathematical (computer, numerical)
modeling. The essence of mathematical modeling lies in the replacement of

10
 Introduction

an object, in particular of a process, by an appropriate mathematical model and

in its further study by using a computer. Operation with the model, instead of the
object, allows to obtain operatively detailed information, showing internal con-
nections of the object and its qualitative and quantitative characteristics. The
mathematical modeling is so popular that, when speaking about it, adjective
“mathematical” may be omitted, as in this book.
In writing the book, we used a personal computer equipped with 32-bit
version of the Windows 7 operating system and Microsoft Office Professional
Plus 2013 Preview. For obtaining information, shown in Fig. 1.8, 2.23 and 3.6,
the reference system of Excel 2010 was used. The system disk name is C and the
computer user name is usr in the book.
We will need the Developer tab in Excel Ribbon (a part of the Excel win-
dow), among tabs Home, Insert, Page Layout, etc. If such a tab does not exist,
we fulfill the following:
1) click on the File button in the top left corner of the Excel window
(in Excel 2007, click on the Office button);
2) click on the Options button;
3) in the Excel Options window opened, click on button Customize Ribbon;
4) in area Customize the Ribbon:
set Main Tabs by using the drop-down list;
then turn on option Developer;
5) click on the OK button.
This operational sequence can be written as the following formula: File >
Options > Customize Ribbon > Main Tabs > turn on Developer > OK. We will
frequently use such formulas.
When opening an Excel workbook containing a macro, the Security Warning
panel can appear. To allow the macro to work, we must click on the Enable
Content button of this panel.
When executing a macro, cycling is possible. To interrupt it, we must press
the Esc key on the computer keyboard.
The enclosed compact disk (CD) contains text files with program modules
and with source data for programs. The texts on the CD correspond to the num-
bered listings in the book. A method of work with these files is described on
pp. 26 and 245.
The program texts on the CD may be used as templates when developing
programs for solving other tasks with the same mathematical formulation as the
tasks considered in the book.
For contact with the author, the following internet resources can be used:
[email protected], http://gzgarber.narod.ru/.

11
 Chapter 1.
Programming in Visual Basic

We review elements of Visual Basic Environment, a part of Microsoft Office,

and constructs of the Visual Basic programming language. The standard window
of operating system Windows, text file and form are used as the user interface
of programs.
In addition, we demonstrate how to work with the program debugger and
reference systems.

12
 1.1. Elements of Visual Basic Environment

1.1. Elements of Visual Basic Environment

For writing and debugging programs, we will use Visual Basic Environment,
which is a part of Microsoft Office.
Program debugging involves detection and correction of errors that, as a rule,
are present in a program text just written.
To go to Visual Basic Environment, we must fulfill the following two opera-
tions:
1) in the Excel window (with the active workbook by name Book1), activate
the Developer tab by clicking on it;
2) click on the Visual Basic button in area Code.
As a result, the Visual Basic Environment window is displayed (Fig. 1.1).
In this window, we can perform various actions: entering and editing the pro-
gram text, as well as debugging and executing the program. Further, we will use
a shorter name for this window and call it “the VB window”.
The program is also called an application or project. It will be in the Excel
workbook (by name Book1).
Let us consider the elements of the VB window.
1. Menu bar. There are standard menus, like in many windows of the
operating system: File, Edit, View, Tools and Help. The Insert menu is used
for organizing a place for program storage (in the workbook). Menus Debug
and Run are respectively used for debugging and running the program.
2. Context menu. It serves for convenience of work in the area (of the VB
window), in which the mouse pointer is located.
For using the context menu:
1) place the mouse pointer in the necessary area of the screen and make the
right click;
2) click (by the left mouse button) on the required command of the displayed
menu.
3. Toolbars: Standard, Edit, Debug and others. Only the standard toolbar is
displayed by default. To add or remove any toolbar, we have to fulfill View >
Toolbars and to click on the required command of the displayed menu. The
check (tick) mark against the command testifies to the presence of the corre-
sponding toolbar on the display screen.

13
 Chapter 1. Programming in Visual Basic

Fig. 1.1. The Visual Basic Environment window including the standard
toolbar, the project explorer window and the properties window

Let us consider the toolbars.

1. Toolbar Standard is displayed by default. It allows us to perform a wide
spectrum of actions.
This toolbar is usually located under the menu bar, however, we can move it
to other areas of the VB window by using the mouse.
2. Toolbar Edit is intended for work with the program text. It realizes possi-
bilities of an elementary text editor:
copying and moving a text fragment to Windows Clipboard;
inserting the text fragment from Windows Clipboard;
search and replacement of words and phrases, etc.
3. Toolbar Debug is intended for debugging the program. Many provisions
are made for debugging:
observation of the current values of the program variables;
step-by-step program execution, in which one operator (statement, in-
struction) or its part is performed on each step, etc.

14
 1.1. Elements of Visual Basic Environment

As a rule, we will review programs without a user-defined form as the

program user interface. Development of such program begins with inserting
a module into the active Excel workbook. For inserting a module, let us fulfill
the following sequence of operations.
1. In the Excel window, Developer > Visual Basic in area Code. As a result,
the VB window appears, including the project explorer window and the proper-
ties window (Fig. 1.1).
If the project explorer window is not displayed, we have to click on Project
Explorer in the View menu. We will need the properties window only in
Section 1.23.
2. Select line VBAProject (Book1) by clicking on it in the project explorer
window.
3. Insert > Module.
As a result, a line corresponding to the inserted module, Module1, appears in
the project explorer window. Besides, an empty window opens; it is the code
window corresponding to Module1 (Fig. 1.2). In this window, we will create the
program text by using the computer keyboard.

Fig. 1.2. The VB window, including the code window,

after inserting Module1 into the Excel workbook

15
 Chapter 1. Programming in Visual Basic

For opening the code window corresponding to the module inserted earlier,
we have to click twice on the name of this module in the project explorer
window.
To delete a module:
1) make the right click on the module name (for example Module1) in the
project explorer window;
2) in the context menu opened, click on the Remove command;
3) click on the No button in the open window with a question about export-
ing the module before removing it.
Before the computer will execute a program, we (as the program developer)
must form its text in the code window. The first and last lines (operators) of the
program are standard:

Sub name()

End Sub

On p. 79, we will consider the origin of word Sub. Word name means the
program name appointed by us.
The name must satisfy the following conditions:
the first character should be a letter;
the name must include only letters, figures and the underscore character;
the name must include less than 256 characters.
As we see, the name cannot include the space character. To use name
consisting of several words, we have:
to begin each word with a capital (uppercase) letter;
to use the underscore character instead of the space character.
Examples of the program name follow:

MyProgram13
my_program
MyProgram_13

Between the first and last lines of the program, we have to place other lines
(operators) of this program. For that, it is possible to use Windows Clipboard and
habitual commands of editing (as in Notepad). After typing a new line, we have
to press the Enter key on the keyboard.
Let us start with a simple program based on the Pythagoras theorem,

c a 2 b2 , (1.1)
for calculating length c of the hypotenuse of a right-angled triangle with legs
a = 3 and b = 4.

16
 1.1. Elements of Visual Basic Environment

In the code window, we type the following program text (Fig. 1.3):

Sub Pythagoras()
a = 3
b = 4
c = Sqr(a ^ 2 + b ^ 2)
End Sub

In this text, Pythagoras is the program name, a, b and c are names of varia-
bles, Sub is a keyword, End Sub is a keyword combination. The program and
variable names are appointed by us.

Fig. 1.3. The VB window with the Pythagoras program in the code window

In a programming language, words used only in the language constructs are

called keywords. We cannot use keywords as names of programs and variables
in programs. By default, Visual Basic Environment is tuned in such manner that
all keywords are highlighted in blue color (at formation of the program text in
the code window), comments are highlighted in green, syntactic errors — in red.
It is visible that the power operation (^) is written as in Excel, Sqr is the
square root function (in Excel, the square root function is SQRT).

17
 Chapter 1. Programming in Visual Basic

For convenience, if we need to place an operator on several lines, for carry-

ing over, we have to type the space character with the subsequent underscore
character. At the finish of typing these characters, we have to press the Enter key
on the keyboard.
If we need to place several operators on one line, we have to type a colon
between these operators.
An apostrophe means that information following it (up to the line end) is
a comment, i.e., a character set, which does not influence the program execution.
Thus, our program can be written as follows:

Sub Pythagoras()
a = 3: b = 4
c = _
Sqr(a ^ 2 + b ^ 2) 'according to Pythagoras
End Sub

If a comment occupies several lines, each line must be preceded by an apos-

trophe. For example, program

Sub Pythagoras()
a = 3: b = 4
c = Sqr(a ^ 2 + b ^ 2) 'according to Pythagoras:
'pythagorean pants are
'equal in all directions
End Sub

is equivalent to the previous program.

To save the program, we fulfill the following:
1) in the Excel window, File > Save As > Browse;
2) in the Save As window opened, choose a folder intended for saving the
Excel workbook, for example, My Documents;
3) by means of drop-down list Save as type, set the following file type: Excel
Macro-Enabled Workbook;
4) click on the Save button.
As a result, the Pythagoras program is saved as a part of the Excel work-
book by name Book1 with extension .xlsm.
For returning to the Pythagoras program:
1) open the Book1 workbook;
2) if the Security Warning panel appears, click on button Enable Content;
3) go to Visual Basic Environment;
4) click twice on Module1 in the project explorer window.
We will execute the Pythagoras program in the next section.

18
 1.2. Main commands of the program debugger

1.2. Main commands of the program debugger

After typing the program text, the detection and correction of errors in the
program follows. At this stage, we can use the debugger.
Let us consider the main commands of the debugger; we can see them in the
Debug menu of the VB window.
1. Step Into — the execution of one program operator or its part. The click
on Step Into is equivalent to pressing the F8 key on the keyboard. This command
is used for the step-by-step program execution.
2. Run To Cursor — the execution of the program up to the blinking cursor.
The click on Run To Cursor is equivalent to pressing Ctrl + F8.
For setting the blinking cursor in the proper place of the program, we have to
click on this place.
If we speak about key presses, the plus sign means the synchronism of these
presses, i.e., “pressing Ctrl + F8” means “simultaneous pressing the Ctrl and F8
keys”.
3. Toggle Breakpoint — the installation or liquidation of the breakpoint at
the place, where the blinking cursor is located. The breakpoint marks the pro-
gram line, where the program execution stops. This command can also be per-
formed by pressing the F9 key.
For the installation or liquidation of the breakpoint, we can click on the left
border of the code window against the proper line.
4. Clear All Breakpoints — the liquidation of all breakpoints. This com-
mand can also be performed by pressing Ctrl + Shift + F9.
5. Add Watch — the current visualization of the value of a variable. We will
review the command usage in Section 1.15.
In addition to commands Step Into and Run To Cursor, two more commands
for the program execution, Step Over and Step Out, are in the Debug menu. We
will review them in Section 1.16.
Let us consider commands Run and Reset located in the Run menu of the VB
window.
1. Run — the start of the program execution (or shorter, of the program) and
transition from one breakpoint to another. If the breakpoints are absent, the pro-
gram is executed completely. This command is represented by arrow ► on the
toolbars of the VB window, in particular, on the standard toolbar.

19
 Chapter 1. Programming in Visual Basic

The program can be started from the Excel window; we will consider this
possibility later (p. 113).
2. Reset — the discontinuation of the program execution. This command is
represented by square ■ on the toolbars.
During the program execution stops (in particular, at the breakpoints), yellow
color highlights the operator, which is not executed yet. If we place the mouse
pointer on a variable, its value is displayed.
For obtaining the hypotenuse length by means of the Pythagoras program
from the previous section, we fulfill the following:
1) click on the left border of the code window against the last line of the
program for marking this line by the breakpoint (Fig. 1.4a);
2) click on arrow ► for executing the Pythagoras program up to the
breakpoint;
3) if window Macros appears, successively click on the Pythagoras line and
the Run button in this window;
4) in the code window, whose state is depicted in Fig. 1.4b, place the mouse
pointer on the c variable in the program text; c = 5 appears (Fig. 1.5);
5) click on arrow ► for terminating the program execution.
After starting the program execution, a window containing message Can’t
execute code in break mode may appear, indicating that we forgot to terminate
(or to discontinue) the previous program execution. For correcting this error, we
fulfill the following:
1) click on the OK button in the message window;
2) click on arrow ► (or on square ■) for termination (or discontinuation) of
the program execution;
3) restart the program.
The program execution means consecutive execution of its operators: at first,
the computer sets the values of variables a and b, and then calculates the value
of c. To be convinced, we have to liquidate the breakpoint (by clicking on it) and
to execute the Pythagoras program step-by-step by the F8 key, watching the
change in variables a, b and c.
Let us assume that an error was committed: when typing operator
c = Sqr(a ^ 2 + b ^ 2)
we pressed the minus key on the keyboard instead of the plus key. In this case,
the program execution stops, and a window appears with the following message:
Run-time error ‘5’: Invalid procedure call or argument.
To understand our error, we click on button Debug in the message window.
As a result, the place, where the stop occurred, is highlighted in yellow color.
Looking at the values of a and b, we can understand the reason of the stop — the
negative value of the argument of the square root function.

20
 1.2. Main commands of the program debugger

Fig. 1.4. The Pythagoras program with the breakpoint

(a) before and (b) after starting the program execution

Fig. 1.5. The visualization of the value of c during the execution stop

21
 Chapter 1. Programming in Visual Basic

For fuller information on the possible reasons of the stop, we fulfill the fol-
lowing:
1) remaining in Visual Basic Environment, start the Excel help system by
pressing the F1 key;
2) type Error 5 in the text box of the Excel Help window;
3) click on the Search button;
4) click on the following line of the open list: Invalid procedure call or
argument (Error 5).
After correcting our error (that is, after changing minus to plus), we have to
restart the program.
During the execution, our program (as a set of zeros and units) is located in
the main memory of the computer. This program is being executed by the pro-
cessor that performs different operations including, among others, arithmetic
operations.
Now we will pass to the Visual Basic programming language.

22
 1.3. Variables. Data types

1.3. Variables. Data types

Variables in programming have about the same meaning as variables in

algebra. We recommend declaring a variable before its usage.
The declaration operator has the following syntax:

Dim variable [As type]

In this language construct:

Dim (from “dimension”) is the keyword testifying the appearance of
a new variable;
variable is the variable name;
As is the keyword;
type is the data type (Appendix 1) of the declared variable.
Here and below, the square brackets indicate an optional part of the syntax,
i.e., the part that may be absent. Such usage of the brackets is acceptable because
the square and curly brackets are not used in constructs of Visual Basic.
In other words, the declaration operator has the following two versions:

Dim variable As type

Dim variable

Initially, we will consider the first version.

When (during the program execution) the computer meets the Dim operator,
it allots a memory cell for the variable by name variable. The cell size (in
bytes) is defined by type — the data type of the variable; type is keyword
Boolean, Byte, Integer, Long, Currency or so on (Appendix 1).
According to the second column of the table in Appendix 1, the memory cell
sizes, corresponding to different variables, can strongly differ.
To understand how much information is contained in one byte, let us notice
that three bytes are usually enough for storing information on one pixel (the
color point on the display screen) — one byte each for intensities of the red,
green and blue colors.
One Dim operator allows declaring several variables if we list them through
a comma. The example operators follow:

23
 Chapter 1. Programming in Visual Basic

Dim my_variable As Double

Dim i As Byte, j As Integer, k As Integer

The restrictions on names of variables are the same as on names of programs

(p. 16), at that, the upper or lower case of letters does not matter. For example,
if the Alpha variable is declared and we try to declare the alpha variable,
Alpha will be automatically replaced by alpha.
Alpha is the English name of a Greek letter. If the keyboard does not support
the Greek language, we recommend to use English names of Greek letters
according to Appendix 2 when naming variables and programs.
For reducing programming errors, we recommend to tune Visual Basic Envi-
ronment so that it demands the declaration of variables. For this purpose, we
fulfill the following:
1) open menu Tools in the VB window;
2) click on Options;
3) activate the Editor tab;
4) turn on option Require Variable Declaration;
5) click on the OK button.
As a result, the code window corresponding to a new module will contain the
following first line:

Option Explicit

We can also put this line into the code window (or remove it) manually, as
a usual program line.
In the presence of line Option Explicit, the computer diagnoses the use
of an undeclared variable in the program text: during the program execution, the
computer displays message Variable not defined.
Data types Byte, Integer, Long, Currency, Single and Double
(Appendix 1) are called numerical data types.
According to the third column of the table in Appendix 1:
in a memory cell, corresponding to a variable of the Byte data type, non-
negative integers (up to 255) can only be stored;
in a cell, corresponding to a variable of the Integer or Long data type,
integers can be stored;
in a cell, corresponding to a variable of the Currency, Single or
Double data type, decimal numbers can be stored.
Let us pass to the second version of the Dim operator (p. 23).
If we do not specify the data type when declaring a variable (for example, by
operator Dim W), the variable (by name W) automatically receives the Variant
data type. It means that any information can be stored in a memory cell, corre-

24
 1.3. Variables. Data types

sponding to this variable; i.e., the Variant data type is similar to the general
format of Excel.
Let us consider operator

Dim i, j As Integer

This operator is equivalent to the following:

Dim i As Variant, j As Integer

If we need the Integer data type for both variables, i and j, we should
declare them as follows:

Dim i As Integer, j As Integer

or

Dim i As Integer
Dim j As Integer

Later we will consider ways of declaring variables without the use of the Dim
keyword (p. 81).
As an example of using the declaration operator, let us consider the following
program for calculating the number of days in the 20th century and defining the
current date and time:
Listing 1.1
Sub Century_20()
Dim D1 As Date, D2 As Date, D3 As Date
Dim N As Long
D1 = #1 Jan 1900# 'beginning date of century
D2 = #31 Dec 1999# 'ending date of century
N = D2 - D1 + 1 'number of days in century
D1 = Time 'current time
D2 = Date 'current date
D3 = Now 'date and time
End Sub

In the second and third lines of this program, Date and Long are the data
types (Appendix 1); #1 Jan 1900# and #31 Dec 1999# mean dates on Janu-
ary 1, 1900 and December 31, 1999. The subtraction of the first date from the
second date determines the number of days between these dates.

25
 Chapter 1. Programming in Visual Basic

Word Time means determination of the current time by means of the VB

function by name Time. In other words:
Time is the call of the Time function of VB;
the value of this function is the current time of the day or, that is the
same, the Time function returns the current time of the day into the program
(“into the program” may be omitted).
In operator

D2 = Date

word Date means the call of the Date function returning the current date into
the program.
We see that Date is a name of the VB function and a name of the data type.
Thanks to various contexts, it does not lead to confusion.
Word Now means the call of the Now function returning the current date and
time together.
Regarding the VB functions, we will talk in more details later, in particular,
in Sections 1.4 and 1.9.
To be convinced of the operational capability of the Century_20 program,
we fulfill the following operations:
1) insert a module into the active Excel workbook (p. 15);
2) enter the Century_20 program text into the code window of the new
module;
3) make the step-by-step execution of this program by means of the F8 key,
watching the values of variables D1, D2, D3 and N.
It should be emphasized that, before the first press of key F8, we must place
the blinking cursor inside the program text, not in line Option Explicit.
Text Listing 1.1 of the Century_20 program can be entered into the code
window by means of the keyboard. We can also copy it from the enclosed CD.
For copying:
1) open file Listing_1_01.txt with the Notepad editor, for example, by dou-
ble click on the pictogram of this file in Windows Explorer;
2) in the Notepad window opened, highlight the program text and copy it
into Windows Clipboard, for example, by pressing Ctrl + C;
3) by the click, locate the blinking cursor in the code window of Visual
Basic Environment;
4) paste the program text from Windows Clipboard into the code window,
for example, by pressing Ctrl + V;
5) close the Notepad window with file Listing_1_01.txt.

26
 1.4. Two main functions for conversion of data types

1.4. Two main functions for conversion of data types

A string is a quoted sequence of characters. The example strings follow:

"Hello, World!"
"13.333"
"37 RUR"
"$ 37"
" "
The last string contains only the space character.
Let us supplement the string definition by the empty string, "", which does
not contain any characters.
Strings " " and "" are used in program Listing 1.7. In Section 1.19, we will
expand the string definition even more.
Converting string to number is often necessary. For this purpose, the Val
function is used. It converts the numerical beginning of string to number. If the
Val function cannot perform this, it returns zero into the program. The argument
of the Val function is a string; this function returns a number.
For the backward conversion (that is, number to string), the Str function is
used. The argument of this function is a number, variable of numerical data type
(p. 24) or arithmetic expression (Section 1.8). The Str function returns a string.
Let us make the step-by-step execution of the program below.
Listing 1.2
Sub StrVal()
Dim strA As String
Dim curB As Currency
strA = "45.77"
curB = Val(strA) 'result: curB = 45.77
strA = Str(curB) 'result: strA = " 45.77"
curB = Val("4.7 = X") 'result: curB = 4.7
curB = Val("4,7 = X") 'result: curB = 4
curB = Val("X = 4.7") 'result: curB = 0
curB = Val("") 'result: curB = 0
End Sub

27
 Chapter 1. Programming in Visual Basic

The first comment corresponds to the case when operator

strA = Str(curB)

is highlighted in yellow color during the step-by-step program execution. This

comment means the following: if the mouse pointer is located on curB, infor-
mation curB = 45.77 appears.
Information curB = 45.77 is the result of executing operator

curB = Val(strA)

which is in the same line with the comment.

The second comment corresponds to the case when operator

curB = Val("4.7 = X")

is highlighted in yellow color during the step-by-step execution. This comment

means the following: if the mouse pointer is located on strA, information
strA = “ 45.77” appears. And so on.
All comments, which begin with word “result” or “Returns” (p. 33), have
similar sense.
Let us remind that, during the stops of the program execution, yellow color
highlights the operator, which is not executed yet.
In addition to the Val and Str functions, there are other functions for con-
version of data types. We will review them in Section 1.8.

28
 1.5. Constants

1.5. Constants

Constants are similar to variables. However, unlike a variable, the content of

the memory cell, corresponding to a constant, cannot be changed during the pro-
gram execution. There are two versions of constants in Visual Basic, named
built-in and user-defined constants.
The user-defined constant can be declared by means of operator

Const invariable [As type] = value

In this operator:
Const is the keyword testifying the appearance of a new constant;
invariable is the constant name;
As is the keyword, as in the Dim operator (p. 23);
type is the data type (Appendix 1) of the declared constant;
value is a value of the declared constant.
The restrictions on names of constants are the same as on names of variables
and programs (p. 16).
Examples of the constant declaration follow:

Const e As Double = 2.718281828

'base of natural logarithm
Const e = 2.718281828
Const phi = 1.618033989 'gold relation
Const Flag As Boolean = False
Const Message = "End of Work"
Const Millennium As Date = #1 Jan 2000#
Const beta As Currency = 2 ^ 0.5

In the fourth example operator, Boolean is the so-called logical data type
(Appendix 1).
When executing the last operator, the rounded square root of 2 (that is,
1.4142) is assigned to constant beta. This example shows that value in the
Const operator can be an elementary arithmetic or logical expression (Sec-

29
 Chapter 1. Programming in Visual Basic

tions 1.8 and 1.10). In this case, the content of the memory cell, corresponding to
the constant, is determined by the expression value rounded according to the
type data type.
One Const operator allows declaring several constants if we list them
through a comma. The example operator follows:

Const Min = 0, Max = 1000, tau As Double = 6.283185307

As an example of using constants, let us consider the following program for

conversion of an angle from degrees to radians:

Sub deg2rad()
Dim angleD As Double
Dim angleR As Double
Const pi As Double = 3.141592654 'pi = tau / 2
angleD = 270 'angle equals 270 degrees
angleR = angleD * pi / 180
'result: angle in radians
End Sub

As we see, operator

Const pi As Double = 3.141592654

declares the pi constant before its usage in operator

angleR = angleD * pi / 180

The rad2deg program below, which converts an angle from radians to

degrees, is similar to the above program.

Sub rad2deg()
Dim angleD As Double
Dim angleR As Double
Const pi180 As Double = 3.141592654 / 180
angleR = 4.5 'angle equals 4.5 radians
angleD = angleR / pi180
'result: angle in degrees
End Sub

On p. 82, we will consider the declaration of a constant by means of keyword

combination Public Const.

30
 1.5. Constants

The built-in constant does not need any declaration. Names of the built-in
constants of Visual Basic begin with prefix vb, for example, vbFriday (this con-
stant equals 6).
For names (in particular, names of constants), the developers of Windows
accepted the following agreement: names of similar data begin with the same
short prefix. In particular, the built-in constants of Visual Basic have prefix vb,
the built-in constants of Excel have prefix xl.
In addition to vbFriday, we will come across the following built-in constants:
vbYesNo, vbYes, vbTab, vbCrLf, vbCr, vbLf, xlR1C1, xlCalculationAutomatic,
xlCalculationManual, xlDialogOpen, xlDialogSaveAs, xlCenter, etc.

31
 Chapter 1. Programming in Visual Basic

1.6. Obtaining information

For obtaining information on a built-in constant, we must press the F2 key

when the VB window is active. As a result, the object browser window appears
(Fig. 1.6). In the top box of this window, we should set <All Libraries>
by means of the drop-down list. In the text box below, we should type what is
interesting for us, for example, vbFriday. Then we click on the binoculars picto-
gram. The answer is in the Search Results area (Fig. 1.7).
For closing the object browser window, we have to click on the little cross at
the right end of the menu bar.

Fig. 1.6. The VB window with the object browser

window instead of the code window

32
 1.6. Obtaining information

Fig. 1.7. Information on the built-in vbFriday constant

The Excel help system, started by pressing the F1 key, is useful too (p. 22).
For accelerating the process of finding the necessary information, the blinking
cursor must be preliminarily located on the word of interest.
By means of the Excel help system, we will study the Val function. For this
purpose, let us fulfill the following.
1. Enter the StrVal program (p. 27) into the code window.
2. Locate the blinking cursor on the Val word by clicking on it.
3. Press the F1 key. As a result, the Excel Help window, containing the full
information on the Val function, is displayed (Fig. 1.8).
4. After studying the last information, copy fragment

Dim MyValue
MyValue = Val("2457") ' Returns 2457.
MyValue = Val(" 2 45 7") ' Returns 2457.
MyValue = Val("24 and 57") ' Returns 24.

from the bottom part of the Excel Help window into the StrVal program as
follows:

33
 Chapter 1. Programming in Visual Basic

1) highlight this fragment by the mouse, as in Notepad;

2) copy it into Windows Clipboard by pressing Ctrl + C;
3) locate the blinking cursor in the code window, against the last line of the
StrVal program;
4) paste the fragment from Windows Clipboard into the program text by
pressing Ctrl + V.

Fig. 1.8. The Excel Help window with information on the Val function

34
 1.6. Obtaining information

As a result, the StrVal program takes the following form:

Sub StrVal()
Dim strA As String
Dim curB As Currency
strA = "45.77"
curB = Val(strA) 'result: curB = 45.77
strA = Str(curB) 'result: strA = " 45.77"
curB = Val("4.7 = X") 'result: curB = 4.7
curB = Val("4,7 = X") 'result: curB = 4
curB = Val("X = 4.7") 'result: curB = 0
curB = Val("") 'result: curB = 0
Dim MyValue
MyValue = Val("2457") ' Returns 2457.
MyValue = Val(" 2 45 7") ' Returns 2457.
MyValue = Val("24 and 57") ' Returns 24.
End Sub

We advise the reader to execute this program step-by-step (by means of the
F8 key), watching the value of the MyValue variable.

35
 Chapter 1. Programming in Visual Basic

1.7. Assignment operator

The assignment operator has the following syntax:

variable = expression

Here, variable is the variable name, expression is an arithmetic or logical

expression (Sections 1.8 and 1.10) or string, which can be considered as an
expression (Section 1.19). A separately taken number, constant, variable or func-
tion is also an arithmetic expression. In the assignment operator, = is the
so-called assignment sign.
The assignment operator works as follows:
1) the value of expression is calculated;
2) the resulting value is assigned to variable, i.e., is written into the cor-
responding memory cell.
If the data type of the variable in the left-hand side of the assignment operator
does not coincide with the type of the expression value in the right-hand side, the
value type is generally converted (transformed) during the execution of the
assignment operator.
Let us make the step-by-step execution of the following program for convert-
ing strings "78.8", "78,8" and "78;8" to numbers of the Currency data
type.

Sub Conversion()
Dim curA As Currency
Dim curB As Currency
Dim curC As Currency
curA = "78.8" 'result: curA = 78.8
curB = "78,8" 'result: curB = 78.8
curC = "78;8" 'result is absent
End Sub

Because of executing the first and second assignment operators, strings

"78.8" and "78,8" are successfully converted to number 78.8.
When executing assignment operator

36
 1.7. Assignment operator

curC = "78;8"

the stop occurs with the following information: Run-time error ‘13’: Type mis-
match. It speaks about the following:
the types of variable on the left and of value on the right of the assign-
ment sign (=) are different;
the computer cannot convert string "78;8" to a number.
We will continue considering the data type conversion in the next section of
the book.
Unlike other programming languages, for example C++, multiple assign-
ments, for example x = y = z = 1.3, are inadmissible in VB. We have to use
several assignment operators with the same right-hand side, i.e., language con-
struct

x = 1.3
y = 1.3
z = 1.3

or

x = 1.3: y = 1.3: z = 1.3

is correct.

37
 Chapter 1. Programming in Visual Basic

1.8. Arithmetic expression

In Visual Basic, an integer is represented by a sequence of figures with the

minus sign or without any sign. Examples of integers are

–18 32 0

If a number has the fractional part, it separates from the integral part by
a point. In this case, we may omit the integral part if it equals zero. Examples
of decimal numbers follow:

0.5 –5.68 –.12 .035 3.

In the last example, 3 with a point means a number with zero fractional part, i.e.,
an integer, in fact.
We reviewed the main form of decimal numbers.
Decimal numbers may also be represented in exponential form. For example,
-1.6E-19 is the electron charge, -1.6·10-19 C (its absolute value is figured in
Section 3.13). Instead of E, letter D may be used in the exponential representa-
tion, i.e., the electron charge, -1.6·10-19 C, may be written as -1.6D-19
One of the main constructs of any programming language is an arithmetic
expression similar to the algebraic expression in mathematics. However, we can-
not omit the multiplication sign in the arithmetic expression. Table 1 below con-
tains equivalent algebraic and arithmetic expressions.

Table 1. Examples of expressions

Algebraic expression Arithmetic expression of VB

5x 12 y 5 * x + 12 * y
x
x / y
y
yx y ^ x
x x
19.55·10-6 19.55E-6 or 19.55D-6

38
 1.8. Arithmetic expression

The arithmetic expression contains the invariables (numbers and constants),

variables and/or functions related by arithmetic operations. As we already men-
tioned in the previous section, an individual number, constant, variable or func-
tion is also an arithmetic expression.
The expressions in Table 1 do not include functions. We will review arithme-
tic expressions with functions in the next section.
Arithmetic operations are denoted as follows: + (addition), - (subtraction or
sign change operation), * (multiplication), / (division), ^ (power operation),
\ (integer division, i.e., division of integers neglecting the integer remainder),
Mod (modulus operation, i.e., determining the integer remainder after division of
integers).
According to the assignment operator syntax, the arithmetic expression is on
the right of sign =. For example, assignment operator

z = 5 * x + 12 * y

includes arithmetic expression

5 * x + 12 * y

Let us consider the following program:

Sub Arithmetic1()
Dim m As Integer
Dim n As Integer
Dim x As Double
m = 5
n = 2
x = m / n 'result: x = 2.5
x = m \ n 'result: x = 2
x = m Mod n 'result: x = 1
End Sub

To verify the correctness of this program, we advise the reader to make the
step-by-step execution (by means of the F8 key), watching the change in the
value of x.
If the arithmetic expression contains several operations, the order of their per-
formance is defined by the following rules of priorities of arithmetic operations:
1) first of all, the power operation (^) is performed;
2) next, the multiplication and division (*, /) are performed in that sequence
as they are in the expression;

39
 Chapter 1. Programming in Visual Basic

3) the integer division ( \ );

4) the modulus operation (Mod);
5) the sign change operation (–);
6) finally, the addition and subtraction (+, –) are performed in that sequence
as they are in the expression.
As we see, the power operation (^) has the highest priority in VB.
The rules of priorities of arithmetic operations in VB differ from the rules of
priorities in Excel regarding the power operation, multiplication, division, the
sign change operation (so-called negation), addition and subtraction: the nega-
tion (–) is performed first, i.e., has the highest priority in Excel.
To be convinced, we put

=-1^2

into the Excel formula box and click on the tick button of the Excel formula bar
(or press the Enter key). As a result, value 1 appears in the active cell on the
worksheet. After the execution of VB operators

Dim i As Integer
i = -1 ^ 2

the i variable is equal to -1.

In both VB and Excel, parentheses are used for changing the sequence of the
operations: the values of the parenthesized arithmetic expressions are calculated
at first.
We will not see the square and curly brackets in the arithmetic expressions of
Table 2 below because, as it was already told, such brackets are not used in the
VB constructs.
When calculating the expression value, results of performance of intermedi-
ate arithmetic operations remain in the processor or are written into cells of the
cache memory (from where they are read when required).
The cache memory has a short time of reference (that is, of writing and read-
ing information); this time is much shorter than the time of reference for the
main memory. The cache memory is intended for temporary storage of interme-
diate results and contents of memory cells often used. The program fragment,
which is being executed, may also be stored in the cache memory.
Let us consider the following assignment operator with the arithmetic expres-
sion in the right-hand side:

z = 5 * x + 12 * y

The processor is performing this operator approximately thus:

40
 1.8. Arithmetic expression

1) multiplies 5 by the content of cell x;

2) writes the result into a cache memory cell, for example, cache1;
3) multiplies 12 by the content of cell y;
4) adds the result to the value of cache1;
5) writes the result into cell z.

Table 2. Examples of expressions

Algebraic
Arithmetic expression of VB
expression
7{3[a+b(c+d)]+8}+2 7 * (3 * (a + b * (c + d)) + 8) + 2
-a ^ b
-ab or
–(a ^ b)
a-b a ^ (-b)
ab-c a ^ (b - c)
10-4.7 10 ^ (-4.7)
104.7 10 ^ 4.7
A·B A * B
A * (-B)
-A * B
A(-B)
or
-(A * B)
c
ab a ^ (b ^ c)
a ^ b ^ c
(ab )c or
(a ^ b) ^ c
a b (a * b) / (c * d)
or
c d a * b / (c * d)
a * 1E4
a * 10E3
a ·104
or
a * 10000

In mathematics and programming, participants of operations are called

operands, both in the case of arithmetic operations and in the case of logical
operations (Section 1.10). For example, arithmetic expression 5 * x includes
the following two operands: integer 5 and variable x.

41
 Chapter 1. Programming in Visual Basic

For a better understanding of the rules of priorities of arithmetic operations,

let us consider the following program:

Sub Arithmetic2()
Dim m As Integer
Dim n As Integer
Dim x As Single
Dim y As Single
x = 3
m = 2
n = -1
y = (-3) ^ m 'result: y = 9
y = -(3 ^ m) 'result: y = -9
y = -3 ^ m 'result: y = -9
y = 10 + (x + 7) ^ (m + n) 'result: y = 20
y = 10 + x + 7 ^ m + n 'result: y = 61
End Sub

We advise the reader to make the step-by-step execution of this program,

watching the value of the y variable and explaining the value origin.
In addition, we advise the reader to verify that

(m \ n) * n + m Mod n

equals m for arbitrary integers m and n (naturally, n ≠ 0). For that, the reader has
to write a program, which is similar to Arithmetic1, and to execute the new
program step-by-step.
Arithmetic expressions may contain variables and invariables of different
types. If the type of the value of arithmetic expression in the right-hand side of
the assignment operator (on the right of sign =) does not coincide with the data
type of the variable in the left-hand side of the assignment operator (on the left of
sign =), the type of the value is converted during the assignment.
Let us consider the situation when the value of arithmetic expression on the
right of sign = has a fractional part and the variable on the left of sign = is of the
Integer or Long data type. During the assignment, the value is transformed
according to the following rules for rounding off:
if the fractional part of the value is equal to 0.5, this value is rounded up
to the even number from two nearest integers;
otherwise, the value is rounded up to the nearest integer.
Because operations \ and Mod are applicable only to integers, the execution
of these operations over numbers with a fractional part begins with the rounding

42
 1.8. Arithmetic expression

off the operands to integers according to the formulated rules. The results of
operations \ and Mod are integers.
An operation with one operand is called a unary operation. Among the arith-
metic operations, only the sign change operation (–) is unary. An operation with
two operands (^, *, /, \, Mod, +, – as subtraction) is called a binary operation.
VB includes special functions for converting data types. Two of these func-
tions (Str and Val) were reviewed in Section 1.4; the remaining functions are
listed in Table 3 below.

Table 3. Functions for converting data types

Function name Resulting data type

CBool Boolean
CByte Byte
CCur Currency
CDate Date
CDbl Double
CInt Integer
CLng Long
CSng Single
CStr String
CVar Variant

Requirements to the argument of these functions and examples of their usage

are given in the Excel help system. For accelerating the process of finding the
necessary information, we must press the F1 key when the VB window is active.
Let us make the step-by-step execution of the program below.

Sub Functions()
Dim intN As Integer
Dim strN As String
Dim curN As Currency
intN = -15
strN = Str(intN) 'result: strN = "-15"
strN = CStr(intN) 'result: strN = "-15"
intN = 15 '8th operator
strN = Str(intN) '9th operator
'result: strN = " 15"
strN = CStr(intN) '10th operator
'result: strN = "15"
curN = 25.5 '11th operator

43
 Chapter 1. Programming in Visual Basic

intN = 1 + CInt(curN) '12th operator

'result: intN = 27
intN = CInt(1 + curN) '13th operator
'result: intN = 26
intN = CInt("78.8") '14th operator
'result: intN = 79
intN = CInt("78,8") '15th operator
'result: intN = 79
End Sub

Note the following:

if the argument of the Str and CStr functions is a non-negative number,
these functions return different strings (see the 9th and 10th operators);
the CInt function rounds the argument value according to the rules for
rounding off (see the 12th and 13th operators);
a string may be the CInt function argument (see the 14th and 15th
operators).

44
 1.9. Mathematical functions. Functions of date and time

1.9. Mathematical functions.

Functions of date and time

Let us start with an analysis of the mathematical functions given in the table
of Appendix 3.
We already used the square root function, Sqr(x), in our first program —
Pythagoras on p. 17.
The argument of trigonometric functions (cosine, sine and tangent) is an
angle in radians, not in degrees.
Function Atn(x) is an inverse trigonometric function, arctan x . The
arctangent returns (into the program) the angle in radians from -π/2 to π/2 whose
tangent is equal to the value of x. Such angle is called the principal angle of
tan x .
The sign function, Sgn(x), returns -1, 0, 1 at x < 0, x = 0, x > 0, respec-
tively.
The Log(x) function is the natural logarithm of x, ln x . According to
the logarithm properties [3], the following expressions are valid for the decimal
logarithm: lg x = ln x / ln 10 = ln x / 2.302585093.
In Appendix 3, in addition to the main mathematical functions of Visual
Basic, the VB operators are given for counting the values of trigonometric func-
tion cot x , of inverse trigonometric functions arcsin x , arccos x and arccot x
and of decimal logarithm lg x .
In addition to the mathematical functions of Appendix 3, let us consider func-
tion Round(x[, n]) intended for rounding off numbers with a fractional part.
As we know, the square brackets separate an optional part of the construct.
In other words, this function of VB has the following two versions:
the Round(x, n) function returns the value of x, rounded up to n deci-
mal places;
the Round(x) function returns the integer obtained by rounding off the
value of x according to the rules formulated on p. 42; this function is identical to
the Round(x, 0) and CInt(x) functions.
Arguments of the mathematical functions are arithmetic expressions.
Assignment operator

45
 Chapter 1. Programming in Visual Basic

c = Sqr(a ^ 2 + b ^ 2)

in the Pythagoras program (p. 17) is an example of this.

The mathematical functions are used in arithmetic expressions, and the func-
tions have a priority as compared to the arithmetic operations when calculating
the expression value. For example, during the execution of operator

e = b / c * Cos(a) ^ 3 – d

the processor first calculates the cosine value, and then:

1) cubes this value;
2) writes the result into the cache1 cell of the cache memory;
3) successively performs the division, multiplication and subtraction for cal-
culating the value of arithmetic expression

b / c * cache1 – d

4) writes the result into the memory cell corresponding to variable e.

To learn the use of the mathematical functions, let us do the following:
1) put operator block

Dim V As Single, W As Single

V = intN + Round(Sqr(2 * intN), 2)
W = Round(V) ^ 2

into the code window containing the Functions program (p. 43), above the
last line;
2) mark the last line of the program by the breakpoint;
3) click on arrow ► for the program execution up to the breakpoint;
4) make sure that the calculated values of V and W are equal to 33.21 and
1089, respectively;
5) explain these results;
6) click on arrow ► for terminating the program execution.
In addition to the considered functions with one and two arguments, there is
the Rnd function (from “random”) without arguments. This function is intended
for generation of random numbers used for modeling random phenomena. The
simplest random phenomenon can be described as follows: there is a 50 %
chance that the “head” or “tail” will be the result of a coin flip.
The idea of modeling random phenomena is known for a long time. Follow-
ing the advent of the electronic computers, this idea was developed in the 1950s
under the name of the Monte Carlo method.

46
 1.9. Mathematical functions. Functions of date and time

The Monte Carlo method is used for modeling financial risks, semiconductor
devices and evolution of stars. It is only a part of the problems demanding
generation of random numbers.
The use of the Rnd function is described in the Excel help system, which
must be started by pressing the F1 key when the VB window is active. The
RandomNumbers program in Section 1.15 and code Listing 6.11 in Section 6.8
are examples of the Rnd function’s usage.
Functions of date and time, Time, Date and Now, are without arguments
too. These functions return the following values of the Date data type: current
time, date and date together with time, respectively. They appeared in program
Century_20 (p. 25). We will encounter these functions of date and time more
than once.

47
 Chapter 1. Programming in Visual Basic

1.10. Logical expression

In addition to arithmetic expressions, logical expressions are also important

in Visual Basic.
The logical expression uses the following well known comparison signs: <
(less than), > (greater than), = (equal to), <= (less than or equal to, ≤ in mathe-
matics), >= (greater than or equal to, ≥ in mathematics), <> (unequal to, ≠ in
mathematics).
The logical expression accepts one of the two Boolean (logical) values, True
(logical unit) or False (logical zero). Separately taken True or False is also
a logical expression.
Examples of the logical expression follow:
5 >= 3
5 < 3
False
The first logical expression accepts True; the second and third logical expres-
sions accept False.
Let us consider the following program:

Sub Logic1()
Dim x As Integer
Dim y As Integer
Dim blnA As Boolean
x = 5: y = 2
blnA = x > y 'result: blnA = True
blnA = x = y 'result: blnA = False
End Sub

In this program, we see four assignment operators and two logical expres-
sions (x > y and x = y). According to the assignment operator syntax, the logical
expressions are on the right of the assignment sign. The values of the logical
expressions for x = 5 and y = 2 are given in the corresponding comments.
In complicated logical expressions, logical operations are used. We will con-
sider three of them: Not, And, Or.

48
 1.10. Logical expression

Operation Not is the so-called logical negation. It is defined as follows:

if A equals True, then Not A equals False;
if A equals False, then Not A equals True.
Operation Not has one operand (A), i.e., it is a unary operation.
Definitions of logical operations And and Or are given by the following two
tables.

Definition of the And operation

A B A And B
True True True
True False False
False True False
False False False

Definition of the Or operation

A B A Or B
True True True
True False True
False True True
False False False

According to these tables, the And and Or operations have two operands
(A and B), i.e., they are binary operations.
Logical expression A And B is equal to True only in that case when both
operands are equal to True. In all other cases, expression A And B is equal to
False. The And operation is called conjunction or logical multiplication.
Expression A Or B is equal to False only in that case when both operands
are equal to False. In all other cases, expression A Or B is equal to True. The
Or operation is called disjunction or logical addition.
In the presence of several logical operations in a logical expression, the order
of their performance is defined by the following rules of priorities:
1) first of all, operation Not (logical negation) is performed;
2) further, And (logical multiplication);
3) in last turn, Or (logical addition) is performed.
For change of sequence of the operations’ performance, parentheses are used,
as in arithmetic expressions. Parentheses may be also used for readability of
logical expressions.

49
 Chapter 1. Programming in Visual Basic

For a better understanding of the logical operations, let us consider the

following program:

Sub Logic2()
Dim x As Double
Dim y As Double
Dim z As Double
Dim blnA As Boolean
x = 1
y = 2.87
z = 3.12
blnA = (x > y) And (y < z) 'result: blnA = False
blnA = x < y And y < z 'result: blnA = True
blnA = x > y Or y > z 'result: blnA = False
blnA = Not (x < y Or Not y < z)
'result: blnA = False
blnA = Not x > y And x > y 'result: blnA = False
blnA = Not (x > y And x > y) 'result: blnA = True
End Sub

In operator

blnA = (x > y) And (y < z)

parentheses are used for readability of logical expressions x > y and y < z.
These parentheses may be omitted.
We advise the reader to verify the correctness of the Logic2 program by
means of the step-by-step execution.
Note that double logical expressions, for example 0 < x <= 1, are inadmissi-
ble in VB. Instead of 0 < x <= 1, we have to write

0 < x And x <= 1

or

x > 0 And x <= 1

For grasping the material of this and the previous sections, we advise the
reader to write a program allowing to define the values of y, for which
CInt(y), Fix(y), Int(y), Round(y) are equal to each other if y accepts
the following values:

50
 1.10. Logical expression

-1.8, -1.25, 1.27, 1.68;

f (a ) , f (0.5a 0.5b) , f (b) , 0.5 f (a) 0.5 f (b) .
Here, functions Fix(y) and Int(y) of VB are described in Appendix 3, func-
tions CInt(y) and Round(y) are from Sections 1.8 and 1.9, function f (x ) is
from Appendix 4, a and b are the boundaries of the f (x ) function’s domain.
In addition, we advise the reader to replace condition “CInt(y), Fix(y),
Int(y), Round(y) are equal to each other” of the previous task by one of the
following conditions:
1) Fix(y) = CInt(y) And Fix(y) <> Int(y)
2) Int(y) = CInt(y) And Int(y) <> Fix(y)
3) Fix(y) = Int(y) Or Fix(y) <> CInt(y)
4) Fix(y) = CInt(y) Or Fix(y) <> Int(y)
5) Int(y) = CInt(y) Or Int(y) <> Fix(y)
6) Fix(y) = Int(y) Or Fix(y) = CInt(y)
7) Int(y) = CInt(y) Or Int(y) = Fix(y)
8) CInt(y) = Fix(y) Or CInt(y) = Int(y)
9) Fix(y) = Int(y) And Fix(y) <> CInt(y)

51
 Chapter 1. Programming in Visual Basic

1.11. GoTo operator

Operators of the previous programs are executed by turn. Such programs are
called linear programs.
The GoTo operator is used to change the order of execution of the program
operators. It has the following syntax:

GoTo lbl

In this syntax, the so-called label, lbl, may be one of the following:
a non-negative integer without a sign (0, 1, 2, 3, …);
a sequence of letters, figures and underscores beginning with a letter, for
example, start_53a.
We have to place the lbl label in front of the operator, to which the jump
must be performed (or, that is the same, to which the control must be trans-
ferred). We have to type a colon behind the label.
After executing the operator with lbl in front, the next operator will be exe-
cuted if the labeled operator is not GoTo.
Examples of GoTo usage are given on p. 54: we see two labels in the IT2
program, 2 and LastLine.
If the label is a non-negative integer, this label may be called an operator’s
(line’s) number. For example, operator

If X > 9 And X < 12 Then GoTo LastLine

in the IT2 program with label 2 in front may be named as operator 2 (line 2).
The GoTo operator is often called the unconditional jump operator. In the
next section, we will consider GoTo as a part of the so-called conditional jump
operator.

52
 1.12. Decision-making constructs

1.12. Decision-making constructs

This is a typical situation when, in a certain place of a program, it is neces-

sary to execute those or other operators depending on some conditions. The
choice of the operators is performed by means of one of two decision-making
constructs, If…Then or Select Case.
The first decision-making construct, If…Then, is called the conditional
operator. Below are several versions of this operator.
The simplest conditional operator follows:

If condition Then statement1

where condition is a logical expression.

The computer calculates the value of condition. If True (False) is the
result, we will say that the condition is true (false).
If the condition is true, operator statement1 is executed; if the condition
is false, operator statement1 is not executed.
Further, the next operator (following the If…Then construct) is executed,
regardless of whether or not statement1 was executed (if statement1 is
not the GoTo operator).
Let us consider the following example program:

Sub IT1()
Dim X As Byte
X = 12 'initial value of X
If (X > 9 And X < 12) Then X = X + 1
X = X + 2
X = X * 2 'final value of X
End Sub

For understanding this program, we advise the reader to do the following:

1) install the breakpoint against the End Sub line;
2) click on arrow ► for the program execution up to the breakpoint;
3) make sure that the calculated value of X is equal to 28 during the stop of
the program execution;

53
 Chapter 1. Programming in Visual Basic

4) explain this result;

5) click on arrow ► for terminating the program execution.
A special case of the reviewed If…Then construct is the conditional jump
operator with the following syntax:

If condition Then GoTo lbl

According to this construct including GoTo, if the condition is true, the jump is
performed to the operator labeled by lbl.
The program below is an example of using the conditional jump operator.

Sub IT2()
Dim X As Integer
X = 12
2: If X > 9 And X < 12 Then GoTo LastLine
X = X - 2
GoTo 2
LastLine:
End Sub

We advise the reader to make the step-by-step execution of this program and
to explain why the value of X changes so, instead of differently.
The quantity of operators, which must be executed when the condition is true,
may be greater than one. In this case, the following construct is used:

If condition Then
statements
End If

where statements is an operator block. This conditional operator is per-

formed as follows.
If the condition is true, block statements (below the Then keyword) is
executed. If the condition is false, block statements is not executed.
Further, the jump is performed to the operator following the If…Then con-
struct, regardless of whether or not block statements was executed (if this
block does not include the GoTo operator).
Let us consider the following If…Then construct:

If condition Then statement1 Else statement2

If the condition is true, operator statement1 (behind keyword Then) is

executed. If the condition is false, operator statement2 (behind keyword

54
 1.12. Decision-making constructs

Else) is executed. Further, the jump is performed to the operator following the
If…Then construct, without dependence of what operator was executed earlier,
statement1 or statement2 (if these operators are not GoTo).
The program below is an example of using the last construct.

Sub IT3()
Dim X As Byte
X = 12 'initial value of X
If (X > 9 And X < 12) Then X = X + 3 _
Else X = X + 2
X = X * 2 'final value of X
End Sub

We will assume that operator blocks statements1, statements2, ...,

statementsN, statements do not include GoTo (for brevity).
Let us consider the following construct, which is similar to the previous
If…Then construct, but contains blocks instead of operators:

If condition Then
statements1
Else
statements2
End If

If the condition is true, block statements1 (which is below keyword

Then) is executed. If the condition is false, block statements2 (which is
below keyword Else) is executed. Further, the jump is performed to the opera-
tor following the End If keyword combination, without dependence of what
block was executed earlier, statements1 or statements2.
Let block statementsN be executed after checking not one but several
conditions. In this case, we use the following If…Then construct:

If condition1 Then
statements1
ElseIf condition2 Then
statements2
∙ ∙ ∙ ∙
ElseIf conditionN Then
statementsN
[Else
statements]
End If

55
 Chapter 1. Programming in Visual Basic

If the first condition is true (that is, logical expression condition1 accepts
True), then block statements1 is executed. If this condition is false, the
second condition (which is behind keyword ElseIf) is checked. If this condi-
tion is true (condition2 = True), then block statements2 is executed,
and so on. If conditionN = True, then block statementsN is executed.
If none of the N conditions is true, then block statements (located below the
Else keyword) is executed.
Further, the jump is performed to the operator following the End If key-
word combination, without dependence of what block was executed earlier,
statements1, statements2, ..., statementsN or statements.
An example of using the last construct is in codes Listings 5.7 and 5.8 of
Section 5.11.
We reviewed several versions of the conditional operator. In addition, let us
consider function

IIf(condition, expression1, expression2)

Arguments of this function have the following sense:

condition is a logical expression;
expression1, expression2 are arithmetic or logical expressions
(Sections 1.8 and 1.10) or strings, which can be considered as expressions
(Section 1.19).
Depending on the value of condition (True or False), the IIf func-
tion returns into the program the value of expression1 or expression2,
respectively.
The program below is an example of the function usage.

Sub IT4()
Dim intA As Integer, strA As String
intA = 6
strA = IIf(intA Mod 2 = 0, "Even", "Odd")
End Sub

The second decision-making construct, Select Case, is called the case

operator. The syntax of this construct is as follows:

Select Case expression

Case value1
statements1
Case value2

56
 1.12. Decision-making constructs

statements2
∙ ∙ ∙ ∙
Case valueN
statementsN
[Case Else
statements]
End Select

where expression is an arithmetic or logical expression or string, value1,

value2, ..., valueN are given numbers, Boolean values or strings. The case
operator is performed as follows.
If expression accepts the value of value1, block statements1 is
executed. If expression accepts the value of value2, block statements2
is executed, and so on. If expression is not equal to any of the N values
(value1, value2, ..., valueN ), then block statements (located below the
Case Else keyword combination) is executed.
Further, the jump is performed to the operator following the case operator,
without dependence of what block was executed earlier, statements1,
statements2, ..., statementsN or statements.
When the same block must be executed at several values of expression,
we have to enumerate these values (through a comma) behind the Case key-
word.
As an example, let us consider the following program:

Sub Choice()
Dim x As Integer
x = 1
Select Case 2 * x + 1
Case 1
x = x + 1
Case 2, 3, 4
x = 10
Case Else
x = 20
End Select
x = x Mod 3
End Sub

We advise the reader to execute this program step-by-step and to explain

why the value of x changes so, instead of differently.

57
 Chapter 1. Programming in Visual Basic

1.13. Cycles

To execute repeatedly an operator block, we can use one of three cycle opera-
tors, For…Next, While…Wend and Do…Loop.
The For…Next cycle is used when the number of the block’s executions is
known in advance, i.e., before the first execution of this block. This construct has
the following syntax:

For counter = beginning To ending [Step growth]

statements
Next [counter]

where counter is a variable of numerical data type (p. 24), beginning and
ending are the boundaries of the counter change, growth is the step of this
change; beginning, ending and growth are the cycle parameters.
Let us consider the For…Next cycle at a positive value of growth.
At first, the value of beginning is assigned to the counter variable.
Further, condition counter > ending is checked. If the result is True, the
cycle is completed, at that, block statements is not executed even once.
If the result of checking condition counter > ending is equal to False,
then block statements is executed for the first time. After that, the jump
occurs to the cycle beginning. Further, the counter variable’s value increases
by growth, and condition counter > ending is checked again. If the result
is equal to False, then block statements is executed for the second time,
and so on.
The cycle is completed when the check of condition counter > ending
gives True. In this case, the operator following the cycle is executed.
As an example of the cycle usage, let us consider the following program for
calculating the factorial of number 6:

Sub Factorial1()
Dim I As Byte
Dim F As Long
F = 1

58
 1.13. Cycles

For I = 1 To 6 Step 1
F = F * I
Next I
End Sub

According to handbook [3], factorial of natural number n (n! is the designa-

tion) is the product of the positive integers from 1 to n: n! 1 2 ... n ( 1! 1 ).
We advise the reader to do the following:
1) execute the Factorial1 program step-by-step, watching the F varia-
ble’s value;
2) replace 6 (parameter ending) by 13 in the cycle operator;
3) run the resulting Factorial1 program for calculating 13!;
4) explain the reason of the stop with information Run-time error ‘6’: Over-
flow, using the description of the Long data type in Appendix 1.
At a negative value of the growth parameter, the For…Next cycle works
as at a positive value, but:
condition counter < ending is being checked;
the counter variable cannot be of the Byte data type.
The following 2nd version of the program for calculating 6! is an example of
using a negative value of the growth parameter.

Sub Factorial2()
Dim I As Integer
Dim F As Long
F = 1
For I = 6 To 1 Step -1
F = F * I
Next I
End Sub

If the Step keyword is omitted, the step of the counter change is equal to
unity by default.
Arithmetic expressions may be used as the cycle parameters (beginning,
ending and growth). It is important that all variables in these arithmetic
expressions had numerical values before the For…Next cycle work.
As an example of such usage of arithmetic expressions, let us consider the
following 3rd version of the program for calculating 6!:

Sub Factorial3()
Const e As Double = 2.718281828
Dim J As Byte

59
 Chapter 1. Programming in Visual Basic

Dim
N As Byte
Dim
F As Long
N =
Round(e)
F =
1
For
J = 1 To N ^ 2 - 3
F = F * J
Next J
End Sub

In this program, arithmetic expression N ^ 2 - 3 is used as parameter ending.

The N variable is equal to 3, and N ^ 2 - 3 is equal to 6.
The absence of the Step keyword says that the step of the J variable change
is equal to 1, and J changes from 1 to 6.
Frequently it is required to leave the cycle before completion of its execution.
In this case, the For…Next cycle has the following syntax:

For counter = beginning To ending [Step growth]

[statements1]
If condition Then Exit For
[statements2]
Next [counter]

The Exit For operator is used for immediate exit from the cycle. It is a part
of the simplest conditional operator. The last cycle works as follows.
For each value of the counter variable, after executing the statements1
block, the computer calculates the value of logical expression condition.
If this value is False, the cycle continues to work. Otherwise, the jump is per-
formed to the operator following the cycle construct (without executing block
statements2).
As an example of using the Exit For operator, let us consider the following
4th version of the program for calculating 6!:

Sub Factorial4()
Dim I As Byte
Dim F As Long
F = 1
For I = 1 To 13
F = F * I
If I = 6 Then Exit For
Next I
End Sub

60
 1.13. Cycles

The While…Wend cycle is used when the number of the block’s executions
is not known in advance. The syntax of this cycle follows:

While condition
statements
Wend

The While…Wend cycle work begins with calculating the value of logical
expression condition. If condition = False, the cycle is completed, i.e.,
the jump is performed to the operator following the Wend keyword.
If condition = True, block statements is executed. After that, the value
of logical expression condition is calculated again, and so on.
The 5th version of the program for calculating 6! follows:

Sub Factorial5()
Dim I As Byte
Dim F As Long
F = 1
I = 1
While I <= 6
F = F * I
I = I + 1
Wend
End Sub

The Do…Loop cycle, as well as the While…Wend cycle, is used when the
number of the block’s executions is not known in advance. Four versions of this
construct exist.
The first version is the Do While…Loop cycle with the following syntax:

Do While condition
statements
Loop

The Do While…Loop cycle work begins with calculating the value of logi-
cal expression condition. If condition = False, the cycle is completed,
i.e., the jump is performed to the operator following the Loop keyword. If
condition = True, block statements is executed. After that, the value of
logical expression condition is calculated again, and so on.
The Do While…Loop cycle is equivalent to the While…Wend cycle
reviewed above.

61
 Chapter 1. Programming in Visual Basic

The 6th version of the program for calculating 6! follows:

Sub Factorial6()
Dim I As Byte
Dim F As Long
F = 1
I = 1
Do While I <= 6
F = F * I
I = I + 1
Loop
End Sub

One more example of using the Do While…Loop cycle is the following

program for the movement along the x axis:

Sub Steps1()
Dim x As Single
Dim h As Single
h = 0.5 'step equals 0.5
x = 44 'initial value of x equals 44
Do While x < 55 'final value of x equals 55
x = x + h 'value of x increases by h
Loop
End Sub

We advise the reader to make the step-by-step execution of the Steps1

program, watching the x variable change.
The Do Until…Loop cycle, which is the second version of the Do…Loop
construct, has the following syntax:

Do Until condition
statements
Loop

The Do Until…Loop cycle work begins with calculating the value of logi-
cal expression condition. If condition = True, the cycle is completed,
i.e., the jump is performed to the operator following the Loop keyword.
If condition = False, block statements is executed. After that, the value
of logical expression condition is calculated again, and so on.

62
 1.13. Cycles

The program with the Do Until…Loop cycle for the movement along the x
axis has the following form:

Sub Steps2()
Dim x As Single
Dim h As Single
h = 0.5 'step equals 0.5
x = 44 'initial value of x equals 44
Do Until x >= 55 'final value of x equals 55
x = x + h 'value of x increases by h
Loop
End Sub

There is a situation when, during the work of the Do While…Loop and

Do Until…Loop cycles, block statements is not executed even once
because the condition of completing the cycle is checked before the block execu-
tion. Sometimes it is inconvenient.
The Do…Loop While cycle, which is the third version of the Do…Loop
construct, has the following syntax:

Do
statements
Loop While condition

The Do…Loop While cycle work begins with executing operator block
statements. After that, the value of logical expression condition is
calculated. If condition = False, the cycle is completed. Otherwise, block
statements is executed again, and so on.
The program with the Do…Loop While cycle for the movement along the x
axis has the following form:

Sub Steps3()
Dim x As Single
Dim h As Single
h = 0.5 'step equals 0.5
x = 44 'initial value of x equals 44
Do
x = x + h 'value of x increases by h
Loop While x < 55 'final value of x equals 55
End Sub

63
 Chapter 1. Programming in Visual Basic

The Do…Loop Until cycle, which is the fourth version of the Do…Loop
construct, has the following syntax:

Do
statements
Loop Until condition

The Do…Loop Until cycle work begins with executing operator block
statements. After that, the value of logical expression condition is
calculated. If condition = True, the cycle is completed. Otherwise, block
statements is executed again, and so on.
The program with the Do…Loop Until cycle for the movement along the x
axis has the following form:

Sub Steps4()
Dim x As Single
Dim h As Single
h = 0.5 'step equals 0.5
x = 44 'initial value of x equals 44
Do
x = x + h 'value of x increases by h
Loop Until x >= 55 'final value of x equals 55
End Sub

During the work of the Do…Loop While and Do…Loop Until cycles,
block statements is executed at least once because the condition of complet-
ing the cycle is checked after the block execution.
All four versions of the Do…Loop cycle can contain the Exit Do operator
intended for immediate exit from the cycle. In the usage, this operator is similar
to the Exit For operator of the For…Next cycle.
For the While…Wend cycle, there is no operator similar to the Exit For
and Exit Do operators.

64
 1.14. Manifestation of the error of real numbers’ computer representation

1.14. Manifestation of the error of real

numbers’ computer representation

In the last program of the previous section, we will reduce the value of h to
one-fifth. We can expect that the number of repeated executions of operator
x = x + h will increase fivefold as a result and the final value of x will remain
equal to 55. However, it is not so: the final value of x is equal to 55.09983.
In order to verify this assertion, we do the following:
1) enter program

Sub Steps5()
Dim x As Single
Dim h As Single
h = 0.1 'this operator distinguishes
'Steps5 from Steps4
x = 44
Do
x = x + h
Loop Until x >= 55
End Sub

into the code window;

2) install the breakpoint against the End Sub line;
3) click on arrow ► for the program execution up to the breakpoint;
4) place the mouse pointer on the x variable in the program text; as a result,
x = 55.09983 appears (Fig. 1.9);
5) click on arrow ► for terminating the program execution.
We see “the phenomenon” of changing the final value of x after reducing the
value of h because “almost all” real numbers in the computer are represented
with an error.
In our case, the value of variable h of the Single data type is slightly less
than 0.1. Therefore, there is the “superfluous” execution of the x = x + h
operator during the cycle. This explains the difference between the observed
value of 55.09983 and the expected value of 55.

65
 Chapter 1. Programming in Visual Basic

Fig. 1.9. The code window with the Steps5

program and with the final value of x

Let us replace Single by Currency in the declaration of the h variable.

The new version of the program follows:

Sub Steps6()
Dim x As Single
Dim h As Currency 'this operator distinguishes
'Steps6 from Steps5
h = 0.1
x = 44
Do
x = x + h
Loop Until x >= 55
End Sub

The final value of x, calculated by means of the Steps6 program, is equal

to 55 (Fig. 1.10). This is because the value of variable h of the Currency data
type is exactly equal to 0.1, therefore, there is not the superfluous execution of
the x = x + h operator during the cycle.
According to what has been said above, when working with real numbers, we
have to consider possible inaccuracy of their computer representation because
the error can lead to unexpected results.

66
1.14. Manifestation of the error of real numbers’ computer representation

Fig. 1.10. The code window with the Steps6

program and with the final value of x

67
 Chapter 1. Programming in Visual Basic

1.15. Arrays

An array is a sequence or table (two-dimensional, three-dimensional, and so

on) with variables of the same data type, which are called elements of the array.
Every reference to an element includes the array name and one index (the ele-
ment number in the sequence) or several indices (coordinates of the element in
the table). If there are two indices, they determine the coordinates of the element
in the two-dimensional table, i.e., the numbers of row and column whose inter-
section is the element.
Before usage, the array must be declared. Besides, we have to specify the
lower and upper boundaries of each index’s change.
There are two kinds of arrays — static and dynamic. For each of them,
we will use the following scheme: we will consider the one-dimensional array
(with one index) followed by the multidimensional array (with several indices)
together with the one-dimensional array.
The static array is used when the quantity of its elements is known in
advance. In the declaration of a static one-dimensional array, for the index, we
have to specify the lower and upper boundaries defining the quantity of ele-
ments, and these boundaries cannot be changed during the program execution.
Static arrays are declared as variables, i.e., by means of keywords Dim and
As. The boundaries are integers in parentheses. The To keyword must be
between the lower and upper boundaries. Examples of the declaration follow:

Dim arrB(1 To 10) As Currency

Dim A(-10 To 10) As String

If a single integer is given in the parentheses, it is the upper boundary. In this

case, the lower boundary is equal to zero by default. For example, operator

Dim arrA(9) As Byte

is equivalent to operator

Dim arrA(0 To 9) As Byte

68
 1.15. Arrays

If the lower boundary of indices must be equal to unity by default, we have

to type

Option Base 1

above the first line of the program. In this case, operator

Dim arrA(9) As Byte

is equivalent to operator

Dim arrA(1 To 9) As Byte

Further, we will consider that line Option Base 1 is absent.

The values of the boundaries should be between the limits for the Long data
type (Appendix 1), i.e., from -2147483648 to 2147483647.
Let us consider the following program:

Sub StaticArrays()
Dim B1(1 To 6) As Byte, S1 As Byte
Dim B2(1 To 6) As Currency, S2 As Currency
Dim B3(1 To 6) As Byte, S3 As Byte
Dim I As Byte
'Determination of first five elements of arrays:
For I = 1 To 5 'first cycle
B1(I) = I 'first power of I
B2(I) = I ^ .333 'cubic root of I
B3(I) = I ^ 3 'I cubed
Next
'Determination of sixth elements of arrays:
S1 = 0
S2 = 0
S3 = 0
For I = 1 To 5 'second cycle
S1 = S1 + B1(I)
S2 = S2 + B2(I)
S3 = S3 + B3(I)
Next
B1(6) = S1
B2(6) = S2
B3(6) = S3
End Sub

69
 Chapter 1. Programming in Visual Basic

In this program:
the first cycle is used for determination of the first five elements of arrays
B1, B2 and B3; upon completing the cycle, these elements of the arrays contain
the numerical values, which are equal to the 1st, 1st / 3rd and 3rd powers of I:
B1(1) = 1, B1(2) = 2, …, B1(5) = 5; B2(1) = 1, B2(2) = 1.2596, …,
B2(5) = 1.7091; B3(1) = 1, B3(2) = 8, …, B3(5) = 125;
the second cycle is used for summation of the earlier determined elements
of the arrays; upon completing the cycle, the sums of the first five elements of
arrays B1, B2 and B3 are respectively equal to S1 = 15, S2 = 6.9971 and
S3 = 225;
the remaining operators of the program assign the calculated values of the
sums to the sixth elements of the arrays: B1(6) = 15, B2(6) = 6.9971 and
B3(6) = 225.
It was noted above that one-dimensional and multidimensional arrays exist.
We reviewed one-dimensional arrays, which are similar to rows and columns on
the Excel worksheet and to vectors in mathematics.
For declaration of multidimensional arrays (with several indices), we use
a construct similar to operator Dim for one-dimensional arrays. The difference
consists in that several boundaries are given through a comma.
For example, operators

Dim A(4, 6) As Byte

Dim B(1 To 5, -7 To -1) As Byte

declare two-dimensional arrays A and B, which contain identical quantities of

elements. This quantity is equal to 5 × 7 = 35.
A two-dimensional array is similar to a range of cells on the Excel worksheet
and to a matrix in mathematics. A three-dimensional array is similar to a range of
cells on several worksheets of the same Excel workbook. A four-dimensional
array is similar to a range of cells on several worksheets of several open work-
books.
Open Excel workbooks are represented by buttons on the taskbar of Win-
dows Desktop.
The reference to an element of a multidimensional array includes the array
name and the indices listed through a comma. Examples of the reference are
figured in the following assignment operators:

A(i, j + 1) = 17
D(K) = A(i, 0)

As we see, the reference may be on the left and/or right of the assignment sign.

70
 1.15. Arrays

Operator

Dim C(1 To 5, -5 To -1, 4) As Byte

declares a three-dimensional array containing 5 × 5 × 5 = 125 elements.

The number of indices is called the dimension of an array. The dimension of
the above C array is equal to 3. The largest dimension equals 60.
An array occupies S × Q + 4 × D + 20 bytes of the main memory, where:
S is the memory size (in bytes) occupied by one element;
Q is the quantity of the elements;
D is the dimension.
It concerns both static and dynamic arrays. The last array is reviewed below.
The dynamic array is used when the quantity of its elements is not known in
advance and must be defined during the program execution. When finishing
work with the dynamic array, it is possible to free the memory cells occupied by
this array. It is important for problems demanding large size of the main
memory.
The declaration of the dynamic array has the following two parts.
1. The array is declared by means of the Dim operator without boundaries of
indices. A pair of parentheses must follow the array name.
2. The boundaries of indices are specified by means of the ReDim operator
in a proper place of the program, at that, we can use not only integers as bounda-
ries, but also arithmetic expressions. It is important that all variables in these
expressions have numerical values before the ReDim operator execution.
The following program is an example of using the dynamic array.

Sub DynamicArray()
Dim A() As Byte 'declaration of array
Dim M As Integer, N As Integer
M = 3
ReDim A(-5 To M ^ 2) 'specification of boundaries
For N = -5 To M ^ 2
A(N) = N + 30
Next
ReDim A(5) 'specification of boundaries
N = 0
Do
A(N) = N ^ 3
N = N + 1
Loop Until N ^ 2 > 10
End Sub

71
 Chapter 1. Programming in Visual Basic

After typing this program in the code window, let us do the following.
1. By clicking, set the blinking cursor on variable A, more precisely, in front
of or behind A. Fulfill operations Debug > Add Watch. The Add Watch window,
containing the variable name in text box Expression, appears.
2. Click on OK. The Watches window, intended for the current visualization,
appears with a line corresponding to the A array.
3. Set the blinking cursor in any place of the program text.
4. Execute the program step-by-step (Fig. 1.11), watching the A array values
by means of the Watches window.

Fig. 1.11. The step-by-step execution of the DynamicArray program

For visualization of the A array values in the Watches window, we must click
on the plus sign in front of A in this window. As a result, plus turns to minus and
the A array values appear below.
We can edit the contents of the Watches window.
Let us admit that we want to watch the value of variable n instead of the A
array values. For this purpose, during the step-by-step program execution
(between presses of the F8 key), we must:

72
 1.15. Arrays

1) click on the minus sign in front of A; as a result, the plus sign appears but
the A array values disappear;
2) highlight A by clicking on it (in the Watches window);
3) type n instead of A highlighted;
4) press the Enter key;
5) continue the step-by-step program execution, watching the value of n.
To remove the line, corresponding to the A array, from the Watches window,
we must:
1) highlight this line by clicking on the glasses pictogram;
2) press the Delete key.
To add a line, we must fulfill Debug > Add Watch, and so on.
To close the Watches window, we must click on the little cross in the top
right corner of this window. To open the Watches window, we must fulfill
View > Watch Window.
The ReDim operator can be used for changing the array’s dimension, as in
the following program:

Sub Dimension()
Dim arrA() As Byte
ReDim arrA(1, 1) 'two-dimensional array
arrA(0, 0) = 13: arrA(0, 1) = 14
arrA(1, 0) = 15: arrA(1, 1) = 16
ReDim arrA(1 To 3, 3, 3) 'three-dimensional array
arrA(1, 0, 0) = 17
End Sub

Note that, at repeated execution of the ReDim operator, the array values will
be lost because the ReDim operator nulls all elements of the array.
To keep the array values, we have to insert the Preserve keyword between
ReDim and the array name. As an example, see operators 0, 1 and 2 in the fol-
lowing program:

Sub Conservation1()
Dim J As Integer
0: Dim arrA() As Integer 'declaration of array
1: ReDim arrA(-5 To 1) 'specification of boundaries
For J = -5 To 1
arrA(J) = J ^ 2
Next J
2: ReDim Preserve arrA(-5 To 4)
'specification of boundaries

73
 Chapter 1. Programming in Visual Basic

For J = 2 To 4
arrA(J) = J ^ 3
Next J
End Sub

Labels 0, 1 and 2 may be omitted.

We advise the reader to execute the Conservation1 program step-by-
step, watching the arrA array values by means of the Watches window.
In the Conservation1 program, the Preserve keyword is used to
change the upper boundary of the one-dimensional array’s index without nulling
the array values. When changing the lower boundary, we cannot use this key-
word.
In the case of a multidimensional array, the Preserve keyword can be used
only when changing the upper boundary of the last index. As an example of such
usage of the Preserve keyword, see operators 0, 1 and 2 in the following pro-
gram:

Sub Conservation2()
Dim I As Integer, J As Integer
0: Dim arrA() As Integer 'declaration of array
1: ReDim arrA(2, -5 To 1)
'specification of boundaries
For I = 0 To 2
For J = -5 To 1
arrA(I, J) = (I + 1) * J ^ 2
Next J
Next I
2: ReDim Preserve arrA(2,-5 To 4)
'specification of boundaries
For I = 0 To 2
For J = 2 To 4
arrA(I, J) = (I + 1) * J ^ 3
Next J
Next I
End Sub

We advise the reader to make the step-by-step execution of the last program,
watching the arrA array values by means of the Watches window.
The values of the lower and upper boundaries of any index can be returned to
the program. For this purpose, functions LBound and UBound are respectively
used. We can look the description of these functions in the Excel help system
started by pressing the F1 key when the VB window is active.

74
 1.15. Arrays

If arrA is a one-dimensional array, the LBound and UBound functions are

used as in the operator block below:

Dim lower As Long, upper As Long

lower = LBound(arrA) 'lower boundary of index
upper = UBound(arrA) 'upper boundary of index

Function UBound is necessary, for example, when the value of the upper
boundary is unknown and, at the same time, we have to increase this value by
certain number.
As it was already told, advantage of the dynamic array over the static is that
we can free the memory cells, earlier occupied by the dynamic array. Operator
Erase is used for this, as in the following example program:

Sub Memory()
Dim A() As Byte
Dim B() As Byte
ReDim A(8) 'memory for A: 9 + 4 + 20 = 33 bytes
Erase A 'memory for A: 0 bytes
ReDim B(2, 3)
'memory for B: 12 + 8 + 20 = 40 bytes
End Sub

After inputting the Memory program into the code window, let us make the
following:
1) generate the lines corresponding to arrays A and B in the Watches win-
dow, fulfilling Debug > Add Watch twice;
2) click on any place of the program text to set the blinking cursor there;
3) execute the program step-by-step, watching the memory distribution by
means of the Watches window.
For solving a series of mathematical problems, arrays of random numbers are
required. They can be generated by means of the Rnd function (p. 46).
Let us consider the following program:

Sub RandomNumbers()
Dim N As Long
Dim I As Long
Dim S() As Single
N = 20
ReDim S(1 To N)

75
 Chapter 1. Programming in Visual Basic

Randomize 'it must be before calls of Rnd

For I = 1 To N
S(I) = Rnd
Next I
End Sub

This program calculates 20 random real numbers from 0 to 1 and writes them
into dynamic array S. The Rnd function is being called in cycle For…Next.
Function Rnd is the built-in generator of random real numbers uniformly
distributed on segment [0, 1] : a number, being returned by the Rnd function,
appears in any place of segment [0, 1] with equal probability.
Before a series of the Rnd function calls, the Randomize operator must be
executed for preparing the random-number generator for work.
We advise the reader to execute the RandomNumbers program step-by-
step, watching the S array by means of the Watches window.
Operator Randomize and function Rnd are used in code Listing 6.11
intended for minimization of the multimodal function (Section 6.8).

76
 1.16. User-defined procedures

1.16. User-defined procedures

A user-defined procedure allows executing an operator block in different

places of our program.
According to this VB construct:
1) we have to write the block, intended for multiple executions, only once;
2) a name with formal parameters or without them must be written in the
header of the block.
This expanded block is called a procedure declaration.
Generally, the user-defined procedure has:
input parameters, which are considered given;
output parameters, which are determined while executing the block.
After forming the procedure declaration, we locate the calls of this procedure
with the actual parameters in those places of the program, which should be occu-
pied by the block being the procedure prototype.
The procedures are divided into functions and subroutines (subprograms).
The function, returning a value, is intended for usage in arithmetic and logical
expressions and in strings. The subroutine usage in expressions and strings is
impossible. This is the main difference between the function and subroutine.
The declaration of the user-defined function has the following syntax:

Function name([formal_parameters]) [As type]

statements
End Function

where name is the function name, type is the data type (Appendix 1) of
name, i.e., of the function value, formal_parameters are the parameter
(argument) names listed through a comma, statements is the operator block.
The parameter names can be accompanied by keywords.
Block statements must include at least one assignment operator whose
left-hand side (on the left of sign =) is the function name.
The function call looks like

name([actual_parameters])

77
 Chapter 1. Programming in Visual Basic

where actual_parameters are variables, arrays, expressions (arithmetic,

logical) and/or strings listed through a comma.
As a result of calling the name function, the function value, corresponding to
actual_parameters, is returned into the program.
Records and arrays of records may be among formal and actual parameters.
We will review the record construct in Section 1.18.
Let us consider the following code:

Sub Program1()
Dim L As Long
Dim W As Double
L = Fact(12)
W = 4.2 + Fact(10) / 2
End Sub

Function Fact(N) As Long 'N < 13

Dim I As Byte
Dim J As Long
J = 1
For I = 1 To N
J = J * I
Next I
Fact = J
End Function

The first group of operators is program Program1; the second group is the
declaration of function Fact. They are in one or different modules of the same
Excel workbook, i.e., we can type the program and the function declaration in
one or different code windows.
The Fact function calculates N!, i.e., factorial of natural number N, which is
the formal parameter.
We see two calls of the Fact function in the program, with 12 and 10 as the
actual parameter.
1. The Fact function call is located in the right-hand side of assignment
operator

L = Fact(12)

As a result of the operator execution, the Fact function value (that is, the value
returned by the Fact function into the program when N is equal to 12) is
assigned to the L variable.

78
 1.16. User-defined procedures

2. The Fact function call is figured in arithmetic expression

4.2 + Fact(10) / 2

The value of this arithmetic expression is assigned to the W variable.

The declaration of the user-defined subroutine has the following syntax:

Sub name([formal_parameters])
statements
End Sub

where name is the subroutine name, formal_parameters are the parameter

names listed through a comma, as in the above declaration of the user-defined
function, statements is the operator block.
Keyword Sub occurs from word “subroutine”. It is in the beginning and end
not only of the subroutine declaration (or simply of the subroutine), but also of
the program because “main subroutine” is synonym of “program”.
There are two equivalent operators of calling the subroutine:

Call name([actual_parameters])
name [actual_parameters]

where actual_parameters is the list of actual parameters, as in the above

call of the user-defined function. In the presence of the Call keyword, the
actual parameters are in parentheses; in the absence of Call, the parentheses are
not used.
The example program and subroutine follow:

Sub Program2()
Dim aa As Single, bb As Single
Dim cc1 As Single, cc2 As Single, cc3 As Single
aa = 3
bb = 4
Call Hypotenuse(aa, bb, cc1)
'1st call of subroutine
Call Hypotenuse(3, 4, cc2)
'2nd call of subroutine
Hypotenuse aa, bb, cc3
'3rd call of subroutine
End Sub

79
 Chapter 1. Programming in Visual Basic

Sub Hypotenuse(ByVal A, ByVal B, ByRef C)

C = Sqr(A ^ 2 + B ^ 2)
End Sub

The first group of operators is program Program2, the second group is the
declaration of subroutine Hypotenuse for calculating the length of the hypote-
nuse of a right-angled triangle. They are located in one or different modules.
We see three operators of calling the Hypotenuse subroutine in program
Program2, and two of them contain the Call keyword.
Formal parameters A and B (in the subroutine declaration) are the input pa-
rameters, lengths of the legs. The ByVal keyword in front of A and B in the first
line of the subroutine declaration (i.e., in the header of the operator block) means
that these parameters must be passed by value (when calling Hypotenuse). In
this case, the values of parameters A and B (3 and 4, respectively) are transferred
to the Hypotenuse subroutine at all three calls.
Formal parameter C is the output parameter, i.e., the hypotenuse length. The
ByRef keyword in front of C means that this parameter must be passed by
reference (when calling Hypotenuse). In this case, the address of the memory
cell, corresponding to variable cc1 (at the first call), cc2 (at the second call)
or cc3 (at the third call), is transferred to the Hypotenuse subroutine.
Keyword ByRef may be omitted.
In the code under consideration, the parameters of the Hypotenuse subroutine
are simple variables (not arrays). If the parameters are arrays, both input and
output parameters must be passed by reference.
Let us consider the following code located in one module:

Dim N1 As Integer

Sub Program3()
Dim xx(50) As Double
Dim yy(50) As Double
Dim i As Integer
N1 = 3
For i = N1 To 30
xx(i) = 0.1 * i
Next i
Call XSINX(30, xx, yy) 'call of subroutine
End Sub

Sub XSINX(ByVal N2, ByRef X() As Double, _

ByRef F() As Double)

80
 1.16. User-defined procedures

Dim j As Integer
For j = N1 To N2
F(j) = X(j) * Sin(X(j))
Next j
End Sub

In addition to usage of arrays as the subroutine parameters, we see something

new in the last code: the N1 variable is declared above the first line of the pro-
gram. This is done for making N1 visible (in respect of the usage possibility) in
both the program and subroutine. Let us return to the variable declaration.
So far, we have been speaking about how to declare variables, but have not
mentioned where to declare. We can declare them in two places:
inside the program or user-defined procedure;
in the general declarations area occupying the top part of the code
window.
The place of the variable declaration defines the area of the variable usage.
For example, if a variable is declared in the user-defined procedure (as variable j
in the last code), only this procedure “sees” this variable. Other procedures
(if they exist) and the program cannot use this variable’s value and change it.
Such variable is called a local variable. We can also say that the variable is
visible at the procedure level.
For letting a variable’s value be accessible to all user-defined procedures of
the given module, we have to declare this variable in the general declarations
area (as variable N1 in the last code). In this case, the program and all user-
defined procedures, declared in the given module, can use this variable’s value
and change it.
Such variable is called a module variable. The Dim keyword in front of N1
may be replaced by the Private keyword:

Private N1 As Integer

All that was said about variables also concerns the user-defined constants, but
the constant’s value, naturally, cannot be changed.
If the declaration of the XSINX subroutine is located in the separate module,
the first line of the last code should be as follows:

Public N1 As Integer

Such variable is called a public variable. The declaration of public variable N1

can be in each of two modules or only in one, the first (with text of Program3)
or second (with text of XSINX).

81
 Chapter 1. Programming in Visual Basic

At the similar declaration of a constant, instead of a variable, the correspond-

ing line of the general declarations area should begin with keyword combination
Public Const.
The debugger command, being fulfilled by pressing the F8 key, is called Step
Into: during the step-by-step program execution by means of the F8 key, the
entrance into the user-defined procedure takes place. If we do not need the step-
by-step execution inside the user-defined procedure, we use the following two
commands of the Debug menu.
1. Step Over — the step-by-step program execution without entrance into
the user-defined procedure. This command can also be performed by pressing
Shift + F8.
2. Step Out — exit from the user-defined procedure. It is used when the pro-
cedure remainder should be executed in the automatic mode. This command can
also be performed by pressing Ctrl + Shift + F8.
We advise the reader to execute programs Program1, Program2 and
Program3 by pressing F8, Ctrl + F8, Shift + F8 and Ctrl + Shift + F8. Before
the execution, the blinking cursor must be located inside the program text (not in
the procedure declaration and not in the general declarations area).
One or several last parameters of the list of parameters of the user-defined
procedure (function or subroutine) can be optional, i.e., they can be absent in the
procedure call.
If the parameter is not obligatory, the Optional keyword must be in front
of this parameter’s name in the first line of the procedure declaration. The
optional parameter must have the Variant data type (Appendix 1).
Let us consider the following example code:

Sub Program4()
Dim bytA As Byte, bytB As Byte
Dim intC As Integer
bytA = 5
bytB = 10
intC = Apt(bytA, bytB)
'result: bytB = 6, intC = 625
bytB = 10
intC = Apt(bytA) 'result: bytB = 10, intC = 625
End Sub

Function Apt(ByVal a, Optional b As Variant)

If Not IsMissing(b) Then b = a + 1
Apt = a ^ 4
End Function

82
 1.16. User-defined procedures

In this example, the b parameter of the Apt function is optional; keyword

Optional in front of this parameter tells about it. Keyword combination
As Variant behind b may be omitted.
Other features of the Apt function declaration:
Not is the logical negation;
IsMissing is the following function.
The value of IsMissing(b) is equal to True or False as follows:
True in the absence of the optional parameter in the Apt function call;
False in the presence of the optional parameter.
We advise the reader to execute the Program4 program step-by-step,
watching the values of variables bytB and intC.
We can specify a value of the optional parameter in the absence of this
parameter in the call of the user-defined procedure.
The example program and procedure follow:

Sub Program5()
Dim bytA As Byte, bytB As Byte, bytC As Byte
bytA = 5: bytB = 2
Call Ept(bytA, bytC, 5 * bytB) 'result: bytC = 15
Call Ept(bytA, bytC) 'result: bytC = 8
End Sub

Sub Ept(ByVal a, c, Optional b = 3)

c = a + b
End Sub

In this code:
the b parameter of the Ept subroutine is optional;
if the call of Ept contains only two actual parameters, b = 3 is used when
executing operator c = a + b.
We advise the reader to make the step-by-step execution of the Program5
program, watching the value of bytC.
Operators

Exit Sub
Exit Function

are intended for immediate terminating the procedure execution. The first opera-
tor should be in the subroutine declaration, the second — in the function declara-
tion. At their execution, the jump is being performed to the same point, as upon
the normal terminating the procedure execution.

83
 Chapter 1. Programming in Visual Basic

The programs in this section use only one user-defined procedure; for exam-
ple, program Program1 uses only the Fact function. However, several user-
defined procedures may be used by one program.
According to the computer terminology, as a program, we can consider the
program itself (the main subroutine) together with the declarations of the user-
defined procedures (used by the program) and/or together with the general decla-
rations area. For example, program Program3 together with operator

Dim N1 As Integer

can also be called program Program3. The following can also be called pro-
gram Program3: the program itself, the XSINX subroutine declaration and the
general declarations area. Sometimes term “program” is used in the extended
sense, which is equivalent to “code”.

84
 1.17. Built-in procedures. Usage of standard windows

1.17. Built-in procedures. Usage of standard windows

Visual Basic includes a considerable quantity of the built-in procedures,

which differ from the user-defined procedures in the following: developers of the
Visual Basic features programmed their declarations. These declarations are hid-
den from us as the program developer.
The built-in procedures, as well as the user-defined procedures, are divided
into functions and subroutines.
We have already encountered the built-in functions of VB. They are:
functions reviewed in Section 1.9;
IIf in Section 1.12;
IsMissing in Section 1.16, etc.
Below is considered built-in function InputBox intended for input of in-
formation (into the program) by means of the standard windows of operating
system Windows.
An example of the built-in subroutine is the MsgBox procedure intended for
output of information (from the program) into the standard windows.
We will use InputBox and MsgBox in the program from Section 1.1, in-
tended for calculating the hypotenuse length, with which we began studying VB.
1. Let us enter the following program into the code window:
Listing 1.3
Sub Pythagoras1()
Dim a As Single
Dim b As Single
Dim c As Single
Dim s As String
1: s = InputBox( _
"Enter length of the first leg and click OK")
2: a = Val(s)
3: s = InputBox( _
"Enter length of the second leg and click OK")
4: b = Val(s)
5: c = Sqr(a ^ 2 + b ^ 2) 'according to Pythagoras
6: s = Str(c)

85
 Chapter 1. Programming in Visual Basic

7: MsgBox s
End Sub

2. Let us run the Pythagoras1 program. The window appears (Fig. 1.12),
offering to input the length of the first leg of a right-angled triangle. Let us put,
for example, 400 (without inverted commas) into the text box of this window by
means of the keyboard and click on the OK button.

Fig. 1.12. The first window (on the Excel worksheet)

for inputting the source data

3. The window appears (Fig. 1.13), offering to input the length of the second
leg. Let us put, for example, 300 and click on OK.

Fig. 1.13. The second window for inputting the source data

4. The window appears (Fig. 1.14), containing the hypotenuse length calcu-
lated. Let us click on the OK button to close this window and terminate the pro-
gram execution.

86
 1.17. Built-in procedures. Usage of standard windows

Fig. 1.14. The window with the result

In the above program, operator 1 includes the InputBox function call,

which is used for entering information from the keyboard. This function returns
the string, which was entered into the text box of the window in Fig. 1.12, i.e.,
"400". This string is assigned to the s variable of the String data type.
Operator 2 converts the s string’s value to number 400 and assigns this
number to the a variable of the Single data type.
Operator 3 assigns string "300", which was entered into the text box of the
window in Fig. 1.13, to the s variable. Operator 4 converts the s string’s value
to number 300 and assigns this number to the b variable.
Operator 5 calculates the hypotenuse length according to the Pythagoras the-
orem. The calculated value of 500 is assigned to variable c. Operator 6 converts
500 to string; as a result, the s string’s value becomes "500".
Operator 7, which is the call of the built-in MsgBox subroutine, opens the
window with value 500 calculated (Fig. 1.14). As in the case of the user-defined
subroutine, operator 7 may be written in the following form:

Call MsgBox(s)

The calls of built-in procedures InputBox and MsgBox have only one
parameter (of the String data type), which is obligatory. However, these pro-
cedures also have optional parameters, whose appointment may be looked in the
Excel help system started by pressing the F1 key when the VB window is active.
Before pressing this key, we recommend to locate the blinking cursor on the
required word (InputBox or MsgBox) in the code window containing program
Pythagoras1.
In the Excel help system, the MsgBox procedure is termed as a function,
instead of a subroutine, because the call of MsgBox may be a part of arithmetic

87
 Chapter 1. Programming in Visual Basic

expressions. In this case, MsgBox returns (into the program) the integer value
depending on the button, on which the user clicked. In the example below, we
will consider using the MsgBox procedure as a function.
For calculating the area of a right-angled triangle by the Pythagoras1
program, we replace operator 7 by three operators labeled by 7, 8 and 9. The
following program is the result:

Sub Pythagoras2()
Dim a As Single
Dim b As Single
Dim c As Single
Dim s As String
1: s = InputBox( _
"Enter length of the first leg and click OK")
2: a = Val(s)
3: s = InputBox( _
"Enter length of the second leg and click OK")
4: b = Val(s)
5: c = Sqr(a ^ 2 + b ^ 2) 'according to Pythagoras
6: s = Str(c)
7: Dim Ret As Integer
8: Ret = MsgBox(s, vbYesNo, _
"Do you want to calculate area?")
9: If Ret = vbYes Then MsgBox Str(a * b / 2)
End Sub

In operator 8, procedure MsgBox is a function. In this case, the procedure

parameters are in parentheses.
When executing the Pythagoras2 program, the window with the hypote-
nuse length calculated has other content (Fig. 1.15). After clicking on the Yes
button, the window with the triangle area calculated is displayed (Fig. 1.16).
By clicking on the OK button, we terminate the program execution.
Operators 7 — 9 of the last program may be replaced by operator

If MsgBox(s, vbYesNo, _
"Do you want to calculate area?") = vbYes _
Then MsgBox Str(a * b / 2)

This also gives the results shown in Fig. 1.15 and 1.16.

88
1.17. Built-in procedures. Usage of standard windows

Fig. 1.15. The first resulting window

Fig. 1.16. The second resulting window

89
 Chapter 1. Programming in Visual Basic

1.18. Records

A record or user-defined data type is a collection of variables, possibly of dif-

ferent data types, grouped together under a single name for convenient handling.
Each variable of the record is called a field.
Before usage of a record, we must create it by means of the Type operator
with the following syntax:

Type struct
field1 As type1
field2 As type2
∙ ∙ ∙ ∙ ∙
fieldN As typeN
End Type

where struct is a name of the record, field1, field2, …, fieldN are the
field names, type1, type2, …, typeN are data types (Appendix 1) for the
corresponding fields.
The record creation operator must be placed in the general declarations area
of the program, which uses this record.
For example, we have to work up results of a university session, where the
following three subjects were evaluated: physics, mathematics and informatics.
In this case, the following fields are necessary:
1) name of the student;
2) number of the test book;
3) mark in physics;
4) mark in mathematics;
5) mark in informatics.
Let the name of the record be Session, and let the field names be Name,
Number, Physics, Math and Inform. In this case, the record creation opera-
tor has the following form:

Type Session
Name As String
Number As Long

90
 1.18. Records

Physics As Byte
Math As Byte
Inform As Byte
End Type

To get access to the created record (by name struct), we have to declare
one or several variables in the program by means of the Dim operator, in the
same way as we declared variables in Section 1.3.
The declaration operator has the following syntax:

Dim variable As struct

where variable is the variable name, struct is the variable’s data type.
For example, operator

Dim Sess As Session, BestSess As Session

declares variables (records) Sess and BestSess of the Session data type.
For the reference to the record field, we use the variable and field names
separated by a point.
For example, assignment operator

Sess.Name = "Maksim Zakharkin"

contains the reference in the left-hand side. As a result of the operator execution,
string

"Maksim Zakharkin"

is assigned to the Name field of the Sess variable (of type Session).
For filling the fields, it is convenient to use the With operator, which has the
following syntax:

With variable
.field1 = expression1
.field2 = expression2
∙ ∙ ∙ ∙ ∙ ∙ ∙
.fieldN = expressionN
End With

In this operator, expression1, expression2, ..., expressionN are

arithmetic or logical expressions (Sections 1.8 and 1.10) or strings, which can
be considered as expressions (Section 1.19).

91
 Chapter 1. Programming in Visual Basic

For example:

With Sess
.Name = "Maksim Zakharkin"
.Number = 02237
.Physics = 4
.Math = 5
.Inform = 5
End With

The assignment operator can be applied both to fields and to entire records,
as in the following examples:

BestSess.Number = Sess.Number
BestSess = Sess

Arrays of records may be used. For example, it is natural to store the marks
obtained by students of group E13 in array SessE13. If this array of records is
static, it is declared as follows:

Dim SessE13(1 To 15) As Session

For example, let us consider the following code for calculating distance
between Tushino and Ostankino on the Moscow map.
Listing 1.4
Type Point
Name As String
x As Single
y As Single
End Type

Sub TushinoOstankino()
Dim P() As Point
ReDim P(0 To 1)
With P(0)
.Name = "Tushino"
.x = 17.6
.y = 29.7
End With

92
 1.18. Records

With P(1)
.Name = "Ostankino"
.x = 44.1
.y = 37.5
End With
MsgBox Str(Distance(P())) 'output of 27.62408
End Sub

Function Distance(positions() As Point) As Single

Distance = _
Sqr((positions(0).x - positions(1).x) ^ 2 + _
(positions(0).y - positions(1).y) ^ 2)
End Function

Code Listing 1.4 includes the operator creating the Point record, program
TushinoOstankino and the Distance function. This code has the follow-
ing peculiarities:
P is the dynamic array of records of the Point data type;
the Distance function argument is the array of records of type Point.
To understand the work of program TushinoOstankino, we advise the
reader to execute it step-by-step, watching the P array in window Watches.
Before the first press of the F8 key, the blinking cursor should be located in the
TushinoOstankino program, not in the operator creating the Point record
and not in the Distance function.

93
 Chapter 1. Programming in Visual Basic

1.19. Work with strings

The string is not only a pair of quotation marks "" and a sequence of charac-
ters enclosed in quotes (p. 27), but also a variable of the String data type,
declared by means of the String keyword. For example, in operator block

Dim A As String
Dim B As String * 15
A = "Informatics"
B = "Informatics"

we see the following strings: A, B, "Informatics".

The string as a variable may have inconstant or constant length (Appendix 1).
The variable-length string occupies a part of the main memory, which can
change during the program execution. In the above example, A is a string of
variable length. According to Appendix 1, after performing assignment operator

A = "Informatics"

the A string occupies 21 bytes of the main memory.

The fixed-length string occupies a fixed part of the main memory. At the
end of the string declaration (behind an asterisk), we must specify the size of the
main memory (in bytes) for this string. In the above example, B is a string of
fixed length, which occupies 15 bytes of the main memory.
The quantity of characters of the value, assigned to the fixed-length string,
may differ from the quantity specified in the declaration, i.e., may be less or
greater than 15. In the first case, instead of missing characters, spaces will be
automatically added to the end of the string. In the second case, superfluous
characters will be automatically removed.
For association of two or more strings, we must use one of signs & (amper-
sand) and + (plus). The resulting string, as a quoted sequence of characters, does
not depend on the sign.
The program below is an example of using signs & and + as the string con-
nector.

94
 1.19. Work with strings

Sub Strings1()
Dim strA As String, strB As String, strC As String
strA = "String ": strB = "variable"
strC = strA & strB
'result: strC = "String variable"
strC = "String " + strB
'result: strC = "String variable"
End Sub

Owing to the presence of the connector, strings

strA & strB
"String " + strB
are called compound strings. They can be considered as expressions similar to
arithmetic and logical expressions.
Term “substring” is used below. It is a string, not containing the space and
tabulation characters. Substrings are put together into a string by means of the
space character or vbTab — the built-in constant corresponding to the tabulation
character. The space character and vbTab (or the tabulation character) are called
the substring connectors.
Let us complete the last program by operators
strA = "String variable"
'result: strA = "String variable"
strB = "String" & " " & "variable"
'result: strB = "String variable"
strC = "String" & vbTab & "variable"
'result: strC = "String|variable"
According to the comments, the resulting first and second strings are the
same quoted set of characters. This string differs from the resulting third string
by only the character between substrings "String" and "variable". It is:
the space character in the first and second strings;
the tabulation character in the third string.
If we place the mouse pointer on strC in the program text (after execution
of the last operator), information String□variable or String variable appears.
It should be emphasized that the connection result is a string, not a substring.
When working with strings, three functions of removing spaces are used:
Trim deletes the beginning and ending spaces of the string that is the
function argument;
LTrim deletes the beginning spaces of the string (on the left);
RTrim deletes the ending spaces of the string (on the right).

95
 Chapter 1. Programming in Visual Basic

The following program is an example of using these functions.

Sub Strings2()
Dim strA As String, strB As String
strA = " String variable "
strB = Trim(strA)
'result: strB = "String variable"
strB = LTrim(strA)
'result: strB = "String variable "
strB = RTrim(strA)
'result: strB = " String variable"
End Sub

As already mentioned in Sections 1.4 and 1.8, for converting number to

string, the CStr or Str function is used; for the inverse conversion, the Val
function is used.
Function Space returns a string of spaces into the program; the quantity of
spaces is determined by the function argument.
The following program uses functions CStr, Val and Space.

Sub Strings3()
Dim strA As String, curB As Currency
Dim strC As String
strA = "X = "
curB = 45.77
strC = strA & CStr(curB)
'result: strC = "X = 45.77"
curB = Val("45.77 = X") 'result: curB = 45.77
curB = Val(strC) 'result: curB = 0
strC = "String" & Space(3) & "variable"
'result: strC = "String variable"
End Sub

It is possible to transform a string so that all letters in it become uppercase or

lowercase. For this purpose, functions UCase and LCase are used, respectively,
as in the following example program:

Sub Strings4()
Dim strA As String, strB As String
strA = "Pavel Ivanov"
strB = UCase(strA) 'result: strB = "PAVEL IVANOV"
strB = LCase(strA) 'result: strB = "pavel ivanov"
End Sub

96
 1.19. Work with strings

To replace any part of a string by certain characters, the Replace function

is used. The example program follows:

Sub Strings5()
Dim strA As String
Dim strB As String
strA = "Pavel Ivanov"
strB = Replace(strA, "Ivanov", "Gusev")
'result: strB = "Pavel Gusev"
End Sub

To determine the quantity of characters in a string (without considering

inverted commas), function Len (from “length”) is used, at that, its argument
is the string. Here is the example program:

Sub Strings6()
Dim strA As String
Dim intA As Integer
strA = "String variable"
intA = Len(strA) 'result: intA = 15
End Sub

Functions Replace and Len have additional possibilities, which can be

studied by means of the Excel help system started by pressing the F1 key when
the VB window is active.
Quite often, we have to extract a part from a string. For this purpose, func-
tions Left, Right and Mid (from “middle”) are used.
The calls of functions Left and Right are as follows:

Left(string, quantity)
Right(string, quantity)

These functions return the string containing the specified quantity of charac-
ters of the beginning and end of string, respectively.
The call of function Mid follows:

Mid(string, number[, quantity])

This function returns the string containing quantity characters of string,

starting from the character whose number equals number. If quantity is
omitted, the Mid function returns all characters up to the end of string.

97
 Chapter 1. Programming in Visual Basic

Functions Left, Right and Mid are used in the following program:

Sub Strings7()
Dim strA As String
Dim strB As String
strA = "My string variable"
strB = Left("My string variable", 9)
'result: strB = "My string"
strB = Right(strA, 8) 'result: strB = "variable"
strB = Mid(strA, 4, 6) 'result: strB = "string"
strB = Mid(strA, 11) 'result: strB = "variable"
End Sub

Let us consider an example of using the compound string in the call of the
MsgBox procedure.
1. In Listing 1.4, we replace operator MsgBox Str(Distance(P()))
on p. 93 by the following operator block:

Dim L As Single
Dim S As String
Dim M As Single
L = Distance(P())
S = InputBox("Enter scale and click OK")
'input of scale, Fig. 1.17
M = Val(S)
L = M * L
MsgBox "Distance from " & P(0).Name & " to " & _
P(1).Name & vbCrLf & "equals" & Str(L) & " km"
'output of distance, Fig. 1.18

2. Let us click on arrow ► to start the program execution.

3. We put 0.4 into the text box of the displayed window (Fig. 1.17) and click
on the OK button.
4. Let us click on the OK button in the emerging window (Fig. 1.18) to
terminate the program execution.
In the above operator block, the calls of InputBox and MsgBox appeared.
As already mentioned, the built-in InputBox function is used to input infor-
mation from the keyboard; this function returns (into the program) the string,
which was put into the text box. The build-in MsgBox procedure is used to
depict information on the display screen. The parameter of this procedure is
a compound string, which includes built-in constant vbCrLf.

98
 1.19. Work with strings

Fig. 1.17. The window for inputting the scale

Fig. 1.18. The window with the calculation result

Because of using the vbCrLf constant, we see two lines in the resulting
window (Fig. 1.18). The vbCrLf constant can be replaced by association
vbCr & vbLf of built-in constants “return” vbCr (from “Carriage return”) and
“new line” vbLf (from “Linefeed”).
The TimeValue function is useful, which converts time from the String
data type to the Date data type. As an example of using this function, let us
consider the following program:

Sub TV()
Dim dTime1 As Date
Dim dTime2 As Date
Dim dTime3 As Date
dTime1 = #2:30:45# + TimeValue("00:15:00")
'1st assignment operator
'result: dTime1 = 2:45:45

99
 Chapter 1. Programming in Visual Basic

dTime2 = Now + TimeValue("00:25:00")

'2nd assignment operator
dTime3 = Time + TimeValue("00:00:10")
'3rd assignment operator
End Sub

Because of executing the 2nd assignment operator, the dTime2 variable has
a value, which is equal to the current date and time plus 25 minutes. Because of
executing the 3rd assignment operator, dTime3 is equal to the current time plus
10 seconds.
In Listing 2.10 on p. 170, we will use the Format function intended for
converting a value to a string of the given form. To see the full information on
this function, we must press the F1 key when the blinking cursor is located on
Format in the code window.

100
 1.20. Work with text files

1.20. Work with text files

A file is an area on a hard disk, compact disk, USB flash drive (UFD) or any
other medium that:
contains single-type information;
has a name.
When working with files, we will use operations for reading information
from a file and writing information into a file.
Several types of files exist; we will consider text files. The contents of such
file are lines of characters with a combination of characters “return” and “new
line” at the end of each line.
For viewing text files, we will use the Notepad editor. At its window, we will
not see characters “return” and “new line”.
To start working with a text file, operator Open is used for opening this file.
This operator has the following syntax:

Open name For purpose As number

In this construct, name is the file full name, i.e., the file name together with its
path (in the file system of Windows) and extension, number is the file number,
purpose is keyword Input, Output or Append (from “appending”).
The last three keywords have the following sense:
Input means that the file must be opened for reading information from
this file;
Output — the file must be opened for writing information into it;
Append — the file must be opened for adding information into it.
As the file number, we recommend to use the number variable of the
Integer data type whose value is the result of executing the following assign-
ment operator:

number = FreeFile

where FreeFile is the built-in function that returns (into the program) the free
file number.

101
 Chapter 1. Programming in Visual Basic

The last assignment operator must be placed above the Open operator, more
precisely, it must be executed before the Open operator.
After finishing the work with the file, it must be closed by the Close opera-
tor as follows:

Close number

For addition of new lines into the file, operator Print is used, which has the
following syntax:

Print #number, line1

where line1 is a string (the digit may be different).

Let the file with specified number be opened by means of the Output
keyword. When performing the Print operator, string line1 (with a combina-
tion of characters “return” and “new line” at the end, not in inverted commas) is
written into the file beginning. When repeated performing the Print operator,
string line2 is added into the file, and so on.
Let the file with specified number be opened by means of the Append
keyword. When performing the Print operator, line1 is added into the file.
Below, we will consider two ways of extracting information from the file
with specified number, which is opened by means of the Input keyword.
1. Extracting information by using operator Line Input with the follow-
ing syntax:

Line Input #number, variable

This operator reads the next line from the file, at that, this line (without “return”
and “new line” at the end) is assigned to variable of the String data type.
2. Extracting information by means of built-in function

Input(quantity, number)

This function returns (into the program) the string, which contains the subse-
quent characters from the file. The quantity of these characters is specified by
quantity.
In programs Creation and Addition, given below, we will use the re-
viewed operators and built-in functions intended for work with text files. Using
these programs as examples, we will also consider other useful operators and
functions.

102
 1.20. Work with text files

The first program follows:

Listing 1.5
Sub Creation()
Dim FName1 As String
Dim FName2 As String
Dim FNum1 As Integer
Dim FNum2 As Integer
Dim n1 As Long
Dim n2 As Long
Dim strA As String 'auxiliary string
1: MkDir("c:\Users\usr\texts")
'creating folder texts
2: FName1 = _
"c:\Users\usr\texts\a.txt"
'full name of 1st file
3: FName2 = _
"c:\Users\usr\texts\b.txt"
'full name of 2nd file
'Creating file a.txt:
FNum1 = FreeFile
Open FName1 For Output As FNum1
'opening file a.txt
strA = "Text file is created,"
Print #FNum1, strA
Print #FNum1, _
"it contains several strings."
'Determining quantity of characters in file a.txt:
n1 = LOF(FNum1)
Close FNum1 'closing file a.txt
MsgBox "In file a.txt" & Str(n1) & " characters"
'Copying information from file a.txt to file b.txt:
FNum1 = FreeFile
Open FName1 For Input As FNum1
'opening file a.txt
FNum2 = FreeFile
Open FName2 For Output As FNum2
'opening file b.txt
Do Until EOF(FNum1) 'cycle of reading-writing
Line Input #FNum1, strA
Print #FNum2, strA
Loop

103
 Chapter 1. Programming in Visual Basic

Close FNum1 'closing file a.txt

'Adding new string into file b.txt:
strA = "New string is added."
Print #FNum2, strA
'Determining quantity of characters in file b.txt:
n2 = LOF(FNum2)
Close FNum2 'closing file b.txt
FNum2 = FreeFile
Open FName2 For Input As FNum2
'opening file b.txt
strA = Input(1, FNum2) 'reading character
Close FNum2 'closing file b.txt
MsgBox "In file b.txt" & Str(n2) & _
" characters, " & _
"and first character is " & _
strA
End Sub

In operators 1, 2 and 3, usr is the computer user name. Before the program
execution, the reader should type his concrete user name instead of usr in these
operators.
During the program execution, operator

MkDir("c:\Users\usr\texts")

creates folder texts inside folder usr; MkDir is the abbreviation of “make
directory”.
The built-in LOF function returns into the program the quantity of characters
in the file, at that, characters “return” and “new line”, which are at the end of
each line, are taken into consideration. The argument of this function is the file
number; LOF is the abbreviation of “length of file”.
The argument of the built-in EOF function is the file number too. This func-
tion, figuring in the condition of the Do Until…Loop cycle termination, returns
(into the program) True at achievement of the file end. The function name is the
abbreviation of “end of file”.
The MsgBox procedure is intended for depicting the string, which is its
parameter, on the display screen. Two calls of MsgBox are in the Creation
program, therefore, two windows, represented in Fig. 1.19 and 1.20, appear
during the program execution.
After termination of the program execution, files a.txt and b.txt have
the following contents.

104
 1.20. Work with text files

Fig. 1.19

Fig. 1.20

File a.txt:

Text file is created,

it contains several strings.

File b.txt:

Text file is created,

it contains several strings.
New string is added.

We advise the reader to look through the contents of files a.txt and
b.txt by using the Notepad editor to be convinced of the correctness of the
program work.
Before restarting the Creation program, the MkDir operator of creating
the texts folder must be omitted, for example, by means of an apostrophe.

105
 Chapter 1. Programming in Visual Basic

The liquidation of files a.txt and b.txt and folder texts may be
required at the end of the program execution. For this purpose, the following
operators are used:

Kill(FName1)
Kill(FName2)
RmDir("c:\Users\usr\texts")

where RmDir is the abbreviation of “remove directory”. We must insert these

operators above operator End Sub.
At the end of this section, we will consider how to tune Windows Explorer
for displaying not only the file names, but also the extension of these names, for
example, extension .txt for the above text files.
The following second program adds a line into file b.txt and runs the
Notepad editor.
Listing 1.6
Sub Addition()
Dim FNum As Integer, n As Long
Dim RetVal As Integer 'for function Shell
FNum = FreeFile
'Opening file b.txt and adding new string:
Open "c:\Users\usr\texts\b.txt" _
For Append As FNum
Print #FNum, "Second new string is added."
'Determining quantity of characters in file b.txt:
n = LOF(FNum)
Close FNum 'closing file b.txt
MsgBox "In file b.txt" & Str(n) & " characters"
'Starting editor Notepad:
0: RetVal = Shell("c:\Windows\notepad.exe",1)
End Sub

The Shell function (operator 0) is intended for running any executable file
whose name has extension .exe. The first argument of the Shell function is the
full name of the executable file (here, the full name must be without spaces). The
second argument, which may be omitted, defines the style of the window being
a result of calling the Shell function.
To see the full information on the Shell function, we must press the F1 key
when the blinking cursor is located on Shell in the code window.

106
 1.20. Work with text files

We will run the Addition program after the execution of the Creation
program (without the operators liquidating the files and folder).
The window, shown in Fig. 1.21, appears during the Addition program
execution. After clicking on OK in this window, operator 0 is executed, which
starts the Notepad editor (Fig. 1.22), and then operator End Sub is executed, i.e.,
the Addition program is terminated.

Fig. 1.21

Fig. 1.22. The Notepad window

After terminating the Addition program, the Notepad window remains

open. By means of this window, we can view the contents of a text file (in par-
ticular, b.txt) and, if necessary, create a new file or edit an existing file.
After the Addition program execution, file b.txt has the following con-
tents:

107
 Chapter 1. Programming in Visual Basic

Text file is created,

it contains several strings.
New string is added.
Second new string is added.

If a file is opened for adding information (by using keyword Append), this
file may not exist. In this case, the file is created. If a file is opened for reading
information (by using keyword Input), this file, naturally, should exist.
To see file names with extension in Windows Explorer, we must fulfill the
following:
1) in Windows Explorer, open the folder, which contains files of interest;
2) on the menu bar of Windows Explorer, fulfill Organize > Folder and
search options;
3) in open window Folder Options, activate tab View;
4) in list Advanced settings, turn off option Hide extensions for known file
types;
5) successively click on buttons Apply and OK.
To expand the chosen operation mode of Windows Explorer to all folders, we
must click on button Apply to Folders before clicking on the Apply button. In
open window Folder Views, we must click on the Yes button.
We advise the reader to write a program for creating a text file including
n 1 lines of the following form:
f( xi ) fi ,
where xi , fi are the values of argument and function, 0 i n; a x0 < x1 <
x 2 < ... < xn 2
< xn 1
< xn b . Function f (x) and corresponding values of
a and b from Appendix 4 must be used.
The constructs of Visual Basic considered above can be used for solving
sufficiently complicated tasks, two of which will be formulated in the next
section.

108
 1.21. Matrix terminology. Formulation of demonstration tasks

1.21. Matrix terminology. Formulation

of demonstration tasks

In the previous sections, we were solving the task of calculating the hypote-
nuse length. However, this task is very simple, and it is impossible to show all
possibilities of Visual Basic on it. Therefore, we will also solve two tasks of
transposing a numerical matrix, relative to the main and auxiliary diagonals. Let
us formalize the concept of matrix transposition.
Let A be a matrix containing m rows and n columns, B be a matrix containing
n rows and m columns:

a11 a12 ... a1 j ... a1n b11 b12 ... b1i ... b1m
a21 a22 ... a2 j ... a2 n b21 b22 ... b2i ... b2 m
. . . . . . . . . . . .
A , B .
ai1 ai 2 ... aij ... ain b j1 b j 2 ... b ji ... b jm
. . . . . . . . . . . .
am1 am 2 ... amj ... amn bn1 bn 2 ... bni ... bnm

If elements of matrix B “are calculated” according to formula

b ji aij (1.2)
at 1 ≤ i ≤ m, 1 ≤ j ≤ n, mathematicians say that matrix B is obtained by transpos-
ing matrix A relative to the main diagonal (or simply, by transposing matrix A).
The main (left-to-right) diagonal of the matrix is an imaginary straight line
from the top left corner of the matrix to the bottom right corner.
If formula (1.2) is replaced by formula
b ji am 1 i, n 1 j
, (1.3)
then matrix B is obtained by transposing A relative to the auxiliary diagonal.
The auxiliary (right-to-left) diagonal of the matrix is an imaginary straight
line from the top right corner to the bottom left corner.
To solve the task of transposing matrix A, we must generate matrix B accord-
ing to formula (1.2) or (1.3). Note that AT is the usual designation of the A ma-
trix transposed relative to the main diagonal, i.e., according to (1.2).

109
 Chapter 1. Programming in Visual Basic

Let us notice the following obvious fact: when transposing the matrix relative
to the main diagonal, its rows and columns interchange their positions.
If m = n, matrix A is called a square matrix. In this case, elements a ii are
located on the main diagonal, elements ai , m 1 i
are located on the auxiliary
diagonal, 1 ≤ i ≤ m.
When transposing square matrix A relative to the main (auxiliary) diagonal,
the mirror reflection of A relatively the main (auxiliary) diagonal takes place.
Further, we will use the following terms and definitions.
The A matrix, containing m rows and n columns, is named as “the A matrix
of size m × n” or “the A matrix m × n”. The matrix, containing one column or one
row, is called a vector.
The product of the A matrix m × n and the C matrix n × s is the AC matrix
m × s, which is a result of the scalar multiplication of the rows of A and the
columns of C, namely:

a11 a12 ... a1n c11 c12 ... c1s

a 21 a 22 ... a 2 n c 21 c 22 ... c 2 s
AC
. . . . . . . .
a m1 am2 ... a mn c n1 cn 2 ... c ns

a11c11 ... a1n cn1 a11c12 ... a1n cn 2 ... a11c1s ... a1n cns
a21c11 ... a2 n cn1 a21c12 ... a2 n cn 2 ... a21c1s ... a2 n cns
.
... ... ... ...
am1c11 ... amncn1 am1c12 ... amncn 2 ... am1c1s ... amncns

More precisely, the elements of R = AC are defined by the following formula:

n
riq ( a ij c jq ) ,
j 1
1 ≤ i ≤ m, 1 ≤ q ≤ s.
The A matrix m × n and the D matrix m × n are called equal if the correspond-
ing elements of these matrices are equal: aij dij , 1 ≤ i ≤ m, 1 ≤ j ≤ n.
Square matrix E is called the unit matrix if the main diagonal of E contains
only units, and all remaining elements of E are equal to zero.
The inverse A matrix is the A-1 matrix of size n × m, which satisfies the fol-
lowing condition: A-1A = E, where E is the unit matrix of size n × n.

110
 1.22. Program for transposing a matrix relative to its auxiliary diagonal

1.22. Program for transposing a matrix

relative to its auxiliary diagonal

The program for solving one of the two tasks formulated in the previous sec-
tion is given below. This program uses file a.txt with the source data (values
of m and n and matrix A) and creates file b.txt with the result (the B matrix).
Listing 1.7
Sub TRANSPA()
Dim FNameA As String, FNameB As String
Dim FNum As Integer
Dim strC As String, strD As String, strE As String
Dim m As Integer, n As Integer
Dim i As Integer, j As Integer
Dim k As Integer, l As Integer
Dim A() As Double, B() As Double
1: FNameA = _
"c:\Users\usr\texts\a.txt"
2: FNameB = _
"c:\Users\usr\texts\b.txt"
3: FNum = FreeFile
'Opening file a.txt:
4: Open FNameA For Input As FNum
'Reading values of m and n from file a.txt:
5: Line Input #FNum, strC
6: strC = Mid(strC, 3)
7: m = Val(strC)
8: Line Input #FNum, strC
9: strC = Mid(strC, 3)
10: n = Val(strC)
'Setting size of matrices:
11: ReDim A(1 To m, 1 To n)
12: ReDim B(1 To n, 1 To m)
'Reading matrix A from file a.txt:
13: For i = 1 To m
14: Line Input #FNum, strC

111
 Chapter 1. Programming in Visual Basic

15: j = 0
16: strD = ""
'string "" is not equal to string " "
17: l = Len(strC)
18: For k = 1 To l
19: strE = Mid(strC, k, 1)
20: If strE <> " " Then strD = strD & strE
21: If strE = " " Or k = l Then
22: j = j + 1
23: A(i, j) = Val(strD)
24: strD = ""
25: End If
26: Next k
27: Next i
'Closing file a.txt:
28: Close FNum
29: FNum = FreeFile
'Opening file b.txt:
30: Open FNameB For Output As FNum
'Forming matrix B, its writing into file b.txt:
31: For j = 1 To n
32: strC = ""
33: For i = 1 To m
34: B(j, i) = A(m + 1 - i, n + 1 - j)
35: strC = strC & Str(B(j, i)) & " "
36: Next i
37: Print #FNum, strC
38: Next j
'Closing file b.txt:
39: Close FNum
End Sub

In operators 1 and 2, usr is the user name.

Program TRANSPA is intended for transposing a numerical matrix relative to
its auxiliary diagonal. In operators 16, 24 and 32, we see string "", not contain-
ing any character. Operators 20, 21 and 35 include string " ", which contains
only the space character.
Operator 34 corresponds to formula (1.3), which defines the operation of
transposing a matrix relative to its auxiliary diagonal.
File

c:\Users\usr\texts\a.txt

112
 1.22. Program for transposing a matrix relative to its auxiliary diagonal

contains the source data, i.e., values of m and n and matrix A. This file may be
created by means of the Notepad editor whose window is represented in
Fig. 1.23.

Fig. 1.23. The source data: elements of the matrix’s

rows separate from each other by one space

For starting the execution, we can use one of the following two ways:
clicking on arrow ► in the VB window;
fulfilling the following operations in the Excel window: Developer (or
View) > Macros > line TRANSPA > Run.
It is possible to appoint a combination of keys starting the program execu-
tion. For that, being in the Excel window, we fulfill operations Developer
(or View) > Macros > line TRANSPA > Options. In open window Macro
Options, we fulfill the following:
appoint the startup key combination, for example, Ctrl + t;
click on the OK button;
close the window by clicking on the little cross in the top right corner.
After that, the simultaneous press of the Ctrl and t keys runs the TRANSPA
program.
The result of the TRANSPA program execution is matrix B, which is in file

c:\Users\usr\texts\b.txt

The Notepad window with matrix B is represented in Fig. 1.24.

Let us consider ways of transition from the VB window to the Excel window
and back.
The transition from the VB window to the Excel window is possible in one of
the following four ways:

113
 Chapter 1. Programming in Visual Basic

clicking on the Excel button at the left end of the standard toolbar in the
VB window;
simultaneous pressing the Alt and F11 keys;
rolling the VB window down by means of the underscore button in the
top right corner;
clicking on the Excel button on the taskbar of Windows Desktop.

Fig. 1.24. The calculation result

For returning to the VB window from the Excel window, we should fulfill
Developer (or View) > Macros > line TRANSPA > Edit. As a result, the VB win-
dow with the TRANSPA program text in the code window appears. Further, we
can correct the program, for example, change names of files with the source data
and for the calculation result.
If we click on button Step Into (instead of button Edit), the VB window will
also be opened. Further, the step-by-step execution of the TRANSPA program
(by means of the F8 key) is possible.
Other ways of transition from the Excel window to the VB window follow:
pressing Alt + F11;
rolling the Excel window down by means of the underscore button;
clicking on the VBA button on the taskbar of Windows Desktop.
The TRANSPA program text almost coincides with the text of the program,
which will be reviewed in the next section. Therefore, the TRANSPA program
will still be required.
For further usage of the TRANSPA program, we have to save it on the com-
puter’s hard disk by fulfilling the following operations similar to the operations
described on p. 18:
1) in the Excel window, File > Save As > Browse;
2) in the Save As window, choose a folder intended for saving the Excel
workbook, for example, c:\Users\usr;

114
 1.22. Program for transposing a matrix relative to its auxiliary diagonal

3) enter BookTRANSPA into text box File name;

4) set file type Excel Macro-Enabled Workbook by means of drop-down list
Save as type;
5) click on the Save button.
The TRANSPA program is saved as a part of the BookTRANSPA.xlsm
workbook of Excel.
As we see, the user interface of the TRANSPA program is a pair of text files.

115
 Chapter 1. Programming in Visual Basic

1.23. User-defined forms

As the program user interface, the form is sometimes convenient. For creat-
ing this rectangle with text boxes, buttons and other control elements, Form
Designer exists in Visual Basic Environment.
Let us create the form for the program intended for calculating the length of
the hypotenuse of a right-angled triangle.
The following control elements are suitable for the form:
place for displaying the calculation result;
text box for inputting (into the program) the length of the first leg;
text box for inputting the length of the second leg;
button for starting the calculation.
Let Book1 be the name of the active Excel workbook. Developing the pro-
gram (project) with the form begins with the following sequence of operations
for inserting the form into the active Excel workbook (instead of inserting the
module).
1. In the Excel window, fulfill Developer > Visual Basic in area Code.
As a result, the VB window (Fig. 1.1) containing the project explorer window
and the properties window appears.
In the absence of the properties window, we can open it by fulfilling View >
Properties Window.
2. In the project explorer window, highlight line VBAProject (Book1) by
clicking on it.
3. Fulfill Insert > UserForm.
As a result, the following features appear (Fig. 1.25):
blank form UserForm1, a rectangle with points;
line UserForm1 in the project explorer window;
the Toolbox window, containing the control elements.
If the last window is not displayed, for displaying it, we must click on the
form and then fulfill View > Toolbox.
The properties window “is tied” to the form because this element is selected
by the frame with 8 markers (five black and three white, Fig. 1.25).
To change the width of the UserForm1 form, we have to drag the form’s right
border by using the mouse, having seized the white marker in the center of this

116
 1.23. User-defined forms

border. Thus, the value of the Width property (in the properties window) also
changes.

Fig. 1.25. The VB window after inserting

UserForm1 into the Excel workbook

Expression “to seize a point” means the following: to place the mouse pointer
on this point, and then to press the left button on the mouse.
The value of property Width (that is, the form width) can also be set in the
properties window.
Similarly, we can change the height of the UserForm1 form (property
Height).
After clicking on property Caption (in the properties window), we will
change it to the following: My first form.
The Toolbox window (Fig. 1.25) contains pictograms of the control elements,
which can be simply used in the project. We will use the following four ele-
ments: Label, TextBox, CommandButton and CheckBox. To see the element
name, we have to place the mouse pointer on this element in Toolbox.
To insert an element into the form and to edit this element, we fulfill the
following operations.

117
 Chapter 1. Programming in Visual Basic

1. Select the form by clicking on it.

2. Click on the required element of the Toolbox window.
3. Move the mouse pointer into the form. At that, the pointer becomes
a crosshair.
4. Place the crosshair into the required part of the form and press the left
button on the mouse. After that, release the button.
Or in another way, without releasing the left button, drag the mouse pointer,
for example, downwards and to the right. After that, release the button.
5. Move the element (if needed), having seized its center.
6. Move the element’s borders (if needed), by turn having seized the white
markers.
7. Set the element’s properties by means of the properties window, which is
tied to this element (selected by the frame with 8 white markers).
Upon termination of the element editing, we must remove its selection by
clicking on area outside the element limits. To return to the element editing, we
must select this element by clicking on its image in the form. For removing the
selected element from the form, we must press the Delete key.
In the form, we should create the message place, where the calculation result
will be displayed during the program execution. For this purpose, we will use
element Label.
To insert element Label into the form and to change several properties of this
element, we fulfill the following.
1. Click on the Label element, which is in the Toolbox window.
2. Move the mouse pointer into the form. At that, the pointer becomes
a crosshair.
3. Place the crosshair into the top left part of the form and press the left but-
ton on the mouse. Without releasing the button, drag the mouse pointer down-
wards and to the right. After that, release the button.
The message place with default name Label1 (that is, the Label1 element) is
the result (Fig 1.26).
4. Use the Font property for setting the font fashion and size. When clicking
on this property, a button appears in the properties window. The dots on this but-
ton mean that a window exists for setting the Font property.
5. Click on the dots image button. At that, the Font window appears for set-
ting the message parameters.
6. Click on line Cambria Math (Fig. 1.27) for setting this font. In the Sample
area, the font fashion is pictured.
7. Set the font size at 10.
8. Click on the OK button.

118
 1.23. User-defined forms

Fig 1.26. The VB window after inserting

element Label1 into form UserForm1

Fig. 1.27. The window for setting font

119
 Chapter 1. Programming in Visual Basic

9. Set the following values of other properties of the message place.

TextAlign: 2 (horizontal center alignment, taken from the drop-down list)
Caption: Enter source data and click button "Account"
Name: LabelMessage (we change the element name)
10. By means of the mouse, change the height and width of the message place
and the height and width of the form to get the message image as in Fig. 1.28.

Fig. 1.28. The form after the program start

Similarly, we can insert the text box into the form. For that, the TextBox
element is used.
We will insert two text boxes (Fig. 1.28), with names TextBoxA and
TextBoxB. On the left of the text boxes, we will create inscriptions a and b by
using the Label element.
To insert the button into the form, we fulfill the following.
1. Click on the CommandButton element, which is in the Toolbox window.
2. Depict the button in the form.
3. Set the following properties of this button.
Caption: Account
Name: ComButCalc
4. Remove the button selection by clicking on area outside its limits.
The development of the form by means of Form Designer is completed. Now
we have to develop the program.
Let us click twice on the Account button, which is in the UserForm1 form.
At that, there appears the code window (corresponding to the form) with the
blank of the program, being started by clicking on the Account button.

120
 1.23. User-defined forms

The following first and last lines of the program are generated automatically:

Private Sub ComButCalc_Click()

End Sub

We already encountered the Private keyword on p. 81. Here, it means the

following: if ComButCalc_Click is used as a subroutine, its call should be in
the same code window, in which the subroutine declaration is located. However,
we are not going to use ComButCalc_Click as a subroutine. Therefore, the
Private keyword may be deleted.
The ComButCalc_Click program for calculating the hypotenuse length
follows:

Private Sub ComButCalc_Click()

Dim a As Single, b As Single, c As Single
Dim s As String
1: s = TextBoxA.Text
2: a = Val(s)
3: s = TextBoxB.Text
4: b = Val(s)
5: c = Sqr(a ^ 2 + b ^ 2) 'according to Pythagoras
6: s = Str(c)
7: LabelMessage.Caption = "c =" & s
End Sub

This program is similar to the Pythagoras1 program (p. 85).

Unlike program Pythagoras1, the ComButCalc_Click program con-
tains the following three constructs similar to the record: TextBoxA,
TextBoxB and LabelMessage.
In operator 1, “field” Text of “record” TextBoxA (of “type” TextBox),
more precisely, property Text of element TextBoxA is a variable of the
String data type whose value coincides with the string being entered into the a
text box of the form during the program usage. In operator 3, property Text of
element TextBoxB coincides with the string being entered into the b text box.
The values of a and b are the results of executing operators 1 — 4.
Operator 5 calculates length c of the hypotenuse; operator 6 converts this
value to string s.
Operator 7 assigns the string (which includes the s string) to property
LabelMessage.Caption (that is, to property Caption of element
LabelMessage). The value of this property is put into the form.

121
 Chapter 1. Programming in Visual Basic

We can start the developed program by clicking on arrow ► of the VB win-

dow. The form (Fig. 1.28), loaded into the Excel window, is the result of this
click. Fig. 1.29 shows the form’s state upon finishing the calculation, i.e., after
putting the lengths of the triangle’s legs into text boxes a and b and clicking on
the Account button.

Fig. 1.29. The form upon finishing the calculation

Further, the calculation may be repeated for other source data, i.e., other
lengths of the triangle’s legs. Upon termination of the series of calculations, we
have to close the form by clicking on the little cross in the top right corner
(Fig. 1.29).
We can simplify the ComButCalc_Click program as follows:

Private Sub ComButCalc_Click()

Dim a As Single, b As Single, c As Single
1: a = TextBoxA.Value
2: b = TextBoxB.Value
3: c = Sqr(a ^ 2 + b ^ 2) 'according to Pythagoras
4: LabelMessage.Caption = "c =" & Str(c)
End Sub

In this program, we used property Value of element TextBox. At that, when

performing operator 1, the information, which is in the corresponding text box of
the form, is interpreted according to the data type of the variable on the left of
sign =, that is, as a number. The same can be said about operator 2.

122
 1.23. User-defined forms

In the previous section, we developed the TRANSPA program, Listing 1.7, for
transposing a number matrix relative to its auxiliary diagonal. This program is
inconvenient to use because names of files are a part of the program text (see
operators 1 and 2). To eliminate this drawback, we will add the form, already
designed, to the TRANSPA program. In this case, the form elements have the
following sense (from top to down, Fig. 1.28):
place for displaying messages;
text box for inputting the full name of the source data file;
text box for inputting the full name of the result file;
button for starting the calculation.
The basis of the new ComButCalc_Click program is the TRANSPA pro-
gram text, which is in the BookTRANSPA workbook. To gain access to this text,
we fulfill the following (p. 18):
1) open the BookTRANSPA workbook with Excel, for example, by means
of the context menu;
2) go to Visual Basic Environment in the standard way;
3) open the code window with the TRANSPA program text by clicking twice
on the corresponding module name in the project explorer window.
The TRANSPA program text is also contained in file Listing_1_07.txt of the
enclosed CD.
By using the TRANSPA program text, Listing 1.7, we will transform the
ComButCalc_Click program developed above to the following view by
means of Windows Clipboard.

Private Sub ComButCalc_Click()

Dim FNameA As String, FNameB As String
Dim FNum As Integer
Dim strC As String, strD As String, strE As String
Dim m As Integer, n As Integer
Dim i As Integer, j As Integer
Dim k As Integer, l As Integer
Dim A() As Double, B() As Double
1: FNameA = TextBoxA.Text
2: FNameB = TextBoxB.Text
3: FNum = FreeFile
'Opening file a.txt:
4: Open FNameA For Input As FNum
'Reading values of m and n from file a.txt:
5: Line Input #FNum, strC
6: strC = Mid(strC, 3)
7: m = Val(strC)

123
 Chapter 1. Programming in Visual Basic

8: Line Input #FNum, strC

9: strC = Mid(strC, 3)
10: n = Val(strC)
'Setting size of matrices:
11: ReDim A(1 To m, 1 To n)
12: ReDim B(1 To n, 1 To m)
'Reading matrix A from file a.txt:
13: For i = 1 To m
14: Line Input #FNum, strC
15: j = 0
16: strD = ""
'string "" is not equal to string " "
17: l = Len(strC)
18: For k = 1 To l
19: strE = Mid(strC, k, 1)
20: If strE <> " " Then strD = strD & strE
21: If strE = " " Or k = l Then
22: j = j + 1
23: A(i, j) = Val(strD)
24: strD = ""
25: End If
26: Next k
27: Next i
'Closing file a.txt:
28: Close FNum
29: FNum = FreeFile
'Opening file b.txt:
30: Open FNameB For Output As FNum
'Forming matrix B and its writing into file b.txt:
31: For j = 1 To n
32: strC = ""
33: For i = 1 To m
34: B(j, i) = A(m + 1 - i, n + 1 - j)
35: strC = strC & Str(B(j, i)) & " "
36: Next i
37: Print #FNum, strC
38: Next j
'Closing file b.txt:
39: Close FNum
40: LabelMessage.Caption = _
"Account is terminated. Input..." 'message
End Sub

124
 1.23. User-defined forms

Property Text of element TextBoxA is a variable whose value coincides

with the full name of the text file, containing values of m and n and matrix A;
this full name is entered into the a text box of the UserForm1 form during the
program usage. Property Text of element TextBoxB is a variable whose value
coincides with the full name of the text file for matrix B; this full name is entered
into the b text box. When executing the program, operators 1 and 2 specify the
files’ full names. Operators 4 and 30 open the corresponding files.
When executing operator 40, string

"Account is terminated. Input..."

is assigned to property LabelMessage.Caption, i.e., this string is put into

the message place of the form.
We can start the developed program by clicking on arrow ► of the VB win-
dow. As a result, the form (Fig. 1.28) is loaded. Fig. 1.30 shows the form’s state
upon finishing the calculation, i.e., after inputting

c:\Users\usr\texts\a.txt
c:\Users\usr\texts\b.txt

into text boxes a and b, respectively, and clicking on button Account (usr is the
user name in these two full names).
Further, the calculation may be repeated for other names of the files and for
other contents of the source data file. Upon termination of the series of calcula-
tions, we have to close the form, depicted in Fig. 1.30, by clicking on the little
cross in the top right corner.
The source data file, which contains values of m and n and matrix A, can be
created and/or edited by using the Notepad editor. We can view the calculation
result by means of the same editor. Fig. 1.23 and 1.24 show the Notepad window
with the source data and calculation result.
We can use element TextBox not only for input of information but also for
output. In this case, property Text of the corresponding TextBox element (for
example, property TextBox3.Text) must be in the left-hand side of the assign-
ment operator.
We advise the reader to modify the form and the text of the last program so
that "Account is terminated. Input..." is being put into the text box of the form.
We learned how to use the following three elements of the Toolbox window:
Label, TextBox and CommandButton. Let us also consider element CheckBox,
which is used in the following cases:
an option should either be turned on or turned off;
one of two alternatives must be chosen.

125
 Chapter 1. Programming in Visual Basic

Fig. 1.30. The form upon termination of the matrix transposition

In the form, the CheckBox element looks like a little square field. At the click
on this field, the check (tick) mark appears in it; at the repeated click, the check
mark disappears.
Property Value of element CheckBox has value True or False as follows:
True when the check mark is present in the field;
False when the check mark is absent.
Let us expand the form, depicted in Fig. 1.28, by the CheckBox element
placed on the left of the button and return to the program, which calculates the
hypotenuse length.
The new version of the ComButCalc_Click program (p. 121) follows:

Private Sub ComButCalc_Click()

Dim a As Single, b As Single, c As Single
Dim s As String, out As String
a = TextBoxA.Value
b = TextBoxB.Value
c = Sqr(a ^ 2 + b ^ 2) 'according to Pythagoras
out = "c =" & Str(c)
If CheckBox1.Value Then
s = Str(a * b / 2)
out = out & vbCrLf & "s =" & s
End If
LabelMessage.Caption = out
End Sub

126
 1.23. User-defined forms

In this program, CheckBox1 is the CheckBox element’s name. In the

presence of the check mark, the triangle area is calculated in addition to the
hypotenuse length.
Fig. 1.31 and 1.32 show the initial and final states of the form. To get the re-
sult, depicted in Fig. 1.32, we must fulfill the following operations:
1) put 3 and 4 into text boxes a and b of the form in Fig. 1.31;
2) set the check mark by clicking on element s;
3) click on the Account button.
Let us save the Excel workbook on the hard disk of the computer under name
BookForm, having done File > Save As > Browse, and so on (p. 114). At that, the
project (with the form), which is a part of the workbook, is saved too.
For returning to the project, the double click on line UserForm1 in the project
explorer window is required. We can delete the form, as well as the module, by
means of the context menu. For that:
1) right click on the form name in the project explorer window;
2) in the open context menu, fulfill the Remove command;
3) click on the No button in the open window with a question about export-
ing the form before removing it.
We advise the reader to use the form with the CheckBox element for the
matrix transposition and to modify the program on p. 123 so that:
in the presence of the check mark, the matrix transposition relative to the
main diagonal would be performed according to formula (1.2) on p. 109;
in the absence of the check mark, the matrix transposition relative to the
auxiliary diagonal would be performed according to formula (1.3).

Fig. 1.31. The form after the program start

127
 Chapter 1. Programming in Visual Basic

Fig. 1.32. The form upon finishing the calculation

128
 1.24. Digression. Developing programs with the form in Microsoft Visual Studio

1.24. Digression. Developing programs with

the form in Microsoft Visual Studio

The programs, which were reviewed above and will be reviewed below, have
the following drawback: we must equip our computer with tabular processor
Excel for calculations by means of these programs. If we want to execute a pro-
gram without using Excel, this program should be developed in Visual Basic
Environment, which is not a part of Microsoft Office. Let us consider singulari-
ties of developing the program (project) with the form in Visual Basic Environ-
ment, which is a part of package Microsoft Visual Studio 2010.
Microsoft Visual Studio 2010 is installed on the computer in the standard
way. To start Visual Studio, we fulfill the following operations on Windows
Desktop: Start > All Programs > Microsoft Visual Studio 2010 > Microsoft
Visual Studio 2010. The Start Page window is the result.
For setting the necessary operation mode of Visual Studio, we must fulfill the
following operations:
1) click on the New Project hyperlink;
2) in the left area of the open New Project window, click on the plus sign
against Other Languages;
3) in the open list, click on the plus sign against Visual Basic;
4) in the open list, click on line Windows;
5) in the list of the central area of the New Project window, click on line
Windows Forms Application.
The necessary operation mode of Visual Studio is the result. The following
information in the right area of the New Project window speaks about it:
A project for creating an application with a Windows user interface.
Further, the project name and the folder, intended for the project, must be
given by means of text boxes Name and Location. By default, the project and
folder have the same name, WindowsApplication1 (the digit in the name may be
different).
After clicking on the OK button, the window of Visual Basic Environment
is displayed. Blank form Form1 is a part of this window. Fig. 1.33 shows the
window after fulfilling View > Toolbox. To open the properties window, we must
fulfill the following operations: View > Other Windows > Properties Window.
The properties window (Fig. 1.34) appears in place of the Toolbox window.

129
 Chapter 1. Programming in Visual Basic

Fig. 1.33. The window of Visual Basic Environment (a part

of Microsoft Visual Studio 2010) with the Toolbox window

The development of forms and corresponding programs is similar to the

development described in the previous section.
The text of the programs, developed in Microsoft Visual Studio 2010, is
close to the text of the programs, developed in Excel. For example, program
ComButCalc_Click for calculating the hypotenuse length (p. 121) becomes
as follows:

Public Class Form1

Private Sub ComButCalc_Click(ByVal sender _
As System.Object, ByVal e As System.EventArgs) _
Handles ComButCalc.Click
Dim a As Single, b As Single, c As Single
Dim s As String
s = TextBoxA.Text

130
 1.24. Digression. Developing programs with the form in Microsoft Visual Studio

a =
Val(s)
s =
TextBoxB.Text
b =
Val(s)
c =
Math.Sqrt(a ^ 2 + b ^ 2)
'according to Pythagoras
s = Str(c)
LabelMessage.Text = "c=" & s
End Sub
End Class

The beginning and end of the program were generated automatically when
we clicked twice on the button inserted into the form.

Fig. 1.34. The properties window as a part

of Visual Basic Environment

The project contains several files and folders. By default, it is located in the
folder with the following full name:

c:\Users\usr\WindowsApplication1

where usr is the user name.

131
 Chapter 1. Programming in Visual Basic

To save the current state of the project, we must fulfill File > Save. For
returning to the project, we can use one of the following two ways:
the double click on the WindowsApplication1.vbproj file in the project
folder;
the click on the WindowsApplication1 hyperlink in area Recent Projects
of the Start Page window of Visual Studio.
When we execute the program during its development, the debug version of
the program executable file is created automatically. The full name of this file
follows:

c:\Users\usr\WindowsApplication1\bin\Debug\ _
WindowsApplication1.exe

Upon termination of the program development, we must create the release

(working) version of the executable file. For that, we fulfill the following opera-
tions:
1) Build > Configuration Manager;
2) in the Configuration Manager window opened, enter Release into the
Active solution configuration box by means of the drop-down list;
3) click on the Close button;
4) Build > WindowsApplication1.
The release version of the executable file is the result. Its full name follows:

c:\Users\usr\WindowsApplication1\bin\Release\ _
WindowsApplication1.exe

The release version of the executable file is more efficient (in respect of the exe-
cution rate) in comparison with the debug version.
For loading the form, we must click twice on the program executable file in
Windows Explorer. Further, we can use the form for the calculation.
If we want to transfer the program with the form to another user, it is enough
to transfer the program executable file.
Let us return to programming in tabular processor Excel.

132
 Chapter 2.
Programming in VBA

We review the main objects of the VBA programming language, which is the
Visual Basic extension in the sense that VBA includes the VB constructs. The
Excel table is considered as the user interface of macros.
Besides, we also consider creating Excel user-defined functions and working
with Excel Macro Recorder, Personal Macro Workbook and the reference sys-
tems.

133
 Chapter 2. Programming in VBA

2.1. Loading the form from the Excel window.

Running the program executable file

The user-defined forms, developed in Section 1.23, were loading from the
VB window, for example, by clicking on arrow ► of the standard toolbar. How-
ever, it is desirable to load the form from the Excel window.
Below, we will consider the form-loading program.
Let us open the BookForm workbook of Excel (p. 127), go to the VB window
and fulfill Insert > Module.
We enter the following simple program into the code window of the inserted
module:

Sub LoadingForm()
UserForm1.Show
End Sub

Here, UserForm1 is the form name.

Now the form, pictured in Fig. 1.31, can be loaded from the Excel window as
follows: Developer (or View) > Macros > line LoadingForm > Run. By fulfilling
Developer (or View) > Macros > line LoadingForm > Options > …, we can
appoint a combination of keys for loading the form.
The loaded form may be used for transposing a numerical matrix or calculat-
ing the length of the hypotenuse of a right-angled triangle and its area.
If we want the blinking cursor to be, for example, in text box b of the loaded
form, it is necessary to insert operator

UserForm1.TextBoxB.SetFocus

above operator UserForm1.Show. The LoadingForm program becomes as

follows (Fig. 2.1):

Sub LoadingForm()
UserForm1.TextBoxB.SetFocus
UserForm1.Show
End Sub

134
 2.1. Loading the form from the Excel window. Running the program executable file

Fig. 2.1. The VB window with lines UserForm1

and Module1 in the project explorer window

Let us save the Excel workbook with the LoadingForm program under the
old name of BookForm.
The following program (in the code window) allows opening the Notepad
window in Excel:

Sub StartEXE()
Dim RetVal As Integer 'for function Shell
RetVal = Shell("c:\Windows\notepad.exe",1)
End Sub

In this program, we see the call of the Shell function, as in the Addition
program on p. 106.
To open the Notepad window, we must run the StartEXE program by ful-
filling Developer (or View) > Macros > line StartEXE > Run. Besides, we can
appoint a combination of keys for opening the Notepad window.
By using the StartEXE macro, we can run any executable file, in particular,
created in Visual Basic Environment, which is a part of Microsoft Visual Studio
(Section 1.24). For that, c:\Windows\notepad.exe must be replaced by
the full name (without spaces) of this executable file.

135
 Chapter 2. Programming in VBA

2.2. Layout of the control elements

on the Excel worksheet

The control elements of Section 1.23 can be placed on the Excel worksheet.
For example, to create the form-loading button, we must fulfill the following.
1. In the Excel window with open workbook BookForm, fulfill Developer >
Insert in area Controls. The window with control elements appears (Fig. 2.2).

Fig. 2.2. The Excel window with control elements below the Insert button

2. Click on the Button element in the Form Controls area of the last win-
dow, and then click on any cell of the Excel worksheet. The Assign Macro win-
dow opens (Fig. 2.3).
3. Click on the New button. The VB window opens (Fig. 2.4), and button
Button1 selected by 6 markers appears on the worksheet (Fig. 2.5).
4. Insert operator UserForm1.Show into the program blank in the code
window depicted in Fig. 2.4. The following program is the result:

Sub Button1_Click()
UserForm1.Show
End Sub

5. Click on area outside Button1 to remove the button selection.

136
 2.2. Layout of the control elements on the Excel worksheet

Fig. 2.3

Fig. 2.4. The VB window with Module1 and Module2 in the project window

137
 Chapter 2. Programming in VBA

Fig. 2.5. The created button on the worksheet

Now, for loading the form pictured in Fig. 1.31, all we have to do is click on
button Button1.
To change the inscription on the button:
1) select the button as follows:
right click on it;
then perform the Cancel command in the context menu opened;
2) edit the inscription as usual text;
3) click on area outside the button to remove its selection.
To remove the selected button, we must press the Delete key.
The Excel worksheet with the control elements plays the role of the form. For
beauty of this “form”, it is possible to remove gridlines from the Excel work-
sheet. For that, in the Excel window, we must fulfill View > turn off Gridlines in
area Show.
If we want to use the button for opening the Notepad window, program
Button1_Click must be changed as follows:

Sub Button1_Click()
Dim RetVal As Integer 'for function Shell
RetVal = Shell("c:\Windows\notepad.exe",1)
End Sub

In this program, we can replace c:\Windows\notepad.exe by the full

name of any other executable file (p. 135). In this case, the above program is
suitable for running this executable file from the Excel window.

138
 2.3. User-defined functions of Excel

2.3. User-defined functions of Excel

Our study of Visual Basic began with the Pythagoras program (p. 17)
intended for calculating the hypotenuse length. Let us consider the following
function for solving this simple task:

Function Hypotenuse(a, b)
Hypotenuse = Sqr(a ^ 2 + b ^ 2)
End Function

This function declaration should be entered into the code window after insert-
ing a module, for example Module1, into the active Excel workbook. As a result,
the Hypotenuse function appears in the User Defined category of the Excel
functions library. To verify this, we must fulfill the following two operations:
1) click on the fx button of the Excel formula bar;
2) in the Insert Function window opened (the first window of Function
Wizard), enter User Defined into box Or select a category by means of the drop-
down list.
We see line Hypotenuse in list Select a function (Fig. 2.6), i.e., the considered
function is available in category User Defined of the Excel functions library.
Let us interrupt the operation of Function Wizard by clicking on the Cancel
button in the Insert Function window.
The created function is used, for example, as follows:
1) into two cells on the Excel worksheet, enter the lengths of the triangle’s
legs, for example, 30.02 and 40 into A2 and B2, respectively;
2) select a cell for the hypotenuse length, for example, C2;
3) in the Excel formula box after =, put the Hypotenuse function whose
arguments are the addresses of the cells with the lengths of the triangle’s legs
(we must type a semicolon between the arguments instead of a comma accepted
in Visual Basic);
4) calculate in one of the following two ways:
by clicking on the tick button of the Excel formula bar;
by pressing the Enter key.
As a result, the hypotenuse length appears in the selected cell (Fig. 2.7).

139
 Chapter 2. Programming in VBA

Fig. 2.6. The first window of Function Wizard

Fig. 2.7. The use of the Hypotenuse function

While using Function Wizard, we have to click on OK in the Function Argu-

ments window (the second window of Function Wizard) for calculation.

140
 2.3. User-defined functions of Excel

The second window of Function Wizard is intended for inputting arguments

of a function, in particular, of a user-defined function. This window appears as
a result of clicking on OK in the first window of Function Wizard (Fig. 2.6). We
will use the second window of Function Wizard in Section 2.15 (Fig. 2.32).
User-defined functions are used in Excel formulas, just like built-in func-
tions.
The Hypotenuse function can be considered as the user-defined function
of Visual Basic (p. 77). Therefore, we can use the Hypotenuse function in
a program as follows:

Sub Pythagoras()
Dim a As Single
Dim b As Single
Dim c As Single
a = 3
b = 4
c = Hypotenuse(a, b)
End Sub

This text must be put into a new module, for example, by name Module2.
By executing the last program step-by-step, we can verify its operational
capability.
In Excel, when transposing a matrix (cell range) relative to the main diagonal
(Section 1.21), its rows and columns interchange their positions. For such trans-
position, Excel has the built-in TRANSPOSE function, which is available in
category Lookup & Reference of the functions library. It is an example of the
function returning an array of numbers, instead of one number (the SQRT func-
tion returns a number, see p. 17).
The TRANSPOSE function is used as follows:
1) enter a matrix (containing, for example, 3 rows and 4 columns) into Excel
cells;
2) select the cell range (containing 4 rows and 3 columns) for the transposi-
tion result;
3) into the Excel formula box after =, enter the TRANSPOSE function
whose argument is the source range (containing 3 rows and 4 columns);
4) calculate in one of the following two ways:
by clicking on the tick button of the Excel formula bar when keys Ctrl
and Shift are simultaneously pressed;
by pressing the Enter key when keys Ctrl and Shift are simultaneously
pressed (that is, by pressing Ctrl + Shift + Enter).
The transposition result appears in the selected cell range.

141
 Chapter 2. Programming in VBA

Function Wizard simplifies the use of the TRANSPOSE function. For trans-
posing the matrix, we have to click on OK in the Function Arguments window
when keys Ctrl and Shift are simultaneously pressed.
It is interesting to create a user-defined function for transposing a matrix (cell
range) relative to its auxiliary diagonal according to formula (1.3) on p. 109.
This function’s declaration follows:
Listing 2.1
Function TRANSPOSEA(massive As Variant) As Variant
Dim m As Integer, n As Integer
Dim i As Integer, j As Integer
Dim R() As Variant 'resulting matrix
m = massive.Rows.Count 'quantity of rows
n = massive.Columns.Count 'quantity of columns
ReDim R(1 To n, 1 To m) 'specification of size
For j = 1 To n
For i = 1 To m
R(j, i) = massive(m + 1 - i, n + 1 - j)
Next i
Next j
TRANSPOSEA = R
End Function

Formal parameter massive may be considered as a record (Section 1.18).

More precisely, massive is a variable of the Range type, but we will talk about
it later.
Upon putting Listing 2.1 into the code window, the TRANSPOSEA function
appears in category User Defined of the Excel functions library.
TRANSPOSEA is used as the TRANSPOSE function:
1) enter a matrix (intended for transposing relative to its auxiliary diagonal)
into Excel cells, for example, A1:D3;
2) select the cell range for the transposition result, for example, A5:C8;
3) into the Excel formula box, enter

=TRANSPOSEA(A1:D3)

4) calculate in one of the following two ways:

by clicking on the tick button of the Excel formula bar when keys Ctrl
and Shift are simultaneously pressed;
by pressing Ctrl + Shift + Enter.
The result of transposing the matrix relative to its auxiliary diagonal appears
in the selected A5:C8 range (Fig. 2.8).

142
 2.3. User-defined functions of Excel

Fig. 2.8. The result of the TRANSPOSEA function usage

While using Function Wizard for the matrix transposition, we have to click
on OK in the Function Arguments window when keys Ctrl and Shift are simulta-
neously pressed.
It is interesting to compare the TRANSPOSEA function with the program of
Section 1.22. By the number of operators, the function is four times shorter than
the program. Besides, the function can be used when creating Excel macros by
means of Excel Macro Recorder (Section 2.5).
Note that the name of the formal parameter of the TRANSPOSEA function
can differ from massive, for example, name diapason may be used.
When operating with the matrices containing a large number of columns, it is
convenient to use the R1C1 reference style, in which the Excel columns are
numbered by natural numbers instead of letters.
For setting the R1C1 reference style, we must fulfill the following operations:
1) click on the File button in Excel;
2) Options > Formulas;
3) turn on option R1C1 reference style in area Working with formulas of the
Excel Options window;
4) click on the OK button.

143
 Chapter 2. Programming in VBA

2.4. Two methods for developing Excel macros

An Excel macro is a set of operators, which can be executed automatically.

The macro is written in a programming language called Visual Basic for Appli-
cations or VBA.
The VBA programming language includes the reviewed constructs of Visual
Basic.
The following two methods are used for creating macros:
1) programming in Visual Basic Environment;
2) creating by means of Excel Macro Recorder.
We used the first method in the previous sections, in particular, for develop-
ing two macros for loading the UserForm1 form (Sections 2.1 and 2.2).
The second method is simpler but less universal in comparison with the first
method; it is normally considered when learning Excel. We will consider the
second method because Excel Macro Recorder is an excellent helper in the first
method for creating macros: if we do not know how to write down any action
(or set of actions) in VBA, we have to create a macro for performing this action
(or set of actions) by using Excel Macro Recorder and then study the VBA code
generated automatically.
This is how we will develop:
the operator blocks for automatic creation of graphs in Listings 3.10, 3.13
and 3.18;
operator 12 in Listing 6.15 for the horizontal center alignment of the
Excel cell content.
Excel Macro Recorder also facilitates the use of the first method for creating
macros. The fact is that the program in VBA can be created not from scratch, but
by starting from a prototype, which was created by means of Excel Macro
Recorder.
It is precisely in this way that we will develop the graph subroutine for
automatic creation of graphs (Section 4.8).
Creating macros by editing the prototype leads to a considerable “thought
saving”.

144
 2.5. Excel Macro Recorder

2.5. Excel Macro Recorder

Excel Macro Recorder operation is similar to recording by means of a video-

disk recorder. Excel Macro Recorder:
records the operations being fulfilled by the user in the Excel window;
then transforms these operations into a set of VBA operators, that is, into
a program.
By means of Excel Macro Recorder, we will create a macro for performing
the following operational sequence:
1) removal of gridlines from the Excel worksheet;
2) setting the R1C1 reference style;
3) assignment of the Currency format to all cells on the worksheet;
4) selection of the R14C5 cell, that is, E14.
For creation of the macro, we fulfill the following operations.
1. Developer > Record Macro (in area Code) or View > arrow Macros
(in area Macros) > Record Macro.
2. Enter a name of the macro and other information into text boxes of the
Record Macro window opened:
Macro name: MR
Shortcut key: Ctrl+m (it will be used for starting the macro execution)
Store macro in: This Workbook (taken from the drop-down list)
Description: Result of using Excel Macro Recorder
3. In the window with the filled text boxes (Fig. 2.9), click on button OK.
At that, the Record Macro button changes its status to Stop Recording. Macro
record is started.
4. Remove gridlines from the active Excel worksheet, for example, by name
Sheet1: View > turn off Gridlines in area Show.
5. Set the R1C1 reference style as follows:
1) click on the File button;
2) Options > Formulas;
3) turn on the R1C1 reference style option in area Working with formulas of
the Excel Options window;
4) click on the OK button.
6. Assign the Currency format to all cells on the active worksheet as
follows:

145
 Chapter 2. Programming in VBA

1) select all cells on the worksheet by clicking on the intersection of the top
line (with numbers of columns) and the left column (with numbers of rows);
2) activate the Home tab;
3) in area Number, set the Currency format by using the drop-down list.
7. Select the R14C5 cell by clicking on it.
8. Developer > Stop Recording (in area Code) or View > arrow Macros
(in area Macros) > Stop Recording.

Fig. 2.9. The Record Macro window before clicking on the OK button

The record is finished; macro MR is the result.

Before verifying the operational capability of the created macro, we return to
the A1 reference style. For that:
1) click on the File button;
2) Options > Formulas;
3) turn off the R1C1 reference style option in area Working with formulas of
the Excel Options window;
4) click on button OK.
Before starting the macro, let us create a worksheet, for example Sheet2, for
which operations written in the macro must be performed. For creating Sheet2,
we use the New Sheet command, depicted by the round plus icon in the bottom
part of the Excel window (see in Fig. 2.5).

146
 2.5. Excel Macro Recorder

For starting the MR macro from the Excel window, one of the following two
ways is used:
simultaneous pressing keys Ctrl and m;
Developer (or View) > Macros > line MR > Run.
When executing the macro, the computer repeats actions enumerated in items
4 — 7.
After restoring the A1 reference style, let us save the Excel workbook with
the MR macro (which is the result of using Excel Macro Recorder) under the
name of BookMacrorecorder. During the save, we have to set the following file
type: Excel Macro-Enabled Workbook.
For removing the MR macro, we must fulfill the following operations:
Developer (or View) > Macros > line MR > Delete > Yes. However, we will not
delete the macro because it will be required in the next sections.

147
 Chapter 2. Programming in VBA

2.6. VBA code generated by Excel Macro Recorder

and its editing

In the previous section, we created the VBA macro corresponding to the

sequence of Excel operations performed with Excel Macro Recorder turned on.
For displaying the macro text, we must fulfill the following in the Excel win-
dow with workbook BookMacrorecorder: Developer (or View) > Macros > line
MR > Edit. At that, the VB window appears with the code window containing
the required text.
The macro code follows:
Listing 2.2
Sub MR()
'
' MR Macro
' Result of using Excel Macro Recorder
'
' Keyboard Shortcut: Ctrl+m
'
ActiveWindow.DisplayGridlines = False
Application.ReferenceStyle = xlR1C1
Cells.Select
Selection.NumberFormat = "#,##0.00$"
Range("E14").Select
End Sub

There is the possibility of editing this code created by using Excel Macro
Recorder.
Let us assume that we want to set content of cell R14C5 (that is, E14) by
means of the standard window. For that, we type the following operator above
the last line of Listing 2.2:

ActiveCell.Formula = _
InputBox("Enter price in dollars" _
& vbCrLf & "into the active cell")

148
 2.6. VBA code generated by Excel Macro Recorder and its editing

Fig. 2.10 shows the expanded text of the MR macro.

Let us create the Sheet3 worksheet and start the MR macro with the additional
operator when Sheet3 is active. At that, the window represented in Fig. 2.11
appears.

Fig. 2.10. The VB window with the macro text

after typing the additional operator

Fig. 2.11. The standard window for inputting

information into the selected cell

149
 Chapter 2. Programming in VBA

We put a value, for example 12, into text box Enter price in dollars into the
active cell. After clicking on the OK button, the text box content appears in cell
R14C5 (Fig. 2.12). The dollar sign in front of 12.00 is because of the Currency
format of the cell.

Fig. 2.12. The result of the MR macro execution

Let us restore the A1 reference style.

Further, we will consider constructs of VBA, which are absent in Visual
Basic.

150
 2.7. Objects and events

2.7. Objects and events

All modern programming languages (in particular, Visual Basic and VBA)
are object- and event-oriented.
The object orientation is based on partitioning the subject area (for which we
are developing a program) and clustering the parts.
The program’s part, which corresponds to the cluster, is named object.
The object’s characteristics are named properties.
Actions performed over the object are named methods.
Examples of the subject area are problems of modeling financial risks, semi-
conductor devices and evolution of stars (p. 47), as well as Visual Basic and
VBA themselves.
Examples of the object are the Visual Basic objects, which were considered
in Section 1.23: UserForm, Label, TextBox, CommandButton and CheckBox.
The event orientation is based on the following concepts:
“event” — the object’s qualitative change, which follows from work of
the user or computer;
“handler for event” — the command set, executed by the computer when
the event occurs.
An example of the event is the click on the Account button in the user-
defined form, for example, depicted in Fig. 1.28. Because of handling this event,
the ComButCalc_Click program execution is started.
The VBA programming language is intended for creating programs in the
Microsoft Office applications, such as word processor Word, tabular processor
Excel, technical editor Visio, database management system Access, etc. VBA
differs from Visual Basic in the presence of specific objects of Microsoft Office
and of its applications.
We will be interested in the so-called Excel objects intended for Excel VBA
Programming. Examples of such objects are Workbook, Worksheet, Range and
ActiveCell.
Operators intended for work with an object make the following:
setting object properties;
returning object properties into the program;
applying object methods.

151
 Chapter 2. Programming in VBA

The syntax of setting object properties follows:

object.property = expression

where expression is an arithmetic or logical expression or string. The com-

puter executes this operator as the normal assignment operator:
1) calculates the value of expression;
2) assigns this value to property object.property.
The syntax of returning object properties follows:

object.property

Property object.property may be a part of operators, in particular, it

may be in the right-hand side of the assignment operator, i.e., the property is
similar to built-in functions of Visual Basic. Often, object.property itself
is an object.
Note that not all object properties can be returned and set. There are proper-
ties, which can be only returned or set. To study possibilities of this or that object
property, we must use the reference systems, which are started by pressing the
F1 and F2 keys when the VB window is active (Sections 1.6 and 2.12).
The operator of applying object methods has the following form:

object.method

The operator of applying the Add method is the exception. It creates a new
object, subobject, and adds it to object. The syntax of this operator
follows:

[Set variable = ]object.Add

In this syntax, Set is the keyword, variable is a variable of the same data
type as subobject.
Applying the Add method is similar to calling the MsgBox procedure of
Visual Basic, which is both a subroutine and function.
An object hierarchy exists. The highest in the hierarchy of the application is
the Application object, i.e., all other objects “are included” in it. The Application
object reminds the Russian nested doll, but (unlike the nested doll) several
objects may be included in each object.
The full object name is a sequence of the object names separated by a point,
at that, this sequence begins with Application. For example,

Application.Workbooks("Archive").Worksheets("Cod"). _
Range("A1")

152
 2.7. Objects and events

in VBA for Excel is the full name of the Range("A1") object or the reference
to the A1 cell on the Cod worksheet of the Archive workbook.
The use of the full object name is not necessary. Often, we can use the
incomplete name, i.e., without names of the objects activated at present. For
example, if the Archive workbook is active, then the full name of the
Range("A1") object may be shortened as follows:

Worksheets("Cod").Range("A1")

As before, this is the reference to the A1 cell on the Cod worksheet of the
Archive workbook.
In the above VBA notations, strings are in the parentheses. These strings may
be compound.
An example of using compound strings is the following operator:

Range("G" & CStr(CInt(Now - #1 Jan 2000#))).Select

Because of its execution, the following cell is selected on the active Excel work-
sheet: the intersection of the G column and the row whose number is equal to the
number of days from the century beginning.
We advise the reader to do the following:
1) type the last operator above line End Sub of program Century_20
(p. 25);
2) execute the obtained program;
3) pay attention to the position of the Excel cell activated.
We see that the object construct is similar to the record (Section 1.18). There-
fore, as the first approximation, the object can be considered as the built-in
record whose creation operator (the Type operator) is hidden from the program
developer.
In addition to properties and methods, some objects of VBA are also charac-
terized by events. However, it does not relate to the main Excel objects, which
will be considered below.

153
 Chapter 2. Programming in VBA

2.8. Object Application

It was mentioned above that the Application object occupies the top level in
the object hierarchy of Excel. This means that the Application object controls the
settings of the application, i.e., such settings as are in window Excel Options
(to open this window, we must fulfill File > Options). Operator

Application.ReferenceStyle = xlR1C1

in the MR macro text, which was created by means of Excel Macro Recorder (see
Listing 2.2 on p. 148), speaks about this role of the Application object.
However, the Application object not only changes parameters of Excel. If we
want to use Excel functions when programming in VBA, the Application object
is also necessary for us.
Let us use built-in functions AVERAGE and SUM of Excel for processing
range A1:A4 on the Sheet1 worksheet.
1. Insert a module into the active Excel workbook. Enter program
Listing 2.3
Sub BuiltinFunctions()
Dim W As Single
1: W = Application. _
Average(Worksheets("Sheet1").Range("A1:A4"))
MsgBox "Average = " & CStr(W)
2: W = Application. _
Sum(Worksheets("Sheet1").Range("A1:A4"))
MsgBox "Sum = " & CStr(W)
End Sub

into the code window.

2. Go to worksheet Sheet1 of the active workbook.
3. Enter number 100 into cell A1, 200 into A2, 300 into A3 and 400 into A4.
4. Run the BuiltinFunctions macro. The window with the average of
the entered numbers appears (Fig. 2.13).

154
 2.8. Object Application

5. To continue the program execution, click on the OK button. The window

containing the sum of the entered numbers appears (Fig. 2.14).
6. To terminate the program execution, click on the OK button.

Fig. 2.13. The window with the first processing result

Fig. 2.14. The window with the second processing result

The built-in Excel functions are properties of the Application object. When
executing operators 1 and 2, properties Average and Sum return with parameter
Worksheets("Sheet1").Range("A1:A4").
In addition to Excel functions, the Application object has other properties.
From the long list of the properties, we will consider the following.
 ActiveWorkbook — the active workbook.
 ActiveSheet — the active worksheet of the active workbook (in our opin-
ion, ActiveWorksheet would be a more suitable name of this property).
 ActiveCell — the active cell on the active worksheet of the active work-
book.
As an example of returning the ActiveCell property, let us consider the fol-
lowing program, which sets the italic font for the active cell and puts text Report
for May into this cell.

155
 Chapter 2. Programming in VBA

Sub ItalicFont()
With Application.ActiveCell
.Font.Italic = True
.Value = "Report for May"
End With
End Sub

The With operator was defined on p. 91.

Properties ActiveWorkbook, ActiveSheet and ActiveCell are objects. We will
consider them in greater detail in the next three sections.
 Calculation — the calculation mode (see area Calculation options of the
Excel Options window after fulfilling File > Options > Formulas).
The main values of this property are:
1) xlCalculationAutomatic — the automatic calculation: recalcula-
tion according to formulas is performed automatically when Excel cells’ contents
change; it is the default operation mode of Excel;
2) xlCalculationManual — the manual calculation, for example,
when pressing the F9 key.
As an example of setting the Calculation property, let us consider the follow-
ing operator intended for setting the automatic calculation mode:

Application.Calculation = xlCalculationAutomatic

 Dialogs — the collection of the Excel dialog boxes.

In VBA, word “collection” means a group of single-type objects.
By means of the Dialogs property, it is possible to display the Excel dialog
boxes. For example, the execution of operator

Application.Dialogs(xlDialogOpen).Show

leads to displaying window Open familiar to us; xlDialogOpen is the

parameter of the Dialogs property.
In the last operator, Application.Dialogs(xlDialogOpen) is the
object (of the Balloon type), Show is the object’s method.
We already encountered the Show method in Sections 2.1 and 2.2. It is
operator

UserForm1.Show

figuring in the programs of loading the user-defined form.

Let us use the Show method for saving an Excel workbook.

156
 2.8. Object Application

1. Above the last line of the BuiltinFunctions macro (p. 154), insert
the following operator of displaying the document saving window:

Application.Dialogs(xlDialogSaveAs).Show

The macro has the following expanded text:

Sub BuiltinFunctions()
Dim W As Single
1: W = Application. _
Average(Worksheets("Sheet1").Range("A1:A4"))
MsgBox "Average = " & CStr(W)
2: W = Application. _
Sum(Worksheets("Sheet1").Range("A1:A4"))
MsgBox "Sum = " & CStr(W)
Application.Dialogs(xlDialogSaveAs).Show
End Sub

2. Go to the Excel window.

3. Enter arbitrary numbers, for example 100, 200, 300 and 400, into range
A1:A4 on the Sheet1 worksheet.
4. Run the above macro (the windows, depicted in Fig. 2.13 and 2.14, will
appear during the execution).
5. In the displayed Save As window, enter file name BookApplication.
6. Set file type Excel Macro-Enabled Workbook by means of drop-down list
Save as type;
7. Click on the Save button.
8. Make sure that the obtained BookApplication workbook contains macro
BuiltinFunctions and the file name has extension .xlsm.
On p. 108, we considered the order of tuning Windows Explorer to see file
names with extension.
In the above ItalicFont macro, operator With was used for setting ob-
ject properties. However, we can use this operator for applying object methods.
For example, operator

Application.Dialogs(xlDialogSaveAs).Show

can be replaced by construct

With Application.Dialogs(xlDialogSaveAs)
.Show
End With

157
 Chapter 2. Programming in VBA

The With operator may include both properties and methods. As an exam-
ple, let us consider the following new version of the BuiltinFunctions
macro:

Sub BuiltinFunctions()
Dim W As Single
1: W = Application. _
Average(Worksheets("Sheet1").Range("A1:A4"))
MsgBox "Average = " & CStr(W)
2: W = Application. _
Sum(Worksheets("Sheet1").Range("A1:A4"))
MsgBox "Sum = " & CStr(W)
With Application
.ActiveCell.Value = "Report for May"
'property Value
.Dialogs(xlDialogSaveAs).Show 'method Show
End With
End Sub

During the execution, text Report for May appears in the active cell.
From the list of methods of the Application object, we will consider the fol-
lowing: Quit, Calculate, OnTime.
 Quit — quit Excel.
The operator of applying the Quit method follows:

Application.Quit

 Calculate — the forced calculation.

As examples of using the Calculate method, let us consider the following
operators:

Application.Calculate
Worksheets("Report7").Calculate
Worksheets("Report7").Range("A1:C10").Calculate

If mode “manual calculation” is set in Excel, then:

the execution of the first operator leads to the calculation according to the
formulas in all open Excel workbooks (which are represented by buttons on the
taskbar of Windows Desktop); this execution is equivalent to pressing key F9;
the execution of the second operator leads to the calculation on the
Report7 worksheet of the active workbook;

158
 2.8. Object Application

the execution of the third operator leads to the calculation in the A1:C10
range on the Report7 worksheet of the active workbook.
As we know, the execution of operator

Application.Calculation = xlCalculationManual

tunes Excel for the manual calculation.

To tune Excel for the manual calculation, we can also fulfill the following
operational sequence in the Excel window: File > Options > Formulas > turn on
Manual > OK.
 OnTime — the start of the macro at the given moment of time; the time
and the macro name are the method parameters.
As an example of applying the OnTime method, let us consider the following
macro, which writes the current time into cell A1 (on the active worksheet) after
starting the macro.
Listing 2.4
Sub MyMacro()
1: Range("A1") = Time
2: Application.OnTime Now + TimeValue("00:00:01"), _
"MyMacro"
End Sub

In the MyMacro macro, operator 1 writes the current time into the A1 cell.
Operator 2 starts MyMacro one second after the current moment; therefore,
operator 1 is executed every second, i.e., the A1 cell content is updated every
second.
For the initial start of MyMacro, we must fulfill Developer (or View) >
Macros > line MyMacro > Run.
Let us depict the time starting from the moment of opening the Excel work-
book (containing the corresponding macro). For this purpose, we use name
Auto_Open instead of MyMacro. The new version of the last program follows:

Sub Auto_Open()
1: Range("A1") = Time
2: Application.OnTime Now + TimeValue("00:00:01"), _
"Auto_Open"
End Sub

We save the workbook, containing this macro, as a macro-enabled workbook,

for example, by name BookOnTime. After that, we close the BookOnTime
workbook.

159
 Chapter 2. Programming in VBA

Further, let us fulfill the following:

1) open the BookOnTime workbook;
2) to allow the macro to work, click on the Enable Content button of the
Security Warning panel.
At that, the Auto_Open macro is started automatically. So is happened
thanks to the macro name (Excel so is arranged).
After the auto start, the Auto_Open macro is working as MyMacro:
every second, Auto_Open starts;
therefore, every second, operator 1 writes the current time into cell A1 on
the active worksheet.
For consolidating the material of this section, we advise the reader to under-
stand the work of the following code, which contains the Auto_Open macro
with two operators of applying the OnTime method.
Listing 2.5
Sub Auto_Open()
Application.OnTime TimeValue("12:30:00"), "MyMa1"
Application.OnTime TimeValue("12:31:00"), "MyMa2"
End Sub

Sub MyMa1()
Range("G1") = 13.333
End Sub

Sub MyMa2()
Range("G2") = Time
Application.OnTime Now + TimeValue("00:00:01"), _
"MyMa2"
End Sub

These three macros should be put into one module or different modules of the
same Excel workbook.

160
 2.9. Objects Workbook, Workbooks and ActiveWorkbook

2.9. Objects Workbook, Workbooks

and ActiveWorkbook

Object Workbook follows right after object Application in the object hierar-
chy. It is easy to understand the following properties of the Workbook object.
 Name — the workbook name with its extension.
The possible extensions of the name follow:
.xlsm if the workbook contains macros;
.xlsx if macros are absent in the workbook, etc.
The file name together with its extension frequently is also called the file name.
As a rule, it does not lead to any confusion.
 Path — the path to the workbook in the file system of Windows.
 FullName — the workbook name with its path and extension, i.e., the
workbook full name.
Let us consider the following three methods of the Workbook object:
 Close — closing the workbook;
 Save, SaveAs — saving the workbook.
The SaveAs method differs from the Save method in that SaveAs has a list of
optional parameters, which includes FileName, FileFormat and Password.
Examples of using the properties and methods of the Workbook object are
given below, when considering objects of the Workbook data type. Here, “data
type” has the same sense as in expression “variables (records) of the Session data
type” on p. 91.
Objects of the Workbook type (“data” is omitted for brevity) have the proper-
ties and methods of the Workbook object.
The Workbooks object is an object containing all open Excel workbooks.
This object is also named as the Workbooks collection.
In VBA,

Workbooks("Personnel_department")

is the Workbook type object corresponding to the open workbook by name

Personnel_department.
Let us consider the main methods of the Workbooks object.
 Activate — activation of the specified workbook (from a number of the
open Excel workbooks) when its first worksheet becomes active.

161
 Chapter 2. Programming in VBA

As an example of applying the Activate method, let us consider the following

operator of activating the above Personnel_department workbook:

Workbooks("Personnel_department").Activate

 Add — the creation of a new workbook, which becomes active at once.

Applying the Add method is accompanied by return (into the program) of
the created workbook, which is the Workbook type object. As an example, see
operator 2 in the NewBook macro (p. 164).
Note that all collections have the Add method. Its application adds (to the
collection) a new object of the corresponding type. In particular, a new object of
the Workbook type is added to the Workbooks collection.
All collections, in particular Workbooks, have useful property by name
Count that allows determining the quantity of objects in the collection.
An example of returning the Count property is in the following program:
Listing 2.6
Sub NumberofBooks()
MsgBox Str(Workbooks.Count)
End Sub

The window, containing the number of open workbooks and the OK button,
is displayed when executing the NumberofBooks macro. To terminate the
execution, we must click on OK.
It was mentioned in the previous section that ActiveWorkbook is a property
of the Application object, corresponding to the active workbook. As the Work-
book type object, the ActiveWorkbook property has the properties and methods
of the Workbook object.
To obtain examples of returning properties Name and FullName of the
Workbook object and of applying the Save method, let us fulfill the following:
1) in the Excel window with the BookMacrorecorder workbook (p. 147),
click on the Enable Content button of the Security Warning panel to allow the
MR macro to work;
2) Developer (or View) > Macros > line MR > Edit;
3) put operator block

1: Dim S As String
2: S = ActiveWorkbook.Name
3: MsgBox S
4: MsgBox ActiveWorkbook.FullName
5: ActiveWorkbook.Save

162
 2.9. Objects Workbook, Workbooks and ActiveWorkbook

above the last line of the MR macro (Fig. 2.15);

4) start the MR macro execution, for example, by clicking on arrow ► in the
VB window.

Fig. 2.15. The VB window with the macro text

after inserting the additional operator block

Because of executing operator 2, the S string has a value that is equal to the
name of the active workbook. Operator 3 displays this string (Fig. 2.16).
Let us click on the OK button (in the window depicted in Fig. 2.16) to con-
tinue executing the MR macro. Operator 4 displays the active workbook’s full
name (Fig. 2.17). After clicking on the OK button, operator 5 saves the active
workbook.
To obtain an example of returning the Path property of the Workbook object,
we will execute the MR macro with the following form of the operator intended
for displaying the full name:

4: MsgBox ActiveWorkbook.Path & "\" & _

ActiveWorkbook.Name

163
 Chapter 2. Programming in VBA

The window, depicted in Fig. 2.17, appears during the execution, as in the case
of the previous version of operator 4.

Fig. 2.16. The window with the active workbook’s name

Fig. 2.17. The window with the full name of the active workbook

In the following example, a workbook is created, a numerical value is entered

into it, this workbook is saved and closed, and the Excel window is closed too.
1. Insert a module into a blank workbook, for example Book1, and put the
following text into the code window (Fig. 2.18):
Listing 2.7
Sub NewBook()
1: Dim wbNewWorkbook As Workbook
2: Set wbNewWorkbook = Workbooks.Add
3: wbNewWorkbook.Worksheets("Sheet1").Range("A1"). _
Value = 100
4: wbNewWorkbook.SaveAs _
"c:\Users\usr\Hour0.xlsx"
5: wbNewWorkbook.Close

164
 2.9. Objects Workbook, Workbooks and ActiveWorkbook

6: MsgBox "Workbook is closed"

7: Application.Quit
End Sub

2. Start the NewBook macro execution.

3. When the window with message Workbook is closed appears (Fig. 2.19),
click on the OK button in it. At that, the NewBook macro closes the Excel win-
dow. When closing the Excel window, we may disagree with the offer to save
Book1.
4. Open the Hour0 workbook in folder

c:\Users\usr

Number 100 is in cell A1 on worksheet Sheet1.

Fig. 2.18. The VB window with the NewBook macro text

The NewBook macro begins with the declaration of the wbNewWorkbook

variable (operator 1). Because of executing operator 2, a workbook is created
and assigned to the wbNewWorkbook variable. Operator 3 puts number 100
into the A1 cell on the Sheet1 worksheet of this workbook.
When creating a workbook, tabular processor Excel gives it a default name,
for example, Book2. Because the workbook name is known inexactly, the
wbNewWorkbook variable is used instead of Workbooks("Book2") in the
macro. Operator 4 saves this workbook under the following full name:

c:\Users\usr\Hour0.xlsx.

165
 Chapter 2. Programming in VBA

Operator 5 closes the Hour0 workbook, and operator 6 displays the message
about it (Fig. 2.19). Operator 7 performs exit from Excel.
In operator 1, we may replace the Workbook type by the Object data type
(Appendix 1) similar to the Variant data type familiar to us.

Fig. 2.19. The window with the OK button for

terminating the NewBook macro execution

166
 2.10. Objects Worksheet, Worksheets and ActiveSheet

2.10. Objects Worksheet, Worksheets

and ActiveSheet

The Worksheet object follows Workbook in the object hierarchy. Let us con-
sider the following properties of object Worksheet:
 Name — the worksheet name;
 Cells — one of the following:
the collection of all cells on the worksheet;
the single cell if we specify (by two integers in parentheses, through a
comma) the numbers of row and column whose intersection defines this cell.
The Cells property will be also considered in the next section, as applied to
the Range object.
We will consider the following two methods of the Worksheet object:
 Activate — activating the worksheet;
 Delete — deleting the worksheet.
Examples of using the properties and methods of the Worksheet object are
given below, when considering objects of the Worksheet type. The last objects
have the properties and methods of the Worksheet object.
The Worksheets object is an object containing all worksheets of the Excel
workbook. This object is also named as the Worksheets collection.
Its number or name can identify each worksheet of the Worksheets object.
For example, Worksheets(1) designates the 1st worksheet of the workbook, and
Worksheets(“Sheet1”) is the worksheet by name Sheet1. Worksheets(1) and
Worksheets(“Sheet1”) are objects of the Worksheet type.
As examples of applying the Activate and Delete methods of the Worksheet
object, let us consider the following operators:

Worksheets(3).Activate
Worksheets("Sheet2").Delete

The first operator activates the 3rd worksheet of the active workbook, and the
second operator deletes worksheet by name Sheet2 of the active workbook.
When working with object Worksheets, the application of the Add method
adds to the collection a new worksheet that becomes active at once. This applica-
tion is accompanied by return (into the program) of the created worksheet, which

167
 Chapter 2. Programming in VBA

is the Worksheet type object. As an example, see operator 2 in the NewSheet

macro on p. 170.
The Worksheets collection (as well as other collections) has the Count
property that allows determining the number of objects in the collection.
An example of returning property Count is in the following macro:
Listing 2.8
Sub NumberofSheets()
MsgBox Str(Worksheets.Count)
End Sub

The window, containing the number of worksheets in the active Excel work-
book and the OK button, is displayed when executing the NumberofSheets
macro (Fig. 2.20). For terminating the execution, we have to click on OK.

Fig. 2.20

As we already mentioned in Section 2.8, ActiveSheet is a property of the

Application object, corresponding to the active worksheet of the active work-
book. As the Worksheet type object, the ActiveSheet property has the properties
and methods of the Worksheet object.

168
 2.10. Objects Worksheet, Worksheets and ActiveSheet

An example of setting the Name property of the Worksheet object is the fol-
lowing operator:

ActiveSheet.Name = "July"

This operator assigns the month name to the active worksheet.

As an example of returning the Cells property of the Worksheet object, let us
consider the following program:
Listing 2.9
Sub Color()
Dim i As Integer
Dim j As Integer
For i = 1 To 4
For j = 1 To 5
0: With ActiveSheet.Cells(i, j)
If .Value < 0 Then
.Font.Color = QBColor(10)
.Font.Italic = True
End If
End With
Next j
Next i
End Sub

This program:
1) examines the values in cells A1:E4 on the active worksheet;
2) sets the green italic font when the value is negative.
Note that the object name can be excluded from VBA notation

ActiveSheet.Cells(i, j)

In this case, line 0 becomes as follows:

With Cells(i, j)

Besides, note that operator Cells.Select in macro Listing 2.2 on p. 148

(programmed by means of Excel Macro Recorder) is equivalent to operator
ActiveSheet.Cells.Select.
As an example of applying the Add method of the Worksheets object, let us
consider the program for inserting a new worksheet into the active Excel work-
book. For that, we fulfill the following:

169
 Chapter 2. Programming in VBA

1) put text
Listing 2.10
Sub NewSheet()
1: Dim wsNewWorksheet As Worksheet
2: Set wsNewWorksheet = Worksheets.Add
3: wsNewWorksheet.Name = Format(Date, "d mmmm yyyy")
End Sub

into the code window;

2) execute the NewSheet macro.
The name of the inserted worksheet is the current date. This worksheet is
active; it is placed in front of the former active worksheet.
The NewSheet macro (which is similar to NewBook on p. 164) begins with
the declaration of variable wsNewWorksheet (operator 1). The created work-
sheet is assigned to this variable (operator 2). Operator 3 sets the worksheet
name by means of the wsNewWorksheet.Name property.
Operator 3 includes the call of the Format function intended for converting
a value (in particular, of the Date data type) to a string of the given form. The
first argument is the Date function returning the current date. The second argu-
ment is the string, which gives the following format of the date: the day of
month, the name of month, the year completely.
Note that the Set keyword is used not only in the operator of applying the
Add method: any assignment operator, where objects are present, begins with the
Set keyword. For example, the NewSheet program will also work if we
replace Worksheets.Add by object Worksheets(1) in operator 2. The
new version of the program follows:

Sub NewSheet1()
1: Dim wsNewWorksheet As Worksheet
2: Set wsNewWorksheet = Worksheets(1)
3: wsNewWorksheet.Name = Format(Date, "d mmmm yyyy")
End Sub

As the execution result, the current date becomes the 1st worksheet’s name.

170
 2.11. Objects Range, Selection and ActiveCell

2.11. Objects Range, Selection and ActiveCell

Object Range follows the Worksheet object in the object hierarchy. It allows
working with the following elements of Excel:
range of cells;
range of columns;
range of rows;
single cell.
Let us consider some properties of the Range object.
 Formula — the Excel formula with operands (cell addresses) in the A1
reference style.
For example, operators

Range("C2:F8").Formula = "=$A$4+COS($A$10)"
Range("D:E").Formula = "=$A$4+COS($A$10)"
Range("2:2").Formula = "=$A$4+COS($A$10)"
Range("B3").Formula = "=$A$4+COS($A$10)"

set the Formula property. These operators are respectively used to put formula
=$A$4+COS($A$10)

into the following parts of the active worksheet:

range C2:F8;
columns D and E;
the 2nd row;
the B3 cell.
 FormulaR1C1 — the Excel formula with operands (cell addresses) in the
R1C1 reference style.
For example, operator

Worksheets("Sheet1").Range("G1:H4").FormulaR1C1 = _
"=SQRT(R5C8)^3+7.3"

sets the FormulaR1C1 property. This operator is used for entering formula

171
 Chapter 2. Programming in VBA

=SQRT($H$5)^3+7.3

into the G1:H4 range on the Sheet1 worksheet (the R1C1 reference style option
may be turned on or off, see p. 143).
 Address — the cell address.
 Offset — the range shifted relative to the selected (active) range accord-
ing to two integers in parentheses.
In the Offset property, the first parameter (in parentheses) is the vertical shift,
and the second parameter is the horizontal shift. A comma is placed between
these parameters.
 Value — one of the following:
the array of the range values;
the cell value, if the Range object corresponds to the single cell.
 Columns — the collection of the range columns.
 Rows — the collection of the range rows.
 Cells — the collection of the range cells.
Note that we considered the Cells property regarding the Worksheet object in
the previous section.
Below, we will list some methods of the Range object.
 Clear — the removal of the range contents.
For example, the following application of the Clear method clears cells
A1:F7 on the Sheet1 worksheet:

Worksheets("Sheet1").Range("A1:F7").Clear

 Select — the selection (activation) of the range.

Objects of the Range type have properties and methods of the Range object.
When working with the Range type objects, it is convenient to use the
For Each…Next cycle, which is similar to cycle For…Next (p. 58). The syntax
of this operator may be studied by means of the Excel help system started by
pressing the F1 key when the VB window is active.
In the following program example, operator For Each…Next is used for
squaring the values of range A1:A6 on Sheet1 of the active workbook:
Listing 2.11
Sub Square()
Dim x As Range
For Each x In Worksheets("Sheet1").Range("A1:A6")
x.Value = x.Value ^ 2
Next
End Sub

172
 2.11. Objects Range, Selection and ActiveCell

The Selection object allows working with the active (selected) cells. As the
Range type object, the Selection object has properties Columns, Rows and Cells.
Thus, properties

Selection.Columns
Selection.Rows
Selection.Cells

are the collections of columns, rows and cells of the selected range, respectively.
The following example program inserts the multiplication table into the
selected range on the active worksheet:
Listing 2.12
Sub MultiplicationTable()
Dim m As Integer, n As Integer
Dim i As Integer, j As Integer
1: m = Selection.Rows.Count 'quantity of rows
2: n = Selection.Columns.Count 'quantity of columns
3: For i = 1 To m
4: For j = 1 To n
5: Selection.Cells(i, j).Value = i * j
6: Next j
7: Next i
End Sub

Operators 1 and 2 contain the Count property whose return allows defining
the quantity of objects in the Selection.Rows and Selection.Columns
collections. Because of executing these operators, the quantities of rows and
columns in the selected range are assigned to the m and n variables, respectively.
Operator 5 may have the following form:

Selection.Cells(i, j) = i * j

Here, .Value is present implicitly.

We advise the reader to do the following:
1) enter program MultiplicationTable into the code window;
2) go to the Excel window and select the range for the multiplication table;
3) run program MultiplicationTable;
4) looking the execution result, verify the program correctness.
The Selection object can be used for recovering the selection. For example, in
the graph subroutine (Section 4.8):
in the beginning, operator

173
 Chapter 2. Programming in VBA

Set wbOldSelection = Selection

assigns the selected range to the wbOldSelection variable of the Range

type;
in the end, operator

wbOldSelection.Select

selects the wbOldSelection range.

It was mentioned in Section 2.8 that ActiveCell is a property of the Applica-
tion object, corresponding to the active cell on the active worksheet of the active
workbook. As the Range type object, the ActiveCell property has the properties
and methods of the Range object.
The PropofRange program, given below, contains examples of using the
properties and methods of the Range object.
1. Put 100 into cell B1, 200 into B2 and 300 into B3 on the Sheet1 work-
sheet.
2. Put formula

=SUM(B1:B3)

into cell B4 on Sheet1, and click on the tick button of the Excel formula bar.
3. Go to Visual Basic Environment and insert a module into the active
workbook.
4. Enter the following text into the code window:
Listing 2.13
Sub PropofRange()
1: Worksheets("Sheet1").Range("A1").Select
2: ActiveCell.Offset(2, 3).Select
3: MsgBox "Current cell — " & ActiveCell.Address
4: MsgBox "Value in cell B4 = " & _
Range("B4").Value
5: MsgBox "Formula in cell B4: " & _
Range("B4").Formula
End Sub

5. Run the PropofRange program execution after activating the Sheet1

worksheet. The window with message Current cell - $D$3 and the OK button
appears (Fig. 2.21a).

174
 2.11. Objects Range, Selection and ActiveCell

6. Click on the OK button. The window with message Value in cell B4 =

600 and the OK button appears (Fig. 2.21b).
7. Click on the OK button. The window with message Formula in cell B4:
=SUM(B1:B3) and the OK button appears (Fig. 2.21c).
8. Click on the OK button for terminating the execution.
Operator 1 of the PropofRange program selects cell A1 on the Sheet1
worksheet. Operator 2 selects the D3 cell, shifted 2 vertically and 3 horizontally
relative to the A1 cell (property Offset and method Select of the Range object are
figured in this operator). Further, three windows with the message and OK but-
ton are sequentially displayed (Fig. 2.21).

Fig. 2.21a. The Excel worksheet with the first window

Fig. 2.21b. The second window

175
Chapter 2. Programming in VBA

Fig. 2.21c. The third window

176
 2.12. Study of objects

2.12. Study of objects

We can obtain information on any object by means of the reference systems

started by pressing keys F1 and F2. For that, we press the F2 key when the VB
window is active. As a result, the object browser window appears.
If the Range object is interesting for us, we highlight Range in list Classes
by click. At that, list Members of 'Range', containing properties and methods of
object Range, appears in the object browser window (Fig. 2.22).

Fig. 2.22. The VB window containing the object browser window

177
 Chapter 2. Programming in VBA

For obtaining information on the Select method of the Range object, we high-
light line Select in list Members of 'Range' and press the F1 key. As a result,
the Excel Help window, containing the necessary information, is displayed
(Fig. 2.23).

Fig. 2.23. The Excel Help window with the description

of the Range.Select method

178
 2.12. Study of objects

We advise the reader to obtain information on the Add method of object

Worksheets in a similar way.
In addition, we advise the reader to obtain information on the ActiveCell
object by means of the binoculars pictogram in the object browser window, such
as we obtained information on vbFriday on p. 32.

179
 Chapter 2. Programming in VBA

2.13. Using the Excel table as the user

interface of programs

We considered the main Excel objects. Now we can use the Excel table as the
user interface of programs.
Initially, we will create a macro for calculating the length of the leg of
a right-angled triangle with given lengths of the hypotenuse and the other leg.
In this macro, we will realize formula

b c2 a2
following from formula (1.1) on p. 16.
Let length a of the leg be in the A4 cell, length c of the hypotenuse be in C4.
Let B5 be intended for resulting length b. The macro has the following text:
Listing 2.14
Sub Leg()
Dim a As Single, b As Single, c As Single
a = Range("A4").Value
c = Range("C4").Value
If c < a Then
Range("B5").Value = "Error"
End 'immediate termination of macro
End If
b = Sqr(c ^ 2 - a ^ 2) 'according to Pythagoras
Range("B5").Value = b
End Sub

We advise the reader to do the following:

1) enter the Leg macro text into the code window of a new module;
2) go to the Excel window;
3) enter 4.3 and 5.1 into cells A4 and C4, respectively;
4) run the Leg macro;
5) make sure that number 2.742261 appears in the B5 cell on the active
worksheet.
As we know, length c of the hypotenuse must be greater than or equal to
length a of the leg. If the C4 cell value is less than the A4 cell value, logical

180
 2.13. Using the Excel table as the user interface of programs

expression c < a accepts True when executing the Leg macro. In this case,
message Error appears in cell B5 and the End operator terminates the macro
execution. We encounter the End operator for the first time.
It is easy to understand the work of the following macro intended for trans-
posing a numerical matrix relative to its auxiliary diagonal according to formula
(1.3) on p. 109:
Listing 2.15
Sub TRANSPA()
Dim m As Integer, n As Integer
Dim i As Integer, j As Integer
m = Selection.Rows.Count 'quantity of rows
n = Selection.Columns.Count 'quantity of columns
For j = 1 To n
For i = 1 To m
Selection.Cells(j + m + 1, i) = _
Selection.Cells(m + 1 - i, n + 1 - j)
Next i
Next j
End Sub

This macro is used as follows:

1) on the Excel worksheet, select the matrix intended for the transposition;
2) run the TRANSPA macro.
The result of the macro execution appears below the original matrix
(Fig. 2.24).

Fig. 2.24. The Excel worksheet with the initial matrix and
the result of its transposing relative to the auxiliary diagonal

181
 Chapter 2. Programming in VBA

2.14. Two more Excel macros.

Personal Macro Workbook

Let us create a macro fulfilling the following operations:

1) at the 1st run, the macro writes the value of cell A3 on worksheet Sheet1
into cell B2 on worksheet Sheet2;
2) at the 2nd run, the macro writes the value of the same cell, Sheet1!A3,
into cell Sheet2!B4;
3) at the 3rd run, the macro writes the value of cell Sheet1!A3 into cell
Sheet2!B6, and so on.
Into cell Sheet1!A3, the values are entered manually before the next run of
the macro.
Let us use Sheet1!A4 as an auxiliary cell. We have to write zero into this cell
before a series of the macro runs.
Into the code window of a new module, we enter the following text of the
macro:
Listing 2.16
Sub Macr1()
Dim i As Integer, M As Range
Set M = Range("Sheet2!B1:Sheet2!B200")
i = Range("Sheet1!A4").Value + 2
M(i) = Range("Sheet1!A3").Value
Range("Sheet1!A4").Value = i
End Sub

In the above Macr1 macro:

i is an auxiliary variable;
M is the array corresponding to range B1:B200 on worksheet Sheet2;
Range("Sheet1!A4").Value, Range("Sheet1!A3").Value
are the values in cells A4 and A3 on worksheet Sheet1.
At the 1st run of the Macr1 macro:
1) number 2 is assigned to variable i;
2) the value of cell A3 on worksheet Sheet1, for example 100, is assigned to
variable M(2), i.e., to cell B2 on worksheet Sheet2;
3) the value of i (number 2) is written into cell A4 on worksheet Sheet1.

182
 2.14. Two more Excel macros. Personal Macro Workbook

At the 2nd run of the Macr1 macro:

1) number 4 is assigned to variable i;
2) the value of cell A3 on worksheet Sheet1, for example 200, is assigned to
variable M(4), i.e., to cell B4 on worksheet Sheet2;
3) the value of i (number 4) is written into cell A4 on worksheet Sheet1.
And so on. We see that the Macr1 macro fulfills the necessary operations
(Fig. 2.25).

Fig. 2.25. Worksheets Sheet1 (a) and Sheet2 (b) after

the fourth run of the Macr1 macro: numbers 100, 200, 300
and 400 are put into cell Sheet1!A3 before the macro runs

183
 Chapter 2. Programming in VBA

The developed macro, Macr1, has two drawbacks:

we can use this macro only in the Excel workbook, where it was created;
the number of the macro runs (in one series) should not exceed 100
because the M array, containing the fixed quantity of elements, is used in this
macro.
For liquidation of the first drawback, we will create Personal Macro Work-
book. For that, let us fulfill the following operations.
1. Open a blank workbook.
2. Open window Record Macro by fulfilling Developer > Record Macro
(in area Code) or View > arrow Macros (in area Macros) > Record Macro.
3. Enter Personal Macro Workbook into box Store macro in by using the
drop-down list (Fig. 2.26) and click on OK.

Fig. 2.26. The Record Macro window before clicking on the OK button

4. Developer > Stop Recording (in area Code) or View > arrow Macros
(in area Macros) > Stop Recording.
5. Open the VB window.
6. Click on line VBAProject (PERSONAL.XLSB) in the project explorer
window, and fulfill File > Save PERSONAL.XLSB.
7. Close the Excel window by clicking on the little cross in the top right
corner. If the window with a question about saving changes in Personal Macro
Workbook is displayed (when closing the Excel window), click on the Yes
button.

184
 2.14. Two more Excel macros. Personal Macro Workbook

Because of these operations, the PERSONAL.XLSB file is created. This file

is Personal Macro Workbook of the computer user by name usr.
Personal Macro Workbook opens automatically when starting Excel. Macros
and user-defined functions, placed in Personal Macro Workbook, can be used
in Excel workbooks that are in folder c:\Users\usr (and in folders enclosed
in it).
The automatic opening of Personal Macro Workbook does not lead to
appearance of any button on the taskbar of Windows Desktop.
Let us fulfill the following operations:
1) open a blank workbook;
2) go to the VB window;
3) click on the VBAProject (PERSONAL.XLSB) line in the project explorer
window;
4) Insert > Module (it means the module insertion into Personal Macro
Workbook);
5) put macro
Listing 2.17
Sub Macr2()
Dim i As Integer
i = Range("Sheet1!A4").Value + 2
Range("Sheet2!B" & CStr(i)) = _
Range("Sheet1!A3").Value
Range("Sheet1!A4").Value = i
End Sub

into the code window.

In Fig. 2.27, Module1 in the project explorer window is the result of opera-
tion of Excel Macro Recorder that was used for creating Personal Macro Work-
book. This module may be removed or used in further work with Personal Macro
Workbook, for example, for storage of the Func9 function (Section 2.15). The
Macr2 macro is placed in the Module2 module.
To save changes made in Personal Macro Workbook, we must fulfill the fol-
lowing in the VB window: File > Save PERSONAL.XLSB. At that, the Macr2
macro, which is a part of file PERSONAL.XLSB, is saved on the hard disk of
the computer.
Macro Macr2, used as macro Macr1, has the following advantages:
macro Macr2 can be used in all Excel workbooks that are in folder
c:\Users\usr;
the number of the macro runs is not limited because the M array is absent
in the macro code.

185
 Chapter 2. Programming in VBA

Fig. 2.27. The VB window with the Macr2 macro

in the code window of the Module2 module

It is possible to transfer Personal Macro Workbook to other user. For this

purpose, the PERSONAL.XLSB file should be copied into the appropriate folder
of this user. Let us consider the location of Personal Macro Workbook on the
hard disk of the computer.
The workbooks, which are in the XLSTART folder, are opening automatical-
ly when starting Excel. The full name of this folder follows:

c:\Users\usr\AppData\Roaming\Microsoft\Excel\XLSTART

where usr is the user name. File PERSONAL.XLSB is located in this folder.
If we want to transfer our Personal Macro Workbook to the user whose name
is usr2, we must copy the PERSONAL.XLSB file into folder

c:\Users\usr2\AppData\Roaming\Microsoft\Excel\XLSTART

of his computer.
Note that folder AppData (from “Application Data”) may be hidden in
Windows Explorer. To make a folder or file visible, we must fulfill the following
operations:
1) open the folder containing the hidden folder or file of interest;
2) Organize > Folder and search options;

186
 2.14. Two more Excel macros. Personal Macro Workbook

3) in open window Folder Options, activate tab View;

4) in list Advanced settings, turn on option Show hidden files, folders, and
drives;
5) successively click on buttons Apply and OK.
To expand the chosen operation mode of Windows Explorer to all folders, we
must click on button Apply to Folders before clicking on the Apply button. In
open window Folder Views, we must click on the Yes button.

187
 Chapter 2. Programming in VBA

2.15. One more user-defined function of Excel

There are value h and two-dimensional numerical array (matrix) A, whose

elements are in cells of the Excel table. We will consider a user-defined function,
which returns the number of the A array values exceeding h.
Let us create an Excel workbook and then open the VB window. Further, let
us insert a module into Personal Macro Workbook, already created, and then put
the following text of the user-defined function into the code window of the
inserted module.
Listing 2.18
Function Func9(massive As Variant, h As Variant) _
As Integer
1: Dim k As Integer, i As Integer, j As Integer
2: k = 0
3: If TypeName(massive) = "Range" Then
4: For i = 1 To massive.Rows.Count
5: For j = 1 To massive.Columns.Count
6: If massive(i, j) > h Then
7: k = k + 1
8: End If
9: Next j
10: Next i
11: Func9 = k
12: Else
13: MsgBox "Func9: Argument is not range"
14: End If
End Function

The current state of the project is depicted in Fig. 2.28.

The built-in TypeName function (in line 3) returns its argument’s data
type as a string. This function is used for verifying the type of input parameter
massive. In the correct call of the Func9 function, the massive variable has
the Range type.

188
 2.15. One more user-defined function of Excel

Fig. 2.28. The VB window with text of the Func9 function in the code window

In line 4, the quantity of rows in the massive range is the result of return-
ing property massive.Rows.Count. In line 5, the quantity of columns is the
result of returning property massive.Columns.Count.
In line 6, massive(i, j) is the reference to the corresponding cell of the
massive range.
The Func9 function is used as follows.
1. Enter values of matrix A, for example, into range D5:E8 on worksheet
Sheet4.
2. Assign a name, for example Test9, to the D5:E8 range. For this purpose:
1) select this range, and fulfill Formulas > Define Name in area Defined
Names;
2) in the open New Name window, type Test9 in text box Name;
3) enter Sheet4 into box Scope by means of the drop-down list (Fig. 2.29);
4) click on the OK button.
3. Enter formula

=PERSONAL.XLSB!Func9(Test9;2)

into any cell on worksheet Sheet4, for example, E11.

189
 Chapter 2. Programming in VBA

In the above formula, an exclamation mark means that the Func9 function is
a part of file PERSONAL.XLSB.
4. Click on the tick button of the Excel formula bar.

Fig. 2.29. The New Name window with automatically filled box Refers to

The resulting number of values in the D5:E8 cells exceeding 2 appears in the
E11 cell (Fig. 2.30).
Below is the second way of using the Func9 function, without assigning
a name to the range with the A array.
Let us use Function Wizard as follows:
1) select a cell for the result, for example, Sheet4!E11;
2) click on the fx button of the Excel formula bar to start Function Wizard;
3) in the User Defined category of the open Insert Function window, high-
light the line corresponding to the Func9 function (Fig. 2.31), and click on the
OK button;
4) in the open Function Arguments window (Fig. 2.32), enter:
the source range (for example, D5:E8 on the Sheet4 worksheet) into the
first text box;
the h value, for example 2, into the second text box;
5) click on the OK button.
We see the former result: five values exceed h = 2 (Fig. 2.30).
Note that cell E11 and range D5:E8 may be on one or different worksheets of
the same Excel workbook.

190
 2.15. One more user-defined function of Excel

Fig. 2.30. The source data and calculation result in the Excel window

Fig. 2.31. The first window of Function Wizard

191
 Chapter 2. Programming in VBA

Fig. 2.32. The second window of Function Wizard

We advise the reader to calculate the values of function f (x) from Appen-

dix 4 and also of functions g ( x) x f ( x) and h( x) x 2 f ( x) at not less than 10

points of segment [a, b] , including a and b, in the following ways:
creating a macro by means of Excel Macro Recorder;
programming an Excel user-defined function in VBA;
programming a macro in VBA.
Use the Excel tools to create the graphs of functions f (x) , g (x) and h(x) .

192
 2.16. Digression. Change of Excel options

2.16. Digression. Change of Excel options

In the next four chapters of the book, languages VB and VBA will be used
for programming numerical methods. Besides, we will consider the Excel proce-
dures, which are a part of the add-ins, for solving some of our tasks.
For installing the necessary Excel add-ins, let us fulfill the following opera-
tions:
1) File > Options > Add-Ins;
2) enter Excel Add-ins into the Manage box by means of the drop-down list,
and click on the Go button;
3) in the open Add-Ins window, turn on options Analysis ToolPak and Solver
Add-in, and click on the OK button.
As a result, the Data Analysis and Solver commands appears in Excel Rib-
bon — on the Data tab, in the Analysis area.
We will need to tune Excel so that formulas are displayed in cells, instead of
results of calculations according to these formulas. For this purpose, we must
fulfill the following operations:
1) File > Options > Advanced;
2) turn on option Show formulas in cells instead of their calculated results
located below box Display options for this worksheet;
3) click on the OK button.
In this way, we change the contents of the cells with formulas on the work-
sheet specified in box Display options for this worksheet. The contents of the
cells without formulas do not change.
Upon turning off option Show formulas in cells instead of their calculated
results, the worksheet returns to the customary form, with values in the cells.
We will also need a workbook with the module for programs, realizing
numerical methods. To obtain it, let us fulfill the following:
1) open a blank workbook, and insert the Module1 module into it;
2) put text

Sub main()

End Sub

193
 Chapter 2. Programming in VBA

into the code window of this module;

3) in the VB window, fulfill File > Save;
4) in the open Save As window, choose the c:\Users\usr folder familiar to us
(usr is the user name);
5) enter the workbook name, BookNM, into text box File name (NM is the
abbreviation of “numerical methods”);
6) set file type Excel Macro-Enabled Workbook by using the drop-down list;
7) click on the Save button;
8) close the VB and Excel windows.
We will enter texts of programs by name main into the Module1 module.
Upon termination of work with the BookNM workbook, we will always save it.

194
 Chapter 3.
Finite Difference Method
for Solving Differential Equations

Two kinds of conditions on the solution of the second-order linear differen-

tial equation are reviewed, namely, the boundary and periodicity conditions.
Replacing the derivatives by their finite difference analogs, we obtain the
so-called finite difference schemes — systems of linear algebraic equations
of special form.
We review several versions of the decomposition method [4] for solving
the finite difference schemes. The simplest scheme is also solved by the Gaussi-
an elimination method. The question of stability of these two methods is investi-
gated in respect of not increasing the computing error during solving this
scheme.
The finite difference method for solving the linear differential equation is
used for solving the nonlinear differential equation by the quasilinearization
method. For demonstration of the finite difference method’s possibilities,
we develop subroutines and programs for solving mathematical and applied
problems. We use the Excel scatter diagrams for visualization of calculation
results.

195
 Chapter 3. Finite Difference Method for Solving Differential Equations

3.1. Finite difference analogs

of derivatives for a uniform grid

Let us introduce an increasing sequence of points,

x k < xk 1 < xk 2 < ... < xr 2 < xr 1
< xr ,
on segment a ≤ x ≤ b, at that, x k a , xr b . In other words, segment [a, b] is
covered with a grid whose nodes have coordinates x k , xk 1
, xk 2
, ..., xr 2
,
xr 1
, xr .
For simplicity, we will consider that the grid is uniform, i.e., the grid step,
xi xi 1 h , does not depend on i, k + 1 ≤ i ≤ r (later, we will consider
a nonuniform grid).
In addition to this grid (Fig. 3.1), called the main grid, we will use the so-
called auxiliary grid with the following nodes:
x[ k ] ( xk xk 1 ) / 2 , x[ k 1] ( xk 1 xk 2 ) / 2 , ..., x[i 1] ( xi 1 xi ) / 2 ,
x[i ] ( xi xi 1
) / 2 , ..., x[ r 2]
( xr 2
xr 1
) / 2 , x[ r 1]
( xr 1
xr ) / 2 .
As we see, numbers (indices) of the auxiliary grid nodes are in square brackets.

Fig. 3.1. Grids on segment [a, b] : main — indices without brackets;

auxiliary — indices in square brackets

Let u k , u[k ] , u k 1
, u[k 1]
, uk 2
, ..., ui 1
, u[i 1]
, ui , u[i ] , ui 1
, ...,

ur 2
, u[r 2]
, ur 1
, u[r 1]
, u r be the u (x) function values at the corre-
sponding nodes of the main and auxiliary grids.
A function, set on a grid, is called a grid function or a function in tabular
form (tabular function). An example of such function is u (x) considered above.

196
 3.1. Finite difference analogs of derivatives for a uniform grid

We will come across grids and grid functions more than once.
Let us assume that function u (x) is continuous in segment [a, b] , has con-
tinuous derivatives of m 1 orders, and the m-th derivative exists, in other
words, function u (x) is differentiable m times. At that, in a neighborhood of any
interior point x of [a, b] , we can use Taylor’s formula [3] in the following form:

du 2 d 2u 3 d 3u
u( x ) u ( x) ( x) ( x) ( x) ...
1! dx 2! dx2 3! dx3
m 1 dm 1u
m
( x ) O( ). (3.1)
(m 1)! dxm 1

m
The O notation, O( ) , has the following sense (for natural m).
m
If ε is a small quantity, as in (3.1), O( ) means the quantity whose absolute
m
value is less or equal to C | | , where C is a positive constant, i.e., C is inde-
m
pendent of ε. In this case, O( ) is called the quantity of m-th order of small-
m
ness or the quantity of -th order.
If Ε is a large positive quantity, O( Ε m ) means the positive quantity whose

value tends to CΕ m when Ε , C is a positive constant.

If the grid step, h, is a small quantity, we have the following chain of equali-
ties according to formula (3.1):
du 1 1
(x ) [u ( xi ) u ( xi )]
dx i 2 2

h ( h / 2) 2 d 2 u h ( h / 2) 2 d 2 u
u xi u ( xi ) ( xi ) u xi u ( xi ) ( xi )
2 2 dx 2 2 2 dx 2
h h
2 2

u[ i ] u[ i 1]
O(h 2 ) O(h 2 ) . (3.2)
h
Similarly, we can obtain two more expressions for the first derivative of the
u (x) function:

197
 Chapter 3. Finite Difference Method for Solving Differential Equations

du ui 1
ui
(x ) O( h 2 ) , (3.3)
dx [ i ] h
du ui 1 ui 1
( xi ) O( h 2 ) . (3.4)
dx 2h
Using formulas (3.2) and (3.3), we can also obtain an expression for the
second derivative of u (x) as follows:

d 2u u ( x[ i ] ) u ( x[ i 1]
)
( xi ) O(h 2 )
dx2 h
ui 1
ui ui ui 1
h h ui 1
2u i ui 1
O(h 2 ) O(h 2 ) . (3.5)
h h 2

After neglecting summands O( h 2 ) in (3.2), (3.4) and (3.5), we have finite

difference analogs of the first two derivatives of the u (x) function at point xi .

198
 3.2. Finite difference scheme for the linear differential equation. The decomposition method

3.2. Finite difference scheme for the linear

differential equation. The decomposition method

Let us consider the following second-order linear differential equation of

general form on segment [a, b] :

d 2u du
g ( x)
e( x)u f ( x) , (3.6)
dx2 dx
where g (x) , e(x ) and f (x) are given functions, u (x ) is an unknown function.
For uniqueness of the solution of this equation, we use the following left and
right boundary conditions:
A0 u ( a ) A1 u ( a ) A , (3.7)
B0 u (b) B1 u (b) B, (3.8)
where A and B are given parameters, A0 B0 1 , A1 B1 0 to begin with.
Let us consider a method for solving this boundary value problem, based on
replacing the derivatives by their finite difference analogs,
du ui 1
ui 1 d 2u ui 1
2u i ui 1
( xi ) , .( xi )
dx dx 2h h 2 2

These expressions follow from (3.4) and (3.5) if we neglect the summands of
second order of smallness, O( h 2 ) .
Let xi be an internal node of the main grid (Fig. 3.1), i.e., node xi does not
coincide with points a and b. Equation (3.6) at this node looks like
d 2u du
( xi ) g ( xi ) ( xi ) e( xi ) u ( xi ) f ( xi ) .
dx2 dx

Multiplying both sides of this equality by h 2 and substituting the finite differ-
ence analogs of the derivatives, we get the following linear algebraic equation:
u
i i 1
u
i i
u
i i 1 i
, (3.9)
where ui 1
, ui , u i 1
are unknown variables, i = k + 1, k + 2, ..., r – 2, r – 1,

199
 Chapter 3. Finite Difference Method for Solving Differential Equations

i
1 0.5 g i h ,

i
ei h 2 2,

i
1 0.5 g i h , (3.10)

i
fi h 2 .
In the last four expressions, g i , ei and fi are the values of the coefficients
and right-hand side of equation (3.6) at node xi : g i g ( xi ) , ei e ( xi ) ,
fi f ( xi ) , i = k + 1, k + 2, ..., r – 2, r – 1.
Boundary conditions (3.7) and (3.8) can be written as follows:
u
k k
u
k k 1 k
, (3.11)
u
r r 1
u
r r r
, (3.12)
where k r
2, k r
0, k
2A , r
2B .
The system of linear algebraic equations (3.9), (3.11) and (3.12) is called the
finite difference scheme for boundary value problem (3.6) — (3.8). More pre-
cisely, this system is called the one-dimensional finite difference scheme because
it corresponds to the problem with one spatial coordinate.
Using the definitions of matrix multiplication and equality (Section 1.21), we
can write scheme (3.9), (3.11), (3.12) as the following matrix equation:
k k
0 0 ... 0 0 0 0
k 1 k 1 k 1
0 ... 0 0 0 0
0 k 2 k 2 k 2
... 0 0 0 0
. . . . . . . . .
0 0 0 0 ... r 2 r 2 r 2
0
0 0 0 0 ... 0 r 1 r 1 r 1
0 0 0 0 ... 0 0 r r

uk k
uk 1 k 1
uk 2 k 2
... ... .
ur 2 r 2
ur 1 r 1
ur r

200
 3.2. Finite difference scheme for the linear differential equation. The decomposition method

The coefficient (system) matrix, which is a part of this matrix equation, has
the so-called tridiagonal form: nonzero elements are only on the main diagonal,
the first “diagonal” below this and the first “diagonal” above the main diagonal.
Finite difference scheme (3.9), (3.11), (3.12) is usually solved by the decom-
position method [4] whose main advantage is its efficiency in comparison with
other methods for solving this system of linear algebraic equations:
for calculating the values of the n r k 1 unknowns ( u k , u k 1 ,
uk 2
, ..., u r 2
, ur 1
, u r ) by the decomposition method, O(n) arithmetic
operations must be performed (p. 203);
the solution of the system of equations (3.9), (3.11) and (3.12) by the
Gaussian elimination method (Sections 3.9 and 3.10) requires O ( n3 ) arithmetic
operations (p. 227).
The decomposition method includes two stages, which are called the forward
and backward sweeps. To obtain formulas for the sweeps, let us connect the
unknowns, u i 1 and ui , through formula
ui 1
Pi ui Qi , (3.13)
where Pi , Q i are the auxiliary unknowns, i = r, r – 1, ..., k + 1.
After substituting the last expression into equation (3.9), we obtain
i i
Q
i i
ui ui 1
P
i i i
P
i i i
or
ui Pi u
1 i 1
Qi 1
,
where Pi 1
and Qi 1
are determined by the following recurrence formulas:

i
Pi 1
, (3.14)
P
i i i

i i
Qi
Qi 1
. (3.15)
i i
P i

According to the last two formulas, if the values of Pk 1

, Qk 1
are known,
the use of these formulas at i = k + 1, k + 2, ..., r – 1 gives the values of Pk 2
,
Qk 2
, Pk 3
, Qk 3
, ..., Pr , Q r .
The values of Pk 1
, Qk 1
are determined by formulas

201
 Chapter 3. Finite Difference Method for Solving Differential Equations

k
Pk 1
, (3.16)
k

k
Qk 1
, (3.17)
k

which follow from left boundary condition (3.11).

The recurrence calculation of unknown Pi , Q i using formulas (3.16), (3.17)
and (3.14), (3.15) is called the forward sweep.
Knowing the values of u r , Pi , Q i (k + 1 ≤ i ≤ r), we can calculate unknown
ur 1
, ur 2
, ..., u k using recurrence formula (3.13).
For determining the value of u r , let us consider the following system of two
linear algebraic equations with unknown u r 1
and u r :
u
r r 1 r r
u r
,
ur 1
Pr u r Qr .
The first equation is right boundary condition (3.12); the second equation is
(3.13) at i = r. Let us solve this system for u r 1 and u r by means of Cramer’s
rule [3].
The determinant of the system looks like
r r
D .
1 Pr
The following determinants are obtained from D by replacing the first and
second columns by the column of the right-hand sides:
r r r r
D1 , D2 .
Qr Pr 1 Qr
According to Cramer’s rule, we have
D1 r r
P r
Qr
ur 1 ,
D r r
P r

D2 r r
Qr
ur . (3.18)
D r r
P r

The recurrence calculation of unknown ui using formulas (3.18) and (3.13)

is called the backward sweep.

202
 3.2. Finite difference scheme for the linear differential equation. The decomposition method

The use of the forward and backward sweeps gives the values of unknown
u k , u k 1 , u k 2 , ..., u r 2 , u r 1 , u r , i.e., the solution of finite difference
scheme (3.9), (3.11), (3.12) for linear differential equation (3.6) with boundary
conditions (3.7) and (3.8). The calculated values of u k , u k 1 , u k 2 , ..., u r 2 ,
ur 1
, u r approximate the exact solution of problem (3.6) — (3.8) at the nodes
of the main grid on segment [a, b] .
A simple analysis of formulas (3.14) and (3.15) of the forward sweep shows
that it requires O(n) arithmetic operations, n . A similar analysis of formu-
la (3.13) shows that the backward sweep also requires O(n) arithmetic opera-
tions. Thus, the solution of finite difference scheme (3.9), (3.11), (3.12) by the
decomposition method requires O(n) arithmetic operations.

203
 Chapter 3. Finite Difference Method for Solving Differential Equations

3.3. Sufficient stability conditions

for the decomposition method

A repeatable algorithm (computing process) is called stable if the error aris-

ing at any step (for example, the rounding-off error) does not increase during the
computing process.
The decomposition method includes the forward and backward sweeps. Each
of these sweeps can be unstable. Let us investigate this question.
It is easiest to write the sufficient condition of the backward sweep stability,
i.e., the stability of the calculation process corresponding to recurrence formula
(3.13). This condition has the following form:
Pi 1 (3.19)
for all values of i from k + 1 to r. If this condition is satisfied, then:
1) small error i (which is a part of the calculated value of ui ) goes into the
value of ui 1
without increase; i 1
Pi i
is the come error of i
-th order;
2) because the additional error ( i 1
generated when calculating the value
of ui 1
) may be positive or negative with probability 0.5, the total error
( i 1
= i 1
+ i 1
) is of i
-th order.
Similarly, the sufficient condition of stability of the calculation according to
recurrence formula (3.15) of the forward sweep is

i
1 (3.20)
P
i i i
for all values of i from k + 1 to r – 1. If this condition is satisfied, then:
1) small error i (which is a part of the calculated value of Q i ) goes into
the value of Qi 1
without increase;

i
i 1 i
P
i i i

is the come error of i

-th order;

204
 3.3. Sufficient stability conditions for the decomposition method

2) because the additional error ( i 1

generated when calculating the value
of Qi 1
) may be positive or negative with probability 0.5, the total error
( i 1
= i 1
+ i 1
) is of i
-th order.

Let us show that the simultaneous satisfaction of inequalities (3.19) and

(3.20) is the sufficient condition of stability of the calculation according to recur-
rence formula (3.14) of the forward sweep.
Let i be a small error in the calculated value of Pi . Using formula (3.14)

and expansion 1 /(1 x) 1 x x2 x3 ... from table “Important Series

Expansions” [3], we have the following chain of equalities:
i i
1
Pi 1 i 1
i
( Pi i
) i i i
P i
1 i i
/( P
i i i
)

i 2 i 2
Pi 1
1 i
O( i
) Pi 1
Pi 1 i
O( i
).
P
i i i
P
i i i

By comparing the beginning and end of this chain, we obtain

i
i 1
Pi 1 i
. (3.21)
P
i i i

If inequality (3.19), which looks like Pi 1

1 , and inequality (3.20) are
satisfied, then
i
Pi 1
1.
P
i i i
According to (3.21) and the last inequality, we can state the following:
1) small error i (which is a part of the calculated value of Pi ) goes into the
value of Pi 1
without increase; the come error, i 1
, is defined by formula
(3.21), i.e., i 1
is of i
-th order;
2) because the additional error ( i 1
generated when calculating the value
of Pi 1
) may be positive or negative with probability 0.5, the total error
( i 1
= i 1
+ i 1
) is of i
-th order.

That is, the algorithm of the calculation according to formula (3.14) is stable.

205
 Chapter 3. Finite Difference Method for Solving Differential Equations

Thus, the simultaneous satisfaction of inequalities (3.19) and (3.20) for all
values of i is enough for stability of the forward and backward sweeps, i.e., of
the decomposition method in general.
By means of conditions (3.19) and (3.20), we will obtain other stability con-
ditions of the decomposition method, which are more convenient to use.
Let us assume that

0 Pk 1
1 (3.22)
and inequalities
i i
, i i i
, (3.23)

i
0 (3.24)
are simultaneously satisfied for all values of i. Then inequalities (3.19) and
(3.20) are satisfied for all values of i, i.e., the decomposition method is stable.
To prove the last assertion, we transform formula (3.14) as follows:
i i i
Pi 1
,
P
i i i i
( i i i
) i
(1 Pi ) i i

where i
( i i i
) i
(1 Pi ) .
Similarly, we obtain the following expression for a part of formula (3.15):
i i
.
P
i i i i i

If 0 Pi 1 , inequality 1 Pi 0 is satisfied. Then (3.23) and (3.24) result

in i
0 and
0 Pi 1
1, (3.25)

i
0 1. (3.26)
P
i i i
According to (3.22), inequalities (3.25) and (3.26) are satisfied for i = k + 1,
k + 2, ..., r – 1. As the consequence of this, inequalities (3.19) and (3.20) are satis-
fied for all values of i, i.e., the decomposition method is stable.
We proved that conditions (3.22) — (3.24), as well as conditions (3.19) and
(3.20), are the sufficient stability conditions for the decomposition method.
Because coefficients i , i and i are defined by expressions (3.10), the
consequence of the obtained stability conditions is the unconditional stability of
the decomposition method for solving boundary value problem (3.6) — (3.8), for
which the following conditions are satisfied:

206
 3.3. Sufficient stability conditions for the decomposition method

1) g (x) 0 for all values of x from a to b, i.e., equation (3.6) looks like

d 2u
e ( x )u f ( x) ;
dx 2
2) e(x) 0 for all values of x;
3) left boundary condition (3.7) is such that 0 Pk 1
1.
The unconditional stability means the stability for arbitrary step h of the grid.

207
 Chapter 3. Finite Difference Method for Solving Differential Equations

3.4. Simplification of the second-order

linear differential equation

Thanks to the unconditional stability of solving the last boundary value prob-
lem (formulated at the end of the previous section), the substitution into equation
(3.6), which excludes the first derivative of u (x ) , is of interest. We will show
that such substitution looks like
x
u ( x) U ( x ) exp 0.5 g ( y ) dy . (3.27)
a

The exponential function (Appendix 3) is figured in this expression.

Let us differentiate function (3.27) twice. Using the basic rules of differentia-
tion [3], we obtain
du dU x
0.5 gU exp 0.5 g ( y ) dy , (3.28)
dx dx a

d 2u d 2U dU x
g (0.25g 2 0.5 g ) U exp 0.5 g ( y )dy . (3.29)
dx 2 dx 2 dx a

By substituting expressions (3.27) — (3.29) into equation (3.6), we obtain the

following equation without the first derivative of unknown function U (x) :

d 2U
E ( x )U F ( x) , (3.30)
dx 2
where
E e( x) 0.25g 2 ( x) 0.5g ( x) , (3.31)
x
F f ( x ) exp 0.5 g ( y ) dy . (3.32)
a

Let expressions (3.27) and (3.28), in which x a , be substituted into condi-

tion (3.7). We obtain the following left boundary condition for U (x) :

208
 3.4. Simplification of the second-order linear differential equation

A2 U ( a ) A1U ( a ) A3 , (3.33)
where
A2 A0 0.5 A1 g ( a ) , (3.34)
A3 A. (3.35)
After substituting expressions (3.27) and (3.28), in which x b , into condi-
tion (3.8), we have the following right boundary condition:
B2 U (b ) B1U (b ) B3 , (3.36)
where
B2 B0 0.5 B1 g (b) , (3.37)
b
B3 B exp 0.5 g ( y ) dy . (3.38)
a

We will use substitution (3.27) not only in this chapter, but in the next chap-
ter too.

209
 Chapter 3. Finite Difference Method for Solving Differential Equations

3.5. Program realization of the decomposition method

For solving equation (3.6) with boundary conditions (3.7) and (3.8) by the
decomposition method, we insert module Module2 into the BookNM workbook
(p. 194) and put the following subroutine declaration into this module:
Listing 3.1
Sub fb(ByVal k, ByVal r, ByVal h, ByVal A, ByVal B, _
ByRef G() As Double, ByRef E() As Double, _
ByRef F() As Double, ByRef U() As Double)
Const BETAK = -2, GAMMAK = 0
Dim DELTAK As Double
Const ALPHAR = 0, BETAR = -2
Dim DELTAR As Double
Dim alpha As Double, beta As Double
Dim gamma As Double, delta As Double
Dim i As Integer, w As Double
Dim P() As Double: ReDim P(k + 1 To r)
Dim Q() As Double: ReDim Q(k + 1 To r)
DELTAK = -2 * A
DELTAR = -2 * B
'Forward sweep:
P(k + 1) = -GAMMAK / BETAK
Q(k + 1) = DELTAK / BETAK
For i = k + 1 To r - 1
w = 0.5 * G(i) * h
alpha = 1 - w
beta = E(i) * h ^ 2 - 2
gamma = 1 + w
delta = F(i) * h ^ 2
w = alpha * P(i) + beta
P(i + 1) = -gamma / w
Q(i + 1) = (delta - alpha * Q(i)) / w
Next i
'Backward sweep:

210
 3.5. Program realization of the decomposition method

U(r) = (DELTAR - ALPHAR * Q(r)) / _

(ALPHAR * P(r) + BETAR)
For i = r To k + 1 Step -1
U(i - 1) = P(i) * U(i) + Q(i)
Next i
End Sub

The subroutine name (fb) occurs from “forward-backward”: in the decom-

position method, we use the sweep from left to right (in the direction of the x
axis) and then from right to left (in the opposite direction).
If necessary, it is easy to obtain formulas of the “backward-forward” decom-
position method, in which the sweep from right to left is used, and then from left
to right.
The fb subroutine parameters have the following sense:
k, r are numbers of the left and right boundary nodes of the main grid on
segment [a, b] (Fig. 3.1);
h is a grid step, h (b a) /( r k ) ;
A, B are values of the solution of equation (3.6) on the left and right
boundaries of segment [a, b] according to boundary conditions (3.7) and (3.8);
G, E are arrays of values of the coefficients of equation (3.6) at the nodes
of the main grid;
F is an array of values of the right-hand side of equation (3.6);
U is an array intended for the solution values.

211
 Chapter 3. Finite Difference Method for Solving Differential Equations

3.6. Examples of using the decomposition method

As an example of using the decomposition method, we will solve equation

d 2u du
x2 3c x u 0 (3.39)
dx 2 dx

on segment 0 ≤ x ≤ b with boundary conditions

u (0) 1 , u (b) 0 (3.40)
for c = 10 and b = 1.5.
According to Task 3 of the second chapter in book [5], equation (3.39) was
obtained while studying temperature characteristics of a radial flow between
parallel round disks. The radial flow is a horizontal movement of liquid, gas or
plasma from the general center or to the center.
Equation (3.39) can be written in form (3.6), where
g ( x) x 2 , e( x ) 3c x , f (x) 0.
It is easy to verify that conditions (3.23) and (3.24) are satisfied for all values
of x from 0 to b if h 2 / b 2 = 0.889 (that is, if r – k ≥ 2). Besides, according
to (3.16) and (3.11), Pk 1 0 , i.e., condition (3.22) is satisfied too. Thus, the
sufficient stability conditions for the decomposition method are satisfied if
r – k ≥ 2.
For solving problem (3.39), (3.40), let us consider a program with the follow-
ing source data table.
c 10
b 1.5
l 15
x u

The first three rows of this table contain the values of parameters c and b and
the number of steps, l = r – k. The bottom row contains titles of the Excel
columns intended for the solution result, i.e., for the u (x) function in tabular
form.

212
 3.6. Examples of using the decomposition method

Instead of text

Sub main()

End Sub

we enter the following program text into Module1 of the BookNM workbook:
Listing 3.2
Sub main()
Dim X() As Double
Dim G() As Double
Dim E() As Double
Dim F() As Double
Dim U() As Double
Dim c As Double, b As Double, l As Integer
Dim h As Double, i As Integer
c = Selection.Cells(1, 2)
b = Selection.Cells(2, 2)
l = Selection.Cells(3, 2)
h = b / l
ReDim X(5 To 5 + l)
ReDim G(5 To 5 + l)
ReDim E(5 To 5 + l)
ReDim F(5 To 5 + l)
ReDim U(5 To 5 + l)
For i = 5 To 5 + l
X(i) = (i - 5) * h
1: G(i) = X(i) ^ 2
2: E(i) = -3 * c * X(i)
3: F(i) = 0
Next i
4: Call fb(5, 5 + l, h, 1, 0, G, E, F, U)
For i = 5 To 5 + l
Selection.Cells(i, 1) = X(i)
5: Selection.Cells(i, 2) = U(i)
Next i
End Sub

It was mentioned above that the source data for this program are the values
located in the Excel table (Fig. 3.2a). We must select this table before the pro-
gram execution (Fig. 3.2b).

213
 Chapter 3. Finite Difference Method for Solving Differential Equations

Fig. 3.2. The Excel worksheet (a) before and

(b) after selection of the source data table

The calculated coordinates of the grid nodes and values of the solution are
located in columns x and u (Fig. 3.3). For obtaining the u (x) graph, located on
the Excel worksheet, we must fulfill the following operations:
1) select the values of x and u, i.e., the B6:C21 range;
2) activate the Insert tab;
3) perform the Insert Scatter command in area Charts;
4) perform command Scatter with Smooth Lines.

214
 3.6. Examples of using the decomposition method

Fig. 3.3. The calculated dependence, u (x) , and its graph

By substitution (3.27),
x
x3
u U ( x ) exp 0.5 y 2 dy U ( x ) exp , (3.41)
0
6
we can bring boundary value problem (3.39), (3.40) to form (3.30), (3.33),
(3.36), i.e.,
d 2U
[0.25x 4 (3c 1) x ] U 0, (3.42)
2
dx
U (0) U (b) 0 .
1, (3.43)
As we see, derivative dU / dx is absent in equation (3.42), i.e., the 1st item
of the conditions on p. 207 is satisfied. The 2nd and 3rd items are also satisfied
because:
the coefficient in front of U is less or equal to zero when x ≥ 0;
Pk 1 0 according to (3.16) and (3.11).
Thus, the decomposition method is unconditionally stable for solving boundary
value problem (3.42), (3.43).
The following program is used as the previous one.

215
 Chapter 3. Finite Difference Method for Solving Differential Equations

Listing 3.3
Sub main()
Dim X() As Double
Dim G() As Double
Dim E() As Double
Dim F() As Double
Dim U() As Double
Dim c As Double, b As Double, l As Integer
Dim h As Double, i As Integer
c = Selection.Cells(1, 2)
b = Selection.Cells(2, 2)
l = Selection.Cells(3, 2)
h = b / l
ReDim X(5 To 5 + l)
ReDim G(5 To 5 + l)
ReDim E(5 To 5 + l)
ReDim F(5 To 5 + l)
ReDim U(5 To 5 + l)
For i = 5 To 5 + l
X(i) = (i - 5) * h
1: G(i) = 0
2: E(i) = -0.25 * X(i) ^ 4 - (3 * c + 1) * X(i)
3: F(i) = 0
Next i
4: Call fb(5, 5 + l, h, 1, 0, G, E, F, U)
For i = 5 To 5 + l
Selection.Cells(i, 1) = X(i)
5: Selection.Cells(i, 2) = U(i) * _
Exp(-X(i) ^ 3 / 6)
Next i
End Sub

This program solves boundary value problem (3.42), (3.43) and then calcu-
lates the u (x) dependence by means of formula (3.41). It differs from program
Listing 3.2 in operators 1, 2 and 5. The result of using program Listing 3.3, the
u (x) function in tabular form, is close to the previous result depicted in Fig. 3.3.
Boundary value problem (3.42), (3.43) will be also used in Section 4.11 to
demonstrate the cubic spline method for solving the second-order linear differen-
tial equation.

216
 3.7. Examples of the computing error. Instability and loss of accuracy

3.7. Examples of the computing error.

Instability and loss of accuracy

The decomposition method usage can lead to an appreciable computing error.

We will consider this question on an example of equation
d 2u du
c c (3.44)
dx 2 dx
on segment 0 ≤ x ≤ b with boundary conditions
u (0) 0 , u(b) b . (3.45)
It is obvious that the solution of this boundary value problem looks like
u ( x) x for any value of the c constant.
Equation (3.44) can be written in form (3.6), where
g ( x) f ( x ) c ,
e(x) 0 .
To solve problem (3.44), (3.45) by the decomposition method, we change
operators 1 — 4 of program Listing 3.2 to get the following program:
Listing 3.4
Sub main()
Dim X() As Double
Dim G() As Double
Dim E() As Double
Dim F() As Double
Dim U() As Double
Dim c As Double, b As Double, l As Integer
Dim h As Double, i As Integer
c = Selection.Cells(1, 2)
b = Selection.Cells(2, 2)
l = Selection.Cells(3, 2)
h = b / l
ReDim X(5 To 5 + l)
ReDim G(5 To 5 + l)
ReDim E(5 To 5 + l)

217
 Chapter 3. Finite Difference Method for Solving Differential Equations

ReDim F(5 To 5 + l)
ReDim U(5 To 5 + l)
For i = 5 To 5 + l
X(i) = (i - 5) * h
1: G(i) = c
2: E(i) = 0
3: F(i) = c
Next i
4: Call fb(5, 5 + l, h, 0, b, G, E, F, U)
For i = 5 To 5 + l
Selection.Cells(i, 1) = X(i)
5: Selection.Cells(i, 2) = U(i)
Next i
End Sub

This program is used as the programs of the previous section. In Fig. 3.4, we
see the execution results for three values of c, equal to 10, 1017 and 1018.
For problem (3.44), (3.45), it is easy to make sure that conditions (3.22) and
(3.23) are satisfied for any positive values of c and h.
For h = b / l = 0.1, we can state the following about
i
1 0.5 g i h 1 0.5ch 1 0.05c ,
condition (3.24) and the stability with respect to the computing error:
for c = 10, the value of i is positive, i.e., condition (3.24) is satisfied,
therefore, the decomposition method is stable;
for c = 1017 and 1018, the value of i is negative, i.e., condition (3.24)
is not satisfied, therefore, the decomposition method can be either stable or
unstable.
The sign change of i occurs at c = 20.
According to Fig. 3.4a and 3.4b, an appreciable deviation of the calculated
u (x) dependence from the exact u (x) dependence, u ( x) x , appears at value
c = 1017, which exceeds value c = 20 by several orders of magnitude. It follows
from the fact that sufficient stability conditions (3.22) — (3.24) are not neces-
sary, i.e., the process of solving the finite difference scheme for boundary value
problem (3.6) — (3.8) by the decomposition method may remain stable with
respect to the computing error if not all conditions from (3.22) — (3.24) are
satisfied.
According to Fig. 3.4c, the computing error of the solution can exceed 100 %
for large values of c.
Let us use substitution (3.27) for solving problem (3.44), (3.45).

218
 3.7. Examples of the computing error. Instability and loss of accuracy

Fig. 3.4. Graphic results of solving problem (3.44), (3.45) by program Listing 3.4
for b = 1.5, l = 15 and the following values of c: 10 (a), 1017 (b), 1018 (c)

219
 Chapter 3. Finite Difference Method for Solving Differential Equations

Substitution
x
u U ( x ) exp 0.5 g ( y ) dy
a

x
U ( x ) exp 0.5 cdy U ( x ) exp( 0.5cx ) (3.46)
0

into (3.44), (3.45) gives the following equation and boundary conditions:
d 2U
0.25c 2 U c exp(0.5cx ) , (3.47)
dx 2
U (0) 0 , U (b) b exp(0.5cb) . (3.48)
For boundary value problem (3.47), (3.48), it is easy to make sure that all the
conditions, formulated on p. 207, are satisfied when x ≥ 0, i.e., the decomposition
method is unconditionally stable.
The following program is used as the previous one.
Listing 3.5
Sub main()
Dim X() As Double
Dim G() As Double
Dim E() As Double
Dim F() As Double
Dim U() As Double
Dim c As Double, b As Double, l As Integer
Dim h As Double, i As Integer
c = Selection.Cells(1, 2)
b = Selection.Cells(2, 2)
l = Selection.Cells(3, 2)
h = b / l
ReDim X(5 To 5 + l)
ReDim G(5 To 5 + l)
ReDim E(5 To 5 + l)
ReDim F(5 To 5 + l)
ReDim U(5 To 5 + l)
For i = 5 To 5 + l
X(i) = (i - 5) * h
1: G(i) = 0
2: E(i) = -0.25 * c ^ 2
3: F(i) = c * Exp(0.5 * c * X(i))

220
 3.7. Examples of the computing error. Instability and loss of accuracy

Next i
4: Call fb(5, 5 + l, h, 0, b * Exp(0.5 * c * b), _
G, E, F, U)
For i = 5 To 5 + l
Selection.Cells(i, 1) = X(i)
5: Selection.Cells(i, 2) = U(i) * _
Exp(-0.5 * c * X(i))
Next i
End Sub

Program Listing 3.5 solves boundary value problem (3.47), (3.48) and then
calculates the u (x) dependence by means of formula (3.46). This program dif-
fers from Listing 3.4 in operators 1 — 5.
Fig. 3.5 shows the results of using program Listing 3.5 at c = 10 and 20.
Unlike program Listing 3.4, program Listing 3.5 was not used at c = 1017 and
1018 because of exceeding 709.782712893 by the argument of the exponential
function in operator 3. In other words, according to Fig. 3.6, the stop of the exe-
cution of program Listing 3.5 would occur at c = 1017 and 1018 with the follow-
ing information: Run-time error ‘6’: Overflow.

Fig. 3.5. Graphic results of using program Listing 3.5:

the continuous line — c = 10, l = 15;
the dashed line — c = 10, l = 30;
the dash-dotted line — c = 20, l = 15

According to the continuous and dashed lines in Fig. 3.5, we must reduce the
number of steps on segment [0, b] for improving the accuracy. According to the
dash-dotted line, the accuracy may be lost even when the computing process is
stable with respect to the computing error. We will speak about measures against
loss of accuracy also in section “Instead of Conclusions”.

221
 Chapter 3. Finite Difference Method for Solving Differential Equations

Fig. 3.6. The Excel Help window with the Exp function description

It is natural to ask about the existence of a method for solving the finite dif-
ference scheme for boundary value problem (3.6) — (3.8), which is more stable
with respect to the computing error than the decomposition method used by us.
Before answering this question, we will consider several methods for solving the
system of linear algebraic equations of general form.

222
 3.8. Solving the system of linear algebraic equations by using Excel functions

3.8. Solving the system of linear algebraic

equations by using Excel functions

We should solve the system of linear algebraic equations

a11 x1 a12 x2 ... a1n xn f1 ,

a 21 x1 a 22 x2 ... a 2 n xn f2 ,
(3.49)
. . . . . . . . .
an1 x1 a n 2 x2 ... a nn xn fn

with coefficient (system) matrix

a11 a12 ... a1n

a 21 a 22 ... a 2 n
A (3.50)
. . . .
a n1 an2 ... a nn

of general form: any element aij can be either zero or nonzero.

According to the definitions of matrix multiplication and equality, given in
Section 1.21, system (3.49) can be written as matrix equation
Ax f , (3.51)
where f is the vector of the right-hand sides, x is the vector of the unknown vari-
ables:

f1 x1
f2 x2
f , x .
... ...
fn xn

System (3.49) or, which is the same, matrix equation (3.51) can be solved by
means of Excel, without programming in VBA. Let us consider two methods of
the solution.
1. The solution of equation (3.51) can be written as x = A-1f, where A-1 is the
inverse A matrix (Section 1.21).

223
 Chapter 3. Finite Difference Method for Solving Differential Equations

The A-1 matrix can be calculated by means of the built-in MINVERSE

function. The subsequent multiplication of the A-1 matrix and the f vector can be
performed by means of the built-in MMULT function. Functions MINVERSE
and MMULT return an array, as well as built-in function TRANSPOSE and
user-defined function TRANSPOSEA developed by us (Section 2.3).
The drawback of this method for solving matrix equation (3.51) is its ineffi-
ciency (with respect to the execution time) for large values of n because the
matrix inversion is essentially the solution of the following n matrix equations:
ATx f 1 , ATx f 2 , ..., A T x fn ,
where A T is the transposed A matrix,
1 0 0
0 1 0
f1 , f2 , ..., f n .
... ... ...
0 0 1
Indeed, if
x11 x12 x1n

x1 x 12 , x 2 x 22 , ..., x n x 2n
... ... ...
x 1n x n2 x nn
are the solutions of the above n matrix equations and

x11 x 12 ... x 1n

B x12 x 22 ... x n2 , (3.52)

. . . .
x1n x 2n ... x nn
it is easy to make sure that BA = E, where E is the unit matrix, that is, B = A-1.
2. Matrix equation (3.51) can be solved by Cramer’s rule [3], which is
reduced to computation of n + 1 determinants of n-th order, D, D1, D2, ...., Dn.
These determinants are calculated by using the MDETERM function. For large
values of n, this method is even less efficient than the previous one.
Many numerical methods need the solution of system (3.49) or matrix equa-
tion (3.51). For this purpose, let us develop two subroutines: the first subroutine
realizes the classical Gaussian elimination method; the second one realizes
a modernization of this method.

224
 3.9. Solving the system of linear algebraic equations by the Gaussian elimination method

3.9. Solving the system of linear algebraic equations

by the Gaussian elimination method

In a summary, the classical Gaussian elimination method can be described as

follows.
Using the first equation of the system of n equations (3.49), the x1 unknown
is eliminated from the remaining n 1 equations. Further, by means of the 2nd
equation of the resulting system (of n equations), x 2 is eliminated from the next
n 2 equations. By using the 3rd equation of the new system (of n equations),
x 3 is eliminated from the next n 3 equations. And so on, until we get the
equation determining the x n unknown. In the last turn, the remaining n 1 un-
knowns are calculated in the reverse sequence: xn 1,
xn 2 , ..., x1 . To do this,
the already calculated unknowns are substituted into the equations, starting with
the penultimate equation.
In more details, the Gaussian elimination method includes two stages, which
are called the forward and backward courses. The forward course consists in the
transformation of matrix equation (3.51) to equation
Bx g (3.53)
with matrix
1 b12 ... b1n
0 1 ... b2 n
B (3.54)
. . . .
0 0 ... 1
called an upper triangular matrix because nonzero elements are on the main
diagonal and above it. The backward course is the solution of matrix equation
(3.53) relative to the x vector of the unknowns.
Let us consider the forward course of the Gaussian elimination method.
We assume that the a11 coefficient is nonzero. In this case, when dividing
both sides of the 1st equation of system (3.49) by this coefficient, we obtain
equation

225
 Chapter 3. Finite Difference Method for Solving Differential Equations

x1 b12 x2 ... b1n xn g1 , (3.55)

a1 j f1
where b1 j , g1, j = 2, ..., n. The last equation is the result of the first
a11 a11
step of the forward course.
Let us multiply both sides of (3.55) by a21 and subtract the resulting equa-
tion from the 2nd equation of system (3.49). We obtain an equation without x1 .
Let us multiply both sides of (3.55) by a31 and subtract the resulting equation
from the 3rd equation of system (3.49). We obtain one more equation without
x1 , and so on.
Let us multiply both sides of (3.55) by a n1 and subtract the resulting equation
from the n-th equation of system (3.49).
The result is the following system of linear algebraic equations without x1 :

a [222 ] x2 ... a [22n] xn f 2[ 2 ] ,

. . . . . . . (3.56)
a [n22] x2 ... a [nn2 ] xn f n[ 2 ] ,

where aij[ 2] aij ai1b1 j , fi[ 2] fi ai1 g1 , i = 2, ..., n, j = 2, ..., n. The index in
square brackets is the number of the current step of the forward course.
Let us assume that the a[222] coefficient is nonzero. Dividing both sides of the
1st equation of system (3.56) by this coefficient, we obtain equation
x2 b23 x3 ... b2 n xn g2 , (3.57)

a [22j] f 2[ 2 ]
where b2 j , g2 , j = 3, ..., n. The last equation is the result of the
a [222 ] a [222 ]
second step of the forward course.
To eliminate x 2 , we fulfill the same operations with equation (3.57) and sys-
tem (3.56) that we previously fulfilled with equation (3.55) and system (3.49).
Equation
xn gn , (3.58)

where g n f n[ n ] / a[nnn ] , is the result of the n-th step of the forward course.

226
 3.9. Solving the system of linear algebraic equations by the Gaussian elimination method

Equations (3.55), (3.57) and (3.58) give the following system of linear alge-
braic equations:
x1 b12 x2 ... b1n xn g1 ,
x2 ... b2 n xn g2 ,
(3.59)
. . . . .
xn gn .
This system is the result of the forward course of the Gaussian elimination
method. We can write (3.59) in the form of matrix equation (3.53) with upper
triangular matrix (3.54).
The backward course consists in the determination of the unknown variables
by solving system (3.59). The following expressions are used:
1) according to the ultimate equation,
xn gn ;
2) according to the penultimate equation,
xn 1 gn 1 bn x ;
1, n n
3) according to the (n – 2)th equation,
xn 2 g n 2 bn 2, n x
1 n 1
bn x ,
2, n n
and so on.
An analysis of the forward course of the Gaussian elimination method shows
that it requires O( n 3 ) arithmetic operations, n . Similarly, the backward

course requires O( n 2 ) arithmetic operations. Thus, the solution of the system of

linear algebraic equations (3.49) by the Gaussian elimination method requires
O( n 3 ) arithmetic operations.

227
 Chapter 3. Finite Difference Method for Solving Differential Equations

3.10. Two subroutines for solving the system

of linear algebraic equations

Into Module3 of the BookNM workbook, we put the following declaration of

the gaus subroutine, realizing the classical Gaussian elimination method.

Listing 3.6
Sub gaus(ByVal n, ByRef A() As Double, _
ByRef F() As Double, ByRef X() As Double)
Dim i As Integer, j As Integer, k As Integer
Dim AA() As Double: ReDim AA(1 To n, 1 To n)
Dim FF() As Double: ReDim FF(1 To n)
Dim B() As Double: ReDim B(1 To n, 1 To n)
Dim G() As Double: ReDim G(1 To n)
'Forward course:
For i = 1 To n
For j = 1 To n
AA(i, j) = A(i, j)
Next j
FF(i) = F(i)
Next i
For k = 1 To n - 1
For j = k + 1 To n
B(k, j) = AA(k, j) / AA(k, k)
Next j
G(k) = FF(k) / AA(k, k)
'end of step No. k
'beginning of step No. k + 1
For i = k + 1 To n
For j = k + 1 To n
AA(i, j) = AA(i, j) - AA(i, k) * _
B(k, j)
Next j
FF(i) = FF(i) - AA(i, k) * G(k)
Next i

228
 3.10. Two subroutines for solving the system of linear algebraic equations

Next k
G(n) = FF(n) / AA(n, n)
'Backward course:
X(n) = G(n)
For k = n - 1 To 1 Step -1
X(k) = G(k)
For j = k + 1 To n
X(k) = X(k) - B(k, j) * X(j)
Next j
Next k
End Sub

The gaus subroutine parameters have the following sense:

n is the number of the unknowns;
A is an array corresponding to matrix (3.50) of matrix equation (3.51);
F, X are arrays corresponding to the f and x vectors in (3.51).
Below, we will consider the program realization of the Gaussian elimination
method with choice of leading coefficient.
The leading coefficient is the coefficient, by which the division is performed.
Thus, at the first step of the classical Gaussian elimination method’s forward
course, a11 is the leading coefficient; at the second step, a[222] is the leading coef-
ficient, and so on. Sometimes, the division is incorrect because the leading coef-
ficient may be very small (in absolute value) or equal to zero. Therefore, the fol-
lowing three modernizations of the Gaussian elimination method are used in
practice.
1. At each step of the forward course, permuting the equations is performed
to get the maximum (in absolute value) element of the first column of the system
matrix as the leading coefficient. Thus, at the first step, matrix (3.50) of system
(3.49) is processed; at the second step, matrix

a [222 ] ... a [22n]

. . .
a [n22] ... a [nn2 ]

of system (3.56) is processed, and so on. In this case, we say that the leading
coefficient is chosen in the column.
2. Renumbering the unknowns is performed to get the maximum element of
the first row of the system matrix as the leading coefficient. In this case, we say
that the leading coefficient is chosen in the row.

229
 Chapter 3. Finite Difference Method for Solving Differential Equations

3. Permuting the equations and renumbering the unknowns are performed

to get the maximum element of the system matrix as the leading coefficient. That
is, the whole matrix is considered when choosing the leading coefficient.
Into Module4 of the BookNM workbook, we enter the following declaration
of the gauss subroutine, which realizes the Gaussian elimination method with
choice of leading coefficient in the matrix.
Listing 3.7
Sub gauss(ByVal n, ByRef A() As Double, _
ByRef F() As Double, ByRef X() As Double, _
Optional c = 0, Optional d = 0)
Dim i As Integer, j As Integer, k As Integer
Dim i_max As Integer, j_max As Integer
Dim m As Integer
Dim w As Double
Dim AA() As Double: ReDim AA(1 To n, 1 To n)
Dim FF() As Double: ReDim FF(1 To n)
Dim O() As Integer: ReDim O(1 To n)
Dim XX() As Double: ReDim XX(1 To n)
Dim B() As Double: ReDim B(1 To n, 1 To n)
Dim G() As Double: ReDim G(1 To n)
1: For j = 1 To n
2: O(j) = j
3: Next j
'Forward course:
4: w = 0
5: For i = 1 To n
6: For j = 1 To n
7: AA(i, j) = A(c + i, d + j)
8: If Abs(AA(i, j)) > w Then
9: w = Abs(AA(i, j))
10: i_max = i
11: j_max = j
12: End If
13: Next j
14: FF(i) = F(c + i)
15: Next i
16: For j = 1 To n
17: w = AA(i_max, j)
18: AA(i_max, j) = AA(1, j)

230
 3.10. Two subroutines for solving the system of linear algebraic equations

19: AA(1, j) = w
20: Next j
21: w = FF(i_max)
22: FF(i_max) = FF(1)
23: FF(1) = w
24: For i = 1 To n
25: w = AA(i, j_max)
26: AA(i, j_max) = AA(i, 1)
27: AA(i, 1) = w
28: Next i
29: m = O(j_max)
30: O(j_max) = O(1)
31: O(1) = m
For k = 1 To n - 1
For j = k + 1 To n
B(k, j) = AA(k, j) / AA(k, k)
Next j
G(k) = FF(k) / AA(k, k)
'end of step No. k
'beginning of step No. k + 1
w = 0
For i = k + 1 To n
For j = k + 1 To n
AA(i, j) = AA(i, j) - AA(i, k) * _
B(k, j)
If Abs(AA(i, j)) > w Then
w = Abs(AA(i, j))
i_max = i
j_max = j
End If
Next j
FF(i) = FF(i) - AA(i, k) * G(k)
Next i
For j = k + 1 To n
w = AA(i_max, j)
AA(i_max, j) = AA(k + 1, j)
AA(k + 1, j) = w
Next j
w = FF(i_max)
FF(i_max) = FF(k + 1)
FF(k + 1) = w
For i = k + 1 To n

231
 Chapter 3. Finite Difference Method for Solving Differential Equations

w = AA(i, j_max)
AA(i, j_max) = AA(i, k + 1)
AA(i, k + 1) = w
Next i
m = O(j_max)
O(j_max) = O(k + 1)
O(k + 1) = m
For i = 1 To k 'permuting columns of
'calculated part of array B
w = B(i, j_max)
B(i, j_max) = B(i, k + 1)
B(i, k + 1) = w
Next i
Next k
G(n) = FF(n) / AA(n, n)
'Backward course:
XX(n) = G(n)
For k = n - 1 To 1 Step -1
XX(k) = G(k)
For j = k + 1 To n
XX(k) = XX(k) - B(k, j) * XX(j)
Next j
Next k
32: For j = 1 To n
33: X(d + O(j)) = XX(j)
34: Next j
End Sub

Among parameters of the gauss subroutine, we see optional parameters

c and d. They are necessary to have the possibility of solving the following
system of linear algebraic equations with non-traditional (shifted) numbering of
the system’s elements (matrix rows and columns, right-hand sides and unknown
variables):

ac 1, d
x
1 d 1
ac 1, d
x
2 d 2
... ac 1, d
x
n d n
fc 1
,
ac 2, d
x
1 d 1
ac 2, d
x
2 d 2
... ac 2, d
x
n d n
fc 2
,
. . . . . . . .
ac n, d
x
1 d 1
ac n, d
x
2 d 2
... ac n, d
x
n d n
fc n
,

where c and d are given integers. Such numbering is used in code Listing 6.15
(see operator 8 corresponding to n = c = d = 2).

232
 3.10. Two subroutines for solving the system of linear algebraic equations

The presence of optional parameters is one of distinctions of the gauss sub-

routine from the previous gaus.
Let us briefly consider the gauss subroutine.
Listing 3.7 includes the O array of original serial numbers of the unknown
variables. This array is necessary for recovery of the unknowns’ numbering,
which changes during the solution of system (3.49). The j-th element of the O
array is the original number ( but not shifted by d ) of the j-th unknown variable.
Operators 1 — 3 prepare array O for the subsequent transformation.
Operators 4 — 15:
1) determine elements of the AA array corresponding to matrix (3.50) of the
system of linear algebraic equations (operator 7);
2) find the maximum (in absolute value) element of the AA array; its serial
numbers in the vertical and horizontal directions are respectively assigned to
variables i_max and j_max (operators 8 — 12);
3) determine elements of the FF array corresponding to the f vector of the
right-hand sides of the system of linear algebraic equations (operator 14).
Operators 16 — 20 interchange the 1st and i_max-th rows of the AA array,
and operators 21 — 23 interchange the 1st and i_max-th elements of the FF
array. It is equivalent to permuting the equations of system (3.49).
Operators 24 — 28 interchange the 1st and j_max-th columns of the AA
array, and operators 29 — 31 interchange the 1st and j_max-th elements of the
O array. It is equivalent to changing the numbering of the unknowns.
Thus, because of executing operators 16 — 31, the maximum element of the
AA array appears in the top left corner of AA. Further, the divisions by this ele-
ment are performed (at the beginning of the k cycle at k = 1). The first step of
the forward course terminates on it.
In the subsequent steps of the forward course:
1) the elements of arrays AA and FF are calculated;
2) during calculation of the AA array values, the maximum element of the
AA array is determined; its serial numbers in the vertical and horizontal direc-
tions are determined too;
3) arrays AA, FF and O are transformed;
4) the columns of the already calculated part of the B array, corresponding to
the B matrix of system (3.59), are permuted because of the unknowns’ renum-
bering;
5) the divisions by the maximum element of the AA array are performed (at
the beginning of the k cycle at the next value of k).
After the considered operators of the forward course, obvious operators of the
backward course follow.

233
 Chapter 3. Finite Difference Method for Solving Differential Equations

The forward and backward courses’ results, contained in the XX array, are the
calculated values of the unknowns. Operators 32 — 34 determine variables
xd 1 , xd 2 , ..., xd n by recovering the original numbering.
The gauss subroutine usage does not guarantee the solution of the system of
linear algebraic equations (3.49) because in general it may not exist if the D
determinant of matrix (3.50) is equal to 0: the division by D is performed in
Cramer’s rule (p. 202).
The gaus and gauss subroutines allow us to invert the A matrix. For that,
we must use formula (3.52), where B = A-1.

234
 3.11. Reduction of the computing error

3.11. Reduction of the computing error

Let us return to solving boundary value problem (3.6) — (3.8).

The computing error can be reduced if we use the Gaussian elimination
method with choice of leading coefficient in the matrix for solving finite differ-
ence scheme (3.9), (3.11), (3.12). To verify this, let us put the following program
into Module1 of the BookNM workbook instead of the program located there.
Listing 3.8
Sub main()
Dim X() As Double
Dim G() As Double
Dim E() As Double
Dim F() As Double
Dim U() As Double
Dim c As Double, b As Double, l As Integer
Dim h As Double, i As Integer
Dim n As Integer, j As Integer, w As Double
Dim AA() As Double
Dim FF() As Double
Dim XX() As Double
c = Selection.Cells(1, 2)
b = Selection.Cells(2, 2)
l = Selection.Cells(3, 2)
h = b / l
n = l + 1
ReDim X(5 To 5 + l)
ReDim G(5 To 5 + l)
ReDim E(5 To 5 + l)
ReDim F(5 To 5 + l)
ReDim U(5 To 5 + l)
ReDim AA(1 To n, 1 To n)
ReDim FF(1 To n)
ReDim XX(1 To n)
For i = 5 To 5 + l

235
 Chapter 3. Finite Difference Method for Solving Differential Equations

X(i) = (i - 5) * h
1: G(i) = c
2: E(i) = 0
3: F(i) = c
Next i
For i = 1 To n
For j = 1 To n
AA(i, j) = 0
Next j
Next i
AA(1, 1) = 1: AA(1, 2) = 0: FF(1) = 0
For i = 2 To n - 1
w = 0.5 * G(i + 4) * h
AA(i, i - 1) = 1 - w
AA(i, i) = E(i + 4) * h ^ 2 - 2
AA(i, i + 1) = 1 + w
FF(i) = F(i + 4) * h ^ 2
Next i
AA(n, n - 1) = 0: AA(n, n) = 1: FF(n) = b
4: Call gauss(n, AA, FF, XX)
For i = 5 To 5 + l
Selection.Cells(i, 1) = X(i)
Selection.Cells(i, 2) = XX(i - 4)
Next i
End Sub

This program solves the finite difference scheme corresponding to boundary

value problem (3.44), (3.45). It is used as program Listing 3.4: the source data
are the values in the table (Fig. 3.2a), at that, we must select this Excel table
before the program execution (Fig. 3.2b).
Operator 4 in the above program is the gauss subroutine call, i.e., finite dif-
ference scheme (3.9), (3.11), (3.12) is solved by the modernized Gaussian elimi-
nation method, with choice of leading coefficient in the matrix. In this case, the
computing error is not observed even for c = 10308 (according to Appendix 1, this
is almost the maximum value for the Double data type), i.e., the calculation
result looks like Fig. 3.4a.
When using the classical Gaussian elimination method (that is, when replac-
ing gauss by gaus in operator 4), the computing error is identical to that
obtained when using the decomposition method. It is not surprising, because
the decomposition method, actually, is an efficient version of the classical
Gaussian elimination method for the system of linear algebraic equations with
the tridiagonal matrix.

236
 3.11. Reduction of the computing error

In practice, the Gaussian elimination method is not used for the solution
of finite difference scheme (3.9), (3.11), (3.12) because of its inefficiency
(with respect to the execution time) for large values of n r k 1 .
When c → ∞ in problem (3.44), (3.45), the computing error of the solution
can be reduced in another way, without usage of the modernized Gaussian elimi-
nation method.
This alternative method for solving problem (3.44), (3.45) at large values of
c includes the following two stages:
1) changing formulation of the boundary value problem by excluding the
second derivative, u (x) , and one of the boundary conditions, u (0) 0 or
u(b) b ;
2) solving the resulting boundary value problem for the first-order linear
differential equation,
du
1.
dx
Thus, we should exercise caution in formulation of the problem and in choice
of the solution method.

237
 Chapter 3. Finite Difference Method for Solving Differential Equations

3.12. Solving the nonlinear differential equation

by the quasilinearization method

On segment [a, b] , we will consider the following differential equation:

d 2u du
F x, u , , (3.60)
dx 2 dx

where F ( x, y, z ) is a continuous nonlinear function of variables x, y and z,

F F
whose first partial derivatives ( x, y , z ) and
( x, y , z ) exist and continu-
y z
ous, and the second partial derivatives with respect to y and z exist. Because of
the function nonlinearity, this equation is called a nonlinear differential equation.
As in the case of linear differential equation (3.6), the solution of nonlinear
differential equation (3.60) must satisfy the left and right boundary conditions,
(3.7) and (3.8).
In Section 6.10, we will solve the formulated boundary value problem by
the shooting method. Solving this problem by the quasilinearization method
considered below is the following iterative process: the boundary value problem
for a linear differential equation is being solved when calculating the ( j 1 )th
solution approximation over the known j-th approximation.
An initial approximation of the solution (corresponding to the zero value
of j ) must be given, and must satisfy boundary conditions (3.7) and (3.8).
Thus, the solution of the nonlinear problem is reduced to solving a series
of linear problems. Let us obtain the linear differential equation of the
quasilinearization method.
Using Taylor’s formula [3] for functions of two variables, we have
F ( x, y y, z z)
2
1 1
F ( x, y , z ) y z F ( x, y , z ) y z F
1! y z 2! y z
F F 2
F ( x, y , z ) y ( x, y , z ) z ( x, y , z ) O ( ), (3.61)
y z

238
 3.12. Solving the nonlinear differential equation by the quasilinearization method

where y and z are increments of the second and third arguments of function
F ( x, y, z ) , 0 is the maximum of quantities | y | and | z | .
The solution of equation (3.60) can be written in the following form:
u ( x) u j ( x) v( x) ,
where u j (x) is the known j-th approximation of solution u (x ) , v (x ) is a small
quantity. By substituting this expression into equation (3.60), we obtain
d 2u j d 2v du j dv
F x, u j v, .
dx 2 dx 2 dx dx

Using expression (3.61) without the summand of second order of smallness,

we lead the last equation to the following form:
d 2u j d 2v du j F du j F du j dv
F x, u j , x, u j , v x, u j ,
dx 2 dx 2 dx y dx z dx dx

or
d 2v dv
g ( x) e( x )v f ( x) , (3.62)
dx 2 dx
where
F du j
g ( x) x, u j , , (3.63)
z dx

F du j
e( x ) x, u j , , (3.64)
y dx

du j d 2u j
f ( x) F x, u j , . (3.65)
dx dx 2
Let u 0 ( x ) be the initial approximation of the solution of nonlinear differen-
tial equation (3.60), satisfying boundary conditions (3.7) and (3.8):
u 0 ( a ) A , u 0 (b ) B . (3.66)
In the quasilinearization method, the ( j 1 )th approximation of the u (x ) solu-
tion ( j = 0, 1, 2, ...) is calculated over the known j-th approximation as follows:

239
 Chapter 3. Finite Difference Method for Solving Differential Equations

1) the values of functions g (x) , e(x ) and f (x) at the internal nodes of the
main grid on segment a ≤ x ≤ b (Fig. 3.1) are calculated according to formulas
(3.63) — (3.65);
2) second-order linear differential equation (3.62) with zero boundary condi-
tions v(a) v(b) 0 is solved by the decomposition method; function v (x ) is
the result;
3) the ( j 1 )th approximation of the u (x ) solution is calculated according
to formula
u j 1 ( x) u j ( x) v( x) .
The iterative process can be terminated under various conditions; we will use
the following:
max f ( x ) , (3.67)
a x b
where φ is a given positive constant, f (x) is function (3.65) tending to zero at
all points of segment [a, b] when j → ∞.

240
 3.13. Solving the Shockley-Poisson equation

3.13. Solving the Shockley-Poisson equation

The quasilinearization method considered above will be used for simulation

of a silicon photosensitive target. Such targets numbering between one and three
are located behind the objective of a camera or movie camera. The incident light
is converted to electrical signals with help of these targets.
The photosensitive target consists of cells of identical structure, and there are
millions of them, as many as the number of pixels defining the maximum resolu-
tion of the camera, or several times more.
According to article [6], as a mathematical model of the cell, we can use the
following nonlinear differential equation:
d 2u qu( x )
q N 0 exp N A ( x) , (3.68)
2 kbT
dx
where u (x ) is the electric potential in volts, ε = 103.545·10-12 C / (V·m) is the
dielectric permittivity of silicon, q = 1.6·10-19 C is the absolute value of the elec-
tron charge, k b = 1.38·10-23 J / C is the Boltzmann constant, T = 300 K is the
absolute room temperature, N A (x ) is the difference between the acceptor and
donor concentrations, N A (a ) = N A (b ) = N 0 is a positive value, a ≤ x ≤ b.
To start, let us set the following coordinates of the boundaries of segment
[a, b] : a = – 0.5·10-6 m, b = 0.5·10-6 m. The boundary conditions are as follows:
u(a) u(b) 0 .
We can encounter this (or similar) equation in the physics of semiconductor
devices. It is the Poisson equation whose right-hand side (the electric charge
density) is a nonlinear function of electric potential u (x ) . Because W. Shockley
obtained the form of the right-hand side in [7], equation (3.68) is called the
Shockley-Poisson equation.
As we know, two types of carriers of electric current exist in a semiconduc-
tor: negatively charged electrons and positively charged holes. The electron has
charge – q, the hole has charge +q. Impurities of a semiconductor, which supply
electrons, are called donors; impurities, which supply holes, are called acceptors.

241
 Chapter 3. Finite Difference Method for Solving Differential Equations

When the donor molecule loses an electron, it becomes positively charged; the
acceptor molecule losing a hole becomes negatively charged.
We can write equation (3.68) in form (3.60), where
q qy
F ( x, y , z ) N A ( x) N 0 exp .
k bT
Because of absence of the z variable in the right-hand side of this expression,
g (x) 0 in differential equation (3.62) according to (3.63). Expressions (3.64)
and (3.65) take the following form:
q2 qu j ( x )
e( x ) N 0 exp ,
k bT k bT

q qu j ( x) d 2u j
f ( x) N A ( x) N 0 exp .
k bT dx 2
According to the previous section, the solution of the boundary value prob-
lem for equation (3.68) is reduced to the repeated solution of linear differential
equation (3.62) with the above dependences, g (x) , e(x ) and f (x) , and with
boundary conditions v(a) v(b) 0 .
It was shown at the end of Section 3.3 that the process of solving such linear
problem by the decomposition method is unconditionally stable.
Let us consider table Listing 3.9 with the following source data for program
Listing 3.10:
coordinates a and b of the boundaries;
the value of φ in condition (3.67) for finishing the iterative process;
the values of dependences N A (x ) and u 0 ( x ) at the nodes of the uniform
grid on segment a ≤ x ≤ b (the number of grid nodes is equal to the number of
values of N A or u 0 in the table, that is, 11).
According to the table, the N A (x ) dependence is symmetric to the origin of
coordinates, x = 0. At the ends of segment [a, b] , the semiconductor contains
acceptors whose concentration is equal to 7·1020 m-3 = N A (a ) = N A (b ) = N 0 .
At the midpoint (x = 0), the semiconductor contains donors whose concentration
is equal to 3·1022 m-3 = N A ( 0) .
The initial approximation of the solution is calculated by means of Excel
according to formula

242
 3.13. Solving the Shockley-Poisson equation

x
u0 ( x ) cos 1.
0.5 10 6
This u 0 ( x ) dependence satisfies boundary conditions (3.66) at A = B = 0.

Listing 3.9
a -5.00E-07
b 5.00E-07
phi 1.00E-01
NA u0
7.00E+20 0.00E+00
7.00E+20 1.91E-01
7.00E+20 6.91E-01
7.00E+20 1.31E+00
-3.00E+22 1.81E+00
-3.00E+22 2.00E+00
-3.00E+22 1.81E+00
7.00E+20 1.31E+00
7.00E+20 6.91E-01
7.00E+20 1.91E-01
7.00E+20 0.00E+00

The program below is intended for solving the boundary value problem for
the Shockley-Poisson equation by the quasilinearization method.
Listing 3.10
Sub main()
Dim NA() As Double
Dim U() As Double
Dim X() As Double
Dim U2() As Double
Dim V() As Double
Dim G() As Double
Dim E() As Double
Dim F() As Double
Dim m As Integer
Dim a As Double, b As Double
Dim phi As Double
Dim h As Double
Dim i As Integer, j As Integer

243
 Chapter 3. Finite Difference Method for Solving Differential Equations

Dim w1 As Double, w2 As Double

Dim w3 As Double, max As Double
Dim sb As String, se As String, sn As String
Const q = 1.6E-19
Const epsilon = 103.545E-12
Const kb = 1.38E-23
Const T = 300
m = Selection.Rows.Count 'quantity of rows
a = Selection.Cells(1, 2)
b = Selection.Cells(2, 2)
phi = Selection.Cells(3, 2)
h = (b - a) / (m - 5)
ReDim NA(5 To m)
ReDim U(5 To m)
ReDim X(5 To m)
ReDim U2(5 To m)
ReDim V(5 To m)
ReDim G(5 To m)
ReDim E(5 To m)
ReDim F(5 To m)
Selection.Cells(4, 3) = "x"
w1 = q / epsilon: w2 = q / (kb * T)
For i = 5 To m
NA(i) = Selection.Cells(i, 1)
U(i) = Selection.Cells(i, 2)
X(i) = (i - 5) * h + a
Selection.Cells(i, 3) = X(i)
G(i) = 0
Next i
For j = 1 To 1000
max = 0
For i = 6 To m - 1
w3 = NA(5) * Exp(-w2 * U(i))
E(i) = -w1 * w2 * w3
U2(i) = (U(i + 1) - 2 * U(i) + _
U(i - 1)) / h ^ 2 'second derivative
F(i) = w1 * (NA(i) - w3) - U2(i)
If Abs(F(i)) > max Then max = Abs(F(i))
Next i
Call fb(5, m, h, 0, 0, G, E, F, V)
For i = 5 To m
U(i) = U(i) + V(i)

244
 3.13. Solving the Shockley-Poisson equation

Next i
If max < phi Then Exit For
Next j
Selection.Cells(4, 4) = "u"
For i = 5 To m
Selection.Cells(i, 4) = U(i)
Next i
1: sb = Selection.Cells(5, 3).Address
2: se = Selection.Cells(m, 4).Address
3: sn = ActiveSheet.Name
4: Range(sb & ":" & se).Select
5: Selection.NumberFormat = "0.0E+00"
6: Charts.Add
7: ActiveChart.ChartType = xlXYScatterSmoothNoMarkers
8: ActiveChart.SetSourceData Source:= _
Sheets(sn).Range(sb & ":" & se), PlotBy:= _
xlColumns
9: ActiveChart.Location Where:= xlLocationAsObject, _
Name:=sn
10: ActiveChart.Axes(xlValue).MajorGridlines.Select
11: Selection.Delete
12: ActiveChart.Legend.Select
13: Selection.Delete
End Sub

We enter this program into Module1 of the BookNM workbook.

Data of Listing 3.9 are contained in text file Listing_3_09.txt, which is on the
enclosed CD. To copy this data into range B2:C16 on the Sheet2 worksheet of
the BookNM workbook, we fulfill the following operations, which are close to
operations described on p. 26:
1) open the Listing_3_09.txt file with Notepad, for example, by double click
on the pictogram of this file in Windows Explorer;
2) in the Notepad window opened, highlight the table text and copy it into
Windows Clipboard, for example, by pressing Ctrl + C;
3) on the Sheet2 worksheet of the BookNM workbook, select the B2 cell by
clicking on it;
4) paste the Windows Clipboard contents into the B2:C16 range, for exam-
ple, by pressing Ctrl + V;
5) close the Notepad window with the Listing_3_09.txt file.
Using the standard features of Excel, the resulting table can be decorated as
in Fig. 3.7.

245
 Chapter 3. Finite Difference Method for Solving Differential Equations

Before the program execution, we must select the Excel table depicted in
Fig. 3.7. The execution results are as follows:
the coordinate and solution values located in the x and u columns, respec-
tively (Fig. 3.8);
the u (x) graph on the Excel worksheet.
Operators 1 — 13, intended for constructing the u (x) graph, were pro-
grammed by means of Excel Macro Recorder (Sections 2.4 and 2.5). Let us con-
sider the appointment of these operators.

Fig. 3.7. The Excel table with the source data

Operator 1 assigns the address of the first cell of the x column (that is, string
"D6") to the sb variable; operator 2 assigns the address of the last cell of the u
column (that is, string "E16") to the se variable; operator 3 assigns the name of
the active Excel worksheet to the sn variable. Operator 4 selects the x and u
columns (that is, range D6:E16); operator 5 assigns the necessary numerical
format to the selected cells. Operators 6 — 9 construct the graph; they corre-
spond to the 2nd, 3rd and 4th operations on p. 214. Operators 10 and 11 delete
the gridlines from the graph area; operators 12 and 13 delete the legend.

246
 3.13. Solving the Shockley-Poisson equation

Fig. 3.8. The program execution results

The calculated spatial distribution of the electric potential, u (x ) , is not realis-

tic. The fact is that the derivative,
du
u ( x) ,
dx
must equal zero at the ends of segment [a, b] according to the physics of semi-
conductor devices, i.e., u (a) u (b) 0 must be, but we do not see it in
Fig. 3.8.
To obtain realistic distribution u (x ) , we add areas with length of 9.5·10-6 m
and N A = N 0 = 7·1020 m-3 to segment [a, b] on the left- and right-hand sides.
We leave the grid step, h, unchanged, at that, the number of steps increases
20-fold.
The left and right boundaries of new segment [a, b] have the following
coordinates: a = -10-5 m, b = 10-5 m. The initial approximation of the solution
of equation (3.68), satisfying boundary conditions (3.66) at A B 0 , is calcu-
lated according to formula
x
u0 ( x ) cos 1 (3.69)
5
10
by means of Excel.
The u (x) graph, constructed by program Listing 3.10, is given in Fig. 3.9.
We see that u (a) u (b) 0 , in other words, function u (x ) is flat at the ends of
segment [a, b] .

247
 Chapter 3. Finite Difference Method for Solving Differential Equations

Fig. 3.9. The realistic spatial distribution of the electric potential: the horizontal
coordinate, x, is in meters; the vertical coordinate, u, is in volts

248
 3.14. Finite difference analogs of derivatives for a nonuniform grid

3.14. Finite difference analogs of derivatives

for a nonuniform grid

The execution time for solving the boundary value problem can be reduced
by replacing the uniform grid on [a, b] with a nonuniform grid whose step,
xi xi 1
hi , depends on i. At the transition to a nonuniform grid, expressions
(3.4) and (3.5) for the first and second derivatives of the u (x) function at node
xi become more complicated.
To obtain new expressions for the derivatives, we introduce axis z parallel to
the x axis (Fig. 3.10). If the origin of coordinates (z = 0) is at the xi node of the
grid, the xi 1
node has coordinate z hi , and the xi 1
node has coordinate
z hi 1
.

Fig. 3.10. The parabola passing through points

( xi 1 , u i 1 ), ( xi , ui ) and ( xi 1 , u i 1 )

249
 Chapter 3. Finite Difference Method for Solving Differential Equations

Let us consider the following second-degree polynomial:

P( z ) z2 z ui , (3.70)
where ui is the u (x) function value at the xi node; P (0) ui .
Coefficients α and β are determined by equations P ( hi ) ui 1
and
P ( hi 1
) ui 1
, which can be written as follows:

hi2 hi ui 1
ui ,

hi2 1
hi 1
ui 1
ui .

Solving this system of linear algebraic equations by Cramer’s rule [3], we obtain
D1 D2
, , (3.71)
D D
where
D hi hi 1 ( hi hi 1 ) ,
D1 (u i 1
ui ) hi (u i 1
ui ) hi 1
, (3.72)

D2 (ui 1
ui ) hi2 (ui 1
ui )hi2 1 .
Differentiating polynomial (3.70) twice, we obtain
dP
( z) 2 z , (3.73)
dz
d 2P
(z) 2 .
2
dz
From here, expressions for the derivatives at z = 0 follow:
dP D2 (u i 1
u i ) hi2 (u i 1
u i ) hi2 1
( 0) , (3.74)
dz D hi hi 1
( hi hi 1
)

d 2P D1 (u i 1
u i ) hi (u i 1
u i ) hi 1
( 0) 2 2 2 . (3.75)
dz 2 D hi hi (h
1 i
hi 1
)
We have the finite difference analogs of the first and second derivatives of
the u (x) function at the xi node.
Using expression (3.1), we can show the following:

250
 3.14. Finite difference analogs of derivatives for a nonuniform grid

du (u i 1
u i ) hi2 (u i 1
u i ) hi2 1
hi hi 1 d 3u
( xi ) ( xi ) ... ,
dx hi hi 1
( hi hi 1
) 6 dx 3

2
d u (u i 1
u i ) hi (u i 1
u i ) hi 1
hi hi 1 d 3u
( xi ) 2 ( xi ) ...
dx 2 hi hi 1
( hi hi 1
) 3 dx 3

or
du (u i 1
ui ) hi2 (u i 1
ui ) hi2 1 2
( xi ) O( hmax ), (3.76)
dx hi hi (h
1 i
hi 1
)

d 2u (u i 1
ui ) hi (u i 1
ui ) hi 1
( xi ) 2 O( hmax ) , (3.77)
dx 2 hi hi (h
1 i
hi 1
)

where hmax max {hi } is the maximum step of the grid, hmax 0.
k 1 i r
The resulting expressions, (3.76) and (3.77), are similar to expressions (3.4)
and (3.5). Naturally, (3.76) and (3.77) become (3.4) and (3.5), respectively, for
the constant step ( hi hi 1 h ).

Expressions (3.5) and (3.77) include summands O( h 2 ) and O ( hmax ) , respec-

tively. That is, at the transition from a uniform grid to a nonuniform grid, the
error of the finite difference analog of the second derivative changes from the
2nd order of smallness to the 1st order. It means deterioration of the finite differ-
ence approximation accuracy of the second derivative at the transition from
a uniform grid to a nonuniform grid.

251
 Chapter 3. Finite Difference Method for Solving Differential Equations

3.15. The decomposition method

for a nonuniform grid

Substituting expressions (3.74) and (3.75) into linear differential equation

(3.6) instead of u (x) and u (x) , respectively, we obtain linear algebraic equa-
tion (3.9),
u
i i 1
u
i i
u
i i 1 i
,
where
hi 1
(2 g i hi 1
)
i
,
hi hi 1

i
ei hi hi 1
g i ( hi hi 1
) 2,
hi ( 2 g i hi )
i
, (3.78)
hi hi 1

i
fi hi hi 1
.

Equation (3.9) for i = k + 1, k + 2, ..., r – 2, r – 1 and boundary conditions

(3.11) and (3.12) still form the system of linear algebraic equations with the
tridiagonal matrix, which can be solved by the decomposition method as follows:
1) at first, the forward sweep is performed, i.e., unknown Pk 1
, Qk 1
,
Pk 2
, Qk 2
, ..., Pr , Q r are calculated according to formulas (3.16), (3.17) and
(3.14), (3.15);
2) then the backward sweep is performed, i.e., unknown u r , u r 1
, ..., u k
are calculated according to formulas (3.18) and (3.13).
Let us put the following declaration of the subroutine, realizing the decompo-
sition method for differential equation (3.6) with boundary conditions (3.11) and
(3.12) for a nonuniform grid, into Module5 of the BookNM workbook.
Listing 3.11
Sub foba(ByVal k, ByVal r, ByRef X() As Double, _
ByRef G() As Double, ByRef E() As Double, _

252
 3.15. The decomposition method for a nonuniform grid

ByRef F() As Double, _

ByVal GAMMAK, ByVal DELTAK, _
ByVal ALPHAR, ByVal DELTAR, _
ByRef U() As Double)
Const BETAK = -2, BETAR = -2
Dim alpha As Double, beta As Double
Dim gamma As Double, delta As Double
Dim i As Integer, w As Double
Dim H() As Double: ReDim H(k + 1 To r)
Dim P() As Double: ReDim P(k + 1 To r)
Dim Q() As Double: ReDim Q(k + 1 To r)
For i = k + 1 To r
H(i) = X(i) - X(i - 1)
Next i
'Forward sweep:
P(k + 1) = -GAMMAK / BETAK
Q(k + 1) = DELTAK / BETAK
For i = k + 1 To r - 1
w = H(i) + H(i + 1)
alpha = H(i + 1) * (2 - G(i) * H(i + 1)) / w
beta = E(i) * H(i) * H(i + 1) - _
G(i) * (H(i) - H(i + 1)) - 2
gamma = H(i) * (2 + G(i) * H(i)) / w
delta = F(i) * H(i) * H(i + 1)
w = alpha * P(i) + beta
P(i + 1) = -gamma / w
Q(i + 1) = (delta - alpha * Q(i)) / w
Next i
'Backward sweep:
U(r) = (DELTAR - ALPHAR * Q(r)) / _
(ALPHAR * P(r) + BETAR)
For i = r To k + 1 Step -1
U(i - 1) = P(i) * U(i) + Q(i)
Next i
End Sub

The foba subroutine parameters have the following sense:

k, r are numbers of the left and right boundary nodes of the grid;
X is an array of grid nodes;
G, E are arrays of values of the coefficients of equation (3.6) at the grid
nodes;

253
 Chapter 3. Finite Difference Method for Solving Differential Equations

F is an array of values of the right-hand side of equation (3.6);

GAMMAK, DELTAK are values of parameters k and k in left boundary
condition (3.11), where k
2;
ALPHAR, DELTAR are values of parameters r
and r
in right boundary
condition (3.12), where r
2;
U is an array intended for the solution values.
Here and below, we consider only the main grid on segment [a, b] with node
coordinates x k , xk 1
, xk 2
, ..., xr 2
, xr 1
, x r (Fig. 3.1).

254
 3.16. Solving the Shockley-Poisson equation on a nonuniform grid

3.16. Solving the Shockley-Poisson

equation on a nonuniform grid

Let us consider the following source data table.

Listing 3.12
maxtime 1
maxiter 1000
phi 1.00E-01
NA u0 x
7.00E+20 0.00E+00 -1.0E-05
7.00E+20 4.89E-02 -9.0E-06
7.00E+20 1.91E-01 -8.0E-06
7.00E+20 4.12E-01 -7.0E-06
7.00E+20 6.91E-01 -6.0E-06
7.00E+20 1.00E+00 -5.0E-06
7.00E+20 1.31E+00 -4.0E-06
7.00E+20 1.59E+00 -3.0E-06
7.00E+20 1.81E+00 -2.0E-06
7.00E+20 1.95E+00 -1.0E-06
7.00E+20 1.98E+00 -6.0E-07
7.00E+20 1.99E+00 -4.0E-07
7.00E+20 2.00E+00 -3.0E-07
7.00E+20 2.00E+00 -2.0E-07
-3.00E+22 2.00E+00 -1.0E-07
-3.00E+22 2.00E+00 0.0E+00
-3.00E+22 2.00E+00 1.0E-07
7.00E+20 2.00E+00 2.0E-07
7.00E+20 2.00E+00 3.0E-07
7.00E+20 1.99E+00 4.0E-07
7.00E+20 1.98E+00 6.0E-07
7.00E+20 1.95E+00 1.0E-06
7.00E+20 1.81E+00 2.0E-06
7.00E+20 1.59E+00 3.0E-06
7.00E+20 1.31E+00 4.0E-06
7.00E+20 1.00E+00 5.0E-06
7.00E+20 6.91E-01 6.0E-06
7.00E+20 4.12E-01 7.0E-06
7.00E+20 1.91E-01 8.0E-06
7.00E+20 4.89E-02 9.0E-06
7.00E+20 0.00E+00 1.0E-05

255
 Chapter 3. Finite Difference Method for Solving Differential Equations

Table Listing 3.12 contains:

maxtime, the limiting execution time in seconds;
maxiter, the limiting number of the quasilinearization method iterations; it
must be less than the maximum value of the Integer data type, that is, 32767
(Appendix 1);
the value of φ in condition (3.67) for finishing the iterative process of the
quasilinearization method;
the values of spatial coordinate x;
the values of dependence N A (x ) ;
the values of dependence u 0 ( x ) , which are calculated according to for-
mula (3.69) by using Excel.
If maxtime is greater than 59 or less than 0, then the execution time is not
limited.
The program below is intended for solving the boundary value problem
for the Shockley-Poisson equation on the nonuniform grid defined by the right
column of table Listing 3.12.
Listing 3.13
Sub main()
Dim NA() As Double
Dim U() As Double
Dim X() As Double
Dim H() As Double
Dim U2() As Double
Dim V() As Double
Dim G() As Double
Dim E() As Double
Dim F() As Double
Dim m As Integer
Dim maxtime As Integer
Dim maxiter As Integer
Dim phi As Double
Dim i As Integer
Dim j As Integer
Dim w1 As Double
Dim w2 As Double
Dim w3 As Double
Dim max As Double
Dim sb As String
Dim se As String

256
 3.16. Solving the Shockley-Poisson equation on a nonuniform grid

Dim sn As String
Dim tm As Date
Const q = 1.6E-19
Const epsilon = 103.545E-12
Const kb = 1.38E-23
Const T = 300
m = Selection.Rows.Count 'quantity of rows
maxtime = Selection.Cells(1, 2)
maxiter = Selection.Cells(2, 2)
phi = Selection.Cells(3, 2)
ReDim NA(5 To m)
ReDim U(5 To m)
ReDim X(5 To m)
ReDim H(6 To m)
ReDim U2(5 To m)
ReDim V(5 To m)
ReDim G(5 To m)
ReDim E(5 To m)
ReDim F(5 To m)
w1 = q / epsilon
w2 = q / (kb * T)
For i = 5 To m
NA(i) = Selection.Cells(i, 1)
U(i) = Selection.Cells(i, 2)
X(i) = Selection.Cells(i, 3)
G(i) = 0
Next i
For i = 6 To m
H(i) = X(i) - X(i - 1)
Next i
If maxtime >= 0 And maxtime < 60 Then
tm = Now + TimeValue("00:00:" & CStr(maxtime))
End If
For j = 1 To maxiter
max = 0
For i = 6 To m - 1
w3 = NA(5) * Exp(-w2 * U(i))
E(i) = -w1 * w2 * w3
U2(i) = 2 * ((U(i + 1) - U(i)) * H(i) + _
(U(i - 1) - U(i)) * H(i + 1)) / _
(H(i) * H(i + 1) * (H(i) + H(i + 1)))
F(i) = w1 * (NA(i) - w3) - U2(i)

257
 Chapter 3. Finite Difference Method for Solving Differential Equations

If Abs(F(i)) > max Then max = Abs(F(i))

Next i
0: Call foba(5, m, X, G, E, F, 0, 0, 0, 0, V)
For i = 5 To m
U(i) = U(i) + V(i)
Next i
If max < phi Then Exit For
If maxtime >= 0 And maxtime < 60 And _
Now > tm Then Exit For
Next j
Selection.Cells(4, 4) = "u"
For i = 5 To m
Selection.Cells(i, 4) = U(i)
Next i
1: sb = Selection.Cells(5, 3).Address
2: se = Selection.Cells(m, 4).Address
3: sn = ActiveSheet.Name
4: Range(sb & ":" & se).Select
5: Selection.NumberFormat = "0.0E+00"
6: Charts.Add
7: ActiveChart.ChartType = xlXYScatterSmoothNoMarkers
8: ActiveChart.SetSourceData Source:= _
Sheets(sn).Range(sb & ":" & se), PlotBy:= _
xlColumns
9: ActiveChart.Location Where:= xlLocationAsObject, _
Name:=sn
10: ActiveChart.Axes(xlValue).MajorGridlines.Select
11: Selection.Delete
12: ActiveChart.Legend.Select
13: Selection.Delete
End Sub

The source data for this program are the values of table Listing 3.12
(Fig. 3.11). Before the program execution, we have to select this Excel table
(range B2:D36, Fig. 3.12).
The execution results are the u solution values, which are located near the
corresponding values of the x coordinate, and the u (x) graph on the Excel work-
sheet (Fig. 3.13).
The u (x) graph is constructed automatically when executing operators 1 —
13. The same operators are present in program Listing 3.10.

258
3.16. Solving the Shockley-Poisson equation on a nonuniform grid

Fig. 3.11. The Excel table with the source data

Fig. 3.12. The worksheet before the program execution

259
 Chapter 3. Finite Difference Method for Solving Differential Equations

Fig. 3.13. The results of the program execution: the horizontal coordinate
of the graph, x, is in meters, the vertical coordinate, u, is in volts

260
 3.17. Use of solution symmetry

3.17. Use of solution symmetry

According to table Listing 3.12, the N A (x ) dependence is symmetric to the

origin of coordinates: N A ( x ) = N A (x ) . Therefore, under symmetric boundary

conditions, in particular u( 10 5 ) = u (10 5 ) = 0, the solution of the Shockley-

Poisson equation is also symmetric: u ( x) = u (x) . We can use the symmetry
for further reducing the execution time.
By setting b 0 , we reduce the length of segment [a, b] to half. The right
boundary condition becomes u (b) 0 . We leave the left boundary and condi-

tion unchanged: a 10 5 , u (a) 0 .

Let us correct the foba subroutine to use it for solving differential equation
(3.6) with left boundary condition (3.11) at given u (b) .
We consider that i = r – 1 in Fig. 3.10. Thus, the right node ( z hr ) coin-
cides with the right boundary of segment [a, b] , that is, with point b.
Let us return to expression (3.73) for the first derivative of second-degree
polynomial (3.70). If z hr , then
dP
( h ) 2 hr .
dz r
Using expressions (3.71) and (3.72) at i = r – 1, we have
dP hr2 u r 2
( hr2 1
2 hr h
1 r
hr2 ) u r 1
( hr2 1
2 hr h ) ur
1 r
( hr ) .
dz hr 1 r
h ( hr 1
hr )
Equating the last expression and u (b) , we obtain equation

hr2 u r 2
( hr2 1
2 hr 1 r
h hr2 ) u r 1
( hr2 1
2 hr 1 r
h ) ur
u (b )
hr h ( hr
1 r 1
hr )
or

261
 Chapter 3. Finite Difference Method for Solving Differential Equations

ur 2
ur 1
ur , (3.79)
where
hr2 ,

hr2 1
2hr h
1 r
hr2 ,

hr2 1
2hr h ,
1 r
(3.80)
hr h ( hr
1 r 1
hr ) u (b) .
According to formula (3.13) for the backward sweep, we have
u r 1 Pr u r Qr ,
ur 2
Pr u
1 r 1
Qr 1
Pr 1
( Pr ur Qr ) Qr 1
.
Substituting these expressions into equation (3.79), we have
[ Pr 1
( Pr ur Qr ) Qr 1
] [ Pr ur Qr ] ur .
The solution of this equation follows:
( Pr 1
Qr Qr 1
) Qr
ur , (3.81)
Pr 1 r
P Pr
where , , and are defined by formulas (3.80).
Let us put the following declaration of the subroutine, which realizes the
decomposition method for differential equation (3.6) with u (b) given, into
Module6 of the BookNM workbook.
Listing 3.14
Sub forbac(ByVal k, ByVal r, ByRef X() As Double, _
ByRef G() As Double, ByRef E() As Double, _
ByRef F() As Double, _
ByVal GAMMAK, ByVal DELTAK, _
ByVal U1B, ByRef U() As Double)
Const BETAK = -2
Dim alpha As Double, beta As Double
Dim gamma As Double, delta As Double
Dim i As Integer, w As Double
Dim H() As Double: ReDim H(k + 1 To r)
Dim P() As Double: ReDim P(k + 1 To r)
Dim Q() As Double: ReDim Q(k + 1 To r)
For i = k + 1 To r

262
 3.17. Use of solution symmetry

H(i) = X(i) - X(i - 1)

Next i
'Forward sweep:
P(k + 1) = -GAMMAK / BETAK
Q(k + 1) = DELTAK / BETAK
For i = k + 1 To r - 1
w = H(i) + H(i + 1)
alpha = H(i + 1) * (2 - G(i) * H(i + 1)) / w
beta = E(i) * H(i) * H(i + 1) - _
G(i) * (H(i) - H(i + 1)) - 2
gamma = H(i) * (2 + G(i) * H(i)) / w
delta = F(i) * H(i) * H(i + 1)
w = alpha * P(i) + beta
P(i + 1) = -gamma / w
Q(i + 1) = (delta - alpha * Q(i)) / w
Next i
'Backward sweep:
alpha = H(r) ^ 2
beta = -H(r - 1) ^ 2 - 2 * H(r - 1) * H(r) - _
H(r) ^ 2
gamma = H(r - 1) ^ 2 + 2 * H(r - 1) * H(r)
delta = H(r - 1) * H(r) * (H(r - 1) + H(r)) * U1B
U(r) = (delta - alpha * (P(r - 1) * Q(r) + _
Q(r - 1)) - beta * Q(r)) / _
(alpha * P(r - 1) * P(r) + beta * P(r) + gamma)
For i = r To k + 1 Step -1
U(i - 1) = P(i) * U(i) + Q(i)
Next i
End Sub

Formula (3.81) is used to start the backward sweep in the forbac subroutine.
The subroutine parameters have the following sense:
k, r are numbers of the left and right boundary nodes of the grid;
X is an array of grid nodes;
G, E are arrays of values of the coefficients of equation (3.6) at the grid
nodes;
F is an array of values of the right-hand side of equation (3.6);
GAMMAK, DELTAK are values of parameters k and k in left boundary
condition (3.11), where k
2;
U1B is a value of u (b) ;

263
 Chapter 3. Finite Difference Method for Solving Differential Equations

U is an array intended for the solution values.

The program, which solves the boundary value problem for the Shockley-
Poisson equation by means of subroutine forbac, differs from Listing 3.13 of
the previous section only in the following operators:

0: Call forbac(5, m, X, G, E, F, 0, 0, 0, V)
5: Selection.NumberFormat = "0.000E+00"

The source data for this program are the values located in table Listing 3.15
(Fig. 3.14). We must select this Excel table (range B2:D21) before the program
execution.
Listing 3.15
maxtime 1
maxiter 1000
phi 1.00E-01
NA u0 x
7.00E+20 0.00E+00 -1.00E-05
7.00E+20 4.89E-02 -9.00E-06
7.00E+20 1.91E-01 -8.00E-06
7.00E+20 4.12E-01 -7.00E-06
7.00E+20 6.91E-01 -6.00E-06
7.00E+20 1.00E+00 -5.00E-06
7.00E+20 1.31E+00 -4.00E-06
7.00E+20 1.59E+00 -3.00E-06
7.00E+20 1.81E+00 -2.00E-06
7.00E+20 1.95E+00 -1.00E-06
7.00E+20 1.98E+00 -6.00E-07
7.00E+20 1.99E+00 -4.00E-07
7.00E+20 2.00E+00 -3.00E-07
7.00E+20 2.00E+00 -2.00E-07
-3.00E+22 2.00E+00 -1.00E-07
-3.00E+22 2.00E+00 0.00E+00

The results of the program execution are the u solution values and the u (x)
graph for negative values of x (Fig. 3.15).
According to the u (x) graph, semiconductor layer -6 μm ≤ x ≤ 6 μm, whose
plane is perpendicular to the x axis, is the potential well for signal electrons: this
layer collects electrons knocked out by photons (light particles) from semicon-
ductor molecules. The light falls on the layer plane.
In Section 4.7, the mathematical modeling of the silicon photosensitive target
will be continued, and we will use the resulting u (x) dependence depicted in
Fig. 3.15.

264
 3.17. Use of solution symmetry

Fig. 3.14. The Excel table with the source data

Fig. 3.15. The results of the program execution: the horizontal coordinate
of the graph, x, is in meters, the vertical coordinate, u, is in volts

265
 Chapter 3. Finite Difference Method for Solving Differential Equations

We advise the reader to develop the forbacs subroutine (similar to

forbac), which realizes the decomposition method for differential equation
(3.6) on segment [a, b] with u (a ) given. The forbacs subroutine must have
the following parameters:
k, r are numbers of the left and right boundary nodes of the grid on seg-
ment [a, b] ;
X is an array of grid nodes;
G, E are arrays of values of the coefficients of equation (3.6) at the grid
nodes;
F is an array of values of the right-hand side of equation (3.6);
U1A is a value of u (a ) ;
ALPHAR, DELTAR are values of parameters r
and r
in right boundary
condition (3.12), where r
2;
U is an array intended for the solution values.
We advise to put the forbacs subroutine declaration into Module6 of the
BookNM workbook below the forbac subroutine declaration developed above.
In addition, we advise the reader to use the forbacs subroutine to create
a picture similar to Fig. 3.15, but for x 0 .

266
 3.18. The cyclic decomposition method

3.18. The cyclic decomposition method

Let us consider linear differential equation (3.6) on the whole x axis,

x , assuming that the coefficients and right-hand side of the equation
are periodic functions with period Π b a :
g ( x Π ) g ( x ) , e( x Π ) e ( x ) , f ( x Π ) f ( x ) .
In this case, the equation solution is also periodic:
u ( x Π ) u ( x) for x .
We supplement the grid on segment a ≤ x ≤ b (Fig. 3.1) by nodes outside this
segment in such manner that xi r k xi Π for i .
Let us replace boundary conditions (3.11) and (3.12) by the following perio-
dicity condition for the solution: u i r k u i , i . As a result, we
have the system of linear algebraic equations
u
k r 1
u
k k
u
k k 1 k
, (3.82)
u
i i 1
u
i i i i
u 1 i
, (3.83)
i = k + 1, k + 2, ..., r – 2,
r
u
1 r 2
u
r 1 r 1 r 1 k
u r 1
. (3.84)
The coefficients and right-hand sides are determined as follows:
the values of i , i , i , i are calculated according to formulas (3.78)
at k + 1 i r – 1;
the values of k
, k
, k
, k
are calculated according to formulas
hk 1
(2 g k hk 1
)
k
,
hr hk 1

k
ek hr hk 1
g k ( hr hk 1
) 2,
hr (2 g k hr )
k
, (3.85)
hr hk 1

k
f k hr hk 1
.

267
 Chapter 3. Finite Difference Method for Solving Differential Equations

The formulated system of r – k equations (3.82) — (3.84) with the r – k un-

knowns ( u k , u k 1 , u k 2 , ..., u r 2 , u r 1 ) is the finite difference scheme for
linear differential equation (3.6) with the solution periodicity condition. We can
write this scheme as the following matrix equation:

k k
0 0 ... 0 0 k

k 1 k 1 k 1
0 ... 0 0 0
0 k 2 k 2 k 2
... 0 0 0
. . . . . . . .
0 0 0 0 ... r 2 r 2 r 2

r 1
0 0 0 ... 0 r 1 r 1

uk k
uk 1 k 1
uk 2 k 2
.
... ...
ur 2 r 2
ur 1 r 1

The matrix of this equation has the so-called cyclic tridiagonal form.
The system of equations (3.82) — (3.84), i.e., the above matrix equation,
is usually solved by the cyclic decomposition method. Let us consider this
algorithm.
Let y (x) , z (x) and u (x ) be grid functions defined on grid x k < xk 1
<
xk 2
< ... < xr 2
< xr 1
< x r (Fig. 3.1), and:

variables y i satisfy the following system of linear algebraic equations

(3.9) with zero boundary conditions:
y
i i 1
y
i i i i
y 1 i
, (3.86)
i = k + 1, k + 2, ..., r – 2, r – 1,
yk yr 0 ; (3.87)
variables zi satisfy the following system of equations (3.9) with zero
right-hand side and unit boundary conditions:
z
i i 1
z
i i
z
i i 1
0, (3.88)

268
 3.18. The cyclic decomposition method

i = k + 1, k + 2, ..., r – 2, r – 1,
zk zr 1 ; (3.89)
variables ui are the following linear combination of y i and zi :
ui yi u k zi , (3.90)
i = k, k + 1, k + 2, ..., r – 2, r – 1, r.
By substituting expression (3.90) into equations (3.83) and (3.84), we can
easily verify that grid function (3.90), u (x) , satisfies these equations at an arbi-
trary value of u k . Let us find the value of u k , at which grid function u (x) satis-
fies equation (3.82). For this purpose, we will consider the sweep formulas for
solution of systems (3.86), (3.87) and (3.88), (3.89).
Let formulas (3.18) and (3.13) of the backward sweep look like
yr 0 , zr 1 , (3.91)
yi 1
Pi yi Qi , zi 1
Pi z i Si , (3.92)
where i = r, r – 1, ..., k + 1. In this case, formulas (3.16), (3.17) and (3.14), (3.15)
of the forward sweep become
Pk 1 0 , Qk 1 0 , S k 1 1 , (3.93)

i i i
Qi S
i i
Pi 1
, Qi 1
, Si 1
, (3.94)
P
i i i i i
P i
P
i i i
where i = k + 1, k + 2, ..., r – 1.
By substituting expression (3.90) at i = r – 1 and i = k + 1,
ur 1 yr 1 uk z r 1 ,
uk 1
yk 1
uk zk 1
,
into equation (3.82), we obtain
k
( yr 1 u k z r 1 ) k k
u k
( yk 1
uk zk 1
) k
or
k k
yr 1 k
yk 1
uk . (3.95)
k
z
k r 1
z
k k 1

At this value of u k , the linear combination of grid functions y (x) and z (x) ,
defined by formula (3.90), satisfies not only equations (3.83) and (3.84), but also
equation (3.82).
According to the cyclic decomposition method, the system of linear algebraic
equations (3.82) — (3.84) is solved as follows:

269
 Chapter 3. Finite Difference Method for Solving Differential Equations

1) the forward sweep is performed according to formulas (3.93) and (3.94),

i = k + 1, k + 2, ..., r – 1;
2) the backward sweep is performed according to formulas (3.91) and (3.92),
i = r, r – 1, ..., k + 2;
3) the value of u k is calculated according to formulas (3.85) and (3.95);
4) the values of ui (k + 1 i r – 1) are calculated according to (3.90).

270
 3.19. Program realization of the cyclic decomposition method

3.19. Program realization of the cyclic

decomposition method

Let us consider a subroutine for solving linear differential equation (3.6)

under the following periodicity condition: u ( x Π ) u ( x) , where Π b a is
the period, x . The coefficients and right-hand side of equation (3.6)
are periodic functions: g ( x Π ) g ( x) , e( x Π ) e( x) , f ( x Π ) f ( x) .
For program realization of the cyclic decomposition method, we put the
following subroutine declaration into Module7 of the BookNM workbook.
Listing 3.16
Sub forwback(ByVal k, ByVal r, ByRef X() As Double, _
ByRef G() As Double, ByRef E() As Double, _
ByRef F() As Double, ByRef U() As Double)
Dim alpha As Double, beta As Double
Dim gamma As Double, delta As Double
Dim i As Integer, w As Double
Dim H() As Double: ReDim H(k + 1 To r)
Dim P() As Double: ReDim P(k + 1 To r)
Dim Q() As Double: ReDim Q(k + 1 To r)
Dim S() As Double: ReDim S(k + 1 To r)
Dim Y() As Double: ReDim Y(k + 1 To r)
Dim Z() As Double: ReDim Z(k + 1 To r)
For i = k + 1 To r
H(i) = X(i) - X(i - 1)
Next i
'Forward sweep:
P(k + 1) = 0
Q(k + 1) = 0
S(k + 1) = 1
For i = k + 1 To r - 1
w = H(i) + H(i + 1)
alpha = H(i + 1) * (2 - G(i) * H(i + 1)) / w
beta = E(i) * H(i) * H(i + 1) - _
G(i) * (H(i) - H(i + 1)) - 2

271
 Chapter 3. Finite Difference Method for Solving Differential Equations

gamma = H(i) * (2 + G(i) * H(i)) / w

delta = F(i) * H(i) * H(i + 1)
w = alpha * P(i) + beta
P(i + 1) = -gamma / w
Q(i + 1) = (delta - alpha * Q(i)) / w
S(i + 1) = -alpha * S(i) / w
Next i
'Backward sweep:
Y(r) = 0
Z(r) = 1
For i = r To k + 2 Step -1
Y(i - 1) = P(i) * Y(i) + Q(i)
Z(i - 1) = P(i) * Z(i) + S(i)
Next i
'Calculation of solution:
w = H(r) + H(k + 1)
alpha = H(k + 1) * (2 - G(k) * H(k + 1)) / w
beta = E(k) * H(r) * H(k + 1) - _
G(k) * (H(r) - H(k + 1)) - 2
gamma = H(r) * (2 + G(k) * H(r)) / w
delta = F(k) * H(r) * H(k + 1)
U(k) = (delta - alpha * Y(r - 1) - _
gamma * Y(k + 1)) / (beta + alpha * Z(r - 1) + _
gamma * Z(k + 1)) 'calculation of U(k)
For i = k + 1 To r - 1
U(i) = Y(i) + U(k) * Z(i) 'calculation of U(i)
Next i
U(r) = U(k)
End Sub

The forwback subroutine parameters have the following sense:

k, r are numbers of the left and right boundary nodes of the grid on
segment [a, b] ;
X is an array of grid nodes;
G, E are arrays of values of the coefficients of equation (3.6) at the grid
nodes;
F is an array of values of the right-hand side of equation (3.6);
U is an array intended for the solution values.
Elements U(k) and U(r) are equal in the resulting U array.

272
 3.20. Solving the oscillation equation

3.20. Solving the oscillation equation

We will use the cyclic decomposition method for mathematical modeling

of the oscillating motion of a bar. For this, we will be guided by the theory of the
fifth chapter in book [8].
Let a bar with mass M be attached to the free end of a spring (Fig. 3.16). The
bar’s position is defined by horizontal coordinate u; the origin of coordinates
(u = 0) corresponds to the equilibrium state, in which the spring is not strained.

Fig. 3.16. A bar sliding along a horizontal surface

If the bar is deviated to the left or right from the equilibrium state, the spring
resilience force (directed opposite to the displacement) acts on the bar. The force
value is defined by formula
Fr Ku .
Positive coefficient K is called the elastic constant of the spring. As is
customary in physics, we consider that the direction of the force vector coincides
with the u axis for positive values of Fr (that is, the force is directed from left
to right) and the force vector is directed opposite to the u axis for negative values
of Fr (that is, the force is directed from right to left).
Let us deviate the bar from the equilibrium state and then release it. The bar
starts to make an oscillatory motion, which is described by the following equa-
tion of Newton’s second law:
d 2u
M Ku ,
dt 2

273
 Chapter 3. Finite Difference Method for Solving Differential Equations

where d 2u / dt 2 is the bar acceleration.

If the initial deviation of the bar (from the equilibrium state) is positive and
equal to A , the solution of the last equation looks like
u (t ) A cos( 0
t) ,
where A is the oscillation amplitude, 0
K / M is the cyclic frequency of
the oscillatory motion. The cosine argument, 0
t , is in radians.
Cyclic frequency 0
(in radians per second) is related with oscillation fre-
quency f 0 (in hertzs) and period T0 (in seconds) as follows:

0
2 f0 2 / T0 .
In the presence of friction between the bar and horizontal surface, we observe
the oscillation damping. If the bar velocity is small, the friction force is propor-
tional to the bar mass, M, and to the first power of the bar velocity, du / dt . The
friction force is directed opposite to the velocity vector.
Taking into account the friction, we write the equation of Newton’s second
law as follows:
d 2u du
M Ku LM ,
dt 2 dt

where L ≥ 0 is the so-called coefficient of the oscillation damping.

In order to exclude the oscillation damping, we introduce force F (t ) with
period T . The equation of Newton’s second law becomes
d 2u du
M Ku LM F (t ) . (3.96)
dt 2 dt

It is obvious that after a while the oscillation of the bar becomes periodic
with period T .
Let us consider that time t changes from a negative value, t 0 , to infinity.
Let t1 be a negative value exceeding t 0 , such that:
t 0 ≤ t < t1 is an area of establishing the oscillation, in which the u (t )
solution is an aperiodic function;
t ≥ t1 is an area of the established oscillation, in which the u (t ) solution
is a periodic function with period T .

274
 3.20. Solving the oscillation equation

Because the periodic oscillation of the bar is interesting for us, we consider
equation (3.96) on segment [0, T ] with the solution periodicity condition.
Equation (3.96) can be written in form (3.6),
d 2u du
g eu f (t ) , (3.97)
dt 2 dt
2
where g L , e 0
K / M , f (t ) F (t ) / M .
Let us consider the following source data table.
Listing 3.17
M 0.001
K 800
L 100
F t
0.00E+00 0.00E+00
0.00E+00 2.00E-03
0.00E+00 4.00E-03
0.00E+00 6.00E-03
0.00E+00 8.00E-03
0.00E+00 1.00E-02
0.00E+00 1.20E-02
0.00E+00 1.40E-02
0.00E+00 1.60E-02
0.00E+00 1.80E-02
0.00E+00 2.00E-02
0.00E+00 2.20E-02
0.00E+00 2.40E-02
0.00E+00 2.60E-02
5.00E+00 2.80E-02
1.00E+01 3.00E-02
5.00E+00 3.20E-02
0.00E+00 3.40E-02
0.00E+00 3.60E-02
0.00E+00 3.80E-02
0.00E+00 4.00E-02
0.00E+00 4.20E-02
0.00E+00 4.40E-02
0.00E+00 4.60E-02
0.00E+00 4.80E-02
0.00E+00 5.00E-02

275
 Chapter 3. Finite Difference Method for Solving Differential Equations

In this table:
M is the bar mass in kilograms;
K is the elastic constant of the spring, in N / m;
L is the coefficient of the oscillation damping, in 1 / s;
t are the values of time for one period, in seconds;
F are the F (t ) function values in newtons.
According to table Listing 3.17:
1) every 0.05 seconds, a positive force acts on the bar, for example, it gets
a kick from left to right;
2) the maximum value of the force equals 10 N;
3) the duration of the force action equals 0.008 s.
The program for solving equation (3.97) under the periodicity condition is
given below.
Listing 3.18
Sub main()
Dim T() As Double
Dim G() As Double
Dim E() As Double
Dim F() As Double
Dim U() As Double
Dim m As Integer
Dim MM As Double
Dim KK As Double
Dim LL As Double
Dim i As Integer
Dim sb As String, se As String
Dim sn As String
m = Selection.Rows.Count 'quantity of rows
MM = Selection.Cells(1, 2)
KK = Selection.Cells(2, 2)
LL = Selection.Cells(3, 2)
ReDim T(5 To m)
ReDim G(5 To m)
ReDim E(5 To m)
ReDim F(5 To m)
ReDim U(5 To m)
For i = 5 To m
T(i) = Selection.Cells(i, 2)
G(i) = LL
E(i) = KK / MM

276
 3.20. Solving the oscillation equation

F(i) = Selection.Cells(i, 1) / MM
Next i
0: Call forwback(5, m, T, G, E, F, U)
Selection.Cells(4, 3) = "u"
For i = 5 To m
Selection.Cells(i, 3) = U(i)
Next i
1: sb = Selection.Cells(5, 2).Address
2: se = Selection.Cells(m, 3).Address
3: Range(sb & ":" & se).Select
4: sn = ActiveSheet.Name
5: Selection.NumberFormat = "0.00E+00"
6: Charts.Add
7: ActiveChart.ChartType = xlXYScatterSmoothNoMarkers
8: ActiveChart.SetSourceData Source:= _
Sheets(sn).Range(sb & ":" & se), PlotBy:= _
xlColumns
9: ActiveChart.Location Where:= xlLocationAsObject, _
Name:=sn
10: ActiveChart.Axes(xlValue).MajorGridlines.Select
11: Selection.Delete
12: ActiveChart.Legend.Select
13: Selection.Delete
14: With ActiveChart
15: .Axes(xlCategory, xlPrimary).HasTitle = True
16: .Axes(xlCategory, _
xlPrimary).AxisTitle.Characters.Text = "t, s"
17: .Axes(xlValue, xlPrimary).HasTitle = True
18: .Axes(xlValue, _
xlPrimary).AxisTitle.Characters.Text = "u, m"
19: End With
20: ActiveChart.Axes(xlCategory).AxisTitle.Select
21: Selection.AutoScaleFont = True
22: With Selection.Font
23: .FontStyle = "regular"
24: .Size = 12
25: End With
26: ActiveChart.Axes(xlValue).AxisTitle.Select
27: Selection.AutoScaleFont = True
28: With Selection.Font
29: .FontStyle = "regular"
30: .Size = 12

277
 Chapter 3. Finite Difference Method for Solving Differential Equations

31: End With

32: ActiveChart.ChartArea.Select
End Sub

In this program, operator 0 is the call of the forwback subroutine, realizing

the cyclic decomposition method.
The source data for the program are the values located in table Listing 3.17
(Fig. 3.17). We must select this Excel table (range B2:C31) before the program
execution. The execution results are the solution values, located in the u column
(near the t column, Fig. 3.18), and the u (t ) graph on the Excel worksheet.
The u (t ) graph is created automatically:
operators 1 — 13 of program Listing 3.18 construct the graph and delete
the gridlines and legend;
operators 14 — 32 superscribe the axes.

Fig. 3.17. The Excel table with the source data

278
 3.20. Solving the oscillation equation

Fig. 3.18. The program execution results,

including the graph of periodic dependence
u (t ) with frequency 1 / T = 20 Hz

The reviewed operators (from 1 to 32) were programmed by means of Excel

Macro Recorder. In Section 4.8, we will consider the graph creation subroutine
based on these operators.
In Section 5.12, we will use periodic dependence u (t ) depicted above in
Fig. 3.18.
We advise the reader to develop a program, similar to Listing 3.18, for calcu-
lating a periodic time dependence of the current, i (t ) , in the electrical circuit
shown in Fig. 3.19. The electromotive force (in volts) of the generator is the fol-
lowing periodic function of time: v(t κ Π ) f (t ) , where t is time in seconds,
a t b , Π b a is the period, is an integer. Function f (t ) and the
values of a and b are given in Appendix 4.
The process of solving the suggested task must include the following two
stages:

279
 Chapter 3. Finite Difference Method for Solving Differential Equations

1) on segment [a, b] , periodic dependence of the capacitor charge, q (t ) , is

calculated by solving equation
d 2q dq 1
L R q v (t ) ;
dt 2 dt C
2) the current is calculated by differentiation of the capacitor charge:
dq
i (t ) .
dt
Dependences v(t ) and i (t ) will be figured in the task on p. 414.

Fig. 3.19. The electrical circuit with the following parameters: electric
resistance R = 100 Ω, inductance L = 0.2 H, capacitance C = 3·10-4 F

280
 Chapter 4.
Cubic Spline

Let us begin with the origin of term “spline”.

Long ago, engineers needed to draw smooth curves through given points. To
do so, they used long elastic wooden strips. Such a strip, called a spline, was
fixed (nailed up to a drawing board) at the given points. As a result, the strip was
bent to provide the smooth curve.
In mathematics, the function describing the bending of an elastic strip is
called a third-degree (cubic) spline [9]. In this chapter, this mathematical
construction is used for interpolation, differentiation and integration of the grid
(tabular) function and also for solving the nonlinear algebraic and linear differen-
tial equations. Besides, we consider two classical methods for solving the non-
linear algebraic equation, namely, the bisection and secant methods.
For demonstration of the spline possibilities, we solve applied problems
concerning the field-effect transistor, silicon photosensitive target and geophysi-
cal cable. The locally one-dimensional scheme [4] is considered for solving the
heat equation of the last applied problem with two spatial coordinates.
In addition to user-defined procedures, realizing the numerical methods,
a subroutine for automatic creation of graphs is developed.

281
 Chapter 4. Cubic Spline

4.1. Definition of cubic spline. Spline moments

Let an increasing sequence of points on segment [a, b] be given as follows:

a xk < x k 1
< xk 2
< ... < xr 2
< xr 1
< xr b . In other words, as in
the previous chapter of the book, segment [a, b] is covered with a grid whose
nodes have coordinates x k , xk 1
, xk 2
, ..., xr 2
, xr 1
, x r . Segments
[ xi 1
, x i ] are called elementary segments, k + 1 ≤ i ≤ r.
Let f (x) be a grid function, and f k , f k 1
, fk 2
, ..., f r 2
, fr 1
, fr
are given values of f (x) at points x k , xk 1 , xk 2 , ..., xr 2 , xr 1 , x r ,
respectively.
A cubic spline (or third-degree spline, Fig. 4.1) is function S (x) , which
satisfies the following conditions:
1) on each elementary segment xi 1 ≤ x ≤ xi (k + 1 ≤ i ≤ r), the spline co-
incides with a third-degree polynomial (generally, the polynomials are different
on different elementary segments);
2) at the grid nodes, the spline has the corresponding grid function values:
S ( xi ) f i ;
3) the spline has a continuous first derivative, i.e., the spline is smooth;
4) the spline has a continuous second derivative;
5) on the boundaries of segment [a, b] , the spline satisfies additional condi-
tions (we will consider these boundary conditions below, closer to the end of the
section).
According to above item (2), the S (x) graph passes through points ( x k , f k ),
( xk , f k 1 ), ( xk 2 , f k 2 ), . . . , ( xr 2 , f r 2 ), ( xr 1 , f r 1 ), ( x r , f r ).
1
According to items (3) and (4), the jumps of the first and second derivatives of
S (x) are absent at the interior grid nodes, i.e., at xk 1 , xk 2 , ..., xr 2 , xr 1 .
The values of the second derivative, S (x) , at the grid nodes are called
moments of the cubic spline:

282
 4.1. Definition of cubic spline. Spline moments

d 2S
( xi ) Mi ,
dx 2
where M i is the spline moment, k ≤ i ≤ r.

Fig. 4.1. The cubic spline graph

Let the spline moments, M i (k ≤ i ≤ r), be given (later we will know how to
calculate them). In this case, items (1) and (4) of the spline definition give
the following expression for the second derivative on elementary segment
[ xi 1 , xi ] :
xi x x xi 1
S ( x) Mi 1
Mi , (4.1)
hi hi
where hi xi xi 1
is the elementary segment’s length or the grid step, k + 1 ≤
i ≤ r.
Below, we will obtain expressions for the spline and its first derivative.
Let us integrate expression (4.1):
( xi x) 2 (x xi 1
)2
S ( x) Mi 1
Mi C1 , (4.2)
2hi 2hi
where C 1 is the integration constant.
Integrating expression (4.2), we have

283
 Chapter 4. Cubic Spline

( xi x)3 (x xi 1
)3
S ( x) Mi 1
Mi C1 x C2 , (4.3)
6hi 6hi
where C 2 is the integration constant.
Integration constants C 1 and C 2 will be determined by means of item (2)
of the spline definition, according to which S ( xi ) f i and S ( xi 1
) fi 1
or

( xi xi 1
)3
xi C1 C2 fi Mi ,
6hi

( xi xi 1
)3
xi C
1 1
C2 fi 1
Mi 1
.
6hi
The solution of this system of two linear algebraic equations (with unknown
C 1 and C 2 ) has the following form:
fi fi 1
Mi Mi 1
C1 hi ,
hi 6
xi 1
xi hi hi
C2 fi fi 1
Mi xi 1
Mi 1
xi .
hi hi 6 6
By substituting the last two expressions for C 1 and C 2 into (4.3) and (4.2),
we obtain the following expressions for the spline and its derivative on segment
[ xi 1 , xi ] :

( xi x)3 (x xi 1
)3 Mi h2
1 i
xi x
S ( x) Mi 1
Mi fi 1
6hi 6hi 6 hi

M i hi2 x xi 1
fi , (4.4)
6 hi

( xi x) 2 (x xi 1
)2 fi fi 1
S ( x) Mi 1
Mi
2 hi 2 hi hi
Mi Mi 1
hi . (4.5)
6

284
 4.1. Definition of cubic spline. Spline moments

Expressions (4.1), (4.4) and (4.5) include moments M i 1

and M i .
Below, we will obtain the system of linear algebraic equations with the
tridiagonal coefficient matrix, which allows us to calculate moments M k ,
Mk 1
, Mk 2
, ..., M r 2
, Mr 1
, Mr .
The expression for the spline derivative at point xi on the left,

( xi xi 1
)2 fi fi 1
Mi Mi 1
S ( xi 0) Mi hi , (4.6)
2hi hi 6
follows from (4.5).
Expression (4.5) for elementary segment xi ≤ x ≤ xi 1
looks like

( xi 1
x) 2 (x xi ) 2 fi 1
fi
S ( x) Mi Mi 1
2hi 1
2hi 1
hi 1
Mi 1
Mi
hi 1
. (4.7)
6
The expression for the spline derivative at point xi on the right,

( xi 1
xi ) 2 fi 1
fi Mi 1
Mi
S ( xi 0) Mi hi 1
, (4.8)
2hi 1
hi 1
6
follows from (4.7).
According to item (3) of the spline definition, the left and right derivatives
are equal:
S ( xi 0) S ( xi 0) .

By means of expressions (4.6) and (4.8), the last equality can be written as
follows:
hi h i hi 1 hi 1 fi 1 fi fi fi 1
Mi 1 Mi Mi 1 ,
6 3 6 hi 1 hi
where i = k + 1, k + 2, …, r – 2, r – 1 are the numbers of the interior grid nodes, or

i
Mi 1
2M i i
Mi 1 i
, (4.9)
where
hi
i
,
hi hi 1

285
 Chapter 4. Cubic Spline

hi 1
i
1 i
, (4.10)
hi hi 1

fi 1
fi fi fi 1
hi 1
hi
i
6 .
hi hi 1
As the boundary conditions in item (5) of the spline definition, we will use
the following linear equations connecting the moments at the ends of segment
[a, b] :
2M k k
Mk 1 k
, (4.11)

r
Mr 1
2M r r
, (4.12)
where k
, k
, r
, r
are given parameters, k and r are numbers of the left and
right grid nodes; xk a and xr b.
If a given value of the function derivative on the left boundary, f k f (a ) , is
the condition in item (5), then (4.8) at i = k leads to the following expressions for
the parameters of equation (4.11):
6 fk 1
fk
k
1, k
fk . (4.13)
hk 1
hk 1

If a given value of the function derivative on the right boundary, f r f (b ) ,

is the condition in item (5), then (4.6) at i = r leads to the following expressions
for the parameters of equation (4.12):
6 fr fr 1
r
1, r
fr . (4.14)
hr hr
A given value of the second derivative on the left boundary ( M k f k ) leads
to the following expressions for the parameters of equation (4.11):
k
0, k
2 fk . (4.15)
A given value of the second derivative on the right boundary ( M r fr )
leads to the following expressions for the parameters of equation (4.12):
r
0, r
2 fr . (4.16)
The constancy of the second derivative at the left end of segment [a, b]
( Mk Mk 1
) leads to

286
 4.1. Definition of cubic spline. Spline moments

k
2, k
0 (4.17)
in equation (4.11).
The constancy of the second derivative at the right end of segment [a, b]
( Mr Mr 1
) leads to

r
2, r
0 (4.18)
in equation (4.12).
Expressions (4.13) — (4.18), as well as equations (4.11) and (4.12), may be
called spline boundary conditions.
Moments M k , M k 1 , M k 2 , ..., M r 2 , M r 1 , M r are determined by
solving the system of linear algebraic equations (4.9), (4.11) and (4.12). In this
case, the decomposition method (Section 3.2) can be used because forms (3.9),
(3.11) and (3.12) are available for equations (4.9), (4.11) and (4.12).
After calculating the moments, the values of the cubic spline and its first and
second derivatives at any point x of segment [a, b] can be calculated according
to formulas (4.4), (4.5) and (4.1), respectively.
The error of interpolating the f (x) function (and its derivatives) by the
S (x) spline (and by its derivatives) is determined by the following expression:

f ( n ) ( x) S ( n ) ( x) 4 n
O(hmax ), (4.19)
where hmax max {hi } is the maximum grid step ( hmax 0 ), n = 0, 1, 2,
k 1 i r

3 is the derivative order, f (0) ( x) f ( x) , S ( 0) ( x) S ( x) . We considered the

sense of the O notation used here in Section 3.1.

287
 Chapter 4. Cubic Spline

4.2. Spline interpolation

Into Module8 of the BookNM workbook, we enter the following declaration

of the subroutine, which realizes the decomposition method for solving the sys-
tem of linear algebraic equations (4.9), (4.11) and (4.12), i.e., for calculating the
spline moments.
Listing 4.1
Sub mos(ByVal k, ByVal r, ByRef X() As Double, _
ByRef F() As Double, _
ByVal GAMMAK, ByVal DELTAK, _
ByVal ALPHAR, ByVal DELTAR, _
ByRef M() As Double)
Dim alpha As Double
Dim gamma As Double, delta As Double
Dim i As Integer, w As Double
Dim H() As Double: ReDim H(k + 1 To r)
Dim P() As Double: ReDim P(k + 1 To r)
Dim Q() As Double: ReDim Q(k + 1 To r)
For i = k + 1 To r
H(i) = X(i) - X(i - 1)
Next i
'Forward sweep:
P(k + 1) = -GAMMAK / 2
Q(k + 1) = DELTAK / 2
For i = k + 1 To r - 1
w = H(i) + H(i + 1)
alpha = H(i) / w
gamma = 1 - alpha
delta = 6 * ((F(i + 1) - F(i)) / H(i + 1) - _
(F(i) - F(i - 1)) / H(i)) / w
w = alpha * P(i) + 2
P(i + 1) = -gamma / w
Q(i + 1) = (delta - alpha * Q(i)) / w
Next i

288
 4.2. Spline interpolation

'Backward sweep:
M(r) = (DELTAR - ALPHAR * Q(r)) / _
(ALPHAR * P(r) + 2)
For i = r To k + 1 Step -1
M(i - 1) = P(i) * M(i) + Q(i)
Next i
End Sub

The subroutine name (mos) occurs from “moments of spline”. The parame-
ters have the following sense:
k, r are numbers of the left and right boundary nodes of the grid on seg-
ment [a, b] ;
X is an array of grid nodes;
F is an array of the f (x) function values at the grid nodes;
GAMMAK, DELTAK correspond to k
and k
in left boundary condition
(4.11);
ALPHAR, DELTAR correspond to r
and r
in right boundary condition
(4.12);
M is an array intended for the spline moments.
The mos subroutine is based on the foba subroutine (Section 3.15).
In practice, not only the spline with boundary conditions (4.11) and (4.12) is
used, but also the periodic spline defined as follows.
Let f (x) be a periodic grid function with period Π b a :
f ( xi r k
) f ( xi ) ,
where xi r k
xi b a, i .
The periodic third-degree spline is function S (x) defined on the whole axis,
x , for which:
1) the first four conditions of the cubic spline definition (p. 282) are satisfied
on segment [a, b] ;
2) S ( xi r k
) S ( xi ) and S ( xi r k
) S ( xi ) for i .
We will use the periodic third-degree spline in Section 5.11, at that, we will
not calculate the moments of this spline. If the reader needs a subroutine for cal-
culating the moments of the periodic spline, its development is not a difficult
task: the forwback subroutine (Section 3.19), realizing the cyclic decomposi-
tion method, should be the basis for the new subroutine.

289
 Chapter 4. Cubic Spline

Let us consider a subroutine of spline interpolation intended for calculating

values of the cubic spline (periodic or with the boundary conditions) and its first
and second derivatives at given point χ of segment [a, b] . This subroutine is
named si from “spline interpolation”.
Into Module9 of the BookNM workbook, we enter the following declaration
of the si subroutine:
Listing 4.2
Sub si(ByVal k, ByVal r, ByRef X() As Double, _
ByRef F() As Double, ByRef M() As Double, _
ByVal chi, ByRef s, _
Optional s1 As Variant, Optional s2 As Variant)
Dim i As Integer
Dim h As Double, hh As Double
Dim h1 As Double, h1h1 As Double
Dim h2 As Double, h2h2 As Double
'Searching elementary segment containing chi:
For i = k + 1 To r
If X(i) > chi Then Exit For
Next i
If i > r Then i = r
'Calculating value of cubic spline at point chi:
h = X(i) - X(i - 1): hh = h * h
h1 = chi - X(i - 1): h1h1 = h1 * h1
h2 = X(i) - chi: h2h2 = h2 * h2
s = (M(i - 1) * h2h2 * h2 + M(i) * h1h1 * h1) / _
(6 * h) + _
((F(i - 1) - M(i - 1) * hh / 6) * h2 + _
(F(i) - M(i) * hh / 6) * h1) / h
'Calculating spline's first derivative at point chi:
If Not IsMissing(s1) Then
s1 = (-M(i - 1) * h2h2 + M(i) * h1h1) / _
(2 * h) + _
(F(i) - F(i - 1)) / h - _
(M(i) - M(i - 1)) / 6 * h
End If
'Calculating spline's second derivative at point chi:
If Not IsMissing(s2) Then
s2 = (M(i - 1) * h2 + M(i) * h1) / h
End If
End Sub

290
 4.2. Spline interpolation

This subroutine has 9 parameters, and the last two parameters, s1 and s2,
are optional. The parameters have the following sense:
k, r are numbers of the left and right boundary nodes of the grid on
segment [a, b] ;
X is an array of grid nodes;
F is an array of the f (x) function values at the grid nodes;
M is an array of the spline moments, for example, determined by the mos
subroutine execution;
chi is given point χ on [a, b] ;
s is a variable (memory cell) intended for the spline’s value at the χ point;
s1, s2 are variables respectively intended for the spline’s first and
second derivatives at the χ point.
In the si subroutine, cycle For…Next is used to find elementary segment
[ xi 1 , x i ] containing the χ point. After finding this segment, the spline’s value
is calculated according to formula (4.4), and, if needed, the spline’s first and
second derivatives are calculated according to (4.5) and (4.1), respectively.

291
 Chapter 4. Cubic Spline

4.3. Use of cubic spline for processing

transistor electrical characteristics

Transistors are the base elements of modern radio electronics. Two types of
transistors exist — bipolar and field-effect transistors. Both those and other such
elements are three-electrode devices based on semiconductors.
Let us consider a concrete example of using the cubic spline construction for
processing electrical characteristics of a field-effect transistor, and such, in which
the main carriers of electric current are electrons rather than holes.
The field-effect transistor electrodes are called source, drain and gate. With-
out delving into the device physics, we note the following regarding the elec-
trodes:
the source injects electrons into the semiconductor, the drain collects
these electrons, the gate regulates the electron flow;
by varying the electric potential difference between the gate and the
source, U gs , we change the drain current, I d .
The major electrical characteristics of the field-effect transistor are the output
current-voltage characteristics (OCVC) representing dependences of the drain
current, I d , on the potential difference between the drain and the source, U ds ,
for various values of U gs . Fig. 4.2 shows the OCVC calculated by means of
mathematical model [10] for electron-hole plasma in the transistor. The simula-
tion based on this model is quite time-consuming: the calculation of the I d value
(for given U gs and U ds ) can take hours on a personal computer.
In the bottom right corner of this book’s cover, we see the two-dimensional
distribution of electron concentration in the transistor for U gs = -3 V and U ds =
14 V. This picture, as well as the OCVC, is from article [10].
Listing 4.3 with tabular representation of the OCVC is given below. In this
table, as well as in Fig. 4.2, the values of potential differences are in volts, the
current is in milliamperes. We see that 18 cells of the table are empty. It is
because the I d value was not calculated for some values of U gs and U ds due
to economic reasons.

292
 4.3. Use of cubic spline for processing transistor electrical characteristics

Fig. 4.2. The output current-voltage characteristics

of the modern transistor with 1 mm gate width
Listing 4.3
The original table of the transistor characteristics

Uds Ugs 0 -1 -2 -3

0 0 0 0 0
0.2 94.5065 39.3361 0 0
0.4 179.9835 0 0
0.5 206.7334 45.3559 0 0
1 207.3824 49.0988 0 0
2 211.1779 55.4772 0 0
4 214.8481 66.276 0 0
6 217.707 76.2717 0 0
7.65 224.2229 0 0
8 0 0
8.35 89.793 0 0
9 0.809 0
10 242.0198 102.0164 5.1056 0
10.8 10.369 0
11 0
11.65 1.0724
12 266.5736 123.9969 22.674
13 37.3677 8.6525
14 298.5628 157.1723 58.0258 42.0182

293
 Chapter 4. Cubic Spline

To build electrical circuits based on the above transistor, we must know the
parameters of its equivalent circuit. To determine these parameters by method
[11], the empty cells of the OCVC table should be filled beforehand. The pro-
gram given below allows doing it by means of the spline interpolation.
Listing 4.4
Sub main()
Dim X() As Double
Dim F() As Double
Dim MOM() As Double
Dim m As Integer, n As Integer
Dim g As Integer, d As Integer
Dim k As Integer, r As Integer
Dim s As Double
m = Selection.Rows.Count 'quantity of rows
n = Selection.Columns.Count 'quantity of columns
ReDim X(2 To n)
ReDim F(2 To n)
ReDim MOM(2 To n)
For g = 2 To m 'setting row number
'Formation of arrays X and F for row No. g:
For r = 2 To n
X(r) = Selection.Cells(1, r)
F(r) = Selection.Cells(g, r)
If F(r) <> 0 Then Exit For
Next r
k = r
For d = k + 1 To n
If Selection.Cells(g, d) <> 0 Then
r = r + 1
X(r) = Selection.Cells(1, d)
F(r) = Selection.Cells(g, d)
End If
Next d
'Calculating array MOM of moments for row No. g:
Call mos(2, r, X, F, 0, 0, 0, 0, MOM)
'Filling all cells of row No. g:
For d = 2 To n
Call si(2, r, X, F, MOM, _
Selection.Cells(1, d), s)
Selection.Cells(g, d) = s
Next d

294
 4.3. Use of cubic spline for processing transistor electrical characteristics

Next g
End Sub

We enter this program into Module1 of the BookNM workbook. The source
data are the values given in table Listing 4.3. The program uses this table in the
transposed form (Fig. 4.3).

Fig. 4.3. The transposed table of the transistor characteristics:

the number format without decimals is set for range C3:U6

For transposing a matrix, Excel includes the TRANSPOSE function (p. 141),
which interprets the contents of empty cells as zero.
After using the TRANSPOSE function, we fulfill the following operations:
1) select the table shown in Fig. 4.3, which is the transposition result;
2) copy this table into Windows Clipboard, for example, by clicking on the
Copy button in the Clipboard area of the Home tab;
3) click on the Paste arrow;
4) in the Paste Values area of the open window, click on the left icon.
The table, depicted in Fig. 4.3, visually does not change. We must select this
table of the transistor characteristics (range B2:U6) before running program
Listing 4.4.
When executing the program, the g cycle parameter accepts the values of
2, 3, 4 and 5. Respectively, the cells of the 2nd, 3rd, 4th and 5th rows of the
selected table are being filled.
Let us consider arrays X and F at a fixed value of g. Before calling the mos
subroutine, arrays X and F contain the following values.
The F array begins with zero, i.e., F(2) = 0. The next elements of the F
array are the values of the g-th row (of the selected table), but not all: the zeros,
which are between the nonzero values, are eliminated. It is these zeros that filled
the empty cells of table Listing 4.3 upon its transposition by the TRANSPOSE
function.

295
 Chapter 4. Cubic Spline

The X array contains the values of the 1st row (of the selected table), which
are located above the g-th row’s values included in the F array.
For example:
for fixed g = 2, array F contains the values of F(2) = 0, F(3) = 95,
F(4) = 180, F(5) = 207, F(6) = 207, F(7) = 211, F(8) = 215,
F(9) = 218, F(10) = 224, F(11) = 242, F(12) = 267, F(13) = 299,
and array X contains the values of X(2) = 0, X(3) = 0.2, X(4) = 0.4,
X(5) = 0.5, X(6) = 1, X(7) = 2, X(8) = 4, X(9) = 6, X(10) = 7.65,
X(11) = 10, X(12) = 12, X(13) = 14 (k = 2, r = 13);
for fixed g = 4, array F contains the values of F(2) = 0, F(3) = 0,
F(4) = 0, F(5) = 0, F(6) = 0, F(7) = 0, F(8) = 0, F(9) = 0, F(10) = 0,
F(11) = 0, F(12) = 0, F(13) = 1, F(14) = 5, F(15) = 10, F(16) = 23,
F(17) = 37, F(18) = 58, and array X contains the values of X(2) = 0,
X(3) = 0.2, X(4) = 0.4, X(5) = 0.5, X(6) = 1, X(7) = 2, X(8) = 4,
X(9) = 6, X(10) = 7.65, X(11) = 8, X(12) = 8.35, X(13) = 9,
X(14) = 10, X(15) = 10.8, X(16) = 12, X(17) = 13, X(18) = 14
(k = 2, r = 18).
For every value of g (2, 3, 4 and 5), the MOM array of the spline moments,
corresponding to arrays X and F, is calculated by means of the mos subroutine.
In the call of this subroutine, boundary conditions (4.15) and (4.16) are used,
where f f 0 . The values, being calculated when executing the si sub-
2 r( g )
routine, fill the cells of the g-th row of the table. Thus, zero or nonzero values
fill the cells with zeros of Fig. 4.3. Fig. 4.4 shows the program execution result.

Fig. 4.4. The transposed table of the transistor characteristics after the execution

After transposing the last table (Fig. 4.4) by means of the TRANSPOSE
function, we have table Listing 4.5, which is similar to Listing 4.3, but without
empty cells. Table Listing 4.5 can be used for calculating parameters of the tran-
sistor equivalent circuit by method [11].

296
 4.3. Use of cubic spline for processing transistor electrical characteristics

Listing 4.5
The completed table of the transistor characteristics

Uds Ugs 0 -1 -2 -3

0 0 0 0 0
0.2 94.5065 39.3361 0 0
0.4 179.9835 47.336 0 0
0.5 206.7334 45.3559 0 0
1 207.3824 49.0988 0 0
2 211.1779 55.4772 0 0
4 214.8481 66.276 0 0
6 217.707 76.2717 0 0
7.65 224.2229 85.364 0 0
8 226.068 87.537 0 0
8.35 228.206 89.793 0 0
9 232.931 94.227 0.809 0
10 242.0198 102.0164 5.1056 0
10.8 250.805 109.515 10.369 0
11 253.214 111.626 12.04 0
11.65 261.644 119.291 18.554 1.0724
12 266.5736 123.9969 22.674 1.102
13 282 139.645 37.3677 8.6525
14 298.5628 157.1723 58.0258 42.0182

297
 Chapter 4. Cubic Spline

4.4. Spline integration

According to the rules of integration and basic integrals [3], the integral of
polynomial (4.4) over segment [ xi 1 , xi ] equals
x
i fi 1
fi Mi 1
Mi
S ( x) dx hi hi3 .
x
2 24
i 1

Therefore, the integral of the spline over segment [a, b] equals

b r fi fi r Mi Mi
S ( x ) dx 1
hi 1
hi3 . (4.20)
i k 1
2 i k 1
24
a

Formulas (4.19) and (4.20) give the following estimation of the spline inte-
gration error:
b b
4
f ( x ) dx S ( x ) dx O( hmax ),
a a

where hmax 0 is the maximum grid step.

Let us enter the following declaration of function ios (from “integral of
spline”) into Module10 of the BookNM workbook.
Listing 4.6
Function ios(ByVal k, ByVal r, ByRef X() As Double, _
ByRef F() As Double, ByRef M() As Double)
Dim i As Integer
Dim h As Double
ios = 0
For i = k + 1 To r
h = X(i) - X(i - 1)
ios = ios + (F(i - 1) + F(i)) / 2 * h - _
(M(i - 1) + M(i)) / 24 * h ^ 3
Next i
End Function

298
 4.4. Spline integration

The ios function returns (into the program) the value of the integral of
spline S(x) over segment [a, b] . Parameters k, r, X, F and M have the same
sense as the corresponding parameters of the si subroutine (p. 291). We will use
the ios function for solving the following task.
Table Listing 4.7 below is taken from Task 3.1 in book [8]. It contains the
mass of a vertically falling plastic foam ball (with radius equal to one inch) and
the results of measurement of its coordinate at different instants of time.
Listing 4.7
Experimental dependence of coordinate
of a falling ball versus time

Ball mass M in kilograms 0.000254

Time t in seconds Coordinate y in meters
-0.132 0
0 0.075
0.1 0.26
0.2 0.525
0.3 0.87
0.4 1.27
0.5 1.73
0.6 2.23
0.7 2.77
0.8 3.35

Because of the air resistance, the ball movement differs from the movement
of a material point with the same mass, M = 0.000254 kg. To estimate this dif-
ference, we have to calculate the ball velocity at moment t = 0.8 s and compare it
with the corresponding velocity of the material point.
For solving this task, we use the laws of mechanics as follows:
according to the work-energy theorem, the change in kinetic energy of the
ball is equal to the work done by the forces acting on the ball, i.e., by the gravita-
tional and air resistance forces;
according to Newton’s second law, the resultant of the forces, acting on
d2y
the ball, is equal to the product of its mass, M, and acceleration, , which is
dt 2
a function of the y coordinate.
Thus, the change in kinetic energy of the ball (which will be calculated by a
program) is equal to

299
 Chapter 4. Cubic Spline

b
W M y 2 ( y ) dy , (4.21)
a
where a = 0, b = 3.35,
d2y
y2 ( y ) . (4.22)
dt 2
Let us consider the following program for calculating the change in kinetic
energy of the ball.
Listing 4.8
Sub main()
Dim T() As Double
Dim Y() As Double
Dim Y2() As Double
Dim MOM() As Double
Dim m As Integer
Dim i As Integer
Dim W As Double
m = Selection.Rows.Count 'quantity of rows
ReDim T(3 To m)
ReDim Y(3 To m)
ReDim Y2(3 To m)
ReDim MOM(3 To m)
'Inputting dependence of coordinate Y versus time T:
For i = 3 To m
T(i) = Selection.Cells(i, 1)
Y(i) = Selection.Cells(i, 2)
Next i
'Calculating spline moments, i.e., acceleration Y2:
1: Call mos(3, m, T, Y, -2, 0, -2, 0, Y2)
'Calculating and outputting change in kinetic energy:
2: Call mos(3, m, Y, Y2, -2, 0, -2, 0, MOM)
3: W = Selection.Cells(1, 2) * ios(3, m, Y, Y2, MOM)
4: MsgBox "W =" & Str(Round(W, 6)) & " J"
End Sub

The source data for this program are the values given in table Listing 4.7
(Fig. 4.5). We have to select this Excel table before running the program.
The program contains two calls of the mos subroutine, operators 1 and 2.
In these calls, we use boundary conditions (4.17) and (4.18) for the constancy of
the spline moments at the left and right ends of segment [a, b] : M 3 M 4 and

300
 4.4. Spline integration

Mm Mm 1
(m = 12). Operator 1 calculates the y2 ( y ) function values accord-
ing to formula (4.22). Operator 2 calculates array MOM of the spline moments,
which are used for integrating the y2 ( y ) function over [a, b] . Operator 3, con-
taining the call of the ios function, calculates the change in kinetic energy of
the ball according to formula (4.21). When executing operator 4, the calculated
value is rounded up to six decimal places and displayed in the standard window
(Fig. 4.6). To finish the program execution, we must click on the OK button in
this window.

Fig. 4.5. The Excel table containing the experimental data

Fig. 4.6. The execution result

301
 Chapter 4. Cubic Spline

According to the program execution result (Fig. 4.6), the change in kinetic
energy of the ball is equal to
W 0.004542 J .
Because the material point’s acceleration is equal to the free fall acceleration,
g = 0.981 m / s2, the change in its kinetic energy on segment [a, b] is equal to the
following value:
Mg (b a) 0.008347 J .
Because of the zero value of the ball velocity at moment t = -0.132 s, the
ball’s kinetic energy at t = 0.8 s is equal to W:
MV2
W 0.004542 J .
2
This gives the following value of the ball velocity at moment t = 0.8 s:
2W
V 5.980 m / s .
M
The corresponding value for the material point is
2 Mg (b a )
V 8.107 m / s .
M
Thus, the air resistance plays an important role in the fall of the plastic foam ball.
To consolidate the material of this and the previous sections, we advise the
reader to write a program for calculating the average value of function f (x) on
segment [a, b] . Formula
b
1
f average f ( x ) dx
b a
a

and the cubic spline on uniform grid a x0 < x1 < x 2 < ... < xn 2
< xn 1
<
xn b must be used. The f (x) function is from Appendix 4; segment [a, b] is
this function’s domain.

302
 4.5. Iterative methods for solving the nonlinear algebraic equation

4.5. Iterative methods for solving

the nonlinear algebraic equation

Let nonlinear function f (x) and segment [u, v] be given with the following
properties:
function f (x) is continuous and monotonous on [u, v] ;
the function values on the left and right boundaries of segment [u, v]
have different signs.
We will consider nonlinear algebraic equation
f (x) 0 . (4.23)
It is obvious that this equation has one and only one solution on [u, v] . We have
to find this solution.
A segment containing a unique solution of equation (4.23), for example
[u, v] , is called the uncertainty segment.
Equation (4.23) can be solved by using the Solver add-in for Excel. Such
usage of Solver will be considered on an example of
f ( x) x cos x 1.5 . (4.24)
For this function, segment [0.5, 2.5] is the uncertainty segment (this is easily
verified).
We enter an initial approximation of the solution, for example 2, into cell G1.
Into cell F1, we enter formula

=G1-COS(G1)-1.5

corresponding to mathematical formula (4.24). Fig. 4.7 shows the worksheet

after clicking on the tick button of the Excel formula bar.
Let us fulfill the following operations:
1) Data > Solver in area Analysis;
2) in the Solver Parameters window opened, enter $F$1 into text box Set
Objective;
3) turn on option Value Of, and put zero into the corresponding text box;
4) enter $G$1 into text box By Changing Variable Cells;
5) enter GRG Nonlinear into box Select a Solving Method by means of the
drop-down list (Fig. 4.8);

303
 Chapter 4. Cubic Spline

6) click on the Solve button;

7) in the Solver Results window opened (Fig. 4.9), click on OK to finish
solving equation (4.23), (4.24).

Fig. 4.7. The Excel worksheet before start of solving equation (4.23), (4.24)

Fig. 4.8. The Solver Parameters window before start of solving the equation

304
 4.5. Iterative methods for solving the nonlinear algebraic equation

Fig. 4.9

The following solution results are given in Fig. 4.10:

cell G1 contains the result of solving equation (4.23), (4.24): x =
1.535395;
cell F1 contains 1.44E-07.
The last value, f (1.535395) = 1.44·10-7, is the so-called residual right-hand
side of equation (4.23), (4.24). It characterizes the accuracy of solving this
equation.

Fig. 4.10. The Excel worksheet upon termination of solving the equation

305
 Chapter 4. Cubic Spline

If the accuracy of the solution is unsatisfactory, we must:

1) open the Options window by clicking on the Options button in the Solver
Parameters window depicted in Fig. 4.8;
2) change settings in the Options window;
3) click on the OK button for returning to the Solver Parameters window.
The limitation of the Solver add-in is obvious: it “must know” how to calcu-
late the value of f (x) at an arbitrary value of x. Therefore, we cannot use this
procedure if f (x) is a tabular (grid) function.
Below, we will develop a program for solving equation (4.23), in which
f (x) is a tabular function. Let us consider the bisection method, which will be
used in this program.
Let f u and f v be the values of f (x) on the corresponding boundaries of
segment [u, v] , and these values have different signs. The bisection method
includes the following steps.
1. The midpoint of segment [u, v] and the function value in this midpoint
are calculated according to formulas
w (u v) / 2 (4.25)
and f w f (w) .
2. If the signs of f w and f u are the same, the left boundary of [u, v] is
shifted to the right: assignments u w and f u f w are performed. Further, the
4th item is fulfilled.
3. If the signs of f w and f v are the same, the right boundary of [u, v] is
shifted to the left: assignments v w and f v f w are performed.
4. The previous three items are repeated (that is, the iteration is repeated)
until condition
v u (4.26)
is satisfied, where is a given positive constant.
5. When condition (4.26) is satisfied, the iterative process of solving equa-
tion (4.23) terminates. The w value is considered as the solution of this equation.
If the residual right-hand side of equation (4.23), f w f (w) , is not small
enough, we have to reduce the value of in condition (4.26).
If the uncertainty segment is not known, it can be found by means of the step-
by-step movement along the x axis in the direction of decrease or increase of the
f (x) function as long as f u and f v have identical signs, i.e., until satisfaction
of condition fu f v 0 . In Section 1.13, we already talked about the movement

306
 4.5. Iterative methods for solving the nonlinear algebraic equation

along the x axis when considering various cycles. In Section 6.2, the step-by-step
movement will be used to find the segment that contains the minimum point of
a nonlinear function of one variable.
The limitation of the bisection method is the same as for the Solver add-in:
the program realization of the method “must know” how to calculate the f (x)
value at an arbitrary value of x. Therefore, the program realization of the bisec-
tion method for tabular function f (x) will use the spline interpolation to deter-
mine the function value at any x value, i.e., for the function continuation.
We will solve the following task by the bisection method.
Time moments t (further, x) and corresponding values of coordinate y of the
vertically falling plastic foam ball are given in table Listing 4.7 from the previ-
ous section. We are interested in the T moment of time when the ball has a given
coordinate, for example Y = 1 m.
The desired value of T is the solution of equation (4.23), where
f ( x) y ( x) Y . (4.27)
Because y (x) is a grid function (defined by table Listing 4.7), f (x) determined
by (4.27) is a grid function too.
According to table Listing 4.7:
at u 0.3 and v 0.4 , segment [u, v] is the uncertainty segment;
fu y (0.3) 1 0.13 , f v y (0.4) 1 0.27 .
To solve equation (4.23), (4.27) by the bisection method, we will use the
cubic spline to determine the function value for any x value (for example,
x 0.35 ), i.e., for making the f (x) function continuous.
The following code is intended for determining moment T when the ball has
a given Y coordinate.
Listing 4.9
Dim ns As Long 'counter of calls
Dim m As Integer
Dim X() As Double
Dim F() As Double
Dim MOM() As Double

Sub main()
Dim s As String, Y As Double
Dim i As Integer
Dim u As Double, fu As Double
Dim v As Double, fv As Double
Dim w As Double, fw As Double

307
 Chapter 4. Cubic Spline

Dim epsilon As Double

1: s = InputBox("Enter value of Y and click OK")
2: Y = Val(s)
m = Selection.Rows.Count 'quantity of rows
ReDim X(3 To m)
ReDim F(3 To m)
ReDim MOM(3 To m)
For i = 3 To m
X(i) = Selection.Cells(i, 1)
F(i) = Selection.Cells(i, 2) - Y
If F(i) = 0 Then
3: MsgBox "T =" & Str(Round(X(i), 6))
End
End If
Next i
If F(3) * F(m) > 0 Then
MsgBox "On boundaries of segment [" & _
Cstr(X(3)) & ", " & Cstr(X(m)) & "], " & _
vbCrLf & "function f(x) has identical signs"
End
End If
'Calculating spline moments:
Call mos(3, m, X, F, -2, 0, -2, 0, MOM)
'Searching uncertainty segment:
For i = 4 To m
4: If F(i - 1) * F(i) < 0 Then Exit For
Next i
5: epsilon = (X(i) - X(i - 1)) * 1E-6
'Solving equation:
u = X(i - 1): fu = F(i - 1)
v = X(i): fv = F(i)
ns = 0
6: Call segment(u, fu, v, fv, w, fw, u, fu, v, fv)
7: If v - u >= epsilon GoTo 6
8: MsgBox "T =" & Str(Round(w, 6)) & vbCrLf & _
CStr(ns) & " iterations"
End Sub

Sub segment(ByVal u As Double, ByVal fu As Double, _

ByVal v As Double, ByVal fv As Double, _
ByRef w As Double, ByRef fw As Double, _
ByRef uu As Double, ByRef fuu As Double, _

308
 4.5. Iterative methods for solving the nonlinear algebraic equation

ByRef vv As Double, ByRef fvv As Double)

ns = ns + 1
If Sgn(fu) = 0 Then
uu = u: fuu = fu
vv = u: fvv = fu
Exit Sub
End If
If Sgn(fv) = 0 Then
uu = v: fuu = fv
vv = v: fvv = fv
Exit Sub
End If
9: w = (u + v) / 2 'bisection method
Call si(3, m, X, F, MOM, w, fw)
If Sgn(fw) = Sgn(fu) Then
uu = w: fuu = fw
Exit Sub
End If
If Sgn(fw) = Sgn(fv) Then
vv = w: fvv = fw
End If
End Sub

The source data are given in table Listing 4.7 (Fig. 4.5). We must select this
Excel table before the code execution. A value of Y must be entered into the text
box of the standard window (see operators 1 and 2 and Fig. 4.11). Uncertainty
segment [ xi 1 , xi ] is determined by means of operator 4.

Fig. 4.11. The window with the coordinate of the falling plastic foam ball

309
 Chapter 4. Cubic Spline

Further, equation (4.23), (4.27) is solved by the bisection method (see opera-
tors 6 and 7). In condition (4.26) for finishing the iterative process, the value of
is equal to a millionth part of the initial uncertainty segment’s length (see
operator 5).
In the iterative process, the segment subroutine is used for reducing the
uncertainty segment length (see operator 6). This subroutine is declared below
the main program.
When executing operator 8, the solution of equation (4.23), (4.27) is rounded
up to six decimal places and displayed in the standard window (Fig. 4.12):
T = 0.334042. The number of iterations is also displayed in this window; it
equals 20. We have to click on the OK button for finishing the execution of code
Listing 4.9.

Fig. 4.12. The window for the bisection method with the
calculated moment of time and the number of iterations

Let us use the secant method (instead of the bisection method) in the itera-
tive process of solving equation (4.23), (4.27). For that, formula (4.25) must be
replaced with the following:
(v u ) f v
w v . (4.28)
fv fu
For obtaining the last formula, let us consider the geometric interpretation of
the secant method depicted in Fig. 4.13. According to this figure, we can deter-
mine a new position ( x w ) of the uncertainty segment boundary (left or right)
as follows:
1) run the secant line through points (u, f u ) and (v, f v );
2) denote the coordinate of the crosspoint of this line with the x axis by w.

310
 4.5. Iterative methods for solving the nonlinear algebraic equation

According to Fig. 4.13, the slope of the secant line, passing through points
(u, f u ) and (v, f v ), is equal to
fu fv
.
w u v w
This ratio leads to formula (4.28).

Fig. 4.13. The geometric interpretation of the secant method

For replacing the iterative method in code Listing 4.9, we have to replace
operator 9 with the following:

9: w = v -(v – u) * fv / (fv – fu) 'secant method

Fig. 4.14 shows the result of executing the new version of code Listing 4.9 for
Y = 1. As we see, the number of iterations is reduced to 12 (from 20).
Let us consider the question of the convergence of the above iterative
processes defined by formulas (4.25) and (4.28). Starting with the bisection
method, we use the following designations:
j 1
w x ;
j
is u x or v x to get the same sign of j as the sign of j 1
defined above.
Here, j is the iteration number, x is the exact solution of the equation, that
is, f (x ) 0.

311
 Chapter 4. Cubic Spline

Fig. 4.14. The window for the secant method with the
calculated moment of time and the number of iterations

The x location near any boundary of the initial uncertainty segment is the
worst situation for the convergence. In this case,
j 1 j
C , (4.29)

where C is a constant close to 0.5 on the left. At an arbitrary location of x ,

double inequality 0 ≤ C < 0.5 is true for the bisection method.
Formula (4.29) allows us to estimate the rate of the convergence of the itera-
tive process. According to this formula, quantity j 1 is proportional to the first
power of quantity j , that is, the bisection method’s iterative process converges
linearly.
Similar consideration of the secant method also gives formula (4.29) for
estimating the rate of the convergence of the corresponding iterative process, but
with inequality 0 ≤ C < 1.
In Section 5.5, we will consider another version of the secant method whose
iterative process has quadratic convergence according to formula (5.26) or, in the
simplest form,
j 1 j j 1
C , (4.30)
where C is a constant, j 1 , j , j 1 are small quantities, j is the iteration
number.
We advise the reader to write a program for solving equation f (x) 0 on
segment [a, b] by the bisection and secant methods. In this equation, f (x ) is
a function from Appendix 4; segment [a, b] is this function’s domain. The user-

312
 4.5. Iterative methods for solving the nonlinear algebraic equation

defined form with the CheckBox element (for choosing the method) must be the
user interface of the program. Segment [a, b] must be the initial uncertainty
segment.
In the next section, we will develop a noniterative method for solving the
nonlinear algebraic equation with grid function f (x) . In Section 5.5, we will
return to iterative methods for solving the nonlinear algebraic equation.

313
 Chapter 4. Cubic Spline

4.6. Noniterative method for solving

the nonlinear algebraic equation

In the previous section, we considered two iterative methods for solving

equation (4.23), (4.27), at that, we used the spline interpolation of grid function
(4.27) in the program realization of these methods. However, equation (4.23),
(4.27) can be solved by means of the cubic spline construction without any itera-
tions.
The fact is that, having calculated the moments of the spline, which corre-
sponds to grid function f (x) on segment [-0.132, 0.8], we obtain the representa-
tion of f (x) in the form of third-degree polynomial S (x) on the uncertainty
segment, for example [u, v] = [0.3, 0.4]. To obtain the value of T, it is enough to
solve cubic equation S (x) 0 , which can be written in canonical form
ax3 bx2 cx d 0. (4.31)
According to handbook [3], equation (4.23), (4.27) is solved in the following
three stages if a 0 .
1. By substituting
b
x z , (4.32)
3a
equation (4.31) is transformed to
z3 3p z 2q 0, (4.33)

3ac b 2 2b 3 bc d
where 3 p , 2q .
3a 2 27a 3 3a 2 a

2. The real roots of equation (4.33) are calculated, one or three (two of
which may coincide), and then the corresponding roots of equation (4.31) are
calculated.
For solving equation (4.33), we use the method represented in the table of
Appendix 5. For subsequent calculations of the real roots of equation (4.31), we
use formula (4.32).
3. The unique solution of equation (4.23), (4.27) is determined.

314
 4.6. Noniterative method for solving the nonlinear algebraic equation

The declaration of the table subroutine, intended for calculating the real
roots of equation (4.31), follows:
Listing 4.10
Sub table(ByVal a, ByVal b, ByVal c, ByVal d, _
ByRef x() As Double)
Dim b3a As Double, p As Double, q As Double
Dim r As Double, gamma As Double, phi As Double
Const beta = 1 / 3, pi = 3.141592654
x(1) = 1E+308
x(2) = 1E+308
x(3) = 1E+308
If 27 * a ^ 3 = 0 Then Exit Sub 'if a = 0
b3a = b / (3 * a)
p = (3 * a * c - b ^ 2) / (3 * a ^ 2) / 3
q = (2 * b ^ 3 / (27 * a ^ 3) - b * c / _
(3 * a ^ 2) + d / a) / 2
r = Sgn(q) * Sqr(Abs(p))
If r ^ 3 = 0 Then 'if p = 0
x(1) = -Sgn(q) * Abs(2 * q) ^ beta - b3a
Exit Sub
End If
If q = 0 Then 'if q = 0
x(1) = - b3a
If p < 0 Then
x(2) = Sqr(-3 * p) - b3a
x(3) = -Sqr(-3 * p) - b3a
End If
Exit Sub
End If
gamma = q / r ^ 3 'gamma > 0
If p < 0 Then
If gamma <= 1 Then
phi = Atn(Sqr(1 - gamma ^ 2) / gamma)
x(1) = -2 * r * Cos(phi / 3) - b3a
x(2) = 2 * r * Cos((pi - phi) / 3) - b3a
x(3) = 2 * r * Cos((pi + phi) / 3) - b3a
Else
phi = Log(gamma + Sqr(gamma ^ 2 - 1))
x(1) = -2 * r * _
(Exp(phi / 3) + Exp(-phi / 3)) / 2 - b3a
End If

315
 Chapter 4. Cubic Spline

Else
phi = Log(gamma + Sqr(gamma ^ 2 + 1))
x(1) = -2 * r * _
(Exp(phi / 3) - Exp(-phi / 3)) / 2 - b3a
End If
End Sub

The subroutine name (table) is due to the fact that the subroutine algorithm
is based on the table in Appendix 5. Let us enter declaration Listing 4.10 into
Module11 of the BookNM workbook.
The table subroutine parameters have the following sense:
a, b, c, d are coefficients a, b and c and constant term d of cubic equation
(4.31);
x is an array for real solutions (one or three; free elements of the x array
are assumed to be equal to 10308).
The following program is intended for determining the T moment when the
falling plastic foam ball (from the two previous sections) has a given Y coordi-
nate.
Listing 4.11
Sub main()
Dim s As String, Y As Double
Dim m As Integer
Dim X() As Double
Dim F() As Double
Dim MOM() As Double
Dim i As Integer, h As Double
Dim a1 As Double, a2 As Double
Dim b1 As Double, b2 As Double
Dim a As Double, b As Double
Dim c As Double, d As Double
Dim T As Double
Dim xxx(1 To 3) As Double, k As Integer
1: s = InputBox("Enter value of Y and click OK")
2: Y = Val(s)
m = Selection.Rows.Count 'quantity of rows
ReDim X(3 To m)
ReDim F(3 To m)
ReDim MOM(3 To m)
For i = 3 To m
X(i) = Selection.Cells(i, 1)
F(i) = Selection.Cells(i, 2) - Y

316
 4.6. Noniterative method for solving the nonlinear algebraic equation

If F(i) = 0 Then
3: MsgBox "T =" & Str(Round(X(i), 6))
End
End If
Next i
If F(3) * F(m) > 0 Then
MsgBox "On boundaries of segment [" & _
CStr(X(3)) & ", " & CStr(X(m)) & "], " & _
vbCrLf & "function f(x) has identical signs"
End
End If
'Calculating spline moments:
Call mos(3, m, X, F, -2, 0, -2, 0, MOM)
'Searching uncertainty segment:
For i = 4 To m
4: If F(i - 1) * F(i) < 0 Then Exit For
Next i
'Forming cubic equation:
h = X(i) - X(i - 1)
a1 = MOM(i - 1) / (6 * h)
a2 = MOM(i) / (6 * h)
b1 = (F(i - 1) - MOM(i - 1) * h ^ 2 / 6) / h
b2 = (F(i) - MOM(i) * h ^ 2 / 6) / h
5: a = a2 - a1
6: b = 3 * a1 * X(i) - 3 * a2 * X(i - 1)
7: c = 3 * a2 * X(i - 1) ^ 2 - 3 * a1 * X(i) ^ 2 + _
b2 - b1
8: d = a1 * X(i) ^ 3 - a2 * X(i - 1) ^ 3 + _
b1 * X(i) - b2 * X(i - 1)
'Solving linear equation:
If 27 * a ^ 3 = 0 And 2 * b = 0 Then
9: T = -d / c
10: MsgBox "T =" & Str(Round(T, 6)) & _
vbCrLf & "- root of linear equation"
End
End If
'Solving quadratic equation:
If 27 * a ^ 3 = 0 Then
11: T = (-c + Sqr(c ^ 2 - 4 * b * d)) / (2 * b)
If X(i - 1) <= T And T <= X(i) Then
12: MsgBox "T =" & Str(Round(T, 6)) & _
vbCrLf & "- root of quadratic equation"

317
 Chapter 4. Cubic Spline

End
End If
13: T = (-c - Sqr(c ^ 2 - 4 * b * d)) / (2 * b)
If X(i - 1) <= T And T <= X(i) Then
14: MsgBox "T =" & Str(Round(T, 6)) & _
vbCrLf & "- root of quadratic equation"
End
End If
End If
'Solving cubic or linear equation:
15: Call table(a, b, c, d, xxx)
If xxx(2) = 1E+308 Then
16: MsgBox "T =" & Str(Round(xxx(1), 6)) & _
vbCrLf & "- root of cubic equation"
End
End If
If X(i - 1) <= xxx(1) And xxx(1) <= X(i) _
And X(i - 1) <= xxx(2) And xxx(2) <= X(i) _
And X(i - 1) <= xxx(3) And xxx(3) <= X(i) Then
17: T = (F(i) * X(i - 1) - F(i - 1) * X(i)) / _
(F(i) - F(i - 1))
18: MsgBox "T =" & Str(Round(T, 6)) & _
vbCrLf & "- root of linear equation"
End
End If
For k = 1 To 3
If X(i - 1) <= xxx(k) And xxx(k) <= X(i) Then
19: MsgBox "T =" & Str(Round(xxx(k), 6)) & _
vbCrLf & "- root of cubic equation"
End
End If
Next k
End Sub

The source data are given in table Listing 4.7 (Fig. 4.5). We must select this
Excel table before the program execution. The value of Y must be entered into
the text box of the standard window (see operators 1 and 2 and Fig. 4.11).
Uncertainty segment [ xi 1 , xi ] is determined by means of operator 4.
Operators 5 — 8 calculate the coefficients and constant term of cubic equa-
tion (4.31) according to the following formulas:
a a2 a1 ,

318
 4.6. Noniterative method for solving the nonlinear algebraic equation

b 3a1 xi 3a2 xi 1
,

c 3a2 xi2 1
3a1xi2 b2 b1 ,

d a1xi3 a2 xi3 1
b1xi b2 xi 1
,
where
Mi 1
Mi
a1 , a2 ,
6 hi 6 hi

Mi h2
1 i 1 M i hi2 1
b1 fi 1
, b2 fi .
6 hi 6 hi

For obtaining these formulas, we must transform expression (4.4) to the follow-
ing form:
S ( x ) ax3 bx2 cx d .
After determining a, b, c and d, the approximate solution of equation (4.23),
(4.27) is calculated by solving one of three algebraic equations — a linear, quad-
ratic or cubic equation.
1. The linear equation is solved if the resulting values of coefficients a and b
are equal to zero. Operator 9 calculates a root of equation cx d 0 according
to formula x d /c .
2. The quadratic equation is solved at the zero value of a ( b 0 ). Two roots
of equation bx2 cx d 0 are calculated by operators 11 and 13 according
to the well known formula [3],

x c c2 4bd /( 2b) .

The solution of equation (4.23), (4.27) is the root ( x or x ) belonging to

uncertainty segment [ xi 1
, xi ] .
3. The cubic equation is solved when a 0 . One or three real roots of equa-
tion ax3 bx2 cx d 0 are the result of calling the table subroutine
(operator 15).
In the ordinary situation, uncertainty segment [ xi 1 , xi ] contains only one
out of one or three real roots of equation ax3 bx2 cx d 0 , and this root is
the solution of equation (4.23), (4.27). However, an extraordinary situation is
possible when the number of real roots of equation ax3 bx2 cx d 0 is

319
 Chapter 4. Cubic Spline

equal to three, and all these roots belong to uncertainty segment [ xi 1

, x i ] . In
this case, operator 17 calculates the solution of equation (4.23), (4.27) according
to the following formula:
f i xi 1 f i 1 xi
x .
fi fi 1
This formula is obtained by solving equation L(x) 0 , in which
xi x x xi 1
L( x) fi 1
fi (4.34)
xi xi 1
xi xi 1

is the linear function that equals fi 1

and fi on the corresponding boundaries of
segment [ xi 1
, xi ] .
As a result of executing operator 3, 10, 12, 14, 16, 18 or 19, the solution
of equation (4.23), (4.27) is rounded up to six decimal places and displayed in
the standard window (Fig. 4.15). After clicking on the OK button, the program
execution is terminated.

Fig. 4.15. The window for the noniterative method

with the calculated moment of time and
the type of the solved algebraic equation

We advise the reader to write a program, similar to Listing 4.11, for solving
equation f (x) 0 on segment [a, b] by the noniterative method. In this equa-
tion, f (x ) is a function from Appendix 4; segment [a, b] is this function’s
domain. Uniform grid a x0 < x1 < x 2 < ... < xn 2
< xn 1
< xn b must
be used.

320
 4.7. Calculating the charge storage capacity

4.7. Calculating the charge storage capacity

Having studied the cubic spline construction, we will continue the modeling
of the silicon photosensitive target (Section 3.13).
It was mentioned on p. 264 that semiconductor layer -6 μm ≤ x ≤ 6 μm,
perpendicular to the x axis, is the potential well for signal electrons generated by
light falling on the layer plane. In this regard, an important electrical parameter
is the charge storage capacity of the target cell.
The charge storage capacity, Q max , is the signal electrons’ maximum charge
(in absolute value), which can be localized in the considered layer of unit area
(meter × meter).
The value of Qmax is calculated in the following five stages:
1) the moments of cubic spline S (x) , determined by the tabulated electric
charge density,

qu( x)
f ( x) q N 0 exp N A ( x) ,
kbT

are calculated, where u (x ) is the solution of Shockley-Poisson equation (3.68),

a ≤ x ≤ 0;
2) the coordinate, x e , at which spline S (x) changes its sign from nega-
tive (at x e ) to positive (at x e ), is calculated;
3) the second derivative, w 2 , of spline S (x) at point x e is calculated
( w 2 is used in the next item when integrating the cubic spline over segment
e ≤ x ≤ 0);
4) the positive charge, Q1 , localized in the semiconductor layer, a ≤ x ≤ 0,
is calculated by integrating S (x) over segment e ≤ x ≤ 0:
0
Q1 S ( x ) dx ; (4.35)
e

321
 Chapter 4. Cubic Spline

5) the calculated value of Q1 is multiplied by 2 because the right-hand side

of the semiconductor layer (x > 0) exists in addition to the reviewed left-hand
side of the layer (x < 0):
Qmax 2Q1 . (4.36)
Let us consider the program for calculating the charge storage capacity,
Qmax .
The source data table below is a part of the table represented in Fig. 3.15.
Listing 4.12
NA u0 x u
7.00E+20 0.00E+00 -1.00E-05 0.000E+00
7.00E+20 4.89E-02 -9.00E-06 1.320E-06
7.00E+20 1.91E-01 -8.00E-06 5.782E-05
7.00E+20 4.12E-01 -7.00E-06 2.529E-03
7.00E+20 6.91E-01 -6.00E-06 1.057E-01
7.00E+20 1.00E+00 -5.00E-06 1.272E+00
7.00E+20 1.31E+00 -4.00E-06 3.521E+00
7.00E+20 1.59E+00 -3.00E-06 6.851E+00
7.00E+20 1.81E+00 -2.00E-06 1.126E+01
7.00E+20 1.95E+00 -1.00E-06 1.676E+01
7.00E+20 1.98E+00 -6.00E-07 1.926E+01
7.00E+20 1.99E+00 -4.00E-07 2.057E+01
7.00E+20 2.00E+00 -3.00E-07 2.124E+01
7.00E+20 2.00E+00 -2.00E-07 2.193E+01
-3.00E+22 2.00E+00 -1.00E-07 2.262E+01
-3.00E+22 2.00E+00 0.00E+00 2.286E+01

The program follows:

Listing 4.13
Sub main()
Dim m As Integer
Dim X() As Double
Dim NA() As Double
Dim U() As Double
Dim F() As Double
Dim MOM() As Double
Const q = 1.6E-19
Const kb = 1.38E-23

322
 4.7. Calculating the charge storage capacity

Const T = 300
Dim w As Double, w1 As Double, w2 As Double
Dim i As Integer, h As Double
Dim a1 As Double, a2 As Double
Dim b1 As Double, b2 As Double
Dim a As Double, b As Double
Dim c As Double, d As Double
Dim e As Double
Dim xxx(1 To 3) As Double, k As Integer
m = Selection.Rows.Count 'quantity of rows
ReDim X(2 To m)
ReDim NA(2 To m)
ReDim U(2 To m)
ReDim F(2 To m)
ReDim MOM(2 To m)
w = q / (kb * T)
For i = 2 To m
NA(i) = Selection.Cells(i, 1)
X(i) = Selection.Cells(i, 3)
U(i) = Selection.Cells(i, 4)
Next i
For i = 2 To m
F(i) = q * (NA(2) * Exp(-w * U(i)) - NA(i))
Next i
'Calculating spline moments:
w = 6 * (F(3) - F(2)) / (X(3) - X(2)) ^ 2
w1 = 6 * (F(m - 1) - F(m)) / (X(m) - X(m - 1)) ^ 2
0: Call mos(2, m, X, F, 1, w, 1, w1, MOM)
'Searching uncertainty segment:
For i = 3 To m
4: If F(i - 1) * F(i) < 0 Then Exit For
Next i
'Forming cubic equation:
h = X(i) - X(i - 1)
a1 = MOM(i - 1) / (6 * h)
a2 = MOM(i) / (6 * h)
b1 = (F(i - 1) - MOM(i - 1) * h ^ 2 / 6) / h
b2 = (F(i) - MOM(i) * h ^ 2 / 6) / h
5: a = a2 - a1
6: b = 3 * a1 * X(i) - 3 * a2 * X(i - 1)
7: c = 3 * a2 * X(i - 1) ^ 2 - 3 * a1 * X(i) ^ 2 + _
b2 - b1

323
 Chapter 4. Cubic Spline

8: d = a1 * X(i) ^ 3 - a2 * X(i - 1) ^ 3 + _
b1 * X(i) - b2 * X(i - 1)
'Solving linear equation:
If 27 * a ^ 3 = 0 And 2 * b = 0 Then
9: e = -d / c
GoTo 21
End If
'Solving quadratic equation:
If 27 * a ^ 3 = 0 Then
11: e = (-c + Sqr(c ^ 2 - 4 * b * d)) / (2 * b)
If X(i - 1) <= e And e <= X(i) Then GoTo 21
13: e = (-c - Sqr(c ^ 2 - 4 * b * d)) / (2 * b)
If X(i - 1) <= e And e <= X(i) Then GoTo 21
End If
'Solving cubic or linear equation:
15: Call table(a, b, c, d, xxx)
If xxx(2) = 1E+308 Then
16: e = xxx(1)
GoTo 21
End If
If X(i - 1) <= xxx(1) And xxx(1) <= X(i) _
And X(i - 1) <= xxx(2) And xxx(2) <= X(i) _
And X(i - 1) <= xxx(3) And xxx(3) <= X(i) Then
17: e = (F(i) * X(i - 1) - F(i - 1) * X(i)) / _
(F(i) - F(i - 1))
GoTo 21
End If
For k = 1 To 3
If X(i - 1) <= xxx(k) And xxx(k) <= X(i) Then
19: e = xxx(k)
GoTo 21
End If
Next k
'Calculating charge storage capacity:
21: Call si(2, m, X, F, MOM, e, w, w1, w2)
X(i - 1) = e
F(i - 1) = 0
MOM(i - 1) = w2
22: w1 = ios(i - 1, m, X, F, MOM) 'calculating Q1
23: w = Round(2 * w1, 4) 'calculating Qmax
24: MsgBox "Qmax = " & CStr(w) & " C/m^2"
End Sub

324
 4.7. Calculating the charge storage capacity

The source data are the values given in table Listing 4.12 (Fig. 4.16). We
must select this table before running the program.

Fig. 4.16. The Excel table with the source data

The program contains the call of the mos subroutine (operator 0) intended
for calculating the spline moments. In this call, conditions (4.13) and (4.14) are
used, corresponding to the zero value of the derivative on the left and right
boundaries of segment [a, 0] : f 2 f ( a ) 0 , f m f (0) 0 (m = 17).
The value of e is the result of solving equation S (x) 0 by the noniterative
method given in the previous section. Regarding the part that solves this equa-
tion, program Listing 4.13 is similar to program Listing 4.11. The solution of the
cubic equation is equal to e = -1.967·10-7 (we observed this value by using the
program debugger of Visual Basic Environment).
Operator 21 calculates the w2 value of the second derivative of the S (x)
spline at point x e by calling the si subroutine intended for the spline inter-
polation. Further, the w2 value is used (as the moment) when integrating S (x)
over segment [e, 0] .

325
 Chapter 4. Cubic Spline

Operator 22 calculates charge Q1 according to formula (4.35) by calling the

ios function intended for integrating the cubic spline. Operator 23 calculates
the charge storage capacity, Qmax , according to formula (4.36) and rounds the
result up to four decimal places. When executing operator 24, the rounded
charge storage capacity is displayed in the standard window (Fig. 4.17).

Fig. 4.17. The window with the program execution result

The calculated value of Qmax is of interest for designing both the photosensi-
tive target and elements of extraction of signal electrons stored in the target’s
cells: the considered device belongs to the class of charge transfer devices.

326
 4.8. Subroutine for automatic creation of graphs

4.8. Subroutine for automatic creation of graphs

Further, we will use the graph creation subroutine whose declaration given
below must be put into Module12 of the BookNM workbook.
Listing 4.14
Sub graph(ByVal sb, ByVal se, ByVal sx, ByVal sy)
Dim wbOldSelection As Range
Set wbOldSelection = Selection
Range(sb & ":" & se).Select
Dim sn As String
sn = ActiveSheet.Name
Selection.NumberFormat = "0.00E+00"
Charts.Add
ActiveChart.ChartType = xlXYScatterSmoothNoMarkers
ActiveChart.SetSourceData Source:= _
Sheets(sn).Range(sb & ":" & se), PlotBy:= _
xlColumns
ActiveChart.Location Where:=xlLocationAsObject, _
Name:=sn
ActiveChart.Axes(xlValue).MajorGridlines.Select
Selection.Delete
ActiveChart.Legend.Select
Selection.Delete
With ActiveChart
.Axes(xlCategory, xlPrimary).HasTitle = True
.Axes(xlCategory, _
xlPrimary).AxisTitle.Characters.Text = sx
.Axes(xlValue, xlPrimary).HasTitle = True
.Axes(xlValue, _
xlPrimary).AxisTitle.Characters.Text = sy
End With
ActiveChart.Axes(xlCategory).AxisTitle.Select
Selection.AutoScaleFont = True
With Selection.Font

327
 Chapter 4. Cubic Spline

.FontStyle = "regular"
.Size = 12
End With
ActiveChart.Axes(xlValue).AxisTitle.Select
Selection.AutoScaleFont = True
With Selection.Font
.FontStyle = "regular"
.Size = 12
End With
wbOldSelection.Select
End Sub

The parameters of the graph subroutine are the following four strings:
sb, se — the strings, which define the Excel range containing the argu-
ment and function values: sb is the address of the top left cell of the range; se is
the address of the bottom right cell;
sx, sy — the names of the horizontal and vertical axes, respectively.
In the parameter names:
letter “s” corresponds to word “string”;
“b” corresponds to word “beginning”;
“e” corresponds to word “end”;
“x” means the horizontal axis;
“y” means the vertical axis.
The basis of the graph subroutine are operators 1 — 32 of program
Listing 3.18. However, we see something new in the above subroutine, namely,
the selection recovery (pp. 173 and 174):
in the subroutine beginning, operator

Set wbOldSelection = Selection

assigns the selected range to the wbOldSelection variable of the Range

type;
in the subroutine end, operator

wbOldSelection.Select

selects the wbOldSelection range.

Because of these two operators, the source data table is selected in Fig. 6.37:
the graph subroutine execution (see operator 13 in Listing 6.15) does not
change the selection. This picture can be considered as an example of the subrou-
tine work.

328
 4.9. Cubic spline usage for solving the second-order linear differential equation

4.9. Cubic spline usage for solving the second-

order linear differential equation

In Chapter 3, for solving the boundary value problem for the second-order
linear differential equation, we considered a method based on the solution ap-
proximation by the second-degree polynomial (Section 3.14). Below, for solving
the same problem, we will consider a method based on the solution approxima-
tion by the cubic spline.
As shown in Section 3.4, equation (3.6) can be written in form (3.30),
d 2U
E ( x )U F ( x) , (4.37)
dx 2
where E (x) and F (x) are given functions, U (x) is an unknown function. Con-
ditions (3.7) and (3.8) on the boundaries of segment [a, b] take form (3.33) and
(3.36),
A2 U ( a ) A1U ( a ) A3 , (4.38)
B2 U ( b ) B1U (b ) B3 , (4.39)
where A1 , A2 , A3 , B1 , B2 , B3 are given parameters.
We will develop a cubic spline method for solving the formulated boundary
value problem, (4.37) — (4.39), on the following grid familiar to us: a x k <
xk 1
< xk 2
< ... < xr 2
< xr 1
< xr b.
Let U k , U k 1
, Uk 2
, ..., U r 2
, Ur 1
, U r be the solution values at points
x k , xk 1
, xk 2
, ..., xr 2
, xr 1
, x r , respectively. The S (x) cubic spline
is considered whose graph passes through points ( x k , U k ), ( xk 1
, Uk 1
),
( xk 2
, Uk 2
), . . . , ( xr 2
, Ur 2
), ( xr 1
, Ur 1
), ( x r , U r ).
Because the spline moments are the values of S (x) at the grid nodes, equa-
tion (4.37) gives the following expression for the moment at the i-th node:
M i Fi EiU i , (4.40)

329
 Chapter 4. Cubic Spline

where Fi and E i are the values of functions E (x) and F (x) at the i-th node,
k ≤ i ≤ r.
After replacing f i with U i in expression (4.10) for i , we have
Ui 1
Ui Ui Ui 1
hi 1
hi
i
. 6 (4.41)
hi hi 1
By substituting expressions (4.40) and (4.41) into (4.9), we obtain the follow-
ing equality:
6 6 6
i
Ei 1
Ui 1
2 Ei Ui
hi (hi hi 1
) hi 1
(hi hi 1
) hi (hi hi 1
)

6
i
Ei 1
Ui 1
F
i i 1
2 Fi F
i i 1
.
hi 1
(hi hi 1
)
Substituting expressions (4.10) for i
and i
, we get
hi 6
Ei 1
Ui 1
hi hi 1
hi (hi hi 1
)

6 6
2 Ei Ui
hi 1
( hi hi 1
) hi ( hi hi 1
)

hi 1 6
Ei 1
Ui 1
hi hi 1
hi 1
( hi hi 1
)
hi hi 1
Fi 1
2 Fi Fi 1
.
hi hi 1
hi hi 1

Let us multiply both sides of the last equality by hi hi 1

/ 3 . The following
linear algebraic equation is the result:
U
i i 1
U
i i
Ui
i 1 i
, (4.42)
where U i 1
, U i , Ui 1
are the unknowns, i = k + 1, k + 2, ..., r – 2, r – 1,

hi 1
( Ei h2
1 i
6)
i
,
3( hi hi 1
)

330
 4.9. Cubic spline usage for solving the second-order linear differential equation

2
i
Ehh 2,
3 i i i 1

hi ( E i h2
1 i 1
6)
i
, (4.43)
3( hi hi 1
)

Fi h2h
1 i i 1 2 Fi h2 h
1 i 1 i
i
Fh h .
3( hi hi 1
) 3 i i i 1
3( hi hi 1
)
We write the boundary conditions in the following form similar to (3.11) and
(3.12):
2U k U
k k 1 k
, (4.44)

r
Ur 1
2U r r
, (4.45)
where k
, k
, r
, r
are given parameters.
To obtain formulas for calculating the values of k
and k
, we set i = k,
xk 1
xk hk 1
, fk 1
Uk 1
, fk U k in expression (4.8):
hk 1
Uk 1
Uk Mk 1
Mk
S ( xk 0) Mk hk 1
.
2 hk 1
6
Let us multiply both sides of the last equality by A1 and substitute expressions
Mk Fk EkU k ,
Mk 1
Fk 1
Ek 1
Uk 1
,
corresponding to (4.40) at i = k and i = k + 1. The following equality is the result:
h U Uk
A1S ( xk 0) A1 ( Fk EkU k ) k 1 A1 k 1
2 hk 1
Fk 1
Ek Uk
1 1
Fk EkU k
A1 hk 1
. (4.46)
6
Because U k U (a ) and S ( xk 0) U ( a ) , expression
A1S ( xk 0) A3 A2U k (4.47)
follows from (4.38).
Equating the right-hand sides of (4.46) and (4.47), we have

331
 Chapter 4. Cubic Spline

hk 1
Uk 1
Uk
A3 A2U k A1 ( Fk EkU k ) A1
2 hk 1
Fk 1
Ek Uk
1 1
Fk EkU k
A1 hk 1
.
6
This equation can be written in form (4.44), where
( Ek h2
1 k 1
6) A1
k
, (4.48)
( Ek hk2 1
3) A1 3hk A
1 2

[(2 Fk Fk 1
)hk 1 1
A 6 A3 ]hk 1
k
. (4.49)
( Ek hk2 1
3) A1 3hk A
1 2
To obtain formulas for calculating the values of r
and r
in (4.45), we set
i = r, xr xr 1
hr , f r U r , fr 1
Ur 1
in (4.6):
hr Ur Ur 1
Mr Mr 1
S ( xr 0) Mr hr .
2 hr 6
Let us multiply both sides of the last equality by B1 and substitute expressions
Mr Fr E rU r ,
Mr 1
Fr 1
Er 1
Ur 1
,
corresponding to (4.40) at i = r and i = r – 1. The following equality is the result:
h U Ur 1
B1S ( xr 0) B1 ( Fr ErU r ) r B1 r
2 hr
Fr E rU r Fr 1
Er Ur
1 1
B1 hr . (4.50)
6
Because U r U (b ) and S ( xr 0) U (b) , expression
B1S ( xr 0) B3 B2U r (4.51)
follows from (4.39).
Equating the right-hand sides of (4.50) and (4.51), we have
h U Ur 1
B3 B2U r B1 ( Fr ErU r ) r B1 r
2 hr

332
 4.9. Cubic spline usage for solving the second-order linear differential equation

Fr E rU r Fr 1
Er 1
Ur 1
B1 hr . (4.52)
6
This equation can be written in form (4.45), where
( Er h2
1 r
6) B1
r
, (4.53)
( E r hr2 3) B1 3hr B2
[(2 Fr Fr 1
) hr B1 6 B3 ]hr
r
. (4.54)
( E r hr2 3) B1 3hr B2
The system of linear algebraic equations (4.42), (4.44) and (4.45) is called the
cubic spline scheme for boundary value problem (4.37) — (4.39).
The values of unknown U k , U k 1 , U k 2 , ..., U r 2 , U r 1 , U r are deter-
mined by solving the formulated cubic spline scheme. In this case, the decompo-
sition method (Section 3.2) can be used because forms (3.9), (3.11) and (3.12)
are respectively available for equations (4.42), (4.44) and (4.45). Formulas
(4.48), (4.49) and k 2 are used in formulas (3.16) and (3.17) to start the
forward sweep. Formulas (4.53), (4.54) and r
2 are used in formula (3.18)
to start the backward sweep.

333
 Chapter 4. Cubic Spline

4.10. Program realization of the cubic spline method

for solving the linear differential equation

Let us develop a subroutine for solving differential equation (4.37) under

conditions (4.38) and (4.39) on the left and right boundaries of segment [a, b] .
The values of parameters k
and k
in equation (4.44), corresponding to the
left boundary condition, are calculated according to formulas (4.48) and (4.49).
These values and k 2 are used in formulas (3.16) and (3.17) for starting the
forward sweep of the decomposition method. Further, the calculation is per-
formed according to recurrence formulas (3.14) and (3.15).
The values of parameters r and r in equation (4.45), corresponding to the
right boundary condition, are calculated according to formulas (4.53) and (4.54).
These values and r 2 are used in formula (3.18) for starting the backward
sweep. Further, the calculation is performed according to recurrence formula
(3.13).
The fobas subroutine for solving the boundary value problem for differen-
tial equation (4.37) has the following form:
Listing 4.15
Sub fobas(ByVal k, ByVal r, ByRef X() As Double, _
ByRef E() As Double, ByRef F() As Double, _
ByVal A1, ByVal A2, ByVal A3, _
ByVal B1, ByVal B2, ByVal B3, _
ByRef U() As Double)
Dim alpha As Double, beta As Double
Dim gamma As Double, delta As Double
Dim i As Integer, w As Double
Const c As Double = 2 / 3
Dim h() As Double: ReDim h(k + 1 To r)
Dim P() As Double: ReDim P(k + 1 To r)
Dim Q() As Double: ReDim Q(k + 1 To r)
For i = k + 1 To r
h(i) = X(i) - X(i - 1)
Next i

334
4.10. Program realization of the cubic spline method for solving the linear differential equation

'Forward sweep:
w = (E(k) * h(k + 1) ^ 2 - 3) * A1 + _
3 * h(k + 1) * A2
gamma = (E(k + 1) * h(k + 1) ^ 2 + 6) * A1 / w
delta = ((2 * F(k) + F(k + 1)) * _
h(k + 1) * A1 + 6 * A3) * h(k + 1) / w
P(k + 1) = -gamma / 2
Q(k + 1) = delta / 2
For i = k + 1 To r - 1
w = 3 * (h(i) + h(i + 1))
alpha = h(i + 1) * _
(E(i - 1) * h(i) ^ 2 + 6) / w
beta = c * E(i) * h(i) * h(i + 1) - 2
gamma = h(i) * _
(E(i + 1) * h(i + 1) ^ 2 + 6) / w
delta = F(i - 1) * h(i) ^ 2 * h(i + 1) / w + _
c * F(i) * h(i) * h(i + 1) + _
F(i + 1) * h(i + 1) ^ 2 * h(i) / w
w = alpha * P(i) + beta
P(i + 1) = -gamma / w
Q(i + 1) = (delta - alpha * Q(i)) / w
Next i
'Backward sweep:
w = (E(r) * h(r) ^ 2 - 3) * B1 - 3 * h(r) * B2
alpha = (E(r - 1) * h(r) ^ 2 + 6) * B1 / w
delta = ((2 * F(r) + F(r - 1)) * _
h(r) * B1 - 6 * B3) * h(r) / w
U(r) = (delta - alpha * Q(r)) / _
(alpha * P(r) + 2)
For i = r To k + 1 Step -1
U(i - 1) = P(i) * U(i) + Q(i)
Next i
End Sub

We enter the above declaration into Module13 of the BookNM workbook.

The set of parameters of fobas is close to the set of parameters of the foba
subroutine in Section 3.15:
k, r are numbers of the left and right boundary nodes of the grid;
X is an array of grid nodes;
E is an array of values of the coefficient of equation (4.37) at the grid
nodes;

335
 Chapter 4. Cubic Spline

F is an array of values of the right-hand side of equation (4.37);

A1, A2, A3 are values of the corresponding parameters in left boundary
condition (4.38);
B1, B2, B3 are values of the corresponding parameters in right boundary
condition (4.39);
U is an array intended for the solution values.
The developed fobas subroutine will be used in the remaining sections of
this chapter for solving two concrete boundary value problems.

336
 4.11. Solving the linear differential equation by the cubic spline method

4.11. Solving the linear differential equation

by the cubic spline method

To demonstrate the use of the cubic spline method described above, we will
solve equation (3.42) with boundary conditions (3.43). This boundary value
problem, concerning temperature characteristics of the radial flow between paral-
lel round disks, can be written in form (4.37) — (4.39), where
E ( x) 0.25x 4 (3c 1) x ,
F (x) 0 ,
A1 B1 B3 0,
A2 A3 B2 1.
As in Section 3.6, we consider that c = 10 and a ≤ x ≤ b, where a = 0, b = 1.5.
The program below is intended for solving the formulated problem by using
the fobas subroutine from the previous section.
Listing 4.16
Sub main()
Dim X() As Double
Dim E() As Double
Dim F() As Double
Dim U() As Double
Dim c As Double, b As Double, l As Integer
Dim h As Double, i As Integer
Dim sb As String, se As String
c = Selection.Cells(1, 2)
b = Selection.Cells(2, 2)
l = Selection.Cells(3, 2)
h = b / l
ReDim X(5 To 5 + l)
ReDim E(5 To 5 + l)
ReDim F(5 To 5 + l)
ReDim U(5 To 5 + l)
For i = 5 To 5 + l
X(i) = (i - 5) * h

337
 Chapter 4. Cubic Spline

E(i) = -0.25 * X(i) ^ 4 - (3 * c + 1) * X(i)

F(i) = 0
Next i
0: Call fobas(5, 5 + l, X, E, F, 0, 1, 1, 0, 1, 0, U)
Selection.Cells(4, 3) = "x"
Selection.Cells(4, 4) = "U"
For i = 5 To 5 + l
Selection.Cells(i, 3) = X(i)
Selection.Cells(i, 4) = U(i)
Next i
sb = Selection.Cells(5, 3).Address
se = Selection.Cells(5 + l, 4).Address
1: Call graph(sb, se, "x", "U")
For i = 5 To 5 + l
Selection.Cells(i, 1) = X(i)
2: U(i) = U(i) * Exp(-X(i) ^ 3 / 6)
Selection.Cells(i, 2) = U(i)
Next i
sb = Selection.Cells(5, 1).Address
se = Selection.Cells(5 + l, 2).Address
3: Call graph(sb, se, "x", "u")
Range("P33").Select
End Sub

This main program is used similarly to programs Listing 3.2 and Listing 3.3
of Section 3.6:
the initial data are the values located in the table (Fig. 3.2a);
we must select this Excel table before the program execution (Fig. 3.2b).
Let us consider several operators of program Listing 4.16.
Operator 0 is the call of the fobas subroutine for solving boundary value
problem (3.42), (3.43) by the cubic spline method. The calculated values of the
solution, U (x) , are put into the U column (Fig. 4.18). The U (x) graph on the
Excel worksheet is created automatically when executing the graph subroutine
from Section 4.8; operator 1 is the subroutine call.
Operator 2 calculates solution u (x) of boundary value problem (3.39),
(3.40) by using formula (3.41). The calculated values of u (x) are put into the u
column (Fig. 4.18). The u (x) graph on the Excel worksheet is the result of the
second call of the graph subroutine (operator 3). This u (x) dependence is
close to u (x) depicted in Fig. 3.3.

338
 4.11. Solving the linear differential equation by the cubic spline method

Fig. 4.18. The solutions of problems (3.42), (3.43) and (3.39), (3.40)
and the corresponding graphs (after their displacement by the mouse)

Note that one of the fobas parameters is the X array of grid nodes, i.e., this
subroutine is usable for both uniform and nonuniform grids. In this section, we
used a uniform grid; in the next section, a nonuniform grid will be used.

339
 Chapter 4. Cubic Spline

4.12. Modeling of heating of a geophysical cable.

Locally one-dimensional scheme

Cables for geophysical works are intended for repeated descent of instru-
ments (fastened to the end of the cable) into a borehole and for electrical connec-
tion of these instruments with terrestrial equipment. By measuring the cable
length, the depth of the strata bedding can be determined. Single- and multi-
strand cables are used in practice.
Let us consider the cross section of an armored single-strand cable for geo-
physical works (Fig. 4.19).

Fig. 4.19. Cross section of an armored single-strand cable: 0 is

the cross section center, a, r1, r2, b are special points of the x radius

According to website http://www.proelectro.ru/lib/kabel/128.html, the cable

for temperatures up to 180 °C has the following design. The central conductor

340
 4.12. Modeling of heating of a geophysical cable. Locally one-dimensional scheme

consists of seven copper wires, each of diameter 0.35 mm. This conductor is
covered with a 1.35 mm polymer layer for electrical insulation. A cotton layer
(which is an electrical insulator too) and double-layer armor (two zinced steel
wires) cover the polymer.
We will consider that the conductor is a copper wire of radius 0.5 mm
(0 ≤ x ≤ a in Fig. 4.20) and the insulator has the following three layers:
the 1.35 mm homogeneous polymer adjoins the central wire (a ≤ x ≤ r1 );
the 0.35 mm homogeneous cotton adjoins the armor ( r2 ≤ x ≤ b);
the 0.3 mm non-homogeneous material is placed between these two layers
( r1 < x < r2 ).

Fig. 4.20. The circular ring, a ≤ x ≤ b, corresponding

to an axially-symmetric insulator
We should calculate the overheating of the conductor and insulator (in rela-
tion to the armor) caused by constant electric current I 0 in the conductor. At
high electric current, this overheating added to the armor (borehole) temperature
damages the cable. The problem will be solved under the assumption that physi-
cal parameters of the conductor and insulator are temperature-independent.
The temperature distribution in the insulator, u (x ) at a ≤ x ≤ b, is described
by the following heat equation in cylindrical coordinates:
u 1 u
cd ( x) x , (4.55)
t x x x

341
 Chapter 4. Cubic Spline

where t is time, x is the radial coordinate, (x) is the thermal conductivity of the
insulator, c (x ) is the specific heat capacity, d (x) is the density of the insulator.
Because I 0 const , the temperature distribution does not depend on time,
u/ t 0 , and equation (4.55) becomes
d du
( x) x 0. (4.56)
dx dx
The boundary conditions are as follows:
du
(a) T (a) , (4.57)
dx
u (b) T (b) , (4.58)
where a, b are the inner and outer radii of the ring corresponding to the compli-
cated insulator (Fig. 4.20), T (a) is the derivative of temperature with respect
to x on the inner boundary of the ring, T (b) is the temperature on the outer
boundary.
To use the fobas subroutine from Section 4.10, we will transform boundary
value problem (4.56) — (4.58). For that, we use the following designation:
( x) x ( x) . (4.59)
Equation (4.56) takes form
d du
( x) 0.
dx dx
By differentiating the left-hand side of the last equation as the product of two
functions, (x) and du / dx , we can write this equation as follows:
d 2ud du
0.
dx2 dx dx
Dividing both sides by (x) , we have the following equation:
d 2u du
2
g ( x) 0, (4.60)
dx dx
where
d / dx
d ln
g ( x) . (4.61)
dx
Excluding the first derivative from equation (4.60) by substitution (3.27),
(a)
u ( x) U ( x) , (4.62)
( x)

342
 4.12. Modeling of heating of a geophysical cable. Locally one-dimensional scheme

we write equation (4.60) in form (4.37),

d 2U
E ( x )U F ( x) ,
2
dx
where
E 0.25g 2 ( x) 0.5g ( x) ,
F 0
according to (3.31) and (3.32). Using formula (4.61), we have
2
d ln d 2 ln
E 0.25 . 0.5 (4.63)
dx dx2
Boundary conditions (4.57) and (4.58) take forms (4.38) and (4.39),
A2 U ( a ) A1U ( a ) A3 ,
B2 U ( b ) B1U (b ) B3 ,
where
A1 1,
d ln
A2 0.5 (a) ,
dx
A3 T (a) , (4.64)
B1 0,
B2 1,

(b )
B3 T (b )
(a)
according to (3.34), (3.35), (3.37) and (3.38).
We need a formula that relates the electric current, I 0 , with the derivative of
temperature (with respect to x) on the inner boundary of the ring, T (a) .
To relate I 0 with T (a) , let us look at a piece of wire of length h. Geometri-
cally, this piece of wire is a right circular cylinder [3] of height h with the base
whose radius is equal to a.
According to the heat equation in integral form, the power of electric current
is equal to the heat flow through the lateral surface of the cylinder:
I 02 R (a)T (a) M , (4.65)

343
 Chapter 4. Cubic Spline

where R is the electric resistance of the piece of wire, M is the lateral area of the
cylinder. Because

R h , M 2 ah,
a2
where is the resistivity of copper, equality (4.65) gives

I 02
T (a) . (4.66)
2 3
2 a (a)
Let us develop a program for calculation of overheating of the conductor and
insulator (in relation to the armor) caused by the I 0 current. This program will
use the source data table given below.
Listing 4.17
a 5.00E-04
I0 25
rho 1.78E-08
r1 1.85E-03
l1 10
lambda1 0.25
r2 2.15E-03
l2 5
b 2.50E-03
l3 5
lambda3 0.05

In Listing 4.17, we see:

a — the radius of the copper wire or the inner radius of the homogeneous
polymer (Fig. 4.20), in meters;
I 0 — the electric current in the copper wire, in amperes;
— the resistivity of copper, in Ω·m;
r1 — the outer radius of the homogeneous polymer, in meters;
l1 — the number of steps on segment a ≤ x ≤ r1 ;

1
— the thermal conductivity of the homogeneous polymer, W / (m·K);
r2 — the outer radius of the non-homogeneous material, in meters;
l2 — the number of steps on segment r1 ≤ x ≤ r2 ;

344
 4.12. Modeling of heating of a geophysical cable. Locally one-dimensional scheme

b — the outer radius of the homogeneous cotton, in meters;

l3 — the number of steps on segment r2 ≤ x ≤ b;

3
— the thermal conductivity of the homogeneous cotton, W / (m·K).
The u (x) dependence should be the main result of the program execution.
The program follows:
Listing 4.18
Sub main()
Dim x() As Double
Dim lambda() As Double
Dim mu() As Double
Dim ln_mu() As Double
Dim ln_mu1() As Double
Dim ln_mu2() As Double
Dim E() As Double
Dim F() As Double
Dim U() As Double
Dim a As Double, I0 As Double, rho As Double
Dim r1 As Double, l1 As Integer, lambda1 As Double
Dim r2 As Double, l2 As Integer
Dim b As Double, l3 As Integer, lambda3 As Double
Dim h1 As Double, h2 As Double, h3 As Double
Dim l As Integer, i As Integer
Dim T1a As Double, Tb As Double
Dim A1 As Double, A2 As Double, A3 As Double
Dim B1 As Double, B2 As Double, B3 As Double
Const pi As Double = 3.141592654
Dim w As Double
Dim sb As String, se As String
a = Selection.Cells(1, 2)
I0 = Selection.Cells(2, 2)
rho = Selection.Cells(3, 2)
r1 = Selection.Cells(4, 2)
l1 = Selection.Cells(5, 2)
lambda1 = Selection.Cells(6, 2)
r2 = Selection.Cells(7, 2)
l2 = Selection.Cells(8, 2)
b = Selection.Cells(9, 2)
l3 = Selection.Cells(10, 2)
lambda3 = Selection.Cells(11, 2)

345
 Chapter 4. Cubic Spline

h1 = (r1 - a) / l1
h2 = (r2 - r1) / l2
h3 = (b - r2) / l3
l = l1 + l2 + l3
ReDim x(13 To 13 + l)
ReDim lambda(13 To 13 + l)
ReDim mu(13 To 13 + l)
ReDim ln_mu(13 To 13 + l)
ReDim ln_mu1(13 To 13 + l)
ReDim ln_mu2(13 To 13 + l)
ReDim E(13 To 13 + l)
ReDim F(13 To 13 + l)
ReDim U(13 To 13 + l)
'Definition of arrays x and lambda:
x(13) = a
lambda(13) = lambda1
For i = 14 To 13 + l1
x(i) = x(i - 1) + h1
lambda(i) = lambda1
Next i
For i = 14 + l1 To 13 + l1 + l2
x(i) = x(i - 1) + h2
lambda(i) = 0.5 * (lambda1 - lambda3) * _
Cos(pi * (x(i) - r1) / (r2 - r1)) + _
0.5 * (lambda1 + lambda3)
Next i
For i = 14 + l1 + l2 To 13 + l
x(i) = x(i - 1) + h3
lambda(i) = lambda3
Next i
Selection.Cells(12, 1) = "x"
Selection.Cells(12, 2) = "lambda"
For i = 13 To 13 + l
Selection.Cells(i, 1) = x(i)
Selection.Cells(i, 2) = lambda(i)
Next i
sb = Selection.Cells(13, 1).Address
se = Selection.Cells(13 + l, 2).Address
1: Call graph(sb, se, "x", "lambda")
'Definition of arrays mu, ln_mu, ln_mu1 and ln_mu2:
For i = 13 To 13 + l
mu(i) = lambda(i) * x(i)

346
 4.12. Modeling of heating of a geophysical cable. Locally one-dimensional scheme

ln_mu(i) = Log(mu(i))
Next i
2: Call mos(13, 13 + l, x, ln_mu, _
0, -2 / a ^ 2, 0, -2 / b ^ 2, ln_mu2)
For i = 13 To 13 + l
3: Call si(13, 13 + l, x, ln_mu, ln_mu2, _
x(i), w, ln_mu1(i))
Next i
'Definition and solution of boundary value problem:
For i = 13 To 13 + l
E(i) = -0.25 * ln_mu1(i) ^ 2 - 0.5 * ln_mu2(i)
F(i) = 0
Next i
T1a = -I0 ^ 2 * rho / _
(2 * pi ^ 2 * a ^ 3 * lambda1)
Tb = 0
A1 = 1
A2 = -0.5 * ln_mu1(13)
A3 = T1a
B1 = 0
B2 = 1
B3 = Tb * Sqr(mu(13 + l) / mu(13))
4: Call fobas(13, 13 + l, x, E, F, A1, A2, A3, _
B1, B2, B3, U)
Selection.Cells(12, 3) = "x"
Selection.Cells(12, 4) = "u"
For i = 13 To 13 + l
Selection.Cells(i, 3) = x(i)
5: U(i) = U(i) * Sqr(mu(13) / mu(i))
Selection.Cells(i, 4) = U(i)
Next i
sb = Selection.Cells(13, 3).Address
se = Selection.Cells(13 + l, 4).Address
6: Call graph(sb, se, "x", "u")
Range("O36").Select
End Sub

The source data for this program are the values located in the Excel table
depicted in Fig. 4.21. We must select this table before the program execution.
Let us consider the main stages of the program execution.
After inputting the source data, the (x) function is defined in domain
[a, b] as follows:

347
 Chapter 4. Cubic Spline

1
if a x r1
1 3
x r1 1 3
( x) cos if r1 x r2
2 r2 r1 2
3
if r2 x b

Operator 1 displays the (x) graph on the Excel worksheet (Fig. 4.22).

Fig. 4.21. The source data table

Further, function (x) is determined according to formula (4.59), and

d ln d 2 ln d 2 ln
ln ( x) , , are determined too. Function
is the result of
dx dx2 dx2
calling the mos subroutine (operator 2) with parameters defined by formulas
(4.15) and (4.16), in which
d 2 ln 1 d 2 ln 1
fk (a) , fr (b) .
dx2 a2 dx 2 b2
d ln
Function is the result of calling the si subroutine (operator 3).
dx

348
 4.12. Modeling of heating of a geophysical cable. Locally one-dimensional scheme

Fig. 4.22. The program execution results: (x) is the thermal

conductivity change; u (x) is the overheating change

Thus, we defined boundary value problem (4.37) — (4.39) by using formulas

(4.63), (4.64) and (4.66), F 0 . Operator 4 is the call of the fobas subroutine
for solving this problem by the cubic spline method. Function U (x) is the result.
Operator 5 calculates solution u (x) of boundary value problem (4.56) —
(4.58) by using formula (4.62). The u (x) graph, also located on the Excel work-
sheet (Fig. 4.22), is the result of the second call of the graph subroutine (opera-
tor 6). The u (x) dependence shows the wire and insulator overheating.
According to Fig. 4.22, the wire overheating is equal to T Ta Tb =
21.8 K for I 0 = 25 A. The insulator around the wire has the same overheating.
For I 0 = 50 A, the program execution gives T (50) = 87.3 K.
It was mentioned above that the considered geophysical cable was designed
for temperatures up to 180 °C. Therefore, for given current I 0 , the cable can be

349
 Chapter 4. Cubic Spline

used in boreholes whose temperature does not exceed 180 T ( I 0 ) degrees

Celsius, where T ( I 0 ) is the result of the program execution for I 0 .
We could use the finite difference method (Chapter 3) in the above program
concerning the geophysical cable. However, the cubic spline method is more
accurate because we use the nonuniform grid. Let us focus on this.
As it was stated at the end of Section 3.14, at the transition from a uniform
grid to a nonuniform grid, the error of the finite difference analog of the second
derivative changes from the 2nd order of smallness to the 1st order. According to
formula (4.19), the second derivative of the difference between the function and
the corresponding cubic spline has the 2nd order of smallness for both uniform
and nonuniform grids. Thus, the cubic spline method is more accurate than the
finite difference method in the case of a nonuniform grid.
Let us return to (4.55). This form of the heat equation is caused by the axial
symmetry of the problem under consideration (Fig. 4.20). In the absence of this
symmetry (Fig. 4.23), the heat equation has the following form:
u u u
cd , (4.67)
t x1 x1 x2 x2
where x1 and x 2 are the Cartesian coordinates, ( x1 , x 2 ) is the thermal con-
ductivity of the insulator, c ( x1 , x 2 ) is the specific heat capacity, d ( x1 , x 2 ) is
the density. We must solve the initial value problem for (4.67), i.e., calculate the
temperature distribution, u (t , x1 , x 2 ) , at t > 0 when u (0, x1 , x 2 ) is given.
For solving this problem, we introduce a uniform grid on the t axis, that is,
we consider moments t k , where is the time step, k = 1, 2, 3, ... Besides,
we cover the equation’s domain (Fig. 4.23) by a two-dimensional spatial grid.
According to the locally one-dimensional scheme (LOS) [4], to calculate the
values of u (t , x1 , x 2 ) over the values of u (t , x1 , x 2 ) , we must successively
solve the boundary value problems for the following differential equations:

u~ ( x1 , x 2, j ) u (t , x1 , x 2, j ) d du~ ( x1 , x2, j )
cd , (4.68)
d x1 dx1

u (t , x1, i , x 2 ) u~ ( x1, i , x 2 ) d du(t , x1, i , x2 )

cd , (4.69)
d x2 dx2

where u~ ( x1 , x 2 ) is an auxiliary function of two variables, x1, i and x2, j are the
coordinates of the gridlines (Fig. 4.23), indices i and j are the gridline numbers.

350
 4.12. Modeling of heating of a geophysical cable. Locally one-dimensional scheme

Fig. 4.23. Cross section of an axially-asymmetric insulator;

ring a x12 x22 b is the equation’s domain

More precisely, the calculation of the temperature distribution at the t mo-

ment, u (t , x1 , x 2 ) , over the known distribution at the previous t moment,
u (t , x1 , x 2 ) , includes the following two stages:
1) calculation of two-dimensional distribution u~ ( x1 , x 2 ) by solving the
boundary value problems for differential equation (4.68) at fixed values of j
( x1 is the independent variable);
2) calculation of two-dimensional distribution u (t , x1 , x 2 ) by solving the
boundary value problems for differential equation (4.69) at fixed values of i
( x 2 is the independent variable).
To solve the boundary value problems for equations (4.68) and (4.69) of the
LOS, the following two methods are used: the finite difference method in [4] and
the cubic spline method in [6]. In other words, the locally one-dimensional finite
difference scheme and the locally one-dimensional cubic spline scheme are
respectively developed in [4] and [6].
The LOS can be used for solving various problems with n ≥ 2 spatial coordi-
nates: the problem solved in [6] has little resemblance to the above one.
In addition to the cubic spline, more simple quadratic and linear splines are
used for the solution of applied problems. Below, we will consider these splines.

351
 Chapter 5.
Quadratic and Linear Splines

In addition to the third-degree (cubic) spline of the previous chapter, we con-

sider the simpler second-degree (quadratic) and first-degree (linear) splines. The
quadratic spline is used for solving the initial value problem (of Cauchy) for the
system of differential equations. The linear spline is used in the least-squares
method.
While solving the initial value problem, the system of nonlinear algebraic
equations should be solved. Therefore, we also consider the Newton method. For
solving a single nonlinear algebraic equation, in addition to the Newton (tangent)
method, we consider two Newton-like methods, namely, the secant and
Steffensen methods (the consideration of the secant method was started in
Section 4.5).
For demonstration of the splines possibilities, programs are developed for the
following purposes:
to simulate the piano mechanism linking a key with hammer;
to determine the dependence of production results versus factors (the so-
called production function) by the least-squares method;
to solve the sound insulation problem.
The last problem also demonstrates possibilities of user-defined subroutines
for the forward and backward discrete Fourier transforms of a periodic tabular
function. These subroutines (Section 5.11) are based on the corresponding
Algol 60 procedures developed by A. L. Zakharov, scientific worker of Pulsar
R&D Manufacturing Company, Moscow, in the end of the 1970s.

352
 5.1. Definition of quadratic spline. Spline slopes

5.1. Definition of quadratic spline. Spline slopes

As in Section 4.1, let us consider segment [a, b] covered with grid a xk <
xk 1
< xk 2
< ... < xr 2
< xr 1
< xr b . Values f k , f k 1
, fk 2
, ...,
fr 2
, fr 1
, f r of grid function f (x) are given.
A quadratic spline (or second-degree spline, Fig. 5.1) is function P(x) ,
which satisfies the following conditions:
1) on each elementary segment xi 1 ≤ x ≤ xi (k + 1 ≤ i ≤ r), the spline
coincides with a second-degree polynomial (generally, the polynomials are
different on different elementary segments);
2) at the grid nodes, the spline has the grid function values: P ( xi ) fi ;
3) the spline has a continuous derivative, i.e., the spline is smooth;
4) on the left boundary of segment [a, b] , the spline satisfies an additional
condition (the boundary condition may be formulated on the right boundary).

Fig. 5.1. The quadratic spline graph

353
 Chapter 5. Quadratic and Linear Splines

According to item (2), the P(x) graph passes through points ( x k , f k ),

( xk 1
, fk 1
), ( xk 2
, fk 2
), . . . , ( xr 2
, fr 2
), ( xr 1
, fr 1
), ( x r , f r ).
According to item (3), the jumps of derivative P (x) are absent at the interior
grid nodes, i.e., at points xk 1
, xk 2
, ..., xr 2
, xr 1
.
The values of P (x) at the grid nodes are called slopes of the quadratic
spline: P ( xi ) Qi (k ≤ i ≤ r).
Let the spline slopes, Qi (k ≤ i ≤ r), be given. In this case, items (1) and (3)
of the quadratic spline definition lead to the following expression for the deriva-
tive on elementary segment [ xi 1 , xi ] :
xi x x xi 1
P ( x) Qi 1
Qi , (5.1)
hi hi
where hi xi xi 1
is the elementary segment’s length or the grid step, k + 1 ≤
i ≤ r.
By integrating (5.1), we obtain the following expression:
( xi x) 2 (x xi 1
)2
P( x) Qi 1
Qi C, (5.2)
2hi 2hi
where C is the integration constant.
Assuming that slope Qi 1 is known, we will determine integration constant
C and slope Qi . The resulting expression for C will be substituted into expres-
sion (5.2) for P(x) .
According to item (2) of the quadratic spline definition, we have the follow-
ing equalities:
P ( xi 1 ) fi 1 ,
P ( xi ) fi .
Using expression (5.2), we get the system of two linear algebraic equations with
unknown C and Qi :
hi
C fi 1
Qi 1
, (5.3)
2
hi
Qi C fi . (5.4)
2

354
 5.1. Definition of quadratic spline. Spline slopes

Substituting expression (5.3) into equation (5.4), we have

f fi 1
Qi 2 i Qi 1 . (5.5)
hi
By substituting expression (5.3) into (5.2), we obtain the following expres-
sion for the spline on segment [ xi 1 , xi ] :

hi ( xi x) 2 (x xi 1
)2
P( x) Qi 1
Qi fi 1
. (5.6)
2 2hi 2hi

According to item (4) of the quadratic spline definition, we define the slope
on the left boundary of segment [a, b] as follows:
Qk f (a ) ,
where f (a) is a given value of the function derivative on the left boundary.
Using recurrence formula (5.5) successively at i = k + 1, k + 2, ..., r, we can
calculate all slopes of the quadratic spline. Further, we can:
interpolate grid function f (x) by means of formula (5.6);
calculate the f (x) value at arbitrary point x of segment [a, b] by using
formula (5.1).
The error of interpolating the f (x) function (and its derivatives) by the
P(x) spline (and by its derivatives) is determined by the following expression:

f ( n ) ( x) P ( n ) ( x) 3 n
O(hmax ),
where hmax max {hi } is the maximum grid step ( hmax 0 ), n = 0, 1, 2
k 1 i r

is the derivative order, f (0) ( x) f ( x) , P( 0) ( x) P( x) . The sense of the used

O notation is explained in Section 3.1.
The above mathematical construction of quadratic spline will be used for
solving the initial value problem for the system of differential equations.

355
 Chapter 5. Quadratic and Linear Splines

5.2. Method for solving the initial value problem

for the system of differential equations

Let the system of two differential equations

du
E ( x, u , v ) , (5.7)
dx
dv
F ( x, u , v ) (5.8)
dx
be given on segment [a, b] . Generally, E and F are nonlinear functions of three
variables.
Let the values of unknown functions u (x) and v (x ) be given on the left
boundary:
u (a) A ,
v( a ) Q,
where A and Q are parameters.
The conditions on the left boundary are called the initial conditions.
For solving this initial value (Cauchy) problem on segment [a, b] , the grid is
constructed (Fig. 5.1) and the u (x) and v (x ) functions are considered as quad-
ratic splines. In this case, according to equations (5.7) and (5.8), the values of
functions E and F at the grid nodes are the slopes of these splines.
With regard to splines u (x) and v (x ) , expression (5.5) for the slope takes
the following forms:
u ui 1
E ( xi , ui , vi ) 2 i E ( xi 1 , ui 1 , vi 1 ) ,
hi
vi vi 1
F ( xi , ui , vi ) 2 F ( xi 1
, ui 1
, vi 1
)
hi
or
hi
ui E ( xi , ui , vi ) , (5.9)
2

356
 5.2. Method for solving the initial value problem for the system of differential equations

hi
vi F ( xi , ui , vi ) , (5.10)
2
where
hi
ui 1
E ( xi 1
, ui 1
, vi 1
),
2
hi
vi F ( xi 1 , ui 1 , vi 1 ) .
1
2
Equations (5.9) and (5.10) form the system of algebraic equations with
respect to ui and vi (at known ui 1 and vi 1 ).
If u k u (a) A and vk v(a) Q are given, unknown functions u (x) and
v (x ) can be calculated at all grid nodes on segment [a, b] by solving the system
of algebraic equations (5.9) and (5.10) successively at i = k + 1, k + 2, ..., r.
As an example, we will solve the system of differential equations
du
v, (5.11)
dx
dv v2
(5.12)
dx 2 u
on segment [0, 1] . The initial conditions look like
u (0) A,
v(0) Q.
This initial value problem was solved in the fifth chapter of book [5] while
modeling a catalytic converter.
The catalytic converter is a device intended for lowering toxicity of waste
gases in the car exhaust system.
Let us write system (5.11), (5.12) in form (5.7), (5.8), where
E v,

v2
F .
2 u
We see that E is a linear function of variable v, F is a nonlinear function of
variables u and v.
Algebraic equations (5.9) and (5.10) take the following form:

357
 Chapter 5. Quadratic and Linear Splines

hi
ui vi , (5.13)
2
hi vi2
vi , (5.14)
2 2 ui
where
hi
ui 1
vi 1
,
2
hi vi2 1
vi 1
.
2 2 ui 1

Let us solve the system of algebraic equations (5.13) and (5.14) for unknown
ui and vi .
Equation (5.13) can be written as
hi
ui
v . (5.15)
2 i
Multiplying both sides of equation (5.14) by 2 ui and substituting expression
(5.15) for ui , we obtain
(2 )
vi . (5.16)
hi
2
2
Formulas (5.16) and (5.15) allow us to calculate the values of vi and ui suc-
cessively at i = k + 1, k + 2, ..., r.

358
 5.3. Program for solving the initial value problem

5.3. Program for solving the initial value problem

Let us develop a program for solving the initial value problem for the system
of differential equations (5.11) and (5.12). Number l = r – k of steps on segment
[0, 1] and the values of A and Q are given in the table below.
l 10
A 0.25
Q -0.1

As a result of the program execution, dependences u (x) and v (x ) should appear

under the original table.
The program has the following form:
Listing 5.1
Sub main()
Dim l As Integer
Dim u As Double, v As Double
Dim h As Double, h2 As Double
Dim i As Integer
Dim alpha As Double, beta As Double
l = Selection.Cells(1, 2)
u = Selection.Cells(2, 2) 'value of A
v = Selection.Cells(3, 2) 'value of Q
h = 1 / l: h2 = h / 2
Selection.Cells(4, 1) = "x"
Selection.Cells(4, 2) = "u"
Selection.Cells(4, 3) = "v"
Selection.Cells(5, 1) = 0
Selection.Cells(5, 2) = u
Selection.Cells(5, 3) = v
For i = 6 To 5 + l 'movement along axis x
Selection.Cells(i, 1) = (i - 5) * h
alpha = u + h2 * v
beta = v - h2 * v ^ 2 / (2 - u)
v = (2 - alpha) * beta / (2 - alpha + _

359
 Chapter 5. Quadratic and Linear Splines

h2 * beta)
u = alpha + h2 * v
Selection.Cells(i, 2) = u
Selection.Cells(i, 3) = v
Next i
End Sub

The program uses the values of the above table (Fig. 5.2). We must select this
Excel table before the program execution. The coordinates of the grid nodes and
the corresponding values of the u (x) and v (x ) dependences are placed in col-
umns x, u, v (Fig. 5.3).

Fig. 5.2. The Excel table with the source data

Fig. 5.3. The program execution results

360
 5.3. Program for solving the initial value problem

We advise the reader to write a program (similar to Listing 5.1) for solving
the Bernoulli differential equation [3],
du
f ( x) u 2 ,
dx
on segment [a, b] under initial condition u(a ) 1 . In this equation, f (x ) is
a function from Appendix 4; segment [a, b] is this function’s domain.

361
 Chapter 5. Quadratic and Linear Splines

5.4. Solving the system of nonlinear algebraic

equations by the Newton method

In Section 5.6, we will need to solve the system of nonlinear algebraic equa-
tions
f1 ( x1 , x2 , ..., xn ) 1
,
f 2 ( x1 , x2 , ..., xn ) 2
, (5.17)
. . . . . .
f n ( x1 , x2 , ..., xn ) n
,

where functions f1 , f 2 , ..., f n are twice differentiable with respect to argu-

ments x1 , x 2 , ..., x n , right-hand sides 1
, 2
, ..., n
are given constants (we
can consider them equal to zero). The solution process is iterative.
To understand the iteration content, we write solution ( x1 , x2 , ..., xn ) x
of system (5.17) in form
x1 x1j z1 , x2 x2j z2 , ..., xn xnj zn ,

where ( x1j , x2j , ..., xnj ) x j is the j-th approximation of the x solution, z1 ,
z 2 , ..., z n are small quantities.

By substituting expressions for x1 , x 2 , ..., xn into (5.17), we obtain the fol-

lowing system of nonlinear algebraic equations with respect to z1 , z 2 , ..., z n :

f1 ( x1j z1 , x2j z2 , ..., xnj zn ) 1

,

f2 ( x1j z1 , x2j z2 , ..., xnj zn ) 2

,
. . . . . . . . . .
fn ( x1j z1 , x2j z2 , ..., xnj zn ) n
.

362
 5.4. Solving the system of nonlinear algebraic equations by the Newton method

Using Taylor’s formula [3] for functions of n variables in a neighborhood of

point x j , we have
f i ( x1j z1 , x2j z 2 , ..., xnj zn ) f i ( x1j , x2j , ..., xnj )
1
z z2 ... zn f i ( x1j , x2j , ..., xnj ) O( z max
2
),
1! 1 x1 x2 xn

1 ≤ i ≤ n. At that, the last algebraic equations with respect to z1 , z 2 , ..., z n

take the following linear form:
f1 f1 f1
z1 (x j ) z2 ( x j ) ... z n (x j ) 1
f1 ( x j ) O ( z max
2
),
x1 x2 xn
f2 f2 f2
z1 (x j ) z2 ( x j ) ... zn (x j ) 2
f2 (x j ) 2
O ( zmax ),
x1 x2 xn
. . . . . . . . . . . . . . . . . . . .
fn f f
z1 ( x j ) z2 n ( x j ) ... zn n ( x j ) n
fn (x j ) 2
O ( zmax ),
x1 x2 xn
where z max is the maximum of small quantities | z1 | , | z 2 | , ..., | z n | ,

zmax max {| zi |} .
1 i n

In the Newton method, the ( j 1 )th approximation of the solution ( j = 0, 1,

2, ...) is calculated over the known j-th approximation as follows:
1) matrix

f1 f1 f1
(x j ) ( x j ) ... (x j )
x1 x2 xn
a11 a12 ... a1n
f2 f2 f2
a 21 a 22 ... a 2 n (x j ) ( x j ) ... (x j )
x1 x2 xn (5.18)
. . . .
. . . .
a n1 an2 ... a nn
fn j
fn j
fn
(x ) ( x ) ... (x j )
x1 x2 xn

of the partial derivatives of functions f1 , f 2 , ..., f n is calculated;

363
 Chapter 5. Quadratic and Linear Splines

2) vector

b1 1
f1 ( x j )
b2 f 2 (x j )
2 (5.19)
... . . . .
bn
n
f n (x j )
is calculated;
3) the system of linear algebraic equations
a11 z1 a12 z 2 ... a1n z n b1 ,
a 21 z1 a 22 z 2 ... a 2n z n b2 ,
(5.20)
. . . . . . . . .
a n1 z1 an2 z 2 ... a nn z n bn

is solved, for example, by the Gaussian elimination method considered in

Sections 3.9 and 3.10;
4) the ( j 1 )th approximation of the solution is calculated according to for-
mulas
x1j 1
x1j z1 ,

x2j 1 x2j z2 , (5.21)

. . . . .
xnj 1
xnj zn .

Various conditions can be applied for finishing the iterative process; we will
use the following:
zmax , (5.22)

where is a given positive constant,

zmax max {| zi |} .
1 i n

Matrix (5.18) of the partial derivatives of functions f1 , f 2 , ..., f n is called

the Jacobian matrix of these functions.

364
 5.5. Newton and Newton-like methods for solving the single nonlinear algebraic equation

5.5. Newton and Newton-like methods for solving

the single nonlinear algebraic equation

It is obvious that the Newton iterative process of the previous section can be
used for solving not only the system of nonlinear algebraic equations, but also
for solving one equation, which can be considered as a system containing one
equation with respect to x1 :
f1 (x1 ) 1
.
In this case, the Newton method has a simple geometric interpretation.
The geometric interpretation will be considered on an example of equation
(4.23),
f (x) 0 ,
satisfying the following conditions in the f (x) function’s domain:
the f (x) function is continuous and monotonous;
the derivative, f (x) , is continuous, i.e., the f (x) function is smooth;
f (x) is different from 0;
the equation solution, x , exists.
In addition, we suppose that the solution’s initial approximation ( x 0 , which must
be given) is located in the f (x) function’s domain.
The Newton iterative process, defined by items (1) — (4) of the previous sec-
tion, can be written as follows:
f (x j )
xj 1
xj , (5.23)
f (x j )
where x j , x j 1
are the j-th and ( j 1 )th approximations of the equation solu-
tion, j = 0, 1, 2, ... Initial approximation x 0 is given.
Fig. 5.4 gives the geometric interpretation of the process of solving equation
f (x) 0 . According to this figure, the ( j 1 )th approximation of the x solu-
tion can be determined as follows:
1) restore the perpendicular to the x axis from point x j ;

365
 Chapter 5. Quadratic and Linear Splines

2) run the tangent line through the point of intersection of the perpendicular
with the f (x) graph;
3) consider the coordinate of the point of intersection of the tangent line with
the x axis as the ( j 1 )th approximation of the equation solution, x j 1 .

Fig. 5.4. The geometric interpretation of the Newton method

This algorithm follows from formula (5.23) and the following property of the
tangent line passing through the point with coordinates x j and f ( x j ) : the slope
of this tangent line is equal to
f (x j )
f (x j ) .
xj xj 1
Due to the above geometric interpretation of the Newton method, it is also
called the tangent method.
In Section 4.5, equation (4.23), (4.24) of form x cos x 1.5 0 was solved
by using the Solver add-in for Excel. Below is a code for solving this equation
by the tangent method.
Listing 5.2
Sub main()
Dim j_lim As Integer
Dim zeta As Double
Dim x As Double

366
 5.5. Newton and Newton-like methods for solving the single nonlinear algebraic equation

Dim f As Double
Dim a As Double
Dim j As Integer
Dim z As Double
1: j_lim = Range("I1").Value
2: zeta = Range("H1").Value
3: x = Range("G1").Value
For j = 1 To j_lim 'Newton iterations
4: Call f_function(x, f)
5: Call fx_jacobian(x, a)
6: z = -f / a
7: x = x + z
8: If Abs(z) < zeta Then Exit For
Next j
9: Range("G2").Value = x
10: Range("F2").Value = f
11: Range("E2").Value = a
12: Range("I2").Value = j
End Sub

Sub f_function(ByVal x As Double, _

ByRef f As Double)
f = x - Cos(x) - 1.5
End Sub

Sub fx_jacobian(ByVal x As Double, _

ByRef fx As Double)
fx = 1 + Sin(x)
End Sub

In this code:
1) j_lim is the limiting number of iterations; its value is taken from cell I1
of the active worksheet (operator 1);
2) zeta is constant in condition (5.22) of finishing the Newton iterative
process; its value is taken from cell H1 (operator 2);
3) x is the current approximation of the equation solution; the initial value of
x (the initial approximation of the solution) is taken from cell G1 (operator 3).
After inputting the values of j_lim, zeta and x, the j cycle of the Newton
iterations is performed. In this cycle:
1) operator 4 calculates value f of the function according to formula
f ( x) x cos x 1.5 by calling the f_function subroutine;

367
 Chapter 5. Quadratic and Linear Splines

2) operator 5 calculates the derivative’s value, a, according to formula

f ( x) 1 sin x by calling the fx_jacobian subroutine;
3) operators 6 and 7 calculate the next approximation of the equation solu-
tion according to formula (5.23);
4) operator 8 checks condition (5.22), | z | , for finishing the Newton
iterations.
When finishing the iterative process (after leaving the j cycle), the calculated
values of x, f and a are respectively put into cells G2, F2 and E2 (operators 9,
10 and 11). Operator 12 puts the number of the Newton iterations into cell I2.
Fig. 5.5 shows the Excel worksheet upon termination of the code execution.
Cell G2 contains the following value of the solution of equation (4.23), (4.24):
x = 1.535394. It practically coincides with the x value calculated by means of
Excel in Section 4.5.

Fig. 5.5. The worksheet upon termination of solving the equation

Let us consider the question of the convergence of the Newton iterative pro-
cess defined by formula (5.23). For that, we use the following designation (simi-
lar to the designations on p. 311): x j x j
, where x is the exact solution
j
of the equation, i.e., f (x ) 0 . Let be a small quantity.
Substituting expressions x j x j
and x j 1
x j 1
into formula
(5.23), we have
f (x j)
j 1 j
.
f (x j)

j j
Let us use expressions similar to (3.1) for f ( x ) and f ( x ) and

take into account f (x ) 0 . At that, we have the following chain of equalities:

f (x ) j f ( x ) 0.5( j )2 f ( x ) O [( j )3 ]
j 1 j

f (x ) j f ( x ) O [( j )2 ]

368
 5.5. Newton and Newton-like methods for solving the single nonlinear algebraic equation

j f ( x ) 0.5( j )2 f ( x ) O [( j )3 ]
j

f (x ) j f ( x ) O [( j )2 ]

0.5 j f (x ) O [( j )2 ] j)
j j 2 0.5 f ( x ) O(
( ) .
f (x ) j f (x ) O [( j )2 ] j)
f (x ) O (
j
Neglecting O ( ) in the numerator and denominator, we have the following
formula:

j 1 f (x ) j 2
0.5 ( ) . (5.24)
f (x )

Because f (x ) 0 , the last formula shows that the Newton iterative process
has quadratic convergence.
The secant method (Section 4.5), defined by formula (4.28) in form
(x j xj 1) f (x j )
xj 1
xj , (5.25)
f (x j ) f (x j 1)

can be considered as a Newton-like method. The reason is that formula (5.25)

takes form (5.23) of the Newton (tangent) method at x j xj 1
0 because
f (x j ) f (x j 1)
f (x j ) .
xj xj 1
As we see, the secant method in form (5.25) needs two initial approximations
of the equation solution, x 0 and x1 .
Fig. 5.6 gives the geometric interpretation of the secant method. According to
this figure, the ( j 1 )th approximation of the x solution over the ( j 1 )th and
j-th approximations can be calculated as follows:
1) restore the perpendiculars to the x axis from points x j 1 and x j of the
x axis;
2) run the secant line through points ( x j 1
, fu ) and ( x j , fv ) of intersection
of the perpendiculars with the f (x) graph;
3) consider the coordinate of the point of intersection of the secant line with
the x axis as the ( j 1 )th approximation of the equation solution, x j 1 .
Formula, intended for estimating the rate of the convergence of the secant
method’s iterative process, is close to formula (5.24):

369
 Chapter 5. Quadratic and Linear Splines

j 1 f (x ) j j 1
0.5 . (5.26)
f (x )
The derivation of this formula is similar to the above derivation of (5.24). For-
mula (4.30) is the simplified form of (5.26).

Fig. 5.6. The geometric interpretation of the secant method:

fu f (x j 1
), fv f (x j )

In Section 4.5, we considered another version of the secant method, which

has slow (linear) convergence according to formula (4.29). However, f (x ) 0
is not the necessary condition for the convergence.
The Steffensen method, defined by formula
f 2 (x j )
xj 1
xj ,
f [x j f ( x j )] f (x j )
can also be considered as a Newton-like method. The reason is that the last for-
mula takes form (5.23) of the Newton method because
f [x j f ( x j )] f (x j ) f (x j ) f (x j )

at f ( x j ) 0.
Unlike the previous Newton-like method, the Steffensen method needs only
one initial approximation of the equation solution, x 0 .

370
 5.5. Newton and Newton-like methods for solving the single nonlinear algebraic equation

Fig. 5.7 shows the possibility of cycling the Newton iterative process without
convergence to solution x if not all conditions, formulated at the beginning of
this section, are satisfied (set j = 0 in Fig. 5.7). If the Newton method is used for
solving the system of nonlinear algebraic equations (5.17), the cycling is also
possible.

Fig. 5.7. Cycling the iterative process of the Newton (tangent) method

The Newton method, described in the previous section, may be considered as

the tangent method generalization to the system of nonlinear algebraic equations
(5.17). The secant and Steffensen methods also have the generalizations.
In Appendix 6, the Excel circular reference is used for realization of the tan-
gent method for solving nonlinear equation x cos x 1.5 0 . In a similar way,
we can realize the bisection, secant and Steffensen methods in Excel without
programming in VBA.
We advise the reader to write a program for solving equation f (x) 0 on
segment [a, b] by the Steffensen method. In this equation, f (x ) is a function
from Appendix 4; segment [a, b] is this function’s domain. The user-defined
form (as the user interface of the program) must include the CheckBox element
for choice of the solution’s initial approximation:
x 0 a in the absence of the check mark;
x0 b in the presence of the check mark.
If point x j f ( x j ) appears outside segment [a, b] during the solution, the

secant method with x j 1

0.5(a b) must be used for continuation of solving
equation f (x) 0 .

371
 Chapter 5. Quadratic and Linear Splines

5.6. Modeling of the piano mechanism

linking a key with hammer

In Section 5.2, we considered a method for solving the initial value problem
for the system of two first-order differential equations, (5.7) and (5.8). However,
the solution method does not change if the number of the system equations is
greater than two, as in the initial value problem below.
Article [12] considers a series of mathematical models of the piano mecha-
nism linking a key with hammer. We will use the model of medium complexity,
in which the key and hammer motion is described by the following system of
second-order differential equations:
d 2 u1
E2 (u1 , u3 ) , (5.27)
dt 2
d 2u3 du3
E 4 u1 , u 3 , , (5.28)
dt 2 dt
where
f k (u1 u3 )
E2 (u1 , u3 ) , (5.29)
m1
2
du3
q k (u1 u3 )
du3 dt
E 4 u1 , u 3 , . (5.30)
dt p qu3
In this model, u1 is the key’s displacement downward, u 3 is the hammer’s
displacement forward, f is the force acting on the key, k is the elastic constant
of the spring, m1 is the key’s mass, p and q are given constants, p qu3 m3 is
the hammer’s mass depending on its displacement (more exactly, m 3 is the
hammer’s effective mass).
We have to develop a code for solving this system of differential equations
on the time segment, 0 ≤ t ≤ b, with the following zero initial conditions:

372
 5.6. Modeling of the piano mechanism linking a key with hammer

du1 du3
u1 (0) ( 0) u3 (0) ( 0) 0.
dt dt
Let us introduce the key and hammer velocities, u 2 and u 4 respectively, as
follows:
du1
u2 ,
dt
du3
u4 .
dt
The system of equations (5.27) and (5.28) takes the following form:
du1
E1 (u2 ) ,
dt
du 2
E2 (u1 , u3 ) ,
dt
du3
E3 ( u4 ) , (5.31)
dt
du4
E 4 (u1 , u 3 , u 4 ) ,
dt
where
E1 u2 ,
f k (u1 u3 )
E2 ,
m1
E3 u4 , (5.32)

qu 42 k ( u1 u3 )
E4 .
p qu3
The initial conditions become
u1 (0) u2 (0) u3 ( 0 ) u 4 ( 0) 0. (5.33)
As in Section 5.2, we construct the grid on segment 0 ≤ t ≤ b for solving the
formulated initial value problem and consider functions u1 (t ) , u 2 (t ) , u 3 (t ) ,
u 4 (t ) at the grid nodes. The grid step is assumed constant: b / l , where l is
a given number of time steps.

373
 Chapter 5. Quadratic and Linear Splines

Let t i be the coordinate of a node, which is not the initial time moment,
i.e., 1 ≤ i ≤ l. To calculate the values of u1 (t ) , u 2 (t ) , u 3 (t ) , u 4 (t ) over known
values of u1 (t ) , u 2 (t ) , u 3 (t ) , u 4 (t ) , we must solve the following
system of nonlinear algebraic equations:

u1 (t ) E1[u2 (t )] u1 (t ) E1[u2 (t )] ,
2 2

u 2 (t ) E2 [u1 (t ), u3 (t )] u2 (t ) E2 [u1 (t ), u3 (t )] ,
2 2

u3 (t ) E3[u4 (t )] u3 (t ) E3[u4 (t )] ,
2 2

u4 (t ) E4 [u1 (t ), u3 (t ), u4 (t )]
2

u4 (t ) E4 [u1 (t ), u3 (t ), u4 (t )] .
2
After designations

u1 (t ) E1[u2 (t )] 1
,
2

u 2 (t ) E2 [u1 (t ), u3 (t )] 2
,
2

u3 (t ) E3 [u4 (t )] 3
, (5.34)
2

u 4 (t ) E4 [u1 (t ), u3 (t ), u4 (t )] 4
,
2
we can write this system of nonlinear algebraic equations in form (5.17) at n = 4:
f1 ( x1 , x2 , x3 , x4 ) 1
,
f 2 ( x1 , x2 , x3 , x4 ) 2
,
f 3 ( x1 , x2 , x3 , x4 ) 3
, (5.35)
f 4 ( x1 , x2 , x3 , x4 ) 4
,
where x1 u1 (t ) , x2 u 2 (t ) , x3 u3 (t ) , x4 u 4 (t ) .
According to expressions (5.32), we have

374
 5.6. Modeling of the piano mechanism linking a key with hammer

f1 x1 E1 ( x2 ) x1 x2 ,
2 2
f k ( x1 x3 )
f2 x2 E2 ( x1 , x3 ) x2 ,
2 2 m1

f3 x3 E3 ( x4 ) x3 x4 , (5.36)
2 2
qx42 k ( x1 x3 )
f4 x4 E4 ( x1 , x3 , x4 ) . x4
2 2 p qx3
For solving the system of nonlinear algebraic equations (5.35) by the Newton
method, we need expressions for the partial derivatives of functions f1 , f 2 , f 3 ,
f 4 with respect to arguments x1 , x 2 , x 3 , x 4 , i.e., for the elements of Jacobian
matrix (5.18). By using the basic rules of differentiation [3], we obtain
f1 f1 f1 f1
1, , 0,
x1 x2 2 x3 x4
f2 k f2 f2 k f2
, 1, , 0,
x1 2m1 x2 x3 2m1 x4
f3 f3 f3 f3
0, 1, , (5.37)
x1 x2 x3 x4 2
f4 k f4
, 0,
x1 2 p qx3 x2

f4 k( p qx3 ) [ qx42 k ( x1 x3 )] q f4 qx 4
. , 1
x3 2 x4 p qx3
( p qx3 ) 2
Let us develop a code for solving the system of differential equations (5.31)
with initial conditions (5.33). In table Listing 5.3 with the source data for the
required code:
l is the number of time steps;
τ is the time step in seconds;
f is the force acting on the key, in newtons;
m1 is the key’s mass in kilograms;

375
 Chapter 5. Quadratic and Linear Splines

p, q, k are constants in expressions (5.29) and (5.30); their dimensions are

as follows: [p] = kg, [q] = kg / m, [k] = N / m;
ζ is a positive constant in condition (5.22) for finishing the Newton itera-
tive process.
Listing 5.3
l 30
tau 1.00E-03
f 10
m1 0.074
p 0.406
q 18.3
k 1.16E+04
zeta 1.00E-09

As results of the execution, we must have the values of time t and the corre-
sponding values of the key’s and hammer’s displacements and velocities, i.e., we
du1 du3
must have u1 , u2 , u 3 , u4 as functions of time t.
dt dt
Below are the main program and the e_functions and fx_jacobian
subroutines for solving the initial value problem.
Listing 5.4
Dim tau As Double
Dim f As Double, m1 As Double
Dim p As Double, q As Double
Dim k As Double
Dim tau2 As Double

Sub main()
Dim l As Integer
Dim zeta As Double
Dim u(1 To 4) As Double
Dim x(1 To 4) As Double
Dim z(1 To 4) As Double
Dim e(1 To 4) As Double
Dim a(1 To 4, 1 To 4) As Double
Dim b(1 To 4) As Double
Dim alpha(1 To 4) As Double
Dim m As Integer, i As Integer, j As Integer
Dim max As Double

376
 5.6. Modeling of the piano mechanism linking a key with hammer

Dim sb As String, se As String

l = Selection.Cells(1, 2)
tau = Selection.Cells(2, 2)
f = Selection.Cells(3, 2)
m1 = Selection.Cells(4, 2)
p = Selection.Cells(5, 2)
q = Selection.Cells(6, 2)
k = Selection.Cells(7, 2)
zeta = Selection.Cells(8, 2)
tau2 = tau / 2
For m = 1 To 4
1: u(m) = 0 'values at t = 0
Next m
Selection.Cells(9, 1) = "t"
Selection.Cells(9, 2) = "u1"
Selection.Cells(9, 3) = "u2"
Selection.Cells(9, 4) = "u3"
Selection.Cells(9, 5) = "t"
Selection.Cells(9, 6) = "u4"
Selection.Cells(9, 7) = "j max"
Selection.Cells(10, 1) = 0
Selection.Cells(10, 2) = u(1)
Selection.Cells(10, 3) = u(2)
Selection.Cells(10, 4) = u(3)
Selection.Cells(10, 5) = 0
Selection.Cells(10, 6) = u(4)
For i = 11 To 10 + l 'movement along time axis
2: Call e_functions(u, e)
For m = 1 To 4
3: alpha(m) = u(m) + tau2 * e(m)
4: x(m) = u(m) + tau * e(m)
Next m
For j = 1 To 1000 'Newton iterations
5: Call fx_jacobian(x, a)
6: Call e_functions(x, e)
For m = 1 To 4
7: b(m) = alpha(m) - (x(m) - tau2 * e(m))
Next m
8: Call gauss(4, a, b, z)
For m = 1 To 4
9: x(m) = x(m) + z(m)
Next m

377
 Chapter 5. Quadratic and Linear Splines

max = 0
For m = 1 To 4
If Abs(z(m)) > max Then _
max = Abs(z(m))
Next m
10: If max < zeta Then Exit For
Next j
For m = 1 To 4
11: u(m) = x(m)
Next m
Selection.Cells(i, 1) = (i - 10) * tau
Selection.Cells(i, 2) = u(1)
Selection.Cells(i, 3) = u(2)
Selection.Cells(i, 4) = u(3)
Selection.Cells(i, 5) = (i - 10) * tau
Selection.Cells(i, 6) = u(4)
Selection.Cells(i, 7) = j
Next i
sb = Selection.Cells(10, 1).Address
se = Selection.Cells(10 + l, 2).Address
12: Call graph(sb, se, "t, s", "u1, m")
sb = Selection.Cells(10, 5).Address
se = Selection.Cells(10 + l, 6).Address
13: Call graph(sb, se, "t, s", "u4, m/s")
Range("O36").Select
End Sub

Sub e_functions(ByRef x() As Double, _

ByRef e() As Double)
e(1) = x(2)
e(2) = (f - k * (x(1) - x(3))) / m1
e(3) = x(4)
e(4) = (q * x(4) ^ 2 + k * (x(1) - x(3))) / _
(p - q * x(3))
End Sub

Sub fx_jacobian(ByRef x() As Double, _

ByRef fx() As Double)
Dim m3 As Double
fx(1, 1) = 1
fx(1, 2) = -tau2
fx(1, 3) = 0: fx(1, 4) = 0

378
 5.6. Modeling of the piano mechanism linking a key with hammer

fx(2, 1) = tau2 * k / m1
fx(2, 2) = 1
fx(2, 3) = -tau2 * k / m1
fx(2, 4) = 0
fx(3, 1) = 0: fx(3, 2) = 0
fx(3, 3) = 1
fx(3, 4) = -tau2
m3 = p - q * x(3)
fx(4, 1) = -tau2 * k / m3
fx(4, 2) = 0
fx(4, 3) = tau2 * (k * m3 - _
(q * x(4) ^ 2 + k * (x(1) - x(3))) * q) / m3 ^ 2
fx(4, 4) = 1 - tau * q * x(4) / m3
End Sub

The u array contains values u1 , u 2 , u 3 , u 4 of the solution. Operator 1 sets

the solution’s values at t 0 according to (5.33). The x array contains the solu-
tion approximations.
By means of the i cycle, the movement along the time axis is performed with
step τ. The u array, used in the call of the e_functions subroutine (opera-
tor 2), contains values u1 , u 2 , u 3 , u 4 of the solution at the t moment
of time. This subroutine calculates corresponding values E1 , E 2 , E 3 , E 4
of the e array according to formulas (5.32). Operator 3 calculates values
1
, 2
, 3
, 4
of the alpha array according to formulas (5.34). Operator 4
calculates the x array of the initial approximation of the solution at the t moment
according to formulas
x1 u1 (t ) E1[u 2 (t )] ,
x2 u 2 (t ) E2 [u1 (t ), u3 (t )] ,
x3 u 3 (t ) E3 [u 4 (t )] ,
x4 u 4 (t ) E4 [u1 (t ), u3 (t ), u4 (t )] .
Further (inside the i cycle being under consideration), the j cycle of the
Newton iterations is performed. In the j cycle:
1) by calling the fx_jacobian subroutine (operator 5), the a array is cal-
culated according to formulas (5.18) and (5.37);
2) after calling the e_functions subroutine (operator 6), the b array is
calculated according to formulas (5.19) and (5.36) by means of operator 7;

379
 Chapter 5. Quadratic and Linear Splines

3) by calling the gauss subroutine (operator 8), the system of linear alge-
braic equations (5.20), defined by arrays a and b, is solved;
4) by means of operator 9, the next approximation of the solution at the t
moment is calculated according to formulas (5.21);
5) operator 10 checks if condition (5.22) is satisfied for termination of the
Newton iterations.
When terminating the iterative process (after leaving the j cycle), values u1 ,
u 2 , u 3 , u 4 of the solution at the t moment (which are in the x array) are
assigned to the u array’s elements by means of operator 11. These values
(together with the value of t and the number of the Newton iterations, jmax ) are
put into cells of Excel.
At the end of the main program (after leaving the i cycle), the graphs of
calculated dependences u1 (t ) and u 4 (t ) are constructed automatically by means
of the graph subroutine. Operators 12 and 13 are the calls of this subroutine.
The declarations of e_functions and fx_jacobian are located below
the main program (see Listing 5.4).
The e_functions subroutine calculates values E1 , E 2 , E 3 , E 4 according
to formulas (5.32). The parameters of this subroutine have the following sense:
x is an array of values u1 , u 2 , u 3 , u 4 ;
e is a one-dimensional array intended for the required values.
The fx_jacobian subroutine calculates the Jacobian matrix of functions
f1 , f 2 , f 3 , f 4 according to formulas (5.18) and (5.37). The parameters have
the following sense:
x is an array of arguments x1 , x 2 , x 3 , x 4 ;
fx is a two-dimensional array intended for the required values of the par-
tial derivatives.
The source data for code Listing 5.4 are the values given in table Listing 5.3
considered above. The corresponding Excel table, depicted in Fig. 5.8, must be
selected before the code execution.
Upon termination of the execution, the values of t, u1 , u 2 , u 3 , u 4 and jmax
are located in the corresponding columns on the Excel worksheet (Fig. 5.9).
Besides, the u1 (t ) and u 4 (t ) graphs are located on this worksheet; Fig. 5.10
shows them completely.
The graphs, depicted in Fig. 5.10, are results of the mentioned calls of the
graph subroutine.

380
5.6. Modeling of the piano mechanism linking a key with hammer

Fig. 5.8. The Excel table with the source data

Fig. 5.9. The code execution results

381
 Chapter 5. Quadratic and Linear Splines

Fig. 5.10. The automatically created graphs: a — u1 (t ) ; b — u 4 (t )

382
 5.7. Definition of linear spline

5.7. Definition of linear spline

The simplest spline, linear, is defined as follows.

Let values f k , f k 1 , f k 2 , ..., f r 2 , f r 1 , f r of grid function f (x) be
given at the nodes of grid x k < xk 1
< xk 2
< ... < xr 2
< xr 1
< xr .
A linear spline (or first-degree spline, Fig. 5.11) is function L(x) , which satis-
fies the following conditions:
1) on each elementary segment xi 1 ≤ x ≤ xi (k + 1 ≤ i ≤ r), the spline
coincides with a first-degree polynomial (generally, the polynomials are different
on different elementary segments);
2) at the grid nodes, the spline has the corresponding grid function values:
L ( xi ) fi .

Fig. 5.11. The linear spline graph

According to condition (2), the L(x) graph passes through points ( x k , f k ),

( xk 1
, fk 1
), ( xk 2
, fk 2
), . . . , ( xr 2
, fr 2
), ( xr 1
, fr 1
), ( x r , f r ).

383
 Chapter 5. Quadratic and Linear Splines

The definition of linear spline leads to the following expression for the L(x)
function on elementary segment [ xi 1
, xi ] :
xi x x xi 1
L( x) fi 1
fi , (5.38)
hi hi
where hi xi xi 1
is the elementary segment’s length or the grid step, k + 1 ≤
i ≤ r. Thus, the linear spline is the linearly interpolated tabular function.
Formula (5.38) was already used in programs Listing 4.11 and Listing 4.13:
see formula (4.34) on p. 320.
The error of interpolating the f (x) function (and its derivative) by the L(x)
spline (and by its derivative) is determined by the following expression:
f ( n ) ( x) L( n) ( x) 2 n
O(hmax ), (5.39)
where hmax max {hi } is the maximum grid step ( hmax 0 ), n = 0, 1 is
k 1 i r

the derivative order, f (0) ( x) f ( x) , L( 0) ( x) L( x) .

It is obvious that the mathematical constructions of the linear and cubic
splines are similar. However, the cubic interpolation is much more exact than the
linear interpolation: the interpolation errors differ by two orders of hmax . To
make sure, we have to compare expressions (4.19) and (5.39).
The linear spline can be used for interpolation, differentiation and integration
of the grid (tabular) function. For this purpose, we can use the si subroutine and
the ios function, developed for the cubic spline construction, but all elements of
the M array (containing the spline moments) must be zero in the calls of these
user-defined procedures.
When the moments are nullified, formula (4.20), realized in the ios func-
tion, takes the following form:
b b r fi fi
1
f ( x) dx L( x) dx hi . (5.40)
a a i k 1
2
fi 1
fi
It is obvious that summand hi is equal to the area of a trapezium with
2
height hi xi xi 1 and bases fi 1 and fi . Therefore, integration formula
(5.40) is widely known as the trapezoidal method.
Formula (5.39) gives the following estimation of the error of numerical inte-
gration by the trapezoidal method:

384
 5.7. Definition of linear spline

b b b r fi fi
1 2
f ( x) dx L( x) dx f ( x) dx hi O( hmax ),
a a a i k 1
2
where hmax 0 is the maximum grid step.
The linear spline can be used in the noniterative method for solving the non-
linear algebraic equation (instead of the cubic spline, Section 4.6). This leads to
a simplification of the program, but also to a deterioration in accuracy of the
equation solution.
We advise the reader to write a program for calculating the values of func-
tion
1
x
u ( x) 1 f ( y ) dy (5.41)
a

at the nodes of a uniform grid on segment [a, b] . In this formula, f (x ) is

a function from Appendix 4; segment [a, b] is the f (x ) function’s domain. The
integration must be performed by the user-defined ios function, as the program
realization of formula (5.40).
Function (5.41) is the analytical solution of the Bernoulli differential equa-
tion [3],
du
f ( x) u 2 ,
dx
on segment [a, b] under initial condition u(a ) 1 . On p. 361, we spoke about
the numerical solution of this initial value problem.
We advise the reader to write a program, similar to Listing 4.11, for solving
equation f (x) 0 on segment [a, b] by the noniterative method based on the
linear spline. In this equation, f (x ) is a function from Appendix 4; segment
[a, b] is this function’s domain. Uniform grid a x0 < x1 < x 2 < ... < xn 2
<
xn 1
< xn b must be used.
In the next section, the linear spline will be used in the least-squares method.
Besides, this mathematical construction appears in the sound insulation problem.

385
 Chapter 5. Quadratic and Linear Splines

5.8. The least-squares method

The linear spline can be used in the least-squares method. We will consider
this question on an example of the following table from Task 8.1 in book [13].

The source data

The wheat
The land plot The land quality productivity ( uj ),
number ( j ) mark ( xj )
centner per hectare
1 30 23.5
2 35 23.7
3 35 24.0
4 38 26.7
5 29 24.3
6 40 28.8
7 45 33.5
8 37 27.6
9 35 23.0
10 40 29.4
11 50 30.5
12 52 35.0

The above table gives the values characterizing the land quality and wheat
productivity for each of ν = 12 land plots. The least-squares method allows us to
establish the functional dependence of the wheat productivity, u, on the land
quality mark, x.
Let F (x) be an unknown functional dependence (of a given type) defined up
to parameters c1 , c 2 , ..., c n . More precisely, this dependence looks like
F (c1 , c2 , ..., cn , x ) , where c1 , c 2 , ..., c n and x are the function parameters and
argument, respectively.
According to the least-squares method under consideration, F (x) is the
required functional dependence of productivity u on mark x if the parameters
of F (x) are equal to the coordinates of the minimum point of function

386
 5.8. The least-squares method

G (c1 , c 2 , ..., c n ) [ F ( x j ) u j ]2 (5.42)

j 1

on c1 , c 2 , ..., c n .
Let us assume that:
1) F (x) is linear spline L(x) at a grid on the x axis;
2) the n number of the grid nodes and their location are given: z1 < ... <
z < ... < z n are the coordinates of the grid nodes on the x axis;
3) parameters c1 , c 2 , ..., c n of F (x) are values L1 , ..., L , ..., Ln of
the L(x) spline at grid nodes z1 , ..., z , ..., z n , respectively.
According to the least-squares method, we have to find the minimum point of
function

G ( L1 , ..., L , ..., Ln ) [ L( x j ) u j ] 2 (5.43)

j 1

on L1 , ..., L , ..., Ln .
In Section 6.12, we will minimize this function iteratively (by subroutines
mini and minim). Here, we will consider a noniterative method to find the
required minimum point, which is based on the concept of fundamental spline.
According to the necessary condition for an extreme value [3], the minimum
points of function (5.43) belong to the set of points that are solutions of the sys-
tem of equations
G
0,
L1
. . .
G
0, (5.44)
Li
. . .
G
0.
Ln
Therefore, for finding the minimum point of function (5.43), we have to solve
the last system and then to analyse the solution results.
The solution of system (5.44) begins with the following concept of funda-
mental spline.

387
 Chapter 5. Quadratic and Linear Splines

The fundamental spline, (x ) , is a spline equaling to 1 at the z node of

grid z1 < ... < z < ... < z n and 0 at all other n – 1 nodes. Because 1 ≤ κ ≤ n,
we have n fundamental splines 1
( x ) , ..., (x ) , ..., n
( x) .
If the fundamental splines are linear, then (according to the fundamental
spline definition) an arbitrary linear spline can be written as follows:
L( x ) L1 1 ( x ) ... L ( x ) ... Ln n ( x )
L1 1 ( x ) ... Li i
( x ) ... Ln n
( x) .
Using the last expression and formula (5.43), we have the following chain of
equalities:
G
0.5 {[ L( x j ) u j ] L( x j ) / Li }
Li j 1

{ i ( x j )[ L( x j ) u j ]}
j 1

{ i ( x j )[ L1 1 ( x j ) ... L ( x j ) ... Ln n
( x j ) u j ]}
j 1

L1 [ i ( x j ) 1 ( x j )] ... L [ i (x j ) ( x j )] ...
j 1 j 1

Ln [ i (x j ) n
( x j )] [ i ( x j )u j ]
j 1 j 1

L1 ( ij 1 j
) ... L ( ij j
) ... Ln ( ij nj
)
j 1 j 1 j 1

( ij
uj), (5.45)
j 1
where
ij i
(x j ) , (5.46)
1 ≤ i ≤ n, 1 ≤ j ≤ ν,
i.e., coefficients ij
form the A matrix of size n × ν:

388
 5.8. The least-squares method

11 12
... 1

21 22
... 2
A . (5.47)
. . . .
n1 n2
... n

Due to (5.45), equations (5.44) become linear algebraic equations

q11 L1 ... q1 L ... q1n Ln r1 ,
. . . . . . . . . . .
qi1 L1 ... qi L ... qin Ln ri , (5.48)
. . . . . . . . . . .
qn1 L1 ... qn L ... qnn Ln rn ,
where

qi ( ij j
) , ri ( ij
u j ). (5.49)
j 1 j 1
The system of linear algebraic equations (5.48) can be written as matrix
equation
QL R , (5.50)
where Q is the system matrix, R is the vector of the right-hand sides, L is the
vector of the unknown variables:

q11 ... q1 ... q1n r1 L1

. . . . . ... ...
Q qi1 ... qi ... qin , R ri , L L . (5.51)
. . . . . ... ...
qn1 ... qn ... qnn rn Ln

The system of linear algebraic equations (5.48) can be solved for unknown
L1 , ..., L , ..., Ln (the linear spline values at the grid nodes) by the Gaussian
elimination method, i.e., by means of subroutine gaus or gauss.

389
 Chapter 5. Quadratic and Linear Splines

5.9. Program to determine the dependence of

the wheat productivity on the land quality

Let us develop a program for computing values L1 , ..., L , ..., Ln of the

linear spline at the nodes of grid z1 < ... < z < ... < z n on the x axis, which is
based on the theoretical material of the previous section.
In the source data table given below:
n is the number of the grid nodes;
the Z column contains the x coordinates of the grid nodes;
the remaining two columns (Mark and Productivity) are the same as in
table “The source data” of the previous section.
Listing 5.5
n 3
Mark Productivity Z
30 23.5 25
35 23.7 45
35 24.0 55
38 26.7
29 24.3
40 28.8
45 33.5
37 27.6
35 23.0
40 29.4
50 30.5
52 35.0

The program for solving this task follows:

Listing 5.6
Sub main()
Dim X() As Double
Dim U() As Double
Dim Z() As Double

390
 5.9. Program to determine the dependence of the wheat productivity on the land quality

Dim lambda() As Double

Dim MOM() As Double
Dim A() As Double
Dim Q() As Double
Dim R() As Double
Dim L() As Double
Dim m As Integer
Dim n As Integer
Dim j As Integer
Dim i As Integer, k As Integer
m = Selection.Rows.Count 'quantity of rows
n = Selection.Cells(1, 2) 'number of nodes
ReDim X(3 To m)
ReDim U(3 To m)
ReDim Z(1 To n)
ReDim lambda(1 To n)
ReDim MOM(1 To n)
ReDim A(1 To n, 3 To m)
ReDim Q(1 To n, 1 To n)
ReDim R(1 To n)
ReDim L(1 To n)
For j = 3 To m
X(j) = Selection.Cells(j, 1)
U(j) = Selection.Cells(j, 2)
Next j
'Preparations for spline interpolation:
For i = 1 To n
Z(i) = Selection.Cells(2 + i, 3)
0: MOM(i) = 0
Next i
'Forming matrix A:
For i = 1 To n
For k = 1 To n 'forming i-th spline
If k = i Then
lambda(k) = 1
Else
lambda(k) = 0
End If
Next k
For j = 3 To m
'calculating values of i-th spline
1: Call si(1, n, Z, lambda, MOM, X(j), _

391
 Chapter 5. Quadratic and Linear Splines

A(i, j))
Next j
Next i
'Forming matrix Q and vector R:
For i = 1 To n
For k = 1 To n
Q(i, k) = 0
For j = 3 To m
Q(i, k) = Q(i, k) + A(i, j) * A(k, j)
Next j
Next k
R(i) = 0
For j = 3 To m
R(i) = R(i) + A(i, j) * U(j)
Next j
Next i
'Solving system of linear algebraic equations:
2: Call gauss(n, Q, R, L)
'Outputting results:
Selection.Cells(2, 4) = "L"
For i = 1 To n
Selection.Cells(2 + i, 4) = L(i)
Next i
End Sub

The A matrix (array A) is formed according to formulas (5.46) and (5.47) by

means of the si subroutine used for the interpolation (operator 1). Because the
fundamental splines, 1 ( x ) , ..., i (x ) , ..., n ( x ) , are linear (not cubic), the
MOM array of the moments is nulled by operator 0.
Matrix Q and vector R are formed according to formulas (5.49) and (5.51).
Matrix equation (5.50) is solved by calling the gauss subroutine (operator 2).
We can use the simpler subroutine by replacing the subroutine name with gaus
in operator 2.
The remaining operators of the program put the L solution of matrix equation
(5.50) into cells of Excel.
The source data for program Listing 5.6 are given in table Listing 5.5
(Fig. 5.12). Before the program execution, we must select this Excel table (range
B2:D15).
The execution results are the linear spline values at the grid nodes, which
are in the L column located near the Z column (Fig. 5.13). Fig. 5.14 shows the
experimental points and the calculated line.

392
5.9. Program to determine the dependence of the wheat productivity on the land quality

Fig. 5.12. The Excel table with the source data

Fig. 5.13. The program execution results

393
 Chapter 5. Quadratic and Linear Splines

Fig. 5.14. Experimental points and the fractured line of functional

dependence of the wheat productivity (in units of centner per hectare)
on the land quality mark

According to Fig. 5.14, we really determined the minimum point of function

(5.43).
The value of spline L(x) at point x, located between nodes zi 1 and z i
(i = 2, 3, ..., n – 1, n), can be calculated by means of the Excel formula corre-
sponding to mathematical formula (5.38) of the following form:
( z i x ) Li 1 ( x z i 1 ) Li
L( x)
zi zi 1
If x z1 , then
(z2 x ) L1 (x z1 ) L 2
L( x) .
z2 z1
If x z n , then
(zn x ) Ln 1
(x zn 1
) Ln
L( x) .
zn zn 1
Note that the cubic spline can be used instead of the linear spline in the least-
squares method. In this case, naturally, the fundamental splines are cubic.
The functional dependence of the wheat productivity on the land quality
mark, determined by the least-squares method, describes the dependence of pro-
duction results on factors. Such dependence is called the production function; it
is the basis for many economic calculations.

394
 5.10. The forward and backward Fourier transforms of a periodic function

5.10. The forward and backward Fourier

transforms of a periodic function

Let f (t ) be a real-valued periodic function of time t with given period T,

which describes an oscillation with frequency 1 / T . This function can be repre-
sented as the Fourier series [3]:
a0
f (t ) [ a k cos( k t ) bk sin( k t )] , (5.52)
2 k 1
where 2 / T is the cyclic frequency, a 0 , a k , bk are the Fourier coeffi-
cients:
2T
ak f (t ) cos( k t ) dt , (5.53)
T 0
k = 0, 1, 2, 3, ...,
2T
bk f (t ) sin( k t ) dt , (5.54)
T 0
k = 1, 2, 3, ...
Let us recall several terms related to periodic function
f k (t ) Ak sin( k t k
), (5.55)
where k is a natural number.
The oscillation, described by this formula, is called a harmonic oscillation
with cyclic frequency k . Quantities Ak , k t k
, k are respectively the
amplitude, phase and initial phase of the harmonic oscillation.
Using this terminology, we can say that (5.52) — (5.54) is the decomposition
of the oscillation, described by the f (t ) function, into harmonic oscillations:
a0 a0
f (t ) fk (t ) Ak sin( k t k
). (5.56)
2 k 1
2 k 1

The Fourier coefficients, amplitude and initial phase are related by the following
formulas:

395
 Chapter 5. Quadratic and Linear Splines

ak Ak sin k
,
bk Ak cos k
, (5.57)

Ak ak2 bk2 ,
where k = 1, 2, 3, ...
The calculation of the coefficients of the function’s expansion into the Fouri-
er series is called the analysis or forward Fourier transform. The analysis is de-
fined by formulas (5.53) and (5.54).
The calculation of the f (t ) function’s values, which correspond to the
known coefficients, a 0 , a k and bk (k = 1, 2, 3, ...), is called the synthesis or
backward Fourier transform. The synthesis is defined by formula (5.52).
Along with the Fourier transform theory, which considers the f (t ) function
known at all points of period [0, T ] , there is a theory, which considers the f (t )
function known only at the nodes of a uniform grid on [0, T ] .
Let n be a given number of equal steps or elementary segments [t j 1
, tj],
j = 1, 2, …, n, on period [0, T ] and f (t ) be a grid function given at points
t0 0 , t1 T / n , t2 2T / n , ..., t n 1
( n 1)T / n , t n T . We respectively
denote the f (t ) function’s values at these points as f 0 , f1 , f 2 , ..., f n 1
, fn ,
at that, f 0 = f n because of the f (t ) periodicity. The calculation of the Fourier
coefficients of grid function f (t ) is called the forward discrete Fourier trans-
form. The calculation of values f 0 , f1 , f 2 , ..., f n 1
, f n , which correspond to
the Fourier coefficients, is called the backward discrete Fourier transform.
The Data Analysis add-in for Excel includes the Fourier Analysis procedure
for performing the discrete Fourier transform. We will consider the use of this
procedure on an example of function
f (t ) 1 sin( t ) cos(2 t ) . (5.58)
Into cells A1:A8 on an Excel worksheet, we put this function’s values at the
first n = 8 points (ωt = 0, π/4, π/2, 3π/4, π, 5π/4, 3π/2, 7π/4) of the period with 9
equidistant points (0, π/4, π/2, 3π/4, π, 5π/4, 3π/2, 7π/4, 2π), i.e., we enter num-
bers 2, 1.707107, 1, 1.707107, 2, 0.292893, -1, 0.292893 (Fig. 5.15).
Let us fulfill the following operations:
1) Data > Data Analysis > Fourier Analysis > OK;
2) in the Fourier Analysis window opened, enter $A$1:$A$8 into text box
Input Range;
3) after activating text box Output Range, enter $B$1 into it (Fig. 5.16);

396
 5.10. The forward and backward Fourier transforms of a periodic function

4) click on the OK button to get the procedure execution results (Fig. 5.17).

Fig. 5.15. The Excel worksheet with the Fourier analysis source data

Fig. 5.16. The Fourier Analysis window before the procedure execution

Fig. 5.17. The Excel worksheet with the Fourier analysis results

397
 Chapter 5. Quadratic and Linear Splines

If we locate the mouse pointer on an exclamation mark (near the B1 cell), the
following information appears: The number in this cell is formatted as text or
preceded by an apostrophe. We should not be afraid of this information.
According to Fig. 5.17, the calculation results (in cells B1:B8) are the coeffi-
cients of the Fourier series in complex form: letter “i” (in cells B2, B4, B6 and
B8) means the imaginary unit, i 1.
The complex representation of the Fourier series [3] can be written as
ik t ik t
f (t ) c0 ck e c k
e ,
k 1
where c0 a0 , c k and c k
are complex Fourier coefficients, k = 1, 2, 3, ...
For using the procedure execution results, we have to know the following:
ik t
1) the values of coefficients c 0 and c k in front of e (k = 1, 2, 3, 4), mul-
tiplied by n = 8, are respectively located in cells B1, B2, B3, B4, B5 (the first half
of the set of complex Fourier coefficients);
ik t
2) the values of coefficients c in front of e k
(k = 1, 2, 3, 4), multi-
plied by 8, are respectively located in cells B8, B7, B6, B5 (the second half of
the set of complex Fourier coefficients);
3) c k = c k = 0 for k > 4.
For verifying it, we will use the following expressions, which follow from the
Euler relation for complex numbers [3]:
ik t ik t ik t ik t
e e ie ie
sin( k t ) , (5.59)
2i 2
ik t ik t
e e
cos( k t ) . (5.60)
2
Using (5.59) at k = 1 and (5.60) at k = 2, we write expression (5.58) in form
i i t 1 i2 t 1 i2 t i i t
f (t ) 1 e e e e . (5.61)
2 2 2 2
Multiplying values c 0 = 1, c1 = i / 2 , c 2 = 1 / 2 , c3 = 0, c 4 = c 4 = 0,
c 3
= 0, c 2
= 1/ 2 , c 1
= i / 2 by 8, we obtain the Fourier analysis results
depicted in Fig. 5.17. Let us pay attention to the following: the value of both 8c4
and 8c 4
is in the B5 cell.

398
 5.10. The forward and backward Fourier transforms of a periodic function

ik t ik t
We see that the complex coefficients in front of functions e and e
in expression (5.61) are conjugate, i.e., differ from each other only in the sign in
front of the imaginary unit. This is the property of the expansion of real-valued
ik t
functions, as f (t ) , in terms of e . The expansion of complex-valued func-
ik t
tions in terms of e does not have such property.
In the bottom left corner of the Fourier Analysis window (Fig. 5.16), we see
a little square field (element CheckBox) called Inverse. When clicking on this
field, the check mark appears in it, meaning the switching from the forward
transform (analysis) to the backward transform (synthesis).
Let the worksheet, depicted in Fig. 5.17, be active. To verify the correctness
of working the Fourier Analysis procedure, we fulfill the following operations:
1) Data > Data Analysis > Fourier Analysis > OK;
2) in the Fourier Analysis window opened, enter $B$1:$B$8 into text box
Input Range;
3) after activating text box Output Range, enter $E$1 into it;
4) by clicking on element Inverse, set the check mark in it;
5) click on the OK button for starting the procedure execution.
The execution results are depicted in Fig. 5.18. We see practical coincidence
of columns A and E, and that is natural because the forward and backward Fou-
rier transforms are performed successively.
If the mouse pointer is located on an exclamation mark (near the E1 cell), the
same information appears as in the case of Fig. 5.17.

Fig. 5.18. The Excel worksheet with results of the analysis and synthesis

The main drawback of the Fourier Analysis procedure consists in the follow-
ing: the number of steps on the period is not arbitrary, it must be a power of 2,
i.e., n must be equal to 2, 4, 8 (as in the above example), 16, 32, 64 or so on.
It is because this procedure realizes the so-called fast Fourier transformation [3].

399
 Chapter 5. Quadratic and Linear Splines

5.11. Subroutines for the forward and backward

discrete Fourier transforms

In this section, we will develop subroutines for the forward and backward
discrete Fourier transforms, free of the drawback formulated at the end of the
previous section. These subroutines will be used in the next section.
At first, let us obtain a discrete analog of formula (5.53) at k = 0, 1, 2, 3, ...
We use designation
f (t ) cos(k t ) g (t ) (5.62)
and consider periodic third-degree spline S (t ) , which respectively assumes
values g 0 , g 1 , g 2 , ..., g n 1
, g n at points 0, t1 , t 2 , ..., t n 1
, tn ( g 0 g n ).
According to expression (4.20), the integral of S (t ) equals
T n gj 1
gj n Mj 1
Mj
S (t ) dt h h3 , (5.63)
0 j 1
2 j 1
24
where h T / n is the grid step, M 0 , M 1 , M 2 , ..., M n 1
, M n are the spline
moments, M 0 Mn.
Let us consider equation (4.9) in form
j
M j 1 2M j j
Mj 1 j
, (5.64)
j = 1, 2, …, n, taking into account the periodicity of g (t ) and S (t ) , i.e., the fol-
lowing equalities:
g n 1 g1 , M n 1 M 1 ,
where g n 1
and M n 1
are the values of the g (t ) and S (t ) functions, respec-
tively, at additional node t n 1
T h . Because of the grid step constancy and
expressions (4.10), we have
1
j j
,
2
3
j
(g j 1
2g j gj 1
).
h2

400
 5.11. Subroutines for the forward and backward discrete Fourier transforms

Equation (5.64) takes the following form:

1 1 3
M j 1 2M j Mj 1 ( g j 1 2g j g j 1) .
2 2 h2
Summing both sides of the last equation over j = 1, 2, …, n, we obtain
1 n n 1 n
Mj 1
2 Mj Mj 1
2 j 1 j 1
2 j 1

3 n n n
gj 1
2 gj gj 1
h2 j 1 j 1 j 1
or
1 n n 1 n n n n
3
Mj 2 Mj Mj gj 2 gj gj
2 j 1 j 1
2 j 1 h2 j 1 j 1 j 1
or
n
Mj 0.
j 1
Expression (5.63) becomes simpler:
T n gj 1
gj n
S (t ) dt h h gj .
0 j 1
2 j 1
Taking into account expression (5.62), we obtain
2T
ak f (t ) cos( k t ) dt
T 0

2 n 2 n 2
h f j cos(k jh ) f j cos k j . (5.65)
T j 1
n j 1
n

This formula is the required discrete analog of formula (5.53) at k = 0, 1, 2, 3, ...

Similarly, we can obtain the following discrete analog of formula (5.54):
2 n 2
bk f j sin k j , (5.66)
n j 1
n
k = 1, 2, 3, ...
According to (5.52), (5.65) and (5.66), the formulas of the backward and
forward discrete Fourier transforms can be written in the following form for an
odd value of n:

401
 Chapter 5. Quadratic and Linear Splines

(n 1) / 2 2 2
fj d0 d 2k 1
sin k j d 2 k cos k j , (5.67)
k 1
n n

1 n
d0 fj , (5.68)
n j 1

2 n 2
d 2k 1
f j sin k j , (5.69)
n j 1
n

2 n 2
d2k f j cos k j , (5.70)
n j 1
n

where j = 0, 1, 2, ..., n in (5.67), k = 1, 2, 3, ..., (n 1) / 2 in (5.69) and (5.70).

Similarly, we have the following for an even value of n.
The backward discrete Fourier transform is performed according to formula
n/2 2 2
fj d0 d 2k 1
sin k j d 2 k cos k j , (5.71)
k 1
n n
j = 0, 1, 2, ..., n.
The forward discrete Fourier transform is performed according to formulas
(5.68) — (5.70) at k = 1, 2, 3, ..., (n 2) / 2 and
1 n
dn 1
f j sin( j ) 0, (5.72)
n j 1

1 n
dn f j cos( j ) . (5.73)
n j 1

Let us pay attention to the following:

the Fourier coefficients do not depend on the period size;
the number of Fourier coefficients is equal to n, i.e., to the number of
equal steps on the period (when n is even, we do not take into account coefficient
d n 1 , which is equal to zero).
Into Module14 of the BookNM workbook, we enter the following declaration
of the subroutine for the forward discrete Fourier transform:
Listing 5.7
Sub fouf(ByVal n, ByRef F() As Double, _
ByRef D() As Double, Optional m)

402
 5.11. Subroutines for the forward and backward discrete Fourier transforms

Dim m2 As Integer, j As Integer, k2 As Integer

Dim a As Double, b As Double
Dim w As Double, z As Double
Dim c0 As Double, s0 As Double
Dim c1 As Double, s1 As Double
Dim c2 As Double, s2 As Double
Const pi As Double = 3.141592654
If IsMissing(m) Then
m2 = n
ElseIf m < 0 Or m > n Then
m2 = n
Else
m2 = m
End If
a = 0
For j = 1 To n
a = a + F(j)
Next j
D(0) = a / n
w = 2 * pi / n
c0 = Cos(w): s0 = Sin(w)
z = 2 / n
c1 = 1: s1 = 0
For k2 = 2 To m2 Step 2
w = c1 * c0 - s1 * s0
s1 = s1 * c0 + c1 * s0: s2 = s1
c1 = w: c2 = c1
a = 0: b = 0
For j = 1 To n
a = a + F(j) * c2
b = b + F(j) * s2
w = c2 * c1 - s2 * s1
s2 = s2 * c1 + c2 * s1: c2 = w
Next j
D(k2 - 1) = b * z
D(k2) = a * z
If k2 = n Then
D(k2 - 1) = 0
D(k2) = D(k2) / 2
End If
Next k2
End Sub

403
 Chapter 5. Quadratic and Linear Splines

The subroutine name (fouf) occurs from words “Fourier” and “forward”.
The subroutine parameters have the following sense:
n is the number of equal steps on the period;
F is an array of the function values;
D is an array intended for the Fourier coefficients;
m is the doubled number of harmonic oscillations, which are of interest
(this optional parameter is used for possible reducing the execution time).
The basis of this subroutine are formulas (5.68) — (5.70), (5.72) and (5.73).
According to formulas (5.69) and (5.70), the forward discrete Fourier transform
requires the calculation of the sine and cosine values, repeated many times.
These values can be calculated in one of two ways:
by the multiple calls of built-in functions Sin(x) and Cos(x);
by the single call of functions Sin(x) and Cos(x) for calculating the
values of sin( 2 / n) and cos(2 / n) , and by subsequent usage of trigonometric
formulas for calculating the values of sin( kj 2 / n) and cos(kj 2 / n) at k > 1
and/or j > 1.
To reduce the execution time, the second way is used, which is based on the
following trigonometric formulas [3]:
sin( ) sin( ) cos( ) cos( ) sin( ) , (5.74)
cos( ) cos( ) cos( ) sin( ) sin( ) , (5.75)
where α and β are angles.
Into Module15 of the BookNM workbook, we enter the following declaration
of the subroutine for the backward discrete Fourier transform:
Listing 5.8
Sub foub(ByVal n, ByRef F() As Double, _
ByRef D() As Double, Optional m)
Dim m2 As Integer, j As Integer, k2 As Integer
Dim w As Double, z As Double
Dim c0 As Double, s0 As Double
Dim c1 As Double, s1 As Double
Dim c2 As Double, s2 As Double
Const pi As Double = 3.141592654
If IsMissing(m) Then
m2 = n
ElseIf m < 0 Or m > n Then
m2 = n
Else
m2 = m
End If

404
 5.11. Subroutines for the forward and backward discrete Fourier transforms

w = 2 * pi / n
c0 = Cos(w): c1 = c0: c2 = c1
s0 = Sin(w): s1 = s0: s2 = s1
For j = 1 To n
z = D(0)
For k2 = 2 To m2 Step 2
z = z + D(k2 - 1) * s2 + D(k2) * c2
w = c2 * c1 - s2 * s1
s2 = s2 * c1 + c2 * s1: c2 = w
Next k2
F(j) = z: w = c1 * c0 - s1 * s0
s1 = s1 * c0 + c1 * s0: s2 = s1
c1 = w: c2 = c1
Next j
F(0) = F(n)
End Sub

The subroutine name (foub) occurs from words “Fourier” and “backward”.
The parameters have the following sense:
n is the number of equal steps on the period;
F is an array intended for the function values;
D is an array of the Fourier coefficients;
m is the doubled number of harmonic oscillations, which are considered.
Formulas (5.67), (5.71), (5.74) and (5.75) are used in the foub subroutine.

405
 Chapter 5. Quadratic and Linear Splines

5.12. Solving the sound insulation problem

The sound waves, emitted by a sound source (by a vibrating body), propagate
in the medium (in a solid body, liquid or gas) in the form of longitudinal oscilla-
tions of the density of the medium. The ear of an adult person perceives sound
oscillations with frequency 1 / T from 17 — 20 Hz to about 20 kHz.
To lower the sound level (i.e., the amplitude of the density oscillation),
soundproof coverings are used. A technique of constructing such coverings with
given acoustic parameters was developed at the Andreyev Acoustics Institute,
Moscow. According to Fig. 5.19 (from page http://www.akin.ru/r_comm15.htm
of the institute website), the covering is characterized by efficiency with respect
to lowering the sound level. As we see, both the calculated and experimental
dependences for the covering’s efficiency are linear splines whose argument is
the logarithm of frequency k / T k / 2 of harmonic oscillation (5.55).

Fig. 5.19. The calculated (dark) and experimental (light) dependences of

the efficiency on the frequency: between the fractures, the efficiency is a linear
function of the logarithm of frequency (the base of logarithm does not matter due
to proportionality of logarithms with different bases [3]: log a x M log b x )

406
 5.12. Solving the sound insulation problem

Let us return to modeling of the bar oscillation in Section 3.20. We will at-
tach weightless membrane Ms (whose area equals A) to the bar and consider that
this membrane is the sound source, at that, dependence u (t ) in Fig. 3.18 is the
Ms membrane deviation. Besides, we will place the mechanism in a box with the
soundproof covering whose efficiency is defined by the calculated (dark) de-
pendence in Fig. 5.19.
We have to answer the following question: by how many decibels is the
sound (from the Ms membrane) attenuated, when it goes through the box’s
soundproof covering?
Outside the box, we place weightless membrane Mr (whose area equals A) of
the sound receiver; u r (t ) is the periodic dependence, which characterizes the Mr
membrane deviation (Fig. 5.20). We have to determine difference Δ (in decibels)
between the intensities of the oscillations of membranes Ms and Mr.

Fig. 5.20. The membranes separated by the soundproof covering

We use the following definition of the intensity of the harmonic oscillation

with frequency k / T :
Pk 10 lg Ak2 10 lg (d 22k 1
d 22k ) ,
where lg is the decimal logarithm, Ak is the oscillation amplitude, d 2k 1
, d 2k
are the Fourier coefficients of this oscillation, k = 1, 2, 3, ...
The intensities of the harmonic oscillations of membranes Ms and Mr are
given by formulas
Pks 10 lg [(d 2sk 1
)2 (d 2sk ) 2 ] , (5.76)

Pkr 10 lg [(d 2rk 1

)2 (d 2rk ) 2 ] , (5.77)

where d 2sk 1
, d 2sk and d 2r k 1
, d 2r k are the Fourier coefficients of dependences
u (t ) and u r (t ) , respectively, k = 1, 2, 3, ...

407
 Chapter 5. Quadratic and Linear Splines

The efficiency (in decibels) of the soundproof covering at frequency k / T is

equal to
Pks Pkr .
k
(5.78)
The sound oscillation of the environment density can be written in form
(5.56). Let us assume that phase k t k
of harmonic oscillation (5.55) does
not change when the sound goes through the soundproof covering, and amplitude
Ak changes only (k = 1, 2, 3, ...). In this case, the sound attenuation is calculated
as follows.
1. By means of the fouf subroutine, the forward discrete Fourier transform
of the u (t ) dependence, depicted in Fig. 3.18 (n = 25), is performed, that is,

Fourier coefficients d 0s , d 2sk 1 , d 2sk of this dependence are calculated, k = 1,

2, ..., 12.
2. The summary intensity of the oscillation of membrane Ms is calculated
according to formula
12
Ps 10 lg [(d 2sk 1
)2 ( d 2sk ) 2 ] . (5.79)
k 1

3. For every frequency k / T , the values of

/ 20
d 2r k 1
d 2sk 1
10 k , (5.80)
/ 20
d 2r k d 2sk 10 k (5.81)
are calculated, where k
is the efficiency of the soundproof covering at the con-
sidered frequency, k = 1, 2, ..., 12. Formulas (5.80) and (5.81) follow from the
logarithm properties [3] and formulas (5.57), (5.76) — (5.78).
4. The summary intensity of the oscillation of membrane Mr is calculated
according to formula
12
Pr 10 lg [(d 2r k 1
)2 ( d 2r k ) 2 ] . (5.82)
k 1

5. The required sound attenuation is calculated according to the following

formula:
Ps Pr . (5.83)
To obtain dependence u r (t ) of the Mr membrane deviation, we have to per-
form the backward discrete Fourier transform by means of the foub subroutine

408
 5.12. Solving the sound insulation problem

according to formula (5.67), which includes d 0r 0 and the values of d1r , d 2r ,

d 3r , d 4r , ..., d 23
r r
, d 24 calculated according to formulas (5.80) and (5.81).
As the source data for a program, intended for solving our problem, we use
table Listing 5.9, which includes:
the results of solving the oscillation equation (Fig. 3.18);
the dependence of the covering’s efficiency on the frequency.

Listing 5.9

M 0.001 N 8
K 800 frequency efficiency
L 100 17 5
F t u 45 1
0.00E+00 0.00E+00 6.62E-04 100 13.8
0.00E+00 2.00E-03 -5.89E-03 126 10
0.00E+00 4.00E-03 5.89E-03 158 8.8
0.00E+00 6.00E-03 -1.60E-03 720 8.8
0.00E+00 8.00E-03 -3.07E-03 1100 10
0.00E+00 1.00E-02 4.66E-03 1600 10
0.00E+00 1.20E-02 -2.57E-03
0.00E+00 1.40E-02 -1.01E-03
0.00E+00 1.60E-02 3.20E-03
0.00E+00 1.80E-02 -2.67E-03
0.00E+00 2.00E-02 2.91E-04
0.00E+00 2.20E-02 1.87E-03
0.00E+00 2.40E-02 -2.27E-03
0.00E+00 2.60E-02 9.54E-04
5.00E+00 2.80E-02 8.20E-04
1.00E+01 3.00E-02 1.65E-02
5.00E+00 3.20E-02 1.77E-02
0.00E+00 3.40E-02 -1.46E-02
0.00E+00 3.60E-02 1.48E-03
0.00E+00 3.80E-02 1.03E-02
0.00E+00 4.00E-02 -1.25E-02
0.00E+00 4.20E-02 5.17E-03
0.00E+00 4.40E-02 4.59E-03
0.00E+00 4.60E-02 -9.23E-03
0.00E+00 4.80E-02 6.32E-03
0.00E+00 5.00E-02 6.62E-04

409
 Chapter 5. Quadratic and Linear Splines

The subtable with the dependence of the covering’s efficiency on the fre-
quency includes:
N, the number of the frequency values;
the frequency values (in hertzs) in column frequency;
the efficiency values (in decibels) in column efficiency, according to the
dark dependence in Fig. 5.19.
The program follows:
Listing 5.10
Sub main()
Dim T() As Double
Dim U() As Double
Dim D() As Double
Dim m As Integer
Dim N As Integer
Dim j As Integer, k2 As Integer
Dim i As Integer
Dim sb As String, se As String
Dim fr As Double, eff As Double
Dim ln_frequ() As Double
Dim efficiency() As Double
Dim MOM() As Double
Dim Ps As Double, Pr As Double
Dim w As Double
m = Selection.Rows.Count 'quantity of rows
N = Selection.Cells(1, 5)
ReDim T(0 To m - 5)
ReDim U(0 To m - 5)
ReDim D(0 To m - 5)
ReDim ln_frequ(1 To N)
ReDim efficiency(1 To N)
ReDim MOM(1 To N)
For j = 0 To m - 5
T(j) = Selection.Cells(j + 5, 2)
U(j) = Selection.Cells(j + 5, 3)
Next j
sb = Selection.Cells(5, 2).Address
se = Selection.Cells(m, 3).Address
0: Call graph(sb, se, "t, s", "u, m")
For i = 1 To N
ln_frequ(i) = Log(Selection.Cells(i + 2, 4))
efficiency(i) = Selection.Cells(i + 2, 5)

410
 5.12. Solving the sound insulation problem

MOM(i) = 0
Next i
1: Call fouf(m - 5, U, D)
D(0) = 0
fr = 1 / T(m - 5) 'value of main frequency 1/T
'Calculating value of Ps:
w = 0
For k2 = 2 To m - 5 Step 2
w = w + D(k2 - 1) ^ 2 + D(k2) ^ 2
Next k2
Ps = 10 * Log(w) / 2.302585093
'Calculating value of Pr:
w = 0
For k2 = 2 To m - 5 Step 2
2: Call si(1, N, ln_frequ, efficiency, MOM, _
Log(k2 / 2 * fr), eff)
D(k2 - 1) = D(k2 - 1) * 10 ^ (-eff / 20)
D(k2) = D(k2) * 10 ^ (-eff / 20)
w = w + D(k2 - 1) ^ 2 + D(k2) ^ 2
Next k2
Pr = 10 * Log(w) / 2.302585093
'Calculating sound attenuation:
3: w = Ps - Pr
4: MsgBox "delta =" & Str(Round(w, 3)) & " dB"
'Synthesis of time dependence:
5: Call foub(m - 5, U, D)
sb = Selection.Cells(5, 6).Address
se = Selection.Cells(m, 7).Address
Selection.Cells(4, 6) = "t"
Selection.Cells(4, 7) = "ur"
For j = 0 To m - 5
Selection.Cells(j + 5, 6) = T(j)
Selection.Cells(j + 5, 7) = U(j)
Next j
6: Call graph(sb, se, "t, s", "ur, m")
Range("O33").Select
End Sub

Operator 0 of this program creates the u (t ) graph of the bar deviation, i.e.,
of the deviation of membrane Ms, which is the sound source. For that, the
graph subroutine (Section 4.8) is used for the first time.

411
 Chapter 5. Quadratic and Linear Splines

Further, dependence u (t ) is expanded into the Fourier series by calling the

fouf subroutine for the forward discrete Fourier transform (operator 1). By
means of the first k2 cycle, the value of Ps is calculated according to formula
(5.79).
By means of the second k2 cycle, the value Pr is calculated according to
formulas (5.80) — (5.82). In this case, the linear spline, determined by the de-
pendence of efficiency on frequency in tabular form, is used. The argument of the
linear spline is ln_ frequ — the logarithm of frequency or the so-called logarith-
mic frequency (see operator 2 and the caption to Fig. 5.19).
Operator 3 calculates the sound attenuation according to formula (5.83). The
resulting value is rounded up to three decimal places and put into the standard
window (operator 4).
By calling the foub subroutine for the backward discrete Fourier transform,
periodic time dependence u r (t ) of the deviation of membrane Mr , which is the
sound receiver, is synthesized (operator 5). Operator 6 creates the u r (t ) graph.
For that, the graph subroutine is used for the second time.
The source data for program Listing 5.10 are the values given in complex
table Listing 5.9 (Fig. 5.21). Before running the program, this Excel table (range
B2:F31) must be selected.

Fig. 5.21. The Excel table with the source data

After starting the program execution:

1) the u (t ) graph appears;

412
 5.12. Solving the sound insulation problem

2) the window with the Δ value of the sound attenuation appears (Fig. 5.22);
3) after clicking on OK, the u r (t ) dependence and its graph appear.
Fig. 5.23 shows the Excel worksheet upon termination of the program execu-
tion. The automatically constructed graphs of dependences u (t ) and u r (t ) are
on this worksheet; Fig. 5.24 shows them completely.

Fig. 5.22. Window with the calculated sound attenuation

Fig. 5.23. The calculated u r (t ) dependence and graphs

413
 Chapter 5. Quadratic and Linear Splines

Fig. 5.24. The automatically created graphs: a — u (t ) ; b — u r (t )

We advise the reader to return to the task on p. 279 for calculating the Fouri-
er coefficients of dependences v(t ) and i (t ) .

414
 Chapter 6.
Numerical Methods
for Nonlinear Programming

For solving a series of applied problems, in particular the ones involving

optimization, we must find minimum point x of linear or nonlinear non-
negative function F (x) = F ( x1 , x2 , ..., xn ) , where x = ( x1 , x 2 , ..., x n ) is
a point (vector) of the n-dimensional space, n ≥ 1. The F (x) function being min-
imized is called an objective function.
Minimizing a linear objective function with linear constraints is called linear
programming. Minimizing a nonlinear objective function (with or without con-
straints) is called nonlinear programming.
The minimization, which does not require knowledge of mathematical
expressions for partial derivatives of the objective function with respect to its
arguments, is called the search for the minimum point. The importance of search
methods follows from the fact that expressions for the partial derivatives are
often unavailable.
In the first section of this chapter, tasks of linear and nonlinear programming
are solved by means of the Solver add-in. In the fifth chapter of book [2], we
demonstrated that Solver for Excel 2007 can give an incorrect result upon mini-
mizing nonlinear function
F ( x1 , x2 ) 100 ( x2 x12 ) 2 (1 x1 ) 2 (6.1)
known as the Rosenbrock function. We did not observe this drawback in the later
versions of Excel (2010 and 2013). However, the minimization method, realized
in Solver for the later versions of Excel, is less efficient than the Powell method
[14] considered in this chapter (p. 454).
The x point is called a local minimum of function F (x) if a neighborhood

of x exists, where inequality F (x) F (x ) is satisfied. Function F (x) can be

unimodal, with one local minimum, or multimodal, with several local minima.
Two subroutines are developed that are intended to find a local minimum of
the F (x) function of one or several variables. In the case of several variables,
these subroutines realize the coordinate-descent and Powell methods.

415
 Chapter 6. Numerical Methods for Nonlinear Programming

Possibilities of the developed minimization subroutines are demonstrated on

the following mathematical and applied tasks:
optimizing the dimensions of a one-liter tin can;
determining the equilibrium state of a four-spring mechanical system;
minimizing a nonlinear function with nonlinear constraints and a tabular
function of two variables;
determining the local minima of a multimodal function of two variables.
Besides, the minimization subroutines are used:
in the shooting method for solving the nonlinear differential equation with
boundary conditions;
in the least-squares method for determining the production function.
It is obvious that the problem of maximizing positive function G (x) is
equivalent to the problem of minimizing function F (x) 1 / G(x) . Therefore, the
minimization subroutines of this chapter can be used to find the maxima of func-
tion G (x) .

416
 6.1. Minimizing linear and nonlinear functions of several variables by the Solver add-in

6.1. Minimizing linear and nonlinear functions

of several variables by the Solver add-in

We used the Solver add-in for solving the nonlinear algebraic equation (Sec-
tion 4.5). Let us consider two more examples of using this add-in — for solving
problems of linear and nonlinear programming.
Initially, we will solve the following transportation problem of linear pro-
gramming from article [15].
There are three points (A1, A2, A3 ) for sending identical loads and four points
( B1, B2, B3, B4 ) for receiving them. Let cij be the expenses for relocation of one
load from Ai to Bj (i = 1, 2, 3, j = 1, 2, 3, 4). We have to find the minimum of
total expenses for the required relocations from points A1, A2, A3 to points B1,
B2, B3, B4.
More precisely, the source data for the problem include:
the matrix of expenses for transportation of one load looking like
c11 c12 c13 c14 3 4 5 6
C c21 c22 c23 c24 4 4 2 1 ;
c31 c32 c33 c34 2 1 6 8
the number of loads at points A1, A2, A3: a1 8 , a2 5 , a3 7;
the requirement for loads at points B1, B2, B3, B4: b1 4 , b2 4,
b3 2 , b4 10 (note that b1 b2 b3 b4 a1 a2 a3 ).
We have to find:
the minimum of total expenses for the load relocations from points A1,
A2, A3 to points B1, B2, B3, B4;
the corresponding number of loads relocated from point Ai to point Bj
(i = 1, 2, 3, j = 1, 2, 3, 4).
For solving the formulated problem, let us put:
the C matrix into range A1:D3 on an Excel worksheet;
values a1 8 , a2 5 , a3 7 into cells E1:E3 of the same worksheet;
values b1 4 , b2 4 , b3 2 , b4 10 into cells A4:D4 (Fig. 6.1).

417
 Chapter 6. Numerical Methods for Nonlinear Programming

Fig. 6.1. The Excel worksheet with the source data:

the shaded range corresponds to the C matrix

We will use matrix

x11 x12 x13 x14
X x21 x22 x23 x24 ,
x31 x32 x33 x34
where xij is the number of loads relocated from Ai to Bj .
Product cij xij denotes the expenses for the load relocation from point Ai
to point Bj (i = 1, 2, 3, j = 1, 2, 3, 4). The total expenses for the relocations are
equal to
F cij xij , (6.2)
i, j
where the summation is over i = 1, 2, 3 and j = 1, 2, 3, 4.
Formula (6.2) gives the linear function of 12 variables,
F F ( x11 , x12 , x13 , x14 , x21 , x22 , x23 , x24 , x31 , x32 , x33 , x34 ) ,
whose arguments are the X matrix elements. We have to minimize this function
with linear constraints
4
x ij ai , (6.3)
j 1
i = 1, 2, 3,
3
xij bj , (6.4)
i 1
j = 1, 2, 3, 4.
Continuing to fill the Excel worksheet, we allot cells A5:D7 for the X matrix
elements, which are unknown, and put arbitrary numbers into these cells (for
example, units). Into cell F2, we enter formula

=SUMPRODUCT(A1:D3;A5:D7)

418
 6.1. Minimizing linear and nonlinear functions of several variables by the Solver add-in

corresponding to mathematical formula (6.2). Into cells E5, E6 and E7, we enter
formulas

=SUM(A5:D5)
=SUM(A6:D6)
=SUM(A7:D7)

corresponding to the left-hand side of constraints (6.3). Into cells A8, B8, C8 and
D8, we enter formulas

=SUM(A5:A7)
=SUM(B5:B7)
=SUM(C5:C7)
=SUM(D5:D7)

corresponding to the left-hand side of constraints (6.4).

Upon tuning Excel so that formulas are displayed in cells (p. 193), the work-
sheet takes the form depicted in Fig. 6.2. We return to the customary tuning of
Excel when the calculated results are in cells.

Fig. 6.2. The Excel worksheet with the formulas in the cells: the shaded
top and bottom ranges correspond to the C and X matrices, respectively

Let us fulfill the following operations:

1) Data > Solver in area Analysis;
2) in the Solver Parameters window, enter $F$2 into text box Set Objective;
3) turn on option Min;
4) enter $A$5:$D$7 into text box By Changing Variable Cells;
5) by means of clicks on the Add button, successively enter conditions
$E$5:$E$7=$E$1:$E$3 and $A$8:$D$8=$A$4:$D$4 into box Subject to the
Constraints (in addition to the equal sign, we can use comparison signs “less
than or equal to” and “greater than or equal to” in the constraints);

419
 Chapter 6. Numerical Methods for Nonlinear Programming

6) enter Simplex LP into box Select a Solving Method by means of the drop-
down list (Fig. 6.3);
7) click on the Solve button;
8) in the Solver Results window opened, click on OK.

Fig. 6.3. The Solver Parameters window before minimizing

linear function (6.2) with linear constraints (6.3) and (6.4)

Results of minimizing function (6.2) with constraints (6.3) and (6.4) appear
in cell F2 and range A5:D7 (Fig. 6.4).
Value 58 in cell F2 gives the minimum of total expenses for the load reloca-
tions. The contents of range A5:D7 suggest the distribution of the load reloca-
tions for minimizing the total expenses.
For example:

420
 6.1. Minimizing linear and nonlinear functions of several variables by the Solver add-in

value x11 1 (in cell A5) means that one load must be relocated from
point A1 to point B1;
value x24 5 (in cell D6) means that five loads must be relocated from
point A2 to point B4.

Fig. 6.4. The worksheet upon termination of minimizing

function (6.2) with constraints (6.3) and (6.4)

According to the source data for the problem solved by us, equality
3 4
ai bj
i 1 j 1
is satisfied. If this equality is not satisfied, that is,
3 4
ai bj ,
i 1 j 1
we should introduce virtual point B5 (for receiving the loads) and consider that
3 4
b5 ai bj ,
i 1 j 1

c15 c25 c35 0.

Further, we use the above method for solving the problem of minimizing the
total expenses for the load relocations from points A1, A2, A3 to points B1, B2,
B3, B4, B5. The calculated values of x15 , x25 , x35 are the numbers of loads
remaining at points A1, A2, A3.

421
 Chapter 6. Numerical Methods for Nonlinear Programming

According to the fourth chapter of book [2], the considered operation mode
of Solver for Excel 2007 can be used for solving the system of linear algebraic
equations (3.49). Solver for the later versions of Excel does not have this ability.
Below, we will minimize nonlinear function (6.1), the Rosenbrock function,
by means of the Solver add-in. The graph of this rather popular test function is
depicted in Fig. 6.5, which is taken from the following page of Wikipedia, the
free encyclopedia: http://en.wikipedia.org/wiki/Rosenbrock_function.
The following is obvious:
nonlinear function (6.1) is continuous and non-negative;
point x of the function minimum has coordinates x1 x2 1 , and
F (1, 1) 0;
point x is in “ravine” (on the dark stripe in Fig. 6.5).
Let us show that the Rosenbrock function does not have other minimum points.

Fig. 6.5. Graphic of Rosenbrock function F ( x1 , x2 ) 100 ( x2 x12 ) 2 (1 x1 ) 2

The partial derivatives of function (6.1) are defined by the following algebra-
ic expressions:

422
 6.1. Minimizing linear and nonlinear functions of several variables by the Solver add-in

F / x1 400x13 400x1 x2 2 x1 2, (6.5)

F / x2 200( x2 x12 ) . (6.6)

According to the necessary condition for an extreme value [3], the minimum
points of function F ( x1 , x 2 ) belong to the set of points ( x1 , x2 ) that are solu-
tions of the system of equations
F / x1 0 ,
F / x2 0
or
400x13 400x1x2 2 x1 2 0,

x2 x12 0.

According to the second equation, x2 x12 . Substituting this expression into the

first equation, we have x1 x12 1 .

1 . Consequently, x2
Thus, function (6.1) has the unique minimum point with coordinates
x1 x2 1 , i.e., the function is unimodal, and high-quality minimization tools
must find this point under any initial approximation.
To minimize function (6.1) by means of the Solver add-in, we assume that:
the G1 and H1 cells contain the values of x1 and x 2 , respectively;
the F1 cell contains the value of F ( x1 , x 2 ) .
Into cell F1, we enter formula

=100*(H1-G1^2)^2+(1-G1)^2

corresponding to mathematical formula (6.1). Into cells G1 and H1, we respec-

tively enter values - 5.5 and 0.5 defining the initial approximation of the mini-
mum point: x1 = - 5.5 and x 2 = 0.5 (Fig. 6.6).

Fig. 6.6. The worksheet before minimizing nonlinear function (6.1)

423
 Chapter 6. Numerical Methods for Nonlinear Programming

Let us fulfill the following operations:

1) Data > Solver in area Analysis;
2) in the Solver Parameters window opened, enter $F$1 into text box Set
Objective;
3) turn on option Min;
4) enter $G$1:$H$1 into text box By Changing Variable Cells;
5) enter GRG Nonlinear into box Select a Solving Method by means of the
drop-down list (Fig. 6.7)
6) click on the Solve button;
7) click on the OK button in the Solver Results window.
The result of minimizing the Rosenbrock function is given in Fig. 6.8.

Fig. 6.7. The Solver Parameters window before minimizing function (6.1)

424
 6.1. Minimizing linear and nonlinear functions of several variables by the Solver add-in

Fig. 6.8. The worksheet upon termination of minimizing function (6.1)

The Excel formula, intended for calculating the values of objective function
F ( x1 , x 2 ) , may include Excel functions, in particular, user-defined functions.
To demonstrate this, let us consider the following user-defined function corre-
sponding to mathematical function (6.1).
Listing 6.1
Function Rosenbrock(x1, x2)
Rosenbrock = 100 * (x2 - x1 ^ 2) ^ 2 + _
(1 - x1) ^ 2
End Function

We put this declaration into Module16 of the BookNM workbook and formula

=Rosenbrock(G1;H1)

into cell F1.

Excel user-defined functions, developed by means of VBA, may be very
complicated. Therefore, the Solver add-in allows us to solve quite complex
problems.
During the minimization, we can see results of iterations. For demonstrating
this possibility of the Solver add-in, we will solve the previous task again (see
Fig. 6.9, which is similar to Fig. 6.6).

Fig. 6.9. The worksheet before minimizing nonlinear function (6.1)

To see results of iterations, we fulfill the following operations:

1) click on the Options button in the Solver Parameters window depicted in
Fig. 6.7;

425
 Chapter 6. Numerical Methods for Nonlinear Programming

2) in the Options window opened, turn on option Show Iteration Results

(Fig. 6.10);
3) click on the OK button;
4) click on the Solve button in the Solver Parameters window (Fig. 6.7).

Fig. 6.10. The Options window with option Show Iteration Results turned on

426
 6.1. Minimizing linear and nonlinear functions of several variables by the Solver add-in

Fig. 6.11a shows the result of the first iteration. After the next 4 clicks on
Continue in window Show Trial Solution, we see the result of the 5th iteration
(Fig. 6.11b). After the next 16 clicks on Continue, the Solver Results window
appears. We click on OK in the last window to terminate the execution.
Fig. 6.12, which is similar to Fig. 6.8, shows the result. Thus, 21 iterations were
performed.
The Powell minimization method (Section 6.5) gives the same result after 5
iterations (p. 454).

Fig. 6.11. The worksheet after the 1st (a) and 5th (b) iterations

Fig. 6.12. The worksheet upon termination of minimizing function (6.1)

427
 Chapter 6. Numerical Methods for Nonlinear Programming

6.2. Method for minimizing a nonlinear

function of one variable

The minimization of a nonlinear function of one variable, f (x) , is an itera-

tive process, which needs an initial approximation of the required minimum
point. When developing the minimization method, we will assume that f (x)
satisfies the following conditions (in the function’s domain including the unique
minimum point, x , and its initial approximation, x 0 ):
the f (x) function is continuous and non-negative;
the derivative, f (x) , is continuous (that is, the function is smooth).
In the algorithm below, the minimization of f (x) includes the following
three stages:
1) finding the so-called uncertainty segment, which contains point x within
itself;
2) reducing the uncertainty segment to one and a half times;
3) reducing the uncertainty segment by means of the parabolic interpolation
of f (x) .
For finding the uncertainty segment, the movement along the x axis must be
in the direction of decreasing f (x) , at that, each step is twice as large as the
previous step (as in the second chapter of book [16]).
When searching the uncertainty segment, only three points are stored, whose
designations are a, b, c, at that, point a is the last, b is the penultimate point,
c precedes b. The function values at these points are also stored: f a f (a ) ,
fb f (b ) , f c f (c ) . The movement is terminated when function f (x)
becomes increasing, i.e., condition f a f b is satisfied (Fig. 6.13). Resulting
[a, c] is the uncertainty segment.
Let d be the midpoint of segment [a, b] : d (a b) / 2 . After calculating
value f d f (d ) (Fig. 6.14), the condition of the termination of searching the
minimum point is checked and, if needed, the uncertainty segment is reduced
according to the following algorithm.

428
 6.2. Method for minimizing a nonlinear function of one variable

Fig. 6.13. Termination of searching the uncertainty segment: [a, c] is

the uncertainty segment; the movement from right to left took place

Fig. 6.14. The result of calculating value f d f (d ) :

segments [ a, d ] , [d , b] and [b, c] are equal in length

1. If condition fb fd fb is satisfied ( is a given positive constant),

the minimization is terminated, and assignments x d , f (x ) f d are per-

formed. Otherwise, the next item is fulfilled.

429
 Chapter 6. Numerical Methods for Nonlinear Programming

2. Of the four points (a, b, c, d ), three neighboring points are chosen, such
that the function value at the midpoint is less than the function values at the
endpoints. These three points are denoted by letters a, b, c; the corresponding
function values are denoted by f a , f b , f c , respectively. As a result, the uncer-
tainty segment, [a, c] , is reduced to one and a half times (Fig. 6.15).
3. The minimum point of the parabola, passing through points A = (a, f a ),
B = (b, f b ), C = (c, f c ), is calculated according to formula

f a (b 2 c2 ) fb (c 2 a2 ) fc ( a 2 b2 )
d 0.5 . (6.7)
f a (b c) fb (c a) fc ( a b)
The derivation of this formula is given below.
4. Value f d f (d ) is calculated (Fig. 6.15). The jump to the first item is
performed.

Fig. 6.15. The results of reducing the uncertainty segment to 1.5 times and
of the parabolic interpolation: [a, c] is the resulting uncertainty segment;
b is the midpoint of [a, c] ; d is the minimum point of the parabola

Let us derive formula (6.7).

According to the Lagrange interpolation formula [3], the equation of the
quadratic parabola, passing through points A, B, C (Fig. 6.15), is as follows:
( x b)( x c ) ( x a )( x c ) ( x a )( x b)
y fa fb f .
( a b)(a c ) (b a )(b c ) (c a )(c b) c

430
 6.2. Method for minimizing a nonlinear function of one variable

Using the basic rules of differentiation [3], we obtain

dy ( x b) ( x c ) ( x a) ( x c) ( x a ) ( x b)
fa fb f .
dx ( a b)(a c ) (b a )(b c ) (c a )(c b) c
We bring the right-hand side of this expression down to common denominator:
dy 1
[ 2 x (b c ) f a (b 2 c 2 ) f a
dx ( a b)(a c )(b c )
2 x(a c) fb (a 2 c 2 ) fb 2 x(a b) f c (a 2 b 2 ) fc ] .
The coordinate of the minimum point of the considered quadratic parabola is
the solution of equation dy / dx 0 , that is,
2 x (b c) fa (b 2 c 2 ) fa 2 x(a c ) fb (a 2 c 2 ) fb
2 x(a b) fc (a 2 b 2 ) fc 0
or
2 x (b c) fa 2 x(a c ) fb 2 x(a b) fc
2 2 2 2 2
(b c ) fa (a c ) fb (a b 2 ) fc .
This solution, denoted by d, has form (6.7).

431
 Chapter 6. Numerical Methods for Nonlinear Programming

6.3. The coordinate-descent method

Let x0 F ( x1 , x2 , ..., xn ) be a continuous and non-negative function of n

variables. Besides, we assume that this function has the unique minimum point
and continuous partial derivatives F / x1 , F / x2 , …, F / xn . Determining
the minimum point of F ( x1 , x2 , ..., xn ) is an iterative process.
In the coordinate-descent method, each iteration consists of successive mini-
mizations of the F function on arguments x1 , x 2 , …, x n : at first, the minimiza-
tion of the F function on x1 is performed, then on x 2 , ..., and on x n . The itera-
tions are repeated until the condition of the termination of searching the mini-
mum point is satisfied.
The declaration of the mini subroutine, realizing the coordinate-descent
method, is given below. We enter it into Module17 of the BookNM workbook.
Listing 6.2
Sub mini(ByVal n, ByRef x() As Double, _
ByRef ss() As Double, ByVal rho, Optional alpha)
Dim fa As Double, fb As Double, fc As Double
Dim fd As Double
Dim a As Double, b As Double, c As Double
Dim d As Double, e As Double
Dim s() As Double: ReDim s(n)
Dim y() As Double: ReDim y(n)
Dim z() As Double: ReDim z(n)
Dim sss() As Double: ReDim sss(n, n)
Dim i As Byte, j As Byte
d = 0
For j = 1 To n
For i = 1 To n
If i <> j Then d = ss(i, j) ^ 2 + d
Next i
Next j

432
 6.3. The coordinate-descent method

If d <> 0 Then
MsgBox "mini: Array ss is non-diagonal"
End 'immediate termination of macro
End If
For j = 1 To n
For i = 1 To n
sss(i, j) = ss(i, j)
Next i
Next j
m1: For i = 0 To n
y(i) = x(i)
z(i) = x(i)
Next i
For j = 1 To n 'j - number of descent direction
For i = 1 To n
s(i) = ss(i, j)
Next i
d = 0
For i = 1 To n
d = s(i) ^ 2 + d
Next i
If d = 0 Then
For i = 1 To n
s(i) = sss(i, j)
Next i
Else
For i = 1 To n
sss(i, j) = s(i)
Next i
End If
'Finding uncertainty segment:
fa = y(0): fb = y(0): fc = y(0)
a = 0: b = 0: c = 0
d = 1: e = 1
s1: For i = 1 To n
x(i) = y(i) + d * s(i)
Next i
Call func(n, x)
If Not IsMissing(alpha) Then
If x(0) < alpha Then GoTo m3
End If
fd = x(0)

433
 Chapter 6. Numerical Methods for Nonlinear Programming

If fd < fa Then
fc = fb: fb = fa: fa = fd
c = b: b = a: a = d
d = 2 * d + e
GoTo s1
Else
If fa = fb Then
fb = fd
b = d
e = -2 * d
d = e
GoTo s1
End If
End If
fc = fb: fb = fa: fa = fd
c = b: b = a: a = d
d = (a + b) * 0.5
For i = 1 To n
x(i) = y(i) + d * s(i)
Next i
Call func(n, x)
If Not IsMissing(alpha) Then
If x(0) < alpha Then GoTo m3
End If
fd = x(0)
'Reducing uncertainty segment:
s2: If Abs(fb - fd) < rho * fb Then GoTo s0
If (c - d) * (d - b) < 0 Then
If fd < fb Then
fc = fb: fb = fd
c = b: b = d
Else
fa = fd
a = d
End If
Else
If fd < fb Then
fa = fb: fb = fd
a = b: b = d
Else
fc = fd
c = d

434
 6.3. The coordinate-descent method

End If
End If
d = fa * (b - c) + fb * (c - a) + fc * (a - b)
If d = 0 Then GoTo s0
d = (fa * (b * b - c * c) + _
fb * (c * c - a * a) + _
fc * (a * a - b * b)) / (2 * d)
For i = 1 To n
x(i) = y(i) + d * s(i)
Next i
Call func(n, x)
If Not IsMissing(alpha) Then
If x(0) < alpha Then GoTo m3
End If
fd = x(0)
GoTo s2
s0: For i = 1 To n
ss(i, j) = x(i) - y(i)
y(i) = x(i)
Next i
y(0) = x(0)
Next j
'Checking condition of minimization termination:
m3: If Not IsMissing(alpha) Then
If test(n, x, z, rho, alpha) Then GoTo m1
Else
If test(n, x, z, rho) Then GoTo m1
End If
End Sub

The mini subroutine under consideration has five parameters, and parameter
alpha is optional.
The obligatory parameters have the following sense:
n is the number of variables;
x is an array with elements x(0), x(1), x(2), …, x(n), at that,
memory cell x(0) contains the F function’s value corresponding to the values
of x1 , x 2 , …, x n being in memory cells x(1), x(2), …, x(n), respectively;
ss is a two-dimensional array n × n, corresponding to diagonal matrix S
of initial steps: memory cell ss(1, 1) contains the initial step along the
x1 axis, cell ss(2, 2) contains the initial step along the x 2 axis, …, cell

435
 Chapter 6. Numerical Methods for Nonlinear Programming

ss(n, n) contains the initial step along the x n axis (the diagonal matrix is
a square matrix that has nonzero elements only on its main diagonal);
rho is the relative change in F, terminating the minimization along an
axis (see in the first item of the algorithm in the previous section).
The above declaration of the mini subroutine, Listing 6.2, contains three
calls of the func subroutine intended for calculation of the F function’s value
(the value of the x(0) element) corresponding to the current values of elements
x(1), x(2), …, x(n).
Optional parameter alpha of the mini subroutine is used when we know
that the minimum value of non-negative function F is equal to zero: when condi-
tion F is satisfied, operator GoTo m3 (following the func subroutine
calls) is performed, and then the test function is called.
In the test function declaration, conditions must be formulated for termi-
nating the minimization. The call of this function is performed when F and
also at the end of every iteration, that is, after minimizations of the F function on
all arguments x1 , x 2 , …, x n .
Function test returns True to the mini subroutine if none of the minimi-
zation termination conditions is satisfied. In this case, the next iteration is per-
formed (that is, the minimizations of the F function on arguments x1 , x 2 , …,
x n are repeated) and the call of function test is again performed. The mini
subroutine execution is terminated when test returns False.
To control the course of the x0 F ( x1 , x2 , ..., xn ) function minimization,
the test function declaration must contain operators for putting the current
values of elements x(0), x(1), x(2), …, x(n) into cells on the Excel work-
sheet.
Note that from iteration to iteration in the course of the minimization:
not only one-dimensional array x is being changed, but two-dimensional
array ss, corresponding to matrix S of initial steps for the next iteration, is also
being changed;
when changing array ss, the S matrix remains diagonal in the considered
coordinate-descent method.

436
 6.4. Examples of using the minimization methods

6.4. Examples of using the minimization methods

The mini subroutine from the previous section will be used for solving the
following two tasks: the optimization of a tin can and the Rosenbrock function
minimization.
The optimization of a tin can.
According to this task, we have to determine a variant of the cylindrical tin
can of a given volume, which is the best in terms of the amount of tin required.
The volume and total area of a right circular cylinder [3] are respectively
defined by formulas
V r 2h ,
A 2 r ( r h) ,
where r is the base radius, h is the height. The first formula leads to
V
h . (6.8)
r2
By means of this expression, we eliminate h from the second formula:
V
A( r ) 2 r 2 2 . (6.9)
r
For given volume V of the tin can, the total area is defined by the last formu-
la. We have to find the minimum point of the A(r ) function for positive values
of the r argument.
Because the objective function, A(r ) defined by (6.9), has a simple form, the
optimization task can be solved analytically. For that, according to [3], we must
fulfill the following:
1) find real roots of equation
dA
0; (6.10)
dr
2) choose those of the roots, which are positive and satisfy the following
inequality:

437
 Chapter 6. Numerical Methods for Nonlinear Programming

d2A
0.
dr 2
Using formula (6.9) and the basic rules of differentiation [3], we obtain
dA 2
(2 r 3 V ) .
dr r 2
Thus, equation (6.10) becomes
2 r3 V 0.
This equation has only one real root:
V
r . 3 (6.11)
2
The second derivative of the A(r ) function equals

d2A 4
( r3 V).
2 3
dr r
We see that the second derivative is positive for any positive value of r. There-
fore, formula (6.11) gives the desired optimal value of r.
According to formula (6.11), for V = 1000 cm3 (that is, in the case of the one-
liter can), the optimal base radius is equal to r = 5.41926 cm. Formulas (6.8)
and (6.9) lead to the following optimal values of the height and total area of the
tin can: h = 10.83853 cm, A = 553.5810 cm2.
For optimizing the one-liter can by means of subroutine mini, we enter the
following text of program main, subroutine func and function test into
Module1 of the BookNM workbook.
Listing 6.3
Dim nf As Long 'counter of calls of func
Dim nt As Long 'counter of calls of test

Sub main()
Dim x() As Double
Dim ss() As Double
Dim n As Byte
n = 1 'number of variables
ReDim x(n)

438
 6.4. Examples of using the minimization methods

ReDim ss(1 To n, 1 To n)
x(1) = Range("Sheet2!G1").Value
ss(1, 1) = Range("Sheet2!G2").Value
If ss(1, 1) ^ 2 = 0 Then
Range("Sheet2!A1").Value = _
"Initial step must be increased"
End
End If
nf = 0
nt = 0
0: Call func(n, x) 'it must be before minimization
1: Call mini(n, x, ss, 1E-6)
End Sub

Sub func(ByVal n, ByRef x() As Double)

Const pi As Double = 3.141592654
nf = nf + 1
x(0) = 2 * pi * x(1) ^ 4 + 2 * 1000 / x(1) ^ 2
End Sub

Function test(ByVal n, ByRef x() As Double, _

ByRef z() As Double, ByVal rho, Optional alpha) _
As Boolean
nt = nt + 1
Range("Sheet2!A" & CStr(nt)) = x(0)
Range("Sheet2!B" & CStr(nt)) = x(1)
Range("Sheet2!C" & CStr(nt)) = nt
Range("Sheet2!D" & CStr(nt)) = nf
2: If Abs(z(0) - x(0)) < n * rho * z(0) Then
test = False
Else
test = True
End If
If Not IsMissing(alpha) Then
If x(0) < alpha Then test = False
End If
3: If n = 1 Or nt = 1048576 Then test = False
End Function

When executing the test function, the values of x(0) = F ( 1

) , x(1) =

1
, nt and nf are put into the nt-th row on the Sheet2 worksheet.

439
 Chapter 6. Numerical Methods for Nonlinear Programming

The sense of the last two variables follows:

nt is the current number of the iterations, i.e., of the test function calls;
nf is the current number of the calculated values of objective function
F ( 1 ) , i.e., of the func function calls.
In other words, the nt variable is the counter of the test function calls, the
nf variable is the counter of the func function calls. The Excel row number
may be considered as the iteration number (we will use this in Section 6.7).
The mini subroutine execution is terminated when the relative change of
the F ( 1 ) function becomes less than ρ = 10-6: conditional operator 2 (where n
is equal to unity) includes variables z(0) and x(0), i.e., the F ( 1
) function
values at the iteration’s beginning and end, respectively.
Due to operator 3, the mini subroutine execution also terminates when the
number of arguments is equal to one or the Sheet2 worksheet does not contain
empty rows. We will encounter operator
If n = 1 Or nt = 1048576 Then test = False
more than once in this chapter.
According to (6.9), objective function F ( 1
) has the following form:

2 4 V
F( 1
) A( 1
) 2 1
2 . (6.12)
2
1

2
We made substitution r 1
in order to have only non-negative values of the r
radius in the course of the A area minimization. The tin can volume is given in
the func subroutine; it is equal to V = 1000.
The initial approximation of the minimum point, x(1) = 1 , and initial step
ss(1, 1) = s11 are respectively taken from cells G1 and G2 on the Sheet2
worksheet. The sign of s11 determines the direction of the initial step, and | s11 |
determines its size. Before the mini subroutine call (operator 1), the initial
value of objective function x(0) = F ( 1 ) must be defined. For that, operator 0
is used, which calculates the x(0) value corresponding to the x(1) value by
means of the func subroutine.
Fig. 6.16 shows the Sheet2 worksheet (a) before and (b) after the execution
of code Listing 6.3. According to Fig. 6.16a, the initial value of the 1 variable
equals 1, initial step s11 equals 0.01.

440
 6.4. Examples of using the minimization methods

According to Fig. 6.16b, only one iteration was performed for the minimiza-
tion of function (6.12) because it is a function of one variable (see operator 3);
the iteration result is in cells A1 and B1. The final value of 1 (which is in the
B1 cell) is equal to 2.328087. In this case, the radius of the tin can base,
2
r 1
= 5.41999, is close to the optimal value, r = 5.41926, calculated
according to formula (6.11). During the search for the minimum point, 13 values
of the objective function were calculated. The final value of the objective
function (which is in the A1 cell) is equal to A = 553.5811.

Fig. 6.16. The Sheet2 worksheet (a) before and (b) after minimization of F ( 1
)

We see conditional operator 2 in code Listing 6.3. It can be replaced with one
of the following two assignment operators with logical expression in the right-
hand side:

test = Not Abs(z(0) - x(0)) < n * rho * z(0)

or

test = Abs(z(0) - x(0)) >= n * rho * z(0)

441
 Chapter 6. Numerical Methods for Nonlinear Programming

The Rosenbrock function minimization.

To find the minimum point of nonlinear function (6.1) by the coordinate-
descent method, we change the text of program main, subroutine func and
function test as follows:
Listing 6.4
Dim nf As Long 'counter of calls of func
Dim nt As Long 'counter of calls of test

Sub main()
Dim x() As Double
Dim ss() As Double
Dim n As Byte
Dim i As Byte, j As Byte
Dim d As Double
n = 2 'number of variables
ReDim x(n)
ReDim ss(1 To n, 1 To n)
For j = 1 To n
x(j) = Worksheets("Sheet2").Cells(1, 6 + j)
For i = 1 To n
ss(i, j) = Worksheets("Sheet2"). _
Cells(1 + i, 6 + j)
Next i
Next j
For j = 1 To n
d = 0
For i = 1 To n
d = ss(i, j) ^ 2 + d
Next i
If d = 0 Then
Range("Sheet2!A1").Value = _
"You must increase" & Str(j) & _
"-th initial step"
End
End If
Next j
nf = 0
nt = 0
Call func(n, x) 'it must be before minimization
1: Call mini(n, x, ss, 1E-6, 1E-6)
End Sub

442
 6.4. Examples of using the minimization methods

Sub func(ByVal n, ByRef x() As Double)

nf = nf + 1
x(0) = 100 * (x(2) - x(1) ^ 2) ^ 2 + _
(1 - x(1)) ^ 2
End Sub

Function test(ByVal n, ByRef x() As Double, _

ByRef z() As Double, ByVal rho, Optional alpha) _
As Boolean
Dim j As Byte
nt = nt + 1
For j = 0 To n
Worksheets("Sheet2").Cells(nt, j + 1) = x(j)
Next j
Worksheets("Sheet2").Cells(nt, n + 2) = nt
Worksheets("Sheet2").Cells(nt, n + 3) = nf
2: If Abs(z(0) - x(0)) < n * rho * z(0) Then
test = False
Else
test = True
End If
If Not IsMissing(alpha) Then
3: If x(0) < alpha Then test = False
End If
If n = 1 Or nt = 1048576 Then test = False
End Function

According to conditional operators 2 and 3, the test function returns

False into the mini subroutine when at least one condition is satisfied of the
following two:
the relative change of objective function F ( x1 , x2 ) on one iteration
becomes less than 2ρ = 2·10-6 (the number of arguments is equal to 2);
the objective function value becomes less than α = 10-6.
In this case, the search of the minimum point of the Rosenbrock function termi-
nates.
When executing the test function, the values of x(0) = F ( x1 , x2 ) ,
x(1) = x1 , x(2) = x 2 and the current values of nt and nf (that is, the num-
bers of calls of test and func, respectively) are put into cells on the Sheet2
worksheet.

443
 Chapter 6. Numerical Methods for Nonlinear Programming

Fig. 6.17 shows worksheet Sheet2 (a) before and (b, c) after executing code
Listing 6.4.
Coordinates x(1) = x1 , x(2) = x 2 of the initial approximation of the
minimum point and also initial steps ss(1, 1) = s11 , ss(2, 2) = s 22 are
respectively taken from ranges G1:H1 and G2:H3 on worksheet Sheet2
(Fig. 6.17a). Before calling the mini subroutine, the initial value of the
objective function is calculated by calling the func subroutine. The result of
the Rosenbrock function minimization is located in range A2066:C2066
(Fig. 6.17c).

Fig. 6.17. The Sheet2 worksheet (a) before and

(b, c) after the execution of code Listing 6.4

444
 6.4. Examples of using the minimization methods

According to Fig. 6.17a, the initial values of the variables equal x10 = -5.5,

x 20 = 0.5 (we used these initial values in Section 6.1 when minimizing the
Rosenbrock function by the Solver add-in), initial steps s11 and s 22 are equal to
0.01, and s12 = s 21 = 0.
Fig. 6.17b and 6.17c show the results of the initial and final iterations.
According to Fig. 6.17c, the result of the execution of code Listing 6.4 is
x1 = x 2 = 1, and during the execution:
2066 iterations were performed;
16781 values of objective function (6.1) were calculated.
Further, we will consider the Powell minimization method. According to
Fig. 6.18 (similar to Fig. 6.17c), the method usage reduces:
the number of iterations to 5;
the number of the calculated values of objective function (6.1) to 376.

445
 Chapter 6. Numerical Methods for Nonlinear Programming

6.5. The Powell minimization method

Let x ( x1 , x2 , ..., xn ) be a point of the n-dimensional space, n ≥ 2. We

have to find the unique minimum point of non-negative and continuous function
F (x) , which has continuous partial derivatives F / x1 , F / x2 , …, F / xn .
As mentioned earlier, the minimization of the F (x) function is an iterative
process. According to the Powell method (based on a good theory [14]), each
iteration consists of successive minimizations of the F (x) function along direc-

tions S 1 , S 2 , ..., S n (which are defined by the S matrix), at that, a set of direc-
tions for the next iteration is formed. More precisely, one iteration includes the
following seven stages.
1. For i = 1, 2, …, n, vectors x i are defined according to recurrence formula
xi xi 1
i
Si ,

where x 0 is the initial approximation of the minimum point for the given itera-
tion, i
is the minimum point of the following function of one variable:

fi ( ) F (xi 1
Si ) .
2. The maximum change in the F (x) function,

max {F (xi 1
) F (xi )} ,
1 i n
is defined. Integer k is defined as the serial number of the direction along which
this change has happened.
3. Values Fa F ( 2x n x0 ) , Fb F ( x n ) , Fc F (x0 ) are defined.
4. If Fa Fc 0 and
2
Fc Fb
2( Fc 2 Fb Fa ) ,
Fa Fc

446
 6.5. The Powell minimization method

then the 5th item is fulfilled. Otherwise, x n is assigned to x 0 , and the jump to
the 7th item is performed without changing the set of minimization directions.
5. Two vectors are defined according to formulas
Sn 1
xn x0 ,
x xn Sn 1
,

where is the minimum point of the following function of one variable:

fn 1
( ) F (x n Sn 1
).
6. The new set of the n minimization directions (matrix S ) for the next
iteration is formed as follows:
S 1 , ..., S k 1
, Sk 1
, ..., S n , S n 1
,
where k is the integer defined in the 2nd item. Assignment x 0 x is per-
formed.
7. If none of the termination conditions of the F (x) function minimization
is satisfied, the following iteration is performed.
The method of Section 6.2 is used for minimizing f1 ( ) , f 2 ( ) , …, f n ( ) ,
fn 1
( ).
The declaration of the minim subroutine, realizing the Powell method, is
given below. We enter it into Module18 of the BookNM workbook.
Listing 6.5
Sub minim(ByVal n, ByRef x() As Double, _
ByRef ss() As Double, ByVal rho, Optional alpha)
Dim fa As Double, fb As Double, fc As Double, _
fd As Double
Dim a As Double, b As Double, c As Double, _
d As Double, e As Double, dm As Double
Dim s() As Double: ReDim s(n)
Dim y() As Double: ReDim y(n)
Dim z() As Double: ReDim z(n)
Dim sss() As Double: ReDim sss(n, n)
Dim i As Byte, j As Byte, k As Byte, m As Byte
For j = 1 To n
For i = 1 To n
sss(i, j) = ss(i, j)
Next i

447
 Chapter 6. Numerical Methods for Nonlinear Programming

Next j
m1: For i = 0 To n
y(i) = x(i)
z(i) = x(i)
Next i
dm = 0
For j = 1 To n 'j - number of descent direction
For i = 1 To n
s(i) = ss(i, j)
Next i
d = 0
For i = 1 To n
d = s(i) ^ 2 + d
Next i
If d = 0 Then
For i = 1 To n
s(i) = sss(i, j)
Next i
Else
For i = 1 To n
sss(i, j) = s(i)
Next i
End If
'Finding uncertainty segment:
fa = y(0): fb = y(0): fc = y(0)
a = 0: b = 0: c = 0
d = 1: e = 1
s1: For i = 1 To n
x(i) = y(i) + d * s(i)
Next i
Call func(n, x)
If Not IsMissing(alpha) Then
If x(0) < alpha Then GoTo m3
End If
fd = x(0)
If fd < fa Then
fc = fb: fb = fa: fa = fd
c = b: b = a: a = d
d = 2 * d + e
GoTo s1
Else
If fa = fb Then

448
 6.5. The Powell minimization method

fb = fd
b = d
e = -2 * d
d = e
GoTo s1
End If
End If
fc = fb: fb = fa: fa = fd
c = b: b = a: a = d
d = (a + b) * 0.5
For i = 1 To n
x(i) = y(i) + d * s(i)
Next i
Call func(n, x)
If Not IsMissing(alpha) Then
If x(0) < alpha Then GoTo m3
End If
fd = x(0)
'Reducing uncertainty segment:
s2: If Abs(fb - fd) < rho * fb Then GoTo s0
If (c - d) * (d - b) < 0 Then
If fd < fb Then
fc = fb: fb = fd
c = b: b = d
Else
fa = fd
a = d
End If
Else
If fd < fb Then
fa = fb: fb = fd
a = b: b = d
Else
fc = fd
c = d
End If
End If
d = fa * (b - c) + fb * (c - a) + fc * (a - b)
If d = 0 Then GoTo s0
d = (fa * (b * b - c * c) + _
fb * (c * c - a * a) + _
fc * (a * a - b * b)) / (2 * d)

449
 Chapter 6. Numerical Methods for Nonlinear Programming

For i = 1 To n
x(i) = y(i) + d * s(i)
Next i
Call func(n, x)
If Not IsMissing(alpha) Then
If x(0) < alpha Then GoTo m3
End If
fd = x(0)
GoTo s2
s0: d = y(0) - x(0)
If d > dm Then
dm = d
m = j
End If
For i = 1 To n
ss(i, j) = x(i) - y(i)
y(i) = x(i)
Next i
y(0) = x(0)
Next j
If n = 1 Then GoTo m3
'Last descent:
For i = 1 To n
x(i) = 2 * y(i) - z(i)
Next i
Call func(n, x)
If Not IsMissing(alpha) Then
If x(0) < alpha Then GoTo m3
End If
fa = x(0): fb = y(0): fc = z(0)
a = fa - fc
If a >= 0 Then GoTo m2
a = (fc - fb - dm) / a
If 2 * (fc - 2 * fb + fa) * a ^ 2 >= dm Then _
GoTo m2
For j = m To n - 1
k = j + 1
For i = 1 To n
ss(i, j) = ss(i, k)
Next i
Next j
For i = 1 To n

450
 6.5. The Powell minimization method

s(i) = y(i) - z(i)

Next i
'Finding uncertainty segment:
fa = y(0): fb = y(0): fc = y(0)
a = 0: b = 0: c = 0
d = 1: e = 1
h1: For i = 1 To n
x(i) = y(i) + d * s(i)
Next i
Call func(n, x)
If Not IsMissing(alpha) Then
If x(0) < alpha Then GoTo m3
End If
fd = x(0)
If fd < fa Then
fc = fb: fb = fa: fa = fd
c = b: b = a: a = d
d = 2 * d + e
GoTo h1
Else
If fa = fb Then
fb = fd
b = d
e = -2 * d
d = e
GoTo h1
End If
End If
fc = fb: fb = fa: fa = fd
c = b: b = a: a = d
d = (a + b) * 0.5
For i = 1 To n
x(i) = y(i) + d * s(i)
Next i
Call func(n, x)
If Not IsMissing(alpha) Then
If x(0) < alpha Then GoTo m3
End If
fd = x(0)
'Reducing uncertainty segment:
h2: If Abs(fb - fd) < rho * fb Then GoTo h0
If (c - d) * (d - b) < 0 Then

451
 Chapter 6. Numerical Methods for Nonlinear Programming

If fd < fb Then
fc = fb: fb = fd
c = b: b = d
Else
fa = fd
a = d
End If
Else
If fd < fb Then
fa = fb: fb = fd
a = b: b = d
Else
fc = fd
c = d
End If
End If
d = fa * (b - c) + fb * (c - a) + fc * (a - b)
If d = 0 Then GoTo h0
d = (fa * (b * b - c * c) + _
fb * (c * c - a * a) + _
fc * (a * a - b * b)) / (2 * d)
For i = 1 To n
x(i) = y(i) + d * s(i)
Next i
Call func(n, x)
If Not IsMissing(alpha) Then
If x(0) < alpha Then GoTo m3
End If
fd = x(0)
GoTo h2
h0: For i = 1 To n
ss(i, n) = x(i) - y(i)
y(i) = x(i)
Next i
y(0) = x(0)
'Writing result of descents into array x:
m2: For i = 0 To n
x(i) = y(i)
Next i
'Checking condition of minimization termination:
m3: If Not IsMissing(alpha) Then
If test(n, x, z, rho, alpha) Then GoTo m1

452
 6.5. The Powell minimization method

Else
If test(n, x, z, rho) Then GoTo m1
End If
End Sub

The minim subroutine parameters have the same sense as the corresponding
parameters of the mini subroutine (p. 435). However, two-dimensional array
ss may be non-diagonal.
In the minim subroutine, elements ss(1, j), ss(2, j), ..., ss(n, j) of
the ss array, corresponding to the j-th column ( S j ) of the S matrix, define the
j-th descent direction and the initial step along this direction, which equals
sj s12j s22 j ... 2
snj , j = 1, 2, ..., n. The descent directions, defined by the
ss array (matrix S ), is generally changing from iteration to iteration during the
F ( x1 , x2 , ..., xn ) function minimization.
Let us test the new subroutine by means of the Rosenbrock function. For this
purpose, we replace line

1: Call mini(n, x, ss, 1E-6, 1E-6)

with line

1: Call minim(n, x, ss, 1E-6, 1E-6)

in Listing 6.4. We leave subroutine func, function test and the initial data
without change.
The result of the Rosenbrock function minimization by the Powell method is
located in cells A5:C5 (Fig. 6.18).

Fig. 6.18. The Sheet2 worksheet upon termination of the code execution

453
 Chapter 6. Numerical Methods for Nonlinear Programming

We see that the obtained minimum point is the same as when using the
Solver add-in and coordinate-descent method: x1 = x 2 = 1. The minimization
of function (6.1) requires:
5 iterations of the Powell method (Fig. 6.18);
21 iterations of Solver (p. 427);
2066 iterations of the coordinate-descent method (p. 445).
Thus, the Powell method is more efficient than the Solver add-in and coordi-
nate-descent method for the Rosenbrock function. This is due to the following:
1) minimum point x (1, 1) is located in the ravine (p. 422);
2) in the Powell method, the initial descent directions, which are parallel to
the x1 and x 2 axes, are being converted to the descent directions, S 1 and S 2 ,
oriented along the ravine bottom (Fig. 6.19).

Fig. 6.19. The level curves of the Rosenbrock function in the neighborhood
of the minimum point and vectors S 1 and S 2 for the 4th iteration

454
 6.5. The Powell minimization method

The iteration results, ( x11 , x12 ) , ( x12 , x22 ) , ( x13 , x23 ) , ( x14 , x24 ) , …, approach

minimum point x (1, 1) along the ravine bottom. This assertion follows from
Fig. 6.18 and 6.19.

455
 Chapter 6. Numerical Methods for Nonlinear Programming

6.6. Determining the equilibrium

state of a four-spring system

Below, we will determine the equilibrium state of the mechanical system

depicted in Fig. 6.20. This system is formed by four weightless springs of the
same length, λ = 2 meter (without a load), but with different elastic constants.
The springs are located in the plane of the paper with axes x1 and x 2 . One ends
of the springs are fastened together, the other ends are attached to tops A1, A2,
A3, A4 of an imaginary square with the 2-meter side.

Fig. 6.20. The four-spring system: the x 3 axis is directed “towards us”
and passes through the springs junction point with zero coordinates

The outside force, vector f directed along the x 3 axis (for example, the force
of gravity), acts on body M attached to the springs junction. The value of this
force, f | f | , is given, and this value is negative if vector f and axis x 3 do
not coincide in direction (according to the common practice in physics), i.e., the
vector is directed “from us” (Fig. 6.20). The elastic constants of the springs equal
K 1 , K 2 , K 3 , K 4 , respectively.

456
 6.6. Determining the equilibrium state of a four-spring system

We have to find coordinates x1 , x 2 , x 3 of the springs junction point in the

equilibrium state, which is the result of a damped oscillation of the mechanical
system. The damping occurs, for example, because of the air resistance: the
resistance force is proportional to the velocity of body M and is directed opposite
to the velocity vector.
According to the principle of minimum potential energy of the elasticity the-
ory, the work of the elasticity forces and of the outside force for relocation of the
M body from the origin of coordinates to the point with coordinates x1 , x 2 , x 3
(that is, the potential energy of body M) assumes its minimum value in the equi-
librium state. We will use this principle.
Up to a constant, the potential energy of body M is equal to
2
4 Km
F ( x1 , x 2 , x3 ) ( x1 a m1 ) 2 ( x2 am2 ) 2 ( x3 a m3 ) 2
m 1
2
f x3 C, (6.13)
where am1 , am 2 are the first two coordinates of point Am , which are equal to ±1,
am 3 = 0 is the third coordinate of point Am (m = 1, 2, 3, 4), C is a positive con-
stant introduced to ensure the positivity of function F ( x1 , x 2 , x3 ) . Difference

( x1 a m1 ) 2 ( x2 am2 ) 2 ( x3 a m3 ) 2 is the change in the length of

the m-th spring.
For the minimization of function (6.13), initially, we use the coordinate-
descent method. The corresponding code is given below.
Listing 6.6
Dim nf As Long 'counter of calls of func
Dim nt As Long 'counter of calls of test
Dim K1 As Double, K2 As Double
Dim K3 As Double, K4 As Double
Dim f As Double, C As Double

Sub main()
Dim x() As Double
Dim ss() As Double
Dim n As Byte, i As Byte, j As Byte
Dim d As Double
K1 = Range("Sheet2!G5").Value
K2 = Range("Sheet2!H5").Value

457
 Chapter 6. Numerical Methods for Nonlinear Programming

K3 = Range("Sheet2!I5").Value
K4 = Range("Sheet2!J5").Value
f = Range("Sheet2!G6").Value
C = Range("Sheet2!H6").Value
n = 3 'number of variables
ReDim x(n)
ReDim ss(1 To n, 1 To n)
For j = 1 To n
1: x(j) = Worksheets("Sheet2").Cells(1, 6 + j)
For i = 1 To n
2: ss(i, j) = Worksheets("Sheet2"). _
Cells(1 + i, 6 + j)
Next i
Next j
For j = 1 To n
d = 0
For i = 1 To n
d = ss(i, j) ^ 2 + d
Next i
If d = 0 Then
Range("Sheet2!A1").Value = _
"You must increase" & Str(j) & _
"-th initial step"
End
End If
Next j
nf = 0
nt = 0
Call func(n, x) 'it must be before minimization
3: Call mini(n, x, ss, 1E-6)
End Sub

Sub func(ByVal n, ByRef x() As Double)

Const lambda As Double = 1.414213562
nf = nf + 1
x(0) = K1 / 2 * (Sqr((x(1) - 1) ^ 2 + _
(x(2) - 1) ^ 2 + x(3) ^ 2) - lambda) ^ 2 + _
K2 / 2 * (Sqr((x(1) + 1) ^ 2 + _
(x(2) - 1) ^ 2 + x(3) ^ 2) - lambda) ^ 2 + _
K3 / 2 * (Sqr((x(1) + 1) ^ 2 + _
(x(2) + 1) ^ 2 + x(3) ^ 2) - lambda) ^ 2 + _
K4 / 2 * (Sqr((x(1) - 1) ^ 2 + _

458
 6.6. Determining the equilibrium state of a four-spring system

(x(2) + 1) ^ 2 + x(3) ^ 2) - lambda) ^ 2 - _

f * x(3) + C
End Sub

Function test(ByVal n, ByRef x() As Double, _

ByRef z() As Double, ByVal rho, Optional alpha) _
As Boolean
Dim j As Byte
nt = nt + 1
For j = 0 To n
Worksheets("Sheet2").Cells(nt, j + 1) = x(j)
Next j
Worksheets("Sheet2").Cells(nt, n + 2) = nt
Worksheets("Sheet2").Cells(nt, n + 3) = nf
If Abs(z(0) - x(0)) < n * rho * z(0) Then
test = False
Else
test = True
End If
If Not IsMissing(alpha) Then
If x(0) < alpha Then test = False
End If
If n = 1 Or nt = 1048576 Then test = False
End Function

The test function returns False when the relative change of function
(6.13) on one iteration becomes less than 3ρ = 3·10-6 (the number of arguments is
equal to 3). In this case, the search of the minimum point terminates.
When executing the test function, the values of x(0), x(1), x(2),
x(3), nt and nf are put into cells on the Sheet2 worksheet.
By means of operators 1 and 2, coordinates x(1), x(2), x(3) of the initial
approximation of the minimum point and initial steps ss(1, 1), ss(2, 2),
ss(3, 3) are respectively taken from ranges G1:I1 and G2:I4 on the Sheet2
worksheet. Assignment operators 1 and 2 interpret the contents of the empty
cells as zero.
Elastic constants K 1 , K 2 , K 3 , K 4 of the springs (in units of N / m) are
respectively taken from cells G5, H5, I5, J5. The value of force f (in newtons) is
taken from G6; constant C is taken from H6.
According to Fig. 6.21a:
the initial values of variables x1 , x 2 , x 3 are equal to zero;

459
 Chapter 6. Numerical Methods for Nonlinear Programming

initial steps s11 , s 22 , s33 are equal to 0.01;

the elastic constants have the following values: K 1 = 100, K 2 = 200,
K 3 = 300, K 4 = 400;
the force: f = -800;
the constant: C = 10000.
According to Fig. 6.21b:
the springs junction point has coordinates x1 = - 0.01278, x 2 = - 0.33154,

x3 = -1.96816 in the equilibrium state (the coordinates are in meters);

the number of iterations equals 4;
88 values of objective function (6.13) were calculated during the minimi-
zation.

Fig. 6.21. The Sheet2 worksheet (a) before and (b) after the code execution

460
 6.6. Determining the equilibrium state of a four-spring system

The use of the Powell method for the minimization of function (6.13) is not
difficult. For this purpose, we have to replace mini with minim in operator 3.
Upon the change in the minimization method, only the number of the calcu-
lated values of objective function (6.13) changes markedly — from 88 to 95.
The above minimization problem is the only one in this chapter whose solu-
tion by the Powell method is not more efficient than by the coordinate-descent
method.

461
 Chapter 6. Numerical Methods for Nonlinear Programming

6.7. Minimization with nonlinear constraints

Minimizing nonlinear function x0 F ( x1 , x2 , ..., xn ) with constraints is

a frequently encountered problem. Often the following m inequalities play the
role of these constraints:
C1 ( x1 , x2 , ..., xn ) 0 ,
C2 ( x1 , x2 , ..., xn ) 0, (6.14)
. . . . . .
Cm ( x1 , x2 , ..., xn ) 0 ,
where C i ( x1 , x 2 , ..., x n ) is a given dependence, 1 ≤ i ≤ m. If this dependence is
nonlinear, the corresponding inequality is called a nonlinear constraint. Later we
will consider a constraint of the equality type.
In the case of simple dependences Ci ( x1 , x2 , ..., xn ) , 1 ≤ i ≤ m, the formu-
lated minimization problem can be solved by replacing the variables.
For n = 1, we already replaced the variable for solving the task of optimizing
a tin can in Section 6.4. In order to have only non-negative values of radius r in
the course of minimization of area A, we minimized a function with argument
2
1
related with r as r 1
, instead of minimizing the A(r ) function.
For n > 1, we will consider the variable replacement method on an example
of Rosenbrock function (6.1) with constraints x1 1 and | x 2 | 0.7 , which can
be written in form (6.14),
1 x1 0, (6.15)
0 .7 | x 2 | 0. (6.16)
We introduce new variables 1
and 2
defined by equations
2
x1 1 1
, (6.17)
x2 0.7 sin 2
. (6.18)

462
 6.7. Minimization with nonlinear constraints

It is visible that variables x1 and x 2 satisfy the required inequalities, (6.15) and
(6.16), for all values of 1
and 2
from to . Substituting expressions
(6.17) and (6.18) into formula (6.1), we get the following objective function:
2 2 2 4
G( 1
, 2
) 100[0.7 sin 2
(1 1
) ] 1
. (6.19)

After obtaining the minimum point of this function, χ ( 1

, 2
) , the required

values, x1 and x 2 , must be calculated by means of formulas (6.17) and (6.18).

The text of program main, subroutine func and function test, intended
for minimizing function (6.1) with constraints (6.15) and (6.16), has the follow-
ing form:
Listing 6.7
Dim nf As Long 'counter of calls of func
Dim nt As Long 'counter of calls of test

Sub main()
Dim x() As Double
Dim ss() As Double
Dim n As Byte
Dim i As Byte, j As Byte
Dim d As Double
n = 2 'number of variables
ReDim x(n)
ReDim ss(1 To n, 1 To n)
For j = 1 To n
x(j) = Worksheets("Sheet2").Cells(1, 6 + j)
For i = 1 To n
ss(i, j) = Worksheets("Sheet2"). _
Cells(1 + i, 6 + j)
Next i
Next j
For j = 1 To n
d = 0
For i = 1 To n
d = ss(i, j) ^ 2 + d
Next i
If d = 0 Then
Range("Sheet2!A1").Value = _
"You must increase" & Str(j) & _

463
 Chapter 6. Numerical Methods for Nonlinear Programming

"-th initial step"

End
End If
Next j
nf = 0
nt = 0
Call func(n, x) 'it must be before minimization
1: Call minim(n, x, ss, 1E-6, 1E-6)
2: Worksheets("Sheet2").Cells(nt + 1, 2) = _
1 - x(1) ^ 2
3: Worksheets("Sheet2").Cells(nt + 1, 3) = _
0.7 * Sin(x(2))
End Sub

Sub func(ByVal n, ByRef x() As Double)

Dim x1 As Double, x2 As Double
nf = nf + 1
x1 = 1 - x(1) ^ 2
x2 = 0.7 * sin(x(2))
x(0) = 100 * (x2 - x1 ^ 2) ^ 2 + x(1) ^ 4
End Sub

Function test(ByVal n, ByRef x() As Double, _

ByRef z() As Double, ByVal rho, Optional alpha) _
As Boolean
Dim j As Byte
nt = nt + 1
For j = 0 To n
Worksheets("Sheet2").Cells(nt, j + 1) = x(j)
Next j
Worksheets("Sheet2").Cells(nt, n + 2) = nt
Worksheets("Sheet2").Cells(nt, n + 3) = nf
If Abs(z(0) - x(0)) < n * rho * z(0) Then
test = False
Else
test = True
End If
If Not IsMissing(alpha) Then
If x(0) < alpha Then test = False
End If
If n = 1 Or nt = 1048576 Then test = False
End Function

464
 6.7. Minimization with nonlinear constraints

From code Listing 6.4 for solving the minimization problem without con-
straints, the above code differs in the following:
in operators of the func subroutine because it is now intended for calcu-
lating values of objective function G ( 1 , 2 ) according to formula (6.19)
instead of (6.1);
in operator 1 because we use the Powell method in the new program;
in the presence of operators 2 and 3, which correspond to formulas (6.17)
and (6.18), respectively.
After obtaining the values of 1
and 2
, operators 2 and 3 calculate the

values of x1 and x 2 , i.e., the solution of the minimization problem for the
Rosenbrock function with constraints (6.15) and (6.16).
Fig. 6.22 shows the Sheet2 worksheet upon termination of the code execu-
tion.

Fig. 6.22. The worksheet upon termination of the code execution

According to Fig. 6.22:

the initial approximation of the minimum point, defined by the values of
0 0
cells G1 and H1, has coordinates 1
= -3 and 2
= 5;
the initial steps, defined by the values of cells G2:H3, are directed along
the 1
and 2 axes and are equal to 0.01;

465
 Chapter 6. Numerical Methods for Nonlinear Programming

the coordinates of the minimum point of function G ( 1

, 2
) are

1
= - 0.403 and 2
= 7.854;

the required values are x1 = 0.837 and x 2 = 0.7; they are located in cells
B14 and C14.
When replacing minim with mini in operator 1 (that is, when using the
coordinate-descent method) the number of iterations increases from 13 to 254,
and the number of the calculated values of objective function G ( 1 , 2 )
increases from 505 to 2372.
The application of the variable replacement method, which was considered,
is very limited. For minimizing nonlinear function F ( x1 , x2 , ..., xn ) with con-
straints (6.14), the penalty function method is more universal. According to this
method, functions
Gk ( x) F ( x) 2 k [ D1 ( x) D2 ( x) ... Dm ( x)] (6.20)
of vector argument x ( x1 , x 2 , ..., x n ) , k ≥ 0, are being sequentially mini-
mized for k = 0, 1, 2, ... The value of k determines the “weight” of sum
D1 ( x) D2 ( x) ... Dm ( x) , i.e., its contribution to the G k (x ) function; m is
the number of the constraints.
We use formula
Di (x) min{Cip (x), 0} ,
where the p power is an odd natural number, for example 3, 1 ≤ i ≤ m.
According to the last formula:
Di (x ) 0 when Ci (x) 0 , i.e., the i-th inequality of (6.14) is satisfied;

Di (x) Cip (x) 0 when Ci (x) 0 , i.e., the i-th inequality of (6.14) is
unsatisfied.
We can combine functions (6.20) into the following single function with
additional argument x n 1 :
G ( x , xn 1
) F ( x) xn 1
[ D1 ( x) D2 ( x) ... Dm ( x)] . (6.21)
The original problem with constraints is reduced to a sequence of problems
without constraints: we must find the minimum point of each of functions
G0 ( x ) G ( x, 1) , G1 ( x) G ( x, 2) , G2 ( x) G ( x, 4) , ..., Gk ( x) G (x, 2k ) , ...

466
 6.7. Minimization with nonlinear constraints

It is obvious that the obtained sequence of minimum points x 0 , x1 , x 2 , ...,

x k , ... converges to required solution x of the minimization problem for func-

tion F (x) with constraints (6.14).
For a fixed value of k, one of the considered methods for unconstrained
minimization (the coordinate-descent or Powell method) minimizes function
Gk (x) G (x, 2k ) , at that, minimum point x k 1
of Gk 1
( x) G (x, 2k 1
) is

used as the initial approximation of the x k minimum point of function

Gk (x) G (x, 2k ) .

We must specify initial approximation x 0 of minimum point x 0 of function

G0 ( x ) G ( x, 1) . The x 0 point may be considered as the initial approximation

of required x .
Functions (6.20) are called the penalty functions. Let also:
function (6.21) of form
G ( x1 , x2 , ..., xn , xn 1 ) F ( x) xn 1[ D1 ( x) D2 ( x) ... Dm ( x)]
be called the penalty function;
summand xn 1[ D1 ( x) D2 ( x) ... Dm ( x)] be called the penalty.
The penalty function method allows to solve the minimization problem for
Rosenbrock function (6.1) with constraint x12 x22 1 , which can be written in
form (6.14),
C1 ( x1 , x2 ) 0 , (6.22)
where
C1 ( x1 , x2 ) x12 x22 1 . (6.23)
To use the Powell method for minimizing function (6.1) with this constraint,
we must enter the following code into Module1.
Listing 6.8
Const DBL_MAX = 1E+308
Dim nf As Long 'counter of calls of func
Dim nt As Long 'counter of calls of test
Dim m As Byte
Dim y0 As Double, z0 As Double

467
 Chapter 6. Numerical Methods for Nonlinear Programming

Sub main()
Dim x() As Double
Dim ss() As Double
Dim n As Byte, i As Byte, j As Byte
Dim d As Double
n = 2 'number of variables
m = 1 'number of constraints
ReDim x(-1 To n + m)
ReDim ss(1 To n, 1 To n)
For j = 1 To n
x(j) = Worksheets("Sheet2").Cells(1, 6 + j)
For i = 1 To n
ss(i, j) = Worksheets("Sheet2"). _
Cells(1 + i, 6 + j)
Next i
Next j
For j = 1 To n
d = 0
For i = 1 To n
d = ss(i, j) ^ 2 + d
Next i
If d = 0 Then
Range("Sheet2!A1").Value = _
"You must increase" & Str(j) & _
"-th initial step"
End
End If
Next j
nf = 0
nt = 0
x(n + 1) = 1
Call func(n, x) 'it must be before minimization
z0 = x(0): y0 = DBL_MAX
1: Call minim(n, x, ss, 1E-3, 1E-6)
End Sub

Sub func(ByVal n, ByRef x() As Double)

Dim c As Double
nf = nf + 1
c = -x(1) ^ 2 - x(2) ^ 2 + 1
2: If c > = 0 Then
x(-1) = 0

468
 6.7. Minimization with nonlinear constraints

Else
x(-1) = x(3) * (-c ^ 3) 'penalty, p = 3
End If
x(0) = 100 * (x(2) - x(1) ^ 2) ^ 2 + _
(1 - x(1)) ^ 2 + x(-1)
End Sub

Function test(ByVal n, ByRef x() As Double, _

ByRef z() As Double, ByVal rho, Optional alpha) _
As Boolean
Dim j As Integer
nt = nt + 1
For j = -1 To n + m
Worksheets("Sheet2").Cells(nt, j + 2) = x(j)
Next j
Worksheets("Sheet2").Cells(nt, n + m + 3) = nf
If Abs(z0 - x(0)) < n * rho * z0 Then
If Abs(y0 - x(0)) < n * rho * y0 Then
test = False
Else
x(n + 1) = x(n + 1) * 2
If x(n + 1) > DBL_MAX Then
test = False
Else
test = True
End If
y0 = x(0)
Call func(n, x)
End If
Else
test = True
End If
z0 = x(0)
If Not IsMissing(alpha) Then
If x(0) < alpha Then test = False
End If
If n = 1 Or nt = 1048576 Then test = False
End Function

The results of the code execution are depicted in Fig. 6.23. As we see, initial
approximation x 0 , defined by the values of cells G1 and H1, has coordinates

469
 Chapter 6. Numerical Methods for Nonlinear Programming

x10 = -5.5 and x 20 = 0.5. According to the contents of cells G2:H3, the initial
steps are directed along the x1 and x 2 axes and equal 0.01.

Fig. 6.23. The Sheet2 worksheet upon termination of the code execution

Columns A and B contain the values of penalty x(-1) = x3 D1 ( x1 , x2 ) and

penalty function x(0) = G ( x1 , x2 , x3 ) , respectively. Columns C, D and E con-
tain the values of coordinates x(1) = x1 and x(2) = x 2 and variable x(3) =

x3 2 k . Column F contains the current value of nf, i.e., the current number of

the func subroutine calls. The coordinates of the obtained minimum point, x ,
are equal to x1 = 0.787 and x 2 = 0.62.

470
 6.7. Minimization with nonlinear constraints

When replacing minim with mini in operator 1 (that is, when using the
coordinate-descent method), the number of iterations increases from 26 to 153,
and the number of the calculated values of the objective function increases from
385 to 1313.
The penalty function method allows solving the minimization problem for the
F ( x1 , x2 , ..., xn ) function when the equality type constraints are present among
constraints (6.14); for example, C1 ( x1 , x2 , ..., xn ) 0 may be the first con-

straint. In this case, we use D1 (x) C1q (x) in expressions (6.20) and (6.21),
where power q is an even natural number, for example 2.
As an example of using this form of the penalty function method, we will
consider the minimization problem for Rosenbrock function (6.1) with constraint
x12 x22 1 , which can be written as follows:
C1 ( x1 , x2 ) 0,

where the left-hand side is defined by (6.23), i.e., C1 ( x1 , x2 ) x12 x22 1.

To minimize function (6.1) with this constraint, we must change the func
subroutine declaration in code Listing 6.8 as follows:
Listing 6.9
Sub func(ByVal n, ByRef x() As Double)
Dim c As Double
nf = nf + 1
c = -x(1) ^ 2 - x(2) ^ 2 + 1
2: If c = 0 Then x(-1) = 0 Else _
x(-1) = x(3) * c ^ 2 'penalty, q = 2
x(0) = 100 * (x(2) - x(1) ^ 2) ^ 2 + _
(1 - x(1)) ^ 2 + x(-1)
End Sub

The code execution results are depicted in Fig. 6.24.

As we see, initial approximation x 0 of the minimum point has coordinates
x10 = -5.5 and x 20 = 0.5, the initial steps are directed along the x1 and x 2 axes

and equal 0.01. The coordinates of the obtained minimum point, x , are equal to
x1 = 0.786 and x 2 = 0.618. When replacing the minim subroutine with mini,

471
 Chapter 6. Numerical Methods for Nonlinear Programming

the number of iterations increases from 18 to 67, the number of the calculated
values of the penalty function increases from 347 to 686.

Fig. 6.24. The Sheet2 worksheet after the code execution

Besides the above methods, the barrier function method is used for the
F (x) function minimization with nonlinear constraints. It differs from the penal-
ty function method in the form of the second summand in formula (6.20):
k
Gk (x) F ( x) 2 [1 / C1p (x) 1 / C2p (x) ... 1 / Cmp (x)] , (6.24)
where p is an even natural number. The G ( x, xn 1
) function has the following
form:
G(x, xn 1
) F ( x) xn 1 1[1 / C1p (x) 1 / C2p (x) ... 1 / Cmp (x)] . (6.25)

Summand xn 1 1[1 / C1p (x) 1 / C2p (x) ... 1 / Cmp (x)] is called the barrier; func-
tions (6.24) and (6.25) are called the barrier functions.
As we see, function (6.25) is close to (6.21) in form. Therefore, the new code
for minimizing function (6.1) with constraint (6.22), (6.23) by the Powell method
differs from Listing 6.8 only in the following operator:

2: x(-1) = 1 / x(3) * (1 / c ^ 2) 'barrier, p = 2

472
 6.7. Minimization with nonlinear constraints

For constraints in the form of inequalities (6.14), initial approximation x 0 of

the minimum point must satisfy these inequalities. For example, x 0 (0, 0)
may be when minimizing function (6.1) with constraint (6.22), (6.23) by the new
version of code Listing 6.8. Therefore, the method being considered is often
called the interior point method.
Fig. 6.25 shows the results of using the new version of code Listing 6.8 (with
minim in operator 1).

Fig. 6.25. The initial (a) and final (b) iterations

As we see in Fig. 6.25, initial approximation x 0 of the minimum point has

zero coordinates, the initial steps are directed along the x1 and x 2 axes and

equal 0.01. The coordinates of the obtained minimum point, x , are equal to
x1 = 0.786 and x2 = 0.616. When replacing minim with mini in operator 1,
the number of iterations increases from 60 to 314, and the number of the calcu-
lated values of the barrier function increases from 703 to 2552.
Let us solve one more minimization problem for Rosenbrock function (6.1),
when constraint x12 x22 1 is imposed on the minimum point. This inequality

can be written in form (6.22), where C1 ( x1 , x2 ) x12 x22 1 .

473
 Chapter 6. Numerical Methods for Nonlinear Programming

For minimizing function (6.1) with this constraint by the penalty function and
Powell methods, we use Listing 6.8 with minim in operator 1 and the following
func subroutine:
Listing 6.10
Sub func(ByVal n, ByRef x() As Double)
Dim c As Double
nf = nf + 1
c = x(1) ^ 2 + x(2) ^ 2 - 1
2: If c >= 0 Then
x(-1) = 0
Else
x(-1) = x(3) * (-c ^ 3) 'penalty, p = 3
End If
x(0) = 100 * (x(2) - x(1) ^ 2) ^ 2 + _
(1 - x(1)) ^ 2 + x(-1)
End Sub

Fig. 6.26 shows the results of using the last version of code Listing 6.8.

Fig. 6.26. The Sheet2 worksheet upon termination of the code execution

As we see in Fig. 6.26, initial approximation x 0 of the minimum point has

coordinates x10 = -5.5 and x 20 = 0.5, the initial steps are directed along the x1
and x 2 axes and equal 0.01. The coordinates of the obtained minimum point,

x , are equal to x1 = 1 and x 2 = 1.

The barrier function method gives an incorrect result.

474
 6.8. Minimization of the multimodal function

6.8. Minimization of the multimodal function

The search methods (the coordinate-descent and Powell methods) considered

above were developed for finding the minimum point of the so-called unimodal
function, x0 F ( x1 , x2 , ..., xn ) , that has a single local minimum. Application
of these methods to the multimodal function, with several local minima, gives
only one minimum point, which depends on the initial approximation of the
minimum point and on the directions and values of initial steps.
Let function F ( x1 , x2 , ..., xn ) be multimodal inside its domain, and all the
minima are to be found. The solution of this problem may be required for subse-
quent definition of the global minimum or for solving the maximin problem, i.e.,
for definition of the function’s maximum value among the local minima.
All local minima can be found if the initial approximation of the minimum
point is defined by means of the random-number generator and, at that, the code
execution is sufficiently long.
For example, let us consider function
F ( x1 , x2 ) (cos2 x1 cos 2 x2 )[100 ( x2 x12 ) 2 (1 x1 ) 2 ] (6.26)
in the rectangle given by inequalities
0.5 x1 5.5 ,
0 .5 x2 5.
To find the local minima of function (6.26) inside its domain (the above rec-
tangle), we use the following text of program main, subroutine func and func-
tion test:
Listing 6.11
Dim nf As Long 'counter of calls of func
Dim nt As Long 'counter of calls of test
Dim np As Long 'counter of minimum points

Sub main()
Dim x() As Double
Dim ss() As Double

475
 Chapter 6. Numerical Methods for Nonlinear Programming

Dim n As Byte
Dim i As Byte, j As Byte
Dim d As Double
Dim sec As Long
Dim min As Byte, hour As Byte
Dim st As String
Dim stp As Date
Dim np_lim As Long
n = 2 'number of variables
ReDim x(n)
ReDim ss(1 To n, 1 To n)
1: sec = Range("Sheet2!G4").Value
2: hour = sec \ 3600
3: min = (sec - hour * 3600) \ 60
4: sec = sec - hour * 3600 - min * 60
5: st = CStr(hour) & ":" & CStr(min) & ":" _
& CStr(sec)
6: stp = Now + TimeValue(st)
7: np_lim = Range("Sheet2!H4").Value
8: Randomize 'it must be before calling Rnd
9: np = 0
beg:
10: np = np + 1
11: x(1) = 0.5 + (5.5 - 0.5) * Rnd
12: x(2) = 0.5 + (5 - 0.5) * Rnd
For j = 1 To n
For i = 1 To n
ss(i, j) = Worksheets("Sheet2"). _
Cells(1 + i, 6 + j)
Next i
Next j
For j = 1 To n
d = 0
For i = 1 To n
d = ss(i, j) ^ 2 + d
Next i
If d = 0 Then
Range("Sheet2!A1").Value = _
"You must increase" & Str(j) & _
"-th initial step"
End
End If

476
 6.8. Minimization of the multimodal function

Next j
nf = 0
nt = 0
Call func(n, x) 'it must be before minimization
Call minim(n, x, ss, 1E-6, 1E-6) 'Powell method
13: For j = 0 To n
14: Worksheets("Sheet2").Cells(np, j + 1) = x(j)
15: Next j
16: Worksheets("Sheet2").Cells(np, n + 2) = nt
17: Worksheets("Sheet2").Cells(np, n + 3) = nf
18: If Now < stp And np < np_lim And np < 1048576 _
Then GoTo beg
End Sub

Sub func(ByVal n, ByRef x() As Double)

nf = nf + 1
x(0) = (Cos(x(1)) ^ 2 + Cos(x(2)) ^ 2) * _
(100 * (x(2) - x(1) ^ 2) ^ 2 + (1 - x(1)) ^ 2)
End Sub

Function test(ByVal n, ByRef x() As Double, _

ByRef z() As Double, ByVal rho, Optional alpha) _
As Boolean
Dim j As Byte
nt = nt + 1
If Abs(z(0) - x(0)) < n * rho * z(0) Then
test = False
Else
test = True
End If
If Not IsMissing(alpha) Then
If x(0) < alpha Then test = False
End If
End Function

From code Listing 6.4, intended for minimizing the Rosenbrock function,
the last code differs in operators of the func subroutine and presence of opera-
tors 1 — 18 in the main program. Let us consider the purpose of these addi-
tional operators.
Operator 1 assigns the specified limiting execution time (in seconds), being
in cell Sheet2!G4, to variable sec of the Long data type. Operators 2 — 5
form the st string of format "hh:mm:ss", in which hh is one or two digits

477
 Chapter 6. Numerical Methods for Nonlinear Programming

defining the number of hours, mm is the number of minutes, ss is the number of

seconds. Operator 6 determines the moment of the execution termination and
assigns it to variable stp (from word “stop”) of the Date data type.
Operator 7 assigns the limiting number of found minimum points to variable
np_lim of the Long data type. This number is taken from cell Sheet2!H4.
The Randomize operator (labeled by 8) prepares the built-in random-
number generator for work; operator 9 nullifies the np counter of the found
minimum points.
Operator 10, below label beg (from “beginning”), increases the np counter
by 1. Operators 11 and 12 define the coordinates of initial approximation x 0 of
the next minimum point by means of the Rnd function returning real numbers
uniformly distributed on segment [0, 1] .

The x 0 point falls into any place of rectangle 0.5 x1 5.5 and
0 .5 x2 5 with equal probability.
Below operator 12, we see familiar operators of the function minimization
by the Powell method. Upon termination of the minim subroutine execution,
operators 13 — 18 are performed.
Operators 13 — 17 put the result of searching the local minimum into the
empty row on the Sheet2 worksheet. The result includes:
x(0) — the value of function F ;
x(1), x(2) — the values of arguments x1 and x 2 ;
nt, nf — the numbers of the test and func calls required for finding
the local minimum.
Operator 18 performs the jump to label beg if the following three conditions
are satisfied simultaneously:
the execution time is less than the limiting time;
the number of found minimum points is less than the limiting number;
empty rows still stay on the Excel worksheet.
After the jump to label beg, the new x 0 point is defined randomly and the
minimization is repeated.
Fig. 6.27 shows the beginning of the Sheet2 worksheet after the code execu-
tion. The limiting execution time and the limiting number of found minimum
points are equal to 1 second and 100, respectively. During the code execution,
100 minimum points were calculated; coordinates of seven of them are given in
table “The minimization results” below. The 93 remaining points practically

478
 6.8. Minimization of the multimodal function

coincide with these seven points or are outside the domain defined by inequali-
ties 0.5 x1 5.5 and 0.5 x2 5 .

Fig. 6.27. The first 16 rows on the Sheet2 worksheet

after the code execution

The minimization results

Minimum’s number x0 F ( x1 , x2 ) x1 x2
1 0 1.57 1.57
2 0 4.71 1.57
3 0 4.71 4.71
4 0 1.57 4.71
5 0 1 1
6 0.397 2.13 4.54
7 0.732 1.88 3.55

When creating table “The minimization results”, we applied the Excel filter
to the worksheet of Fig. 6.27 as follows:
1) set the number format with two decimal places for columns B and C;
2) select columns A:E;
3) Data > Filter in area Sort & Filter (Fig. 6.28);

479
 Chapter 6. Numerical Methods for Nonlinear Programming

4) in the first row of the B column, fulfill Number Filters > Custom Filter by
means of the drop-down list;
5) in the Custom AutoFilter window opened, set the following: ≥ 0.5 and ≤
5.5 (Fig. 6.29);
6) click on the OK button;
7) in the first row of the C column, fulfill Number Filters > Custom Filter;
8) in the Custom AutoFilter window opened, set the following: ≥ 0.5 and ≤ 5
(Fig. 6.30);
9) click on the OK button.

Fig. 6.28. The first 16 rows on the Sheet2 worksheet

after starting the filter

The first four minima in table “The minimization results” are located at the
points, where the first multiplicand of objective function (6.26) is equal to zero:
cos2 x1 cos2 x2 0

480
 6.8. Minimization of the multimodal function

because / 2 = 1.57, 3 / 2 = 4.71, i.e., cos (1.57) = cos ( 4.71) = 0.

The 5th minimum has coordinates x1 = x 2 = 1, for which the second multi-
plicand of objective function (6.26) is equal to zero:
100 ( x2 x12 ) 2 (1 x1 ) 2 0 .
The 6th and 7th minima are not so obvious: objective function (6.26) has
nonzero values at points (2.13; 4.54) and (1.88; 3.55).

Fig. 6.29. Setting the constraints for x1 during the filter usage

Fig. 6.30. Setting the constraints for x 2 during the filter usage

481
 Chapter 6. Numerical Methods for Nonlinear Programming

Strictly speaking, the above method for minimizing the multimodal function
does not guarantee finding all local minima inside the function’s domain. How-
ever, the probability of this is close to unity if the code execution is sufficiently
long.
Often, we do not believe in the adequacy of the obtained solution of a mini-
mization problem, even when we know that the objective function is unimodal.
In this case, the use of this section’s method is a good idea.
We advise the reader to write a program for finding the global minimum of
function
F ( x1 , x2 ) f12 ( x1 ) f 22 ( x2 )
inside rectangle a1 x1 b1 and a2 x2 b2 , where f1 ( x1 ) and f 2 ( x2 ) are
functions from Appendix 4, segments [ a1 , b1 ] and [ a2 , b2 ] are the domains of
functions f1 ( x1 ) and f 2 ( x2 ) , respectively.
In addition, we advise the reader to use the random-number generator for
defining both the initial approximation of the minimum point and the initial steps
along the x1 and x 2 axes.
By the way, the last minimum in table “The minimization results” (with co-
ordinates x1 = 1.88 and x 2 = 3.55) is the solution of the following maximin
problem: to determine the maximum value of function (6.26) among the local
minima inside rectangle 0.5 x1 5.5 and 0.5 x2 5 . The global minimum
is equal to zero; it is reached at the first five minimum points.

482
 6.9. Minimization of the tabular function

6.9. Minimization of the tabular function

To demonstrate the possibilities of the numerical methods for nonlinear pro-

gramming, we used simple functions that can be differentiated analytically. For
example, we easily derived expressions (6.5) and (6.6) for the partial derivatives
of the Rosenbrock function. In this case, methods of the third chapter in book
[16], using the first and second partial derivatives of the objective function, are
effective for the minimization.
In practice, it is often necessary to minimize a function in tabular form or
a function whose values are calculated implicitly, for example, by solving any
equation. In these cases, the analytical differentiation of the function being min-
imized is impossible, and the search methods take on special significance.
We will consider the minimization of a positive tabular function of two vari-
ables. The values of arguments x1 and x 2 and of function x 0 F ( x1 , x 2 ) are
given in table Listing 6.12 and Fig. 6.31.
Listing 6.12
The positive tabular function of two variables

x1 x2 0 0.2 0.4 0.6 0.8

2 1.703125 1.529781 1.981909 3.548149 7.270101

2.2 1.673125 1.282539 1.514134 2.903724 6.656206
2.4 1.883125 1.23895 1.201694 2.35404 6.07493
2.6 2.333125 1.399014 1.04459 1.899096 5.526274
2.8 3.023125 1.762731 1.042822 1.538892 5.010238
3 3.953125 2.330101 1.196389 1.273429 4.526821
3.2 5.123125 3.101124 1.505292 1.102706 4.076024
3.4 6.533125 4.075801 1.969531 1.026724 3.657846
3.6 8.183125 5.25413 2.589105 1.045483 3.272287
3.8 10.07313 6.636113 3.364015 1.158982 2.919348
4 12.20313 8.221749 4.294261 1.367221 2.599029

483
 Chapter 6. Numerical Methods for Nonlinear Programming

Fig. 6.31. The Excel table with source data

According to Fig. 6.31:

cells B3:B13 contain values x1, i of argument x1 (0 ≤ i ≤ k, k = 10), i.e.,

grid nodes x1,0 , x1,1 , ..., x1, k on axis x1 ;

cells C2:G2 contain values x2, j of argument x 2 (0 ≤ j ≤ r, r = 4), i.e.,

grid nodes x2,0 , x2,1 , ..., x2, r on axis x 2 ;

cells C3:G13 contain the F ( x1 , x 2 ) function values.

The code intended for minimizing the F ( x1 , x 2 ) function follows:

Listing 6.13
Const DBL_MAX = 1E+308
Dim x1() As Double 'grid nodes on axis x1
Dim x2() As Double 'grid nodes on axis x2
Dim k As Integer 'number of segments on x1
Dim r As Integer 'number of segments on x2
Dim ff() As Double 'values of function F
Dim f1() As Double 'values of splines
Dim f2() As Double 'values of function of x2
Dim mm() As Double 'values of moments about x2
Dim m1() As Double 'values of moments about x1

484
 6.9. Minimization of the tabular function

Dim m2() As Double 'values of moments about x2

Dim nf As Long 'counter of calls of func
Dim nt As Long 'counter of calls of test
Dim no As Integer 'shift for output

Sub main()
Dim x(2) As Double
Dim ss(1 To 2, 1 To 2) As Double
Dim i As Integer, j As Integer
Dim min As Double
k = Selection.Rows.Count - 2
r = Selection.Columns.Count - 2
ReDim x1(k)
ReDim x2(r)
ReDim ff(k, r)
ReDim f1(k)
ReDim f2(r)
ReDim mm(k, r)
ReDim m1(k)
ReDim m2(r)
For i = 0 To k
x1(i) = Selection.Cells(2 + i, 1)
Next i
For j = 0 To r
x2(j) = Selection.Cells(1, 2 + j)
Next j
For i = 0 To k
For j = 0 To r
ff(i, j) = Selection.Cells(2 + i, 2 + j)
Next j
Next i
'Calculation of 2D array of moments about x2:
For i = 0 To k
For j = 0 To r
f2(j) = ff(i, j)
Next j
0: Call mos(0, r, x2, f2, 0, 0, 0, 0, m2)
For j = 0 To r
mm(i, j) = m2(j)
Next j
Next i
'Specifying initial approximation of minimum point:

485
 Chapter 6. Numerical Methods for Nonlinear Programming

min = DBL_MAX
For i = 0 To k
For j = 0 To r
If ff(i, j) < min Then
min = ff(i, j)
x(1) = x1(i)
x(2) = x2(j)
End If
Next j
Next i
no = i + 2
'Specifying initial steps:
ss(1, 1) = 0.01: ss(1, 2) = 0
ss(2, 1) = 0: ss(2, 2) = 0.01
'Searching minimum point:
nf = 0
nt = 0
Call func(2, x) 'it must be before minimization
Selection.Cells(no, 1) = x(0)
Selection.Cells(no, 2) = x(1)
Selection.Cells(no, 3) = x(2)
Call minim(2, x, ss, 1E-6) 'Powell method
'Outputting minimum point:
Selection.Cells(no + 1, 1) = x(0)
Selection.Cells(no + 1, 2) = x(1)
Selection.Cells(no + 1, 3) = x(2)
Selection.Cells(no + 1, 4) = nt
Selection.Cells(no + 1, 5) = nf
End Sub

Sub func(ByVal n, ByRef x() As Double)

Dim i As Integer
Dim j As Integer
nf = nf + 1
For i = 0 To k
For j = 0 To r
f2(j) = ff(i, j)
m2(j) = mm(i, j)
Next j
1: Call si(0, r, x2, f2, m2, x(2), f1(i))
Next i
2: Call mos(0, k, x1, f1, 0, 0, 0, 0, m1)

486
 6.9. Minimization of the tabular function

3: Call si(0, k, x1, f1, m1, x(1), x(0))

End Sub

Function test(ByVal n, ByRef x() As Double, _

ByRef z() As Double, ByVal rho, Optional alpha) _
As Boolean
Dim j As Byte
nt = nt + 1
If Abs(z(0) - x(0)) < n * rho * z(0) Then
test = False
Else
test = True
End If
If Not IsMissing(alpha) Then
If x(0) < alpha Then test = False
End If
End Function

We enter this code into Module1 of the BookNM workbook. The source data
are in the Excel table (Fig. 6.31); this table (range B2:G13) should be selected
before the code execution.
Let us consider the three main stages of the code execution.
First of all, the moments of cubic splines S 0 ( x 2 ) , S1 ( x 2 ) , . . . , S k ( x 2 ) ,
defined by grid functions f 0 ( x2 ) F ( x1,0 , x2 ) , f1 ( x2 ) F ( x1,1 , x2 ) , . . . ,
f k ( x2 ) F ( x1, k , x2 ) , are calculated by means of operator 0:
f 0 ( x2 ) and S 0 ( x 2 ) correspond to range C3:G3;
f1 ( x2 ) and S1 ( x 2 ) correspond to range C4:G4;
f k ( x2 ) and S k ( x2 ) correspond to range C13:G13.
These moments are stored in two-dimensional array mm. Later this array is used
(by means of one-dimensional array m2) in the func subroutine for the spline
interpolation (operator 1).
Further, coordinates x(1) and x(2) of initial approximation x 0 of the x
minimum point are specified as follows:
the minimum of range C3:G13 (Fig. 6.31) is found: min = 1.026724
(cell F10);
the point, where function F equals min, is taken as the initial approxima-
tion: x(1) = 3.4, x(2) = 0.6.

487
 Chapter 6. Numerical Methods for Nonlinear Programming

In last turn, two-dimensional array ss of initial steps is specified and the

search for the minimum point is performed by the Powell method. The following
data are put into the Excel worksheet:
initial array x into cells B14:D14 (Fig. 6.32);
final array x and the final values of nt and nf into cells B15:F15.

Fig. 6.32. The code execution results

Let us use the following notations: 1

and 2
are the current values of vari-
ables x(1) and x(2), respectively.
The value of objective function 0
F( 1
,
2
) is the result of the func
subroutine execution. This value is calculated as follows:
1) by means of the si subroutine (operator 1), grid function f ( x1 ) is
determined according to the following formulas:
f ( x1,0 ) = S 0 ( 2 ) , f ( x1,1 ) = S1 ( 2 ) , . . . , f ( x1, k ) = S k ( 2
);
2) by means of the mos subroutine (operator 2), the moments of spline
S ( x1 ) , corresponding to f ( x1 ) , are calculated;
3) by means of the si subroutine (operator 3), the value of S ( 1
) is calcu-
lated, which is considered as the objective function’s value:

488
 6.9. Minimization of the tabular function

x(0) = 0
= F( 1
,
2
) = S( 1
).
According to Fig. 6.32, the found minimum point of the F ( x1 , x 2 ) function,

given by table Listing 6.12, has coordinates x1 = 3.111 and x 2 = 0.525, and the

minimum value of F ( x1 , x 2 ) is equal to F ( x1 , x2 ) = 0.905.

We advise the reader to develop a noniterative method and corresponding
program for minimizing a positive tabular function of two variables (similar to
the method and program of Section 4.6). Listing 6.12 must be used for testing the
program.
Besides splines S 0 ( x 2 ) , S1 ( x 2 ) , . . . , S k ( x 2 ) considered above, cubic

splines S 0 ( x1 ) , S 1 ( x1 ) , . . . , S r ( x1 ) must be used. The last set of splines is

defined by grid functions f 0 ( x1 ) F ( x1 , x 2, 0 ) , f 1 ( x1 ) F ( x1 , x 2,1 ) , . . . ,

f r ( x1 ) F ( x1 , x 2, r ) :

f 0 ( x1 ) and S 0 ( x1 ) correspond to range C3:C13 in Fig. 6.31;

f 1 ( x1 ) and S 1 ( x1 ) correspond to range D3:D13;

f r ( x1 ) and S r ( x1 ) correspond to range G3:G13.

In the next two chapters, we will minimize a function whose values are calcu-
lated implicitly, more precisely, by solving the initial value problems for the sys-
tem of differential equations.

489
 Chapter 6. Numerical Methods for Nonlinear Programming

6.10. Solving the nonlinear differential

equation by the shooting method

Minimization of an implicit function may be required to solve the following

boundary value problem on segment [a, b] :

d 2u du
F x, u , , (6.27)
dx 2 dx
u ( a) A , (6.28)
u (b) B , (6.29)
where A and B are given parameters, F is a nonlinear function of variables x, y
and z.
Problem (6.27) — (6.29) was solved by the quasilinearization method in Sec-
tion 3.12. Below, this problem will be solved by the shooting method.
We introduce unknown function
du
v( x) .
dx
This expression can be written in the following equation form:
du
E ( x, u , v ) , (6.30)
dx
where E is a function of simple form: E v .
Equation (6.27) becomes
dv
F ( x, u , v ) . (6.31)
dx
In Section 5.2, the method for solving the system of equations (6.30) and
(6.31) was developed for initial conditions (6.28) and
v( a) Q . (6.32)
According to the shooting method, problem (6.27) — (6.29) can be solved by
the repeated solution of system (6.30), (6.31) with initial conditions (6.28) and
(6.32) for different values of Q until satisfaction of condition (6.29) at point b.
Let us consider the shooting model, which is a good illustration of the shoot-
ing method (this explains the method name).

490
 6.10. Solving the nonlinear differential equation by the shooting method

If we neglect the air resistance, then Newton’s second law gives the projectile
trajectory, u (x) , described by equation (6.27), where

F g (1 Q 2 ) / V 2 .
In this expression:
du
Q (a) is the slope of the gun barrel located at point ( a, A) with
dx
coordinates x = a and u = A;
V is the projectile velocity at the moment of leaving the barrel;
g is the free fall acceleration.
Let us assume that the target is at point (b, B) with coordinates x = b and
u = B. Solving the system of equations (6.30) and (6.31) with initial conditions
(6.28) and (6.32) at various values of Q, we simulate the shooting when varying
the slope of the gun barrel (Fig. 6.33).

Fig. 6.33. Graphic image of the u (x) solution of system (6.30), (6.31)
with initial conditions (6.28) and (6.32) at two values of Q

To obtain slope Q , at which the projectile hits the target at point (b, B), we
have to solve algebraic equation
u (Q, b) B , (6.33)
where u(Q, x) is the result of solving the system of differential equations (6.30)
and (6.31) with initial conditions (6.28) and (6.32) at given Q.

491
 Chapter 6. Numerical Methods for Nonlinear Programming

The above algebraic equation can be solved by the bisection, secant or

Steffensen method (Sections 4.5 and 5.5). We will solve it by minimizing the
following function:
G(Q) [u (Q, b) B]2 . (6.34)
The method for solving algebraic equation (6.33) by minimizing function
(6.34) is based on the following obvious assertion: the problem of solving the
system of nonlinear algebraic equations (5.17) is equivalent to the problem of
finding point x , at which non-negative function
F ( x) [ f1 (x) 1
]2 [ f 2 ( x) 2
]2 ... [ fn (x) n
]2 (6.35)
is equal to 0. Here, x ( x1 , x2 , ..., xn ) is a point of the n-dimensional space,
n ≥ 1.
Minimizing the F ( x1 , x2 , ..., xn ) function, defined by (6.35), is a popular
method for solving the system of nonlinear algebraic equations (5.17).

492
 6.11. Modeling of the hammer motion in the piano mechanism

6.11. Modeling of the hammer motion

in the piano mechanism

As an example of using the shooting method, we will solve the following task
of modeling of the hammer motion in the piano mechanism when rupture of the
spring occurs:
in addition to the initial time moment, a = 0, of rupture of the spring,
another moment, b = 20 ms, is given;
we know that the hammer’s displacement forward is equal to A3 = 7 mm
and B3 = 15 mm at moments a and b, respectively;
we have to determine values A4 and B4 of the hammer velocity at these
moments.
For solving this task, we use the simplified mathematical model of the piano
mechanism: the elastic constant of the spring, k, equals zero in model (5.27) —
(5.30). In this case, the hammer’s displacement is described by equation
d 2 u3 du3
E4 u3 , , (6.36)
dt 2 dt
where
2
du3
q
du3 dt
E4 u3 , .
dt p qu3
Second-order nonlinear differential equation (6.36) is considered on time
segment 0 ≤ t ≤ b. We have to solve this equation with boundary conditions
u 3 (0) A3 , (6.37)
u 3 (b ) B3 . (6.38)
As in Section 5.6, let us introduce the hammer velocity, u 4 du3 / dt . We
obtain the following system of first-order differential equations:
du3
E3 ( u 4 ) , (6.39)
dt

493
 Chapter 6. Numerical Methods for Nonlinear Programming

du4
E4 ( u3 , u 4 ) , (6.40)
dt
where
E3 u4 ,

qu 42
E4 .
p qu3
According to the shooting method, boundary value problem (6.36) — (6.38)
can be solved by the repeated solution of the system of differential equations
(6.39) and (6.40) with initial conditions (6.37) and
u 4 (0) A4 . (6.41)

For obtaining desired value A4 of the hammer velocity at the initial time
moment, we have to minimize the following function similar to (6.34):
G( A4 ) [u3 ( A4 , b) B3 ]2 , (6.42)
where function u3 ( A4 , t ) is the result of solving the system of equations (6.39)
and (6.40) with initial conditions (6.37) and (6.41) at given A4 .
Function (6.42) will be minimized by the mini subroutine. We could use the
minim subroutine because mini and minim work equally in the case of a one-
variable function (n = 1).
In the source data table given below, values l, τ, p, q and ζ have the same
sense as in table Listing 5.3 (p. 376).
Listing 6.14
l 20
tau 1.00E-03
A4 0
p 0.406
q 18.3
zeta 1.00E-09
A3 0.007
B3 0.015

The sense of the remaining values in table Listing 6.14 is as follows:

A4 is the initial approximation of the minimum point of function (6.42)
in units of m / s;

494
 6.11. Modeling of the hammer motion in the piano mechanism

A3 is the given value of u3 (0) in meters;

B3 is the given value of u3 (b) in meters, b l .
The code for solving the problem of minimizing function (6.42) has the fol-
lowing form:
Listing 6.15
Dim x(0 To 1) As Double 'array for minimization
Dim nf As Long 'counter of calls of func
Dim nt As Long 'counter of calls of test
Dim l As Integer, tau As Double
Dim A4 As Double
Dim p As Double, q As Double
Dim zeta As Double
Dim A3 As Double, B3 As Double
Dim tau2 As Double
Dim uu() As Double
Dim jj() As Integer

Sub main()
Dim ss(1 To 1, 1 To 1) As Double
Dim i As Integer
Dim sb As String, se As String
l = Selection.Cells(1, 2)
tau = Selection.Cells(2, 2)
A4 = Selection.Cells(3, 2)
p = Selection.Cells(4, 2)
q = Selection.Cells(5, 2)
zeta = Selection.Cells(6, 2)
A3 = Selection.Cells(7, 2)
B3 = Selection.Cells(8, 2)
tau2 = tau / 2
ReDim uu(1 To 4, 0 To l)
ReDim jj(0 To l)
x(1) = A4 'initial approximation of A4
ss(1, 1) = 1E-6 'initial step along A4 axis
nf = 0: nt = 0
Call func(1, x) 'it must be before minimization
Call mini(1, x, ss, 1E-12, 1E-12)
MsgBox "A4 = " & CStr(Round(x(1), 3)) & " m/s"
MsgBox "B4 = " & CStr(Round(uu(4, l), 3)) & " m/s"
Selection.Cells(9, 1) = "t"

495
 Chapter 6. Numerical Methods for Nonlinear Programming

Selection.Cells(9, 2) = "u3"
Selection.Cells(9, 3) = "u4"
Selection.Cells(9, 4) = "j max"
For i = 0 To l 'movement along time axis
Selection.Cells(10 + i, 1) = i * tau
Selection.Cells(10 + i, 2) = uu(3, i)
Selection.Cells(10 + i, 3) = uu(4, i)
Selection.Cells(10 + i, 4) = jj(i)
12: Selection.Cells(10 + i, 4). _
HorizontalAlignment = xlCenter 'alignment
Next i
sb = Selection.Cells(10, 1).Address
se = Selection.Cells(10 + l, 2).Address
13: Call graph(sb, se, "t, s", "u3, m")
End Sub

Sub e_functions(ByRef x() As Double, _

ByRef e() As Double)
e(3) = x(4)
e(4) = q * x(4) ^ 2 / (p - q * x(3))
End Sub

Sub fx_jacobian(ByRef x() As Double, _

ByRef fx() As Double)
Dim m3 As Double
fx(3, 3) = 1: fx(3, 4) = -tau2
m3 = p - q * x(3)
fx(4, 3) = -tau2 * (q * x(4) / m3) ^ 2
fx(4, 4) = 1 - tau * q * x(4) / m3
End Sub

Sub func(ByVal n, ByRef x() As Double)

Dim m As Integer
Dim i As Integer, j As Integer
Dim u(3 To 4) As Double
Dim xx(3 To 4) As Double
Dim z(3 To 4) As Double
Dim e(3 To 4) As Double
Dim a(3 To 4, 3 To 4) As Double
Dim b(3 To 4) As Double
Dim alpha(3 To 4) As Double
Dim max As Double

496
 6.11. Modeling of the hammer motion in the piano mechanism

nf = nf + 1
1: u(3) = A3: u(4) = x(1) 'values at t=0
For m = 3 To 4
uu(m, 0) = u(m)
Next m
jj(0) = 0
For i = 1 To l 'movement along time axis
2: Call e_functions(u, e)
For m = 3 To 4
3: alpha(m) = u(m) + tau2 * e(m)
4: xx(m) = u(m) + tau * e(m)
Next m
For j = 1 To 1000 'Newton iterations
5: Call fx_jacobian(xx, a)
6: Call e_functions(xx, e)
For m = 3 To 4
7: b(m) = alpha(m) - _
(xx(m) - tau2 * e(m))
Next m
8: Call gauss(2, a, b, z, 2, 2)
For m = 3 To 4
9: xx(m) = xx(m) + z(m)
Next m
max = 0
For m = 3 To 4
If Abs(z(m)) > max Then _
max = Abs(z(m))
Next m
10: If max < zeta Then Exit For
Next j
For m = 3 To 4
11: u(m) = xx(m)
uu(m, i) = u(m)
Next m
jj(i) = j
Next i
14: x(0) = (u(3) - B3) ^ 2
End Sub

Function test(ByVal n, ByRef x() As Double, _

ByRef z() As Double, ByVal rho, Optional alpha) _
As Boolean

497
 Chapter 6. Numerical Methods for Nonlinear Programming

nt = nt + 1
If Abs(z(0) - x(0)) < n * rho * z(0) Then
test = False
Else
test = True
End If
If Not IsMissing(alpha) Then
If x(0) < alpha Then test = False
End If
If n = 1 Or nt = 1048576 Then test = False
End Function

This code for minimizing function (6.42), actually, is a combination of two

codes — for optimizing a tin can (Section 6.4) and for simulation of the piano
mechanism (Section 5.6). Operator 14 defines the objective function’s form.
The source data for the code are specified in table Listing 6.14 (Fig. 6.34).
We must select this table before the code execution.

Fig. 6.34. The Excel table with the source data

During the execution:

1) the window with calculated velocity A4 = A4 = 0.294 m / s appears
(Fig. 6.35);
2) after clicking on the OK button, the window with calculated velocity
B 4 = B4 = 0.623 m / s appears (Fig. 6.36);

498
 6.11. Modeling of the hammer motion in the piano mechanism

3) after clicking on button OK, the following results, depicted in Fig. 6.37,
appear:
the values of t, u 3 and u 4 ;
the numbers of the Newton iterations, j max ;
the graph of dependence u 3 (t ) .
The last graph is the result of the graph subroutine execution (operator 13).
On p. 328, we used Fig. 6.37 to demonstrate the features of this subroutine
intended for automatic creation of graphs.
In section “Instead of Conclusions”, we will need the dependences of the cal-
culated values of A4 and B 4 versus the number of steps on segment 0 ≤ t ≤ b.
These dependences, depicted in Fig. 6.38, were obtained by executing code
Listing 6.15 for l = 2, 3, 5, 10, 20, 30. The time step is equal to b / l , where
b = 0.02.

Fig. 6.35. Window with the calculated value

of the hammer velocity at moment a = 0

Fig. 6.36. Window with the calculated value

of the hammer velocity at moment b = 20 ms

499
 Chapter 6. Numerical Methods for Nonlinear Programming

Fig. 6.37. The code execution results, which include the u 3 (t ) graph

Fig. 6.38. Velocities A4 (dashed curve) and B 4 (continuous curve)

versus the number of time steps, i.e., dependences A4 (l ) and B4 (l )

500
 6.12. Nonlinear programming and the least-squares method

6.12. Nonlinear programming

and the least-squares method

In Section 5.8, we considered the least-squares method to solve the task of

determining the production function, more precisely, the functional dependence
of the wheat productivity on the land quality. Let us return to this question.
According to the least-squares method, to get required values L1 , ..., L , ...,
Ln of the linear spline at the grid nodes, z1 < ... < z < ... < z n , we have to
find the minimum point of non-negative function (5.43) of form

G ( L1 , ..., L , ..., Ln ) [ L( x j ) u j ] 2 .
j 1
In Section 5.9, we minimized this function by solving the system of linear alge-
braic equations (5.48). In this section, we will use the Powell method for the
minimization.
Below is a code for finding required values L1 , ..., L , ..., Ln .

Listing 6.16
Dim nf As Long 'counter of calls of func
Dim nt As Long 'counter of calls of test
Dim m As Integer
Dim XX() As Double
Dim UU() As Double
Dim ZZ() As Double

Sub main()
Dim x() As Double
Dim ss() As Double
Dim n As Integer
Dim i As Integer
Dim j As Integer
Dim d As Double
m = Selection.Rows.Count 'quantity of rows
n = Selection.Cells(1, 2) 'number of nodes

501
 Chapter 6. Numerical Methods for Nonlinear Programming

ReDim XX(3 To m)
ReDim UU(3 To m)
ReDim ZZ(1 To n)
ReDim x(n)
ReDim ss(1 To n, 1 To n)
For j = 3 To m
XX(j) = Selection.Cells(j, 1)
UU(j) = Selection.Cells(j, 2)
Next j
For i = 1 To n
ZZ(i) = Selection.Cells(2 + i, 3)
Next i
For j = 1 To n
x(j) = Cells(1, 6 + j)
For i = 1 To n
ss(i, j) = Cells(1 + i, 6 + j)
Next i
Next j
For j = 1 To n
d = 0
For i = 1 To n
d = ss(i, j) ^ 2 + d
Next i
If d = 0 Then
Cells(1, 1).Value = _
"You must increase" & Str(j) & _
"-th initial step"
End
End If
Next j
nf = 0
nt = 0
Call func(n, x) 'it must be before minimization
0: Call minim(n, x, ss, 1E-6) 'Powell method
'Output of results:
For j = 0 To n
Selection.Cells(m + 2, j + 1) = x(j)
Next j
Selection.Cells(m + 2, n + 2) = nt
Selection.Cells(m + 2, n + 3) = nf
Selection.Cells(2, 4) = "L"
For i = 1 To n

502
 6.12. Nonlinear programming and the least-squares method

Selection.Cells(2 + i, 4) = x(i)
Next i
End Sub

Sub func(ByVal n, ByRef x() As Double)

Dim j As Integer, L As Double
nf = nf + 1
x(0) = 0
For j = 3 To m
If XX(j) <= ZZ(2) Then
L = ((ZZ(2) - XX(j)) * x(1) + _
(XX(j) - ZZ(1)) * x(2)) / (ZZ(2) - ZZ(1))
Else
L = ((ZZ(3) - XX(j)) * x(2) + _
(XX(j) - ZZ(2)) * x(3)) / (ZZ(3) - ZZ(2))
End If
x(0) = x(0) + (L – UU(j)) ^ 2
Next j
End Sub

Function test(ByVal n, ByRef x() As Double, _

ByRef z() As Double, ByVal rho, Optional alpha) _
As Boolean
Dim j As Byte
nt = nt + 1
If Abs(z(0) - x(0)) < n * rho * z(0) Then
test = False
Else
test = True
End If
If Not IsMissing(alpha) Then
If x(0) < alpha Then test = False
End If
End Function

This code has features of program Listing 5.6.

In addition to table Listing 5.5 (with the source data for program Listing 5.6),
the G1:I4 range contains the source data for code Listing 6.16 (Fig. 6.39):
range G1:I1 contains zero initial approximations of required coordinates
x(1) = L1 , x(2) = L2 , x(3) = L3 of the minimum point;

503
 Chapter 6. Numerical Methods for Nonlinear Programming

range G2:I4 contains the S matrix of initial steps.

Arrays XX, UU, ZZ are used in Listing 6.16 instead of arrays X, U, Z in
Listing 5.6. Before the code execution, we must select the B2:D15 range.

Fig. 6.39. The source data

According to Fig. 6.40, four iterations were performed and 104 values of
objective function (5.43) were calculated during the minimization by the Powell
method (by means of the minim subroutine, see operator 0). The calculated

values L1 , L2 , L3 of the linear spline at grid nodes z1 < z 2 < z 3 are respec-

tively placed in cells E4, E5 and E6 (and also in cells C17, D17 and E17). Natu-
rally, these values are the same as when using program Listing 5.6.
When using the minim subroutine in the least-squares method, we can easily
switch to another form of the required functional dependence, for example, to
quadratic form
F (C1 , C2 , C3 , x) C1 C2 x C3 x 2 . (6.43)
The values of constants C1 , C 2 and C3 must be determined by minimizing the
following non-negative function similar to (5.42):

G (C1 , C 2 , C3 ) [ F (C1 , C 2 , C3 , x j ) u j ] 2 , (6.44)

j 1

504
 6.12. Nonlinear programming and the least-squares method

where x j and u j are the values given in columns Mark and Productivity of the
source data table depicted in Fig. 6.39, 1 ≤ j ≤ ν (ν = 12 is the number of land
plots).

Fig. 6.40. The execution results

The following code is intended for finding the values of constants C1 , C 2

and C3 by means of the Powell minimization method.

Listing 6.17
Dim nf As Long 'counter of calls of func
Dim nt As Long 'counter of calls of test
Dim m As Integer
Dim XX() As Double
Dim UU() As Double

Sub main()
Dim x() As Double
Dim ss() As Double
Dim n As Integer
Dim i As Integer
Dim j As Integer

505
 Chapter 6. Numerical Methods for Nonlinear Programming

Dim d As Double
m = Selection.Rows.Count 'quantity of rows
n = Selection.Cells(1, 2) 'number of nodes
ReDim XX(3 To m)
ReDim UU(3 To m)
ReDim x(n)
ReDim ss(1 To n, 1 To n)
For j = 3 To m
XX(j) = Selection.Cells(j, 1)
UU(j) = Selection.Cells(j, 2)
Next j
For j = 1 To n
x(j) = Cells(1, 6 + j)
For i = 1 To n
ss(i, j) = Cells(1 + i, 6 + j)
Next i
Next j
For j = 1 To n
d = 0
For i = 1 To n
d = ss(i, j) ^ 2 + d
Next i
If d = 0 Then
Cells(1, 1).Value = _
"You must increase" & Str(j) & _
"-th initial step"
End
End If
Next j
nf = 0
nt = 0
Call func(n, x) 'it must be before minimization
0: Call minim(n, x, ss, 1E-6) 'Powell method
'Output of results:
For j = 0 To n
Selection.Cells(m + 2, j + 1) = x(j)
Next j
Selection.Cells(m + 2, n + 2) = nt
Selection.Cells(m + 2, n + 3) = nf
Selection.Cells(2, 4) = "C"
For i = 1 To n
Selection.Cells(2 + i, 4) = x(i)

506
 6.12. Nonlinear programming and the least-squares method

Next i
End Sub

Sub func(ByVal n, ByRef x() As Double)

Dim j As Integer
Dim F As Double
nf = nf + 1
x(0) = 0
For j = 3 To m
F = x(1) + x(2) * XX(j) + x(3) * XX(j) ^ 2
x(0) = x(0) + (F - UU(j)) ^ 2
Next j
End Sub

Function test(ByVal n, ByRef x() As Double, _

ByRef z() As Double, ByVal rho, Optional alpha) _
As Boolean
Dim j As Byte
nt = nt + 1
If Abs(z(0) - x(0)) < n * rho * z(0) Then
test = False
Else
test = True
End If
If Not IsMissing(alpha) Then
If x(0) < alpha Then test = False
End If
End Function

This code slightly differs from Listing 6.16.

For code Listing 6.17, the source data are almost the same as for code
Listing 6.16:
column Z may be absent (Fig. 6.41);
range G1:I1 contains zero initial approximations of required coordinates
x(1) = C1 , x(2) = C 2 , x(3) = C3 of the minimum point;
range G2:I4 contains the S matrix of initial steps.
Before the code execution, we must select the B2:C15 range.
According to Fig. 6.42, four iterations were performed and 92 values of
objective function (6.44) were calculated during the minimization by the Powell
method. The calculated values, C1* , C2* and C3* , are respectively placed in cells

507
 Chapter 6. Numerical Methods for Nonlinear Programming

E4, E5 and E6 (and also in cells C17, D17 and E17). Fig. 6.43 shows the graph
of quadratic dependence (6.43) with the calculated constants:
F (C1 , C2 , C3 , x) C1 C2 x C3 x 2 ,

where C1* = 10.2961, C2* = 0.37988 and C3* = 0.00158.

Fig. 6.41. The source data

Fig. 6.42. The execution results

508
 6.12. Nonlinear programming and the least-squares method

Fig. 6.43. The experimental points and the quadratic line of the functional
dependence of the wheat productivity on the land quality mark

509
Instead of Conclusions

For solving mathematical and applied tasks, an alternative to the numerical

method is the analytical method aimed at obtaining a solution in the formula
form. It is interesting to compare these two methods.
The obvious advantage of numerical methods over analytical ones is their
generality. However, numerical methods have a disadvantage consisting in the
approximate nature of the numerical solution of a task. Therefore, for improving
the accuracy, we have to run the program (realizing the numerical method)
several times while changing the values of the numerical method parameters.
The program execution must be repeated until the execution results become
independent of the parameter values. It is these results that are reliable.
As an illustration, let us return to the task of modeling of the hammer motion
in the piano mechanism when the spring is ruptured (Section 6.11).
In this task:
the numerical method parameter is l — the number of steps on the time
segment considered, 0 ≤ t ≤ b = 20 ms;
the results of solving the task are A4 and B4 — the hammer velocities at
the initial time moment and in 20 ms, respectively.
It is obvious that the accuracy of the numerical solution of the formulated
task improves with increasing l.
We solved the task for different values of l and depicted dependences A4 (l )
and B4 (l ) in Fig. 6.38. According to these dependences, the values of A4 and
B4 do not depend on l beyond 10. Therefore, the values obtained at l = 20 ( A4 =
0.294 m / s and B4 = 0.623 m / s) are authentic within the limits of the used
mathematical model of the hammer motion when the spring is ruptured.
The question of losing the numerical solution accuracy was also considered
at the end of Section 3.7.
Analytical and numerical methods are used successively for solving many
problems. For example, the integral of a complex function may be a result of the
analytical solution, and the numerical integration is required for obtaining the
values of this integral: according to formula (5.41) on p. 385, the numerical inte-

510
 Instead of Conclusions

gration of function f (x ) must be done to complete the analytical solution of the

initial value problem for the Bernoulli differential equation.
One more example of successive usage of analytical and numerical methods
is in Section 4.12, where:
1) the first derivative was excluded from differential equation (4.60) by sub-
stitution (4.62);
2) the resulting problem was solved numerically by the cubic spline method;
3) the original problem for differential equation (4.60) was solved by using
formula (4.62).
Guided by the material of Section 2.14, we can create Personal Macro Work-
book with our program modules. However, we can go even further, namely, can
modify our program modules for using them as the Excel add-ins. For this pur-
pose, package Microsoft Visual Studio can be used. Let us consider how to tune
Microsoft Visual Studio 2010 for this.
According to Section 1.24, the Start Page window opens when starting
Microsoft Visual Studio 2010. Further, we must:
1) click on the New Project hyperlink;
2) in the left area of the New Project window opened, click on the plus sign
against Other Languages;
3) in the open list, click on the plus sign against Visual Basic;
4) in the open list, click on the plus sign against Office, and further click on
line 2010 (or 2007);
5) in the list of the central area of the New Project window, click on line
Excel 2010 Add-In.
As a result, the necessary operation mode of Visual Studio is set. The follow-
ing information in the right area of the New Project window speaks about it:
A project for creating a managed code add-in for Excel 2010.
Further, we must specify the project name and the project folder location by
using text boxes Name and Location. At clicking on the OK button, the window
of Visual Basic Environment appears. This window includes the code window
with automatically generated lines.
Information on creating an add-in for Excel is available in the following
internet resource: http://msdn.microsoft.com/en-us/library/.
The idea of writing this book was spurred by book [17], the best seller of the
end of the 1960s. The last book convinced the author that learning numerical
methods is more effective if it follows learning programming and builds on the
skills of programming. In this case, the learner has the possibility to grasp diffi-
cult material on numerical methods by developing the program modules that
realize these methods.
In conclusion, the author would like to share the experience of using Excel
for solving scientific and engineering problems.

511
 Instead of Conclusions

In Pulsar R&D Manufacturing Company, Moscow, tabular processor Excel,

equipped with macros, is used as a preprocessor and postprocessor for compli-
cated programs written earlier in the Visual C++ programming language (a part
of Microsoft Visual Studio) for mathematical modeling of microwave transistors
[18]. In other words, the source data for the modeling are prepared in Excel and
the results are processed in Excel, in particular, graphs are constructed.
As we know, the Visual Basic programming language includes the Shell
function (p. 106), which allows running, from Excel, the executable file of an
arbitrary program, in particular, of a program written in Visual C++. By using
the Shell function, the author created the system of mathematical modeling
based on Excel. In this system, programs (written in Visual C++) and Excel
exchange data by means of text files.
When developing a program in Visual C++ (or in any other programming
language), we should pay attention to the following: the text file with the tabula-
tion character, as the substring connector (p. 95), is not only the text file, but also
the Excel workbook containing one worksheet. When we open such file with
Excel, substrings appear in separate cells, and Excel interprets a substring, which
contains digits and other features of a number, as the corresponding number.
The author hopes that he has managed to achieve the goals formulated at
the beginning of the book. It remains to wish the reader successful solution of
interesting problems.

512
Appendix 1.
Data Types of Visual Basic and VBA

Values of variable / constant

Data type Memory cell size
or comment
Boolean True (logical unit),
2B
(logical) False (logical zero)
Byte
(short integer 1B Integers from 0 to 255
unsigned)
Integer
2B Integers from -32768 to 32767
(integer)
Long Integers from -2147483648
4B
(long integer) to 2147483647
Numbers with four decimal places
Currency
8B from -922337203685477.5808
(scaled integer)
to 922337203685477.5807
Numbers with a fractional part
from -3.402823·1038
to -1.401298·10-45
Single
4B for negative values
(single-precision)
and from 1.401298·10-45
to 3.402823·1038
for positive values
Numbers with a fractional part
from -1.79769313486231·10308
to -4.94065645841247·10-324
Double
8B for negative values
(double-precision)
and from 4.94065645841247·10-324
to 1.79769313486232·10308
for positive values
Date from 1 January 100
Date
8B to 31 December 9999
(date-time)
and time from 0:00:00 to 23:59:59

513
 Appendix 1. Data Types of Visual Basic and VBA

Values of variable / constant

Data type Memory cell size
or comment
10 B +
1 B per character
String length from 0 to 231 characters
for string of
String variable length

1 B per character
for string of String length from 1 to 216 characters
fixed length

16 B
if the choice by
context gives
data type Values correspond to the Boolean,
Boolean, Byte, Byte, Integer, Long,
Integer, Long, Currency, Single, Double or
Currency, Date data type
Variant
Single, Double
or Date

22 B +
1 B per character Values correspond to String of
if the choice variable length
is not made
The memory cell stores the object
Object 4B
reference
Depends
on the quantity
User-defined Created by the Type operator
of fields and
their data types

514
Appendix 2.
Greek and Russian Alphabets
Denoted by Latin Letters

The Greek alphabet with English names of the letters

Lowercase Uppercase English

(small) letter (capital) letter name
α Α Alpha
β Β Beta
γ Γ Gamma
δ Δ Delta
ε Ε Epsilon
ζ Ζ Zeta
η Η Eta
θ Θ Theta
ι Ι Iota
κ Κ Kappa
λ Λ Lambda
μ Μ Mu
ν Ν Nu
ξ Ξ Xi
ο Ο Omicron
π Π Pi
ρ Ρ Rho
σ Σ Sigma
τ Τ Tau
υ Υ Upsilon
φ Φ Phi
χ Χ Chi
ψ Ψ Psi
ω Ω Omega

515
Appendix 2. Greek and Russian Alphabets Denoted by Latin Letters

The Russian alphabet denoted by Latin letters

Russian Latin Russian Latin

а; А a; A п p
б; Б b; B р r
в v с s
г g т t
д d у u
е; Е ye, e; Ye, E ф f
ё; Ё yo; Yo х kh, h
ж zh ц ts
з z ч tch, ch
и i ш sh
й y щ sch
к k ы y
л l э e
м m ю yu
н n я ya
о o ъ, ь apostrophe

516
Appendix 3.
The Main Mathematical Functions

The main mathematical functions of Visual Basic

Call of Return value, mathematical designation

the function and domain of the function
Abs(x) Absolute value of x, | x |
Atn(x) Arctangent of x, arctan x
Cos(x) Cosine of x, cos x
Exp(x) Exponential function, e x exp x (Fig. 3.6)
Fix(x) The result of truncating the fractional part of x
Int(x) The greatest integer not exceeding x
Log(x) Natural logarithm of x, ln x , at x > 0
Sgn(x) Sign function: -1, 0 or 1, depending on the sign of x
Sin(x) Sine of x, sin x
Sqr(x) Square root of x, x , at x ≥ 0
Tan(x) Tangent of x, tan x , at x (0.5 k ) , k is an integer

Examples of using the above functions

The Visual Basic operators for calculating the values of trigonometric func-
tion cot x , of inverse trigonometric functions arcsin x , arccos x and arccot x
and of decimal logarithm lg x are given below.

cot_x = Cos(x) / Sin(x) 'if Sin(x) <> 0

arcsin_x = Atn(x / Sqr(1 - x ^ 2)) 'if Abs(x) < 1

arccos_x = Atn(Sqr(1 - x ^ 2) / x)
'if x > 0 And x <= 1
arccot_x = Atn(1 / x) 'if x > 0

lg_x = Log(x) / 2.302585093 'if x > 0

517
Appendix 4.
Material for Tasks

The table below, taken from [19], contains 31 functions f (x) and their
domains [a, b] with the following properties:
function f (x) is continuous and monotonous on segment [a, b] ;
the signs of f (x) on the left and right boundaries of [a, b] are different.
This table is used in tasks for the reader (the tasks begin with words “we
advise the reader”). The f (a) values, given in the table, are intended to help the
reader debug his programs.
By using this table, a teacher can create 31 variants of tasks for exams. The
variant number may be a student’s birthday.

No. of
Function f(x) a b f(a)
variant
1 (3.8 3 sin x ) / 0.35 x 2 3 0.3905776

2 1 0 0.85 0.3333333
[3 sin(3.6 x)] x

3 cos 1 0.3 x 3 x 0 1 0.5403023

4 sin 1 0.4 x 2 x 0 1 0.841471
5 3 2 3 -1.2502
0.25x x 1.2502
6 2 1 2 0.1
0.1x x ln x
7 3x 4 ln x 5 2 4 -1.772589
8 ex e x
2 0 1 -2
9 3 0.4 1 -0.7307382
x x x 2.5
10 tan x (tan3 x ) / 3 (tan5 x ) / 5 1 / 3 0 0.8 -0.3333333
11 cos(2 / x) 2 sin(1 / x) 1 / x 1 2 -1.099089
12 sin(ln x) cos(ln x) 2 ln x 1 3 -1
13 ln x x 1.8 2 3 0.4931472

518
 Appendix 4. Material for Tasks

No. of
Function f(x) a b f(a)
variant
14 0.4 arctan x x 1 2 0.1853982
15 x tan x 1 / 3 0.2 1 -0.2927913
2
16 tan(0.55x 0.1) x 0.4 1 0.171389
17 2 sin(1 / x) x 1.2 2 0.0598231
18 1 sin x ln(1 x) x 0 1.5 1
0.52
19 cos( x 2) x 0.5 1 0.4029458
20 ln(1 x) 3 x 2 3 0.024503
21 ex ln x 10x 3 4 -8.815851
22 x x 1 3 -8.649598
3x 14 e e
23 2 1 3 -5
2 ln x 6 ln x 5
24 2x sin x cos x 0.4 1 -0.6095263
25 cos x exp( x 2 / 2) x 1 1 2 -0.0662284
26 1 x tan x 0 0.9 1
27 sin( x 2 ) cos( x 2 ) 10 x 0 1 1
28 ex 1 e 2x 2 -1 0 -0.5665994
29 1 x cos 1 x 0 0.9 0.4596977
30 tan(x / 2) cot(x / 2) x 1 2 -0.2841852
31 x cos x 0.5 2.5 -0.3775826

519
Appendix 5.
Analytical Method for Solving
the Cubic Algebraic Equation

Below, we will present Method 3 of handbook [3] for solving cubic equation
z3 3 p z 2q 0 , where p and q are nonzero real numbers.

Let us consider variable r p , the sign coinciding with the sign of q,

3
i.e., q/r 0 . The auxiliary value, φ, and the roots, z1 , z 2 and z 3 , are
determined according to the following table.

p 0

q 2 p3 0 q 2 p3 0 p 0
or 0 1 or 1
arccos Arcosh Arsinh

1 2
2 ln 2 1
arctan ln 1

z1 2r cos z1 2r ch z1 2r sh
3 3 3

z2 2r cos z2 r ch i 3 r sh z2 r sh i 3 r ch
3 3 3 3 3

z3 2r cos z3 r ch i 3 r sh z3 r sh i 3 r ch
3 3 3 3 3

As we see, the above formulas for calculating the roots depend on the signs
of p and q 2 p3 (i is the imaginary unit). The following mathematical functions
are used:
ch, sh — the hyperbolic cosine and sine:
ch x cosh x (e x e x ) / 2 , sh x sinh x (e x e x ) / 2 ;
Arcosh, Arsinh — the area-hyperbolic cosine and sine.

520
Appendix 6.
Realization of the Tangent Method
by Using the Excel Circular Reference

To use the circular reference feature, let us tune Excel as follows:

1) click on the File button;
2) Options > Formulas;
3) turn on option Automatic in area Calculation options of the Excel Options
window;
4) turn on option Enable iterative calculation;
5) for example, set 100 for the limiting number of iterations in text box
Maximum Iterations;
6) for example, set 0.001 for the final relative error in text box Maximum
Change;
7) click on the OK button.
If a formula, placed in an Excel cell, contains the reference to the same cell
(maybe not directly but indirectly, through a series of other references), we say
that the circular reference exists. The cyclic reference is used when we want to
realize an iterative or recurrence process.
Let us use the cyclic reference for solving nonlinear algebraic equation
f (x) 0 , where f ( x) x cos x 1.5 , by the tangent method (Section 5.5).
For that, we fulfill the following operations on an Excel worksheet.
1. Into cell G2, intended for variable x, enter an initial approximation of the
solution, for example 2.
2. Into cell F2, enter formula

=G2-COS(G2)-1.5

corresponding to mathematical formula f ( x) x cos x 1.5 . When clicking

on the tick button of the Excel formula bar, value 0.916147 appears in cell F2.
3. Into cell E2, enter formula

=1+SIN(G2)

corresponding to mathematical formula f ( x) 1 sin x . When clicking on the

tick button, value 1.909297 appears in the E2 cell.

521
 Appendix 6. Realization of the Tangent Method by Using the Excel Circular Reference

4. Into cell G2, intended for variable x, enter formula

=G2-F2/E2

corresponding to mathematical formula

f (x j )
xj 1
xj
f (x j )
of the tangent method. When we click on the tick button, values 1.999373, 0 and
1.535394 appear in cells E2, F2 and G2, respectively.
The multiple circular reference to cell G2 gives the last three values. Thus,
G2 contains the result of solving equation x cos x 1.5 0 , which is equal to
x = 1.535394.
In a similar way, the cyclic reference can be used for solving equation
f (x) 0 by other iterative methods of Sections 4.5 and 5.5.

522
References List

1. G. Z. Garber, Bases of Programming in Visual Basic and VBA for Excel 2007
(in Russian), Moscow: SOLON-PRESS, 2008.
2. G. Z. Garber, Bases of Programming in VBA for Excel and of Numerical
Methods (in Russian), Moscow: PRINTKOM, 2009.
3. I. N. Bronshtein, K. A. Semendyayev, G. Musiol, H. Muehlig, Handbook of
Mathematics, 5th edition, Springer, 2007.
4. A. A. Samarskii, The Theory of Difference Schemes, Marcel Dekker, 2001.
5. T. Y. Na, Computational Methods in Engineering: Boundary Value Prob-
lems, Academic Press, 1979.
6. G. Z. Garber, E. V. Kostyukov, Yu. A. Kuznetsov, Two-dimensional model-
ing of a cell of photosensitive silicon charge-coupled device for color televi-
sion cameras (in Russian), Electronic Engineering. Series 2. Semiconductor
Devices, 1989, no. 3, pp. 40 – 44.
7. W. Shockley, Electrons and Holes in Semiconductors, with Applications to
Transistor Electronics, Van Nostrand Reinhold, 1950.
8. H. Gould, J. Tobochnik, An Introduction to Computer Simulation Methods:
Applications to Physical Systems, Addison-Wesley, 1988.
9. J. H. Ahlberg, E. N. Nilson, J. L. Walsh, The Theory of Splines and Their
Applications, Academic Press, 1967.
10. G. Z. Garber, Model for simulation of AlGaAs-GaAs power heterostructure
FETs, Proceedings of EUROCON 2005 – The International Conference on
“Computer as a Tool”, Belgrade: IEEE, 2005, vol. 1, pp. 867 – 870.
11. G. Z. Garber, Method for calculating a small-signal equivalent circuit of
extremely high frequency heterostructural field-effect transistors, Journal of
Communications Technology and Electronics, 2005, vol. 50, no. 7, pp. 822 –
825.
12. A. Oledzki, Dynamics of piano mechanisms, Mechanism and Machine Theo-
ry, 1972, vol. 7, pp. 373 – 385.

523
 References List

13. S. N. Volkov, Land Management. Economic and Mathematical Methods and

Models (in Russian), Moscow: KOLOS, 2001.
14. M. J. D. Powell, An efficient method for finding the minimum of a function
of several variables without calculating derivatives, The Computer Journal,
1964, vol. 7, no. 2, pp. 155 – 162.
15. M. I. Korobochkin, Solving optimization problems of linear and nonlinear
programming in Excel (in Russian), Land Management, Cadastre and Moni-
toring of Lands, 2006, no. 12, pp. 73 – 76.
16. D. M. Himmelblau, Applied Nonlinear Programming, McGraw-Hill, 1972.
17. D. D. McCracken, W. S. Dorn, Numerical Methods and Fortran Program-
ming, John Wiley and Sons, 1965.
18. G. Z. Garber, Experience of using Microsoft Excel for mathematical model-
ing of microwave transistors (in Russian), Electronic Engineering. Series 2.
Semiconductor Devices, 2012, no. 2, pp. 22 – 27.
19. V. B. Glagolev, Visual Basic 6.0. Collection of tasks for laboratory and prac-
tical classes (in Russian), Publishing House of Moscow Power Engineering
Institute, 2000.

524
Subject Index

A CCur 43
Abs 230, 517 CDate 43
Activate 161, 162, 167 CDbl 43
ActiveCell 148, 155, 156, 158, 174 Cells 148, 167, 169, 172, 173
ActiveChart 245 Charts 245
ActiveSheet 155, 168, 169, 245 CheckBox 125 – 127, 151
ActiveWorkbook 155, 162, 163 CInt 43, 44, 51, 153
Add 152, 162, 164, 167, 170, 245 Clear 172
Add Watch 19, 72, 73, 75 Clear All Breakpoints 19
Address 172, 174, 245 CLng 43
And 48 – 52 Close 102 – 104, 161, 164
Append 101, 102, 106, 108 Color 169
Application 148, 152, 154 – 160, 165 Columns 142, 172, 173, 188, 189
Array CommandButton 120, 151
static 68 Const 29, 30, 210
dynamic 71 Constant
As 23, 29, 77, 78, 81, 90, 101, 103 built-in 31
Atn 45, 315, 517 user-defined 29
Auto_Open 159, 160 Cos 46, 517
Average 154, 155 Count 142, 162, 168, 188, 189
Course
B backward 225
Boolean 23, 29, 43, 48, 439, 513 forward 225
ByRef 80 CSng 43
Byte 23, 24, 43, 513 CStr 43, 44, 96, 153
ByVal 80 Currency 23, 24, 27, 29, 43, 513
CVar 43
C Cycle
Calculate 158 Do…Loop 61
Calculation 156, 159 Do Until…Loop 62, 63, 103
Call 79 Do While…Loop 61, 62
Caption 117, 120, 121 Do…Loop Until 64
Case 56, 57 Do…Loop While 63
Case Else 57 For Each…Next 172
CBool 43 For…Next 58
CByte 43 While…Wend 61

525
 Subject Index

D fouf 402, 404, 411

Date 25, 26, 29, 43, 47, 170, 513 FreeFile 101, 103
Delete 167, 245 FullName 161, 162
Dialogs 156 – 158 Func9 188 – 190
Dim 23 – 25, 80, 210 Function 77, 78, 139
Do 61 – 64
Double 24, 29, 43, 513 G
gaus 228, 229, 234
E gauss 230, 232, 234, 236, 497
Else 54, 55, 188, 403 GoTo 52, 54, 308
ElseIf 55, 403 graph 173, 327, 328, 338
End 180, 181, 433
End Function 77, 78, 139 H
End If 54, 55, 188, 403 Height 117
End Select 57 Hypotenuse 139 – 141
End Sub 16, 79, 80
End Type 90, 92 I
End With 91 – 93, 156 – 158 If 53 – 55, 188, 403
EOF 103, 104 IIf 56
Erase 75 In 172
Event of object 151 Input 101 – 104, 108
exe 106, 132, 135, 138 InputBox 85, 87, 148
Exit Do 64 Int 51, 517
Exit For 60, 245 Integer 23, 24, 43, 513
Exit Function 83 ios 298 – 300
Exit Sub 83, 309 IsMissing 82, 83, 290
Exp 216, 220 – 222, 517 Italic 156, 169

F J
False 29, 48, 49, 126, 513 Jacobian matrix 364
fb 210, 211, 213
FileFormat 161 K
FileName 161 Kill 106
Fix 51, 517
foba 252, 253, 258 L
fobas 334, 335, 338, 347 Label 118, 151
Font 118, 156, 169 LBound 74, 75
For 58 – 60, 101, 103 LCase 96
For Each 172 Left 97, 98
forbac 262 – 264 Len 97
forbacs 266 Line Input 102, 103
Format 100, 170 LOF 103, 104
Formula 148, 171, 174 Log 45, 315, 517
FormulaR1C1 171 Long 23, 24, 25, 43, 513
forwback 271, 272, 277 Loop 61 – 64
foub 404, 405, 411 LTrim 95, 96

526
 Subject Index

M P
MDETERM 224 Password 161
Method Path 161, 163
barrier function 472 PERSONAL.XLSB 184 – 186, 189, 190
bisection 306, 309 Preserve 73, 74
interior point 473 Print 102, 103
of object 151, 152 Private 81, 121
penalty function 466 Procedure
secant 310, 311 built-in 85
Steffensen 370 Fourier Analysis 396
tangent 366 user-defined
trapezoidal 384 function 77
variable replacement 462 subroutine 79
Mid 97, 98 Property of object 151, 152
mini 432, 435, 436, 439, 442 Public 81
minim 447, 453, 464, 486 Public Const 82
MINVERSE 224
MkDir 103 – 105 Q
MMULT 224 QBColor 169
Mod 39, 40, 42, 56 Quit 158, 165
mos 288, 289, 294
MsgBox 85 – 88, 98 R
Randomize 76, 476, 478
N Range 148, 152 – 154, 171, 172, 178
Name 120, 161, 162, 167, 169 ReDim 71, 73 – 75, 142, 210
Next 58 – 60, 172 ReferenceStyle 148
Not 48 – 50, 82 Replace 97
Now 25, 26, 47, 100, 153, 159, 476 Reset 20
NumberFormat 148, 245 Right 97, 98
RmDir 106
O Rnd 46, 75, 76, 476, 478
Object 151, 166, 514 Rosenbrock 425
Offset 172, 174 Round 45, 46, 51, 300
OnTime 159, 160 Rows 142, 172, 173, 188, 189
Open 101, 103 RTrim 95, 96
Operator Run 19
case 56 Run To Cursor 19
conditional 53
conditional jump 54 S
unconditional jump 52 Save 161, 162
Option Base 69 SaveAs 161, 164
Option Explicit 24 Select 148, 153, 172, 174, 178, 328
Optional 82, 83, 230 Select Case 56, 57
Or 48 – 51, 112 Selection 148, 173, 174, 181, 328
Output 101 – 103 Set 152, 164, 170, 174, 182, 328

527
 Subject Index

SetFocus 134 True 48, 49, 126, 513

Sgn 45, 309, 517 Type 90, 92
Shell 106, 135, 138, 512 TypeName 188
Show 134, 136, 156 – 158
si 290, 291, 294, 324 U
Sin 81, 517 UBound 74, 75
Single 24, 42, 43, 513 UCase 96
Space 96 Until 62 – 64, 71, 103
Sqr 17, 46, 517 UserForm 116, 151
Step 58 – 60
Step Into 19, 82, 114 V
Step Out 82 Val 27, 33 – 35
Step Over 82 Value 122, 126, 156, 172 – 174
Str 27, 43, 44, 300 Variable
String 27, 43, 94, 514 local 81
Sub 16, 79, 80 module 81
Sum 154, 155, 174 public 81
Sweep Variant 24, 25, 43, 82, 514
backward 202 vbCrLf 98, 99, 126, 148
forward 202 vbTab 95

T W
table 315, 316, 318, 324 Wend 61
Tan 517 While 61 – 63
Text 121, 123, 125 Width 117
TextAlign 120 With 91 – 93, 156 – 158
TextBox 120 – 122, 125, 151 Workbook 161, 164
Then 53 – 55, 188, 403 Workbooks 152, 161, 162, 164
Time 25, 26, 47, 100, 159 Worksheet 167 – 170
TimeValue 99, 100, 159, 476 Worksheets 152 – 155, 167 – 170
To 58 – 60, 68, 71
Toggle Breakpoint 19 X
TRANSPOSE 141, 142, 224, 295 xlCalculationAutomatic 156
TRANSPOSEA 142, 143, 224 xlCalculationManual 156, 159
Trim 95, 96 XLSTART 186

Подписано в печать 21.06.2013

Формат 60х88/16. Печ. л. 33
Заказ № 125
Отпечатано в полн. соотв. с электронной версией заказчика
в ООО «ИПЦ «Маска»
117246, Москва, Научный проезд, 20