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

The Common Reset in the Cubic Law

October 10, 20268 min read
Companion paper: The Unit Sampling Defect in Digit-Collision Energy →
Rings of clocks, each broken at the same narrow seam. Their gold endpoints line up along it, the one place where every clock starts over together.
Rings of clocks, each broken at the same narrow seam. Their gold endpoints line up along it, the one place where every clock starts over together.

Revised November 2, 2026.

A finite carry table and a continuous sawtooth field describe the same readings. Their mean-square energies nevertheless approach a difference of one as the table grows. The difference comes from a single entry at which every sawtooth resets and the table reads zero.

The earlier result established that this unit survives. The revised argument gives a much stronger bound on everything left over. After correcting the common reset, the finite average approaches the continuous energy at a quarter-power rate, up to an arbitrarily small loss in the exponent. That measures how accurately the original arithmetic table reads the continuous field behind it.

Counting the carries

A digit collision means that the selected digit positions agree. In a two-digit base-three word, the possibilities are 00, 11 and 22, occupying positions zero, four and eight on a circle of nine places. Multiplication by two moves those marks to zero, eight and seven. A further step of two takes the last two marks across the join, including the mark that lands exactly on zero. The carry count is two.

A step of two crosses the join from two of the nine possible starting positions. For three selected marks, the proportional allowance is therefore two thirds. Subtracting that allowance leaves a centered reading of four thirds. Doing the same for every multiplier fills the carry table. Squaring its centered readings and adding gives the collision energy.

A nine-place wheel shows positions zero, eight and seven after multiplication by two. Two gold paths cross the join. Nine reordered signed clock readings lead to the exact averages one and sixteen twenty-sevenths. A nine-place wheel shows positions zero, eight and seven after multiplication by two. Two gold paths cross the join. Nine reordered signed clock readings lead to the exact averages one and sixteen twenty-sevenths.
Base three. A step of two carries two marks past zero, against an allowance of two thirds. The nine readings below, squared, average 16/27, and the curve averages 1.

The table has an exact harmonic description. Start with a repeating ramp that rises linearly, drops at each integer and takes the value zero at the drop. Center the ramp around zero; this is the sawtooth function. I call each dilated copy a clock. In base ppp, there are p−1p-1p−1 clocks, completing one, two, up to p−1p-1p−1 cycles per unit interval. Double their readings and add them before squaring.

Reading that field at the p2p^2p2 equally spaced grid points gives precisely the centered carry table in a different order. The identity includes the resets and holds in composite bases too. The rate theorem uses odd prime bases.

At base three, the nine squared readings have average 16/2716/2716/27. The continuous squared field has average 111. Their difference is 11/2711/2711/27. For larger prime bases, that difference can lie above or below one; the theorem controls its approach to one.

The sample at zero

At base five, all four clocks reset at zero. Just to the right, each doubled clock approaches minus one. Just to the left around the periodic join, each approaches plus one. Their sum approaches minus four on one side and plus four on the other, so its square approaches sixteen from both sides.

At zero itself, the assigned readings are all zero. The table samples that point exactly.

At base five, four rising and resetting clocks have gold zeros at the origin. The lower graph has one-sided squared height sixteen and an actual zero sample, with nearby sample values on each side. At base five, four rising and resetting clocks have gold zeros at the origin. The lower graph has one-sided squared height sixteen and an actual zero sample, with nearby sample values on each side.
Base five. All four doubled clocks reset together at zero. Their squared sum climbs toward 16 from both sides, but the sample at zero reads 0, which costs the average 16/25.

An integral is unchanged if we alter the value at one point. The grid average gives that point a share of 1/251/251/25. Replacing its zero squared reading by sixteen therefore raises the average by 16/2516/2516/25.

In base ppp, the same correction is

(p−1)2p2⟶1.\frac{(p-1)^2}{p^2}\longrightarrow1.p2(p−1)2​⟶1.

The height beside the reset grows with the square of the number of clocks, while the grid’s weight on one point shrinks as the inverse square of the base. Their product survives. This is why refining the grid does not remove the discrepancy. The field being sampled is growing at the same time.

Cancellation between the resets

Identifying the common reset leaves the main difficulty. The interior contains many other resets, and their number grows with the base. At each one, some clocks restart while others continue. Their combined contribution to the sampling error must be controlled with its sign.

At base five, replace the origin’s squared reading by sixteen and leave the other twenty-four samples alone. Draw each reading as a rectangle of width 1/251/251/25. Some rectangles overshoot the squared field; others fall short. Their signed differences add to

239−48+1625=−1225.\frac{23}{9}-\frac{48+16}{25}=-\frac1{225}.923​−2548+16​=−2251​.

The base-five squared response is compared with twenty-five sample rectangles after replacing the origin sample by sixteen. Teal and violet regions mark positive and negative differences. Twenty-five signed bars sum to minus one over 225. The base-five squared response is compared with twenty-five sample rectangles after replacing the origin sample by sixteen. Teal and violet regions mark positive and negative differences. Twenty-five signed bars sum to minus one over 225.
Base five, with the zero at the origin replaced by 16. Violet marks where a sample rectangle overshoots the curve, teal where it falls short. The 25 signed areas add to −1/225.

The cancellation visible in this small example has to be proved uniformly as the whole table grows. An exact sampling identity separates the common reset, the interior interaction and a smaller curvature term. It also applies to weighted mixtures and grids passing through additional resets, recording their contributions explicitly.

The interior interaction can be written in classical Dedekind sums. Their Fourier frequencies sometimes repeat patterns from smaller grids. Grouping copies of the same pattern cancels complete cycles before any estimate is applied. Two geometric-sum estimates control the large-frequency tails.

For the remaining frequencies, reciprocity and Fourier completion turn the arithmetic phases into a family of linear twists. These twists have equally spaced slopes. The argument counts those whose slope approaches an integer, where cancellation can be weak. The others receive a stronger bound. This distinction allows all inverse frequencies to be included without charging each one the largest possible cost.

Balancing the resulting estimates gives the quarter-power rate. The carry table supplies the exact finite observation and its endpoints; the Dedekind and exponential-sum tools control the error of that observation.

The stronger rate

Write EpE_pEp​ for the finite carry energy and J(p−1)\mathcal J(p-1)J(p−1) for the continuous mean-square energy. Their difference after normalization is the sampling defect,

Sp=J(p−1)−Epp2.\mathcal S_p=\mathcal J(p-1)-\frac{E_p}{p^2}.Sp​=J(p−1)−p2Ep​​.

The revised theorem states that, for every ε>0\varepsilon>0ε>0, through odd primes,

Sp=1+Oε ⁣(p−1/4+ε).\mathcal S_p=1+O_\varepsilon\!\left(p^{-1/4+\varepsilon}\right).Sp​=1+Oε​(p−1/4+ε).

The earlier version gave the much smaller exponent 1/4586961/4586961/458696, with a logarithmic factor. The present proof improves the error estimate for the complete table. The bound is asymptotic, with a constant depending on ε\varepsilonε; it does not supply a numerical error certificate at a specified base.

Equivalently, correcting the common reset gives

Ep+(p−1)2p2=J(p−1)+Oε ⁣(p−1/4+ε).\begin{aligned} \frac{E_p+(p-1)^2}{p^2}&=\mathcal J(p-1)\\ &\quad+O_\varepsilon\!\left(p^{-1/4+\varepsilon}\right). \end{aligned}p2Ep​+(p−1)2​​=J(p−1)+Oε​(p−1/4+ε).​

The correction is prescribed by the carry construction. It is not fitted to the observed energies. The theorem controls every odd prime base, including the exceptional alignments that a finite panel of computations might miss.

The Cubic Law and The Secondary Term describe the leading structure of this energy. The sampling theorem keeps the continuous energy exact and measures how faithfully the finite table reads it. Further terms in the continuous energy and sharper sampling rates are distinct questions about the same construction.

The square-root conjecture asks for the stronger error Oε(p−1/2+ε)O_\varepsilon(p^{-1/2+\varepsilon})Oε​(p−1/2+ε). The present estimates reach their best balance at 1/41/41/4, so reaching 1/21/21/2 requires further cancellation in the signed interior sum. That question remains open.

What is established is the boundary correction and a uniform rate for the error after it is removed. The exact bridge identifies what the table reads. The cancellation theorem determines how accurately its average recovers the continuous energy.

The complete argument is in The Unit Sampling Defect in Digit-Collision Energy, with earlier versions preserved under the paper’s DOI.

Companion paper: The Unit Sampling Defect in Digit-Collision Energy →
Share

Discussion

Sign in to join the discussion.

← All articlesRead the paper →
← Previous: Unity, Refinement, and Signed Capture
Next: Collision Calculus →