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

The Collision Spectrum

March 31, 202615 min read
Companion paper: The Collision Spectrum →
Blue and gold paths briefly accompany one another before opening into unequal continuations. Each collision coefficient combines a finite digit-boundary factor with an L-value, so both shape its strength.
Blue and gold paths briefly accompany one another before opening into unequal continuations. Each collision coefficient combines a finite digit-boundary factor with an L-value, so both shape its strength.

Each coefficient of the centered base-five collision table is a value L(1,χ)L(1,\chi)L(1,χ) of a Dirichlet LLL-function times a short sum read off the diagonal of equal digit pairs. The new fact here concerns those diagonal sums. In every odd prime base, the squares of those diagonal sums add up to exactly 2b(b−1)22b(b-1)^22b(b−1)2. Scaled by that total they become weights that sum to one. Then the table’s squared entries, which total 484848 at base five, become an exact weighted average of ∣L(1,χ)∣2|L(1,\chi)|^2∣L(1,χ)∣2, equal to 6π2/256\pi^2/256π2/25 at base five. The LLL-values and the identity linking them to Bernoulli numbers are classical, and the digits supply the weights.

This is the third of the three preprints going up on arXiv this month, arXiv 2604.00054, beside The Collision Invariant and The Collision Transform. It distills the base-five results of The Collision Spectrum and the L-Function Landscape, the factorization of each coefficient, the collision energy of 484848 and the fourth moment 192π4/625192\pi^4/625192π4/625, and proves the diagonal total in every base.

Your browser does not support embedded video.
From digit matches to the sizes of L-values. A video presentation of The Collision Spectrum, 7 min 37 sec.

Eight sums drawn as walks

Each of those LLL-values is an infinite sum, and a sum of complex numbers can be drawn. Put each term down as an arrow, tail at the tip of the one before, and the sum is wherever the arrows end up.

For L(1,χ)L(1,\chi)L(1,χ) the nnnth arrow has length 1/n1/n1/n, and the character χ\chiχ sets its direction. Take the character that turns eighteen degrees at each factor of two. Every nnn that five doesn’t divide is a power of two modulo 252525, say n≡2kn\equiv2^kn≡2k, and arrow nnn points 18k18k18k degrees around from straight right. Arrow 111 points straight right with length 111. Arrow 222 turns 181818 degrees with length 1/21/21/2. Since 3≡273\equiv2^73≡27 modulo 252525, arrow 333 turns 126126126 degrees with length 1/31/31/3, arrow 444 turns 363636 degrees with length 1/41/41/4, and arrow 555 is skipped. Past 252525 the directions repeat, and over every block of twenty-five they point all around the circle and cancel. The shrinking arrows therefore settle toward one point, about 1.379+0.713i1.379+0.713i1.379+0.713i. That point is L(1,χ)L(1,\chi)L(1,χ). The other seven characters turn by other multiples of eighteen degrees and draw seven more walks.

Eight reciprocal-step paths approach four conjugate pairs of endpoints in the complex plane. Three enlargements show increasingly late steps near the endpoint of path three. Eight reciprocal-step paths approach four conjugate pairs of endpoints in the complex plane. Three enlargements show increasingly late steps near the endpoint of path three.
Eight walks with steps of length 1/n, skipping multiples of five, each turning by its own character. Below, walk 3 through three enlargements, still circling its limit.

All eight walks take the same first step, one unit to the right, because every character points the number 111 straight ahead. That is why their lines all leave from the point at 111 on the axis. From the second arrow on they split up. The endpoints come in mirror pairs, four above the axis and four below, because two characters that turn by the same amount in opposite directions draw mirror-image walks. The open circles mark the limits computed independently, and after 62,50062{,}50062,500 arrows every walk is still circling its own. The lower panels follow walk 333 from arrow 252525, from arrow 250250250 and from arrow 2,5002{,}5002,500, each view ten times narrower than the last, and the path is still turning. An infinite sum has no last arrow, so the drawing only ever gets closer.

Eight matches at twenty-nine

The finite side starts with long division in base five, whose digits are 000 through 444. Divide each number from 111 through 282828 by 292929 and stack the repeating blocks, which are fourteen digits long, one under another.

A colored twenty-eight-by-fourteen field gives the complete repeating base-five blocks at denominator twenty-nine. Eight gold boxes mark equality in the first two columns. A colored twenty-eight-by-fourteen field gives the complete repeating base-five blocks at denominator twenty-nine. Eight gold boxes mark equality in the first two columns.
All twenty-eight fractions of 29 in base five, one full period per row. The eight boxed rows begin with two equal digits.

Read down the first two columns and box the rows whose two digits agree. There are eight, at 1/291/291/29, 7/297/297/29, 8/298/298/29, 14/2914/2914/29, 15/2915/2915/29, 21/2921/2921/29, 22/2922/2922/29 and 28/2928/2928/29, beginning 000000, 111111, 111111, 222222, 222222, 333333, 333333 and 444444. Read from the top and bottom at once and they pair up. The lone 000000 faces the lone 444444, the two 111111s face the two 333333s, and the two 222222s meet in the middle.

The first digits split the fractions into five bins of five or six, and eight matches against a smaller bin of five make a collision score of 333. That score depends only on the last two base-five digits of the denominator. Twenty-nine is 104104104 in base five, so it reads cell 040404, whose column averages −1/5-1/5−1/5, and the centered entry is

3−(−1/5)=16/5.3-(-1/5)=16/5.3−(−1/5)=16/5.

Doing the same for the other nineteen cells gives the centered table of The Collision Transform. Every column adds to zero, and every reflected pair of cells is opposite in sign.

The complete count and the two tables

The eight numerators whose first two base-five digits agree are 1, 7, 8, 14, 15, 21, 22 and 28. Their two-digit words are 00, 11, 11, 22, 22, 33, 33 and 44. The figure shows every fraction.

The raw lag-one collision table has twenty cells. The row supplies the first base-five digit of the address; the column supplies the last.

First digit Final 1 Final 2 Final 3 Final 4
0 0 1 0 3
1 0 -1 2 -1
2 0 1 -2 -1
3 0 -3 0 -1
4 -4 -1 -2 -1

The column means are −4/5,−3/5,−2/5,−1/5-4/5,-3/5,-2/5,-1/5−4/5,−3/5,−2/5,−1/5. Subtract them to get

First digit Final 1 Final 2 Final 3 Final 4
0 4/5 8/5 2/5 16/5
1 4/5 -2/5 12/5 -4/5
2 4/5 8/5 -8/5 -4/5
3 4/5 -12/5 2/5 -4/5
4 -16/5 -2/5 -8/5 -4/5

The entries belong to the residue classes modulo 25 coprime to five. The formula applies to every denominator above 25 in those classes, prime or not.

The same amount, differently arranged

Square each centered entry and add. Squaring stops the negative entries from cancelling the positive ones, so the total measures how much variation the table holds. The twenty squares add to exactly 484848, and that 484848 survives a change of coordinates. The table can be read address by address, or as a mixture of repeating patterns, the Dirichlet characters modulo 252525. Only eight patterns are needed, and the eight coefficients rebuild the whole table.

Twenty signed address sectors and eight character sectors form two different radial shapes. Each colored shape has exact area forty-eight, on the same geometric scale. Twenty signed address sectors and eight character sectors form two different radial shapes. Each colored shape has exact area forty-eight, on the same geometric scale.
The same table two ways, twenty addresses above and eight patterns below. Sector areas are the squared sizes, and each shape encloses 48.

The two shapes in the figure look nothing alike and enclose the same area, the dashed circle’s 484848. In the upper one each of the twenty addresses gets a sector whose area is its squared entry, so the long spokes at 444 and 212121 are the two entries of size 16/516/516/5. In the lower one the area belongs to the eight patterns. The pairs 333 and 171717 and 111 and 191919 take nearly all of it, and 777, 999, 111111 and 131313 get the small wedge at the bottom. That conservation of squared size is Parseval’s identity, and it lets the total be computed in whichever coordinates are easiest. Twenty small fractions are very easy coordinates.

Eight patterns and the factor twenty

The patterns are Dirichlet characters modulo 25. Write χj(2)=e2πij/20\chi_j(2)=e^{2\pi i j/20}χj​(2)=e2πij/20. A character respects multiplication, so this choice fixes its value at every eligible address. It is zero at multiples of five.

Each nonprincipal character has total weight zero over a period. Its partial sums are therefore bounded, and Dirichlet’s test proves convergence after division by nnn. The finite paths in the opening illustrate that theorem. They do not establish it by getting close to an endpoint.

Reflection removes the even characters. Family centering removes those inherited from modulo five. The eight primitive odd characters have indices

j=1,3,7,9,11,13,17,19.j=1,3,7,9,11,13,17,19.j=1,3,7,9,11,13,17,19.

With the average normalization

f^(χ)=120∑af(a)χ(a)‾,\widehat f(\chi)=\frac1{20}\sum_a f(a)\overline{\chi(a)},f​(χ)=201​a∑​f(a)χ(a)​,

Parseval reads

∑af(a)2=20∑χ∣f^(χ)∣2=48.\sum_a f(a)^2=20\sum_\chi|\widehat f(\chi)|^2=48.a∑​f(a)2=20χ∑​∣f​(χ)∣2=48.

The spectral shape therefore assigns each sector area 20∣f^(χ)∣220|\widehat f(\chi)|^220∣f​(χ)∣2. The original shape assigns area f(a)2f(a)^2f(a)2. Sector areas, rather than radii, represent the quantities. All sectors are shown on the same area scale.

The endpoints enter the count

The eight patterns are also the direction rules of the eight walks. At step nnn the character gives the direction and 1/n1/n1/n the length, so summing forever gives L(1,χ)L(1,\chi)L(1,χ). The connection runs deeper than a shared label. Each coefficient factors into the LLL-value, a factor fixed by the boundaries of the matching-digit intervals, and a phase that turns the product without changing its length.

In base five the boundary factor comes from four small integers, the right-hand ends 111, 777, 131313 and 191919 of the matching intervals among the twenty-five equal pieces of the unit interval. No large prime enters that calculation. The boundary factors give the eight squared distances unequal weights, some counting far more than others, and the total weight has an exact formula. Normalized to one, it turns the conserved 484848 into the weighted average of the squared distances,

∑χwχ∣L(1,χ)∣2=6π225.\sum_\chi w_\chi|L(1,\chi)|^2=\frac{6\pi^2}{25}.χ∑​wχ​∣L(1,χ)∣2=256π2​.

No walk had to be followed to its end. One exact statistic of all eight came straight from the finite table.

Four radial spectra grow from eight shares at base five to two thousand four hundred fifty at base seventy-one. Each has the same total area as its gold reference circle. Four radial spectra grow from eight shares at base five to two thousand four hundred fifty at base seventy-one. Each has the same total area as its gold reference circle.
The weights at bases 5, 13, 31 and 71, one spoke per character. Every shape and every gold circle has the same total area.

Every shape in the figure, and every gold circle, encloses the same area, one full unit of weight. At base five the eight shares make a fan with two broad wedges and two thin ones. At thirteen there are 727272 shares and the spokes start to vary, and at thirty-one there are 450450450. At seventy-one the 2,4502{,}4502,450 shares burst into a ragged star, its longest spokes reaching far past the circle while most of the rest sit inside it. The total never changes. Only its allocation does, and the shapes at larger bases are finite records with no limiting shape claimed.

The exact weights and weighted mean

For an odd prime base bbb, the diagonal boundary factor is

SG(χ)=2∑r=0b−2χ(r(b+1)+1)‾.S_G(\chi)=2\sum_{r=0}^{b-2}\overline{\chi(r(b+1)+1)}.SG​(χ)=2r=0∑b−2​χ(r(b+1)+1)​.

Summing over all primitive odd characters gives the exact total

∑χ∣SG(χ)∣2=2b(b−1)2.\sum_\chi|S_G(\chi)|^2=2b(b-1)^2.χ∑​∣SG​(χ)∣2=2b(b−1)2.

Thus wχ=∣SG(χ)∣2/[2b(b−1)2]w_\chi=|S_G(\chi)|^2/[2b(b-1)^2]wχ​=∣SG​(χ)∣2/[2b(b−1)2] sums to one. The coefficient factorization and Parseval give

∑χwχ∣L(1,χ)∣2=π2∑af(a)22b2(b−1).\sum_\chi w_\chi|L(1,\chi)|^2 =\frac{\pi^2\sum_a f(a)^2}{2b^2(b-1)}.χ∑​wχ​∣L(1,χ)∣2=2b2(b−1)π2∑a​f(a)2​.

At five, the total boundary mass is 160 and the table energy is 48. Substitution gives 6π2/256\pi^2/256π2/25.

Character index Share Squared distance
1 19.07164% 2.409338
3 23.13314% 2.922431
7 1.86686% 0.235843
9 5.92836% 0.748935
11 5.92836% 0.748935
13 1.86686% 0.235843
17 23.13314% 2.922431
19 19.07164% 2.409338

The total-mass proof expands a squared boundary sum into ordered pairs. Character orthogonality retains equal endpoints modulo b2b^2b2, rejects opposite endpoints at that modulus, and subtracts the inherited modulo-bbb contribution. The two lower-modulus counts cancel in total. At five, the endpoint sum has squared total 40+0−8+8=4040+0-8+8=4040+0−8+8=40. Doubling each endpoint sum gives the boundary-factor total 160. The Collision Spectrum has the full proof.

A second copy of the distance

Base five has one more exact relation. Its normalized weight is proportional to the squared distance itself,

wχ=2532π2∣L(1,χ)∣2,w_\chi=\frac{25}{32\pi^2}|L(1,\chi)|^2,wχ​=32π225​∣L(1,χ)∣2,

so a walk that ends twice as far from the start gets four times the weight. The weighted average then holds two copies of the same squared distance and becomes a fourth power,

∑χ∣L(1,χ)∣4=192π4625,\sum_\chi|L(1,\chi)|^4=\frac{192\pi^4}{625},χ∑​∣L(1,χ)∣4=625192π4​,

about 29.924129.924129.9241, though the fraction is the exact value. A finite polynomial identity proves that extra relation at base five. Nothing here assumes the two factors ought to resemble each other in other bases, and I don’t extend it on the strength of a picture.

Check the fourth-power identity

At base five the exact relation is

wχ=2532π2∣L(1,χ)∣2.w_\chi=\frac{25}{32\pi^2}|L(1,\chi)|^2.wχ​=32π225​∣L(1,χ)∣2.

Insert it into the weighted mean to get

2532π2∑χ∣L(1,χ)∣4=6π225.\frac{25}{32\pi^2}\sum_\chi|L(1,\chi)|^4=\frac{6\pi^2}{25}.32π225​χ∑​∣L(1,χ)∣4=256π2​.

For the finite certificate, set

P(x)=1+x+x2+x7,P(x)=1+x+x^2+x^7,P(x)=1+x+x2+x7,

Q(x)=1+2x+3x2+x3−2x4+x5+x6+2x7.\begin{aligned} Q(x)&=1+2x+3x^2+x^3-2x^4\\ &\qquad+x^5+x^6+2x^7. \end{aligned}Q(x)​=1+2x+3x2+x3−2x4+x5+x6+2x7.​

Write P∗(x)=P(x−1)P^*(x)=P(x^{-1})P∗(x)=P(x−1) and Q∗(x)=Q(x−1)Q^*(x)=Q(x^{-1})Q∗(x)=Q(x−1). At primitive twentieth roots, ∣Q(x)∣2=5∣P(x)∣2|Q(x)|^2=5|P(x)|^2∣Q(x)∣2=5∣P(x)∣2. The finite certificate is

x7(QQ∗−5PP∗)=Φ20(x)R(x),x^7(QQ^*-5PP^*)=\Phi_{20}(x)R(x),x7(QQ∗−5PP∗)=Φ20​(x)R(x),

where

Φ20(x)=x8−x6+x4−x2+1\Phi_{20}(x)=x^8-x^6+x^4-x^2+1Φ20​(x)=x8−x6+x4−x2+1

and

R(x)=−3+x2+5x3+x4−3x6.R(x)=-3+x^2+5x^3+x^4-3x^6.R(x)=−3+x2+5x3+x4−3x6.

The factor Φ20\Phi_{20}Φ20​ vanishes at those roots. The Collision Spectrum identifies the two polynomials with the Bernoulli and boundary factors. Their magnitude relation gives the displayed weight formula, character by character.

Keep following the patterns

The digit fields below show every numerator and one complete period at six denominators. They share the same five-color digit key. At larger denominators the individual cells become too small at reading width. Open the full-resolution plate to inspect them.

Six complete colored fraction fields show every row and a full period at denominators twenty-nine, fifty-nine, one hundred nine, one hundred forty-nine, two hundred sixty-nine and five hundred nine. Six complete colored fraction fields show every row and a full period at denominators twenty-nine, fifty-nine, one hundred nine, one hundred forty-nine, two hundred sixty-nine and five hundred nine.
Every numerator and one full period at six denominators, 157,454 digits on one color key.

The second atlas keeps every primitive odd character at eighteen prime bases. Sector area represents normalized weight, with the same total area and radial scale throughout. The unit circle marks equal weights.

Eighteen complete equal-area radial spectra show every primitive odd character at prime bases five through seventy-one, all on one radial scale. Eighteen complete equal-area radial spectra show every primitive odd character at prime bases five through seventy-one, all on one radial scale.
All 14,372 weights at eighteen prime bases, each sector’s area its share, against the gold circle of equal shares.

Three additional plots retain the individual numerical comparisons.

All eight normalized weights as bars, with the equal-share level marked. Character indices use the generator 2 modulo 25.
The eight base-five weights as bars, with the equal-share level marked.
The complete centered table in base five. The four column sums of squares are 64/5, 56/5, 56/5 and 64/5.
The centered base-five table, with column sums of squares 64/5, 56/5, 56/5 and 64/5.
The eight fourth powers individually. Their exact total is 192π⁴/625, approximately 29.9240727656. Labels are rounded.
The eight fourth powers, totaling 192π⁴/625, about 29.9241.

The exact total of the weights holds at every base, and their allocation is the part still to understand. At seventy-one the boundary factors and the LLL-values share 2,4502{,}4502,450 channels unevenly, and how they divide the weight as the base grows is open. At five the two factors are exactly proportional, which they are at no other base through seventy-one, and that proportionality is what makes the sum of the eight fourth powers exactly 192π4/625192\pi^4/625192π4/625.

Companion paper: The Collision Spectrum →
Share

Discussion

Sign in to join the discussion.

← All articlesRead the paper →
← Previous: The Collision Transform
Next: The Structure That Survives →