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

The Collision Spectrum and the L-Function Landscape

March 10, 202517 min read
Companion paper: The Collision Spectrum and the L-Function Landscape →
A small blue-and-gold fold opens into a much larger, quiet sweep. At base five, a finite collision table determines an exact fourth moment of Dirichlet L-values.
A small blue-and-gold fold opens into a much larger, quiet sweep. At base five, a finite collision table determines an exact fourth moment of Dirichlet L-values.

Averages of ∣L(1,χ)∣|L(1,\chi)|∣L(1,χ)∣ over families of Dirichlet characters are classical objects in analytic number theory, closely tied to class numbers. At the single modulus twenty-five, one fourth-power average of that kind can be computed exactly from long division, a fixed-modulus identity that rests on a cyclotomic relation special to base five. Add the fourth powers of ∣L(1,χ)∣|L(1,\chi)|∣L(1,χ)∣ over the eight primitive odd characters modulo 252525, eight values of infinite series, and they total exactly 192π4/625192\pi^4/625192π4/625. The number underneath it is 484848, the sum of the squares of twenty fractions in the centered base-five collision table.

The comparison in The Spectral Repulsion set each collision coefficient against a sum over primes. Inside a single coefficient there is a cleaner structure. It factors exactly into two finite sums. One is a generalized Bernoulli number, which is L(1,χ)L(1,\chi)L(1,χ) up to a known constant, and the other is a four-term character sum from the diagonal of equal digit pairs. At base five the two factors are locked in a fixed ratio of 5\sqrt55​. Every coefficient is then a fixed multiple of ∣L(1,χ)∣2|L(1,\chi)|^2∣L(1,χ)∣2, and squaring the table turns 484848 into the fourth moment. At larger bases the lock loosens. The correlation between the two factors falls from 0.8440.8440.844 at seven to 0.6750.6750.675 at seventy-one and stays positive across all eighteen bases checked.

Twenty fractions

The table comes from The Centered Collision Sum. For each denominator, count the fractions whose first two base-five digits agree, take away the smaller bin’s size, and then subtract the average for the denominator’s final-digit family. For primes past 252525 the result is fixed by the remainder modulo 252525, so twenty entries cover them all.

Twenty signed fractions occupy four columns and five rows. Every column sums to zero. Below, five squares on a common scale show the five entry magnitudes, with their multiplicities. The sum of twenty squared entries is 48. Twenty signed fractions occupy four columns and five rows. Every column sums to zero. Below, five squares on a common scale show the five entry magnitudes, with their multiplicities. The sum of twenty squared entries is 48.
The twenty centered base-five entries by remainder modulo five, each column summing to zero. Below, the five sizes after squaring, with their counts, adding to 48.

The four columns are the four remainders modulo five, and each one sums to zero, teal against violet. Squaring throws the signs away, so the cancellation stops helping. The first column has four copies of 4/54/54/5 and one −16/5-16/5−16/5 at 212121, and that single violet cell supplies most of the column’s 64/564/564/5. The second and third columns give 56/556/556/5 apiece and the fourth another 64/564/564/5. The row of squares at the bottom sorts the twenty squared entries by size, from four small squares at 2/52/52/5 to the two large ones at 16/516/516/5, and together they add to

E5=64+56+56+645=48.E_5=\frac{64+56+56+64}{5}=48.E5​=564+56+56+64​=48.

The twenty entries, ready to add
Quotient Remainder 1 Remainder 2 Remainder 3 Remainder 4
0 4/54/54/5 (at 1) 8/58/58/5 (at 2) 2/52/52/5 (at 3) 16/516/516/5 (at 4)
1 4/54/54/5 (at 6) −2/5−2/5−2/5 (at 7) 12/512/512/5 (at 8) −4/5−4/5−4/5 (at 9)
2 4/54/54/5 (at 11) 8/58/58/5 (at 12) −8/5−8/5−8/5 (at 13) −4/5−4/5−4/5 (at 14)
3 4/54/54/5 (at 16) −12/5−12/5−12/5 (at 17) 2/52/52/5 (at 18) −4/5−4/5−4/5 (at 19)
4 −16/5−16/5−16/5 (at 21) −2/5−2/5−2/5 (at 22) −8/5−8/5−8/5 (at 23) −4/5−4/5−4/5 (at 24)

The first column contains four copies of 4/54/54/5 and one −16/5-16/5−16/5. Its sum of squares is

4(45)2+(165)2=645.4\left(\frac45\right)^2+\left(\frac{16}5\right)^2=\frac{64}5.4(54​)2+(516​)2=564​.

The second gives

2(85)2+2(25)2+(125)2=565.2\left(\frac85\right)^2+2\left(\frac25\right)^2+\left(\frac{12}5\right)^2=\frac{56}5.2(58​)2+2(52​)2+(512​)2=556​.

Reflection supplies the other two. The entry at 25−a25-a25−a is the negative of the entry at aaa, so their squares agree.

Eight patterns out of twenty

A Dirichlet character modulo 252525 gives each allowed residue a complex number of size one, and it respects multiplication. Since 222 generates all twenty allowed residues, the character’s value at 222 settles everything, and there are twenty choices, e2πij/20e^{2\pi ij/20}e2πij/20. Comparing a character with the table gives its Fourier coefficient,

cχ=120∑a∈U25f(a)χ(a)‾,c_\chi=\frac1{20}\sum_{a\in U_{25}}f(a)\overline{\chi(a)},cχ​=201​a∈U25​∑​f(a)χ(a)​,

where the bar reverses the phase and U25U_{25}U25​ is the set of residues not divisible by five.

Twenty colored character patterns are aligned with their coefficient magnitudes. Eight highlighted rows have nonzero bars. A complex-plane plot beneath shows all eight coefficients and the twelve vanishing coefficients at the origin. Twenty colored character patterns are aligned with their coefficient magnitudes. Eight highlighted rows have nonzero bars. A complex-plane plot beneath shows all eight coefficients and the twelve vanishing coefficients at the origin.
The twenty characters modulo 25, one per row, colored by phase. Twelve rows are faded, and the eight bright ones carry the coefficients shown in the bars and the dial.

Each row of the figure is one character, its colors cycling through the phases as the powers of two go by. Twelve rows are faded, and the bars beside them are empty. Ten of those characters give aaa and 25−a25-a25−a equal values. The table gives those two residues opposite values, so the coefficients cancel. Two more are constant down each column of the table, and every column sums to zero. The eight bright rows are the primitive odd characters. Odd means reflection reverses their sign, and primitive means a prime’s remainder modulo five isn’t enough to know their value. Their bars are far from equal. Characters 333 and 171717 carry the longest, 777 and 131313 the shortest, and in the dial underneath each conjugate pair mirrors across the horizontal axis.

Twenty terms and four

Each of those eight coefficients factors into two finite sums. The first runs over all twenty residues,

Bχ=125∑a∈U25aχ(a)‾,B_\chi=\frac1{25}\sum_{a\in U_{25}}a\overline{\chi(a)},Bχ​=251​a∈U25​∑​aχ(a)​,

and it is the generalized Bernoulli number B1,χˉB_{1,\bar\chi}B1,χˉ​​. The name is classical, and the definition on the page is an ordinary finite sum. The second comes from the digit boundaries. In base five the two-digit words with equal digits are 000000, 111111, 222222, 333333 and 444444, which are the integers 000, 666, 121212, 181818 and 242424. Adding up the change in the character across each of those five intervals gives a sum DχD_\chiDχ​, and reflection folds it in half,

∣Dχ∣=2∣χ(1)‾+χ(2)‾+χ(3)‾+χ(4)‾∣.|D_\chi|=2\left|\overline{\chi(1)}+\overline{\chi(2)}+ \overline{\chi(3)}+\overline{\chi(4)}\right|.∣Dχ​∣=2​χ(1)​+χ(2)​+χ(3)​+χ(4)​​.

A five-by-five grid highlights the five equal-digit cells. Beneath it, a path of twenty weighted complex arrows gives the Bernoulli sum; a second path of four unit arrows gives the short character sum. Gold segments join each starting point to its sum. A five-by-five grid highlights the five equal-digit cells. Beneath it, a path of twenty weighted complex arrows gives the Bernoulli sum; a second path of four unit arrows gives the short character sum. Gold segments join each starting point to its sum.
The diagonal of equal digit pairs in base five, then the twenty-arrow Bernoulli sum and the four-arrow short sum for the character with χ(2) = e^{2πi/20}.

In the figure the gold cells mark the diagonal, the five places with matching digits. Beneath it, for the character with χ(2)=e2πi/20\chi(2)=e^{2\pi i/20}χ(2)=e2πi/20, the left path adds twenty weighted arrows that wander out and back before landing at ∣Bχ∣≈2.4704|B_\chi|\approx2.4704∣Bχ​∣≈2.4704. The right path adds only four unit arrows, the character at 111, 222, 333 and 444, and reaches 2.76202.76202.7620. The coefficient multiplies the two,

cχ=−BχDχ‾20,c_\chi=-\frac{B_\chi\overline{D_\chi}}{20},cχ​=−20Bχ​Dχ​​​,

phase included. The proof turns each floor difference in the collision count into the same Bernoulli factor, after which the endpoint and bin-size terms cancel and only the diagonal factor is left. The Collision Spectrum and the L-Function Landscape carries out that accounting for every odd prime base, with 202020 replaced by b(b−1)b(b-1)b(b−1).

The five boundary intervals and the exact product

Extend the character by zero at multiples of five. With G={0,6,12,18,24}G=\{0,6,12,18,24\}G={0,6,12,18,24},

Dχ=∑n∈G[χ(n+1)‾−χ(n)‾].D_\chi=\sum_{n\in G}\bigl[\overline{\chi(n+1)}-\overline{\chi(n)}\bigr].Dχ​=n∈G∑​[χ(n+1)​−χ(n)​].

The left endpoints are 0,6,12,18,240,6,12,18,240,6,12,18,24. Reflection modulo 252525 sends the right endpoints 1,7,13,19,251,7,13,19,251,7,13,19,25 to 24,18,12,6,024,18,12,6,024,18,12,6,0. Oddness reverses their character values. Thus

Dχ=−2χ(6)‾∑k=14χ(k)‾.D_\chi=-2\overline{\chi(6)}\sum_{k=1}^4\overline{\chi(k)}.Dχ​=−2χ(6)​k=1∑4​χ(k)​.

The factor at 666 rotates the short sum without changing its length. For the character with χ(2)=e2πi/20\chi(2)=e^{2\pi i/20}χ(2)=e2πi/20, the two factors and their product are approximately

Bχ=−1.525731+1.942954i,Dχ=5.520147−0.206396i,cχ=0.441164−0.520524i.\begin{aligned} B_\chi&=-1.525731+1.942954i,\\ D_\chi&=5.520147-0.206396i,\\ c_\chi&=0.441164-0.520524i. \end{aligned}Bχ​Dχ​cχ​​=−1.525731+1.942954i,=5.520147−0.206396i,=0.441164−0.520524i.​

The product formula gives the same coefficient as averaging all twenty weighted table entries directly.

The infinite series enters

For a nontrivial character the values cancel enough for

L(1,χ)=χ(1)+χ(2)2+χ(3)3+⋯L(1,\chi)=\chi(1)+\frac{\chi(2)}2+\frac{\chi(3)}3+\cdotsL(1,χ)=χ(1)+2χ(2)​+3χ(3)​+⋯

to converge, and the classical functional equation ties that value to the finite Bernoulli sum. For primitive odd characters modulo b2b^2b2,

∣Bχ∣=bπ∣L(1,χ)∣.|B_\chi|=\frac b\pi|L(1,\chi)|.∣Bχ​∣=πb​∣L(1,χ)∣.

So one factor in every collision coefficient carries an LLL-value, and the other is still the four-term sum from the digit diagonal. At most bases both have to be carried along.

Five has an extra relation. For each of its eight characters

∣Dχ∣=5 ∣Bχ∣,|D_\chi|=\sqrt5\,|B_\chi|,∣Dχ​∣=5​∣Bχ​∣,

and the arrows above bear it out, since 2×2.7620=5.52402\times2.7620=5.52402×2.7620=5.5240 is 5\sqrt55​ times 2.47042.47042.4704. With the two factors locked together, both carry the same LLL-value, and their product is a square,

∣cχ∣=554π2∣L(1,χ)∣2.|c_\chi|=\frac{5\sqrt5}{4\pi^2}|L(1,\chi)|^2.∣cχ​∣=4π255​​∣L(1,χ)∣2.

Four squares have side lengths equal to the four distinct L-value magnitudes at modulus 25. Their areas represent squared magnitudes. Below, their collision coefficients lie on one exact quadratic curve. Four squares have side lengths equal to the four distinct L-value magnitudes at modulus 25. Their areas represent squared magnitudes. Below, their collision coefficients lie on one exact quadratic curve.
Each conjugate pair as a square with side |L(1, χ)|. Below, all four pairs on the exact curve, with no constant fitted.

Each conjugate pair gets a square whose side is ∣L(1,χ)∣|L(1,\chi)|∣L(1,χ)∣, from 0.4860.4860.486 for the pair 777 and 131313 up to 1.7101.7101.710 for 333 and 171717. Below, the four pairs sit exactly on the curve 55 x2/(4π2)5\sqrt5\,x^2/(4\pi^2)55​x2/(4π2), and nothing about the curve is fitted. Powers of a twentieth root of unity give both finite sums, and an exact polynomial identity forces the ratio of their sizes.

The identity behind the square law

Put z=χ(2)‾z=\overline{\chi(2)}z=χ(2)​. For the eight primitive odd characters, zzz runs through the primitive twentieth roots of unity. Since 3≡273\equiv2^73≡27 and 4≡22(mod25)4\equiv2^2\pmod{25}4≡22(mod25), the four-term sum is

p(z)=1+z+z2+z7.p(z)=1+z+z^2+z^7.p(z)=1+z+z2+z7.

Pairing opposite residues in the Bernoulli sum and reducing powers with Φ20(z)=0\Phi_{20}(z)=0Φ20​(z)=0 gives Bχ=−(2/5)q(z)B_\chi=-(2/5)q(z)Bχ​=−(2/5)q(z), where

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

The exact certificate is

x7[q(x)q(x−1)−5p(x)p(x−1)]=Φ20(x)h(x),\begin{aligned} x^7\bigl[q(x)q(x^{-1})&-5p(x)p(x^{-1})\bigr]\\ &=\Phi_{20}(x)h(x), \end{aligned}x7[q(x)q(x−1)​−5p(x)p(x−1)]=Φ20​(x)h(x),​

with

Φ20(x)=x8−x6+x4−x2+1,h(x)=−3+x2+5x3+x4−3x6.\begin{aligned} \Phi_{20}(x)&=x^8-x^6+x^4-x^2+1,\\ h(x)&=-3+x^2+5x^3+x^4-3x^6. \end{aligned}Φ20​(x)h(x)​=x8−x6+x4−x2+1,=−3+x2+5x3+x4−3x6.​

At x=zx=zx=z, the right side is zero and z−1=zˉz^{-1}=\bar zz−1=zˉ. Hence ∣q(z)∣2=5∣p(z)∣2|q(z)|^2=5|p(z)|^2∣q(z)∣2=5∣p(z)∣2. Together with ∣Dχ∣=2∣p(z)∣|D_\chi|=2|p(z)|∣Dχ​∣=2∣p(z)∣, this gives the claimed factor 5\sqrt55​.

Exponents ∣Bχ∣\lvert B_\chi\rvert∣Bχ​∣ ∣Dχ∣\lvert D_\chi\rvert∣Dχ​∣ ∣L(1,χ)∣\lvert L(1,\chi)\rvert∣L(1,χ)∣ ∣cχ∣\lvert c_\chi\rvert∣cχ​∣
1 / 19 2.470410 5.524004 1.552204 0.682328
3 / 17 2.720773 6.083832 1.709512 0.827636
7 / 13 0.772914 1.728289 0.485636 0.066791
9 / 11 1.377344 3.079834 0.865411 0.212099

Each row names a conjugate pair by its exponents in χ(2)=e2πij/20\chi(2)=e^{2\pi i j/20}χ(2)=e2πij/20. The two characters in a pair have equal magnitudes. The decimals are rounded, while the identity is exact.

Spend the forty-eight

Fourier coordinates describe the same table in a different basis, and the total squared size doesn’t change. With the averaged coefficients used here, Parseval gives

∑χ∣cχ∣2=4820.\sum_\chi|c_\chi|^2=\frac{48}{20}.χ∑​∣cχ​∣2=2048​.

Twenty squared table entries and eight Fourier contributions form two bar charts on the same vertical scale. Each chart totals 48. The lower bars include the factor twenty required by the averaged Fourier coefficients. Twenty squared table entries and eight Fourier contributions form two bar charts on the same vertical scale. Each chart totals 48. The lower bars include the factor twenty required by the averaged Fourier coefficients.
The same 48 two ways, as twenty squared entries above and as eight character contributions below.

The top panel is the table’s twenty squared entries, and the tall bars at 444 and 212121 are the two copies of 16/516/516/5, squared. The bottom panel is the same 484848 spread over the eight surviving characters, twenty times each squared coefficient. The pair 333 and 171717 takes about 13.713.713.7 apiece and the pair 111 and 191919 about 9.39.39.3, while 777 and 131313 barely register. Each coefficient already contains ∣L(1,χ)∣2|L(1,\chi)|^2∣L(1,χ)∣2, so every bar in the bottom panel is a fixed multiple of ∣L(1,χ)∣4|L(1,\chi)|^4∣L(1,χ)∣4. Keeping the normalization visible,

12516π4∑χ primitive odd∣L(1,χ)∣4=4820,\frac{125}{16\pi^4} \sum_{\chi\ \mathrm{primitive\ odd}}|L(1,\chi)|^4 =\frac{48}{20},16π4125​χ primitive odd∑​∣L(1,χ)∣4=2048​,

and the exact fourth moment at modulus 252525 is

∑χmod25χ primitive odd∣L(1,χ)∣4=192π4625.\boxed{\displaystyle \sum_{\substack{\chi\bmod25\\\chi\ \mathrm{primitive\ odd}}} |L(1,\chi)|^4=\frac{192\pi^4}{625}.}χmod25χ primitive odd​∑​∣L(1,χ)∣4=625192π4​.​

It adds the eight fourth powers outright, with no averaging. The fractions supply 484848, the Fourier normalization supplies 202020, and the factors of π\piπ come in through the classical LLL-value identity. Nothing is estimated from a sample of primes.

The weighted moment at every odd prime base

Write Eb=∑a∈Ub2∣f(a)∣2E_b=\sum_{a\in U_{b^2}}|f(a)|^2Eb​=∑a∈Ub2​​∣f(a)∣2. The factorization and Parseval give

∑χmodb2χ primitive odd∣L(1,χ)∣2∣Dχ∣2=π2b(b−1)b2 Eb.\begin{aligned} &\sum_{\substack{\chi\bmod b^2\\\chi\ \mathrm{primitive\ odd}}} |L(1,\chi)|^2|D_\chi|^2\\ &\hspace{2em}=\frac{\pi^2 b(b-1)}{b^2}\,E_b. \end{aligned}​χmodb2χ primitive odd​∑​∣L(1,χ)∣2∣Dχ​∣2=b2π2b(b−1)​Eb​.​

The diagonal factor is the weight in this second moment. Base five’s additional identity turns that weight into another squared LLL-value, producing the unweighted fourth moment above. The general factorization does not remove the diagonal factor.

The line opens into a cloud

At seven the two factors no longer sit on one line. The comparison covers every odd prime base from five through seventy-one, all 14,37214{,}37214,372 primitive odd characters, counting conjugate partners separately.

Four complete scatterplots compare the two finite magnitudes in bases five, seven, nineteen and seventy-one. Four points on a line become clouds of nine, eighty-one and 1,225 conjugate pairs. A lower chart records all eighteen correlations and highlights the upward steps. Four complete scatterplots compare the two finite magnitudes in bases five, seven, nineteen and seventy-one. Four points on a line become clouds of nine, eighty-one and 1,225 conjugate pairs. A lower chart records all eighteen correlations and highlights the upward steps.
Bernoulli size against short-sum size at bases 5, 7, 19 and 71, then the correlation at all eighteen bases, teal where it steps up.

Each panel plots the Bernoulli size across and the short-sum size up, both divided by their own root-mean-square size so the panels share a scale. At five the four pairs sit on the dashed diagonal. At seven nine pairs scatter to both sides of it, at nineteen eighty-one pairs form a loose band, and at seventy-one the 1,2251{,}2251,225 pairs make a cloud that still leans along the diagonal. The correlation drops from 111 at five to 0.84400.84400.8440 at seven and 0.67530.67530.6753 at seventy-one. The curve at the bottom shows the decline isn’t steady, since its four teal segments are steps up, from twenty-nine to thirty-one, thirty-seven to forty-one, fifty-nine to sixty-one and sixty-seven to seventy-one.

Both magnitudes come from finite sums, so these are complete families with no prime cutoff to push further. The open question is how the four-term sum, then the six-term sum, then the longer initial character sums behave against the Bernoulli factor as the base grows, and eighteen families don’t settle it.

All eighteen families, at full resolution
Eighteen scatterplots contain all 7,186 conjugate pairs from the prime bases five through seventy-one. Common axes reveal the widening clouds and their different arrangements. Each panel states its base, pair count and correlation. Eighteen scatterplots contain all 7,186 conjugate pairs from the prime bases five through seventy-one. Common axes reveal the widening clouds and their different arrangements. Each panel states its base, pair count and correlation.
All eighteen families on one shared scale, 7,186 conjugate pairs in all, at full resolution.
Base Characters Correlation
5 8 1.000000
7 18 0.844014
11 50 0.804214
13 72 0.800442
17 128 0.792520
19 162 0.763013
23 242 0.751734
29 392 0.720323
31 450 0.726701
37 648 0.708004
41 800 0.711126
43 882 0.698702
47 1,058 0.687538
53 1,352 0.685638
59 1,682 0.675935
61 1,800 0.679046
67 2,178 0.669012
71 2,450 0.675317

The two magnitudes are each divided by their own root-mean-square size within a family. This puts every panel on the same scale without changing its correlation. One dot represents a conjugate pair. The ledger counts both characters. The diagonal in each panel is a reference line of slope one.

Factoring the coefficient places the value at one inside it, beside a finite weight from the digit diagonal, and says nothing yet about the LLL-functions elsewhere in the critical strip. At five the account is exact. Twenty squared fractions sum to 484848, and 484848 times 4π4/6254\pi^4/6254π4/625 equals the sum of ∣L(1,χ)∣4|L(1,\chi)|^4∣L(1,χ)∣4 over the eight primitive odd characters modulo twenty-five.

The table, the square law and the ledger in their first drawings
The twenty base-five entries in their original column layout. Small labels identify the residues; large fractions give the centered values.
The twenty base-five entries in their first drawing, residues small and centered values large.
The original comparison of Bernoulli and diagonal magnitudes, followed by the coefficient’s quadratic relation to the L-value magnitude. Each point represents a conjugate pair.
The first drawing of the two magnitudes side by side, and of the coefficient against the L-value.
The original eighteen-base correlation ledger. The downward trend includes reversals and supplies no fitted growing-base law.
The first drawing of the eighteen-base correlation ledger, reversals included.
Companion paper: The Collision Spectrum and the L-Function Landscape →
Share

Discussion

Sign in to join the discussion.

← All articlesRead the paper →
← Previous: The Spectral Repulsion
Next: The Analytic Collision Transform →