24_padic.tex

ABC: source and evidence reader

24_padic.tex

Download original file

This is the original TeX manuscript source. Use the paper’s PDF for typeset equations.

Read the typeset paper

Skip source text

1 \clearpage
2 \section{Exact local determinant valuation and denominator budget}\label{sec:padic}
3 \subsection{Depletion gives an integral local matrix}
4 Fix a prime $p$ and an integer $r\ge1$. Define
5 \[
6  F_{s,p}(t)=\sum_{p\nmid k}\frac{t^k}{k^s},\qquad
7  \mathcal H_{r,p}(t)=\det(F_{i+j+1,p}(t))_{0\le i,j<r}.
8 \]
9 These series converge at every $t\in\Q_p$ with $v_p(t)>0$. They are integral power series. The Cauchy--Binet expansion from Section~\ref{sec:hankel} now ranges only over increasing $p$-unit indices.
10 
11 Let $u_1<\cdots<u_r$ be the first $r$ positive integers not divisible by $p$. Then
12 \[
13  u_j=j+\left\lfloor\frac{j-1}{p-1}\right\rfloor,
14  \qquad S_{p,r}=\sum_{j=1}^r u_j.
15 \]
16 For $h\ge1$, put $s_h=(p-1)p^{h-1}$ and write
17 \[
18  r=q_hs_h+b_h,\qquad 0\le b_h<s_h.
19 \]
20 Define the finite sum
21 \[
22  E_{p,r}=\sum_{h\ge1}\left(s_h\binom{q_h}{2}+b_hq_h\right).
23 \]
24 The terms are zero once $s_h\ge r$.
25 
26 \subsection{The residue-class energy minimum}
27 For $r$ distinct $p$-unit integers $k_1,\ldots,k_r$, the valuation of their Vandermonde product is
28 \[
29  \sum_{i<j}v_p(k_j-k_i)
30  =\sum_{h\ge1}\#\{i<j:k_i\equiv k_j\pmod{p^h}\}.
31 \]
32 There are $s_h$ unit residue classes modulo $p^h$. If their occupation counts are $n_1,\ldots,n_{s_h}$, the number of coincident pairs is $\sum_j\binom{n_j}{2}$.
33 
34 For a fixed sum $r$, this quantity is minimized when the counts differ by at most one. If $n_i\ge n_j+2$, moving one item from class $i$ to class $j$ lowers the pair count by $n_i-n_j-1>0$. At the balanced minimum, $b_h$ classes have count $q_h+1$ and the others count $q_h$. The minimum is therefore
35 \[
36  (s_h-b_h)\binom{q_h}{2}+b_h\binom{q_h+1}{2}
37  =s_h\binom{q_h}{2}+b_hq_h.
38 \]
39 The first $r$ positive $p$-units realize this minimum for every $h$ simultaneously. Ordered unit residues repeat in blocks modulo $p^h$, and each full block contains each unit class exactly once. A prefix adds at most one occurrence to each class. Hence their Vandermonde valuation is exactly $E_{p,r}$.
40 
41 \begin{theorem}[Exact depleted valuation]\label{thm:depleted-valuation}
42 If $v_p(t)=e\ge1$, then
43 \[
44  \boxed{v_p(\mathcal H_{r,p}(t))=eS_{p,r}+2E_{p,r}.}
45 \]
46 In particular, the depleted determinant is nonzero at every such argument.
47 \end{theorem}
48 \begin{proof}
49 The denominators in the depleted expansion are $p$-units. Every $r$-element index subset has Vandermonde valuation at least $E_{p,r}$. The subset $(u_1,\ldots,u_r)$ uniquely minimizes the sum of its indices. Its term has valuation $eS_{p,r}+2E_{p,r}$. Every other subset has a strictly larger index sum and no smaller Vandermonde valuation, so its term has strictly larger valuation.
50 
51 The expansion converges $p$-adically. A unique term of minimum valuation cannot be cancelled by the remaining terms. Its valuation is therefore the valuation of the whole determinant.
52 \end{proof}
53 For rank two,
54 \[
55  v_p(\mathcal H_{2,p}(t))=
56  \begin{cases}3e,&p\ne2,\\4e+2,&p=2.\end{cases}
57 \]
58 At two the first unit indices are $1,3$, of sum four, and their difference contributes one valuation before squaring. For $p>r$ in arbitrary rank, the first $r$ indices are $1,\ldots,r$ and all their differences are units. Thus the general formula reduces to $es_r$.
59 
60 \subsection{The universal undepleted expression at good primes}
61 The raw $\mathcal H_r$ is one universal expression in ordinary polylogarithms. Its local power-series interpretation converges when $v_p(t)>0$, but its coefficients can have $p$-denominators. For fixed rank, the possible denominator loss grows only logarithmically with each index, whereas its numerator power of $t$ gives linearly growing valuation. Thus the terms tend to zero and the finite Cauchy--Binet expansions pass to the convergent local limit.
62 \begin{theorem}[Exact good-prime valuation]\label{thm:good-valuation}
63 If $p>3r-1$ and $v_p(t)=e\ge1$, then
64 \[
65  v_p(\mathcal H_r(t))=e\,r(r+1)/2.
66 \]
67 \end{theorem}
68 \begin{proof}
69 Write $a_r=2r-1$ and $s_r=r(r+1)/2$. A term indexed by $k_1<\cdots<k_r$ has valuation
70 \[
71  e\sum_i k_i+2\sum_{i<j}v_p(k_j-k_i)-a_r\sum_i v_p(k_i).
72 \]
73 For every $h\ge1$, the hypothesis gives $p^h-r>a_rh$. At $h=1$ this is $p-r>a_r$. Successive increments of $p^h$ exceed $a_r$, proving the inequality for all $h$.
74 
75 If $h_i=v_p(k_i)>0$, then $k_i-i\ge p^{h_i}-r>a_rh_i$. If $h_i=0$, then $k_i-i\ge0$ and there is no denominator loss. Thus every subset other than $(1,\ldots,r)$ has valuation strictly above $es_r$. The first subset has a unit coefficient because its indices and differences are smaller than $p$. Its term has valuation $es_r$, and a unique minimum again proves equality.
76 \end{proof}
77 The strict prime threshold is retained. A good-prime equality is stronger than the all-prime lower bound below.
78 
79 \subsection{An explicit fixed correction at the remaining primes}
80 Put $a_r=2r-1$. For each prime $p$ and $1\le i\le r$, define
81 \[
82  C_{r,p,i}=\max\!\left(0,\sup_{h\ge1}\left\{
83  a_rh-\left(p^h\left\lceil\frac i{p^h}\right\rceil-i\right)\right\}\right).
84 \]
85 This is a finite integer maximum because $p^h$ eventually grows faster than $h$. Put $C_{r,p}=\sum_{i=1}^rC_{r,p,i}$. The good-prime inequality just proved implies $C_{r,p}=0$ for $p>3r-1$. Hence
86 \[
87  B_r=\prod_{p\le3r-1}p^{C_{r,p}}
88 \]
89 is a fixed positive integer depending only on the rank.
90 
91 \begin{theorem}[All-prime contact lower bound]\label{thm:budget}
92 For every prime $p$ and $v_p(t)=e\ge1$,
93 \[
94  \boxed{v_p(B_r\mathcal H_r(t))\ge es_r.}
95 \]
96 Here $v_p(0)=+\infty$. This theorem alone does not establish nonvanishing at the small primes.
97 \end{theorem}
98 \begin{proof}
99 For an ordered index $k_i\ge i$ with $h_i=v_p(k_i)>0$, the smallest multiple of $p^{h_i}$ at least $i$ is $p^{h_i}\lceil i/p^{h_i}\rceil$. Therefore
100 \[
101  a_rh_i-(k_i-i)\le C_{r,p,i}.
102 \]
103 For $h_i=0$, the same bound holds because its left side is nonpositive. The numerator Vandermonde only improves valuation. Since $e\ge1$, every term in the convergent expansion has valuation at least
104 \[
105  es_r+\sum_i[(k_i-i)-a_rh_i]\ge es_r-C_{r,p}.
106 \]
107 The ultrametric inequality gives the same bound for the sum. Multiplication by $B_r$ restores the missing $C_{r,p}$.
108 \end{proof}
109 For rank two, $B_2=2^7\cdot3^3=3456$ is a valid budget. For instance, at $p=2$ the two component maxima, in index order $i=1,2$, are three and four. At $p=3$ they are one and two. The other primes contribute zero. The formula need not be minimal.
110 
111 The budget is a theorem about the explicitly defined local analytic series. It is not a theorem about the denominator of a motivic class, which has not been constructed here. Even a vanishing correction budget need not imply exact equality at a prime outside the good-prime range. Additional cancellation may raise the valuation.
112 
113 \subsection{The all-place comparison is still absent}
114 These results give exact depleted nonvanishing, universal good-prime equality and universal all-prime lower contact. They do not identify the real value, the various $p$-adic values, or different sector arguments as the places of one nonzero algebraic number. An arithmetic realization and its norm maps remain separate requirements.

Back to source controls

Other PDFs for this work

Original file and source record

This copy matches the recorded source file. Its publication here does not change the work’s review status.

SHA-256: fa7b5e23ab4fe8cf5a62376b6bae62216d34f4ad028057b07fcb20bd669a5355

Original source on GitHub

Back to ABC: source and evidence reader