math0501295/dens.tex
1: \section{Density of primitive lattice points}\label{S:Dens}
2: The main result of this section is Corollary~\ref{C:Dens2}.  
3: It will be needed in \S\ref{S:NEsub} to find, given a vector 
4: $w\in W$, vectors $w'\in W$ such that $w'=w+gv$ for some $v\in Z$ 
5: satisfying certain given inequalities on $|v|$ and $|w\times v|$.  
6: 
7: \subsection{Continued fractions in vector form}
8: Recall each $\alpha\in\bbR$ admits an expansion of the form 
9: \begin{equation}\label{Cfrac}
10: \alpha = a_0 + \cfrac{1}{a_1+\cfrac{1}{a_2+\dots}}
11: \qquad a_0\in\bbZ, a_1,a_2,\dots\in\bbN
12: \end{equation}
13: whose terms are uniquely determined except for a two-fold 
14: ambiguity when $\alpha$ is rational; e.g. $22/7=3+1/7=3+1/(6+1/1)$.  
15: The $k$th \textit{convergent} of $\alpha$ is the reduced fraction 
16: $p_k/q_k$ that results upon simplifying the expression obtained by 
17: truncating (\ref{Cfrac}) so that the last term is $a_k$.  
18: The convergents of $\alpha$ satisfy the recurrence relations 
19: \begin{equation}\label{RR}
20: \begin{tabular}{cc}
21: $p_{k+1} = a_{k+1}p_k + p_{k-1}$ & $\;p_0=a_0,\,p_{-1}=1$\\
22: $q_{k+1} = a_{k+1}q_k + q_{k-1}$ & $\;q_0=1,\;\;q_{-1}=0$
23: \end{tabular},
24: \end{equation}
25: the identity 
26: \begin{equation}\label{PR}
27:       p_kq_{k+1}-p_{k+1}q_k=(-1)^{k+1}
28: \end{equation}
29: and the inequalities 
30: \begin{equation}\label{CFI}
31: \frac{1}{q_k(q_{k+1}+q_k)} 
32:         < \left|\alpha-\frac{p_k}{q_k}\right|
33:         < \frac{1}{q_kq_{k+1}}.  
34: \end{equation}
35: A rational $p/q$ is said to be a \textit{best approximation of the 
36: second kind} if 
37: \begin{equation}\label{BA2}
38: |q\alpha-p|\leq|n\alpha-m|\qquad\text{for all}\;\;m\in\bbZ,\;n=1,\dots,q
39: \end{equation}
40: and this property characterises the convergents of $\alpha$ modulo 
41: the $0$th convergent $a_0$, which is a best approximation to $\alpha$ 
42: iff the fractional part of $\alpha$ is $\leq1/2$.  The following is 
43: a useful test for a rational to be a convergent: 
44: \begin{equation}\label{SCC}
45: \left|\alpha-\frac{p}{q}\right| \leq \frac{1}{2q^2},\;\;\gcd(p,q)=1 
46: \quad\Rightarrow\quad (p,q)=\pm (p_k,q_k) \;\text{for some}\;k\geq0.  
47: \end{equation}
48: 
49: It will be convenient for us to recast the above facts in vector form.  
50: Setting $v_k:=\vec{p_k,q_k}$, the recurrence relations (\ref{RR}) and 
51: the identity (\ref{PR}) become 
52: \begin{gather}\label{VRR}
53:    v_{k+1} = a_{k+1}v_k + v_{k-1}\qquad 
54:                v_0=\vec{a_0,1}, \quad v_{-1}=\vec{1,0} \\
55: \intertext{and}\label{VPR}
56:    v_k\times v_{k+1} := p_kq_{k+1}-p_{k+1}q_k = (-1)^{k+1}.  
57: \end{gather}
58: Although (\ref{CFI}) can easily be rewritten in vector notation, the 
59: resulting expression looks awkward because of the distinguished nature of 
60: the coordinate directions.  Instead, we shall use the following analog of 
61: (\ref{CFI}) which is expressed in terms of the vector $w:=\vec{\alpha,1}$ 
62: and its Euclidean length $|w|$:  
63: \begin{equation}\label{VCFI}
64: \frac{1}{|v_{k+1}+v_k|} < \frac{\left|w\times v_k\right|}{|w|} 
65:                         < \frac{1}{|v_{k+1}|}  
66: \end{equation}
67: To see (\ref{VCFI}) recall that convergents alternate on both 
68: sides of $\alpha$ and (\ref{CFI}) follows from the fact 
69: that the rational $(p_k+p_{k+1})/(q_k+q_{k+1})$ always lies on the 
70: same side of $\alpha$ occupied by $p_k/q_k$.  (\ref{VCFI}) follows 
71: similarly from a comparison of the components of $v_{k+1}+v_k$, $w$ 
72: and $v_{k+1}$ in the direction perpendicular to $v_k$.  
73: 
74: \begin{Def}\label{D:Spec}
75: If $\theta=|w|^{-1}w$ where $w=\vec{\alpha,1}$ as above, then 
76: we define $$\Spec(\theta)=\{v_0,v_1,\dots\} = \{v_k\}_{k\geq0}$$  
77: and call the vectors $v_k$ the \textit{convergents} of $\theta$.  
78: The definition is extended to all unit vectors by requiring 
79: $$\Spec(1,0)=\{\vec{1,0}\},\quad \Spec(-\theta)=-\Spec(\theta)$$ 
80: and to all nonzero vectors by $\Spec(w)=\Spec(|w|^{-1}w)$.  
81: We shall also denote by $\spec(w)$ the sequence of Euclidean 
82: lengths of vectors in $\Spec(w)$.  
83: \end{Def}
84: 
85: The next lemma was motivated by (\ref{SCC}) and will be needed 
86: in \S\ref{S:NEsub}.  
87: \begin{Lem}\label{L:VSCC}
88: Let $w$ be a vector that makes an angle $\phi$ with the $y$-axis.  
89: Then for any $v\in\bbZ^2$ such that $|v|\cos\phi>1$ we have 
90: \begin{equation}\label{VSCC}
91: \frac{|w\times v|}{|w|} \leq \frac{1}{2|v|},\quad\gcd(v)=1. 
92: \quad\Rightarrow\quad \pm v\in\Spec(w).  
93: \end{equation}
94: \end{Lem}
95: \begin{proof}
96: Let $P=P(v,w)$ be the closed parallelogram that has $\pm v$ as two 
97: of its vertices, one pair of sides parallel to $w$ and the other pair 
98: parallel to the $x$-axis.  The characterisation of convergents given 
99: in (\ref{BA2}) is equivalent to the statement that every nonzero 
100: $u\in P\cap\bbZ^2$ belongs to the union of the two sides of $P$ 
101: parallel to $w$.  Hence, it is enough to verify this statement under 
102: the given hypotheses.  
103: 
104: Apply Lemma~\ref{L:SVC} with $\theta=|w|^{-1}w$ (and $V$ the set 
105: of integer lattice points which are not scalar multiples of $v$) 
106: to conclude there is some $T$ for which $g_T^\theta v$ is the 
107: shortest vector in $V'=g_T^\theta(\bbZ^2-0)$.  Let $E$ be the 
108: inverse image under $g_T^\theta$ of the largest closed disk 
109: centered at the origin whose interior is disjoint from $V'$.  
110: The boundary of $E$ is an ellipse passing through the points 
111: $\pm v$ while the interior contains no integer lattice points 
112: other than the origin.  
113: 
114: Without loss of generality, we assume $w$ lies in the first 
115: quadrant and $v$ in the upper half plane.  There are two cases.  
116: First, if $v$ lies to the right of $w$, then $E$ contains $P$ 
117: and we are done.  Now, if $v$ lies to the left of $w$, 
118: then let $x$ be the length of a horizontal side of $P$ and 
119: $y$ the vertical distance between $v$ and its reflection in 
120: the line $\bbR w$.  It is enough to show $x<1$ and $y<1$.  
121: If $z$ is the distance between $v$ and its reflection then 
122: $z=2|v|\sin\angle{wv}=2|w\times v|/|w|\leq1/|v|<\cos\phi$ 
123: so that $x=z\sec\phi<1$ and $y=z\sin\phi<1$.  
124: \end{proof}
125: 
126: 
127: 
128: \subsection{Density of rationals in intervals}
129: For any $\Omega\subset\bbR^2$ with $0<\area(\Omega)<\infty$ let 
130: $$\dens(\Omega) = \frac{^\#Z\cap\Omega}{\area(\Omega)}.$$  
131: \begin{Lem}\label{L:8/27}
132: If $\Omega$ is a compact convex subset of the first quadrant 
133: containing $(0,0)$, $(1,0)$ and $(0,1)$ but not $(1,1)$ then 
134: $\dens(2\Omega\setminus\Omega)>8/27$.  
135: \end{Lem}
136: \begin{proof}
137: By convexity there is a function $y=f(x), 0\leq x\leq1$ whose graph 
138: is contained in $\partial\Omega$ and $f(0)\geq1$.  Similarly, there is 
139: a $x=g(y), 0\leq y\leq1$ whose graph is contained in $\partial\Omega$ 
140: and $g(0)\geq1$.  Without loss of generally, assume $$f(1/2)\geq g(1/2).$$  
141: Let $\Omega_1\subset\Omega$ be the part below the graph of $f$ and 
142: $\Omega_2=\Omega\setminus\Omega_1$.  Since $\Omega_1$ lies below 
143: any tangent line at $(1/2,f(1/2))$, $\area(\Omega_1)\leq f(1/2)$.  
144: There are three cases.  First, if $f(1/2)<1$ then the assumption 
145: above implies $g(0)<3/2$ so that $\area(\Omega_2)\leq1/8$.  
146: Since $(1,1)\in2\Omega\setminus\Omega$ we have 
147: $$\dens(2\Omega\setminus\Omega)\geq 
148: \frac{1}{3}\left(\frac{1}{f(1/2)+1/8}\right)>8/27.$$
149: Second, if $f(1/2)\geq1$ and $f(1)<1/2$ then it still follows that 
150: $g(0)<3/2$ while the number of vectors in $2\Omega\setminus\Omega$ 
151: of the form $(1,n)$ is $\lfloor2f(1/2)\rfloor$ so that 
152: $$\dens(2\Omega\setminus\Omega)>
153: \frac{1}{3}\left(\frac{2f(1/2)-1}{f(1/2)+1/8}\right)\geq8/27.$$
154: Finally, if $f(1/2)\geq1$ and $f(1)\geq1/2$ then the assumption above 
155: implies $g(0)<2$ so that $\area(\Omega_2)\leq1/2$.  Apart from points of 
156: the form $(1,n)$ we also have $(2,1)\in2\Omega\setminus\Omega$.  Hence, 
157: $$\dens(2\Omega\setminus\Omega)>
158: \frac{1}{3}\left(\frac{2f(1/2)}{f(1/2)+1/2}\right)\geq4/9>8/27.$$
159: \end{proof}
160: 
161: Note if $\Omega_a=\{(x,y): x+y\leq a, x\geq0, y\geq0\}$ then 
162: \begin{align*}
163: \dens(2\Omega_a\setminus\Omega_a) &= 2/3a^2 
164:              \quad\text{for}\quad 1\leq a<3/2\quad\text{and}\\
165: \dens(\gamma\Omega_1\setminus\Omega_1) &= 0
166:              \quad\text{for}\quad \gamma<2
167: \end{align*}
168: show that the constants in the preceding lemma are sharp.  
169: 
170: Let $S^1(\bbQ)$ be the set of unit vectors of the form $v/|v|$ for some 
171: $v\in Z$ and $S^1_b(\bbQ)$ the subset formed by those with $|v|\leq b$.  
172: Let $\cI_b$ denote the collection of intervals in $S^1$ with endpoints 
173: in $S^1_b(\bbQ)$.  For any interval $I\subset S^1$ let 
174: $$\Omega(I,b) = \{v\in\bbR^2 : \bbR_+v\cap I\not=\emptyset, |v|\leq b\}.$$
175: 
176: \begin{Prop}\label{P:8/27}
177: For any $I\in\cI_b$, $\dens(2\Omega(I,b)\setminus\Omega(I,b))>8/27$.  
178: \end{Prop}
179: \begin{proof}
180: First we show if $I$ is minimal, i.e. $S^1_b(\bbQ)\cap\text{int}I=\emptyset$, 
181: then its endpoints correspond to a pair of vectors $v,v'\in Z$ such that 
182: $|v\times v'|=1$.  Indeed, there is a linear map $\gamma$ that sends $v$ 
183: to $(0,1)$ and $v'$ to $(a,a')\in\bbZ^2$ with $0\leq a'<a=|v\times v'|$.  
184: If $a>1$ then $\gamma^{-1}(1,1)\in S^1_b(\bbQ)\cap\text{int}I$; hence, $a=1$.  
185: Now observe $\gamma\Omega(I,b)$ is a compact convex set satisfying the 
186: hypothesis of Lemma~\ref{L:8/27} and since $\gamma$ preserves density, 
187: the proposition holds for minimal $I$ in $\cI_b$.  Since every interval 
188: in $\cI_b$ is a (finite) disjoint union of minimal ones, this completes 
189: the proof.  
190: \end{proof}
191: 
192: %\begin{Lem}\label{L:2/3}
193: %If $\Omega_+=\{(x,y)\in\Omega : y\geq x\}$ where 
194: %$\Omega$ satisfies the hypothesis of Lemma~\ref{L:8/27}, 
195: %then $\dens(2\Omega_+\setminus\Omega_+)>2/3$.  
196: %\end{Lem}
197: %\begin{proof}
198: %Let $b$ be the $y$-coordinate of the unique point where $\partial\Omega$ 
199: %intersects $x=1/2$.  The area of the part of $\Omega_+$ above (resp. below) 
200: %$y=1$ is at most $b-1$ (resp. $1/2$) so that $\area(\Omega_+)\leq b-1/2$; 
201: %the number of points of the form $(1,n)$ in $2\Omega_+\setminus\Omega_+$ 
202: %is $\lfloor2b\rfloor>2b-1$.  
203: %\end{proof}
204: 
205: \begin{Thm}\label{T:Dens}
206:   Let $\Omega=\Omega(I,b)$ where 
207:   $I = \{\theta'\in S^1 : \sin\angle\theta\theta'<\eps/b\}$, 
208:   $\theta\in S^1$, $\eps>0$ and $b\geq1$.  
209:   Then $\spec(\theta)\cap[\eps^{-1},b]\not=\emptyset$ 
210:   implies $\dens(2\Omega\setminus\Omega)>4/27\pi$.  
211: \end{Thm}
212: \begin{proof}
213: Let $v_k\in\Spec(\theta)$ be the convergent with length 
214: $|v_k|=\max\spec(\theta)\cap[\eps^{-1},b]$.  Then the RHS of 
215: (\ref{VCFI}) implies (the direction of) $v_k$ lies in $I$.  
216: Without loss of generality, assume $v_k$ lies to the left of $\theta$.  
217: Let $\theta'$ be the right endpoint of $I$ and $v'_l\in\Spec(\theta')$ 
218: the convergent with length $|v'_l|=\max\spec(\theta)\cap[1,b]$.  
219: 
220: Note that $v_k$ does not lie strictly between $v'_l$ and $v'_{l+1}$ 
221: since the length of any such vector in $Z$ is at least $|v'_{l+1}|>b$.  
222: Nor can $v_k=v'_l$ since the RHS of (\ref{VCFI}) would imply $v'_l$ 
223: lies strictly to the right of $v_k$.  Therefore, $v'_l$ lies strictly 
224: to the right of $v_k$.  Let $\Omega'=\Omega(J,b)$ where $J$ is 
225: the interval with left endpoint $v_k$ and right endpoint $v'_l$.  
226: 
227: The interval $I$ has length $|I|=2\sin^{-1}(\eps/b)\leq\pi\eps/b$.  
228: If $|v'_l|\geq2\eps^{-1}$ then the RHS of (\ref{VCFI}) implies 
229: $|J| > \sin^{-1}(\eps/b) - \sin^{-1}(\eps/2b) > \eps/2b$ (we may 
230: assume $\eps<b$ for otherwise the theorem is easily seen to hold) 
231: while if $|v'_l|<2\eps^{-1}$ then 
232: $|J| > \sin\angle v_kv'_l \geq \frac{1}{|v_k|\,|v'_l|} > \eps/2b.$  
233: In either case, we have $|J|>|I|/2\pi$.  
234: 
235: If $v'_l$ lies to the left of $\theta'$ or $|v'_{l+1}|>2b$ then 
236: $Z\cap(2\Omega'\setminus\Omega')\subset Z\cap(2\Omega\setminus\Omega)$ 
237: since any vector strictly between $v'_l$ and $v'_{l+1}$ has length 
238: greater than $2b$.  In this case, Proposition~\ref{P:8/27} implies
239: $\dens(2\Omega\setminus\Omega)>4/27\pi$.  If $|v'_l|\geq2\eps^{-1}$ 
240: then arguing as before we see the angle between $v_k$ and $v'_{l+1}$ 
241: is at least $\eps/2b>|I|/2\pi$ so that we may again conclude that 
242: $\dens(2\Omega\setminus\Omega)>4/27\pi$.  Therefore, we may assume 
243: $v'_{l+1}$ lies between $v_k$ and $\theta'$, $|v'_{l+1}|\leq2b$ and 
244: $|v'_l|<2\eps^{-1}$ in the remaining.  Assuming $\eps<b/\sqrt{2}$ 
245: as we may, since the theorem is easily seen to hold otherwise, 
246: we obtain the following criterion for a vector left of $\theta'$ 
247: to lie in $I$: 
248: \begin{equation}\label{I:vinI}
249: \frac{|v\times\theta'|}{|v|} < \frac{\sqrt{2}\eps}{b}
250: \end{equation}
251: Note that a vector of the form $av'_l+v_{l-1}$ lies to the left of 
252: $\theta'$ if $a\leq a_{k+1}$.  We will show the number of vectors 
253: of length $\geq b/2$ satisfying the above conditions is at least 
254: $c_0b\eps$ for some absolute constant $c_0>0$.  This will complete 
255: the proof since between any two vectors of length $\geq b/2$ (but 
256: $\leq b$) there is a vector whose length is $>b$ and $\leq2b$.  
257: Using the RHS of (\ref{VCFI}) we have 
258: \begin{equation}\label{I:Ifrac}
259: |v\times\theta'| = |v'_{l+1}\times\theta'-(a_{l+1}-a)v'_l\times\theta'|
260:                \leq \frac{1+|a_{l+1}-a|}{|v'_{l+1}|}.  
261: \end{equation}
262: Assuming $|v|\geq b/2$ ($v=av'_l+v_{l-1}$) we see (\ref{I:vinI}) 
263: holds for $\lfloor b\eps/\sqrt{2}\rfloor$ integers $a\leq a_{l+1}$.  
264: Among these there are at least $b/2|v'_l|\geq b\eps/4$ which make 
265: $|v|\geq b/2$.  Since $\area(\Omega)<\pi b\eps/2$, we conclude 
266: $\dens(2\Omega\setminus\Omega)>1/6\pi$.  
267: \end{proof}
268: 
269: 
270: \subsection{Density in a strip}
271: \begin{Thm}\label{T:Dens2}
272:   Let $\Sigma=\{v\in\bbR^2:|v\times\theta|<\eps,b<|v|\leq2b,v\cdot\theta>0\}$ 
273:   where $\theta\in S^1$, $0<\eps\leq1$ and $b\geq1$.  
274:   Then $\spec(\theta)\cap[\eps^{-1},b]\neq\emptyset$ 
275:   implies $\dens(\Sigma)>2/27\pi$.  
276: \end{Thm}
277: \begin{proof}
278: Let $v_k\in\Spec(\theta)$ with $|v_k|=\max\spec(\theta)\cap[\eps^{-1},b]$.  
279: There are two cases.  
280: If $|v_k|\geq2\eps^{-1}$ let $\Omega=\Omega(\theta,\eps,b)$ be the 
281: region in Theorem~\ref{T:Dens} and put $\Omega'=\Omega(\theta,\eps/2,b)$.  
282: Then Theorem~\ref{T:Dens} implies $\dens(2\Omega'\setminus\Omega')>4/27\pi$.  
283: Since $2\Omega'\setminus\Omega'\subset\Sigma$ and $\area(\Omega')>b\eps/2$ 
284: it follows there are at least $2b\eps/9\pi$ vectors in $Z\cap\Sigma$.  
285: If $|v_k|<2\eps^{-1}$ let $v=av_k+v_{k-1}$ and use (\ref{I:Ifrac}) for 
286: $\theta$ and the hypothesis $|v_k|\geq\eps^{-1}$ to deduce 
287: $$|v\times\theta| \leq \frac{1+|a_{k+1}-a|}{|v_{k+1}|} 
288:                   < \frac{1+|a_{k+1}-a|}{a_{k+1}|v_k|} \leq \eps$$
289: for all $a$ such that $1\leq a<2a_{k+1}$.  Since $|v_k|<2\eps^{-1}$ 
290: the number $v$ with $b<|v|\leq2b$ is $\geq b\eps/2$.  
291: In either case, we have found $2b\eps/9\pi$ vectors $Z\cap\Sigma$.  
292: By an elementary analysis we obtain 
293: \begin{equation}\label{I:area}
294: 3b\eps/2\leq 2b\eps\left(1-\frac{\eps^2}{4b^2}\right) \leq 
295: \area(\Sigma) \leq 2b\eps\left(1+\frac{\eps^2}{2b^2}\right) \leq 3b\eps.  
296: \end{equation}
297: Using the RHS we get $\dens(\Sigma)>2/27\pi$.  
298: \end{proof}
299: 
300: \begin{Cor}\label{C:Dens2}
301: There are universal constants $\rho_1>0$ and $c_1>0$ such that 
302: $\spec(w)\cap [\eps^{-1},b]\neq\emptyset$ (and $0<\eps\leq1\leq b$) 
303: implies there are $\rho_1b\eps$ vectors $v\in Z$ satisfying the 
304: inequalities $$w\cdot v>0,\quad 
305: b\leq|v|\leq2b,\quad c_1\eps|w|\leq|w\times v|\leq\eps|w|.$$  
306: \end{Cor}
307: \begin{proof}
308: Let $\Sigma=\Sigma(\eps,b)$ be as in Theorem~\ref{T:Dens2} with 
309: $\theta=|w|^{-1}w$.  First we claim $\dens(\Sigma)$ is bounded 
310: above by some universal constant.  Indeed, from the cross product 
311: formula we see there's a universal constant $C>1$ such that the 
312: largest (resp. smallest) angle between two vectors in $Z\cap\Sigma$ 
313: is $<C\eps/b$ (resp. $>C^{-1}/b^2$).  Hence, the number of vectors 
314: in $Z\cap\Sigma$ is at most $C^2b\eps$ so that the LHS of (\ref{I:area}) 
315: implies the claim.  
316: 
317: Now let $\Sigma_1=\Sigma(c_1\eps,b)$ and $\Sigma_2=\Sigma\setminus\Sigma_1$.  
318: Observe that the claim implies for any $\rho<2/27\pi$, $c_1$ can be chosen 
319: sufficiently small so that $\dens(\Sigma_2)>\rho$.  Since (\ref{I:area}) 
320: implies $\area(\Sigma_2)\geq cb\eps$ for some universal $c>0$, the corollary 
321: follows.  
322: \end{proof}
323: 
324: