1:
2:
3:
4:
5: \documentclass[12pt]{amsart}
6: \usepackage{latexsym}
7: \usepackage{amsthm}
8: \usepackage{amsmath}
9: \usepackage{epsfig}
10: \usepackage{amsfonts}
11: \usepackage{amssymb}
12: \newtheorem{theorem}{Theorem}
13: \newtheorem{lemma}{Lemma}[theorem]
14: \newtheorem{defn}{Definition}
15: \newtheorem{prop}{Proposition}[theorem]
16: \newtheorem{cor}{Corollary}[theorem]
17: \newtheorem{example}{Example}
18:
19:
20:
21: \def\R{\mathbb R}
22: \def\L{\mathcal{L}}
23:
24:
25:
26:
27:
28: \begin{document}
29:
30:
31:
32: \title{Total Curvature and Packing of Knots}
33:
34:
35: \author{Gregory Buck} \address{G. Buck, Department of Mathematics, St. Anselm
36: College,\newline Manchester, NH 03102.}
37: \author{Jonathan Simon}\address{J. Simon, Department of Mathematics, University of Iowa, \newline Iowa City IA 52242.}
38: \thanks {Research supported by NSF Grants DMS\,0107747 (Buck) and DMS\,0107209 (Simon). Email jsimon@math.uiowa.edu or gbuck@anselm.edu. We thank J. McAtee and R. Weiler for helpful comments.}
39:
40:
41: \begin{abstract} We establish a new relationship between total
42: curvature of knots and crossing number. If $K$ is a smooth knot in
43: $\R^3$, $R$ the cross-section radius of a uniform tube neighborhood
44: $K$,
45: $L$ the arclength of $K$, and $\kappa$ the total curvature of $K$, then
46:
47: \begin{center} crossing number of $K$ $< 4\,\frac{L}{R}\;
48: \kappa$\;.
49: \end{center}
50:
51: The proof generalizes to show that for smooth knots in $\R^3$, the crossing
52: number, writhe, M\"obius Energy, Normal Energy, and Symmetric Energy are all bounded by the product of total curvature and rope-length.
53:
54:
55:
56: One can construct knots in which the crossing numbers grow as fast as the
57: $(4/3)$ power of $\frac{L}{R}$. Our theorem says that such families must
58: have unbounded total curvature: If the total curvature is bounded, then the
59: rate of growth of crossings with ropelength can only be linear.
60:
61:
62: Our proof relies on fundamental lemmas about the total
63: curvature of curves that are packed in certain ways: If a long smooth curve
64: $A$ with arclength $L$ is contained in a solid ball of radius $\rho$, then
65: the total curvature of $K$ is at least proportional to
66: $L/\rho$. If $A$ connects concentric spheres of radii $a \geq 2$ and $b\geq
67: a+1$, by running from the inner sphere to the outer sphere and back again,
68: then the total curvature of $A$ is at least proportional to $1/\sqrt{a}$.
69: \end{abstract}
70:
71:
72:
73:
74: \maketitle
75:
76:
77:
78: \markboth{G. Buck and J. Simon}{Total Curvature and Packing of Knots}
79:
80:
81:
82:
83:
84:
85:
86:
87:
88:
89:
90:
91:
92:
93:
94:
95:
96: \section{Introduction} The total curvature of smooth closed curve in
97: $\R^3$ must be at least $2\pi$; this is a theorem of Fenchel
98: \cite{Ch,Fen51}. If the curve actually is a nontrivial knot, then the
99: Fary-Milnor theorem \cite{Ch,Fary, Fen51, MilnorTotalCurvature} says the
100: total curvature must be $> 4\pi$. Are there properties of the knot that
101: could guarantee larger total curvature? Successive composition \cite{Fox50}
102: or or other kinds of satellite constructions (\cite{Schubert54} together
103: with \cite{MilnorTotalCurvature}) will work. On the other hand, topological
104: complexity in the form of high crossing-number is not enough: it is well
105: known at least since \cite{Milnor53} that one can construct knots with
106: arbitrarily large minimum crossing-number represented by curves with
107: uniformly bounded total curvature. Here is one way to build them.
108:
109: \begin{example}[Knots with bounded total curvature] Fix any odd integer
110: $n$. Construct a smooth knot $K_n$, with minimum crossing number $n$, as
111: the union of four arcs $H,A,C_1, C_2$, where the total curvatures are
112: $\kappa(H) \to 0$ as $n \to \infty$,
113: $\kappa(A)=0$, and
114: $\kappa(C_1) \approx \kappa(C_2) \approx 2\pi$. Let $H$ be the circular
115: helix in $\R^3$ parametrized as $[\cos(t), \sin(t), n^2t]$, $t=0\ldots
116: n\pi$. {\rm The height coordinate $n^2t$ makes $\kappa(H)$ behave like
117: $1/n$ for large $n$. Using any exponent larger than 1, i.e.
118: $n^{1+\epsilon}t$, still makes
119: $\kappa(H) \to 0$.} Let $A$ be the central axis of the cylinder on which
120: $H$ runs. Let $C_1$ and $C_2$ be curves that smoothly connect the top of
121: $H$ to the bottom of $A$ and vice-versa. {\rm For large $n$, the tangent
122: vectors at the beginning and end of $H$ are nearly vertical. The arcs
123: $C_1$ and $C_2$ can be chosen to be almost planar-convex curves, with total
124: curvatures $\kappa(C_i) \approx 2\pi$.} {\rm Similarly,} for any $(p,q)$,
125: torus knots or links of type ($p$ meridians,
126: $q$ longitudes), can have total curvature close to
127: $2\pi q$ if they are drawn on a standard torus that is long and thin enough.
128: \end{example}
129:
130: In this paper, we show that examples of the preceeding kind are, in a sense,
131: the only kind possible. In order to represent an infinite family of knot
132: types with uniformly bounded total curvature, the knots must be ``long and
133: thin"; if we imagine them made of actual ``rope", then the ratio of length to
134: rope-thickness must grow without bound.
135:
136:
137: %%%%%%%%%%%%
138: %%%%%%%%%%%%
139: %If we form
140: %knots by successive composition, e.g.
141: %$K\# K\ldots \# K$, then any
142: %curves representing these knot types will have the property that total
143: %curvature
144: %$\to \infty$ as the number of factors increases (\cite{Fox50} or
145: %\cite{Schubert54}, combined with
146: %\cite{MilnorTotalCurvature}). By the same reasoning, if we form knots
147: %using some other kind of repeated satelite construction, then the bridge
148: %numbers, hence total curvature, grow without bound.|
149: %%%%%%%%%%%%
150: %%%%%%%%%%%%
151:
152: \begin{defn} Suppose $K$ is a smooth knot in $\R^3$. For $r>0$, consider
153: the disks of radius
154: $r$ normal to $K$, centered at points of $K$. For $r$ sufficiently small,
155: these disks are pairwise disjoint and combine to form a tubular neighborhood
156: of $K$. Let R(K), the {\em thickness radius} of $K$, denote the supremum of
157: such ``good" radii. The {\em ropelength} of $K$, denoted
158: $E_L(K)$, is the ratio
159: $$E_L(K)= \frac{\text {\rm total arclength of } K}{R(K)}\;. $$
160: \end{defn}
161:
162: The fundamental properties of thickness radius were developed in
163: \cite{LSDR}. The idea of using the ratio of length--to--radius to measure
164: knot complexity was introduced in \cite{BO1}, and this ratio, denoted
165: $E_L(K)$, is connected to other knot ``energies" in \cite{BS2, BS3} and
166: \cite{RSBound}. Variations on thickness are developed in
167: \cite{Mo,devrthickness,DEJ2,devrthickness2, GonzalezMaddocks99,KS2,
168: RawdonIdealKnots98, Rawdon2000}.
169:
170:
171: \begin{defn} Let $K$ be a smooth knot. From almost every direction, if we
172: project $K$ into a plane, the projection is regular, in particular there are
173: only finitely many crossings. We can average this crossing-number over all
174: directions of projection (i.e. over the almost-all set of directions that
175: give regular projections). This {\em average crossing number} is denoted
176: $\rm{acn}(K)$.
177: \rm {Certainly, the minimum crossing-number of the knot-type,}
178: $\rm{cr}[K]$, \rm {satisfies}
179: $\rm{cr}[K]
180: \leq
181: \rm{acn}(K)$. We shall rely on the formulation of $\rm{acn}(K)$ developed
182: in \cite{FHW}.
183: \end{defn}
184:
185:
186:
187:
188: Our main result is the following:
189: \begin{theorem}\label{MainTheorem} If $K$ is a smooth knot in $\R^3$, then
190: $$\rm{acn}(K) < 4\; E_L(K) \;\kappa(K)\;.$$
191: \end{theorem}
192:
193: The coefficient $c=4$ has been rounded up for simplicity. What matters is that the crossing number is essentially bounded by ropelength times curvature. In heuristic discussions, we may omit coefficients altogether.
194:
195: If we are given some family of knots in which total
196: curvature is uniformly bounded, while crossing number is growing, then the
197: ropelength must be growing at least as fast as the crossing numbers.
198: Alternatively, if the crossing numbers are growing faster than ropelength,
199: then the total curvatures must be growing fast enough to make up the
200: difference. We showed in \cite{BS2,BS3} that $\rm{acn}(K)
201: \leq E_L(K)^{4/3}$, and there are examples \cite{B2, CKS} where the $4/3$
202: power is achieved. In the particular examples of \cite{B2, CKS}, the knots
203: and links have evident growing total curvature; our theorem says that some
204: unbounded amount of total curvature must occur in any situation of
205: more-than-linear growth of crossings with ropelength.
206:
207: If we model a knot made of actual ``rope" as a smooth curve with a uniform
208: tube neighborhood, then the thickness (radius) $r$ of that rope is $\leq
209: R(K)$, so $E_L(K) \leq \frac{L}{r}$. Thus the theorem also holds with
210: $\frac{L}{r}$ in place of $E_L$.
211:
212:
213:
214:
215: \section{Lemmas on total curvature}
216:
217: The three lemmas in this section establish fundamental properties of smooth
218: space-curves, relating total curvature to packing, to oscillation relative
219: to a given point, and to the ``illumination'' of a given point. We deal in
220: this section with smooth space-curves, not assuming the curves are simple
221: or closed; and we make no use of thickness. Also we do not assume the
222: curves have finite length.
223:
224: To keep the arguments as simple as possible, we assume throughout the paper
225: that ``smooth" means smooth of class
226: $C^2$. The lemmas and theorem can be adapted for curves that are piecewise smooth. For a smooth curve
227: $A$, we denote the total curvature of $A$ by $\kappa(A)$.
228:
229: It is intuitively clear that if a long rope is packed in a small box, then
230: the rope must curve a lot. This fundamental lemma is an important
231: ingredient in our analysis of the interplay between ropelength, crossing
232: number, and total curvature.
233:
234: A ball of radius $\rho$ contains, of course, a diameter of length $2\rho$.
235: But once we postulate length $> 2\rho$, an arc in the ball must curve. In
236: this version, we use $L\geq 3\rho$, but any constant larger than $2$ will
237: produce some guaranteed amount of total curvature. Inequality (\ref{packinginequality1}) and the proof below are taken from \cite{chakerian64}, with a slight adjustment for non-closed curves. If the curve is closed, then the number $2$ can be omitted from (\ref{packinginequality1}).
238:
239: \begin{lemma}[Packing and curvature] \label{packing} Suppose $A$ is a smooth
240: connected curve of length
241: $L$, contained in a round 3-ball of radius $\rho$. Then
242: $\kappa(A)$ is approximately proportional to at least $L/\rho$. More precisely, letting $\kappa$ denote $\kappa(A)$, we have the following,
243:
244: \begin{equation}\label{packinginequality1}
245: L \leq \rho(\kappa +2)\;,\end{equation}
246: which gives
247: \begin{equation} \label{packinginequality2} L \geq 3\rho \implies \kappa
248: \geq 1\;. \end{equation}
249: \end{lemma}
250:
251:
252: \begin{proof}
253:
254: Translate the ball and curve so the center of the ball is at the origin.
255: Let $s \to x(s)$, $s \in [0,L]$, be a unit speed parametrization of $A$. Since $|x'(s)|=1$, we can write
256: $$L = \int_{s=0}^L x'(s) \cdot x'(s)\;.$$
257: Integrate by parts to get
258: $$L = x'(s)\cdot x(s) \mid_0^L - \int_{0}^L x(s)\cdot x''(s)\;.$$
259: Since $|x'(s)| = 1$, and $|x(s)| \leq \rho$, the first term is at most $2\rho$ and the second term is at most $\rho \kappa$.
260: \end{proof}
261:
262: Another basic way that a long curve is forced to have a guaranteed amount of total
263: curvature is if its distance from some given point oscillates. This is
264: captured in the next lemma.
265:
266: \begin{lemma}[Oscillation and curvature] \label{oscillation} Let $S_a, S_b$
267: be concentric spheres with radii $a<b$. Let $A$ be a smooth curve that
268: starts at a point of $S_a$, somewhere touches the sphere $S_b$, and ends at
269: a point of $S_a$. Then the total curvature is at least approximately on the order of $1/\sqrt{a}$.
270:
271: More precisely,
272: \begin{equation} \label{arcsininequality}
273: \kappa(A) \geq \pi-2 \arcsin(a/b)\;.
274: \end{equation} If $b \geq a+1$, then
275: \begin{equation} \label{oscillationinequality}
276: \kappa(A) >\frac{ 2
277: \sqrt{2}}{\sqrt{a+1}}\;.
278: \end{equation} If $a \geq 2$, the bound (\ref{oscillationinequality}), along
279: with simplifying the coefficient, gives
280: $$
281: \kappa(A) >\frac{2}{\sqrt{a}} \; .
282: $$
283: \end{lemma}
284:
285: {\bf Remark}. One can think of the arc in this lemma as being just contained in the
286: spherical shell bounded by the two given spheres, or as reaching out past
287: $S_b$, just so it returns back to end on $S_a$.
288:
289:
290: \begin{proof} We rely on \cite{MilnorTotalCurvature} to reduce the proof of
291: (\ref{arcsininequality}) to analyzing a certain triangle, and then calculate
292: (\ref{oscillationinequality}).
293:
294: Let
295: $p,q$ be the endpoints of
296: $A$,
297: $z$ a point of
298: $A
299: \cap S_b$,
300: $\bar{pz}$ and $\bar{zq}$ the line segments from $p$ to $z$ and from $z$ to
301: $q$, and let $P$ be the two-edge polygon $\bar{pz} \cup \bar{zq}$. Since
302: the polygon $P$ is an inscribed polygon of $A$, we know from
303: \cite{MilnorTotalCurvature} that $\kappa(A) \geq \kappa(P)$, so it suffices
304: to establish the desired lower bound for $\kappa(P)$. If either of the edges
305: of $P$ is not tangent to the sphere $S_a$, we can pivot the edge at point
306: $z$ to move the edge to tangency in a way that opens the angle
307: $\widehat{pzq}$, so reducing $\kappa(P)$. Thus, it suffices to prove the
308: lower bound for tangent two-edge polygons.
309:
310: The three points $p,z,q$ determine a plane; we wish that plane also would
311: include the center of the spheres. If not, then (keeping the two edges
312: tangent to $S_a$, and allowing the points of tangency and the angle
313: $\widehat{pzq}$ to change), rotate the plane of
314: $p,z,q$ (with axis of rotation the line through $z$ parallel to
315: $\bar{pq}$) until it does contain the center of the spheres. This
316: deformation also would increase the angle
317: $\widehat{pzq}$, and so decrease the total curvature of $P$. Thus, we are
318: reduced to the situation where $p,z,q$ and the center of the spheres are
319: coplanar, and $P$ consists of tangent lines to $S_a$ drawn symmetrically
320: from point $z$ on $S_b$. We then have a right-triangle with
321: $$\sin (\frac{\widehat{pzq}}{2}) = \frac{a}{b}\;,$$ which gives
322: (\ref{arcsininequality}).
323:
324: To derive (\ref{oscillationinequality}), first note that
325: $\pi-2 \text{arcsin }(a/b)$ increases as $(b-a)$ gets larger. So if we show
326: $\pi-2 \text{arcsin }(a/b) \geq \frac{2\sqrt2}{\sqrt{a+1}}$ for
327: $b=a+1$, then we will have that inequality for all $b\geq a+1$.
328:
329: Rewrite (\ref{arcsininequality}) in terms of
330: $a$ and
331: $b=a+1$,
332: $$
333: \kappa(A)\geq \pi-2\arcsin\left(1-\frac{1}{a+1}\right)\;.
334: $$
335: Now let $t=\sqrt{\frac{1}{a+1}}$ and check (analytically or
336: graphically) that
337: $$
338: \pi-2\arcsin(1-t^2) >(2 \sqrt{2})\;t\;.
339: $$
340:
341: \end{proof}
342:
343:
344: In the next lemma, we call the curve $Y$ instead of $A$, to help clarify
345: how the lemmas will be used later: We will prove Lemma
346: \ref{illumination} by applying Lemmas \ref{packing} and
347: \ref{oscillation} to subarcs $A$ of $Y$.
348:
349: Suppose
350: $Y$ is a smooth curve in
351: $\R^3$, and
352: $x_0$ is a point some finite distance from $Y$. The integral
353: \begin{equation}
354: \int_{y \in Y} \frac{1}{|y-x_0|^2}
355: \end{equation} can be thought of as measuring the ``illumination" of $x_0$
356: by $Y$.
357: \begin{lemma}[Illumination and curvature]\label{illumination} Suppose $Y$ is
358: a smooth curve in $\R^3$, and $x_0$ is a point such that
359: $\forall y \in Y$, $|y-x_0|\geq 2$. Then the illumination of $x_0$ by $Y$ is
360: bounded by the total curvature of $Y$. More precisely,
361:
362: \begin{equation} \label{illuminationbound}
363: \int_{y \in Y} \frac{1}{|y-x_0|^2} \leq c_1+c_2 \;\kappa(Y)\;,
364: \end{equation} where $c_1,c_2$ are universal constants independent of $Y$ {\rm (values $c_1=16$ and $c_2=43$ are sufficient).}
365: \end{lemma} Lemma \ref{illumination} is perhaps the most intricate part of
366: the paper. Before proving it, we present four special cases. The general
367: argument does not reduce to these special cases -- rather we include them
368: to give an intuitive sense of why the proposition might be true (the first
369: four), and some of the issues one needs to confront in building a proof (the
370: fifth).
371:
372: \subsection{Special cases for Lemma \ref{illumination}}
373: \subsubsection{A spiral to show the lemma is sharp in the power of $\kappa(Y)$.}\label{sharp} Let $Y$ be
374: the polar coordinates curve
375: $r=3-1/\theta$,
376: $\theta = 1
377: \ldots\Theta$. As $\Theta$ increases, the illumnination (of $x_0=$ the
378: origin) is aymptotic to $\frac{1}{3}\kappa(Y)$.
379:
380: \subsubsection {$Y$ is a ray} \label{raycase} Suppose $Y$ is a straight
381: line, starting at a point $2$ units from $x_0$ and aiming radially away from
382: $x_0$. Then the line integral is just
383: $\int_2^{\infty}{1/s^2} \;ds = 1/2$.
384: \hfill \qed
385:
386: \subsubsection {$Y$ is a straight line} Suppose $Y$ is a straight line,
387: infinite in both directions, and tangent to the sphere of radius $2$
388: centered at
389: $x_0$. Then
390: $$\int_{y \in Y} \frac{1}{|y-x_0|^2} =
391: \int_{-\infty}^{\infty}\frac{1}{4+s^2}\;ds\;\;=\;\;\frac{\pi}{2}\;.
392: $$ If the line is a finite segment, or the minimum distance from $Y$ to
393: $x_0$ is $>2$, then the integral is $< \pi/2$. \hfill \qed
394:
395: \subsubsection{$Y$ is a certain kind of polygon}\label{polygoncase} Suppose
396: $Y'$ is a polygonal path (or closed curve) consisting of
397: $e$ edges (of possibly varying lengths), such that each pair of consecutive
398: edges meets at a right angle. Form a smooth curve $Y$ by replacing the
399: corners of $Y'$ with small quarter-circles. Then, by the second special
400: case, each edge of
401: $Y$ contributes $< \pi/2$ to the illumination integral, so
402: \begin{align}\nonumber \int_{y \in Y} \frac{1}{|y-x_0|^2} < e
403: \;\frac{\pi}{2}
404: &= \kappa(Y) \;\;\;\text{ if the polygon has endpoints, or }
405: \\ \nonumber &{\;\;\;\;\;}\kappa(Y) + \pi/2 \;\;\;\text{ if the polygon is
406: closed.}
407: \end{align}
408: \hfill \qed
409:
410: \subsubsection{$Y$ is a monotone arc}\label{monotonecase} Suppose $Y$ is a
411: smooth curve, starting at a point
412: $y_0$ with $|y_0-x_0|=2$, with the property that the distance function
413: $|y-x_0|$ is monotone increasing on $Y$.
414:
415: For $n=2, 3, \dots$, let $B[n]$ denote the round ball of radius $n$ centered
416: at $x_0$, and let $S[n, n+1]$ denote the spherical shell with radii $n$ and
417: $n+1$. By our assumption of monotonicity, each intersection
418: $ Y \cap S[n,n+1]$ is a connected arc, which we denote $Y_n$. Then
419: $$\int_{y\in Y}\frac{1}{|y-x_0|^2} = \sum_{n=2}^{\infty}
420: \int_{y\in Y_n}\frac{1}{|y-x_0|^2}\leq
421: \sum_{n=2}^{\infty}\frac{\ell(Y \cap S[n,n+1])}{n^2}\;.$$
422: We would like to bound each of the numbers $\ell(Y \cap S[n,n+1])$, in
423: terms of total curvature of $Y$, somehow using Lemma
424: \ref{packing}. That lemma gives upper bounds for the lengths
425: $\ell(Y\cap B[n])$ in terms of total curvature, but doesn't explicitly bound
426: the amounts in given shells. We get around this problem by bounding (not the
427: illumination integral from $Y$ itself, but rather) the illumination integral
428: for a hypothetical curve $Y^*$ that is packed around
429: $x_0$ in such a way as to make the illumination integral as large as
430: possible subject to the constraints imposed by Lemma \ref{packing}. (In
431: this intuitive discussion of the monotone case, we will continue with the
432: image of a ``hypothetical curve". In the actual proof of Lemma
433: \ref{illumination}, we will be more rigorous.)
434:
435: For brevity, let $\kappa$ denote $\kappa(Y)$. Since $Y \cap {\rm int}\,B[2] = \emptyset$, we start with
436: $Y_2 = Y\cap B[3]$. By inequality (\ref{packinginequality1}),
437: $$ \ell(Y \cap B[3]) \leq 3 (\kappa + 2)\;.\\
438: $$ Similarly,
439: \begin{align} \nonumber
440: \ell(Y \cap B[4]) &\leq 4 (\kappa + 2)\;, \\ \nonumber
441: \ell(Y \cap B[5]) &\leq 5 (\kappa + 2)\;, \\ \nonumber &\text{etc}.
442: \end{align}
443: If the curve $Y$ does not actually achieve these bounds, then add extra
444: length (the ``hypothetical" curve $Y^*$) in each of the shells as needed to
445: actually reach these bounds. Since we are adding length to the $Y$ that
446: already exists, the illumination integral can only increase. Thus the
447: illumination for $Y^*$ is an upper bound for the illumination for $Y$.
448:
449: We have
450: \begin{align}\nonumber
451: \ell(Y^* \cap B[3]) &= 3 (\kappa + 2)\;,\\ \nonumber
452: \ell(Y^* \cap B[4]) &= 4 (\kappa + 2)\;,\\ \nonumber
453: \ell(Y^* \cap B[5]) &= 5 (\kappa + 2)\;,\\ \nonumber \text{etc.}
454: \end{align}
455:
456: Thus
457: \begin{align}\nonumber
458: \ell(Y^* \cap S[2,3]) &= 3 (\kappa + 2)\;,\\ \nonumber
459: \ell(Y^* \cap S[3,4]) &= 4 (\kappa + 2)\;-\;3 (\kappa +
460: 2)\;=\;(\kappa + 2)\;,\\
461: \nonumber
462: \ell(Y^* \cap S[4,5]) &= 5 (\kappa + 2)\;-\;4 (\kappa +
463: 2)\;=\;(\kappa + 2)\;,\\ \nonumber \text{etc.}
464: \end{align}
465:
466: And so,
467: $$ \int_{y \in Y} \frac{1}{|y-x_0|^2} \leq \int_{y \in Y^*}
468: \frac{1}{|y-x_0|^2} < \frac{3 (\kappa + 2)}{2^2}
469: \;+\;\sum_{n=3}^{\infty}\frac{ (\kappa + 2)}{n^2} < 2\kappa+3\;.
470: $$ \qed
471:
472:
473:
474:
475:
476:
477:
478: \subsection{Proof of Lemma \ref{illumination}}
479:
480:
481:
482: We begin as we did in section \ref{monotonecase}. For $n=2, 3, \dots$, let
483: $B[n]$ denote the round ball of radius $n$ centered at $x_0$, and let $S[n,
484: n+1]$ denote the spherical shell with radii $n$ and $n+1$. We need to bound
485: the total arclength of $Y$ contained in each shell $S[n,n+1]$, but we cannot
486: do this directly since we are not assuming monotonicity as in section
487: \ref{monotonecase}. The shell-intersections might consist of long arcs, or
488: might consist of unions of many short arcs, as
489: $Y$ meanders in space, close to, or far from,
490: $x_0$.
491:
492: When $Y$ is contributing to the integral by having long arcs close to
493: $x_0$, we can infer curvature from Lemma
494: \ref{packing}. If $Y$ is contributing to the integral by oscillating in and
495: out from $x_0$, we can use Lemma \ref{oscillation} to infer curvature.
496:
497: To implement this plan, and handle the problem of very small oscillations,
498: we are going to translate the problem into discrete combinatorics.
499:
500: \subsubsection{Assume finite length} If $Y$ has infinite length, express $Y$
501: as an increasing union of curves of finite length. Since the constants
502: $c_1, c_2$ do not depend on the curve $Y$, we can apply inequality
503: (\ref{illuminationbound}) to each of these and observe that both sides of
504: inequality (\ref{illuminationbound}) converge appropriately.
505:
506: \subsubsection{Cut $Y$ into small pieces} Pick any integer $M > \ell(Y)$,
507: and cut $Y$ into $M$ consecutive arcs $Y_i$ of equal length. Let $ \epsilon
508: $ denote the length of each sub-arc, and note $\epsilon<1$.
509:
510: We assign to each arc $Y_i$ a label $1,2,3,\ldots$ representing the shell
511: $S[n,n+1]$ that (perhaps only approximately) contains $Y_i$. Specifically,
512: if
513: $Y_i \subset S[n,n+1]$, assign label $n$. If $Y_i$ is not entirely contained
514: in one shell, then it must intersect a sphere $S[n]$; because
515: $\epsilon < 1$, $Y_i$ can intersect at most one sphere $S[n]$; we assign
516: that label $n$ to the arc. Note that if an arc $Y_i$ carries label
517: $n$, then $Y_i \subset S(n-1, n+1]$ and $Y_i \cap S[n, n+1] \neq \emptyset$.
518: The set of possible labels is $\left\{2, \dots, M+1\right\}$.
519:
520:
521: \subsubsection{Discretize the problem} For each integer $n$, let $\phi(n)$
522: denote the total number of arcs
523: $Y_i$ that are labeled $n$. Thus
524: \begin{equation}\label{firstsum}
525: \int_{y \in Y} \frac{1}{|y-x_0|^2} = \sum_i \int_{y \in Y_i} \frac{1}{|y-x_0|^2}
526: \;<\;\sum_{n\geq 2} \phi(n) \frac{\epsilon}{(n-1)^2}\;.
527: \end{equation} We also need the auxiliary function that counts the total
528: number of arcs that are labeled between $2$ and $n$. Define
529: $$ \Phi(n) = \sum_{j=2}^{n} \phi(j)\;.
530: $$
531: We proceed as follows:
532: \begin{enumerate}
533: \item Abstract the arc $Y$ as the string of integers $\mathcal{L}_Y=<a_1,
534: a_2,
535: \ldots, a_M>$, where $a_i$ is the shell label of $Y_i$.
536: \item Show $\mathcal{L}_Y$ is constrained in certain ways.
537: \item Find a bound for $\Phi(n)$, in terms of $\kappa(Y)$, using Lemma
538: \ref{packing} and Lemma \ref{oscillation}.
539: \item Note that the functions $\phi$ and $\Phi$ make sense for any finite
540: string $\mathcal{L}$ of integers
541: \item For any finite string $\mathcal{L}$ of integers $\geq 2$, define an
542: ``energy"
543: $$ E(\mathcal{L}) = \sum_{n\geq 2} \phi(n) \frac{\epsilon}{(n-1)^2}\;.
544: $$
545: \item Construct a string $\mathcal{L}^{*}$ of integers $\in \left\{2,
546: \dots, M+1 \right\}$ for which we know bounds on the numbers $\phi(n)$, and
547: for which we know
548: $E(\mathcal{L}) \leq E(\mathcal{L}^{*})$.
549:
550:
551: \item Find a bound for the value $E(\mathcal{L}^{*})$, which is an upper
552: bound for the final sum in (\ref{firstsum}), of the form we want.
553: \end{enumerate}
554:
555:
556: Let $\mathcal{L}_Y$ be the the string of shell labels associated to $Y$. In
557: order to bound $\Phi(n)$, we first establish certain properties of
558: $\mathcal{L}_Y$.
559:
560: \subsubsection{Constraints on $\mathcal{L}_Y$}
561: To shorten formulas in the rest of the
562: proof of Lemma \ref{illumination}, we will use $\kappa$ to denote $\kappa(Y)$.
563:
564: Since the arcs $Y_i$ are listed in their order along $Y$, each intersection
565: $Y_i \cap Y_{i+1}$ is nonempty. A substring such as $\langle 34434543 \rangle$ is possible. On the other hand there
566: cannot be a substring such as $\langle 34435543 \rangle$, because subarcs
567: with labels $3$ and $5$ are contained in the disjoint half-open shells $S(2,4]$ and
568: $S(4,6]$. So the first constraint is:
569: \begin{itemize}
570: \item $\mathcal{L}_Y$ must consist of contiguous labels.
571: \end{itemize}
572:
573: For a given value of $n$, there cannot be too many long substrings of
574: $\mathcal{L}_Y$ consisting of labels $\leq n$. A substring of
575: $\mathcal{L}_Y$ consisting of $q$ symbols
576: $\leq n$ represents a connected arc $A\subset Y$ of length $q\epsilon$
577: contained in the ball $B[n+1]$. If $q$ is such that $q\epsilon \geq
578: 3(n+1)$, then, by Lemma \ref{packing}(\ref{packinginequality2}),
579: $\kappa(A) \geq 1$. Thus, for such $q$, we can have no more than $\kappa$ such strings. So the second constraint is:
580: \begin{itemize}
581: \item For each $n$, the string $\mathcal{L}_Y$ contains at most $\kappa$
582: pairwise disjoint substrings of length $\frac{3(n+1)}{\epsilon}$ consisting
583: of integers $\leq n$.
584: \end{itemize}
585:
586: In section (\ref{BoundPhiN}), we will apply this formula to substrings of $\mathcal{L_Y}$ with entries up through (n+1). We also want to phrase the bound in terms
587: of what $\mathcal{L}_Y$ {\em cannot} contain. Specifically,
588: \begin{itemize}
589: \item For each $n$, the string $\mathcal{L}_Y$ cannot contain $(\kappa + 1)$
590: pairwise disjoint substrings of length $\frac{3(n+2)}{\epsilon}$ consisting
591: of integers $\leq (n+1)$.
592: \end{itemize}
593:
594:
595: We obtain a third constraint, this time on ``jumps". If, in the string
596: $\mathcal{L}_Y$, we observe a substring $\langle n \ldots n+1 \ldots n
597: \rangle$, we cannot infer any particular contribution to total curvature that is independent of $\epsilon$. But if we
598: see
599: $\langle n \ldots n+2 \ldots n \rangle$, then we can. Let us call a
600: substring
601: $\lambda = \langle n \ldots n+2 \ldots n \rangle$ of $\mathcal{L}_Y$ a {\em
602: jump at level $n$}. Two jumps are {\em non-overlapping} if they are
603: disjoint, or meet in at most one term $a_i$ (of necessity, then, an endpoint
604: of each).
605:
606: An arc $Y_i$ with label $n$ has nonempty intersection with $S[n,n+1]$; an
607: arc with label $n+2$ intersects $S[n+2, n+3]$. Thus if $\lambda$ is a jump
608: at level $n$, then the subarc of $Y$ determined by $\lambda$ has a subarc $A$ that starts at the sphere
609: S[n+1] and reaches as far out as some $S[b]$, $b \geq n+2$, before heading
610: back to end at $S[n+1]$. By Lemma \ref{oscillation}, such an arc
611: contributes more than $\frac{2}{\sqrt{n+1}}$ to total curvature. Thus we
612: have our third constraint:
613: \begin{itemize}
614: \item For each $n$, the string $\mathcal{L}_Y$ cannot have $\frac{1}{2} \,
615: \kappa \, \sqrt{n+1}$ non-overlapping jumps of level $n$.
616: \end{itemize}
617:
618:
619: \subsubsection{Bound $\Phi(n)$}\label{BoundPhiN}
620: We now combine the constraints on substrings and jumps in
621: $\mathcal{L}_Y$ to get bounds on $\Phi(n)$.
622: \begin{prop} \label{Phibound} For the string $\mathcal{L}_Y$, for each
623: $n\geq 2$,
624: \begin{equation} \label{PhiboundInequality}\Phi(n) < 8 \kappa
625: \, \frac{n^{3/2}}{\epsilon}\;+\;6\frac{n}{\epsilon}.
626: \end{equation}\end{prop}
627: \begin{proof} Suppose, to the contrary, that for some $n$, $\mathcal{L}_Y$
628: does have that many symbols $2, 3, \ldots, n$. The bound
629: (\ref{PhiboundInequality}) was chosen to be a simple expression that, for
630: $n \geq 2$, dominates
631: $$
632: (\kappa + 1) \left( \frac{3(n+2)}{\epsilon} \right) \;+\; \left(
633: \frac{1}{2} \;\kappa \, \sqrt{n+1}\right) \left(\frac{3(n+2)}{\epsilon}\right).
634: $$
635:
636: Visualize the string $\mathcal{L_Y}$ so that, temporarily, only the symbols
637: $2,3,\ldots,n$ are visible. Parse these into pairwise disjoint substrings
638: of length $\frac{3(n+2)}{\epsilon}$. By assumption, we have (many) more than $\kappa+1$ of these substrings. So in the actual string $\mathcal{L}_Y$, a
639: number of these substrings must get broken up by inserted symbols $\geq
640: n+1$. Now make all the symbols $a_i=n+1$ in $\mathcal{L}_Y $ visible as
641: well. These certainly can break up substrings consisting only of symbols
642: $2,3,\ldots,n$, but they offer no improvement on our situation of exceeding
643: the total curvature bound:
644: we chose the lengths of the substrings to be large enough that even if they
645: were made from symbols $2,3,\ldots,n+1$, they would still each be
646: contributing $\geq 1$ to total curvature, so we cannot have more than
647: $\kappa$ of these. Thus we must have some symbols
648: $\geq n+2$ in the original string $\mathcal{L}_Y $ to break up a number of
649: the ``offending" substrings. How many of the substrings can be broken by
650: inserting symbols $a_j \geq n+2 \,$? Each offending substring that gets
651: broken this way represents at least one jump at level n. So we must have
652: have fewer than $\frac{1}{2} \, \kappa \, \sqrt{n+1}$ such interruptions.
653: We are assuming $\Phi(n)$ is large enough that the number of offending substrings is
654: greater than the number of possible interruptions plus the maxumum number we
655: could tolerate to be uninterrupted. We conclude that $\Phi(n)$ cannot be
656: that large.
657: \end{proof}
658:
659:
660: \subsubsection{Construct $\mathcal{L}^*$}
661:
662: We have completed Step 3 of our plan, and now proceed. The functions $\phi$
663: and $\Phi$ make sense for abstract finite strings $\mathcal{L}$ of integers:
664: $$\phi(n)= \text{number of symbols } a_i \text { of } \mathcal{L} \text {
665: that are}\; n\;;$$
666: $$ \Phi(n) = \sum_{j=2}^n \phi(j)\;.
667: $$
668: And we can define the ``energy"
669: $$
670: E(\mathcal{L}) = \sum_{n \geq 2} \phi(n) \frac{\epsilon}{(n-1)^2}\;.
671: $$
672:
673: We want to construct a string $\mathcal{L}^{*}$ whose energy we can bound,
674: but also whose energy is larger than $E(\mathcal{L}_Y)$. We do this by
675: successive modification of $\mathcal{L}_Y$. We will denote the new strings
676: $\L_2, \L_3, \ldots$, and denote the corresponding functions $\phi_2,
677: \Phi_2, \phi_3, \Phi_3$, etc. Let us also introduce notation for the bounds
678: in inequality (\ref{PhiboundInequality}):
679: $$
680: \beta(n) = 8 \kappa
681: \, \frac{n^{3/2}}{\epsilon}\;+\;6\frac{n}{\epsilon}\;.
682: $$
683:
684: The strings $\L_m$ will have the following properties:
685: \begin{itemize}
686: \item $\Phi_m(n)\leq \beta(n)$ for all $n$.
687: \item $\Phi_m(n) = \beta(n)$ for $n=2\ldots m$.
688: \item $E(\L_Y) \leq E(\L_m)$.
689: \end{itemize}
690:
691:
692: If we take a string $\L$ and change some symbol $a_i$ to a lower integer,
693: that increases $E(\mathcal{L})$. Also, if we introduce a new additional
694: symbol
695: $a_j$ somewhere in $\mathcal{L}$, that increases $E(\mathcal{L})$.
696:
697: We begin the construction by adjoining to $\L_Y$ enough terms $a_j = (M+1)$
698: to raise $\Phi(M+1)$ to equal $\beta(M+1)$. Call this string $\L_1$. All
699: the other strings $\L_m$ will be obtained by changing various terms of
700: $\L_1$ to lower values, thus raising energy while keeping $\Phi(M+1)$
701: unchanged.
702:
703:
704:
705: We know in $\L_Y$ that $\phi(2)=\Phi(2) < 8 \kappa
706: \, \frac{2^{3/2}}{\epsilon}\;+\;6\frac{2}{\epsilon}$. To construct $\L_2$,
707: change enough 3's in $\L_Y$ to 2 to bring the number of 2's up to
708: $\beta(2)$. [Note: For simplicity, we will use the bound itself, rather
709: than rounding up if it is not an integer.] Changing a 3 to a 2 has no
710: effect on $\Phi(n)$ for $n\geq 3$. If there are not enough 3's (i.e. if
711: $\Phi(3) < \beta(2)$), we change 4's to 2's. This increases $\Phi(3)$ up to
712: $\beta(2)$, still $< \beta(3)$; and values $\Phi(n)$ are unchanged for $n
713: \geq 4$. If there are not enough 4's, we change 5's, etc. Continuing in
714: this way, we obtain a string $\L_2$ with the properties
715: \begin {enumerate}
716: \item $\phi_2(2) = \Phi_2(2) =\beta(2)$,
717: \item $\Phi_2(n) \leq \beta(n)$ for $n \geq 3$, and
718: \item $E(\L_2) \geq E(\L_Y)$.
719: \end{enumerate}
720:
721: We next want to make $\phi_2(3)$ large enough that $\Phi_2(3)=\beta(3)$. So
722: we again change higher labels, first change 4's to 3's, then (if necessary)
723: 5's to 3's, etc.
724:
725: We continue inductively to construct $\L_3, \ldots, \L_{(M+1)} = \L^*$, so
726: that
727: \begin{enumerate}
728: \item For $n=2\ldots m$, $\Phi_m(n) = \beta(n)$.
729: \item For $n = (m+1)\ldots (M+1)$, $\Phi_m(n) \leq \beta(n)$.
730: \item $E(\L_m) \geq E(\L_Y)$.
731: \end{enumerate}
732:
733: Because we know exactly the values $\Phi^*(n)$, we can compute the values
734: $\phi^*(n)$ and so bound the energy.
735: $$
736: \phi^*(2) = \beta(2) = 8 \kappa
737: \, \frac{2^{3/2}}{\epsilon}\;+\;6\frac{2}{\epsilon}\;,
738: $$
739: and for $n \geq 3$,
740: $$
741: \phi^*(n) = \Phi^*(n)-\Phi^*(n-1) = \beta(n)-\beta(n-1) = 8 \kappa
742: \, \frac{n^{3/2}- (n-1)^{3/2}}{\epsilon}\;+\;\frac{6}{\epsilon} .
743: $$
744:
745: \subsubsection{Bound $E(\L^*)$}
746:
747: We bound the energy of $\L^*$ by passing to an infinite sum, so the value of
748: $M$ is immaterial. Since we know the values $\phi^*(n)$, we have
749: \begin{align}
750: \nonumber E(\L^*) &= \sum_{n=2}^{M+1} \phi^*(n) \frac{\epsilon}{(n-1)^2}
751: \\ &< \nonumber 8 \kappa 2^{3/2} + 12 + 8 \kappa \sum_{n=3}^\infty
752: \frac{n^{3/2}-(n-1)^{3/2}}{(n-1)^2} + 6 \sum_{n=3}^\infty \frac{1}{(n-1)^2}
753: \\ \nonumber &< 16 + 43\kappa.
754: \end{align}
755: \qed
756:
757:
758:
759: \section{Proof of Theorem \ref{MainTheorem}}
760:
761: As a preliminary step, rescale the knot so the thickness radius $R(K)=1$.
762: This has no effect on the total curvature or on the average crossing number,
763: and simplifies the ratio
764: $E_L(K)$ to just the length, $L$. We want to show
765: $$
766: \rm{acn} (K) \leq c\cdot L\cdot\kappa(K)\;,
767: $$
768: where $c$ is some coefficient that works for all knots.
769:
770: The average crossing number of a knot can be expressed as an integral over
771: the knot
772: \cite{FHW}, similar to Gauss's double integral formula for the linking
773: number of two loops. Specifically,
774: $$
775: \rm{acn}(K) = \frac{1}{4 \pi}\int_{x \in K}
776: \int_{y
777: \in K}
778: \frac{|<T_x\,, T_y\,, x-y>|}{|x-y|^3} \;\;,
779: $$
780: where $T_x, T_y$ are the unit tangents at $x,y$ and $<u,v,w>$
781: is the triple scalar product $(u \times v)\cdot w$ of the three vectors
782: $u,v,w$.
783:
784: Write the double integral
785: as a sum of two terms:
786: $$\text{Near}(K) = \int_{x \in K} \int_{\rm{arc}(x,y) \leq \pi}
787: \frac{|<T_x\,, T_y\,, x-y>|}{|x-y|^3} \;\;,$$ and
788: $$\text{Far}(K) = \int_{x \in K} \int_{\rm{arc}(x,y) \geq \pi}
789: \frac{|<T_x\,, T_y\,, x-y>|}{|x-y|^3} \;\;.$$
790:
791: We shall analyze these contributions separately in the next two sections,
792: and find bounds of the form
793: $$
794: \nonumber \text{Near}(K) \leq b_1\;L
795: $$
796: $$
797: \text{Far}(K) \leq c_1\;L + c_2\;L\;\kappa(K)\;.
798: $$
799: where the coefficients are independent of $K$. In each case,
800: we bound the inner integral and then multiply by $L$ to bound the double
801: integral.
802:
803: Combining Near and Far, we get a bound for any smooth curve $K$ of the form
804: $$\text{acn}(K) \leq aL + bL \kappa(K).$$ But if $K$ is a {\em closed}
805: curve, then by Fenchel's theorem, $\kappa(K) \geq 2\pi$. Thus letting
806: $c=\frac{1}{4\pi}\left(b+\frac{a}{2\pi}\right)$, we have $$\text{acn}(K) \leq c \; L \;\kappa(K)\;.$$
807: Using the values of $c_1$ and $c_2$ from Lemma (\ref{illumination}) and $b_1$ from section (\ref{BoundingNearK}), we get $c \approx 3.8$.
808: \qed
809:
810:
811:
812:
813: \subsection{Bounding $\text{Near}(K)$} \label{BoundingNearK}
814:
815: We shall show that the inner integral is uniformly bounded, independent of
816: $K$.
817:
818: For any smooth curve with thickness radius
819: $R$, it is shown in
820: \cite{LSDR} that the curvature at each point is at most $1/R$. So in the
821: present situation, we know that the curvature of $K$ is everywhere $\leq 1$.
822:
823: Let $\theta\to x(\theta)$ be a unit speed parametrization of $K$. So
824: $x'(\theta) = T_x$ and $|x''(\theta)| \leq 1$. We are studying points
825: $y$ for which $\text{arc}(x,y) \leq \pi$, so we can take for the parameter
826: set the interval $[0,\pi]$, with our starting point $x=x(0)$ and
827: $y=y(\theta)$ for some $\theta \in [0,\pi]$. Using the same parameter set,
828: let $\theta\to p(\theta)$ be an arclength preserving parametrization of the
829: unit semi-circle. Since the curvature of $K$ is everywhere bounded by the
830: curvature of the unit circle, Schur's theorem \cite{Ch} tells us that for
831: each $\theta$,
832: $$
833: |x(\theta) - x(0)| \geq |p(\theta) - p(0)|\;,
834: $$
835: That is,
836: $$
837: |y-x| \geq \sqrt{2-2\cos \theta}\;.
838: $$
839: Thus
840: \begin{eqnarray} \nonumber
841: \frac{|<T_x\,, T_y\,, x-y>|}{|x-y|^3} \leq
842: \frac{|<T_x\,, T_y\,,
843: \frac{x-y}{|x-y|}>|}{2-2\cos \theta} \\ \nonumber =
844: \frac{|<T_x\,, T_y\,,
845: \frac{x-y}{\theta}>|}{2-2\cos \theta}\;\;\frac{\theta}{|x-y|}
846: \end{eqnarray} Using Schur's theorem again, we have $|x-y| \geq
847: |p(\theta)-p(0)| =
848: \sqrt{2-2\cos \theta}$. The function $\frac{\theta}{\sqrt{2-2\cos \theta}}$
849: is increasing on
850: $[0,\pi]$, with maximum value $\pi/2$. So
851: $$
852: \frac{|<T_x\,, T_y\,, x-y>|}{|x-y|^3} \leq \frac{\pi}{2} \;
853: \frac{|<T_x\,, T_y\,,
854: \frac{x-y}{\theta}>|}{{2-2\cos \theta}}\;.
855: $$
856: The vectors $T_y$ and $\frac{x-y}{\theta}$ are each
857: first-order (in terms of $\theta$) close to
858: $T_x$. Specifically, we have for $T_y$,
859: $$
860: T_y = T_x + \int_{t=0}^\theta x''(t)\;.
861: $$
862: Since $|x''| \leq 1$, this says we can write $T_y$ as $T_x +
863: V$, where
864: $|V| \leq \theta$.
865:
866: On the other hand,
867: the fundamental
868: theorem of calculus, applied first to $x(\theta)$ and then again to $x'(s)$, says
869: $$ y = x(\theta)= x(0) + \int_{s=0}^\theta
870: x'(s)\;ds\;=\;x(0)+\theta x'(0) +
871: \int_{s=0}^\theta
872: \int_{u=0}^s x''(u)\;du\;ds\;,
873: $$
874: so we can write
875: $\frac{|x-y|}{\theta}$ as $ T_x +W$, where $|W| \leq
876: \frac{1}{2}\theta$.
877:
878:
879:
880: We now have
881: $T_x \times T_y = T_x \times \left(T_x + V\right) = T_x \times V$, which is a vector perpendicular to $T_x$ with
882: length $\leq \theta$. When we take the dot product of this vector with
883: $T_x+W$, we just get the dot product with
884: $W$, so a number whose size is at most $\frac{1}{2}\theta^2$.
885:
886: We now have
887: $$
888: \frac{|<T_x\,, T_y\,, x-y>|}{|x-y|^3} \leq \frac{\pi}{4}\;\;
889: \frac{\theta^2}{{2-2\cos \theta}}\;\leq
890: \frac{\pi}{4}\left(\frac{\pi}{2}\right)^2\;,
891: $$
892: so the inner integral is bounded by $b_1=(2\pi)\cdot
893: (\frac{\pi}{4})\cdot(\frac{\pi}{2})^2$, since the points $y$ run from (what
894: we might denote as) $x-\pi$ to $x+\pi$.
895:
896: Multiply this bound for the inner integral by $L$ to bound the double
897: integral.
898: \qed
899:
900:
901: \subsection{Bounding $\text{Far}(K)$}
902:
903: As in the previous case, we shall bound the inner integral,
904: $$ \int_{\rm{arc}(x,y) \geq \pi}
905: \frac{|<T_x\,, T_y\,, x-y>|}{|x-y|^3} \;\;,$$ then multiply by $L$ to bound
906: the double integral.
907:
908: As before, we write the integrand as the triple scalar product of three unit
909: vectors, divided by $|x-y|^2$. Since the numerator has magnitude at most
910: $1$, it suffices to bound
911: $$\int_{\rm{arc}(x,y) \geq \pi}
912: \frac{1}{|x-y|^2} \;\;.$$
913:
914: For any smooth curve with thickness radius
915: $R$, it is shown in
916: \cite{LSDR} that points
917: $x,y$ with $\text{arc}(x,y)\geq \pi R$ must have $|x-y|\geq 2R$. So in our
918: situation, when $\text{arc}(x,y)\geq \pi$, we know $|x-y|\geq 2$.
919:
920:
921: Fix $x$ and let $Y=\left\{y \in K \;|\; \text{arc}(x,y) \geq \pi \right\}$.
922: By Lemma \ref{illumination},
923: $$\int_Y \frac{1}{|y-x|^2} \leq c_1 + c_2 \kappa(Y) \leq c_1 + c_2
924: \kappa(X)\;.$$ Thus
925: $$\text{Far}(K) \leq c_1 L + c_2 \kappa(X) L\;.$$
926: \qed
927:
928:
929: \section{Knot energies} The analysis of knot energies $E_N(K)$ and $E_S(K)$
930: in \cite{BS2, BS3} and the Mobius energy $E_O$ in \cite{RSBound} are similar
931: to the analysis of average crossing number here: All involve bounding
932: ``Near" and ``Far" contributions, and all rely on bounding $\int_y
933: \frac{1}{|y-x|^2}$ for the ``Far" part. We can use Lemma \ref{illumination}
934: to show that each of these energies is bounded by [some constant, that
935: depends on the energy but not on $K$, times] $E_L(K) \kappa(K)$.
936:
937: \section{Can the theorem be improved?} Our theorem says (throughout
938: this section, we will suppress coefficients)
939: $$\rm{acn}(K) \leq E_L(K) \;\kappa(K)\;.$$ Is it possible to lower the
940: exponent (from 1) on one or both of
941: $E_L(K)$, $\kappa(K)$? As noted in Example \ref{sharp}, Lemma
942: \ref{illumination} is sharp. However, if we include thickness, and postulate
943: that the knot is long and distributed homogeneously in space, then we can
944: argue heuristically that there would be a lower-order bound. This leads to
945: the conjecture that in fact $\rm{acn}(K) \leq E_L(K) \;\kappa(K)^{1/2}\;.$
946:
947: Scale the knot $K$ so it has thickness radius $r(K)=1$. Then $E_L(K)$
948: is just the arclength, $L$, of $K$. Suppose
949: $K$ is distributed in space so that relative to each point
950: $x_0
951: \in K$, each spherical shell $S[n,n+1]$ about $x_0$ contains on the order of
952: $n^\beta$ arclength of $K$. Here $\beta$ is constant, independent of the
953: choice of
954: $x_0$, and is a measure of the density of packing of $K$. Fix some $x_0\in
955: K$. The shells run from $n=0$ to whatever value $N$ (for that $x_0$) is
956: needed to engulf all of
957: $K$.
958:
959: The amount of arclength of $K$ in each shell has to include at least enough
960: to reach from one sphere to the other, a constant, so $\beta\geq 0$. On the
961: other hand, since $r(K)=1$, an arc (or union of arcs) of $K$ of some total length $\ell$ carries with it a proportional amount of excluded volume $=\pi \ell$. Since the volume of a spherical shell is approximately proportional to the area of a boundary sphere, we must have $\beta \leq 2$.
962:
963: The total arclength $L$ of $K$ is the sum of the amounts in the shells, so
964: if the amount in each shell is on the order of $n^\beta$, then $L\approx N^{\beta+1}$.
965:
966: Assuming $K$ is long enough, relative to $N$, to apply Lemma \ref{packing},
967: we have $\kappa(K) \geq L/N \approx N^{\beta + 1}/N = N^\beta$.
968:
969: We proceed as in the proof of the main theorem:
970: $\text{Near}(K)$ is bounded by a constant and we will bound $\text{Far}(K)$ by bounding the inner Illumination integral and multiplying by $L$.
971:
972: If the amount of length of $K$ in each shell $S[n,n+1]$ is on the order of $n^{\beta}$, then
973:
974: $$\text{Illumination} \leq \sum_{n=2}^N \frac{n^\beta}{n^2}
975: \approx \begin{aligned}
976: &\text{constant} - N^{\beta-1}\;\;\; &(0 \leq \beta < 1) \\ &\log N \;\;\; &(\beta=1)
977: \\ &N^{\beta-1}\;\;\; &(1 < \beta \leq 2) \;.
978: \end{aligned}
979: $$
980:
981: In the first situation $(0 \leq \beta < 1)$, we get a bound on average crossing number proportional to $L$, that is $\text{acn}(K)\leq a_1 + a_2 E_L(K)$. This situation includes long, thin knots, as well as knots such as iterated composites of congruent curves where crossing number and ropelength grow at the same rates.
982: In the second situation $(\beta=1)$, we have $\text{acn}(K) \leq a_1+a_2 E_L(K) \log \kappa(K)$.
983: In the third situtation $(1 < \beta \leq 2)$, we have
984: $$\text{acn}(K) \leq a_1 + a_2 E_L(K) \kappa(K)^{\frac{\beta-1}{\beta}}\;.$$
985: When $\beta=2$, we have the densest possible spatial packing of $K$ as in the examples \cite{B2, CKS}, where the growth rate $\text{acn}(K)\approx E_L(K)\kappa(K)^{1/2}$ is attained.
986:
987:
988:
989:
990:
991:
992:
993:
994:
995:
996:
997:
998:
999:
1000:
1001:
1002:
1003:
1004:
1005:
1006:
1007:
1008:
1009: \newpage
1010:
1011: \bibliographystyle{siam}
1012:
1013:
1014:
1015: \bibliography{JonsEricBib_TotCurv013103}
1016:
1017:
1018:
1019:
1020:
1021:
1022:
1023:
1024:
1025:
1026:
1027:
1028:
1029:
1030:
1031:
1032:
1033:
1034:
1035:
1036:
1037:
1038:
1039: \end{document}
1040:
1041:
1042:
1043:
1044:
1045:
1046:
1047:
1048:
1049:
1050: