1: \section{The Level of Distribution} \label{chap:lod}
2:
3: With $q_i$, $\bz$, and $D$ as in the statement of Theorem
4: \ref{thm:main}, we make the following definitions:
5: \begin{align*}
6: \ld &:= \{ \x \in \Z^2\, : \,d_i | q_i(\x) \quad(i=1,\ldots,g) \},&
7: \rho(\bd) &:= \# (\ld \cap [0,d_1\ldots d_g)^2), \\
8: \lstard &:= \{ \x \in \ld\, : \, (\x;d_1\ldots d_g)=1\}, &
9: \rstar(\bd) &:= \# (\lstard \cap [0,d_1\ldots d_g)^2), \\
10: \Psi &:= \{ \x \in \Z^2\, :\, \x \equiv \bz \pmod{D} \}, & \Psi_b &:=
11: \{ \x \in \Z^2\, : \, b \x \in \Psi\}.
12: \end{align*}
13:
14: \begin{theorem}[Level of Distribution] \label{thm:lod}
15: Let $q_i(\x)$ and $D$ be defined as in Theorem \ref{thm:main}. For
16: any real numbers $M,Q_1,\ldots,Q_g\ge 1$, let
17: $$
18: T(M,\bQ) := \sum_{\substack{d_i \le Q_i \\ (d_i;D)=1}}
19: \sup_{\partial(\cR) \le M} \left| \# (\ld \cap \cR \cap \Psi) - \frac{\vol(\cR)
20: \rho(d_1,\ldots,d_g)}{(d_1 \ldots d_g D)^2} \right|.
21: $$
22: Then, writing $Q := Q_1\ldots Q_g$, there exist constants
23: $\nu_1, \nu_2 >1 $, depending only on $g$ such that
24: $$
25: T(M,\bQ) \ll Q (\log 2 Q)^{\nu_1} + M \sqrt{Q} (\log 2 Q)^{\nu_2}.
26: $$
27: \end{theorem}
28:
29:
30: \subsection{Transition from $\lstard$ to $\ld$} \label{sec:transition}
31:
32:
33:
34: We begin with the following bridging result, which will be employed in
35: Section \ref{sec:lodunstarred} to
36: express the unstarred sum in
37: terms of the starred sum, leading to Theorem
38: \ref{thm:lod}.
39:
40: \begin{lemma}[Transition Formula] \label{lem:transition}
41: Let $D \in \N$ and suppose that $(d_i;D)=1$ for $i=1,\ldots,g$. Then
42: we have
43: $$
44: \# \left( \ld \cap \cR \cap \Psi \right) = \sum_{b | \psi(\bd)} \#
45: \left( \lstarc \cap \cR / b \cap \Psi_b\right),
46: $$
47: where $c_i := d_i / (d_i ; b^2)$, for $i=1,\ldots,g$, and the
48: multiplicative function $\psi$ is defined by
49: $
50: \psi(p^{\alpha_1},\ldots,p^{\alpha_g}) :=
51: p^{\ceil{\max(\alpha_1,\ldots,\alpha_g)/2}}.
52: $
53: \end{lemma}
54:
55: By definition, $\# (\ld \cap \cR \cap \Psi) = \# \{ \x \in \cR : d_i | q_i(\x),\
56: i=1,\ldots,g,\ \x \in \Psi \}$. As in our pairs of forms article
57: \cite{Mar06}, the Lemma follows by partitioning
58: this set according to $(\x;\psi(\bd))$.
59: \begin{explain}
60: Such a partition lets us write $\# (\ld \cap \cR \cap \Psi)$ as:
61: $$
62: \sum_{b | \psi(\bd)} \# \{ \x \in \cR : q_i(\x) \equiv 0
63: \smod{d_i},\ (\x ; \psi(\bd)) = b, \ \x \in \Psi \}.
64: $$
65: We claim that if $b | \psi(\bd)$ then $\psi(\bd)/b =
66: \psi(\bc)$. Indeed, if $b = \prod p^\beta$, and $d_i = \prod
67: p^{\alpha_i}$, then $\psi(\bd)/b =
68: \prod_p p^{\lceil (\max(\alpha_1,\ldots,\alpha_g) -2 \beta)/2\rceil }$.
69: On the other hand, $$\psi(\bc) = \prod_p p^{\lceil \max_{i=1,\ldots,g}(\alpha_i -
70: \min(\alpha_i,2\beta))/2 \rceil}.$$
71: Now for each $i$, $$\alpha_i - \min(\alpha_i,2\beta) = \begin{cases}0 &
72: \text{ if } \alpha_i \le 2 \beta, \\ \alpha_i - 2 \beta & \text{ if }
73: \alpha_i \ge 2 \beta.\end{cases}$$
74: Therefore $$\max_{i=1,\ldots,g}(\alpha_i - \min(\alpha_i,2\beta)) =
75: \max_{i=1,\ldots,g}(\alpha_i)-2\beta.$$
76:
77: It is clear that $(\by;\psi(\bc))=1 \iff (\by;c_1\ldots c_g)=1$.
78: Write $\x = b \by$; we have $b | \psi(\bd)$, so $(b;D)=1$. We may
79: write $b^{-1} \x \equiv \by \smod{D}$, and deduce
80: $$
81: \# (\ld \cap \cR \cap \Psi) = \sum_{b|\psi(\bd)} \# \{ \by \in \cR/b
82: : c_i |q_i(\by), \ (\by;c_1\ldots c_g)=1, \by \in \Psi_b\},
83: $$
84: hence the result.
85: \end{explain}
86: \subsection{Upper Bounds for $\rho$}
87:
88: \subsubsection{The Function $\rstar$}
89:
90: \begin{lemma} \label{lem:rstarbound}
91: For every prime $p$ one has that $\rstar(p^{e_1},\ldots, p^{e_g})
92: \ll p^{\max(e_1,\ldots,e_g)},
93: $
94: and if $(p;D)=1$ then
95: \begin{equation} \label{eqn:rstarsplit}
96: \rstar(p^{e_1},\ldots,p^{e_g}) = 0, \text{ if } e_i, e_j>0 \text{
97: for some }i \ne j.
98: \end{equation}
99: \end{lemma}
100:
101: Assume $\rstar(p^{e_1},\ldots,p^{e_g}) \ne 0$. As in \cite{Mar06}, we
102: may demonstrate that if $i \ne j$, then
103: $p^{\min(e_i,e_j)}|\Res(q_i,q_j)$, and hence $p^{\min(e_i,e_j)}|D$.
104: \begin{explain}
105: To see this, note that there exists $\x
106: = (x_1,x_2)$ such that $(\x;p)=1$ and $p^{e_i} | q_i(\x)$ for
107: $i=1,\ldots g$. Without loss of generality, $p \nmid x_2$. Define
108: $Q_i(Y) := q_i(Y,1)$, and $y \equiv x_1 x_2^{-1}
109: \smod{p^{\max(e_1,\ldots,e_g)}}$. Then $0 \equiv q_i(x_1,x_2) \equiv
110: x_2^2 Q_i(y) \smod{p^{e_i}}$. Therefore $p^{e_i} | Q_i(y)$ for
111: $i=1,\ldots, g$. So for all $i \ne j$, $Q_i(\by) \equiv Q_j(\by)
112: \equiv 0 \smod{p^{\min(e_i, e_j)}}$. We deduce
113: $p^{\min(e_i,e_j)} | \Res(Q_i,Q_j)=\Res(q_i,q_j)$.
114: \end{explain}
115: Suppose $(p;D)=1$; then $\min(e_i,e_j)=0$ for every pair $i\ne j$.
116: This suffices for the second part of the Lemma. The first part is
117: proved by induction on $g$. Our previous paper \cite{Mar06} contains
118: a proof for the base case, $g=2$. In general, one has
119: \begin{align*}
120: \rstar(p^{e_1},\ldots,p^{e_g}) &\le \# \{ \x \smod{p^{e_1+\ldots + e_g}}
121: : p^{e_i}| q_i(\x), (\x;p)=1, i=2,\ldots,g\} \\
122: & = p^{2e_1} \# \{ \x \smod{p^{e_2+\ldots+e_g}} : p^{e_i} | q_i(\x),
123: i=2,\ldots, g ; (\x;p)=1\} \\
124: & \ll p^{2 e_1} p^{\max(e_2,\ldots,e_g)} \text{ by induction on }g \\
125: & \le p^{2 e_1 \max(e_1,\ldots,e_g)}
126: \end{align*}
127: Generalising, we have that for all $i$, $\rstar(p^{e_1},\ldots,
128: p^{e_g}) \ll p^{2e_i} p^{\max(e_1,\ldots,e_g)}$. Thus
129: $$
130: \rstar(p^{e_1},\ldots,p^{e_g}) \ll p^{2 \min(e_i, e_j)}
131: p^{\max(e_1,\ldots e_g)} \ll D^2 p^{\max(e_1,\ldots,e_g)} \ll p^{\max(e_1,\ldots,e_g)}.
132: $$
133:
134: \subsubsection{The Function $\rho$}
135:
136:
137: \begin{lemma} \label{lem:rbound}
138: Let $p$ be a prime and let $e_1, \ldots, e_g$ be non-negative
139: integers. Let $\sigma$ be a permutation in $S_g$ such that
140: $e_{\sigma(1)} \le \ldots \le e_{\sigma(g)}$. Then
141: $$
142: \rho(p^{e_1},\ldots,p^{e_g}) \ll (e_{\sigma(g)} - e_{\sigma(g-1)} +1 )p^{2
143: e_{\sigma(1)}+\ldots+2e_{\sigma(g-1)}+e_{\sigma(g)}}.
144: $$
145: Also,
146: $$\rho(p,\ldots,p) = \rstar(p,\ldots,p)+p^{2(g-1)} = p^{2(g-1)} + O(p),$$
147: and for all but a finite set of primes $p$, one has
148: $$
149: \rho(p,1,\ldots,1) \le 2p ;\ \rho(1,p,1,\ldots,1) \le 2p,\ \ldots, \
150: \rho(1,\ldots,1,p) \le 2p.$$
151: \end{lemma}
152:
153: An application of Lemma \ref{lem:transition} provides us with the
154: following formul\ae:
155: \begin{equation}
156: \rho(\bd) = \sum_{b | \psi(\bd)} \# (\lstarc \cap [0,d_1 \ldots d_g /
157: b)^2) = \sum_{b | \psi(\bd)} \rstar(\bc) \left( \frac{(d_1 ; b^2)
158: \ldots (d_g;
159: b^2)}{b} \right)^2, \label{eqn:rhotransition}
160: \end{equation}
161: where $c_i = d_i / (d_i ; b^2)$.
162: In particular, $\rho(p^{e_1},\ldots,p^{e_g})$ can be written as
163: \begin{equation}
164: \sum_{0 \le \beta \le \ceil{e_g/2}} \rstar \left(
165: \frac{p^{e_1}}{(p^{e_1} ; p^{2 \beta})}, \ldots,
166: \frac{p^{e_g}}{(p^{e_g} ; p^{2 \beta})}
167: \right) \left( \frac{(p^{e_1} ; p^{2 \beta})\ldots(p^{e_g} ; p^{2
168: \beta})}{p^\beta} \right)^2. \label{eqn:rhorstar}
169: \end{equation}
170:
171: Using these formul\ae and our upper bound for $\rstar$,
172: we may derive a proof of the first two statements, much as in \cite{Mar06}.
173: \begin{explain}
174: Indeed, without loss of generality, we shall assume $e_1 \le \ldots \le e_g$.
175: Split the range of summation as $0 \le 2 \beta < e_1$, $e_1 \le 2 \beta <
176: e_2, \ldots, e_{g-1} \le 2 \beta < e_g$ and $\beta=\ceil{e_g/2}$. We
177: have the following expression for
178: $\rho(p^{e_1},\ldots, p^{e_g})$:
179: \begin{align*}
180: &\sum_{0 \le \beta < e_1/2}
181: \rstar(p^{e_1-2\beta},\ldots,p^{e_g-2\beta}) p^{\beta(4g-2)} \\
182: &+ \sum_{e_1/2 \le \beta < e_2/2}
183: \rstar(1,p^{e_2-2\beta},\ldots,p^{e_g-2\beta}) p^{2 e_1}
184: p^{\beta(4g-6)} \\
185: &+ \sum_{e_2/2 \le \beta < e_3/2}
186: \rstar(1,1,p^{e_3-2\beta},\ldots,p^{e_g-2\beta}) p^{2e_1 +2e_2}
187: p^{\beta(4g-10)} \\
188: &+\ldots+ \sum_{e_{g-1}/2 \le \beta < e_g/2}
189: \rstar(1,\ldots,1,p^{e_g-2\beta}) p^{2e_1+\ldots 2e_{g-1}}
190: p^{2\beta} \\
191: &+ p^{2 e_1+\ldots + 2e_{g-1}+e_g}.
192: \end{align*}
193: Via an application of our upper bound for $\rstar$, we see that for
194: each $0\le i \le g-2$, the
195: sum over the range $e_i/2 \le \beta < e_{i+1}/2$ is bounded by
196: $p^{2e_1+\ldots+2e_i+e_g}p^{2e_{i+1}(g-i-1)}$. This is an increasing
197: function of $i$, so $\rstar$ is bounded above by $p^{2e_1+\ldots+2
198: e_{g-1}+e_g} + (e_g-e_{g-1})p^{2e_1+\ldots+2e_{g-1}+e_g}$, as
199: required.
200:
201: An application of equation (\ref{eqn:rhorstar}) gives
202: $$
203: \rho(p,\ldots,p) = \rstar(p,\ldots,p)+p^{2(g-1)} = p^{2(g-1)} +O(p),
204: $$
205: \end{explain}
206: The final statement of Lemma \ref{lem:rbound} is proved using the
207: one-form result of \cite{Mar06}.
208:
209:
210: \subsection{Level of Distribution---Starred Version}
211:
212: \begin{lemma} \label{thm:lodstar}
213: Define
214: \[
215: T^{*}(M,\bQ) := \sum_{\substack{d_i \le Q_i \\ (d_i;D)=1}} \sup_{\cR
216: : \partial R \le M}
217: \left| \# (\lstard \cap \cR \cap \Psi) - \frac{\rstar(d_1,\ldots, d_g)}
218: {(d_1 \ldots d_g D)^2}
219: \vol(\cR)\right|.
220: \]
221: Then there exist constants $\nu_1'$
222: and $\nu_2'$ depending only on $g$ such that
223: \[ T^{*}(M,\bQ) \ll M \sqrt{Q} ( \log(2 Q))^{\nu_1'} +
224: Q ( \log (2 Q) )^{\nu_2'}
225: \]
226: uniformly for $M > 0$ and $Q_1, \ldots, Q_g \ge 1$.
227: \end{lemma}
228:
229: \subsubsection{The Quantities $\lstard$}
230:
231: Assume that $\bd = (d_1, \ldots, d_g)$ is fixed and
232: define $a := d_1 \ldots d_g$. Let $\cU(a)$ be the set of equivalence classes
233: of $\x \in \Z^2$ under multiplication with $(x_1 ; x_2 ; a)=1$, as in
234: \cite{Mar06}. For a given $\cA \in \cU(a)$, one has that $\cA \subset
235: \lstard$ or $\cA \cap \lstard = \emptyset$. Define $\cU'(\bd) := \{
236: \cA \in \cU(d_1\ldots d_g) : \cA \subset \lstard \}.$ Then $\lstard$
237: is the disjoint union
238: $
239: \lstard = \bigcup_{\cA \in \cU'(\bd)}{\cA}.
240: $
241:
242: Now $\# (\cA \cap [0,a)^2) = \phi(a)$, so
243: \begin{equation} \label{eqn:rhoeqn}
244: \rstar(\bd) = \# \cU'(\bd) \phi(d_1 \ldots d_g).
245: \end{equation}
246:
247: Using this equation, the summand can be bounded from above:
248: \begin{equation} \label{eqn:lodstardd}
249: \begin{split}
250: & \left| \# (\lstard \cap \cR \cap \Psi) - \frac{\rstar(\bd)}
251: {(d_1 \ldots d_g D)^2}
252: \vol(\cR)\right| \\
253: \le & \sum_{ \cA \in \cU'(\bd)} \left| \# (\cA \cap \cR \cap \Psi) -
254: \frac{\phi(d_1 \ldots d_g)}{(d_1 \ldots d_g D)^2} \vol(\cR) \right|.
255: \end{split}
256: \end{equation}
257:
258: \subsubsection{Estimating $\#(\cA \cap \cR \cap \Psi)$}
259:
260: Choose
261: $\cA \in \cU(a)$ and define $G(\cA)$ by:
262: \[
263: G(\cA) := \{ \x \in \Z^2 : (\exists \lambda \in \Z) (\exists \by \in
264: \cA) (\x \equiv \lambda \by \smod{a}) \}.
265: \]
266: As in \cite{Mar06}, the lattice $G(\cA)$ has a non-zero element of minimal
267: length $\bv(\cA)$. Moreover, $\bv(\cA)$ satisfies $|\bv(\cA)|^2 \ll
268: a$, and
269: $$
270: \#(\cA \cap \cR \cap \Psi)
271: = \frac{\phi(a)}{a^2D^2} \vol(\cR) + O \left( d(a) \left(
272: \frac{M}{|\bv(\cA) |} + 1 \right) \right). \label{eqn:caeqn}
273: $$
274: Substituting this into equation (\ref{eqn:lodstardd}), we
275: have:
276: \begin{equation} \label{eqn:lodstar5}
277: \begin{split}
278: T^{*}(M, \bQ) & \ll \sum_{\substack{d_i \le Q_i \\ (d_i;D)=1}}
279: \sum_{\cA \in
280: \cU'(\bd)} d( d_1 \ldots d_g) \left(
281: \frac{M}{|\bv(\cA)|} + 1 \right) \\
282: & \ll M \sum_{d_i \le Q_i} d(d_1 \ldots d_g) \sum_{\cA \in
283: \cU'(\bd)}
284: \frac{1}{| \bv(\cA) |} + \sum_{d_i \le Q_i} d(d_1 \ldots d_g)
285: \# \cU'(\bd) \\
286: & = M T_1^{*}(\bQ) + T_2^{*}(\bQ), \text{ say.}
287: \end{split}
288: \end{equation}
289:
290: \subsubsection{Evaluating \protect{$T_1^{*}(\bQ)$}}
291:
292: We shall prove:
293: \begin{lemma} \label{lem:t1bound}
294: The quantity
295: $$
296: T_1^*(\bQ) := \sum_{d_i \le Q_i} d(d_1 \ldots d_g) \sum_{\cA \in \cU'(\bd)}
297: \frac{1}{| \bv(\cA) |}
298: $$
299: satisfies the upper bound $ T_1^*(\bQ) \ll \sqrt{Q} ( \log 2 Q)^{\nu_1'},$
300: for some constant $\nu_1'$ depending only on $g$, and where $Q := Q_1
301: \ldots Q_g$.
302: \end{lemma}
303:
304: For $\cA \in \cU'(\bd)$, we have $| \bv(\cA)| \ll \sqrt{d_1 \ldots
305: d_g}$. Consequently,
306: $$
307: T_1^{*}(\bQ) \le \sum_{0 < | \bv| \ll \sqrt{Q}} \frac{1}{| \bv
308: |} \sum_{\substack{d_i \le Q_i \\ d_i | q_i(\bv)}} d(d_1 \ldots d_g)
309: \# \cU'(\bd).
310: $$
311:
312: Write $d_i = \prod p^{e_i}$ and apply equation (\ref{eqn:rhoeqn}):
313: $$
314: \# \cU'(\bd) = \frac{\rstar(\bd)}{\phi(d_1 \ldots d_g)}
315: = \prod_p \frac{\rstar(p^{e_1},\ldots,p^{e_g})}{\phi(p^{e_1+\ldots+e_g})},
316: $$
317: by multiplicativity of $\rstar$. By Lemma \ref{lem:rstarbound}, if
318: $(p;D)=1$ and if at least two of the
319: $e_i$ are positive,
320: then $\rstar(p^{e_1},\ldots,p^{e_g})=0$. Thus $\# \cU'(\bd)=0$ unless
321: for all $p$
322: satisfying $(p;D)=1$, we have $e_i=0$ for all but at most one $i$. In which
323: case, we may write:
324: $$
325: \# \cU' (\bd) \le
326: \prod_{\substack{p : \\(p;D)=1}}
327: \prod_{i=1,\ldots,g}
328: \frac{\rstar(p^{e_i \delta_{1i}},\ldots,p^{e_g \delta_{g_i}})}
329: {\phi(p^{e_1 \delta_{1i}+\ldots e_g \delta_{gi}})}
330: \prod_{\substack{p : \\ p | D}} C
331: \frac{p^{\max(e_1,\ldots,e_g)}}{\phi(p^{e_1+\ldots e_g})} \ll 2^{\nu(d_1\ldots d_g)},
332: $$
333: much as in \cite{Mar06}.
334:
335: \begin{explain}
336: Observing that $2^{\nu(d_1 \ldots d_g)} \le d(d_1 \ldots d_g)$, and using the
337: submultiplicativity property of $d$, we deduce that
338: $
339: d(d_1 \ldots d_g) \# \cU'(\bd) \ll d(d_1)^2 \ldots d(d_g)^2,
340: $
341: whence
342: $$
343: T_1^{*}(\bQ) \ll \sum_{0 < | \bv| \le \sqrt{Q}} \frac{1}{|\bv|}
344: \sum_{d_i | q_i(\bv)} d(d_1)^2 \ldots d(d_g)^2.
345: $$
346: \end{explain}
347: Defining the function $h$ by
348: $
349: h(n) := \sum_{a | n} d(a)^2,
350: $
351: we have
352: \begin{equation} \label{eqn:lodstar6}
353: T_1^{*}(\bQ) \ll \sum_{0 < |\bv| \le \sqrt{Q}} \frac{1}{|\bv|}
354: h(q_1(\bv)) \ldots h(q_g(\bv)).
355: \end{equation}
356: Regarding this function, we have the following Lemma, to be found
357: in \cite{Mar06}:
358:
359: \begin{lemma} \label{lem:hdivisor}
360: The function $h$ is multiplicative. Moreover, $h$ is
361: submultiplicative in the sense that for all $m_1, m_2$, we have
362: $h(m_1 m_2) \le h(m_1) h(m_2)$. Furthermore, we have $h(p) \ll 1$
363: uniformly in $p$.
364: Let $\eta \ge 1$, then for every natural number $n$, there exists a
365: positive integer $m$ satisfying $m|n$, $m \le n^{1 /\eta}$, and
366: $$
367: h(n) \ll_\eta h(m)^{1+\lfloor \eta\rfloor}.
368: $$
369: \end{lemma}
370:
371:
372: We apply Lemma \ref{lem:hdivisor} to equation
373: (\ref{eqn:lodstar6}), with $\eta = 2g$, to obtain:
374: $$
375: T_1^*(\bQ) \ll \sum_{\substack{j \ge 0 \\ P = 2^j \ll \sqrt{Q}}}
376: \frac{1}{P} \sum_{P \le |\bv| \le 2P} \sum_{\substack{m_i | q_i(\bv)
377: \\ m_i \ll P^{1/g} \\ i=1,\ldots,g}} h(m_1)^{2g+1} \ldots h(m_g)^{2g+1};
378: $$
379: \begin{explain}
380: where we have split the range for $| \bv |$ into dyadic intervals, and
381: used the fact that $q_i(\bv) \ll | \bv|^2$ to deduce that $m_i \le
382: |q_i(\bv)|^{1/(2g)}$ implies $m_i \ll P^{1/g}$;
383: \end{explain}
384: we have
385: $$
386: T_1^*(\bQ) \ll \sum_{\substack{j \ge 0 \\ P = 2^j \ll \sqrt{Q}}}
387: \frac{1}{P}
388: \sum_{m_i \ll P^{1/g}} h(m_1)^{2g+1} \ldots h(m_g)^{2g+1} \sum_{\substack{|\bv| \le
389: 2P \\ q_i(\bv) \equiv 0 \smod{m_i} \\ i=1, \ldots g}} 1.
390: $$
391: The innermost sum
392: \begin{explain}
393: is of order $\rho(m_1,\ldots, m_g)\{ P^2 / (m_1
394: \ldots m_g)^2 +
395: P/(m_1 \ldots m_g)\}$,
396: where the second term accounts for the error at the boundary. We have
397: arranged that $m_1 \ldots m_g \ll P$, whence the second term is subsumed by
398: the first, and hence the innermost sum
399: \end{explain}
400: is of order $\rho(m_1, \ldots, m_g) P^2
401: / (m_1 \ldots m_g)^2$, so
402: \begin{align*}
403: T_1^*(\bQ) & \ll \sum_{\substack{j \ge 0 \\ P=2^j \ll \sqrt{Q}}} P
404: \sum_{m_i \ll P^{1/g}}\frac{h(m_1)^{2g+1} \ldots h(m_g)^{2g+1}
405: \rho(m_1, \ldots, m_g)}{m_1^2 \ldots m_g^2}\\
406: & \ll \sqrt{Q} \sum_{m_i \ll Q^{1/(2g)}}
407: \frac{h(m_1)^{2g+1} \ldots h(m_g)^{2g+1} \rho(m_1, \ldots,
408: m_g)}{m_1^2 \ldots m_g^2}.
409: \end{align*}
410: The summand is multiplicative, so we have the following upper bound:
411: $$
412: T_1^*(\bQ) \ll \sqrt{Q} \prod_{p \ll Q^{1/(2g)}}
413: \sum_{\substack{e_i=0 \\i=1,\ldots,g}}^\infty \frac{h(p^{e_1})^{2g+1}
414: \ldots h(p^{e_g})^{2g+1} \rho(p^{e_1},\ldots,p^{e_g})}{p^{2e_1+\ldots+2e_g}}.
415: $$
416: Let $k(e_1,\ldots,e_g)$ denote the summand. Using our upper bound for
417: $\rho$, we estimate the
418: sum $S= \sum_{e_i=0}^\infty k(e_1,\ldots,e_g)$ as follows:
419: \begin{align*}
420: S \le & \sum_{e_1=0}^\infty \sum_{e_2,\ldots,e_g \le e_1}
421: k(e_1,\ldots,e_g) + \sum_{e_2=1}^\infty \sum_{\substack{e_i \le e_2 \\
422: i \ne 2}} k(e_1,\ldots,e_g) \\ & + \ldots + \sum_{e_g=1}^\infty
423: \sum_{\substack{e_i \le e_g \\ i \ne g}} k(e_1,\ldots,e_g).
424: \end{align*}
425:
426: Now
427: $$
428: \sum_{e_1=0}^\infty \sum_{\substack{e_i\le e_1 \\ i \ne 1}}
429: k(e_1,\ldots,e_g) = 1 + \frac{C_1}{p}+\frac{C}{p^2} +
430: \sum_{e_1=2}^\infty \sum_{\substack{e_i \le e_1 \\ i \ne 1}}
431: k(e_1,\ldots,e_g).
432: $$
433: \begin{explain}
434: The first term in this expression arises from $k(0,\ldots,0)$, the
435: second from $k(1,0,\ldots,0)$, and the third from those sets of $e_i$
436: where $e_1=1$ and at least one other $e_i=1$, the bound deriving from
437: the $n$-variable result $\rho(p,\ldots,p) \ll p^{2(n-1)}$.
438: \end{explain}
439: We have
440: $$
441: S' := \sum_{e_1=2}^\infty \sum_{\substack{e_i \le e_1 \\ i \ne 1}}
442: k(e_1,\ldots,e_g) \ll \sum_{e_1=2}^\infty \sum_{e_i \le e_1}
443: \frac{e_1^{6g+3} \ldots e_g^{6g+3} e_1}{p^{e_1}},
444: $$
445: using $h(p^e) \ll e^3$ and $\rho(p^{e_1},\ldots,p^{e_g}) \ll e_1 p^{2
446: e_2 + \ldots 2 e_g +e_1}$.
447: So $S' \ll \sum_{e_1=2}^\infty \frac{e_1^A}{p^{e_1}}$, for some $A \in
448: \N$, depending on $g$.
449: Therefore $p^2 S' \ll \sum_{e_1=2}^\infty \frac{e_1^A}{p^{e_1-2}} \le
450: \sum_{e_1=2}^\infty \frac{e_1^A}{2^{e_1-2}} < \infty,$ by the ratio
451: test.
452:
453: Thus
454: $$
455: \sum_{e_1=0}^\infty \sum_{\substack{e_i\le e_1 \\ i \ne 1}}
456: k(e_1,\ldots, e_g) \le 1 + \frac{C_1}{p}+\frac{C}{p^2},
457: $$
458: for a new constant $C$. By a similar argument,
459: $$
460: \sum_{e_i=1}^\infty \sum_{\substack{e_j \le e_i \\ j \ne i}}
461: k(e_1,\ldots,e_g) \le \frac{C_1}{p} + \frac{C}{p^2},
462: $$
463: for $i=1,\ldots,g$. Thus, with a possible change of constants $C_1$
464: and $C$,
465: $$
466: \sum_{\substack{e_i=0 \\ i=1,\ldots,g}}^\infty k(e_1,\ldots,e_g) \le 1 +
467: \frac{C_1}{p}+\frac{C}{p^2}.
468: $$
469:
470: The constant $C_1$ could potentially depend on the forms in question.
471: However, by a more careful analysis, one can
472: remove this dependence and ensure that $C_1$ depends only on $g$.
473:
474: We have:
475: $$
476: T_1^*(\bQ) \ll \sqrt{Q} \prod_{p \ll Q^{1/(2g)}} \left( 1
477: + \frac{C'}{p} + \frac{C}{p^2} \right ).
478: $$
479: Much as in \cite{Mar06}, this implies our desired upper bound.
480: \begin{explain}
481: Indeed consider the following Lemma:
482: \begin{lemma} \label{lem:mertensresult}
483: Let $Q > 1$ and $C > 0$ be real numbers. Let $k$ be a natural
484: number and define
485: $$ S' = \prod_{p \le Q} \left( 1 + \frac{k}{p} +
486: \frac{C}{p^2} \right).$$
487: Then
488: $$
489: S' \ll_{k, C}(\log Q)^k.
490: $$
491: \end{lemma}
492: \end{explain}
493:
494: \subsubsection{Evaluating \protect{$T_2^{*}(\bQ)$}}
495: We shall prove
496: \begin{lemma} \label{lem:t2bound}
497: The quantity $T_2^*$ satisfies the upper bound
498: $$
499: T_2^*(\bQ) \ll Q \log(2 Q)^{\nu_2'}.
500: $$
501: for some constant $\nu_2'$, depending only on $g$.
502: \end{lemma}
503:
504: \begin{explain}
505: Recall
506: $$
507: T_2^*(\bQ) := \sum_{\substack{d_i \le Q_i \\ i = 1,\ldots,g}}d(d_1
508: \ldots d_g) \# \cU'(\bd).
509: $$
510: \end{explain}
511: In our analysis of the sum $T_1^*(\bQ)$, we demonstrated that $\#
512: \cU'(\bd) \ll 2^{\nu(d_1 \ldots d_g)}$. We have that $2^{\nu(a)} \le
513: d(a)$ for any $a$ and that the $d$ function satisfies $d(ab) \le d(a)
514: d(b)$ for any $a$ and $b$. Thus,
515: \begin{align*}
516: T_2^*(\bQ) & \ll \left( \sum_{d_1 \le Q_1} d(d_1)^2 \right) \ldots \left(
517: \sum_{d_g \le Q_g}d(d_g)^2\right) \ll Q (\log Q_1)^3 \ldots (\log
518: Q_g)^3 \\
519: & \ll Q (\log 2 Q)^{3g},
520: \end{align*}
521: where we use the AM--GM inequality in the last line.
522: This proves Lemma
523: \ref{lem:t2bound}. Combining this with Lemma \ref{lem:t1bound}
524: gives us our starred level of distribution formula, Lemma
525: \ref{thm:lodstar}.
526:
527:
528: \subsection{Level of Distribution---Unstarred Version}
529: \label{sec:lodunstarred}
530:
531: Recall our convention that the symbol $c_i$ represents $d_i / (d_i ;
532: b^2)$. We apply Lemma \ref{lem:transition} and equation
533: (\ref{eqn:rhotransition}) to give the following expression for $T(M,\bQ)$:
534: \begin{align*}
535: &\sum_{\substack{d_i \le Q_i \\ (d_i ; D)=1}}
536: \sup_{\substack{\cR: \\ \partial(\cR) \le M}}
537: \left| \sum_{b | \psi(\bd)} \left\{ \#(\lstarc \cap \cR / b \cap \Psi_b) -
538: \frac{\rstar(\bc)}{(c_1 \ldots c_g D)^2} \vol(\cR/b) \right\} \right| \\
539: &\le \sum_{\substack{c_i \le Q_i \\ (c_i;D)=1 }} \sum_{\substack{b
540: \le Q \\ (b_i ; D ) =1}} \delta(\bQ,\bc,b)
541: \sup_{\substack{\cR : \\ \partial(\cR) \le M}}
542: L(\bc,b,\cR),
543: % \left| \#(\lstarc \cap \cR / b \cap \Psi_b) -
544: % \frac{\rstar(\bc)}{(c_1 \ldots c_g D)^2} \vol(\cR/b) \right|,
545: \end{align*}
546: where we write $L(\bc,b,\cR)$ in place of
547: $$
548: \left| \#(\lstarc \cap \cR / b \cap \Psi_b) -
549: \frac{\rstar(\bc)}{(c_1 \ldots c_g D)^2} \vol(\cR/b) \right|,
550: $$
551: and where $\delta(\bQ,\bc,b) = \# \{ (d_1, \ldots d_g) : d_i \le Q_i,\ c_i = d_i /
552: (d_i ; b^2),\ b | \psi (\bd) \}$.
553:
554: We shall derive an upper bound for $\delta$. Note that
555: $\delta(\bQ,\bc,b) \le \prod_{i=1}^g \Delta(c_i,b)$, where $\Delta(c_i,b) := \#
556: \{d_i: c_i = d_i/(d_i;b^2) \}.$ Suppose $d_i = \prod p^{\alpha_i}$,
557: $b = \prod p^\beta$, and $c_i = \prod p^{\gamma_i}$. We can estimate
558: $\Delta(c_i,b)$ by determining how many choices there are for each
559: $\alpha_i$. We require $\alpha_i - \min(\alpha_i, 2 \beta) =
560: \gamma_i$. For each $i$, we have either $\gamma_i=0$ or
561: $\gamma_i>0$. If $\gamma_i=0$, then $\alpha_i \le 2 \beta$, so there
562: are $2 \beta+1$ choices for $\alpha_i$. If $\gamma_i >0$, then
563: $\alpha_i = \gamma_i+2\beta$, so there is only one choice for
564: $\alpha_i$. In either case, there are at most $2\beta+1$ choices for
565: $\alpha_i$. Thus $\Delta(c_i,b) \le \prod_{p^\beta || b }3 \beta$,
566: and hence $\delta(\bQ,\bc,b) \le \prod_{p^\beta||b} (3 \beta)^g
567: =: \theta(b)$.
568:
569: By analogy with \cite{Mar06}, we have $\sum_{b \le B}
570: \delta(\bQ,\bc,b) \ll B (\log B)^{A_1}$, where $A_1 := 3^g-1$.
571: \begin{explain}
572: Indeed, by
573: induction on $\beta$, one has $3 \beta \le \binom{\beta+2}{2}$ (for
574: the induction step, note $\binom{\beta+3}{2} = \binom{\beta+2}{2} +
575: \binom{\beta+2}{1}).$
576: One has $d_3(b) = \prod_{p^\beta || b} \binom{\beta+2}{2}$ (see \cite{IK04},
577: section 1.4) and $\sum_{b \le B} d_3(b)^g \ll B (\log B)^{3^g -1}$
578: (ibid. section 1.6), so $\sum_{b \le B} \delta(\bQ,\bc,b) \le \sum_{b
579: \le B} \theta(b) \ll B (\log B)^{A_1}$.
580: \end{explain}
581: If $\delta(\bQ,\bc,b) \ne 0$, then we may deduce that $c_1 \ldots c_g
582: b \le Q$, restricting the range of summation in the level of
583: distribution formula.
584: \begin{explain}
585: Indeed, suppose that $\delta(\bQ,\bc,b)
586: \ne 0$, then there exist $d_1, \ldots d_g$ such that $b | \psi(\bd)$ and
587: $c_i = d_i / (d_i ; b^2)$. One may verify, by restricting to prime
588: powers, that $b | \psi(\bd)$ implies $b| (d_1 ; b^2)\ldots (d_g ;
589: b^2)$. This may be rewritten as $c_1 \ldots c_g b | d_1 \ldots d_g$,
590: from which it follows that $c_1 \ldots c_g b \le Q$.
591: \end{explain}
592:
593: Our sum $T(M,\bQ)$ is estimated by
594: \begin{explain}
595: \begin{align*}
596: T(M,\bQ) & \ll \sum_{\substack{c_i: \\ (c_i;D)=1 \\ c_i \le Q_i}}
597: \sum_{\substack{b : \\ (b;D)=1 \\ c_1 \ldots c_g b \le
598: Q}}\delta(\bQ, \bc, b) \sup_{\substack{\cR : \\\partial(\cR)
599: \le M }} L(\bc,b,\cR) \\
600: & \le \sum_{\substack{j_i : \\ C_i = 2^{j_i} \le Q_i}}
601: \sum_{\substack{C_i \le c_i \le 2 C_i \\ (c_i;D)=1}}
602: \sum_{\substack{b : \\ (b;D)=1 \\b \le
603: \frac{Q}{c_1 \ldots c_g}}} \theta (b)
604: \sup_{\substack{\cR : \\ \partial(\cR) \le M }}
605: L(\bc,b,\cR).
606: \end{align*}
607:
608: If we further split the range for $b$ into dyadic intervals, then
609: \end{explain}
610: $$
611: T(M,\bQ) \ll \sum_{\substack{j_i : \\ C_i = 2^{j_i} \le Q_i}}
612: \sum_{\substack{C_i \le c_i \le 2 C_i \\ (c_i;D)=1}}
613: \sum_{\substack{k: \\ B = 2^k \le \frac{Q}{c_1 \ldots c_g}}}
614: \sum_{\substack{b : \\ (b;D)=1 \\B \le b \le 2 B}} \theta(b)
615: \sup_{\partial(\cR) \le M} L(\bc,b,\cR).
616: $$
617: \begin{explain}
618: Our aim is to use the estimate for $\sum_b \theta(b)$, but we need to
619: handle sensitively the factor of $\sup L(\bc,b,\cR)$.
620: \end{explain}
621: For each choice
622: of $B$, define $b(B)$ by requiring $B \le b(B) \le 2B$, $(b(B);D)=1$
623: and requiring that for all $b$ with $B \le b \le 2B$ and $(b;D)=1$, one has
624: $$
625: \sup_{\partial(\cR) \le M} L(\bc,b,\cR) \le \sup_{\partial(\cR) \le M
626: } L(\bc,b(B),\cR).
627: $$
628: % A potential problem arises if there are no $b$ in the range $B \le b
629: % \le 2B$ such that $(b;D)=1$. This is more a problem of notation than
630: % a real issue, as the problematic $B$s would leave us with a sharper
631: % upper bound. None the less,
632: Let $S$ denote the set of integers $B$
633: such that there are no $b$ in the range $B \le b \le 2B$ with
634: $(b;D)=1$. We have the upper bound:
635: \begin{align*}
636: T(M,\bQ)
637: & \ll \sum_{\substack{j_i : \\ C_i = 2^{j_i} \le Q_i}}
638: \sum_{\substack{C_i \le c_i \le 2 C_i \\ (c_i;D)=1}}
639: \sum_{\substack{k: \\ B = 2^k \le \frac{Q}{c_1 \ldots c_g} \\ B
640: \not\in S}}
641: B (\log 2B)^{A_1} \sup_{\partial(\cR) \le M} L(\bc,b(B),\cR) \\
642: & \le \sum_{\substack{j_i : \\ C_i = 2^{j_i} \le Q_i}}
643: \sum_{\substack{k: \\ B = 2^k \le \frac{Q}{C_1 \ldots C_g} \\ B
644: \not\in S}} B (\log 2B)^{A_1} \sum_{\substack{C_i \le c_i \le 2
645: C_i \\ (c_i;D)=1}} \sup_{\partial(\cR) \le M} L(\bc,b(B),\cR).
646: \end{align*}
647: Writing $\cR' := \cR/b(B)$, we may now apply our starred level of
648: distribution formula (Lemma
649: \ref{thm:lodstar}) to the inner sum, which is bounded from above by
650: \begin{align*}
651: &\sum_{\substack{C_i \le c_i \le 2 C_i \\ (c_i;D)=1}}
652: \sup_{\partial(\cR') \le M/B }
653: \left| \#(\lstarc \cap \cR' \cap \Psi_{b(B)}) -
654: \frac{\rstar(\bc)}{(c_1 \ldots c_g D)^2} \vol(\cR') \right| \\
655: & \ll \frac{M}{B} \sqrt{C_1 \ldots C_g} ( \log 2^{g+1} C_1 \ldots
656: C_g)^{\nu_1'} + C_1 \ldots C_g (\log 2^{g+1} C_1 \ldots C_g)^{\nu_2'}.
657: \end{align*}
658: Applying the same reasoning as in \cite{Mar06}, we may then deduce the
659: level of distribution formula.
660:
661: % Local Variables:
662: % TeX-master: "master"
663: % End:
664: