FUNDAMENTALS OF MULTICOMPONENT DISTILLATION Charles D. Holland Texas A&M University McGraw-Hill Book CompanyChapter 1 CONTENTS Preface Introduction to the Fundamentals of Distillation Fundamental Principles Involved in Distillation Separation of Multicomponent Mixtures by Use of One Equilibrium Stage “Multiple Stage Separation of Binary Mixtures Separation of Multicomponent Mixtures at Total Reflux Notation Problems References Development and Application of the Theta Method of Convergence to Conventional Distillation Columns Equations Required to Deseribe Conventional Distillation Columas Formulation and Application ofthe @ Method of Convergence, the K, Method, andthe Constant- ‘Composition Method Convergence Characteristics ofthe 0 Method of Convergence, the K, Method, nd the Constant-Composition Method Other ropis: Pressure Efects, Nonidea! Solutions, and (Other Specifications Notation Problems ReferencesChapter 3 Chapter 4 4 42 43 44 Chapter 5 st Chapter 6 61 62 6a Application of the Theta Method of Convergence to Complex Columns and to Systems of Columns Complex Disillation Columns Systems of lteeonnected Distillation Columns A Generalzed Formulation fora Sytem of Columas in Which ve Tot [Ae Specified Notation Problems References The 2N Newton-Raphson Method Formulation of the 2’ Newton-Raphson Method for Single Columns [Numerical Methods for Solving the 2 Newton-Raphson Equations Systems of Columns Comparison of Procedures 1,2, 3, and the Cohumn Modular and System Modiulae Methods Problems References Almost Band Algorithms of the Newton- Raphson Method Almost Band Algorithms for Absorbers and Strippers, Independent Variables {yy} and (7) ‘Almost Band Algorithms for Conventional and Complex Disillaton Columns Modifications of Broyden's Method The Boston Sullivan Algorithm Problems References Appendix $1 Development ofthe Broyden-Householder ‘Algorithm for Sparce Matrices ‘Systems of Azeotropic and Extractive Distillation Columns ive Characteristics of Azcotropic and Extractive Processes Solution of Problems Involving Single Columns Used to et Atcouoic and Extractive Disiltions sms of Columns in the Service of Sept Nonidea Sutona Notation Problems References 87 87 101 m2 us 116 0 ni m ur 156 m 175 176, 18 180 185 195 210 2 a 216 216 2 24 246Chapter 7 nM 72 B 14 Chapter 8 ot 92 93 Chapter 10 to 102 Solution of Systems of Columns with Energy Exchange Between Recycle Streams Columns Having a Common Condenser and Reboiler EEnoray Exchange Between Recycle Streams of Systems of Distillation Columns Energy Transer by Use of rim Heat Exchangers ‘Other Applications ofthe Theta Methods Notation Problems References Distillation Accompanied by Chemical Reaction ‘Application ofthe Theta Method of Convergence 10 Distillation Columns in which Chemical Reactions Occur Formulation of the N{r + 2) Newlon-Raphson Method Formulatien of an Almost Band Algorithm Notation Problems References Optimum Design and Operation of Conventional and Complex Distillation Columns Procedure 1. Determination ofthe Minimum Number of Stages Required to Eflet Specified Separation ata Given Reflux Ratio for Conventional and Complex Distillation Columns Procedure 2. Optimum Economic Design of Conventional and Comp:x Distillation Columns Procedure 3. Minimization of the Reflux Ratio for Existing Conventional and Complex Distillation Columns Notation Problems References Appendix 1 Approximate Solutions at Total Reflux ‘Appendix $-2 Approximate Solutions at Operating Reflxes ‘Appendix 83 A’Modiied Form ofthe Complex Method of Box for Otjective Functions Having No Constraints Appendix 4 Formulation ofthe Eeonomic Objective Function Characteristics of Continuous Distillation Columns at Various Modes of Operation ‘Continuous Distition Columns at Total Reflux, Type 1 (Lj¥joy =1, F =D + B,D 40, B40) ‘Continuous and Batch Distillation Columns at Total Reflux in Both Sections, Type 2: (D = 0, 8=0, F ~0) 28 249 263 268 m m m4 as 29 302 as x20 325 26 327 328 33 335 no 340 30103 Chapter 11 ne 2 Chapter 12 4 2 Chapter 13 nt 2 83 Chapter 14 Mt 142 Operating Characteristics of Continuous Distillation Columas over Wide Ranges of Distillate and Reflux Rates, Notation Problems References Conventional and Complex Distillation Columns at Minimum Reflux Determination of Produet Distributions for Columns with Infinitely Many Stages by Use of the @ Method Determination of the Minimum Reflux Ratio by the 0 Method [A Newion-Raphson Formulation: Two Independent Variables Per Stage ‘An Almost Band Algorithm Formulation of the Newton Raphson Method Proof ofthe Existence ofthe Reetilying and Stripping Section Pinches Notation Problems References Design of Sieve and Valve Trays Single-Pass Sieve Trays with Crossflow Valve Trays Notation Problems References Application of the Fundamentals of Mass Transfer to Plate Efficiencies and Packed ‘Columns Fundamental Relationships for Mass Transfer between Vapor and Liquid Phases Plate Eficienies Packed Distillation Columns Notation Problems References Thermodynamic Relationships for Multicomponent Mixtures Fundamental Priniples of Thermodynamics Needed in the Calculation of Vapor-Liquid Equilibria and Enthalpes of Multicomponent Mixtures Equations of State and Their Use in the Prediction of K Values and Enthalpies ass 36 302 364 365 365 a2 393 or 410 43 ais a7 432 49 4a aa 45, 403 454 470 486 49 9 a2 493 suChapter 15 Ist 132 133 Use of Multiple Equations of State forthe Caleulation of K Values Estimation ofthe Fugacity Coefcients for the Vapor Phase of Polar Gas Mixtures Calculation of Liquid-Phase Activity Coefcints Notation Problems References Selected Topics in Matrix Operations and Numerical Methods for Solving Multivariable Problems Storage of Large Sparse Matrices Sealing of Mateices Numerical Methods fr Solving Multivariable Problems Problems References Appendixes Matrix Norms and Theorems Equilibrium and Enthalpy Data Indexes ‘Author Index ‘Subject Index 528 sat 50 558 539PREFACE This book constitutes ar in-depth treatment of the subject of multicomponent distillation. It begins with first principles and goes to the frontiers of the subject. Each topic is introduced in an elementary and fundamental manner which makes the book suitable for the undergraduate student, the graduate student, and the practicing engineer. The subject matter is presented in the order of increasing “ifficulty and complexity. ‘The gap between the treatment of binary and multicomponent mixtures is closed in Chap. 1. This chapter is initiated by presenting the fundamental relationships and techniques needed for making bubble-point and dew-point calculations, and it is concluded by the presentation of techniques for solving 2 variety of special types of problems such as the separation of a multicomponent mixture by a single-stage flash process and the separation of a multicomponent mixture by use of multiple stages at the operating condition of total reflux. In Chaps. 2 through 5, the theta methods and variations of the Newton- Raphson method are applied to all types of single columns and systems of ‘columns in the service of separating both ideal and nonideal solutions. Applica- tions of the techniques presented in Chaps. 2 through 5 to systems of azeotropic and extractive distillation columns are presented in Chap. 6. An extension of these same techniques as required for the solution of problems involving energy exchange between recycle streams is presented in Chap. 7. Special types of separations wherein the distillation process is accompanied by chemical reactions are treated in Chap. 8 In Chap. 9, all ofthe techniques developed in Chaps. 1 through 8 are brought to bear in the design and operation of conventional and complex distillation ‘columns. To complete the in-depth treatment of multicomponent distillation, the special topics of total reflux, minimum reflux, design of valve and sieve trays, plate efficiencies, design of packed columns, thermodynamic relationships, and selected numerical methods ate presented in Chaps. 10 through 15. A Sulu tions Manual may be obtained (without cost) by Faculty members by writing directly to me or to McGraw-Hill,Since the writings of any author are influenced by all that he has met, 1 am deeply indebted to all of my teachers, fellow faculty members, students, and the many past and present workers in the field of distillation, In particular, I wish to thank Professors P. T. Eubank and D. T. Hanson of the Department of Chemical Engineering as well as Professors Emeriti K, C. Klipple and H. A. Luther of the Department of Mathematics for their many helpful suggestions. The direction and advice generously offered by N. J. Tetlow and Ron McDaniel of Dow Chemical Company is appreciated immersely. I also want to thank Dr. Tetlow for his contributions to Chap. 14 (Thermodynamic Relationships for Multicomponent Mixtures) as coauthor. The helpful suggestions provided by W. L. Bolles of Monsantoare very much appreciated. For the support ofthe research (upon which ‘much of this book is based) by David L. Rooke, Donald A. Rikard, and Holmes H.MeClure (allof Dow ChemicalCompany), lam most thankful. Also, support of research provided by the Texas Ergineering Experiment Station and the Center for Energy and Mineral Resources is appreciated. To both former and present graduate students (Najeh S. Al-Haj-Ali, G, W. Bentzen, Andy Feng, S, E. Gallun, Alejandro Gomez M. J. R. Haas, F. E. Hess, Alicia Izarraraz, P. E. Mommessin, and G. P. Pendon) who participated inthis research, [salute you for your many contributions and I shall always be indebted to you. Finally, I want to pay special tribute to my ‘Staff Assistant, Mrs. Wanda Greer, who helped make this book possible through her loyal service and effective asistance over the past eight years in the discharge of my administrative responsibilities Charles D. HollandCHAPTER ONE INTRODUCTION TO. THE FUNDAMENTALS OF DISTILLATION {In this chapter, the fundamental principles and relationships involved in making ‘multicomponent distillation calculations are developed from first principles. To enhance the visualization of the application of the fundamental principles to this, separation process, a variety of special cases are considered which include the ‘determination of bubble-point and dew-point temperatures, single-stage flash sep- rations, multiple-stage separation of binary mixtures, and multiple-stage sepa- ration of multicomponent mixtures at the operating conditions of total reflux. ‘The general objective of distillation is the separation of compounds that have different vapor pressures at any given temperature. The word distillation as used herein refers to the physical separation of a mixture into two or more fractions that have different boiling points. + Ifa liquid mixture of two volatile materials is heated, the vapor that comes ~ off will have a higher concentration of the lower boiling material than the liquid from which it was evolved. Conversely, if a warm vapor is cooled, the higher boiling material has a tendency to condense in a greater proportion than the lower boiling material. The early distillers of alcohol for beverages applied these fundamental principles. Although distillation was known and practiced in an- tiquity and a commercial still had been developed by Coffey in 1832, the theory of distillation was not studied until the work of Sorel"* in 1893. Other early wor- kers were Lord Rayleigh" and Lewis.* Present-day technology has permitted the large-scale separation by distillation of ethylbenzene and p-xylene, which have only a 39°F difference in boiling points (Re. 1).2 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION Clear oF voper-feee fiquid Figure 1-1 Interior fa column equipped with seve trays, AA distillation column consists of a space for contacting vapor and liquid streams for the purpose of effecting mass transfer between the two. phases ‘Although the contacting of two phases is generally eflected by a series of plates (or trays), packed columns are becoming more widely used as discussed in Chap. 13. However, in the development ofthe fundamentals of the various caleu Jational procedures in this and subsequent chapters, itis supposed that the column is equipped with plates, Tn normal operation, there is a certain amount of liquid on each plate, and some arrangement is made for ascending vapors to pass through the iquid and ‘make contact with it. The descending liquid flows down from the plate above through a downcomer, across the next plate, and then over a weir and into another downcomer to the next lower plate as shown in Fig. 1-1. For many Years, bubble caps were used for contacting the vapor with the liquid. A variety _of designs of bubble caps are shown in Fig, 1-2. These contacting devices pro- * mote the production of small bubbles of vapor with relatively large surface areas. Over the past 20 years, most of the bubble-cap trays have been replaced by other types of contacting devices. New columns are usually equipped with either __taloe trays (ee Fig, 1-3) or sieod Ways (see Fig. -1) sometimes called perforated > “trays. In valve trays, the valve opens wider as the vapor velocity ineveases and closes as the vapor velocity decreases. This feature of opening and closing allows the valve to remain immersed in liquid and thereby preserve a liquid seal over Wide ranges of liquid and vapor flow rates. : Distillation columns have been built as high as 338 et, Diameters as large as 50 feet have been used. Operating pressures for distillation columns have been reported which range from 15 mmHg to 500 Ibjin? abs. A typical commercial in- stallation is shown in Fig. -4INTRODUCTION TO THE FUNDAMENTALS OF DISTILLATION 3 Figure 1-2 Various types of bubble caps used i dsiltion columns (By courtesy of Glisch Ine) Figure 1-3 Portion of a Giisch Vt ballast tr: (By courtesy of Glseh ne.)INTRODUCTION TO THE FUNDAMENTALS OF DISTILLATION 5. {As indicated in Fig. 1-5, the overhead vapor V3, upon leaving the top plate, centers the condenser where itis either partially or totally condensed. The liquid formed is collected in an accumulator from which the liquid stream L, (called reflux) and the top product stream D (called the distillate) are withdrawn. When the overhead vapor ¥; is totally condensed to the liquid state and the distillate D is withdrawn as a liquid, the condenser is called a total condenser. If V2 is partially condensed to the liquid state to provide the reflux L, and the distillate D iis withdrawn as a vapor, the condenser is called a partial condenser. The i ie ty Ea Wf Trees fe fy sf (Feed plate) Figere 1-5 Sketch ofa conventional cola ia which the totabiow rates are constant within the resting and stripping sections.{6 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION amount of liquid reflux i commonly expressed in terms ofthe reflux ratio, L/D. Although the internal liquidto-vapor ratio L/V is sometimes referred t0 a8 the internal refx ratio, the term reflux ratio will be reserved herein to mean L,/D. The liquid that leaves the bottom plate of the column enters the reboler, where itis partially vaporized. The vapor produced is allowed to flow back up through the column, and the liquid is withdrawn from the reboiler and called the bottoms oF bottom produet B. In practice, the reboiler is generally located ex- ternal fom the column 1-1 FUNDAMENTAL PRINCIPLES INVOLVED IN DISTILLATION To compute the composition of the top product D and the bottom product B which may be expected by use of a given distillation column operated at a given set of conditions itis necessary to obtain a solution to equations of the follow- ing types: 1. Equilibrium relationships 2. Component-material balances 3. Total-material balances 4, Energy balances Consider first the subject of equilibrium relationships. Physical Equilibrium A two-phase mixture is said to be in physical equilibrium ifthe following condi- tions are satisfied (Re. 3) 1. The temperature T" of the vapor phase is equal to the temperature T* of the liquid phase, 2. The total pressure P" throughout the vapor phase is equal to the total pressure P' throughout the liquid phase. (a) 3. The terdency of each component to escape from the liquid phase to the vapor phase is exactly equal to its tendency to escape from the vapor phase to the liquid phase. In the following analysis it is supposed that a state of equilibrium exists, Th=Te=T, PP and the escaping tendencies are equal. Now consider the special case where the third condition may be represented by Raoult’ law Py. = Pix, (1-2)INTRODUCTION TO THE FUNDAMENTALS OF DISTILLATION 7 where x; and y, are the mole factions of component 7 in the liquid and vapor phases, respectively, and P; is the vapor pressure of pure component iat the temperature T of the system ‘The separation of a binary mixture by distillation may be represented in two-dimensional space while n-dimensional space is required to represent the separation of a multicomponent mixture (i> 2).The graphical method proposed bby MoCabe and Thiele for the solution of problems involving binary mixtures is presented in a subsequent section. The MeCabe-Thiele method makes use of ‘an equilibrium curve which may be obtained from the “boiling-point diagram.” Construction and Interpret for Binary Mixtures jon of the Boiling-Point Diagram ‘When a state of equilibrium exists between a vapor and a liquid phase composed ‘of two components A and B, the system is described by the following set of independent equations Pg = Pads Eulbrium Prem Pade relationships (13) Yat yond eet eet where itis understood that Raoult’ law is obeyed. Since the vapor pressures P and Pp depend upon T alone, Eq. (I-3) consists of four equations in six un- knowns, Thus, to obtain a solution to this et of equations, two variables must be fixed. [Observe that this result is in agreement with the Gibbs phase rule: P+ =e+ 2. For the above case, the number of phases # = 2, the number of ‘components ¢ = 2, and thus the number of degrees of freedom ¥° = 2, that is, the number of variables which must be fixed is equal to 2,] In the construction of the boiling-point diagram for a binary mixture, the total pressure P is fixed and a solution is obtained for each of several temperatures iying between the boiling point temperature T, of pure A and the boiling-point temperature 7, of pure B at the total pressure P. That is, when T = Ty, P, = Pand when T = Ty, Pa = P. ‘The solution of the set of equations (Eq. (1-3)] for x, in terms of Pa, P and P is effected as follows. Addition of the first two equations followed by the elimination of the sum of the y’s by use of the third expression yields P= Paxat PoXxe (1-4) Elimination of x» by use of the fourth equation of the set given by Eq, (1-3) followed by rearrangement of the result so obtained yields xa pote (5)Figure 1-6 The boting point diagram, From the definition of a mole fraction (0 <.x4 <1), Bq. (1-8) has & meaning Solution at a given P for every T lying betwecn the botling-point temperatures , and Ty of pure 4 and pure B, respectively. Alice x, has bec computed by tse of Eq (1-5) atthe specifed P and T, the corresponding valve oly which sin esilibrium with the value of sy 50 obtained i computed by ike ofthe frst expression of Eq (13). namely, Pa vam (Bu (16) By plotting 7 versus x, and T versus y,, the lower and upper curve, respec tively, of Fig. 1-6 are typical of those obtained when component is more volatile than B. Component A is said to be more volatile than component B, if for all T in the closed interval T, << Tg, the vapor pressure of A is greater than the vapor pressure of B, that is, P, > Py. The horizontal lines such as CE that join equilibrium pairs (x, y), computed at a given T and P by use of Eqs. (1-3) and (1-6), are commonl; called te lines Example 1-1 (Taken from Ref. 6 by courtesy Instrument Society of America). By use of the following vapor pressures for benzene and toluene {taken from The Chemical Engineer's Handbook, 2d ed, J. H. Perry (ed.) MeGraw-INTRODUCTION 10 THE FUNDAMENTALS OF DISTILLATION 9 Hill, New York, 1941}. compute the three equilibrium pairs (x,y) om a boiling-point diagram which correspond to the temperatures T = 80.02°C, 100°C, and T= 1104°C. The total pressure is fixed at P= 1 atm Given: “Temperature P (benzene) P (oluene) coy (rots) (mg) 1 3000 482 ano as ano ans 590 ess roast 7000 1m the more recent editions, the vapor pressure ‘01704 Simm for toluene at 108°C inaceuraey ited 57405 06 141 mm, SoLUTION ALT = 8002°C, P, gives 100, and P = 760, Then Eq, (1-5) ‘Thus yan Ax gel Therefore at the temperature versus y, coincide at (1, 80.02). 80002°C, the curves T versus x, and T At T= 1104°C, P,= 1748, Py 7160, and P= 760, Then by Eq, (1-5) Xa Tas — 760 v and thus 748 vam [Fe] =0 Hence, the curves T versus x, and T versus y, again coincide at the point (0, 104) ‘At any temperature between T, and Ty, say T = 100°C, the calculations are carried out as follows 760-559 _ 201 *4= 344 — 559 ~ 785, meadand 1344 v= Ge, ‘These results give the point (0.256, 100) on the T versus x. curve and the point (0.453, 100) on the T versus y, curve. Other points on these curves for temperatures lying between T, and T, are located in the same manner. 02s) = 0453 A boiling-point diagram is @ most convenient aid in the visualization of phase behavior. For dfiniteness, suppose P is fixed at 1 atm. Consider fist the case ofthe liquid mixture of A and Bata temperature Tj, ata pressure of | atm, and with the compesition x, = xy. X4=1—X,. As indicated by Fig 1-6, such a mixture is inthe single-phase region, Suppose the pressure i held fixed att atm throughout the course of the following changes. Firs, suppose the mixture is heated to the temperature 7, At this temperature, the frst evidence of a vapor phase, a “bubble of vapor.” may be observed, The temperature 7, is called the bubble-point temperature of aliquid withthe composition x, The mole fraction of A in the vapor in equilibrium with this liquid is seen to be y. As the mixture is heated from 7, to 7;. vaporization continues. Since A has a greater escaping tendency than B, the liquid becomes leaner in A (xy < x,) The relative amounts of A and B vaporized also depend on thee relative amounts in the liquid phase As the liquid phase becomes richer in B, the vapor phase also becomes richer in B (y2 <1). Point D (the intersection of the horizontal fine passing through T; And the vertical line passing though x, is sen to li in the two-phase region. As outlined in Prob. 1-22. i can be shown that the ratio of the moles of vapor to the mole of liguid formed from a feed of composition x, at T, is equal to the fatio of CD/DE. Also, note that all intial liquid mixtures (atthe temperature T,) with the mole fraction of lying between x, and ys will have the same equilib- sium composition (x3. 13) atthe temperature Ty and pressure P= 1 atm. Ifthe Particular mixture x, = vy at Ty is heated until point Fis reached, the equil brium mixture (ry, 13) at; is obtgined. The temperature i called the dew- Point temperature, AtF the last point in the two-phase region, al of the liquid is ‘aporized with the exception of say, one drop, Thus, the dew-point temperature is seen tobe that temperature at which the fist drop of liquid is formed when & vapor with the composition 3 = xy is cooled from a temperature greater than its dew-point temperature to its dew-point temperature. Ty Generalized Equilibrium Relationships Unfortunately, the phase behavior of many mixtures is not adequately described by Raoult’s law. A more precise statement of the third condition ef Eq, (I-1) is that the partial molar-free energy of each component in the vapor phase is equal to its partial molar-free energy in the liquid phase (see Chap. 14), From this condition the following alternative but equivalent statement may be deduced (1)INTRODUCTION TO THE FUNDAMENTALS OF DISTILLATION L1. where and j-are the fugacites of component inthe vapor and quid phases, respectively, evaluated atthe compositions ofthe respective phases and atthe B fand T ofthe system, Equation (1-1) may be restated in the following equivalent form i fly = ab Sts, (18) where J {T= fugacities of pure component i in the liquid and vapor states, respectively, evaluated at the total pressure P and temperature T of the system ‘= mole fractions of component j in the liquid and vapor phases, respectively = activity coefficients of component i in the liquid and vapor phases, respectively. yf = yf(P, T, x1. --1 xs 8 =H(P, Ts din Vd) IF it may be assumed that the vapor forms an ideal solution, then y} = 1 for each {and Eq, (1-8) reduces to w= WK, (19) where K,= if the ideal solution K value. The expression given by Eq. (1-9) is recognized as one form of Henry's law. If the liquid phase also forms an ideal solution (7!"= I forall) then Eq, (19) reduces to ye Kx, (1-10) In some ofthe literature, the activity coefcients 3f and fare absorbed in Ky thats, the produet y£K;/7 is called K,and an equation of the form of Eq. (I-10) is obtained which is applicable o systems described by Eq. (I-8). If the elect of total pressure on the liquid fugacity #s negligible in the neighborhood of the vapor pressute of pure component i, then (eu) where P; is the vapor pressure of pure component i at the temperature T. If in addition to the assumptions required to obtain Eqs. (I-10) and (I-11), one also assumes that the vapor phase obeys the perfect gas law (Po= RT), then Eq, (I-10) reduces to Raoult’s law, Eq. (1-2). Determination of the Bubble-Point and Dew-Point Temperatures ‘of Multicomponent Mixtures In the interest of simplicity, the equilibrium relationship given by Bq. (1-10) is used in the following developments. The state of equilibrium for a two-phase (vapor and liquid) system is described by the following equations where any12. FUNDAMENTALS OF MULTICOMPONENT DISTILLATION ‘number of components ¢ are distributed between the two phases w= Kix, Equilibrium | Sym ‘elationships a (1-12) Since K; isa function of the total pressure P and the temperature T [K, = K(P, T)] itis evident that the expressions represented by Eq, (1-12) eon- sists of ¢ + 2 equations in 2c + 2 unknowns. Thus, o obtain a solution to these equations, variables must be fixed. For the particular case where ¢ ~ 1 values of x, and the total pressure P are fixed, the temperature T required to satisty these equations is called the bubble. point temperature. The cth mole fraction may be found by use of the (¢ 1) fixed Values of x; and the last expression given by Eq, (1-12). When the first expression is summed over all components and the sum of they; eliminated by use of the second expression given by Eq. (I-12), the following result is obtained DKxet (143) Equation (1-13) consists of one equation in one unknown, the temperature. The form of the implicit function K(T) generally requires that the solution of Eq, (1-13) for the bubble-point temperature be effected by a trial-and-error procedure. Of the many numerical methods for solving such a problem, only Newton's method*.* is presented. In the application of this method, it is con- venient to restate Eq. (1-13) in functional form as follows y= $k ‘Thus, the bubble-point temperature is that T that makes f(T) =0. In the solu- tion of this problem by use of Newton's method, an expression for f°(7) is needed. Term-by-term differentiation of Eq, (I-14) yields fi) Newton's method is initiated by the selection of an assumed value for T, say T, Then the values of /(Z,) and f'(T,) ate determined. The improved valve of T, denoted by T+ is found by application of Newton's formula (see Prob. 1-6) Sh) F(T) The value so obtained for 7,., becomes the assumed value forthe next tial This procedure is repeated until | f(0)| is less than some small preassigned (144) (14s) T= T— (1-16)INTRODUCTION TO THE FUNDAMENTALS OF DISTILLATION. 13 positive number « Observe that when the T has been found that makes S(T) =0, each term K,x, of the summation in Eq. (I-14) is equal to y,. In illustrative Example 1-2 as well as those which follow, synthetie functions for the K values and the enthalpies were selected in order to keep the arithmetic simple. Example 1-2 (a) If for a three-component mixture, the following informa- tion is available, compute the bubble-point temperature at the specified pressure of P= I atm by use of Newton’s method. Take the first assumed value of T, to be equal to 100°F. (6) Find the composition of the vapor in equilibrium with the liquid Component 1 10 yer 464710? 2 BID 464710 3 x10 a6utr x10" 13 SouvTiON (a) The formula for (whichis nde in Newton's metho is tained as follows a(T)_ dK, gs ~nry| Ec Sr Bar > RCo“ Gs) = E (Kanes?) Trial 1. Assume T = 100°F = S60°R Ko 50078 Component and taim Ks, EP (K,aiN BT) 1 1 Be ‘000403 2 2 2 000986 3 3 33 nis 01a a= 00296 Thus S156) = 5 «4K «029614 FUNDAMENTALS OF MULTICOMFONENT DISTILLATION Then AT) 1 1 t Pay 73 ~ aon = 20° Trial 2. Assume T= 526°R Ku 268 Component and baim= Kix, E/T (UK x MET) 1 oss 0195 00168 ono 2 1170 ooi6s 000555, 3 1388 00168 3 170 on197 Thus J(526) = 1.17 = 1=0.17 526) = 0.9197 Then _ (0.17) 1 T Baie =F Trial 3. Assume T= $174°R and repeat the steps shown above. Tye results so obtained are as follows s(5174) = 10102 — £5174) = 00175 0102 ‘Then 00102) _ siggp T= S174 ~ Gon73) Trial 4. Assume T= 5168°R 1(5168) = 0.999 ~ 1 = -0001 which is within the desired accuracy of the calculations. Thus, the bubble- point temperature is S168°R = S68°F. (6) At equilibrium y; = K,x,. Thus Ko eeR Comporent am yokes 1 os a4 2 10 ons 3 1s 0300INTRODUCTION TO THE FUNDAMENTALS OF DISTILLATION 15 When the {y,) and P are fixed rather than the {xi} and P, the solution temperature of the expressions given by Eq (I-12) is called the dew-point temper. ature. By earranging the first expression of Eq. (1-12) to the form x, = y,/K, and carrying out steps analogous to those described above, the dew-point function F(T) is obtained (1-17) The dew-point temperature that T that makes F(T) =O. In this ease <9 dk, rin - 3 eae (118) When the T is found that makes (7) Eq, (1-17) is equal t0 x, Now observe that if after the bubble-point temperature has been determined for a given set of x/s, the set of y's so obtained are used to determine the ddew-poiut temperature at the same pressure, it will be found that these two temperatures are equal. For a binary mixture, this result is displayed graphically in Fig. 1-6. For example, a bubble-point temperature calculation on the basis of the {x,) yields the bubble-point temperature T, and the composition of the vapor {y;,}. Then a dew-point temperature on the set {y,,} yields the dew-point temperature T, and the original set of x,/s. = 0, each term y;/K, of the summation in Use of the Ky, Method for the Determination of Bubble-Point and Dew-Point Temperatures Robinson and Gilliand'? pointed out that if the relative values of the K/s or P's are independent of temperature, the expressions given by Eq, (1-12) may be rearranged in a manner such that trial-and-error calculations are avoided in the determination of the bubble-point and dew-point temperatures. The ratio K,/Ky is called the relative volatility a, of component i with respect to component h, that is, z (119) where K, and K, are evaluated at the same temperature and pressure. Compo- nent b may or may not be a member of the given mixture under consideration. ‘When the {x} and the pressure P are given and itis desired to determine the bbubble-point temperature, the formula needed may be developed by first rewrit- ing the first expression of Eq, (1-12) as follows kos aK yx (1-20)Summation of the members of Eq. (1-20) over all components i, followed by Fearrangement yields L in Since the as are independent of temperature, they may be computed by use of the values of K, and K, evaluated at any arbitrary value of T and at the specified Pressure, After Ky has been evaluated by use of Eq. (I-21), the desired bubble- Point temperature is found from the known relationship between K, and T. Ifthe y's are known instead of the x; then the desired formula for the determination of the dew-point temperature is found by first rearranging Eq, (1-20) to the following form K, (12) Kx, and then summing over all components to obtain k= 52 (2) This equation is used to determine the dew-point temperature in a manner ‘analogous to that described for the determination of the bubble-point tempera- ture by use of Eq (1-21), Many families of compounds are characterized by the fact that their vapor pressures may be approximated by the Clausius-Clapeyron equation, and by the fact that their latent heats of vaporization are approximately equal. The loga- rithm of the vapor pressures of the members of such families of compounds fall ‘on parallel lines when plotted against the reciprocal of the absolute temperature. For any two members i and 6 of such a mixture, itis readily shown that a; is independent of temperature, Although there exists many systems whose a;'s are very nearly constant and Eqs. (1-21) and (1-22) are applicable for the determination of the bubble-point and dew-point temperatures, respectively, the greatest use of these relationships lies in their application in the iterative procedures for solving multicomponent distillation problems as described in Chap. 2 Example 1-3 Use the K, method to solve Example 1-2. SOLUTION Since Ky may be selected arbitrarily, take K, T= 100°F = S60'R’ Ky. AssumeSTRODUCTION TO THE FUNDAMENTALS oF DISTILLATION 17 Thus yes Since Ky = Cye"'"", it follows that Ey _ 46487 x 10° InC\/Ky In&x Tr 168°R -2 SEPARATION OF MULTICOMPONENT MIXTURES BY USE OF ONE EQUILIBRIUM STAGE Each of the separation processes considered in Secs. 1-2 and 1-3 consist of special cases of the general separation problem in which a multicomponent ‘mixture i to be separated into two or more parts through the use of any number of plates. ‘The boiling-point diagram (Fig, 1-6) is useful for the visualization of the necessary conditions requited for a flash to occur. Suppose that feed to be flashed has the composition X, = xy, (x, « and x, ») and further suppose that this liquid mixture at the temperature T, and the pressure P= 1 atm is to be flashed by raising the temperature to the specified lash temperature T; = T; at the specified flash pressure P= 1 atm, First observe that the bubble-point tem- perature of the feed Typ at P= 1 atm is T,. The dew-point temperature, Top, of the feed at the pressure P= 1 atm is seen to be Ts. Then it is obvious from Fig. 1-6 that a necessary condition for a flash to occu at the specified pressure is that Tar < Ty < Tow (1-23) In practice, the flash process is generally carried out by reducing the pressure on the feed stream rather than by heating the feed at constant pressure as described above. To determine whether the feed will flash at a given T; and P, the above inequality may be used by determining the bubble-point and dew-point tempera- tures of the feed at the specified pressure P. In the determination of the bubble- point temperature of the feed at the specified P of the lash, the {x,} in Eq, (1-14) are replaced by the {X;) of the feed, and in the determination of the dew-point temperature at the specified pressure, the {yj} in Eq. (I-17) are replaced by the (X)}. Alternatively, the inequality given by Eq (1-23) i satisfied if at the specified Ty and P S(T.) > 0 F(T) >0 (1-24)Figure 1-7 The ash proses where H(t) =F KyX.- 1 eX, FT) x K, The symbol Ky, represents the K value of component i evaluated at T, and P. ‘The two types of fash calculations which are commonly made are generally referred to as isothermal and adiabatic flashes. -1 (1-25) Isothermal Flash Process ‘The name “isothermal lash” is commonly given to the singlestage separation process shown in Fig. 1-7 for which the fash temperature Ty and pressure P are specified as well as the total low rate F and composition {X,} of the feed. The name “isothermal fash originated, no doubt, from the fact that the tempera ture of the contents of the flash drum as well as the vapor and liquid streams formed by the flash is fixed at T,..The lash temperature T, is not necessarily equal to the feed temperature prior to its lashing. For the set of specifications stated above, the problem is to find the total flow rates Vand Ly and the respective compositions {y,} and {xp} ofthe vapor and liquid streams formed by the flash process. In addition to the c + 2 equations required to describe the state of equilib- rium between the vapor and liquid phases [see Eq, (I-12) ¢ additional component-material balances which enclose the flash chamber are required (0 describe the isothermal lash process. Thus, the independent equations required to describe this lash process are as followsINTRODUCTION TO THE FUNDAMENTALS OF DISTILLATION 19 Xr (12.8) Seem | 7H Taare | 129 Material balances FX, (12.0 Equation (1-26) is seen to represent 2c + 2 equations in 2c +2 unknowns [¥ Les Web (vil This system of nonlinear equations is readily reduced to one equation in one unknown (say ¥,) in the following manner. First observe that the total material balance expression (a dependent equation) may be obtained by summing each ‘member of the third expression of Eq, (1-26) over all components to give FLX, ‘The relationships given by Eq. (1-26) may be reduced to one equation in one unknown in a varity of ways, and a variety of forms ofthe lash function may be obtained. One form of the fash function is developed below and a diferent form is developed in Chap. 4 in the formulation of multiple-stage problems. Elimination ofthe ys from the lst expression given by Eq, (1-26) by use of the first expression, followed by rearrangement, yields x, “Smt brSmn of FaVetby (120) oe 28 "LIF VR alF a Elimination of Ly from Ea, (1-28) by use of Ea, (1-27) yeas Xs stacey 129) where % Yor ‘When each side of Eq, (1-29) is summed over all components i and the result so obtained is restated in functional form, the following expression of the flash function is obtained x, Pw) & U-"0-Kal~ 1 (1-30) a0 5 Xl = Kn) 2 PO 2 40 RP om20 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION AA graph of the branch of the function PCY) which contains the postive root is presented in Fig 18. An examination of ths curve shows that Newton's method always converges tothe desired root when W = 1 is taken to be the frst assumed value of the root. After the postive root that makes P(¥) = 0 has been found, both ¥ and Ly may be calculated by use ofthe fact that = Vp/F and the total ‘material balance given by Eq. (I-27). Also, a comparison of Eqs. (1-29) and (1-30) shows that each term in the summation of P() = 0 is one of the solution Values of xj. After the solution set of x's has been computed, the correspond- ing set of ysis found by using the ist expression of Eq. (1-26), ys = KrsXvs Example 1-4 (Taken from Ret. 6 by courtesy Instrument Society of America.) It is proposed to flash the following feed at a specified temperature T, = 100°F and a pressure P = 1 atm, caret % 1 wrt ig : ea + Tis in °F and Pisin aim Ir the feed rate to the flash drum is F = 100 mol/, compute the vapor and liquid rates V; and Ly leaving the fash as well as the respective mole fractions {yr} nd {xp} ofthese steams SOLUTION First, the specified value of T, will be checked to determine ‘whether or not it lies between the bubble-point and dew-point temperatures of the feed. Component x % me Kk, 1 » m4 2 1 n'y 3 Be 3m i Tae00) 1 Yeyse ne 18 Graphieal representation of the Rash fuetion PY) Thus Ss) = YK nX,~ t= 194— F(T) and thus Typ < 100 < Top. Trial 1. Assume ¥ = Component Ky t-Fn Wl —Ke)22. FUNDAMENTALS OF MULTICOMPONENT DISTILLATION Components 0) P(1) = 18487 __ (02619 18487, ¥ ) = 101417 = 0858s Trial 2. Assume Y = 08583 and repeat the steps shown in the first trial ‘The results so obtained are as follows P(08583) = 1.0651 — P'(08583) = 1.0358, v, = 0.8583 — (2065 Continuation of this procedure gives the solution value of = 0.787. Thus, Ve= 78.7, Lp = 21.3, and the solution sets {xp} and {yp} are as follows Compo _ , - - net = in I= Ke) IMU Kn) x agg r 0667 052s oars 701 0238 2 =100 0787 L787 ous? ox 3 2500 1968 2968 ou 0392 Up to this point no mention has been made of the manner of satisfying the energy requirement of the flash. The specification of T; implies that the feed either possesses precisely the correct amount of energy for the flash to occur at T, at the specified P or that energy is to be added or withdrawn at the flash sowterte ge) wr may be restated as For a partial condenser, y, =X, and hence y,;= follows DK, PX a= (Pear (214) or hy Aud e215) where Au= LalKD) ‘The expression given by Eq. (2-15) may be used to represent both a partial condenser and a total condenser, provided that Ay, is set equal to Ly /D for a total condenser.+ Also, the form of Ay, differs slightly rom that for 4, because of the double representation of the reboiler by the subscripts NV and B. ‘Thus, the equilibrium relationship yy, = Kix may be restated in the form 5 Pt 216) Varna = 6 where Ax= BK Vs) When the ly’s and b, are eliminated from Eq. (2-12) by use of Eqs. (2-10), (2-15), and (2-16), the following result is obtained (Ay + IM, + 0 Aud (Aas + Wor + 0y =0 Material Amn iPjnna = (Au + Do + Oar Sf 2) ares Apesityens = (Apart Dyna Op Equilibrium | Ay-tity-tu~ (Ant Dent Opens Telatonships wath Ue Dg Opes UH S414 2 N= 1) An yeni (Asi Digs 0 em) * This notation of convenience should no be taken to mean that K,,= Ifo total condenser For he boiling point temperature, T ofthe distillate leaving a total condenser is computed by wse of the equation Eount= BeateDEVELOPMENT AND APPLICATION OF THE THETA METHOD OF CONVERGENCE 53 ‘This set of equations may be stated in the matrix form, “h (8) ‘The remainder ofthe development of the calculational procedure is ordered in the same sequence in which the calculations are carried out. The caleulational Procedure is initiated by the assumption ofa set of temperatures (7) and a set of vapor rates {¥} from which the corresponding set of liquid rates {L,) is found by use of the total material balances presented below. This particular choice of independent variables was fist proposed by Thiele and Geddes." On the basis ‘of the assumed temperatures and total-low rates, the absorption factors {4} appearing in Eq. (2-18) may be evaluated for component i on each plate j. Since matric A, in Eq, 2-18) is of tridiagonal form, this matrix equation may be solved for the calculated values of the vapor rates for component i [denoted by (ey )a] by us of the Thomas algorthm* which follows. Consider the following set of linear equations in the variables xy, x3, .-.xy-1. Xy whose coefficients form @ tridiagonal matrix, by tere = dy 8, + Bax + 2X5 = dy Osx + yxy + eyXe= ds (2-19) yap Ryn + Bye Xyas + Onan = dyn ayxyey + byxy= dy‘Sf FUNDAMENTALS OF MULTICOMPONENT DISTILLATION ‘These equations may be solved by use of the following recurrence formulas, which are applied in the order stated cy/by gy = a/b h (6=2.3...N-1) ba hon (2-20) Me KERN) After the f's and o's have been computed, the values of x4, y= 4y os 3584 Ate ‘computed as follows xy an maa—hme KEN 1N 2 21) 21) ‘The development of the recurrence formulas is outlined in Prob. 2-3. An improved form of these expressions was recently proposed by Boston and Sullivan.’ For the special case of a convestional distillation column in which ‘model 2 (see Fig. 2-2) for the feed plate is assumed, the procedure proposed by Boston and Sullivan (see Prob. 2-3) may be used,to reduce the above formulas to the following form matt Ay my = Aymy +1 ma Aum tl (223 .N=1) (0-2) 9s-2=0 977 Ut Apa 9- Vt 1 Annie EE (k= S41, f+ 2, 0005) o Bh (k=f+Lf- ) Again, after the 's and g's have been computed, the values of yy yeas oo 835 x, are computed by use of Eq. (2-21). ‘Aller the recurrence formulas have been applied for each component i and the set of component vapor rates {(0j)q} have been found, the corresponding set of liquid rates (lj) are then found by use of Eq (2-10). These ses of calculatedDEVELOPMENT AND APPLICATION OF THE THETA METHOD OF CONVERGENCE 55 method of convergence and the Ky ‘method in the determination of an improved et - Formulation of the 0 Method of Convergence In this application of the @ method of convergence, itis used to weight the mole fractions which are employed in the K, method for computing a new tempera- ture profile. The corrected product rates are used as weight factors in the calcu- lation of improved sets of mole fractions. The corrected terminal rates are seleeted such that they are both in overall component-material balance and in agrerment with the specified value of D, that is, PX,= (dea + (bideo (2-23) and ), oft) 235) ( (The subscripts co and ea are used throughout this discussion to distinguish between the corrected and calculated values of a variable, respectively.) Elimina- tion of (b,(d) from Eqs. (2-23) and (2-25) yields the formula for (4)... namely, eles 1+ O(bi/d bw ‘Since the specified values of (4). are to have a sum equal to the specified value (of D, the desired value of @ is that 8 > 0 that makes g(0) = 0, where 9) = Yd). ‘A graph of this function for @ > 0 is shown in Fig. 2-4 In the determination of 8 by Newton's method, the following formula for the first derivative, '(0), is needed (e= 26) D (2.27) a g _(bildibeaF Xs =~ T+ ObiiddaP ‘After the desired value of has been obtained, (bi). may be computed by use of Eq, (2-25). (Note, Newton's method converges t0 the positive root of g(?), provided @ =0 is taken to be the first tril value; see Prob. 2-11), For the case where the dew-point temperature of the distillate is specified instead of the distillate rate D, the g function has the form shown in Prob. 2-13. (228)'S6 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION. ie) Figure 24 Geometrical Representation of the function (0) in the neighborhood of the positive 000, The corrected mole fractions for the liquid and vapor phases are computed as follows (elder Ealddaldde y= able ™ 2 (ude ‘The development of these formulas as well as the proof ofthe fact that they are ‘consistent with the definition of @ is left as an exercise for the student (see Prob. 2-15}. Determination of a Set of Improved Temperatures by Use of the K, Method On the basis of the mole fractions given by Eq. (2-29) and the lat temperature Profile (the one assumed to make the mth trial), the new temperature profile is found by vse of the K, method!” in the following manner, For any plate jVELOPMENT AND APPLICATION OF THE THETA METHOD OF CONVERGENCE ST Eq, (I-21) or (1-22) may be applied as follows i) ea where ay = K,/Kja the relative volatility of component i atthe temperature of plate j. The quantity Ky is the K value of the base component, evaluated at the temperature of plat It can be shown that the x,’s and y,’s defined by Eq. (2:29) form a consis- teat set in that they give the same value of K, (see Prob. 2-7). Component b represents a hypothetical base component wnose K value is given by InKy where the constants a and b are evaluated on the basis of the values of K at the upper and lower limits of the curve fits of the midboiling component of the mixture or one just lighter. Thus, after Kj, has been computed by use of Eq, (2-30), the temperature T,,,, to be assumed for the next trial is calculated directly by use of Eq. (2-31) The corrected compositions and the new temperatures are used in the ‘enthalpy balances to determine the total flow rates to be used for the next trial through the column. Determination of a Set of Improved Total-Flow Rates by Use of the Constant-Composition Method In the constant-composition method, one of the total-low rates (V, or Lj) is ‘eliminated from the enthalpy balance given by Eq. (2-1) for each stage by use of ‘the component-material balances for the respective stage. The restatement of the ‘enthalpy balances given by Eq. (2-1) in the form called the constant-composition method may be initiated by first observing that YE Harem Sag and Use of relationships of this type permits the enthalpy balance to be restated in terms of the component-flow rates, For example, the enthalpy balance enclosing plate j ViesHjer—Ljhy—DHy~ Oem 0 G HN f-2) (232)‘may be restated in terms of the component-flow rates as follows Elenite Mal Hdd ~ Qe (233) hy, for a distillation column having a total condenser and, Hy = Hy, for a distillation column having a partial condenser. When the component-material balance enclosing plate j [see Eq. (2-12)] ene betds Gb nf 2) is used to eliminate r,,,,, from Eq, (2-33), the following result is obtained ° Elta = halt yes Had] ~ @ LX Ayvrcs = Malu t DY (Hyori HodX ns — Qe ‘The desired expression for calculating L, is then given by = DE Uys HDX bys fet) 238) Ena hss Similarly DE Hp. HaDX n+ Vp Sys — Hare as) od . c= bE (tay hs + D 36) The flow rates in the stripping section may be determined by use of the enthalpy balances which enclose either the top or the bottom of the column and the given plate. When the reboiler is enclosed, the following formula is obtained 0 BE (has yd Mere ft ee NI) (237) 2 Hie Mado ace ‘This expression is developed in a manner analogous to that demonstrated above for Eq. (2-34). The above formulas are given the name “ constant-composition method” because cach of the summations appearing in Eqs. (2-34) through (2-37) may be represented by a thermodynamic process which occurs at constantDEVELOPMENT AND APPLICATION OF THE THETA METHOD OF CONVERGENCE $9 composition. The reboiler duty Qy is found by use of the overall enthalpy balance (the ast expression given by Ea. (2-1) “The totaiow rates of the vapor and liquid streams are related by the following total materia balances Vjer=Ly+D Vea Vym by y+ f=2) (2:38) F=D4+B [Aller the 1,8 forthe retfying section and the Vjs for the stripping section have been determined by use ofthe enthalpy balances, the remaining total-low rates are found by use of Eq, (2-38). These most recent sets of values of the variables Travsh Wavsh and {Lyne} are used to make the next ral through the column. The procedure described is repeated until values ofthe desired accuracy have been obtained, A summary of the steps of the proposed ealculational procedure follow Caleulatioral Procedure for Ideal Solutions 1. Assume « set of temperatures {7} and a set of vapor rates (¥). [The set of liquid rats corresponding to the Set of assumed vapor rates are found by use of the total material balances; see Eq, (238)] 2. On the basis of the temperatures and flow rates assumed in step 1, compute the component-flow rates by use of Eqs. (2-18) through (2-21) [or (2-22)] for each component i 3. Find the # > O that makes g(@) = 0; see Eqs. (2-26) through (2-28) (Newton's smethod* always converges to the desited 8, provided that the first assumed value of @ is taken to be equal to 2er0.) 4. Use Eq, (2-29) to compute the corrected x's or y's for each component i and plate 5. Use the results ofstep 4 to compute the Kg for each stage j by use of either one ofthe expressions given by Eq, (2-30). Use the K's so obtained to compute a new set of emperatures {7q» 1] by use of Eq, (2-31). 6. Use the results of steps 4 and $ to compute new sets of totalflow rates, (Wwe a} and {L, 4 By use of Eqs. (2-34) through (2-38). 7. 10, the T;s, and. js are within the prescribed tolerances, convergence has been achieved; otherwise, repeat steps 2 through 6 on the basis of the most recent set of Ts and V's. In the above calculational procedure, it is supposed that the pressure drop from plate to plate is negligible relative t9 the total pressure, If this assumption is not valid, the caleulational procedure is modified as described in Sec. 2-4. ‘The solution of the component-material balances and equilibrium relation- ships by use of the above recurrence formulas is demonstrated by the following ‘numerical exampie.(60 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION Overhead wp0r, Ye *100) endenser | —4 ine Kaccamutanae 2st L280] Tes0 Cust J 0-50 F100 | yz 100 Figure 24 Flow diagram for Example 2-4 Example 2-1 (a) On the basis of the initial set of temperatures (Ty = T; = 1 $60°R) and the total-low rates displayed in Fig. 2-5, solve Eq. (2-18) for the component-flow rates by use of the above recurrence formulas given by Eqs. (2-20) and (2-21). (b) Repeat (a) by use of the recur- tence formulas given by Eqs. (2-22) and (2-21), Ky= Cem (“E/T Comporent X, C, z Specifications ' 3 4 102/R; 46447 « 10” Tota condenser, P= 1 atm poling point 2 US Sx 1OYP 46447 10% Liquid feed (y= FX N= Af = 3 U3 12e 10%P 46447 109 F= 1001 mol, D= Ly = Ly = 5016 ‘moth, Ly = 10 1b moll, Y= Hy = Y= 100 1 moth 1 Tia®,DEVELOPMENT AND APPLICATION OF THE THETA METHOD OF CONVERGENCE 61 SOLUTION (a) Use of Eqs. (2-20) and (2-21) The correspondence of the symbols in the recurrence formulas and the elements of A, and follow b= (Aut 1) dy=0 a= Ay bys (Ay +1) 4-0 a= Ay by= (An +) dy= FX, m= Ay be= ~ (Aut 1) d=0 CCaleuaticn ofthe 4s follows Ka Rae Ru osark 4 and L atm 40° RK IR 1 1 1 2 x2 2 ' 2 a a 3 ' 3 te 1a 1 2 2 ua a 6 Application of the recurrence formulas for tridiagonal matrix equations follows: Component bh By a by v 2 10 8 ' 2 210 4 ' 3 21 oo 4 ' Componest ey dy a be : 1 momo 4s —1s0000 2 1 33303 075 125000 3 1 m3 0s = 116687 Compo: 4, ay fy an not « SP mah pe dees of ah 1100000 100000 0 0.30000 200000 0 2 07m —13353 0 033333 = 141667 0 3 066667 = 1500000 “o2s00 —125000 0(62. FUNDAMENTALS OF MULTICOMPONENT DISTILLATION. Component a= : : -a pont at, ing Wh beth A 7431046 49538 002797 SoOL6AH —sgOPI2—_saoD4ES (cae ated) ° 3686828 09772162 —_LOOI3RS —_LomORI3 LOMAS 1000240 Calculated values of the temperature, “F ‘Teal Sage 7 * 9 0 a 2 1 ais 779 ——-TRIT at 772266 4772266 2 SSAOTIS $8610 —S861279 S612 SSHIDE —SS6LI28 3 46313 AST —OASWO—GDASTIN DASE. ASE 4 GBOBISS —6808B62——GRORTIG —GRONGSS——GSOKEIE—_GRONGIE D——sogorss— S000131 90MM so.ON29 — soONDI —somDDs (caleu. lated) ° 1000128 1.000063 1.000031 1000014 LomaONs 1.000008 Final flow rates Component a 1 261560 7176038 2 1514837188893 3 Seppm08 2463411 2-3 CONVERGENCE CHARACTERISTICS OF THE 0 METHOD OF CONVERGENCE, THE K, METHOD, AND THE CONSTANT-COMPOSITION METHOD ‘The 0 methad of convergence may be classified as an iterative procedure, and the success of such a method depends upon the proper arrangement of all equations involved in the entire calculational procedure. The complete calculational procedure was developed by trial by finding what appeared to be the best formD:VELOPMENT AND APPLICATION OF THE THETA METHOD OF CONVERGENCE TH forall equations. In particular, it was found that the method was most stable when the temperatures were determined by the Ky method and the total-low rates by the constant-composition method. The use of other procedures for the determination of either the temperatures or the total-flow rates caused some problems to diverge which could be solved by use of the K, method and the constant-composition method as discussed below. In summary, it was found tha ifthe combination ofthe 9 method of eonver- ‘gence, the K; method, and the constant-composition method were used, conver- {gence of almost all problems involving conventional and complex columns as well as systems of distillation columns could be achieved. This combination of caleulational procedures is also one ofthe fastest known methods." Although a proof ofthe convergence ofthe combination ofthe 0 method, Ky, method, and constant-composition method cannot be constructed, it can be shown (and is shown below) that the method constitutes an exact solution to certain total reflux problems. To illustrate the characteristics of the combination of these three ealeula- mal procedures, Example 2-7 was selected. The statement ofthis example is presented in Table 2-2. The temperature profiles obtained by use of the Ky method on the basis of the corrected compositions found by use ofthe 0 method [Eqs. (2-27) and (2-28)] are presented in Table 2-3. The constants a and b inthe expression for K, [Eq, (2-31)] were found by use of the K values for -CyHyo at 510 and 960°R. ‘The values of O and the calculated values of D at the end of ‘ach tral are also listed in this table. The vapor rates computed by use of the constant-composition method, the temperatures found by the K, method, and the corrected compositions are presented in Table 2-4. The solution sets {dj and {b) are presented in Table 2-5. To saisty the convergence criterion, twelve trials were required and 2.60 seconds of computer time on an AMDAHL 470 V/6 computer using a WATFIV compiler. Table 2-2 Statement of Example 2-7 Compo vent FX, cH 48 (all fow rate are in Hb mal) boling-pont liquid fed Ct partial condenser, column presure = 300 Ibi? abs, N= 12 and f = 5. Eguilibir and enthalpy data fr al components are given in Tables B-1 and B-2ofthe Append. The filet to be taken neat ‘vith plate number between, = 610°R and Ty) = 910°R. Take the iii te rote oe = 968 = 2.3... 1}. an the coneponding ce Component .CaHys #35 nd bin bg here stermie!o8 190 the basis ofthe values forthe K of -CaHys at $10 and 9€0°R. as 10 1 Commonly rlerred 1 a the 400°F —aormal boiling action.‘72 FUND/MENTALS OF MULTICOMPONENT DISTILLATION ‘Table 23 Temperature profiles, theta, and calculated values of D “Temperature profiles (78) Triad Stage L 2 3 4 5 6 L(disle «S759 STS 56786 «5675656156 ~——s6757 Tate) 2 ssc 5067 s94s¢_ 59413 sona 5433, 3 7 681 SRB 4 si2gs 785 NOI 25 Sed) 6170295726582 GIOS3 eISSN ‘ 64702 T0968 S673 «NID 3 mos 71792793596 A3T 8 ® 99742093) Thaas 9 qa? 707938? 0 msoy 788496 THD THLE u W173 993709 TaSIT Taek? 2 amar 774196684 Te T6770 TS 13 (bowons) 8368 83341547763 Dias. 41575 30aws DIMI LSTO eH iste) @ Rats om676 23462 omISIM LOH 0968934 “Temperature profies (2) ‘Tea ce Stage 7 5 9 0 " 2 FP Nite 567576757677 —«S6757 6737 ate) 2 S438 59436 suas 547 sous 3 611956192 gttg oti 1193 4 6x27 63026 02707 62026 Seed) Gnd 6183 63 osrat ‘ ess? 6s gags aT 60 7 dois 70336 9a 0330 so328 5 ns 71419 Tha 714 nas 9 m7 m95 mFS moo 0 me sD BLID man u 7es0s 748TH tats mas R Wei98 Texan 76799680 76800 1 (botioms) 8265) 426594265668 32657 Plea 3461781925 S160 «15986 3Le0ST 3159977 ed) ° 101356 _o9pais2 100243 ages La002 100000, — SiS __osnaiia 100265 soivay 10002 100000DEVELOPMENT AND APPLICATION OF THE THETA METHOD OF CONVERGENCE 73. ‘Table 2-4 Vapor rates obtained by the constant-composit ‘method for Example 2-7 ee Vapor rats(b-l/) ‘Trl Sa 1230 5 6 2 9430 9480 94s 9480 a 920898299329 919 4 sea wats wal 5 wn 7925 Host ‘ uns nist 1 9st 12679 . nest 10 9 14539 14460 0 183 1394 " 1909 19.83 B 1075 0 B m2 1333 ‘Vapor cates (Ib mol) “hia sage 7 89 oon oD 2 94809480 940 940948094 39329 929 91299329 onay nas 4 soar asain goa? goa Swot ome wo? sot sot sot © 9921075 tis? 11079 11081 11080, 7 12636 12625 1262912624 1262813627 S 13691 68k 1368713686686 He 9 taka eat tate asa? Laat 104839 aR? 14888 nase 14887 Hea) twat 947 194 1947 2 18m 1am 182 1372 372 Ba ase 12382 13s 332 By solving a wide variety of examples it was found that the boiling range, the thermal condition of the feed, the number of plates, the initial temperature, and vapor rate profiles had no appreciable effect on the number of trials required to achieve convergence. However, as the boiling range of the feed is increased, it may become necessary to average the temperature profiles as well as the vapor-rate profiles for successive trials, ‘The use of a component much lighter than the midboiling component of the ‘mixture can introduce significant overcorrections of the temperature profiles while the use of a component much heavier than the midboiling component can lead to undercorrections of the temperature profiles. The use of the midboiling74 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION Table 2-5 Solution sets of the component flow rates in the distillate and bottoms of Example 2-7 ‘Component 4 200000 « 10 11163 x 10°" 0995950 « 10 011627 «10° 03920 «10 0.27665 « 10°# O126eO 10" 015358 o7a6 027578 « 053699 014462 « 108 020188 « 107? O1sI97 x 108 94035 x 10-* 011299 « 10" (094025 x 10-7 089599 «10 (06347 « 10-7 080559 «10 065162 10-0999 10 = 39608 5 10° ah Q,— 1278 x 10° Bru emergence exterion [gH] = 10-* ‘component of the feed mixture or the component just lighter for the calculation of the constants a and b of the K, method generally give satisfactory results, Recently, however, improvements in the K, method were proposed by Billingsley? Wherein a different K, was used for each plate and composition effects were also included, To obtain the convergence characteristics exhibited in Table 2-3 through 2-5, it cannot be overemphasized that the complete set of calculational Procedures (the @ method, the K, method, and the constant-composition method) must be used. When the 9 method was introduced initially,"™ other procedures were used for the determination of the temperatures and total-flow ‘ates, and the convergence characteristics differed from those presented in Tables 23 through 2-5. ‘Comparison of the 0 Method with the Method of Direct Iteration Some of the convergence characteristics of the @ method are demonstrated by comparing the method with the method of direct iteration. The method of Airect iteration differs from the @ method only by the procedure used to compute the compesitions. Instead of the expressions given by Eq, (2-29), the following ‘expressions are used in the method of direct iteration © Eltdaltde (ould) SE loulddaldd (2-49) (2-50)DEVELOPMENT AND APPLICATION OF THE THETA METHOD OF CONVERGENCE 75. By comparison of these expressions with those given by Eqs. (2-25), (2-26), and (2-29). iis evident that the method of direct iteration amounts to seiting 0 = 1 in Eq. (2-26) for all tials. The results obtained for Example 2-2 with the fow rates held fined atthe set stated in this example are presented in Table 2-1. When the method of direc iteration was used, a calculated value of $2.14 was obtained for D at theend of the third tril, and [1 trials were required to obtain tempera- tures which were correct to eight digits. Ata set of fixed L/V's, the calculated values of D follow the same type of variation with respect to temperature as might be expected for an actual column, Ir the assumed temperature profile is too high, then the calelated value of D wil be greater than the specified value; and if the assumed temperature profile is too low, the calculated value of D will be less than the specified value. More precisely, by use of the expressions presented in Prob. 2-4, itis readily shown that by, deteases with an increase in the temperature of any one plate ofthe column, and conversely, Thus, by Eq, (2-46), d, increases as any one temperature is increased, and conversely. Now consider Eqs. (2-26) and (2-27). Ifthe sum of the calculated values of d, is greater than the specified value of D, 8 postive value of 0 that is greater than nity is required to satisfy Eq. (2-27) (that is, make g(0) = 0), and ifthe suin of the calculated values of ds less than the specified value of D, a postive value of less than unity is required to satisfy Eg. (2-27). Thus, ifthe temperature profile for the previous trial was too low. a value of @ less than unity will be abtained and if too high, a value of @ greater than unity will be obtained. ‘The use of corrected ds based on values of < 1 and 0 > 1 gives lower and higher temperatures, respectively, than those predicted by the method of direct iteration. Th is eally shown by consideration of the variation ofthe corected 4s with 8 for very light and very heavy components. For a very light component bids very small so that Eq, (2-26) reduces to (bo = FX, est) For a very heavy component, b,/d, is very large. Thus Fx, a> aesti 2-32) Now consider the case where the temperature profile of the previous trial was too high. This condition leads to a value of @ greater than unity. In view of Eqs. (2-51) and (2-52), itis seen that the formulas for the mole factions for each plate [Eq, (229)] give sets of compositions with a relatively smaller proportion of heavies than those obtained by direct iteration. Therefore, the temperatures computed on the basis of the corrected compositions are less than the corre- sponding temperatures calculated by the method of direct iteration, since the latter are calculated on the basis of the calculated d/s. ‘An Exact Solution Given by the 8 Method of Convergence Ifa set of b,/d’s corresponding to a given D are known for a column at total ‘reflux and a system for which the a's are constant, then the @ method of conver-76. FUNDAMENTALS OF HULTICOMPONENT DISTILLATION ‘gence may be used to compute the solution set of b,/d's corresponding to any other value of D. That the 0 method of convergence is applicable to a problem ofthis type is ‘demonstrated in the following manner. Consider the case of a distillation column with N stages and a rotal condenser which is operated at two different distillate rates, D, and D; [or at two different specified values of by/d,, denoted by (byids), and. (bs/dy)s}- For these two different sets of operating conditions at {otal reflux, the Fenske equation* [see Eq, (1-58)] gives In (bv/di)2 = In (ba /dy)o ~ (N — 1) In a In (b/d), = In (By/d,), — (N= 1) In, (2-53) Elimination of Ina, fiom these two equation gives In (b,/d,)2 = In 8 + In (b,/d,), (r,t), a where boldede fas), If the subscripts 1 and 2 in Eq. (2-54) are replaced by ca and co, respectively, then Eq. (2-25) is obtained. A graphical representation of the expressions given by Eqs. (2-53) and (2-54) is presented in Fig. 2-6, That the @ method is an exact. oD Ina, igure 24 A graphical representation of 8 is obtained by considering two arbitrarily specie values fora base component it a column a total refxDEVELOPMENT AND APPLICATION OF THE THETA METHOD OF CONVERGENCE 77 Figure 27 Shit ofthe b/d protes bythe 8 method. solution for certain problems involving columns at total reflux is demonstrated by Prob. 2-5 While the @ method constitutes an exact solution to certain total reflux problems, it represents only an approximate solution to problems wherein oper- ating conditions other than total reflux are employed, The @ method for @ conventional distillation column at any operating condition other than total ‘may be represented with the aid of Fig. 2-7 as the shifting of the most recently calculated by/d; profile up or down the same distance (|in |) for each compo- nent as required to obtair a new set of by/d's which ere in agreement with the specified value of D. 2-4 OTHER TOPICS: PRESSURE EFFECTS, NONIDEAL SOLUTIONS, AND OTHER SPECIFICATIONS Pressure Effects In the calculational procedure demonstrated above, the effect of the varia- tion of the column pressure from plate to plate on the K's, H's, and hy's was‘78 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION. neglected. The effect of pressure on these variables may be taken into account as described below. For mixtures whici behave as ideal solutions, the K values and the vapor and liquid enthalpies of the pure components {K,,, Hs, hy} depend upon the temperature T; and pressure P, of each plate J. Methods for taking the effect of pressure into account in the evaluation of the K,’s, H,’s, and hy’s are to be found in standard thermodynamic texts, handbooks, and the scientific literature as described in Chap. 14, The effort required to account for the effect of pressure ‘on the K values and enthalpies may be reduced by taking advantage of the following observations. Fist, for columns operating at relatively high pressure, the pressure drop across the column is small relative to the total pressure. Thus, in many instances, itis possible to use a single column pressure in the evaluation of the K,’s, Hy’s, and fy’s, and these quantities may be curve fit asa function of temperature alone, For columns operated at relatively low pressures, such as vacuum distilla- tions, the effect of pressure on the enthalpies can be neglected because the vapor tends to behave as a perfect gas, but the eflect of pressure on the K values cannot bbe neglected. However, the fact that the vapor behavior approaches that of a perfect gas gives rise tc a simple method for including the effect of pressure on the K values. In particular, at low pressures, xP =P K, (88) "Now suppose that the K values for component i are evaluated at some pressure Py at each of several temperatures and that these values of K, are curve fit as a function of T at the pressure P,. The value of Ky at P, and 1, (K,(P,, 7)] may be approximated with good accuracy by use of the value given by the curve fit [K, (Py, T,)] and the folowing relationship which is suggested by Ea. (2-55) KylPy. T) = 22 KylPs. 1) (2-56) Nonideal Solutions Most problems involving the separation of nonideal solutions may be solved by use of the @ method of convergence. When used to solve such problems, the method does become slower and it may be necessary to place certain restraints ‘on the calculational procedure. [The “Almost Band Algorithm” which is pre- sented in Chap. 5 may be used to solve any problem for which the 0 method fails) To apply the @ method to problems involving nonideal solutions, minor modifications of the equations presented above for ideal solutions are necessary. ‘The development of the appropriate expressions for the component-material balances, the Ky method, and the constant-composition method is outlined in Probs. 26 through 2-10DEVELOPMENT AND APPLICATION OF THE THETA METHOD OF CONVERGENCE 79 ‘Modification of the Formulas to Account for Separated Components and Inert Components The formulas for the 0 method were developed for distributed components, where a distributed component is defined as one which appears in both the distillate and bottoms; that is. for a distributed component d, > 0 and b, > 0 ‘A separated component is one which appears in only one product stream. A separated light component is defined as one which appears only in the distillate (6,=0, d,= FX,) and a separated heavy component is one which appears only inthe bottoms (d, = 0, b= FX.) In order to avoid numerical difficulties resulting from the divisions by small values of (dj). in Eqs. 2-26) and (2-29), the following expressions for (dy should be used. These expressions are obtained by a simple rearrangement of Eq. (226), namely, (deo (di )ePs (57) where p, is defined by ‘Then for a separated heary component for which (4, Eq, (2-58) that p, is finite, Since p, is finite and (d)-4= (ie 0 ) Eq. (2-57) gives Consequently, it follows by material balance that for a separated heavy com- ponent ‘Thus, if separated heavies ae present, the ratios (a). /(di)c in the formulas given by Eq. (2-29) for x, and y, become indeterminate. As suggested by Tetlow,'® these indeterminate terms may be eliminated by use of Eqs. (2-57) and (2-58), ‘The resulting expressions may be used for all types of components, separated lights, separated heavies, and distributed components (idea (2-59) Lear where pis to be computed by use of Ea. (2-58). Similarly, the expression for g(0) may be expressed in terms of the p's and used to avoid numerical difficulties 0)= 5 dari —D ew)Although inert gases and inert liquids may be treated as two separate classes ‘of components, as demonstrated in Chap. |, it is perhaps simpler from a compu- tational point of view to treat them as volatile components. This treatment may be carried out by assigning the inert gases a K value which is large relative to the light components and the inert liquid a K value which is small relative to the heavy components. Other Types of Specifications In all of the formulations considered, it is supposed that the following specifications have been made: the number of plates in each section; the quan- tity, composition, and thermal condition of the feed; and the column pressure. ‘Two other specifications may be made in addition to these, and the formulation of the functions corresponding to these two additional specifications may be divided into three classes in addition to the class considered above (the specification of the reflux rate L, and the distillate rate D or the temperature of the distillate; see Prob. 2-13). CLASS 1. SPECIFICATION OF THE REFLUX RATE L; (OR THE BOILUP RATE 4) AND ONE ADDITIONAL SPECIFICATION SUCH AS D, ‘A PARTIAL SUM OF d's OK X's ‘When any one of several specifications are made in lieu of the distillate rate D, a corresponding g function may be formulated. Some care must be given, however, to the formulation of this function in order to obtain one which has desirable behavior, stich as monotonic behavior with respect to the variable 8. Suitable functions are given in Probs. 2-12 through 2-14 for a number of specifications. CLASS 2. TWO PURITY SPECIFICATIONS If instead of L, and D, two purity specifications such as (Xu, Xa (Xms Xa} {dy by} oF {dy, dj} where the subscript I is used to denote the light key and h the hheavy key component. To solve problems of this type, the optimization Procedure described in Chap. 9 may be applied as described in Chap. 11, CLASS 3. SPECIFICATION OF THE REFLUX RATIO L,/D. AND THE BOILUP RATIO 14/2 This particular set of specifications gives rise to a set of g functions which involve the material balances and the energy balances enclosing the column as ‘well as the equilibrium relationships for the terminal streams. This application of the @ method to single columns is deferred until Chap. 7 because of its similarity to calculational procedures developed therein.NOTATION 4 absorption factor; defined by Eq. (2-11) “ a square matrix for each component i; defined below Eq. (2-18) h 2 feed vector in the component-materal balances: defined below Eq, (2-18) (0) a function of 0; defined by Eq. (2-27) W enthalpy per mole of feed, regardless of state Hy enthalpy per mole of distillate, regardless of state. For a total con denser Hp = hy and Hp, = hy,. For a partial condenser, Hy = Hy and Hp: = Hs hy enthalpy per mole of bottoms Ay partial molar enthalpy of component i, evaluated at the temperature and vapor composition of plate j A, Virtual value of the partial molar enthalpy, evaluated at the temper- ature and vapor composition of plate j hy molar flow rate at which component iin the liquid phase leaves the jth plate Ny total number of plates p ratio of distillate rates; defined by Eq. (2-58) P, vapor pressure of component i P total pressure Sy stripping factor for component i; defined by Eq. (2-11) u ‘molar flow rate at which component iin the vapor eaves plate j " column vector of components flow rates in the vapor phase; defined by Bq. (2-18) Subscripts ca caleulated value 0 corrected value f feed plate F variables associated with a partially vaporized feed i component number, i= 1, 2,....€ i stage number; for the accumulator j = 1: for the top plate j= 2, for the feed plate j = f, for the bottom plate j = N~ I, and for the reboiler j= IN; thatis, j= 1, 2,3,....f0-.N—1,N k trial number n ‘vial number reek Symbols Pn defined beneath Eq. (2-18) @ ‘4 multiplier defined by Eq. (2-25){2 FUNDAMENTALS OF MULTICOMFONENT DISTILLATION Mathematica Symbols [ei x2 xy]? transpose of a row vector. The transpose of a row vector is z summation over all components k, where k denotes any set of pene nd equal to a column vector; that is, [sy x xs} ‘components less than the total number ¢, say k PROBLEMS 24 (c) Complete the trial calelations inated by Example 2-2 by o shown, (b) Complete the trial calculations inated in Example 2-4 by nding T.. 22 I the folowing st of K va ig the solution value of 8 and by Component Kt 1 oor, 2 0002727 3 oor + T isin “F and Pis in atm, 1s used instead ofthe st given in Example 21, show that (a) 0 = 3687276, and (8) T, = 253°, T,= ARS0°F T, = 210°F, T, = 6O6S"F. Use the Ky method and take Ky = Ky 253 The recurrence formulas given by Eqs. (2-20) and (2-2) for solving equations which art ‘iagonal in form may be developed as outlined below by use ofthe gausian elimination, The system ‘of ineae equations given By Eq (219) is represented by the following mati equation «0 0 0 opay pa bao 0 offs] fa ah 4 8 offal] a a D0 ony ber Crea aera] Paar oo ow ob) La (c) By use ofthe faliowing definitions af fay smay be tanformed tothe following frm J and gy given ia the text, show that Eq (A) 1h 000 OUfa] fa ot goo ofls| | a oot g,o~ offs] | a a ® ° Or flee 0 oo TIPS (0) Commencing with the boom row of Eq (B) show that the matrix multiplication rule may be applied to give Berge NBA (ANAL NA 2.21) oDEVELOPMENT AND APPLICATION OF THE THETA METHOD OF CONVERGENCE 83 (6) For the special case of a conventional distilation column, b= (14 4y). 6) 1 82 Ay eed = ted —lye ad dy OU # {> U-show that when these quantities are sbsbted successively into the formulas for and g, [, (2-20). the recurrence formulas given by (22) are obtained 24 Suppose that the solution st of operating conditions are known. For definiteness, suppose that the column basa total condenser and that the fed enters the column a hguid a is boing point temperature at the column pressure (a) Suppose that the st (bd) and the total dilate rate Dare computed by using the core, set of EVs and the coret Ts except for one particular plate, sy plate k and fr ths particular pit, a temperature less than the correct one i wed in making the caeslatons. Show thatthe ‘alcuisted value of D so obtained is less than the ovect value (0) Repeat (a) forthe ease where the temperature used for plate Kin the caleltion of the (djs roster than the correct temperature, and show that inthis case the calculated value of Dis {rater than the corect valve int: Show that lh Aa Aas Ant Aa Apt Ap aad Ap and that eplbs = SpSpene SA Spon” Sucre +S Spot 25 The following set of b/s weee found by we of Fenske's equation ia Prob. 24a}, and the Corresponding distil rate was D = 4462. On the bats of his known eto i compute the (ds adds at aD» 562629 forthe following example by use ofthe @ method of convergence. ‘Goer ® Component x, a, Other specications 1B 1 16 _N=3panial condenser, 2 wo2 4 F = 100 moth and 3 13301975 tal rex operation “This problem is based on material given in Re. 6, and 9, For the rst trial assume @ Eq. (254) to explain the results obtained 246 Begin withthe equilbrium relationship given by Eq. (1) andthe defiton ofthe ea soltion 1K value given by Eq, (I-10) and obtain the folowing formulas forthe K, method for monies! solutions Sand use 1 om Ky Zant St Eaton i (Ghia 247 Show thatthe same value of Ky is obtained regardless of whether the corrected 3 ofthe ‘correced values af th p's given by Eq, (2.29) are sed. A more precise statement ofthe problem folows. ‘Given: Keni e— Enna84 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION (2) xp and ys are defied by Ea (229). 0 one (ee 0 Show that where m denotes the values ofthe variables assumed 19 make the mh ri. Prob, 27 forthe ease where both the vapor and liquid phases form nonideal solutions. ate defined by E9229) © ike~ (Figg) de © Ga= (X), Kae ¥ OED 29 (e) By we of the deni ofthe Hea solution K vale Kn Sih (ere /jand are the facts of pure component evaluate at the temperate and pressure of plate) show thatthe equitriam relationship given by Eq (-8) Son wy ray be restated in the following forms y= Ayty 0d = Syl ere , Aen = Uy ° Ginsy, ~ () Stow that the componsst materi blancs are of he same general form for wore solutions as those for sel sltiony prviled tat the abortion (or siping) acer hak te Sein en 2.40 Forte ese of nonifesl soliions, show that he expressions forthe enthalpy bales te of the same fon as those pen by Eas (23) trough (257 exp forthe fata sea sltion ental ofthe pure componcns fy. Hy) ae replace by te paral molar vals fy i or the vital vale fhe paral lems fg fs Chap 241 (a) Show that the fnton dened by (227) at cel ro01 on ose oo and eo pathe oot () Show hati 9 =O selec asthe fat sumed vale often Newton's method a converges tothe poste roo of a0.{CATION OF THE THETA METHOD OF CONVERGENCE 8S 212 For the case where the relux rate and a partial sum of the component disillaion rats, Elles A speed, show thatthe corspoading g function, decreases monotonically with 0. The corected distillate sates [dala defined by Eq (257) 213 Show that when ceux rte L, andthe dew point temperatuteT, ofthe distillate are specified, the coresponding form of the g futon i obtained wo-E [Suu and that i decreases monotonically wih 0, The corrected distillate rate ((d)) ate defined by a, (257) and Pam Elle 14 I the refx rato Ly and a partial sm of moe faction are specified show that the following form ofthe corresponding 9 function varies monotonically with 8 LXw 0° Salm ‘Sketch the fanetion (0) for the case where the K components are relatively light and for the ease when the k components are relatively ear. Note: X= (a /Da 215 Let the cortected component flow rates be defined as follows he nf i) de (one= ($f) (te ere and a wndetermined muir wich are ob picked such hat Ede= te Slee Whe (0) Us the defion ofthe moe faton andthe above eto ofthe comet flow ets twobian the expesion gen by Eq, (229) foie eatin ofthe mle Facts Oy te 8 method (0) Show that ifthe undetermined multipiee ny forthe reboler is denoted by 8, then the ‘efning equation fr 8 (Ea, (2-2) follows fom the above definition of (yh forj = N, the rebole. REFERENCES 1.3. F. Boston and S,L. Sullivan Jr: "An Improved Algorithm for Solviag Mass Baiance Equa tion ia Matstage Separation Procesie,” Con J. Chom. Eng. $0: 663 (1972). 2. D.S.Biligsley: "On the Equations of Holland inthe Solution of Problems in Multicomponent Disition” 1BM J. Res. Des, 14:33 (1970)186 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION a “On the Numerical Solaton of Problems ia Muliomponest Distillation at Steady State Ih ATCKE J. 16-41 (1970) 4. Brice Carnahan, H. A. Lube, and J. ©. Wilkes: Applied Numerical Methods. John Wiley & Sons, Ine. New York, 1964 5. MR Fenske: "Fritonaton of Staight-Run Pennsyvania Gatling” Ind Eng. Chem, 24: 482 (0933, C.D. Holland and G. P. Pendon: “Solve More Distillation Problems: Part I Improvements Give Exact Answers” Hdroarbon Paces, $3: 148 (1974) nd P.T. Eubank: “Solve More Distillation Problems: Part I. Partial Mola Enthalpice (Caleulated.”Hyrocrbon Process $3:176 1974) “Introduction to the Fundamentals of Dstilation” Proceedings of the Fourth Anal Educational Symposium ofthe ISA, Ape, 5-7, 1972, Wilmington, Delaware, “Exact Solutions Given by the @-Method of Convergence for Complex and Conve: ‘ional Distillation Columns” Chem. Eng. J, Loughborough University of Technology 7:10, (972) 10. W. K. Lewis and G. L. Matheson, “Studs in Distillation Design of Rectiing Columns for [Natural and Refinery Gasoling” Ind. Eg. Chem, 24:496 1932) 11, W.N.Lyster, §,L, Selivan, J, D. . Billingsley, and C. D. Holand: “Figure Distillation This New Way: Part I-New Convergence Method. Will Handle Many Cases” Pet, Refier 346): 221 (1989) * and —— “Figure Distillation This New Way: Past Purity Can Set Conditions for Cola,” Pet. Refer, 39): 151 (1959, 13. TA. Narter, 1M. Srygley, and C.D. Holland, *Solton of Problems Involving Sytems of Disilation Columns” Con. Chem. Eng, 44:217 (1966). 14 EW. Thile and R. L. Geddes, “Computation of Disilation Apparatus for Hydrocarbon ‘Mintures" Ind Eng. Chem. 25289 (1933). 15. NJ. Tellow (of Dow Chemical Company) Personal Communication ProductCHAPTER THREE APPLICATION OF THE THETA METHOD OF CONVERGENCE TO COMPLEX COLUMNS AND TO SYSTEMS OF COLUMNS ‘The calculation procedures {the @ method, Ky method, and constant composi- tion method] developed in Chap. 2 for conventional distillation columns are applied to complex distillation columns in Sec. 3-1. For solving problems involv ing systems of columns interconnected by recycle streams, a variation of the theta method, called the “capital © method” of convergence is presented in Secs. 3.2 and 3-3. For the case where the terminal flow rates are specified, the capital © method is used to pick a set of corrected component-flow rates which satisfy the component-material balances enclosing each column and the specified values of the terminal rates simultaneously. For the case where other specifications are made in lieu of the terminal rates, sets of corrected terminal rates which satisfy the material and energy balances enclosing each column as well as the equilibrium relationships of the terminal streams are found by use of the capital @ method of convergence as described in Chap. 7, 3:1 COMPLEX DISTILLATION COLUMNS A complex distillation column is defined as one which has either more feeds introduced or streams withdrawn or a combination of these than does a conven- tional distillation column. To demonstrate the application of the @ method and associated calculational procedures, the complex column shown in Fig. 3-188 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION wLiquioy F, fe — ' fe te ' te w, (VAPOR) Ne Figure 31 Complex column with two eed plates and two sidesreams Which has two feed plates and two sidestreams (in addition to the top and bottom products) was selected. In the application of the aforementioned caleulational procedures, the specifications commonly made are as follows: the column pressure; the number ‘of plates; the rate, composition, and thermal condition of each feed; as well as the locations of the feed plates and sidestreams, The number of additional specifications that may be made is equal to the total number of streams with- drawn (the distillate, bottoms, and sidestreams). For the column shown in Fig. 31, the additional specifications ¥, (or L,), D, W, and W, may be made, These in turn fix the dependent variable B. mn order to make the first trial, temperature and L/V profiles forthe columnare assumed. This allows one to solve the component-material balances for the component-flow rates. After the @ method of convergence for the complex column has been applied, the compositions are computed. The compositions s0 obtained are used to compute an improved set of temperatures by use of the Ky ‘method. These temperatures and compositions are used to find a new set of total- flow rates by solving the enthalpy balances after they have been stated in the form of the constant-composition method. ‘Component-Material Balances ‘The component-material balances are stated in terms of a single set of flow rates, the vapor rates, by use of the equilibrium relationship ly, = A,,2, (Eq. (2-10)} Except for the plates from which the sidestreams are withdrawn, the equations {or the remaining stages of the complex column shown in Fig. 3-1 are formulated in precisely the same manner which was shown in Chap. 2 for conventional columns. ‘To demonstrate the notation employed, an enlarged view of plate p (see Fig. 3-2) from which the liquid sidestream W, is withdrawn as shown in Fig. 31. ‘The component-material balance enclosing plate pis seen to be Is — Mas G) as Ip, and since Ip; = Aye (Ea. (2-10)}, it Opeth Mpa ace Oot Since wy, has the same composi follows that (32) Figue 32 Notation used forthe steams catering and leaving» plat tom which gud siestream ie withdrawn,90 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION Figure 3.3 Notation used forthe streams entering and leaving a plat fom which a vapor siestream ie thd, ‘Then after the equilibrium relationship [Eq. (2-10)] has been used to state liquid flow rates of Eq, (3-1) in terms of the vapor rates, one obtains - [i+ sn(t+PE)feut sperm 63) ‘An enlarged view of plate q from which the vapor sidestream W) is with- drawn is shown in Fig. 3-3. The component-material balance for plate q is siven by tent leant feta 64) Since Wand Vg have the same composition, i follows that a= Bg 65) ‘Then by use of this relationship and the equilibrium relationship (Eq. (2-10)}, it is possible to restate Eq. (3-4) in terms of the vapor rates as follows . % =0 66 ‘The complete set of component-material balances may be represented by a matrix equation of the same form as the one used to represent these balances for ‘ conventional column, namely, en[APPLICATION OF THETA METHOD OF CONVERGENCE TO COMPLEX COLUMNS 91 ‘The elements of A, are the same as those displayed beneath Eq, (2-18) except that in this case, the following expressions for p,, must be used for plates p and q from which sidestreams W, and W, are withdrawn pale Ay Le W, G8) % pun lt gt ‘When model 2, shown in Fig, -2, is used for the feed plates, the column vector f; contains the vapor and liquid rates (vp, and [,) for each feed. The elements of ¥, and / may be displayed as follows We Ede oan Oa Byte Bye een Ba OM fo= [0 Orit 00 tyes OF o ‘The elements oy, and beac, Op and Pers Oe 1y1 AN Oras f= Lt 1st, respectively ‘The component-material balances Ay;= 71, may be solved for the vapor rates and d, by use of the recurrence formulas given by Eqs. (2-20) and (2-21) in the same manner as was demonstrated in Example 2-1 ‘After the component-material balances have been solved for the component- flow rates dj, Bais Psi. ---« By the corresponding set of flow rates for the liquid may be calculated by use of the equilibrium relationship given by Eq. (2-10). ‘Then the @ method is applied for the purpose of finding a set of terminal- component flow rates which are in component-material balance and in agree ‘ment with the specified values of the total-low rates of the terminal streams. and fr, lie in rows Formulation of the 0 Method of Convergence ‘The formulation of the @ method of convergence for a complex column follows that originally proposed by Lyster et al First, a @ multiplicr is defined for each stream withdrawn from the column which may be specified independently. For the column shown in Fig. 3-1, any three of the four streams D, W,, W,, and B may be specified independently. For definiteness, suppose that D, W;, Ws are specified. Then B may be found by an overall material balance. The @ multipliers are defined by the following equations (6-10)92. FUNDAMENTALS OF MULTICOMPONENT DISTILLATION. where 85 >0, 0, > 0, and 0,>0. Again the corrected rates are those which satisfy simultaneously, the specifications D, W,, and W, (where B is taken to be the dependent variable) and the overall material balance FX, = (did + (Wider + Waihe + (Pade G11) where FX =FiXut Xa ‘A combination of Eqs. (3-10) and (3-11) yields the following expression for the corrected distillate rate — FX; — + Dolbildiba + Ox(wis/de + O3(Wai/d x Again, in a manner analogous to that shown for conventional columns, ‘numerical problems resulting from the divisions by small values for (d;),, may be Avoided by use of the following expression for (d,). (eo = (dideaPs (3-13) (dee G12) where p, is defined by ———EE (aida + olbahe + O10 a + 820 a:)en ‘When the defining equations for the 6's given by Eq, (3-10) are restated in terms of pj, the following expressions are obtained for the calculation of the corrected terminal rates, namely, G14) (dee = BolbideaPi (wade = OC addaPs (3-15) (wile = O2(W2cleeP: ‘The requirement that the corrected rates satisfy the specifications D, W;, and W, leads to the g functions Golo 05-03) § (dle = D 91(80, 81, 02) (3-16) 44100 03)= § she ~ Ma ‘The desired solution is the set of positive values of @, 8,, and 0 that makes 0 = 91 = 92 = 0, simultaneously. When the solution set of 0's has been found, the set of corrected ds (or p's) will have been computed. These values of (da. ‘and p; may be used to compute the set of improved compositons. In order 0APPLICATION OF THETA METHOD OF CONVERGENCE TO COMPLEX COLUMNS 93, reduce numerical difficulties, in the application of the @ method, the g functions should be stated in normalized form. For example 9% bE heh Corrected Mole Fractions Expressions for the corrected mole fractions for the column shown in Fig. 3-1 ‘may be developed from first principles as outlined in Prob. 2-15. The expressions so obtained are the same as those given in Chap. 2, namely xy — Lltaldee gg = Hida G17) Elida Y (dati where (dj). is given by Eq, (3-13) and p; by Eq, (3-14). The formulas for y,, are obtained by replacing J, in the above expressions by 0, Calculation of a Corrected Set of Temperatures by Use of the K, Method Alter either a corrected set of xy’s or y’s has been computed as described above, a new set of temperatures may be computed by use of the Ky method in the same manner that was described in Chap. 2 for conventional distillation columns. Calculation of a Corrected Set of Total-Flow Rates by Use of the Constant-Composition Method and the Total-Material Balances ‘The development of the expressions for the constant-composition form of the enthalpy balances is carried out in the same manner demonstrated in Chap. 2 for conventional distillation columns. For example, the energy balance enclosing the top of the column and some plate j located between plates p and f — 1 of the column shown in Fig. 3-1 can be expressed as follows E lero sitien hile byes — Hox ~ (18) Elimination of, by se ofthe componcntmatral balance Pye, It Wit dy (3-19) yields Seles hale peas bade Uys e~ Hol] ~ 2e=094 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION Since ly = jx, Wy, = W; Xpn di = DX py the above form of the energy balance may be solved for L, to give’ = WY (Apsara +117 Ho)X n+ Oc oy, (3-20) Eres hades The complete set of enthalpy balances for the column which are obtained in a manner analogous to that demonstrated above. After the Ljs have been computed by use of these enthalpy-balance expressions, the corresponding vapor rates may be computed by use of the total-material balances. Geometrical Representation of the g Functions To simplify the geometrical representation of the g functions, consider the case where only one sidestream W, is withdrawn in addition to the distillate D and the bottoms B as shown in Fig. 3-4. For this column, only two multipliers exist, 0p and 0). These two multipliers are defined by the first two expressions of Eq, (3-10). 1 D and W, are specified, then the corresponding g functions are dela, 0.)= ¥ (dla = D 621) 91(805 81) = ¥ (Wridee — Me (3-22) tere Fx, (Ne TF Bea + BA Again, (d,).. may be stated in terms of p, as demonstrated above by Eqs. (3-13) and (314). — ‘Traces of the functions gy in the 505. Gad;. and the 0;, planes are shown in Fig. 3-8, and traces of gy in the Gy G10, and Oy, planes are shown in Fig. 3-6. The desired solution isthe intersection ofthe traces of gq and 4 in the 86, plane where go = gy = 0 as shown in Fig. 3-7. Examination of these graphs reveals that gy and g, are continuous, monotonic functions forall positive values of 0g and dy, Such behavior is desirable in the numerical solution of problems. ‘The set of @' that makes go = g; = 0 may be found by use ofthe Newton- Raphson method? as described in App. A. The Newton-Raphson method con- sists of the successive solution of the equations corresponding tothe linear terms[APPLICATION OF THETA METHOD OF CONVERGENCE TO COMPLEX COLUMNS 9§ isp mi pel Figue 34 4 complex column with one sdestream.96 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION a2) 29(001% 18 Figure 38 Traces ofthe fnetion (0 0) in the 5 Fos. and ve Tey planes 26H _ a torer Figure 3.6 Traces ofthe function g, inthe 3505,[APPLICATION OF THETA METHOD OF CONVERGENCE TO COMPLEX COLUMNS 97 Figure 3:7 Toes of he fonctions gy and 1,19 the 80 plane 8, ‘of the Taylor series expansions of the functions go and g, abcut the assumed values of the variables 8, and 0,, namely = 9g + 288 ang + 280 = 50-4 pe Ao + Spe AO (323) 20 2a ont ge Mo + Expressions for the partial derivatives are given in Prob. 3-1. The Newton- Raphson method is illustrated by the following example. Example 3-1 Make two trials by use of the Newton-Raphson method for the set of values x and y which make f, = fz = 0 simultaneously Ais ae y el Fs ax ty 2 Take einai ofa tte =o sous Tel ory tie 1 1t=t etjmtvi-20 and th fz Lary Ban Bary Ths for sg» yt Newton Rpon equations tone on 14241-24y 004281424) 2 of these equations yields O=1+4ax Ax=-1/4 and xy xo t Axe 1/4 = 3/4{98 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION Substitution of this value of Ax into either of the Newton-Raphson equa- tions yields ay= 4 and Yi Yot Aye 1+ 1/4 = 5/4 Trial 2. For x, = 3/4 and y, = 5/4 (3/4, 54) = (3/4)? — (5/4 + = fi(3/4, 514) = (3/4)? + (5/4)? — 2 Thus = (2)3/4) = 3/2 7 ON) = 32 Ha (2514) = 52 % (25/4) = 572 and 0= 0+ (3/2) Ax ~ (5/2) dy = 1/8 + (3/2) Ax + (5/2) Ay Solution of these equations for Ax and Ay yields Ax = -124 x, =3/4— 1/24= 17/24 = 0.7083, y= 1/40 yz = 5/4 — 1/40 = 49/40 = 1.225 ‘These values compare well with the solution set of values which are L <= Jg 07011 and y= /3B= 12247 Convergence Characteristics of the 0 Method of Convergence for Complex Columns To demonstrate some of the numerical characteristics of the @ method for com- plex columns, Example 3-2 was selected. The statement of this example is given in Table 3-1 and the geometry of the column is depicted in Fig. 3-4 In addition to the distillate and bottoms, the column has one sidestream withdrawn. Thus, the @ method has two 8 multipliers, 0) and 0, which are defined by the fist two expressions given by Eq. (310). Since D and W, are specified (see Table 3-1) the two g functions to be used for computing these 0s are given by Eqs. (321) and 6-22), ‘The convergence characteristics of this example are presented in Table 3-2, and the solution sets of Ts, V/s, d'5, w,’s and bj are presented in Table 3-3. ToAPPLICATION OF THETA METHOD OF CONVERGENCE TO COMPLEX COLUMNS 99 ‘Table 3-1 Statement of Example 3-2, 2 complex distillation problem Fx, Component (0 mot) cH 20 cally 00 cme 60 ct 125 ry 35 CH 150 152 m3 30 as 0 Other specifications “The coun presnire i 30 in? abs. The K values and enthalpies are given in Tables B- and BD The feed enters the columa on plate f = 6 at is bubble-poit temperature at the Column pressure. The cof has 1 plates, a reboile, and a part Sita atthe rate B= 32298 [biol The sidstcam i withdrawn fom pate p= 10.58 Tiida the rate Wy, = 250 mol A ref ato L,/D = 225i to be used, Take the initial {Choperattes to be linear between, T, = 200° and Tj. ~ 200°F, and take Ty = 200°F and ‘To JOO" Take the ital values of LY, = L for De ee Se ‘Table 3-2 Convergence characteristics exhibited by the 0 method inthe solution of Example 32 ‘Temp. of Temp. of Calculated Calculated Teal 4 DP BF valueofD valve of oe oe 1076s 0862279029304 24.4287 2 ttetes—LeeIs S13 aA aRI93 247670 5 ogeans? 96261311425 ars 320d 247236, & Gms) tozrat 114336 323555250021 S$ deeme —o9ep619 1143 aAS6S_— BR DTTAs 2495864 6 1007s 1006038 4653323102508 4 agents osy7367 fos 322901724982 & oor LooL3a8 wea 3230084250018 9 asppiaz2 0999384 446393228664 249829 10 100344000269 ‘on 3229861 25.0039, 109998121 a9ee87=9 deal (3279772249945 1210000 .0000 est 298112499993 Convergence fiterion: (1) For each tra wales ofp and, were found such that bot [ol and [| wete equal to or ls shan 10"? (2) Cnvergece for the problem was sid to have bee Achieved when jo(, 1] and [C1] were each equal fo ores than 10° “The convergence eile were satised atthe end ofthe 12h til and 281 seconds on an AMDATIL #70 V6 computer withthe WATFIV compile.100 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION. Table 3.3 Solution of Example 3-2 1 inal temperature, qui, and vapor profes 7, 4 ¥ Sage cry Cb moyn) th mot) i nan 267 a8 2 ass sim 37 3 16540 500 103.0 ‘ is3a7 53993 10038 5 203 19495 91391 ‘ 2930 21235 amas 7 25355 mas 1446s 5 26468 36 15475 5 Deis m3 159.66 10 ann 19697 13961 " 51929 18610 1547 2 3618 tant 1839 B a1 am 12201 2. Final pdt ow rates 4 wu % Component (tb molt) (maim) (mot) cH 030000 10! 052125 10-* ana «10 ct, (095585 «10! 15077 x10"? 01926. 10-* cm (039356 «10! 61044 10-1 033848 10°? cm O12 x10 2a (018250 x 10-* ee 011910! O97 10! O40 ot 0102210" 1042 x 038483 « 10 mH, 036429 10°? 064835 x 10! 087129 x 108 mH, 018929 10"F 028282 x osr747 x 108 mes O1s019 <10rF REID x 07618 x 108 mH, 012374 10-F 010869 « 074031 « 108 ‘eo O46 x 10°" 040727 i927 x 108 Qc = 047283 « 10 Bram 0g = 015519 x 107 Bayh satisfy the convergence criterion stated in Table 3-2, a total of 12 complete trials and 281 seconds of computer time (AMDAHL 470 V6 WATFIV) were required, Exact Solutions Given by the @ Method of Convergence for Complex Columns For a distillation column such as the one shown in Fig. 3-4, the @ method Constitutes an exact solution at total reflux; that is, the sets of b/s and w, /ds at two different sets of values for D and W, are related by the single multipliers{5 and 0, as indicated by the frst two expressions of Eq. (3-15). The proof of this statement is established by solving Probs. 3-3 and 3-4. That the 0 method is an exact solution to certain problems at total reflux is further illustrated by solving Probs. 3-5 and 34. 32 SYSTEMS OF INTERCONNECTED DISTILLATION COLUMNS The particular difficulty associated with the solution of problems involving systems of distillation columns becomes apparent upon close examination of the system shown in Fig. 38. First observe that the composition of the recycled bottoms from the second column to the first is not known until a solution to the for | a Figere 38 A system of wo interconnected distilation columns102 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION second column has been obtained. On the other hand, the composition of the feed to the second column is not known until a solution has been found for the first column, since the feed to the second column is the bottoms of the first column, For solving problems ofthis type, Tomme and Holland"? and Tomme! pro- posed an extension of the @ method of convergence, called the “ Capital © Method for Systems.” The “capital ©” is used to distinguish this method for systems from the 6 method for single columns. The capital © method for systems is similar to the @ method for complex columns in that for each external stream which may bbe specified independently, there exists a © multiplier. These © multipliers are to be picked such that the overall component-material balances and the specifications are satistied simultaneously. For each component i, there exists fone overall component-material balance per column which is independent. The strong convergence characteristics of the © method for systems of inter- connected distillation columns can be attributed to the fact that there exists many problems for which the © method for systems constitutes an exact solu- tion for a system of columns at total reflux. ‘The capital © method for systems is introduced by the formulation of the ‘equations for this method for the simple system of two conventional distillation columns shown in Fig. 3-8. In a subsequent section, the formulation is gener- alized for the case of any number of columns with any number of sidestreams withdrawn, Formulation of the Equations for the © Method for a System of Two Distillation Columns AAs shown in Fig. 38, the number of the column from which a stream is will drawn is denoted by a subscript on D and B. To identify the variables of each column, the column number is carried as the last subscript. For example, d, and d;, » denote the flow rates of component iin the distillate streams withdrawn, from columns 1 and 2, respectively. For the system of columns shown in Fig. 38, there exists one independent component-material balance per column, namely FX, +b, —dis Baa burda bur In the interest of simplicity, the subscript co used to identify the corrected flow rates for conventional and complex distillation columns has been omitted in the above expressions and from those which follow. For each total-flow rate of the set D;, By, D3, and By which may be fixed independently, there exists a capital © multiplier. If, is fixed, then D, is seen to be dependent since it is uniquely determined by a total material balance which encloses the entire system. From the total material balances enclosing ‘each column, itis seen that if B, is fixed, then either Ds, D;, or B, may be fixed (3-24)[APPLICATION OF THETA METHOD OF CONVERGENCE To COMPLEX COLUMNS 103 ea ep bas (bi. (is), -e(is), lis), where the calculated values are those found by use of the caleulational procedure described below. The subscript co is dropped in the interest of simplicity in the remainder of the development. The g functions may be stated in terms of any two of the flow rates Dy, Dy, By, and B, which may be specified independently. Thus, if B, and D, are clected as the specified flow rates, then (3.25) 41(01,O2)= Sb, . (3-26) The desired set of 6's is that set of positive numbers that makes gy = g2 = 0, simultaneously. These @'s may be found by use ofthe Newton-Raphson method in a manner analogous to that described in See. 3-1 for complex disilation columns Formulas for {d,.} and {d,} are obtained as follows. Let the expressions siven by Eq. (325) be restated in the form aia (3-27) bam riadia ea where ra = Oy(b. lds, thee Oxfbi, 2d, 2)eu After b,,. and b, ; have been eliminated from the component-material balances by use of the relationships given by Eq. (3-27), the equations so obtained may be rearranged to give (ht rab + rads, noida —(14 Fak = 0 This set of equations may be solved for the corrected distillate rates to give FX, (28) FX(L+ na) din fee rancaere (3-29) a, = Pet (0) Teritha104 FUNDAMENTALS oF MULTICOMPONENT DISHLLATION Figure 39 Behavior ofthe function g in the seighborhoad of the postive ‘oo Graphs of the functions g, and g> in the neighborhood of the solution set of @'s are shown in Figs. 3-9, 3-10, and 3-11. An outline of the proof of the existence of 4 set of positive ©'s that satisfy the g functions simultaneously is presented in Prob. 3-13. A proof for a more general case has been given by Billingsley." ? Calculational Procedure ‘The calculational procedure recommended by Nartker et al consists of apply ing the @ method of convergence (for conventional and complex distillation columns) one time to each column of the system in succession. The terminal Fates so obtained for each column of the system are called the “calculated values for the system,” and denoted by the subscript ca. To initiate the calculational Procedure for the system, the compositions for the minimum number of streams are selected as the independent variables for the system. For example, to initiate the calculational procedure for column 1 of the system shown in Fig. 3-9 the ‘composition of the stream B, is assumed. One complete column tral is made on column 1. A complete column trial consists of the application of the @ method for single columns, the calculation of a new set of temperatures by use of the Ky method, and the calculation of a new set of total-flow rates for use of constant. composition method. These temperatures and total-low rates are stored for use in the next trial calculation for the system. The set of b, :'s obtained by theAPPLICATION OF THETA METHOD OF CONVERGENCE TO COMPLEX COLUMNS 105 Figure 3.10 Behavior ofthe fonction gs nthe neighborhood ofthe poe res, column trial on column 1 become the set of assumed flow rates for making the column trial on column 2. After the column trial on column 2 has been completed, the capital © method for systems is applied, and the set of b 2's So obtained is used to initiate the next system trial by making one column trial on column 1 Figure 311 Intersections of the suraces ofthe functions g, and g, withthe 8,8; plane forthe case where F= B, > 0.106 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION Sia ) ; os : [w, 3 a ic : : : s |" é Bie g 7 7 7 die Figure 312 Sketch ofa sstem of interconnected dit tion columns. Convergence Characteristics of the Capital © Method To demonstrate the convergence characteristics of the capital © method, Example 3-3 was selected. This example involves the system of two complex ‘columns shown in Fig. 3-12. A statement of this example is given in Table 3-4 and the solution is given in Table 3-5. The variation of the 6's and @'s with trial number is presented in Table 3-6. Exact Solutions Given by the @ Method of Convergence for Systems of Distillation Columns For systems of interconnected distillation columns, the © method constitutes an exact solution for certain problems; provided that the a,'s are constant in each[APPLICATION OF THETA METHOD OF CONVERGENCE TO COMPLEX COLUMNS. 107 Table 3-4 Statement of Example 3-3 Fx, Component No hemor) “Toluene ' oso Ethyenzene 2 05100 x 10? Sitene 3 04777 x 108 sopropylbenzne 4 05000 x 10°" methyl s 01000 x 10° ‘thylbenzene emethystyrene ‘ oxo elpropylbenzene 7 2000 Otter specifications Column 1. The column bas 50 plates, a total condenser (tage j = 1) anda reboler (stage j= N= 52) The feed F enters on plate j~ 10 asa subeooled quid a 31775 K at 270 mami “The dita is withdraw as 2 bguid at ie boiling point atthe rate Dy ~ $2 g mol. The sidesteam is withdrawn asa guid trom plate j= 21 atthe rate W, = Il g molh. The top Product steam of column 2 is recycled to plate j= 45 in column I. The pressure in the ‘Condenser i mmHg and it may be trumed to vary Hnealy with plate amber between age /~ 241 a peste of $0 mmYlg and the rebole at 270 mmllg. A refx rato L,/D = 23 isto be usd. Use the equibrium and enthalpy data given in Tables B-3 and B-4 Inia temperature profile. Linear betwscn 525.15 K (the boiling poit of pre ethyben- ene at 40 mmf) and 381143 K (the boiling point of pare styrene at 270 mB) nial sapor rates. Vj = 1820 g moth = 2.3, 52) nil ligt rates. L, = 1300'g moth Y= 1, 2-9} Ly = 2300 g moth j= 10,11, 21), 1, = 2190 g moth = 22,3,» 4 L, = 2380 g moth U = 45 46. 51) ‘Base component. Base the values of «and b inte expression for Ky given by Eq (237) on the Kvales for styrene at 298.15 K and 40 mmHg and #700 K and 370 mmlle Column 2. ‘The “other specications” for column 2-ate the same as those tated for column 1 except fr the folowing tems. D; = 16 g mol/e, Wy=7 g mols. The siesream 1; withdrawn asa gui fom plate) = 2. The bottoms B, of column 1 is fed to column 2 fon plate j= 3. The pressure in the condenser is 0 mmblg, ad it may be assumed to Vary linearly with plate number between sage 2 at a pressure of $0 mim and the reboller at 250 mmHg A feux ratio of Ly /D of 28 to be vse Trial temperature, eapor, and lgud rate profiles. Same as sated for cole 1 nial composition profesor he recycle steam Ds. To iia theft el for eolums 1 assume that ese all ofthe ethylbenzene in the feed leaves inthe stream D,, and take d= Dy — FX, 2-8 ‘Assume the reeyele stream D, is composed mostly of styrene, and Ape Dy dy y= 61s Take das das aa dns=0.108 FUNDAMENTALS oF MULTICOMPONENT DISTILLATION Table 3-5 Solution of Example 3-3 | Final temperature and vapor rate profes for coum 1? 1 % 1, ¥ Stage 0K) the mom en 1 2669 73am mao 2 saat Fa daos am 3 3450 » sess 0473, 4 53693 x0 3637503 3 nese n 36002570 ‘ S06 2 sero1 205 1 300 3 3078 2587 5 3383 aM Sasi 20613 9 34828 3s 3010 20638 10 34866 3% 30 er " san96 spas 2686 2 5919 sis 20708 6 38035 smsi 228 4 3st sna amas 5 35256 snr 2766 6 33300 snss 2783 0 38461 suse mse ® 33859 wise oa » 38654 7598 20825 » sia? mises 21197 2 53636 sm38 21210 2 33924 ysis 222 a 360.10 sm91 22.50 Ey 36093 spn 21236 25 361.75 st boss 21239 26 3258 2 baas 21237 Solution sets of product ates for column 1 4 we ba Component (kg mal) (8 mol) (morn) 1 08000 S816 x 10°9 —_OeTET x 10 2 040954 x 10? 5781S x 10! 10108 «102 3 010248 x 208 (051932 x 10! 0.42530 x 102 ‘ 020858 x 10°? 0.45860 10" —_045420 x 10-* 5 020724 x 10+ 072101 x 10-?——_0.9as8t 10? 6 (051283 5 10 0x77 x10"? as 7 041536 x10 © ONITTS x 10° o1se7t 4+ The convergence criteria forthe g fonctions gy and g foreach column were the same 35 stated in Table 33 for Example 31, except that fn this ease the eritrion of 10° instead of 10" was used. 1 = 018268 x 10! calm x = 020182 x 10" calm[APPLICATION OF THETA METHOD OF CONVERGENCE TO COMPLEX COLUMNS 109 ‘Table 3-5 (continued) {UL Final temperature and vapor rate profiles for cola 2 x 4 1, % sage (K) he molt) sug (K) ie mot) 1 33055 16000 a 3639 S608 2 33613 S600 28 Msi 56290 3 sue27 Sasi » 3658356236 4 2016 Seats x» sees S600 5 MLS s6ans a 36724 S616 6 sai s6a72 2 S619 52405 7 sass sears 3 Bess) 52564 a M4622 Soars au seat 20 ° BaTsi s6as2 8 SepdS2834 10 Mas seas 6 smosT 52966 ul oon S6a8s 3” srios $3095 2 3510556450 a mist S771 B ss 56492 » sma sh34s 4 389 $6052 » 36 S3467 a 340 56092 a mas 51547 6 3351956490 a yun 53.703 ” sere S09 8 mass S38t8 6 35707 seat “ msis $3990 ® 35796 seas 6 msm 54038 » 33684 Sea “ 3629 S445 a 53964 seas o 68854287 2 302 $640 * sae Saas 236133 s6an7 ° mos 54039 Fs Serie Sea04 0 whee 54507 3 3623956300 5 3929 $4006 6 3636556337 2 S001 54665, 1V. Solution sets f produc ates or column? dus was ba Component (¢ mol) femoim) (mot) fl 01968 10" OTBHBR x O-!——0192 « 10-* 2 05799 x 108 012868 x 10 020208 «10 3 0.10198 « 10 05707910 036624 «108 4 021138 10-2 OSH192 «10-2 057496 10! 5 038839 10 021680 x 10"? 080409 «10°? ‘ ©1979 x 10-* 2231 x10"? 1983 Qc += 057000 « 10° cam Ox : = 052253 x 10" cathHO FUNDAMENTALS OF MULTICOMPONENT DISTILLATION ‘Table 3.6 Convergence characteristics of the © method for Example 3-3 1. Colum | Colum 1 Colum 2 Tei a 4 ® r pan 94863 30038 2 oop1383 o1ta6s ox oso 2 2omt 2am 1403 12106 4 Os6381 oats sens i891 5 13090 13016 10657 10892 ‘ o8s68 sez 096894 osrs97 1 10902 104s 10165 10126 4 oasis ose oss osesi2 9 1.066 004s 1.0022 1.0017 0 oper 099741 osm 095835 4 10012 001 1.004 1.004 B osmss 099935 osser3 095580 B 95505 1000 1.0001 1.0001 10 10 10 10 2 system O's Teil °, 8 8, t 1m 10173 2 11s 10238 3 oseri6 spar 4 10161 10099 5 1.006 1.004 ‘ assis ose, 7 1.008 1.006 4 10 10 9 oss584 095589 0 10 10 099953 osser3 2 10 10 5 to 10 3. The convergence criteria were stsed at the end of the Ith teal. 1684 seconds of compoter time were required on an AMDAHL 470 V/6 computer with a WATFIV compiles, ‘The convergence ctiteron forthe system was thatthe euclidean normafheg fonctions forthe system 1074 that, ea ies 4 PP c10-* when the g functions are evaluated at @, = ©, = @,= O,= 1column. A proof ofthis statement is established by solving Probs. 3-7 and 3.8, and a demonstration of the truth of the statement is alflorded by solving Probs, 3:9 and 310 Systems Containing Mixers and Proportional Dividers The @ method of convergence for systems is readily extended to include other types of units such as mixers and proportional dividers which are commonly found in systems of distillation columns. To demonstrate the application of the © method to systems containing units such as these, consider the system shown in Fig. 3-13. Suppose the specifications for the system are taken to be D3 and By, In addition to these, specifications such as the total-flow rate, the thermal condi- tion and composition of the feed F, the reflux rate the column pressure, the type of condenser, and the plate configuration for unit 2 are made. The remaining flow rates, D,, Ds, and B, are computed from the set of three overall material balance equations. For any component i, the component-material balances are tows mo Waa) fi paren 8 Mel--fe] oan 0 atolls When only one stream is withdrawn from a unit such as a mixer, there exists no question about the product distribution for the given unit. For any choice of inputs to the mixer, the corresponding output is uniquely determined by the component-material balance for the mixer (FX,=d,1—r, 3d, 5) Thus, for a mixer witha single output, a © does not exist. Furthermore, © for the propor tional divider is unity. Thus, r,y may be set equal t0 (6, 5/4, shas and the unknown ©; found by use of the following g function 9= Eltsle Ds 6x) where the value of (d, 2 is found by solving Eq, (3-31) ah fp % fmm war? wer i bsmson ron | coun Diver C2 J bs Figure 3.13 A system containing a mixer, lation column and proportional divide.112 FUNDAMENTALS oF MULTICOMPONENT DisMLLATION 3:3 A GENERALIZED FORMULATION FOR A SYSTEM OF COLUMNS IN WHICH THE TOTAL-FLOW RATES OF THE EXTERNAL STREAMS ARE SPECIFIED To illustrate the formulation of the capital © method for the general case of a system having any number of columns interconnected by any number of recycle streams, consider the system of four columns shown in Fig. 3-14. The formula- tion is simplified by the introduction of a new system of nosation, For each column added to the system, there exists an additional capital © ‘multiplier, and each sidestream withdrawn from a column leads to an additional © multiplier. Thus, the number of © multipliers is equal to the number of columns plus the number of sidestreams. The number of g functions is equal to the number of © multipliers, and the number of overall component-material balances to be satisfied by the solution set of the ©'s is equal to the number of ‘columns. The treatment presented below is restricted to systems of distillation col- ‘umns. Systems containing both absorbers, strippers, or reboiled absorbers and distillation columns may also be solved by use of the capital © method of ‘convergence as demonstrated in Chap. 4 AAs shown in Fig. 3-14, the column from which any stream is withdrawn is carried as a superscript, and the column to which an independent feed is in- ‘troduced is also carried as a superscript. This notation follows closely that developed by Nartker et al.'? For the system of four columns, there are four independent overall material balances for each component, namely FIX} +d? —d! — wh, —whi=0 war~ d? ~ wi, wi = 0 Pxp dt hwo ee) what wh, —hi=0 EAE fede 4 “ e Figure 314 A genera sytem of fur interconnected separation unis,[APPLICATION OF THETA METHOD OF CONVERGENCE TO COMPLEX COLUMNS 113 For the system of columns shown in Fig. 3-14, two independent multipliers per ‘column exist, and they are defined as follows Bie oy() (1.2.3.4) ae (34) eof) wonaae x, more compa, ) (k= 1,2,3,4;5=1,2) (635) where k denotes the column number and s denotes the sidestream number. ‘To avoid the numerical problems resulting from the division of the w;s by small to zero values for one or more of the ds, the following definitions of the 1/8 ate used instead of those given by Eq. (3-27) T= Olea (k= 12, 3,45 9 = 1, 2) (3-36) ‘The corrected values of the w,'s for each column may be resteted as follows by combining Eqs. (3-35) and (3-36) whe riot (k= 12,34; 5= 1,2) 337) where Ha fn (K= 1.2.3.4) By use of the relationships given by Eq. (3-37), the component-material balances [Eq. (3-33)] may be restated in terms of the p's as follows PIXE + (i)eaP? ~ [doa + rhs + rip! = 0 Hip) — (dea + s+ Ap? = 0 PPX} + (dap? — (dea + 72s + Php? = 0 (3-38) Tht — Udi ea + rhs + ridpt = 0 ‘These equations may be represented by the matrix equation n= fi (39) where -Ri jg 0 0 pt Fix? nef? -RF Boo] yall ya] 9 ) 0 0 = RP (tty de ex? AO 0 RE Pt ° Ra@at hth (k=234)114 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION Again the number of external streams which may be specified independently is equal to the number of columns plus the number of sidestreams withdrawn (in addition to the distillate and bottoms). Thus, for the system shown in Fig. 3-14, only eight of the twelve external streams may be specified independently. Furthermore, care must be exercised in the selection of the streams to be specified in order to obtain an independent set. For example, an examination of this system shows that all of the W's may be specified independently, and for this set of specifications, the normalized g functions become a= S(whsWe)— 1 (k= 12,34; = 1,2) (3-40) and in view of Eq. (3-37) the expression represented by Eq. (3-40) may be restated in terms of the 1’ and p/s to give a= SU oW-1 k=. 345 2) Gt) After the last column trial of a given system trial has been performed, the capital © method for systems is applied to find a new set of product-component- flow rates which satisfy all of the system component-material balances simulta- neously while being in agreement with the specified values of the terminal flow rates W and 4 (k = 1,2, 3,4). The solution set of p/s is determined by finding the set of positive O's that make gf 0, (k= 1, 2, 3, 4) simultaneously. This set of @'s may be found by use of the Newton-Raphson method; the ‘equations for this method are represented by Js0=-¢ (3-42) where the jacobian matrix J and the column vectors g and A@ have the follow- ing matrix representations at aot J oa 603, 2 gigi 93)" 40 = [40] -- A@t AE} -- AOS” After the functions and their partial derivatives with respect to the @'s have been evaluated, Eq, (3-42) may be solved for A@ by use of any one of many procedures such as gaussian elimination.» ‘The functions and their partial derivatives which appear in g and J, respec tively, may be evaluated by any one of the techniques described in Chap. 4. Also, the matrix equation (Eq. (3-42)] may be solved by use of any one of the matrix techniques described in Chap. 4[APPLICATION OF THETA METHOD OF CONVERGENCE TO COMPLEX COLUMNS 115, NOTATION (See also Chapters 1 and 2) thedue ‘molar flow rate of component iin the distillate from ‘unit number k DAD, total molar flow rate ofthe distillate from unit k Fe total molar flow rate of the independent feeds to unit k Go.9 9 functions of the 0 method for a single column a 9 function for terminal stream number 5 of unit k 8 column vector of g functions; see Eq. (3-42) J Jacobian matrix; defined by Eq. (3-42) ™ (dio /(dibei See Eq. (3-13) vt Gd P ‘column vector ofthe p's rn OX{w))s Value of r for stream s of unit k a ‘a square matrix; defined by Eq. (3-39) ws {otal molar flow rate of sidestream s from unit k i ‘molar flow rate of component i in sidestream s leaving unit k w, total molar flow rate of sidestream s leaving a complex column saci S number of the first feed plate; see Fig. 3-1 ? ‘numberof the plate from which the ist sidestream is withdrawn; see Fig. 31 4 number of the plate from which the second sidestream is withdrawn; see Fig. 3-1 ' number of the second feed plate relative to the top of the column; see Fig. 31 sapere k ‘column or unit number reek etre 6 ‘multiplier for the 0 method for single columns oe ‘multiplier for sidestream s of unit k of a system Mathematica Symbol Ger x2 xaJ7 transpose of a row vector; the transpose of a row vector is equal to a column vector, [x1 x2 el116 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION PROBLEMS 341 For the cate of 4 complex ditilation column which has one siestream withdrawn in addition {othe dsl and bottoms, show that he partial derivatives are given by the following expressions t) TAL Oe + A Da? dak X, OE + n+ LP lb aFX, The alba + Olde ult + Ta 06 )e+ [3.2 (a) Formulate the constant composition form ofthe enthalpy balances forthe cormplex column shown in Fig. (6) Develop the total material balanoes forthe complex column shown in Fig 3 ‘33 Suppose that the complex column shown in Fig. 3-4 isto be operated at total eux at ite and ‘nonzero values of D, Band W, Further suppose thatthe cokimn has a oatial condenser and that the relative volatilities remain costant throughout the column. By w= of he same approach eed 10 0. ‘Step 1! Construct the graphs shown ia Figs. 3-8 and 346 for gy and 9, in the neighborhood where both 8p and are positive ‘Step 2 Show that gy alvaye has a trace inthe TD; plane. Step 3 Show that gy always has a trace inthe Joy plane. Step 4 In otder te prove tht a point of interetion exists for every pir of fonction, i sulicen 10 show that ach Im order to prove this, verily the following relationships and then employ them as required. (0.0) = 84%, f(0.0)| = 4, 312 Obtain expresions forthe partial derivatives ofthe g function given Eq, (1-26 for the sytem ‘own in Fig 38 2313 Show that itis always possible to nd ase of po rmethod forthe system shown in Fig. 38 lie © forthe g functions of the capital © 1. Show thatthe trace of he function g, inthe B; 5; plane exist for al ets of postive and firite values fe 8, and @, 2 Show thatthe ace og exist foal ite and positive values of ©, and for al postive values of (greater than the asymptoti ale. 3, Show that pont ain Fig. -1 is always tthe lt of point b that sa 0 to the left of the minimum is used, Newton's method will predict a negative . The customary procedure used in n-dimensional space consists ofthe successive reduction of the corrections (AY? for the function P(¥)] by factors of 1/2 until positive values of the variables are obtained. For the function P(¥), this prooedure fails because all values of to the lft of the minimum are outside of the region of convergence far the positive nonzero root. In this ease, trials would be made at successively smaller values of Y and the trivial solution ¥¥ = 0 would be approached as this procedure is applied indefinitely. To obtain the desired solution (the ¥ > 0 which makes P(Y) = 0) a new starting value which is to the right of the mini- trum value P(¥) must be selected. Selection of an initial set of the variables which are in the region of convergence can prove difficult for n-dimensional problems unless the functions are very carefully formulated ‘Another type of serious difficulty arises when f(x) exhibits the following type of behavior. Suppose that after having passed through the x axis at some x > 0, it then passes through a minimum (or a maximum) and then approaches 72r0 asymptotically. Thus, for larger and larger values of x to the right of the minimum (oF maximum), the function f(x) becomes smaller and smaller. The function in Prob. 4-5(b) behaves in this manner. Although itis difficult to deduce128. FUNDAMENTALS OF MULTICOMPONENT DISTMILATION the behavior of a function in n-dimensional space, the traces of the function can bbe examined in two-dimensional space and an attempt should be made to formu- late functions whose traces are monotonic. Reboiled Absorbers ‘The sketch for a typical reboiled absorber is shown in Fig, 4-2. To demonstrate the formulation of the 2N Newton-Raphson method for reboiled absorbers, two different sets of specifications are considered. ‘Specification set 1 P, F, {X,), thermal condition of F, Lo, (xo, To, f, N and Qy In order to solve a problem of this type by the 2N' Newton-Raphson method, Vf (LEAN Gas) u 2 1 (Lean oi) 2 EF f-4 —____} (RICH Gas) f 41 Nez (eorroms) Figure 42 Steich of eboiled absorber{THE 2N NEWTON-RAPHSON METHIOD 129 the following sets of 2N independent functions are selected X= 0, 0, Oy T Th TT fe [Fy Fy Fy Gy Ga Gal! The matrix A; of the component-materia balances are of the same general form as the A, given by Eq. (2-18) for distillation columns. When plate 1 is assumed to behave according to model I (see Fig 1-13) and plat fis assumed to have the behavior characterized by model 2 (see Fig. 2-2), the eomponent- material balances may be represented by Eq, (2-18); provided that the elements of f are taken to be the following set A= Ul 0.0 0 vp fy, 0-0} (414) where vy, lies in row f—1 and I), lies in row f “The second set of specifications differs from the frst in that the boilup ratio V,/B is specified instead of the reboiler duty. Specification set 2 P, F,(X;) thermal condition of F, Lo. {sor Tox fo Ne WB For this set of specifications, the 2N independent variables are given by X= 0, Op O4 Th Te Tras Th Oa? (415) and the 2N independent functions f ate the same set listed for Specification Set 1. The N vapor-liquid equilibrium functions are given by Eq. (4-3), and the enthalpy balance functions are given by Eq. (44) forall stages except j = f—1,f N. For stage f— 1 and f, the functions G,_, and G, contain the additional term Sout, and Shahn in their denominators, respectively. For stage N, the normalized form of Gy is given by YlowAl + obs) z = (416) El nitend + Oe From the sketches of conventional distillation column (Fig. 2-1) and a reboiled absorber (Fig. 42) it is seen that the geometrical configuration of a conven- tional distillation column is obtained by replacing plate 1 ofthe reboiled absor- ber by a condenser-accumulator section (stage 1) and by eliminating the feed Lo The condenser-accumulator section is assigned the stage number 1, and when the condenser duty Qe is specified the independent variables corresponding to this stage are 0, (where 0, = Ly D) and T,,10 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION ‘The matrix equation representing the component-material balance is again fiven by Eq, (218), and the elements of the matrices A,, vj, and 7% have the meanings stated below Eq, (2-18) For a column having & partial condenser, the dew-point functions are given by Eq, (43) for j= 2, 3, .--. N. For j= 1, D and d, play the same role as V, and 1, in Eq, (43) and the dew-point function F, is given by -1)4 (417) ay Dé For a column having a total condense, the bubble-poit funetion for the iste sed for Fy, namely, FBS Gud, (418) xcept for stage 1, the enthalpy balance functions are given by Eqs. (4-4) and (4-16). For stage j'= 1, the normalized form of the enthalpy balance function is given by Eta + hid + @c Leute where Hy, H,, for a partial condenser and Hy = hy for a total condenser. For sages Jo 2, 3, oS—2 ft ly £42, «and N~1, the enthalpy balance functions G, are given by Eq. (4-4). The functions G,_., and G, are formulated as described below Eq, (4-15) and Gy is given by Eq. (4-16). ‘The independent variables for different sets of specifications are listed in Table 4.2. Ge -1 (419) ‘Complex Columns Complex columns were defined in Chap. 3 and illustrated by Figs. 3-1 and 3-4 To illustrate the application of the 2N’ Newton-Raphson method to the solution af problems involving complex columns, consider the simple case where the Sidesteam Wis vithdrawn in the liquid phase from some interior plate p. The withdrawal of the sidestream W gives rise to one specification in addition to those stated for conventional columns, in items 1 through 4 of Table 4-2. When this additional specification is taken to be either the totalflow rate W or the ratio W/L, the sets of specifications, independent variables, and functions for this complex column are the same as those stated in Table 4-2 except that either W, or W/L, should be added to each set of specifications. When W;/L,. i specified, the totalmaterial balance for plate p, Yoo t Lyi HL, ~ Wi = 0 may be restated as fllows RyaNy (14 Rp +bR,) hor =0 (420)‘Tue 2N NEWTON-RAPHSON METHOD 131 ‘Table 4-2 Specifications, independent variables, and functions for conventional distilation columns 1. Specification P, F,(X), thermal condition of Ff. N. Qe. ‘model ofthe ed pate Independent vatabes Oy O35 Op 89s Ty To os Troy Te Functions: Fy, Fy. o1 Fy yo Fys Oy, Opes Guy Gy. Fy given by Eq (417) 2 Specification: P,F, [Xj themal condition ofthe fed Ff N, L/D. VB, and a pariah, condenser Independent variables: QO. Oys Bq 4+ One Ti Tes Fontions Same a em 1 partial condenser, and the Fac Te 2 Speciation: Same as item 1 above except that a foal condenser instead of @ patat ‘condenser iso be wed Independent variables: Same a item 1 Function: Same atte t except F, i giten by Eq, (418) 4. Speciation: Same as itn 2 above except that a foal condenser instead of a partial ‘condense to be wed Independent variables: Same as item 2 Functions: Same as tem 3. Thus, the matrix equation RV = -¥ (Eq, (410)} applies, provided that the clement lying on the central diagonal of row p of Ris changed from (1 + R,) to (14 Ry + (W/E Ry) ‘When W; is specified, the total material balance for plate p takes the form Rye Vonr (1+ Rep + Yoo = Me (421) In this case, the constant W, appears in row p of the column vector ¥. For ‘example, if W, is withdrawn from plate p above the feed plate f, then ¥ is of the form [0 = 0 (-M) 0-0 Ye Le OOF (422) ‘The component-material balances are formulated in precisely the same manner as demonstrated in Chap. 3 for complex columns. Three procedures are presented in this chapter for solving the Newton- Raphson equations. Procedure 1 is presented below and procedures 2 and 3 are presented in Sec, 43 Procedure 1. Solution of the 2 Newton-Raphson Equations by ‘Use of the Calculus of Matrices and LU Factorization ‘Tae anal expressions forthe paral derivatives ofthe Fj and Gs with respect to the 6's and T;s are readily obtained by termwise differentiation of ths unetons. For example ALS (La) a0, VL Ay YO, "20,132 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION ‘The partial derivatives of the flow rates which appear in these expressions may be evaluated by solving the matrix equations given below. The partial derivatives of Fs with respect to the 0,'s may be computed by use of the matrix equation obtained by partial differentiation of the members of Eq. (4-10) with respect 10 any 6, say (k= 1,2,...,.N) (424) where 04, _ [220% am 0, 0” OO. ‘Similarly, partial differentiation of the component-material balances [Eq, (2-18)}, A With respect to T; gives on, Age (425) where 24 fPesiders dons!” ol, (eT, eT, OT, Since the matrix A, has at most only two elements which depend upon a partic lar @ or T, the right-hand sides of Eqs. (4-24) and (4-25) reduce to relatively simple column vectors shown in Table 3 for absorbers. ‘The ekments i these column vectors are found by carrying et the mati operations indeed on the right-hand sides of Eqs. (4-24) and (4-25), The partial derivatives ofthe tla ow rates with respect tothe ys maybe found by use ofthe following expression wheh is obtained by tems patel siferentation of Ea (10) 0 se wv__ a Raa Alter the matrix operations implied bythe righthand sie ofthis equation have been performed, the elements of the colamn matrix shown in Table 4) are hind. ‘The paral derivatives of the Vs with spect the temperatures at all eaual to sero becuse every cement ofthe mata Ris independent of tempon ture. The truth of ths statement is demonstrated y frst ferentiation a each member of Eq 4-10) with espe to any Th to obtain v_ aR am oT (426) V=0v=0 and thus R-0=0{THE 2N NEWTON-RAPHSON METHOD 133, Table 4-3 Elements of the column vectors on the right-hand sides of Eqs. (4-24), (4-25), and (4-26) for absorbers 1. Blemens of G, = (24/204, [eu (Cy) 0 OF Ca (C4) 0 OFF Core 0 C4 HO Cyn 0 Cyr ABR (= 1,2 5) Oa Pal—Ds) 0 OF mm Elements of = —(@R/00)V B,=(6, (-£) 0 oF By = 10 F, (-£,) 0 oF Bey a 0 Ee Co B= [0-0 6)" Consequently Hog Ue bdo) : m 0 k tN) (4-27) Comparison of Eqs. (37), (424), and (4-25) shows thatthe same tridiagonal tmatrix A, appears in the expressions for computing the component-flow rates as appears in the expressions for calvulating the partial derivatives of the o's with respect to the temperatures and the 6's. Similarly, the tridiagonal matrix R which appears in the total-material balances (Eq. (4-10}] also appears in the expression for calculating the partial derivatives of the ¥js with respect to the O's (Eq. (4-26)]. This characteristic gives rise to sereral possible procedures for solv- ing for the component-flow rates, the totalflew rates, and their partial deriva- tives. Three of the possible methods for solving these equations are (1) the use of|134. FUNDAMENTALS OF MULTICOMPONENT DISTILLATION the recurrence formulas for tridiagonal matrices [see Eqs. (2-20) through (2-22)}, (2) the use of inverse matrices, and (3) the use of the procedure called “LU factorization.” Only the fastest of these, LU factorization, is described. The nethod of LU factorization is best deseribed by use ofthe following numerical example, Example 4-1 (Hess et al,"? by courtesy Hydrocarbon Process.) Lean oil at the rate Lo = I g mol/s enters the top plate of a three-plate absorber. Rich as at the rate Vs=2.g molfs enters the absorber at the bottom of the column on plate 3 (see Fig. 43) (a) For the fist tril, take (L,/V)),= 1/2 for all j and take the assumed values for the 0/3 to be 0, = 0; — 05 = 1. Find the corresponding values of the total flow rates V,, Vz, and Vs. Solve by gaussian elimination, and store the negative values of each multiplier in the same location as the zero produced by the given multiplier. (b) Show that the resulting matrix produced by the operations deseribed in part (a) may be restated in terms of a lower triangular matrix Land an upper triangular tmatrix U which are related to the original matrix R as follows Lu=R () Show that the multipliers saved in part (a) may be used to operate on the clements of the original matrix (—#) to produce the form possessed by this matrix at the end of the gaussian elimination process. (d) Show that the ‘multipliers saved in part (a) and the final form of the R matrix at the end of the gaussian elimination process may be used to compute the partial deriva tives for the ¥js with respect to 6, Jk 3 It peer eereceeel‘THE 2N NEWTON-RAPHSON METHOD 135 SoLUTION (a) The total-material balanoes may be stated as follows in terms ofthe vapor rates and the Rj (1+ RM +H; = Lo RM -(1+ Ra) + K=O RVs — (14 R= ‘These equations may be stated as a matrix equation of the form RV=C where =. following of Eq, (4-10) is replaced by C in this example to give the [ (14 Ri) 1 0 I: ['] Ry (14 Ra) 1 Vy 0 0 Rk, -(+R MLK) Le, Ry = 1/2, and the fact ‘On the basis of the values given above, Ry = Ry that Lo = 1 and V, = 2, matrices Rand C become 32 1 oT ffen [ ia an [2 “ -2 0 2 -32 Step 1. To eliminate 1/2 (the first element in the second row) multiply row 1 by 1/3 and add this result to row 2. Then replace row 2 by the result ‘ottained by this addition to give -1 -1p -2 32 10 (-13) -76 1 0° 12 ~372, ‘The negative value of the multiplier 1/3 has been stored in the location where a zero was obtained by use of this multiplying factor. To emphasize that @ zero and not 1/3 is to be used in subsequent row operations, the ‘multiplier is enclosed by parentheses. ‘Step 2. To eliminate 1/2 (the second element of row 3) multiply row 2 by 3/7 and add this result to row 3. Then replace row 3 by the result obtained by this addition and store the multiplier 3/7 as before to give the final matrix equation -2 1 ow) p-a iy aie 1 WI “0 ® 0 (3m) -asna}bey L157 RV=C (©) which may be represented symbolically as,136 FUNDAMENTALS OF MULTICOMPONENT DISTLLATION ‘where R'is the upper triangular form ofthe matrix R, and C” isthe resulting form of C obtained by transforming R to R. Application of the multiplication rule” to row 3 {with the understand- ing thatthe actual element where (3/1) is stored is zero] gives (15/14) = 15/7 Thus War Application of the multiplication rule t9 row 2 gives (1/6), + y= 18 After Vs = 2 has been substituted into this equation, one finds Vy=2 ‘Similarly, application of the multiplication rule to the top row gives -G2MK+h=-1 and thus War (6) Let U denote the upper triangular form of R obtained in part (a), [e 1 0 | u=| 0 -76 1 0 0 =1s/6 Let L be the lower triangular matrix formed by use of the multipliers of part (a) and by the use of elements of unity along the central diagonal, that i, 1 0 0 L=|-13 1 0 0-37 1 Now carry out the multiplication of L times U 1 0 of-32 1 oo tu=|-13 1 off 0 -7%6 1 o -37 tL 0 0 =15p4 -32 1 0 -| 12 -32 4 0 1p 32, ‘The matrix obtained by multiplication is seen to be the original matrix R. ‘Thus the calculational procedure described in part (a) and the formulation of the matrices L and U from these results does produce a factorization of R, and for this reason the process is called LU factorization,{THE 2N NEWION-RAPHSON METHOD 137 (c) It will now be demonstrated that most of the operations which would be required to solve any other set of equations that differ from the ‘original set by only the elements of the C matrix have already been per~ formed and need not be repeated. The results of these operations are con- tained in the triangularized form of Rin which the multipliers have been saved as demonstrated by Eq. (B) First observe that the operations required to transform R into an upper triangular matrix are the same regardless of the particular set of elements appearing in the C matrix. Next, examine steps 1 and 2 of part (a) and observe that the elements of the C matrix are operated on by only the ‘multipliers used to transform R into the upper triangular matrix R’. Thus, the final set of elements shown in the derived matrix C’ in Eqs. (B) and (C) can be obtained by commencing with the original set of elements in C and the multipliers stored in R’ and performing the operations shown below. Let the elements of C and C be denoted by Cy. Cz. Cs and Ci, Cz, Cys respectively. Examination of Eqs. (A) and (B) shows that G=C=-1 By examination of the operations in step 1, it is seen that C; may be computed as follows Cr = C2 = (HUB), = 0 = (- 1-1) = -13 Similarly, an examination of the operations in step 2 shows that Cy = Cy —(-37)C = -2- (-3/7)(- 173) = = 1577 Thus, for any set of initial values of Cy, C,, and Cs, the final set of values. Ci, Ca, and Cy may be found by performing the above calculations. The ‘method described is readily extended for the general case where the matrix R contains any number of elements. {(@) By use of the value obtained for ¥, in part (a) and the formulas given in Table 43 for the matrix E, of the matrix equation R aV/00 = Ey, the elements of E, are evaluated as follows aR, 00, a ch 6. To determine the values which the elements of E, would take on by the ‘gaussian elimination process required (o transform R into the upper triangu- Jar matrix Rone may perform on E, the same set of operations performed ‘on C in part (a) Let the elements of E, be denoted by Ey, 1, E1,2,£,,3 and the elements of Ej by Ej, Ei, 2, Ei, By use ofthe elements of E, and the ‘multipliers stored in R’ {see Eq, (B)] the ements of E},,, Ei,2, and Ei, 1138 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION may be computed as follows a= (U3)E; —1/3X1) = -23 1.3 — (=3/TE}, 2 = 0 — (~ 3/1 2/3) = -2/7 (Observe that the form of R’ obtained by performing gaussian elimination on RV = C to give RV = C’is the same as that obtained by performing gaus- sian elimination on R 2V/a@, = E, to give R’aV/d0, = Ej. Thus, the final result may be obtained by use of R’ as given by Eq, (B) and the elements of E; found above -32 1 0 1 (1p) -16 1 =| -23 0 (39) -15/4 2 Application of the multiplication rule to the bottom row gives vy = Sam (27-14) = 47s Similarly, forthe second row a, Ys 6 SG + See 28 and. ~2/3 — 4/15)(-677) = 4/5 and for the first row (1 4/5)(-2/3) = -2/15 Use of the LU factorization technique (demonstrated above) for the calcula- tion of the partial derivatives of the 1's with respect to the 9,5 and T/s ‘materially reduces the time required to compute the partial derivatives of the Fjs and Gs which appear in the jacobian matrix. Additional speed is also achieved by performing numerical operations on only those elements lying on the principal and two adjacent diagonals of the tridiagonal matrices. The re- ‘maining elements are zero at the outset of the gaussian elimination process and are not altered by this process, A summary of the steps of the proposed calcula tional procedure follows.‘THE 2N NEWTON-RAPHSON METHOD 139 Caleulational Procedure 1, Assume a set of temperatures {7} and flow ratios ((L;/¥),}- For each trial after the first one, take the assumed values of the js and the (Ly¥)s to be equal to the most recent values found in step 6 Assume 0) = 0; =" =0y= 1 2, Solve the total-material balances Eq. (4-10) for the V's. Save the upper trian- gular matrix R’ and the multipliers used to obtain it 3. Use R’ and the vectors Ey (k= 1,2, .... N) given in Table 4-3 to compute the partial derivatives OV/@0, (k= 1, 2, ..., N) by use of Eq, (426) and LU factorization 4, Solve the component-material balances (Eq. (2-18)] for the e's. Then com- pute the 1s by use of the equilibrium relationship, fy = Ayo, where A, is defined by Eq. (4-11). Save the upper triangular mattix Aj for each compo- nent and the multipliers used to obtain it 5. Use the upper triangular matrix Aj and the expressions given for the column vectors Cy, (k= 1, 2,..., N) in Table 4-3 to compute the partial derivatives ay, (00, (k= 1, 2, .-., N) by LU factorization, The partial derivatives 0¥,/2% (= 1, 2... N) may be computed by LU factorization by use of the upper triangular matrix A; and the expressions given for the column vectors Dy, (= 1, 2,05 N) in Table 4.3, 6. Use the results obtained by steps 1 through 5 to evaluate the functions {Fj}, {G;}, and their partial derivatives. Then solve Eq, (4-13) for the set of correc” tions AX. If the convergence criteria are satisfied by the values of the fune- tions or the values of the variables so obtained, convergence is said to have been achieved. If the convergence criteria are not satisfied, then correct the assumed set of variables X, to give the next assumed set X,, for step 1 as follows Kars = Xq + BAX First a value of f = 1 is used, and if any one of the corresponding values of the independent variables is not positive or if any one of the temperatures is not ‘within the range of the curve fits, a value of f= 1/2 is used. The reduction of {by a factor of 1/2 is continued until the 6's are all positive and all of the temperatures are within the range of the curve fits. A variety of methods were tested for adjusting the calculated values of the variables, but none of them appeared to be significantly better than the relatively simple halving method described above in step 6. In al of the applications of the 2N Newton-Raphson method which follow, ‘beth the functions and the variables were normalized for the purpose of reducing roundoff error. The functions (F,} and {Gj} are stated in a normalized form. The definition of the 6's contains a normalizing factor, namely, (L,/¥)_. Tempera- tures were normalized by dividing each temperature by some base temperature. Although other more precise methods of normalization may be used such as the140. FUNDAMENTALS OF MULTICOMPONENT DISTILLATION fone presented in Chap. 15, the above relatively simple procedure was satisiae- tory for the examples presented in this chapter. When both the F/s and G/s are normalized, the sum of their squares is ‘meaningful as a measure of how far a given set of trial values of the variables are from the solution set, In the examples presented in this chapter, the convergence criterion used was that ¢, the square root of the average of the euclidean norm, ‘must be less than some small preassigned number «, say ¢= 10~*. 1S page 2 6-35 H+ aN (428) where Nis equal tothe total number of stages. Numerical examples The statement of a typical absorber problem is pre sented as Example 4-2 in Table 4-4. To demonstrate the characteristics ex- hibited by procedure 1, the calculated values of the temperatures and flow rates are presented in Table 4-5. A typical stripper problem is presented as, Example 4-3 in Table 4-6, and the intermediate tril results as well as the solution sets of temperatures, flow ratcs, and product flow rates are pre~ sented in Table 47. To demonstrate the application of the 2NV Newton-Raphson method to reboiled absorbers, Example 4-4 was solved. The statement and solution of this example appears in Table 4-8. The application of the 2N Newton: Raphson method to conventional and complex distillation columns is illus- trated by Examples 4-5 and 4-6. Example 4-5 which involves a conventional, distillation column, is a restatement of Example 2-7 (see Table 2-2) The ‘Table 4-4 Statement of an absorber problem, Example 4-2 Rich gas Component lb moth) Other specitatons co, 0403 00 29, Tuy = OF N= Band N ons 09 P= 800 Tin abs. ital 87000 00 temperature profile 1 be constant 6747 oo a0T, = 28°F for all) G= 1. 27786 ons IN} The initial vapor eae 06375 0006 broil is 0 be constant at 03655 mois Vi= 9088 = 1.2, anise ‘0067 the guid rates ae 00505 0061 Y= hd. Thand oor 01995 Ure the K valves and enthalpies aos) 05736 sven in Tables BS and B6 0020 uae on 1866 00 20619 nom omTHE 2N NEWTON-RAPHSON METHOD 141 ‘Table 4-5 Convergence characteristics exhibited by procedure 1 in the solu- Gon of the absorber problem, Example 42 Temperatures Teal Tia Plate pofle, 2 3 4 s 1 850 was TID “8794 2 as2 goal 8103, 3106 3 aas7 50069087, oo 4 51 aD 8956 a1 5 iss agg are ‘ souk aaa 504 8511 7 ams ang? aa 148 ® ‘30 a6 3335 7533 8s Computed at the end ofthe il indicated Test Pate 2 a “ 1 mate LOT «10013 2 1466911946 = 104.0010, 3 1a ot 4 tae 1242103013 5 rig Latte ‘ 15296 «1251640578008 7 1557 12eR 1.079.056 8 1035310255, 1.000655 I, Soiation sets of product rates Component eu be 034800 012190 018037 019106 x 10° 082441 « 108 (058600 x 10! 046475 10" o1ss72 x 10! 67786 (021022 « 10! (090316 « 107 062315 011987 x 107? 0.36886 our = 10° oie OLIBR2 x 10°? 0.55469 x 10° 0.1098 « 10" 015298 ou7ns x 10" O64 (023692 10" 0.12997 « 108 010240 10" (046763 x 108 0.60846 x 10° 020558 « 10! ae T7398142. FUNDAMENTALS OF MULTICOMPONENT DISTILLATION. ‘Table 4-6 Statement of a stripper problem, Example 4-3 1 Component (ib mah) lh mot) ‘Steam Ba ° cH, ° oor Gali, ° 107 om ° 130 neds 0 28 ne, ° 15 Cath ° 235 Che ° 23s 500 ° 224 a 30 Other specications “The sepper has eight plates and it tobe operated ata column presure of 50 bia? abs. ‘The K values and the enthalpies of the hydrocarbons were taken (fom Table B-7 and B-2 ‘The K values for steam given By (K/T)"™ = 0191302 - 093641 10-*T + ORGGNTIS x 10°*7* ~ 06887865 x 10-*7° (where Tis in “R) were used, This curv Bt is based onthe K values given by Haden. The following curve fs were used forthe vapor and quid ental- Pies of team: ” "189976 + 09787701 10°27 ~ 01599299 x 10747? HP? 1183686 + 08152869 « 10° 7 — 0.6788283 x 10-47? (where 7 is in °R and h and HY age in Btu per 1b mol). These curve fis were based on data Pescnted by Smith and Van Ness"™ ‘The cich oi enters at S70°F andthe be linear between T, = 30°F and Ty Gann 8) at 500°F. Take the iia temperature profile to T5°F Take the vapor probe to be constant a ¥, = BO corresponding solution sets of temperatures, vapor rates, and product rates presented in Tables 2-3 through 2-5 were obtained. Similarly, Example 46 (see Table 4-10) involves a complex distillation column and consists of a restatement of Example 3-2 (see Table 3-1) ina form which is convenient for the application of the 2N Newton-Raphson method. The solution sets of temperatures, total-flow rates,and component flow rates are the same as those given in Table 3-3. ‘Also, to demonstrate the effect of the number of plates on the computer time required to obtain a solution, the frst column ofthe system of columns of Example 3-3 was solved as a complex column and referred to as Example 47. The recycle stream B, (see Fig. 312) was taken to be an independent feed and its composition was taken to be the solution set of values given in Table 3-5. The solution sets of temperatures, total-flow rates, and composition were the same as those presented for Example 3-3 in Table 35, ‘A comparison of the computer times and number of trials required toTHE 2N NEWTON-RAPHSON METHOD 143, Table 4-7 Convergence characteristics exhibited by the Newton-Raphson ‘method (procedure 1) in the solution of the stripper problem, Example 4-3, 1 Terperatore profiles ('R) Trial Piste 2 7 5 H 00) or) 26 2628 2 080 a7 AAs a3 3 00 art mn sre ‘ sso ans 26 amar 5 ao ane mat ana 6 2250 79 733 m3 7 S00 nas 129 8126 ® 30 ma 136 sme 1. js computed atthe end of trial nda Triad Plate ' 2 7 * 1 16876 ‘000877 093998 2 23098 093609 033896 3 13663 ospise 033950 ‘ 14993 ossr22 099986 5 12503 920 0.00067 6 11696 10065 000 7 1126 1087 10000 5 11593 10016 1.0000 TI. Solation sets of product ees ‘Component a he Seam 11336 23 cH ooo oo cH arr oo ci 130 oo Calo 2109 001506 nC 13479 02004 Cth 2399 09 cot 25499 00 0 10992 5110)1144 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION ‘Table 48 Statement and solution of the reboiled absorber problem, Example 4-4 1. Sia Fx, tw Component (ib mot) ib moth) cH, 6s ° Gali, B ° on 1 ° Citing ' ° nGH, 0 nH ° 100 (Other specifications 10°F. The eed F enters on plate f= Sasa vapor at its dew pont temperatre at the colin pressure, The column ha 10 pats plu a reboer ‘The reboler duty is specie at Qy= 3 10" Buu/h. The equilibrium data ae given in Table B-t and the enthalpy data are given in Table 8.2. For the st il ake the vapor rats tobe: 6 7928 anda he earespond Tigi rats to be those obtain by making oa meal Blane Take te tempeatres to a0°F 1, Solution % ¥ sie (8) matt) 4 i saas 7 2 seis? 6007 3 est toe 4 Gos sae $ eins 4on2 6 ise aia 7 wer 8 Gi iene 5 Seta ” sear Sua n oT solve these examples by procedure 1 and by procedures 2 and 3 (presented below) is given as a summary at the end ofthis chapter. ‘The results shown for these examples are in good agreement with the fact that the Newton-Raphson method is said to exhibit quadratic conver- ‘gence. For a single variable problem, quadratic convergence means that the error for the nth trial is proportional to the square of the error for the previous tral. The error forthe nth trial is defined asthe correct value of the variable minus the value predicted by the nth trial. For a multivariable Problem, quadratic convergence means that the norm of the errors given byTHE 2N NEWTON-RAPHSON METHOD 148 lation column ‘Table 4-9 Statement of Example 4-5: a conventional |. Feed. The componens composition, and flow rates are the same a sated in Table 22 for Example 27 2 Other specications. Same as tated in Table 2-2 for Example 27 except fr the flow ing changes. Instead of specijing D as was done in Example 246 the rato B/h, Specied at B/Yy = 0.5365, whichis the solution value obtained for Example 32 ‘Agia asin Example 2-7 the refloxeatiospecied at L/D = 2 For thet til the temperatures ae taken tobe linear between T, = 138°F and the rboer temperature 1))= 307F. The values of (L,/¥), for the fat wil are 1 be taken equal to wnt for stages = 2,3, 1 the mth trial is proportional to the square of the norm forthe (n ~ 1st ra In summary, the Newton-Raphson method, under the mild restrictions stated in App. A, converges quadratically provided that the initial choice of variables ies within the region of convergence. The latter condition is ‘commonly replaced by the less precise statement “provided that the initial choioe of variables is close enough to the solution se.” Since the 2N Newton-Raphson method is an exact application of the Newton-Raphson method for all columns in the service of separating ideal Solutions, convergence forall such problems can be assured; provided that suitable starting values for the Ts and L,/¥js are selected (sce App. A). The fact that no difficulty was encountered in the selection of the starting values of the variables of the 2 Newton-Raphson method suggests that the func- tions and variables selected possess a wide region of convergence. The use of other functions in the 2NV Newton-Raphson method could reduce its region of convergence, and it could become difficult to find a suitable set of starting values. For example, the alternative form of the dew-point function given in Prob. 4-5(5) did result ina smaller region of convergence than that resulting from the use ofthe form of the dew-point function given by Eq. (4-4). Applications of approximate formulations of the Newton-Raphson method such as the ones proposed by Sujata? and Holland!” may fail to converge for some examples. For example, the 2N Newton-Raphson method converged for the test problem called Example 3 by Boyum* while the approximate methods of Sujata and Holland failed ‘Table 4-10. Statement of Example 4-6: a complex column Fe. The components composition and fw rales ar he same a tated in Table for Example 32, 2. Other specifications. Same as sated ia Tablet for Example 3:2 except for the fllow- ing changes. The siesteam Wi withdrawn fom plate p= 11. The refx caio 1L,JD = 225. For BMY and W/L, the solution values obtained for Example +2 are specie amy, B/¥y = 031 and W/L, ~ 01202 For the fist tial assume that thé temperatures are fncar between T= 20°F and Ty =300°F, and take T= 20°F and Tip = 907F Take the iil vlos of Mee 16 j= 2,3,146. FUNDAMENTALS OF MULTICOMPONENT DISTILLATION Other Formulations of the Newton-Raphson Method for Ideal Solutions ‘A number of formulations of the Newton-Raphson method for the solution of distillation problems has been proposed. A brief description of each of several ‘of these formulations wherein the number of independent variables ranges from N to 2N js given below. (Other formulations of the Newton-Raphson ‘method in which compositions or component-flow rates are among the indepen- dent variables are described in Chap. 5.) Greendstadt et al" were among the first to apply the Newton-Raphson ‘method to the solution of distillation problems. In 1961, Sujata”? proposed an approximate application of the Newion-Raphson method for the solution of absorber and stripper problems. Sujata regarded the temperatures as the independent variables. For each set of assumed temperatures, the component- ‘material balances and equilibrium relationships were solved by direct iteration (or successive iteration) for the solution sets of compositions and flow rates. ‘These compositions and flow rates were then used in the enthalpy balances in the calculation of an improved set of temperatures by use of the Newton- Raphson method. In Sujata’s application of the Newton-Raphson method to the enthalpy balances, the dependency of the flow rates and compositions on the temperature was neglected in the calculation of the partial derivatives. Aversion ‘of the 0 method, called the “single method” has been proposed.” ‘Another formulation of the Newton-Raphson method was proposed by Newman’? in 1963 in which the total-low rates of the liquid {L,} were taken as the independent variables and the corresponding sets of temperatures needed to satisly the component-material balances and equilibrium relationships was found by successive application of the Newton-Raphson equations. The compo- sitions and temperatures so obtained were used to solve the enthalpy balances explicitly for a new set of liquid rates. The procedure was then repeated by ‘commencing with this most recent set of liquid rates {L,} Following Newman, Boynton? also took the liquid rates {L,} to be the independent variables, and for each choice of these variables the temperatures required to satisfy the component-material balances, and equilibrium relation- ships were found by successive application of the Newton-Raphson method. The results so obtained were then used in the enthalpy balances to compute a new set ‘of liquid rates by use of one application of the Newton-Raphson method. For the case where both the vapor and liquid phases form ideal solutions, Boynton’s ‘method constitutes an exact application of the Newton-Raphson method. All of the matrix equations solved by Boynton were of order N. Another exact formulation of the Newton-Raphson method, called the ‘ult-9 method, was proposed by Holland.'? Two procedures for applying this ‘method were presented. In both of the procedures, the matrices were all of order 1 All derivatives were evaluated by use of the calculus of matrices In a series of papers, Tierney et al proposed an application of the Newton-Raphson method in which the vapor rates {V)} and the temperatures {T;} were taken to be independent variables. The formulation was exact and the partial derivatives were evaluated by use of the calculus of matrices.‘THE 2N NEWTON-RAPHSON METHOD 147 4.2 NUMERICAL METHODS FOR SOLVING THE 2N NEWTON-RAPHSON EQUATIONS, Instead of using analytical expressions for the evaluation of the partial deriva- tives of the F's and Gs, they may be evaluated numerically. Procedures 2 and 3, presented below, make use of the numerical approximations of the partial deri: vatives. Also presented is a calculational procedure for systems of absorber-type and distillation-type columns which are interconnected by recycle streams. Procedure 2. Solution of the Newton-Raphson Equations by Use of Broyden's Method® ‘Tomich?® was the first o apply Broyden’s method (developed in chap. 15) to the solution of distillation problems. Broyden's method is based on the use of numerical approximations of the partial derivatives appearing in the jacobian matrix. The approach proposed by Broyden permits the inverse of the jacobian matrix to be updated each trial alter the first through the use of Houscholder's formuta.® Thus, itis necessary to invert the jacobian matrix only onee. Since approximate values for the partial derivatives are used, procedure 2 generally requires more trials than does procedure 1. However, since the evaluation of the partial derivatives and the inversion of the jacobian matrix are not generally required after the first trial of procedure 2, it requires less computer time per trial than does procedure 1 For the general case of m independent equations in m unknowns, the Newton-Raphson method may be formulated in terms of n functions inn unknowns. For the kth trial, the resulting set of Newton-Raphson equations may be represented as follows J, AX, = fy (429) where Ji the square jacobian matrix of order n and AX, and f, are coaform- able coltmn vectors a ax, a, Ox, AX = Neer Xe Xe Petey Many os Sue)” Lins fans fal” In the class of methods proposed by Broyden, the partial derivatives af /Ox, in the jacobian matrix J, of Eq. (4-29) are generally evaluated only once. In each successive trial, the elements of the inverse of the jacobian matrix are corrected by use of the computed values of the functions. An algebraic example will be given after the calculational procedure proposed by Broyden has been presented,148 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION ‘Step 1. Assume an initial set of values for the variables, X, and compute fo= Xo) ‘Step 2. Approximate the elements of Hg, where Hy is defined as follows Hy = 35! Broyden obtained a first approximation of the elements of Jo by use of the formula OF, 5 Sis; h) — fs) ax; hy where h, was taken to be equal t0 0.001 x, Step 3. On the basis of the most reoent values of H and f, say Hy and f,, compute AX, = HG Step 4. Find the s, such that the euclidean norm of f(y + 5 AX,) is less han ha of) Ft y= tad ie ong ingly std LAG +s axy}"< Eres)" procee to step 5, Otherwise, compute, by wse ofthe lowing formula which was developed by Broyden se [C+ 69) ~ 1184 where & 7206.4 5 ax) © 7%) If the norm is not reduced by use of 5, after a specified number of trials through the complete procedure, return t0 step 2 and reevaluate the partial derivatives of J, on the basis of X,. As pointed out by Broyden, other methods for picking s, may be used. For example, 4 may he picked such that the ceaclidean norm of fis minimized. Step 5. In the course of making the calculations in step 4, the following vectors will have been evaluated Xie X45, AN fuer =/ Be) Test f+ for convergence. I convergence has not been achieved, compute Y &{THE 2N NEWTON-RAPHSON METHOD 149 Step 6. Compute (AY. 54 AN) AXP, Has XH, Y and return to step 3. Example 48 consists of a simple algebraic example which illustrates the application of this method. Example 48 (Hess et al," by courtesy Hydrocarbon Process.) Itis desired to find the pair of positive roots that make f(x, y)=0 and f(x, ») simultaneously Sils, =x? = ay? = 2 Jol y) = 2x — Bay? 43 Take xp=1 and yo= 1, and make one complete trial calculation as preseribed by steps 1 through 6. Sorution Step I. Since xy = 1, yo= 1 Xo= [LU and fio =filXo)= fill, Y= 1-1-2= 2 Jao = Silo) = fall, I= 2-3 43=2 ‘Step 2. Take the increment h for computing the derivatives with respect to xtobe = (0001)xo = 0.001 Then A, _ (L001, 1) ~ f(t, 1) ox oor 1001 For computing the derivatives with respect t0 y, take ‘n= (0.001)yq = 0.001 Then Hy _ fll, 1.001) ~ fill, 1) y 0007 “70 and Gs _ f(L.001, 1) — (1.1) _ ox 0.001 ca Ay _ fll 1.001) ~ 01) _ 693 oy ‘0.001180 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION Then jo= [HOO ~2001 °= {1002-6003 ‘The inverse of Jo is found by gaussian elimination as follows. Begin with 1.001 -2001] | [1 0 11002 —6003] | [o 1 land carry out the necessary row operations to obtain, b 1] 1.4992 —0.49975 lo25023 -0.25000, Then 1.4992 -0.49975 [0.25023 -0.25000| and fT 1.4992 0.49975 v 025023 0.25000 ‘Step 3. On the basis of the most recent values Hand f, the correction AX is computed as follows wv [28 SSI 0.25023 0.25000 1.0005, Step 4. Take so, = 1. Then noose [>=] and Silo + AXe) = f(49979, 2.0005) = 2.9774 {lp + AXq) = f(49979, 2.0005) = ~7.0468 Since (29774) + (—70468)* > (—27 + @F compute 0) + S3(Ko + AXo) _ (2.9774)? + (~7.0468)? FiXo) + F3%o) y+ = 731529 and sag = Lt oP =! oasor4 oa uw ¥‘THE 2N NEWTON-RAPHSON METHOD ISL Then 5 __ [1], [uose39 ] _ [203839 xe oases []+ suas] ~ [tases] and Filo + 50,2 AX) = /(203839, 1.259865) = 1.08040 J(u + 50,2 AXg) = /(203829, 1.259865) = 1.60372 Thus, the criterion on f,, namely, (—Loso4o)? + (1.6037)? < (—2)° + (2}* has been satisied. Step 5. Ifthe convergence criterion is taken to be that the sum of the squares f, and fy isto be reduced to some small preassigned number, say 10-'°, then this erterion has not been satisfied by x= 20384 and 25986. Then compute 0.91960 visa) [a 1.60372 Yoxti-to= | Step 6. Compute the following products which are needed to find Hy, sav [9% 049915)/ 091560 ‘oe | 0.25023 0.25000 | | -0.39628 tee -14992 03905 sgn (907 ang, it), 3 Mara o~ Sst] * oases 091360 aI — — 6.63: sxayyoot-s2s 2281] 998] - sine “ (He Yo + 504X) AXEHy . “AXSH Yo. it flows ta 1, ~ [-14992_ 049975) _ [ -050674 0.18245 "= [0.25023 0.25000] ~ |-0.06525 0.023491 0.99246 031730 0.18498 0.22651, and the next trial is commenced by returning to step 3 with H,182. FUNDAMENTALS OF MULTICOMPONENT DISTILLATION Examples 4-3 through 4-7 as well as examples presented below were solved. by this procedure, These results are discussed in Seo. 4.4, Procedure 3. Broyden-Bennett Algorithm In the method proposed by Broyden, Householder’s formula was used to obtain the formula for the inverted matrix shown in step 6. The second term on the right-hand side of the expression in step 6 contains the correction proposed by Broyden. Instead of applying Householder’s formula, the calculation of an inverse of the jacobian may be avoided altogether by use of the algorithm proposed by Bennett for updating the LU factors of the jacobian matrix. Example 4-9 will show that fewer numerical operations are required to compute the LU factors than are required to compute the inverse of a matrix, Bennett's algorithm is applied to the Broyden equations as follows. Broyden's algorithm consists of successively updating of the jacobian matrix, of the Newton-Raphison equations by use of the correction matrix xCy’, that is, Sere It CHT where — 5 AX AX, (l=) men AX Since the jacobian matrix J may be stated in terms of its factors Land U as ‘demonstrated in procedure 1, the above expression for Jy. may be restated in the following form (a scalar) Tae User = Uy + Cot Bennett proposed the algorithm presented in Fig. 4-4 for updating the matrices L, and U, to obtain the updated matrices 1,,, and U,,,. When Bennett’ algorithm is used to make the Broyden correction, the following calcu- lational procedure is used, Step 1. Same as step 1 of procedure 2. Step 2. The partial derivatives of Jg are found in the same manner as shown in step 2 of procedure 2. Then find the factors Le and Uy of Jp such that Jo= LoVe as described by Hess et al: see also Conte and de Boor.” ‘Step 3. On the basis of Ly, U,, and f, (the most recent values of L, U, and f) compute AX, as follows LU, AX, Step 4. Same as step 4 of procedure 2THE 2N NEWTON-RAPHSON METHOD 153 -vonemio;e) (71 we Bunepdn oy waged 20038 24 ry amd ae E AINE Bonen a154 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION Step 5. Test f,+4 for convergence. If convergence has not been achieved, compute 5 aT, wef (i = ax! Step 6. Use the algorithm proposed by Bennett (see Fig. 4-4) to obtain the updated matrices L,,, and U,,, of Ly and Uy, where Les Uns1 = Uy + CWE ‘Then return to step 3 ‘The use of Bennett's algorithm is illustrated by the following numerical ‘example: Example 4-9 (Hess et al,!* by courtesy Hydrocarbon Process.) Given the matrix Ay and its factors L, and Uy 2 4 ~6 A=LU=|-4 -7 10 207-9 where o 0 1 0 0 ta, 0 [=]-2 1 0 Eaten Lats 1301 Ua Uns 2 4 -6 Uz2 Ursaf=| 0 1 -2 O° Uys, 0 0 3 ‘The upper and lower triangular matrices U, and L, may be found by use of the technique demonstrated in Example 4-1. It is desired to find Ly, and ‘Uys1 when Ay is corrected by adding the matrix x,Cyj to it; that is, find Lys and Uy. . where Lae Une =TAUs + CoE EEE EEE <‘THE 2N NEWION-RAPHSON METHIOD 15S SOLUTION The calculations are carried out according to the algorithm shown in Fig, 4-4. For this example, n = 3. Begin by making the calculations shown in the second square for i= ‘When Wasi =32 C= (MU) = Us4 + Py = 2+ (1)Q)=5 = CU... = OMIYS = 95 Sinoe i < n(1 <3) take j= i+ 1 ‘Then for =(-ayny=4 Lay bay t= 2+ (943/5) = 2/5 "v2 = (1Q)=2 ~ (RNa) = =4 xy2xy— Lats =3 (IKI) =2 bss 1+ 23/5) = 14/5 "ys = (It) = (32-6) = 10 6+1=-5 Y= Is— MU Wyss Un at Za Since j > n, compute c=c-or i+1=2 2 = (9/S\1) = 2/5 and take Then for Mia= y2lUa.2 P= x, = (42/5) = 8/5 Una = Una t Py = 1+ 8/4) = = y2CIUs,2 = (~4N2/5)(-27/5) Since i n, compute 2/5 — (8/27)8/5) = —2/27 and take Then for i= 3 + PaxC= Since i=3, the process has been completed and the final result is given by 1 0 oyfs 6 -s Aaxcy'=LU=] 2/5 1 offo 27/5 14 ays y27 iflo 0 tater Note that, since the elements of the lower triangular factor L along the diagonal are always equal to unity, it was not necessary to make any correc- tion to them in the algorithm, Less time is consumed by provedure 3 than by procedure 1. Calculation of the LU factors of the matrix J in step 2 of procedure 3 requires approximately [3 operations, whereas the calculation of the inverse of J in step 2 of Procedure 2 requites approximately n* operations, where the matrix J is @ Square matrix of order n. To update the LU factors in step 6 of procedure 3 by use of Bennetts algorithm requires approximately 2n? operation, whereas ap- roximately 31? operations are required to update the inverse of J by use of Householders formula as proposed by Broyden in step 6 of procedure 2. 4-3 SYSTEMS OF COLUMNS ‘Two approaches exist for solving problems involving systems of columns, the “column modular method” and the “system modular method” proposed by Hess.'5 In the column modular approach, the equations for each column of @ system are solved in succession, and in the system modular approach, the com- plete set of equations for the system are solved simultaneously. While the systemTHE 2N NEWTON-RAPHON METHOD 157 modular approach may be the ultimate method for solving problems involving systems of interconnected columns, the column modular approach appears to be the most realistic approach at the present time. Solution of Systems of Columns by Use of the Column Modular Method In the column modular approach, the equations for each column are solved by use of the most efficient procedure for each column. After one trial has been made on each column, the terminal flow rates are placed in component-material balance and in agreement with the specified values of the terminal flow rates by use of the “capital © method” for systems, The entire calculational process is repeated until convergence has been achieved. To illustrate the application of the column modular approach, consider the particular system of columns shown in Fig, 4-5, which consists of a reboiled absorber (column 1) and a distillation column (column 2), For such a system, a combination of the @ method and the 2N Newton-Raphson method is recom- mended. The 2N Newton-Raphson method is used for solving the reboiled absorber and the @ method is recommended for the distillation column. At the {end of one complete trial for each of the two columns, the capital © method is applied to place the system in component-material balance and in agreement with the specified values of the terminal flow rates a Qt Qa, Fi y 0 My Liz Be \ We coum Coun —— \ Ff _ |Repoueo] losmicanon ASSOREER COLUMN Op: Lay Tw Ta Figure 45 A system ofa reboiled absorber and a disilation column,158 FUNDAMENTALS oF MULTICOMPONENT DISTILLATION ‘Suppose that the total flow rates Lyi, Dz, Wj, and W are specified, These specifications in turn fix V, and By. The specifications for the reboiled absorber are as follows: F, {X}, the thermal condition of F, ,N, and the column pressure. ‘The specifications for the distillation column are as follows: N, type of conden- set, f, Ly, D, and the common pressure. Tn the application of the 2N Newton-Raphson method, V; is regarded as specified and the temperature 7, of B, entering column 1 or the intercooler duty Qc, is taken to be the new independent variable. In this case the 2N + 1 var~ inbles X and the 2N + I functions f are as follows X= [5 B2y sees Ons Tos Try Tay soos Tad 1 = Dos Fs Fas oos Fs Gis Gas sons Ga) (430) Corresponding to the new independent variable To for the absorber, the new funation fo ftroduced to express the condition thatthe specified value of Yi, demoted by (Vj): must be equal tothe calculated value at Convergence, that is My ow, “Thus, the 2N +1 functions denoted by f, consist of fy the I dew: point fane- tions, and the N energy balance functions given by Eqs (4) and (65) “The proposed calculation prosedue consis of fist making one trial cal- culation on the reboiled absorber by use of the 24 Newion-Rapon method and then ene trial On the dstilation column by use ofthe 0 method. Then the capita © method is applied onetime othe system inorder to place iin overall component material balance and in agreement withthe speiied values ofthe terminal ow rates To inate the eaeulatonal procedure the composition of ny roryle stream whichis needed is asumed. After the fist Wal through the System: the composition af such reece steams found by the © method are Used. The seps ofthe proposed calutational procedure follow Step 1 On the bas ofthe mos recent set vals (6, yh) found instep 3, make de tal on the reboied absorber by use ofthe 3N Newton-Raphion method. [tn order to iit the fst tra a set of © is assumed] Save the teuling value of 7 and L,/¥ or making the next talon this column, Aer One tri, go to sep 2. Step 2 Use the [a found in step 1 as the fed tothe distillation column, Mate ne tral by uss af he O method. Save the resulting values of the Ts and ms Step 3. Apply the capital © method tothe system. Find the st of © mu plis Fequired to plac the system in componentmaterial balance and in Egreement with the pected tfminal rates Ly Dyy My and Wy. Return (0 sept ‘To ilstrate the use ofthe combination of the 2N Newton-Raphson methods the 0 method, and the capital © method, Example 410 which involves the system shown in Fig. 4-5 was solved. The statement and solution of this example 1 (431)‘THE 2N NEWTON-RAPHSON METHOD 159 are presented in Tables 4-11 and 412, respectively. After one trial had been made on the reboiled absorber by use of the 2N’ Newton-Raphson method {procedures 1, 2, or 3), one trial was made on the distillation column by use of the @ method for single columns (see Chap. 2). Then the system was placed in component-material balance and in agreement with the specified values of the total-flow rates by use of the capital © method of convergence (see Chap. 3). The formulation of the four g functions and the material balances of the capital © method for Example 4-10 are left as an exercise; see Prob. 4-2 Solution of Systems of Columns by Use of the System Modular Method In this formulation of the Newton-Raphson method, the equations for each column of the system are solved simultaneously." To demonstrate this approach, the pipestill exampie is used; soe Fig. 46, To solve problems involving ‘Table 4-11 Statement of Example 4-10 Tn vate x ‘Component (Ib moth) (Ib mol/h) om 350 cai 20 ca 30 ci 30 io te so 3s Gi, 8035 fea’ oo 8 Gy MO CH, 20 Che 10 tas GH, 10 as m0 ther specifeations Column 1—Reboled ABsorer N= 12, f= 7, colon pressure = 300 Toi? abe, W, = 10 To moth, W, i 19 be withdrawn as 4 guid from plate j~ 10, and B, Vs, = L435. Tie feed F enters ab a vapor at T2477"R at 300 lyin abs. For the inital values ofthe temperatures [7] and. vapor rates (¥) use the following. 7, 666R, Ty = 62445°R, T= Ty +1846 Gm 3h, NI) Ty BPR, Hp G, Y= 120 =e Ro 8) Vy 3568 1b molt Column ?~Distilion Column column presire = 30 jin? ab Dy 1, = 5, the column has a total condenser and the 20, W,= 10 Tb mol, Ly, 4/D, = 65, and W) sto be itraw as guid from plate = 2. For thi nial values of (7 ad (use the fllowing. T= TSR T= 7-4 #34 24 NI) Ty 4SER, Vy 2, V,— 150 = 2,2 1) ¥ip=3641 1 moth, Ue the equilibrium and enthalpy data give in Tables Bl and Ba, System specification. W, = 10 B, 100, D; = 2, and W, = 10 8 mol‘Table 4-12 Solution sets of temperatures, vapor rates, and product distribu- tions for Example 4-10 1. Final temperature and vapor rate profiles ‘Cohama 1 Colm 2 i ¥ 1, ¥ sue 6 mot) « (i> mot) Hl 6938 000 a8 2000 2 ons osi9 1873 15000 3 7268 10886 78289 19.13 ‘ 71998 73 7693 1036 5 m9 1238 79585 13957 6 6st iRaas +0099 wa? 7 79196 25341 #0336 14501 5 75951 46852 10979 a7 2 76888 39501 1563 191 0 m3 65105 52690 602 in rat 805 ss812 13580 2 50333 687 2 Prodet distributions for colama 1 Column 1 Flow rates (6 molt) 030599 107 O47E20% 10"? OSI x 10-7 019563 « 10 01917210? 0.7859 10"? 026196 x 10 095018 x 10 028156 040833 x 10 219i 0.69425 026137 019099 065209 0.16885 x 10 051893 028765 x 10, 029343 « 10 0.16889 x 10 ‘O10 x 10° 0278 x 10 015129 x 10 014529 « 102 071083 x 10 (036612 * 10 037907 « 102 o97ses 033 (010823 « 10? 036977 oor ‘990232 « 10 (0.12398 x 10° (Qc, = 0.10389 « 107 Bayh Qn, = 062979 x 10" Bush{HE 2N NEWTON-RAPHSON METHOD 161 Table 4-12 (continued) 3. Product dstibtion foe column 2 Column 2 Flow rates (lb mol) Component as A 7 O10 10"? OMD «10 Oa7 m1 016218 10" 424 10-7 08617 x LO 020057 0439 «10° O1SIB4 x 10" seks 1090 0.6803 « 10" o4n2ss o1sti0 020051 « 10°! 021583 « 10 ease 060467 10°» (075406 « 10 0275 « 10 ost 039072 «10 2mm «10 88146 « 10 053588 «10 0.38747 «10 028672 x 10? osiia 10"! 0106 10802 « 108 61508 «10-2 O.6652 = 10°! 0.9003 x 10 O61IBS 510-7 032479 10-7 0.1294 10? =m OH 10509» 10" Boh 1928 10? Blah systems of columns, the column modular method is generally recommended over the system modular method except for the possible exception of pipestills. Since, however, each sidestripper is small relative to the main column, it is perhaps easier to treat this particular system by the column modular method rather than the system modular method. To illustrate the system modular method, an ab- breviated development of the pipestill problem, Example 4-11, is given The pipestill problem solved herein was originally solved by Cecchetti etal by use of the original @ method of convergence which is described in Chap. 3. This problem is based on data from field tests which were made on the pipestill shown in Fig, 4-6, The @ method for distillation columns may fil o converge for some absorber-type problems, such as the pipestill. The pipestill is classified as an absorber-type problem because the main column has a condenser but no reboiler; the first sidestripper has a reboiler but no condenser; and all of the remaining strippers are of the conventional type. For the pipestill shown in Fig. 4-6, Cecchetti et al found a theoretical analogue column by trial, The theoretical analogue column is defined as that column having perfect plates which gives calculated results that are in good agreement with field test results for the pipestill The theoretical analogue column shown in Fig. 4-7 was proposed by Hess et al!? This is essentially the same as the analogue proposed by Ceechitti etal162 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION i frets i Sau 8 ‘SIOESTRIPPER 3 8 mn rt 2 exe, [C2 Mette ee — sem 25 ™ ‘ x = — 3 FEED. # ~ ‘cnvoe) 3% No. 38 z Figure 46 Actua stage numbers inthe min coloma and the sie strippers of the piesil [Hess et aly Hydrocarbon Process 53): 241 (1977) by courtesy Hydrocarbon Processing] ‘The minor differences between the theoretical analogue shown in Fig. 4-7 and the one used by Cecchetti et al. are a reflection of the different manner in which water was treated. In the present analysis, which follows that of Hess et al,'? water was regarded as being distributed between the vapor and liquid phases on stages 2 through 37. whereas Cecchetti et al. regarded water to be in the vapor phase alone on these stages. On stage 1 (the accumulator), two immiscible liquid phases (a water and a hydrocarbon phase) are assumed to be in equilibrium with the vapor phase, In Fig, 47, the withdrawal rate of the liquid water phase is Wo, and the withdrawal rate of the liquid hydrocarbon phase is W.THE 2N NEWTON-RAPHSON METHOD 163 sere ye alt, LEGEND: The calculated 38 Fesults are enclosed by fe rackets beneath the 0- Served test results, 8/0 “barrels per day of hydrocarbons Figure 47 Theoretical analogue ofthe pipet [ess ta Hydrocarhon Process. 848) 241 (1977) Courtesy Hydrocarbon Processing}164 FUNDAMENTALS oF MULTICOMPONENT DISTILLATION ‘The physical properties (normal boiling points, densities, and molecular weights) of the 34 pseudo components selected to represent the true boiling Point curves of the feed, distillate, and sidestreams are presented by Cecchetti et al On the basis of these data, a feed composed of 34 pseudo components and having the molar compositions and total flow rate shown in Table 4-13 was selected by Cecchetti etal. The specifications for the theoretical analogue column are also given in this table. The calculated sets of temperatures, flow rates, and compositions obtained by the calculational procedure developed below are presented in Tables 4-14 through 4-16. These results were obtained by use of the data of Cecchetti etal for the K values and enthalpies of the pseudo components. Curve fits of these data are presented in Tables B-14 and B-16, The K values for water i= 35 was taken from nomographs given by Hadden and Grayson,"? and the enthalpies from the steam tables given by Smith and Van Ness." In the condenser, the K, was taken equal to its vapor pressure (as given in the steam tables) divided by the total pressure in the condenser. A SYSTEM MODULAR FORMULATION OF THE (N41) NEWTON-RAPHSON METHOD FOR A PIPESTILL “The pipestll shown in Fig 47 is used forthe purpose of separating the hydro- carbon feed Fy into even fractions (Vi, Wi, Way Wa, Wa, Ws, Lag). The remain- ing feeds, F,' through Fs, consist of steam which is used as the stripping ‘medium. The stages are numbered down from the top of the main column. The condenser-accumulator section is assigned the number 1, the top plate the ‘number 2, and the bottom plate of the main column the number 28. The stages of the sidestrippers are numbered consecutively 29 through 37, as shown in Fig. 47 For the theoretical analog column, the following variables are regarded as fixed. 1. The number of theoretical plates in the main column and side-strippers as well as the locations of all stream withdrawals and return positions ‘Quantity, composition, and thermal conditions of all feeds Column pressure and the pressure drop per stage ‘The reflux ratio L., /V, (or alternatively, the condenser duty Qcx) Distillate and sidestream withdrawal rates, W,, Ws, Wy, We, We The boilup ratio Vs,/Lss for the sidestripper No. 1 which has a reboiler (or alternatively, the reboiler duty Qa for this stripper could be specified) ‘The intercooler duty Q, for the pumparound stream W, ‘The pumparound rate W, and its withdrawal and return positions ‘The numerical values for the specifications are listed in Table 4-13.THE 2N NEWION-RAPHSON METHIOD 165. ‘Table 4-13 Composition ofthe feed stream F, and other speci- fications for the pipestll, Example 4-11 (Taken from Ref. 13 by ‘courtesy Hydrocarbon Process.) Other speciation forthe Component FX, ‘heoreial analogue column ‘079000 x 10" The hydrocarbon ed F, enters the 7 2 1025700 x 10" column at 637°F with 038", of 3 1038000 «10? the ford vaporized anda total 4 (43800 x 10 enthalpy of 219890 = 10" Bh, 5 ‘95700 x 10? The steam eaters the main column 6 71400 x 10? and the sdestrippers as super 1 (963300 x 10? heated steam at S72°F and with ® (963300 « 10? the ates Fy = 66 Fy = 694, 9 (076250 x 108 Fy = 268, and F, = 158 b mol (072290 x 10? The sidesiecams ae tobe wth (043990 x 10% deawn atthe aes WH = 15, 049950 «108 Wy = 313.4, Wy = 13608, and (086500 « 108, = 36840 mol/. The pump (029400 » 10? around steam io be witheawn (020400 » 10% at the rae of Ih, = 823 Bb mol ‘051000 x 10% and the interoate duty Q, = ‘034000 x 102 180 «10° Buyh. The refi ratio 034000 x 10? Ly/F; = 1095 and the Bil ratio 030680 x 10? Hylly = 0.13245. The pressure in (030680 x 10? the acumulaoe i 231 Tyin® abs, (67400 x 10? ad the pressure on plate 28 is (6s400 x 10? 2924 Ta? abs. An equal (042400 x 10? pressure drop per plate may be (071200 x 10? sumed forthe main cola and (0467300 x 10? the siestrippers. Te presure (012780 x 10? on the top plate of each side- 0.1360 % 10° Stripper may be taken equal to ‘097100 x 10% the pressure ofthe plate in the (81200 x 10 main cokimn whee the sidesteam (0.67800 x 10? feed to the stiper originated 47700 x 10? 057300 « 108 (029400 108 (028300 «10? 026100 10? = 022032» 10" tb mol166, FUNDAMENTALS OF MULICOMPONENT DISTILLATION CHOICE OF INDEPENDENT VARIABLES ‘Although the temperatures and the L/V's are commonly selected as the indepen- dent variables in the 2N Newton-Raphson method, a heating or cooling duty may be selected as an independent variable, and the L/V for that stage specified For example, by specification of the reflux ratio L,/¥, the condenser duty Qc and the temperature T, become the independent variables for the frst stage of the main column. The variable L;/¥; is replaced by the new variable 8, which is defined by Ea. (438). For the set of specifications stated above and in Table 4-13, the correspond- ing sot of 2N +1 indepentent variables (where Nis equal tothe total number of stages) is: Bo, Qeie 93+ Bay --» 030» Onais 0320-0 O37» Ths Tas: Tay, Except for dy, the 0/5 ate defined by Eq, (4-6) The new variable Opis introduced to account for the two liquié phases in the accumulator, and it is defined as follows WolY, = Oulho/¥ide (432) ‘TOTAL MATERIAL BALANCES ‘The total material balances are formulated in a manner analogous to that demonstrated for complen columns. The returning vapor streams from the side strippers to plates 5, 10, 15, and 22 give rise to elements which lie above the three diagonals of the matrix R of Eq. (4-10) ‘The presence of two perfectly immiscible liquid phases on stage 1 and the typical two-phase (vapor and liquid) behavior on all subsequent stages calls for a special treatment of water on stages 1 and 2. The development of the ‘component-material balances for all components except water are developed for the first and second stages, and then the component-material balances for water ‘on these two stages are developed COMPONENT-MATERIAL BALANCES For stage 1, the component-material balance for any component except water (or water of steam i= c= 35)is given by rye him wu=O (Fe) Since W, and L, have the same composition WLM (#0) Use of this relationship and the equilibrium relationship given by Eq. (4-11) permits the component-material balance to be restated in the following form HL + Ant (Mi Li)Audeu + P20 GF For the second stage, the component-material balance is given by 0 (ite) vat hy ~ bas = bs‘THE 2N NEWTON-RAPHSON METHOD 167 ‘and by use of the equilibriam relationship, the component-material balance is restated in the following form Ayoy = (1+ Andou t=O A (433) ‘The component-material balances for the remaining stages are developed in an analogous manner, and the complete set of equations so obtained may be stated in matrix form as follows Aw=-f% (FO) (434) where w= Bus Oa aa” A= [0 0 ery: bras Oo OF ‘The square matrix A, differs from a tridiagonal matrix by the appearance of only four nonzero elements to the right and four nonzero elements to the left of the tridiagonal band of elements. The four nonzero clements to the right of the tridiagonal band result from the return of the vapor streams from the sidestrip- pets, and the four elements to the left of the tridiagonal band of elements result from the introduction of the sidestreams W,, Ws, W,, and Ws to the sidestrip- pers. The hydrocarbon feed F, enters on plate 27, and vp; is i row 26 while In; is in row 27. COMPONENT-MATERIAL BALANCES FOR WATER Since the two liquid phases (the water phase and the hydrocarbon phase) in the accumulator are taken to be immiscible, it follows that Woe = Wo he ‘Thus the component-material balance for water for stage 1 (the condenser- accumulator section) is given by Since the partial pressure of water vapor above the two liquid phases in the accumulator is equal to its vapor pressure, itis evident that (o1 102 26 011928 « 10° n 0.11338 x 10° Fy . (097097 x 10? 2 81199 x 10? » : (067799 x 10? Ft 1047699 x 10? 2 (057299 % 10 z . 1029599 « 10? o (038299 x 10? 3s 038488 = 10 (077105 « 10" ovat 35684 13500 19430) 77886‘THE 2Nv NEWION-RAPHSON METHOD 171 ‘Table 4-16 Product distribution forthe sidestrippers of Example 4. from Ref. 13) (Taken Compo- reat dns hae hn ba O69 x 10"® 0257S 10% OBIE 10" 094406 x 10°" ‘019430 = 10-2 018269 x 10"? ODS x 10-9 075324 « 10-+ ‘91788 x 10"? 024606 x 10°? OSTHGA 10"? O34SST x 10-7 ons O1S945 «10°! ONTSTS x 10"! 085512 x 10°? O17? « 10! Ose 1816s 057896 x 10°? 051627 x 10" 031236 om 010062 x 10°! 029863 « 10"? 8628 099136 20813 x 10°! 049639 «10°? 010066 x 10! 097587 027918 « 10°! (090039 10% 022821 x 10" 049077 x 10" 030768 x 10-* (069003 = 10% 038008 x 10" 0.26698 10" 070253 10 (036073 «1080421 10" 020027 10" 057568 «10 01744 x10! ORS «100TH x 10H OLRISK 012172 «10? OSa7IS «1007S x 10? OHS 8682 016380 «10 11270 10? 017098 10401 (096060 «10! 018373 x 1002646 ‘16173 «10% 0656 x10! OTHRS x 10? 09569 (049269 < 10-2 OSE x10! ORI x 10? 056579 C231 x10" 028438 o.n2mi x 10" 0.29865 050597 x 10-0199, (026966 x 10” oxs782 Qys022 x 10°* 027583 x 10°! 02596 xO? 012388 « 108 13224 10-7 L837 x OSS 10? 0.42919 « 10" 6624 10°" OMB x 10-2 O4NGT X10? 099107 « 10" 26H 10°" 020398 x 10-* —OUDITH 10? OTST x 10 (0735 « 10" 039168 x 10° 029284 10) 038165 « 10° (67203 10-7" 022268 10-8 O@DITR 0" 0277RB x 10 030380 5 10° O14S3E 10-2 08SIOD x 10" 021479 x 10028489 10° 021711 012939 «10° 031133 « 10? 64525 «10-2 0.20290 x 10-* S640 x 10-7 030036 = 10°? 53617 « 10" o26ra9 x 10 ans asec sons ALLY $97 or aanL ms Het tevontastoy sp neo poumm e —-SuRUy siunuoduoo safe ulinos ayduerg NOON JOON oad TomgRy-UOIRAN NE _aindo> 9/A0ey THYGINW Ue Wo sen jo auMU pur sou HORNER 7 sano 2805 “1 spoqiaus snoqses 20y pownbox sou soindwory LI-F age,174, FUNDAMENTALS OF MULTICOMPONENT DISTILLATION ‘pout s:v9pKore 0} paqusep sinpasoud euoeTRae 24 0 ‘dae ut pounbos swonenteas uonsppe Sum sid z dtu 30} po women reat aur apni Mes Ueigoet a Jo soomEneNs Jo Z9guNN OT sawne> omens apogs 20) € 2unP=Oid pr zgi05ge WNVELIOS —pomau uosydey-uoMAN NE — IP ARLES ¥ a Prno9e ovr pe ¢ asnpaaosd oy sou ¢ ovens E10 wm H Nvaluos eee ur se sad Wy e140 wm H NVaLWOS spoureu @ UE 9 UL T vot “ vp waists uy pre powew ‘syeusy sabes suauodion ‘umes awe BON PON oxic, Sounoo jo suaKS (panunuor).1-p 1481,‘THE 2N NEWTON-RAPHSON METHOD 175 PROBLEMS {41 An adiabatic ash proess may be regarded as an absorber with one equilibrium stage in which te two entering feds are combined to pve a feed fate Fad composition [that evn l= EM aod fuente ye Fe (2) Restate Eas. (41) trough (413) forthe special case ofan absorber with one equilibrium stage. (6) to the terest of simplicity, et (Ly) be St equal unity. Show that Eg (410) educes {oa sealrtoal material balance which may be Solved for Fi give F +0 and that A,x,= =f redues toa scalar-component- material balance which may be sled fr, 10 TOK, 442 (a) Obtain expressions forthe partial derivatives of, and, (Se Prob. 41) with respect 0 7, and, (6) Obtain expesios for the partis deivtves ofthe functions F, and G, (ee Prob 41) ity respect tT, ad 43 Show thatthe function F(0,. 7) (2 Prob. 4-1) decreases monotonically with both 0, and 7, forall 0, >0 and all T, lying between the bubble-point and the dew-point temperatures of the ‘combined fod 4 Repeat Prob. 43 forthe fneton G (0,7) (ee Prob. 41) 445 Produce «sketch ofthe following fms of he phase equilirum function in he postive domain of ata fed value of 7; Iying betwen the bubbe-point and dew-poin temperatures of the led 1S tu HER, 0 FE (tps ae fe) Fe where the formulas for v,and Hae given in Prob 44. 446 A singlestage absorber ist be operated adiabatically a 1 atm pressure. The an ol team Ly enters the singlstage absorber asa quid at 1396°F, and the rch gas ens the abvorber at it dw: point temperature of 200°F at | atm, Use the equlibviam and enthalpy data sated in Examples |-4and 18. The componentslow ates ofthe rich gi Vy, aad kan oil Ly a as follows, Componest ens to ' °| 2 oo i oo176 FUNDaMevTaLS OF MULTICONPONENT DISTILLATION Kt i desired to determine Vy Ly, ant thei compositions by use ofthe Newton Rephion metho, Evaluate the functions F and Gy ad the partial derivatives JF 00, and 0G, 20, These ae two of ‘he four partial derivatives which are needed to make the fist til and calculation by use ofthe Newion-Raphson method. For the fit tal tke y= I and, = 105, practice, both the varibles and the functions should be normed Let anew normalized ‘emperatre be defined as fllows TF) T(oormatizes) = 72 47 Bein with fest principles and formulate the enthalpy balance fnctions 6, fora conventional Aistiation soko, “48 For the functions {F;! and {Gj gven by Eqs. (43) and (4-4 respectively, develop the expres ‘sons for the following ses of partial iatives for allj = 12. N)and ab fee LN) PAL PEL P61 poy) al lant lool er) 449 Pesorn the matrix diferetiaton implied by the right-hand side of Ea. (424) and verify the Forms given in Table 43 for C ‘440 Sopp the mising steps requite to fnd the inverse ofthe jacob matre J ia Example $8 4m yt 2 4 ~6 w=] -4 7 Helles Aa en (6) Obtain the LU factorization of Ay. and compare the result so obtained with that fund in Example 49, 4412 Formulate the four fustions and the mates equation forthe overall component-nateria balances for Example 10. 419 Show thatthe Broyden correction may be sated inthe form given in procedure 3, namely, Compute dered wor ‘Hin: Make uso he Broydenrationships given in Chap. 15, REFERENCES, 1. M, Bem: “Triangular Facto of Modified) Matioss” Numersche Mathematik, 7:217 (1965, 2 D. S. Blingsies: = Numerical Solution of Problems in Multicomponent at Steady State” AIChE J. 16:43 (1970. 3. G. W. Boynton: “eration Solws Dstilation” Hydrocarbon Process 48:183 (January 1970) 4 A.A. Boyum: PRD. Dissertation, Polytechnic Inttte of Brooklys, Brookiym NY. (1965) §.G.G. Brown and M, Souders, 1 "Fundamental Design of Absorbing and Siping Columns for Complex Vapors” Ind. Eng. Chem. 24519 (1932)2N NEWTON-RAPHSON METHOD 177 6 C.G. Broyden: “A Class of Methads for Solving Nonlinear Simultaneous Equations” Math Comp. 19577 (1965) 7. 8 Carnahan. H. A. Luter, and J.D, Wilkes: Applied Numerical Methods, John Wiley & Sons, Inc, New York, 1968, 1% R.Ceochet, RH, Johnston, J L. Nedswieki, and C.D, Holland: * Pipestil Product Verily “These Computer Estimates Hydrocarbon Process and Pe. Rehm 42197159 (196) 9S. D. Conte and Carl deBoor: Elmentary: Numerical Analy. 28 ed, McGraw-Hill Book Company.‘New York, 1972, 10. §E Gall: “Aa Extension of the Mult? Method to Disilation Columns and an Alnos Band Solution of Equliium Stage Prob” MS. Thess, Teaas A&M Univer. 1975 1H John Grevnsiadt, Yonathan Bard, and Bunt More: “Multicomponent Disilation on the BM 704" Ind ng. Chem, 50: 1644 (1958), ST. Hadden and H. G. Grayson: "New Chars for Hydrocarbon Vapor-Liqud Equi,” Hydrocarbon Process. and Pet. Refit. 49}. 207 (1961), 1B. FE Hess C.D. Holland, Ron MeCarie, and N.J. Tetlow: “Solve More Distillation Problems, Part 7—Absorber-Type Pipes” Myrocarhon Pras, 863) (1977). SE Gallon G. W. Benen C. D. Holand, Ron MeDanich, and N, J, Telow: “Solve More Disilation Problems, Part Which Method to Use Hyirecrbun Process, 66) 181 asm, 15. —— PRD. Disertaton, Texas AM University 1977 16 CB. Holland, G. P. Pendon, and S. E Gallon: "Solne More Disilation Problems, Part 3-Application to Absorbers” Hjdrcarbon Processing (1): 101 (1975) Fundamentals and Modeling 2f Separation Proceses— Absorption, Distillation, Epona tion. and Exiaction, Prentice-Hall re, Englewood Clif, NJ. 191% 18, A. Kremer" Theoretical Analysis of Adsorption Process,” Nat. Pet. News. 22:4 (May 21, 1930. 19, M. Kubik, V. Hlavatek, and F. Prochaska: "Global Modulir Newion Raphson Technique for Simulation of an Interconnected Plant Applied to Complex Reciying Columns” Chem. Eng Set 3:277 (1996, 20.1.8 Newman: “Temperature Computed for Distillation.” Hydrocarbon Process. 434) 141 (1963) 21.1. M, Smith and H.C. Van Ness: “Inioduction to Chemical Enginecring Thermodynamics McGraw-Hill Book Company, New York 24 ed, 1959, 22 A.D, Sujata: “Absorber Stripper Cautions Made Ease” Hydrocarbon Proce, 40:137 (0960, 23.4. W. Tierey and J. L. Yanosik: “Simultaneous Flow and Temperature Correction in the Equilibrio Stage Problem” ATCKE J, 15897 (1969). 24... Tiemey and J, A. Bruno: “Equlibriam Stage Calculations” AICHE J. 13:556 (1967, 25.3. F Tomich:"A New Simulation Method for Equilibrium Stage Processes” AICHE J. 16:229 (970)CHAPTER, FIVE ALMOST BAND ALGORITHMS OF THE NEWTON-RAPHSON METHOD The Newton-Raphson formulations, called the Almost Band Algorithms are recommended for solving problems involving columns inthe service ofseparal- ing highly nonideal mixtures. In the Almost Band Algorithms, the independent variables inthe Newton-Raphson method are taken tobe either one or both sets of the component-low rates {I} and (0), the temperatures (7) and in some of the formulations one or nore of the total low rates (L}-In See. 5-1 the indepen- dent variables are taken to be the component-flow rates, {ly}, fy), and the temperatures {T,}. The formulation is presented in Sec. 5-1 for absorbers and strippers, and in Sec. 5-2 the formulation for conventional and complex columns is presented. Two modifications of Broyden's method are presented in Sec. 5-2. ‘The modifications of Broyden’s method preserve the sparsity ofthe inital jacob- ian matrix, whereas the original method as proposed by Broyden does not. The treatment of systems of columns in the service of separating highly nonideal solutions is presented inthe next chapter ‘Although the Almost Band Algorithms use a large number of independent variables, far less computer time is required to obtain a solution to a given distillation problem than might be expected, The computational speed results from the use of selected techniques of sparse matrices and the characteristics of homogeneous functions Sparsity of the jacobian matrix is achieved by a suitable ordering of the variables and functions. The particular choice of variables and functions and their ordering (discussed below) leads to the unique form of the jacobian matrix shown in Fig. 5-1, The well-known method of gaussian elimination may be applied in a stepwise fashion in the transformation of the matrix shown in Fig. 5-1 to the one shown in Fig. 5-2. At any one time, only four of the (c + 2) square matrices along the diagonal and the two corresponding column matrices ate considered in the gaussian elimination process instead of the complete ‘N(c+ 2) square matrix. No arithmetic is ever performed on any of the 2er0 elements lying outside ofthe squares in Figs. -1 and 5-2. 18ALMCST BAND ALGORITHMS OF THE NEWTON-RAPHSON METHOD 179 ‘[oloje" ao -[elolo|e cree aac s| [elelo + gameenee ole ween cacaneenee 4 Figure $4 Structure of the jacobian matrix for an absorb. i 2s win at N@efo] Slresene™ Ol] zepa ne 28 E X SL Figure $2 Upper triangular mais for an absorberHighly nonideal solutions are characterized by the fact that the activity coefficients and the partial molar enthalpies are strongly dependent upon composition. In order to compute the partial derivatives of these quantities which are needed in the application of the Newton-Raphson method, it is con- venient to choose compositions or component-flow rates as members of the set of independent variables. Numerous choices of the independent variables have ‘been made.** -20 To demonstrate the formulation of the Newton- Raphson method, the choice of independent variables proposed by Naphtali and ‘Sandholm'? is used. The Almost Band Algorithm may be formulated for other choices of independent variables as shown by Gallun and Holland.”-®-* 5-1 ALMOST BAND ALGORITHMS FOR ABSORBERS AND STRIPPERS, INDEPENDENT VARIABLES: {labs (ou) AND {7} AAs shown in Fig. 4-1, the plates of the absorber are numbered down from the top of the column, the top plate is assigned the number 1 and the bottom plate the number N. The variables regarded as fixed (or specified) in the developments which follow are: 1. {lod liquid at Ty and at the inlet pressure Po 2. {owe ,sh Vapor at Ty, and at the inlet pressure Py, 3, the column pressure or the pressure on each stage ‘The N(2c + 3) equations required to describe the column may be stated in the following form: (21.2.8) (#12000) Equilibrium _ relationship \ GH Ny (= 1208) (1) Component (fy.14 bens OH— .y) material balances Energy a Balance glenwhere Hy and fy denote the virtual values of the partial molar enthalpies; see Chap. 14" In the above statement ofthe equations, the component-materal bal- anos and the energy balances enclose each stage j ‘Use of the second and third expressions of Eq, (S-1) to eliminate the total flow rates from the equilibrium relationships yields a total of N(2c+ 1) equa- tions for the description of an absorber, When the independent variables are chosen as shown above, itis convenient to state each component-material bal- ance and each energy balance forthe enclosure of a single stage. Thus, after the totalflow rates have been eliminated from Eq, (5-1)'as described above, the resulting st of N(2e+ 1) independent equations required to desribe the column may be stated in functional form to give (5-2) (5:3) m= Oya by Slealiet uh Lewd) (54) In the expressions for the activity coeficients (yf, 9f) and the virtual values of the partial molar enthalpies {Hf}, the mole fractions must have the sum of unity. This eondition is satisfied by use of the following expressions for these mole fractions (5-5) In order to obtain a jacobian matrix having the form shown in Fig. 5-1 for an absorber, both the functions ard the variables must be appropriately ordered. ‘The functions must be ordered as follows Paar Sua Sie mua ma Me GYyor a)” (5-6) ‘where the subscript j= 1, N means that the argument is to be repeated for J=1, 2.1.5 N. The variables must be ordered as follows hea bea Ie Ma M2 °° Oe Dyerad (5-7) By the ordering of the variables is meant the order in which the differentia- tion of each function is carried ovt in the Newton-Raphson method which is the same as the order in which the variables appear in the vector given by Eq. (5-7).182 FUNDAMENTALS OF MLLTICOMPONENT DISTILLATION For example, the Newton-Raphson equation for any one function, say fy (the equilibrium function for plate j and component k, where k denotes a particular fone of the e components}, is (Sn /Ols, 1) Ali, + (Ofe/Ols,2) Aly, 2 +o + (alBly,c) Ali. + (fui 01,4) Avs, +--+ (AflO0s,<) APs, 6+ (Afu/OT) AT, 4 (Spl0 0) Ay, + (plOT) AT ‘The complete set of Newton-Raphson equations may be stated in the following matrix form Jax=-f (5:8) Gis ther a2 ame eite Oh: OT, = ay a ha a Ay hs oTy Ax [CAN 1 B20 Ale Ady Avy, 2° Ady ATi x] and fis defined by Eq. ($6). ‘The complete set of Newton-Raphson equations may be solved by trans- forming the matrix shown in Fig. 5-1 into the upper triangular matrix shown in Fig. 5.2. The triangularization procedure, to be described next, is based on gaus- sian elimination, ‘The matrix shown in Fig. 5-1 has several desirable features which arise from ordering. First, the elements lying outside of the shaded areas in Figs. $-1 and ‘S:-2are always zero. Second, most ofthe elements below the principal diagonal in Fig. 5-1 are zero, and those which are nonzero are clustered along the diagonal. This characteristic makes it possible to consider only a relatively small number of squares of elements (or submatrices) at any one time. For example, in the frst step of the triangularization of the jacobian matrix in Fig. 5-1, the submatrices 1, 2, 3, 4, 6, and 7 are considered. To initiate the (e) and d; should be replaced wherever it appears by its equivalent (D/L,)h,..n the case of a total condenser, the variable v,, does not exist, and it is replaced in f, , and x by ys, the mole faction of component i in the vapor above the liquid in the accumulator, that is, (5-19) and X= LQe bea ee Bar Ye Th aa Ue On Oe Tiyea.w Onl (5-20) Also, wherever y,, appears in the expression for the activity coefficients {y¥) and the enthalpies {#1 it is replaced by its normalized value y4./S¥-1 Ju The set of N[(2c + 1) + 2] functions is again given by Eq. (5-14), However, a different form of the function S, must be used because the expression given by Eg. (5-12) is no longer independent. For it may be obtained by a linear combina- tion of the other functions, namely, si=-|¥ Smet Loss| /0 (521) ‘Thus, the following expression should be used for $, for columns with total condensers Si= Dont (5:22) When the condenser duty Qc and the reboiler duty Qq are specified rather than Ly and Ly, then Qc and Qy are replaced in the vector x of independent variables by Ly and Ly, respectively. The specification functions are given By Eqs. (5-12) and (5-22)188. FUNDAMENTALS OF MULTICOMFONENT DISTILLATION ‘The [N(2c + 1) + 3] Formulation of the Almost Band Algorithm for a Complex Column with One Sidestream Consider the case where a liquid sidestream is withdrawn at the specified rate of W, mokes per hour from plate p as shown in Fig. 3-4. This problem may be solved by introducing a new independent function and a new independent var- iable 4, which is equal to the multiplier of the flow rate L, required to give the specified flow rate W,. The new independent function may be stated as follows eB % The functions and variables x are ordered as indicated by Eqp. (5-14) and (£115) for conventional columes excep for stage Ps and for this sage the Kc: tions aed ther ordering follow -1 (5:23) Sys Sor “Spe Mon Mya” Mye Gy Fe ‘The variables for stage p and their ordering are Fn Iz Ine Ont U2 Ome Te where each w,. is replaced by ¢y in the material balances. The 2c + 2 functions and 2c +2 variables for stage p are reflected by the 2c +2 by 2c + 2 squares shown in Fig. $3. * BET . wow -E A emsns oaie ot ne : aa . La Figure 53 Jacobian matrix for «compir column with qui sidesteam withdr from plateIn the application of the Almost Band Algorithm to problems involving sharp separations, it was required that all of the corrected flow rates be positive and that the temperatures le within the range of the curve fits. In this procedure, the vector correction AX was reduced by an appropriate scalar a until all of the corrected flow rates were positive and the corrected temperatures were within the range of the curve fits, that is, Nox to Ax (5:24) where a= I, 1/2, 1/4 1/8... Fsst a value of a = 1 is tried, and ifthe conditions enumerated are not satisfied, the value of is reduced successively by a factor of 1/2 unl all flow rates are positive and the temperatures lie within range of the curve fis ‘When only the [ys or 1s are selected as the independent variables, the picking of « becomes more difficult, particularly in the case of complex columns, because the « must be selected such the dependent component-flow rates given by the constraining equations (the component-material balances) are positive. ‘The disadvantage of the choice of the {lft and {Tj} as the independent variables over the choice ofthe {Ij} and {7} 3s the independent variables isthe additional storage requirement for the N x ¢ vapor rates {o}. Other formula- tions involving different sets of low rates are considered in Prob. 5-5. Formulations involving the use of the mole fractions as independent var- iables have been proposed by Bruno et al* and Ishii and Otto.'® Bruno et al formulated the problem in terms of N(c + 1) independent variables: the (V), {7}, and {x2 Xj) “* x). To solve an extractive distillation problem which involved nine plates and three components, Bruno et al. reported an execution time of 1.5 minutes on an IBM 360-50. Gallun® solved the same problem (except for a ‘minor diflerence in specifications) in nine seconds of IBM 360-50 execution time with six iterations. The difference in execution time of the two methods was attributed to the efficient matrix solving techniques used by Gallun. Ishii and Otto! presented a very fast algorithm based on the Newton Raphson method. The problem was formulated in terms of N(c + 2) indepen- dent variables, {V)}, {7}, {xj}. In contrast to the algorithm described herein, the algorithm was based upon making several approximations in the evaluation of the partial derivative. These approximations could lead to failure in the solution of problems involving highly nonideal solutions. However, ifthe approximations they proposed are not made, ther algorithm for solving the jacobian matrix is no longer applicable. When their approximations were made in the algorithm presented herein, Example 5-1 appeared to be converging but was far from convergence at the end of 20 trials. NUMERICAL EXAMPLES Statements of Examples 5-1 and 5-2 are presented in Table 5-1, and the solutions are presented in Tables 5-2 through 5-4. The curve fits of all data used in the190 FUNDAMENTA'S OF MULTICOMPONENT DISTILLATION ‘Table 5-1 Statement of Examples 5-1 and 5-2 Example St Example $2 ‘Components x, FAX Xs Methane! 15 6s Acetone « 2 Ethanol s Water 0 5 Methyl ae 3 Benzene » Chloroform » Conventional dsiltion Complex distillation colamn ‘Type ofcolunn columa with two feeds Column presure. atm 1 1 Feed plate location f-6 , enters on plate 6 enters on plate 21 ” . 0 2° ‘Thermal condi of ood Liguid at Fs igud at 20°F one Fis guid at 100°F “Type of condenser Tol Total 1D 9s 3 Bimal Gorn ne ‘Table $-2 Solutions of Examples 5-1 and 5-2 faa Ses — = area aaa Fg eee cnet oor byemotm fala tpn, Bata ave ae oud yih= 0.2 for alli “yy, = 0.25 for all 1 Ou, Bru ax 110[ALMOST BAND ALGORITHMS OF THE NEWTON-RAPHSON METHOD 191 ‘Table §-3 Selution values of selected variables for Example 5-1 Pa 77F Lmovh Pte TF Lymolih 1 1260 satst 0 16646203 2 13302 soa " ban 4a 3 130 300.39 2 bans 46330 4 13378 sass 5 159546515, 5 1409 e198 4 1s aR ‘ 1456 6196 15 iol 7510 7 bass 616s 16 1039 ae 5 be 6178 0 1459 61914 ° 1463 61 4 Component (mot) ou Methanol 12015) (033186 Acstone nase 049304 Methyl ace 29950, 008770 Benzene aso 06688 Chorofee, ro (002052 (Qe = 5609271 «10° Btw (y= 571635 x 10° Bra solution of thése examples are presented in Tables B-11 through B-I8. These curve fits were taken {rom Gallun.*-7 ‘The enthalpy of the liquid phase was approximated by the assumption of ideal solution behavior. The enthalpy func tion @ for the vapor phase was evaluated by use of the first two terms of the virial equation of state as described in Table B-18, The second virial coefficients were approximated as described by Prausnitz etal. The critical properties and parameters needed are presented in Table B-17. These were taken from Table BA of App. Bl, page 213, of Ret. 18, Vapor pressures were expressed by Antoine equations, and the constants for these equations are given in Table B13. Activity coefficients for each component in the liquid phase were approx- imated by use ofthe Wilson equation as described in Chap. 14. The energy terms appearing in this equation are given in Table B-15 for Example 5-1 and in Table B.16 for Example 5-2. The molar volumes appearing in the Wilson equation were curve fit on the basis of the data given in Table B-14. The fugacity coefficients 3f J"/P for the vapor phase were approximated by use of Chap. 3 (Eqs. 3-10 through 3-12) and pages 143-144 of App. A of Ref. 18 as described in Table B-I8, “The assumption of ideal solution behavior forthe calculation of the enthalpy ‘of the liquid phase was made for both examples. For the vapor phase, the192 Funpamen tats oF MULTICOMPONENT DISTILLATION Table 5-4 Solution values of selected va bles for Example 5-2 4 4 Plate CF) (ol) Phe) (moh) 1 1208 7800 2» uss 22300 2 1238 ma 2% M4ssk 22300 3 13280 7658 2s W559 m9 4 1331 1321 26 130 298 5 13508 nx» a 14382-22296 6 toss 12540 » 136s 294 7 10508 2» 14s 2290 8 i123 1246 0 \s77 maa 9 mist Dae u usiy m274 0 is? 2 4609 239 uw 26s 271 3 4639 2236 2 1371 ie u sas 22203 a 452712086 35 uss 136 1 09st 36 183s 097 5 wa a7 a was 029 16 iiss 1823 3 isise 21958 ” 1611799 » 13338 1896 1532911799 o 1352921835 » ise a3 a 151502738 2» 1138987 2 weist 12400 2 1557-22300 n Mss) 22300 Component — dol) yy, Methanol 08396003631 ‘Aecone == 238181020418. Ethanol ‘0000s goon? Water 100389 c= 13887 «108 Bah p= 13230 = 10° SkahALMOST BAND ALGORITHMS OF THE NEWTON-RAPHSON METHOD 193 ‘Table 5-5 Convergence characteristics for Examples 5-1 and 5-2 Execution Number of time Final value Example iterations (1M 360-65) ofet st © ae 10x 10-1 5 9 4s ae108 2 1 Convergence was sai o have bee achieved when the value of @ computed by og rine rad, iocre Sal vas equal to or fs than Ix 10°". The functions appearing in were all normalize. approximation which is analogous to the one represented by Eq. (5-11) was made, It was farther assumed that 00}/07 The inital set of assumed values of the variables shown in Table 5-1 were selected in a relatively arbitrary fashion so that they could not be regarded as good first guesses in the sense that they were close to the solution set. On the basis ofthese first guesses, forthe variables, Examples 5-1 and 5-2 were each solved in less than one minute of IBM 360-65 computer time (see Table 5) by use of the calulational procedure presented hercin, The computer times listed forall examples presented inthis chapter were obtained by use of Almost Band Algorithms involving (N(e-+ 1) + 2} independent variables. The computer times required to solve problems by use of these algorithms were essentially the same a those by the algorithm presented in the text. Comparison of the 2 Newton-Raphson Method with the Almost Band Algorithm for Mixtures which Form Ideal Solutions ‘To compare the characteristics of the 2N Newton-Raphson method and the [N(@c + 1) +2] Almost Band Algorithm for solving problems involving ideal solutions, the sequence of examples shown in Table 5-6 were solved by both ‘methods. For small numbers of components, procedure 2 of the 2V Newton- Raphson method is faster than the Almost Band Algorithm and conversely for large numbers of plates and a small number of components, the Almost Band Algorithm is faster than the 2N Newton-Raphson method as indicated by the results shown in Table 5-7. Since absorber-type problems generally involve large ‘numbers of components relative to the number of plates, and since the mixtures, encountered in most absorber applications do not deviate significantly from ‘deal solutions, the 2 Newton-Raphson method is recommended for solving such problems.194 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION ‘Table $6 Statement of examples used in the comparison of the 2N Newton- Raphson and the Almost Band matrix methods ‘The disillaion column had total condenser andthe ed plate was located in the mie ofthe cotuma, N2, where Nix equal tothe toll numberof tages An equimolar fed war used foreach example andthe total ow rate ofthe feed was fed at 100 mols per hour. Examples were solved wth 4 6,8 10, and 12 components. The identity ofthe particular et of compo- ‘nents sed foreach example ten i tabular form below. The temperature ofthe fend foreach ‘xample was 100F. and a column pessure of 300 lin? abs was used forall examples. The Teflon ratio was held fed at 2, and the product rates were set at 50% ofthe eed rate fr al ‘xamples, The ideal soltion K values and enthalpies were taken from Tables B-1 and B-2. ‘Number of components Component 4 6 ® 0 cH 5 cone x * x oH, x x He PCH nie x x tos x nC x x neat Cie nce * 0 ‘Table 5-7 Comparison of the 2N Newton-Raphson and the Almost Band matrix methods Camp Deven apa aint) li id ee cn ee ee wo soo hOB 1 These results were obtained by use of procedure 2, Broyden's method as modified by Benet AMDAHL, FORTRAN H OPT 2[ALMOST BAND ALGORITHMS OF THE NEWTON-RAPHSON METHOD 195 5-3 MODIFICATIONS OF BROYDEN’S METHOD {As originally propesed, the sparsity of the jacobian matrix is destroyed by Broy- den's method, Two procedures (or modifications) which preserves the sparsity of the jacobian matrices are presented. The procedures are demonstrated by use of simple algebraic examples and applied to the solution of distillation problems whose jacobian matrices are sparse. Since all derivatives may be evaluated numerically in Broyden’s method, > the necessity for programming the expressions needed for the derivatives appear- ing in the Newton-Raphson equations is avoided by use of these methods. The wide variety of thermodynamic packages which are available make these approaches very attractive ‘After the Broyden correction for the independent variables has been computed, Broyden proposed that the inverse of the jacobian matrix of the Newton-Raphson equations be updated by use of Houscholder’s formula, Herein lies the difficulty with Broyden's method. For Newton-Raphson formulations such as the Almost Band Algorithm for problems involving highly nonideal solutions, the corresponding jacobian matrices are exceedingly sparse, and the inverse of a sparse matrix is not necessarily sparse. The sparse characteristic of these jacobian matrices makes the application of Broyden’s method (wherein the inverse of the jacobian matrix is updated by use of Householder’s formula) impractical ‘Two methods have been proposed for retaining the desirable characteristics of Broyden’s method and eliminating the undesirable characteristic of the loss of sparsity of the jacobian matrix through the use of inverses. In both of these modifications of Broyden’s method, the necessity for the development of analyti- cal expressions for the partial derivations is eliminated. To initiate the calcula- tional procedure in each of these modified versions of Broyden's method, the partial derivatives appearing in the jacobian matrix ate evaluated numerically, and the jacobian matrix is updated in subsequent trials through the use of functional evaluations. The first modified form of Broyden's method is the one proposed by Gallun and Holland,” and the second modification is the one proposed by Schubert.?* Method 1. The Broyden-Houscholder Algorithm AAs shown in Chap. 4, Broyden proposed the following formula for updating. the jacobian matrix J, to obtain Ju. th Mf) dora det S Axy Ax, 625) Let the scalar ¢,. and the vectors u,,, and prs be defined as follows 1 SS aa eS tere (lsh, (527) Pier = Axl (6-28)196 FUNDAMENTALS oF MULTICOMPONENT otsTLLATION Use of these definitions permits Eq. (5:25) to be restated in the following form Seer = Set ee Cee Phas (5-29) In this algorithm, Broyden’s method is applied by updating the jacobian ‘matrices by use of Householder's formula."? Let Jy be the intial approximation of the jacobian matrix with which the iterative procedure is started. Then Jo Ax = ~fo (5:30) and Axo = —Ja'fy (531) Although the inverse of Jo appears in Eq. (5-31), it should be noted that the explicit expression of Ja ' need never be developed; only the LU factorization is required. If Jy is sparse, its inverse J is not necessarily sparse, but its factoriza- tion Le Uo is sparse. Thus, throughout the remainder of the development, in- verses are shown but the actual numerical solutions are to be found by use of the LU factorizations rather than the inverses of the jacobian matrices. Alter Axo has been used to find x, as described above, the updated jacobian matrix Jy is found as follows i= Jot uct (5-32) where After Ax, has been used to find x2, the updated jacobian matrix J as follows found J, = 3, + 0:0) pF (5-33) After Jy in this equation has been replaced by its equivalent as given by Eq, (5-32), one obtains Jos Jo t+ ment + uscapf (634) By continuation ofthis procedure, the matrix J, is found as follows Jerr =Jo+ Swept (535) ‘hus, sis posible to state the jacobian matrix Jy, in terms of the inital Jacobian matrix Jp andthe Broydensortections foreach ofthe sucesne erations. 3 ‘An algorithm is givendelow fr solving the Newton-Raphson equations by use of only the LU factorization of Jo and the Broyden update terms given by Eqs. (529), (5-30) and (531) As shown in App. 5-1 this algorithm is based ow the suczessive application of Householder’ formula to Eq, (29)ALMOST HAND ALGORITHMS OF THE NEWFON-RAPHSON METHOD 197 THE BROYDEN-HOUSEHOLDER ALGORITHM FOR SPARSE MATRICES () i=0 SOLVE Jp Axy = fy COMPUTE u,, p/ (2) SOLVE Jow= —f, Jot= ther (3) IF k= 0, GO TO (5) (8) Doy vk B= api y= ape wow By retin (8) waz 1 Oleg) + Pre ior Bm a4. Pres Ax = Wt Be (6) COMPUTE x144, fis tks ss AND RETURN TO (2) To demonstate the application ofthis algorithm, te following algebraic example i sed. In order to reduce the arithetie required to demonstrate the ‘pplication ofthe algorithm, a very simpl example was selected whose solution is seen by inspection to be 2) = lve yf and x5 Example 5-3 (Gallun and Holland,’ by courtesy Comput. Chem. Eng.) It is desired to find the set of positive values of xy, x2, and x, which make A(X) = fal) = fx) = 0, where Sis) =a = 1 fils) = 9-2 Sia)=3—3 (On the bass of the inital st of assured values xo= fl 1 ah ‘compute x, and x; by use of the Broyden-Householder algorithm for sparse matrices Souumion, (1) J =0, SOLVE Jp Axy = —fo Since the analytical expression for the partial derivatives are so easily obtained (namely, f/ax, = 2x, and dfy/Oxq = 0, m ¥ n), they are used198 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION to evaluate the elements of Jo. It is easily verified that J,(xo) = 0, also) = 1, and f(x9) = —2. Thus Since Jy is a diagonal matrix, the solution is seen to be ~ ET First, try 50 = 1 (see procedure 2, Broyden’s method in Chap. 4). Thus 1) poy fa X= Xo + Axo=] 1] +] 12] =] 32 ijt 2 LlX)=0 fole= U4 fxs) = 1 ‘Thus, the inequality (see procedure 2, Chap. 4) is satisfied, since (ay + OPP? < (1p + (-27 2? and the full step size (sy = 1) may be used to compute x,. By Eq, (5-27). w=f=lo 14 1 and and pl=axf=(0 12 1)" @ Thus Next solve Jo = a s& : 1 4 L/S][ALMOST BAND ALGORITHMS OF THE NEWTON-RAPHSON METHOD 199) “esd le By Eq, (5-46) toe 1 ti 4 SoAxg Axo FO) (ATS (yfo 12 4) | 1 . 6 "| Ga) + (1716) + (12) ~ 29 (5/4) +10 172 afin 12 ° B= aplw=(—16290 172 af] =172. = (=16729)(—(1/16) ~ (1/2) = 9729 ° ° Ax, = w+ fry = |—18| + 729)] 1/8 172. 12. ° o —2029)| 18 | = | -s/s8 12} [1029 (6) Fors, =1 someon [i] + [2x - lis] oss] [tess Si&2)=0 — falx2)= 0 fal) = —0.261 ‘Thus, the inequality given of Broyden’s method (see procedure 2, Chap. 4) is satisfied, that is, (0)? + (—0.2617]"? < f(ayay? + (1)? ° u=th=] 0 0261] Hence, Now return to (2)200 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION aE Ll "BHA mee LT | 4 ° z[Rl=] 0 2] | -os30] (3). Since k= 1, go vo (4 (4) j= tandk=t 007 ° B= aypiw=(—1629)0 1/2 4 . ax y= opie ° wewsiiel 0 Jae oli L-s | ose ° resem] | on) ie]=| ome = tel Loose 0 167290 1/2 al 8 0 ol onmr[ALMOST BAND ALGORITHMS OF THE NEWTON-RAPHSON METHOD 201, Since p; = Ax, o Piv2 = [0, —0.0862, -0.345]} 0.009 =0094 —— 1 L633 2 (177.908) + 00316 ~ 0.158 ~ o B= a,p!w = (—633)0, -0.0862, -0.345]] -0.009 | = 0.200 0.094 Ax, = w+ Bs om{ seo se |-[a (6) Try = won vine[ital [i « [9] Sils)=0 fla) = -002 — flxs) = —0.007 Thus, the inequality of Broyden’s method is satisfied (002) + (0007) < (OF + (-0261)) 0 us =f, = | -0.020 and Hence 0.007 Return to (2). ‘Schubert's Modification of Broyden's Method Jn the formulation of Schubert’ method,2" it is convenient to denote the kth approximation ofthe jacobian by G™, where the iteration number is carried as a superseript enclosed by parentheses. Then Broyden's formula for computing the ‘next approximation of the jacobian is given by [eer — 1 = 4) ayer aT agi ‘Schubert proposed a modification of Broyden’s method which takes advan- {age of the fact that in the case of sparse jacobian matrices, most of the elements GH aghs (5:36)202, FUNDAMENTALS OF MULTICOMFONENT ISTILLATION are either equal to zero or fixed constants, These known elements would be ‘modified from trial to trial by Broyden’s method. Schubert imposed the condi- tion that these known elements should remain unchanged in the jacobian revi- sion. By use of this and certain other conditions,*" Schubert obtained the following row-by-row analog of Eq. (5-36) ie ag, UB = (= 9) 0) anor ee Gab aem 637) where ns the order of the jacobian matrix, and af! =a row vector which contains the elements of the ith row of the jacobian G* Ax! =a column vector derived from Ax" by seting to zero each element of Ax" that corresponds to an element of g"” which is a known constant Jf}? = the ith element of ‘The application of Schubert’s method is demonstrated by the following ‘numerical example, Example $-4 (Gallun and Holland,” by courtesy Comput. Chem. Eng.) Make ‘one trial on the protlem stated in Example 5-3 by use of Schubert's method. SoLuTION The calculation of x” by this method is precisely the same as shown for x, in Example 5-3. On the basis of the set of assumed values xe fl 1 ae the Newton-Raphson equations Gave —9 20 olfax, 0 0 2 offax,|=—-]—1 0 0 2]faxy, -2 are solved for 4x to give 0 ax =| 12] 1 Again as shown in Example 5-3, the inequality of Broyden's method (see procedure 2, Chap. 4} is satisfied by taking s = 1. Thus, as in Example 5-3 i lB x a x) 4 AX and asin Example Sf,(1) @) [ALMOST BAND ALGORITHMS OF THE NEWTON-RAPHSON METHOD 203. int, =[2 0 0} “The known constants of gare the two zeros (These elements will have the value zero ‘or any choice of x.) Then the element 1/2 and 1 of 8X” are replaced by zer0s to give fl Since Axi!” Ax” =0, the denominator of the corrective term is zero. To avoid this division by zero, set the correction term of Eq. (5-37) equal to zero to give ei'=[2 0 0] Since the first and third elements of g{ are known to be zero for all choices of x, the corresponding elements of Ax are set equal to zero to give ° Axy? =| 1/2 ° Since $= 1, and f= 1/4 4 gp LAT (ayo 129) =e + Gora = 2 + 3 fo 12 of] =(0 2 q+ 12 9 a Thus fo 52 0} in3 i= [0 0 2] ° wo-[] 1 Since = 1, and f?? J ax" (yoo) sag? + Coreg = 0 y+ “fl fo o lo 1 =[0 0 3+ 0204 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION Thus g'=[0 0 3] Since the rows of Gare given by gf!, gi! and g{P, the required numerical values are available for solving Gi ax Namely 2 0 olfax, 0 0 52 off ax|=—| 14 0 0 alas] 1 ‘Thus 0 1/10 “3 Try 1 1 o 1 xP axis ant = 13214] —a/10] =] 14 2) [-13} [167 Then 0 {x)= 004 f(x) = -0.21 and since the inequality of Broyden’s method ((-004) + (-0.217}"? < (ay? + (IPP? is satisfied, the x obtained by use of s(!’= 1 is taken to be the assumed value of x" for the next trial Example 5-5 was used by Gallun and Holland? to compare Rroyden’s ‘method implemented with the new algorithm to the Newton-Raphson method and to Schubert's? modification of Broyden’s method. The statement of this ‘example is given in Table,5-8 and the solution is presented in Table 5.9, Sau ft ei ily meme br wold hve ‘required an excessive amébnt of computer time. The example is described by 452 Newton-Raphson equafions whose jacobian matrix contains only 3532 nonzero elements out of a possible 204,304. Of these 3532 nonzero elements, 798 are known to be constants, generally 1 or —1 due to the linearity of the equations.[ALMOST BAND ALGORITHMS OF THE NEWTON-RAPHSON METHOD 208 ‘Table 5-8 Statement of Example 5-5 am FXy FX Component (emots) (ermots) (mois) Methyl aeohot ° 02s 650 ‘Acetone ° 030 20 Ethanol ° so 30 Water so wis 50 Other specications “he column has a total condense, 50 sta, F, enters on plate 4 Fon plate 6 and F, on plate 21. The column presure is 760 mm. Feeds Fy. Fy, and F enter with enthalpies (33993877 x 10, 02859018 « 10%, and 00813942 elig mol reapetvely. Aelux ratio (L,/0) o 25 i to be vse and the bottoms 0 be withdrawn atthe rae of 285 Tb mol pet hour. The equibriam and thermodynamic data to be used are the sie as stated or Example $3 ‘Table 5-9 Solution of Example 5-5 1, Final profes, temperature, and vapor aad liquid rates Tw 4 TY 4 Pate CF) (mots) (ibmolh) Plate (*F) (moth) ib mom) 1 1336 S86 ASML T2362 2 el $138 Ses shat 382987 mse 4 tates 7702 Seat 28 stat Tat 3 1970 7456230 skelTaaE 6 16908 05525650 skat 8a 32 7 1933 18886 m2 3632 8 950 1M 560 ma 36x 9 wR ems ma ue 0 19r ne 566 3S mate Mat 62565 6 mat 36 Ro ws 16565 mat 363 Bos 71613565 7821 3432 23 12566 mex 6 13 mas 7586 man 6a 6 10 T5660 20-4632 7 ms nese) mis 3632 womst sD. mi 363 109 18s mie 363 0 es RIS BRS no (369 2 ise 385 sae aw 196367 Bie oma eT mn 363 Bisel wa 6a m2 367 is6l om KR. 76733606 Bisel m2 RD 7566880 (Continued on page 206)206 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION Table 5-9 (continued) 2, Product ditrbation ® 4, Component o mot) (to moun) ‘Methyl azohot (09832228 x (06516167 x 10? Acetone 01936700 x 10? 06133089 x 10| Ethanol 01995224 08104776 x 10 Water 01899599 « 10 0.054004 x 10 Qc= 1115040 Buh Qj = 1262069 Buy 3. Comparison of the Broyden-Hooscholier and the Schubert Algorithms Jacobian Method of sale: ion of Final derivar —Evalua- Factoriaera-aquared Execution Method tives tone ton tions norm timelst Newton-Raphion Anabtil 1413 Bo 12x) 999 Newton-Raphson Numerical 1312 2 40x10" 5994 Broyéen- Householder Analytical 44 St Agro 277 Broyéen- Householder Numerical 5s $6 11610" 3590 Schubert, Analytial 1 ” ST 89110" 77 Seubert, Namerical 1 M 34 20 10-1 2689 ‘YAMDAHL, FORTRAN H, OPT 2 54 THE BOSTON-SULLIVAN ALGORITHM? ‘This algorithm is based on the use of newly defined energy and volatility par- lametets as the primary successive approximation variables, and Broyden's method is used to iterate on these parameters. A brief review of the Boston- Sullivan Algorithm follows. The component-material balances are formulated in a manner analogous to that shown in Chap. 2 exgept forthe fact that Boston and Sullivan’ stated these bane tm fu ates (ate ha te vp ates Teme peratures were computed By use of a variation of the method wherein a different base component is used for each plate as suggested by Billingsley? Partial molar enthalpies were also expressed in terms ofthe devition function 2.[ALMOST BAND ALOORITHMS OF THE NEWTON-RAPHSON METHOD 207 ‘A formulation of the model equations follows. The S,'s appearing in the ‘component-material balance for each stage j Gans + SiMe t Speriherc =O (5-38) may be stated in terms of the “S parameters” as follows Ki it = SiS (5:39) where Ky = KylP;, Tj. (x (pl) (O0te the vapor- and liquid-phase activity coeficients are contained in this definition of the vapor-liqui equ librium ratio) Kp Viby 1 = 5) /Sq the relative $ parameter s-(1 s)”. the base $ parameter, where fl 5 Saeco See ‘The relative volatility as used in this development is defined as follows KylP To teyh Op) = Kal) a4(Py Ty beh Ol) (40) Thus Y= Ky aKa an) It follows a shown in Chap. 2 that (5-42) and thus nas (53) Lauds where = by Stn 4) ‘The K value for the base component was taken to be an exponential function of temperature as follows InKy=4,~7 (5-45) ‘The vapor enthalpies were expressed in terms of the enthalpy departure function & (see Chap. 14) as follows a= Sth Evi +0)= Syn +0 46)208 ruNDaMeNrALS oF MULTICOMPONENT DISTILLATION where © = 0(P, 7, y) and H? isthe enthalpy of component ‘in the perfect gas state atthe temperature T: The enthalpy departure function vas separated into a ‘composition correction term and a temperature-dependent ‘erm by introducing new variables @, and ©, as follows OUP, T ly) = 0, +O (TT) (5-47) The variables @, and @, are defined by Gy = [O4P, T{yt}) — MP, T*, RYT — T*) (5-48) ©, =O + AP, T, {y}) ~ (P, T, ty) (5-49) where O* = (P, T*, {y?)) T+ = reference temperature {yf} = reference set of mole fractions Next the following set of variables was introduced = (5-50) where Hj; is the perfect gas enthalpy of component i evaluated at an arbitrary reference temperature T,. Let the new variables T' and @ be defined as follows e- Fn9, a (51) r-T-5 The total vapor enthalpy may now be expressed as follows H=10+0, 4040-7) + ¥ yt (32) The variable © was partitioned into two factors ©, and , a follows 0=0,0, (5-53) where ©, is strongly dependent on vapor composition and ©, is a weak function of both vapor composition and temperature. These functions ate defined as follows = Fria = S20 = lim & (5-54) Woe / rum rdu | OyALMOST BAND ALGORITHMS OF THE NEWTON-RAPHSON METHOD 209° ‘The enthalpy of the liquid phase was treated in a manner similar to that of ‘the vapor phase. In this case, the partial molar enthalpy fi is related to the excess partial molar enthalpy as follows ohh the (5-55) where hy is the enthalpy of a pure component evaluated a: P and T of the ‘mixture, and fi is the excess partial molar enthalpy. Next a new set of variables was introduced. First, the variable 4, is defined = (hy AT ~ Th) (5-56) where fi isthe liquid enthalpy evaluated at the temperature T,. Then the total enthalpy per mole of liquid mixture is given by hea hts Yahi, (557) where em Y xt ‘The new variable 0 was partitioned in the same manner as @, namely, 0= 0,0, where = Sib ‘The energy balance enclosing any interior stage j other than the feed plate or plate above itis given by Fis sat Sensei — Sahn Selly 0 (5-58) ‘The results obtained above for vapor and liquid enthalpies together with other equations including the total-material balance equations and the component- ‘material balance equations were substituted into Eq, (5-58) to obtain the follow- ing form of the energy balance equations which were employed to calculate the liquid phase rates for any stage j (E+ a vapt So | fer) (12) Since itis supposed that an LU factorization of Jy is available, the result in- dicated by multiplication of Js ' by (—f,, ,) can be obtained by solving Jow= hes (3) for w. After w has been obtained, it is necessary to perform k +1 matrix multi- plications in order to obtain Ax... The development of the algorithm is Simplified by frst supposing that ay ¥,, and p,. (j= I, 2... k-+ 1) are known aswell as the LU factorization of Jy. Let ~ be the replacement operator. Then () j=0 SOLVE Jow= —f,5, QQ) DOj=1,2..,k 41 B= aim wows By, @) deem Although it was assumed in the development of the above algorithm that a, and v; (j= 1, 2, ..., k + 1) were available, the values of these variables may be developed in parallel with the solution of Eq. (10). First observe that when Eq, (4) is post multiplied by (~f,.), one obtains Meer = (+ ae Mes iP Wi (fies) (14) Now begin Eq. (12) and apply the above algorithm for j =0, 1, ..., k to obtain AX EF oe Mee PE A) (1s) Upon comparison of Eqs. (14) and (15), it is seen that Jew his (16) ‘The development of a formula for v4, in terms of Jj! is initiated by first solving Eq. (5) for ¥,. to give Me IE Mk (7) Application of Houscholder’s formula to Eq. (7) permits Eq, (17) to be written as follows. Yer = Et Pte (18) Clearly, the process may be continued to give Neer aa a ra) (C4 aver Ia is defined by tsi (19) Jot ther (20)ALMOST BAND ALGORITHMS OF THE NEWTON-RAPHSON METHOD 215 itis seen by comparison of Eqs. (12) and (19) that an algorithm for the calcula- tion of ¥,+» of the same form as that shown above for Ax,, , exists, and that 2 has the same role in the algorithm for the calculation of ¥,, as w has in the algorithm for the calculation of Ax, y. From the above equations, itis seen that the algorithm presented in the text is appropriate for the solution of Eq, (10) with simultaneous development of the vector ¥.., and the scalar a,CHAPTER SIX SYSTEMS OF AZEOTROPIC AND. EXTRACTIVE DISTILLATION COLUMNS, Azeotropic and extractive distillation are old processes which have become widely used since about 1930. In 1908 Emile Guillaume patented an extractive distillation process for the removal of fuse! oil from fermentation alcohol."° The name “extractive distillation” is said to have been introduced by Dunn et al. Azeotropic and extractive distillation columns are examples of columns in the service of separating highly nonideal solutions. Algorithms for solving prob- Jems involving highly nonideal solutions are described in Chap. 5. Azeotropic and extractive distllarion are the names given to processes in which advantage is taken of the nonideal behavior exhibited by certain mixtures in the presence of selected solvents. In Sec. 6-1, the qualitative aspects of azeotropic and extractive distillation are presented, In Sec. 6-2, several topics are presented which include the quantitative behavior of solvents, three-phase mixtures, and the solution of systems of interconnected columns. In the first of two methods proposed for systems of columns. a “column modular” metiod is presented in which the sets ‘of equations for each column are solved sequentially. In the second procedure, called the “system modular method,” the complete set of equations for the system are solved simultaneously. 6-1 QUALITATIVE CHARACTERISTICS OF AZEOTROPIC AND EXTRACTIVE DISTILLATION PROCESSES because many very close piling mixtures may be separated economically by use of these techniques. The s@aration of such mixtures by conventional distillation ‘methods is usually uneconomical because of the large number of stages which would be required to effect such separations. Applications of oe extractive distillation have continued to expand neSYSTEMS OF AZEOTROPIC AND EXTRACTIVE DISTILLATION COLUMNS 217 Because of the tendency toward nonideal behavior of mixtures, it is generally possible to find some component which when added to a given mixture will increase the difference between the volatilities of the light and heavy key com- ponents to be separated. The component or material added to the mixture to be Separated is called the solvent. When the solvent added to the mixture is with- drawn from the column, usually in the distillate, as an azeotrope with one or both of the key components, the separation process is called azeotropic distilla- tion The name azeatropic distillation has ako sometimes been given to processes where no azcotrope is formed and the solvent is withdrawn almost exclusively in the distillate. ‘When the solvent added to'the mixture is withdrawn almost exclusively in w bottom product without forming an azeotrope, the process is called extrac- tice distillation. Behavior of Solvents An effective solvent for an extractive distillation is one which is attracted to one 6r more of the components. This attraction of the solvent for these components reduces the volatility of the solvent as well as the volatilities ofthe components to which itis attracted, It is desirable that the attraction occur in the natural direction, that is, thatthe solvent be attracted w the relatively heavy compo- nents, However, this is not a necessary condition ‘or the behavior of the solvent. ‘Many separations are carried out in which one ofthe relatively light components js attracted by the solvent and removed in the battom product with the solvent ‘A variety of theories have been advanced for the roles of the solvent in azeotropic and extractive distillation. In the ease of extractive distillation, attrac- tion of the solvent for the certain components of the mixture is commonly attributed to one or more or a combination of the following phenomenon: (1) hydrogen bonding, 2) polar characteristics of the solvent and members of the mixture, 3) the formation of weak unstable chemical complexes, (4) chemi- cal reactions between the solvent’and one or more of the components of the mixture. A more complete statement of theories has been summarized by Berg.? In the ease of azeotropic distillation, the solvent should have the capacity to reduce the tendency of attraction between molecules. For example, a nonpolar solvent may be added to a mixture of polar mokeuks in order to increase the ‘lattes of the more polar compounds relative tothe less polar conipounds. ‘Although any one theory does not suffiiertly explain all applications of azeotropic and extractive distillation, the theories do provide qualitative rules for the selection of solvents. The role of polarity bas been elucidated by Hopkins and Fritsch!” who described the use of prodvcts obtained by oxidation of selected hydrocarbons. Because of the dissimilarities in molecular structure, the oxidation products can be arranged in the order of increasing polarity, 7 namely, estes, oxide, aldehydes, Ketones, acetals, and alcohols. In any class of compounds, the polarity is inversely proportional to the molecular weight, the polarity of straight-chain molecules is greater than that of branched-chain struc-218 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION tures, and olefinic compounds are more polar than their corresponding paraffin derivatives, Separation of Hydrocarbon Oxidation Products by Azeotropic and Extractive Distillation ‘The products obtained by the oxidation of hydrocarbons have small differences in boiling points. Typical oxidation products are shown in Table 6-1. Straight fractionation to recover a pure product from one of these groups is uneconomi- cal because too many plates would be required, One hydrocarbon reaction product produced by Celanese contains over forty components which are capable of forming more than fity binary azeo- tropes with each other.” Several ternary azeotropes are also known to exist Azcotropes of eleven of the oxidation products are listed in Table 6-2. An ex- amination of Table 6-2 shows that the separation of any one product from the ‘mixture of eleven components would be most difficult to effect by straight fractionation. If the solvent is highly polar, the volatility of the more polar compound is lowered and it is withdrawn with the solvent from the bottom of the column, The less polar compound is recovered in the top product, If a nonpolar com- ound is added, the volatility of the more polar compounds may be increased in some instances enough to permit the component with the higher polarity to be removed as an overhead product.” Separation of Acetone and Methanol The use of a polar and a nonpolar solvent to separate acetone and methanol from a mixture of tetramethylene oxide and other oxides has been described by Hopkins and Fritsch.” A schematic drawing of this purification process is shown in Fig. 1. The ternary azeotrope of acetone, methanol, and teamethy- lene, a cyclic ether, may be broken by an extractive distillation using the highly polar solvent, water. The volatility of the methanol is lowered by the water to Such an extent that the azcotrope of acetone and tetramethylene oxide may be distilled overhead in the extractive distilation column, and the methanol is Withdrawn with the water from the bottom of the column. A second column is used to separate the azeotropic mixture of acetone and tetramethylene oxides by use of the relative nonpolar solvent, pentane, An azcotrope of pentane and acetone boiling at 32°C i removed from the top of the column. The azeotrope is broken by adding water which results inthe formation of two phases, a pentane Phase and an acetone-water phase. Purification of Methyl Bhi Ketone Another example presenfed by Hopkins and Fritsch!” consists of the use of azeotropic and extractive distillation to recover part of the methyl ethyl ketoneSYSTIMS OF AZEDTROPIC AND EXTRACTIVE DISTILLATION COLUMNS 219 Table 6-1 Typical groups of oxidation products of hydrocarbons [1V. C. Hop: kins and J. J. Fritsch, Chem. Eng. 51(8):361 (1955), by courtesy ‘McGraw-Hill Book Company.) ‘Normal boing Compound Propionaldchyde a Acraein 25 Ethyl acetate ma Methyl propionate 797 ‘My aleobat 366 ‘Props alol oe See-bulsl seo 98s Table 6-2 Boiling points of selected oxygenated ‘chemicals and their binary azeotropes [L. H. Hors- ley, Azeotropic Data, by courtesy American Chemi- cal Society, Washington, D.C, 1952] Noemal bol Pure compound or binary azetrope point "C ‘Acstone methyl aeohol S85 Aestone 568 Methyl alconol methyl ety! ketone 635 Meth alcohol 67 Methyl etl ketone, water na thy alcohol, meth ethyl ketone 337 Methyl etigt tone, isopropyl aleohol 72.9) nig alcohol, water ma Eth alco ma Methyl eth ketone, tert-butyl alcohol 78.7 “Methyl exh ketone m6 tert-butyl water 09 Isopeopy alcoho, water 3 Tropcop9 alcohol ws ty alcohol 29 ‘propyl alohol, water 70 Seesby alcohol, water as uobuty alcoho, water 8 Water, butyl alcohol 927 propyl alcohol ws See. ale Bs Water 1000 Isobuty also 1080) butyl aleohot ms220 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION Figne 61 Separation of misture containing acetone and methyl alohol by azeotropic and extrac: tive distilion. (WC pins and J.J- Frisch, Chem. Eng. S18)36 (985). by courtesy McGraw “Hl Book Compan) BermacrON pioriccatw oe cera Figwe 62 Purification of methyl ethyl Ketone by azcotopic and extractive distillation [1 C. ‘Hopkins and J. Frivch, Chem. Eng. 518):381 (1958) by courtesy MeGrawsHil Book Compan]SYSTENS OF AZEDTROPIC AND EXTRACTIVE DISTILLATION COLUMNS 21 from a stream which also contained methyl-tetrahydrofuran, formals, acetals, and oxide impurities. The sequence of steps is shown in Fig. 6-2. All but a small portion of the oxide impurities are removed by azeotropic distillation, The azeo- trope is taken overhead at 65°C. In the second column, the solvent is separated from the remaining components by water extraction. The temaining impurities are separated from the methyl ethyl ketone by a water extractive distillation with a water concentration on the trays of approximately 60 percent by weight. The polar solvent, water, reduces the volatilities of the acetals. and the azeotrope of ‘methyl Ketone and water is removed overhead."® The overhead, methyl ethyl ketone and water, is then dried by pentane extraction, Separation of Ethanol and Water An early application of azeotropic distillation was proposed by Guinot and Clark" for the separation of ethanol and water by the use of benzene as the solvent. This process is based on the fact that benzene forms a ternary azeotrope ethanol and water, which has a higher ratio of water to ethanol than does the ethanol-water azeotrope. In the frst column, shown in Fig. 6-3, an azeotro- pic distillation is carried out. A two-phase liquid separation at 20°C in the decanter is used to concentrate the benzene in the reflux to the first column, ‘The solvent benzene is recovered in the second column and water is removed in the third column, Figute 6-3 Azcotropic separation of ethanol and wate by us fa benzene 0 WS Clark, Tears, ts. Chem. Eng. (London) 16:187 (198) by courtesy Th Engincers (Lenton)222 FUNDAMENTALS OF MULTICOMPONENT DISTILLATION Solvents for Hydrocarbon Separations ‘The high demand for relatively pure butadiene and toluene at the outbreak of World War II was met through the use of azeotropic distillation. Nitration grade toluene was needed for the production of explosives and butadiene was needed for the production of synthetic rubber. As a result of the need for these chemi- cals, azcotropic and extractive distillation became large-scale industrial rocesses, PovAn eoumeration of some of the specications which must be met by an efficient solvent follows. Obviously, the solvent should be noncorrosive to the equipment and should not react with the feed to form undesired products. It should produse a sufficient change in the volatilities of the components to be ‘separated so that these components may be separated with a reasonable number Of plates at an economical reflux ratio. The solvent should have an appropriate boiling point relative to the components of the feed to be separated. An azeotro- pic solvent stould have a volatility near the major component desired in the ‘overhead product and an extractive solvent should have a volatility lower than the major component to be withdrawn in the bottom product. The solvent should not be toxic, and it should be available in sufficient quantities at a reasonable price, Azeotropic Separation of Butadiene from 4 Mixture of the C, Hydrocarbons by Use of Ammoni ‘One of the first processes employed to separate butadiene from a C, hydrocar- bbon stream was an azeotropic distillation which used liquid ammonia as the solvent. A description of this process has been presented by Poffenberger et al, who also gave a typical analysis of the C, stream together with the boiling Points of hydrocarbons and their azeotropes. Other solvents such as furfural and acetonitrile are presently employed to effect this separation.'? Other Extractive and Azeotropic Separations Many soe tse te inetd forthe spurt of tice and ther stomata fom pst mire, Bon eat ang oes hn pel ib af pstiav ent Ther of phn! rater tora Scul y Dunn eat A soho ted ta af appoint ee te Te facil pan oem or ester Dhol exact war soured nd pr ao openiont 1a) ae Howson Raney of Shi Company Beau tenes a exe that tine tp msn ew dey fom he bboy oe fsa The pediion of adn fom a butane ed general equa ed