Numerical Calculations for

Process Engineering Using

Excel VBA

Numerical Calculations for Process Engineering Using Excel VBA provides numeri­
cal treatment of process engineering problems with VBA programming and Excel
spreadsheets. The problems are solving material and energy balances, optimising
reactors and modelling multiple-factor processes. The book includes both basic and
advanced codes for numerical calculations. The basic methods are presented in dif­
ferent variations tailored to particular applications. Some macros are combined with
each other to solve engineering problems. Examples include combining the bisection
method and binary search to optimise an implicit correlation, combining golden sec­
tion search with Euler’s method to optimise a reactor and combining bisection code
and Euler’s method to solve steady-state heat distribution. The text also includes non-
conventional examples such as harmony search and network analysis. The examples
include solutions to common engineering problems such as adiabatic flame tempera­
ture, plug flow reactor conversion, batch reactor, heat diffusion and pinch analysis of
heat exchanger networks. The VBA code is presented with mathematical equations
and flowcharts, enabling the audience to adopt the solutions to different problems.
The book contains many demonstrations of numerical techniques to guide users. It
also includes useful summaries of VBA commands/functions and Excel-predefined
functions accessible in VBA.
While the book is developed primarily for undergraduate students, the book is a
helpful resource for postgraduate students and engineers.
Numerical Calculations for
Process Engineering Using
Excel VBA

Chi M. Phan
Designed cover image: © Shutterstock
First edition published 2024
by CRC Press
2385 NW Executive Center Drive, Suite 320, Boca Raton FL 33431
and by CRC Press
4 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN
CRC Press is an imprint of Taylor & Francis Group, LLC
© 2024 Chi M. Phan
Reasonable efforts have been made to publish reliable data and information, but
the author and publisher cannot assume responsibility for the validity of all
materials or the consequences of their use. The authors and publishers have
attempted to trace the copyright holders of all material reproduced in this
publication and apologize to copyright holders if permission to publish in this form
has not been obtained. If any copyright material has not been acknowledged please
write and let us know so we may rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be
reprinted, reproduced, transmitted, or utilized in any form by any electronic,
mechanical, or other means, now known or hereafter invented, including
photocopying, microfilming, and recording, or in any information storage or
retrieval system, without written permission from the publishers.
For permission to photocopy or use material electronically from this work, access
www.copyright.com or contact the Copyright Clearance Center, Inc. (CCC), 222
Rosewood Drive, Danvers, MA 01923, 978-750-8400. For works that are not
available on CCC please contact [email protected]
Trademark notice: Product or corporate names may be trademarks or registered
trademarks and are used only for identification and explanation without intent to
infringe.
ISBN: 978-1-032-42828-4 (hbk)
ISBN: 978-1-032-42829-1 (pbk)
ISBN: 978-1-003-36451-1 (ebk)
DOI: 10.1201/9781003364511
Typeset in Times
by Apex CoVantage, LLC
Contents
Preface.......................................................................................................................ix
About the Author ......................................................................................................xi

Chapter 1 Introduction to Excel and VBA ............................................................ 1

1.1 Microsoft Excel .........................................................................1
1.1.1 Visual Basic for Applications ....................................... 1
1.1.2 Objects in Excel............................................................ 2
1.1.3 Sub and Function Procedures in VBA ......................... 3
1.2 Key Characteristics of VBA Programming .............................. 5
1.2.1 Variables in VBA .........................................................5
1.2.2 Static and Dynamic Arrays .......................................... 6
1.2.3 Returning Multiple Values in a Spreadsheet ................ 7
1.2.4 For . . . Next Loop Structure ........................................8
1.2.5 Do . . . Loop Structures .............................................. 10
1.2.6 Options in VBA Code ................................................ 11
1.2.7 Get, Set or Change Spreadsheet Cell Value ............... 12
1.2.8 Spreadsheet Functions ................................................ 13
1.3 Avoiding Errors in VBA Procedures ....................................... 15
1.3.1 Errors .......................................................................... 15
1.3.2 Endless Loop .............................................................. 16
1.3.3 Displacing Immediate Value ...................................... 17
1.3.4 Breakpoints ................................................................ 17
1.4 Useful Excel Tools for Engineering ........................................ 18

Chapter 2 Roots of an Equation ..........................................................................20

2.1 Introduction .............................................................................20
2.2 Bisection Method.....................................................................20
2.2.1 Bisection Method with Control Error in Variable ...... 22
2.2.2 Bisection Method with Scanning ...............................25
2.2.3 Failures of the Bisection Method ...............................26
2.3 Newton’s and Secant Methods................................................. 27
2.4 The Regula Falsi Method ........................................................28
2.5 Concluding Remarks ............................................................... 29
2.6 Further Examples for Self-Practice ......................................... 30

Chapter 3 Applications of Roots of an Equation ................................................ 31

3.1 Introduction ............................................................................. 31
3.2 Advanced Example: Adiabatic Flame Temperature ................ 31

v
vi Contents

3.3 Advanced Example: Tear Analysis for Material Balances

with a Recycle Stream .............................................................34
3.4 Advanced Example: Determining the Vapour Fraction
from Internal Energy and Specific Volume ............................. 37
3.5 Further Examples for Self-Practice .........................................40

Chapter 4 One-Dimensional Optimisation ......................................................... 43

4.1 Introduction ............................................................................. 43
4.2 Region Elimination.................................................................. 43
4.3 Quadratic Interpolation ...........................................................46
4.4 Optimisation with Scanning .................................................... 47
4.5 Advanced Example: A Combination of Binary Search
and Bisection Method ..............................................................48
4.6 Further Examples for Self-Practice ......................................... 51

Chapter 5 Linear Programming .......................................................................... 54

5.1 Introduction ............................................................................. 54
5.2 Simplex Algorithm .................................................................. 54
5.3 Advanced Example: Pinch Analysis of Heat Exchanger
Network ................................................................................... 59
5.4 Further Examples for Self-Practice ......................................... 63

Chapter 6 Roots of Systems of Linear Equations ............................................... 65

6.1 Introduction ............................................................................. 65
6.2 Successive Substitution Methods............................................. 65
6.3 Successive Substitution Methods with a Relaxation Constant ..... 70
6.4 Concluding Remarks ............................................................... 71
6.5 Further Examples for Self-Practice ......................................... 72

Chapter 7 Applications of Systems of Linear Equations .................................... 74

7.1 Introduction ............................................................................. 74
7.2 Input Validation ....................................................................... 74
7.3 Advanced Example: A Combination of Bisection and
Jacobi Methods ........................................................................ 75
7.4 Advanced Example: 2-Dimensional Steady-State Heat
Transfer .................................................................................... 77
7.5 Advanced Example: Unsteady-State Heat Transfer after
Microwave Heating..................................................................80

Chapter 8 Solving Systems of Non-Linear Equations ........................................ 85

8.1 Introduction ............................................................................. 85
8.2 Newton’s and Secant Methods................................................. 85
Contents vii

8.3 Advanced Example: Tear Analysis with Two Independent

Unknowns ................................................................................ 88
8.4 Advanced Example: Tear Analysis with Stochastic
Optimisation ............................................................................ 91
8.5 Further Examples for Self-Practice .........................................96

Chapter 9 Solving Ordinary Differential Equations ...........................................97

9.1 Introduction .............................................................................97
9.2 One-Step Methods with a Constant Step Size .........................97
9.2.1 Second-Order Methods ..............................................99
9.2.2 Fourth-Order Methods ............................................. 100
9.3 One-Step Methods with an Adaptive Step Size .................... 101
9.4 Justification of the Selected Step Size ................................... 106
9.5 Concluding Remarks ............................................................. 107

Chapter 10 Applications of ODEs....................................................................... 108

10.1 Introduction ........................................................................... 108
10.2 Advanced Example: Optimisation of a Semi-Batch Reactor..... 108
10.3 Advanced Example: Boundary Value ODEs ......................... 113
10.4 Further Examples for Self-Practice ....................................... 116

Chapter 11 Integration ........................................................................................ 118

11.1 Introduction ........................................................................... 118
11.2 Area under a Curve from Discrete Data Points..................... 119
11.3 Advanced Example: Plug Flow Reactor Performance .......... 120

Chapter 12 Network Analysis ............................................................................. 123

12.1 Introduction ........................................................................... 123
12.2 Converting a Network to Table Form .................................... 123
12.3 Advanced Example: Shortest Pathway Problem.................... 124
12.4 Advanced Example: Travelling-Salesman Problem .............. 129
12.5 Example for Self-Practice ...................................................... 130

Appendices������������������������������������������������������������������������������������������������������������ 133
List of Useful VBA Functions Which Returns a Numerical Value... 133
Other Useful VBA Functions ........................................................... 133
List of Useful Worksheet Functions (Using the
WorkSheetFunction Object in VBA)................................................ 134

References ������������������������������������������������������������������������������������������������������������� 137

Index����������������������������������������������������������������������������������������������������������������������139
Preface
I came across Excel VBA as a fresh engineering graduate in 2000. For an engineer
with a strong software background, VBA seems too primitive. Yet the software allows
a powerful connection with databases in Excel spreadsheets and effective communi­
cation with others. Over the next two decades, Excel VBA became an essential part
of my research and teaching. It was a pleasure to explore and create new calculation
methods for new and not-too-new problems. Occasionally, I review my previous code
and see how it can be improved by twisting it around. While there are many guides to
writing simple VBA for a particular numerical method, combining this simple code
can open unlimited opportunities to solve a complex problem. Such combinations
take the coder’s vision beyond computer programming and mathematical algorithms.
This exciting creativity was the primary motivation for this book. I hope the readers
enjoy the journey, too.
I want to thank Dr Gordon Ingram, Dr Anita Hyde and Dr Cuong Nguyen for
their insightful discussion on the art of numerical methods. Finally, I thank Quy
Ngo, Safin Phan and Calleigh Phan for their unconditional support during the book’s
development.

ix
About the Author
Chi M� Phan (PhD, BEng) has taught in the Department of
Chemical Engineering at Curtin University for over 15 years.
He has taught classes in process analysis, process calculations,
process modelling and simulations. In addition to Curtin Uni­
versity, he taught at the University of Newcastle (Australia) and
the University of Hyogo (Japan). He has developed many nu­
merical models for his research in surface science.

xi
 1 Introduction to Excel and
VBA
1.1 MICROSOFT EXCEL
Microsoft Excel is a spreadsheet program written and distributed by Microsoft for com­
puters. It features an intuitive interface and capable calculation and graphing tools, which
have made Excel one of the most popular microcomputer applications. It is overwhelm­
ingly the dominant spreadsheet application available. All Microsoft Office versions
(including Apple Mac OS versions) currently have Excel with Visual Basic applications.

1.1.1 Visual Basic for applications

Since the 1990s, Excel has included Visual Basic for Applications (VBA), a pro­
gramming language based on Visual Basic, which adds the ability to automate tasks
in Excel and to provide user-defined procedures for use in worksheets. The procedure
is attached to the particular Excel file in which it was developed and can only run
on that Excel file. Excel workbooks of this type have to be saved as macro enabled
(.xlsm) in the latest Microsoft Excel versions. Standard Excel files, without macros,
can be saved as. xls files. It is necessary to save as. xlsm after VBA code is created.
All attached macros will be removed by wrongly saving the file as. xls. These macros
can only run within the file if multiple Excel files are opened simultaneously. Macros
can be copied and pasted to different Excel files if needed. VBA macros can also

FIGURE 1.1 VBA editor window.

DOI: 10.1201/9781003364511-1 1
2 Calculations for Process Engineering Using Excel VBA

be created by recording. Macro recording can produce VBA code replicating user
actions, thus allowing simple automation of regular tasks. However, writing, editing
and debugging macros in the VBA editor is more effective for numerical calcula­
tions. This editor screen can be accessed by hitting the keystroke Alt + F11.
By default, new Excel workbooks have no module. The users must insert the mod­
ules. Once inserted, the module can be renamed. All macros in different modules of
the same Excel workbook can be used in any spreadsheet.
VBA implements Microsoft’s Visual Basic, built into all Microsoft Office appli­
cations (including Apple Mac OS versions). VBA is closely related to Visual Basic
but can only run code from within a host application rather than as a stand-alone
application. As a result, VBA can be used to control one application from another
application (for example, automatically creating a Word report from Excel data).

1.1.2 oBjects in excel

VBA manipulates objects in Excel. The objects can be workbooks, worksheets, cell
ranges and charts. Objects are arranged in a hierarchy, and one object can act as a con­
tainer for other objects. At the top of the object hierarchy is Excel, which is an object
called Application, and it can contain other objects such as workbook objects and Com­
mand Bar objects. For example, the workbook object contains objects such as work­
sheet objects and chart objects. A worksheet object, in turn, contains Range objects.
Objects of the same type form a collection. For example, the Worksheet col­
lection consists of all the worksheets in a particular workbook. You can refer to an
object by specifying its position in the object hierarchy, using a dot as a separator.
For example, you refer to Range A1 in worksheet Sheet1 in workbook Book1.xls as

Application.Workbooks(“Book1.xls”).Worksheets(“Sheet1”).
Range(“A1”)

If you omit specific references, Excel uses the active objects. For example, if
Book1 and Sheet1 are active, then the preceding reference can be simplified to:

Range(“A1”)

Objects have properties. For example, a Range object has such properties as Value
and Address. A Chart object has such properties as Title and Type. You can use VBA
to determine object properties and to change properties.
The user can refer to a property of an object by combining the object name with
the property name, separated by a period. For example, you can refer to the value in
cell A1 on Sheet1 by

Worksheets(“Sheet1”).Range(“A1”).Value
Introduction to Excel and VBA 3

You can assign values to variables. A variable is a named element that stores data.
You can use variables to store such things as values, text or property settings. To as­
sign the value in cell A1 on Sheet1 to a variable called X_value, use the following
statement:

X _ value = Worksheets(“Sheet1”).Range(“A1”).Value

In addition to properties, objects have methods. A method is an action Excel per­

forms with an object. For example, one of the methods for a Range object is Clear-
Contents. This method clears the contents of the range. You specify a method by
combining the object with the method, separated by a dot. For example, the follow­
ing statement clears the contents of cell A1:

Worksheets(“Sheet1”).Range(“A1”).ClearContents

While VBA code can manipulate the cells in an Excel spreadsheet directly,
it is not always used in numerical calculations. Instead, a macro is often used
for the purpose. The purpose and limitations of macros are explained in the next
section.

1.1.3 suB and function procedures in VBa

In Excel, the users can perform actions in VBA by writing (or recording) code
in a VBA module. The users can also view and edit existing VBA modules in the
Visual Basic Editor (VBE). In Microsoft Excel’s VBE, users can create two types
of procedures: sub and function. Both are sets of commands that are used to per­
form specific tasks in VBA. A sub, also known as a subroutine or sub-procedure,
is a code used to perform a specific task, but it does not return any value. On the
other hand, a function, also known as a user-defined function procedure, executes
a specific task determined by the Excel user and returns a result. Within the func­
tion, the result is stored as the name of the function. The function can take cell
values in an Excel spreadsheet as arguments. VBA subs and functions can include
all the constructs of modern programming languages, including arrays, branches
and loops.
Many macros can be written into the same module. The program automatically
separates them at the end of each macro, with either End Sub or End Function. Codes
1.1 and 1.2 show examples of sub and function. The first function, named double1,
can be used as a standard Excel built-in function, that is, enter “=double1(cell loca­
tion)” in any spreadsheet cell. The cell location can be changed to any other cell
(“=double1(A1)” or a single numerical value (“=double1(103)”). The second func­
tion, named double2, calculates the double value of a specific cell, cell A1, in the
active worksheet.
4 Calculations for Process Engineering Using Excel VBA

Code 1�1� Two VBA functions for doubling a number�

Function double1(x)
double1 = x * 2
End Function
Function double2()
Dim x
x = Worksheets(“Sheet1”).Range(“A1”).Value
double2 = x * 2
End Function

VBA subs are more useful for “action” procedures (e.g. formatting cells, displac­
ing value in Message Box). To use a sub to export results to an Excel spreadsheet,
a set value can be used. However, the range of cells is specified within the sub and
cannot be moved as with functions. One effective way to use the sub in an Excel
spreadsheet is to create a shape and assign a sub to it. Then the user can click on
the shape. Alternatively, the sub can be assigned to a particular keystroke. Using the
Excel menu and keyboard commands, subs can be created by recording. However,
the recorded macros cannot incorporate logic, branching and looping.

Code 1�2� Two VBA subroutines for doubling a number�

Sub double3()
Dim xx
xx = Worksheets(“Sheet1”).Range(“A1”).Value * 2
MsgBox “The answer is “& xx
End Sub
Sub double4()
Dim x
x = Worksheets(“Sheet1”).Range(“A1”).Value * 2
Worksheets(“Sheet1”).Range(“A2”).Value = x
End Sub

To use double3 or double4 in Code 1.2, one can create a Command Button in an
Excel spreadsheet and then assign the subroutine to it (Figure 1.2). Subsequently, the
user needs to click on the message box to close it. The subroutines do not take the
input value as an argument. In contrast, functions allow both (i) a numerical value as
an argument and (ii) the value stored in a specific cell as the argument.
The output from the subroutine cannot be used by other VBA functions or subrou­
tines. This disadvantage of subroutines limits their usage in numerical calculations.
As seen, both subs and functions can do the same calculations. The two essential
function features are the return of a value (or a set of values) and multiple uses in the
spreadsheet. Once created, functions are available in the Excel library. VBA func­
tions work in the same way as formulas and built-in functions in Excel. A function
Introduction to Excel and VBA 5

FIGURE 1.2 Using a subroutine in Excel spreadsheet.

can be used multiple times within the same Excel workbook. Users can create cus­
tom functions for any action and then access the functions from a cell or a direct
reference from a cell. In addition, the function return value can be combined with
another sub/function. For this reason, VBA functions are more convenient for numer­
ical calculations and will be used mainly in this book.

1.2 KEY CHARACTERISTICS OF VBA PROGRAMMING

1.2.1 VariaBles in VBa
For numerical calculations, it is essential to declare variables. VBA variables can be
a single value or a set of values, as in an array or a matrix. The Dim statement can
specify the variable type within a VBA function by typing “As . . .”.

Dim x As Integer
Dim y1,y2 as string
Dim A(2,10) as long
Dim B() As double

If omitted, the default type of variable is Variant, which means the variable can
be any value (text, numbers, dates, time or objects). For numerical types of variables,
VBA has predefined ranges of values (as in Table 1.1). Therefore, it is important to
6 Calculations for Process Engineering Using Excel VBA

TABLE 1.1
Range of Values for Numerical Data Types
Data Type Range of Values
Byte 0 to 255
Integer –2,147,483,648 to 2,147,483,647
Short –32,768 through 32,767 (short integer)
Long –9,223,372,036,854,775,808 through 9,223,372,036,854,775,807 (long integer)
Single –3.402823×1038 to –1.401298 ×10–45 for negative values, 1.401298×10–45 to
3.402823×1038 for positive values
Double –1.79769313486232×10308 to –4.94065645841247×10–324 for negative values,
4.94065645841247 ×10–324 to 1.79769313486232 ×10308 for positive values
Decimal +/–79,228,162,514,264,337,593,543,950,335 for no decimal points, +/–7.922816
2514264337593543950335 for 28 places to the right of the decimal

consider the expected range of the value. If any calculated value within the code is
outside the defined range, the VBA code will give an error (#VALUE) without spec­
ifying the problem.
The ranges might change with the Excel version. The specified values can be
checked in detail on the Microsoft website.
https://docs.microsoft.com/en-us/dotnet/visual-basic/language-reference/data­
types/

Code 1�3� Size of a matrix (since the function name is Variant, the code
can return a text string or a numerical value)�
Function SM(A)
‘return the size of the matrix
Dim N as integer
N = A.Rows.Count
If A.Columns.Count <> N Then
SM = “Not a square matrix”:
Else
SM = N
End If
End Function

1.2.2 static and dynamic arrays

An array is a set of variables. Excel/VBA allows a multi-dimensional array with up
to 60 dimensions. If omitted, the array variable is Variant, and elements within the
array can be any type. The elements don’t have to be the same types in this case.
Introduction to Excel and VBA 7

Array variables can be either static or dynamic. A dynamic array is an array that
can be resized, as opposed to a static array, which cannot be resized. For static array,
the size is fixed at declaration, with the statement Dim. Dynamic data can be reini­
tiated by the statement Redim. An existing array can be extended while keeping the
existing value using the ReDim Preserve statement.
ReDim can be used to redefine the variable type as well. However, it would
create confusion for code users. More often, ReDim is used to redefine the dimen­
sion of an array or matrix. It is constructive for output with a variable number of
answers.
The values of an array can also be reset with Erase instead of using Redim. While
Redim resizes the dimension of the array, the Erase statement reinitialises the ele­
ments of fixed-size arrays and releases dynamic-array storage space. The initialisa­
tion depends on the type of variables. For example, Erase resets the fixed numeric
array to zero and the fixed variant to Empty.

1.2.3 returning multiple Values in a spreadsheet

Excel allows users to call VBA functions to spreadsheets in matrix format. Unlike
other software, in which the output dimension is defined (for example, precisely the
same as the input), Excel/VBA allows a flexible output.
For example, if the function gives the answers as a 10 × 10 matrix, that is a two-
dimensional array. If the user can take a 2 × 2–range Excel spreadsheet and enter the
function, the output will be the first four elements of the answers. If the user selects
an 11 × 11 range and enters the function, all the extra cells will have a default value
of zero. It is important to note that ReDim cannot be used with the function’s name.
Consequently, a practical way to manage multiple outputs is to declare a dynamic
variable, for example, output(), and re-dimension as needed. After calculations, the
desired outcomes are allocated to elements of output. As the output is variant, its
element can be of any data type independently. For example, output(1,1) can be text,
and output(1,2) is numerical. If an array is used, the outcomes are in a horizontal
range of cells. Finally, the function name can be assigned to an array. Hence, size
and elements are included. The output array should have one or two dimensions only.
Only the first indexes are used if the array has more than two dimensions (remember,
VBA allows up to 60 dimensions!).
The standard way to select a range of cells is to type in the function in the formula
bar and then hit the keystroke Ctrl+Shift+Enter. The bracket {} is automatically
included. If the user selects a range larger than the output, all extra cells will display
“%NA”. This fixed-size result is also known as the LEGACY ARRAY formula.
In some later Excel versions, the user can use the SPILLED ARRAY formula. In
these versions, the user only needs to select one cell (this corresponds to the output
array’s first element) and ensure that other cells (below and to the right) are empty.
The data will automatically spill over the required range. This is similar to the spill­
ing behaviour of a dynamic array. The advantage of this method is that you don’t
need to know the dimension of the data matrix. If the neighbouring cells are not
empty, Excel will return #SPILL! error.
8 Calculations for Process Engineering Using Excel VBA

Code 1�4� Output a matrix in an Excel spreadsheet�

Function mymatrix(n)
Dim MyArray()
ReDim MyArray(1 To 10, 1 To n)
mymatrix = MyArray
End Function
Function mymatrix2(n)
Dim MyArray()
ReDim MyArray(1 To 10, 1 To n, 1 To 20)
mymatrix2 = MyArray
End Function
Function mymatrix3(n)
Dim MyArray()
ReDim MyArray(1 To 10, 1 To n, 1 To 20)
mymatrix3 = MyArray(1, 1, 2)
End Function

The previous presents three functions of the array. Using function mymatrix(5) in
the latest Excel version will return a 5 × 10 table of zero. Typing mymatrix2(5) will
return the #VALUE (VBA cannot express a three-dimensional matrix in an Excel
spreadsheet). Typing mymatrix5(5) will return a single value of zero. If the VBA
function returns a two-dimensional matrix, another VBA function can use that output.

1.2.4 for . . . next loop structure

The numerical method is solving mathematical or physical equations using comput­
ers. In other words, numerical methods solve complicated problems by using many
repetitive steps of simple calculations. Computer software code is developed to serve
this purpose, that is, making computers repeat simple calculations. As a result, the
loop structure is an essential part of most numerical methods.
VBA has two types of iterative loops: For Next and Do � � � While structures. For
Next loops have two options: For Each Next and For Next. These structures are
useful when the number of repetitions is predetermined. For example, in the For
Each Next loop, the iteration loops through a collection of items. The collection can
be a collection of objects or a list of numbers. Examples of collections of objects can
be cells in a range, worksheets in a workbook and any other object you interact with
within Excel. In the case of cells, the order of the loops is left to right, then down
(starting in the first cell in the range and looping to the next column in the same row,
then going to the next row).
Example of For Each Loop: the periodic table is tabulated in an Excel file. The table
has four properties for each element: atomic number, atomic symbol, English element
name and atomic mass. Write a VBA to find the molecular weight of a symbol.
Since the element symbols are not in one-dimensional rows or columns, the For
Each Loop is needed. The function in code (Code 1.5) can be used in any spreadsheet
 Introduction to Excel and VBA
TABLE 1.2
Periodic Table (Short Version) in Excel Spreadsheet
A B C D E F G H I J K L M N O
1 1
2 H
3 hydrogen
4 1.008
5 3 4 5 6 7
6 Li Be B C N
7 lithium beryllium boron carbon nitrogen
8 6.968 9.012 10.81 12.01 14.01
9 11 12 13 14 15
10 Na Mg Al Si P
11 sodium magnesium aluminium silicon phosphorus
12 22.99 24.31 26.98 28.09 30.97
13 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33
14 K Ca Sc Ti V Cr Mn Fe Co Ni Cu Zn Ga Ge As
15 potassium calcium scandium titanium vanadium chromium manganese iron cobalt nickel copper zinc gallium germanium arsenic
16 39.10 40.08 44.96 47.87 50.94 52.00 54.94 55.85 58.93 58.69 63.55 65.38 69.72 72.63 74.92
17 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51
18 Rb Sr Y Zr Nb Mo Tc Ru Rh Pd Ag Cd In Sn Sb
19 rubidium strontium yttrium zirconium niobium molybdenum technetium ruthenium rhodium palladium silver cadmium indium tin antimony
20 85.47 87.62 88.91 91.22 92.91 95.95 98 101.1 102.9 106.4 107.9 112.4 114.8 118.7 121.8
21 55 56 71 72 73 74 75 76 77 78 79 80 81 82 83
22 Cs Ba Lu Hf Ta W Re Os Ir Pt Au Hg Tl Pb Bi

9
23 cesium barium lutetium hafnium tantalum tungsten rhenium osmium iridium platinum gold mercury thallium lead bismuth
24 132.9 137.3 175.0 178.5 180.9 183.8 186.2 190.2 192.2 195.1 197.0 200.6 204.4 207.2 209.0
10 Calculations for Process Engineering Using Excel VBA

within the Excel workbook. The code can be used in other Excel work in the same
folder by using the following:
Workbooks(“example1–1.xlsm”). Worksheets(“PeriodicTable”). Range(“C1:BP46”)

Code 1�5� Example of For Each Next loops to calculate the molecular
weight�
Function MW(name)
Dim Cell As Range
MW = “Element name is not found”
For Each Cell In Worksheets(“PeriodicTable”).
Range(“C1:BP46”)
If StrComp(name, Cell.Value) = 0 Then
MW = Cell.Offset(2, 0). Value ‘MW is stored
2 cells below the Symbol
End If
Next Cell
End Function

The For Next loop follows a sequence of numbers (the data is in Table 1.2). The
default step (1) is used if the step is omitted. Alternatively, any integer (including
negative values) can be used. The For Next loop is more useful for numerical calcu­
lations and will be used frequently in this book.

1.2.5 do . . . loop structures

The second type of VBA repetition loop is the Do loop structure. The critical differ­
ence between Do . . . loops and the For Next loop is that the termination condition
(and the number of repeated iterations) is determined within the loop. There are four
options for Do loops, as presented in Table 1.3. The conditions are adjusted to the
loop structures. In addition, VBA allows the While � � � Wend structure (structure 5
in Table 1.3), which is equivalent to the Do While � � � loop.

TABLE 1.3
Different Do Loop Formats
Loop format Description Example
Do While . . . Loop Runs 0 or more times while the Do While “condition is met”
condition is true Loop
Do . . . Loop While Runs 1 or more times while the Do . . .
condition is true Loop While “condition is met”
Do Until . . . Loop Runs 0 or more times until the Do Until “condition is met”
condition is true Loop
Do . . . Loop Until Runs 1 or more times until the Do . . .
condition is true Loop Until “condition is met”
While . . . Wend Same as Do While . . . loop
Introduction to Excel and VBA 11

Code 1.5 shows two loop structures that are used to find the smallest number in the
Fibonacci series which is larger than x.

Code 1�6� Examples of Do While � � � loop and Do � � � Loop Until�

Function LFn(x)
Dim f0, f1, f2
f0 = 0: f1 = 1
Do While f2 < x
f2 = f0 + f1
f0 = f1: f1 = f2
Loop
LFn = f2
End Function
Function LFn _ 2(x) ‘Excel don’t accept LFn2
Dim f0, f1, f2
f0 = 0: f1 = 1
Do
f2 = f0 + f1
f0 = f1: f1 = f2
Loop Until f2 > x
LFn _ 2 = f2
End Function

A common problem with the loop is the requirement for initial conditions. For
example, the numerical method requires code to repeat the iterations, so the calcu­
lated error is smaller than 0.01. The calculated error is updated within the loop. One
can use the condition as “calculated_error > 0.01” with structure 1 or 2. Alterna­
tively, one can use “calculated_error < 0.01” and structure 3 or 4. Either way, the
initial value of calculated_error must be initialised before the loop.

1.2.6 options in VBa code

1.2.6.1 Option Base
The VBA default base of an array and matrix is 0, meaning the first element’s index is
zero. This might cause confusion when doing numerical coding, in which the standard
practice uses an index from 1. However, this can be overcome by using Option Base 1.

1.2.6.2 Option Explicit

Using Option Explicit forces the coder to declare all their variables. Using Option
Explicit at the start of your Excel VBA code is strongly recommended. The option
will avoid typos in variables, which are difficult to find visually. A typical mistake is
mistyping “xl” as “x1”.
The option can be turned on automatically for all newly created VBA modules by
selecting “Require variable declaration” in VBA Options (Figure 1.3).
12 Calculations for Process Engineering Using Excel VBA

FIGURE 1.3 Screenshot of options in VBA.

1.2.7 get, set or change spreadsheet cell Value

A sub or a function can obtain the value of the cells. The previous example can be
rewritten so that the sub will export the output to a specific cell on an active spread­
sheet. Hence, the cell’s value can be set or changed (if it is not empty). This can be
done using the Range or Cells object.

Range(“A2”).Value=1
Cells(2,1).Value =1

It should be noted that the first index in Cells is the row number, and the column
number is the second index (1 corresponds to column A). For example, Cells(10,2)
corresponds to B10, and Cells(2,10) corresponds to cell J2. To get the value from cell
A2 of the active spreadsheet, one can use the following:

x = Range(“A2”).Value
x = Cells(2,1).Value

The command is useful for selecting values as inputs for a numerical procedure.
Alternatively, one includes the cells as arguments for the function.
Introduction to Excel and VBA 13

Code 1�7� Calculate the average value of three numbers stored in cells
A1, A2 and A3, and export the value to cell B1�
Sub ave _ 3()
Dim ave
ave = (Range(“a1”).Value + Range(“a2”).Value +
Range(“a3”).Value) / 3
Range(“B1”).Value = ave
End Sub

Function ave _ 4()

Dim ave
ave = (Range(“a1”).Value + Range(“a2”).Value +
Range(“a3”).Value) / 3
Range(“B1”).Value = ave
‘Worksheets(“Sheet1”).Range(“B1”).Value = ave
ave _ 4 = ave
End Function

Sub ave _ 5()

Dim anything
anything = ave _ 4()
End Sub

Please note that the function ave_4 does not work by being directly called from
an Excel spreadsheet, that is, using it as a user-defined function. The reason is how
Excel handles VBA functions. When Excel uses a user-defined function, the function
can only access the ranges handed to it via parameters. Any access to other ranges
will result in a circular reference and potential cancellation of the execution. How­
ever, if a function is called within a subroutine, Excel does not mind. Hence, using
sub ave_5 in combination with function ave_4 will generate the same result as sub
ave_3.

1.2.8 spreadsheet functions

Some, but not all, worksheet functions in Excel can be called to use inVBA using the Work-
SheetFunction object. The full list can be found on the Microsoft website: https://docs.
microsoft.com/en-us/office/vba/excel/concepts/events-worksheetfunctions-shapes/
list-of-worksheet-functions-available-to-visual-basic.
The appendix of this book also contains a list of useful functions (mostly numer­
ical) and their explanations. Some of the useful functions are discussed here. Visual
Basic functions do not use the WorksheetFunction qualifier. Therefore, a VBA func­
tion may have the same name as a Microsoft Excel function and work differently.
For example, Application�WorksheetFunction�Log and Log will return different
14 Calculations for Process Engineering Using Excel VBA

values. In addition, Excel and VBA have an extended list of text-related functions
not included in this book.

1.2.8.1 Pi()
Excel has the function Pi(), which returns the value of Pi accurately to 15 digits. The
function can be called in VBA.

Dim pi
pi = WorksheetFunction.pi()

1.2.8.2 Sum, Average, Min, Max

These functions return the operation on a set of values. The values can be included
individually or entered as an array.

x _ aver= WorksheetFunction.average(a(1), a(3), (a(5))

x _ sum = WorksheetFunction.sum(x())

1.2.8.3 Round, Roundup, Rounddown

Round, Roundup and Rounddown return a number to the nearest decimal. These
functions require two arguments, the value and the position of the decimal place.
The position can be negative or positive. If the position is zero, the function returns
the nearest (either up or down) integer. The ROUND function rounds a number up
or down, depending on the value of the last digit. Examples of round, roundup and
rounddown are shown in the following.

x= WorksheetFunction.roundup (12.01,0)’ return 13

y= WorksheetFunction.rounddown(14.91, 1) ‘return 14.9
z= WorksheetFunction.round(101.91,-1)’ return 100
w= WorksheetFunction.round(101.91,1)’ return 101.9

1.2.8.4 Ln, Log and Log10

VBA has its own Log function, which returns the natural logarithm of a number.
Excel has three logarithm functions: LN, LOG and LOG10. LN and LOG10 require
one argument, and the base is fixed (at the natural base and 10, respectively). Spread­
sheet function LOG requires two arguments, in which the second argument is the
base of the logarithm.
Within VBA is a function LOG, which requires only one argument. This VBA
LOG function is the same as WorksheetFunction�LN. To evaluate the logarithm of
any base without using a spreadsheet function, one can use the indirect methods in
Code 1.8.
Introduction to Excel and VBA 15

Code 1�8� Two examples of logarithms in VBA�

Function log _ anybase(x,base)
log _ anybase= log(x)/log(base)
End Function
Function log _ any2(x,base)
log _ any2= WorksheetFunction.log(x,n)
End Function

1.2.8.5 Tanh, Cosh, Sinh, Atanh, Acosh, Asinh

These are hyperbolic and inverse hyperbolic functions and can be used in standard
mathematic formulae. Excel and VBA have normal trigonometric functions (the sine,
the cosine, the tangent and their reciprocals).

1.3 AVOIDING ERRORS IN VBA PROCEDURES

1.3.1 errors
Making errors is part of programming. For Excel/VBA, errors can come in many forms
(Table 1.4). Finding errors can be frustrating and take much work. For numerical meth­
ods, sometimes the error does not come from the coding but from the values of inputs.
Two common errors in using VBA are #REF! and #VALUE!. The #REF! error
can happen when the VBA function name has some letter followed by a number, for
example, PF2. This can be easily avoided by adding a special symbol, as in PF_2.
The #VALUE! error can happen when the function has invalid calculations, such as
being divided by zero, taking the square root of a negative number or having a calcu­
lated value outside the range of a predefined variable. The source of the error can be
tricky to find when it arises from the secondary function.
One way to prevent errors is to use IsError, IsNull or the Excel spreadsheet func­
tions IsNA and IsNumber. A list of these functions and their operation can be found
in the appendix.

TABLE 1.4
Errors When Using VBA Functions in Excel Spreadsheets
Error Meaning
#NAME? This error is displayed when Excel does not recognise text in a formula. For example, the
name of a function may need to be spelled correctly.
#N/A Excel displays this error when a value is unavailable to a function or formula.
#NULL! A space was used in formulas referencing multiple ranges; a comma separates range
references.
#NUM! A formula has invalid numeric data for the type of operation (including VBA functions).
#REF! A reference is invalid.
#VALUE! The wrong type of operation or function argument is used.
16 Calculations for Process Engineering Using Excel VBA

Code 1�9� Avoiding dividing by text input or zero�

Function dividing(x)
If WorksheetFunction.IsNumber(x) Then
If x <> 0 Then
dividing = 1 / x
Else
dividing = “input is zero”
End If
Else
dividing = “input is not a number”
End If
End Function

1.3.2 endless loop

This is the worst problem for coding. Since most numerical methods employ a repeat­
ing loop, it is quite a common problem. The first approach is to use a For Next loop
instead of a While loop. When it is necessary to have a While loop, a limitation on the
number of iterations should be used. Examples of loops are shown in Code 1.10. These
two codes will return 89 for any x greater than 55. The answers are incorrect (as with
the functions in Code 1.7), but the code will stop. In this book, the technique is applied
often. It is a simple and effective method to avoid endless loops. It is noteworthy that
the two combined conditions use different logical functions (And and Or).

Code 1�10� Two While loops with a maximum number of iterations�

Function LFn _ 3(x)
Dim f0, f1, f2, i
f0 = 0: f1 = 1
Do While f2 < x And i < 10
f2 = f0 + f1
f0 = f1: f1 = f2
i = i + 1
Loop
LFn _ 3 = f2
End Function

Function LFn _ 4(x)

Dim f0, f1, f2, i
f0 = 0: f1 = 1
Do
f2 = f0 + f1
f0 = f1: f1 = f2
i = i + 1
Introduction to Excel and VBA 17

Loop Until f2 > x Or i > 10

LFn _ 4 = f2
End Function

1.3.3 displacing immediate Value

The immediate window (Figure 1.4) can be turned on by pressing Ctrl+G. VBA
allows displacing immediate values, which are calculated during the numerical
model. This is very helpful for debugging the program. One way is using the mes­
sage box Msgbox (as demonstrated in Figure 1.2). The other way is using the Debug�
Print tool. Unlike Msgbox, Debug�Print eliminates the need for any confirmation
or acknowledgment every time. It is also used to show the values of variables by
displaying a log of returned values in the immediate window.
The tool effectively evaluates a code to confirm that it is working properly and
detect any bugs in the code. One of the common uses is to include the index and
calculated values of each iteration within the loop to see if the iterations are going in
the right direction.

1.3.4 Breakpoints
The breakpoint specifies the line in your code where VBA should pause the execu­
tion of the macro operation when debugging the code. The shortcut key for adding
a breakpoint is F9. Create a breakpoint, position the cursor on the specific line of
your code where you want VBA to pause and press F9. Alternatively, locate the
line of code where you want to add the breakpoint and click in the left grey margin
beside the line. A dark red dot will appear on the margin beside the specified line to
indicate that the breakpoint has been created. The specified line of code will also be
highlighted with the same colour (Figure 1.4).

FIGURE 1.4 Screenshot of the immediate window in VBE.

18 Calculations for Process Engineering Using Excel VBA

To remove the breakpoint, click on the dot again or press F9 with the cursor posi­
tioned in the highlighted line of code. If there are many breakpoints, the user can
remove them simultaneously by pressing Ctrl+Shift+F9.
Once the code runs to a breakpoint, VBA stops and highlights the breakpoint in
yellow. Then the user can continue the code by pressing F5. Figure 1.4 shows the
immediate values of the loop by inserting “Debug�Print i,f2” in function LFn_4
(Code 1.10).

1.4 USEFUL EXCEL TOOLS FOR ENGINEERING

Excel has two powerful calculation tools, Solver (Figure 1.5) and Goal Seek
(Figure 1.6). These tools can be used to find the root of the equation and the maxi­
mum and minimum.

FIGURE 1.5 Screenshot of Solver.

Introduction to Excel and VBA 19

FIGURE 1.6 Screenshot of Goal Seek.

Solver has three algorithms. While the tools are practical in many cases, they do
not show the underlying calculations. Thus, Solver and Goal Seek are not covered
in this book.
 2 Roots of an Equation
2.1 INTRODUCTION
Solving an equation is the most basic mathematical challenge. This chapter presents
some methods to solve an equation, their shortcomings, and their advantages. More
importantly, numerical methods often require solving a pre-arranged equation and
then using the output for further operation. Consequently, the methods are presented
in VBA forms, which can be routinely employed with other methods to solve com­
plicated problems in subsequent chapters.

2.2 BISECTION METHOD

The bisection method is one of the simplest numerical methods. While the method
does not give an exact solution, it can control the accuracy. It should be noted that the
method does not solve for an exact root of the equation. Instead, the method identifies
an interval that contains a root. The method controls the size of the interval. When the
interval is diminished, the error becomes insignificant, and the root is the interval’s
midpoint. The method starts with an initial interval (x1,x2) with a change of sign.
Subsequently, the value of the function at the midpoint, x3, is calculated.

x1 + x2
x3 = Eq. 2.1
2
Depending on the value of f(x3), a new interval is selected, either (x1,x3) or (x3,x2).
The halving procedure is repeated until the desired accuracy is met. The application
will depend on the nature of the problem. In this chapter, the method is presented
with flowcharts and codes. The methods can be combined with other functions to
solve complicated problems. For example, the combined shooting/bisection method
is used for heat transfer—the combined bisection/tear analysis for solving material
balance with a recycling stream.
For coding, there are many variations of the method. Here a simple example is
used to demonstrate these variations.

EXAMPLE 2�1� SOLVING A NON-LINEAR EQUATION�

PROBLEM STATEMENT�
Find a positive root, between 0.5 and 2, of the following equation:

x 5 ­ 3x + 2 sin ( x ) = 0 Eq. 2.2

The required error of the solution is less than 0.001.

20 DOI: 10.1201/9781003364511-2
Roots of an Equation 21

Solution
The halving procedure can be applied with three options in each iteration.

Code 2�1� Bisection method with a Do While loop�

Option Explicit
Function Bisection(a, b)
Dim x, Error, f1, f2, f3
Error = 0.001 ‘error
f1 = a ^ 5–3 * a + 2 * Sin(a): f2 = b ^ 5–3 * b + 2 *
Sin(b)
If f1 * f2 <= 0 Then ‘The root is between a and b
Do While (b–a) > 2 * Error
x = (a + b) / 2 ‘A first estimate of the root by the
midpoint
f3 = x ^ 5–3 * x + 2 * Sin(x)
If f3 = 0 Then
GoTo SkipPoint
ElseIf f1 * f3 < 0 Then ‘The root is between a
and (a + b)/2
b = (a + b) / 2 ‘Replace the upper estimate
Else ‘The root is between (a + b)/2 and b
a = (a + b) / 2 ‘Replace the lower estimate
End If
Loop
SkipPoint:
Bisection = x
Else
Bisection = “Wrong initial estimates” ‘The root is
not between a and b
End If
End Function

There are some important notes on the VBA function in Code 2.1. First, the
function is evaluated within one line to be typed directly. However, the func­
tion evaluation can be set as a separate VBA function and called in this function
several times. Such an arrangement can help to reduce typos. Furthermore, the
separate function can be called in an Excel spreadsheet directly and checked if the
selected interval (a,b) contains a change-of-sign. Sometimes function evaluation
requires multiple calculation steps and cannot be typed within one line. Hence,
it is advised to set the equation as a separate VBA function, as in the following
example.
The second noteworthy comment is about the first option in the If . . . Elseif struc­
ture, f ( x ) = 0 . This option may help to reduce the number of iterations/calculations,
22 Calculations for Process Engineering Using Excel VBA

but it rarely happens. If there are only two options, the code will be much simpler.
Finally, the usage of SkipPoint: (or Exit Loop) is not recommended. Thus, the option
f ( x ) = 0 should be removed in most cases.

2.2.1 Bisection method with control error in VariaBle

If the required condition is an error in x, the number of iterations can be determined
by

x1 ­ x2
n= Eq. 2.3
tolx

Hence, the For � � � Next structure can be used. The problem is solved, with the
required error, tolx, being an argument, using the code to find a solution between
(0.5, 2) with an error of 0.0001.

Code 2�2� A simple bisection using a For � � � Next loop�

Function BisectionX(Lo, Up, tolx)
Dim Mi, c1, c3, i, n
n = Abs(Up–Lo)/tolx
‘*****************************************
For i = 1 To n
Mi = (Lo + Up) / 2
c1 = f1(Lo): c3 = f1(Mi)
If c1 * c3 < 0 Then
Up = Mi
Else
Lo = Mi
End If
Next i
‘****************************************
BisectionX = Mi
End Function
Function f1(a)
f1 = a ^ 5–3 * a + 2 * Sin(a)
End Function

As the function is separated from the bisection function instead of using the
direct formula as in Code 2.1. The previous format will be more useful for
other cases. Notably, within each iteration, this VBA code requires two function
evaluations, f1(Lo) and f1(Mi). The extra evaluations can be inefficient when
the function evaluation is computationally intensive (see examples in the next
chapter).
Roots of an Equation 23

EXAMPLE 2�2� DETERMINING THE FRICTION FACTOR FROM

THE COLEBROOK EQUATION�
PROBLEM STATEMENT
The Colebrook equation is an implicit equation that combines experimental results
of studies of turbulent flow in smooth and rough pipes (Shashi Menon 2015). For
conduits that are flowing full of fluid at large Reynolds numbers, it is given as:

1 æ e / Dh 2.51 ö
= ­2log10 ç + ÷ Eq. 2.4
ç ÷
f è 3.7 Re f ø

where:
f is the Darcy friction factor, ε (m) is roughness height, Dh (m) is the inside
diameter of the pipe and Re is the Reynolds number.
Part A� Calculate f with a required error less than 0.001 for the following condi­
tions: e = 0.001 m, Dh = 0.05 m and Re = 4000 . The initial range is between
0.001 and 1.
Part B� Calculate f when the required difference between the left-hand and
right-hand sides of the equation is less than 0.001.

Solution
For part A, the error is specified for the friction factor, f. Hence, the initial values can
determine the number of iterations, and a For . . . Next loop structure (Figure 2.1)
can be used. The While . . . Wend loop structure (Figure 2.2) can also be used, with
the condition of err > tolx.

FIGURE 2.1 Flowchart for bisection code with a For . . . Next loop.
24 Calculations for Process Engineering Using Excel VBA

FIGURE 2.2 Flowchart for bisection code with a While . . . Wend loop.

Code 2�3� Bisection code with a For � � � Next loop and a single function
evaluation�
Function BisectionX(Lower, Upper, e, D, Re)
Dim k, n, M, err, tol, c1, c2, c3
tol = 0.001
n = (Upper–Lower)/tol
c1 = colebrook(Lower, e, D, Re)
For k = 1 To n
M = (Lower + Upper) / 2: c3 = colebrook(M, e, D, Re)
If c1 * c3 < 0 Then
Upper = M
Else
Lower = M:c1=c3
End If
Next k
BisectionX = M
End Function
Function colebrook(f, e, D, Re)
colebrook = (1 / f ^ 0.5) + 2 * Log((e/(D * 3.7)) +
(2.51 / (Re * (f ^ 0.5)))/Log(10))
End Function
Roots of an Equation 25

The previous bisection code has removed the f(x3) = 0 option. Furthermore, the calcula­
tion steps are rearranged so that each iteration requires only one function evaluation, c3=
f(x3). It should be noted that the initial values have to be selected so that there is a change
of sign. As a result, evaluation of c2 is not required. However, a validation of sign change
(such as in Code 2.1) might be needed in the general case. For parameter values, Code
2.3 returns f = 0.010743043. While the answer is sufficiently good, there is no guarantee
that the difference between the left and right sides of Eq. 2.4 is smaller than 10–4.

Code 2�4� Bisection code with limitation in function value, using the
same objective function in Code 2�3�
Function BisectionF(Lower, Upper, e, D, Re) ‘return f
Dim M, tol, c1, c2, c3, k
tol = 10 ^ -5
M = (Lower + Upper) / 2
c1 = colebrook(Lower, e, D, Re)
c3 = colebrook(M, e, D, Re)
k = 0
While Abs(c3) > tol And k < 100
If c1 * c3 < 0 Then
Upper = M
Else
Lower = M: c1=c3
End If
M = (Lower + Upper) / 2: c3 = colebrook(M, e, D, Re)
k = k + 1
Wend
BisectionF = M
End Function

Part B� To achieve the required accuracy, the error in function colebrook is used as the
termination condition, as in Code 2.4. In this case, the number of iterations will depend
on the specific tolerance of the function; thus, the number of iterations is unknown.
Consequently, the For . . . Next loop is not suitable. Code 2.4 uses the While . . . Wend
structure of Figure 2.2, with a small variation in testing conditions. The method can be
applied to part A as well (by replacing Abs(c3) with Abs(Upper-Lower)).

2.2.2 Bisection method with scanning

In the previous examples, the bisection method needs an appropriate range before
halving. The two limits are manually selected as initial guesses in the question. For
a general problem, the selection of the two limits is not obvious. Consequently, it is
advantageous to determine the limits identified via the scanning method. In a scanning
procedure, only one starting point is needed. The scanning can be done with a constant
or increasing step size. Hence scanning can identify the limits, which include a change
26 Calculations for Process Engineering Using Excel VBA

of sign. Since the scanning procedure requires function evaluations, the last interval
from the scanning step can be used as the starting region for the bisection method.

Code 2�5� Scanning and bisection code�

Function BisectionS(Lower, e, D, Re)
Dim step, Upper, M, tol, c1, c3, j, jmax
step = Lower / 2: tol = 10 ^ -7: jmax = 100
Upper = Lower + step
j = 0
Do Until c1 * c3 < 0 And j < jmax
step = step * 2
Lower = Upper: Upper = Lower + step
c1 = colebrook(Lower, e, D, Re): c3 = colebrook(Upper,
e, D, Re)
j = j + 1
Loop
If j = jmax Then
BisectionS = “There is no suitable interval for the
bisection method”
Else
M = (Lower + Upper) / 2: c3 = colebrook(M, e, D, Re)
While Abs(c3) > tol
If c1 * c3 < 0 Then
Upper = M
Else
Lower = M: c1 = c3
End If
M = (Lower + Upper) / 2: c3 = colebrook(M, e, D, Re)
Wend
BisectionS = M
End If
End Function

The previous code can be applied with all the parameters in Example 2.1, with a
starting point of 0.001. After the scanning part, the code will find a suitable interval
between 0.0085 and 0.0165 after four scanning steps. In the case of multiple solu­
tions, scanning can target the solution.

2.2.3 failures of the Bisection method

There are special cases where the Bisection method fails. A classic example is the
function:

1
f ( x) = Eq. 2.5
x
Roots of an Equation 27

The function has a change-of-sign interval between a positive and a negative

point, yet it does not have a solution. Trying to solve this equation with Code 2.4
without a limit on the number of iterations will result in an endless loop.
Another kind of failure of bisection happens when the target function has a solu­
tion but no change-of-sign interval. An example function is:

f ( x ) = x2 Eq. 2.6

This equation has a root x = 0, but both sides of the root are positive. There is
an effective way to overcome these failures. In these cases, one can try to minimise
the square of f(x) within a specific range. A minimum of zero will indicate a root,
whereas a non-zero minimum indicates no feasible root. To find a minimum of f 2(x),
one of the optimisation methods in Chapter 4 can be used.

2.3 NEWTON’S AND SECANT METHODS

The bisection method is straightforward but sometimes inefficient. Alternatively,
there are other numerical methods to get the equation’s root faster, requiring fewer
iterations. Newton’s method is one example. In this method, the next guessed value
is determined from the function and first derivative of the function at the previous
value. The iteration formula is:

f ( xi )
xi +1 = xi ­ Eq. 2.7
f ’( xi )

Unfortunately, the equation requires a derivative formula of the function f. In prac­

tice, this can be problematic. Some functions and correlations are not in the explicit
form (such as the Colebrook equation in Eq. 2.4). Consequently, a finite-difference
approximation of the derivative can be used:

f ( xi ) ­ f ( xi ­1 )
f ’( xi ) = Eq. 2.8
xi ­ xi ­1

Hence, Eq. 2.7 can be reduced to:

( xi ­ xi ­1 )
xi +1 = xi ­ f ( xi ) Eq. 2.9
f ( xi ) ­ f ( xi ­1 )

The previous method is named the Secant method (Chapra and Canale 2010).
While this method requires two values to calculate the following value (Figure 2.3),
these two values should be close to each other (to avoid fluctuation and divergence).
The termination condition can be based on the error in x or f. The solution will de­
pend on the initial point and step size selection.
28 Calculations for Process Engineering Using Excel VBA

FIGURE 2.3 Two gradient-based methods, Secant and Regula Falsi.

A VBA function of the secant method is presented in Code 2.6., which is applied
to the same Colebrook equation.

Code 2�6� Secant method�

Function secant(x1, x2, tol, e, D, Re)
Dim k, f1, f2, x3
f1 = colebrook(x1, e, D, Re): f2 = colebrook(x2, e, D, Re)
k = 0
While Abs(f2) > tol And k < 100
x3 = x2–f2 * (x2–x1)/(f2–f1)
x1 = x2: x2 = x3: f1 = f2: f2 = colebrook(x3, e, D, Re)
k = k + 1:
Wend
secant = x3
End Function

Using the code with x1 = 0.001, x2 = 0.002 and tol = 10–7, one can get the same
answer after eight iterations. It should be noted that the answer is extremely sensi­
tive to the initial guesses. A bad selection, for example, 0.001 and 0.1, may lead to
a negative value of x3 and return #VALUE! (since the Colebrook function valuates
log(x) and x0.5).

2.4 THE REGULA FALSI METHOD

As seen previously, the Secant method can significantly reduce the number of iter­
ations. However, it cannot verify whether it will lead to a solution. In that aspect,
bisection is more reliable once the values at the two limits are validated. The Regula
Falsi, or false position, is a compromise between the two approaches. Historically,
this method was documented in cuneiform tablets from the Old Babylon period,
1800 BC (Chabert 1999).
Roots of an Equation 29

The Regula Falsi method can be considered a mix between the bisection and secant
methods. The method starts with a change-of-sign interval, as with the bisection
method. Subsequently, the dividing point is selected by linear regression between the
two points (Figure 2.3) instead of the mid-point. The iterative formula is:

f ( x1 )
x3 = x1 + ( x1 ­ x2 ) Eq. 2.10
f ( x1 ) ­ f ( x2 )

By comparing the signs of the f(x3), the next interval can be either (x3,x1) or (x2,x3).
The linear regression formula can solve the friction factor problem. The method can­
not be used with the For . . . Next loop because the number of iterations cannot be
determined in advance.

Code 2�7� Regula Falsi method with a maximum limit on iterations�

Function PF(Lower, Upper, e, D, Re)
Dim M, tol, c1, c2, c3, k, output(1 To 2, 1 To 3)
tol = 10 ^ -7: k = 0
c1 = colebrook(Lower, e, D, Re):
c2 = colebrook(Upper, e, D, Re)
M = Lower–c1 * (Upper–Lower)/(c2–c1):
c3 = colebrook(M, e, D, Re)
While Abs(c3) > tol And k < 1000
If c1 * c3 < 0 Then
Upper = M: c2 = c3
Else
Lower = M: c1 = c3
End If
M = Lower–(Upper–Lower) * c1 / (c2 – c1):
c3 = colebrook(M, e, D, Re)
k = k + 1
Wend
PF = M
End Function

2.5 CONCLUDING REMARKS

There are many methods, with advantages and disadvantages, to numerically cal­
culate a root of an equation. The selection will depend on the specific nature of the
equation and the initial guesses. In general cases, either the bisection or false-position
method is recommended, depending on the required error. If the control error is in
the variable, use a bisection code (similar to Code 2.3). If the control error is in the
function, the first choice should be the false-position method (Code 2.7).
30 Calculations for Process Engineering Using Excel VBA

2.6 FURTHER EXAMPLES FOR SELF-PRACTICE

EXAMPLE 2�3� DETERMINE THE NUMBER OF ITERATIONS

OF THE BISECTION CODE�
Modify the code in Codes 2.3 and 2.4 so that the number of iterations is shown
in the intermediate windows.

Answer
Insert this line (after the loop).

Debug.Print k

For Code 2.3, k = 1000. For Code 2.4, k = 25.

EXAMPLE 2�4� COMPARING EFFICIENCIES OF BISECTION

AND REGULA FALSI METHODS�
Modify the code in Codes 2.4 and 2.7 to include the number of iterations in
the output. Compare and discuss the results when the initial values are (0.001,
1) and (0.01, 1).

Answer
Change the function name to PF_2 and insert the following lines after the loop:

Dim output(1 To 2, 1 To 2)
output(1, 1) = “solution”: output(2, 1) = M
output(1, 2) = “# iters”: output(2, 2) = k
PF _ 2 = output

Also, do the same for the bisection method. The modified code will produce the
number of iterations. Using the same initial range (0.001, 1) and the required error,
10–7, the bisection method requires 30 iterations, while the Regula Falsi method re­
quires 93 iterations. In this particular instance, the bisection method is three times
faster. However, if the range is changed to (0.01, 1), the respective iterations are 27
and 7, which means the Regula Falsi method is about three times faster. The change
in efficiency is due to the non-linear nature of the Colebrook function within this
range. The Colebrook function for different values of f can also be calculated and
plotted to demonstrate the correlation between f and the VBA function graphically.
 3 Applications of Roots
of an Equation
3.1 INTRODUCTION
The previous chapter demonstrates some numerical methods to solve a single explicit
equation. This chapter presents some applications of the methods to engineering
problems. In these problems, the nature of the correlation/data input is more compli­
cated than an explicit equation. Therefore, the VBA code in the previous chapter is
modified to adapt to these problems.

3.2 ADVANCED EXAMPLE: ADIABATIC FLAME TEMPERATURE

The adiabatic flame temperature is the highest temperature of combustion, in which all
generated energy is used to heat the exhaust gases (Himmelblau and Riggs 2012). The
determination of this temperature requires material and energy balances. First, the mate­
rial balance is needed to determine the exhaust gas quantities. Second, the heat capacities
are integrated to determine the change of enthalpy. Finally, the temperature is re-evalu­
ated. Without a computer program, the temperature needs to be solved by a trial-and-er­
ror approach. In this section, VBA code is used to solve the problem. Furthermore, the
advanced combination of code can determine the excess air at any given flame temperature.

EXAMPLE 3�1� DETERMINING THE ADIABATIC FLAME

TEMPERATURE OF METHANE�
PROBLEM STATEMENT�
Consider the complete combustion of methane in an adiabatic reactor:

CH 4 + 2O2 ® CO2 + 2H 2 O

Both fuel and air enter the reactor at 25°C. Heat capacity equations (Cp in kJ/
gmol/°C and T in °C) for gases are determined from the following equation:

Cp = a + b ´ T + c ´ T 2 + d ´ T 3 Eq. 3.1

The heat capacity constants and heat of formation are given in Table 3.1. Assume
complete combustion and air containing 21% oxygen and 79% nitrogen.
Part A� Determine the adiabatic flame temperature of methane combustion
with air; the air is supplied 50% excess.
Part B� Determine the required excess air so the adiabatic flame temperature
is 1500°C.

DOI: 10.1201/9781003364511-3 31
32 Calculations for Process Engineering Using Excel VBA

TABLE 3.1
Heat Capacity Constants and Heat of Formation under Standard Conditions
(Himmelblau and Riggs 2012)
A B C D E F
1 Heat of formation Heat capacity constants
2 a b c d
3 CH4 –49.963
4 CO2 –393.25 36.110 4.233E-02 –2.887E-05 7.464E-09
5 H2O –241.835 33.460 6.880E-03 7.604E-06 3.593E-09
6 O2 0 29.100 1.158E-02 –6.076E-06 1.311E-09
7 N2 0 29.000 2.199E-03 5.723E-06 –2.787E-09

Solution
Part A� For complete combustion, all fuel CH4 reacts and forms CO2. Using the basis
of 1 gmol of CH4, the amount of the gases is easily calculated. It should be noted
that water is formed in vapour form. Following the material balances, the quantities
of the gas output are:

xCO2 = 1 gmol
xH2O = 2 gmol
xO2 = 2x gmol
xN2 = 2 × (79 / 21) × (1 + x) gmol

In this system, x is the percentage of excess air. For part A, x equals 0.5. The heat
of the reaction is calculated from the heat of formation:

J
°
DHrxn = DHˆ °f ( CO2 ) + 2DHˆ °f ( H 2 O ) ­ DHˆ °f ( CH 4 ) = ­826957 Eq. 3.2
mol
Consequently, the adiabatic flame temperature Td is satisfied by the following
equation:

å xi Cpi ( T ) dT =826957
Td
ò25
Eq. 3.3

where i = CO2, H2O, O2 and N2.

The integration in Eq. 3.3 is applied to all species with the heat capacity given in
Table 3.1. The values of the four coefficients are given in Table 3.1. Hence, three
functions are developed (Code 3.1). The first VBA function calculates the sum of the
integrations in Eq. 3.3 using temperature as an argument. The second VBA function,
ffT, calculates the difference between the left and right sides of Eq. 3.3. The third
VBA function calculates Td so that ffT equals zero. Input A of the two VBA functions
is the 4 × 4–cell range containing the heat capacity constants in Table 3.1. The code
determines Td at 1522.3°C when x =0.5.
Applications of Roots of an Equation 33

Code 3�1� Determining the adiabatic flame temperature of methane�

Option Base 1
Function enthalpyT(MC, T, xCO2, xH2O, xO2, xN2)
Dim i, ent(4)
For i = 1 To 4
ent(i) = MC(i, 1) * T + MC(i, 2) * T ^ 2 / 2 + MC(i,
3) * T ^ 3 / 3 + MC(i, 4) * T ^ 4 / 4
Next i
enthalpyT = ent(1) * xCO2 + ent(2) * xH2O + ent(3) *
xO2 + ent(4) * xN2
End Function

Function ffT(A, T, xCO2, xH20, xO2, xN2) ‘note use

H20 not H2O within this section- still works fine
ffT = enthalpyT(A, T, xCO2, xH20, xO2, xN2)–
enthalpyT(A, 25, xCO2, xH20, xO2, xN2)–826957
End Function

Function BisectionT(Lo, Up, A, excess, tolx)

Dim Mi, c1, c2, c3, i, n, xCO2, xH2O, xO2, xN2
n = Abs(Up–Lo)/tolx
xCO2 = 1: xH2O = 2: xO2 = 2 * excess: xN2 = (79 / 21)
* 2 * (1 + excess)
c1 = ffT(A, Lo, xCO2, xH2O, xO2, xN2)
For i = 1 To n
Mi = (Up + Lo) / 2: c3 = ffT(A, Mi, xCO2,
xH2O, xO2, xN2)
If c1 * c3 < 0 Then
Up = Mi:
Else
Lo = Mi: c1 = c3
End If
Next i
BisectionT = Mi
End Function

Part B� The BisectionT function in Code 3.1 can determine Td at a given excess,
but the question requests the reverse correlation. Hence, another code can be com­
bined with the previous VBA function. Code 3.2 (in combination with Code 3.1) pro­
duces x = 0.531. The combination takes a significant time if the limits are far apart.
It should be noted that this problem can be solved analytically by material bal­
ances. However, the calculation process is tedious with many constants. Writing two
short VBA functions is more efficient (as we already have the VBA functions in Code
3.1). In this case, all VBA functions must be in the same Excel file. For example,
34 Calculations for Process Engineering Using Excel VBA

the Excel spreadsheet needs to have heat capacity constants, as in Table 3.1. Using
multiple VBA functions in this fashion is necessary for many complicated numeri­
cal problems. In this instance, the code used bisection on another bisection method.
Since both functions require multiple iterations, an appropriate selection of initial
values can significantly reduce the calculation time.

Code 3�2� Calculating air excess to control adiabatic flame temperature�

Function fffT(excess, Tmax, A)
fffT = BisectionT(1000, 2000, A, excess, 0.1)–Tmax
End Function
Function bisectionXT(Lo, Up, Tmax, A, tolg)
Dim Mi, c1, c2, c3, j, m
m = Abs(Up–Lo)/tolg
c1 = fffT(Lo, Tmax, A): c2 = fffT(Up, Tmax, A):
For j = 1 To m
Mi = Lo + (Up–Lo) * c1 / (c1–c2)
c3 = fffT(Mi, Tmax, A)
If c1 * c3 < 0 Then
Up = Mi:
Else
Lo = Mi: c1 = c3
End If
Next
bisectionXT = Mi
End Function

3.3 ADVANCED EXAMPLE: TEAR ANALYSIS FOR MATERIAL

BALANCES WITH A RECYCLE STREAM
In some cases, the objective function is not expressed in a formula. Instead, the eval­
uations are done via multiple steps. One of the examples is sequential modelling for
process engineering. In this setup, the bisection method is applied to solve the sys­
tem. For example, implicit correlation can be solved by the bisection method.

EXAMPLE 3�2� TEAR ANALYSIS WITH A RECYCLE STREAM�

PROBLEM STATEMENT�
Ethyl chloride (CH3CH2Cl) can be manufactured by an exothermic reaction of
hydrogen chloride (HCl) with ethylene (CH2=CH2):

C2 H 4 + HCl ® C2 H 5 Cl
Applications of Roots of an Equation 35

The process can be simplified in the following diagram, adapted from exam­
ple 10.2–3 (Felder and Rousseau 2005). The fresh feed contains 50 kgmol/h
of HCl and C2H4 each (equal molar flow). The fresh feed is mixed with the
recycle stream before entering the reactor. The single-pass ethylene conver­
sion is 90%. After the reactor, the stream is separated into Product (containing
C2H5Cl only) and Stream S-4 (containing C2H4 and HCl). Stream S4 is split
into Waste and Recycle streams. Streams S4, Waste and Recycle have the same
composition.
Find the molar flow for the recycle stream if the waste stream has a flow
rate of 8 kgmol/h.

Solution
First, a mathematical model is developed to describe all material balances of the
process. Generally, each stream will have three chemicals (HCl, C2H4 and C2H5Cl).
Since the flow rates of HCl and C2H4 are always equal on all streams, the unknown
variables can be reduced to 1 and thus become a single-variable function. For other
cases, the tear analysis may produce a multiple-dimensional system (see Example
8.2) and require a multiple-dimensional method.
The modelling of this process is based on a tear analysis. Tear analysis is per­
formed by identifying recycle or loop streams in the process flow diagram and tem­
porarily “tearing” them to create separate input and output streams. This simplifies
the system into smaller sub-processes, allowing for more straightforward calculation
and simulation. Once the steady-state conditions of these smaller sub-processes have
been determined, the torn streams are reconnected.
For this particular problem, the tear stream can be either S1, S2, S4 or the recycle
stream. For convenience, the recycle stream is selected. Hence, the flow rate of the
recycle stream is assigned a random value, Rg. From Rg, the component flow rates
in recycle stream are calculated. Subsequently, the materials of other streams are
calculated from streams S1 to S2 to S4 (Figure 3.1). Consequently, the other side
of the tear is calculated from S4, resulting in a value of Rcal. The next step of tear
analysis is to adjust Rg so that Rcal = Rg. This can be done by a VBA function, as in
Codes 3.3 and 3.4.

FIGURE 3.1 Acetylene production with a recycle stream.

36 Calculations for Process Engineering Using Excel VBA

Code 3�3� Calculate the difference between the guessed and calculated
tear steam value (one variable)�
Function diffR(Rg, feed, Conv, W)
Dim S1(3), S2(3), S4(3), Pro(3), Waste(3), R(3)
Dim Rcal
R(1) = Rg / 2: R(2) = Rg / 2 ‘calculate component of R
S1(1) = feed / 2 + R(1): S1(2) = feed / 2 + R(2)
S2(1) = S1(1) * (1–Conv): S2(2) = S1(2) * (1–Conv):
S2(3) = S1(1) * Conv ‘we don’t need S(3) for calc
S4(1) = S2(1): S4(2) = S2(2)
R(1) = S4(1)–W / 2: R(2) = S4(2)–W / 2:
Rcal = R(1) + R(2)
diffR = Rg–Rcal
End Function

The function in Code 3.3. calculates the difference between Rg and Rcal. In this
code, the component flow rates of each stream are presented by an array of two
elements. The flow rates are calculated from each unit operation’s material balance.

Code 3�4� Solving tear analysis�

Function BisectionR(LowerR, UpperR, feed, Conv, W)
Dim x1, x2, x3, RelativeError, CalculatedError, f1,
f2, f3
RelativeError = 0.001 ‘Required relative error in %
CalculatedError = 1
x1 = LowerR: x2 = UpperR
x3 = (x1 + x2) / 2 ‘A first estimate of the root
by the midpoint
If diffR(x1, feed, Conv, W) * diffR(x2, feed, Conv, W)
<= 0 Then ‘The root is between L and U
Do While CalculatedError > RelativeError
f1 = diffR(x1, feed, Conv, W): f3 = diffR(x3,
feed, Conv, W):
If f1 * f3 < 0 Then ‘The root is between x1 x3
x2 = x3
Else
x1 = x3
End If
x3 = (x1 + x2) / 2 ‘Estimate of the root by the
midpoint
CalculatedError = Abs(x2–x1)
Loop
BisectionR = x3
Applications of Roots of an Equation 37

Else
BisectionR = “Wrong initial estimates”
End If
End Function

By applying this code, the flow rate of the recycle stream is determined at R = 2.22
mol/h.

3.4 ADVANCED EXAMPLE: DETERMINING THE VAPOUR

FRACTION FROM INTERNAL ENERGY AND SPECIFIC VOLUME
In some cases, the function cannot be evaluated by a formula, as in the previous
examples. Instead, the data are in discrete points, often presented in a table. Conse­
quently, the return values of variables can be evaluated from a data set by interpreta­
tion. VBA code can be developed for these cases to return the interpolated data. The
principle of bisection and false-position methods is still applicable in combination
with the linear interpretation.

EXAMPLE 3�3� CALCULATING STEAM QUALITY AT A

SPECIFIC ENERGY�
PROBLEM STATEMENT�
This problem is adapted from Example 9.16 (Himmelblau and Riggs 2012). In
this case, the system contains a mixture of steam and saturated water liquid.
The total internal energy and volume are specified:

Btu
Û t = 322.9 Eq. 3.4
lb
ft 3
Vt = 6.25 Eq. 3.5
lb
The steam and vapour saturated values are given in Table 3.2.
Determine the composition (vapour fraction) and temperature of the mixture.

Solution
The unknowns of the process are vapour fraction xv, total pressure P and temperature
T. The degree of freedom of the system is zero. There is a unique solution, a specific
combination of xv, P and T, to satisfy the two conditions in Eqs. 3.4 and 3.5. On the
other hand, the data are available in standard steam tables. In such tables, U and H are
given for both phases as functions of T (or P). An example of a steam table is given
in Table 3.2, which can be stored in an Excel spreadsheet.
38 Calculations for Process Engineering Using Excel VBA

TABLE 3.2
Steam Table for Saturated Vapour
Temperature Sat Pressure Specific Volume (ft3/lbm) Internal energy (Btu/lbm)
(F) (lb/in2) Sat. Liquid Sat. Vapour Sat. Liquid Sat. Vapour
32.018 0.08871 0.01602 3299.9 0 1021
35 0.09998 0.01602 2945.7 3.004 1019
40 0.12173 0.01602 2443.6 8.032 1015.6
45 0.14756 0.01602 2035.8 13.05 1012.2
50 0.17812 0.01602 1703.1 18.07 1008.9
55 0.21413 0.01603 1430.4 23.07 1005.5
60 0.25638 0.01604 1206.1 28.08 1002.1
65 0.30578 0.01604 1020.8 33.08 998.76
70 0.36334 0.01605 867.18 38.08 995.39
75 0.43016 0.01606 739.27 43.07 992.02
80 0.50745 0.01607 632.41 48.06 988.65
85 0.59659 0.01609 542.8 53.06 985.28
90 0.69904 0.0161 467.4 58.05 981.9
95 0.81643 0.01612 403.74 63.04 978.52
100 0.95052 0.01613 349.83 68.03 975.14
110 1.2767 0.01617 264.96 78.01 968.36
120 1.6951 0.0162 202.94 88 961.56
130 2.226 0.01625 157.09 97.99 954.73
140 2.8931 0.01629 122.81 107.98 947.87
150 3.7234 0.01634 96.929 117.98 940.98
160 4.7474 0.01639 77.185 127.98 934.05
170 5.9999 0.01645 61.982 138 927.08
180 7.5197 0.01651 50.172 148.02 920.06
190 9.3497 0.01657 40.92 158.05 912.99
200 11.538 0.01663 33.613 168.1 905.87
210 14.136 0.0167 27.798 178.15 898.68
212 14.709 0.01671 26.782 180.16 897.24
220 17.201 0.01677 23.136 188.22 891.43
230 20.795 0.01684 19.374 198.31 884.1
240 24.985 0.01692 16.316 208.41 876.7
250 29.844 0.017 13.816 218.54 869.21
260 35.447 0.01708 11.76 228.68 861.62
270 41.877 0.01717 10.059 238.85 853.94
280 49.222 0.01726 8.6439 249.04 846.16
290 57.573 0.01735 7.4607 259.26 838.27
300 67.028 0.01745 6.4663 269.51 830.25
Applications of Roots of an Equation 39

Mathematically, the problem requires a combination of T and xv (0< xv <1) so that:

xvUˆ v + (1 ­ xv ) Uˆ l = Uˆ t Eq. 3.6

xvVv + (1 ­ xv ) Vl = Vt Eq. 3.7

The four values (Ul, Uv, Vv, Vl) are determined by linear interpolation from the
table. The solution is based on the principle of the bisection method using two VBA
functions. In the first function, the temperature is an input, and the corresponding
total volume is the output. The function includes three steps:

1 For a given value of T, the corresponding Uv, Ul, Vv, Vl are obtained by
interpolating data.
2 xv is calculated by an interpolation formula of Uv, Ul to satisfy Eq. 3.6, U = 322.9
Btu/lbm. Note that if Uv < Ut or Ul >Ut, there is no physically feasible value of
xv. However, the formula can assign xv = 0 and xv = 1 under those conditions.
3 The total volume Vt is calculated by Eq. 3.7 from xv and Vv,Vl.

A bisection (or false-position method) is then applied to find T so that the previous
function, function Tx in Code 3.5, equals Vt. Finally, the determined Tx is combined
with another interpolation to find P.

Code 3�5� Calculating steam temperature and vapour fraction� The discrete
data (steam table) are stored in cells A2:F100 of the active spreadsheet�
Function Tx(t, U)
Dim Ul, Uv, i, fraction, volume, output(1, 1)
output(0, 0) = “total V”: output(0, 1) = “Vapour fraction”
For i = 3 To 200
If t <= Cells(i + 1, 1) And t >= Cells(i, 1) Then
Ul = Cells(i, 5) + (t–Cells(i, 1))/(Cells(i + 1, 1)–
Cells(i, 1)) * (Cells(i + 1, 5)–Cells(i, 5))
Uv = Cells(i, 6) + (t–Cells(i, 1))/(Cells(i + 1, 1)–
Cells(i, 1)) * (Cells(i + 1, 6)–Cells(i, 6))
Select Case U
Case Is < Ul
fraction = 0
Case Is > Uv
fraction = 1
Case Else
fraction = (U–Ul)/(Uv–Ul)
End Select
output(1, 0) = Cells(i, 4) * fraction + Cells(i, 3) *
(1–fraction)
output(1, 1) = fraction:
40 Calculations for Process Engineering Using Excel VBA

Tx = output
Exit Function
End If
Next i
Tx = 2000
End Function

Function BisectionTx(Lo, Up, tolt, U, V)

Dim Mi, c1, c3, i, n, ent1, ent3, output(1, 1)
n = Abs(Up–Lo)/tolt
ent1 = Tx(Lo, U): c1 = ent1(1, 0)–V
For i = 1 To n
Mi = (Lo + Up) / 2: ent3 = Tx(Mi, U): c3 = ent3(1, 0)–V
If c1 * c3 < 0 Then
Up = Mi
Else
Lo = Mi: c1 = c3
End If
Next i
output(0, 0) = “Temperature”: output(0, 1) = “Vapour
fraction”
output(1, 0) = Mi: output(1, 1) = ent3(1, 1)
BisectionTx = output
End Function

In this case, the vapour and enthalpy are determined via interpolation (instead of
the explicit equations from P or T). Using this code, the solution is T = 210°C and
x = 0.20. It should be noted that the function Tx returns a 2 × 2 matrix. The following
bisection code only requires an element of this output matrix. Since the types of two
variables, entl and ent3, are not specified, the VBA function assigns the whole 2 × 2
matrix to these variables. Subsequently, element (1,0) is used for the bisection proce­
dure. The method may not work for an array of other dimensions (see Example 3.6).

3.5 FURTHER EXAMPLES FOR SELF-PRACTICE

EXAMPLE 3�4� COMBINATION OF BISECTION AND FALSE­

POSITION METHODS�
Change Code 3.2 to the false-position method. Note the number of iterations.

Answer
The false-position method needs only one iteration. The reason is that the
Td-versus-x correlation is linear.
Applications of Roots of an Equation 41

EXAMPLE 3�5� AN ALTERNATIVE SOLUTION TO THE STEAM

PROBLEM IN EXAMPLE 3�3�
In step 2, xv can be found by interpolating the specific volume (instead of spe­
cific energy). What is the final answer?

Answer
The answers are Td = 206.7°C and xv = 0.21. These are more accurate than the
previous one. The reason is that the linearity of U-versus-T is better than V­
versus-T. The solution can be improved using the curve fitting for four values
in Table 3.2.

EXAMPLE 3�6� USING A ONE-DIMENSIONAL ARRAY AS THE

OUTPUT�
Redo Code 3.5 using the output of function Tx as a 1-dimensional array and
apply it to the bisection code. What would happen?

Answer and discussion

Replace output(1,1) with output(1).

output(0) = Cells(i, 4) * fraction + Cells(i, 3) *

(1–fraction)
output(1) = fraction:

For the second function, use the following:

ent1 = Tx(Lo, U): c1 = ent1(0) – V

The first function, Tx, still returns a correct answer (2 × 1 matrix) in an Excel
spreadsheet. The second function returns #VALUE. Maybe other Excel versions can
work with a 1-dimensional array! One way to overcome the error is using output (0,3).

EXAMPLE 3�7� DETERMINE THE COMPOSITION OF A BINARY

MIXTURE FROM THE REFRACTIVE INDEX�
The refractive index of the water-ethanol mixture is given in Table 3.3 (Herráez
and Belda 2006). Create VBA codes to find mole fraction, x, as a function of
the ethanol fraction, nD. Apply the code for finding x for nD = 1.3532.
42 Calculations for Process Engineering Using Excel VBA

TABLE 3.3
The Refractive Index of the Water-Ethanol Mixture
Ethanol Mole Fraction Refractive Index, nD
0 1.3326
0.101 1.3457
0.2001 1.3535
0.2987 1.3574
0.3332 1.3582
0.4087 1.3595
0.5051 1.3604
0.603 1.3614
0.6664 1.3618
0.7032 1.3617
0.7956 1.3614
0.9075 1.3609
1 1.3593

Hint
Modify the first function in Code 3.5.
 4 One-Dimensional
Optimisation
4.1 INTRODUCTION
Finding the optimum, either maximum or minimum, is a common problem in engi­
neering. The objective function can be cost, material, energy or time. In compari­
son, the input can be any physical variable. This chapter will deal with numerical
optimisation. Analytically, the optimum of an equation can be found to differenti­
ate the equation and solve for the root, f’(x) = 0. Various methods in the previous
chapter can find the root of the equation. However, such an approach has two
problems. First, one must clarify whether the optimum is a minimum or a maxi­
mum. Second, the differentiation might be challenging or impossible (for example,
implicit equations). The numerical methods in this chapter will overcome those
challenges.

4.2 REGION ELIMINATION

Similar to the bisection/Regula Falsi methods, the region elimination approach can
be applied to optimisation problems. The method starts with an initial region, which
contains the optimum. Consequently, two internal points are selected, and the values
of the target function at these two points are compared. Depending on the rela­
tive values and nature of the optimisation, a section of the original region is elimi­
nated. The iteration is repeated until the required tolerance is met. The comparison
is simple.
Let a and b be initial ends (a < b). Two values, x1 and x2, are selected between a
and b (x1 < x2). The function is evaluated at these two values for comparison. For
minimisation problems, the three elimination options are:

• If f(x1) < f(x2): the region (x2,b) is eliminated; x2 becomes the new b for the
next iteration.
• If f(x1) > f(x2): the region (a,x1) is eliminated; x1 becomes the new a for the
next iteration.
• If f(x1) = f(x2): two regions (a,x1) and (x2,b) are eliminated; x1 and x2 become
new a and b for the next iteration, respectively.

For maximisation, the opposite of conditions (1) and (2) is applied. Option 3 is
sometimes omitted (Figure 4.1). Similar to the bisection method, there are two im­
portant shortcomings of the elimination method. First, it does not guarantee that an
optimum exists. Second, it does not guarantee global optimum within the region

DOI: 10.1201/9781003364511-4 43
44 Calculations for Process Engineering Using Excel VBA

FIGURE 4.1 Golden section search.

(a,b). Generally, a graphic demonstration of the function can be used to verify the
number of optimums between (a,b).
In the literature, there are three ways to select the two interior points: (i) binary
search, (ii) Fibonacci search and (iii) golden section search. In the first search, the
two interior points are selected as close as possible to the midpoint of the region. In
this case, each iteration eliminates nearly half of the original region. While the binary
search is the most efficient way to divide the region, it requires two function evalua­
tions at each iteration. For some problems, the function evaluation is a time-consuming
step. Consequently, reducing the number of function evaluations while slightly
increasing the number of iterations is preferable.
The other two methods, Fibonacci and golden section, require only one function
evaluation per iteration (except the first iteration). Even though these two methods
require more iterations than the binary search, they might reduce the computing time.
In these instances, the two dividing points are selected so that one of them would be
the next dividing point of the subsequent iteration. The selection is compensated by
reducing the elimination efficiency and increasing the number of iterations. Fibo­
nacci has the lowest number of iterations but requires the series of Fibonacci num­
bers. The golden section method relies on the “golden ratio”, which has been studied
since Euclid (Chapra and Canale 2010). While Excel/VBA does not have an in-built
function for the golden ratio, it can be calculated as (1 + 50.5)/2. After each iteration,
the region is reduced by 0.382.
With a predetermined error (the width of the final region), the number of itera­
tions of the three methods can be determined. Hence the For Next loop can be used.
Among the three elimination methods, however, the golden section method is the
only one that does not require a predetermined error to calculate the two interior
points (Chapra and Canale 2010). In this section, VBA code for the golden section is
shown. Furthermore, the combination of the golden section method will be demon­
strated in Chapter 10.
One-Dimensional Optimisation 45

EXAMPLE 4�1� MINIMISATION BY GOLDEN SECTION SEARCH�

PROBLEM STATEMENT�
The operating cost of a pump and a motor is given:

25
f ( x ) = 400 +1.2x + Eq. 4.1
x
where x is the motor power, expressed in horsepower (h.p.).
The manufacturer can provide a size from 1 to 8 h.p. with an increment of
0.1 h.p. Find the optimum pump size to minimise the cost of operating a pump
and motor.

Solution
In this problem, we need to find a minimum of Eq. 4.1 between 1 and 8. Since the
size must be a multiple of 0.1, the error in x should be 0.1. To get the nearest number
with one digit after the decimal point, the in-built function Round is used.

Code 4�1� Golden section search�

Function GoldenSection _ R(Dlower, Dupper)
Dim D1, D2, f1, f2, err, i, R
err = Abs(Dupper–Dlower)
R = (5 ^ 0.5–1) / 2
i = 0
D2 = Dlower + R * (Dupper–Dlower): D1 = Dupper–R *
(Dupper–Dlower)
f1 = ff(D1): f2 = ff(D2)
While err > 0.1 And i < 100
i = i + 1
If f1 > f2 Then
Dlower = D1: f1 = f2: D1 = D2
D2 = Dlower + R * (Dupper–Dlower): f2 = ff(D2)
Else
Dupper = D2: f2 = f1: D2 = D1
D1 = Dupper–R * (Dupper–Dlower): f1 = ff(D1)
End If
err = Abs(Dupper–Dlower)
Wend
46 Calculations for Process Engineering Using Excel VBA

GoldenSection _ R = WorksheetFunction.Round((D1 + D2)

/ 2, 1)
End Function
Function ff(x)
ff = 400 + 1.2 * x + 25 / x
End Function

The code returns an answer of 4.6 after 9 iterations. Without using Round, the
code will return an answer of 4.5569.

4.3 QUADRATIC INTERPOLATION

The previous methods only compare the values of internal points without consid­
ering the magnitude of these values. The elimination can be very inefficient. The
quadratic interpolation can take the relative slope of the values into the optimisation.
The method starts with three points and assumes that the curve follows a quadratic
function. Subsequently, the method calculates the bottom of the parabola for a min­
imum. The method starts with three initial values. The next value is determined for
each iteration by fitting the quadratic equation. Let us name the three initial points
as x1, x2 , x3 ( x1 < x2 < x3 ) , and the three corresponding evaluations are f1 , f2 , f3 (for
minimisation, f2 is smaller than both f1 and f3). The peak of the parabola is:

x4 =
( 2 2
) ( 2 2
)
1 f1 x2 ­ x3 + f2 x3 ­ x1 + f3 x1 ­ x2 ( 2 2
) Eq. 4.2
2 f1 ( x2 ­ x3 ) + f2 ( x3 ­ x1 ) + f3 ( x1 ­ x2 )

The new point is combined with two of the previous points. Then, either x1 or x3 will
be replaced. The selection is based on the position of x4. Once the three new values
are updated, the iteration can be repeated.

EXAMPLE 4�2� MINIMISATION BY QUADRATIC INTERPOLATION�

Problem statement� Repeat Example 4.1 by quadratic interpolation.

Solution

Code 4�2� Quadratic interpolation�

Function quadratic(x1, x2, x3, tol) ‘note x2 is
between x1-x3
Dim f1, f2, f3, err, i, x4
err = Abs(x3–x1)
One-Dimensional Optimisation 47

While err > (tol / 2) And i < 100

i = i + 1: f1 = ff(x1): f2 = ff(x2): f3 = ff(x3)
x4 = 0.5 * (f1 * (x2 ^ 2–x3 ^ 2) + f2 * (x3 ^ 2–x1 ^ 2)
+ f3 * (x1 ^ 2–x2 ^ 2))/(f1 * (x2–x3) + f2 * (x3–x1)
+ f3 * (x1–x2))
Select Case x4
Case Is >= x2
x1 = x2: x2 = x4
Case Is < x2
x3 = x2: x2 = x4
Case Else
quadratic = “No solution”
End Select
err = Abs(x1–x3)
Wend
quadratic = Round((x1 + x3) / 2, 1)
End Function

With x1 = 1, x2 = 2 and x3 = 8, the code returns the same answer after six iterations.

4.4 OPTIMISATION WITH SCANNING

Similar to the bisection method, region elimination needs two good guesses. Hence,
one can combine the scanning method (such as Code 2.5) with one of the previous
methods. It is also essential to have an appropriate initial range. For example, to find
a minimisation by quadratic method, the three initial starting points need to be in the
proper order: f(x2) is smaller than f(x1) and f(x3). The three points can be routinely
achieved by a scanning method. A simple scanning macro can be developed and
combined with the previous quadratic interpolation (Code 4.3).

Code 4�3� Quadratic interpolation with scanning (combined with

Code 4�2)�
Function scanning(x1, delx, tol)
Dim i, x2, x3, f1, f2, f3
x2 = x1 + delx: f1 = ff(x1): f2 = ff(x2): i = 0
While f1 > f2 And i < 1000
x1 = x2: f1 = f2: delx=delx*2: x2 = x1 + delx:
f2 = ff(x2)
i = i + 1
Wend
If i = 1000 Then
scanning = “no minimum”
48 Calculations for Process Engineering Using Excel VBA

Else
scanning = quadratic(x1 – delx/2, x1, x2, tol)
End If
End Function

The code starts with a small initial value and requires no assumption on the upper
limit. The code also doubles the scanning step after each iteration. The code stops
the scanning procedure when a region containing a minimum, a combination of three
points satisfying the initial conditions of the quadratic interpolation, is found. With
x1 = 0.1 and delx = 0.2 , the code stops scanning after five iterations and starts inter­
polating with three values (3.1, 6.3, 12.7).
In the previous example, the VBA function scanning calls the quadratic interpo­
lation at the end. A similar approach can be applied to the golden section search and
other region elimination methods. While these methods only require two initial points,
it is essential to know that at least there is an optimum between those two points (that
is, an interior point whose function is smaller than the function values at two ends).

4.5 ADVANCED EXAMPLE: A COMBINATION OF BINARY

SEARCH AND BISECTION METHOD
Optimisation can be done with a complicated function, such as an implicit function.
In such instances, the solving equation may require another numerical program, such
as a bisection or secant method.

EXAMPLE 4�3� MINIMISATION OF FRICTION LOSS FACTOR�

PROBLEM STATEMENT�
Part A� For fluid with a kinetic viscosity of 0.00123 m2/s and flow of 2 m3/s,
calculate the friction loss, f, as a function of pipe diameter between 0.04 and
0.6 m. The roughness is assumed at 0.001 m.
Plot the results.
Part B� Find the minimum value of f and the corresponding diameter.

Solution
Part A� Here, the Colebrook function of Example 2.1 is employed again. Since the equation
is in implicit form, a bisection method is needed. However, we want to see the correlation
between f and D. As D increases, velocity and Reynolds number decrease. Consequently,
the value of f changes. A numerical method is needed to find the optimum value of f.
First, fundamental equations of fluid mechanics are needed. For a fluid flow in a
circular pipe (diameter D), the Reynolds number is defined as:

VD
Re = Eq. 4.3
n
One-Dimensional Optimisation 49

where V is the velocity (m/s), and n is the kinetic viscosity (m2/s).

The velocity is a function of the volumetric flowrate (Q) and cross-section area:

4Q
V= Eq. 4.4
p D2
The previous equations can be combined with the Colebrook equation to determine
the friction factor. Bisection code similar to the previous one is used for Part A. How­
ever, the inputs include D and Q (instead of D and Re, as in Code 2.5).

Code 4�4� Bisection method to find friction loss as a function of the pipe
diameter�
Function Bisection2(Lower As Double, Upper As Double,
e, D, Q)
Dim M, err, tol, c1, c2, c3, V, Re ‘V and Re are
variables
V = Q * 4 / (D ^ 2 * 3.1416)
Re = D * V / 0.00123
tol = 10 ^ -15
c1 = colebrook(Lower, e, D, Re)
M = (Lower + Upper) / 2: c3 = colebrook(M, e, D, Re)
err = abs(Upper – Lower)
While err > tol
If c1 * c3 = 0 Then
Bisection2 = M
Exit Function
Else
If c1 * c3 < 0 Then
Upper = M
Else
Lower = M: c1 = c3
End If
End If
M = (Lower + Upper) / 2: c3 = colebrook(M, e, D, Re)
err = Upper–Lower
Wend
Bisection2 = M
End Function

The previous code can be applied to values of D from 0.04 to 0.6 and is plotted
in Figure 4.2. The curve indicates that the correlation has a minimum of around
D = 0.2 m .
50 Calculations for Process Engineering Using Excel VBA

FIGURE 4.2 Friction loss as a function of velocity, as shown in Example 4.2.

Part B� To get the minimum in Figure 4.2, a binary search can be used.

Code 4�5� Combined binary search and bisection methods (the Bisection2
function is in Code 4�4)�
Option Base 0
Function BinaryFunction(Dlower, Dupper, Q, e)
Dim V, Re, tol, D1, D2, B1, B2, i, err
tol = 10 ^ -7: err = Abs(Dupper–Dlower)
i = 0
While err > tol And i < 50
D1 = (Dupper–Dlower) / 2–tol + Dlower: D2 = (Dupper–
Dlower) / 2 + tol + Dlower
B1 = Bisection2(0.0001, 1, e, D1, Q):
B2 = Bisection2(0.0001, 1, e, D2, Q)
i = i + 1
If B1 = B2 Then
Dlower = D1: Dupper = D2
Else
If B1 > B2 Then
Dlower = D1
Else
Dupper = D2
End If
End If
One-Dimensional Optimisation 51

err = Abs(Dupper–Dlower)
Wend
BinaryFunction = (D1 + D2) / 2
End Function

Using the previous code, D* = 0.207. To get the minimum of f, one can call the
function Bisection2 and get fmin = 0.007059. The code can run for a substantially
long time if the initial region is large. One of the disadvantages of the previous bisec­
tion code is that two initial limits are fixed. These two limits are unnecessarily wide
apart to ensure a solution exists in all values of D. The large initial region will require
many iterations during the bisection method. Applying the scanning method with
double step size might significantly reduce the computing time in this case.

4.6 FURTHER EXAMPLES FOR SELF-PRACTICE

EXAMPLE 4�4� COMPARISON BETWEEN BINARY SEARCH AND

GOLDEN SECTION SEARCH�
Calculate the number of iterations between different search methods.

Answer
Insert the index into the code so the program gives the number of iterations.
For example, for Code 4.1, change the last line to

Dim output(1,2)
output(1,1) = Round((D1 + D2) / 2, 1)
output(1,2) = i
GoldenSection =output

EXAMPLE 4�5� FIBONACCI SEARCH�

Write VBA code for binary search and Fibonacci code.

Hints
First, the Fibonacci numbers are generated up to the required limit: ( b ­ a ) / e, where
ε is the tolerance. In Example 4.1, the limit is 80. So the required Fibonacci numbers
are 1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89.
52 Calculations for Process Engineering Using Excel VBA

For each iteration, the selection of x1 and x2 is based on the two neighbouring Fib­
onacci numbers: x1 = b ­ Fi /Fi+1 ( b ­ a ) and x2 = a + Fi /Fi+1 ( b ­ a ) . For example,
the first iteration is based on F10 (55) and F11 (89).

EXAMPLE 4�6� CHANGE THE OUTPUT OF THE BINARY

SEARCH
Change Code 4.5 so that the binary code returns the value of the corresponding
D (instead of f ).

Hints
Modify the last line of the code “BinaryFunction = ( B1 + B2 ) / 2 ”. Alternatively, the
code can give fmin and D * by employing an output matrix, as in Example 4.4.

EXAMPLE 4�7� CHANGE THE REQUIRED INCREMENT

IN OUTPUT�
Repeat Example 4.1, with a different requirement: the manufacturer can pro­
vide a size from 1 to 8 h.p. with an increment of 0.25 h.p.

Solution
Excel has a rounding function, MROUND, which allows rounding a number to a
specified multiple. For example, using the following code with tol = 0.25 will result
in D = 4.5 .

Function GoldenSection(Dlower, Dupper, tol)

Dim D1, D2, f1, f2, err, i, R
err = Abs(Dupper–Dlower)
R = (5 ^ 0.5–1) / 2
i = 0

D2 = Dlower + R * (Dupper–Dlower): D1 = Dupper–R *

(Dupper–Dlower)
f1 = ff(D1): f2 = ff(D2)

While err > tol And i < 100

i = i + 1
If f1 > f2 Then
One-Dimensional Optimisation 53

Dlower = D1: f1 = f2: D1 = D2

D2 = Dlower + R * (Dupper–Dlower): f2 = ff(D2)
Else
Dupper = D2: f2 = f1: D2 = D1
D1 = Dupper–R * (Dupper–Dlower): f1 = ff(D1)
End If
err = Abs(Dupper–Dlower)
Wend
GoldenSection = WorksheetFunction.Mround((D1 + D2) /
2, tol)
End Function
 5 Linear Programming

5.1 INTRODUCTION
While the previous methods in Chapter 4 deal with one-dimensional optimisation,
on the other hand, many practical problems require multiple-dimensional optimisa­
tion. In general cases, these multiple-dimensional problems are challenging to solve.
However, there is a class of optimisation in which all equations are linear. This type
of optimisation can be solved by “linear programming” algorithms. The most popu­
lar one is the simplex algorithm.
The simplex algorithm is widely used for solving linear programming problems
and is still considered one of the most efficient methods for solving linear program­
ming problems. A linear programming problem is an optimisation problem where
we seek to maximise or minimise a linear objective function subject to a set of linear
constraints. The simplex algorithm works by iteratively moving from one basic fea­
sible solution to another in a way that maximises or minimises the objective function.
One of the strengths of the simplex algorithm is its ability to handle large-scale linear
programming problems.

5.2 SIMPLEX ALGORITHM

The simplex algorithm is better demonstrated in the table form. All equations,
including the objective function, are converted into a table form. The problem can be
summarised by using the following example:

EXAMPLE 5�1� MAXIMISE A LINEAR FUNCTION�

PROBLEM STATEMENT�
Maximise the following function:

z = 20x1 +16x1 +12x3 +10 Eq. 5.1

Subject to:
x1 < 4
2x1 + x2 + x3 < 10
2x1 + 2x2 + x3 < 16
All variables are non-negative.

54 DOI: 10.1201/9781003364511-5
Linear Programming 55

Solution
The first step is to convert the inequalities to equalities and then present the table
form. The inequality constraints are converted to equations by adding slack variables.

x1 + x4 = 4 Eq. 5.2

2x1 + x2 + x3 + x5 = 10 Eq. 5.3

2x1 + 2x2 + x3 + x6 = 16 Eq. 5.4

The objective function is rewritten:

­ z + 20x1 +16x1 +12x3 = ­10 Eq. 5.5

Consequently, the objective and equations can be presented in the matrix form.
The simplex algorithm can be applied by improving the feasible solution. The prob­
lem is presented in the matrix form, Table 5.1, and entered into an Excel spreadsheet.
To simplify the explanation, the variables can be classified as either basic or
non-basic variables. In Table 5.1, the three variables without coefficients (x4, x5, x6) are
basic variables. The other variables (x1, x2, x3) are non-basic variables. The simplex
method employs pivoting steps in a repeating manner. The purpose of each pivoting
step is to exchange one basic variable and one non-basic variable (Nawrocki 2013).
The pivoting includes three elementary steps:

1 Determine the non-basic variable to be pivoted. Find the largest positive C

(in row 1) and the corresponding column index (AcC). If all values in row
1 are negative, stop the iteration.
2 Determine the basic variable to be exchanged. Within the column AcC,
from A(AcC,2) to A(AcC,m), find row AcR so that B(AcR)/A(AcC, AcR)
has the smallest positive value. If all A(i, AcC) are non-positive, stop (that
problem is unbounded).
3 Pivot on the basic variable, A(r,c): after this step, x(AcC) becomes the basic
variable, and one of the basic variables becomes a non-basic variable.

TABLE 5.1
The Input of the Simplex Problem
–z x1 x2 x3 x4 x5 x6 B
1 20 16 12 –10
1 0 0 1 0 0 4
2 1 1 1 10
2 2 1 1 16
56 Calculations for Process Engineering Using Excel VBA

The pivoting is repeated until all c are negative. It should be noted that cycling can
happen when steps 2 and 3 have several equal choices. To prevent cycling, one can
use Bland’s anticycling rule:

• Select largest a(1,c) with smallest c (first value)

• Select smallest a(1,c) with smallest r (first value).

The procedure is implemented in the following code. The input is a table (contain­
ing all elements in Table 5.1 except the headings), and the output is a table of the
same size. The code is divided into three parts for the readers to follow. The first one
is validating the dimension and properties of the inputs.

Code 5�1� Linear programming, section 1� Declaring dimensions and

checking column B for negative numbers�
Function LinearP(A)
Dim output(), checkB(), checkC(), checkR(), AcC, AcR,
MinC, MinB, i, j, temp, k, P1, P2
Dim M(), nRow, nCol
Dim RowDim As Integer
Dim ColDim As Integer
nRow = A.Rows.Count: nCol = A.Columns.Count
ReDim checkB(2 To nRow): ReDim checkR(1 To nRow) ‘to
find the r (leaving variable)
ReDim checkC(1 To nCol) ‘to find the column (enter­
ing variable)
ReDim M(1 To nRow, 1 To nCol) ‘M is output matrix,
same size as A
For i = 1 To nRow
For j = 1 To nCol
M(i, j) = A(i, j)
Next j
Next i
‘check B, from 2 to n for a negative number
For i = 2 To nRow
checkB(i) = A(i, nCol)
Next i
MinB = WorksheetFunction.Min(checkB())
If MinB < 0 Then
LinearP = “negative B”
Exit Function
End If

After checking the matrix, the code needs to check if there is a positive coefficient
in the first row (Code 5.2). If there is none, the current solution cannot be improved.
Linear Programming 57

Code 5�2� Linear programming, section 2: determining AcC (entering

non-basic variable) and AcR (row to be pivoted)�
For j = 2 To (nCol–1)
If M(1, j) > 0 Then
checkC(j) = M(1, j)
End If
Next j ‘if there are equal M(1,j), the last (largest
j) is selected
MinC = WorksheetFunction.Max(checkC()) ‘Debug.Print MinC
If MinC > 0 Then
k = 0
Do While (MinC > 0) And (k < 10)
temp = 0
For j = 2 To (nCol–1)
If MinC = M(1, j) Then
AcC = j
End If
Next j:
ReDim checkR(1 To nRow) ‘Erase checkr() set to zero,
ReDim sets to empty
For i = 2 To nRow
If (M(i, AcC) > 0) Then
checkR(i) = M(i, nCol)/M(i, AcC)
End If
Next i
temp = WorksheetFunction.Min(checkR()):
For i = 2 To nRow
If temp = checkR(i) Then
AcR = i
End If
Next i

In the next step, a basic and non-basic variable will be determined to be exchanged
with each other (Code 5.3).

Code 5�3� Linear programming, section 3� Pivoting�

‘cell (AcR, AcC) is the target- normalise ActiveRow
so that M(AcR,AcC) equals to 1
P1 = M(AcR, AcC): ‘Debug.Print AcR, ratio1, M(nrow,
ncol)
For j = 2 To nCol
58 Calculations for Process Engineering Using Excel VBA

M(AcR, j) = M(AcR, j)/P1

Next j: ‘Debug.Print AcR, AcC, M(AcR, AcC), M(AcR,
ncol)

For i = 1 To nRow ‘pivoting

If i <> AcR Then
P2 = M(i, AcC)
For j = 2 To nCol
M(i, j) = M(i, j)–P2 * M(AcR, j): ‘Debug.
Print i, j, M(i, j), M(i, AcC), M(AcR, j)
Next j
End If
Next i
‘------
ReDim checkC(1 To nCol)
For j = 2 To (nCol–1)
If M(1, j) > 0 Then
checkC(j) = M(1, j)
End If
Next j
MinC = WorksheetFunction.Max(checkC())
k = k + 1
Loop
LinearP = M
Else
LinearP = “No improvement”
End If
End Function

After the first iteration, the output will become Table 5.2. After pivoting on A(1,2),
the basic variables are { x1 , x5 , x6 }.

TABLE 5.2
Table after First Iteration
–z x1 x2 x3 x4 x5 x6 B
1 0 16 12 –20 0 0 –90
0 1 0 0 1 0 0 4
0 0 1 1 –2 1 0 2
0 0 2 1 –2 0 1 8

The procedure is repeated until all coefficients in the first row are non-positive.
Linear Programming 59

TABLE 5.3
Table after the Fourth Iteration: The Procedure Stops since all c Values Are
Non-Positive
–z x1 x2 x3 x4 x5 x6 B
1 –4 0 0 0 –8 –4 –154
0 2 0 1 0 2 –1 4
0 0 1 0 0 –1 1 6
0 1 0 0 1 0 0 4

The final output is in Table 5.3. In this table, the three non-basic variables are x1,
x5 and x6. Next, the user can work out the optimised solution from the final form. In
Table 5.3, the three variables with zero coefficients (x2, x3, x4) are basic variables. The
solution corresponds to the combination in which the values of non-basic variables
are zero, and the values of basic values are determined by the other rows of the table.
The values of basic variables are equal to the corresponding values in column B
(since all coefficients of basic variables are 1). For Table 5.3, the B column results in
x2 = 6, x3 = 4, and x4 = 4. Since we only need the actual variables in the original prob­
lem, the solution is x1 = 0, x2 = 6, x3 = 4. The optimised value of z equals –B(1) = 154.

5.3 ADVANCED EXAMPLE: PINCH ANALYSIS OF HEAT

EXCHANGER NETWORK
Pinch analysis is a method used in chemical engineering to optimise the design and
operation of process systems. It is based on the principle of minimising energy con­
sumption and maximising energy recovery by utilising a heat exchange network. The
benefits of pinch analysis include reduced energy consumption, increased energy
recovery, improved process efficiency and reduced environmental impact. It is widely
used in the chemical process industries, such as petroleum refining, petrochemicals
and food processing. Pinch analysis can also be used for the design and optimisation
of water networks, where the focus is on minimising water consumption and maxim­
ising water reuse within a process. A heat exchanger network connects the hot and
cold streams within a processing plant (Serth and Lestina 2014).

EXAMPLE 5�2� DETERMINE A HEAT EXCHANGE NETWORK TO

MINIMISE ENERGY CONSUMPTION�
PROBLEM STATEMENT�
Considering a process with two hot streams, two cold streams and two utili­
ties (adapted from example 8.4 (Serth and Lestina 2014). The heat data are
tabulated in Table 5.4.
Find the optimal heat exchanger network to minimise the heat requirement.
60 Calculations for Process Engineering Using Excel VBA

TABLE 5.4
Heat Data for Heat Exchange Networks
Stream Type Ts Tt (°C) Flow Duty (kW)
H1 Hot 150 60 2.0 180
H2 Hot 90 60 8.0 240
C1 Cool 20 125 2.5 262.5
C2 Cool 20 100 3.0 225
UH1 Hot utility 200 — —
UC1 Cool utility 20 — —

The problem can be converted into a linear optimisation (Pillai and Bandyo­
padhyay 2007) with eight flow variables, four equality constraints and two ine­
quality constraints. The two utilities, UH and UC, are numbered at heating node 1
and cooling node 1. Other streams are numbered subsequently. Let aij be the flow
from hot stream i to cool stream j. Each non-zero aij represents a heat exchanger.
The required temperatures can be achieved by using the hot utility to heat the
two cool streams and the cold utility to cool the two heat streams. This scenario
is represented by Figure 5.1(a). In such a setup, the total energy requirement is
unnecessarily high. To save energy, the plant operator may want to cross some
hot and cool streams. The net energy saving can be achieved by minimising the
utility energy.
The objective is to minimise the duty of the cold utility:

Z = a21 + a31 Eq. 5.6

Since the standard form of linear programming is maximisation, the objective

equation is converted to

­ Z = ­a21 ­ a31 Eq. 5.7

FIGURE 5.1 Heat exchange network. (a) All heating/cooling is done by the two utility
streams. (b) Minimisation of heating duty.
Linear Programming 61

Equation forms for meeting energy requirements:

a12 + a22 + a32 = 262.5 Eq. 5.8

a13 + a23 + a33 = 225 Eq. 5.9

a21 + a22 + a23 = 180 Eq. 5.10

a31 + a32 + a33 = 240 Eq. 5.11

Furthermore, one needs to consider the physical constraints. For instance, the prob­
lem can include new constraints for temperatures between hot and cool streams.
The hot steam H2 has a very high heat duty, but it cannot heat cool streams to above
90°C (whereas the required temperatures for C1 and C2 are above 90°C). Hence, the
heat duty of the exchanger for H2 and C1 is limited to (90 – 20) × 2.5. Similarly, the
heat duty between H2 and C2 is limited to (90 – 20) × 3. Hence we have two other
inequalities:

a22 < ( 90 ­ 20 ) ´2.5 =140 Eq. 5.12

a23 < ( 90 ­ 20 ) ´ 3 = 210 Eq. 5.13

We have two slack variables, S22 and S23, for these equations.
The nice feature of Tables 5.5 is that all coefficients are 1. Yet this isn’t easy to
work out the basic variables. The problem starts with four basic variables, which will
be replaced. However, after the pivoting, only two non-zero (negative) coefficients
are in row 1. It is not apparent which variables are non-basic. Hence, the code can be
modified to include the entered basic variables, which is AcC. The code inserts three
additional rows in the output matrix (Table 5.6).

TABLE 5.5
Heat Data for Heat Exchange Network: Initial Table
Z a12 a13 a21 a22 a23 a31 a32 a33 S12 S22 B
1 1 1 1 1 420
1 1 1 262.5
1 1 1 225
1 1 1 180
1 1 1 240
1 1 175
1 1 210
62 Calculations for Process Engineering Using Excel VBA

TABLE 5.6
Heat Data for Heat Exchange Networks: Final Answer
Z a12 a13 a21 a22 a23 a31 a32 a33 S12 S22 B
1 0 0 –1 0 0 –1 0 0 0 0 0
0 1 1 –1 0 0 –1 0 –0 0 0 67.5
0 0 1 –1 –1 0 0 0 1 0 0 45
0 –0 –0 1 1 1 0 0 0 –0 –0 180
0 –0 –1 1 1 0 1 1 0 –0 –0 195
0 –0 –0 0 1 0 0 0 0 1 –0 175
0 0 0 –1 –1 0 0 0 0 0 1 30

Function LinearP2(A)
Dim checkV(), sols()
ReDim M(1 To nRow + 3, 1 To nCol)
‘-----same as function LinearP
‘-----­
M(nRow + 1, 1) = “No. of iter:”: M(nRow + 1, 2) = k
ReDim checkV(2 To nCol–1): ReDim sols(2 To nCol–1)
M(nRow + 2, 1) = “Variable”: M(nRow + 3, 1) = “Value “
For i = 2 To nCol–1
If M(1, i) = 0 Then
For j = 2 To nRow
If M(j, i) = 1 Then
checkV(i) = checkV(i) + 1
sols(i) = M(j, nCol) ‘if
more than 1 positive coef, sols(i) is the last one
End If
Next j
End If
If checkV(i) = 1 Then
M(nRow + 2, i) = “BV”: M(nRow +
3, i) = sols(i)
Else
M(nRow + 2, i) = “NBV”: M(nRow +
3, i) = ““
End If
Next i
LinearP2 = M
Else
LinearP2 = “No improvement”
End If
End Function
Linear Programming 63

TABLE 5.7
Pinch Analysis of Heat Exchange Networks, with Answers
Z a12 a13 a21 a22 a23 a31 a32 a33 S12 S22 B
1 0 0 –1 0 0 –1 0 0 0 0 0
0 1 1 –1 0 0 –1 0 –0 0 0 67.5
0 0 1 –1 –1 0 0 0 1 0 0 45
0 –0 –0 1 1 1 0 0 0 –0 –0 180
0 –0 –1 1 1 0 1 1 0 –0 –0 195
0 –0 –0 0 1 0 0 0 0 1 –0 175
0 0 0 –1 –1 0 0 0 0 0 1 30
No. of iter: 3 0 0 0 0 0 0 0 0 0 0
Variable BV NBV NBV NBV BV NBV BV BV BV BV 0
Value 67.5 180 195 45 175 30 0

Code 5�4� To include the variables/answers in the output�

The answers, including the values for basic variables, are given in Table 5.7.
This answer corresponds to the diagram in Figure 5.1(b). From these data, one can
calculate other properties of the proposed heat exchange network.

5.4 FURTHER EXAMPLES FOR SELF-PRACTICE

EXAMPLE 5�3� MAXIMISE A LINEAR FUNCTION WITH A

REAL NUMBER�
Using the previous code to solve the following problem:
Maximise the following function:

z = 20x1 +16x1 +12x3 +10

Subject to:

x1 < 4
2x1 + 2x3 < 10
2x1 + 4x2 + x3 < 16
x3 > ­5

All variables, except x3, are non-negative.

64 Calculations for Process Engineering Using Excel VBA

Solution
Since x3 can be both positive and negative, it can be replaced by x7 ­ x8 (which are
both positive). Thus, a new slack variable is introduced:
­ x7 + x8 + x 9 = 5
The previous equation is combined with other equations to form the table.

­z x1 x2 x7 x8 x4 x5 x6 x9 B
1 20 16 12 –12 –10
1 0 0 1 0 0 4
2 2 –2 1 10
2 4 1 –1 1 16
–1 1 1 5

Applying the code will get x1 = 4, x2 = 1.75 and x3 = 1 (from the values of x7
and x8 ).

EXAMPLE 5�3� MAXIMISE A PROFIT�

A truck is to be used to transport two products to their customers. The volume
of the truck is 10,000 L, and the maximum weight that can be carried is 20 t.

• Product A occupies 50 L and weighs 200 kg.

• Product B occupies 80 L and weights 100 kg.
• The profit margins are $5 per unit of Product A and $7 per unit of
Product B.

As the truck has limitations on both weight and volume, determine how
much of each product to load to maximise the profit.

Solution
Introducing two slack variables, one can generate the following table.

–z A B x1 x2 B
1 5 7 0
Volume 50 80 1 10000
Weight 200 100 0 1 20000

Applying the previous code, the answers are B = 90.91 and A = 54.55. Since the
numbers of the transported products have to be integers, there will be four combina­
tions around this combination. The best combination is (90, 55).
 6 Roots of Systems of Linear
Equations
6.1 INTRODUCTION
Systems of linear equations are very common in engineering. For example, one can
find them with a series of reactors. However, the most common problems are related
to process heat transfer. A system of linear algebraic equations can be solved simul­
taneously using a direct method such as Gaussian elimination. The Gaussian method
consists of two steps: forward elimination and back substitution. Forward elimination
reduces all elements in the bottom half to zero (such as in Eq. 6.1). Subsequently, the
solution is found from the last variable (xn to x1).

Eq. 6.1
Another direct method is the Gauss–Jordan method. In this method, the final
matrix contains nonzero values on the diagonal rather than a triangular matrix. The
rows are also normalised by dividing each one by its pivot element. The result is that
the identity matrix is formed, where all diagonal coefficients are the value 1. The
Gauss–Jordan method requires more work and more operations than Gauss elimi­
nation. Gaussian elimination requires order n3/3 multiplications followed by back
substitution requiring order n2 multiplications. Gauss–Jordan elimination requires
order n3/2 multiplications (Lindfield and Penny 2019).
The methods can be problematic due to round-off errors and zero on the diagonal. For
many problems in process engineering, a successive substitution is more advantageous.
In these methods, the initial guesses (a combination of n values of unknowns) are gener­
ated and then improved iteratively. These iterative substitutions require an initial guess of
the solution. Subsequently, this set of solutions is improved until the errors in all variables
are acceptable. The number of iterations of these methods depends on the initial guess
and the required error. An “educated” guess can significantly reduce the iterations and
multiplications. In addition, the number of iterations can be further reduced by employ­
ing a relaxation factor. The VBA codes in this book will only cover the iterative methods.

6.2 SUCCESSIVE SUBSTITUTION METHODS

Two successive substitution methods to improve the solutions are Jacobi and Gauss–
Seidel. In both methods, the variables are updated one at a time. In the Jacobi method,

DOI: 10.1201/9781003364511-6 65
66 Calculations for Process Engineering Using Excel VBA

all variables in the jth iteration are updated from the values of the ( j – 1)th iteration.
In the Gauss–Seidel, each variable is updated using the latest available values of the
other variables. In general, the Gauss–Seidel method provides faster convergence
than the Jacobi method.
The substitution formulas for a 4 × 4 system are presented by four separate
equations:

b1 ­ a12 x2j ­1 ­ a13 x3j ­1 ­ a14 x4j ­1

x1j = Eq. 6.2a
a11
b2 ­ a21 x1j ­ a23 x3j ­1 ­ a24 x4j ­1
x2j = Eq. 6.2b
a22
b3 ­ a31 x1j ­ a32 x2j ­ a34 x4j ­1
x3j = Eq. 6.2c
a33

b4 ­ a41 x1j ­ a42 x2j ­ a43 x3j

x4j = Eq. 6.2.d
a44
Note that the previous equation form is not easy to generalise. The code must
contain the four separate equations in each iteration. Furthermore, the form is more
complicated for systems with different numbers of equations. Consequently, the for­
mula is re-arranged to apply to larger systems of equations. For instance, the first
equation in system 6.2 is re-arranged as:

b1 a11 x1j ­1 a11 x1j ­1 + a12 x2j ­1 + a13 x3j ­1 + a14 x4j ­11
x1j = + ­ Eq. 6.3
a11 a11 a11

Which can be generalised for all substitutions as:

bi aii xij ­1 å aik xkj ­1

xij = + ­ Eq. 6.4
aii aii aii
While the previous equation looks more complicated than equations in 6.2, the
coding is simplified using a second loop (to calculate the last term of the right-hand
side). Furthermore, the procedure can be used for any dimension (rather than n = 4).
The usage of the generalised formula can be seen in the following example codes.

EXAMPLE 6�1� SIMULATING A SERIES OF CONTINUOUS

STIRRED-TANK REACTORS�
PROBLEM STATEMENT
The chemical process consists of four continuous stirred-tank reactors (CSTRs)
and two recycle streams, as in Figure 6.1. In the reactors, the chemical conver­
sion is given by
A ¾k¾
®B
Roots of Systems of Linear Equations 67

FIGURE 6.1 Series of four CSTRs with two recycle streams.

where k is the rate constant (1/hr), which depends on the operating condition
of the reactor.
The volumes and rate constants of the four reactors are tabulated in
Table 6.1. The feed conditions are C0 = 1 M, U0 = 1000 L/hr.
Part A� Prepare the system in a matrix form in an Excel spreadsheet.
Part B� Develop code to solve the system using the Gauss–Seidel method.
Part C� Repeat part B using the Jacobi method.

TABLE 6.1
Reactor Size and Conversion Constants
Reactor 1 2 3 4
Vol (L) 1000 1500 200 1000
k (h−1) 0.1 0.2 0.4 0.3

Solution
Part A. The problem requires a system of linear equations. At steady state, the mass
balances for each reactor are:

Uin = U out Eq. 6.5

UinCin = U out Cout + ( k V ) Cout Eq. 6.6

where: V is the reactor volume (L)

Uin U out are inflow and outflow, respectively (L/hr)
Cin , Cout are A concentrations in inflow and outflow, respectively (mol/L)
k is the rate constant (1/hr)
68 Calculations for Process Engineering Using Excel VBA

TABLE 6.2
Inputs for a System of Linear Equations
A B C D E F
1 Matrix a Cguess B
2 1100 1 1000
3 1000 –1450 150 1 0
4 1150 –1305.6 75.59866 1 0
5 1075.599 –1375.6 1 0

Consequently, a system of 4 × 4 equations can be constructed. The elements of the

matrices A and B are determined from V, k and corresponding flow rates. In the table,
the guess values of all concentrations are 1 M (that is, the value of C0). The table will
be used as input for the VBA functions.
Part B� To solve the previous system, VBA code is developed based on the Gauss–
Seidel method. It is noteworthy that the code calls A, B and Cguess as three separate
arguments instead of using a single input as a 4 × 6 table (Table 6.2). The advantage
of the arrangement is that the code be used for other dimensions.

Code 6�1� Gauss–Seidel method�

Option Base 1
Function GS(a, B, Cguess, tol)
Dim err(10), row, col, n, sum, Cin(10), cnew(10),
MaxErr, k
MaxErr = 1: k = 0: n = a.Rows.Count ‘n = number of
rows in A
For row = 1 To n
Cin(row) = Cguess(row)
Next row
While MaxErr > tol And k < 100
k = k + 1: MaxErr = 0 ‘increase k after each
iteration
For row = 1 To n
sum = 0 ‘reset sum
For col = 1 To n
sum = sum + a(row, col) * Cin(col) ‘
Next col

cnew(row) = (B(row)–sum + Cin(row) * a(row,

row))/a(row, row) ‘calculate Cnew
err(row) = Abs(cnew(row)–Cin(row)) ‘there
will be n errors, the code stops when ALL errors <
tol, or sum of all errors < tol . . .
Roots of Systems of Linear Equations 69

Cin(row) = cnew(row) ‘updating values,

cin(1,2,3,4) will be the input of the next iteration
If err(row) > MaxErr Then
MaxErr = err(row)
End If ‘calculate MaxErr for each iteration
Next row
Wend
Cin(n + 1) = k: Cin(n + 2) = MaxErr
GS = Cin
End Function

Using the VBA function in Code 6.1, a solution is found after 13 iterations
(tol = 10–10).
Part C� For the Jacobi method, the updating loop “Cin(row) = cnew(row)” is sep­
arated from the calculating loop. Mathematically, the two methods require the same
number of calculations. In this instance, the numbers of iterations are also the same.
The function for the Jacobi method is presented in Code 6.2.

Code 6�2� Jacobi method�

Function Jacobi(a, B, Cguess, tol)
Dim err(10), row, col, n, sum, Cin(10), cnew(10),
MaxErr, k
MaxErr = 1: k = 0: n = a.Rows.Count
For row = 1 To n
Cin(row) = Cguess(row)
Next row
While MaxErr > tol And k < 100
k = k + 1: MaxErr = 0 ‘
For row = 1 To n
sum = 0
For col = 1 To n
sum = sum + a(row, col) * Cin(col)
Next col
cnew(row) = (B(row)–sum + Cin(row) *
a(row, row))/a(row, row)
err(row) = Abs(cnew(row)–Cin(row))
If err(row) > MaxErr Then
MaxErr = err(row)
End If
Next row
For row = 1 To n
70 Calculations for Process Engineering Using Excel VBA

Cin(row) = cnew(row) ‘updating all values

at the same time
Next row

Wend
Cin(n + 1) = k: Cin(n + 2) = MaxErr
Jacobi = Cin
End Function

Using a tol of 10–10 and input in Table 6.2, the function in Code 6.1 gives the
answers after 25 iterations.

6.3 SUCCESSIVE SUBSTITUTION METHODS

WITH A RELAXATION CONSTANT
To enhance the convergence, an iterative program can introduce a relaxation, in
which the value at a particular iteration is made up of a combination of the old value
and the newly calculated value:

xij = l xij + (1­ l ) xij ­1 Eq. 6.7

where λ is a weighting factor assigned a value between 0 and 2.

The relaxation can be classified into three scenarios:

• 0 < l < 1: under-relaxation

• l = 1: no relaxation; the code is the same as the Gauss–Seidel or Jacobi
technique
• 1 < l £ 2: overrelaxation

EXAMPLE 6�2� REDUCE THE CALCULATION TIME OF THE

GAUSS–SEIDEL METHOD�
PROBLEM STATEMENT
Repeat Part B of Example 6.1 with a relaxation factor.

Solution
The VBA function in Code 6.1 is modified to include a relaxation factor. Using the
function in Code 6.3, one can vary λ to see the impact on the number of iterations.
Roots of Systems of Linear Equations 71

Code 6�3� A Gauss–Seidel method with relaxation�

Function GSL(a, b, Cguess, tol, lambda)
Dim err(10), row, col, n, sum, Cin(10), cnew(10),
MaxErr, k
MaxErr = 1: k = 0: n = a.Rows.Count ‘n = number of
rows in matrix a
For row = 1 To n
Cin(row) = Cguess(row)
Next row
While MaxErr > tol And k < 100
k = k + 1: MaxErr = 0 ‘increase k after each
iteration
For row = 1 To n
sum = 0 ‘reset sum
For col = 1 To n
sum = sum + a(row, col) * Cin(col)
Next col
cnew(row) = (b(row)–sum + Cin(row) * a(row,
row))/a(row, row) ‘calculate Cnew
err(row) = Abs(cnew(row)–Cin(row))
Cin(row) = lambda * cnew(row) + (1–lambda)
* Cin(row)
If err(row) > MaxErr Then
MaxErr = err(row)
End If ‘calculate MaxErr for each iteration
Next row
Wend ‘next iteration
Cin(n + 1) = k: Cin(n + 2) = MaxErr
GSL = Cin
End Function

The number of iterations can be changed significantly using the VBA function in
Code 6.3 with different values of λ. For this particular system, an over-relaxation
with λ = 1.1 can reduce the number of iterations to 11.

6.4 CONCLUDING REMARKS

A system of linear equations with the same number of variables can be solved numeri­
cally by many different methods. VBA functions based on substitution methods are sim­
ple to develop and can be adapted to many problems. Users can improve the efficiency
of the methods. Two key aspects are the formation of coefficients and the selection of
the initial guesses. Formation equations can be optimised during the transformation
72 Calculations for Process Engineering Using Excel VBA

of the mathematical equations from the physical system. The initial guess can also be
optimised from the physical characteristics. In some processes, the initial guess for the
new operating conditions can be used from previous conditions. These examples will
be demonstrated in the next chapter.

6.5 FURTHER EXAMPLES FOR SELF-PRACTICE

EXAMPLE 6�2� OPTIMISE THE RELAXATION CONSTANT�

Identify the optimal λ between 0 and 3 to minimise the number of iterations of
the Jacobi method.

Hints
Re-arrange the function in Code 3.3 so that the output is only the number of iter­
ation Cn(n + 1). Consequently, combine the code with an optimisation method in
Chapter 4. The smallest number of iterations should be 8, with λ between 1.0381 and
1.0389. The relations between λ and k can also be plotted in Figure 6.2.

EXAMPLE 6�3� OPTIMISE THE INITIAL SOLUTION�

In previous solutions, the values for Cguess were input by the users. Write the
code to generate Cguess automatically from A and B.

FIGURE 6.2 Relationship between relaxation factor and the number of iterations.
Roots of Systems of Linear Equations 73

Hints
An obvious solution is Cg(i) = B(i)/A(i,i). If A is a diagonally dominant matrix, these
guesses are closer to the final answer than random guesses. The code will need A and
B matrices only. Hence, the initial values for the first iteration can be replaced.

For row = 1 To n
Cin(row) = b(row)/a(row, row)
Next row
 7 Applications of Systems
of Linear Equations

7.1 INTRODUCTION
The solution of linear equations can be combined with other methods to solve com­
plex problems. Furthermore, the system of linear equations can be used to impose
relations between the various partial derivatives of a multivariable function. Such
problems are often found in the process of heat transfer. This chapter will present
these problems and solutions.

7.2 INPUT VALIDATION

As can be seen in the previous chapter, the operation will require the dividing of
diagonal elements. Hence, it is essential to confirm that the diagonal elements are
dominant and non-zero. Consequently, it is important to check the coefficient matrix
before proceeding with substitution. To validate if a matrix is a square, one can use
the rows.count and column.count functions. The validation code, such as Code 7.1,
can be used in conjunction with other iterations (Code 6.1, 6.2 or 6.3).

Code 7�1� Validating coefficient matrix before applying the Gauss–Seidel

algorithm�
Function validate(a)
Dim i, n
validate = “Square matrix without zero on diagonal”
n = a.Rows.Count
If n <> a.Columns.Count Then
validate = “Not a square matrix”
Else
For i = 1 To n
If a(i, i) = 0 Then
validate = “Some diagonal elements are zero”
End If
Next i
End If
End Function

In practice, these diagonal elements should be dominant, abs(a(i,i)) > abs(a(i,j)).

Consequently, the sorting procedure may be needed. In the literature, partial

74 DOI: 10.1201/9781003364511-7
Applications of Systems of Linear Equations 75

pivoting is used to swap two rows of coefficients when necessary (to place the max­
imum coefficient of a given column in the proper row). However, as seen in the fol­
lowing examples, some systems already meet the requirement during the formulation
process from the physical equations.

7.3 ADVANCED EXAMPLE: A COMBINATION

OF BISECTION AND JACOBI METHODS
Once the system of equations has been solved, the code can be combined with other
methods to predict other calculations. The following is an example of modelling a
series of reactors.

EXAMPLE 7�1� MANIPULATION OF A SERIES OF CSTRS�

PROBLEM STATEMENT
Reconsider the problem in Example 6.1. Adjust recycle stream 2 so that
C4 = 0.5 M.

Solution
In this case, a bisection code can be used. All elements of matrices A and B in the previ­
ous example have to be calculated within the VBA code. Please note that only two ele­
ments in matrix A depend on U4R. Subsequently, the bisection method can be applied.
First, the data are entered into the specified cells of an Excel spreadsheet (Table 7.1)
The first section of the VBA code is used to generate the matrices A, B and Cguess
before running the Gauss–Seidel or Jacobi method. The elements of A and B are cal­
culated according to the mass balances of the reactors. The output of the VBA code

TABLE 7.1
Reactor Sizes, Constants, and Flow Conditions
A B C
1 Reactor volume and kinetic constant
2 Reactor Vol (L) k (h−1)
3 1 1000 0.1
4 2 1500 0.2
5 3 200 0.4
6 4 1000 0.3
7 Flow conditions
8 CA0 (mol/L) 1
9 U0 (L/hr) 1000
10 U3R (L/hr) 150
76 Calculations for Process Engineering Using Excel VBA

is the concentration of stream 4 (C4). Consequently, the bisection code is used to

determine U4R so that C4 = 0.5 M. This function is shown in Code 7.2.

Code 7�2� Modifying Gauss–Seidel code for optimisation�

Option Base 1
Function GSv(U4r, tol)
Dim err(100), row, col, n, sum, Cin(100), cnew(100), MaxErr, k
Dim a(4, 4), Cguess(4), B(4), V(4), kin(4), Ca0, U0,
U3rV(1) = Range(“B3”).Value: kin(1) = Range(“C3”).Value
V(2) = Range(“B4”).Value: kin(2) = Range(“C4”).Value
V(3) = Range(“B5”).Value: kin(3) = Range(“C5”).Value
V(4) = Range(“B6”).Value: kin(4) = Range(“C6”).Value

Ca0 = Range(“B8”).Value
U0 = Range(“B9”).Value
U3r = Range(“B10”).Value

a(1, 1) = U0 + V(1) * kin(1)

a(2, 1) = U0: a(2, 2) = -(U0 + U3r)–V(2) * kin(2): a(2,
3) = U3r
a(3, 2) = U0 + U3r: a(3, 3) = -(U0 + U3r + U4r)–V(3) *
kin(3): a(3, 4) = U4r
a(4, 3) = U0 + U4r: a(4, 4) = -(U0 + U4r)–V(4) * kin(4)
B(1) = U0 * Ca0
For k = 1 To 4
Cguess(k) = Ca0
Next k
n = 4
‘****** repeat Jacobi or Gauss–Seidel code (Code 6.1.)
‘******
GSv = Cin(4)
End Function

Function BisectionX(Lo, Up, tol)

Dim Mi, c1, c3, i, n
n = Abs(Up–Lo) / 0.01
‘*****************************************
c1 = GSv(Lo, tol)–0.5
For i = 1 To n
Mi = (Lo + Up) / 2
c3 = GSv(Mi, tol)–0.5
If c1 * c3 < 0 Then
Up = Mi
Else
Applications of Systems of Linear Equations 77

Lo = Mi: c1 = c3
End If
Next i
‘*****************************************
BisectionX = Mi
End Function

Using the previous code with a tolerance of 10–4, the required U4R is determined
at 75.2794 L/hr and C4 = 0.50000 M. With a tolerance of 10–6, the optimised U4R is
determined at 75.5018 L/hr and C4 = 0.50000 M. The sensitivity of the correlation
between C4 and U4R is a key advantage of employing recycle streams in reaction
engineering.

7.4 ADVANCED EXAMPLE: 2-DIMENSIONAL STEADY-STATE

HEAT TRANSFER
In 2-dimensional steady-state heat transfer, the temperature distribution in a solid/
fluid object is calculated when the heat transfer rate is constant over time. This type
of heat transfer is important in various engineering applications, such as designing
heating and cooling systems for buildings and calculating the temperature distribu­
tion in electronic devices.
The heat transfer equation for 2-dimensional steady-state heat transfer is given by:

Ñ 2T = ­Q / k Eq. 7.1

where T is the temperature at a point in the body, Q is the heat transfer rate per unit
volume and k is the material’s thermal conductivity.
In Eq. 7.1, Ñ 2T = ­Q / k represents the Laplacian of T, which is the sum of the
second partial derivatives of T with respect to the x and y coordinates. The equation
can be solved using numerical methods, such as finite difference, finite element or
boundary element methods. These methods discretise the object or fluid into a grid of
points, and the temperature at each point is calculated based on the temperature and
heat flux at neighbouring points. Since the temperatures of all points are interlinked,
a system of equations must be solved simultaneously.

EXAMPLE 7�2� STEADY-STATE TEMPERATURE PROFILE

(WHITE 2002)�
PROBLEM STATEMENT
A metal plate is exposed to the surrounding environment, as shown in
Figure 7.1.
Calculate the temperature of the internal nodes.
78 Calculations for Process Engineering Using Excel VBA

FIGURE 7.1 The discretisation of the 2-dimensional heating plate.

Solution
For 2-dimensional steady-state heat problems, there is no heat accumulation at any
point. The temperature of the boundary nodes is the same as the surrounding envi­
ronment. Therefore, the values of temperature at the four corners are ignored. For any
interior nodes, the temperature is given the balance of the four surrounding nodes.
For instance, the temperature of node (1,1) is given by:

4T1,1 = T0,1 + T1,0 + T2,1 + T1,2 Eq. 7.2

With boundary nodes having constant temperatures, a system of nine equations

is established:
é ­4 1 0 1 0 0 0 0 0 ù é T1,1 ù é ­75 ù
ê 1 ­4 1 ê ú
ê 0 1 0 0 0 0 úú ê T2,1 ú êê 0 úú
ê0 1 ­4 0 0 1 0 0 0 ú ê T3,1 ú ê ­50 ú
ê úê ú ê ú
ê1 0 0 ­4 1 0 1 0 0 ú ê T1,2 ú ê ­75 ú
ê0 1 0 1 ­4 1 0 1 0 ú êT2,2 ú = ê 0 ú Eq. 7.3
ê úê ú ê ú
ê0 0 1 0 1 ­4 0 0 1 ú êT3,2 ú ê ­50 ú
ê0 0 0 1 0 0 ­4 1 0 ú êê T1,3 úú ê ­175 ú
ê ú ê ú
ê0 0 0 0 1 0 1 ­4 1 ú êT2,3 ú ê ­100 ú
ê úê ú ê ú
ë0 0 0 0 0 1 0 1 ­4 û ëT3,3 û ë ­150 û

One can use the previous VBA functions, such as Code 6.1 (Gauss–Seidel method),
to solve the previous system. A small modification can be done, as shown in Code
Applications of Systems of Linear Equations 79

7.3, to have the number of outputs exactly the same as the number of equations and
displaced in the vertical direction.

Code 7�3� Modifying Gauss–Seidel code for 2-dimensional steady-state

heat transfer�
Function GSv(a,b,Cguess,tol)
Dim output()
‘-----same as Code 6–1
‘------
ReDim output(n, 1)
For row = 1 To n
output(row, 1) = Cin(row)
Next row
GSv = output

Using the VBA function in Code 7.3, the answer will be given in Table 7.2. The
results are also presented graphically in Figure 7.2.

TABLE 7.2
The Temperature of Internal Nodes
T11 42.76562
T21 33.16739
T31 33.8828
T12 63.07809
T22 56.15845
T32 52.40958
T13 78.52565
T23 76.07029
T33 69.61997

FIGURE 7.2 Steady-state temperature distribution.

80 Calculations for Process Engineering Using Excel VBA

7.5 ADVANCED EXAMPLE: UNSTEADY-STATE HEAT TRANSFER

AFTER MICROWAVE HEATING
The unsteady-state heat conduction is a partial differential equation. The equation
is also applied to molecule diffusion. The numerical methods can be classified as
implicit, explicit and Crank–Nicholson methods (Chapra and Canale 2010). In short,
the implicit method predicts the new values in the time domain from the previous
values. The Crank–Nicholson method can be considered an average of the implicit
and explicit methods. The Crank–Nicholson method is unconditionally stable. In
terms of numerical calculation, both explicit and Crank–Nicholson methods require
solving a system of linear equations.

EXAMPLE 7�3� COOLING OF AN AQUEOUS SOLUTION AFTER

MICROWAVE HEATING�
PROBLEM STATEMENT
Consider the layer of water of 5 cm, which was microwaved to 80°C. The
layer was cooled by exposure to an oil layer at a constant temperature of 23°C.
Assume that heat loss to the container wall is negligible. Water thermal dif­
fusivity is given (10 –6 m2/s):

k = 0.1319 + 6.1231 ´10 –4 ´T ­ 2.5966 ´10 –6 ´T 2

Part A� Calculate the temperature profile in the vertical direction, from the oil
surface to 1.2 cm depth, after 1, 2, 5 and 10 mins.
Part B� Calculate the temperature at the points of 1.2 mm from the surface and
compare it with experimental data (Shibata et al. 2018).

Solution
The heat diffusion equation is given by:

¶T ¶2 T
=k 2 Eq. 7.3
¶t ¶x
where k is the thermal diffusivity of water, which varies with temperature.
The vertical domain of the water body is divided into n nodes. Node 1 is at the
1.2 cm depth, and node n is the water surface. At t = 0, all temperatures, except node
n, equal 80°C.
The dimensionless parameter, λ, is given:

Dt
l=k Eq. 7.4
Dx 2
Applications of Systems of Linear Equations 81

For internal nodes (from 2 to n – 1), the heat diffusion is discretised as:

­lTi l­1+1 + 2 (1+ l ) Ti l +1 ­ lTi l+1+1 = lTi l­1 + 2 (1+ l ) Ti l + lTi l+1 Eq. 7.5

The bottom node is modelled as an insulated boundary condition, dT/dx ( x = xb ) = 0 .

Imagine there is a frictional node outside the domain T(t,–1); the insulated condition
indicates that T­1l +1 = T1l +1 . Consequently, the equation for the first node is:

­lT1l +1 + 2 (1 + l ) T0l +1 ­ lT1l +1 = lT1l + 2 (1 + l ) T0l + lT1l Eq. 7.6

Finally, the equation for node n:

Tnl +1 = 23 Eq. 7.7

The previous system is arranged into a system of n equations. The n × n matrix has
non-zero values in the next diagonal and is often known as the tridiagonal matrix.
The procedure is repeated m times to get the answers in m seconds (Code 7.2). This
code consists of a separate function, SS(T, delx, delt), to calculate the value of λ at
each node temperature.

Code 7�4� One-dimensional heat transfer after microwave heating�

Option Base 1
Function GS _ 3t(n, m, Tpre, T0, delt, delx, tol) ‘C-N,
n is number of nodes
Dim a(), B(), err(), row, col, sum, T(), Tin(),
Tnew(),3t _ t(n, m, Tpre, T0, delt, delx, tol) ‘C-N, n
is number of nodes MaxErr, k, i, s
ReDim a(1 To n, 1 To n), B(1 To n), err(1 To n), Tin(1
To n + 2), Tnew(1 To n), T(1 To n + 2, 1 To m)

‘--- To is the surface temperature- constant value

For row = 1 To n–1
T(row, 1) = Tpre: Tin(row) = Tpre
Next row
Tin(n) = T0: T(n, 1) = T0

For i = 2 To m
MaxErr = 1: k = 1
‘-----------­
While MaxErr > tol And k < 1000
k = k + 1: MaxErr = 0
‘--------------­
a(1, 1) = 2 + 2 * SS(Tin(1), delx, delt): a(1, 2) = -2
* SS(Tin(1), delx, delt)
82 Calculations for Process Engineering Using Excel VBA

B(1) = 2 * (1–SS(T(1, i–1), delx, delt)) * T(1, i–1) + 2

* SS(T(1, i–1), delx, delt) * T(2, i–1)
For row = 2 To n–1
s = SS(Tin(row), delx, delt)
a(row, row) = 2 + 2 * s: a(row, row–1) = -s:
a(row, row + 1) = -s
B(row) = 2 * (1–s) * T(row, i–1) + s * T(row–1,
i–1) + s * T(row + 1, i–1)
Next row
a(n, n) = 1: B(n) = T0
‘------- . . . . . . Gauss–Seidel (code 6.1) or Jacobi
(code 6.2)
For row = 1 To n
sum = 0 ‘reset sum
For col = 1 To n
sum = sum + a(row, col) * Tin(col)
Next col
Tnew(row) = (B(row)–sum + Tin(row) *
a(row, row))/a(row, row)
err(row) = Abs(Tnew(row)–Tin(row))
Tin(row) = Tnew(row)
If err(row) > MaxErr Then
MaxErr = err(row)
End If
Next row
Wend
Tin(n + 1) = k: Tin(n + 2) = MaxErr
For row = 1 To n + 2
T(row, i) = Tin(row)
Next row
Next i
For i = 1 To n ’4 profiles at 60, 120, 300 and 600
seconds
Output(i, 1) = T(i, 60): Output(i, 2) = T(i, 120)
Output(i, 3) = T(i, 300): Output(i, 4) = T(i, 600)
Next i
GS _ 3t = Output
End Function
Function SS(T, delx, delt)
Dim k
k = 0.1319 + 6.1231 * 10 ^ (-4) * T–2.5966 * 10 ^ (-6)
* T ^ 2
SS = k * delt/delx ^ 2
End Function
Applications of Systems of Linear Equations 83

FIGURE 7.3 Temperature profiles near water surface after microwave.

It should be noted the boundary condition is mathematically the same as an insulated

boundary condition. The value of λ is calculated based on the temperature of Tl+1. Alter­
natively, λ can be calculated from the average of Tl and Tl+1. However, such an improve­
ment makes the code more complicated without significantly improving accuracy. The
code will give temperature values in an Excel spreadsheet and the plot in Figure 7.3.
For Part B, the output is modified to give the temperate profile of node 1 (Code
7.5). The results are plotted in Figure 7.4.

Code 7�5� One-dimensional heat transfer after microwave heating, pro­

ducing a temperature at 12 mm depth (the main part of the code is the
same as in Code 7�4)�
Function GS _
Dim a(), B(), err(), row, col, sum, T(), Tin(), Tnew(),
Output(), MaxErr, k, i, s
‘Dim Cout(1, 1000)
ReDim a(1 To n, 1 To n): ReDim Output(1 To m, 1 To 1)
‘-----­
‘------­
For i = 1 To m
Output(i, 1) = T(1, i)
Next i

GS _ 3t _ t = Output ‘ouput as array (cells are in hor­

izontal . . .)
End Function
84 Calculations for Process Engineering Using Excel VBA

FIGURE 7.4 The temperature at 12 mm from the surface; experimental data were obtained
from Shibata and coauthors (2018).
 8 Solving Systems of
Non-Linear Equations
8.1 INTRODUCTION
A system of non-linear equations is a set of equations where at least one is non-
linear, that is, contains a non-linear function such as a polynomial, trigonometric or
exponential function. Unlike linear equations, these equations cannot be solved using
simple algebraic techniques, and numerical methods must be used to obtain solu-
tions. The chapter presents several methods to solve systems of non-linear equations.
While some of the methods can be generalised, the success of a particular method
will depend on the nature of the equations and initial guesses. Furthermore, it should
be noted that many systems cannot be solved. Applying a numerical code to a general
system could lead to unfruitful outcomes.

8.2 NEWTON’S AND SECANT METHODS

While the number of equations and unknowns don’t have to be the same, math­
ematical systems require zero degrees of freedom to have a unique solution. The
necessary condition of the unique solution is the number of equations being equal
to the number of unknowns. Newton’s method in Chapter 2 can be extended for
such a system by employing an m × m Jacobian matrix, where m is the number of
equations/unknowns. The Jacobian matrix is the matrix in which all elements are its
first-order partial derivatives. The iterations will calculate the next set of variables
from the Jacobian of the previous set. The formula is detailed in the following for a
2 × 2 system.
For example, consider a system of two equations and two unknowns:

u ( x, y ) = 0 Eq. 8.1a

v ( x, y ) = 0 Eq. 8.1b

where x and y are independent variables.

The iteration for this 2 × 2 system is given by:

¶vi ¶u
ui ­ vi i
¶y ¶y
xi+1 = xi ­ Eq. 8.2a
¶ui ¶vi ¶ui ¶vi
­
¶x ¶y ¶y ¶x

DOI: 10.1201/9781003364511-8 85
86 Calculations for Process Engineering Using Excel VBA

¶u i ¶v i
vi ­ui
y i +1 = y i ­ ¶x ¶x
Eq. 8.2b
¶u i ¶v i ¶u i ¶v i
­
¶x ¶y ¶y ¶x

where the partial derivatives are evaluated at the values of xi and yi.
An analog to the secant method in Chapter 2 can be used to avoid differential
equations. In this case, two different step sizes, delx and dely, can be applied. VBA
code was developed in the following example to calculate the binary adsorption at
the air/water surface (Le et al. 2012). It should be noted that the step sizes (del1 and
del2) and initial values need to be selected appropriately. If these steps are too large,
the solution can be unstable.

EXAMPLE 8�1� SOLVING TWO NON-LINEAR EQUATIONS BY

THE SECANT METHOD�
PROBLEM STATEMENT
The co-adsorption of two surfactants can be modelled by two equations:

q1
K1C1 =
(1 ­ q1 ­ q2 ) 1
r { (
exp ­2A1q1 ­ 2A12q 2 + (1 ­ r1 ) A1q12 + A2q 22 + 2A12q1q 2 )} Eq. 8.3

q2
K 2 C2 =
(1 ­ q1 ­ q2 )
r2 { (
exp ­2A2q 2 ­ 2A12q1 + (1 ­ r2 ) A1q12 + A2q 22 + 2A12q1q 2 )} Eq. 8.4

where Ci and Γi are the bulk and surface concentrations of surfactants (i = 1,

2), respectively, qi = Gi / G m ,i is the monolayer coverage. G m ,i , K i and Ai are the
adsorption constants of the surfactants, which are predetermined individually.
A12 is a parameter accounting for the cross-interaction between adsorbed mol­
ecules of the two surfactants.
The two surface area fractions are defined as: ri = 1 / wG m , i
Gi G
+ 2
G m ,i G m,2
where the mean molar area, w , is defined as w =
G1 + G 2
From the binary adsorption, the surface tension (γ) is given as:

RT
g = 72 ­ ln (1 ­ q1 ­ q 2 ) A1q12 + A2q 22 + 2A12q1q 2 Eq. 8.5
w
where R and T are the gas constant and temperature, respectively.
For a particular system of surfactant mixture, the parameters are given
in Table 8.1. Calculate the surface tension of this mixture at 25°C when
c1 = 9.6´10 –3 M and c1 =1.7 ´10 – 4 M .
Solving Systems of Non-Linear Equations 87

TABLE 8.1
Parameters for Binary Adsorption of Two Surfactants
A B C D E F G
1 Γm,1 (mol/m2) Γm,2 (mol/m2) K1 (M−1) K1 (M−1) A1 A2 A12
2 1.5 × 10–5 1.1 × 10–5 40.1 2596 0 0.2 –0.963

Solution
The problem requires solving two equations simultaneously. These two equations are
not in an explicit form. However, the two functions are the same, with interchanging
roles between c1 and c2. Hence, a common VBA function, f457, can be used.

Code 8�1� Secant method for a system of two non-linear equations�

Function jacobian457(c1, c2, k1, k2, w1, w2, a1, a2,
a12, theta1, theta2)
Dim i, Err, del1, del2, u, v, dudx, dudy, dvdx, dvdy, a(0, 3)
Err = 1: del1 = theta1 / 200: del2 = theta2 /200: i = 0

While Err > (0.0000000001) and i < 1000

u = f457(theta1, theta2, w1, w2, c1, b1, a1, a2, a12)
v = f457(theta2, theta1, w2, w1, c2, b2, a2, a1, a12)

dudx = (u–f457(theta1–del1, theta2, w1, w2, c1, k1, a1,

a2, a12))/del1 ‘d(f1)/d(theta1)
dudy = (u–f457(theta1, theta2–del2, w1, w2, c1, k1,
a1, a2, a12))/del2
dvdx = (v–f457(theta2, theta1–del1, w2, w1, c2, k2,
a2, a1, a12))/del1
dvdy = (v–f457(theta2–del2, theta1, w2, w1, c2, k2,
a2, a1, a12))/del2

theta1 = theta1–(u * dvdy–v * dudy)/(dudx * dvdy–

dudy * dvdx)
theta2 = theta2–(v * dudx–u * dvdx)/(dudx * dvdy–
dudy * dvdx)

Err = u ^ 2 + v ^ 2: i = i + 1
Wend
a(0, 0) = theta1: a(0, 1) = theta2: a(0, 2) = i:
a(0, 3) = (theta1 + theta2)/(theta1 / w1 + theta2 / w2)
jacobian457 = a
End Function
88 Calculations for Process Engineering Using Excel VBA

Function f457(thetai, thetaj, wi, wj, ci, ki, ai, aj,

aij)
Dim w, ni, ff
w = (thetai + thetaj)/(thetai/wi + thetaj/wj)
ni = wi/w
ff = (thetai/(1–thetai–thetaj) ^ ni) * Exp(-2 * ai *
thetai–2 * aij * thetaj)
ff = ff * Exp((1–ni) * (ai * thetai ^ 2 + aj * thetaj
^ 2 + 2 * aij * thetai * thetaj))
f457 = ff–ci * ki
End Function

Using theta1 = 0.282 and theta2 = 0.321, the solution is obtained after six iter­
ations: theta1 = 0.183, theta2 = 0.192 and ω = 77271 (m2/mol). From Eq. 8.5, the
surface tension is calculated as 55.0 mN/m.

8.3 ADVANCED EXAMPLE: TEAR ANALYSIS WITH TWO

INDEPENDENT UNKNOWNS
In the previous problem, the two equations are symmetric; thus, the same VBA func­
tion can be used for evaluation. In the following examples, the two equations are
different. A similar approach is applicable.

EXAMPLE 8�2� TEAR ANALYSIS WITH TWO INDEPENDENT

VARIABLES�
PROBLEM STATEMENT
Ethyl chloride (CH3CH2Cl) can be manufactured by an exothermic reaction of
hydrogen chloride (HCl) with ethylene (CH2=CH2):

C2 H 4 + HCl ® C2 H 5 Cl

The process can be simplified in the following diagram, adapted from example
10.2–3 (Felder and Rousseau 2005). The fresh feed contains 100 kgmol/h of
HCl and C2H4 mixture. The fresh feed is mixed with the recycle stream before
entering the reactor. The single-pass ethylene conversion is 90%. After the
reactor, the stream is separated into Product (containing C2H5Cl only) and
Stream S-4 (containing C2H4 and HCl). Stream S4 is split into Waste and
Recycle streams. Streams S4, Waste and Recycle have the same composition.
The ethylene molar fraction in the feed is 52%, and the waste stream has a flow
rate of 8 kgmol/h.
Find the molar flows for two species in the recycle stream.
Solving Systems of Non-Linear Equations 89

FIGURE 8.1 Acetylene production with a Recycle stream and two unknowns.

Solution
The flowsheet is the same as in Example 3.2. For example, if the molar fraction is
50%, the solution is the same as in Example 3.2 and can be solved by a bisection
code. In this case, however, since the molar fractions of the feed are not equal, the
solution will have unequal flows in the Recycle stream (Figure 8.1).
For any combination of the R1 and R2, the material balances allow the calculation
of the tear stream (R1cal and R2cal). Consequently, the program needs to solve the
following equations simultaneously:

dif 1 ( R1g,R2 g ) = R1g ­ R1cal Eq. 8.6

dif 2 ( R1g,R2g ) = R 2g ­ R 2cal Eq. 8.7

Hence, two VBA functions are developed correspondingly.

Code 8�2� Two primary functions for tear analysis�

Function difR1(Rg1, Rg2, feed, fraction, Conv, W)
Dim S1(4), S2(4), S4(4), Pro(4), Waste(4), R(4), Rt(4)
Dim ratio

R(1) = Rg1: R(2) = Rg2 ‘tear stream

S1(1) = feed * fraction + R(1): S1(2) = feed * (1–frac­
tion) + R(2)
S2(1) = S1(1)–S1(1) * Conv: S2(2) = S1(2)–S1(1) * Conv:
S2(3) = S1(2) * Conv
S4(1) = S2(1): S4(2) = S2(2)
ratio = S4(1)/(S4(1) + S4(2))
Rt(1) = S4(1)–W * ratio: Rt(2) = S4(2)–W * (1–ratio):

difR1 = Rt(1)–R(1)
End Function

Function difR2(Rg1, Rg2, feed, fraction, Conv, W)

Dim S1(4), S2(4), S4(4), Pro(4), Waste(4), R(4), Rt(4)
90 Calculations for Process Engineering Using Excel VBA

Dim ratio
R(1) = Rg1: R(2) = Rg2 ‘tear stream
S1(1) = feed * fraction + R(1): S1(2) = feed * (1–frac­
tion) + R(2)
S2(1) = S1(1)–S1(1) * Conv: S2(2) = S1(2)–S1(1) * Conv:
S2(3) = S1(2) * Conv
S4(1) = S2(1): S4(2) = S2(2)
ratio = S4(1)/(S4(1) + S4(2))
Rt(1) = S4(1)–W * ratio: Rt(2) = S4(2)–W * (1–ratio):
difR2 = Rt(2)–R(2)
End Function

It should be noted that these functions include the same calculation until the last
line. With these two functions, the four elements of a Jacobian matrix can be obtained
and used to calculate the next set of R1 and R2. The iteration is repeated until the
total error is sufficiently small.

Code 8�3� System of two non-linear equations (in combination with tear
analysis)�
Function jacobian(R1i, R2i, feed, fraction, Conv, W)
Dim i, Err, del1, del2, u, v, dudx, dudy, dvdx, dvdy, a(0, 2)

Err = 1: del1 = 0.001: del2 = 0.001: i = 0

While Err > (0.000000001) And i < 100

u = difR1(R1i, R2i, feed, fraction, Conv, W)
v = difR2(R1i, R2i, feed, fraction, Conv, W)

dudx = (u–difR1(R1i–del1, R2i, feed, fraction, Conv,

W))/del1
dudy = (u–difR1(R1i, R2i–del2, feed, fraction, Conv,
W))/del2
dvdx = (v–difR2(R1i–del1, R2i, feed, fraction, Conv,
W))/del1
dvdy = (v–difR2(R1i, R2i–del2, feed, fraction, Conv,
W))/del2

R1i = R1i–(u * dvdy–v * dudy)/(dudx * dvdy–dudy * dvdx)

R2i = R2i–(v * dudx–u * dvdx)/(dudx * dvdy–dudy * dvdx)
Err = u ^ 2 + v ^ 2: i = i + 1
Wend
a(0, 0) = R1i: a(0, 1) = R2i: a(0, 2) = i
jacobian = a
End Function
Solving Systems of Non-Linear Equations 91

With initial guesses of (5,5), the code gives R1 = 3.111 kgmol/h and R2 = 9.333
kgmol/hr after five iterations. If the fraction is changed to 0.5, the code should give the
same answer as in Example 3.2. While the method gives a quick solution for this case,
extending it to more variables (for example, three or four species in the tear stream) will
be difficult. Therefore, in the next section, an empirical search method is introduced.

8.4 ADVANCED EXAMPLE: TEAR ANALYSIS WITH STOCHASTIC

OPTIMISATION
Stochastic optimisation is an algorithm that uses random samples or simulations to
optimise an objective function. In contrast to deterministic optimisation algorithms,
which use exact formulas and methods to optimise an objective function, stochastic
optimisation algorithms allow randomness in the prediction. Stochastic optimisation
algorithms typically use an iterative procedure to find the optimal solution. At each
iteration, the algorithm samples a subset of the data or a random point in the search
space and uses it to update the current estimate of the optimal solution. The algorithm
then repeats the process until a convergence criterion is met. Stochastic optimisation
methods rely on random variables. VBA can utilise the random function (RAND) for
such a purpose.
Harmony search is a stochastic optimisation (Ingram and Zhang 2010). It is
inspired by the process of musicians improvising music together to produce a har­
monious melody. In harmony search, a potential solution to the optimisation problem
is represented as “harmony,” which consists of a set of values for the decision vari­
ables. The algorithm starts with an initial population of randomly generated harmo­
nies. Then each iteration creates a new harmony by blending existing harmonies in
the population. The blending is done randomly, aiming to produce a new harmony
with a better objective function value than the existing ones. The blending process in
harmony search involves three options: memory consideration, pitch adjustment and
randomisation. In each iteration, one of these three options is selected randomly. The
selection and calculation of these options are explained in the following example.

EXAMPLE 8�3� HARMONY SEARCH WITH TWO INDEPENDENT

VARIABLES�
Problem statement
Repeat Example 8.2 using the harmony search method.

Solution
In this case, the previous example is used to demonstrate the method. Instead of
solving two equations simultaneously, the harmony search aims to minimise the sum
of squares of difference:

dif f 2 = ( R2g ­ R 2cal ) + ( R1g ­ R1cal )

2 2
Eq. 8.8
92 Calculations for Process Engineering Using Excel VBA

The function in the previous equation is a total difference between guessed and cal­
culated values of all unknowns in the tear stream. The function is evaluated in Code
8.4, using the same calculation steps in Code 8.2.

Code 8�4� Function to calculate the sum of squares of differences (with

tear analysis)
Function diffR2(Rg1, Rg2, feed, fraction, Conv, W)
Dim S1(4), S2(4), S4(4), Pro(4), Waste(4), R(4), Rt(4)
Dim ratio
R(1) = Rg1: R(2) = Rg2 ‘tear stream
S1(1) = feed * fraction + R(1): S1(2) = feed * (1–fraction)
+ R(2)
S2(1) = S1(1)–S1(1) * Conv: S2(2) = S1(2)–S1(1) * Conv:
S2(3) = S1(2) * Conv ‘we don’t need S(3) for calc
S4(1) = S2(1): S4(2) = S2(2): ratio = S4(1)/(S4(1) +
S4(2))
Rt(1) = S4(1)–W * ratio: Rt(2) = S4(2)–W * (1–ratio):
diffR2 = (Rt(1)–R(1)) ^ 2 + (Rt(2)–R(2)) ^ 2
End Function

Consequently, the harmony search can be applied (Codes 8.5 to 8.6). The step­
by-step is explained in the following. First, a set of 100 (HMS) combinations of x1
and x2 are generated randomly. It should be noted that the VBA function RND has
a small range of randomness. Therefore, one can improve the process by adding the
function RANDOMIZE before using RND. Furthermore, the code can indirectly use
the Excel spreadsheet function by typing RAND() into a cell and then EVALUATE
that cell. Next, the objective functions of these combinations are evaluated and stored
in the HM array (Code 8.4). The code has three key parameters: HMS (Harmony
Memory Size), HMS ≥ 1; HMCR (Harmony Memory Considering Rate), 0 ≤ HMCR
≤ 1; and PAR (Pitch Adjusting Rate), 0 ≤ PAR ≤ 1 (Ingram and Zhang 2010).

Code 8�5� Harmony search with two variables (the function diffR2 is
defined in Code 8�4)� Section 1: generate the initial sets of solutions�
Function harmony2(LowerR, UpperR, feed, fraction,
Conv, W)
Dim limit(5, 3), x(5), ND
Dim HM(500, 3), HMS, HMCR, PAR, MaxIte
Dim d1, sol, i, j, iter, rand
Dim hmax, hmax _ n, hmin, hmin _ n, out(4, 4)

ND = 2 ‘number of decision variables

‘lower limit, upper limit & # of initial segments

Solving Systems of Non-Linear Equations 93

limit(1, 1) = LowerR: limit(1, 2) = UpperR: limit(1,

3) = 1000
limit(2, 1) = LowerR: limit(2, 2) = UpperR: limit(2,
3) = 1000

HMS = 100: HMCR = 0.9: PAR = 0.6: MaxIte = 100000

‘--------generate the 1st sets of 100 combinations
For i = 1 To HMS
For j = 1 To ND
x(j) = limit(j, 1) + (limit(j, 2)–limit(j, 1))
* Rnd ‘generate random values of x1 and x2 between
lower and upper limits
HM(i, j) = x(j)
Next j
HM(i, 3) = diffR2(x(1), x(2), feed, fraction, Conv, W)
‘evaluate the function
Next i
‘------------------­

Subsequently, the minimum and maximum values within HM are obtained in

Code 8.6.

Code 8�6� Harmony search with two variables� Section 2: find the maxi­
mum and minimum values from the initial sets�
hmax _ n = 1: hmax = HM(1, 3)
For i = 2 To HMS
If HM(i, 3) > hmax Then
hmax _ n = i
hmax = HM(i, 3)
End If
Next i

hmin _ n = 1: hmin = HM(1, 3)

For i = 2 To HMS
If HM(i, 3) < hmin Then
hmin _ n = i: hmin = HM(i, 3)
End If
Next i
out(1, 1) = “Min from the initial sets”: out(1,
2) = “Max from the initial sets”
out(2, 1) = hmin: out(2, 2) = hmax
‘--------------­
94 Calculations for Process Engineering Using Excel VBA

In the next step, a new combination is obtained randomly. A random value is gen­
erated and compared with the two limits, HMCR and PAR. Depending on the value
of RND, either one of three operations will be carried out:

• If RND > HMCR: a new set will be found randomly

• If PAR < RND < HMCR: a new combination is found by randomly selecting
a value from the existing sets.
• If PAR > RND: adjust the obtained combination in option 2.

At the end of the selection, the objective function of the new combination is evalu­
ated and compared against the worst combination in the existing sets. Once the re­
quired number of iterations is reached, the improvement process stops. Finally, the
best combination with sets is identified and exported.

Code 8�7� Harmony search with two variables� Section 3: generating and
comparing the new combination�
---------------
For iter = 1 To MaxIte-1
Randomize: rand = Rnd
For j = 1 To ND
If rand >= HMCR Then ‘Random Searching
x(j) = limit(j, 1) + (limit(j, 2)–
limit(j, 1)) * rand
Else
d1 = Int(HMS * rand) + 1
x(j) = HM(d1, j)
If rand < PAR Then ‘Pitch Adjusting
d1 = (limit(j, 2)–limit(j, 1))/
limit(j, 3)
x(j) = x(j) + d1 * (rand–PAR /
2)
End If
End If
Next j
sol = diffR2(x(1), x(2), feed, fraction, Conv, W)

‘----find hmax and corresponding index hmax _ num

hmax _ n = 1: hmax = HM(1, 3)
For i = 2 To HMS
If HM(i, 3) > hmax Then
hmax _ n = i: hmax = HM(i, 3)
End If
Next i
‘----replacing hmax with the new solution if improved
Solving Systems of Non-Linear Equations 95

If sol < hmax Then ‘if new Sol is smaller than hmax,
replace hmax
For j = 1 To ND
HM(hmax _ n, j) = x(j)
Next j
HM(hmax _ n, 3) = sol
End If

Next iter

hmin _ n = 1: hmin = HM(1, 3)

For i = 2 To HMS
If HM(i, 3) < hmin Then
hmin _ n = i: hmin = HM(i, 3)
End If
Next i
out(3, 1) = “Number of iter”: out(3,
2) = “x1”: out(3, 3) = “x2”: out(3, 4) = “Error”
out(4, 1) = iter: out(4, 2) = HM(hmin _ n,
1): out(4, 3) = HM(hmin _ n, 2): out(4, 4) = HM(hmin _ n,
3)
harmony2 = out
End Function

In this code section, RANDOMIZE is used before calling the RND again. The
value (“rand”) is used for the three calculations in the three improving options. Tech­
nically, different random numbers can be generated and used for these calculations,
which might improve the solution.
With limits at 3 and 10, the code can generate the answers as in Table 8.2. It
should be noted that the answers will be different when the code is repeated.
The two flow rates are close to the answer from the secant method and can be
improved by increasing the number of iterations and other parameters (HMS,
HMCR and PAR).

TABLE 8.2
Harmony Search Output
Min from the Initial Sets Max from the Initial Sets 0 0
0.099774994 113.9915897 0 0
Number of iter x1 x2 Error
10000 3.105824475 7.288284783 0.098218223
96 Calculations for Process Engineering Using Excel VBA

8.5 FURTHER EXAMPLES FOR SELF-PRACTICE

EXAMPLE 8�4� REDUCING CALCULATIONS OF TEAR

ANALYSIS�
In Example 8.2, each iteration must call functions difR1 and difR2 three times
each. The calculations within these two functions are a sequence of mass bal­
ances for the modular operations of the process. The calculation of the two
functions is essentially the same and thus offers an opportunity to reduce com­
putational time. Rewrite the codes so each iteration only needs three mass bal­
ance calculations.

Answer
First, a new function is needed to replace both difR1 and defR2. The func­
tion will return two values simultaneously as two elements of an output array
(2-dimensional). Consequently, the main function will call the corresponding
elements in a Jacobian procedure. Therefore, the output array, even with only
two elements, needs to be in a 2-dimensional form.

EXAMPLE 8�5� Adjusting parameters of the harmony search�

In Example 8.3, the parameters are HMS = 100, HMCR = 0.9, PAR = 0.6, and
MaxIte = 100000. Adjust HMS and MaxIte to improve the process.

Answer
By increasing the number of sets, the best solution is improved significantly.
For instance, Table 8.3 shows that HMS = 500 and MaxIte = 100 can provide
a better solution. However, in this case, the harmony search does not improve
the best solution from the random set. Perhaps some of these problems can be
solved by many random trials of guesses, similar to the search algorithm in
Example 12.3.

TABLE 8.3
Harmony Search Output
Min from the Initial Sets Max from the Initial Sets 0 0
0.006903335 126.048336 0 0
Number of iter x1 x2 Error
100 3.161868572 9.779379845 0.006903335
 9 Solving Ordinary Differential
Equations
9.1 INTRODUCTION
An ordinary differential equation (ODE) is a differential equation containing one
or more functions of one independent variable and the derivatives of those func­
tions. The equations are used to describe many physical processes. For example,
the equations play a central role in predicting the specie balances for chemical reac­
tions. Consequently, solving ODEs is fundamental to reactor design and process flow
sheeting. The order of the ODEs is defined as the highest order of any differential
term. Any nth-order ODE can be expressed in the system of n first-order ODEs.
Numerical methods for solving ordinary differential equations are techniques used
to approximate the solution of an ODE. These methods involve approximating the
derivative or differential equation with a mathematical expression that can be solved
numerically. Many numerical schemes are available for solving ODEs, each with
advantages and disadvantages.

9.2 ONE-STEP METHODS WITH A CONSTANT STEP SIZE

Assuming that we need to approximate the following ODE from x0 to xf:

dy
= f ( x, y ) Eq. 9.1
dx
where y = y0 at x = x0 .
The one-step method will divide the domain, from x = x0 to x f , into m steps with
a constant step size of h. The relationship between h and m is defined as:

h= ( xf ­ x0 ) / m Eq. 9.2

The method will then approximate the values of y at those steps sequentially. These
one-step approximation methods are named the Runge–Kutta (RK) methods for dif­
ferential equations. There are many variations, which are sub-classified according
to the order of the scheme. For example, an nth-order RK method will require n
evaluations within each step and has local and global errors on the order of hn+1 and
hn, respectively. There is only one first-order RK method, Euler’s, which is the most
straightforward numerical method for solving ODEs.
Mathematically, Euler’s prediction is:

y ( x + h ) = y ( x ) + f ( x, y ) h Eq. 9.3

DOI: 10.1201/9781003364511-9 97
98 Calculations for Process Engineering Using Excel VBA

This equation can be demonstrated in the following example.

EXAMPLE 9�1� EULER’S METHOD FOR A SINGLE

FIRST-ORDER ODE�
PROBLEM STATEMENT
Solve the equation by the Euler method:

dy/d x = cos ( x ) from x = 0 to x = p

with an initial value y = 0 at x = 0 .

Apply three different step sizes: π/2, π/6 and π/12. Then plot the results and
compare them with the exact solution y = sin ( x ).

Solution
Using Euler’s method, the solution can be found with simple VBA code (Code 9.1).

Code 9�1� Euler’s method for a single first-order ODE�

Function RK _ 1(x0, y0, dx, xmax)
Dim matrix(), i, n

n = xmax/dx: ReDim matrix(n, 1)

matrix(0, 0) = x0: matrix(0, 1) = y0:

For i = 1 To n
matrix(i, 0) = matrix(i–1, 0) + dx
matrix(i, 1) = matrix(i–1, 1) + ff(matrix(i–1, 0),
matrix(i–1, 1)) * dx
Next i
RK _ 1 = matrix
End Function

Function ff(x, y)
ff = Cos(x)
End Function

It should be noted that the function ff in this example depends on x only, so y

can be removed from the argument of function ff. One of the apparent problems
with Euler’s method is balancing the accuracy and computational time. This can be
demonstrated by the three solutions in Figure 9.1.
Solving Ordinary Differential Equations 99

FIGURE 9.1 Influence of step size on the accuracy of Euler’s method.

Figure 9.1 shows that solution accuracy increases with the reduced step size, h.
However, smaller step sizes will increase the number of iterations and, thus, com­
putational time. The curves also demonstrate two types of errors in the method: the
local truncation error and the accumulative truncation error. For Euler’s method, the
local error is on the order of h2, and the global error is on the order of h (Chapra
and Canale 2010). For many problems, the required step size might be too small
and result in numerous iterations. Consequently, there are other RK methods in the
literature.

9.2.1 second-order methods

In any other order higher than 1, the RK method has many variations. For exam­
ple, the second-order Runge–Kutta method has three well-known variations: Heun’s,
improved polygon and Ralson’s schemes (Chapra and Canale 2010). These varia­
tions have similar accuracy and computation complexity. Here equations and VBA
code for the improved polygon are introduced.
Mathematically, the improved polygon method predicts:

yi+1 = yi + f ( x1+0.5 ,yi+0.5 ) h Eq. 9.4

where yi+0.5 is the slope of the midpoint between xi and xi+1 and is given by:

yi+0.5 = yi + f ( xi , yi ) h / 2 Eq. 9.5

VBA code for the improved polygon method is given in Code 9.2.
100 Calculations for Process Engineering Using Excel VBA

Code 9�2� Improved polygon method for a single first-order ODE�

Function RK _ 2(x0, y0, dx, xmax) ‘Improved polygon
Dim matrix(), i, n, x _ m, y _ m

n = xmax/dx: ReDim matrix(n, 1)

matrix(0, 0) = x0: matrix(0, 1) = y0:

For i = 1 To n
matrix(i, 0) = matrix(i–1, 0) + dx
x _ m = matrix(i–1, 0) + dx / 2
y_m = matrix(i–1, 1) + ff(matrix(i–1, 0),
matrix(i–1, 1)) * dx / 2
matrix(i, 1) = matrix(i–1, 1) + ff(x _ m, y _ m) * dx
Next i
RK _ 2 = matrix
End Function

9.2.2 fourth-order methods

There are many variations of Runge–Kutta methods in third, fourth and higher orders.
Yet there is one outstanding fourth-order method. This method is sometimes referred
to as the classical fourth-order RK method. The scheme is given as:

é1 ù
yi +1 = yi + ê ( k1 + 2k2 + 2k3 + k4 ) ú h Eq. 9.6
ë6 û

where

k1 = f ( xi , yi ) Eq. 9.7a

æ 1 1 ö
k2 = f ç xi + h, yi + k1h ÷ Eq. 9.7b
è 2 2 ø

æ 1 1 ö
k3 = f ç xi + h, yi + k2 h ÷ Eq. 9.7c
è 2 2 ø

k4 = f ( xi + h, yi + k3 h ) Eq. 9.7d

VBA code for the method is given in Code 9.3. The code and VBA functions in
Codes 9.1 and 9.2 compare the three methods, as demonstrated in Figure 9.2.
Solving Ordinary Differential Equations 101

FIGURE 9.2 Comparison between Euler’s, improved polygon and fourth-order Runge–
Kutta methods.

Code 9�3� The classical fourth-order RK method�

Function RK _ 4(x0, y0, h, Xmax)
Dim m(), i, n
Dim k1, k2, k3, k4
n = Xmax/h: ReDim m(n, 1)
m(0, 0) = x0: m(0, 1) = y0:

For i = 1 To n
m(i, 0) = m(i–1, 0) + h
k1 = ff(m(i–1, 0), m(i–1, 1))
k2 = ff(m(i–1, 0) + h / 2, m(i–1, 1) + k1 * h / 2)
k3 = ff(m(i–1, 0) + h / 2, m(i–1, 1) + k2 * h / 2)
k4 = ff(m(i–1, 0) + h, m(i–1, 1) + k3 * h)
m(i, 1) = m(i–1, 1) + (k1 + 2 * k2 + 2 * k3 + k4) * h / 6
Next i
RK _ 4 = m
End Function

9.3 ONE-STEP METHODS WITH AN ADAPTIVE STEP SIZE

The previous section shows that increasing the orders of Runge–Kutta methods can reduce
the numerical steps without compromising accuracy. However, these methods still require
a constant step size. In these instances, the number of steps and the final step are selected
regardless of the behaviour function. However, the actual local error depends on the local
behaviour. For instance, if the function is almost linear, a moderate step size is adequate.
However, a smaller step size might be required if the equations change dramatically.
102 Calculations for Process Engineering Using Excel VBA

A general equation might have sections with dramatic and gradual changes.
Consequently, the coder must balance step size (and the number of steps) and accu­
racy. One way to achieve this is by using an adaptive step size. In this approach,
the step size of each increment is determined by estimating the local absolute error.
If only one RK method is used, the absolute error is unknown. However, com­
bining two RK methods with different orders can control the error. In theory, any
two RK methods can be selected. However, there are certain combinations in which
the evaluations are nested to reduce the number of function evaluations. One of the
well-known combinations is the Runge–Kutta Fehlberg method, which combines a
fourth- and fifth-order Runge–Kutta method.
On the other hand, the easiest way is combining a first-order and second-order
method. This section uses the combination to demonstrate the adaptive step size.
Within each iteration, the method starts with a guessed step size, hpresent. Conse­
quently, the absolute difference between the full-step and the two half-step values is
the error estimate, errc. By comparing with the user-specified tolerance, errd, the step
size is adjusted (Press and Teukolsky 1992):
a
é errd ù
hnew = 0.95 hpresent ê ú Eq. 9.8
ë errc û
In Eq. 9.8, α is the adjustment factor, which equals 0.2 when errc > errd, and 0.25
when errc < errd. Using the formula, the step size is adjusted until the desired ac­
curacy is given by

errd = erra y ( x + h ) Eq. 9.9

where erra is the required relative accuracy.

EXAMPLE 9�2� RK METHODS WITH AN ADAPTIVE STEP SIZE�

PROBLEM STATEMENT
Compare different Euler’s methods for the following ODE:

­10 ( x ­2 )
2

y¢ = 15e – 0.8y

from x = 0 to x = 4 with an initial value y = 0.5 at x = 0.

Part A� Using Euler’s method with a step size of 0.4.
Part B� Using the classical fourth-order Runge Kutta method with a step of 0.5.
Part C� Using an adaptive step size from Euler’s and improved polygon
methods.
Plot the results and compare the results.

Solution
The Euler method in Code 9.1 can be modified to combine with function fff in Code
9.4. The fourth-order Runge–Kutta and adaptive methods are presented in Code 9.4.
Solving Ordinary Differential Equations 103

Code 9�4� Adaptive step size with Euler and second-order method�
Option Base 0
Function fff(x, y)
fff = 15 * Exp(-(x–2) ^ 2 / 0.1)–0.75 * y
End Function
Function RK _ 42(x0, y0, h, Xmax)
Dim m(), i, n, k1, k2, k3, k4

n = Xmax/h: ReDim m(n, 1)

m(0, 0) = x0: m(0, 1) = y0:

For i = 1 To n
m(i, 0) = m(i–1, 0) + h
k1 = fff(m(i–1, 0), m(i–1, 1))
k2 = fff(m(i–1, 0) + h / 2, m(i–1, 1) + k1 * h / 2)
k3 = fff(m(i–1, 0) + h / 2, m(i–1, 1) + k2 * h / 2)
k4 = fff(m(i–1, 0) + h, m(i–1, 1) + k3 * h)
m(i, 1) = m(i–1, 1) + (k1 + 2 * k2 + 2 * k3 + k4) * h / 6
Next i

RK _ 42 = m
End Function

Function adaptive _ E _ 2(x, y, h, errA)

Dim del, y1, y05, x05, output(3), k, ErrD, hnew
ErrD = 1: del = ErrD * 2: hnew = h: k = 0

While del > ErrD And k < 200

h = hnew: k = k + 1
y1 = y + fff(x, y) * h
y05 = y + fff(x, y) * h / 2
del = Abs(fff(x + h / 2, y05) * h / 2 + y05–y1)
ErrD = errA * Abs(y1)

If Abs(del) > ErrD Then

hnew = 0.95 * h * Abs(ErrD/del) ^ 0.25
Else
hnew = 0.95 * h * Abs(ErrD/del) ^ 0.2
End If
Wend

output(0) = h: output(1) = x + h: output(2) = y1:

output(3) = k
adaptive _ E _ 2 = output
End Function
104 Calculations for Process Engineering Using Excel VBA

TABLE 9.1
Adaptive Step Size
A B C D
h x y k
0.5 0 0.5
0.23497 0.23497 0.411886106 4
0.23497 0.469941 0.339300329 4
0.23497 0.704911 0.279506296 4
0.234811 0.939721 0.230283213 4
0.219514 1.159236 0.192413513 4
0.119156 1.278392 0.176739477 4
0.064968 1.34336 0.173465187 5
0.048407 1.391767 0.17690453 5
0.03784 1.429607 0.185924458 6
0.033484 1.463091 0.200662556 6
0.030896 1.493988 0.221957736 6
0.027883 1.521871 0.249633028 7
0.027085 1.548956 0.285865638 7
0.026812 1.575768 0.33270589 7
0.02698 1.602748 0.392888656 7
0.027554 1.630302 0.470063829 7
0.028543 1.658845 0.569148304 7
0.030004 1.68885 0.696886125 7

Please note that the adaptive_E function in Code 9.4 only does one step. To get
the whole function from 0 to 4, the function in Code 9.4 is called continuously in a
table form. A section of the table is shown in Table 9.1. The VBA function also gives
the number of adjustments, k, at each step. The predetermined error plays a clear role
in this value. In Table 9.1, the starting step, h, is fixed at 0.5, and the required relative
error, ErrA, is 1%. The code is designed so that this input can be changed easily.
The adaptive method is plotted against Euler’s, fourth-order RK and analytical
solutions in Figure 9.3.
At first glance, the adaptive step method requires many iterations, with some
improved accuracy over Euler’s method. However, the combined first- and second-
order methods have errors on the order of h3, which is still larger than a fourth-order
Runge–Kutta method. Hence, the adaptive method is less accurate than the fourth-
order RK, as shown in Figure 9.3.
On the other hand, the adaptive method has an advantage when the whole function
is needed. For instance, when function behaviour is used to integrate a curve under
the function, the accuracy along the function is more important than the final value at
the upper limit. This is demonstrated in the plug-flow reactor design in Chapter 10.
Solving Ordinary Differential Equations 105

FIGURE 9.3 RK method with an adaptive step size.

FIGURE 9.4 Euler’s method with a changing step size.

Finally, the calculation can be optimised by manually selecting the step size.
For example, after applying Euler’s method with a constant step size of 0.4, dra­
matic changes are observed in the region between 1 and 3. Consequently, the whole
region between 0 and 4 can be divided into three sections: (0,1), (1,3) and (3,4). The
step sizes in these three regions are selected at 0.2, 0.05 and 0.2, respectively. The
resulting prediction with Euler’s method is demonstrated in Figure 9.4. The method
106 Calculations for Process Engineering Using Excel VBA

has a better prediction than the adaptive method while having a very simple calcu­
lation. The approach requires a visual justification of the important regions. Such
an approach is very helpful for a system of ODEs, such as reactor modelling in
Chapter 10.

9.4 JUSTIFICATION OF THE SELECTED STEP SIZE

In the previous sections, it can be seen that the step size plays a critical role in deter­
mining the accuracy of the numerical methods. It is difficult to justify the step size
without knowing the actual function (as in Figures 9.3 and 9.4). Generally, a smaller
step size is better. Yet there is no clear definition of a sufficiently good step size.
Selecting an unnecessarily small step size would significantly increase the number
of steps.
One simple approach is applying several different step sizes until no significant
changes occur. Here, the approach is applied to the previous example. Assume
that Euler’s method with three constant sizes is selected. For the two regions,
(0,1) and (3,4), h0 is used. Between 1 and 3, the step size is reduced four times,
h0/4. Subsequently, the method is applied for four values of h0: 0.4, 0.2, 0.1 and
0.05 (Figure 9.5). In Figure 9.5, it can be seen that the last two curves are almost
identical. Consequently, an h0 value of 0.1 should be sufficiently adequate for the
problem.

FIGURE 9.5 Euler’s method with four different step sizes.

Solving Ordinary Differential Equations 107

9.5 CONCLUDING REMARKS

While there are many different numerical methods for solving ODEs, all methods
have advantages and disadvantages. The users need to balance between accuracy,
computational time and complication of the coding. The third factor will be demon­
strated in the next chapter, in which the solutions are combined with other numerical
methods to solve engineering problems. Euler’s method with a reasonable step size
should be used. If the method is inadequate, the second preferable option should be
the fourth-order RK (Code 9.3). Finally, if the function has a significant variation in
a certain region, a manual step size variation should be employed.
10 Applications of ODEs
10.1 INTRODUCTION
Numerical methods for solving ODEs are essential for designing and analysing engi­
neering systems. Numerical methods are used to simulate these models and make
predictions about the behaviour of the physical system. With a routine solution of a
system of ODEs, the VBA code can be modified to solve complicated problems, such
as optimising a reactor. In particular, combining other methods, such as bisection or
binary search, can provide a powerful tool to solve these physical processes.

10.2 ADVANCED EXAMPLE: OPTIMISATION OF A SEMI-BATCH

REACTOR
A semi-batch process, also referred to as a semi-continuous process, is an operational
mode used in chemical and process engineering. It combines elements of both batch
and continuous processing. In a batch process, all the raw materials are added at
the start, the reactions occur and the product is removed at the end. In a continuous
process, raw materials are constantly added, and products are constantly removed.
A semi-batch process is a middle ground between these two extremes. In this pro­
cess, one or more reactants are fed into the reaction mixture over time. This might be
done to control the reaction rate, manage heat release from an exothermic reaction
or optimise the yield or selectivity of a reaction. For example, if a reaction is highly
exothermic (i.e. it releases a lot of heat), adding all the reactants at once could cause
the reaction to be uncontrolled, potentially leading to a dangerous situation (Fogler
2006). By adding one reactant slowly, the heat release can be managed safely. Simi­
larly, if a reaction produces an undesired by-product, slowly adding one reactant can
help minimise its production. Modelling a semi-batch process often requires simul­
taneously solving a system of differential and ordinary equations.

EXAMPLE 10�1� SEMI-BATCH REACTOR�

PROBLEM STATEMENT
Consider the series of two reactions, adapted from Example 9.7 (Fogler 2006):
k1
2A® B heat of reaction DH1 = ­6500 cal/mol A, rate of reaction k1a
k2
B ®3C heat of reaction ΔH2 = 8000 cal/mol B, rate of reaction, k2b

CPA = 30 cal/mol K, CPB = 60 cal/mol K, CPC = 20 cal/mol K, CPCatalyst = 35

cal/mol K

108 DOI: 10.1201/9781003364511-10

Applications of ODEs 109

The feed is added continuously at a volumetric flow rate of 240 L/h and tem­
perature of 305 K. The feed contains A at a concentration of 4 mol/L. The
reactor has a heat exchanger with UA = 35000 cal/h K and an exchange (air)
temperature of 298 K. Initially, there is 100 dm3 solution in the reactor, with 1
mol/dm3 of A and 1 mol/L of the catalyst.
Part A� Calculate and plot the chemical concentrations and temperature as a
function of time from zero to 1.5 hours.
Part B� Calculate the maximum concentration of B and the corresponding
reaction time.

Solution
Part A� The problem can be solved using a system of four ODEs and several equa­
tions. The four ODEs account for the changes of three chemicals and temperature.
Heat balance equations describe the transient heat of the two reactions and the heat
input. First, the values of parameters are input into an Excel spreadsheet (Table 10.1).

TABLE 10.1
Parameters of Reactions, Flow Rate, Temperature and Initial Conditions
A B C
1 Reactor Variables Unit
2 Cao 4 mol/dm3
3 To 305 K
4 vo—flowrate 240 dm3/h
4 Ta—temp of air 298 K
5 V_in 100 dm3
6 R 1.987 cal·K−1·mol−1
7 E1 9500 cal/mol
8 E2 7000 cal/mol
9 Tref1 320 K
10 Tref2 300 K
11 k1a 1.25 h−1
12 k2b 0.08 h−1
13 Energy variables
14 delHra −6500 cal/mol A
15 delHrb 8000 cal/mol B
16 UA 35000 cal/hK
17 Cpa 30 cal/molK
18 Cpb 60 cal/molK
19 Cpc 20 cal/molK
20 Cpcatalyst 35 cal/molK
21 N_cat 100 mol
110 Calculations for Process Engineering Using Excel VBA

While these parameters can be included in the VBA function as arguments, such
an arrangement will make the function more complicated and prone to error when
calling the macro. In this instance, the values of parameters are assigned to the VBA
variable using the Range object. The positions of the cells need to be fixed.

Code 10�1� ODEs to calculate the concentration and temperature profiles

of a semi-batch reactor�
Option Base 0
Function dabcTdt(dt, tmax)
Dim M()
Dim i, n, rA, rB, rC, dadt, dbdt, dcdt, dTdt, r1, r2,
CAO, CB0, CC0, T0, Ta, R
Dim E1A, E2B, T1, T2, k1a, k2b, delH1, delH2, UA
Dim CpA, CpB, CpC, CpCa, v0, V_in, CAf, H_ex,
H_A, H_r
‘----------input
CAf = Range(“B2”).Value
T0 = Range(“B3”).Value
v0 = Range(“B4”).Value
Ta = Range(“B5”).Value
V _ in = Range(“B6”).Value
R = Range(“B7”).Value
E1A = Range(“B8”).Value: E2B = Range(“B9”).Value:
T1 = Range(“B10”).Value: T2 = Range(“B11”).Value:
k1a = Range(“B12”).Value: k2b = Range(“B13”).Value:
delH1 = Range(“B15”).Value: delH2 = Range(“B16”).Value:
UA = Range(“B17”).Value:
CpA = Range(“B18”).Value: CpB = Range(“B19”).Value:
CpC = Range(“B20”).Value: CpCa = Range(“B21”).Value:
‘------------­
n = tmax/dt: ReDim M(0 To n, 0 To 5)
M(0, 1) = V _ in: M(0, 2) = 1: M(0, 3) = 0: M(0, 4) = 0:
M(0, 5) = 290

For i = 1 To n
M(i, 0) = M(i–1, 0) + dt: M(i, 1) = M(i–1, 1) + dt * v0
r1 = -k1a * Exp(E1A/R * (1 / T1–1 / M(i–1, 5))) *
M(i–1, 2)
r2 = -k2b * Exp(E2B/R * (1 / T2–1 / M(i–1, 5))) * M(i–1, 3)

dadt = r1 + (CAf–M(i–1, 2)) * v0 / M(i–1, 1): M(i,

2) = M(i–1, 2) + dadt * dt
dbdt = -r1 / 2 + r2–M(i–1, 3) * v0 / M(i–1, 1): M(i,
3) = M(i–1, 3) + dbdt * dt
Applications of ODEs 111

dcdt = -3 * r2–M(i–1, 4) * v0 / M(i–1, 1): M(i,

4) = M(i–1, 4) + dcdt * dt

H _ ex = UA * (298–M(i–1, 5)) ‘heat loss to heat exchanger

H _ A = CAf * v0 * CpA * (M(i–1, 5)–T0) ‘heat of inflow
of pure A
H _ r = (delH1 * r1 + delH2 * r2) * M(i–1, 1)
‘—need to calculate two terms for energy balance
dTdt = (H _ ex–H _ A + H _ r)/((CpA * M(i–1, 2) + CpB *
M(i–1, 3) + CpC * M(i–1, 4)) * M(i, 1) + 100 * CpCa)
M(i, 5) = M(i–1, 5) + dt * dTdt
Next i
dabcTdt = M
End Function

Part B� From the graph in Figure 10.1, the concentration of B has a maximum of
around 0.6 h. To determine this point precisely, we can combine the ODE code with
region elimination code.
First, the function in Code 10.1 is revised so that the output only contains con­
centration B at time tmax. The new function can be named dablTdt_2 (Code 10.2).
The function dabcTdt_2 is essentially the same as in Code 10.1, except for output
(M(n,3) instead of M). Consequently, the function is a subfunction within a golden
section search. The final answers are CB_max = 1.191 and tm = 0.6438 h.

FIGURE 10.1 Concentrations and temperature of a semi-batch reactor.

112 Calculations for Process Engineering Using Excel VBA

Code 10�2� Combination of golden section search and ODEs to optimise a

semi-batch reactor�
Option Base 0
Function dabcTdt _ 2(dt, tmax)
‘----------input
n = Round(tmax/dt, 0): ReDim M(0 To n, 0 To 5)

‘------------same as in Code 10.1

dabcTdt _ 2 = M(n, 3)
End Function

Function golden _ section(t _ l, t _ h, imax, delt)

Dim Gr, d, D1, D2, f1, f2, i, output(2)
Gr = (5 ^ 0.5–1) / 2:
d = t _ h–t _ l: i = 0
d = d * Gr: D2 = t _ l + d: D1 = t _ h–d:
f1 = dabcTdt _ 2(delt, D1): f2 = dabcTdt _ 2(delt, D2)

While d > delt / 2 And i < imax

d = d * Gr
If f1 < f2 Then
t _ l = D1: D1 = D2: D2 = t _ l + d
f1 = f2: f2 = dabcTdt _ 2(delt, D2)
Else
t _ h = D2: D2 = D1: D1 = t _ h–d
f2 = f1: f1 = dabcTdt _ 2(delt, D1)
End If
i = i + 1
Wend

output(0) = (t _ l + t _ h) / 2: output(1) =
dabcTdt _ 2(delt, output(0))

golden _ section = output

End Function

It should be noted that the “For Next” loop in VBA will stop at the rounded number
(for example, n = 12.6, the loop stops at 12). So, if the step size of ODE is larger than the
tolerance of the golden section, the two internal points D1 and D2 can be the same and
may lead to a wrong maximum. The second notation is that the ReDim of M(0 to n) gives
the dimension of the nearest integer to n (for example, if n = 12.6, M has 13 columns).
Similarly, the index is rounded to the nearest integer, M(12.6,3) = M(13,3). Con­
sequently, using round in Code 10.2 is needed to avoid the error. Since n is rounded,
Applications of ODEs 113

the step in the golden section needs to be larger than delt/2 (to differentiate the values
of f1 and f2).

10.3 ADVANCED EXAMPLE: BOUNDARY VALUE ODEs

A boundary value problem (BVP) for ordinary differential equations is a type of
problem where the solution of the ODE is sought subject to specified boundary con­
ditions. In contrast to initial value problems, where the values of the solution and
its derivatives are specified at a single point, in BVPs, the values of the solution are
specified at different points. BVPs arise in many areas of science and engineering,
including mechanics, physics, chemistry and biology.
The two-point BVP is the most common boundary value problem for ODEs,
where the solution is required to satisfy specified conditions at different points. To
solve a BVP, we can combine the numerical methods for initial value problems with
a root-of-equation method. This approach requires the (computational) solution of an
initial value problem. The method assumes the initial values of the unknown varia­
bles and subsequently adjusts the value to satisfy the condition at the other boundary.
The adjustment can be done by using the bisection method. The procedure is often
referred to as a “shooting method”.

EXAMPLE 10�2� SHOOTING METHOD FOR 1-D HEAT DISTRIBUTION�

PROBLEM STATEMENT�
Considering a rod 10 cm long, the temperatures of two ends are maintained at
40 and 200°C. The rod is exposed to air at a constant temperature, Ta = 20°C.
At a steady state, the heat transfer is governed by the equation:

d 2T
= h ( Ta ­ T ) Eq. 10.1
dx 2
where h is the radiative heat loss parameter (cm-1).
Calculate the temperature profile along the rod.

Solutions
The previous second ODE can be expressed as a system of two first ODEs:

dT
=z Eq. 10.2
dx
dz
= h ( Ta ­ T ) Eq. 10.3
dx
In this problem, we will use two different VBA codes. First, code is developed
to calculate the temperature profile for a given set of initial values (Code 10.3). The
code treats the problem as a system of two first ODEs. The code can provide the
temperature profile at the given heat transfer dz/dx (x = 0).
114 Calculations for Process Engineering Using Excel VBA

Code 10�3� Solving the temperature contribution along a long rod for a
given initial conditions�
Option Base 0
Function temp(t0, z0, Ta, h, dx, xmax)
Dim output(), i, n, dxdt, dzdt
n = xmax/dx: ReDim output(n, 2)
output(0, 0) = 0: output(0, 1) = t0: output(0, 2) = z0

For i = 1 To n
output(i, 0) = output(i–1, 0) + dx
output(i, 1) = output(i–1, 1) + output(i–1, 2) * dx
output(i, 2) = output(i–1, 2) + h * (output(i–1, 1)–Ta) * dx
Next i
temp = output
End Function

The second code (Code 10.4) uses using bisection method to work out the required
slope (z0 or dT/dt at x = 0) so that the temperature at the other end equals 200°C.

Code 10�4� Solving the temperature contribution along a long rod by a

bisection method�
Function f(t0, z0, Ta, h, dx, xmax, Te)
Dim m, n
n = xmax/dx
m = temp(t0, z0, Ta, h, dx, xmax)
f = Te–m(n, 1)
End Function

Function BisectionX(Lo, Up, tolt, t0, Ta, h, dx, xmax, Te)

Dim Mi, c1, c3, i, k
k = Abs(Up–Lo)/tolt
c1 = f(t0, Lo, Ta, h, dx, xmax, Te)
For i = 1 To k
Mi = (Lo + Up) / 2: c3 = f(t0, Mi, Ta, h, dx, xmax, Te)
If c1 * c3 < 0 Then
Up = Mi
Else
Lo = Mi: c1 = c3
End If
Next i
BisectionX = Mi
End Function
Applications of ODEs 115

The function f in Code 10.4 uses the temp function in 10.3. The output of f is
the difference between the required temperature (200°C) and the predicted value
at x = 10. Using the function bisectionX, one gets a value of 13.413 so that f = 0.
Using this value as input z0, Code 10.3 will give the answer in Table 10.2 and
Figure 10.2.

TABLE 10.2
Temperature Distribution along a Conductive Rod
A B C
1 Distance (m) Temperature (°C) Gradient (°C/m)
2 0 40 13.41394
3 1 53.41394 13.61394
4 2 67.02788 13.94808
4 3 80.97596 14.41836
5 4 95.39433 15.02812
6 5 110.4224 15.78206
7 6 126.2045 16.68629
8 7 142.8908 17.74833
9 8 160.6391 18.97724
10 9 179.6164 20.38363
11 10 200 21.9798

FIGURE 10.2 Temperature distribution along a conductive rod.

116 Calculations for Process Engineering Using Excel VBA

The exact solution for the temperature is given (Chapra and Canale 2010).

T ( x ) = 73.4523 e 0.1x ­ 53.4523e ­0.1x + 20 Eq. 10.4

The user can compare the numerical solution to this exact solution.

10.4 FURTHER EXAMPLES FOR SELF-PRACTICE

EXAMPLE 10�3� TEST THE REDIM OF AN ARRAY�

Reconsidering Code 10.2, if the value of n in function dabcTdt_2 is calculated
as tmax/delt, what would happen to the function golden_section?

Answer
The program may give an error depending on the value of n, which is decided
by the selected value of dt. For example, if n = 12.51, the array will have
13 rows. Hence, M(n,5) becomes M(13,5). However, the For Next loop stops
at 12.

EXAMPLE 10�4� OPTIMISING A FIRST-ORDER ADIABATIC

BATCH REACTOR, ADAPTED FROM EXAMPLE 9�1
(FOGLER 2006)�
For a first-order adiabatic batch reactor, the rate of change of conversion is
given by:

dX
= k (1 ­ X )
dt
where k is the reaction rate constant, which is dependent on the temperature

é 32, 400 æ 1 1 öù
( )
k = 2.73 ´ 10 ­4 exp ê ç ­ ÷ú
ë 1.987 è 535 T ø û
where reactor temperature, T, is the function of conversion

T = 515 + 90.1 X

Calculate the residence time and reactor temperature for the conversion of
X = 52%.
Applications of ODEs 117

Hints
VBA code is needed to solve the system of three equations. Although the first equa­
tion is an ordinary differential equation, the other two are explicit equations (these
two will be substituted into the ODE). Hence, code for a single ODE can be used.
Furthermore, the code can produce the T and X for plotting (Figure 10.3).
The code can be combined with a bisection method to find time and temperature
when X is 52 %. A bisection code can be used. Hence, the bisection code must eval­
uate X at different values of t. It should be noted that each of these evaluations uses
the same calculation from the beginning, t = 0. Consequently, the scanning method
can be used to reduce the calculations.

FIGURE 10.3 Temperature and conversion as functions of time.

 11 Integration
11.1 INTRODUCTION
Numerical integration is the process of approximating the definite integral of a func­
tion using numerical methods. It is a widely used technique in mathematics, science
and engineering, where analytical solutions may be difficult or impossible.
For a positive function, f(x) > 0, the integration of f(x) from a to b equals the area
under curve f(x) between a and b. This chapter applies the integration to find the area
under a curve. The calculation can be applied to other integration problems.
The most common numerical integration methods are based on approximating
the area under the function’s curve using geometric shapes, such as rectangles, trap­
ezoids and Simpson’s rules. The choice of method depends on the smoothness and
complexity of the function being integrated and the desired level of accuracy.
Amongst these three, the Simpson rule (or Simpson’s 1/3 rule) requires equal step
size. The square method is suitable if the curve is relatively flat. The trapezoid method
is suitable for steep but mostly linear curves. The equation for the trapezoid method is:

h
A= ( f (x1 ) + f (x2 ) ) Eq. 11.1
2
where h is the step size (distance between x1 and x2).
Alternatively, the value of the function at the midpoint can be used:

A = hf (( x1 + x2 ) / 2) Eq. 11.2

Eq. 11.2 has an advantage over Eq. 11.1; it requires only one evaluation. The for­
mula is more accurate than the simple rectangular method, which only requires f(x1).
Simpson’s rule involves approximating the function by a quadratic polynomial
and integrating this polynomial over two neighbouring elements of the integration.
In addition, Simpson’s rule relies on quadratic interpolation over three points of the
curve, thus requiring an even number of equal-distanced elements.
The area of two elements is given by:

h
A=
6
( f ( x1 ) + 4 f ( x1 + h / 2) + f ( x1 + h ) ) Eq. 11.3

The equation is applied to the entire range between a and b.

In addition to the standard Simpson’s rule, there is Simpson’s 3/8 rule. The inte­
gration is based on a cubic interpolation over three elements in this case. The formula
for the 3/8 rule is:

h
A=
8
( f ( x1 ) + 3 f ( x1 + h / 3) + 3 f ( x1 + 2h / 3) + f ( x1 + h ) ) Eq. 11.4

118 DOI: 10.1201/9781003364511-11

Integration 119

11.2 AREA UNDER A CURVE FROM DISCRETE DATA POINTS

The integration can be obtained when the function is not given an explicit form. Such
data are conveniently stored in an Excel spreadsheet.

EXAMPLE 11�1� CALCULATE THE ACCUMULATIVE FLOW RATE�

PROBLEM STATEMENT
The available flow rate data is given in Table 11.1. Calculate the accumulative
volume over the period.

TABLE 11.1
Volumetric Flowrate Data
A B
1 Time (min) Flow (L/min)
2 0 3.5
3 1.8 5.1
4 3.5 5.6
5 6 5.8
6 7 5.9
7 8.2 6.2
8 9.1 6.3
9 10.5 6.1
10 11.6 5.5
11 12 5

Solution
Since the recorded intervals are irregular, we cannot use Simpson’s method. Hence,
the trapezoid method (Eq. 11.1) can be used.

Code 11�1� Integration of area under a curve from discrete points�

Function Area(A)
Dim n, i
n = A.Rows.Count
For i = 2 To n
Area = Area + (-A(i–1, 1) + A(i, 1)) * (A(i–1, 2) +
A(i, 2)) / 2
Next i
End Function
120 Calculations for Process Engineering Using Excel VBA

Applying the code to Table 11.1, one can get a final answer of 7.74 L.

11.3 ADVANCED EXAMPLE: PLUG FLOW REACTOR PERFORMANCE

A plug flow reactor (PFR), or a tubular reactor, is a chemical reactor in which reac­
tants are introduced at one end of a tube, and product(s) are collected at the other end
of the tube. The tube typically has a uniform cross-sectional area. In addition, it is
well mixed in the axial direction, meaning that the composition of the reactants and
products does not vary significantly along the length of the tube.
In a PFR, the concentration of reactants and products at any point along the length of
the tube is determined by the rate of the chemical reaction and the rate of mass transfer
into and out of the tube. A plug flow reactor can be modelled by differential equations
that describe the concentration profile of the species in the reactor as a function of time
and axial position. Such systems can be solved using the methods in Chapters 9 and 10.
For many gas-phase reactions, however, the reactions can generate significant
changes in gas volume. Therefore, the differential methods can be complicated for
such varying-volume or varying-density systems. Alternatively, one can model this
process by using the concept of space-time and the fractional change of volume
(Levenspiel 1999).

EXAMPLE 11�2� MODELLING A REACTOR CONVERSION FROM

SPACE-TIME�
PROBLEM STATEMENT�
This problem is adapted from problem 5.4 (Levenspiel 1999).
Consider a homogenous gas reaction A → 3R. The reaction has a rate at 215°C is:
­rA = 10 ­2 C A 0.5 mol/L s
The feed contains 50% A and 50% inert gas with CAO = 0.0625 mol/L.
Part A� Determine the space-time needed for 80% conversion.
Part B� If the flow rate is increased so that space-time equals 35 s, what is the
conversion of the reactor?

Solution
Part A� Space-time is a mathematical concept for flow reactors and is defined as the
ratio between the feed and reactor volume. For example, a space-time of 30 seconds
means the reactor treats one reactor volume of feed every 30 s.
For a plug flow reactor in this question, the space-time, τ, is related to the reaction
rate and concentration of feed by:

X Af 0.5
C Ao
0.5
æ 1+ e A X A ö
t=
10 ­2 ò
0
ç
è 1­ X A ø
÷ dX A Eq. 11.5
Integration 121

FIGURE 11.1 Plug flow reactor performance curve.

where XA is the conversion and εA is the fractional change in the volume of the sys­
tem. Since the feed contains 50% inert and the reaction triples the gaseous volume,
εA equals 1. Consequently, Eq. 11.5 becomes:

0.5
C 0.5 XA æ 1+ X A ö
t = Ao­2
10 ò
0
ç ÷ dX A
è 1­ X A ø
Eq. 11.6

The function inside the integral can be evaluated and plotted for XA between 0 and
100%, as in Figure 11.1. Consequently, the integral in Eq. 11.5 is represented by the
shaded area. The integral can be obtained by VBA code (Code 11.2).

Code 11�2� Space-time of a plug flow reactor by a trapezoid method�

Function trapez(x1, x2, n)
Dim i, dx
dx = (x2–x1)/n
For i = 0 To n–1
trapez = trapez + fXa(x1 + (i + 0.5) * dx) * dx
122 Calculations for Process Engineering Using Excel VBA

Next i
End Function
Function fXa(Xa)
fXa = ((1 + Xa)/(1–Xa)) ^ 0.5
End Function

In the code, fXa is defined in a separate VBA function. Using the code with n = 8
will return the shaded area as 1.324. Applying Eq. 11.5, τ = 33.1 s. Increasing n will
increase the accuracy of the method.
Part B� In part A, the integration can be obtained numerically for a given Xa.
However, the question requires a value of Xa so that τ = 35s. To get the solution, two
macros are used. The first VBA code is integration, as in Code 11.2. The second one
is applying the Regula Falsi method (Code 11.3). Using these VBA functions, the
conversion for 35s is determined at Xa = 0.824.

Code 11�3� Sizing a plug flow reactor (to be used with the function trapez
in Code 11�2)�
Function PF(Lower, Upper, tau, n)
Dim M, tol, c1, c2, c3, k
c1 = 25 * trapez(0, Lower, n)–tau
c2 = 25 * trapez(0, Upper, n)–tau
M = Lower–c1 * (Upper–Lower)/(c2–c1)
c3 = 25 * trapez(0, M, n)–tau
tol = 10 ^ -7: k = 0
While Abs(c3) > tol And k < 1000
If c1 * c3 < 0 Then
Upper = M: c2 = 25 * trapez(0, Upper, n)–tau
Else
Lower = M: c1 = 25 * trapez(0, Lower, n)–tau
End If
M = Lower + (Upper–Lower) * c1 / (c1–c2)
c3 = 25 * trapez(0, M, n)–tau
k = k + 1
Wend
PF = M
End Function

As in other chapters, the integration can be applied to functions that require com­
plicated numerical evaluations. An example is calculating the total heat produced in
the reactors (Chapter 10).
 12 Network Analysis
12.1 INTRODUCTION
Network analysis problems are a group of problems dealing with nodes and connec­
tions. The problems include the set of nodes and the “weight” factor for each con­
nection between two particular nodes. In geographic terms, the nodes can represent
locations, and weights are the distance. Nowadays, the problems seem too simple
with Google Maps and Apple Maps. Yet network problems can be found in more
complex applications besides finding pathways in a city. For example, the weight
factor can be either time or cost in addition to distance. Of course, this kind of opti­
misation has puzzled many scientists over the centuries. Perhaps the most celebrated
one is Leonard Euler’s “The Seven Bridges of Königsberg”.
While the network is graphically presented in a map form, the network data (con­
nections and weight factors) can be easily stored in an Excel spreadsheet. Conse­
quently, VBA code can be routinely employed.

12.2 CONVERTING A NETWORK TO TABLE FORM

The first step in the network analysis is to convert the 2-dimensional map (such as
Figure 12.1) into a table form (Table 12.1). It is important to note that the nodes should
have a numerical index for analysis. In the sample network, the node names (A, B,
C, . . .) can be converted into numbers. However, for complex texts, such as location
names, assigning a number to each node at the beginning might be more appropriate.

FIGURE 12.1 A network of eight nodes with directional connections.

DOI: 10.1201/9781003364511-12 123

124 Calculations for Process Engineering Using Excel VBA

TABLE 12.1
Distance between Points (the Empty Cells Indicating No Connection)
A B C D E F G H J
1 1 2 3 4 5 6 7 8
2 1 6.6 17.5 2.4 2.9 1
3 2 11.6 0.7
4 3 16.7 4.8 11.5
5 4 4 0.8
6 5 0.7 0.2 4
7 6 3 0.2
8 7 0.5
9 8

Please note that the cells represent one-way travel from row to column. Since all
connections are dimensional, the other direction is zero (for example, from B to A).

12.3 ADVANCED EXAMPLE: SHORTEST PATHWAY PROBLEM

The most common problem is finding the shortest pathway between the two nodes
in a network (Golden 1976). Amongst these methods, Dijkstra’s method is one of
the simplest and most effective. The method creates two sets of nodes: SSet (source
nodes) and FSet (the remaining nodes). Initially, SSet contains a single node, the
starting point (SN). Subsequently, the algorithm includes repetitive iterations, within
which one node is moved from FSet to SSet. Finally, the iteration stops when the
destination node (DN) moves into SSet (Foulds 1981).
Within the iterations, there are several calculation steps:

1 Each node in SSet has a cost value d(i). Initial, SSet contains only SN, and
d(SN) = 0. If node i is not in Sset, Cost(i) = infinity.
2 For each node in FSet, the distances to each node in SSet are obtained.
These values are c(i,j), where i is the node in SSet, and j is the node number
in FSet.
3 The minimum of d(i) + c(i,j) is calculated for node j (MinCost(j)). The cor­
responding node in SSet is recorded as PreNode( j).
4 The minimum is determined from all nodes in FSet. Then, the correspond­
ing node, jmin, is moved from Fset to Sset.
5 The process stops when jmin is DN.

Once the iteration stops, the shortest pathway and corresponding sequence of
nodes are output.
Network Analysis 125

EXAMPLE 12�1� THE SHORTEST PATHWAY BETWEEN TWO

SPECIFIC NODES IN A NETWORK�
PROBLEM STATEMENT�
Find the shortest pathway between points A (column 1) and C (column 3) in
Table 12.1. All connections are two way, and the weights are the same for both
directions.

Solution
First, the input matrix must be checked and filled with two-way connections. The
concept of “infinity” is not available in numerical coding. As a result, a huge but
finite number can be used. The sum of all elements is used to ensure that this number
is larger than all numbers in the matrix (Code 12.1). Here, the Excel function sum
can be used within VBA.
It should be noted that VBA has a conversion function for alphabet and number
(Chr and Asc). However, such a function can be worked within a small range. In
general, one may need another code section to convert the node names into integers.

Code 12�1� Shortest pathway, part 1: setting up matrix M, filling out the
bottom half and all empty cells (no link means infinity distance)�
Option Base 1
Function shortestpath(A, SN, DN)
‘SN is starting node, DS is the destination node
Dim M(), nNode, shortestNNode, N, infinity, i, j, k,
best_j
Dim temp As Variant
Dim Sset(), Fset() As Integer
Dim FCostNode As Variant
Dim MinCost, Prenode, Output, Cost, BestPath As Variant

‘--check square matrix

nNode = A.Rows.Count
If A.Columns.Count <> nNode Then
shortestpath = “Not a square matrix”
Exit Function
End If

infinity = 2 * Application.WorksheetFunction.Sum(A)
ReDim M(1 To nNode, 1 To nNode)
For i = 1 To nNode
For j = i + 1 To nNode
126 Calculations for Process Engineering Using Excel VBA

If A(i, j) > 0 Then

M(i, j) = A(i, j)
Else
M(i, j) = infinity
End If
M(j, i) = M(i, j)
Next j
Next i

The next section of the code defines the nodes in Sset and Fset. In this case, two
arrays are used (SSet and Fset). Initially, all nodes are in Fset, so Fset(i) = i and
Sset(i) = 0. If a particular node is moved from Fset to Sset, the values of the node in
Fset and Sset are interchanged, Sset(i) = i and Fset(i) = 0.

Code 12�2� Shortest pathway, section 2: sizing SSet and FSet and initial
values
ReDim Sset(1 To nNode): ReDim Fset(1 To nNode)
ReDim FCostNode(1 To nNode): ReDim Cost(1 To nNode)
ReDim MinCost(1 To nNode): ReDim Prenode(1 To nNode)
ReDim BestPath(1 To nNode)

For i = 1 To nNode
Fset(i) = i
Cost(i) = infinity
Next i
Sset(SN) = SN: Fset(SN) = 0: Cost(SN) = 0

In the second section of the code, the matrix sizing is required for the For Next
loop. Initially, all nodes are in Fset with a cost value of infinity. Then node SN is
moved to Sset.

Code 12�3� Shortest pathway, part 3: finding the shortest distance�

k = 1: shortestNNode = 1
Do While k < nNode
temp = infinity
For j = 1 To nNode
If Fset(j) > 0 Then ‘only calculate if j is Fset
MinCost(j) = infinity
For i = 1 To nNode
If (Sset(i) > 0) And
(MinCost(j) > (Cost(i) + M(i, j))) Then
Network Analysis 127

MinCost(j) = (Cost(i) +
M(i, j))
Prenode(j) = i
End If
Next i ‘get the shortest cost to
connect j to a node in S- value is stored at
Prenode and MinCOst
If temp > MinCost(j) Then
temp = MinCost(j)
best _ j = j ‘j moves to the
next value, and may not change best _ j
End If
‘reset cost j for next iteration
End If
Next j ‘after j reaches Nnode, best_j is
identified
‘-------move target j to Sset
Sset(best _ j) = best _ j
Fset(best _ j) = 0
Cost(best _ j) = MinCost(best _ j)
For j = 1 To nNode
If Fset(j) > 0 Then
Cost(j) = infinity
End If
Next j
If best _ j = DN Then
k = nNode
Else
k = k + 1
End If
Loop

In the third section of the code, the best node best_j (gives the shortest distance to
SN) is found and moved from Fset to Sset. The step is repeated until the best node is DN.

Code 12�4� Shortest pathway, part 4: finding the node on the shortest
path backward�
‘find nodes on the shortest path
BestPath(nNode) = DN:
k = nNode
Do While k > 0
BestPath(k–1) = Prenode(BestPath(k))
‘Output(2 + nNode–k, 1) = BestPath(k–1)
128 Calculations for Process Engineering Using Excel VBA

If BestPath(k–1) = SN Then
shortestNNode = (nNode–k + 1)
k = 0
Else
k = k–1
End If
Loop

Since the best path from DN to SN is unknown (the path may not include all of
the nodes in SSet), code is needed to find the nodes and sequence of the shortest path
(Code 12.4). Consequently, the output is exported in Code 12.5.

Code 12�5� Shortest pathway, part 5: exporting the shortest pathway (can
use “set cells” to improve visibility)
ReDim Output(1 To shortestNNode + 6, 1 To 1)
Output(1, 1) = “Shortest Distance”
Output(2, 1) = Cost(DN)
Output(3, 1) = “Number of nodes in the shortest path”
Output(4, 1) = shortestNNode
Output(5, 1) = “Nodes on the shortest path from S to D”
Output(6, 1) = SN

For i = 1 To (shortestNNode)
Output(i + 6, 1) = BestPath(nNode–shortestNNode + i)
Next i

shortestpath = Output
End Function

The answers to the problem are given in Table 12.2.

TABLE 12.2
Answers to the Shortest Distance
Shortest Distance
6
Number of nodes in the shortest path
4
Nodes on the shortest path from S to D
1
8
6
3
Network Analysis 129

12.4 ADVANCED EXAMPLE: TRAVELLING-SALESMAN PROBLEM

The mathematical travelling salesman is one of the classical network problems. The
problem is finding the shortest route to pass all nodes (once each) and return to the
starting point (Conley 1988). The critical difference between this and the shortest
path problem is that the pathway has to pass all points on the network. The starting
point is not required. In general cases, this problem may not have a deterministic
answer. The optimisation scheme is complicated. However, we can try Monte Carlo
simulations with randomised numbers. With many trials, the code can get a good
solution.

EXAMPLE 12�2� SHORTEST PATHWAY COVERING ALL NODES

IN A NETWORK�
PROBLEM STATEMENT
An engineer needs to check all listed devices in a processing plant. The time
to move from one piece of equipment to another is listed in Table 12.1. Find a
fast sequence of inspection routines for all devices.

Solution
In the beginning, all nodes are included in SSet. Subsequently, node 1 is added to the
travel path and removed from SSet (SSet = 1). After that, the code randomly adds the
remaining nodes to the travel path. In this case, the code needs to generate a random
number for N – 1 times. Each time, a corresponding node from SSet will be added
to the travel path and removed from SSet. At the end of the trial, the total distance is
compared with distancemin. If the trialled distance is smaller than distancemin, then
the distance becomes distancemin and is used to benchmark the subsequent trial.
Otherwise, distancemin remains the same.

Code 12�6� Random search for a short pathway covering all nodes of a
network�
Function shortesttravel(A, kmax)
Dim M(), nl, sel, distance, distancemin, nNode, k _
sol, N, infinity, i, j, k
Dim Sset(), Pathnode() As Integer
Dim Sol() As Variant
‘------­
‘using Code 12–1- setting up matrix M, filling out
the bottom half and all empty cells
‘-----­
k = 1: distancemin = infinity * nNode
ReDim Sol(1 To nNode + 2)
130 Calculations for Process Engineering Using Excel VBA

While k < kmax

ReDim Pathnode(1 To nNode): ReDim Sset(1 To nNode)
Pathnode(1) = 1: Sset(1) = 1: distance = 0:
nl = nNode:
For j = 2 To nNode
nl = nl–1: sel = Round(Rnd() * (nl–1), 0) + 1:
For i = 1 To nNode
If Sset(i) = 0 And sel > 0 Then
sel = sel–1
End If
If sel = 0 Then
Sset(i) = 1: Pathnode(j) = i: sel = nNode
distance = distance + M(Pathnode(j–1), Pathnode(j))
End If
Next i
Next j
If distancemin >= distance Then
distancemin = distance: k _ sol = k
For i = 1 To nNode
Sol(i) = Pathnode(i)
Next i
End If
k = k + 1
Wend
Sol(nNode + 1) = distancemin: Sol(nNode + 2) = k _ sol
shortesttravel = Sol
End Function

The code results in the sequence of nodes, the obtained shortest distance and the
trial number that produced this answer. In case multiple trials obtain the same answer,
the last trial is shown. In the code, kmax is the maximum number of trials. With a
reasonable kmax (around 20000) and some luck, the code can have the shortest cost
of 18.5 with a corresponding sequence of {1,8,6,4,5,7,2,3}. One can also change the
last line to “shortesttravel = M” to see if a correct matrix M is produced.

12.5 EXAMPLE FOR SELF-PRACTICE

EXAMPLE 12�3� THE SHORTEST PATHWAY COVERS ALL

NODES AND CONTAINS A PARTICULAR CONNECTION�
For a given network,
Network Analysis 131

1 Find the shortest connection between any two points.

2 Find the shortest pathway that contains the previous connection.

Answer
Modify Code 12.6 to:

1 Find the shortest distance and the corresponding two points.

2 Use the starting point and run a Monte Carlo simulation from the
second point.

Note that this solution is not always better than the solution from Code 12.6.
Appendices
List of Useful VBA Functions Which Return a Numerical Value

ABS Returns the absolute value of a number

ASC Returns an integer value representing a character code corresponding to the
function’s first character of a supplied string
ATN Returns the arctangent of a number
COS Returns the cosine of an angle
EXP Returns e raised to the nth power
FIX Returns the integer portion of a number
FormatNumber Returns an expression formatted as a number
INT Returns the integer portion of a number
LOG Returns the natural logarithm of a number
RANDOMIZE Used to change the seed value used by the random number generator for the RND
function
RND Generate a random number between 0 and 1
ROUND Returns a number rounded to a specified number of digits
SGN Returns the sign of a number
SIN Returns the sine of an angle
SQR Returns the square root of a number
TAN Returns the tangent of an angle
MOD Returns the remainder after division operator (Integer)
XOR The bitwise exclusion operator

Other Useful VBA Functions

Array Creates an array, containing a supplied set of values

Filter Returns a subset of a supplied string array based on supplied criteria
ISEMPTY Used to check for blank cells or uninitialised variables
ISERROR Used to check for error values
ISNULL Used to check for a NULL value
ISNUMERIC Used to check for a numeric value
IsArray Tests if a supplied variable is an array
IsMissing Tests if an optional argument to a procedure is missing
IsObject Tests if a supplied variable represents an object variable
Join Joins a number of substrings into a single string

133
134 Appendices

LBound Returns the lowest subscript for a dimension of an array

UBound Returns the highest subscript for a dimension of an array
REDIM Initialises and resizes a dynamic array
BEEP Produces a single beep noise
CALL Transfers control to a subroutine or function
ERASE Reinitialises the elements of an array (numerical variables are set to zero)
GOTO Transfers control to the subroutine indicated by the line label

List of Useful Worksheet Functions (Using the WorkSheetFunction Object in VBA)

Atan Calculates the arc tangent of a given number
Acos Similar to Asin, Atan
Acosh Similar to Asinh, Atanh
Average Returns the average (arithmetic mean) of the arguments
Averageif Calculates the average of the numbers in a range that meet supplied criteria
Cosh Similar to Sinh, Tanh
Count Returns number of numerical values within an array, ignoring text and empty
elements
Degrees Converts radians into degrees
Even Rounds numbers up to the next even integer. The EVEN function always rounds
numbers away from zero, so positive numbers become larger and negative numbers
become smaller (i.e. more negative)
Fact Returns a factorial to a number
ISErr Returns a Boolean value indicating whether an expression is an error value, except
#N/A
IsError Returns a Boolean value indicating whether an expression is an error value (#N/A,
#VALUE!, #REF!, #DIV/0!, #NUM!, #NAME?, or #NULL!)
IsEven Returns TRUE if number is even or FALSE if number is odd. ISEVEN will return
the #VALUE error if a value is not numeric
IsLogical Returns TRUE when a cell contains the logical values TRUE or FALSE, and returns
FALSE for cells that contain any other value, including empty cells
IsNA Returns TRUE when a cell contains the #N/A error and FALSE for any other value
or any other error type
IsNonText Returns TRUE when a cell contains any value except text. This includes numbers,
dates, times, errors and formulas that do not return text
IsNumber Returns TRUE when a cell contains a number and FALSE if not
IsOdd Returns TRUE when a value is an odd number and FALSE when a value is an even
number. ISODD will return the #VALUE error if a value is not numeric
IsRef Returns TRUE when a cell contains a reference and FALSE if not. You can use the
ISREF function to check for a reference in a formula
Ln Returns the natural logarithm of a number
Appendices 135

Log Returns the logarithm of a number to the specified base

Log10 Returns the base-10 logarithm of a number
LogEst Calculates an exponential curve that fits your data and returns an array of values that
describes the curve
Max Returns the largest numeric value in the data provided
Min Returns the smallest numeric value in the data provided
Median Returns the median (middle number) in the supplied set of data
Mround Rounds a given number up or down to the specified multiple. (See Example 4.7.)
Odd Returns the next odd integer after rounding a given number up. The ODD function
always rounds numbers away from zero, so positive numbers become larger and
negative numbers become smaller (i.e. more negative)
Pi() Returns value of Pi
Power Returns a number raised to a given power
RandBetween Returns a random integer number between the two specified numbers
Radians Converts degrees to radians
Round Rounds a number to a specified number of digits. (See Code 4.1.)
RoundDown Rounds a number down to a specified number of digits
RoundUp Rounds a number up to a specified number of digits
Slope Returns the slope of a regression line based on known y and x values
SQRT Returns the square root of a positive number. SQRT returns an error if number is
negative
Sum Sums all elements of an array
References
Chabert, J.-L., 1999. Methods of false position, in: Chabert, J.-L. (Ed.), A History of Algorithms.
Springer, Berlin, Heidelberg, pp. 83–112. https://doi.org/10.1007/978-3-642-18192-4_4
Chapra, S.C., Canale, R.P., 2010. Numerical Methods for Engineers, 6th ed. McGraw-Hill
Higher Education, Boston, MA.
Conley, W., 1988. Travelling salesman problem solved with simulation techniques. International
Journal of Systems Science 19, 2115–2122.https://doi.org/10.1080/00207728808964103
Felder, R.M., Rousseau, R.W., 2005. Elementary Principles of Chemical Processes, 3rd ed.
Wiley, Hoboken, NJ.
Fogler, S., 2006. Elements of Chemical Reaction Engineering, 4th ed. Pearson Education,
Upper Saddle River, NJ.
Foulds, L.R., 1981. A multi-commodity flow network design problem. Transportation Research
Part B 15, 273–283. https://doi.org/10.1016/0191-2615(81)90013-8
Golden, B., 1976. Technical note—shortest-path algorithms: A comparison. Operations
Research 24, 1164–1168. https://doi.org/10.1287/opre.24.6.1164
Herráez, J.V., Belda, R., 2006. Refractive indices, densities and excess molar volumes of
monoalcohols + water. Journal of Solution Chemistry 35, 1315–1328. https://doi.
org/10.1007/s10953-006-9059-4
Himmelblau, D.M., Riggs, J.B., 2012. Basic Principles and Calculations in Chemical Engi­
neering, 9th ed. International Series in the Physical and Chemical Engineering Sci­
ences. Pearson, Boston, MA.
Ingram, G., Zhang, T., 2010. An introduction to the harmony search algorithm. Advances in
Process Systems Engineering, 301–335. https://doi.org/10.1142/9789814299213_0009
Le, T.N., Phan, C.M., Nguyen, A.V., Ang, H.M., 2012. An unusual synergistic adsorption of
MIBC and CTAB mixtures at the air-water interface. Minerals Engineering 39. https://
doi.org/10.1016/j.mineng.2012.06.003
Levenspiel, O., 1999. Chemical Reaction Engineering, 3rd ed. Wiley, New York.
Lindfield, G.R., Penny, J.E.T., 2019. Numerical Methods: Using MATLAB, 4th ed. Elsevier,
London.
Nawrocki, D., 2013. VBA Programming Techniques in Operations Research. 26th Interna­
tional Conference on Technology in Collegiate Mathematics VBA, pp. 255–263.
Pillai, H.K., Bandyopadhyay, S., 2007. A rigorous targeting algorithm for resource alloca­
tion networks. Chemical Engineering Science 62, 6212–6221. https://doi.org/10.1016/j.
ces.2007.07.031
Press, W.H., Teukolsky, S.A., 1992. Adaptive stepsize Runge–Kutta integration. Journal of
Computational Physics 6, 188. https://doi.org/10.1063/1.4823060
Serth, R.W., Lestina, T.G., 2014. Process Heat Transfer—Principles, Applications and Rules
of Thumb, 2nd ed. Elsevier, Oxford.
Shashi Menon, E., 2015. Fluid flow in pipes, in: Transmission Pipeline Calculations and Simu­
lations Manual. Elsevier, Oxford, pp. 149–234. https://doi.org/10.1016/B978-1-85617­
830-3.00005-5
Shibata, Y., Hyde, A., Asakuma, Y., Phan, C., 2018. Thermal response of a non-ionic surfac­
tant layer at the water/oil interface during microwave heating. Colloids and Surfaces
A: Physicochemical and Engineering Aspects 556, 127–133. https://doi.org/10.1016/j.
colsurfa.2018.08.010
White, D. 2002. ChE 224 Process Analysis Study Guide. Curtin University, Bentley.

137
Index
Note: Page numbers in italics indicate a figure and page numbers in bold indicate a table on the
corresponding page.
3/8 rule, 119 binary mixture (composition determination),
refractive index (usage), 41
A binary search, 44, 108
bisection method, combination, 48–51
AcC, determination, 57 golden section search, comparison, 51
accumulative flow rate, calculation, 119–120 output change, 52
acetylene production, recycle stream usage, 50
unknowns, presence, 89 Bisection2, 51
usage, 35 bisection code, 26
acosh (spreadsheet function), 15 flowchart, 23, 24
AcR, determination, 57 function value limitation, 25
active spreadsheet, discrete data (storage), 39–40 iterations, number (determination), 30
adaptive_E, 104 removal, 25
adaptive step size, 104 usage, 24, 117
usage, 101–106 bisection method, 20–22, 34, 43
adiabatic flame temperature binary search, combination, 48–51
control, air excess (calculation), 34 Do While loop, usage, 21
determination, example, 31–34 efficiencies, comparison, 30
equation roots, application, 31–34 failures, 26–27
air excess, calculation, 34 false-position method, combination, 40
algorithm, 91 For . . . Next loop, usage, 22
Gauss-Seidel algorithm, coefficient matrix Jacobi method, combination, 75–77
(validation), 74 scanning, usage, 25–26
simplex algorithm, 54–59 usage, 49, 114
anticycling rule (Bland), 56 variable, control error (presence), 22
Apple Maps, usage, 123 bisection/tear analysis, combination, 20
Application.WorksheetFunction.Log, 13–14 BisectionT function, 33
aqueous solution (cooling), microwave heating bisectionX function, usage, 115
(relationship), 80–84 boundary condition, 83
array boundary nodes, 78
definition, 6 boundary value ODEs, 113–116
one-dimensional array, output usage, 41 boundary value problem (BVP), 113
redim, test, 116 branching, incorporation, 4
values, resetting, 7 breakpoints, 17–18
variables, 7
asinh (spreadsheet function), 15 C
atanh (spreadsheet function), 15
ave_4 function, 13 calculated error, 11
average (spreadsheet function), 14 change-of-sign interval, 27, 29
Change spreadsheet cell value, 12–13
B classical fourth-order RK method, 101
ClearContents method, 3
basic variables, 55 coefficient matrix, validation, 74
values, equivalence, 59 cold streams, process, 59
behaviour function, 101–102 Colebrook equation, 27–28, 48–49
best_j (node), 127 friction factor, determination, 23–25
binary adsorption, 86 column.count function, 74
parameters, 87 component flow rates, 36

139
140 Index

composition (determination), refractive index Do While . . . loop, example, 11

(usage), 41 Do While loop, usage, 21
conductive rod, temperature distribution, 115, Do . . . While structure, 8
115 dynamic arrays, 6–7
constants, 75
constant step size, usage, 97–100 E
continuous stirred-tank reactors (CSTRs), series
conversion constants, 67 elimination
manipulation, 75–77 efficiency, reduction, 44
reactor size, 67 forward elimination, 65
recycle streams, presence, 67 Gaussian elimination, 65
simulation, 66–70 inefficiency, 46
control error, presence, 22 options, 43
convergence, enhancement, 70 region elimination, 43–46
conversion empty cells, connection (absence), 124
constants, 67 endless loop, coding problem, 16
function, 117 energy
correlation, sensitivity, 77 consumption (minimisation), heat exchange
cosh (spreadsheet function), 15 network (determination), 59–63
Crank-Nicholson method, 80 recovery, maximisation, 59
Ctrl+G, usage, 17 requirements, meeting, 61
Ctrl+Shift+Enter, 7 engineering, Excel tools, 18–19
Ctrl+Shift+F9, 18 enthalpy, change, 31
curve (area) equal-distanced elements, 118
discrete data points, usage, 119–120 equation, roots, 20
integration, discrete points (usage), 119 applications, 31
cycling, prevention, 56 Erase, usage, 7
ethyl chloride (CH3CH2Cl), 34–35, 88
D Euler, Leonard, 123
Euler’s method
dabcTdt_2, 111, 116 accuracy, improvement, 104
dablTdt_2, 111 accuracy, step size (impact), 99
Debug.Print, impact, 17–18 adaptive step size, relationship, 103
Debug, usage, 17 application, 105
delx (step size), 86 comparison, 101, 102
dely (step size), 86 constant size selection, assumption, 106
derivative, finite-difference approximation, 27 modification, 102, 104
destination node (DN), 124, 128 predictions, 105–106
diagonal coefficients, 65 problems, 98
Dijkstra method, 124 step size, change, 105
Dim (statement), 7 usage, 98–99, 102–106
dimensionless parameter, 81 accuracy, step size (impact), 99
dimensions, declaration, 56 Euler’s prediction, 97
directional connections, 125 Excel. see Microsoft Excel
discrete data exhaust gas quantities, determination, 31
points, usage, 119–120 Exit Loop, usage, 21–22
storage, 39–40
distancemin, comparison, 129 F
divergence, avoidance, 27
Do Loop formats, 10 F9 (breakpoint), 17–18
Do loop structure, 10 false position, 28
Do . . . Loop structures, 10–11 false-position method, bisection method
Do . . . Loop Until, example, 11 (combination), 40
double1 function, 3 ff (function), 98
double2 function, 3 ffT (VBA function), 32
double3 function, usage, 4 Fibonacci search, 44, 51–52
double4 function, usage, 4 finite-difference approximation, 27
Index 141

first-order adiabatic batch reactor, optimisation, golden_section (function), 116

116 golden section search, 44, 48
first-order method, second-order method binary search, comparison, 51
(combination), 102 code, 45–46
flow conditions, 75 ODEs, combination, 112
flow rate subfunction, 111
accumulative flow rate, calculation, 119–120 usage, 45
data, 119 Google Maps, usage, 123
parameters, 109 gradient-based methods, 28
volumetric flowrate data, 119 guessed/calculated tear stream value, difference
flow variables, 60 (calculation), 36
fluctuation, avoidance, 27
For Each Next H
loop, example, 10
option, 8 Harmony Memory Considering Rate (HMCR),
For Next 92, 94
loop, usage, 16, 113 harmony search (HMS)
option, 8 application, 92
structure, 8 improvement, absence, 96
For . . . Next loop, 29 independent variables, usage, 91–95
bisection code flowchart, 23 output, 95, 96
bisection code, usage, 24 parameters, adjustment, 96
structure, 8, 10 variables, usage, 92–95
unsuitability, 25 heat capacity constants, 32
usage, 22 heat data, 60–62
For . . . Next structure, usage, 22 heat diffusion
forward elimination, 65 discretisation, 81
fourth-order methods, 100 equation, 80
fourth-order Runge-Kutta (RK) method heat exchange network
classical fourth-order RK method, 101 determination, 59–63
comparison, 101 heat data, 60–62
friction factor, determination, 23–25 heating/cooling, utility streams execution,
friction loss 60
factor, minimisation, 48–50 pinch analysis, 63
finding, bisection method (usage), 49 heat exchanger network, pinch analysis, 59–63
function, 50 heat of formation, 32
FSet node, distances, 124 heat release, management, 108
FSet sizing, 126 heat transfer, 20, 113
function procedures (VBA), 3–5 hot streams, process, 59
function value, limitation, 25 hydrogen chloride (HCl), 34–35, 88
fXa, defining, 122
I
G
If . . . Elseif structure, 21–22
Gaussian elimination, 65 immediate value, displacement, 17
Gauss-Jordan elimination, 65 immediate window (screenshot), 17
Gauss-Seidel algorithm, coefficient matrix improved polygon comparison, 101
(validation), 74 independent unknowns, tear analysis, 88–91
Gauss-Seidel code, modification, 76–77, 79 independent variables
Gauss-Seidel method, 65–69, 75, 78 tear analysis, 88–91
basis, 68 usage, 91–95
calculation time, reduction, 70–71 infinity distance, 125–126
relaxation, impact, 71 input validation, 74–75
Gauss-Seidel technique, 70 insulated boundary condition, 81
Get spreadsheet cell value, 12–13 integration, 118
Goal Seek, 18 internal energy, impact, 37–40
screenshot, 19 internal nodes
142 Index

heat diffusion, discretisation, 81 bisection method, usage, 114

temperature, 79 initial conditions, 114
interpolation formula, usage, 39 looping, incorporation, 4
IsError, 15
IsNA, 15 M
IsNull, 15
IsNumber, 15 material balances, recycle stream usage (tear
iterations analysis), 34–37
maximum limit, 29 matrix
maximum number, usage, 16–17 output, 8
number, determination, 30 size, 6
number, increase, 44 square, validation, 74
table, samples, 58, 59 max (spreadsheet function), 14
iterative loops, types, 8 maximum values, finding, 93
MaxIte, 96
J mean molar area, 86
memory consideration, 91
Jacobian matrix, 85 Message Box, 4
Jacobi method, 65–66, 69–70 methane (CH4), adiabatic flame temperature
bisection method, combination, 75–77 (determination), 31–34
iterations, number (minimisation), 72 Microsoft Excel, 1–5
solution, optimisation, 72–73 built-in function, 3
updating loop, separation, 69 objects, manipulation, 2–3
usage, 67 tools, 18–19
Jacobi technique, 70 Microsoft Excel spreadsheet
cells, manipulation, 3
K change cell value, 12–13
functions, 13–15
kinetic viscosity, 49 get cell value, 12–13
kmax (code), 130 matrix, output, 8
multiple values, returning, 7–8
L periodic table (short version), 9
set cell value, 12–13
LEGACY ARRAY formula, 7 subroutine, usage, 5
linear equations, systems VBA functions, usage errors, 15
applications, 74 microwave heating
inputs, 68 one-dimensional heat transfer, relationship,
roots, 65 81–82
linear function, maximisation, 54–59 unsteady-state heat transfer, 80–84
real number, usage, 63–64 min (spreadsheet function), 14
linear objective function, 54 MinCost(j), 124
linear programming, 54 minimisation
AcC, determination, 57 friction loss factor, minimisation,
AcR, determination, 57 48–50
algorithms, 54 golden section search, usage, 45
dimensions, declaration, 56 problem, 43
pivoting, 57–58 quadratic interpolation, usage,
linear regression, 29 46–47
Ln (spreadsheet function), 14 minimum values, finding, 93
local error, 99 molecular weight (calculation), For Each Next
local truncation error, 99 loops (examination), 10
Log (spreadsheet function), 14 Monte Carlo simulation, 129, 131
Log10 (spreadsheet function), 14 MROUND function, 52–53
logarithms, examples, 15 Msgbox, usage, 17
logic, incorporation, 4 multiple-dimensional problems, solution
long rod, temperature contribution (challenges), 54
(solving) m x m Jacobian matrix, 85
Index 143

N single first-order ODE, Euler method (usage),

98–99
net energy saving, achievement, 60 solving, 97, 108
network system, 106
analysis, 123 usage, 110–111
nodes, pathway (shortest distance), 125–128, output(), 7
129–130 output
table form conversion, 123–124 change, 52
Newton’s method, 27–28, 85–88 matrix, 40
nodes required increment, change, 52
coverage, shortest pathway (usage), 130–131 overrelaxation, 70
nodes, finding, 127–128
nodes network, directional connections, 125
non-basic variables, 55
P
entry, 57 parameter values, 25, 109
non-linear equation, solving partial pivoting, 74–75
example, 20–21 Pi() (spreadsheet function), 14
Secant method, usage, 86–88 pinch analysis, 59–63
non-linear equation systems answers, 63
Secant method, usage, 87–88 benefits, 59
solving, 85 pipe diameter function (friction loss
tear analysis, combination, 90 determination), bisection method (usage),
non-positive values, 59 49
non-zero coefficients, 61 Pitch Adjusting Rate (PAR), 92, 94
non-zero minimum, 27 pitch adjustment, 91
number, average value (calculation), 13 pivoting, 57–58
number doubling partial pivoting, 74–75
function, 4 plug flow reactor (PFR)
subroutine, 4 design, 104
numerical data types, values (range), 6 sizing, 122
numerical integration, 118 space-time, trapezoid method (usage),
numerical methods, usage, 77 121–122
plug flow reactor (PFR) performance, 120–122
O curve, 121
points, distance, 124
objective function, usage, 24, 25 polygon method, improvement, 99, 100
one-dimensional array, output usage, 41 PreNode(j), 124
one-dimensional heat distribution, shooting profit maximisation, 64
method (usage), 113–116
one-dimensional heat transfer, microwave
Q
heating
relationship, 81–82 quadratic equation, 46
temperature, production, 83 quadratic function, 46
one-dimensional optimisation, 43 quadratic interpolation, 46–48, 118
one-step methods usage, 46–47
adaptive step size, usage, 101–106 quadratic method, minimisation, 47
constant step size, usage, 97–100 quadratic polynomial, integration, 118
optimisation
Gauss-Seidel code, modification, 76–77
R
scanning, combination, 47–48
Option Base, 11 random function (RAND), usage, 91
Option Explicit, 11 randomisation, 91
ordinary differential equation (ODE) RANDOMIZE function, 92, 95
applications, 108 Range object, 110
approximation, 97–100 properties, 2
code, 111 reactions, parameters, 109
golden section search, combination, 112 reactor. see semi-batch reactor
144 Index

conversion, space-time modelling, 120–122 application, 51

first-order adiabatic batch reactor, usage, 117
optimisation, 116 Secant gradient-based method, 28
plug flow reactor (PFR) performance, Secant method, 27–28, 85–88
120–122 impact, 28
series. see continuous stirred-tank reactors. usage, 48, 86–88, 95
size, 67, 75 VBA function, 28
real number, usage, 63–64 second-order methods, 99
recycle stream adaptive step size, relationship, 103
presence, 67 first-order method, combination, 102
tear analysis, 34–37 semi-batch process, 108
unknowns, presence, 89 semi-batch reactor
usage, 35 concentration (calculation), ODEs (usage),
Redim (statement), 7, 113 110–111
ReDim Preserve statement, 7 concentrations, 111
#REF! error, 15 example, 108–113
refractive index, 42 optimisation, 108–113
usage, 41 optimisation, golden section search/ODEs
region elimination, 43–46 (combination), 112
Regula Falsi, gradient-based method, 28 reactions/flow rate/temperature/conditions,
Regula Falsi method, 28–29, 43 parameters, 109
application, 122 temperature, 111
efficiencies, comparison, 30 temperature profiles (calculation), ODEs
iterations, maximum limit, 29 (usage), 110–111
relaxation set cells, usage, 128
factor, iterations number (relationship), 72 Set spreadsheet cell value, 12–13
impact, 71 “Seven Bridges of Königsberg, The” (Euler), 123
introduction, 70 shooting method, usage, 113–116
relaxation constant shortest distance, answers, 128
optimisation, 72 shortest path, node (finding), 127–128
usage, 70–71 shortest pathway
Reynolds numbers, 23, 48 exporting, 128
roots, applications, 20, 31 FSet sizing, 126
round, 14, 113 matrix setup, 125–126
rounddown, 14 network nodes coverage, 129–130
round-off errors, 65 node, 125–128
roundup, 14 node coverage, 130–131
Round, usage (avoidance), 46 problem, 124–128
rows.count function, 74 shortest distance, finding, 126–127
Runge-Kutta (RK) Fehlberg method, 102 SSet sizing, 126
Runge-Kutta (RK) methods, 97 short pathway, random search, 129–130
adaptive step size, usage, 102–106, 105 simplex algorithm, 54–59
fourth-order Runge-Kutta method, simplex method, 55
comparison, 101 simplex problem, input, 55
variations, 100 Simpson’s method, usage (impossibility), 119
Simpson’s rule, 118
S single first-order ODE
Euler method, usage, 98–99
saturated vapour, steam table, 38 polygon method, improvement, 100
scanning single function evaluation, bisection code (usage),
code, 26 24
macro, development, 47 SkipPoint, usage, 21–22
optimisation, combination, 47–48 slack variables, 55, 61
procedure, 26 introduction, 64
step, doubling, 48 solutions, initial sets (generation), 92–93
usage, 25–26, 47–48 Solver (screenshot), 18
scanning method, 25 space-time
Index 145

concept, usage, 120 solving, 36–37

feed concentration, relationship, 120 stochastic optimisation, usage, 91–95
modelling, 120–122 usage, 34–37, 92
reaction rate, relationship, 120 tear stream value, difference (calculation), 36
specific energy, steam quality (calculation), temperature
37–40 contribution, solving, 114
specific volume, impact, 37–40 distribution, 115, 115
SPILLED ARRAY formula, 7 functions, 117
#SPILL! error, 7 parameters, 109
spreadsheet. see Microsoft Excel spreadsheet profiles, 83
SSet node, 127–129 termination condition, 27
cost value, 124 text input division, avoidance, 16
SSet sizing, 126 theta, usage, 88
starting point (SN), 124, 128 time
static arrays, 6–7 space-time, 120–122
steady-state conditions, 35 temperature/conversion functions, 117
steady-state temperature trapezoid method
distribution, 79 equation, 118, 119
profile, 77–79 suitability, 118
steam usage, 121–122
table, 38 travelling-salesman problem, 129–130
temperature, calculation, 39–40 trial-and-error approach, 31
steam quality trigonometric functions, usage, 15
calculation, 37–40 two-dimensional heating plate, discretisation, 78
solution, alternate, 41 two-dimensional steady-state heat problems, 78
step size two-dimensional steady-state heat transfer,
change, 105, 106 77–79
constant step size, usage, 97–100 Gauss-Seidel code, modification, 79
impact, 99
justification, 106 U
stirred-tank reactors, series (simulation), 66–70
stochastic optimisation, usage, 91–95 under-relaxation, 70
Stream S-4 (Stream S4), 35, 88 unsteady-state heat transfer, 80–84
sub procedures (VBA), 3–5 updating loop, separation, 69
subroutine, usage, 5 user-specified tolerance (errd ), 102
substitution methods, 65–70 utility energy, minimisation, 60
relaxation constant, usage, 70–71 utility streams, execution, 60
sum (spreadsheet function), 14, 125
sum of squares of differences, calculation V
function, 92
surface area fractions, 86 #VALUE, 28, 41
surface temperature, 84 error, 15
surfactants return, 8
binary adsorption, parameters, 87 vapour fraction
co-adsorption, 86 calculation, 39–40
mixture, 86 determination, internal energy/specific
systems, roots, 65 volume (impact), 37–40
variables
T control error, presence, 22
independent variables, tear analysis, 88–91
tanh (spreadsheet function), 15 non-basic variable, entry, 57
tear analysis, 88–91 usage, 5–6, 92–93
basis, 35 values, assigning, 3
bisection combination, 20 Variant (variable default type), 5–6
calculations, reduction, 96 velocity, friction loss function, 50
non-linear equations system, combination, 90 visibility (improvement), set cells (usage), 128
primary functions, 89–90 Visual Basic for Applications (VBA), 1–2
146 Index

function procedures, 3–5 volumetric flowrate, 49

function scanning, 48 data, 119
functions, list, 133–134
iterative loops, types, 8 W
logarithms, examples, 15
number doubling function, 4 water-ethanol mixture, refractive index, 42
number doubling subroutines, 4 water surface, temperature profiles, 83
Option Base, 11 While loops
Option Explicit, 11 avoidance, 16
options (screenshot), 12 iterations, maximum number (example),
procedures, errors (avoidance), 16–17
15–18 While . . . Wend
programming, characteristics, 5–15 loop, bisection code flowchart, 24
sub procedures, 3–5 structure, 23
usage errors, 15 Worksheet collection, 2
variables, usage, 5–6 WorksheetFunction.LN, 14
WorkSheetFunction Object, usage, WorkSheetFunction Object, usage, 134–135
134–135 WorksheetFunction qualifier, 13
Visual Basic for Applications (VBA) code worksheet functions, list, 134–135
development, 68
Gauss-Seidel method basis, 68 Z
options, 11
Visual Basic for Editor (VBE), 3 zero coefficients, 59
immediate window (screenshot), 17 zero division, avoidance, 16