Petty's Notebook
ArticlesPapersnfieldAbout
Get notified when new posts are published. No spam, just math.
Alexander S. Petty  |  ©2009-2026
← Back
Response

The Resets That Meet

October 20, 202610 min read
Companion paper: The Sampling Defect and Reset Forests →
Several reset branches can end at the same surviving clock.
Several reset branches can end at the same surviving clock.

Revised November 15, 2026.

An integral can pass over one exceptional point without changing its value. A finite average has to read it. At the common reset of the collision field, the difference at that one point carries an entire unit in the limit.

The Common Reset in the Cubic Law accounts for the unit and gives a quarter-power bound on the error left over. The half-power target asks for a closer agreement between the growing collision table and its continuous energy. What in the arithmetic permits that improvement?

Whole families of resets already reach the target, even when every contribution is charged at its absolute size. Keep their signs and the bound improves again. The question now has a boundary. Inside these families the stronger comparison works. The complementary reading remains open.

The defect at five

At prime five, four ramps run around the unit interval. The first completes one circuit, the second two, the third three and the fourth four. Each rises from minus one toward plus one, then starts over. At a reset its assigned value is zero.

Add the four readings, then square their sum. Integrating that square gives 23/923/923/9. Reading it at the 25 equally spaced points of the carry table and averaging gives 48/2548/2548/25. The same field has two energy readings, separated by

239−4825=143225.\frac{23}{9}-\frac{48}{25}=\frac{143}{225}.923​−2548​=225143​.

All four clocks reset at zero. Their sum approaches minus four from the right and plus four from the left around the circle. Its square approaches sixteen from either side. The sample at zero reads zero.

Sixteen is fixed by those traces. Restore that one squared reading and leave the other twenty-four alone. The grid average rises by 16/2516/2516/25. The difference becomes

239−(4825+1625)=−1225.\frac{23}{9}-\left(\frac{48}{25}+\frac{16}{25}\right)=-\frac1{225}.923​−(2548​+2516​)=−2251​.

One changed reading takes the discrepancy from 143/225143/225143/225 to −1/225-1/225−1/225. The integral stays put. The corrected table reads slightly high.

The squared four-clock field and its 25 sampled values. The origin has limiting height 16 and sampled value zero. Continuous energy 23/9 minus grid average 48/25 is 143/225. Adding the origin correction 16/25 to the grid average leaves minus 1/225. The squared four-clock field and its 25 sampled values. The origin has limiting height 16 and sampled value zero. Continuous energy 23/9 minus grid average 48/25 is 143/225. Adding the origin correction 16/25 to the grid average leaves minus 1/225.
Four clocks and 25 samples. The curve shows the squared field between resets; filled points show grid readings and hollow points show one-sided reset values. The gold origin sample is zero where the squared field approaches sixteen. Restoring it leaves the signed difference −1/225.

At a general prime ppp, there are p−1p-1p−1 clocks and p2p^2p2 samples. The common reset contributes

(p−1)2p2.\frac{(p-1)^2}{p^2}.p2(p−1)2​.

This approaches one. The squared reset height grows at the same scale as the number of samples, so refinement keeps the missed contribution alive.

At the other resets, some clocks start over while others keep reading. All their values combine before the square is taken. The boundary formula in Collision Calculus retains these interactions, their signs and the curvature between resets. To improve the comparison, we need to see which of these boundaries belong together.

Divisor forests

At prime thirty-one, the field has thirty clocks. At one third, clocks three, six, nine and every other multiple of three reset together. At one fifth, the multiples of five reset. At two fifths, those same clocks reset again.

Reduce an address to lowest terms. Its denominator names the clocks that meet there. At one third, ten clocks reset and twenty keep reading. All thirty contribute to the field at that address.

Now multiply an address repeatedly by thirty and reduce it after each step. Powers of two, three and five disappear from its denominator. Twelve eventually reaches one. Fourteen leaves seven. Twenty-one also leaves seven.

Fourteen and twenty-one have the same surviving clock. Twenty-eight feeds through fourteen to seven. Separate denominators enter the same family through different branches.

The sampling grid moves the addresses too. Comparing its motion with the source’s motion gives return conditions on the surviving denominators. One selection at thirty-one includes every denominator from two through thirty except seventeen, twenty-three and twenty-nine.

Reduced reset addresses at prime 31. Gold marks show 26 selected denominator families. Three violet rows at denominators 17, 23 and 29 remain outside. Every mark reads the complete thirty-clock field. Reduced reset addresses at prime 31. Gold marks show 26 selected denominator families. Three violet rows at denominators 17, 23 and 29 remain outside. Every mark reads the complete thirty-clock field.
The return-selected forest at prime thirty-one contains 26 denominator families. The highlighted rows at seventeen, twenty-three and twenty-nine remain outside. Each mark in either group reads all thirty clocks. This small complement is specific to the finite example.

Those surviving clocks divide an explicit integer built from the source size and grid size. That integer may grow with the prime, provided it stays below a fixed power. The estimate then includes every reduced address in every selected denominator family, with all the incoming branches above its surviving clock.

At thirty-one, twenty-six of the twenty-nine nonorigin denominator families are covered together. Only three remain outside. That small complement does not persist as the cutoff grows. Under the stated height condition the forest population is sparse, and the complement is large.

Cancellation within a family

Read the selected forest at thirty-one. Keep the full thirty-clock field at every address, together with each reset’s multiplicity and sampling weight.

The positive contributions add to about 0.150180.150180.15018. The negative contributions add to about −0.13008-0.13008−0.13008. Charging every contribution at its absolute size gives about 0.280260.280260.28026.

Keep the signs and only about 0.020100.020100.02010 remains.

At prime 31, positive and negative address contributions are grouped by selected denominator. Gold points show each signed total. The absolute sum is about 0.28026 and the signed forest contribution is about 0.02010. At prime 31, positive and negative address contributions are grouped by selected denominator. Gold points show each signed total. The absolute sum is about 0.28026 and the signed forest contribution is about 0.02010.
Positive and negative contributions at the 211 addresses of this selected forest. Each gold point gives a denominator’s signed total. Taking absolute sizes gives about 0.28026; keeping signs leaves about 0.02010. Every contribution retains the full thirty-clock field and its sampling weight. These are finite values at prime thirty-one.

The theorem controls these sums as the source and grid grow. After sampling normalization, the absolute forest sum has a half-power bound. Its signed sum has a first-power bound. Restoring the forest’s reading to complete clocks has the first-power bound too.

For comparison, a sixteenfold increase in ppp divides p−1/4p^{-1/4}p−1/4 by two, p−1/2p^{-1/2}p−1/2 by four and p−1p^{-1}p−1 by sixteen. The estimates include constants and arbitrarily small exponent losses. They bound the errors without requiring each successive value to be smaller.

The target holds throughout these selected families before cancellation is even used. Their signs leave a stronger allowance. The remaining half-power question lies in the complementary reset reading, still taken from the same complete field on the same grid.

The forest bound and the complementary reading

For an odd prime ppp, put N=p−1N=p-1N=p−1 and Q=p2Q=p^2Q=p2. The midpoint sawtooth is ψ(x)={x}−1/2\psi(x)=\{x\}-1/2ψ(x)={x}−1/2 away from integers and ψ(x)=0\psi(x)=0ψ(x)=0 at integers. The field and its sampling defect are

BN(x)=2∑h=1Nψ(hx),Sp=∫01BN(x)2 dx−1Q∑a mod QBN(a/Q)2.B_N(x)=2\sum_{h=1}^N\psi(hx),\qquad \mathcal S_p=\int_0^1B_N(x)^2\,dx-\frac1Q\sum_{a\bmod Q}B_N(a/Q)^2.BN​(x)=2∑h=1N​ψ(hx),Sp​=∫01​BN​(x)2dx−Q1​∑amodQ​BN​(a/Q)2.

Let rough⁡N(k)\operatorname{rough}_N(k)roughN​(k) be the largest divisor of kkk coprime to NNN. For a positive integer DDD, the selected forest is

F(N,D)={2≤k≤N:rough⁡N(k)∣D}.\mathcal F(N,D)=\{2\leq k\leq N:\operatorname{rough}_N(k)\mid D\}.F(N,D)={2≤k≤N:roughN​(k)∣D}.

For any set of reset denominators A\mathcal AA, its signed boundary reading is

AA=∑k∈A⌊Nk⌋∑1≤a<k(a,k)=1BN(a/k)ψ(Qa/k).\mathcal A_{\mathcal A}=\sum_{k\in\mathcal A}\left\lfloor\frac Nk\right\rfloor\sum_{\substack{1\leq a<k\\(a,k)=1}}B_N(a/k)\psi(Qa/k).AA​=∑k∈A​⌊kN​⌋∑1≤a<k(a,k)=1​​BN​(a/k)ψ(Qa/k).

If D≤CpKD\leq Cp^KD≤CpK for fixed C,KC,KC,K, the absolute boundary sum over the forest is Oε(p3/2+ε)O_\varepsilon(p^{3/2+\varepsilon})Oε​(p3/2+ε), and its signed sum is Oε(p1+ε)O_\varepsilon(p^{1+\varepsilon})Oε​(p1+ε). Multiplication by 4/p24/p^24/p2 gives the normalized half-power and first-power bounds.

The figure uses the return window L=3L=3L=3 and

D=15(NQ+1)(N2−N−1)(Q−N)(Q3−N3).D=15(NQ+1)(N^2-N-1)(Q-N)(Q^3-N^3).D=15(NQ+1)(N2−N−1)(Q−N)(Q3−N3).

At p=31p=31p=31, its omitted denominators are exactly 17,23,2917,23,2917,23,29.

Write Aother\mathcal A_{\rm other}Aother​ for the same signed reading over the complementary denominators. The exact decomposition and proved estimates give

Sp−1=4p2Aother+Oε(p−1+ε).\mathcal S_p-1=\frac4{p^2}\mathcal A_{\rm other}+O_\varepsilon(p^{-1+\varepsilon}).Sp​−1=p24​Aother​+Oε​(p−1+ε).

The full half-power target is therefore equivalent to Aother=Oε(p3/2+ε)\mathcal A_{\rm other}=O_\varepsilon(p^{3/2+\varepsilon})Aother​=Oε​(p3/2+ε). This bound remains open. The established global exponent remains 1/41/41/4.

The selected source

A reset belongs to several clocks at once. Keeping a family of resets therefore raises a source question. Which coefficients produce the selected totals?

Give six clocks coefficient one. Their totals at reset denominators one through six are

(6,3,2,1,1,1).(6,3,2,1,1,1).(6,3,2,1,1,1).

The three at denominator two comes from clocks two, four and six. Keep only denominator families dividing six. The record becomes (6,3,2,0,0,1)(6,3,2,0,0,1)(6,3,2,0,0,1).

Work backward. The unique source with that record has coefficients

(2,2,1,0,0,1).(2,2,1,0,0,1).(2,2,1,0,0,1).

Clocks one and two must double. Deleting clocks four and five alone would leave the wrong totals.

Four rows show original coefficients 1 1 1 1 1 1, reset totals 6 3 2 1 1 1, selected totals 6 3 2 0 0 1, and recovered coefficients 2 2 1 0 0 1. The first two coefficients must double. Four rows show original coefficients 1 1 1 1 1 1, reset totals 6 3 2 1 1 1, selected totals 6 3 2 0 0 1, and recovered coefficients 2 2 1 0 0 1. The first two coefficients must double.
Read each row by its index from one through six. Selecting denominator families changes the source required to produce their totals. Divisor inversion doubles the first two coefficients and retains the selected reset record exactly.

Divisor inversion performs this reconstruction. The selected boundaries have a source of their own, and it still interacts with the rest of the field. Its mixed work remains in the calculation when the reading is restored to the original grid.

These interactions carry most of the leading energy. The Cubic Law assigns one third to the harmonic diagonal and two thirds to interactions between different clocks. Removing those terms removes two thirds of the leading budget. The continuous energy also has its own secondary terms. The sampling defect compares the two complete readings.

The packet fringe

The complete cutoff interior admits a second grouping into packets of two or three clocks. Its unfinished packets form a fringe, and the entire fringe fits within the half-power allowance.

At thirty-one, the complete packet is {23,22,19}\{23,22,19\}{23,22,19} and its fringe is {24,20}\{24,20\}{24,20}. The forest’s complement is {17,23,29}\{17,23,29\}{17,23,29}. These are different selections of the same arithmetic. The packet estimate controls its unfinished edges; the forest estimate controls its selected reset families.

The grouping uses the source word, the short ramp sum left after complete periods and reflection. The upper word advances by two here, the lower by one. An upper word opens a cell, and the lower words in that cell complete its packet.

The five interior clocks at prime 31, placed by source-word length. Upper clock 19 and lower clocks 23 and 22 form the complete packet at words 7 and 8. Lower clock 24 at word 6 and upper clock 20 at word 9 form the cutoff fringe. The five interior clocks at prime 31, placed by source-word length. Upper clock 19 and lower clocks 23 and 22 form the complete packet at words 7 and 8. Lower clock 24 at word 6 and upper clock 20 at word 9 form the cutoff fringe.
The complete cell contains upper clock 19 and lower clocks 23 and 22. Lower clock 24 and upper clock 20 remain at the unfinished edges. The horizontal coordinate is the source-word length; the labels beside the dots are clock denominators. All five interior clocks appear once. This packet split differs from the forest selection.

The proofs are developed in The Sampling Defect and Reset Forests. They write out the remaining signed readings and give a sufficient bound on the cost of observing them.

At five, restoring one sample leaves −1/225-1/225−1/225. At a growing cutoff, the corresponding remainder gathers readings from more and more reset addresses. The forest bound controls complete families of those addresses. The packet bound pays for the unfinished edges of a different grouping.

Beyond those estimates, the remaining clocks still read the same field. Their return relations are fixed by the arithmetic. How much cancellation do those relations force?

Companion paper: The Sampling Defect and Reset Forests →
Share

Discussion

Sign in to join the discussion.

← All articlesRead the paper →
← Previous: Collision Calculus