1: %% This document created by Scientific Word (R) Version 2.0
2: %% Starting shell: mathart1
3:
4:
5: \documentclass[12pt,thmsa]{article}
6: \usepackage{amsfonts}
7: %%%%%
8: \usepackage{sw20bams}
9:
10: %TCIDATA{TCIstyle=Article/art1.lat,bams,mathart1}
11:
12: \input tcilatex
13: \begin{document}
14:
15: \author{Steven R. Finch}
16: \title{The Logarithmic Spiral Conjecture}
17: \date{January 10, 2005}
18: \maketitle
19:
20: \begin{abstract}
21: When searching for a planar line, if given no further information, one
22: should adopt a logarithmic spiral strategy (although unproven).
23: \end{abstract}
24:
25: \footnotetext{
26: Copyright \copyright\ 2005 by Steven R. Finch. All rights reserved.}This
27: brief paper is concerned entirely with geometry in the plane and continues a
28: thought in \cite{FW}. If a line intersects a circle in one or two points, we
29: say that the line \textbf{strikes} the circle. If a line intersects a circle
30: in exactly one point (that is, if the line is tangent to the circle), we say
31: that the line \textbf{touches} the circle.
32:
33: Let $f$ be a nonnegative, continuously differentiable function on $\Bbb{R}$
34: satisfying
35: \[
36: \begin{array}{ccc}
37: \lim\limits_{\theta \rightarrow -\infty }f(\theta )=0, & &
38: \lim\limits_{\theta \rightarrow \infty }f(\theta )=\infty .
39: \end{array}
40: \]
41: The polar curve $r=f(\theta )$ intersects every line in the plane, that is, $%
42: f$ is a \textbf{spiral}. (Reason: for each $R>0$, there exists $\Theta $ so
43: large that $\theta >\Theta $ implies $f(\theta )>R$. Any line striking the
44: circle $r=R$ must therefore intersect the curve $r=f(\theta )$. Since $R$
45: was arbitrary, the statement follows.) Existence of intersection points is
46: only the beginning of our study.
47:
48: Consider the set $\Sigma $ of all lines that strike the circle $r=R$. The
49: spiral $r=f(\theta )$ possesses a first intersection point $\theta $ with
50: each line in $\Sigma $; let $\theta _1$ denote the supremum of all such $%
51: \theta $. Loosely put, $\theta _1$ constitutes the worst case scenario when
52: seeking all members of $\Sigma $ via the search strategy $r=f(\theta )$.
53: Clearly $\theta _1$ depends on $R$ and $\theta _1=-\infty $ when $R=0$.
54:
55: The cost of finding all lines in $\Sigma $, starting from the origin, can be
56: quantified by the arclength
57: \[
58: \Lambda (f)=\dint\limits_{-\infty }^{\theta _1}\sqrt{f(\theta )^2+f^{\prime
59: }(\theta )^2}\,d\theta .
60: \]
61: We naturally wish to minimize $\Lambda (f)$ as a function of $f$, for fixed $%
62: R$. Our focus is on the following asymptotic inequality.
63:
64: \begin{conjecture}
65: \[
66: \lim_{R\rightarrow \infty }\frac{\Lambda (f)}R\geq 13.8111351795...
67: \]
68: with equality if and only if $f(\theta )\sim Ce^{\kappa \theta }$ as $\theta
69: \rightarrow \infty $, where $\kappa =0.2124695594...$ and $C>0$ is arbitrary$%
70: .$
71: \end{conjecture}
72:
73: The two numerical constants appear precisely in \cite{Fin}, along with
74: detailed treatment of the special case of a logarithmic spiral $f(\theta
75: )=e^{\kappa \theta }$. Difficulties arise in the general case, owing to the
76: vast variety of spirals permitted
77:
78: A sketch of a geometric proof of Conjecture 1 was published in \cite{BY1,
79: BY2, BY3}. The first part claimed that an optimal spiral must be similar
80: with respect to both rotations and dilations about the origin; the second
81: part claimed that such a highly symmetric spiral must necessarily be a
82: logarithmic spiral. The second part, in fact, is true via the solution of a
83: well-known functional equation \cite{You}. We doubt, however, that any
84: purely geometric proof of the first part can be rigorously correct (although
85: appealing). A more careful analysis, based on the calculus of variations, is
86: perhaps mandatory.
87:
88: \subsection{Examples}
89:
90: We repeat certain steps employed in \cite{Fin}, suitably generalized.
91:
92: \begin{lemma}
93: The distance between the line $Ax+By+C=0$ and the origin is $|C|/\sqrt{%
94: A^2+B^2}$.
95: \end{lemma}
96:
97: \begin{lemma}
98: The equation of a line tangent to the spiral $r=f(\theta )$ is $y-f(\theta
99: )\sin (\theta )=m(x-f(\theta )\cos (\theta ))$, where $\theta $ corresponds
100: to the point of tangency and the slope is given by
101: \[
102: m=\frac{f^{\prime }(\theta )\sin (\theta )+f(\theta )\cos (\theta )}{%
103: f^{\prime }(\theta )\cos (\theta )-f(\theta )\sin (\theta )}.
104: \]
105: \end{lemma}
106:
107: \noindent \textbf{Proof of Lemma 3.} Clearly
108: \[
109: \frac{dy}{dx}=\frac{dy/d\theta }{dx/d\theta }=\frac{(f(\theta )\sin (\theta
110: ))^{\prime }}{(f(\theta )\cos (\theta ))^{\prime }}=\frac{f^{\prime }(\theta
111: )\sin (\theta )+f(\theta )\cos (\theta )}{f^{\prime }(\theta )\cos (\theta
112: )-f(\theta )\sin (\theta )}.
113: \]
114:
115: \begin{theorem}
116: Let $L$ denote the first line that is both tangent to the spiral $r=f(\theta
117: )$ and tangent to the circle $r=R$. The tangency point $\theta _0$ of $L$
118: with the spiral satisfies the equation
119: \[
120: R^2(f(\theta )^2+f^{\prime }(\theta )^2)=f(\theta )^4.
121: \]
122: \end{theorem}
123:
124: \noindent \textbf{Proof of Theorem 4.} Apply Lemma 2 with $A=m$, $B=-1$ and $%
125: C=f(\theta )(\sin (\theta )-m\cos (\theta ))$ to obtain $(1+m^2)R^2=f(\theta
126: )^2(\sin (\theta )-m\cos (\theta ))^2$. Substituting the expression for $m$
127: from Lemma 3 gives the desired equation.\\
128:
129: We emphasize that, on the one hand, $\theta _0$ is where the spiral first
130: intersects a line that touches the circle $r=R$ (the touching occurs
131: elsewhere). On the other hand, $\theta _1$ is just above where the spiral
132: last intersects a new line that strikes the circle $r=R$ (the striking,
133: again, occurs elsewhere). If the function $f$ is strictly increasing, then
134: in the interval $\theta _0<\theta <\theta _1$, the spiral intersects all
135: other lines that touch $r=R$; at $\theta =\theta _1$, repetition begins so
136: we stop there. Suppose that we are given a spiral $r=f(\theta )$ for which $%
137: f(\theta )\not \sim Ce^{\kappa \theta }$ for any $\kappa >0$, $C>0$. Clearly
138: \[
139: \frac{\Lambda (f)}R\geq \frac 1R\dint\limits_{-\infty }^{\theta _0}\sqrt{%
140: f(\theta )^2+f^{\prime }(\theta )^2}\,d\theta ,
141: \]
142: and thus if we demonstrate that the right hand side $\rightarrow \infty $ or
143: is at least $>13.82$, then this is consistent with Conjecture 1.
144:
145: As a first example, consider Archimedes' spiral
146: \[
147: f(\theta )=\left\{
148: \begin{array}{lll}
149: \kappa \theta & & \text{if }\theta \geq 0, \\
150: 0 & & \text{if }\theta <0.
151: \end{array}
152: \right.
153: \]
154: From Theorem 4, it follows that $R^2(1+\theta ^2)=\kappa ^2\theta ^4$ and
155: hence
156: \[
157: \theta _0=\frac R\kappa \sqrt{\frac 12\left( 1+\sqrt{1+\frac{4\kappa ^2}{R^2}%
158: }\right) }\geq \frac R\kappa \sqrt{\frac 12\left( 1+1\right) }\geq \frac
159: R\kappa .
160: \]
161: Consequently, the normalized arclength is bounded from below by
162: \[
163: \frac \kappa R\dint\limits_0^{\theta _0}\sqrt{1+\theta ^2}\,d\theta \geq
164: \frac \kappa R\dint\limits_0^{\theta _0}\theta \,d\theta =\frac \kappa
165: {2R}\theta _0^2\geq \frac R{2\kappa }\rightarrow \infty
166: \]
167: as $R\rightarrow \infty $. Alternatively, we can avoid solving for $\theta _0
168: $ altogether:\ From $R^2(1+\theta ^2)=\kappa ^2\theta ^4$, deduce that
169: \[
170: R=\frac{\kappa \,\theta ^2}{\sqrt{1+\theta ^2}}\leq \kappa \,\theta ^2
171: \]
172: and hence that $\theta \rightarrow \infty $ as $R\rightarrow \infty $. Here
173: we obtain
174: \[
175: \frac \kappa R\dint\limits_0^{\theta _0}\theta \,d\theta =\frac \kappa
176: {2R}\theta _0^2=\frac \kappa 2\frac{\sqrt{1+\theta _0^2}}{\kappa \,\theta
177: _0^2}\theta _0^2=\frac 12\sqrt{1+\theta _0^2}\geq \frac{\theta _0}%
178: 2\rightarrow \infty
179: \]
180: as $\theta _0\rightarrow \infty $ (and thus as $R\rightarrow \infty $). This
181: latter device will be useful in the following examples. See Figure 1 for an
182: illustration.\FRAME{ftbpFUX}{4.7495in}{4.8083in}{0pt}{\Qcb{The first contact
183: point that the spiral $r=\theta $ has with a line tangent to the circle $R=6$
184: is at $\theta _0=348.4^{\circ }$. The second contact point with the line is
185: at $\theta _1=641.5^{\circ }$. Incidently, the line is tangent to $R=6$ at $%
186: 339.1^{\circ }<\theta _0$.}}{}{spr.jpg}{\special{language "Scientific
187: Word";type "GRAPHIC";maintain-aspect-ratio TRUE;display "USEDEF";valid_file
188: "F";width 4.7495in;height 4.8083in;depth 0pt;cropleft "0";croptop
189: "0.9991";cropright "0.9997";cropbottom "0";filename
190: 'C:/SPR.JPG';file-properties "XNPEU";}}
191:
192: Consider next the spiral
193: \[
194: f(\theta )=\left\{
195: \begin{array}{lll}
196: e^{\theta ^a} & & \text{if }\theta \geq 0, \\
197: e^{-\left| \theta \right| ^a} & & \text{if }\theta <0
198: \end{array}
199: \right.
200: \]
201: for a fixed exponent $a>0$. From Theorem 4, it follows that $R^2(1+a^2\theta
202: ^{2a-2})=e^{2\theta ^a}$, that is,
203: \[
204: R=\frac{e^{\theta ^a}}{\sqrt{1+a^2\theta ^{2a-2}}}\leq e^{\theta ^a}.
205: \]
206: Hence $\theta \rightarrow \infty $ as $R\rightarrow \infty $. If $a>1$, the
207: normalized arclength is bounded from below by
208: \begin{eqnarray*}
209: &&\frac 1R\left( \dint\limits_{-\infty }^0\sqrt{1+a^2\left| \theta \right|
210: ^{2a-2}}\,e^{-\left| \theta \right| ^a}d\theta +\dint\limits_0^{\theta _0}%
211: \sqrt{1+a^2\theta ^{2a-2}}\,e^{\theta ^a}d\theta \right) \\
212: &\geq &\frac 1R\left( \dint\limits_{-\infty }^0a\,\left| \theta \right|
213: ^{a-1}e^{-\left| \theta \right| ^a}d\theta +\dint\limits_0^{\theta
214: _0}a\,\theta ^{a-1}e^{\theta ^a}d\theta \right) \\
215: &=&\frac 1R\left( 1+e^{\theta _0^a}-1\right) =\sqrt{1+a^2\theta _0^{2a-2}}%
216: \geq a\,\theta _0^{a-1}\rightarrow \infty
217: \end{eqnarray*}
218: as $\theta _0\rightarrow \infty $ (and thus as $R\rightarrow \infty $). If $%
219: 0<a<1$, the normalized arclength is bounded by
220: \begin{eqnarray*}
221: &&\ \frac 1R\left( \dint\limits_{-\infty }^0\sqrt{1+a^2\left| \theta \right|
222: ^{2a-2}}\,e^{-\left| \theta \right| ^a}d\theta +\dint\limits_0^{\theta _0}%
223: \sqrt{1+a^2\theta ^{2a-2}}\,e^{\theta ^a}d\theta \right) \\
224: \ &\geq &\frac 1R\left( \dint\limits_{-\infty }^0e^{-\left| \theta \right|
225: ^a}d\theta +\dint\limits_0^{\theta _0}e^{\theta ^a}d\theta \right) \geq
226: \frac 1R\left( 0+\dint\limits_0^{\theta _0}e^{\theta ^a}d\theta \right)
227: \end{eqnarray*}
228: and we have asymptotics
229: \[
230: \frac 1R\dint\limits_0^{\theta _0}e^{\theta ^a}d\theta \sim \frac 1R\left(
231: \frac 1a\theta _0^{1-a}e^{\theta _0^a}\right) =\frac 1a\sqrt{1+a^2\theta
232: _0^{2a-2}}\,\theta _0^{1-a}\rightarrow \infty
233: \]
234: as $\theta _0\rightarrow \infty $ (and thus as $R\rightarrow \infty $). Only
235: the case $a=1$ remains, which is covered in \cite{Fin}. This is compelling
236: (but not completely convincing) evidence that the Logarithmic Spiral
237: Conjecture is valid.
238:
239: Consider finally the spiral
240: \[
241: f(\theta )=\left\{
242: \begin{array}{lll}
243: \theta ^be^\theta & & \text{if }\theta \geq 0, \\
244: 0 & & \text{if }\theta <0
245: \end{array}
246: \right.
247: \]
248: for a fixed exponent $b>0$. From Theorem 4, we obtain
249: \[
250: R=\frac{\theta ^{b+1}e^\theta }{\sqrt{\theta ^2+(b+\theta )^2}}\leq \frac 1{%
251: \sqrt{2}}\theta ^be^\theta ,
252: \]
253: hence $\theta \rightarrow \infty $ as $R\rightarrow \infty $. Clearly $%
254: \theta _1\geq \theta _0+\pi $ on geometric grounds. Therefore the normalized
255: arclength is bounded from below by
256: \[
257: \frac 1R\dint\limits_0^{\theta _0+\pi }\sqrt{\theta ^2+(b+\theta )^2}%
258: \,\theta ^{b-1}e^\theta d\theta \geq \frac{\sqrt{2}}R\dint\limits_0^{\theta
259: _0+\pi }\theta ^be^\theta d\theta
260: \]
261: and we have asymptotics
262: \begin{eqnarray*}
263: \frac{\sqrt{2}}R\dint\limits_0^{\theta _0+\pi }\theta ^be^\theta d\theta
264: &\sim &\frac{\sqrt{2}}R\left( (\theta _0+\pi )^be^{\theta _0+\pi }\right) \\
265: \ &=&\sqrt{2}\,\frac{(\theta _0+\pi )^be^{\theta _0+\pi }}{\theta
266: _0^{b+1}e^{\theta _0}}\sqrt{\theta _0^2+(b+\theta _0)^2} \\
267: \ &\rightarrow &2e^\pi >13.82
268: \end{eqnarray*}
269: as $\theta _0\rightarrow \infty $.
270:
271: The logarithmic spiral appears with regard to another planar search problem
272: \cite{GC}, but the techniques of Gal \&\ Chazan do not seem to apply here. A
273: min-mean analog of our Conjecture 1 could also be formulated, starting with
274: \cite{Fin}.
275:
276: \subsection{Acknowledgements}
277:
278: I am grateful to Ricardo Baeza-Yates, Li-Yan Zhu and John Shonder for their
279: assistance. A discussion during my Visiting Lecture at Oberlin College in
280: November 2004 was also very helpful.
281:
282: \begin{thebibliography}{9}
283: \bibitem{FW} S. R. Finch and J. E. Wetzel, Lost in a forest, \textit{Amer.
284: Math. Monthly} 11 (2004) 645-654.
285:
286: \bibitem{Fin} S. R. Finch and L.-Y. Zhu, Searching for a shoreline,
287: math.OC/0501123.
288:
289: \bibitem{BY1} R. A. Baeza-Yates, J. C. Culberson and G. J. E. Rawlins,
290: Searching in the plane, \textit{Inform. and Comput.} 106 (1993) 234--252;
291: MR1241311 (94h:90019).
292:
293: \bibitem{BY2} R. A. Baeza-Yates, J. C. Culberson and G. J. E. Rawlins,
294: Searching with uncertainty (extended abstract), \textit{First Scandinavian
295: Workshop on Algorithm Theory}, Proc. 1988 Halmstad conf., ed. R. Karlsson
296: and A. Lingas, Lecture Notes in Comput. Sci. 318, Springer-Verlag, 1988, pp.
297: 176--189.
298:
299: \bibitem{BY3} R. A. Baeza-Yates, J. C. Culberson and G. J. E. Rawlins,
300: Searching with uncertainty (full report), Research Report CS--87--68, Dept.
301: of Computer Sci., Univ. of Waterloo, 1987.
302:
303: \bibitem{You} R. M. Young, \textit{Excursions in Calculus. An Interplay of
304: the Continuous and the Discrete}, Math. Assoc. Amer., 1992, pp. 144--153;
305: MR1184144 (93k:00007).
306:
307: \bibitem{GC} S. Gal and D. Chazan, On the optimality of the exponential
308: functions for some minimax problems, \textit{SIAM J. Appl. Math.} 30 (1976)
309: 324--348; errata 32 (1977) 520; MR0391997 (52 \#12815) and MR0456579 (56
310: \#14803).\\
311:
312: \begin{tabular}{lll}
313: & Steven R. Finch & \\
314: & Clay Mathematics Institute & \\
315: & One Bow Street & \\
316: & Cambridge, MA 02138 & \\
317: & \textit{Steven.Finch@inria.fr} &
318: \end{tabular}
319: \end{thebibliography}
320:
321: \end{document}
322: