A mathematical theory of pure exchange economies without
the no-critical-point hypothesis
Citation for published version (APA):
Geldrop, van, J. H. (1980). A mathematical theory of pure exchange economies without the no-critical-point
hypothesis. Stichting Mathematisch Centrum. https://doi.org/10.6100/IR94027
DOI:
10.6100/IR94027
Document status and date:
Published: 01/01/1980
Document Version:
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers)
Please check the document version of this publication:
• A submitted manuscript is the version of the article upon submission and before peer-review. There can be
important differences between the submitted version and the official published version of record. People
interested in the research are advised to contact the author for the final version of the publication, or visit the
DOI to the publisher's website.
• The final author version and the galley proof are versions of the publication after peer review.
• The final published version features the final layout of the paper including the volume, issue and page
numbers.
Link to publication
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal.
If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, please follow below link for the End User Agreement:
www.tue.nl/taverne
Take down policy
If you believe that this document breaches copyright please contact us at:
openaccess@tue.nl
providing details and we will investigate your claim.
OF PURE EXCHANGE ECONOMIES
WITHOUT THE
OF PURE EXCHANGE ECONOMIES
WITHOUT THE
NO-CRITICAL-POINT HYPOTHESIS
PROEFSCHRIFT
~ .
TER VERKRIJGING VJ\N DE GRAAD VJ\N DOCTOR IN DE TECHNISCHE WETENSCHAPPEN AAN DE TECHNISCHE HOGESCHOOL EINDHOVEN, OP GEZAG VAN DE RECTOR
MAGNIFICOS, PROF. IR. J. ERKELENS, VOOR EEN COMMISSIE AANGEWEZEN DOOR HET COLLEGE VAN DEKANEN IN HET OPENBMR TE VERDEDIGEN OP
VRIJDAG 14 NOVEMBER 1980 TE 16.00 UUR
DOOR
.JOHANNES HUBERTUS VAN GELDROP
GEBOREN TE ROTTERDAM
1980
Prof.Dr. S.T.M. Ackermans
en
Aan de naqedachtenis van mijn ouders,
Chapter 1. Introduction
1.1. Commodities, prices and .preferences 1.2. The set of equilibria
1.3. The set of local Pareto optima 1.4. Disastrous allocations
1.5. Survey 1.6. Some examples
Chapter 2. Preliminaries on manifolds Introduction
2.1. Differentiable mappings and submanifolds 2.2. Sard's Theorem
2.3. The Whitney c"" topology 2. 4. Transversal! ty
2. 5. Characterization of local Pareto optima 2.6. The submanifold
r.
\
Chapter 3. The set T Introduction
3.1. A first definition ofT 3.2. An alternative definition of T 3.3. Comparison with the results of Smale 3.4. T is not dependent on scale transformations
"" 1 m 3.5. T is a dense subset of C (JR ,B)
.. 1 m
3.6. Is T open in c (JR ,:R) ?
Chapter 4. '!'he set of equilibria in a pure exchange economy Introduction 4. 1. Regular economies 4.2. An example 1 3 6 7 7 8 12 13 18 19 21 22 28 31 31 36 39 42 44 52 55 55 62
Chapter 5. The set of local strict Pareto optima in a pure exchange economy
Introduction 5.1. Regular pairs
5.2. Local structure of ecro:i=1 m 5.3. A local optimal part of·e
cr 5.4. Some examples
Chapter 6. Trade curves Introduction
zi, u)
6.1. Construction of a vector field, generating trade curves
6.
2. An example Epilogue References Index Samenvatting Curriculum vitae 69 70 74 80 84 94 96 103 105 106 108 109 110CHAPTER 1
INTROVUCTION
In this monograph we consider a
pure erechange economy
without producers. There are R. durable goods and m agents. We assume R. 2: 2, m 2: 2.' R.m
Each point x = (x1, ••• ,xm) € E represents an
allocation,
wherexi
=
(x~,
••• ,x:) € E1 is thecommodity bundle
ofagent
i (1s
is
m) • We assume that for each agent the whole of E1 is hisconswrrption set,
the set of possible commodity bundles. For a more elaborate discussion of the terms "goods", "consumption sets" and the sign convention, concerningx~
< 0, xi = O,x~
> O, see Debreu [1].With each commodity, say the h-th one, is associated a real number, its
• h h .
pnce
p • The price p may be positive (scarce commodity), null (free com-modity) or negative (noxious comcom-modity). Thepl'ice system
is the R.-tuple1 R. R.
p
=
(p '• • • ,p ) € .E •1 R.
The
vatue
of a bundle a = (a , ••• ,a ) , relative to the price system p, is · the standard inner productR.
p•a :=
L
h=1
h h
P a •
We assume p {: 0. Two price systems p and q are
equivalent
i f there is sane positive A € E such that q=
Ap. Hence, if we take a price system p, weR.-1
always choose p € S , i.e. p•p
=
1.R.m R.-1
Each point (x,p) € E x s · defines a
state
of the economy.Given two bundles a and a' in ER. one and only one of the following three alternatives is assumed to hold for agent i:
(1) a
is preferred to
a';'· (2) a
is indifferent to
a';(3) a'
is preferred to
a.It is convenient to introduce a
preference relation
f
on E1 for agent i. Then the foregoing alternatives read as follows:{1) a' .S a and -, (a .S a') i i (2) a' :ka and a~ a' ; i ~ (3) a :;, a' and -,(a' :S a) • i i
The binary relation :k is assumed to be
Pefle:dve
andtransitive.
i t
The preference relation :;, is said to be
continuous
if for each a' e: JR the.
t
it
sets {a e: JR
I
a~ a'} and {a e: JRI
a'~ a} are closed.·~ . ~ t
If
f
is continuous, there is a continuous function ui: JR -+ JR satisfying for ·all a, a':For a proof see Debreu [1], page 56-59.
Suph a function ui is called a
utiZity function,
representing the preferen-ces of agent i. In order to use the calculus of differentiable manifoldsand maps we assume utility functions to be
smooth.
See Chapter 2.In economic literature several assumptions are proposed about preference relations and, consequently, utility functions, the relevance of each of them being a matter of taste or realism. We mention here:
(1)
nonsa 1-a 1-on,
t . t.
i.e. for each a e: :R t there is some a' e: :R t preferred to a,. (2)convereity,
i.e. for each a e: :Rt the set {a' e: JRtI
ui(a')~
ui(a)} isconvex;
(3)
monotonicity,
i.e. u. (a') > u. (a) whenever a' ;. a and a'h ~ ah for allh.~ l.
We
donot make any of these assumptions. In
OU!'modeZ the aZass of util.ity
functions coincides with the clAss of smooth functions
:Rt -+JR.Given some u'!:;ility function ui and some bundle a, the set of points a' , preferred to a by agent i, can have locally different shapes depending on the
gradient
To get same insight in the possible situations we assume for the moment u
1 {a) = 0 and i = 2. Then, up to degeneracies one has the following
A} Here Du
1 (a) .; 0. Agent i has a clear idea in which directions to move in order to increase his utility.
B} 'l'he same picture as (A) but Du
1 (a} = 0. It is possible that some other utility function, representing the same preference relation, has a gradient .; 0 at a. C)
0
D)0
E)1 •
2. The
.6e.t
o
6
eqlUU..bJti.a.Here DUi (a) = 0 and the point a represents a local minimum for ui. Each direction improves the posi-tion of i.
The point a represents a local maximum for u
1• This situation is described as
Zocat satiation.
The point a represents a
point of
doubt.
If some direction improves the position of i, its opposite direction equally does.Now we assume that aqent i is endowed with some
initiat
bundle r i e BR,.R.-1
Then, given some price system p e S , he faces his
budget set
It will be his aim to maximize his utility function ui on this budget set. If such a maximizing bundle exists we can find it in the set Ei (p,ri) of points x
Given initial bundles r = {r1, •.• ,rm) and utility functions u
=
{u1, ••• ,um) the set Eex(r,u) of
e:ctendsd equil-ibrium states
is defined by:{x,p) e E (r,u) if and only if ex 1.2.2.
!
I
xi =I
ri • i=1 i=1 m xi£ Ei(p,ri)]i=l • See also Smale [15].The first· condition is inspired by the definition of pure exchange econo~ mies, in which allocations are admissible if and only if they can be
ob-m
tained by redistri):>ution of the total, resowoces l:i=l ri, given by the ini-tial endowments.
The second condition is the first order criterion which implies that each agent i finds himself endowed with a bundle in his budget set which is. a possible local maximum for ui. Hence the adjective "extended".
Given an economy, defined by (r,u) = {r
1, ••• ,r ,u1, ••• ,u ), we al~o
con-tm 1-1 m m
sider the set Ecr (r,u) c: lR x S of critical equilibrium states (x,p) for which the following holds:
m m
I
xi =I
ri ,i=1 i=l
1.2.3.
Then, as is easily seen, Eex(r,u) c: Ecr(r,u).
'
The system {1.2.3) consists of 1+m-l +m1 equations in m+m1+1-l unknowns p,,x,p). Roughly speaking· there is in general locally one solution (or none), .so the set Ecr(r,u) is in general a discrete one.
'l'o make more exact the notion of the vague statement "in general" we use methods of GLobal, JmaZysis, especially of transversaUty theorems. See for instaJ).ce Golubitsky and Guillemin [6], . Hirsch (9], or Dierker [2J.
introduced I in Chapter 3. For this set T we are able to prove the
. General :vesult I:
Tis dense and far each
u
£ Tthere is a dense.set of initial endowments
r
far which
Ecr(r,u)is
disCl'ete.
Since the set of smooth functions in its turn is dense in the set of
con-I
tinuous fuhctions it follows that the assumption of smoothness for utility functions is not too bold.
Moreover, the question whether T is also open, is in this context less in-terestinq,:since in the set of continuous functions no neighbourhood is filled up ~Y functions satisfying sane differentiability condition, whereas
o~nness
ot
T by its very nature may only be studied in the set of at leasttwice differentiable functions. See Chapter 3.
Smale [15] restricts the consumption space of each agent to the closure of the positive orthant of
Il'
and, using only utility functions with non-zero gradients troves also for a certain set Y of utility tuples that i t has the desired pr~ties: there is a dense set of r for which Eex(r,u) is discrete.Y itself is dense.
His methods are based upon the existence of the
nornred gradients
V E JaR, •
As will be shown in several examples there are many pairs (r,u) for which, according to our results, the set Ecr(r,u) is discrete, and contains states
{x,p) for whiCh some bundles xi are stationary points for the corresponding utility function ui.
Moreover, ~e shall show that our methods, applied to utility functions satisfying Smale's
no-critical-point hypothesis,
lead to the same set of economies as obtained by Smale. So~our results generalize those of Smale.
A second topic in pure exchange econanies is the set of local
strict Pa:l'eto
optima:
Given sane w e
Jr/·,
being the total resources in the economy, and utility functions ui one considers the set 8(w,u) of admissible allocations x e Etmfor which the following holds:
There is an open neighbourhoOd
0
c JR!m of x such that for each admissible allocation y e0,
yF
x, there is at least one i such that ui(yi) < ui(xi}. In this context prices, and consequently budget sets, are not involved and the only. criterion for redistribution of w is non--decreasing of utility. So, given sane admissible allocation r, agents are willing to accept sane admissible allocation x if and only if ui(x1) ~ u
1
(ri)J~.1
• If we take intoaccount that such redistribution has to be realized by exchanging small amounts of goods, it seems reasonable to assume that no trade takes place from r e B.
1-1
It will be shown that x e 8 implies (x,p) e Eex (x,u) for sane p e S • Acting in the same spirit as in the definitions of equilibria we introduce sets Sex(w,u) and acr(w,u} as follows:
(1)
(2}
X € 6 (w,u) if and only if X is admissible and (x,p) E Eex(x,u} for
ex t-1
sanepeS
X € 6 (w,u) if and only if X is admissible and {x,p) € E
0r(x,u) for cr t-1 sanepes Bence 8 c 6ex c 6cr" The conditions 1. 3.1.
I
Y
xi • w , i=1 mDui (xi) = li p]i=1
constitute a system of 1 + tm equations in m + tm + R. - 1 unknowns (A ,x,p) , the solutions of which determine points x e acr· In general the set of solutions is parametrized by m- 1 variables.
If u
E T th~Peis a dense set of total reeouraes
w~for
~hiahe
0r(w,u)
is a
submanifoZdl of dimension
m-1.r
For the definition of submanifold see Chapter 2.
It will be shown that eex is the intersection of ecr with a closed set. Hence the set e ex and a fortiori the set 9 may"not be a submanifold. Moreover, i;E one restricts the consumption sets to the open or closed
posi-i !/, •
tiVe Orthant in JR 1 the preCiSe deSCrJ.ptiOn Of the StrUCture Of
a
Or e', • , ex,
is a compli~ated affair. It turns out to be necessary that one invokes the theory of stratified manifolds with corners. See for instance Wan [23], Smale [17], or Schecter [11]. We do not enter into these problems. We only remark that, as for the structure of ecr our results generalize those of Smale [16] :in the same sense as described in the case of equilibria.
i
Omitting the no-critical-point hypothesis leads to some special effect, If all of the /!utility functions ui have som.e critical point, say zi' then
!/,-1 m
(z,p) € E~{z,u) for all p € S and z E ecr{w,u) for w
=
ti•l zi.We may coruiider such a point z as
disastrous
for the economy. None of the agents has some specifiedshort run demand,
indicated by the direction of the gradients Dui (zi), being all zero.It will be shown that for the set of pairs (r,u), (w,u) respectively, in-dicated in the general results I and II there are no admissible disastrous allocatioruk.
1.
5.
SU/I.vey o6 :the conten:t.s o6:th.U.
rnonogJt.aphIn Chapteri 2 we give a summary of some standard. material of global analysis, contained in Sections 2.1 until 2.4. Section 2.5 is Cl.evoted to a proof of a well-known theorem on local Pareto optima. This proof is based upon methods, used by the author in his paper [5]. In Section 2,6 a s~ifold f is introduced, which plays an important role in the sequel.
Chapter 3 bontains the Cl.efinition of the set T and the proof that T is Cl.ense.
Fur~ermore,
a discussion about the openness of T is given in 3.6.The proof of the general result I is the main topic of Chapter 4. Moreover, we introduce some subsets of T, which are open in the set of utilities u. chapter 5 is devoted to the proof of the general result II and some criteria for points in ecr on which one can decide whether they are points in
e
or not, are discussed.Chcipter 6 contains some elements of trtade aUl'Ve8 • The introduction to this
topic is postponed to the first section of Chapter 6.
1. 6. Some.
examptu
Before we end this chapter we present some standard illustrative pictures, intended to give an impression of the relationship between equilibria and
'
Pareto optima.
We assume
t
= m=
2 and use the so-calledEdge~Portth-bo:r:
in :rt2• See also Debreu [1], Hildenbrand-Kix:man [8], Dierker [2], Smale [16], and many other authors.The horizontal axis represents quantities of commodity I and the vertical 2
quantities of commodity II. Let w e :R be the total resources in the eco-nomy. We measure quantities for consumer 1 from the origin and quantities for consumer 2 from w. Then each point in :rt2 represents an admissible location. For instance, the origin corresponds with the allocation
4 4
(O,w) e JR , whereas w represents the allocation (w,O) e :R • Points within the open rectangle through 0 and w and sides parallel to the axe.s corre-spond with allocations (x
1,x2) ,in the positive orthant. The closure of this rectangle is generally denoted as the Edgeworth-box. We do. not confine our~ selves to these allocations, so our set of allocations is the whole of JR2• Given utility functions u1 and u
2, throuqh each point in JR 2
there pass two curves, the indiffertenae curvee for u1 and u
2, indicated by a solid curve for u
1 and a dotted curve for u2• If we consider equilibria we indicate the initial bundles (r1 ,r2) by the initial bundle r1•
Each allocation (x
1,x2) is indicated by the bundle x1 • . '(1) If u
1 and u2 both satisfy strong convexity assumptions we have pic-tures as in Fiqure 1 •
---+---tr
Figure 1.Edge-worth box for strongly conv~x utility functions ..
Obviously, each point where the indifference curves are tangent is a point; of
eex
and even a point of 8.The set 8 is the curve, denoted by Edgeworth's cont:r>aat curve (o-o-o) t passihg throur;rh.all these tangent points.
Given r1, one finds the equilibria (x,p) by selecting those points x 1 on 9 where the common tangent passes through r 1, as is the case .for x1
and y1, but not for z1•
(2) In Figure 2 the utility functions satisfy weak convexity conditions. The s$t of equilibria contains a one-dimensional set. Intuitively one sees that a slight perturbation of the utility functions breaks down the whole structure, in accordance with general result I.
I I
.
"
) I
Figure 2~
(3) In Figure 3 the utility functions do not satisfy convexity conditions •
• w
~I
Figure 3.
Edgeworth box for concave utility functions.
The indifference curves u
1 == 2 and u2
=
1 are tangent at x1 and x1 is a point of ecr' Clearlyx
1 is not a point of 9 since each point in theshaded region is better for both of the agents. Equally x
1 does not correspond with an equilibrium since u
1 and u2 both increase along the line x
1r1•
(4) In Figure 4 the function u
1 satisfies the convexity condition but u2 does not.
I I
Fiqure 4.
Edgeworth box for u
(5)
As on~ sees the situations in x
1 and y 1, both being points of diffetent from each other.
x1 is!not a point of 9, since all points in the shaded region
eex,
are are better for both of the agents. The concavity of u2 overrules the con-vexity of u1 at x1•
y1 is a point of 8, due to the fact that the convexity of u1 is
stron~er than the concavity of u2 at y1•
In FiJure 5 we consider utility functions defined by:
where we assume a 1
+
a2 tl w. I I U2'--;#' ; u1::.-J_..,1-'~.__ I I I I I I I I l l l Figure S. •wEdgeworth box for utility functions wtth satiation points.
6 = 9 ex is the closed segment between a1 and w - a2• If r l lies between th~ verticals through a1 and w -a2, the set E ex (r, u) consists of only one p¢int. Otherwise E (r,u) is empty.
1
ex
In cake w = a1 + a
2 the set
e
ex consists of only one point, namely a1, but nowe
is the whole of :at2. Moreover 1 for every r1 the point cr
(a1 ,p) where p .1. a
1- r1 represents an equilibrium. Clearly the point a1 is disastrous and we do not have the general situation as stated i~ general results I and II.
CHAPTER 2
PRELIMINARIES ON MANI'FOLVS
The first four sections of this chapter contain a summary of standard topics from global analysis, in a form adapted to the context they will be used in. For instance, all of,the manifolds considered here are submani-folds of some Euclidean space.
IJ) Section 2.5 we prove a well-known theorem on local Pareto optima. This theorem has been proved formerly by Smale [20], and Wan [22], but the proof given here is basically different from theirs. See also [5].
In Section 2.6 we introduce a subset
r
of :Rtm and show thatr
is a sub-manifold. This setr
plays an important role in the sequel.OUr main references for this chapter are [6] and [9] •. We do not, at least not before Section 2.5, present proofs of the statements we make. They can be found in [6] or [9].
· n 1 n .1 nT
Points in E. are given as a row x
=
(x , ••• ,x ) or as a column (x , ••• ,x ) ,~ context making clear which form is chosen.
The topology in En will always be the metric topology, induced by the ·standard inner product, defined by
n X. y :=
I
h=l h h X y ' llxll := (x • x)! • n-1 n nThe unit sphereS c E. is the set of points x € E. satisfying llxll • 1. Furthermore, given a subset
u
of En its interior is denoted byu
or intu,
and its closure byu.
2.
7.V.£6 nte.n.t<.a.ble. mapp.in.g.6
and
.6ubma.n.i..6o£.cl6
~ n
Let 0 be an open subset of :R , and k a nonnegative integer.
2 .1.1. DEFINITION. ck (U, lRm)
is the set of az:t maps
f: u + E.mbeing
ktimes
diffePentiable UJith alZ. del'ivatives
upto ord.eP
kcontinuous on
o .
..
2.1.2. DEFINITION. c"' (U,JRm) :=
n
Ck(O,JRm)is the sat of smooth maps
We shall extend the notion of differentiability up to order k to maps X + Y, where X and Y are submanifolds, and we shall define later on the sets
k ..
C,(X,Y) aqd C (X,Y).
2 m
Let f E: C i(U,.JR) and x
0 E:
u.
Then by Taylor's theorem there exists a unique1 n m .
linear m~ JR + JR , denoted by Df (x
0) , a un1que symmetric bilinear form D2f (x
0) : ~n x JRn + JRm, and a map p: u +
lffl
such that for all x ;;: JRn suffi-cientlyc~ose
to x0 the following holds:
i
where
1 n n 1 m m
With respect to the coordinates (x , ••• ,x ) on :R and (y , ••• ,y ) on JR the
del'ivative
Df(x0) has the matrix[
()fi
l
i=l , ••• ,m- j
<xo> .
ax
J=l, •••,n
Equally, tlb.esecond del'ivative
D2f (x0) for f: JRn + lR is denoted by 2.1.3.
~[ 32f Ji=1, ...
,n i i j . . !ax ax
J=l, ••• ,n • 1 m 1 k m hCHAIN RULE,
Let
f ;;: C (U,.JR )~ q £ C (V,.JR ) and f(U) <= V c: JR •Ten
the composition
q • f E c1 (U, lRk) and D (q o f) (x0) .. Dq (f cx0> ) o Df (x0)
fOP, each x 0 e
o.
Before stating the implicit function theorem we need some notations.
n m 1 k
Let U c lR and V c lR be open and F E
c
(U xv, lR ) • Then, given x0 E u, Yo € v, we define (1) F E cl (U,JRk) by F (x) := F(x,y 0) , Yo Yo (2) F E cl (V ,JRk) by F (y) := F(x 0,y) • xo xo
2.1.4. IMPLICIT FUNCTION THEOREM. Let U c lRn and V c lRm be open sets. Let
oo m
(x0,y0) E u x v and FE
c
(uxv, lR) be such that rank DFx0<y0) =.m.
Then there is an open neighbourhood
u•
cu
of x0 and a map
cp E C00 (U' ,JRm) satisfying
(2) F(x,cp(x)) = F(x
0,y0) for aZZ x E u•
See [9]~ page 214.
As a consequence of the implicit function theorem (and vice versa) one has
, n oo , n
" 2.1.5. INVERSE FUNCTION THEOREM. Let U c lR be open and f E C (U,JR ) • Let
x0 E
u and
rank Df(x0) = n. Then there are open neighbourhoodsu
1 ofand oo n • •
x
0
~ u2 of Yo := f(x0> a map g Ec
<u 2,JR ) satisfy1-ng(2) f(g(y)) = y for aZZ y E u
2, and g(f(x)) = x for aZZ x E u1•
See [9]~ page 214.
Now we come to the definition of a submanifold of dimension k as a subset of some Euclidean n-space where n ~ k. Intuitively a k-dimensional
submani-k
fold has locally the structure of an open subset of E .
2.1.6. DEFINITION. A subset y c lRn is a submanifoZd of JRn of dimension k
there e:dsts an open neighbOU'I'hood
u
of
y0 and
a
q~ E:c""
(U, lRn)such
: +
thqt
q~(y0
) = 0~ rank Dql(x) = nfor all
x E: u~an4
q~ (V)=
Yn u,
: { 1 n n
I
k+l n }zvh~re V :• (z , ••• ,z ) E: E z = ... • z "' 0 •
Moreover~
the pair
(U,cp)is called a submanifotd chart for
Yat
y0•
So the submanifold chart (U,cp) provides a local parametrization of Y by means of the first k coordinates of cp(x), in a neighbourhood of y
0• For example, each open subset of JRn is an n-dimensional submanifold of lRn, and a Q-dimensional submanifold of En is a discrete set.
It shouU
beemphasised that the topology on a submanifotd of
lRnis the
one~
induced by the topol-ogy on
En.J:f not nec;:essary we do not specify in the future the Euclidean space in w~ich a sqbmanifold is contained, nor the dimension of the submanifold. The definition of submanifold is not always as manageable as desired in order to find out whether a subset 'of JRn is a submanifold or not. The followingitheorem will be useful in the sequel.
2.1.7.
~OREM.
A subset
Y c:x(l
is a submanifotd of
JRnof dimension kif
ani/. only if for every point Yo
E: Ythere e:dsts
anopen
neighbour-hood 0 of
y0 and
a
ljl E:c ..
(O,mn-k)such that
rank Dljl(x)=
n- kfor
aZZ
x E:0
and
Tji+(O) = Y n0.
See [6]~ page 9.
As one sees the submanifold Y is locally defined as the solution set of the equation 1jl (x)
=
0, constituting n - k equations in n unknowns. Due to the implicit function theorem the set Y is locally parametrized in a smooth wayby sane k""tuple fran the n coordinates in En. +
The kernel Dljl (y
0) (0) of Dljl(y0) has dimension k.
on
the other hand, if (U,cp) is a submanifold chart for Y at y0 one has the k-dimensional subspace
-1 , n
Dq~ (()) (V) of JR • Fran the definitions i t follows
foljl.(y
0> <O>
= Dql-1co> <v> .
i
2.1. 8.
~INITION. Let
Y c ».nbe a k-dimeneiona't submanifo'td. Let
Yobe a ,,
POfnt in
Y and (U,cp)a submanifotd chart for
Yat
Yo·Then
Ty0Y
is
t~ set of pairs (y
0
,~y)~ zvhere
ByE: Dcp-l(O) (V).The set
Tyaarted the tangent spaae to
Yat
y0•
It
hasthe stz>Uetu:l'e of a
k-dimensional veetor
~>paae~
isomorphie to
Dip -1 (0) (V).Gecmetrically Ty Y is the set of all tangents to smooth curves on Y 1
pass-O .
ing through y0, considered as a Euclidean space. In general, speaking of Ty
0Y, we only give the second ccmponent of the pair (y0,6y), i.e. oy e ~n. Now we introduce differentiability and derivatives for maps, defined on submanifolds.
2 .1. 9. DEFINITION.
Let
xi c: E. nibe
ki-dimensional submanifo
Zd.8for
1 "' 1, 2.Let
xi e xi and (Ui ,cp1)be submanifold charts for
xiat
xi (i = 1,2) and f:x
1 -+ x2be a
map auehthat
f <x1 l = x2• Furthermore~v
1 andv
2are defined
asin
2.1.6.Let
kbe a nonneqative inteqer.
(1) Themap
f: x1 -+ x2
is said to be of
classd"
at
x1if the
map-1 . . • k
cp
2 o f o cp 1 : V 1 + V 2 1.8
of
a wasB Cat
0 e V 1 •(2) ck<x1 ,x2)
is the set of maps
f: x1 -+ x2being of elaas
ckat
every point
x1 e x1• 00 .. k (3) C (X 1,x
2) :=n
C (X1,x
2) k=O(4)
Given
f ec
1cx
1
,x
2
>~
x1 ex
1 and (x1,ox1) eTx
1
x
1
~ the map
Tx1 f: Tx1 x1 + Tx2 x2
is defined
asfo
l'!A::Ms:
or shortly:
(5) Tx
1f<x1,6x1> := (f(x
1
),Df(x1
)6x1
>~ ~herethe definition of
DfCx1> foll~sfrom
(4).In general, speaking of the derivative Tx/• we only give the second part, i.e. Df(x1)ox1, or shortly Df(x1).
2.1 !10. LEMMA. The
Cartesian product
x1 x x2of wo submanifolds is a
sub-manifold
and dim{XI
'1'(1 ) (Xl X X2) = { (Xl ,x2,ox1 ,ox2)
j
oxl € 'l'x1X1' OX2 € 'l'x2X2} • f'l'x2See [ 6]~ page 5.
. 0 1..- •
2.1.11. ~INI'l'ION.
Let
f €c
(X,lR)Ulru:re
X-z.s some submanifoZd. Then the
support
Supp fis the a losure of
f + ( lR \ { 0}) •2.1.12. DEFINITION.
Let
{ua}ae:Abe a family of subsets of a submanifoZd
x~suah that
UA ua •x.
In other
U~ord.s~ {Ua}a£Ais a aover of x.
ae:
Then
{Ua}ae:Ais said
tobe loaa'Lly finite if for every
x € Xthere
is
anopen neighbourhood
0 c:x
of x
suah that
0 n ua =rl
for aU
but a finite number of
a'sin
A.2/1.13. TBEOREM (Existence of a partition of unity, subordinate to an open cqver of X.)
L~t
Xbe a submanifoZd
and {ua}a€Aan open aover of
x~
i.e. all of
t~e ua
are open
and U u = X.There is a family
{f } Aof smooth
! aEA B· a ae:
mdpe
x + lReatis fyi7lf!
I
q>
fa(x) e: [0,1]for aU
a € A andall
x e: x;(2) Supp fa c: u<X
for aZZ
<X € A;(3) {Supp f } A
is a loaaZly finite aover of x;
a <X€(4)
L
fa(x)=
1for all
x €x.
(Thie is a finite
sum, due to
aE:A
locally finiteness. Moreover, this shOUis that the interiors of
Supp faform
anopen, locally finite aover of
x.)see [9]~ page 43.
2.1.14. COROLLARY.
Let
x
be a eubmanifoZd. Let
u
andv
be open subsets of
XU~ith
u
c:v.
Then there is
an f € c~(X,lR)suah that
ffx) • {
~
• f(x) •
1 isre
[6]~ page 17.if
X € U ,if
Xf
V ~otheruise.
2 .1 .15. COROLLARY.
Let
c
be a a tosed subset of
:m.n.
Then there ewists a
81T/Ooth function
f: JRn + Esuah that
f (x) ~ 0everyuJhere and
....
c
= f (0) •See [6], page 17.
2.2.
Sand'~Theo4em
Let a € JRn and b E: En and bk. > ak, k = 1, ... ,n. '!'hen C (a,b) is the closed
n k k k
block consisting of all points x E: E , satisfying a s x s b , k = 1, ••• ,n. '!'he volume of C(a,b) is
n~=l
(bk- ak).2.2.1. DEFINITION.
A subset
s
c 'lil.nis thin in
JR0if fo'l' evecy
£ > 0the'l'e
is a aountabte aove'l'i'Yifl of
s
with btoaks in
JR03the sum of whose
volumes is less than
E.2.2.2. DEFINITION.
Let
xbe an n-atmensionat submanifold.
and sbe a subset
of
x.
Then
s
is said to be thin in
Xif ther>e e:cists a aountab le
open aove'l'i'Yifl
u1 ,u2, •••of
s andahart-ma:ps
!fl1 ,!f12, ...so that
!fl1(u
1 n S)
is thin in
En,fo'l' atl i.
As a consequence of 2.2.2 one has the following: if
s
is thin inx,
then sdoes not contain an open subset of x, so its complement is dense in X• See 'lso 2.2.7.
2.2.3. LEMMA.
Let
m < n and Y c Embe a submanifold. Then
f(Y)is thin in
R0,
for> aU
f E: c"" (Y, JRn).See [6], pa_ge 31.
2.2.4. DEFINITION.
Let
X c: JRnbe a k-dimensionat submanifold.
andY c 'lil.mbe a p-dimensional submanifold.. Let
f €c
1 (X,Y). (1) corank Df (x0) := m.i.n (dim
x,
dim Y) - rank Df (x0) ,for>
x0 E: X. (2) x0
is said to be a a'l'itiaal point of
fif
corank Df Cx0) >o.
Othel'IJise
x0is aatled a
Nguta'l'point of
f.The set of a'l'itiaal
points of
fis denoted by
C[f].(3
Yo
€ Yis said to be a ariticaZ vaZue off if
yo
€ f(C[f]).0ther'1Pi8e
y0
iB said
to be a roeguta:z. vaZue of
f.2.2.5. REMARK.
(1)
Foro the definition of
Df(x0) ..
Bee
2.1.9.(2)
Sinae ..
a8stated in
2.1.9 ..one may inte:r!proet
Df(x0)
aa a. Unearo
map
E.k -+-~.. a ari tiaaZ point
x0
is a point rvheroe
Dfhaa
not
fuU
roank.
{3}
FX'om the thirod parot of
2.2.4it fotZows that evexy pointy
€ Ynot being in
f(X}iB a roeguta:z. value of
f.2.2~6. THEOREM (Sa.rd).
Let
X and Ybe submanifolds
and f €c""
(X,Y).Then
the set of aritiaal values of
fis thin in
Y.Se' [6] .. page 34.
2,2.7. cOROLLARY (Brown).
Let
x andYbe submanifolds
and f " c""(x,Y).Then
th~
set of roeguta:z. values of
fis dense in
Y.se~ [6] .. page 36.
2.3. The
Whitney
c""
Topotogy
Let
A~
be the vector space of polynomials in n variables of degrees
k,which have their constant term equal to zero.
A~
is isomorphic to some Euclidean space.Given f € Ck(JR.n,E.), we define.the continuous map
as follows:
k
• j f(x) := (x,f(x), Dk f(x)) ,
where Dk
~
(x) is the polynomial of degrees
k given by the Taylor expansionI '
of f at
xi
: up to order k after the first term. Since E.n , x E. K Ak n is iso-morphic to some Euclidean space, i t has the metric of that space, denotedk
2.3.1. DEFINITION,
Let
fe
C00(lRn,JR)and
o
e
c0(JRn,JR+),ltJhere
:R+is the.
set ofpositive
rea~s.Then
It can be shown that these sets, given n and k, form a base for a topology
co n k
on c (JR ,:R), called the Whitney c top<)logy, or shortly
wk.
So0
e ~ i fand only if for every f e
0
there is a o e c0(JRn,JR+) such that~~(f~o}
c0.
Then w~ cwk
fort
~k •
..
• co co n2.3.2. DEFINITION. w := U wk
is the Whz.tney
c
topology on
c
(JR ,:R). k=O,
0
""
n ""Hence a subset c C (JR ,JR) is open in the Whitney c topology, or shortly c""-open, if i t is open in the Whitney ck-topology, or shortly
~-open,
for some k <: 0.Within the same terminology: F c C00(lRn,JR) is C00-dense, if and only if i t is ck-dense for each k <: 0.
2.3.3. LEMMA.
Let
{fm}meJNbe a sequence of functions in
C00(lRn,:R).Then
fm + fin the Whitney
cktopol-ogy~ if
andon~y if there is, a compact
subset
K c lRn andan
m0 e N
suah that
f (xl = f (xlfor all.
x (. K,k k m
m <: m~ and j fm + j f
unifoi'ITily on
K.See [6), page 43.
2.3.4. THEOREM. C00(lRn,JR)
is a Bcdre apace in the Whitney
c""topology.
(So, the intei'Bection of a countab
~e co~ ~ctionof
c ..
-open
and-dense
subsets of
c'"'
(lRn ,JR)is
c""
-dense.)
See [6), page 44,
2.3.5. THEOREM.
Given
n,m e: E.the set
c""(lRn,:R)mis a Bcdre apace in
theproduct
topology~induced by the Whitney
c"" topo~yon the factors
.., nc
(JR ,:R) • See [6], page 47.2. 4. T
l!.aYl4:veJ!..4a.U;tyI I
I
2.4.1.
DEF~NITION.
Let
X, Y and Zbe submanifolde
and f~ c
1(X,Y).Let
z
c
Y and xa point in
x.
Then
fis said to
inte~sectz
t~aneve~s~l~at
x
(denoted by
fm
z
at x) if
eithe~ (1) f(x)I
z
o~As an interpretation of this definition,2.4.1 we give the following. Since Z c Y, the tangent space to Z at z ~ Z is a linear subspace of the tangent space to Y at z. Now, given x €
x,
we have:f
m
z at~is equivalent to: If
f(x) € z~then
f~ eve~dy
€ Tf(x)Y the~ea:rte
oz
~ Tf(x)z and cSx € Tx Xsuch that
!
2.4.2.
J
1c5y
=
dz + Df(x)c5x •,INITION.
Let
x~ Y andz be subma.nifolds
andz
c Y.L(Jt
f €
c
1(X,Y), and Ba subset of
z.
Then
fis said to
inte~seat
z
t~anave~saZly
on
B iff~every
x € X eithe~(1) f(X)
I
B 0~(2) f(x) € B and Tf(x)y = Tf(x)z + Df(x.) (TXX) •
2.4.3. DEFINITION.
Let
x ..
Y and Zbe submanifolde
and Z c Y.Let
f € c1 (X, Y) •
Then
fis said to intersect
ztrunsve~saZZy
(denoted
by
f AtzJ
if
fm z
at
eve~ypoint
x €x.
2.4.4. TBEOREM.
Let
x ..
Y andz
be submanifolde ..
whe~ez
c Y.Let
f € C01(X,Y) and f
m z.
Then
f+(Z) cx
is a submanifold
anddub f+(Z)
=
dim X- dim Y +dim Z •! • + .. _
FurothemoN .. g1-ven
x .:: f {Z)tr"" tangent epaae to
f+(Z)consists of
al~ cSx € T Xsatisfying
Of (x) cSx € Tf (x) Z.Theorem 2.4.4 provides a powerful tool in order to construct submanifolds. We shall use it frequently in the sequel.
Now we give, in a very specialized setting, a theorem concerning transver-sality of a parametrized family of maps.
(.t+1)m .. Let R, ~ 1, m ~ 1 be integers and Ban open subset of E • Let
be a smooth map, and W a sublilanifold of E2.tm+m. We define for each b E B
the smooth map
by:
In this context we have:
2.4.5. THEOREM. If ~
.n
w the set {b E BI
~b.n
w} is dense in B.See [6], page 53.
Before ending this introduction to some topics of global analysis we give a theorem in which second derivatives are involved.
2. 4. 6. DEFINITION. Let f E c"" (:IRn, E) • Then f is said to be a Morse function
if Df(x) = 0 implies D2f(x) is nonsinguZar, for aU x E En.
2.4.7. THEOREM (Morse). The set of Morse functions is open and dense in
C00 (:IRn ,E). See [6], page 63.
2.5.
Ch~ete4lzationo6
local Pa4eto optima
This section is devoted to a well-known theorem on local Pareto optima. The proof is essentially the same as in the author's paper [5], where a slight-ly generalized form has been presented.
00 R, . R, id
Let R, ~ 2, m ~ 2 and ui E C (JR ,E), i = 1, ••• ,m. Given w. E:R we 'Cons er
the set Aw consisting of the points x E E.tm satisfying
E{=1
xi = w. So Aw is the set of admissible allocations in a pure exchange economy with total resources w.As pointed out in 1.31 a first order condition for a point z e ~ to be a local Pareto optimum is that there are nonnegative reals /... 1 and some
R.-1 J.
peS such that Dui(zi) = !.
1p 1 i = l1 ••• ,m. (See also Chapter 5.) We restrict ourselves here to the case that Dui(z
1) p 01 i = 1, ••• ,m.
R.m
2.5.1. OEE'INITION. Let z e :R 1 and Dui (zi) p 0, i = 1, ••• ,m.
'
I
I { •,tm m
N : = <v1 , ••• 1 v ) e :It
I
L
z
m
i=12.5.2. DE]!'INI'l'ION.
Let
ze
lR R.m , we
lR , R. z € Aw' p .e
S R.-1 , Dui (zi) • AiP1wiirh
Ai > 0, i = 1, •••,m.
Then
Hz: Nz -+ :Ris defined
as fo~lA::Ms: IThis map Hz acts as a generalized second derivative we use in order to estabiishlwhether a point z, satisfying the first order condition for a
I
local str~ct Pareto optimum, is optimal or not. (In Chapter 5 we extend the notions Nz and Hz to the case that Dui (ail
=
0 for some i.)We prove the following theorem, using properties of implicit functions:
2.5.3. TBEOREM. Let we JR1 and z e A
satisfy
Dui (zi) =!.
1
pJ~.
1
~
where
R.-1 w
p ~
s
and
/..i > 0for
a~l i.(1)
If
Bz(v) < 0for aU
ve
Nz' v p 0~then
zie a
loca~strict
Pa:t'eto
optimum.(2)
If
Bz (v) > 0for some
v € Nz,then
zis not a local strict
Pa:peto
optimum.Before giving the proof of 2.5.3 we need some properties of the first and second derivatives of implicit functions.
CONVENTIONS •
(2) Given • f E:
c
.. (lR , JR) , Jl. ~ E: JR , Jl. t.Je def't,ne •t
h en
2
f(~)
1 1 x (JI. -1) matrix
~
1 (JI. -1)ax" ax 2
2...Lm
- .t
ax ax
(JI.-1) x (JI.-1) matri:c
a
2 fx 1 matri:c --.:--[ by:
ax ax
2.5.4. LEMMA. Let f € C 00
(lRR..,JRl and!;; E: JRJI. such that
a~
(!;;)#-
o.
axThen there is a neighboU!'hood 0 of ~ := <~1, ••• ,~!1.-1) and a smooth
function g: 0 + lR satisfying: (1) g(~)
=
!;;JI. ; (2l f(x,g(x)) = f(t;;) for a7:l x € 0 • - af - - -1 - - -(3) Dg(x)= -
----r
(x,g(x)) Df(x,g(x)) for aU x € 0 •ax
- - -
a
2 - - Ta
2 f (4)n
2f(x) + _ _ f_ (x) Dg(x) + Dg(x)t-=-
(x) + axax!l.
ax ax
+ 32 £<x> ng<x>
Tng<i>
+~
<x>
n2g<x>
=o
ax1 ax1 ax!l. for aUi
€ 0 •PROOF. (1), (2) and (3) follow"directly from 2.1.4. Writing out (3) in components we get:
for i = 1, ••• , Jl.-1. Differentiating with respect to xj, j = 11 •••
~·JI.-1,
leads toa£
a
2+ g ""0
1
j i •ax ax ax
Now we come to the proof of 2. 5. 3.
Since p €
F~-l,
at least one of the coordinates of pis not equal to zero. We assume p > 0. Other cases can be treated in a similar way. Sinceau
1 ~--~ (z
1l = Ai p > 0, and according to the implicit function theorem there
a
xiare SIIIOOth functiOnS ip i 1 defined On a neighbourhood
0
i Ofz
1 SUCh that
(1) i = 1, ••• ,m ;
(2) Ui (X:I.,!pi (Xi)) "'Ui (z1) 1 i"' 1, ... ,m 1
(3) ui(xi)
~
ui(zi) is equivalent toX~~
ipi(Xi) for all xi with xi € oi Ii
=
1, ••• ,m •L~t 0~
:= oi X lit andO'
:=Oi
by X • • . X0'
m'
Let l E c""(0'
n A , w :R) be defined mt
<x> :•
l:
lj) i<xi>
i=l t -wSince Oi is an open neighbourhOod hood of
z~,
the setO'
nAw
is an We observe thati
it, (z) - w ~
- !-1
of zi in :R and ll. an open neighbour-open neighbourhood of z in Aw.
0 •
If ui(xi)
~
u1(zi) for all i, and x E0'
n Aw, thenx~ ~
ipi(xi) for all i and, consl!!quentlyso
m
w~ ~
l:
IPi(xi)
or l(x) $ 0 . i=lFor the ploof of part 1 of the theorem we assume Hz (v) < 0 for all v E: Nz,
v .;. 01 and claim that l has a local strict minimum 0 at z. If so, then z is
To prove our claim, we take oz
=
(oz1, ••• ,&zm) where oz1 is sufficiently m
,small and Ei=l oz1
=
0.£ (z + oz) - w t =
m
i!t c,i czi>
+Dq~i
czi>Eii +i
o2~~'i
czi> (Eii,&zi> +where and - R. +pi{ozi)} -w = + p (oz) - w .. R. lim p(oz) =
o
llozii+O ll&zll2,In order to derive this Taylor expansion of£ in z we have used the deri-vatives of implicit functions as given in 2.5.4, together with the fact that p•p
=
1., m m - - p
Since Ei=l ozi = 0, i t follows Ei=l ozi
=
0 and v € Nz' So the function;,.has the following properties:
(1) ..C (z) = 0 ;
m
(2) Df(z)oz
=
0 for alloz,
satisfyingL
ozi=
0; i=l(3) the second derivative o2! (z) is definite positive as a quadratic fo:rm on the set T A • z w
This proves our claim and the first part of the theorem. Turning to the second part, let v " Nz and Hz (v) > 0. Since
thete are ~reals a
1, ••• ,am such ,that
m and
L
aii=l
0 •
Let b := (b1, ••• ,bm)' where bi aip. We consider the curve x(t) in Aw
through z:
Since uli is smooth, Dui (zi) Taylor's theorem: where l
um
t -2 p. (t) =o ,
t+O ~ I since vi € 1 Dui(zi)J.. and p•pSo there ~s an E > 0 such that
I
~ip' ~i > 0, and p•p
=
1, we find using2 2
+
l
t D u. (zi) (v. ,vi) + p, (t)~ ~ ~
1. Hence it follows
for all t € (O,e), i = l, ••• ,m.
Obviously this implies that z is not a local strict Pareto optimum.
0
I
If all of lthe functions ui are strictly convex at zi' then each second derivative
o
2ui(zi) is definite negative on the kernel of the first deri-vative. In that case Hz(v) < 0 for all v # 0, v € Nz' so z € 9.If each ui is strictly concave at zi' then Hz(v) > 0 for all v # 0, and
z ' 9.
If some oi' the ui are convex, others concave, then convexity may dominate and Hz is definite negative, or not. See also the examples in 1.6.
2.6. The .t.u.bma.n.i.&o!d
r
In 1.2 first order necessary conditions for an equilibrium (x,p) are formu-lated, one of them being Dui{xi) = Aip]~=
1
• Moreover, in 1.4 we mentioned that disastrous allocations are not welcome in our model. So we are looking.
m
for allocations x, where DUi(xi)
=
Aip]i=l' and where at least on~ Ai is not equal to zero. This leads to the definition of the setr
c m ,con-R.m
sisting of those points v Em , for which there are reals A., not all
R.-1 • m . ~
zero, and some pES , satisfying vi= Aip]i=1• See 2.6.2.
2.6.1. DEFINITION.
T:tie set
r
is parametrized by m + R. - 1 parameters and hence the following is not unexpected.2. 6. 2. LEMMA.
r
is a submanifoZd of dimension
m + R.-1and for eaah point
x
=
(A1p, ••• ,AmP) Er
the tangent spaae
Txrtor
consistsof all
those vectors
(ox1, ••• ,oxm} E mR.m
for whiah there are some
opE pLand
reals
oA1 , ••• ,oAmsatisfYing
oxi = oAip +Aiop]~=l'
PROOF. The proof is based upon 2.1.7.
Let z
=
(z1, ••• ,z) be a point ofr.
At least one of the zi is not O, so weR. m R.
assume zm .;. 0. Then ~ ;. 0 on a neighbourhood U of z in mR.m. We consider the map~:
U
+B(m-l)(R.-1 ) given bywhere
i = 1, ••• ,m-1 •
Here, as in 2.5, the bar denotes that the last coordinate has been skipped. Obviously ~+(0)
=
r
oU.
our claim is that rank D1)J(x) (m- 1) (R.-1) for all x E U. Since R.m - (m + R.-1)
=
{m -1) (R. -1), application of 2.1. 7 settles the proof.R.m . Given ox
=
(ox1, ••• , oxm) E m we find ~ (x) (ox} from
R. 1 - 'l":':" R. 1 -
-r-'!jli (Xi + OXi t X m + OX ) m
=
(X mIll + OX ) (Xi + QX, ) - (Xi + OXi) (X + QX ) =where
So
and
;= $i(x< ,xm) + oxm" x< + xR. ox, - x: 6x - ox:
x
+ p< (ox) ,~ ~ m ~ ~ m ~ m ~
1
llo~~o
lloxll llpi(ox) II o •D$ , (x. ,x ) (ox. , ox ) : ~ ~ m ~ m
Dljl(x) (ox) (D$
1(x1,x )(oxm 1,ox ), ••• ,D$ m m-1cx m-1,x )(ox m m-1,ox) m
29
With respect to cartesian coordinates, D$(x) is represented by the follow-ing matrix, also denoted by D$ (x) :
Dljl(x)
/).
m0
0
/).
0
m
0
where {).i ils the (R.- 1) x R. matrix:
0 -x. 1
~
0
i 1, •.•
,m,
and
0
is the (R.- 1) x R. matrix whose entries are all zero.Since xR. :f 0 in U the matrix t:. has rank R. - 1 and consequently the
m m
(m-1)(R.-1) x mR. matrix Dljl(x) has rank (m-1)(R.-1) for all x E U. So r is a submanif:old of dimension m + R. - 1 •
Since Txr lis the kernel of Dljl(x), i t consists of all those ox satisfying
t:.m oxi = {).i oxm , i = 1, ••• ,m-1
p := R. p 0 0 0 -p 1 R.-1 -p
is an {!-1) x R. matrix, the kernel of which is spanned by p.
R.m m-1
Hence Txr consists of all those ox E JR for which there are reals
a.J.
1 ~ ~= such thatGiven ox E Txr there is a unique decomposition
Then
where
J.
On the other hand, given op E p and OA1 I • • • ,o>.ml the vector
ox= (6x11•••1oxm), with oxi
=
Ai op + 6;\PI satisfies ox E Txr' as iseasily seen.
D
The proof of 2. 6. 2 is based upon the assumption that at least one of the A i is nonzero.
I
m R.-1The set {(>.
1p, ••• ,A.mp) A E JR 1 pES } is not a submanifold for R. ~ 2,
m 2: 2. For, assume that there is a smooth map F: 1lm-+ P.n, n ::; R.m, defined on a neighbourhood 0 of 0 E P.R.m such that F(A.
1p, ••• ,AmP) = 0 for all p E
st-
1 and A in a neighbourhood of 0 E~
and rank DF(x} = n on0
(see2.1.7). Then, differentiating the coordinates Fj];=
1 with respect, to
Ai]~=l
we obtainR.-1
In x E r the vector p E
s
is, up to its sign, uniquely determined, where~ R.-1as ~ch p E S satisfies (A1p, ••• ,Amp) = 0 for (A1, ••• ,Am) = 0.
aFj t-1
Hence it follows that ---- (0) •p = 0 for all pES and all i,j. So
axi
rank DF(O) = 0 and consequently F is constant on a neighbourhood of 0, in contradiction with 2.6.2.
CHAPrER 3
THE SET
TThe equations
in the un~nowns (A,x,p) may have several: kinds of solution sets, depending
.I ., R. m
Ort the m-1tUple U E C (:m ,JR) • In general one expects that mR. equations in m + mR. + R. -, 1 unknowns have an m + R. :... 1 dimensional solution set, according to the implicit function theorem. In this chapter we define a subset T of
oo R. m
C (JR ,JR) having the property that the equations above establish an
m + R. - 1 dimensional set, for each u E T.
Section 3J1 contains the definition of T and a manageable second order criterionlfor tuples u to be an element ofT.
'
In Section 3.2 the number of unknowns is reduced in replacing the equation Dui(xi)
=
Aip by Dui(xi)=
(Dui(xi)•p)p. This leads to a redefinition ofT. Section 3,3 is devoted to the proof that our results generalize those of Smale.In Sectiot 3.4 we justify the use of utility functions instead of prefer-ence relations by proving that.two utility tuples, representing the same preferences, are either both in T or both not in T.
Section 3.5 contains the proof that. T is dense, and in Section 3.6 we discuss the question whether T is open or not.
!
3.1.
A
6~t de6~ono6
Tco R. m
Let R. ~ ~' m ~ 2 and u E C (lR ,JR) • Then u induces a smooth map - m mR. R. R.m
i gU: JR X ::R X :R -+ :R X :R
I as followf!~:
3.1.1. DEFINITION
I n o er wor : T th ds ~s · th e se o m-tup es u or w c gu as t f 1 f hi h h 0 £ ~ ,.,tm+l as a ...
regular value.
Given
(~>..~x.~p)
E: Em x Etm xE~,
the derivative ogu(>.,x,p) (o>.,ox,op)equals
tm
Rearrangement of coordinates in E x E gives us the following matrix re-presentation of ngu(>.,x,p):
oq
u p,,x,p) (o>.,ox,op) 2 D u1 (x1) 0 0•
0 oT 1 0 -p T 01 D2um(xm) 01 oT 1 00
is the ~ x II. matrix with all entries zero.01 is the II. x 1 matrix with all entries zero. I is the ~ x ~ identity matrix.
01 01 ->.1I
.
0~ o>.1 T -1. I -p m 0 2p o>.m opFrom the definition ofT it follows that u € T if and only if gu(l.,x,p) • 0
implies rank
oq
(>.,x,p) • tm + 1. u ~m2 0 0 T 01 01 -!. I D u 1<x1) -p
...
1 (a1, ••• ,a.m,a> 0 D um(xm) 2 01 -p T -A. I m oT 1 oT 1 0 0 2pSo, the {mi'. + 1) x (m +mR. + .9.) matrix, representing ~u (!.,x,p) has not full
cank if and only if there is some {a
1, ••• ,a.m,B) '~ 0 such that
m
..
0 for all i, p•ai .. 0 for all i,
l
A.1ai "' 2Sp • i=13 .1. 2. DEFINITION.
(S is
a compact
submanifoZd of dimension 1m+ 1-m-2.)1-1 1 1
Given p E
s
we denote byn :
lR + lR the orthogonal projection on the. .l p
orthoplement p of p.
, "" 1 m
3.1.3. ~INITION. Let u e
c
(lR ,lR) • ThenI I G . € C«> (JRi',m X $ t JRR,m X lRR,m X JRR,) u is defined as foUoos: ( m 2 m m )
Gufx,p,a) := ITP Dui(xi)]i=1 ,D ui(xi)ai]i=1 '1!1 (Dui(xi) •p)ai •
'
Using the map Gu we give the following characterization of the set T.
3.1.4. ~OREM. u € T i f and onl;y if Gu (x,p,a) rf: 0 for aZ"l
1m (x(p,a) E lR x S.
J!.-1 PROOF. Let Gu(x,p,u)
Do.
Since rrp Dui(xi)=
0 for all i, and p €s
there are reals A.1 such that Dui(x1) = A.ip, i = l, ••• ,m. So gu(A.,x,p) = 0. srnce A..= Du.(xi) •p, it follows that the nonzero tuple (a
1, ••• ,a ,0)
1 1 m
belongs to the kernel of the transposed of the matrix
nq
(A.,x,p). Henceu
u
t
T. On the other hand, we assume ut
T. Then there is a point~+ ~
(A,x,p) E g (0) where Dg (A.,x,p) is not of full rank. Hence there is a
u J!.mu
nonzero pair (a,8) € :R x :R satisfying:
It follows that
m
26p • p =
l
A.i ai • p=
0 •1=1
So 6
=
0 since p•p=
1 and consequently a ~ 0. Sincewe have
26p •
For;maps between Euclidean spaces the matrix representation provides a
0
1 straightforward tool in determination of the rank of the derivative,
where-as in general for maps between manifolds rank determination can be compli-cated (see definition 2.1.9). This was the reason why the map
gu
was intro-duced, making possible the straightforward proof of 3.1.4 •.t-1
Now we return to the convention p € S •
"" J!. m
3.1.5. DEFINITION. Let u E C (JR ,JR) • Then the smooth map
is
defined as foll~e:m g {A,X1p) := (Du1(x.) -A.1pJ.
1) •
u 1 1=
3.1.6. LEMMA.
co R. m