You are on page 1of 24

Models of Entanglement

Gregory BUCK and Jonathan SIMON

Abstract. There are many ways to pack, and arrange, laments in space. The packing can be quantied in terms of how much lament length is contained in a typical unit of volume. How something is arranged is harder to dene, but one can think in terms of systems that are random as opposed to those that have some kinds of regularity. We identify three fundamental regimes of packing (linear, laminar, globular) and illustrate how several dierent real and virtual lament systems fall into these three regimes. The amount of entanglement in a lament system is an essential property, strongly inuencing physical behavior. There are dierent ways to try to quantify entanglement; we focus here on average crossing number. We present several models of laments and analyze the typical tangling behavior in the dierent packing regimes. We conclude with an example comparing the tangling eects of random and orderly packing.

Mathematics Subject Classication 2000: Primary 57M25; Secondary 92E10, 53A04, 74K05 Keywords: Average Crossing Number, Curvature, Filaments, Knots, Packing, Polymers, Tangling

1. Introduction We are interested in the general problem of measuring the entanglement for lament systems, such as in Figure 1. Our goal is to describe how the topological/geometric complexity of the entanglement can be bounded, or even better estimated, if we understand how densely the lament is packed in space. In this
1

GREGORY BUCK AND JONATHAN SIMON

Figure 1. Basic Question: For complicated laments (here one curve or a pair of curves), how does the complexity of entanglement grow with length? discussion, we think of measuring entanglement complexity in terms such as average crossing number, linking number, or average writhe. Just to have a particular example to think about (later), here is the beginning of a paper [9] proposing a particular kind of packing for proteins at medium densities. Soluble proteins are well-packed, and their packing densities may be as high as that of crystalline solids. Yet there are numerous packing defects or voids in protein structures, whose size distributions are broad. The volume (v) and area (a) of protein does not scale as v a3/2 , which would be expected for models of tight packing. Rather, v and a scale linearly with each other. In addition, the scaling of protein volume and cluster-radius is characteristic of random sphere packing. Such scaling behavior indicates that the interior of proteins is more like Swiss cheese with many holes than tightly packed jigsaw puzzles. The picture in Figure 2 is a model we might make of their proposed packing. Imagine a lament that is packed so as to occupy nearly all the space represented by the surfaces of a bunch of packed spheres (not occupying much of the interiors). The left picture shows the (virtual) spheres. The right picture shows a 1-skeleton of those surfaces. Imagine a lament that wanders around in space so it looks
G.B. Research partially supported by NSF Grant DMS 0107747. J.S. Research partially supported by NSF Grant DMS 0107209. Special thanks to Prof. A. Kawauchi and the Osaka City University Advanced Mathematics Institute for their hospitality. This paper is a revised version of the lecture notes for JS talk in the OCAMI International Workshop on Knot Theory for Scientic Objects, March 2006 .

MODELS OF ENTANGLEMENT

approximately like that 1-skeleton. What is the topological complexity of that lament?

Figure 2. What is the average crossing number of the 1-complex that models protein laments packed densely along spherical shells? Our basic thesis is that in any lament system, there is a characteristic rate of spatial packing: how much length is contained in how much volume. And the expected growth rate of entanglement, relative to the length L of the lament, should be estimated rst (in particular, bounded) by a power of the lament length determined from the packing rate. There are three basic regimes of packing: Linear, Laminar, and Globular. The bounds (which are predictions if the systems are random) for the associated growth rates of average crossing numbers are L1 , L log(L), and L4/3 . When the growth rates dier from these, we can conclude that the lament system involves some bias in how the lament is arranged within the given packing regime. In the protein example above, the packing is laminar, and we expect the average crossing number to be on the order of L log(L). There is another regime for packing that we can view as a theoretical upper bound of density: ignore any physical thickness of the laments and allow an innite length of string to be packed in a nite volume. In this Innite Density regime, we see tangling on the order of L2 . In Figure 3, we see a few of the many ways laments can be packed. The chart in Figure 4 lists the geometry expectations/bounds for these and other situations. Example (N), a neat coil of rope, is an extreme situation where the packing is spatially dense, yet so orderly that there is essentially no tangling. On the other hand, the orderly

GREGORY BUCK AND JONATHAN SIMON

packing in Example (F) achieves the upper bound L4/3 , which would also be achieved by rope that is randomly arranged but tightly packed in a space-lling way in a conned box.

Figure 3. There are many ways that laments can be packed in space.

MODELS OF ENTANGLEMENT

Figure 4. We want to understand the topological complexity associated to each kind of packing.

GREGORY BUCK AND JONATHAN SIMON

1.1. How to measure tangling? The topological knot or link type seems intuitively to be the fundamental lower bound on how much entanglement there is in some lament system. [We still can wonder about the question of dening how tangled is a given knot type: algebraic invariants? minimum crossing number? unknotting number?] But if a lament is an open string or complicated unknot (as in Figure 5), then there is no topological complexity even though the lament(s) may appear to be very much tangled.

Figure 5. This curve is topologically trivial but, intuitively, very tangled. (Example suggested for knot-energy ows by M. Freedman, with some added twists) So topological knot-type does not perfectly capture our intuitive idea of entanglement. But it is a fundamental lower bound. The other extreme of seemingly natural measures of entanglement is the observed crossing-number: count the number of times a lament crosses over itself (or over another lament if we are studying several interacting strings). However, this number varies according to how we look at the curve. For example, the curve shown in Figure 6 has 42 crossings as seen from one direction, and just 5 when seen from another.

Figure 6. This curve looks very dierent when seen from dierent directions

MODELS OF ENTANGLEMENT

For this reason, people think in terms of averaging the number of crossings over all directions of viewing, the so-called average crossing number. In the example of Figure 6, the average crossing number is approximately 25.7. 1.2. Dene ACN. The denition is made precise as an integral over the 2-sphere. For each direction vector v S 2 , let c(K, v) denote the number of crossings in the curve K as seen from direction v. Assuming K is [piecewise] smooth, this function is nite for all v except perhaps for a set of measure zero in S 2 . So the following integral makes sense: Definition (Average Crossing Number). 1 c(K, v) dS2 . ACN(K) = 4 vS2 The average crossing number for any particular curve of a given knot type is obviously an upper bound for the minimum crossing number of the knot type; in fact, it is always strictly larger [7]. We can to rewrite the ACN integral in a form that is useful for getting bounds [4]. Dene a map : K K S 2 by xy (x, y) = . |x y| View this function as a parameterization of S 2 from the domain K K, and think of calculating the area of S 2 using this parameterization. The function over-counts points of S 2 : a point v S 2 is hit q times precisely when c(K, v) = q. The weighted area of S 2 , divided by the area of S 2 , is thus equal to the average crossing number. The integrand in the integral below is the Jacobian stretching factor (with respect to the area element dx dy on K K) for the given parameterization . We have (1) ACN(K) = 1 4 | < Tx , Ty , x y > | dy dx , |x y|3

xK

yK

where Tx , Ty are the unit tangents at x, y and < u, v, w > is the triple scalar product (u v) w of the three vectors u, v, w. The truth of entanglement of a given curve lies somewhere between minimum crossing number of a topological knot-type and the average crossing number of the given curve. We are focusing in this paper on calculating, or estimating, ACN. According to ACN, the curve in Figure 7 is approximately three times as tangled as the curve in

GREGORY BUCK AND JONATHAN SIMON

Figure 9. However, we also should point out that the curve in Figure 8, which is a torus [un-]knot of type (12, 1) has approximately the same ACN.

Figure 7. This curve is topologically trivial, geometrically very tangled, and has average crossing number 75.5

Figure 8. This untangled looking curve has the same ACN as the tangled curve in Figure 7

Figure 9. This curve has ACN 25.7

MODELS OF ENTANGLEMENT

1.3. ACN is related to writhe and linking number. If we give the curve(s) orientation(s), then each crossing has a sign, 1. For any projection, we can count the net total number of crossings, which might be positive, negative or zero. The average of this signed crossing number, taken again over all spatial directions, is the average writhe for one curve, or the linking number for two. The double-integral formula for ACN is just a slight adaptation of Gausss formula for the linking number of two (oriented) curves J, K: Linking number(J, K) = < Tx , Ty , x y > 1 dy dx . 4 xJ yK |x y|3

The only dierence is that ACN uses the absolute value in the numerator. Because ACN is so close conceptually to writhe and linking number, and the integrals are so similar, it is not surprising that any method for bounding ACN probably also can be adapted to give a bound for writhe or linking number. Typically, the writhe (resp. linking number) is smaller in a fundamental way than ACN. We will discuss this more later. 1.4. Dene Rope-Length. Suppose we are studying a situation involving some lament(s). If the total length of the lament(s) is very small, then, intuitively, there cannot be very much tangling. So the length of the lament(s) is an important parameter in controlling the entanglement; call this L. We want to nd bounds for the tangling of the form This task seems hopeless as stated. We can make knots of arbitrarily high complexity out of a tiny length of string: we just need to make the string thin enough. So to model solid physical laments, we need to keep track of the thickness of the lament, along with its length. For any smooth curve K in 3-space R3 , there is some > 0 such that the disks of radius centered at points of K, normal to K, are pairwise disjoint. Call such a radius good. In that case, the disks combine to form a tubular neighborhood of K. Let R(K), the thickness radius of K, be the supremum of all good . We dene the rope-length of K as arc length of K EL (K) = . R(K) ACN function of L .

10

GREGORY BUCK AND JONATHAN SIMON

Figure 10. Dening the thickness radius of a curve If the the rope-length of a lament in space is small, i.e. the thickness radius is a large fraction of the actual length, then the lament looks like the center-line of a short, thick piece of rope. So it is intuitively plausible that we can nd bounds for how complicated the tangling can be (in particular, the average crossing number of K) in terms of EL (K). Remark. One can extend the denition of thickness radius and ropelength to more general situations such as polygons or other curves that are less than C 2 smooth, as well as modify the denition to pay more or less attention to local vs. global behavior of a curve [6, 11, 8, 5]. 2. Questions to ask about ACN Once we agree that the laments in some real, or virtual, experiments have some amount of thickness, then we can ask questions about how the tangling complexity grows as the rope-length increases. In particular, What is an upper bound for ACN in terms of rope-length EL ? What growth rates do we expect for random laments? What growth rates do we see in particular controlled situations? It turns out that there are fundamentally dierent ways that a lament can be packed in space. Each of these regimes of packing produces its own answers to the above questions. We also will think about bounds in the case of zero thickness. In this paper, we will consider some of the possible situations, to illustrate how one might

MODELS OF ENTANGLEMENT

11

do these kinds of analyses. In particular, we will describe some simple models that convey the right intuition and clarify the questions one should ask about real physical lament systems. In a real physical system, it might not be clear just what packing regime is involved. For example, if one is studying a collection of wiggly strands (e.g. some protein studies), can they best be understood as having zero thickness or some positive thickness? Do the laments pack densely so as to ll space or tend to escape and stretch out? Referring to table 4, example (G) illustrates positive thickness + dense packing; example (H) is positive thickness + avoid packing; example (B) is zero thickness + forced dense spatial packing; and example (P) is zero thickness + no forcing of packing. The Gaussian random ight in example (P) is perhaps the most subtle, since is exhibits similar behavior that seems to be a tipping point between globular and linear packing; it is somehow similar to the orderly laminar packing illlustrated in example (Q). See the paper [3], and references therein (also earlier mentioned [9]), for quantitative discussions of competing models for actual protein packing and entangling. We will rst dene the several regimes, then describe the ACN behavior in each. The discussions here emphasize intuitive understanding and simple models. For more careful general proofs, see [2] or [12], and also see [1]. The papers [1, 12] contains detailed proofs for bounding ACN in terms of ropelength in the globular regime. It is easy to slightly modify the nal section of [2] to get the desired bounds for ACN in the other regimes: In that section of [2], just multiply the Illumination bound by EL instead of multiplying it by the total curvature. 3. Four regimes of packing The same piece of rope can be packed in space in many ways. In the presence of lament thickness, there are three basic homogeneous packing regimes: Linear, Laminar, and Globular, each having a characteristic behavior of ACN. The fourth regime is laments with zero thickness, which can provide ultimate upper bounds. 3.1. Linear regime. The lament is packed in space in such a way that the amount of lament within any xed distance from a point on the curve is roughly proportional to that distance. This is illustrated in Figure 11, which shows a long iterated connected sum of trefoil knots. In this situation, the amount of lament that lies

12

GREGORY BUCK AND JONATHAN SIMON

between spatial distance r and r + 1 relative to any given point is constant (or, more realistically, is bounded between two constants).

Figure 11. Example of linear regime packing. 3.2. Laminar regime. See Figure 12. The lament is packed in space in an essentially 2-dimensional way. The amount of lament length within any xed distance r of a point is roughly proportional to the area of a patch of surface of area r r. The amount of lament length that lies between distance r and r + 1 relative to any given point is essentially proportional to r.

Figure 12. Example of laminar regime packing 3.3. Globular regime. See Figure 13. The lament is packed in a (nearly) space-lling way, so as to surround each of its points with a positive density of length per volume. That is, the amount of lament length that lies within a solid ball of radius r around any point is essentially proportional to r 3 , and the amount of lament length between distance r and r + 1 relative to any given point is proportional to r 2 . 3.4. Zero-thickness, or innite density, regime. In the example shown in Figure 14, we generate paths by connecting consecutive points (chosen randomly) on the boundary of a given ball. If the segments have no thickness, then we can keep on drawing sticks forever, with the path (generically) never crossing itself.

MODELS OF ENTANGLEMENT

13

Figure 13. Example of globular regime packing

Figure 14. Innitely dense packing in the zerothickness regime 4. Answers: How does ACN grow with length in the various regimes. The case of zero thickness is perhaps easiest to understand. In our example above, where a polygonal curve is growing by successive random chords in a ball, each stick interacts strongly with all the others, so the expected ACN should look like the square of the number, N, of sticks. Here is how we can quantify that intuition and get the desired estimate for ACN in terms of length. Consider the original denition of ACN. How many crossings do we expect to see from some direction? Once N is large, then looking from any direction, we expect to see at least a certain fraction, p, of the sticks having projected length greater than, say, 1/2. Then we have a variation of the Buon needle problem: Two sticks of length 1/2 contained in the unit disk have some probability q > 0 of crossing. So, for each pair of sticks, the probability of seeing them

14

GREGORY BUCK AND JONATHAN SIMON

cross from any given direction is > p2 q. This gives N 2 > ACN > N(N 1) 2 pq. 2

which says ACN is of order N 2 . We can replace N by L, since the total length L of such a polygon is indeed proportional to N. The 4 average length of a random chord of the unit sphere is 3 , so L 4 N. 3 The integral formula for ACN helps explain why the situation is dierent for thick laments. If the sticks have some nite thickness, and they are not allowed to overlap, then eventually we would have to move to a larger ball to contain more sticks. In the integral formula (1) for ACN, the integrand decays with |x y|2. So instead of having ACN grow like L2 , we will see a lower power growth rate. Here are some answers that we will see in the subsequent discussion. For simplicity, assume that the world has been rescaled to have always R(K)=1. This makes the rope-length = the actual length, L, of K. max random regime ACN ACN 1 2 max innite L 2 linear L laminar L log L globular L4/3 random Writhe max

5. Deriving bounds We have described earlier the references (and modications) for careful proofs of the ACN bounds in various packing regimes. Instead of just repeating those proofs here, we instead develop some very simple models that we hope will give a better intuitive understanding. In particular, the models will allow us to estimate the behavior of some random systems. 5.1. An initial step for all the regimes. Before we can pass to simpler models, we need to handle one technical detail. First, rewrite the ACN integral in a slightly dierent form: Let Uxy denote the unit vector xy Uxy = . |x y| So the formula is

MODELS OF ENTANGLEMENT

15

ACN(K) = 1 4
xK yK

The ACN integral includes two kinds of contributions, which are controlled dierently. We distinguish between pairs of points (x, y) where x and y are close in the sense of arc-length along the curve (specically where arc(x, y) ) and pairs (x, y) where x and y are further apart. When the point y is close to x, the denominator is close to 0, and one even has to worry if this improper integral converges. The key is that when x and y are close, then Tx and Ty are close, so the cross product Tx Ty is small in size. At the same time, that cross product is perpendicular to the tangent vectors, and so nearly perpendicular to U. That makes the numerator approach zero as fast as the denominator. If we are assuming the laments have xed thickness 1, this gives a bound on the curvature of the curves, which gives a bound on how much U and Tx can dier. That, in turn, gives a uniform bound on the ACN integrand for the case where x and y are close in arc-length along a curve. A careful analysis gives that for smooth curves, the set of pairs 3 {(x, y) : arc(x, y) } contributes 8 (L) to ACN. In the linear regime, the coecient 8 would have to be added to whatever coecient we get for L coming from points that are further apart along the curve. In the other regimes, this contribution from pairs of nearby points gets eaten up by the rest of ACN, which involves a higher power of L. (So again, one has to adjust the coecient of that power of L to consume the linear contribution.) We are left with the problem of understanding the (dominant) part of the ACN integral that comes from pairs (x, y) where arc(x, y) . Another theorem about thickness of laments [10] says that for such points, the spatial distance |x y| is 2. On the other hand, the numerator in the ACN integral (5.1) is the triple scalar product of three unit vectors; so it is 1. Thus the task of bounding ACN reduces to the task of bounding an integral of the form
x,y
3

| < Tx , Ty , U > | dy dx , |x y|2

1 dx dy , |x y|2

where x and y are points on the lament(s) where we knot the spatial distance between them is bounded away from zero.

16

GREGORY BUCK AND JONATHAN SIMON

Note that if we are studying polygons (with thickness) instead of smooth curves, then ACN of consecutive sticks is 0. So polygon ACN also reduces to understanding something that looks like 1/|x y|2. Just to keep the arithmetic as simple as possible, we will only assume |x y| 1 rather than the available bound 2. 6. A very simple model Here is an extremely simplied model of entanglement in the zero-thickness, densely packed regime. Consider a system of n sticks drawn as horizontal diameters of a vertical cylinder. Each stick is at a dierent height, so no two sticks intersect. Furthermore, the sticks are rotated so no two are parallel. Seen from above, each stick crosses over or under each other stick. If we want to model the crossings between two dierent laments, then assign to each stick an attribute color = red or blue. If we want to study linking number (or writhe), assign to each stick an attribute orientation. See Figure 15.

Figure 15. Model entanglement with two kinds (colors) of oriented sticks in a cylinder. In this simple model, each stick crosses over/under every other stick. This model is an abstraction and simplication of the situation where we have one or two long laments, somehow entangled. The total number of times sticks cross over each other is supposed to be analogous to the average crossing number (or, with orientations, the writhe or linking number) of a real lament ensemble. We will do some exercises with this model. In the following example, we compare the crossing number of two laments with the linking number of the two laments, interpreted here as the algebraic crossing number of oriented red and blue sticks.

MODELS OF ENTANGLEMENT

17

Although we are not including thickness at this point, this model is a template for studying that situation as well. 6.1. Crossing number vs. linking number in the stack-of-thin-sticks model. We have N sticks; there are 2N ways to assign color red or blue to each stick. Of these, there are N ways r to select r sticks to be red, with the rest blue. Since each red stick interacts with each blue stick in the sense that they cross (as seen from above), for each of the N color assignment types, we see r (r)(N r) crossings. Assuming the colors are assigned randomly, that is randomly with each color assignment equally likely, we would like to know the average over all possible color assignments of the (dierent-color) crossing numbers. Let us denote this average as CRB . Combining the above counts, we have CRB =
N N r=0 r

(r)(N r) 2N

1 = N(N 1) . 4 In particular, according to this model, the average number of crossings of one lament vs. another is expected to be on the order of the square of the total length of the laments and half the total crossing number of the (uncolored) system. This is an extremely naive model, but as indicated in the example at the beginning of Section 4, we believe it captures the essential behavior of more complicated models. Now consider what happens if we count the crossings with orientations. Each crossing gets a value, 1 in the usual way. We have N sticks. There are 4N ways to assign colors and orientations to the sticks. Let us denote color red by the number +1 and blue by number 1, and let ci be the color assigned to stick i. Similarly, use xi = 1 to denote the orientation of stick i. The, for a given assignment of colors and orientations, the algebraic sum of the crossings, i.e. the linking number of the ensemble, is is 1 Lk() = 2
n1 n

j=1 i=j+1

xi xj (ci cj ) .

Now consider all of the 4N possible labelings of N sticks. We get a distribution of values that we can analyze statistically. The analysis is

1 Remark. The 2 is not to correct for duplication; it compensates for the fact that (ci cj ) = 2.

18

GREGORY BUCK AND JONATHAN SIMON

elementary (we thank statistics colleague R. Russo) but takes a bit of algebra. For each pair (j, i), the number xi xj (ci cj ) is 2, with equal probability. If the values for dierent pairs (j, i) were independent of each other, then this would be simple coin-ipping. However, the values are not independent: for example, if the (j, i) term and (j, k) term are both zero [same colors], then so is the (k, i) term . Nevertheless, the distribution behaves a lot like coin ipping. The mean is, of course zero, and we can calculate the standard deviation exactly . By symmetry, the mean is 0. Write out the square of Lk() explicitly as an iterated sum. Simplify, using the fact that each of the number bits is 1. Use symmetry and a little induction to show that most of the terms add to zero; do a little more arithmetic to nish calculating the variance 2 . The distributions are (based on calculating exactly the distributions for several values of N) enough like normal distributions to justify assuming that the standard deviation is a good approximation for the mean absolute deviation, i.e. for the average magnitude of the linking number. We obtain the following: For close packing of two laments with 0 thickness, Total crossings = N (N 1) . 2 CRB = (N )(N 1) . 4 Lk() = 0. Lk 2 () = 2 = (N )(N 1) . 2 | Lk()| =
(N )(N 1) 2

1 2

CRB .

The last statement is another instance of the well known phenomenon that for many random processes, the average drift away from the mean in n steps is on the order of n. 6.2. Crossing number in the stack-of-THICK-sticks model. Here we model the Linear regime by assuming that the sticks in the previous model have some xed amount of thickness, let it be 1. We can use the previous formulas for the crossing and linking numbers of the zero-thickness stick ensemble to give expressions for the various kinds of tangling in the case of thick sticks. But now, the summands in our estimate of ACN now have a damping-out eect proportional to 1/dist2 , giving formulas as follows:

MODELS OF ENTANGLEMENT

19

1 Lk() = 2

N 1 j=1

xi xj (ci cj ) . (i j)2 i=j+1 (ci cj ) . (i j)2 i=j+1 xi xj . (i j)2 i=j+1


N N N

CRB () =

1 2

N 1 j=1

W rithe() =

N 1 j=1

ACN () =

N 1 j=1

The ACN estimate gives a bound for all the others, and is simple to calculate, since it does not depend on the labels chosen for . By collecting terms, we have ACN () = = (N)
2

1 . (i j)2 i=j+1

2 N 6

N 1 N k k=1 k 2 N 1 1 N 1 1 k=1 k 2 k=1 k

ln(N) .

The 6 N term dominates for large N, so we see that in the linear regime, with laments having positive thickness, average crossing number is at most proportional to rope-length. Remark. Our stack-of-sticks model allows all segments of the lament(s) to interact with full strength, limited only by their distance apart. In actual ACN calculations, there also is the eect of relative angles between the sticks that would aect the numerator in the ACN integral. So the behavior of the stack-of-sticks model is an upper bound for other situations in the linear regime. But so long as the direction vectors for the laments have a lot of freedom, we expect the stack-of-sticks model to give the correct exponent for L, with appropriate corrections needed for the coecients.

20

GREGORY BUCK AND JONATHAN SIMON

6.3. Model for the laminar regime. (See Figure 16) Proposition. In the laminar regime, the bound on ACN (and the expected behavior for random systems) is . ACN (L)(log L)

Proof. The amount of lament that lies within a given distance d from any one point x on the lament is proportional to d2 . And the amount of lament that lies at distance d from x is proportional to d. Let D be the radius around x that is just large enough to enclose the lament. So the length L (or number of sticks N) is proportional to D. Thus we can estimate ACN as
D

N
k=1

1 N log(N) . d2

Figure 16. Sticks model of the laminar regime The proposed protein packing in Figure 2 does not look like Figure 16. But it still is an example of the laminar regime. The amount of lament that lies within any xed distance D of a typical point on the lament is approximately proportional to D 2 .

MODELS OF ENTANGLEMENT

21

6.4. Model for the globular regime. (See Figure 17)

Figure 17. Sticks model of the globular regime Proposition. In the globular regime, the bound on ACN (and the expected behavior for random systems) is . ACN L(4/3)

Proof. The amount of lament that lies within a given distance d from any one point x on the lament is proportional to d3 . And the amount of lament that lies at distance d from x is proportional to d2 . Let D be the radius around x that is just large enough to enclose the lament. So the length L (or number of sticks N) is proportional to D 3 . Thus we can estimate ACN as
D

N
k=1

d2

1 N (4/3) . 2 d

22

GREGORY BUCK AND JONATHAN SIMON

6.5. Orderly packing vs. random packing. We conclude with an example where packing pattern and packing density both matter for ACN. We have noted before that the actual behavior of a system does not depend only on the packing regime. Within a given regime, it is possible to have laments packed so there is much less entanglement than random packing in that regime might suggest. Suppose a lament is packed, in the globular sense, but the packing is an orderly coil (e.g. as in Figure 18). How dierent would the ACN be, compared to a lament packed in a more random space-lling way?

Figure 18. Neat coiling in the globular regime It turns out that the answer seems to depend on the shape of the container. A neat coil of rope, stored in a tall thin container, will have higher ACN than if the rope were jumbled around. On the other hand, in a short wide container, if we allow the sticks (i.e. line elements of a smooth curve) to have random orientations in 3-space, we get higher ACN than if the sticks are all parallel to the base plane. For laments that ll the container, the local contributions to ACN for random orientations vs. neatly coiled laments compare as follows. The ratio of local contributions to ACN(random dense packing) ACN(coiled dense packing looks like the graph below. (The horizontal axis is the ratio of horizontal separation-to-vertical separation). When we average this local ratio over containers of dierent shapes, we nd that for a cube (presumably then also a symmetric spherical ball), the ACN for random packing would be around 3 times as large

MODELS OF ENTANGLEMENT

23

Figure 19. ACN local advantage of random vs neat coiling as a function of (width)/(height).

as what we would see for an orderly spool (as in Figure 18). In a container whose hight is around 20 times its width, we would nd the ACN for orderly coiling and for random wiggling about the same, with the orderly coiling producing more ACN if the container were even more eccentric. 6.5.1. Why this happens. Think of a sticks model for the problem. Assuming the density distributions are the same, the dierence between the ACNs in the two situations lies in the numerator of the ACN integrand. Suppose one stick is near the origin, with a given direction u in the xy-plane. Consider another stick at some point x = (x, y, z) R3 , with direction vector v. Let w be the unit vector (1/|x|)x. The ACN numerator is a volume, |(u v) w|. These volumes have one average when the vector v ranges over all 3-dimensional directions, and a dierent average when v ranges over horizontal 2-dimensional directions. The ratio of the two average volumes should tells us the ratio of the two local contributions to ACN. This ratio is sensitive to y/z. When the point x is directly above the base stick, the horizontal direction vectors produce more crossing with the base stick. When y (0.37)z, the two averages become equal. And when the point x is laterally further away from the base stick, then 3-dimensional distribution of directions for the x stick produces more ACN.

24

GREGORY BUCK AND JONATHAN SIMON

7. Additional Acknowledgements Several gures and calculations of average crossing number were done using R. Schareins KnotPlot [13]. W. Hager helped conrm the behavior of ACN in several situations. R. Russo provided a very helpful statistical argument. References
[1] G. Buck and J. Simon, Thickness and crossing number of knots, Topology and its Applications (91), 1999, 245257. [2] G. Buck and J. Simon, Total curvature and packing of knots, Topology and its Applications (154) 2007, 192204. [3] A. Dobay, J. Dubochet, A. Stasiak, and Y. Diao, Scaling of the average crossing number in equilateral random walks, knots, and proteins, Physical and Numerical Models in Knot Theory, Eds. J. Calvo, K.C. Millett, E.J. Rawdon, and A. Stasiak, World Scientic Publ. (Series on Knots and Everything vol. 36), 2005, 219232. [4] M. Freedman, Z.-X. He, and Z. Wang,On the Mbius energy of knots and o unknots, Annals of Math. (139), 1994, 150. [5] Y. Diao, C. Ernst, and E. J. Janse van Rensburg, Thicknesses of knots, Math. Proc. Camb. Phil. Soc. (126), 1999, 293310. [6] O. Gonzalez and J.H. Maddocks, Global curvature, thickness and the ideal shapes of knots, Proceedings of the National Academy of Sciences, USA, 96 (1999), 47694773. [7] This proposition is attributed to N. Kuiper: Given a space curve K representing a nontrivial knot type [K], there is some direction from which the apparent crossing number of K is strictly larger than the minimum crossing number of the knot-type. [8] R. Kusner and J. Sullivan, On distortion and thickness of knots, Topology and Geometry in Polymer Science, Springer Verlag Publ., 1998, 6778. [9] J. Liang, J. Zhang and R. Chen, Statistical geometry of packing defects of lattice chain polymer from enumeration and sequential Monte Carlo method, Journal of Chemical Physics (117), 2002, 35113521. [10] R. Litherland, J. Simon, O. Durumeric, and E. Rawdon, Thickness of knots, Topology and its Applications (91), 1999, 233244. [11] E. J. Rawdon, Approximating smooth thickness, J. Knot Theory and its Ramications (9), 2000, 113145. [12] E. Rawdon and J. Simon, Mbius energy of thick knots, Topology and its o Applications (125) 2002, 97109. [13] R. Scharein, KnotPlot software system. See http://www.pims.math.ca/knotplot/KnotPlot.html G. Buck. Department of Mathematics, St. Anselm College, Manchester NH 03102. gubck@anselm.edu J. Simon. Department of Mathematics, University of Iowa,Iowa City IA 52242 jsimon@math.uiowa.edu

You might also like