%% PNAStmpl.tex

%% Template file to use for PNAS articles prepared in LaTeX

%% Version: Apr 14, 2008

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%% BASIC CLASS FILE 

%% PNAStwo for two column articles is called by default.

%% Uncomment PNASone for single column articles. One column class

%% and style files are available upon request from pnas@nas.edu.

%% (uncomment means get rid of the '%' in front of the command)

%\documentclass{pnasone}

\documentclass{pnastwo}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%% Changing position of text on physical page:

%% Since not all printers position

%% the printed page in the same place on the physical page,

%% you can change the position yourself here, if you need to:

% \advance\voffset -.5in % Minus dimension will raise the printed page on the 

                         %  physical page; positive dimension will lower it.

%% You may set the dimension to the size that you need.

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%% OPTIONAL GRAPHICS STYLE FILE

%% Requires graphics style file (graphicx.sty), used for inserting

%% .eps files into LaTeX articles.

%% Note that inclusion of .eps files is for your reference only;

%% when submitting to PNAS please submit figures separately.

%% Type into the square brackets the name of the driver program 

%% that you are using. If you don't know, try dvips, which is the

%% most common PC driver, or textures for the Mac. These are the options:

% [dvips], [xdvi], [dvipdf], [dvipdfm], [dvipdfmx], [pdftex], [dvipsone],

% [dviwindo], [emtex], [dviwin], [pctexps], [pctexwin], [pctexhp], [pctex32],

% [truetex], [tcidvi], [vtex], [oztex], [textures], [xetex]

%\usepackage[dvips]{graphicx}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%% OPTIONAL POSTSCRIPT FONT FILES

%% PostScript font files: You may need to edit the PNASoneF.sty

%% or PNAStwoF.sty file to make the font names match those on your system. 

%% Alternatively, you can leave the font style file commands commented out

%% and typeset your article using the default Computer Modern 

%% fonts (recommended). If accepted, your article will be typeset

%% at PNAS using PostScript fonts.

% Choose PNASoneF for one column; PNAStwoF for two column:

%\usepackage{PNASoneF}

%\usepackage{PNAStwoF}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%% ADDITIONAL OPTIONAL STYLE FILES

%% The AMS math files are commonly used to gain access to useful features

%% like extended math fonts and math commands.

\usepackage{amssymb,amsfonts,amsmath,amsthm}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%% OPTIONAL MACRO FILES

%% Insert self-defined macros here.

%% \newcommand definitions are recommended; \def definitions are supported

%\newcommand{\mfrac}[2]{\frac{\displaystyle #1}{\displaystyle #2}}

%\def\s{\sigma}

\input{preamble}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%% Don't type in anything in the following section:

%%%%%%%%%%%%

%% For PNAS Only:

\contributor{Submitted to Proceedings

of the National Academy of Sciences of the United States of America}

%\url{www.pnas.org/cgi/doi/10.1073/pnas.0709640104}

\copyrightyear{2008}

\issuedate{Issue Date}

\volume{Volume}

\issuenumber{Issue Number}

%%%%%%%%%%%%

\begin{document}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%% For titles, only capitalize the first letter

%% \title{Almost sharp fronts for the surface quasi-geostrophic equation}

\title{Higher categories, colimits, and the blob complex}

%% Enter authors via the \author command.  

%% Use \affil to define affiliations.

%% (Leave no spaces between author name and \affil command)

%% Note that the \thanks{} command has been disabled in favor of

%% a generic, reserved space for PNAS publication footnotes.

%% \author{<author name>

%% \affil{<number>}{<Institution>}} One number for each institution.

%% The same number should be used for authors that

%% are affiliated with the same institution, after the first time

%% only the number is needed, ie, \affil{number}{text}, \affil{number}{}

%% Then, before last author ...

%% \and

%% \author{<author name>

%% \affil{<number>}{}}

%% For example, assuming Garcia and Sonnery are both affiliated with

%% Universidad de Murcia:

%% \author{Roberta Graff\affil{1}{University of Cambridge, Cambridge,

%% United Kingdom},

%% Javier de Ruiz Garcia\affil{2}{Universidad de Murcia, Bioquimica y Biologia

%% Molecular, Murcia, Spain}, \and Franklin Sonnery\affil{2}{}}

\author{Scott Morrison\affil{1}{Miller Institute for Basic Research, UC Berkeley, CA 94704, USA} 

\and Kevin Walker\affil{2}{Microsoft Station Q, 2243 CNSI Building, UC Santa Barbara, CA 93106, USA}}

\contributor{Submitted to Proceedings of the National Academy of Sciences

of the United States of America}

%% The \maketitle command is necessary to build the title page.

\maketitle

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

\begin{article}

\begin{abstract}

We explain the need for new axioms for topological quantum field theories that include ideas from derived 

categories and homotopy theory. We summarize our axioms for higher categories, and describe the ``blob complex". 

Fixing an $n$-category $\cC$, the blob complex associates a chain complex $\bc_*(W;\cC)$ to any $n$-manifold $W$. 

The $0$-th homology of this chain complex recovers the usual TQFT invariants of $W$. 

The higher homology groups should be viewed as generalizations of Hochschild homology. 

The blob complex has a very natural definition in terms of homotopy colimits along decompositions of the manifold $W$. 

We outline the important properties of the blob complex, and sketch the proof of a generalization of 

Deligne's conjecture on Hochschild cohomology and the little discs operad to higher dimensions.

\end{abstract}

%% When adding keywords, separate each term with a straight line: |

\keywords{n-categories | topological quantum field theory | hochschild homology}

%% Optional for entering abbreviations, separate the abbreviation from

%% its definition with a comma, separate each pair with a semicolon:

%% for example:

%% \abbreviations{SAM, self-assembled monolayer; OTS,

%% octadecyltrichlorosilane}

% \abbreviations{TQFT, topological quantum field theory}

%% The first letter of the article should be drop cap: \dropcap{}

%\dropcap{I}n this article we study the evolution of ''almost-sharp'' fronts

%% Enter the text of your article beginning here and ending before

%% \begin{acknowledgements}

%% Section head commands for your reference:

%% \section{}

%% \subsection{}

%% \subsubsection{}

\dropcap{T}he aim of this paper is to describe a derived category analogue of topological quantum field theories.

For our purposes, an $n{+}1$-dimensional TQFT is a locally defined system of

invariants of manifolds of dimensions 0 through $n{+}1$. In particular,

the TQFT invariant $A(Y)$ of a closed $k$-manifold $Y$ is a linear $(n{-}k)$-category.

If $Y$ has boundary then $A(Y)$ is a collection of $(n{-}k)$-categories which afford

a representation of the $(n{-}k{+}1)$-category $A(\bd Y)$.

(See \cite{1009.5025} and \cite{kw:tqft};

for a more homotopy-theoretic point of view see \cite{0905.0465}.)

We now comment on some particular values of $k$ above.

A linear 0-category is a vector space, and a representation

of a vector space is an element of the dual space.

Thus a TQFT assigns to each closed $n$-manifold $Y$ a vector space $A(Y)$,

and to each $(n{+}1)$-manifold $W$ an element of $A(\bd W)^*$.

For the remainder of this paper we will in fact be interested in so-called $(n{+}\epsilon)$-dimensional

TQFTs, which are slightly weaker structures in that they assign 

invariants to mapping cylinders of homeomorphisms between $n$-manifolds, but not to general $(n{+}1)$-manifolds.

When $k=n{-}1$ we have a linear 1-category $A(S)$ for each $(n{-}1)$-manifold $S$,

and a representation of $A(\bd Y)$ for each $n$-manifold $Y$.

The TQFT gluing rule in dimension $n$ states that

$A(Y_1\cup_S Y_2) \cong A(Y_1) \ot_{A(S)} A(Y_2)$,

where $Y_1$ and $Y_2$ are $n$-manifolds with common boundary $S$.

When $k=0$ we have an $n$-category $A(pt)$.

This can be thought of as the local part of the TQFT, and the full TQFT can be reconstructed from $A(pt)$

via colimits (see below).

We call a TQFT semisimple if $A(S)$ is a semisimple 1-category for all $(n{-}1)$-manifolds $S$

and $A(Y)$ is a finite-dimensional vector space for all $n$-manifolds $Y$.

Examples of semisimple TQFTs include Witten-Reshetikhin-Turaev theories, 

Turaev-Viro theories, and Dijkgraaf-Witten theories.

These can all be given satisfactory accounts in the framework outlined above.

(The WRT invariants need to be reinterpreted as $3{+}1$-dimensional theories with only a weak 

dependence on interiors in order to be

extended all the way down to dimension 0.)

For other non-semisimple TQFT-like invariants, however, the above framework seems to be inadequate.

For example, the gluing rule for 3-manifolds in Ozsv\'ath-Szab\'o/Seiberg-Witten theory

involves a tensor product over an $A_\infty$ 1-category associated to 2-manifolds \cite{1003.0598,1005.1248}.

Long exact sequences are important computational tools in these theories,

and also in Khovanov homology, but the colimit construction breaks exactness.

For these reasons and others, it is desirable to 

extend to above framework to incorporate ideas from derived categories.

One approach to such a generalization might be to simply define a

TQFT via its gluing formulas, replacing tensor products with

derived tensor products (c.f. \cite{1011.1958}).

However, it is probably difficult to prove

the invariance of such a definition, as the object associated to a manifold

will a priori depend on the explicit presentation used to apply the gluing formulas.

We instead give a manifestly invariant construction, and

deduce from it the gluing formulas based on $A_\infty$ tensor products.

This paper is organized as follows.

We first give an account of our version of $n$-categories.

According to our definition, $n$-categories are, among other things,

functorial invariants of $k$-balls, $0\le k \le n$, which behave well with respect to gluing.

We then show how to extend an $n$-category from balls to arbitrary $k$-manifolds,

using colimits and homotopy colimits.

This extension, which we call the blob complex, has as $0$-th homology the usual TQFT invariant.

(The name comes from the ``blobs" which feature prominently

in a concrete version of the homotopy colimit.)

We then review some basic properties of the blob complex, and finish by showing how it

yields a higher categorical and higher dimensional generalization of Deligne's

conjecture on Hochschild cochains and the little 2-disks operad.

Of course, there are currently many interesting alternative notions of $n$-category and of TQFT.

We note that our $n$-categories are both more and less general

than the ``fully dualizable" ones which play a prominent role in \cite{0905.0465}.

They are more general in that we make no duality assumptions in the top dimension $n{+}1$.

They are less general in that we impose stronger duality requirements in dimensions 0 through $n$.

Thus our $n$-categories correspond to $(n{+}\epsilon)$-dimensional {\it unoriented} or {\it oriented} TQFTs, while

Lurie's (fully dualizable) $n$-categories correspond to $(n{+}1)$-dimensional {\it framed} TQFTs.

At several points we only sketch an argument briefly; full details can be found in \cite{1009.5025}. 

In this paper we attempt to give a clear view of the big picture without getting 

bogged down in technical details.

\section{Definitions}

\subsection{$n$-categories} \mbox{}

In this section we give a definition of $n$-categories designed to work well with TQFTs.

The main idea is to base the definition on actual balls, rather combinatorial models of them.

This has the advantages of avoiding a proliferation of coherency axioms and building in a strong

version of duality from the start.

%\nn{maybe say something about goals: well-suited to TQFTs; avoid proliferation of coherency axioms;

%non-recursive (n-cats not defined n terms of (n-1)-cats; easy to show that the motivating

%examples satisfy the axioms; strong duality; both plain and infty case;

%(?) easy to see that axioms are correct, in the sense of nothing missing (need

%to say this better if we keep it)}

%

%\nn{maybe: the typical n-cat definition tries to do two things at once: (1) give a list of basic properties

%which are weak enough to include the basic examples and strong enough to support the proofs

%of the main theorems; and (2) specify a minimal set of generators and/or axioms.

%We separate these two tasks, and address only the first, which becomes much easier when not burdened by the second.

%More specifically, life is easier when working with maximal, rather than minimal, collections of axioms.}

We will define two variations simultaneously,  as all but one of the axioms are identical in the two cases.

These variations are ``isotopy $n$-categories", where homeomorphisms fixing the boundary

act trivially on the sets associated to $n$-balls

(and these sets are usually vector spaces or more generally modules over a commutative ring)

and ``$A_\infty$ $n$-categories",  where there is a homotopy action of

$k$-parameter families of homeomorphisms on these sets

(which are usually chain complexes or topological spaces).

There are five basic ingredients 

\cite{life-of-brian} of an $n$-category definition:

$k$-morphisms (for $0\le k \le n$), domain and range, composition,

identity morphisms, and special behavior in dimension $n$ (e.g. enrichment

in some auxiliary category, or strict associativity instead of weak associativity).

We will treat each of these in turn.

To motivate our morphism axiom, consider the venerable notion of the Moore loop space

\cite[\S 2.2]{MR505692}.

In the standard definition of a loop space, loops are always parameterized by the unit interval $I = [0,1]$,

so composition of loops requires a reparameterization $I\cup I \cong I$, and this leads to a proliferation

of higher associativity relations.

While this proliferation is manageable for 1-categories (and indeed leads to an elegant theory

of Stasheff polyhedra and $A_\infty$ categories), it becomes undesirably complex for higher categories.

In a Moore loop space, we have a separate space $\Omega_r$ for each interval $[0,r]$, and a 

{\it strictly associative} composition $\Omega_r\times \Omega_s\to \Omega_{r+s}$.

Thus we can have the simplicity of strict associativity in exchange for more morphisms.

We wish to imitate this strategy in higher categories.

Because we are mainly interested in the case of strong duality, we replace the intervals $[0,r]$ not with

a product of $k$ intervals (c.f. \cite{0909.2212}) but rather with any $k$-ball, that is, 

any $k$-manifold which is homeomorphic

to the standard $k$-ball $B^k$.

By default our balls are unoriented,

but it is useful at times to vary this,

for example by considering oriented or Spin balls.

We can also consider more exotic structures, such as balls with a map to some target space,

or equipped with $m$ independent vector fields.

(The latter structure would model $n$-categories with less duality than we usually assume.)

\begin{axiom}[Morphisms]

\label{axiom:morphisms}

For each $0 \le k \le n$, we have a functor $\cC_k$ from 

the category of $k$-balls and 

homeomorphisms to the category of sets and bijections.

\end{axiom}

Note that the functoriality in the above axiom allows us to operate via

homeomorphisms which are not the identity on the boundary of the $k$-ball.

The action of these homeomorphisms gives the ``strong duality" structure.

For this reason we don't subdivide the boundary of a morphism

into domain and range in the next axiom --- the duality operations can convert between domain and range.

Later we inductively define an extension of the functors $\cC_k$ to functors $\cl{\cC}_k$ 

defined on arbitrary manifolds. 

We need  these functors for $k$-spheres, for $k<n$, for the next axiom.

\begin{axiom}[Boundaries]\label{nca-boundary}

For each $k$-ball $X$, we have a map of sets $\bd: \cC_k(X)\to \cl{\cC}_{k-1}(\bd X)$.

These maps, for various $X$, comprise a natural transformation of functors.

\end{axiom}

For $c\in \cl{\cC}_{k-1}(\bd X)$ we define $\cC_k(X; c) = \bd^{-1}(c)$.

Many of the examples we are interested in are enriched in some auxiliary category $\cS$

(e.g. vector spaces or rings, or, in the $A_\infty$ case, chain complexes or topological spaces).

This means that in the top dimension $k=n$ the sets $\cC_n(X; c)$ have the structure

of an object of $\cS$, and all of the structure maps of the category (above and below) are

compatible with the $\cS$ structure on $\cC_n(X; c)$.

Given two hemispheres (a ``domain" and ``range") that agree on the equator, we need to be able to 

assemble them into a boundary value of the entire sphere.

\begin{lem}

\label{lem:domain-and-range}

Let $S = B_1 \cup_E B_2$, where $S$ is a $k{-}1$-sphere $(1\le k\le n)$,

$B_i$ is a $k{-}1$-ball, and $E = B_1\cap B_2$ is a $k{-}2$-sphere (Figure \ref{blah3}).

Let $\cC(B_1) \times_{\cl{\cC}(E)} \cC(B_2)$ denote the fibered product of the 

two maps $\bd: \cC(B_i)\to \cl{\cC}(E)$.

Then we have an injective map

\[

	\gl_E : \cC(B_1) \times_{\cl{\cC}(E)} \cC(B_2) \into \cl{\cC}(S)

\]

which is natural with respect to the actions of homeomorphisms.

%(When $k=1$ we stipulate that $\cl{\cC}(E)$ is a point, so that the above fibered product

%becomes a normal product.)

\end{lem}

If $\bdy B = S$, we denote $\bdy^{-1}(\im(\gl_E))$ by $\cC(B)_E$.

\begin{axiom}[Gluing]

\label{axiom:composition}

Let $B = B_1 \cup_Y B_2$, where $B$, $B_1$ and $B_2$ are $k$-balls ($0\le k\le n$)

and $Y = B_1\cap B_2$ is a $k{-}1$-ball (Figure \ref{blah5}).

Let $E = \bd Y$, which is a $k{-}2$-sphere.

%Note that each of $B$, $B_1$ and $B_2$ has its boundary split into two $k{-}1$-balls by $E$.

We have restriction maps $\cC(B_i)_E \to \cC(Y)$.

Let $\cC(B_1)_E \times_{\cC(Y)} \cC(B_2)_E$ denote the fibered product of these two maps. 

We have a map

\[

	\gl_Y : \cC(B_1)_E \times_{\cC(Y)} \cC(B_2)_E \to \cC(B)_E

\]

which is natural with respect to the actions of homeomorphisms, and also compatible with restrictions

to the intersection of the boundaries of $B$ and $B_i$.

If $k < n$,

or if $k=n$ and we are in the $A_\infty$ case, 

we require that $\gl_Y$ is injective.

(For $k=n$ in the isotopy $n$-category case, see Axiom \ref{axiom:extended-isotopies}.)

\end{axiom}

\begin{axiom}[Strict associativity] \label{nca-assoc}\label{axiom:associativity}

The gluing maps above are strictly associative.

Given any decomposition of a ball $B$ into smaller balls

$$\bigsqcup B_i \to B,$$ 

any sequence of gluings (where all the intermediate steps are also disjoint unions of balls) yields the same result.

\end{axiom}

%This axiom is only reasonable because the definition assigns a set to every ball; 

%any identifications would limit the extent to which we can demand associativity.

%%%% KW: It took me quite a while figure out what you [or I??] meant by the above, so I'm attempting a rewrite.

Note that even though our $n$-categories are ``weak" in the traditional sense, we can require

strict associativity because we have more morphisms (cf.\ discussion of Moore loops above).

For the next axiom, a \emph{pinched product} is a map locally modeled on a degeneracy map between simplices.

\begin{axiom}[Product (identity) morphisms]

\label{axiom:product}

For each pinched product $\pi:E\to X$, with $X$ a $k$-ball and $E$ a $k{+}m$-ball ($m\ge 1$),

there is a map $\pi^*:\cC(X)\to \cC(E)$.

These maps must be

\begin{enumerate}

\item natural with respect to maps of pinched products,

\item functorial with respect to composition of pinched products, 

\item compatible with gluing and restriction of pinched products.

\end{enumerate}

%%% begin noop %%%

% this was the original list of conditions, which I've replaced with the much terser list above -S

\noop{

These maps must satisfy the following conditions.

\begin{enumerate}

\item

If $\pi:E\to X$ and $\pi':E'\to X'$ are pinched products, and

if $f:X\to X'$ and $\tilde{f}:E \to E'$ are maps such that the diagram

\[ \xymatrix{

	E \ar[r]^{\tilde{f}} \ar[d]_{\pi} & E' \ar[d]^{\pi'} \\

	X \ar[r]^{f} & X'

} \]

commutes, then we have 

\[

	\pi'^*\circ f = \tilde{f}\circ \pi^*.

\]

\item

Product morphisms are compatible with gluing.

Let $\pi:E\to X$, $\pi_1:E_1\to X_1$, and $\pi_2:E_2\to X_2$ 

be pinched products with $E = E_1\cup E_2$.

Let $a\in \cC(X)$, and let $a_i$ denote the restriction of $a$ to $X_i\subset X$.

Then 

\[

	\pi^*(a) = \pi_1^*(a_1)\bullet \pi_2^*(a_2) .

\]

\item

Product morphisms are associative.

If $\pi:E\to X$ and $\rho:D\to E$ are pinched products then

\[

	\rho^*\circ\pi^* = (\pi\circ\rho)^* .

\]

\item

Product morphisms are compatible with restriction.

If we have a commutative diagram

\[ \xymatrix{

	D \ar@{^(->}[r] \ar[d]_{\rho} & E \ar[d]^{\pi} \\

	Y \ar@{^(->}[r] & X

} \]

such that $\rho$ and $\pi$ are pinched products, then

\[

	\res_D\circ\pi^* = \rho^*\circ\res_Y .

\]

\end{enumerate}

} %%% end \noop %%%

\end{axiom}

To state the next axiom we need the notion of {\it collar maps} on $k$-morphisms.

Let $X$ be a $k$-ball and $Y\subset\bd X$ be a $(k{-}1)$-ball.

Let $J$ be a 1-ball.

Let $Y\times_p J$ denote $Y\times J$ pinched along $(\bd Y)\times J$.

A collar map is an instance of the composition

\[

	\cC(X) \to \cC(X\cup_Y (Y\times_p J)) \to \cC(X) ,

\]

where the first arrow is gluing with a product morphism on $Y\times_p J$ and

the second is induced by a homeomorphism from $X\cup_Y (Y\times_p J)$ to $X$ which restricts

to the identity on the boundary.

\begin{axiom}[\textup{\textbf{[for isotopy  $n$-categories]}} Extended isotopy invariance in dimension $n$.]

\label{axiom:extended-isotopies}

Let $X$ be an $n$-ball and $f: X\to X$ be a homeomorphism which restricts

to the identity on $\bd X$ and isotopic (rel boundary) to the identity.

Then $f$ acts trivially on $\cC(X)$.

In addition, collar maps act trivially on $\cC(X)$.

\end{axiom}

\smallskip

For $A_\infty$ $n$-categories, we replace

isotopy invariance with the requirement that families of homeomorphisms act.

For the moment, assume that our $n$-morphisms are enriched over chain complexes.

Let $\Homeo_\bd(X)$ denote homeomorphisms of $X$ which fix $\bd X$ and

$C_*(\Homeo_\bd(X))$ denote the singular chains on this space.

\begin{axiom}[\textup{\textbf{[for $A_\infty$ $n$-categories]}} Families of homeomorphisms act in dimension $n$.]

\label{axiom:families}

For each $n$-ball $X$ and each $c\in \cl{\cC}(\bd X)$ we have a map of chain complexes

\[

	C_*(\Homeo_\bd(X))\tensor \cC(X; c) \to \cC(X; c) .

\]

These action maps are required to restrict to the usual action of homeomorphisms on $C_0$, be associative up to homotopy,

and also be compatible with composition (gluing) in the sense that

a diagram like the one in Theorem \ref{thm:CH} commutes.

\end{axiom}

\subsection{Example (the fundamental $n$-groupoid)} \mbox{}

We will define $\pi_{\le n}(T)$, the fundamental $n$-groupoid of a topological space $T$.

When $X$ is a $k$-ball with $k<n$, define $\pi_{\le n}(T)(X)$