i

> 1

Journal of Knot Theory and Its Ramifications, Vol. 2 , No. 3 (1993) 251-284 @World Scientific Publishing Company

REIDEMEISTER MOVES FOR SURFACE ISOTOPIES AND T H E I R INTERPRETATION AS MOVES T O MOVIES

[

J. SCOTT CARTER Department of Mathematics, University of South Alabama, Mobile, Alabama 36688, [email protected]

1

MASAHICO SAITO

Department of Mathematics, University of Texas, A u s t i n , Tezas 78712, saitoQmath.utezas.edu

I

i

I

ii

tI

r:I .[ I[

:i

:I :[ 'I-

t

Received 6 March 1993 ABSTRACT

A movie description of a surface embedded in Cspace is a sequence of knot and link diagrams obtained from a projection of the surface to 3-space by taking 2-dimensional cross sections perpendicular to a fixed direction. In the cross sections, an immersed collection of curves appears, and these are lifted to knot diagrams by using the projection direction from 4-space. We give a set of 15 moves to movies (called movie moves) such that two movies represent isotopic surfaces if and only if there is a sequence of moves from this set that takes one to the other. This result generalizes the Roseman moves which are moves on projections where a height function has not been specified. The first 7 of the movie moves are height function parametrized versions of those given by Roseman. The remaining 8 are moves in which the topology of the projection remains unchanged.

1. Introduction Knotted surfaces can be studied diagrammatically by their projections in 3space or by their movies. Having chosen a projection from 4-space into bspace, the projected surface can be cut by 2-dimensional planes that are perpendicular to a fixed direction. The general intersection of such a plane with the projected surface is a collection of immersed curves. Over and under crossing information is depicted on these curves by breaking the resulting diagrams at their double points. Here, over and under refers the relative height in 4-space as measured with respect to the projection from 4 to 3-space where 3-space is regarded as a subspace of 4space. A movie is this sequence of classical link diagrams. The difference between two successive stills in a movie is either a topological equivalence of the underlying graph or an elementary string interaction: A birth or death, a Reidemeister move of type I, 11, or 111, or a surgery. Between the stills that so differ, a critical point of the surface or a critical point of the double points of the surface occurs. The critical data are depicted via their effect on the cross sectional diagrams.

252

J. S. C a r t e r €4 M. Saito

From such a movie, a projection can be constructed. And a projection together with crossing information yields a movie. That the movie can be slowed to be realized as sequence of elementary string interactions is one consequence of the work herein. From the diagrammatic point of view one is lead to ask: "What are the Reidemeister moves for movies?" (This question was posed to the first named author by Louis Kauffman during the "New Mathematical approaches to DNA" conference held in Santa Fe, New Mexico in January 1990. We are grateful for the question and his encouragement .) Movie diagrams can be considered as projections of knotted surfaces with an explicit height function on the 3-space and so Reidemeister moves on movies must include the ability to change height functions. In t h e classical c a s e (knots and links of circles in 3-space) this problem is explicitly solved by including the A s (or simply A-type) moves [lo] [6]. These are moves in which the topology of the 4-valent graph underlying the diagram remains unchanged. Here we give a proof of the 4-dimensional Reidemeister Theorem due to Roseman [ll, 12, 131. (See Homma-Nagase [8] for the case of PL maps in 3-manifolds.) Moreover, we give a list of movie moves that are sufficient to express any isotopy. In the process, we prove the sufficiency of the type I, type 11, type 111, and A-type moves to express classical knot isotopies in the presence of a height function. (See Figures 10 through 14.) That these techniques could be applied to the classical case was first noticed by Roseman [ll]. One might expect that the known solutions [4] [14] to the Zamolodchikov equation yield invariants of knotted surfaces analogous to the Jones polynomial. This is because the Zamolodchikov equation is essentially the quadruple point move of Figures 29 and 22. For these solutions to define invariants of knotted surfaces, one must have a set of moves in hand to check invariance under isotopy. In fact, some of the movie moves are problematic if we use Zamolodchikov's assignment of amplitudes to the elementary string interactions. See [2] for further discussion, and see [3] for an interpretation of the movie moves in the language of 2-categories. 1.1. Statements of results and summary

Section 2 contains the background on general position maps that will be assumed throughout. The notation needed to distinguish various multiple points and their covers is introduced. The main result of the section is the commutative diagram 2.1 that is an immediate consequence of the definitions. This material is folklore that should be in the literature. In Sec. 3, we prove that a given general position map (of curves, surfaces, or solids) or a general position immersion in higher dimensions possesses a function that is Morse on all of its multiple points. In Sec. 4, we show how to project t o a plane and obtain fold lines that iniersect the self intersection manifolds transversely.

Movie Moves

253

The results of these two sections are the key technical steps in finding a sufficient set of movie moves. Section 5 contains the moves on knots that we call elementary string interactions. These moves are analogous to crossings and optima of classical knots. General position surfaces in 3-manifolds can be decomposed as elementary string interactions. These decomposition theorems appear in Theorem 5.2 and the remarks 5.2.1. Theorem 5.1 is the classical Reidemeister Theorem. In Sec. 6 the Roseman moves of isotopy are listed. We use the results of Secs. 3 and 4 to assume the isotopy between knottings is in an appropriate position. Then it is decomposed as a sequence of the Roseman moves. The proof follcws from studying the involution on the double decker set of a projection of the isotopy. In Sec. 7, Roseman's moves and some others are interpreted in terms of movies. Finally, these 15 moves are sufficient to construct any isotopy of movies of a knotted surface (Theorem 7.1). This, the main point of the paper, was announced in [2]. 2. The M u l t i p l e P o i n t S e t s

A unified treatment to the various multiple point sets is given here. To define these we can assume all the maps are nice in the following sense. First, all maps are smooth throughout the paper. Next, it is known that any smooth map can be approximated (by a homotopy) by a map with simple singularities. Specifically, we can assume that a map O : Xn -+ Yntl has, at worst, branch points: In case n = 2, in a neighborhood of a branch point the map looks like Whitney's umbrella which is parametrized by (x, y) I+ (x, xy, y2). In case n = 3, in a neighborhood of the branch point the map looks like this Whitney's umbrella crossed with an open interval. See [7] for details. Furthermore, any immersion can be approximated by a nice one which locally looks like the intersection of coordinate planes. a general position map if for any point in the image of O there Thus we call is a neighborhood in which the image of O is either (1) the intersection of the coordinate planes (including the embedded neighborhood), or (2) a standard (as above) branch point. In case n = 3 we further assume that the branch point set (which is 1 dimensional) intersects the image of O transversely. We give notations for multiple point sets and their preimages for general position maps. For the rest of the section all maps are general position maps. Define T@(r)to be the closure of the configuration space

-

-

C@(r)= { ( x l , . . . , x r ) E X x . . . x X : xi

-

# x,

for i

# j, & O(xl) = . . . @(x,))

.

r

+

The set T@(r)is a manifold of dimension n 1 - r . Since O may have branch points, it is convenient to include these limit points; in this way if X is compact then T@(r)is compact while C@(r)may not be.

-

J . S . Carter tY M. Saito

254

Evidently, C@(r)admits a free action of the symmetric group C, that extends I V

to an action on T+(r). The fixed points of this action project to branch points of the map @. Define a space

-

This is a quotient of the product space modulo the given action of C, on T+(r) and the permutation action on the set of all k-fold subsets of {1,2,. . . , r}. Specifically, an equivalence relation is defined by: (x,,, . . . , xnV)x {ii, . . . , ik}

N

>

I

( x m l , .. . , xm,) x {jl,. . . , j k }

if and only if there is a permutation a E C, such that ~ ( { i l ,.. . , ik}) = {jl,. . . ,j k ) and (xm17 . . . , xmP)= (xo-l(nl)>.. . 3 xo-1(nv)) . For 0 5 k defined by

< r there is a (branched) covering

-

I

T+(r, r ) = T+(r)

~ , , k : T+(r, k) -+

rr,k: [(xi,. .. , x r ) ; {jl,... , j k l l

[ X I , . . ., x r ]

where T@(r)has been identified with T@(r)/C,. The cardinality of the general fibre is (L). : T@(r,k) -+ T+(k) that is defined by There is also an immersion

Any of the sets T@(r)map into Y via @, : [(XI,.. . , x,); {jl,. . . ,jk}] H @(XI).We adopt the conventions that T+(O) = Y , T+(1) = X , and cPo is the identity map on Y. In the next section, we let X, denote the image @,(T+(r)). That these maps are well defined follows directly from the definitions except possibly at branch points. But by dimension restrictions, the branch points occur only for r = 2, and so they lie on the diagonal of X x X. 2.1. Proposition. For any r = 1 , 2 , . . . , ( n commutative diagram:

+ 1 ) and for

Proof. The proof follows directly from the definitions.

any 0

< k 5 r , there is a

I I

Movie Moves

255

The set T@(r,1) = T@(r,r - 1) is called the r-decker set (or manifold) and the set T+(r) is called the r-tuple set (or manifold). They are immersed manifolds from the assumption that cP is a general position map. Consider for example the intersection of the four coordinate 3-balls in the 4ball. This is the image of @ : B: U B; U Bi U Bi -t B4.(Here Bf = {(x2,x3, x4) : C xj" 5 11, Bz = {(XI,xg, 24) : C x; 5 11, and so forth; the map P h i includes B? into B4 by setting the ith coordinate equal to 0.) The quadruple point manifold, T@(4),is a single point; its quadruple decker manifold, Ta(4, I), is the disjoint union of four points. The triple point manifold, T@(3)consists of four coordinate arcs; the triple decker manifold Ta(3, 1) then has twelve components - three over each component of Ta(3). The six coordinate disks form the double point manifold, Ta(2), and the double decker manifold Ta(2, 1) contains two copies of each of these. The six points in Ta(4,2) immerse in T@(2)as the set of double points of @3,2(Ta(3,2)). In Fig. 1 coordinate squares in the unit cube in 3-space are depicted. The triple point Ta(3) is denoted by Ut1,2,3). The double point set Ta(2) consists of three lines indicated by Utl,2) (resp. U{2,31,U{1,3)) each of which is the intersection line between the squares Utl1 and Ut2) (resp. Ut2) and Ut3), UilI and Ut3)).

Fig. 1. Coordinate neighborhoods of multiple points 3

2

Remark. In studying the bordism group of immersions, one can include normal bundle information with these covers and obtain a geometric realization of transfer maps in stable homotopy.

! I

i

256

J. S. Carter €4 M. Saito

3. E v e r y w h e r e M o r s e Functions In this section X and Y are compact manifolds. Let O : ( X , doX, d l X ) -+ (Y, doY, 8lY) denote a proper general position map if n = 1, 2, 3 or a general position immersion if n 4. Both the boundary of X (resp. Y) are written as the union of two disjoint sets each of which is the union of components of the boundary of X (resp. Y): d X = doX U dlX, (resp. dY = doY U dlY). A map f : (Y, doY,dlY) + (I,(01, {I)), from the space Y to the unit interval I = [O, 11, is a proper everywhere Morse function if

>

1. f oO, is Morse on T*(r) for all r = 0 , 1 , ... , n + l ) , 2. (f o 0,)-'(i) = diX, for i = 0,1, and 3. (f o a,) has no critical points in a neighborhood of d X r , for all r = 0 , 1 , . . . , n f 1. If X and Y are closed manifolds, then is called an everywhere Morse function if O satisfies the first condition in the above. By definition, the 0-tuple manifold, T@(O)is Y, and a. is the identity map. 3.1. T h e o r e m . Let O : ( X ; doX,d l X ) + (Y; doY, dlY) be a proper general position map from a compact n-manifold, X, to a compact ( n 1)-manifold Y. Then the set of proper everywhere Morse function f : (Y; doY, dlY) + ( I ; {O), (1)) is dense in the C 2 topology.

+

3.1.1. R e m a r k . By the commutativity of Diagram 2.1, an everywhere Morse function is equivariantly Morse on each of the T@(r,k). Theorem 3.1 holds when the boundaries of X or Y are empty. 3.1.2. C o o r d i n a t e Neighborhoods. Before proving 3.1, we establish some notation with regard t o coordinate neighborhoods. Let U denote a coordinate neighborhood of a point y € Y , and let h : U --+ Rn+l,denote a coordinate function such that h(y) = 0. If y E dY, then without loss of generality we may assume that h maps U to the intersection of the interior of the unit ball and the upper-half space: {(XI,.. . , x n + l : xn+l 0). If y is a branch point of O, then there is a unique x E X such that (1) O(x) = y, and (2) there is a coordinate neighborhood (W, k) of x such that hak-'(xl, xz, (x3)) = (xl, ~ 1 x 2x,i , (x3)) when n = 2 (respectively n = 3). If y is not a branch point, the coordinate neighborhood, U, can be chosen so that O-l(U) is a collection of disjoint coordinate neighborhoods of points in X. Thus O-'(u) = u ~ U . . U U , , for 0 5 r 5 n + 1 ,

>

and if y E d Y then 0

5 r 5 n. Furthermore, there are coordinate maps

Finally, with U a s above, each of the multiple point manifolds has similar coordinate neighborhoods. Specifically, if S c {1,2,. . . , r ) is any subset, define

Movie Moves

257

Us = { Z E Rn+l: zj = 0 for j E S). Then Us is a coordinate neighborhood of a point in T @ ( J S Jthat ) maps into U . We adopt the conventions that UB = U . For more intuition about these coordinatizations see Fig.1.

Proof of 3.1. In the Lemmas 3.1.4 through 3.1.10 we show that in the local coordinates, the set of functions that are Morse on each of the self intersection sets is open and dense in the set of all functions. This is relatively clear from the local description of general position multiple or branch points, for almost all projection directions are Morse on all the self intersection strata. From these local gener:; position coordinates we can construct a properly everywhere Morse function as follows. (The technical details of the proof are similar to those given in Milnor [9] Theorems 2.7 and 2.5.) First, the set of properly everywhere Morse functions is open in the C2 topology by Lemmas 3.1.7 and 3.1.8. Lemma 3.1.3 gives a function without any critical points on the boundaries of T @ ( r )for r = 0,. . . , n + 1. We adjust this function in stages. First cover Y by a finite number of coordinate neighborhoods in which @ has a standard form. Choose a compact refinement of this cover. Then find a function fl that is everywhere Morse on a given compact neighborhood, C1, and that is C2 close to the original function. This can be done because the set of everywhere Morse functions is dense for the local coordinates. Any function that is sufficiently close to fl is also everywhere Morse on C l . Furthermore, there is a function f2 that is close to fl and that is everywhere Morse on a compact coordinate neighborhood C 2 . The process continues for a finite number of stages until we find a neighborhood of the given function such that any function g therein is everywhere Morse on all of Y. That is, g o Or is Morse for all r = 0 , . . . , n 1. Thus the proof follows from the lemmas below.

+

When the proof of one of the following lemmas is an easy translation of the proof given in Milnor [9], the details are omitted.

I

3.1.3. Lemma. There is a smooth function f : Y

I

Proof. See Milnor [9] Lemma 2.6.

-

[O, 11 such that

1. ( f o ~ , ) - ' ( i ) = d i T a ( r ) for r = 0 , 1 , . . . n, and i = 0 , l ; 2. There are no critical points o f f , f a , and f a , in a neighborhood of each of the boundaries.

3.1.4. Lemma. Let U denote an open ball around 0 in Rn+'. Let f : U -+ R denote any C2 map. T h e n almost all linear maps L : Rn+l --+ R, satisfy the following condition: the composition ( f L ) Q s is Morse for all S C {1,2,. . . , r ) where r n 1 and as : Us U is the inclusion.

< +

-

+

Proof. The proof is similar the proof of Lemma A in Milnor [9], but details are provided.

258

J. S. Carter

€4 M. Saito

+

Let y E U be in the image of some Os. If IS1 = n 1, then by definition, f is Morse. Consider the manifold N = Us x H o r n R ( ~ " ,R ) where St = {1,2, . . . , n 1)\S, and R" denotes the linear subspace of Rn+' in which xi = 0 if i E S . The manifold N is of dimension 2(n 1- ISI). Let M s = {(x, L) E N : D(f L)as = 0); this is a submanifold that is homeomorphic to Us by the map x I+ (x,-D(fOs)(x)). Each point (x,L) E Ms corresponds to a critical point of (f L)Os that is degenerate when the matrix

+

+

+ +

is singular. The projection T : M + H o m ( ~ " , R ) onto the second factor is the map x H -D(fQs). Thus (f L)@s has a degenerate critical point if and only if LIRSl is the image of a critical point of T. Now by Sard's theorem the set of maps, L', that are the image of critical points of T : M + H O ~ ( R ~R ') ,has measure zero in R". Thus if (f + L)Qs has a degenerate critical point, then LIRsr is in this measure zero subset of R ~ ' .So for fixed S, the set of Ls (ells) that yield degeneracy is of measure zero in R n t l . The union over all S C {1,2,. . . , r } of these measure zero sets is still of measure zero, and this completes the proof.

+

3.1.5. Lemma. Let O : R2 + R3 be defined by @(x,y ) = (x,xy, y2). Let f : R 3 + R be a C m function. T h e n for almost all linear maps L : R3 + R , the map (f L)O has only non-degenerate critical points.

+

Proof. The only critical point of O is (0,O). By Lemma 3.1.4, we may assume that (f L) is Morse, and so (f L) o O is Morse except possibly a t (0,O). The map f can be C 2 approximated by a second degree polynomial function f = 3 air; + aijxizj. Let L = biz; denote a linear map. Then L) has a critical point at (0,O) if and only if bl = -al. It is degenerate if and (f only if b3 = -a3. SO the set of linear maps that give degenerate critical points is of measure zero.

+

+

+

xi=l x:,j=l

x:=,

3.1.6. Lemma. Let O : R3 + R~ be defined by @(x,y, r ) = (x, xy, y2, r). Let f : R4 + R be a C m function. T h e n for almost all linear maps L : R4 + R the map (f L)@ has only non-degenerate critical points.

+

Proof. As in Lemma 3.1.5 assume that (f + L) o @ has a critical point along the line (0,0,z). If this is degenerate, then each bi depends linearly on r . So the set of linear maps L that yield degenerate critical points is of measure zero. 3.1.7. Lemma. Let S c {1,2, . . . , n). Let Ks be a compact subset of the open ball Us in n S r . Suppose that f : U + R has only non-degenerate critical points on each K s T h e n there is a positive number S such that if g : U + R is a C 2 function that satisfies

I

Movie Moves

259

and

for i , j E S' at all points of K s , then gJUs has only non-degenerate critical points on

Proof. See Lemma B in [9]. 3.1.8. Lemma. Let I< be a compact subset o f R n + l . Let I 1) to the plane is approximated by a map only with two types of singularity: cusps and folds (see [7] for definitions). The rank of the derivative of the map drops by 1 at a point on these types of singularity. More specifically, the neighborhood of a point on the fold is parametrized by (u, x, (y)) t, (u, x 2 ( fy2)) in case of n = 2 (resp. n = 3). The set on which the rank drops is 1-dimensional and called a fold lane. A neighborhood of a cusp is parametrized by (u, x, (y)) H (u,x3 ux(+y2)); CUSPS are isolated

+

Fig. 2. Fold lines

Movie Moves

261

CUSP

Fig. 3. A Cusp

Our goal in this section is to prove Theorem 4.1. We need this theorem for the later sections. We briefly explain our motivations here. In order to show that the set of movie moves is sufficient to express any isotopy between knotted surfaces, we consider a 1-parameter family of "height" functions on the range. Theorem 4.1 allows us to assume that this family of height functions is as nice as possible. Notice here that a critical point for one such height function extends to a 1-dimensional set in the family (at least locally), and this forms a fold line. The image of this fold in the projection gives rise t o those isotopy moves that involve folds.

I

I

I

I

I

4.1. Theorem. Let n = 1 or 2. The set of functions q : Yn+'x I + R x I satisfying the following properties i s open and dense i n the C2 topology. (1) For ( n , k) = (1, I), (2, I ) , or (2,2), the singularities of the composition qk = q o Q k consist only of folds and cusps. (2) These singularities intersect the image of the multiple decker sets of @ transversely. More precisely, Let Fk be the fold locus of q k . This is the submanifold of T@(k) on which the rank of the derivative of qk drops by 1. Then Fk intersects the image of Q,,k ( T @ ( rk)) , transversely. 4.1.1. Remarks. In case n = 1, the map Q has double and triple points. The double point manifold in this case is 1-dimensional, and the triple point manifold is 0-dimensional. Thus there are no folds along these strata. In case n = 2, the map @ has double, triple, and quadruple points. The triple point manifold is 1dimensional and the quadruple point manifold consists of isolated points. These self intersection sets do not have folds. Thus the fold lines exist in only the cases when (n, k) = (1,I ) , (2,2), and ( 2 , l ) . The double point manifold may have cusps on its boundary, as in Fig. 6. Theorem 4.1 will be used to give a proof of the sufficiency of the Reidemeister moves and A-type moves t o express any isotopy between knots in the smooth category (Sec. 5). By this theorem we can assume without loss of generality that any isotopy between two knot projections can be approximated by an isotopy satisfying the conditions in Theorem 4.1.

II

262

J. S. Carter t3 M. Saito

Fig. 4. Moving crossing points past folds: unbranched

SCC

FIG 32

4

Fig. 5. Moving crossing points across folds: branched

-

-

-

Movie Moves

Fig. 6. Cusps on double point set

Fig. 7. Moving triple points across fold lines

263

264

J. S. Carter U M. Satto

In case n = 1, the singularities of q@ are fold lines and simple cusps. The fold lines are properly embedded submanifolds of TB(l) that intersect the image under 0(2,1) of the double decker manifold of a projection of a knot isotopy. The double decker manifold is 1-dimensional, and when these pass over a fold, there is either a branch point (type I) move, or one of the A-type moves (Fig. 14). The triple decker manifold, in this case, consists of isolated points, and so by transversallty the fold lines miss these points. In case n = 2, then we use Theorem 4.1 t o prove that the movie moves suffice to construct isotopies between knotted surfaces. In this case, @ is the projection t o R3 x I of the knot isotopy. The singularities of q@ are isolated cusp points or properly embedded fold lines. These intersect the image of the surface of double decker points of F x I (where F is the knotted surface) in isolated points. Furthermore, the double point manifold will have simple cusps and folds, and these may intersect the image of the triple decker manifold These intersections give rise to the movie moves. Details are found In Sec. 7.

4.1.2. Proof of 4.1. The proof is analogous t o the proof of Theorem 3.1. We first show that locally the fold lines intersect the multiple decker manifolds transversely for an open dense subset of maps. Then we adjust the map, q, so that qQ,is nice on bigger and bigger portions of the manifold X x I. T h a t is, in a neighborhood of q there is a small open set of maps such that any map therein is nice on all of X x I. The open set is the intersection of neighborhoods in which maps are subsequently nicer on larger and larger portions of the manifolds. Thus the proof follows from Lemmas 4.1.3 through 4.1.6.

+

4.1.3. Lemma. Let n = 1 , 2 , let U be an open ball an Rn+', and let 0 5 r 5 n 2. Let S c { I , . . . , r ) , and for i E S let U, = { I E Rn+' : t,= 0). The set of smooth maps f : U -+ R2 such that f lU, has folds and sample cusps whzch zntersect the remaznzng U3 and thezr zntersectzons transversely zs an open dense set zn the CCO topology.

Proof. The map f, = f lU, has folds and simple cusps if it is 2-generic in the sense of [7]. The szngular set, Sl(f,), is defined t o be the set of x E U,on which the rank of the derivative drops by 1. Thus for most f,, the fold lines of f, = f lU, form a submanifold of U,on which the rank of fz drops by 1, [7], p. 87. For any i E S, the set of maps f such that f, has folds is open and dense. Similarly, the set of maps that has folds on the intersection of these subsets is open and dense. An argument about codimensions shows t h a t there are non-empty folds only on the double point manifold when n = 2. (See [7], p. 143.) The intersection of this finite collection of open dense subsets is also open and dense. Now if a fold line does not intersect the image of the multiple decker manifold transversely, or if it intersects at a cusp, then the submanifold of fold points can be perturbed t o a transverse intersection. Moreover, there is a map that is close t o the

L

--

Movie Moves

265

original f , and the fold locus of the new map is the transversely intersecting fold. This completes the proof.

-

= (x, xy, y2). T h e n for R 2 , the composition LQ has folds containing (0,O).

4.1.4. L e m m a . Let O : R2 + R3 be defined by @(x,y)

almost all linear maps L : R3

Proof. The fold lines are the points at which the rank of DLQ drops by 1. But the determinant of DLO can be written as Ax2 + By + Cy2. where the constants A, B , C depend on the linear map L. For most values of A , B , C, det = 0 on a quadratic curve containing (O,O), so for an open dense set of linear maps the folds will be generic. 4.1.5. L e m m a . Let 0 : R3 --+ R4 be defined by @(x,y,z) = ( x , x y , y 2 , z ) . For most

linear maps L : R4 -+ R 2 , the fold locus of the map LO is a submanifold of R3.

Proof. This is a direct calculation. 4.1.6. L e m m a . Let O be the map defined in Lemma 4.1.4 or Lemma 4.1.5. Then for most maps f from R3 (resp. R4) to R2, the folds of the composition fO intersect the image of the double decker set transversely.

1

Proof. The fold lines are the points at which the derivative drops rank by 1, but by the Lemmas 4.1.4 and 4.1.5 this happens for almost all linear maps and hence for almost all maps f . In this section, we include a set of illustrations depicting the types of cusps and folds that arise for maps of 3-manifolds, and the intersection of these fold with the higher multiplicity decker sets. These, the projections of the image of 0 , are the Figs. 2 through 7. 5. T h e R e i d e m e i s t e r T h e o r e m a n d E l e m e n t a r y S t r i n g I n t e r a c t i o n s

In this section, we give a proof in the smooth category of the classical Reidemeister theorem in the case a height function is fixed during the isotopy. The elementary string interactions are defined. And we show how to use Theorem 3.1 to decompose knottings of surfaces. 5.1. Theorem. Suppose two classical link diagrams of isotopic links are given. Fix a height function on the plane and assume that these diagrams have isolated maximal, minimal, and crossing points. T h e n there exists a finite sequence of Reidemeister moves and A-type moves that transforms one of the diagrams into the other. These Reidemeisler moves and the A-type moves are depicted in Figs. 10 through 14.

Proof. Let L denote a closed 1-manifold (finite collection of circles). Let H : L x I --+ R3 x I denote an isotopy between knottings of L. Consider a projection R~ + R, and let p : R3 x I --+ R2 x I denote this projection times the identity map on I. The map p may be perturbed so that Q = pH is a proper general position

I

266

J . S. C a r t e r €4 M. S a i t o

map on the surface L x I. By Theorem 3.1, we may assume that the projection q': R2 x I I is everywhere Morse. Furthermore, the projection q : R2 x I -+ R x I may be assumed to satisfy the conclusion of Theorem 4.1. That is, when restricted to the image of a, the projection q generically has folds and simple cusps, the fold lines intersect the double decker manifold transversely (and hence in isolated points), and the only folds on the double decker set are the branch points of the generic map a . -+

Fig. 8. BirthIDeath

Fig. 9. Surgery

Fig. 10. Type I

Fig. 11. Type I1

Movie Move3

267

Fig. 12. Type I11

Fig. 13. Switch Back move

Fig. 14. Moving crossing across a fold

Consider the moves t o link diagrams that are depicted in Figs. 10 through 14. In these pictures, the height direction - in which the top of the page is up is fixed. On the right side of each figure is a depiction of the projection p of the corresponding move. The map q can be thought of as the projection t o the plane of the picture.

A type I move (Fig. 10) is a critical point of the Morse function q1O2 on the double point manifold. At this local optimum, the involution on the double decker manifold has a fixed point. A type I1 move (Fig. 11) is a critical point of q1Q2 a t which the involution on the double decker manifold is acting freely. A type 111 move (Fig. 12) represents an isolated triple point of O . Locally, the fold lines intersect the double point manifold transversely as in Fig. 10 or 14. And if the fold line have critical points (with respect t o q'), then there is a cusp singularity in the surface and this is depicted in Fig. 13. The moves depicted exhaust the possibilities for critical points and transverse intersections. This completes the proof.

268

J. S. Carter €4 M. Saito

Note in the above proof that a branch point (resp. a maximal or minimal point of the double point set, an isolated triple point) of the projection of an isotopy of a classical link corresponds to the Reidemeister type I (resp. type 11, type 111) move if the projection is a general position map and a height function on 3-space is fixed. In general, for a general position map from a closed surface to 3-space, there may also be maximal, saddle, and minimal points such that in a neighborhood of such a point the surface is embedded. Consider the intersection of the surface with a pair of planes on either side of such a critical point where the planes are perpendicular to the fixed height function. Then the curves of intersection differ by the birth of a small circle, the death of a small circle, or a surgery between two arcs. These events correspond to a local maximum, a local minimum, and a saddle point, respectively. 5.1.1. Definition. The elementary string interactions (ESIs) are birth, surgery, death, Reidemeister type I, type 11, and type I11 moves. These are depicted in Figs. 8 through 12, and they correspond to passing through critical points of a surface or its singular set as described in the paragraph above. 5.2. T h e o r e m . Let h : F + R 4 be an embedded closed surface. Consider a projection p : R4 + R3 to a subspace and a height function k : R~ + R. For almost all projections p and height functions k , we get a projection p o f ( F ) such that (1) p o f is a general position map, and (2) the slices of p o f ( F ) by parallel planes i n R3 that are perpendicular to the height direction k gives a sequence of elementary string interactions up to topological equivalence of link diagrams.

Proof. By Theorem 3.1 we have an everywhere Morse function on a projection of F which we may assume is a general position map. Each of the elementary string interactions represents a critical point of the surface, its double point manifold, or a triple point. This completes the proof. 5.2.1. R e m a r k s . 1. If a height direction is also fixed in all of the 3-dimensional cross sections, then one must include the A-type moves as elementary string interactions. There might be a good reason to do this from the point of view of constructing invariants of knotted surfaces. But in this case, the list of movie moves becomes 2. Theorem 5.2 can be generalized to properly embedded compact surfaces in R 3 x I. 3. Theorem 5.2 can also be generalized to generically mapped surfaces in 3manifolds. 6. Roseman's T h e o r e m 6.1. T h e o r e m ( R o s e m a n ) . Suppose two projections in 3-space of isotopic knotted surfaces in 4-space are given. Assume that they are the images of general position maps. A set of moves between generically mapped surfaces are depicted in Figs. 15

I

Movie Moves

269

through 22. Then, there exists a finite sequence of moves from this set which transf o m s one of the projections t o the other up t o topological equivalence.

Fig. 15. An optimum on the surface of self intersections with fixed points

Fig. 16. A saddle on the surface of self intersections with fixed points

270

J. S. Carter €4 M. Saito

f st-tF/G

zg

Fig. 17. An optimum on the surface of self intersection with no fixed points

4

5f-cF/G

r6

Fig. 18. A saddle on the surface of self intersections with no fixed points

SEE

NG

Fig. 19. An optimum on the triple point set with no fixed points

Movie Moves

Fig. 20. Moving a branch point through a triple point

Fig. 21. Mixing type I1 and type I deaths on the double decker set

271

272

J.

S. C a r t e r €4 M. Saito

SEE

JSC Fig. 22. A quadruple point or tetrahedral move

R3 denote Proof. Let H : F x I -+ R4 x I denote an isotopy. Let p : R4 a projection. We may choose the direction of p such that the composition O = p ( H x id) is a generic map of the 3-manifold F x I. Let q : R~ x I --t I denote the projection. This may be chosen so that q is everywhere Morse on O ( F x I). Without loss of generality these choices can be done preserving the property that H is a n isotopy since isotopies form an open dense set (see [13]). We proceed to analyze the critical points of qOz, q03, and qQ4. (The composition qO has no critical points because H is an isotopy, and in perturbing q to be Morse, we did so in such a way that the property of being an isotopy is preserved.) Consider the critical points of the double decker manifold. Recall from remark 3.1.1 that since q is Morse on the double point manifold, it is equivariantly -+

Movie Moves

273

Morse on the double decker manifold. The double point manifold of a generically mapped 3-manifold is a surface. And critical points of surfaces come in two flavors: Optima (maxima or minima) and saddles. We think of the former as a quadratic cone and the latter as a saddle. Since there is an involution on the double decker manifold there is either a line of fixed points on the saddle or quadratic cone, or there are a pair of saddles or cones that are interchanged. The line of fixed points can be any axis of symmetry. In Fig. 15 the move that corresponds to an optimum with a line of fixed points is depicted. Figure 17 depicts the move that corresponds to a free action on the quadratic cones. A saddle of double decker points with a line of fixed points is depicted in Fig. 16. In Fig. 18 there is a pair of saddles that are interchanged by the involution. Next we consider the critical points on the triple decker set. The triple decker manifold is a collection of curves immersed in F x I as the set of double points of the double decker manifold. Morse critical points of curves are optima. Since these curves are the double points of an immersed surface, they are the trace of either type I1 moves or type I moves. Here the remark 5.2.1 (2) is being applied to the double decker set which is an immersed surface. In case three disjoint type I1 moves occur on the double decker surface, we have two triple points of the knot projection cancelling. This is depicted in Fig. 19. Locally, at a triple point of a surface, 3 arcs of double points intersect. If an arc at a triple point returns to the triple point in an a fashion, then the arc of double points with which the alpha is involved is embedded in F. (This matter is illustrated in detail in [I].) Thus, if a triple point dies in the isotopy via a type I move in the double decker set, there is a type I1 move on the companion sheet. Thus the illustration given in Fig. 21 gives the only possible way a type I death of a triple point can occur. This projects to the Fig. 20. Finally, the map can have quadruple points; such points are isolated. The corresponding move is depicted in Fig. 22. It represents a version of the Zamolodchikov equation [14]. Since all possible Cg equivariant critical points of surfaces, all possible Cg equivariant critical points of curves, and all possible C4 equivariant actions on 4 points have been listed, this completes the proof.

7. Movie Moves In this, our final section, we list moves to movies that are sufficient to construct any isotopy of knotted surfaces. These moves were announced in [2]. Let h : F -. R 4 be an embedded closed surface (i.e., a knotted surface or a knotting). Without loss of generality we may assume that p o h is a general position map where p : R 4 -. R3 is a projection. Furthermore from Theorem 5.2 we may assume that p o h is a sequence of ESIs when we fix a height function on 3-space

274

J. S. Carter F5 M. Saito

and take slices by parallel planes perpendicular to this fixed height direction. Such a projection p o h ( F ) with a fixed height function on 3-space is called a movie description (or simply a movie) of the knotted surface. The slices described above are also called a movie. Two such projections are topologically equivalent if there is an isotopy between them sending the double point set and ESIs of one t o those of the other. When we take slices, a set of moves between sequences of ESIs are depicted in Figs. 23 through 37. These are called movie moves. The goal of this section is to prove that any two projections of isotopic knotted surfaces are related by a sequence of these moves. Each movie move corresponds to one of the following: 1. A critical p o i n t of the self intersection set of t h e k n o t projection. These are movie parametrizations of Roseman's original seven moves, where we interpret an isolated quadruple point as a critical point. 2. Passing the surface of d o u b l e p o i n t s transversely t h r o u g h a fold line. At most, those moves that correspond to the unbranched moves illustrated in Fig. 4, and the top two moves illustrated in Fig. 5 are necessary. 3. Passing a t r i p l e p o i n t transversely t h r o u g h a fold line of the d o u b l e decker set. The projection of this move is illustrated in Fig. 7. 4. Cusps. There is a cusp on the fold line that corresponds t o the cancellation of a birthldeath and a surgery. There are also two types of cusps on the surface of self intersections. These are illustrated in projection in Fig. 6.

Some comments on movie moves follow. Note from Theorem 3.1 that any movie is described by a sequence of elementary string interactions. These correspond crossings and maximal/minimal points in classical case. A movie of a knot that is given as a sequence of elementary string interactions can be transformed to another by a sequence of the moves to movies that are depicted. Thus the movie moves are elementary moves t o sequences of elementary string interactions that are sufficient to construct any isotopy. In the depiction of the movie moves we have adopted the standard convention of listing movie moves with only one possible set of crossing data. We leave it as an exercise for the reader to find all the crossings for which the movies moves remain valid. Furthermore, the movie moves can be read up or down. We have only depicted maxima, but minima occur upon turning the movie move upside down. Moves in which critical points can change level are not listed. These are analogous to the braid relations uiuj = uj ui if li - jl > 1. One such move is used in Fig. 38. 7.1. T h e o r e m . Suppose two projections in 3-space of isotopic knotted surfaces in 4-space are given. W e may assume that they are described as sequences of ESIs with respect to a fixed height function on 3-space. T h e n there exists a finite sequence from the set of movie moves depicted i n Figs. 23 through 3 7 which transforms one to the other up t o topological equivalence.

i

Movie Moves

Fig. 23. Removing/adding branch points through an optimum on the double decker set

Fig. 24. Removing/adding branch points through a saddle in the double decker set

275

I

276

J. S. Carter €4 M. Sarto

I

I

Fig. 25. Removingladding circles of double points through an optimum

Fig. 26. A saddle move on the double decker set

Movie Moves

277

Fig. 27. Removing/adding a pair of triple points through an optimum on triple point manifold

Fig. 28. Moving a branch point past a sheet via an optimum on the triple point manifold

278

J. S.

C a r t e r €4

M.

Saito

Fig. 29. The Zamolodchikov Move: a quadruple point of the isotopy

Movie Movea

Fig. 30. A cusp in the isotopy

Fig. 31. Moving a branch point through an optimum fold

279

280

J. S. Carter €5 M. Saito

Fig. 32. Moving a branch point through a saddle

Fig. 33. Moving a plane past an optimum fold

I Movae Moves

Fig. 34. Moving a plane through a saddle fold

Fig. 35. A cusp of the double point manifold

281

282

J. S. Carter f3 M. Saito

Fig. 36. A cusp of the double point manifold that involves a branch point

Fig. 37. Moving a triple point past a fold in the double point manifold

Movie Moves

I Fig. 38. Factoring the unusual moves of branch points past a saddle

Proof. By Theorem 3.1 and 4.1 we may assume that the projections of the isotopy are everywhere Morse and fold lines intersect the multiple points transversely. The Roseman moves exhaust the set of Morse critical points that can occur on the multiple point sets. And the list of folds intersecting the multiple point set is complete listing of these transverse intersection. All that remains t o be shown is that the trace of the moves that are illustrated in the bottom frame of Fig. 5 can be factored into other moves. This factorization is given in Fig. 38. The third frame of Fig. 5 can clearly be written as a cusp on the double point set (Fig. 36 and Fig. 6).

I

Acknowledgements Jose Barrionuevo helped us with proving Morse functions are dense on neighborhoods of branch points. We have had valuable conversations with Cameron Gordon, Kunio Murasugi, Dan Silver, arid Richard Hitt. Jacob Towber and John Fischer impressed upon us the urgency of writing down these results. We have received encouragements from L. H. Kauffman and J . D. Stasheff throughout the period we

284

J. S. Carter €4 M. Saito

have been engaged in this and related research. We especially thank Dennis Roseman for giving us permission to give a proof of his work. Many of the ideas for this paper are found in his manuscripts. Although we have not explicitly listed the work [5] as a reference, we drew upon it at several points to better understand folds and cusps. Yuuichiro Sugimoto gave a careful reading of an earlier manuscript, and we benefited from his suggestions. Finally we thank the referee for valuable comments. References [ I ] J. Scott Carter, Extending immersions of curves t o properly immersed surfaces, Topology and its Applications 40 ( 1 9 9 1 ) , 287-306. [2] J. Scott Carter and Masahico Saito, Syzygies among elementary string interactions i n dimension 2 1 , Lett. in Math. Phys. 23 ( 1 9 9 1 ) , 287-300. [3] Fischer Jr. and E. John, Braided tensor 2-categories and 2-knots, Preprint. [4] Igor Frenkel and Gregory Moore, Simplex equations and their solutions, Comm. Math. Phys. 138 ( 1 9 9 1 ) , 259-271. [5] George K . Francis, A Topological Picturebook, Springer Verlag, New York ( 1 9 8 7 ) . [6] Peter J. Freyd and David Yetter, Braided compact categories with applications t o low dimensional topology, Advances in Math. 77 ( 1 9 8 9 ) , 156-182. [7] Martin Golubitsky and Victor Guillemin, Stable Mappings and T h e i r Singularities, GTM Vol. 14, Springer Verlag, New York (1973). [8] Tatsuo Homma and Teruo Nagase, O n elementary deformations of the m a p s of surfaces i n t o 3-manifolds I, Yokohama Math. J. 33 ( 1 9 8 5 ) , 103-119. [9] John Milnor, Lectures o n the h-Cobordism Theorem, Mathematical Notes, Vol. 1 , Princeton University Press, Princeton, NJ (1965). [ l o ] Kurt Reidemeister, Knotentheorie, Springer Verlag, Berlin (1932). K n o t Theory, BSC Associates, Moscow, Idaho (1983). [ l l ] Dennis Roseman, Reidemeister-type moves for surfaces i n four dimensional space, Preprint. [12] Dennis Roseman, Projections of codimension 2 embeddings, Preprint. [13] Dennis Roseman, Elementary moves for higher dimensional knots, Preprint. [14] A. B. Zamolodchikov, Tetrahedron equations and the relativistic S - m a t r i x of straightstrings i n 2 1-dimensions, Commun. Math. Phys. 79 ( 1 9 8 1 ) , 489-505, reprinted in Yang-Baxter Equation i n Integrable Systems, ed. Jimbo, World Scientific Publishing Co., Singapore (1989).

+

+