Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated

Canonical cyclic lifts of one circuit have two actual colors


The canonical cyclic lift flattens any integer relation. Unlike the earlier designated-template splitting bound, the theorem below colors all ternary relations of specified covers of a single rational circuit. It does not bound arbitrary covers or simultaneous constructions using several interacting circuits.

1. The canonical polynomial matrix

Let A∈{0,1}X×VA\in\{0,1\}^{X\times V} have distinct nonzero columns forming a rational circuit. Put n=∣V∣n=|V| and choose its primitive relation

ker⁡QA=Qa,a∈ZV,gcd⁡v∣av∣=1.\ker_{\mathbb Q}A=\mathbb Q a,\qquad a\in\mathbb Z^V,\quad \gcd_v|a_v|=1.

All av≠0a_v\ne0. Write Lv=∣av∣L_v=|a_v| and

px=∑av>0Axvav=∑av<0Axv∣av∣.p_x=\sum_{a_v>0}A_{xv}a_v =\sum_{a_v<0}A_{xv}|a_v|.

For each row, tile [0,px)[0,p_x) separately by its positive and negative incident lengths, in any fixed orders. Let gxv≥0g_{xv}\geq0 be the tile start and G=max⁡Axv=1gxvG=\max_{A_{xv}=1}g_{xv}. Set

A(z)xv=Axvzgxv,hv(z)=sgn⁡(av)SLv(z),SL(z)=1+⋯+zL−1.A(z)_{xv}=A_{xv}z^{g_{xv}},\qquad h_v(z)=\operatorname{sgn}(a_v)S_{L_v}(z),\qquad S_L(z)=1+\cdots+z^{L-1}.

The two tilings of every row give the exact identity

A(z)h(z)=0over Z[z].(1)A(z)h(z)=0\quad\hbox{over }\mathbb Z[z]. \tag{1}

For any degree m>max⁡xpxm>\max_xp_x, use incidence shifts b↦b+gxvb\mapsto b+g_{xv} modulo mm. The coefficients of hh, assigned to sheets, form a nonzero ternary relation in the resulting cover. All used sheets are distinct within a type because Lv≤max⁡xpx<mL_v\leq\max_xp_x<m.

2. Full dissociation of the two parity classes

Theorem. Suppose max⁡vLv≥2\max_vL_v\geq2. For every odd prime PP with

P−1>nG,2P>max⁡xpx,(2)P-1>nG,\qquad 2P>\max_xp_x, \tag{2}

the degree-2P2P canonical cover is properly dissociated-colored by c(v,b)=b mod 2c(v,b)=b\bmod2. In particular its actual coloring number is exactly two.

Proof. Since A(1)=AA(1)=A has rank n−1n-1, while (1) supplies a kernel,

ker⁡Q(z)A(z)=Q(z)h(z).\ker_{\mathbb Q(z)}A(z)=\mathbb Q(z)h(z).

The vectors h(z)h(z) and h(−z)h(-z) are not proportional. To verify this,

SL(−z)SL(z)=1−z1+z{1,L even,(1+zL)/(1−zL),L odd.(3)\frac{S_L(-z)}{S_L(z)} =\frac{1-z}{1+z} \begin{cases} 1,&L\ \text{even},\\ (1+z^L)/(1-z^L),&L\ \text{odd}. \end{cases} \tag{3}

An even and odd length have unequal ratios. Two distinct odd lengths also have unequal ratios (compare their first nonconstant terms after canceling the common factor). Thus proportionality requires either all lengths even, or all the same odd length. Primitivity excludes the first case and reduces the second to Lv=1L_v=1 for all vv, contrary to the hypothesis.

Therefore the stacked matrix

M(z)=[A(z)A(−z)]M(z)=\begin{bmatrix}A(z)\\ A(-z)\end{bmatrix}

has full column rank over Q(z)\mathbb Q(z). Choose a nonzero nn-row minor D(z)D(z). Its polynomial degree is at most nGnG.

Suppose fv(b)∈{0,±1}f_v(b)\in\{0,\pm1\} is a relation supported on sheet parity ε\varepsilon, and put Fv(z)=∑b=02P−1fv(b)zbF_v(z)=\sum_{b=0}^{2P-1}f_v(b)z^b. At every 2P2P-th root of unity,

A(z)F(z)=0,F(−z)=(−1)εF(z).A(z)F(z)=0,\qquad F(-z)=(-1)^\varepsilon F(z).

Applying the first identity at −z-z also gives M(z)F(z)=0M(z)F(z)=0. Every 2P2P-th root other than 1,−11,-1 has order PP or 2P2P and degree P−1P-1 over Q\mathbb Q. For completeness, irreducibility of 1+z+⋯+zP−11+z+\cdots+z^{P-1} follows by substituting z+1z+1 and applying Eisenstein at PP; the order-2P2P case is obtained by z↦−zz\mapsto-z. Equation (2) thus ensures D(z)≠0D(z)\ne0 at all these roots. Consequently F(z)=0F(z)=0 there.

Write Fv(z)=zεQv(z2)F_v(z)=z^\varepsilon Q_v(z^2), where deg⁡Qv<P\deg Q_v<P. It vanishes whenever z2z^2 is a nontrivial PP-th root. Fourier inversion on Z/PZ\mathbb Z/P\mathbb Z shows that its coefficient sequence is constant:

fv(ε+2j)=λv∈{0,±1}(0≤j<P).(4)f_v(\varepsilon+2j)=\lambda_v\in\{0,\pm1\} \quad(0\leq j<P). \tag{4}

Sum the covering relation over all sheets. It gives Aλ=0A\lambda=0, so λ=qa\lambda=q a for rational qq. Since aa is primitive and λ\lambda integral, Bezout's identity gives q∈Zq\in\mathbb Z. The bounds ∣λv∣≤1|\lambda_v|\leq1 and max⁡∣av∣≥2\max|a_v|\geq2 force q=0q=0. Thus (4) vanishes and each parity class is fully dissociated. The nonzero canonical relation from Section 1 proves that one color does not suffice. □\square

3. Fixed voltages preserve type independence at large prime degree

Here is a separate useful stabilization fact. Fix any nonnegative integer translation voltages on any finite binary base, and a rationally independent set UU of its column types. Some ∣U∣|U|-row minor of A(z)UA(z)_U is nonzero at z=1z=1, and has degree at most ∣U∣G|U|G. For a prime cover degree QQ with Q−1>∣U∣GQ-1>|U|G, that minor is nonzero at every QQ-th root. Fourier decomposition of the cyclic cover then shows that all copies of the types in UU are rationally independent.

For a rational circuit, partition its types into any two nonempty proper sets and apply this fact to each. For its canonical voltages, when

Q−1>nG,Q>max⁡xpx,Q-1>nG,\qquad Q>\max_xp_x,

the degree-QQ cover has χ1=χQ=2\chi_1=\chi_{\mathbb Q}=2, including when the base circuit is already ternary. The upper bound is a type coloring; the canonical relation gives the lower bound. This is a theorem for fixed voltages and growing prime degree, not for all covers of those degrees.

4. Adaptive cuts also exist, but do not defeat these colorings

Multiplying (1) by (z−1)Y(z)(z-1)Y(z), for any binary sheet function yy, gives the ternary relation

fv(b)=sgn⁡(av)(y(b−Lv)−y(b)).(5)f_v(b)=\operatorname{sgn}(a_v) (y(b-L_v)-y(b)). \tag{5}

These are cuts of the circulant multigraph with a type-vv edge between b−Lvb-L_v and bb, labeled by the covered column (v,b)(v,b). This graph is connected because gcd⁡vLv=1\gcd_vL_v=1. A coloring avoids all nonzero relations (5) exactly when deleting the edges of each individual color leaves a connected graph: a constant-on-components indicator gives the reverse implication.

Thus the interval template is only one member of a genuinely adaptive family. Nevertheless Section 2 colors every ternary relation, including all these cuts, in the stated nonflat-circuit covers. A successful cover amplifier must go beyond independently chosen canonical one-circuit lifts, for example by coupling different relation spaces. No such arbitrary-color coupling theorem is established here.

The polynomial identities, Fourier argument, primitive coefficient step, and quantifier restrictions are author-recorded, checked against the supplied proof. No computation or new external paper theorem is used; the review is internal.