MAXIMA AND MINIMA
RAGNAR FRISCH MAXIMA AND MINIMA THEORY AND ECONOMIC APPLICATIONS IN COLLABORA TION WITH A. NATAF SPRJNGER-SCIENCE+BUSJNESS MEDIA, B.V.
MAXIMA ET MINIMA Theorie et applications economiques Dunod, Paris, 1960 Translated from the French by Express Translation Service, London SOLE DISTRIBUTORS FOR U.S.A. AND CANADA RAND McNALLY & COMPANY, CHICAGO ISBN 978-94-017-6410-0 ISBN 978-94-017-6408-7 (ebook) DOI 10.1007/978-94-017-6408-7 Softcoverreprint ofthe hardcover1st edition 1966 1966 All rights reserved No part of this book may be reproduced in any form, by print, photoprint microfilm, or any other means without permission from the publisher
PREFACE TO THE ENGLISH EDITION It is always difficult to be simple without becoming over-simple when one expounds a difficult subject, and I know of no field more deserving that description than econometrics. Here we are concerned with applying mathematical methods to the solution of economic and statistical problems. By their very nature, and especially because the formulae occurring in this type of problems contain a very large number of variables, these problems often involve mathematical ideas and techniques of a relatively advanced type, unfamiliar in general to those who might be most naturally interested in studies of this kind. I have attempted to resolve this contradiction - which seems unsurmountable at first sight - in the following way: When I have had to approach a problern requiring the use of ideas of a more advanced Ievel than that of college mathematics, I have started with a detailed discussion of a few particularly simple special cases. But I have not applied to these simple cases the elementary and easy method which might have been sufficient to solve the problern in these particular cases. On the contrary, I have applied to these simple cases the more elaborate method suitable for dealing with the general case. In this way, the reader is brought - in a quite intuitive manner, usually without the aid of any proof - to understand the steps leading to the generalisation. This generalisation then becomes almost evident. Experience has shown that proceeding in this way one is able to get astonishingly far with a very moderate mathematical apparatus, and to reach an understanding of the situation which is sufficient for practical needs. This is the spirit in which the present exposition of the Theory of Maxima and Minima was conceived. Maximum and minimum problems of the type that occurs in linear and non-linear and even non-convex programming have not been attacked in an extensive way in the present book. Their peculiarity is that the variables are constrained by inequalities, expressed by upper and lower bounds. In this case the gradient on the boundary of the admissible domain changes discontinuously at certain points, which makes the method of Lagrange multipliers - so convenient in other circumstances - of limited applicability, at least in its Straightforward and simple form. Various aspects of the programming type of maximum and minimum V
MAXIMA AND MINIMA problems are discussed in a nurober ofmemoranda and studies published in recent years by the University of Oslo Institute of Economics, most of them bearing my name. My main activity on the research front at this moment is to perfect my nonplex method for the solution of non-convex programming problems. This work has been wholeheartedly supported by the Norwegian Computing Center. It is hoped that a report of the results ofthis research may be published in the not too distant future. I wish to express my gratitude to all those who have cooperated in the publication of the present book. In the first place, I am greatly indebted to my friend and colleague and former pupil, Professor Leif Johansen, now my successor as director of the University of Oslo Institute of Economics. Relying on notes taken by him as a student in 1953 during my coursein the Oslo University, he has checked and improved the Norwegian text. This textwas subsequentlyenlarged by some new developments which I added at the time of one of my sojourns in Paris, in the spring of 1957. This material formed the basis of the French edition published in 1960. The English edition is in all essentials identical with the French edition, with the exception that a nurober of misprints have now been corrected. Hopefully we may believe that not too many new ones have been introduced. I also owe a debt of gratitude to my young friend Mr. Hävard Alstadheim, amanuensis in the University of Oslo Institute of Economics. He knows all the ins and outs ofmy published and as yet unpublished thoughts on an advanced form of macroeconomic planning, and he is thoroughly familiar with the type of reasoning contained in the present book, so essential as a basis for the theory of macroeconomic programming. Mr. Alstadheim is responsible for the checking and correcting of the MS of the English translation, and for correcting the page proofs. My thanks are also due to Professor A. Nataf who furnished the examples given at end of each chapter. Last but not least my thanks are due to Express Translation Service who presented a first draft ofthe English translation, and to Mr. Antonius Reidel and his associates in the D. Reidel Publishing Company of Dordrecht-Holland for the excellent job they have clone in producing this edition. Oslo, November 1965 RAGNAR FRISCH VI
TABLE OF CONTENTS CHAPTER I PRELIMINARY OBSERVATIONS The role of mathematics Distinction between necessary and sufficient conditions Maximum and minimum: extremal problems Neighbourhood Strong and weak sense Remarks on the necessary and sufficient conditions EXAMPLE 1 3 4 4 5 6 7 CHAPTER II MAXIMUM AND MINIMUM IN THE DISCRETE CASE - INTRODUCTION TO THE PROBLEM OF LINEAR PROGRAMMING Discrete values 8 EXAMPLE 9 Formulation of the problern of linear programming 11 Fundamental theorem concerning the solutions of the problern of linear programming 12 CHAPTER 111 PRELIMINARY REMARKS ON THE DETERMINATION OF THE EXTREMA OF A CONTINUOUS FUNCTION OF ONE VARIABLE Horizontaltangent 15 VII
MAXIMA AND MINIMA Stationary Points End-points. Open interval and closed interval Bounds. Exact bounds 16 18 19 CHAPTER IV EXACT CONCEPTS CONCERNING TAYLOR'S FORMULA AND THE EXTREMA OF A FUNCTION OF ONE VARIABLE Taylor's formula in the case of one variable The necessary condition of the first order for the case variable A sufficient condition in the case of one variable A strengthened necessary condition A necessary and sufficient condition EXAMPLE of one 22 22 23 24 25 25 CHAPTER V THE NECESSAR Y CONDITION OF THE FIRST ORDER IN THE CASE OF TWO OR MORE VARIABLES, WITHOUT CONSTRAINTS Graphical illustration. Intuitive concepts concerning the necessary conditions in the case of two variables 27 Taylor's formula in the case of two variables 29 Exact concepts concerning the necessary conditions of the first order in the case of two variables 31 The generalisation of Taylor's formula for the case of n variables. The necessary conditions of the first order in the case of n variables 33 Further remarks conceming the Coordinates of points situated on the segment joining the two points ( x ~, Further geometric remarks concerning the coordinates of the points situated inside a circle of radius R 35 x ~ ) and (x 1, x 2) 34 EXAMPLES 36 VIII
TABLE OF CONTENTS CHAPTER VI THE NECESSAR Y CONDITION OF THE FIRST ORDER IN THE CASE OF TWO OR MORE VARIABLES WITH CONSTRAINTS Intuitive representation and graphical illustration of a constraint 37 The terminology associated with 'terrain' and 'path' 38 The direct formulation of the condition of the first order 38 Formulation of the first-order condition by means of a Lagrange multiplier 39 Definition of m constraints 40 Direct formulation of the first-order conditions in the case of n variables with m constraints 41 Formulation by means of Lagrange multipliers 43 SUPPLEMENT 44 Remarks 46 EXAMPLES 47 CHAPTER VII SIMULTANEOUS SEARCH FOR THE EXTREMA OF SEVERAL FUNCTIONS- PARETO OPTIMALITY Abandonment of the investigation of a composite function. Definition of the Pareto optimality 49 Constrained Pareto optimality 49 Various applications of the notion of Pareto optimality, notably in the theory of production 51 EXAMPLES 52 Intuitive explanation of the Pareto optimality 58 CHAPTER VIII LINEAR EQUATIONS One linear equation with one variable Two linear equations with two variables Three linear equations with three variables A system of n equations in n unknowns 60 61 62 66 IX
MAXIMA AND MINIMA The elements of the adjoint 66 Cramer's formula and its application in the non-singular case 67 Theorems concerning homogeneous systems which have the same number of equations as unknowns 69 Definition of a general linear system with m equations and n unknowns. Fundamental theorems concerning the existence of solutions of a general system 73 Case of non-zero solutions 74 Generaltheorems concerning homogeneous systems 75 EXAMPLES 76 Summary of the results concerning linear equations in two unknowns CHAPTERIX LINEAR RELATIONS BETWEEN VECTORS, BETWEEN FUNCTIONS, AND BETWEEN EQUATIONS Definition of a 'vector' and a 'system of vectors' 86 Linear dependence of two vectors 87 Graphical representation and mathematical expression of the linear dependence of three vectors 88 Linear dependence of m vectors 89 Linear relations between functions 91 Linear relations between linear equations 92 The rank R of the complete matrix of the system of equations. Degree of reduction and degree of freedom 94 SUPPLEMENT 95 EXAMPLES 98 CHAPTER X SECOND-ORDER CONDITION AND SUFFICIENT CRITERION FOR LOCAL EXTREMUM IN THE CASE OF TWO VARIABLES WITHOUT CONSTRAINTS The quadratic form, the dominant part in the neighbourhood of an extremum, for a function of two variables 101 In the case of two variables, the quadratic form can be positive definite, negative definite, indefinite, and serni-definite 102 84 X
TABLE OF CONTENTS These four cases correspond to the cases considered in the discussion of the extrema of a function of one variable 103 Preliminaries to the study of quadratic forms 104 Variations of the value Q when the argument varies along the unit circle 106 The necessary first-order conditions for the extrema of a quadratic form 108 The characteristic equation in the case of two variables 109 The characteristic roots and the complete classification of all possibilities in the case of two variables 110 SUPPLEMENT 112 EXAMPLES 113 CHAPTERXI SECOND-ORDER CONDITIONS AND SUFFICIENT CRITERIA FOR THE LOCAL EXTREMUM IN THE CASE OF SEVERAL VARIABLES WITHOUT CONSTRAINTS Complete formulation of the sufficient conditions in the case of n variables 119 EXAMPLE 120 CHAPTER XII SECOND-ORDER CONDITIONS AND SUFFICIENT CRITERIA FOR THE LOCAL EXTREMUM IN THE CASE OF SEVERAL VARIABLES WITH CONSTRAINTS Preparatory discussion of the case of two variables with one constraint 122 The structure ofthe quadratic form in the case oftwo variables with one constraint 124 The quadratic form in the case of n variables with m constraints 125 The characteristic equation in the above case (n variables, m constraints) 126 Does the quadratic form take effectively a maximum value and a minimum value? 127 XI
MAXIMA AND MINIMA Provisional analysis on the hypothesis that the values satisfying the fi.rst-order conditions are taken by the form and that there are no difficulties concerning the endpoints 128 Graphica1 analysis tending to show that the hypothesis is confirmed 131 Exact analysis confirming the hypothesis 133 EXAMPLES 136 CHAPTER XIII A BRIEF ACCOUNT OF THE THEOR Y OF MATRICES AND OF THE CALCULATION OF DETERMINANTS Definition of a matrix and examination of the special case of square matrices 144 Symbolic operations on matrices 145 Definition of a determinant 147 Determinants of order one, of order two, of order three 148 The sub-determinants of order two contained in a determinant of order three 149 Some remarkable relations 150 The adjoint elements; fundamental formulae- notably (13.25) - for the expansion of a determinant along a row or a column 151 Elementary theorems about determinants 152 Definition of the rank of a matrix 155 Examples of the calculation of rank 156 Sylvester's formula for the sub-determinants of the adjoint 157 Gram's formula 158 Hadamard's theorem 160 Theorem concerning the determinant of the product of matrices 161 Characteristic polynomial of a matrix 162 The absolute minimum of necessary knowledge about matrices and determinants 166 EXAMPLES 167 Theoretical Appendix on Complex Nurobers 171 Bibliography 174 Index 175 XII