Morfismos, Vol 14, No 1, 2010

VOLUMEN 14NÚMERO 1

ENERO A JUNIO DE 2010ISSN: 1870-6525

MorfismosComunicaciones EstudiantilesDepartamento de Matematicas

Cinvestav

Editores Responsables

• Isidoro Gitler • Jesus Gonzalez

Consejo Editorial

• Luis Ernesto Carrera • Beatris Adriana Escobedo• Samuel Gitler • Onesimo Hernandez-Lerma

• Hector Jasso Fuentes • Daniel Lopez• Miguel Maldonado • Raul Quiroga Barranco• Enrique Ramırez de Arellano • Enrique Reyes

• Armando Sanchez • Martın Solis

Editores Asociados

• Ricardo Berlanga • Emilio Lluis Puebla• Isaıas Lopez • Guillermo Pastor

• Vıctor Perez Abreu • Carlos Prieto• Carlos Renterıa • Luis Verde

Secretarias Tecnicas

• Roxana Martınez • Laura Valencia

Morfismos, Volumen 14, Numero 1, enero a junio de 2010, revista semestral, editadapor el Centro de Investigacion y de Estudios Avanzados del Instituto PolitecnicoNacional (Cinvestav), a traves del Departamento de Matematicas. Avenida InstitutoPolitecnico Nacional 2508, Colonia San Pedro Zacatenco, C.P. 07360, Mexico, D.F.Reserva de Derechos al Uso Exclusivo No. 04-2008-100210441300-102. Licitud deTıtulo No. 14729. Licitud de Contenido No. 12302, ambos otorgados por la ComisionCalificadora de Publicaciones y Revistas Ilustradas de la Secretarıa de Gobernacion.

Morfismos puede ser consultada electronicamente en “Revista Morfismos” en la di-

reccion http://www.math.cinvestav.mx. Para mayores informes dirigirse al telefono

(55) 5747-3871. Toda correspondencia debe ir dirigida a la Sra. Laura Valencia, De-

partamento de Matematicas del Cinvestav, Apartado Postal 14-740, Mexico, D.F.

07000 o por correo electronico: [email protected].

VOLUMEN 14NÚMERO 1

ENERO A JUNIO DE 2010ISSN: 1870-6525

Informacion para Autores

El Consejo Editorial de Morfismos, Comunicaciones Estudiantiles del Departamento deMatematicas del CINVESTAV, convoca a estudiantes de licenciatura y posgrado a presentarartıculos para ser publicados en esta revista bajo los siguientes lineamientos:

• Todos los artıculos seran enviados a especialistas para su arbitraje. No obstante, losartıculos seran considerados solo como versiones preliminares y por tanto pueden serpublicados en otras revistas especializadas.

• Se debe anexar junto con el nombre del autor, su nivel academico y la instituciondonde estudia o labora.

• El artıculo debe empezar con un resumen en el cual se indique de manera breve yconcisa el resultado principal que se comunicara.

• Es recomendable que los artıculos presentados esten escritos en Latex y sean enviadosa traves de un medio electronico. Los autores interesados pueden obtener el for-mato LATEX2! utilizado por Morfismos en “Revista Morfismos” de la direccion webhttp://www.math.cinvestav.mx, o directamente en el Departamento de Matematicasdel CINVESTAV. La utilizacion de dicho formato ayudara en la pronta publicaciondel artıculo.

• Si el artıculo contiene ilustraciones o figuras, estas deberan ser presentadas de formaque se ajusten a la calidad de reproduccion de Morfismos.

• Los autores recibiran el archivo pdf de su artıculo publicado.

• Los artıculos deben ser dirigidos a la Sra. Laura Valencia, Departamento de Matemati-cas del Cinvestav, Apartado Postal 14 - 740, Mexico, D.F. 07000, o a la direccion decorreo electronico [email protected]

Author Information

Morfismos, the student journal of the Mathematics Department of the Cinvestav, invitesundergraduate and graduate students to submit manuscripts to be published under thefollowing guidelines:

• All manuscripts will be refereed by specialists. However, accepted papers will beconsidered to be “preliminary versions” in that authors may republish their papers inother journals, in the same or similar form.

• In addition to his/her a!liation, the author must state his/her academic status (stu-dent, professor,...).

• Each manuscript should begin with an abstract summarizing the main results.

• Morfismos encourages electronically submitted manuscripts prepared in Latex. Au-thors may retrieve the LATEX2! macros used for Morfismos through the web sitehttp://www.math.cinvestav.mx, at “Revista Morfismos”, or by direct request to theMathematics Department of Cinvestav. The use of these macros will help in theproduction process and also to minimize publishing costs.

• All illustrations must be of professional quality.

• Authors will receive the pdf file of their published paper.

• Manuscripts submitted for publication should be sent to Mrs. Laura Valencia, De-partamento de Matematicas del Cinvestav, Apartado Postal 14 - 740, Mexico, D.F.07000, or to the e-mail address: [email protected]

Lineamientos Editoriales

“Morfismos” es la revista semestral de los estudiantes del Departamento de Mate-maticas del CINVESTAV, que tiene entre sus principales objetivos el que los estu-diantes adquieran experiencia en la escritura de resultados matematicos.

La publicacion de trabajos no estara restringida a estudiantes del CINVESTAV;deseamos fomentar tambien la participacion de estudiantes en Mexico y en el extran-jero, ası como la contribucion por invitacion de investigadores.

Los reportes de investigacion matematica o resumenes de tesis de licenciatura,maestrıa o doctorado pueden ser publicados en Morfismos. Los artıculos que apare-ceran seran originales, ya sea en los resultados o en los metodos. Para juzgar esto,el Consejo Editorial designara revisores de reconocido prestigio y con experiencia enla comunicacion clara de ideas y conceptos matematicos.

Aunque Morfismos es una revista con arbitraje, los trabajos se conside-raran como versiones preliminares que luego podran aparecer publicadosen otras revistas especializadas.

Si tienes alguna sugerencia sobre la revista hazlo saber a los editores y con gustoestudiaremos la posibilidad de implementarla. Esperamos que esta publicacion pro-picie, como una primera experiencia, el desarrollo de un estilo correcto de escribirmatematicas.

Morfismos

Editorial Guidelines

“Morfismos” is the journal of the students of the Mathematics Department ofCINVESTAV. One of its main objectives is for students to acquire experience inwriting mathematics. Morfismos appears twice a year.

Publication of papers is not restricted to students of CINVESTAV; we want toencourage students in Mexico and abroad to submit papers. Mathematics researchreports or summaries of bachelor, master and Ph.D. theses will be considered forpublication, as well as invited contributed papers by researchers. Papers submittedshould be original, either in the results or in the methods. The Editors will assignas referees well–established mathematicians.

Even though Morfismos is a refereed journal, the papers will be con-sidered as preliminary versions which could later appear in other mathe-matical journals.

If you have any suggestions about the journal, let the Editors know and we willgladly study the possibility of implementing them. We expect this journal to foster, asa preliminary experience, the development of a correct style of writing mathematics.

Morfismos

Contenido

Introduction to the manifold calculus of Goodwillie-Weiss

Brian A. Munson . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

Impacto de la polıtica fiscal en un ambiente con inflacion estocastica: un mo-delo de control optimo

F. Ortiz-Arango, F. Venegas-Martınez, y C. E. Castillo-Ramırez . . . . . . . . 51

Morfismos, Vol. 14, No. 1, 2010, pp. 1–50

Introduction to the manifold calculus ofGoodwillie-Weiss !

Brian A. Munson

Abstract

We present an introduction to the manifold calculus of functors,due to Goodwillie and Weiss. Our perspective focuses on the rolethe derivatives of a functor F play in this theory, and the analogieswith ordinary calculus. We survey the construction of polynomialfunctors, the classification of homogeneous functors, and resultsregarding convergence of the Taylor tower. We sprinkle exam-ples throughout, and pay special attention to spaces of smoothembeddings.

2000 Mathematics Subject Classification: 57-02, 57R40, 57R42.Keywords and phrases: Manifold calculus, Taylor towers, analyticityand connectivity of derivatives of spaces of embeddings.

Contents

1 Introduction 21.1 Further reading . . . . . . . . . . . . . . . . . . . . . . . 41.2 Conventions . . . . . . . . . . . . . . . . . . . . . . . . . 51.3 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Derivatives 82.1 Comparison with classical calculus . . . . . . . . . . . . 82.2 Cubical diagrams and total homotopy fibers . . . . . . . 102.3 Criticism of analogies . . . . . . . . . . . . . . . . . . . 14

!Invited article.

1

2 Brian A. Munson

3 Polynomial Functors 153.1 Definitions and examples . . . . . . . . . . . . . . . . . . 153.2 Characterization of polynomials . . . . . . . . . . . . . . 183.3 Approximation by polynomials . . . . . . . . . . . . . . 203.4 The Taylor tower . . . . . . . . . . . . . . . . . . . . . . 22

4 Homogeneous Functors 224.1 Definitions and examples . . . . . . . . . . . . . . . . . . 224.2 Classification of homogeneous polynomials . . . . . . . . 24

5 Convergence and Analyticity 255.1 Convergence of the series . . . . . . . . . . . . . . . . . 265.2 Convergence to the functor . . . . . . . . . . . . . . . . 27

6 Convergence for Spaces of Embeddings 306.1 Connectivity of the derivatives of embeddings . . . . . . 316.2 Connectivity estimates for the linear and quadratic stages

for embeddings . . . . . . . . . . . . . . . . . . . . . . . 326.3 Some disjunction results for embeddings . . . . . . . . . 35

7 Appendix 417.1 Homotopy limits and colimits . . . . . . . . . . . . . . . 417.2 The functor Map(!,!) . . . . . . . . . . . . . . . . . . 447.3 The Blakers-Massey Theorem . . . . . . . . . . . . . . . 45

1 Introduction

We intend to explain some of the intuition behind one incarnation ofcalculus of functors, namely the “manifold calculus” due to Weiss andGoodwillie [17, 34]. Specifically, we will highlight some analogies be-tween the ordinary calculus of functions f : R " R and the manifoldcalculus of functors. The trouble with analogies is that they are notequivalences, and some may lead the reader to want to push them fur-ther. Some may indeed be pushed further than we are currently aware,and some may lead to direct contradictions and/or bad intuition. An-other risk is that it is considered bad manners to tell people how tocategorize various ideas: part of our mathematical culture seems to bethat we leave intuition for talks and personal communications and rigorand precision for our papers, and with good reason: we cannot antic-ipate the ways in which our work may be useful in the future, and so

Manifold calculus 3

it may be best to convey it in as concise and precise a way as possi-ble. We feel the relatively small risk of misleading the reader and thefaux pas of making permanent intuitive notions by publishing them isa small price to pay for the possibility that this may entice some readerto learn more about these ideas and try to use them. Finally, we wouldlike to emphasize that this is not meant to be a rigorous introductionto calculus of functors. We will frequently omit arguments which woulddistract us from our attempts to be lighthearted. We hope this workmakes digesting the actual details from the original sources easier fornewcomers.

The philosophy of calculus of functors is to take a functor F and re-place it by its Taylor series, and we will begin our discussion of ordinarycalculus there and work backwards. Associated to a smooth functionf : R ! R is its Taylor series at zero (we choose zero for convenience;any center will work just fine):

(1) f(0) + f !(0)x+ f !!(0)x2

2!+ · · ·+ f (n)(0)

xn

n!+ · · · .

There are two natural questions to ask about this power series: (1)does it converge, and if so, for what x?; and (2) if it converges, doesit converge to f? The Taylor series is computationally much easierto work with than the function. A typical application is to truncatethe series at degree k, thus obtaining the kth degree Taylor polynomialTkf of the function f . If one is lucky and f (k+1)(x) can be controlledto be small in some neighborhood of zero, then one can use Taylor’sinequality to estimate the remainder. Specifically, if |f (k+1)(x)| " M ina neighborhood of zero, then the remainder satisfies the inequality

Rk(x) = |f(x)# Tk(x)| " M|x|k+1

(k + 1)!.

Our first goal is to construct the analog of a Taylor series for a functorF which associates to each open set U in a smooth manifold M a topo-logical space (we will specify the categories and hypotheses on F soon).A simple example to keep in mind is the space of maps, U $! Map(U,X)for some space X. Our second goal will be to explore issues of conver-gence in the special case of spaces of embeddings; here the functor ofinterest is U $! Emb(U,N), the space of smooth embeddings of U in asmooth manifold N . Here are some natural questions that arise basedon the above:

4 Brian A. Munson

1. What is the definition of the derivative of a functor, and howshould we compute it?

2. What is the definition of a polynomial functor?

3. Can we write the Taylor polynomial of a functor as a polynomialwhose coe!cients are the derivatives?

4. What is a “good” approximation, and what should we mean byconvergence?

We will answer these questions and more. Despite our attempt atlightheartedness, there will be no avoiding certain constructions un-pleasant for the purposes of an introductory paper. The main culprithere is homotopy limits and colimits, and we will assume the reader ismore or less familiar with these. If the reader has not seen these beforeor has only a nodding acquaintance with them, let her not despair; wewill try to give some intuition about what role these objects play, thoughit may still remain largely indigestible. Nevertheless, we have done twothings: (1) Provided an appendix with the statements and attempts atexplanation of results we have used in proofs, and (2) We have tried togive alternate, hopefully simpler, constructions whenever possible, andfocused on special cases where intimate knowledge of homotopy limitsand colimits is not necessary. We also assume the reader is familiar withthe basics of di"erential topology, namely handlebody decompositionsof smooth manifolds, and the basics of transversality.

This paper is organized into two main parts. The first, Section 1.3to Section 4, is concerned with developing the notion of derivatives andpolynomials, and tells us how to build a Taylor series for a functor, andTheorem 4.2.1 even gives a reasonable description of its homogeneouspieces. The usefulness of the definitions developed in these sectionspay o" in the proof of Theorem 3.2.1, and this proof contains a usefulorganizational principle important to later arguments, namely inductionon the handle dimension. The second part is devoted to the questionof convergence. We make a few general remarks about convergence inSection 5, and then move on to the specific case of spaces of embeddingsin Section 6.

1.1 Further reading

This work is an introduction, not a sample of the state-of-the art, butit is right for the reader to ask whether there is any point to this en-

Manifold calculus 5

deavor, so what follows are some references to applications of manifoldcalculus. Any omissions are due to the ignorance of the author. Weisshimself wrote a more rigorous survey [33] with a di!erent perspectivethan this one. A survey with emphasis on spaces of embeddings andrelated spaces is [11]. Another survey with many ideas from di!erentialtopology which are useful for studying spaces of embeddings using man-ifold calculus is [6]. For a survey on homotopy calculus, which sharesmany of the same tools as manifold calculus, see [20]. As for applica-tions of manifold calculus to spaces of embeddings, there are severalrecent works, including [1], [2], [3], [5], [7], [13], [14], [15], [21], [23], [25],[26], [27], [29], [28], [31], and [32]. For applications to spaces of linkmaps and connections with generalizations of Milnor’s invariants, see[16], [22], and [24].

1.2 Conventions

We will not be too careful about the category of spaces in which wewill work. For some purposes, the category Top of compactly generatedspaces will be fine. For other purposes, such as spaces of maps, wework in the category of simplicial spaces (a k-simplex in Map(X,Y ) isa map "k ! Map(X,Y )). We will, by abuse, always let Top denotethe target category. We write k in place of {1, 2, . . . , k}. We let int(X)stand for the interior of a subset X of some topological space. A spaceX is k-connected if !i(X) vanishes for 0 " i " k for all choices ofbasepoint in X. Every space is (#2)-connected, and nonempty spacesare (#1)-connected. A map of spaces f : X ! Y is k-connected if it isan isomorphism on !i for 1 " i < k and a surjection when i = k for allpossible choice of basepoints. Its homotopy fibers are therefore (k# 1)-connected. Conversely, if for all choice of basepoints in Y , the homotopyfiber of f is (k#1)-conected, then f is k-connected. In particular, everymap is (#1)-connected.

The union of a smooth manifold Lm with boundary with a “j-handle” Hj = Dj $ Dm!j is obtained by choosing an embedding e :Sj!1 ! "L and forming the identification space L %f Hj by attachingHj to "L along "Dj $Dm!j & Hj . We refer to j as the dimension ofthe handle, and refer to Dj $ {0} as the core of the handle. All smoothcompact manifolds admit a “handle decomposition”, which is a descrip-tion of M as a union of handles of various dimensions together withattaching maps to tell us how to embed the boundary of one handle inthe boundary of another (see [19]). We define the handle dimension of

6 Brian A. Munson

M to be the smallest integer j such that M admits a handle decompo-sition with handles of dimension less than or equal to j. A handlebodydecomposition of a smooth manifold Mm is the analog of a cell struc-ture on M , with j-handle playing the role of j-cell. Note that if P p is asmooth compact submanifold of M , then the disk bundle of its normalbundle is a smooth compact codimension 0 submanifold of M of handledimension at most p.

A j-dimensional handle Hj is a manifold with corners. That is,!Hj = !Dj ! Dn!j " Dj ! !Dn!j , and this union happens along thecorner set !Dj ! !Dn!j . More generally we will eventually encounterwhat is called a smooth manifold triad. Roughly speaking, this is atriple (Q, !0Q, !1Q), which is a smooth manifold Q of dimension q whoseboundary is decomposed as !0Q"!1Q and whose corner set is !0Q#!1Q.Boundary points have neighborhoods which look locally like [0,$) !Rq!1, and points in the corner set have neighborhoods which look locallylike [0,$) ! [0,$) ! Rq!2. In particular, we regard a j-handle Hj

as a smooth manifold triad with !0Hj = !Dj ! Dn!j and !1Hj =Dj ! !Dn!j . We refer the reader to [17] for details.

1.3 Preliminaries

We need to discuss the axioms necessary to impose on our functors toobtain an interesting and computable theory.

Definition 1.3.1 Let M be a smooth closed manifold of dimension m.Define O(M) to be the category (poset) of open subsets of M . Itsobjects are open sets U % M , and morphisms U & V are the inclusionmaps U % V .

Manifold calculus studies contravariant functors F : O(M) & Topwhich satisfy two axioms. Before we state them, let us consider a fewexamples, all of which are basically some sort of space of maps.

Example 1.3.2 Let X be a space. The functor Map(', X) : O(M) &Top given by the assignment U (& Map(U,X) is a contravariant functor,since an inclusion U % V gives rise to a restriction map Map(V,X) &Map(U,X).

Example 1.3.3 Let N be a smooth manifold. The embedding functorEmb(', N) : O(M) & Top is given by U (& Emb(U,N). This is thespace of smooth maps f : U & N such that (1) f is one-to-one, and

Manifold calculus 7

(2) df : TU ! TN is a vector bundle monomorphism. A related ex-ample is the space of immersions Imm(", N) : O(M) ! Top, given byU #! Imm(U,N). This is the space of smooth maps f : U ! N whichsatisfy (2). We think of an immersion as a local embedding.

The axioms we impose on our functors amount to something likecontinuity. The first tries to say that our functors should take equiva-lences to equivalences. At first glance, a category of open subsets of asmooth manifold should have di!eomorphism be the notion of equiva-lence. Of course, an inclusion map will never be a di!eomorphism, sowe ask for the next best thing. Let U, V $ O(M) with U % V . Theinclusion map i : U ! V is called an isotopy equivalence if there is anembedding e : V ! U such that the compositions i & e and e & i areisotopic to the identities of V and U respectively.

Definition 1.3.4 A contravariant functor F : O(M) ! Top is good if

1. it takes isotopy equivalences to homotopy equivalences, and

2. for any sequence of open sets U0 % U1 % · · · % Ui % · · · , thecanonical map F (

!i Ui) ! holimi F (Ui) is a homotopy equiva-

lence.

Another informal expression of the first axiom is that F behaves wellon thickenings. The reader may safely ignore the homotopy limit in thesecond axiom in favor of this explanation: the functor F is determinedby its values on open sets U which are the interior of smooth compactcodimension 0 submanifolds of M . Indeed, for any open set U , one canselect an increasing sequence V0 % V1 % · · · % Vi % · · · % U such that!

i Vi = U , and each Vi is the interior of a smooth compact codimension0 submanifold of M . This is a sensible thing to impose in light of ourmain example of interest, Emb(", N) : O(M) ! Top. After all, weare only interested in the values of Emb(U,N) when U is the interiorof some smooth compact manifold. It is also necessary for many of ourarguments to assume that U is of this form.

The structure of the category O(M) is much richer than the usualtopology on the real line R, so analogies between functions f : R ! Rand functors F : O(M) ! Top may seem a little weak. Still, there area few things to say that may be helpful. First, in light of the secondaxiom above, we could consider the full subcategory of all open sets Uwhich are the interiors of smooth compact codimension 0 submanifolds

8 Brian A. Munson

of M , which we call OMan(M). We like to think of OMan(M) ! O(M)as the analog of the dense subset Q ! R (after all, every continuousfunction is determined by its values on a dense subset). For more onthis, see Theorem 3.2.1. We will work almost exclusively in the cate-gory OMan(M), and so we will make a few remarks about its structure.The objects U " OMan(M) can be coarsely categorized based on theirhandle dimension. This should be thought of as a more refined notionof dimension of a manifold, and it plays a more important role in thistheory than does the ordinary dimension. In particular we will oftenrefer to the handle dimension of an open set U , which means the handledimension of the compact codimension 0 submanifold whose interior isU . Another important subcategory is the full subcategory of open sub-sets di!eomorphic with at most k open balls. This is the subcategoryof OMan(M) consisting of those sets U of handle dimension 0.

Definition 1.3.5 Let k # 0. The objects of the full subcategoryOk(M) ! O(M) are those open sets U which are di!eomorphic withthe union of at most k disjoint open balls in M .

We will return to the categories Ok(M) later, and their importancewill become clear once we define the notion of a polynomial functor.

2 Derivatives

2.1 Comparison with classical calculus

In order to build the Taylor series of a function f , we must discussderivatives. For a smooth function f : R $ R, its derivative at 0 isdefined by

f !(0) = limh"0

f(h)% f(0)

h.

For our analogy, we will ignore the denominator of the di!erence quo-tient in favor of the di!erence f(h)%f(0). We must decide three things:what plays the role of 0, what plays the role of h, and what plays therole of the di!erence f(h) % f(0). As for 0 and h, their analogs are,respectively, the empty set &, and the simplest non-empty open set: aset B which is di!eomorphic with an open ball. It is simplest in thesense that it has a handle structure with a single 0-handle.

As for the di!erence f(h) % f(0), since & ! B, for a functor F wehave a map F (B) $ F (&). There are a few ways of computing the

Manifold calculus 9

di!erence between two spaces with a map between them. The rightthing to do is to compute the homotopy fiber.

Definition 2.1.1 We define the derivative of F at ! to be

F !(!) = hofiber(F (B) " F (!)).

One reason this is natural is because the homotopy fiber, via thelong exact sequence in homotopy groups, describes the di!erence be-tween two spaces in homotopy. If M is connected, then our first axiom(together with a trick allowing us to relate two disjoint open balls inthe same path component) implies that the homotopy type of F !(!) isindependent of the choice of B.

Example 2.1.2 Let F (U) = Map(U,X). Let B be an open ball inM . Then F !(!) = hofiber(Map(B,X) " Map(!, X)) # X, sinceMap(!, X) = $ and Map(B,X) # X.

Example 2.1.3 Consider the functor E(U) = Emb(U,N) and let Bbe an open ball in M . We have E!(!) = hofiber(E(B) " E(!)). Anembedding of B is determined by its derivative at a point in B by theinverse function theorem, and so E(B), and hence E!(!), is equivalentto the space of injective linear maps Rm " Rn.

This process can be iterated, just as in ordinary calculus. Choose abasepoint in F (M), which endows F (U) with a basepoint for all U %O(M) via the map F (M) " F (U). For our purposes it is more usefulto have formulas for the higher derivatives only in terms of the functorF , not its derivatives. Consider the following non-standard formula forthe second derivative of f : R " R at 0:

f !!(0) = limh1,h2"0

f(h1 + h2)& f(h1)& f(h2) + f(0)

h1h2.

Once again for an analogy, we will throw away the denominator andfocus on the iterated di!erence f(h1 + h2) & f(h1) & f(h2) + f(0) =f(h1 + h2) & f(h1) & (f(h2) & f(0)). Now all we need is an analog of+, for which we will use disjoint union, so h1+h2 becomes B1

!B2 for

two disjoint open balls B1, B2 ' M . Then we iterate homotopy fibersand define

(2) F !!(!) = hofiber"hofiber(F (B1

#B2) " F (B1))

&" hofiber(F (B2) " F (!))$.

10 Brian A. Munson

This iterated homotopy fiber is, by definition, the “total homotopyfiber” of the following square diagram:

F (B1!

B2) !!

""

F (B1)

""F (B2) !! F (!)

The kth derivative of F at ! is the total homotopy fiber of a k-dimen-sional cubical diagram involving k disjoint open balls. In order to makethis precise, we require a brief discussion of cubical diagrams. They areubiquitous in calculus of functors, and we will use them frequently.

2.2 Cubical diagrams and total homotopy fibers

Details about cubical diagrams can be found in [12, Section 1]. Otheraspects important to this work not appearing in this section have beenplaced in the appendix to cause minimal distraction. For a finite setT , let |T | be its cardinality and P(T ) denote the poset of non-emptysubsets of T . For instance, if T = 1 = {1}, this poset looks like ! " {1},and if T = 2, then we can diagram this poset as a square

! !!

""

{1}

""{2} !! {1, 2}

Here we have only indicated those morphisms which are non-identitymorphisms and minimal in the sense that they cannot be written as acomposition of multiple non-identity morphisms. A 0-cube is a space,a 1-cube is a map of spaces and a 2-cube is a square diagram. Ingeneral, the 2|T | subsets can be arranged to form a |T |-dimensionalcube whose edges are the inclusion maps as above. Experience suggestsunderstanding statements for k-cubes in the cases k = 2, 3 is usuallyenough. We will focus almost exclusively on square diagrams.

Definition 2.2.1 Let T be a finite set. A |T |-cube of spaces is a co-variant functor

X : P(T ) #" Top .

We may also speak of a cube of based spaces; in this case, the target isTop!, the category of based spaces.

Manifold calculus 11

We can view a |T |-cube X as a map (i.e. a natural transformationof functors) of (|T | ! 1)-cubes Y " Z as follows. Fix t # T . DefineY : P(T !{t})" Top! by Y(S) = X (S). Define Z : P(T !{t})" Top!by Z(S) = X (S $ {t}). There is clearly a natural transformation offunctors Y " Z, and we may write X = (Y " Z).

Definition 2.2.2 The total homotopy fiber, or total fiber, of a |T |-cubeX of based spaces is the space tfiber(X ) given by the following iterativedefinition. For a 1-cube X" " X1, the total homotopy fiber is definedto be the homotopy fiber of the map X" " X1. For a k-cube X ,write it as a map of (k ! 1)-cubes Y " Z, and define tfiber(X ) =hofiber(tfiber(Y)" tfiber(Z)).

This is well defined because the homotopy type of tfiber(X ) is inde-pendent of the choice of Y and Z above by [12, Proposition 1.2a]. Thiscan be shown to be equivalent to the following definition, which is moreconcise, obviously well defined, but requires knowledge of homotopylimits.

Proposition 2.2.3 ([12, 1.1b]) For a |T |-cube X of based spaces, thetfiber(X ) is the homotopy fiber of the map

a(X ) : X (%) !" holimS #="

X (S).

The reader is encouraged to prove this in the case of a square dia-gram.

Definition 2.2.4 Let X be as above. If a(X ) is k-connected, we saythe cube is k-cartesian. In case k = &, (that is, if the map is a weakequivalence), we say the cube X is homotopy cartesian.

For a space (0-cube) X, the convention is that k-cartesian means(k ! 1)-connected, and for a map (1-cube) X " Y to be k-cartesianmeans it is k-connected, so its homotopy fibers are (k ! 1)-connected.A square

X" !!

""

X1

""X2

!! X12

is homotopy cartesian if the map X" " holim(X1 " X12 ' X2) is ahomotopy equivalence. Such a square is often referred to as a homotopy

12 Brian A. Munson

pullback square because holim(X1 ! X12 " X2) is the space of all(x1, !, x2) such that xi # Xi for i = 1, 2 and ! is a path in X12 betweenthe images of x1 and x2. In contrast, the pullback of X1 ! X12 " X2

is the space of all (x1, x2) # X1 $ X2 such that the images of the xiin X12 are equal. There is a useful relationship between pullbacks andhomotopy pullbacks. If

X! !!

""

X1

""X2

!! X12

is a pullback square, then it is a homotopy pullback if either X1 ! X12

or X2 ! X12 is a fibration. That is, in this case the map from thepullback to the homotopy pullback is an equivalence. A similar criterioncan be formulated for general cubes, though it is more complicated.A useful and familiar example of a homotopy pullback is obtained bysetting X2 to a point and letting X1 ! X12 be a fibration whose fiberover the image of X2 in X12 is X!.

Viewing a (|T |+ 1)-cube Z as a map of |T |-cubes X ! Y as in ouriterative definition of total homotopy fiber, choose a basepoint y # Y(%),which bases each Y(S), and define a |T |-cube Fy(S) = hofiber(X (S)!Y(S)).

Proposition 2.2.5 ([12, 1.18]) With X ,Y,Z as above, the (|T | + 1)-cube X is k-cartesian if and only if for each choice of basepoint y # Y(%),the |T |-cube S &! Fy(S) is k-cartesian.

For |T | = 1, this says that a map of spaces X ! Y is k-connectedif and only if all of its homotopy fibers are k-cartesian, which means(k ' 1)-connected in the case of a 0-cube. We present one final factwhich will be useful in the proof of Theorem 6.2.1.

Proposition 2.2.6 ([12, 1.22]) Let X ,Y be |T |-cubes, and suppose wehave a map X ! Y such that for all S (= %, X (S) ! Y(S) is k-connected. Then the map holimS "=!X (S)! holimS "=! Y(S) is (k' |T |+1)-connected.

Returning to our discussion of derivatives, we can now make a sen-sible definition of the derivatives of F at %.

Definition 2.2.7 Let B1, . . . , Bk be pairwise disjoint open balls in M .Define a k-cube of spaces by the rule S &! F (

!i/#S Bi). Define the


kth derivative of F at the empty set, denoted F (k)(!), to be the totalhomotopy fiber of the k-cube S "# F (

!i/!S Bi).

Example 2.2.8 We can compute the derivatives of the functor F (U) =Map(U,X). We have already seen that F "(!) $ X. Let B1, B2 bedisjoint open balls. F ""(!) is the total homotopy fiber of the square

F (B1"

B2) !!

""

F (B1)

""F (B2) !! F (!)

Since each Bi is homotopy equivalent to a point %i, and Map(&, X) pre-serves homotopy equivalences, this is equivalent to the total homotopyfiber of the square

F (%1"

%2) !!

""

F (%1)

""F (%2) !! F (!)

Clearly F (%1"

%2) = X 'X, and by our calculation above, we see thatF ""(!) is the total homotopy fiber of the square

X 'X !!

""

X

""X !! %

Here the vertical map X 'X # X is projection onto the second coor-dinate, and the horizontal map is projection onto the first coordinate.Using our iterative definition of homotopy fiber (and taking fibers verti-

cally), we see that F ""(!) = hofiber(Xid# X) $ %. Alternately, we could

observe that this square is both a pullback and a homotopy pullback.A similar computation shows that F (k)(!) $ % for k ( 3. That is, allderivatives but the first of F (U) = Map(U,X) vanish, which suggeststhis should be a linear functor. It is, as we will see in the next section.

Example 2.2.9 Let us compute the first two derivatives of F (U) =Map(U2, X). We have F "(!) = hofiber(Map(B2, X) # Map(!, X)) $

14 Brian A. Munson

hofiber(X ! ") # X. F !!($) is the total homotopy fiber of the square

Map((B1!

B2)2, X) !!

""

Map(B21 , X)

""Map(B2

2 , X) !! Map($, X)

.

Since Map((B1!

B1)2, X) = Map(B21 , X) % Map(B2

2 , X) % Map(B1 %B2, X) % Map(B2 % B1, X), we have F !!($) # Map(B1 % B2, X) %Map(B2 % B1, X) # X2. All of the higher derivatives are contractible.In a similar fashion, one can compute the first k derivatives of F (U) =Map(Uk, X); all derivatives of order greater than k are contractible.

2.3 Criticism of analogies

We justify our definition of derivatives based on the classification theo-rem for homogeneous functors which appears below as Theorem 4.2.1,in which the derivatives at the empty set play a central role. Despite theimportance of the derivatives as we have defined them, we have reasonto be interested in the derivative of F at an arbitrary open set. We evenhave reason to be interested in something which formally resembles aderivative (the homotopy fiber of a restriction map) as described above,but which does not simply involve studying di!erences based on tak-ing disjoint unions with open balls. It is natural to make the followingdefinition.

Definition 2.3.1 For an open set V & O(M) and an open ball B dis-joint from V , define

F !(V ) = hofiber(F (V"

B) ! F (V )).

Although the disjoint union is an obvious candidate for the analogof sum, it is not at all clear that we should ignore more general unions,for example, the attaching of a handle. In fact, we should not. Aswe have mentioned, it is enough for us to understand the values of afunctor on open sets V which are the interior of a compact codimension0 submanifold L of M ; that is, when V & OMan(M). For the purposes ofthis informal discussion, we will replace V with L. We wish to considermore generally hofiber(F (L 'f H i) ! F (L)). The special case of i = 0is the disjoint union of L with an m-dimensional disk. Similar criticismsapply to the study of higher derivatives. More general di!erences will


become important when we tackle the question of convergence and theanalog of a bound on f (k+1)(x) for x close to zero, which is importantin understanding the remainder Rk(x) = |f(x)! Tkf(x)|.

3 Polynomial Functors

3.1 Definitions and examples

A function f : R " R is linear if f(x + y) = f(x) + f(y) for all x, y.More generally, we might say a function is linear if f(x + y) ! f(x) !f(y) + f(0) = 0. Making analogies as we did in Section 2, and beingmore flexible about the analog of sum (and using an arbitrary unionin place of the disjoint union), this leads one to say that a functorF : O(M) " Top is linear if for all open V,W in M the total homotopyfiber of

F (V #W ) !!

""

F (W )

""F (V ) !! F (V $W )

is contractible. This implies that the second (and higher) derivativesof F vanish by letting V and W be disjoint open balls, but linearity isclearly a stronger condition. Linear functors are also called polynomialof degree % 1, or excisive. We pause for an example before formalizingthis definition.

Example 3.1.1 Let X be a space. The functor U &" Map(U,X) islinear. This follows from the fact that Map(!, X) sends (homotopy)pushout squares to (homotopy) pullback squares. See Proposition 7.2.2.

We can reformulate this in a way more suitable to our needs, andalthough it may seem a bit strange at first, the proof of Theorem 3.2.1should help the reader understand why the definition is presented thisway.

Definition 3.1.2 A functor F : O(M) " Top is polynomial of degree% 1 if for all U ' O(M) and for all disjoint nonempty closed subsetsA0, A1 ( U , the diagram

F (U) !!

""

F (U !A0)

""F (U !A1) !! F (U ! (A0 #A1))

16 Brian A. Munson

is homotopy cartesian.

To relate this back to the definition above, note that if we put W =U !A0 and V = U !A1, then U = V "W , and U ! (A0"A1) = V #W .The reason for this is that it is convenient for the purposes of inductivearguments (we will see this first in the proof of Theorem 3.2.1) to thinkabout “punching holes” in an open set to reduce its handle dimension.The definition of polynomial of higher degree generalizes the notion oflinearity.

Definition 3.1.3 A functor F : O(M) $ Top is called polynomial ofdegree % k if for all V & O(M) and for all pairwise disjoint nonemptyclosed subsets A0, A1, . . . , Ak+1 ' V , the map F (V ) $ holimS !=" F (V !!

i#S Ai) is a homotopy equivalence; in the same way, the diagram S ($F (V !

!i#S Ai) is homotopy cartesian.

To compare this with the definition of the kth derivative, let V bethe union of k+1 disjoint open balls and let the Ai be the componentsof V . Thus a polynomial of degree % k has contractible derivatives oforder k + 1 and above.

Proposition 3.1.4 If F is polynomial of degree % k, then it is polyno-mial of degree % k + 1.

This is certainly something that had better be true if this definitionis to make any sense. It is not completely trivial, but follows from thefact that if two opposing (k+1)-dimensional faces of a (k+2)-cube arehomotopy cartesian, then that (k+2)-cube is itself homotopy cartesian.Now let us consider several more examples.

Example 3.1.5 The functor U ($ Map(Uk, X) is polynomial of degree% k (but not polynomial of lower degree). This basically follows fromthe pigeonhole principle. Let A0, . . . , Ak be pairwise disjoint nonemptyclosed subsets of U . For a point (x1, . . . , xk) & Uk, each xi is in at mostone Aj , hence there is some l such that xi & U!Al for all i by the pigeon-

hole principle. Therefore Uk =!k

i=1(U ! Ai)k. It follows immediatelythat, Uk = colimS !="(U!

!i#S Ai)k, and one can show that in fact Uk )

hocolimS !="(U!!

i#S Ai)k. Since Map(!, X) preserves equivalences andturns homotopy colimits into homotopy limits (Proposition 7.2.2), wehave an equivalence Map(Uk, X) ) holimS !="Map((U !

!i#S Ai)k, X).


Example 3.1.6 We can generalize the previous example without doingany extra work as follows. Let C(k, U) ! Uk be the configuration spaceof k points in U (those (x1, . . . , xk) " Uk such that xi #= xj for i #= j).The group !k acts on C(k, U) by permuting the coordinates, and we let!Uk

"= C(k, U)/!k denote the quotient by this action. This gives us the

space of unordered configurations of k points in U . The same argumentas in the previous example shows that both U $% Map(C(k, U), X) andU $% Map(

!Uk

", X) are polynomial of degree & k.

Example 3.1.7 The functor U $% Emb(U,N) is not polynomial of de-gree & k for any k. We will indicate why for k = 1. Let A0, A1 ! Ube pairwise disjoint closed subsets, and put Ui = U ' Ai, and U12 =U1 ( U2. We are asked to check whether the map Emb(U1 ) U2, N) %holim(Emb(U1, N) % Emb(U12, N) * Emb(U2, N)) is an equivalence.That is, given fi " Emb(Ui, N) with a homotopy between their restric-tions to U12, is this enough to determine an element of Emb(U1)U1, N)?It is not, due to an obstruction, namely that f1(U1) and f2(U2) mightintersect in N . It is, however, true that the map Emb(U1 ) U2, N) %holim(Emb(U1, N) % Emb(U12, N) * Emb(U2, N)) has a certain con-nectivity; see Section 6.3 and Theorem 6.3.5

Example 3.1.8 The functor U $% Imm(U,N) is polynomial of degree& 1. Let A0, A1 ! U be pairwise disjoint closed, and put Ui = U ' Ai,and U01 = U0 ( U1. Then the square

Imm(U0 ) U1, N)

!!

"" Imm(U0, N)

!!Imm(U1, N) "" Imm(U01, N)

is clearly a pullback, since being an immersion is a local condition,and immersions of U0 and U1 which agree on their intersection makean immersion of the union. It is a homotopy pullback because therestriction map Imm(U0, N)% Imm(U01, N) is a fibration. This fact is areformulation of the Smale-Hirsch theorem. This isn’t quite technicallycorrect; the Smale-Hirsch theorem does not apply to the restriction mapof open sets. However, this can be overcome without too much di"culty.See [34, Lemma 1.5].

18 Brian A. Munson

3.2 Characterization of polynomials

Theorem 3.2.4 below is a structure theorem for polynomials, and laterwe will discuss a structure theorem for homogeneous polynomials, Theo-rem 4.2.1. Theorem 3.2.1, a structure theorem for linear functors (poly-nomials of degree ! 1), which contains aspects of the proofs of bothTheorem 3.2.4 and Theorem 4.2.1, will be given below, and it has asimple parallel for ordinary linear functions f : R " R. The techniquesof its proof are used many times in this paper.

Consider the following proof that every continuous linear functionf : R " R is of the form f(x) = ax. Let a = f(1). Linearity impliesf(n) = an for n a natural number. If p and q are natural numbers withq #= 0, then ap = f(q p

q ) = qf(pq ) by linearity, and so f(pq ) = apq . By

density of Q in R and continuity of f , this implies f(x) = ax for all realnumbers x.

Let p : Z " M be a fibration, and let !(M,Z; p) be its spaceof sections. For example, if Z = M $ X and p is the projection,!(M,Z; p) = Map(M,X). The following theorem says that all lin-ear functors F such that F (%) = & are the space of sections of somefibration. Or, more roughly, that they are all (twisted) mapping spaces.

Theorem 3.2.1 Let F : O(M) " Top be a good functor such thatF (%) = & and which is polynomial of degree ! 1. Then there is afibration p : Z " M for some space Z and a natural transformationF (U) " !(U,Z; p) which is an equivalence for all U ' O(M).

Proof. First we make the natural transformation F (U) " !(U,Z; p).Let O(1)(V ) denote the category of open subsets of V which are di"eo-morphic to exactly one open ball. Note that all inclusions in this cat-egory are isotopy equivalences, and that the realization |O(1)(V )| ( V .Let Z = hocolimU!O(1)(M) F (U). Since F takes isotopy equivalences

to homotopy equivalences, Z quasifibers over |O(1)(M)| ( M withspace of sections equivalent to holimU!O(1)(M) F (U) by Theorem 7.1.6.There is a natural transformation F (V ) " holimU!O(1)(V ) F (U) since

F (V ) ( holimU!O(V ) F (U) by Theorem 7.1.5 and O(1)(M) " O(M) in-duces the map in question. We define !(V ) = holimU!O(1)(V ) F (U). Wenow must show F (V ) " !(V ) is an equivalence. To do so, it is enoughby the second part of Definition 1.3.4 to check that it is an equivalencewhen V is the interior of a compact codimension zero submanifold L ofM . We will proceed by induction on the handle dimension of V .


Let k be the handle dimension of L. The base case to consider isk = 0, when V is a disjoint union of finitely many open balls. For this,we will induct on the number of components l of V . The base case isl = 1, and in this case V is a final object in the category O(1)(V ), and sothe map F (V ) ! !(V ) is an equivalence by Theorem 7.1.5. For l > 1,let A0, A1 be two distinct components of V , and put VS = V "

!i!S Ai

for S # {0, 1}. Consider the following diagram

F (V ) !!

""

!(V )

""holimS "=# F (VS) !! holimS "=# !(VS)

Since both F and ! are polynomial of degree $ 1, the vertical mapsare equivalences, and by induction, each map F (VS) ! !(VS) is anequivalence, and hence the induced map of homotopy limits over S is anequivalence by Theorem 7.1.4. Therefore the top arrow is an equivalenceas well.

The general case proceeds in a similar fashion. Let k > 0 be the han-dle dimension of V , and let l denote the number of handles of dimensionk. Let e : Dk %Dm$k ! L be one of these k-handles. Let D0, D1 # Dk

be disjoint disks, and put A%i = Di % Dm$k. Then A0 = V & A%

0 andA1 = V & A%

1 are nonempty disjoint closed subsets of V , and if we putVS = V " 'i!SAi, then for S (= ), VS is the interior of a compact codi-mension zero submanifold LS which can be given a handle structure withfewer than l handles of dimension k (see Figure 1 in Subsection 6.2 fora picture in a slightly di"erent case). Once again consider the followingdiagram.

F (V ) !!

""

!(V )

""holimS "=# F (VS) !! holimS "=# !(VS)

The vertical arrows are equivalences because F and ! are polynomialof degree $ 1. For S (= ), the map F (VS) ! !(VS) is an equivalence byinduction on l, and hence so is the bottom horizontal arrow. It followsthat the top arrow is an equivalence as well. !

Remark 3.2.2 The idea of this proof is philosophically similar to thatwhich classifies continuous linear functions. We first constructed thedesired functor ! by averaging (taking a homotopy limit) the values

20 Brian A. Munson

of F on single open balls (akin to a = f(1); we took an average toensure functoriality), and we see from the proof that !, and hence F ,is completely determined by the value of F on an open ball. Thenwe showed using linearity with a handle induction argument that thisimplied that F (V ) ! !(V ) was an equivalence for V " OMan(M) (ouranalog of Q). Finally we used continuity to conclude the result forgeneral open sets V .

Remark 3.2.3 We have already seen that Imm(M,N) is polynomialof degree # 1. We may ask how to express it as a space of sections. Inthis case, an immersion f is a section of a bundle over M whose fiber atx " M is the space of vector bundle monomorphisms TxM ! Tf(x)N .This is, once again, a version of the Smale-Hirsch Theorem.

A proof similar to that in Theorem 3.2.1 characterizes polynomialsin terms of their values on finitely many open balls, and it also utilizesa similar handle induction argument.

Theorem 3.2.4 ([34, Theorem 5.1]) Suppose F1 ! F2 is a naturaltransformation of good functors and that Fi is a polynomial of degree# k for i = 1, 2. If F1(V ) ! F2(V ) is an equivalence for all V " Ok(M),then it is an equivalence for all V " O(M).

Note that a polynomial p : R ! R of degree k such that p(0) = 0 isdetermined by its values on k distinct points; similarly, our polynomialfunctors F are completely determined by their values on the categoryof disjoint unions of at most k open balls.

3.3 Approximation by polynomials

Now we will construct the kth Taylor polynomial TkF for a functor F .Proceeding with an ordinary Taylor polynomial in mind, we would liketo construct a functor TkF which has the following properties:

• The derivatives F (i)($) and (TkF )(i)($) agree for 0 # i # k.

• TkF is polynomial of degree # k.

• There is a natural transformation F ! TkF , so that we maydiscuss the “remainder” RkF = hofiber(F ! TkF ).


Looking back at our discussion of derivatives, we computed F (i)(!)by looking at the total homotopy fiber of a cubical diagram of the valuesof F on at most i disjoint open balls. One way to ensure that thederivatives of order at most k of F and TkF agree is to make the valuesof F (V ) and TkF (V ) agree when V is a disjoint union of at most k openballs. With this in mind, for V " O(M), recall the poset Ok(V ) of opensubsets of U which are di!eomorphic with at most k open balls in V . Itis a subposet of O(V ), and we want the values of F and TkF to agreeon these subcategories.

Definition 3.3.1 Let TkF (V ) = holimU!Ok(V ) F (U).

This is a (homotopy) Kan extension of F along the inclusion of thesubcategory Ok(V ) # O(V ). It says that the value of TkF at a givenopen set V is an “average” of the values of F on at most k open ballscontained in V . Note that if V itself is di!eomorphic with at mostk open balls, then V is a final object in Ok(V ), and so TkF (V ) =holimU!Ok(V ) F (U) $ F (V ), so we really have correctly prescribed thevalues of TkF the way we said we would.

It is not clear from Definition 3.3.1 that TkF is a polynomial ofdegree % k, but it turns out that this is so. The proof is not trivial.Let us content ourselves with knowledge that an ordinary polynomialof degree k such that p(0) = 0 is completely determined by its valueson at most k points, and it is clear from the definition of TkF as anextension over the subcategory of at most k “points” that the analog ofthis is true.

There is a natural transformation F # TkF given by observing thatthe inclusion Ok(V ) # O(V ) induces a map of homotopy limits

F (V ) $ holimU!O(V ) F (U) # holimU!Ok(V ) F (U) = TkF (V )

and noting that the first equivalence follows since V is a final object inO(V ) (see Theorem 7.1.5 in the appendix).

Note that O0(V ) contains only the empty set for all V , and soT0F (V ) = F (!) for all V .

Example 3.3.2 Since F (V ) = Map(V,X) is polynomial of degree % 1,F (V ) # T1F (V ) is an equivalence by Theorem 3.2.4, since their valuesagree when V is a single open ball.

22 Brian A. Munson

Example 3.3.3 The linearization of embeddings is immersions. Thatis, T1 Emb(V,N) ! Imm(V,N). The natural transformation

Emb(V,N) " Imm(V,N)

is an equivalence when V is a single open ball, and hence T1 Emb(U,N)= holimV !O1(U) Emb(V,N) is equivalent to holimV !O1(U) Imm(V,N) !Imm(U,N), with the last equivalence given by the fact that Imm(#, N)is polynomial of degree $ 1, as in the previous example.

3.4 The Taylor tower

Armed with a definition of TkF , we can now form the “Taylor tower”of F , the analog of the Taylor series. The inclusion Ok"1(V ) " Ok(V )induces a map TkF (V ) " Tk"1F (V ), and so we obtain a tower of func-tors

· · · " TkF " Tk"1F " · · · " T1F " T0F.

Since V is a final object in O(V ), we may identify F (V ) withholimO(V ) F , and the inclusion Ok(V ) " O(V ) induces maps F " TkFwhich are compatible with one another. Hence there is a natural trans-formation F " holimk TkF , and we would like to know under what cir-cumstances this map is an equivalence; that is, when the Taylor tower ofF converges to F . This is the subject of Section 5. Before we embark onquestions of convergence, it will be useful to understand the di!erenceshofiber(TkF " Tk"1F ).

4 Homogeneous Functors

4.1 Definitions and examples

An explicit description of polynomial functors is perhaps too much tohope for, so we will content ourselves with a classification of homoge-neous functors. Fortunately there is a parallel with ordinary calculushere too. For f : R " R, consider the kth homogeneous piece of itsTaylor series, Lkf(x) = Tkf(x)# Tk"1f(x) = f (k)(0)x

k

k! . The classifica-tion of homogeneous functors shares a similar form. Roughly speaking,it is the space of sections of a fibration over

!Mk

"whose fibers are the

derivatives F (k)(%). We will state this more precisely below, but firstwe define what it means for a functor to be homogeneous and considersome examples.


Definition 4.1.1 A functor E : O(M) ! Top is homogeneous of de-gree k if it is polynomial of degree " k and Tk!1E(V ) # $ for all V .

Example 4.1.2 For a good functor F , choose a basepoint in F (M).This bases F (V ) for all V % O(M). The functor LkF = hofiber(TkF !Tk!1F ) is homogeneous of degree k. That it is polynomial of degree" k follows from the fact that TkF and Tk!1F are both polynomial ofdegree " k. To see that Tk!1LkF (V ) # $ for all V , first observe thatTk!1 commutes with homotopy fibers (see Theorem 7.1.3; homotopylimits commute), and next observe that Tk!1TkF # Tk!1F . Indeed,Tk!1TkF (V ) = holimW"Ok!1(V ) holimU"Ok(W ) F (U), and sinceW is dif-feomorphic with at most k& 1 open balls, it is a final object in Ok(W ),and so holimU"Ok(W ) F (U) # F (W ).

Example 4.1.3 The functor U '! Map(U2, X) is polynomial of degree" 2, so its quadratic approximation T2Map(U2, X) # Map(U2, X).However, it is not homogeneous of degree 2, because, as we showedabove, it has a non-trivial first derivative, which would necessarily van-ish were it homogeneous. In fact, T1Map(U2, X) # Map(U,X). LetU ! U2 be the diagonal map. This gives rise to a restriction

Map(U2, X) ! Map(U,X).

Note that when U is a single open ball, Map(U2, X) ! Map(U,X) isan equivalence, and since Map(U,X) is polynomial of degree " 1, itfollows from Theorem 3.2.4 that T1Map(U2, X) # Map(U,X). There-fore L2Map(U2, X) = hofiber(Map(U2, X) ! Map(U,X)). Similarly,U '! Map(Uk, X) is not homogeneous of degree k unless k = 1.

Example 4.1.4 We compute

L3Map(U3, X) = hofiber(T3Map(U3, X) ! T2Map(U3, X)).

As in the previous example, T3Map(U3, X) # Map(U3, X). Let!(U) (U3 denote the fat diagonal, !(U) = {(x1, x2, x3) |xi = xj for some i )=j}. We would like to claim that U '! Map(!(U), X) is a model forT2Map(U3, X), and while this is in spirit the case, our answer will beslightly di"erent.

We proceed as follows: For S ( {1, 2, 3}, let

!S(U) = {(x1, x2, x3) |xi = xj for all i, j % S}.

24 Brian A. Munson

Then !(U) = colim1<|S|!S(U) (the union of these spaces covers !(U),

and we define !(U) = hocolim1<|S|!S(U). Thus, since Map(!, X)turns homotopy colimits into homotopy limits by Proposition 7.2.2, con-sequently Map(!(U), X) = holim1<|S|Map(!S(U), X). It is clear thatMap(!S(U), X) is a polynomial of degree " 4! |S|, and since 1 < |S| "3, for all S under consideration, holim1<|S|Map(!S(U), X) is polyno-mial of degree " 2 because each functor in the diagram is polynomialof degree " 2. Note that U3 # hocolim1!|S|!S(U), and hence there

is a natural transformation of functors Map(U3, X) $ Map(!(U), X)given by the obvious inclusion of categories.

By inspection, when U is a union of at most two disjoint open balls,the map Map(U3, X) $ Map(!(U), X) is an equivalence, and so byTheorem 3.2.4, T2Map(U3, X) # Map(!(U), X). It follows that

L3Map(U3, X) # hofiber(Map(U3, X) $ Map(!(U), X)).

Spaces of maps are special cases of sections of bundles, and we cangeneralize further to include examples such as these.

Example 4.1.5 Let p : Z $!Mk

"be a fibration with a section. Let

"(!Mk

", Z; p) denote its (based) space of sections. The assignment U %$

"(!Uk

", Z; p) is polynomial of degree " k. Define

"

#!

#U

k

$, Z; p

$= hocolim

N"N"

##U

k

$&Q,Z; p

$.

One may think of this as the space of germs of sections near the fat

diagonal. It turns out that Tk#1"(!Uk

", Z; p) # "(!

!Uk

", Z; p), and hence

"c

##U

k

$, Z; p

$= hofiber

#"

##U

k

$, Z; p

$$ "

#!

#U

k

$, Z; p

$$.

is homogeneous of degree k. We refer to "c as the space of compactlysupported sections.

4.2 Classification of homogeneous polynomials

The last example in the previous section is quite general, according tothe classification of homogeneous functors.


Theorem 4.2.1 ([34, Theorem 8.5]) Let E be homogeneous of degreek. Then there is an equivalence, natural in U ,

E(U) !" !c

!!U

k

", Z; p

",

where !c is the space of compactly supported sections of a fibration p :Z "

#Uk

$. The fiber over S of the fibration p is the total homotopy fiber of

a k-cube of spaces made up of the values of E on a tubular neighborhoodof S. In particular, if E(U) = hofiber(TkF (U) " Tk!1F (U)), then thefibers of the classifying fibration are the derivatives F (k)(#).

This has a pleasing analogy with the kth homogeneous componentxk

k! f(k)(0) of the Taylor series centered at 0 for a smooth function f ,

where#Uk

$plays the role of xk

k! , and, of course, F(k)(#) plays the role of

f (k)(0). We will not discuss the proof of Theorem 4.2.1, but remark thatmost of the required tools are on display in the proof of Theorem 3.2.1.The classifying fibration p : Z "

#Uk

$is the pullback of a fibration

p : Z "#Mk

$, induced by the inclusion U " M .

5 Convergence and Analyticity

Now that we can construct a Taylor tower for a functor F and under-stand a bit about its structure, we are ready to ask whether or notit approximates the functor F in a useful way. The Taylor series of afunction f : R " R need not converge to f ; in fact, the series neednot converge at all. We will discuss the extent to which an approxima-tion by polynomial functors does a suitable job of approximating thehomotopy type of the values of a given functor. The reader may alreadysuspect that a “suitable” approximation is one which approximates thehomotopy type of through a range. On R, |x!y| measures the di"erenceof x and y, and in Top, a useful “metric” for measuring the di"erencebetween spaces X and Y with respect to a map f : X " Y is to ask forthe connectivity of the homotopy fiber hofiber(f).

Two natural questions to ask are:

1. Does the Taylor tower of a functor F converge to anything?

2. Does the Taylor tower converge to F?

Information about the first question can be obtained from Theo-rem 4.2.1, the characterization of homogeneous functors, and there is

26 Brian A. Munson

an easy answer if one can compute the connectivity of the derivativesof a functor. The second is much more di!cult. This section will firstdiscuss some generalities regarding convergence, including the usefulnotion of !-analyticity, where the integer ! is analogous to a radius ofconvergence. Section 6 will tackle the convergence question for spacesof embeddings, to give the reader a sense of what types of argumentsgo into proving convergence results in a specific example.

5.1 Convergence of the series

For a smooth function f : R ! R with Taylor series!

akxk

k! , the radius

of convergence r is the largest value of r such that!

akxk

k! convergesabsolutely for |x| < r. Thus there are two possibilities for the conver-gence of the series: either it converges only at 0, or it converges on anopen interval centered at 0.

We would not speak of convergence of the Taylor series of a functorF unless the homotopy type of TkF stabilizes with k; that is, unless themaps TkF ! Tk!1F have connectivity increasing to infinity with k. Fora functor F , we are interested in the homotopy type of holimk TkF , andwhether the homotopy type of TkF “stabilize” as k increases. One wayto detect this is to study the maps TkF ! Tk!1F . If their connectivitiesincrease to infinity with k, then we would say that the Taylor seriesconverges, and Theorem 4.2.1 is useful in giving us a means to attackthis. In particular, if the derivatives have increasing connectivity, thiswill ensure these maps are highly connected.

Proposition 5.1.1 For a good functor F , if F (k)(") is ck-connected,then LkF (M) is (ck # km)-connected. More generally, if U has handledimension j, then LkF (U) is (ck # kj)-connected.

The homogeneous classification theorem tells us that LkF (M) =hofiber(TkF (M) ! Tk!1F (M)) is equivalent to the space of compactlysupported sections of a fibration over

"Mk

#whose fibers are the deriva-

tives F (k)("). Thinking of a section space as a twisted mapping space,standard obstruction theory arguments (see Proposition 7.2.1) showthat if ck is the connectivity of F (k)("), then LkF (M) is (ck # km)-connected (see Proposition 7.2.1 for the basic idea). In any case, theTaylor tower of F converges for all U of handle dimension $ j if ck#kjtends to infinity with k.

We can see that the analog of the radius of convergence has some-thing to do with handle dimension, although we have not yet tackled


this in a serious way. This is organized more systematically below asthe notion of !-analyticity of a functor.

5.2 Convergence to the functor

We would certainly say that TkF converges to F if the canonical mapF ! holimk TkF is an equivalence. In this case, the connectivity ofLkF informs us about the connectivity of the “remainder” RkF =hofiber(F ! TkF ).

Proposition 5.2.1 For a good functor F , if F ! holimk F is an equiv-alence and Lk+1F is ck-connected, where ck is an increasing function ofk, then F ! TkF is ck-connected.

Proof. Since Lk+1F = hofiber(Tk+1F ! TkF ) is ck-connected, Tk+1

! TkF is (ck + 1)-connected, and since ck is an increasing function ofk, it follows that TlF ! TkF is (ck + 1)-connected for all l > k. SinceF ! holimk TkF is an equivalence, F ! TkF is ck-connected as well. !

Although it may be di!cult to establish a homotopy equivalenceF ! holimk TkF , in practice it is feasible to understand the connec-tivity of LkF by Proposition 5.1.1, since it reduces to computing theconnectivity of the derivatives F (k)("). Hence even with a lack ofknowledge of convergence, we can formulate conjectures about the con-nectivities of the maps F ! TkF based on the connectivity of LkF .Understanding the di"erence between F and TkF is a natural ques-tion in ordinary calculus as well. We are often interested in the errorRk(x) = |f(x) # Tkf(x)| for certain x. For f smooth on [#r, r] and

satisfying |f (k+1)| $ Mk on (#r, r), we have Rk(x) $ Mkrk+1

(k+1)! . If

Mkrk+1

(k+1)! ! 0 as k ! %, then we would say that f is analytic on

(#r, r); that is, its Taylor series converges to it. We wish, therefore, toanswer the following questions:

1. What is the analog of the radius of convergence?

2. What should we mean by a bound on f (k+1) within the radius ofconvergence?

3. How can we estimate the “error” RkF = hofiber(F ! TkF )?

28 Brian A. Munson

Briefly, the answer to the first question is that the radius of conver-gence is a positive integer !. An open set V which is the interior of asmooth compact codimension 0 submanifold L of M is within the radiusof convergence if the handle dimension of L is less than !. The answerto the second lies in our criticism given in the last section of Section 2 ofour definition of the derivatives of F . Our definition of derivative onlyallows the attaching of a handle of dimension 0 (disjoint union), whilewe will need to understand what happens for more general unions. Asimilar comment applies to higher derivatives. We will expand on all ofthis below.

To answer the third question, note that we are asking about theextent to which a given functor F fails to be polynomial of degree ! k.We have two options available to us. The first is to study the homotopyfiber of F (V ) " TkF (V ). This has the advantage that it is a naturaltransformation of functors, and it is the connectivity of this map weare ultimately interested in. Unfortunately, the target is a homotopylimit over a category not very accessible to computation. The otheroption is to study the extent to which the functor F fails to satisfy thedefinition of a polynomial. This is much more computationally feasible,because it involves values of the original functor on certain kinds ofcubical diagrams.

Suppose F : O(M) " Top is a functor and ! > 0 is an integer. Fork > 0, let P be a smooth compact codimension 0 submanifold of M , andQ0, . . . , Qk be pairwise disjoint compact codimension 0 submanifolds ofM # int(P ). Suppose further that Qi has handle dimension qi < !. LetUS = int(P $QS).

Definition 5.2.2 The functor F is !-analytic with excess c if the (k+1)-cube S %" F (US) is (c+

!ki=0(!# qi))-cartesian.

This is the analog of a bound on f (k)(x) for x close to 0. In this case,being close to zero means having small handle dimension, and the (k+1)-cube S %" F (US) certainly resembles a more general (k+1)st derivative-like expression. We will see shortly that ! gives the radius of convergenceof the Taylor tower of F . Note that this definition is concerned withsomething close to the kth derivative of F at P , although we allowourselves to study multirelative di!erences not just involving disjointopen balls, but arbitrary manifolds with bounded handle dimension. Itis this definition that gives us our answer to the second question above,as we will see in the next theorem, which is the estimate for the errorRkF = hofiber(F " TkF ).


Theorem 5.2.3 ([17, Theorem 2.3]) If F is !-analytic with excess c,and if U ! O(M) is the interior of a smooth compact codimension 0submanifold of M with handle dimension q < !, then the map F (U) "TkF (U) is (c+ (k + 1)(!# q))-connected.

Corollary 5.2.4 ([17, Corollary 2.4]) Suppose F is !-analytic with ex-cess c. Then for each U ! O(M) which is the interior of a com-pact codimension 0 submanifold of handle dimension < !, the mapF (U) " holimk TkF (U) is an equivalence.

This follows since the connectivities of the maps F (U) " TkF (U)increase to infinity with k if the handle dimension of U is less than !.Thus we see how the handle dimension can be thought of as the radiusof convergence, where an open set is measured by its handle dimension.

We will not give the proof of Theorem 5.2.3, although we wouldlike to make a few remarks. The strategy of the proof is similar to theinductive proof of Theorem 3.2.1. We are interested in the connectivityof the map F (U) " TkF (U), and as usual, it su!ces to study thespecial case where U is the interior of a smooth compact codimension 0submanifold L of M . Using a handle decomposition, we select pairwisedisjoint closed subsets A0, . . . , Ak such that for S $= %, US = U#

!i!S Ai

is the interior of a compact smooth codimension 0 submanifold whosehandle dimension is strictly less than the handle dimension of L. Wethen consider the diagram

F (U) !!

""

TkF (U)

""holimS "=# F (US) !! holimS "=# TkF (US).

The right vertical arrow is an equivalence since TkF is polynomial ofdegree & k, and by induction we can get a connectivity estimate for thebottom horizontal arrow. We have a connectivity for the left verticalarrow by assuming F is !-analytic. Together these give an estimate forthe connectivity of F (U) " TkF (U). The next section is devoted tounderstanding how to obtain connectivity estimates for the left verticalarrow in the case k = 1, 2 for the functor F (U) = Emb(U,N). Inparticular, the di!cult task is verifying that a given functor is !-analyticfor some !, which gives a connectivity estimate for the left vertical arrow.Before we embark on this, let us state one more corollary regarding

30 Brian A. Munson

convergence. This next result is the analog of the uniqueness of analyticcontinuation.

Corollary 5.2.5 ([17, Corollary 2.6]) Suppose F1 ! F2 is a naturaltransformation of !-analytic functors, and that F1(U) ! F2(U) is anequivalence whenever U " Ok(M) for some k. Then F1(V ) ! F2(V )is an equivalence for each V which is the interior of a smooth compactcodimension 0 submanifold of handle dimension less than !.

Proof. Suppose V " O(M). Consider the following diagram.

F1(V ) !!

""

F2(V )

""holimk TkF1(V ) !! holimk TkF2(V )

Since F1(U) ! F2(U) is an equivalence whenever U is in Ok(M) for anyk, it follows from Theorem 3.2.4 that TkF1 ! TkF2 is an equivalence forall k. Hence the lower horizontal arrow is an equivalence for all V . Ifthe handle dimension of V is less than !, then F1(V ) ! holimk TkF1(V )and F2(V ) ! holimk TkF2(V ) are equivalences by Corollary 5.2.4, soF1(V ) ! F2(V ) is an equivalence. !

6 Convergence for Spaces of Embeddings

The following is a theorem due to Klein and Goodwillie about the con-vergence of the Taylor tower of the embedding functor. A version forspaces of Poincare embeddings has appeared in [14], which is an impor-tant step in proving the result below, which will appear in [13].

Theorem 6.0.6 The functor U #! Emb(U,N) is n $ 2 analytic withexcess 3$ n. Hence, if M is a smooth closed manifold of dimension m,and N a smooth manifold of dimension n, then the map

Emb(M,N) $! Tk Emb(M,N)

is [k(n $ m $ 2) + 1 $ m]-connected. In particular, if n $ m $ 2 > 0,then the canonical map

Emb(M,N) $! holimk Tk Emb(M,N)

is a homotopy equivalence.


The proof of this theorem goes beyond the scope of this work, al-though we wish to present some of the ideas involved in arriving atsuch estimates. Note that the estimate for the map Emb(M,N) !Tk Emb(M,N) can be conjectured using Proposition 5.2.1; we will com-pute the connectivity of the derivatives of embeddings below. Note alsothat in the case m = 1 and n = 3 (essentially knot theory), we do nothave convergence (although the theorem still gives a non-trivial answer).

One can obtain the connectivity of Emb(M,N) ! T1 Emb(M,N)“by hand” without too much work, and some of the ideas that go intoone version of this computation (the second proof of Theorem 6.2.1below) are important in obtaining estimates for all k. We will alsodiscuss a weaker estimate for the map Emb(M,N) ! T2 Emb(M,N).The techniques required for the results above are far beyond the scopeof this work, and involves important relationships between embeddings,pseudoisotopies, and di!eomorphisms, as well as some surgery theory.

6.1 Connectivity of the derivatives of embeddings

The first step in understanding some of the ideas that go into establish-ing the analyticity of the embedding functor is to compute the connec-tivity of the derivatives of the embedding functor.

Theorem 6.1.1 Let U =!

iBi " M be a disjoint union of k openballs. For S " k, let US = U #

"i!S Bi. The k-cube S $! Emb(US , N)

is ((k # 1)(n # 2) + 1)-cartesian. That is, if E(U) = Emb(U,N), thenE(k"1)(%) is (k # 1)(n# 2)-connected.

Let us begin with an observation that will simplify things. For asubset S of k, the projection map

#i/!S Bi&Emb(US , N) ! Emb(US , N)

is an equivalence because balls and products of balls are contractible(if S = k, we take

#i/!S Bi to be a point). Let C(j,N) denote the

configuration space of j points inN . The map#

i/!S Bi&Emb(US , N) !C(k # |S|, N)& Imm(US , N) which is induced by the map which sends((x1, . . . , xk), f) to ((f(x1), . . . , f(xk), (dfx1 , . . . , dfxk)) is an equivalencefor all S (where again the product of balls is taken to be a point ifS = k). Hence S $! Emb(US , N) is j-cartesian if and only if S $!C(k# |S|, N)& Imm(US , N) is j-cartesian. The cube S $! Imm(US , N)is homotopy cartesian whenever k ' 2 because Imm(#, N) is polynomialof degree ( 1. Therefore S $! Emb(US , N) is j-cartesian if and only ifS $! C(k# |S|, N) is j-cartesian for k ' 2. For illustration, we will onlyprove this in the case where k = 2. The cases k ' 3 are straightforward

32 Brian A. Munson

enough, and all that they require is an application of the Blakers-MasseyTheorem 7.3.2.Proof. For k = 2, we are looking at the square

Emb(U,N) !!

""

Emb(U0, N)

""Emb(U1, N) !! Emb(!, N)

By the remarks preceding the proof, this square is j-cartesian if andonly if

C(2, N) !!

""

C(1, N)

""C(1, N) !! C(!, N)

is j-cartesian. The maps in this diagram are fibrations, and taking fibersvertically over p " C(1, N) yields the 1-cube N # {p} $ N , which is an(n#1)-connected map, and hence the original square is (n#1)-cartesianby Proposition 2.2.5. !

As we mentioned, the Blakers-Massey Theorem 7.3.2 needs to beapplied for higher k. For instance, the case k = 3 ends with fiberingover (p, q) " C(2, N) and observing that the square

N # {p, q} !!

""

N # p

""N # q !! N

is a homotopy pushout and is (2n# 3)-cartesian by the Blakers-MasseyTheorem 7.3.2.

6.2 Connectivity estimates for the linear and quadraticstages for embeddings

We will give two proofs of the following theorem. The second requiresa disjunction result from the next section, but beyond this, it is almostidentical to the proof of Theorem 3.2.1.

Theorem 6.2.1 The map Emb(M,N) $ T1 Emb(M,N) is (n# 2m#1)-connected. In fact, if V % M is the interior of a compact codimension0 handlebody with handle dimension k, then the map is (n # 2k # 1)-connected.


The first proof is much easier and employs general position argu-ments, although it only gives the connectivity estimate in terms of thedimension of M , not the stronger statement involving the handle di-mension. The second uses a bit more machinery, but reduces the proofto the special case where M is the disjoint union of balls via an induc-tion argument on the handle dimension, but requires a disjunction resultfrom the next section. Hopefully this further convinces the reader of theimportance of derivatives. Its methods are also important in organizingthe proof of the connectivity estimate for Emb(M,N) ! Tj Emb(M,N)for all j.

First Proof. We have already mentioned that

T1 Emb(M,N) " Imm(M,N).

Let h : Sk ! Imm(M,N) be a map with adjoint H : M # Sk ! N .Consider the map H : M #M # Sk ! N #N defined by H(x, y, s) =(H(x, s), H(y, s)). We can arrange, by a small homotopy, for H to besmooth and H to be transverse to the diagonal. Let D = H!1(!(N))be the inverse image of the diagonal. It is a submanifold of M#M#Sk

of dimension 2m+k$n, which is empty if k < n$2m, and in this case,the map h clearly has image in Emb(M,N). A similar argument showsthat a homotopy h : Sk # I ! Imm(M,N) lifts to Emb(M,N) if k <n$2m$1, and it follows that the inclusion Emb(M,N) ! Imm(M,N)is (n$ 2m$ 1)-connected. !Second Proof. We will induct on k. For the base case k = 0, let l bethe number of components of V . The result is trivial, and the map inquestion is an equivalence, when l = 0, 1. Suppose l % 2. Consider thesequence

(3) Emb(V,N) ! Tl Emb(V,N) ! Tl!1 Emb(V,N) !· · · ! T1 Emb(V,N).

The map Emb(V,N) ! Tl Emb(V,N) is an equivalence since V is afinal object in Ol(V ). By the classification Theorem 4.2.1 of homoge-neous functors, we have that Lj Emb(V,N) = hofiber(Tj Emb(V,N) !Tj!1 Emb(V,N)) is equivalent to "c(

!Vj

",Emb(j)(&)). Since V has han-

dle dimension 0,!Vj

"also has handle dimension 0, and the fibers are

thus (j$1)(n$2)-connected; in other words, the map Tj Emb(V,N) !Tj!1 Emb(V,N) is ((j$1)(n$2)+1)-connected. This is true no matter

34 Brian A. Munson

D2

A0

A2

A3

!D1!D1L

D0 A1D3

D1

Figure 1: A picture of four disks Di in the core of a 1-handle D1 !D1

attached to L along !D1 !D1 and their corresponding thickenings Ai.The Di are subsets of the core D1 ! {0} (which itself is depicted asthe curve in the middle of the handle), and Ai = Di !D1 " D1 !D1.Note that removing k # 1 of the Ai leaves a manifold with (k$1) extra0-handles, but one fewer 1-handle.

what basepoint is chosen, provided m < n. It follows that the composedmap Emb(V,N) % T1 Emb(V,N) is (n$ 1)-connected.

Now suppose k > 0. Let V = int(L). For j = 1 to s, let ej :Dk ! Dn!k % L denote each of the s k-handles. Assume e!1

j (!L) =

!Dk!Dn!k for all j. Since k > 0, we may choose pairwise disjoint closeddisks D0, D1 in the interior of Dk, and put Aj

i = ej(Di ! Dn!k) & V .

Then each Aji is closed in V , and if we set Ai = 's

j=1Aji , then for each

nonempty subset S of {0, 1}, VS = V $'i"SAi is the interior of a smoothcompact codimension 0 submanifold of M of handle dimension strictlyless than k. See Figure 1 for a low-dimensional picture where there arefour disks Di instead of just two.

In the square diagram

Emb(V,N) !!

""

T1 Emb(V,N)

""holimS #=$ Emb(VS , N) !! holimS #=$ T1 Emb(VS , N),

the right vertical arrow is again an equivalence because T1 Emb($, N) ispolynomial of degree( 1, and by induction for all S )= *, Emb(VS , N) %T1 Emb(VS , N) is (n$2(k$1)$1)-connected, and by Proposition 2.2.6,the map of homotopy limits has connectivity n$ 2(k$ 1)$ 1$ 2+ 1 =


n!2k. By Theorem 6.3.5, the left vertical map is (n!2k!1)-connected,and it follows that the top horizontal map is (n! 2k ! 1)-connected.!

The base case of the induction on handle dimension above requiredan argument which was di!erent from the inductive step. In particularit required knowledge of the higher derivatives, and we do not see a wayaround this. Attempts to mimic the inductive step for the base caseyield connectivity estimates which are less than those desired.

Theorem 6.2.2 The map Emb(M,N) " T2 Emb(M,N) is (2n!3m!3)-connected. In fact, if V # M is the interior of a compact codimen-sion 0 submanifold of M whose handle dimension is k, then the mapEmb(V,N) " T2 Emb(V,N) is (2n! 3k ! 3)-connected.

The second proof of Theorem 6.2.1 can be adapted with very fewchanges. The only changes (besides the connectivity estimates them-selves) are that the pairwise disjoint closed subsets chosen are three innumber, and instead of referencing Theorem 6.3.5, we reference Theo-rem 6.3.6. However, Theorem 6.3.6 is weaker than what we need, andwe can really only claim to prove a weaker version of Theorem 6.2.2,stated below. The issue here is that there is a stronger version of The-orem 6.3.6 which we are unable to prove by elementary means.

Theorem 6.2.3 With hypotheses as in Theorem 6.2.2, the map

Emb(V,N) " T2 Emb(V,N)

is (2n! 4k ! 3)-connected.

6.3 Some disjunction results for embeddings

For the second proof of Theorem 6.2.1 we needed an estimate for howcartesian the square E

Emb(V,N) !!

""

Emb(V0, N)

""Emb(V1, N) !! Emb(V01, N)

is. Here V = V! is the interior of some smooth compact codimension0 submanifold of M with handle dimension k, and, for S $= %, theVS are the interiors of compact codimension 0 submanifolds of handle

36 Brian A. Munson

dimension less than k. As in the proof of Theorem 6.2.1, let V = int(L).We chose each Ai to be a union of products of a k-dimensional diskwith an (m ! k)-dimensional disk. Note that LS = L !

!i!S Ai is not

compact, but its interior is the interior of a smooth compact codimension0 submanifold of M . This is important to note because below we willwork not with the open sets that appear in E , but with their closedcounterparts L and the Ai.

Let us first consider a formally similar situation. Suppose Q0 andQ1 are smooth closed manifolds of dimensions q0 and q1 respectively,and let QS =

!i/!S Qi for S " {0, 1}. Consider the square S = S #$

Emb(QS , N):

Emb(Q0 %Q1, N) !!

""

Emb(Q0, N)

""Emb(Q1, N) !! Emb(&, N).

It is enough by Proposition 2.2.5 to choose a basepoint in Emb(Q0 %Q1, N) and take fibers vertically and compute the connectivity of themap of homotopy fibers. By the isotopy extension theorem, the mapEmb(Q0%Q1, N) $ Emb(Q1, N) is a fibration with fiber Emb(Q0, N !Q1). We will show that Emb(Q0, N ! Q1) $ Emb(Q0, N) in Theo-rem 6.3.5 below, and hence the square S, is (n! q0 ! q1 ! 1)-cartesian.Although the squares E and S are formally similar, it is not clear howto use Theorem 6.3.5 to give an estimate for how cartesian the squareE is.

First note that we can generalize the situation in the square S to arelative setting. That is, suppose Q0, Q1 and N have boundaries, andthat embeddings ei : !Qi $ !N have been selected to have disjointimages. Let Emb!(QS , N) be the space of embeddings f : QS $ Nsuch that the restriction of f to !QS is equal to eS , and such thatf"1(!N) = !QS . Then it is also true that

Emb!(Q0 %Q1, N) !!

""

Emb!(Q0, N)

""Emb!(Q1, N) !! Emb!(&, N)

is (n! q0 ! q1 ! 1)-cartesian; in particular, the proof of this is identicalto that of Theorem 6.3.5 with the exception of repeating the phrase“relative to the boundary” over and over.


We can make a further generalization to the case of compact man-ifold triads (defined in Section 1.2). Suppose the Qi are compact n-dimensional manifold triads of handle dimension qi, where n ! qi " 3,and Y is an n-dimensional smooth manifold with boundary. In this caseembeddings ei : !0Qi # !N have been chosen, and we let Emb!0(QS , N)stand for the obvious thing.

Theorem 6.3.1 ([17, Theorem 1.1]) The diagram

Emb!0(Q0 $Q1, N) !!

""

Emb!0(Q0, N)

""Emb!0(Q1, N) !! Emb!0(%, N)

is (n! q0 ! q1 ! 1)-cartesian.

This can be generalized to the case where the dimension of the Qi ism & n, essentially by a thickening of the m-dimensional Qi by the diskbundle of an (n!m)-plane bundle.

Proposition 6.3.2 ([17, Observation 1.3]) If dim(Qi) = m & n thenTheorem 6.3.1 is true.

The rough idea of the proof is to assume that Y is embedded in Rn+k

and let Grn!m = colimk Grn!m+k(Rn+k) be a limit of Grassmannians.Consider the map Emb(QS , Y ) # Map(QS , Grn!m) given by sending anembedding f to its normal bundle "f . The homotopy fiber of this mapover some # can be identified with the space of embeddings of the diskbundle of # overQS . Since S '# Map(QS , Grn!m) is homotopy cartesian(because Map(!, X) is polynomial of degree & 1), by Proposition 2.2.5,the square of homotopy fibers is (n! q0 ! q1 ! 1)-cartesian if and onlyif the square S '# Emb!(QS , Y ) is (n ! q0 ! q1 ! 1)-cartesian. Note,however, that this introduces more corners, since the closed disk bundleof a smooth manifold with boundary is already a compact manifold triaditself. The new corners due to the disk bundle are introduced along thecorner set of the original compact manifold triad. It will do no harm toignore this.

Without changes whatsoever we can assume the Qi are submanifoldsof an m-dimensional manifold M . Now we are in a position to describea situation which is directly related to the square E , and we generalizethis situation further by introducing a new manifold P . Suppose that

38 Brian A. Munson

P is a smooth compact codimension 0 manifold triad in M , Q0, Q1 aresmooth compact codimension 0 manifold triads in M! int(P ), and thatthe handle dimension of Qi satisfies n! qi " 3. Put QS =

!i/!S Qi.

Proposition 6.3.3 The square S #$ Emb(P%QS , N) is (n!q0!q1!1)-cartesian.

Proof. The square S #$ Emb(P,N) is homotopy cartesian since all mapsare equivalences, and hence S #$ Emb(P %QS , N) is (n! q0 ! q1 ! 1)-cartesian if and only if the square of homotopy fibers

S #$ hofiber(Emb(P %QS , N) $ Emb(P,N))

is (n! q0 ! q1 ! 1)-cartesian for all choices of basepoint in Emb(P,N).The map Emb(P % QS , N) $ Emb(P,N) is a fibration with fiberEmb!0(QS , N!P ), which is (n!q0!q1!1)-cartesian by Theorem 6.3.1.!

We finally arrive at the technical statement which relates the opensets in square E with the closed sets we have been considering.

Corollary 6.3.4 ([17, Corollary 1.4]) Let P,Q0, Q1 be as in Propo-sition 6.3.3, and set VS = int(P % QS). Then S #$ Emb(VS , N) is(n! q0 ! q1 ! 1)-cartesian.

To connect this explicitly with the square E , we choose the Qi tobe the Ai considered in Theorem 6.2.1, and P to be the closure ofL!(A0%A1). We now proceed to give the promised disjunction results.

Theorem 6.3.5 Suppose P and Q are smooth compact submanifolds ofan n-dimensional manifold N of dimensions p and q respectively. Theinclusion map Emb(P,N!Q) $ Emb(P,N) is (n!p!q!1)-connected.

An important special case is when both P and Q = & are points,which says that N ! & $ N is (n ! 1)-connected. The rough idea,expanded in the proof below, is that any map Sk $ N misses a point ifk < n, and that the same is true of any homotopy Sk ' I if k < n! 1.The former proves the map of homotopy groups is surjective if k < nand the latter that it is injective if k < n! 1.

Proof. We will not fuss about basepoints. The following argument canbe adapted to accomodate them. Let Sk $ Emb(P,N). We may regardthis as a map Sk ' P $ N , and by a small homotopy we can make it


both smooth and transverse to Q ! N . If k + p < n " q, equivalently,k < n " p " q, transverse intersection means empty intersection, andhence we have a map Sk #P $ N "Q, which gives us our desired mapSk $ Emb(P,N " Q). A similar argument shows that any homotopySk # I $ Emb(P,N) lifts to Emb(P,N "Q) if k < n" p" q" 1, hencethe map in question is (n" p" q " 1)-connected. !

We can piggyback on the previous result to obtain the followinggeneralization.

Theorem 6.3.6 The square of inclusion maps

Emb(P,N " (Q0 %Q1)) !!

""

Emb(P,N "Q0)

""Emb(P,N "Q1) !! Emb(P,N)

is (2n" 2p" q1 " q2 " 3)-cartesian.

Once again the special case where P , Q0 = &0 and Q1 = &1 are allpoints is a good one to consider before embarking on the proof. In thatcase, we claim that the square of inclusion maps

N " (&0 % &1) !!

""

N " &0

""N " &1 !! N

is (2n " 3)-cartesian. The square is clearly a homotopy pushout, andsince the maps N " (&0 % &1) $ N " &i are (n " 1)-connected for i =0, 1, by the Blakers-Massey Theorem, the square is (2n " 3)-cartesian.Unfortunately, in the general case the square will not be a homotopypushout, but it is close to being one, and we will use a generalization ofthe Blakers-Massey Theorem to complete the proof.

Proof. We claim that the square in question is (2n" 2p" q0 " q1 " 1)-cocartesian (see Definition 7.3.1 in the appendix). Given this, since themaps Emb(P,N " (Q0 %Q1)) $ Emb(P,N "Qi) are (n" p" qj " 1)-connected for i = 0, 1 and j = 1, 0 respectively by Theorem 6.3.5,it follows by the generalized Blakers-Massey Theorem 7.3.3 that thesquare is min{2n" 2p" q0" q1" 1, 2n" 2p" q0" q1" 3}-cartesian, andthe result follows.

40 Brian A. Munson

Now we will establish the estimate for how cocartesian this squareis. Let h : Sk ! Emb(P,N) be a map, and H : P " Sk ! N itsadjoint. We wish to construct a lift to hocolim(Emb(P,N # Q0) $Emb(P,N # (Q0 %Q1))! Emb(P,N #Q1)). It is enough if H has theproperty that for all s & Sk, H(P " {s}) ' N #Qi for some i (possiblyboth).

By a small homotopy, we can make the map H : P"Sk ! N smoothand transverse to Q0 and Q1. For i = 0, 1, let Wi = H!1(Qi). Then W0

and W1 are disjoint submanifolds of P "Sk of dimensions p+k+ q0#nand p + k + q1 # n respectively. Consider the map d : W0 " W1 !Sk " Sk induced by projection to Sk. Again by a small homotopy of Hwe can arrange for this map to be transverse to the diagonal, and letD = d!1(!(Sk)). The dimension of D is 2p + q0 + q1 # 2n + k, whichis negative (and hence D is empty) if k < 2n # 2p # q0 # q1, so that hlifts if k ( 2n# 2p# q0# q1# 1. A similar argument shows a homotopyh : Sk" I ! Emb(P,N) lifts if k ( 2n# 2p# q0# q1# 2, and hence thesquare is (2n# 2p# q0 # q1 # 1)-cocartesian. !

Remark 6.3.7 Theorem 6.3.6 tells us that the 3-cube S )! (Emb(P %QS , N)! Emb(QS , N)) is (2n#2p#q0#q1#3)-cartesian. Relabel andput Q2 = P and q2 = p, and for R ' {0, 1, 2} write R )! Emb(QR, N),whereQR =

!i/"R Qi. The estimate 2n#2q2#q0#q1#3 is not symmetric

in the qi; that is, it depended upon us choosing a way to view this 3-cube as a map of squares. We could therefore slightly improve ourresult by letting P = Qi, where Qi minimizes the handle dimensionamong Q0, Q1, Q2. Although this is a slight improvement, it is not thebest possible, which is probably not surprising in light of the asymmetryin the qi. One can prove that the 3-cube R )! Emb(QR, N) is (2n #q0 # q1 # q2 # 3)-cartesian.

Let us end by remarking that a lack of convergence does not nec-essarily mean the Taylor tower does not contain anything interesting.In fact, for embeddings of I in R2 " I relative to the boundary (whereTheorem 6.0.6 doesn’t give convergence because the codimension is 2),[32] shows that the Taylor series for the embedding functor containsfinite type invariants of knots. On a related note, one can study mul-tivariable functors such as (U, V ) )! Link(U, V ;N). Here U and V areopen subsets of smooth closed manifolds P and Q, and Link(U, V ;N)is the space of “link maps” U ! N , V ! N whose images are disjoint.It is not known whether its (multivariable) Taylor series converges to


it, but it is clear that its polynomial approximations are interesting,since, for example, hofiber(Link(P,Q;N)! T1 Link(P,Q;N)) containsthe information necessary to define the generalized linking number. See[24] and [16].

7 Appendix

This section contains a collection of facts about homotopy limits andcubical diagrams. It is by no means exhaustive, even for the purposesof this paper. What we have included is most of what one needs to beable to understand this work, and we have tried to include only thosefacts which seem more widely useful in calculus of functors. Proofs aregenerally omitted, and if given, are very sketchy and are just meantto outline the major ideas and/or give an intuitive understanding. Asmany references as possible are given.

7.1 Homotopy limits and colimits

The standard reference for homotopy limits and colimits is [4]. Othersinclude [8], [9], [18], and [30]. We begin with an explicit description ofthe homotopy limit and homotopy colimit. We assume the reader has apassing familiarity with simplicial sets and their realizations, as well asover/under categories. We present these models because they involveexpressions which are easy for a beginner to grasp, and while othermodels for homotopy limits and colimits are better for other purposes,these are good for getting one’s hands dirty with categories with finitelymany objects and morphisms. For a category D, we write |D| for therealization of its nerve.

Definition 7.1.1 ([4, XI.3.2 and XII.2.1]) Let C be a small categoryand F : C ! Top a covariant functor. The homotopy limit of F , denotedholimC F , is

(4) lim(!

c

Map(|C " c|, F (c))

!!

c!c!

Map(|C " c|, F (c"))#!

c!

Map(|C " c"|, F (c"))).

Dually, the homotopy colimit of F , denoted hocolimC F , is

colim("

c

|c " C|$ F (c)#"

c!c!

|c" " C|$ F (c)!"

c!

|c" " C|$ F (c")).

42 Brian A. Munson

The maps in these diagrams are induced by the identity map andfunctoriality of various functors. That is, c ! c! induces maps F (c) !F (c!), |c! " C| ! |c " C|, and |C " c| ! |C " c!|. Note that Map(#,#) iscontravariant in the first variable and covariant in the second. One nicefeature of this definition is that it only requires the reader to understandlimits and colimits of very simple diagrams, namely lim(X1 ! X12 $X2) and colim(X1 $ X" ! X2), which are the fiber product of X1 withX2 over X12 and the union of X1 with X2 along X" respectively.

The ordinary limit (inverse limit) and colimit (direct limit) of F canbe defined by replacing all realizations of over/under categories abovewith a point. Indeed, the limit is a subspace of the product

!c#C F (c)

and the colimit is a subspace of the coproduct"

c#C F (c).An important special case is when C = P(2) is the poset of subsets

of {1, 2}. Let X : P({1, 2})! Top be a covariant functor, and write XS

in place of X (S). Let P0(2) be the subposet of nonempty subsets, andlet P1(2) be the subposet of proper subsets. We depict the diagram ofspaces as follows.

X"f1 !!

f2""

X1

g1""

X2 g2!! X12

Although it is somewhat tedious, it is straightforward to show thefollowing from the definition of homotopy limit and homotopy colimit.

Proposition 7.1.2 We have

(5) holimP0(2)

F = {(x1, !, x2) % X1 &Map(I,X12)&X2

| !(0) = g1(x1), !(1) = g2(x2)}

andhocolim

P1(2)F = (X1

#X" & I

#X2)/ '

where for x % X", (x, 0) ' f1(x) and (x, 1) ' f2(x). In particular, ifX2 = ( is the one-point space, then holimP0(2) F is the homotopy fiberof g1 over g2((), and hocolimP1(2) F is the homotopy cofiber of f1.

Thus homotopy (co)fiber is a special case of homotopy (co)limit. Itis also true that homotopy (co)limits commute.


Theorem 7.1.3 Suppose F : C ! D " Top is a bifunctor. Then thereare homeomorphisms

holimC

holimD

F #= holimC!D

F #= holimD

holimC

F

andhocolim

Chocolim

DF #= hocolim

C!DF #= hocolim

Dhocolim

CF.

In particular, homotopy limits commute with homotopy fibers andhomotopy colimits commute with homotopy cofibers. The followingtheorem establishes the homotopy invariance of homotopy limits andcolimits.

Theorem 7.1.4 ([4, XI.5.6 and XII.4.2]) Suppose F " G is a natu-ral transformation of functors from C to Top. If F (c) " G(c) is anequivalence for all c $ C, then this induces equivalences

holimC

F %" holimC

G

andhocolim

CF %" hocolim

CG.

This fact is not true for ordinary limits and colimits; one can viewthe construction of the homotopy limit/colimit as a way to remedy thisfailure. Nevertheless, there are conditions under which the categorical(co)limit is equivalent to the homotopy (co)limit; we will not pursue thishere, but point out that we have encountered this situation already forsquare diagrams, where it was enough that the square be a categoricalpushout/pullback and a map from the initial object/to the final objectbe a cofibration/fibration.

Theorem 7.1.5 ([4, XI.4.1 and XII.3.1]) Suppose F : C " Top is co-variant. If C has an initial object ci, then holimC F & F (ci). If C has afinal object cf , then hocolimC F & F (cf ).

If F is contravariant, then we need to switch “initial” with “final”in the above statement. The corresponding facts about ordinary limitsand colimits are obvious if one defines such notions in terms of universalproperties. The following is a useful result we used in the proof ofTheorem 3.2.1, and is also central to the proof of Theorem 4.2.1. Itdescribes a close relationship between homotopy limits and homotopycolimits when the functor F is especially well-behaved. It has a similarflavor to Quillen’s Theorems A and B.

44 Brian A. Munson

Theorem 7.1.6 ([10]) If F : C ! Top takes all morphisms to homo-topy equivalences, then hocolimC F quasifibers over |C|, and the space ofsections of the associated fibration is equivalent to holimC F .

The quasifibration statement is at least relatively easy to believe. Ifwe let " : C ! Top denote the functor which takes all objects to theone-point space, then hocolimC " # |C|, and the natural transformationF ! " induces the map hocolimC F ! |C|. Since F takes all morphismsto homotopy equivalences, the fibers F (c) all have the same homotopytype.

7.2 The functor Map($,$)

We can regard Map($,$) : Top%Top ! Top as a bifunctor which iscontravariant in the first variable and covariant in the second variable.We are mostly interested in Map($, Z) and its variants for a fixed Z,but it is also useful to consider Map(X,$) for a fixed X.

Proposition 7.2.1 For a finite complex X of dimension k, the func-tor Map(X,$) takes j-connected maps to (j $ k)-connected maps. Inparticular, if Z is j-connected, then Map(X,Z) is (j $ k)-connected.

This can be proved by standard obstruction theory arguments. It istrue of more general mapping spaces too, such as the space of sectionsof a fibration p : E ! B, which we loosely think about as the space ofmaps from B into the fiber of p.

Proposition 7.2.2 ([4, XII.4.1]) The functor Map($, Z) : Top ! Toptakes (homotopy) colimits to (homotopy) limits. That is, if C is a smallcategory and X : C ! Top a functor, then Map(hocolimc!C X (c), Z) #holimc!C Map(X (c), Z). In particular, it takes coproducts to products,and turns homotopy cofiber sequences into homotopy fiber sequences.

Proof. We will indicate some of the ideas that go into showing a specialcase of this: if

X" !!

""

X1

""X2

!! X12


is a homotopy pushout square, then for any space Z,

Map(X12, Z) !!

""

Map(X1, Z)

""Map(X2, Z) !! Map(X!, Z)

is a homotopy pullback square. First, every homotopy pushout squareadmits an equivalence from a pushout square; that is, a square of theform

X! !!

""

X1

""X2

!! X1 !X! X2

where X! " Xi is a cofibration for each i = 1, 2. If we apply Map(#, Z)to this square, one checks by inspection that the resulting square is apullback; that is, that Map(X1 !X! X2, Z) = Map(X1, Z) $Map(X!,Z)

Map(X2, Z). Hence the square

Map(X12, Z) !!

""

Map(X1, Z)

""Map(X2, Z) !! Map(X!, Z)

is a pullback. The functor Map(#, Z) takes a cofibration A " X toa fibration Map(X,Z) " Map(A,Z), and so the pullback square is infact a homotopy pullback. !

The proof of the general statement is in fact easier given an explicitdescription of the homotopy colimit of a functor X : C " Top in termsof under categories. See [4].

7.3 The Blakers-Massey Theorem

We will only make statements for square diagrams, as those are the onlytypes of cubical diagrams we have seriously studied in this work. All ofwhat we say here has generalizations to higher dimensional cubes, andwe refer the reader to [12] for details. We have already made use of thenotion of a k-cartesian cube. Its dual notion, namely what it meansfor a cube to be k-cocartesian, is useful because the Blakers-MasseyTheorem tells us how cartesian a k-cocartesian cube is.

46 Brian A. Munson

Definition 7.3.1 For a |T |-cube X , the total homotopy cofiber is thehomotopy cofiber of the canonical map

b(X ) : hocolimS!T X (S) !" X (T ).

If b(X ) is k-connected, we say the cube is k-cocartesian, and if k =#,we say the cube is homotopy cocartesian.

Thus a squareX! !!

""

X1

""X2

!! X12

is homotopy cocartesian if the map hocolim(X1 $ X! " X2) " X12

is an equivalence. The homotopy colimit appearing here is the doublemapping cylinder.

Theorem 7.3.2 Suppose the square

X! !!

""

X1

""X2

!! X12

is homotopy cocartesian, and that the maps X! " Xi are ki-connected.Then the square is (k1 + k2 ! 1)-cartesian.

Here is a useful generalization, whose proof we omit.

Theorem 7.3.3 Suppose the square

X! !!

""

X1

""X2

!! X12

is k-cocartesian, and that the maps X! " Xi are ki-connected. Thenthe square is min{k ! 1, k1 + k2 ! 1}-cartesian.

We will provide only a very bare sketch of Theorem 7.3.2, if onlyto point out that one way of proving this uses disjunction techniquesreminiscent of our arguments in Section 6.3.


Proof. One can reduce to the case where Xi is the union of X! with a(ki + 1)-cell eki+1, and X12 is the union of X! with both cells. We aretherefore interested in the square

X! !!

""

X! ! ek1+1

""

X! ! ek2+1 !! X! ! ek1+1 ! ek2+1.

For i = 1, 2, let ai be points in the interior of the two cells, and rewritethis square as

Y " {a1, a2} !!

""

Y " {a2}

""Y " {a1} !! Y.

The claim is that the map

(6) Y " {a1, a2}# holim(Y " {a2}# Y $ Y " {a1})

is (k1 + k2 " 1)-connected. Recall the description of the codomain wegave just after Definition 7.1.1. A map

! : Sk # holim(Y " {a2}# Y $ Y " {a1})

corresponds by adjointness to a map ! : Sk% I # Y such that !(s, 0) &=a2 and !(s, 1) &= a1 for all s ' Sk. Extend ! to ("", 1+") to avoid talkingabout manifolds with corners. By a small homotopy, make ! smoothnear the ai and transverse to them (we may speak of smoothness becausethe interior of a cell has a smooth structure; transversality here meansthe ai are regular values of !). Now consider the map

! : Sk % {(t1, t2)|ti &= t2}# Y % Y

given by (s, t1, t2) (# (!(t1), !(t2)). Again a small homotopy will makea1 % a2 a regular value of !. Note that !"1(a1 % a2) has dimensionk" k1" k2 and hence will be empty if k < k1+ k2. This means that forsuch k, ! is homotopic to a map !#(s, t) which misses the ai for all t, andso !# lifts to a map Sk # Y "{a1, a2}. Hence the map in equation (6) issurjective on homotopy groups for k ) k1 + k2 " 1. A similar argumentestablishes injectivity when k < k1 + k2 " 1. !

48 Brian A. Munson

AcknowledgementThe author thanks Jesus Gonzalez for helpful conversations during

the preparation of this manuscript, Greg Friedman and Peter Landweberfor helpful feedback, and Wellesley College for their hospitality.

Brian A. MunsonDepartment of Mathematics,Wellesley College,Wellesley, [email protected]

References

[1] Greg Arone, Pascal Lambrechts, Victor Turchin, and Ismar Volic,Coformality and rational homotopy groups of spaces of long knots,Math. Res. Lett. 15 (2008), no. 1, 1–14.

[2] Gregory Arone, Derivatives of embedding functors I: The stablecase, arXiv:math.AT/0707.3489.

[3] Gregory Arone, Pascal Lambrechts, and Ismar Volic, Calculus offunctors, operad formality, and rational homology of embeddingspaces, Acta Math. 199 (2007), no. 2, 153–198.

[4] A. K. Bousfield and D. M. Kan, Homotopy limits, completions andlocalizations, Springer-Verlag, Berlin, 1972, Lecture Notes in Math-ematics, Vol. 304.

[5] Ryan Budney, Little cubes and long knots, Topology 46 (2007),no. 1, 1–27.

[6] , A family of embedding spaces, Groups, homotopy, and con-figuration spaces, Geom. Topol. Monogr., vol. 13, Geom. Topol.Publ., Coventry, 2008, pp. 41–83.

[7] Ryan Budney, James Conant, Kevin P. Scannell, and Dev Sinha,New perspectives on self-linking, Adv. Math. 191 (2005), no. 1,78–113.

[8] W. G. Dwyer and J. Spalinski, Homotopy theories and model cat-egories, Handbook of Agebraic Topology, North-Holland, Amster-dam, 1995, pp. 73–126.


[9] William G. Dwyer, Homotopy theory and classifying spaces, tenunpublished lectures (consisting of 112 pages) given in Copen-hagen, Denmark, in June 2008. Notes and slides available online atwww.math.ku.dk/!jg/homotopical2008/Dwyer.CopenhagenNotes.pdf.

[10] , The centralizer decomposition of BG, Algebraic topology:new trends in localization and periodicity (Sant Feliu de Guıxols,1994), Birkhauser, Basel, 1996, pp. 167–184.

[11] T. Goodwillie, J. Klein, and M. Weiss, Spaces of smooth embed-dings, disjunction, and surgery, Surveys on surgery theory, vol. 2,Ann. of Math. Stud., no. 149, Princeton Univ. Press, Princeton,NJ, 2001, pp. 221–284.

[12] Thomas G. Goodwillie, Calculus II: Analytic functors, K-Theory5 (1991/92), no. 4, 295–332.

[13] Thomas G. Goodwillie and John R. Klein, Multiple disjunction forspaces of smooth embeddings, in preparation.

[14] , Multiple disjunction for spaces of Poincare embeddings,Journal of Topology 1 (2008), 761–803.

[15] Thomas G. Goodwillie, John R. Klein, and Michael S. Weiss, AHaefliger style description of the embedding calculus tower, Topol-ogy 42 (2003), no. 3, 509–524.

[16] Thomas G. Goodwillie and Brian A. Munson, A stable range de-scription of the space of link maps, Algebr. Geom. Topol. (2010),1305–1315.

[17] Thomas G. Goodwillie and Michael Weiss, Embeddings from thepoint of view of immersion theory II, Geom. Topol. 3 (1999), 103–118 (electronic).

[18] Philip S. Hirschhorn, Model categories and their localizations,Mathematical Surveys and Monographs, vol. 99, Amer. Math. Soc.,2003.

[19] Antoni A. Kosinski, Di!erential manifolds, Dover Books on Math-ematics, Dover, 2007 (originally published in Pure and AppliedMathematics, 138. Academic Press, Inc., Boston, MA, 1993).

50 Brian A. Munson

[20] Nicholas J. Kuhn, Goodwillie towers and chromatic homotopy: anoverview, Proceedings of the Nishida Fest, Geom. Topol. Monogr.,vol. 10, Geom. Topol. Publ., Coventry, 2007, pp. 245–279.

[21] Pascal Lambrechts, Victor Turchin, and Ismar Volic, The rationalhomology of spaces of long knots in codimension > 2, submitted,arXiv:math.AT/0703649.

[22] Brian A. Munson, Deriviatives of the identity and generalizationsof Milnor’s invariants, submitted, arXiv:0909.3074.

[23] , Embeddings in the 3/4 range, Topology 44 (2005), no. 6,1133–1157.

[24] , A manifold calculus approach to link maps and the linkingnumber, Algebr. Geom. Topol. 8 (2008), no. 4, 2323–2353.

[25] Brian A. Munson and Ismar Volic, Cosimplicial models for spacesof links, arXiv:0906.2589[math.AT].

[26] , Multivariable manifold calculus of functors, arXiv:0904.4185[math.AT].

[27] Kevin P. Scannell and Dev P. Sinha, A one-dimensional embeddingcomplex, J. Pure Appl. Algebra 170 (2002), no. 1, 93–107.

[28] Dev P. Sinha, Operads and knot spaces, J. Amer. Math. Soc. 19(2006), no. 2, 461–486 (electronic).

[29] , The topology of spaces of knots, Amer. J. Math. 131(2009), no. 4, 945–980.

[30] Rainer M. Vogt, Commuting homotopy limits, Math. Z. 153 (1977),no. 1, 59–82.

[31] Ismar Volic, Configuration space integrals and Taylor towers forspaces of knots, Topology Appl. 153 (2006), no. 15, 2893–2904.

[32] , Finite type knot invariants and the calculus of functors,Compos. Math. 142 (2006), no. 1, 222–250.

[33] Michael Weiss, Calculus of embeddings, Bull. Amer. Math. Soc.(N.S.) 33 (1996), no. 2, 177–187.

[34] , Embeddings from the point of view of immersion theory I,Geom. Topol. 3 (1999), 67–101 (electronic).

Morfismos, Vol. 14, No. 1, 2010, pp. 51–68

Impacto de la polıtica fiscal en unambiente con inflacion estocastica:

un modelo de control optimo

Francisco Ortiz-Arango Francisco Venegas-MartınezClaudia Estrella Castillo-Ramırez

ResumenelbixefloibmacedopitnocatreibayaneuqepaımonoceanunEaledotcapmileaulaveesacitsacotsesenoicaflnialednodney

polıtica fiscal sobre los contribuyentes. La funcion de utilidad totalesperada se aplica solo a individuos, adversos al riesgo, que concierta probabilidad (positiva) estan vivos en el momento en quepresentan su declaracion fiscal. Se supone que los contribuyentesson sujetos a tasas impositivas sobre la riqueza, el ingreso y elconsumo.

Abstract

In a small open economy with flexible exchange rate and whereinflation is stochastic, we assess the impact of fiscal policy ontaxpayers. The total expected utility function applies only torisk-averse individuals who, with a certain (positive) probability,are alive when submitting their fiscal statement. It is assumedthat taxpayers are subject to tax rates on wealth, income andconsumption.

2010 Mathematics Subject Classification: 91B70, 93E20.Keywords and phrases: Stochastic optimal control, mathematical eco-nomics.

1 Introduccion

Sin duda, muchos esfuerzos ha hecho la autoridad fiscal federal enMexico para favorecer la regularizacion de los contribuyentes, por ejem-

:seroiretnaselacsfisonaedsenoicaralcedarapodanodnocnahes,olp

51

52 F. Ortiz-Arango, F. Venegas-Martınez, y C. E. Castillo-Ramırez

creditos fiscales derivados de contribuciones, cuotas compensatorias ysus accesorios, ası como multas por incumplimiento de obligaciones fis-cales distintas de pago. Tambien se ha incrementado el monto por elcual los asalariados no estan obligados a presentar declaracion anual yse han otorgado estımulos a diversas industrias. No obstante, falta elproposito de alentar la competitividad, reducir la dependencia de los in-gresos petroleros, fomentar con estımulos fiscales la inversion extranjeradirecta, eliminar la elusion de los grandes corporativos con la consoli-dacion de sus resultados (a diferencia de la evasion en donde de maneraconciente no se pagan tributaciones, en la elusion se buscan impreci-siones, huecos o fallas en la legislacion fiscal que permitan disminuirel monto de las contribuciones), incorporar a la economıa informal alpago de contribuciones sin que con ello se reconozca a los ilegales, in-crementar, en la medida de lo posible, las potestades tributarias de losestados y los municipios, y poner mas atencion en la redistribucion dela riqueza.

Los regımenes tributarios en Latinoamerica, tıpicamente, han puestoenfasis en la seleccion de los niveles de tasas impositivas al ingreso(renta) y al consumo. Ademas, en paıses de America Latina comoVenezuela, Brasil, Mexico, Colombia, Ecuador y Peru, y de MedioOriente como Arabia Saudita, Iraq, Iran, Kuwait, en donde la ventade petroleo y gas representa una proporcion importante de los ingresosfiscales de sus gobiernos, las reformas hacendarias se han dirigido a re-ducir dicha dependencia petrolera. Es tambien importante destacar queen varias de estas economıas no ha sido posible instrumentar reformastributarias dirigidas a incentivar (con estımulos fiscales al sector empre-sarial domestico) el crecimiento de la demanda interna a fin de reducircon ello el efecto de choques externos de corto y largo plazo.

Un tema actual en el desempeno de la economıa mexicana es elimpacto fiscal sobre diversas variables fundamentales, particularmentesobre el consumo y la inversion en ambientes de riesgo e incertidum-bre. La literatura que relaciona el tema de incertidumbre con polıticaeconomica ha marcado una tendencia creciente en las ultimas decadas.Por ejemplo, se destacan los trabajos de: Brennan y McGuire (1975);Giovannini (1988); Alesina y Tabellini (1989); Elder (1999); Venegas-Martınez (2008), (2006), (2001), (2000a) y (2000b); y Venegas-Martınezy Gonzalez-Arechiga (2000).

El modelo propuesto en la presente investigacion supone que losagentes perciben una tasa impositiva incierta sobre la riqueza, la cuales conducida por un movimiento geometrico Browniano. Tambien, en

Impacto de la polıtica fiscal en un ambiente con inflacion estocastica 53

el modelo se consideran impuestos sobre la renta y el consumo de losagentes; el consumo es gravado mediante una tasa ad valorem. Lafuncion de utilidad total esperada se aplica solo a individuos, adver-sos al riesgo, que con cierta probabilidad (positiva) estan vivos en elmomento en que presentan su declaracion fiscal.

Se supone ademas que los individuos que pueblan esta economıatienen expectativas de depreciacion gobernadas por un proceso combi-nado de difusion con saltos. En este contexto, los pequenos movimientosdel tipo de cambio, que estan siempre presentes, se modelan a travesde un movimiento Browniano, y una depreciacion extrema y repentina(un salto en el tipo de cambio), que ocasionalmente ocurre, se modelamediante un proceso de Poisson. La mezcla de un movimiento Brown-iano con un proceso de saltos proporciona una distribucion con excesode curtosis, colas pesadas y sesgo para el tipo de cambio, lo que permiteproducir dinamicas mas realistas en el tipo de cambio que no puedenser generadas utilizando unicamente el movimiento Browniano. Estehecho no solo es una sofisticacion teorica, sino un aspecto relevante queincorpora mayor realismo en el modelado del comportamiento del tipode cambio.

En el modelo que se desarrolla en el presente trabajo, bajo el su-puesto de agentes adversos al riesgo, se examina la dinamica de equi-librio del consumo y la riqueza cuando la polıtica fiscal es incierta. Eneste contexto, tambien se discuten varios temas especıficos de polıticaeconomica. Por ejemplo, se estudian los efectos sobre el consumo yel bienestar economico de cambios permanentes en los parametros quedeterminan las expectativas de la polıtica fiscal. Con respecto a es-tudios sobre los efectos de la polıtica fiscal en el bienestar economicoen ambientes estocasticos es importante mencionar los trabajos de Ag-ell, Persson y Sacklen (2004) y Amilon y Bemin (2003). Varios de losresultados obtenidos en esta investigacion proporcionan elementos quedeben ser incorporados en la generacion de recomendaciones en materiade polıtica fiscal a fin de elaborar una reforma tributaria integral.

La organizacion de esta investigacion es como sigue. En la seccion2 se introduce la dinamica estocastica del nivel general de precios contipo de cambio flexible. En el transcurso de la seccion 3 se enlistan losactivos disponibles y sus rendimientos. Durante la seccion 4 se define elimpuesto sobre la riqueza. En la seccion 5 se introducen los impuestosISR e IVA. Por su parte, la seccion 6 define una restriccion del tipocash-in-advance (el dinero se utiliza para financiar el consumo). Enla seccion 7 se establece el problema de decision (control optimo) del


consumidor representativo. En la seccion 8 se obtienen las condicionesnecesarias de optimo. En la seccion 9 se resuelven las condiciones deprimer orden (CPO). A traves de la seccion 10 se llevan a cabo diversosejercicios de estatica comparativa. En la seccion 11 se examinan losimpactos de los impuestos sobre el bienestar economico. En la seccion12 se estudia la dinamica de la riqueza y consumo bajo el esquemapropuesto de contribuciones. Por ultimo, en la seccion 13, se presentanlas conclusiones.

2 Nivel de precios con tipo de cambio flexible

En esta seccion se desarrolla un modelo estocastico de una economıapequena y abierta poblada con agentes identicos (en gustos y dota-ciones) de vida infinita. La economıa produce y consume un solo bienperecedero. Se supone que el bien es comerciable internacionalmente,sin barreras arancelarias, y el nivel general de precios domesticos, Pt, esdeterminado por la condicion de poder de paridad de compra, a saber,

Pt

P !t

= et,

donde P !t es el precio en moneda extranjera del bien en el resto del

mundo, y et es el tipo de cambio nominal. Se supone, por simplicidad,que P !

t es igual a 1. Tambien, se supone que el valor inicial del tipo decambio, e0, es conocido e igual a 1. Como siempre, el nivel general deprecios es el promedio ponderado de los precios de una muestra de losbienes y servicios que se producen en una economıa, la ponderacion tomaen cuenta la importancia relativa que las unidades familiares asignan algasto. En el caso de Mexico, el nivel general de precios es el Indice Na-cional de Precios al Consumidor (INPC), el cual genera periodicamente(quincenalmente) el INEGI como la relacion entre el promedio pon-derado actual de precios y el promedio ponderado de los precios de unano base (2002).

Asimismo, se supone que el numero de saltos, movimientos extremosy repentinos, en el tipo de cambio, por unidad de tiempo, siguen unproceso de Poisson qt con intensidad !, de tal manera que

IP(q){un salto unitario durante dt} = IP

(q){dqt = 1} = !dt (1)

y

IP(q){mas de un salto durante dt} = IP

(q){dqt > 1} = o(dt).


De esta manera,

IP(q){ningun salto durantedt} = IP

(q){dqt = 0} = 1 ! !dt ! o(dt), (2)

donde o(dt)/dt " 0 cuando dt " 0. Ası, E(q) [dqt] = Var(q)

[dqt] = !dt.El numero inicial de saltos se supone igual a cero, es decir, q0 = 0.

Se supone ahora que el consumidor percibe que la tasa de inflacionesperada, dPt/Pt, y por lo tanto la tasa esperada de depreciacion, det/et,sigue un movimiento geometrico Browniano con saltos de Poisson des-crito por

dPt

Pt=

det

et= µdt + "P dzt + #dqt, (3)

donde µ es la tasa media esperada de depreciacion (o inflacion) condi-cionada a que no se presenten saltos, "P es la volatilidad instantaneadel nivel general de precios, y # es el tamano medio esperado de unsalto en el tipo de cambio. El proceso (de Wiener) dzt es normal conmedia E(z) [dzt] = 0 y varianza Var

(z)[dzt] = dt. Asimismo, se supone

que dzt es (estocasticamente) independente de dqt. En lo que se sigue,µ, "P , ! y # son constantes positivas. La presencia de saltos pemitiramodelar la existencia de movientos bruscos e inesperados en el nivelgeneral de precios; situacion que esta mas alla de la distribucion nor-mal, i.e., mas alla del supuesto de un proceso de difusion browniana.Es importante destacar que las difusiones con saltos son modelos muycomunes en aplicaciones financieras en teorıa del control y juegos; comose puede ver en Oksendal y Sulem (2007) y Mataramvura y Oksendal(2008). Asimismo, se debe mencionar que en lugar de un proceso dedifusion con saltos, serıa interesante utilizar un proceso de difusion concoeficientes que pueden cambiar aleatoriamente, µ(yt) y "(yt), donde yt

es una cadena de Markov (finita) cuyos estados representan los distintoscambios en una economıa o un mercado financiero; veanse, por ejemplo,Yin y Zhu (2010) y Escobedo-Trujillo (2007).

3 Dinero, bonos y sus rendimientos

Se supone que en esta economıa, el agente mantiene saldos monetariosreales,

mt = Mt/Pt,

donde Mt es el acervo nominal de dinero. La tasa de retorno estocasticapor la tenencia de saldos reales, dRm, esta dada por el cambio porcentual


en el precio del dinero en terminos de bienes. Al aplicar el lema deIto para procesos de difusion con saltos al inverso del nivel de precios,con (3) como el proceso subyacente, se obtiene (vease el lema de Itoextendido a saltos, por ejemplo, en Venegas-Martınez (2001) o (2008))

dRm =dmt

mt= (!µ + !2

P)dt ! !P dzt!

! "

1 + "

"dqt. (4)

Por simplicidad se supone que el agente solo tiene acceso a un bonogubernamental domestico, bt, que paga una tasa de interes real libre deriesgo, r, constante para todos los plazos. En este caso, se satisface

dbt = rbtdt, (5)

donde b0 es dado. Ası, el bono paga r unidades del bien de consumopor unidad de tiempo. Los agentes toman r como dada. La ecuacion(5) se puede interpretar como una cuenta bancaria, en la que se realizaun deposito inicial con valor b0 al tiempo cero, y que gana a una tasainstantanea libre de riesgo, r, en cada instante t.

4 Impuestos sobre la riqueza

Se supone que la riqueza del consumidor representativo es objeto delpago de un impuesto. Se supone que este es gravado a una tasa, #t, deacuerdo con la ecuacion diferencial estocastica siguiente:

d#t

#t= #dt + !!d#zt, #0 > 0, (6)

cond#zt = $dzt +

$1 ! $2dut, dut " N (0, dt), (7)

yCov

%dzt,d

%$zt +

$1 ! $2ut

&&= $dt, (8)

donde # es la tasa media esperada de crecimiento del impuesto sobrela riqueza, !! es la volatilidad de la tasa impositiva en la riqueza, y$ # [!1, 1] es la correlacion entre los cambios en la inflacion y los cambiosen los impuestos sobre la riqueza. Observe que un incremento en el tipode cambio deprecia los saldos monetarios reales. Esto, a su vez, reduceel valor real de los activos, situacion que puede llevar a la autoridadfiscal a modificar su polıtica fiscal. Los procesos qt, zt, y ut se suponenindependientes por pares.


5 Dinero para financiar consumo

Considere ahora una restriccion del tipo cash-in-advance de la forma:

mt = !ct, (9)

donde ct es el consumo y ! > 0 es el tiempo que se mantiene el dineropara financiar el consumo. De esta forma, la depreciacion en el tipo decambio actua como un impuesto estocastico en los saldos monetariosreales.

6 Riqueza, ISR e IVA

En esta seccion se introducen los impuestos directos (ISR) e indirectos(IVA) y se caracterizan las decisiones optimas de consumo y portafoliode un agente representativo. La acumulacion de la riqueza, at, delconsumidor representativo en terminos de las decisiones de portafolio,wt = mt/at, 1!wt = bt/at, y de consumo, ct, esta dada por el siguientesistema de ecuaciones diferenciales estocasticas:

dat = atwtdRm + at(1 ! wt)dRb ! ("tat + (1 + !")ct) dt + (1 ! "")ydt,

d"t = " "tdt + #!"t#$dzt +

$1 ! $2dut

%, "0 > 0, (10)

donde dRb = dbt/bt, !" es una tasa impositiva ad valorem (al valoragregado) del consumo, y es un flujo constante de ingreso (en terminosreales), "" es el impuesto sobre la renta y a0 = m0 + b0 > 0. Usualmenteel porcentaje que aportan los individuos con mayor riqueza inicial, a0,en la recaudacion total del impuesto al valor agregado es mayor que elde los de menor riqueza. Lo mismo sucede con el impuesto sobre larenta.

Si se sustituyen las ecuaciones (4), (5) y (9) en la primera ecuaciondel sistema (10), se tiene que

dat = at

&[r!!wt!"t+(1!"")y] dt ! wt#P dzt ! wt

' %

1 + %

(dqt

), (11)

donde! = (1 + !")!!1 + r + µ ! #2

P.

La expresion (11) es la ecuacion diferencial estocastica que conducela riqueza del individuo despues de incorporar los rendimientos de losdiferentes activos e impuestos.


7 Problema del agente representativo

La funcion de utilidad del tipo von Neumann-Morgenstern al tiempo0, u0, de un agente representativo, competitivo (precio aceptante) yadverso al riesgo esta dada por:

u0 = E!" !

0u(ct) e"rtf(t)dt

### F0

$. (12)

donde F0 es la informacion disponible al tiempo t = 0, la cual esen-cialmente contempla los valores iniciales a0 y !0. Observe que la tasasubjetiva de descuento del agente ha sido igualada a la tasa de interes,r, a fin evitar dificultades tecnicas innecesarias en la dinamica de equi-librio. La funcion de densidad de que el individuo este vivo al tiempot se supone de la forma exponencial, es decir, f(t) = "e"!t donde 1/"representa el numero promedio de individuos vivos al tiempo t. De estamanera la funcion de utilidad total esperada se aplica solo a individuos,adversos al riesgo, que con cierta probabilidad estan vivos en el momentoen que presentan su declaracion fiscal. Se empleara la funcion de utili-dad logarıtmica, u(ct) = log(ct), con el proposito de generar solucionesanalıticas que hagan mas simple el analisis posterior.

8 Condiciones necesarias de optimo

La ecuacion de Hamilton-Jacobi-Bellman para el problema de progra-macion dinamica estocastica en tiempo continuo en el que se maximizala utilidad esperada del agente, sujeto a su restriccion presupuestal in-tertemporal, es (veanse, al respecto, Venegas-Martınez (2008) o Ok-sendal y Sulem (2007)):

#I(at, !t, t) ! It(at, !t, t) ! I" (at, !t, t)! !t ! 12I"" (at, !t, t)!2

t $2"

!Ia(at, !t, t)at [r ! !t + (1 ! %!)y]

= maxw

&log(%"1atwt)e"(r+!)t!Ia(at, !t, t)at!wt+1

2Iaa(at, !t, t)a2t w

2t $

2P

!Ia" (at, !t, t)at!twt$P $"& + #I

'at

'1 + '(1 ! wt)

1 + '

(, !t, t

(), (13)

donde los subındices en I, Ia, Iaa, etc., significan derivadas parciales.La funcion de valor

I(at, !t, t) = maxw

Et

&" !

tlog

*%"1asws

+e"(r+!)sds

### at, !t

)


es la funcion de utilidad indirecta (o funcion de bienestar economico)del consumidor, e Ia(at, !t, t) es la variable de coestado. Dado el factorde descuento exponencial en la utilidad indirecta, es conveniente definira I(at, !t, t) en forma separable como

I(at, !t, t) ! F (at, !t)e!(r+!)t. (14)

Por lo tanto, la ecuacion (14) se transforma en

(" + r + #)F (at, !t) " F" (at, !t)! !t " 12F"" (at, !t)!2

t $2"

"Fa(at, !t)at [r " !t + (1 " !!)y]

= maxw

"log(%!1atwt) " Fa(at, !t)at!wt + 1

2Faa(at, !t)a2t w

2t $

2P

"Fa" (at, !t)at!twt$P $"& + "F#at

#1 + '(1 " wt)1 + '

$, !t

$%. (15)

La ecuacion anterior representa una condicion necesaria del problemade decision del consumidor y sera resuelta a continuacion.

9 Solucion de la CPO

Se postula como posible candidato de solucion de (15) a

F (at, !t) = (0 + (1 log&

at

!t

'+ H(!t; (2, (3), (16)

donde (0, (1 y H(!t; (2, (3) se tienen que determinar a partir de laecuacion (15). Las constantes (2 y (3 se determinan de tal maneraque H(!0) = 0 y H "(!0) = 0. Al sustituir la ecuacion (16) en (15), seobtiene

(r + #) ((0 + (1 log (at)) + (1(! " r " (1 " !!)y " 1

2$2"

)

+(r + #)H(!t) " H "(!t)!t! " 12H ""(!t)!2

t $2" " (r + #)(1 log(!t) + (1!t

= maxw

"log(%!1atwt) " (1!wt " 1

2(1w2t $

2P

+ "(1 log#1 + '(1 " wt)

1 + '

$%.

(17)Las condiciones de primer orden del problema de optimizacion intertem-poral del agente representativo conducen a una proporcion de riqueza


asignada a la tenencia de saldos reales invariante en el tiempo, wt ! w,ası como a la relacion

1!1w

" "#

1 + #(1 " w)= (1 + !$)%!1 + r + µ " &2

P+ w&2

P. (18)

Ahora se tiene que determinar H($t) como solucion de la ecuacion dife-rencial ordinaria de segundo orden

(r+')H($t)"H "($t)$t$" 12H ""($t)$2

t &2!"(r+')!1 log($t)+!1$t = 0. (19)

Los coeficientes !0 y !1 son determinados de (15) despues de sustituir elvalor optimo w#. Ası,

!1 = (r + ')!1,

lo que produce que el coeficiente de log(at) en la ecuacion (17) sea cero,y

!0 =1

r + 'log(%!1w#)" 1

(r + ')2

"((1 + !$)%!1 + r + µ " &2

P)w#

+ 12(w#&P )2 + $ " r " (1" #$)y " 1

2&2! "" log

$1 + #(1 " w#)1 + #

%&. (20)

El supuesto de utilidad logarıtmica conduce a que w dependa solamentede los parametros que determinan las caracterısticas estocasticas de laeconomıa, y por lo tanto w es constante. Es decir, la actitud del con-sumidor hacia el riesgo cambiario es independiente de su riqueza, i.e., elnivel de riqueza resultante en cualquier instante no tiene relevancia paralas decisiones de portafolio. Mas aun, debido a la utilidad logarıtmica,el coeficiente de correlacion, ( # ("1, 1), no juega papel alguno en lasdecisiones del consumidor. Por ultimo, es importante senalar que laecuacion (18) es cubica, por lo que tiene al menos una raız real.

Se puede demostrar que la solucion de la ecuacion (19) satisface

H($t) = !2$"1t + !3$

"2t +

1$

log($t)'1+

2(&2

! + 2$)$t

(+

1$

'1"&2

!

2$

(, (21)

donde)1 =

4(r + ')(2$ " &2

! ) +)

(2$ " &2! )2 + 8(r + ')&2

!

(22)

y

)2 =4(r + ')

(2$ " &2! ) "

)(2$ " &2

! )2 + 8(r + ')&2!

. (23)


Los coeficientes !2 y !3 se determinan de tal manera que H("0) = 0 yH !("0) = 0. La primera condicion inicial, H("0) = 0, asegura que elbienestar economico,

W ! I(a0, "0, 0) = F (a0, "0) = !0 +1

r + #log

!a0

"0

", (24)

sea independiente de la seleccion de H. La segunda condicion inicial,H !("0) = 0, garantiza que la funcion de bienestar sea decreciente conrespecto del impuesto a la riqueza, esto es,

$I

$"

###!=!0

= " 1(r + #)"0

< 0, (25)

y tambien asegura que H sea la unica solucion de la ecuacion (19).La ecuacion (18) es cubica con una raız negativa y dos raıces positi-

vas. Esto puede verse si se intersecta la lınea recta definida por el ladoderecho de la ecuacion (18) con la grafica definida por el lado izquierdode (18). En este caso, hay solamente una interseccion que proporcionaun estado estacionario (unico) de la riqueza que el consumidor asigna ala tenencia de saldos reales w" # (0, 1), lo que elimina la posibilidad deventas en corto

10 Experimentos de estatica comparativa

En esta seccion se obtienen los primeros resultados relevantes del mo-delo propuesto. Un aumento permanente en el impuesto ad valorem alconsumo producira una reduccion permanente en la proporcion de lariqueza asignada al consumo futuro, ya que

$w"

$$"= " 1

%!< 0, (26)

donde! =

%r

(w")2+

&'2

[1 + '(1 " w")]2+ (2

P

&.

11 Impactos sobre el bienestar economico

A continuacion se evaluan los impactos de choques exogenos en el bienes-tar economico. Como siempre, el criterio de bienestar, W , del individuorepresentativo es la utilidad indirecta con una riqueza real inicial, a0, y


una tasa impositiva inicial de la riqueza, !0. Por lo tanto, en virtud delas ecuaciones (14), (20) y (24) el bienestar esta definido por:

W (µ,", #, y, ! , !! , "! ; a0, !0) ! I(a0, !0, 0)

=1

r + $[1 + log(a0/!0) + log(%!1w")]

" 1(r + $)2

#((1+ !!)%!1 + r +µ"&2

P)w" + 1

2(w"&P )2 + ! " (1""!)y" 12&2

!

"" log$1 + #(1 " w")

1 + #

%&, (27)

donde se ha utilizado los siguientes resultados:

H(!0) = 0, (28)

I(a0, !0, 0) = F (a0, !0) (29)

y

F (a0, !0) = '0 +1

r + $log

'a0

!0

(. (30)

A continuacion se calculan los impactos en el bienestar economico pro-ducidos por cambios permanentes en la tasa impositiva media esperadaa la riqueza, el impuesto esperado ad valorem al consumo, y el impuestosobre la renta. En este caso, se tiene

(W

(!= " 1

(r + $)2< 0, (31)

(W

(!!= " 1

(r + $)2%!1w" < 0 (32)

y(W

("!= " 1

(r + $)2y < 0.

Por lo tanto, aumentos en la tasa impositiva media esperada sobre lariqueza, la tasa impositiva en el consumo y en el impuesto sobre la rentaconducen a una reduccion en el bienestar economico.


12 Dinamica de la riqueza y consumo

Ahora se obtiene el proceso estocastico que genera la riqueza real delconsumidor cuando se aplica la regla optima. Despues de sustituir w!

en la ecuacion (11), se obtiene

dat = at

!" !"w!

1 + "(1 ! w!)+ (w!#P )2 ! $t + (1 ! #$)y

$dt

!w!#P dzt+"1 + "(1 ! w!)

1 + "! 1

$dqt

%, (33)

donde$t = $0 exp

&($ ! 1

2#2! )t + E#

"t'

, (34)

y E # N (0, 1). La funcion densidad de probabilidad de $t, dado $0,satisface

f!t|!0

(x|$0) =1"

2µt#!xexp

()

*!12

+log (x/$0) ! ($ ! 1

2#2! )t

#!"

t

,2-.

/ . (35)

Ademas, se tieneE[$t|$0] = $0e

!t (36)

yVar[$t|$0] = $2

0 e2!t0e"2

! t ! 11. (37)

La solucion a la ecuacion diferencial estocastica (33), condicionada pora0, es

at = a0e#t , (38)

donde

%t = &t + 't, &t|$t # N [[F (w!) ! $t + (1 ! #$)y]t,G(w!)t],

't = L(w!)qt,

yqt # P(!t).

Es decir, qt es un proceso de Poisson con intensidad !. Los componentesestacionarios de los parametros de las distribuciones antes mencionadasson:

F (w!) =!"w!

1 + "(1 ! w!)+

(w!#)2

2,


G(w!) = (w!!)2,

y

L(w!) = log!1 + "(1 ! w!)

1 + "

".

Asimismo, observe que

E[#t|$t] = [F (w!) ! $t + (1 ! #$)y + L(w!)%]t

yVar[#t|$t] = [G(w!) + [L(w!)]2%]t.

Mas aun, se sigue que

E[#t] = E{E[#t|$t]} = [F (w!) ! $0e! t + (1 ! #$)y + L(w!)%]t, (39)

yVar[#t] = Var{E[#t|$t]} + E{Var[#t|$t]}

= t2$20 e2!t

$e"2

t t ! 1%

+ [G(w!) + [L(w!)]2%]t. (40)

Estas dos ultimas ecuaciones, de acuerdo con (38), determinan la mediay la varianza de la velocidad a la que crece la riqueza real del individuo.

En virtud de las ecuaciones (9) y (38), el proceso estocastico para elconsumo se puede escribir como

c!t = &"1w!a0e#t . (41)

Esto indica que, en ausencia de mercados de productos derivados fi-nancieros, el riesgo de depreciacion tiene un efecto en la riqueza a travesde la incertidumbre en #t, es decir, la incertidumbre cambia el conjuntode oportunidades que enfrenta el consumidor. Por otra parte, el riesgode depreciacion tambien afecta la composicion del portafolio por mediode sus efectos en w!. De este modo, un cambio en la polıtica economicaestara acompanado tanto del efecto riqueza como del de sustitucion.

13 Conclusiones

En muchas economıas, la ausencia, desde hace muchas decadas, de re-formas fiscales a fondo ha acumulado una carga fiscal sin precedentessobre los contribuyentes. Esta investigacion desarrollo, en un ambientede riesgo e incertidumbre, un modelo que permite analizar el impactofiscal sobre las decisiones de consumo y portafolio. Se considero una


economıa en donde los agentes son sujetos a tasas impositivas sobre lariqueza, el ingreso y el consumo.

Las concordancia que las variables fundamentales deben guardar conlos niveles de riesgo en el equilibrio, son relaciones fragiles que requierende un manejo responsable en el diseno e implementacion de una reformafiscal. En este sentido, la incertidumbre puede conducir a cambios cuan-titativos y cualitativos significativos con respecto al analisis simplista delmarco determinista.

La mayor parte de la investigacion documentada sobre los efectosde la polıtica fiscal en el desempeno de las economıas ignora la incer-tidumbre, proporcionando justificaciones elaboradas para menospreciarla inclusion de factores de riesgo. Sin embargo, como se ha demostradoen esta investigacion, la consideracion de incertidumbre en vısperas deuna reforma fiscal permite analizar dinamicas transicionales mas realis-tas. Varios de los resultados obtenidos en el transcurso de este tra-bajo proporcionan elementos que pueden ser incorporados en diversasrecomendaciones en materia de polıtica fiscal a fin de elaborar una re-forma tributaria integral.

El modelo desarrollado supone que los agentes perciben incertidum-bre en la polıtica fiscal. En en analisis se consideraron impuestos sobrela riqueza, la renta y el consumo. Se supuso ademas que las expecta-tivas de depreciacion son conducidas por un proceso estocastico. Bajoeste ambiente de riesgo e incertidumbre, se examinaron las decisionesde consumo e inversion de un agente representativo. Se ha mostradoque el agente asigna proporciones constantes de su riqueza a los dife-rentes activos disponibles en la economıa, a fin de transferir consumohacia el futuro. Dichas proporciones constantes dependen solamentede los parametros que determinan las caracterısticas estocasticas de laeconomıa. Ası, la actitud del consumidor hacia el riesgo cambiario esindependiente del nivel de riqueza en cualquier instante.

Ası mismo se evaluaron los impactos de choques exogenos de va-riables fundamentales en el bienestar economico, incluyendo cambiospermanentes en las diferentes impuestos. Entre los resultados se destacaque un aumento en cualquiera de los impuestos considerados conduce auna reduccion en el bienestar economico (utilidad indirecta) cuando eldestino del gasto no desempena papel alguno en la funcion de utilidad(directa) o en la restriccion presupuestal de las unidades familiares. Porultimo, se ha discutido el caso cuando el bienestar depende del gasto,mostrando como se modifican los resultados obtenidos anteriormente.

Nuestro paıs cuenta con diversas disposiciones fiscales y es nece-


sario modificarlas y adecuarlas (incluso eliminarlas) a fin de impulsarel crecimiento y la competitividad, reducir la dependencia de los ingre-sos petroleros, eliminar la elusion de corporativos con la consolidacionde resultados financieros, incorporar a la economıa informal al pago decontribuciones (el impuesto a los depositos en efectivo es muy limitado),incrementar las potestades tributarias de los estados y los municipios,y poner mas atencion en la redistribucion de la riqueza con desarrollosocial.

AgradecimientosLos autores desean agradecer los valiosos comentarios y multiples

sugerencias de dos dictaminadores anonimos. Por supuesto, las opi-niones que se vertieron y los errores que persistieron son responsabilidadexclusiva de los autores.

Francisco Ortiz-ArangoEscuela deCiencias Economicas y EmpresarialesUniversidad PanamericanaAugusto Rodin 498Col. Insurgentes MixcoacDel. Benito JuarezMexico, D. F. [email protected]

Francisco Venegas-MartınezEscuela Superior de EconomıaInstituto Politecnico NacionalPlan de Agua Prieta, No. 66Col. Plutarco Elıas CallesDel. Miguel HidalgoMexico, D. F. [email protected]

Claudia Estrella Castillo-RamırezDepartamento de SistemasUniversidad Autonoma Metropolitana-AzcapotzalcoAv. San Pablo No. 180Col. Reynosa TamaulipasDel. AzcapotzalcoMexico, D. F. [email protected]

Referencias

[1] Agell, J. M. Persson, and H. Sacklen (2004).“The e!ects of tax re-form on labor supply, tax revenue and welfare when tax avoidancematters”. European Journal of of Political Economy, Vol. 20, pp.963-982.

[2] Alesina, A. and G. Tabellini (1989). “External debt, capital flightand political risk”, Journal of International Economics, Vol. 27,


pp. 199-220.

[3] Amilon, H. and H. P. Bemin (2003). “Welfare e!ects of control-ling labor suply: an application of the stochastic Ramsey Model.Journal of Economic Dynamics and Control, Vol. 28, pp. 331-348.

[4] Barro, R. J. (1990). “Government spending in a simple model ofendogenous growth”. Journal of Political Economy, Vol. 98, pp.S103-S125.

[5] Brennan, G. and T. McGuire (1975). “Optimal policy choice underuncertainty”. Journal of Public Economics, Vol. 4, pp. 205-209.

[6] Elder, E. (1999). “Dynamic fiscal policy with regime-duration un-certainty: The Tax-Cut Case”. Journal of Macroeconomics, Vol.21, pp. 29-55.

[7] Giovannini, A. (1988). “The real exchange rate, the capital stock,and fiscal policy”, European Economic Review, Vol. 32, pp. 1747-1767.

[8] Mataramvura, S. and B. Oksendal (2008). “Risk minimizingportfolios and HJBI equations for stochastic di!erential games”.Stochastics, Vol. 80, pp. 317-337.

[9] Oksendal, B. and A. Sulem (2007). Applied Stochastic Control ofJump Di!usions. Springer, Berlin.

[10] Venegas-Martınez, F. (2001). “Temporary stabilization: a stochas-tic analysis”. Journal of Economic Dynamics and Control, Vol. 25,pp. 1429-1449.

[11] Venegas-Martınez, F. (2000a). “On consumption, investment, andrisk”. Economıa Mexicana, Nueva Epoca, Vol. 9, pp. 227-244.

[12] Venegas-Martınez, F. (2000b). “Utilidad, aprendizaje y estabi-lizacion,”. Gaceta de Economıa, Vol. 10, pp. 153-169.

[13] Venegas-Martınez, F. (2006). “Stochastic temporary stabilization:undiversifiable devaluation and income risks”. Economic Mod-elling, Vol. 23, pp. 157-173.


[14] Venegas-Martınez, F. (2008). Riesgos Financieros y Economicos(Productos Derivados y Decisiones Economicas bajo Incertidum-bre). 2a. edicion. Cengage Learning (anteriormente InternationalThomson Editors).

[15] Venegas-Martınez, F. y B. Gonzalez-Arechiga (2000). “Mercadosfinancieros incompletos y su impacto en los programas de estabi-lizacion de precios: El caso mexicano”. Momento Economico, Vol.111, pp. 20-27.

[16] Venegas-Martınez, F y F. Ortiz-Arango (2010). Evaluacion delimpacto fiscal en las decisiones de consumo y portafolio: un en-foque estocastico. En Avances Recientes en Valuacion de Activos yAdministracion de Riesgos, Vol. 1, Francisco Ortiz Arango Coordi-nador. Universidad Panamericana, UP.

Apoyo tecnico: Omar Hernandez Orozco.

Morfismos, Comunicaciones Estudiantiles del Departamento de Matematicas delCINVESTAV, se imprime en el taller de reproduccion del mismo departamento locali-zado en Avenida Instituto Politecnico Nacional 2508, Colonia San Pedro Zacatenco,

edotsogaedsemlenerimirpmiedonimretesoremunetsE.F.D,ocixeM,06370.P.C2010. El tiraje en papel opalina importada de 36 kilogramos de 34 ! 25.5 cm constade 500 ejemplares con pasta tintoreto color verde.

Contenido

Introduction to the manifold calculus of Goodwillie-Weiss

Brian A. Munson . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

Impacto de la polıtica fiscal en un ambiente con inflacion estocastica: un mo-delo de control optimo

F. Ortiz-Arango, F. Venegas-Martınez, y C. E. Castillo-Ramırez . . . . . . . . 51

Documents

Morfismos, Vol 14, No 1, 2010