Lossfunk

01 · model

The problem and the region identity

A constrained A

For an angle φ, write e(φ)=(cos φ,sin φ). Given an apex P∈ℝ², a radius r>0, and distinct unit vectors u,v, define

A(P,r;u,v) = {P+tu : t≥0} ∪ {P+tv : t≥0} ∪ [P+ru, P+rv].

The first two pieces are the legs; the last is the crossbar. Equal leg lengths are a metric constraint, so arbitrary affine deformations do not preserve the problem.

Generic arrangement

A cross-shape intersection is proper when it lies strictly beyond every participating ray apex and strictly inside every participating crossbar. An arrangement is generic when supporting lines from different A's are nonparallel, all actual intersections are proper, no vertex lies on a foreign component, and no three shapes meet at one point.

Regions are crossings plus a fixed contribution

Proposition

F = I + 2n + 1.

Here I is the number of finite proper crossings and F is the number of complementary regions.

To prove it, compactify the plane by one point at infinity. A single A becomes a connected graph with four vertices (apex, two bar endpoints, infinity) and five edges, hence first Betti number two. The union of n disjoint A's, with their infinity points identified, has first Betti number 2n. Every proper transverse crossing identifies two interior arc points and increases the first Betti number by one. The compactified union therefore has Betti number 2n+I, and Euler's formula on the sphere gives F=(2n+I)+1.

02 · extremal graph

The universal upper bound

Cutler, Karlsson, and Sloane proved two local facts: two constrained long-legged A's have at most eight proper finite intersections, and no three A's can be pairwise eight-crossing.

Form a graph G₈ with one vertex per A and an edge exactly when the pair has eight crossings. The three-object obstruction says that G₈ is triangle-free. If it has e edges, then every edge contributes at most eight crossings and every nonedge at most seven:

I ≤ 8e + 7(C(n,2) − e)
= 7C(n,2) + e.

Mantel's theorem gives e≤⌊n²/4⌋. Combining this with the region identity yields

a(n) ≤ 7C(n,2) + ⌊n²/4⌋ + 2n + 1.

Equality is rigid: the eight-crossing graph must be the balanced complete bipartite graph K⌊n/2⌋,⌈n/2⌉, and every pair within a part must have seven crossings. The lower construction now has a precise target.

03 · lower construction

A narrow family and a wide nested fan

Fix part sizes p,q≥1, put m=max(p,q), and choose a sufficiently small ε>0. The construction uses O=(0,0), C=(1,0), and h=3/2.

Ni

The narrow family

For 1≤i≤p, set

ai=3i,   αi=aiε, ri=cos αi, Pi=C−rie(αi).

Its leg directions are di±=e(αi±ε). Center the bar as

ni=e(αi+π/2), Mi=C−ri(1−cos ε)e(αi), si=risin ε.

Then Bi={Mi+tni:−si≤t≤si} and its supporting line is e(αi)·X=cos αi cos ε.

Wj

The wide family

For 1≤j≤q, set

qj=4−j,   θj=2 arctan qj, Dj=4j,   Lj=4m+j, λj=Ljε.

The low and high directions are j=e(λj) and uj=e(π−θjj). Put the high bar endpoint at Hj=C+ε(−Dj,h), then define

Rj=(1+hε)/sin(θj−λj), Qj=Hj−Rjuj, Kj=Qj+Rjj.

Thus (Qj)y=−1 and the bar joins Kj to Hj.

At ε=0, the relevant wide-family abscissae are

Aj = 1 + cot θj = 1 + (Dj−qj)/2,    Bj = 1 + cot(θj/2) = 1+Dj,

with Qj→(Aj,−1), Kj→(Bj,−1), and Hj→C.

The blow-up at C and the blow-up at y equals minus one
Left: the chart near C, where narrow legs/bar and the wide high-leg/bar fan become lines. Right: the chart near y=−1, where the wide low legs are ordered. The picture is schematic; the proof uses only strict slope and order inequalities.

The three required incidence masks

Order each shape's components as first leg, second leg, crossbar. A 1 means that the indicated proper component intersection is required; a 0 makes no assertion.

N–N · 7110
110
111
N–W · 8111
111
011
W–W · 7111
011
011

04 · incidence verification

Why the three pair types work

Blow-up stability

If, after an invertible affine rescaling, two supporting lines and their directed component domains converge to nonparallel lines meeting in the relative interior of both limiting domains, then their original components intersect properly for all sufficiently small positive ε. This follows because the two line-intersection parameters are continuous rational functions and the limiting domain inequalities are strict.

Narrow–narrow pairs: seven crossings

Take i<k, write a=ai and b=ak, so b−a≥3. Under the blow-up (x,y)↦(x,y/ε) near O, the σ-leg of Ni tends to

Y = −a + (a+σ)x,   x>0,   σ∈{−1,+1}.

The σ-leg of Ni and τ-leg of Nk meet at limiting abscissa x=(b−a)/(b−a+τ−σ)>0, because the denominator is at least (b−a)−2≥1. Hence all four leg–leg intersections are proper.

For the bar of Ni against the τ-leg of Nk, put A=aε, B=bε. Writing the point as Mi+t ni, direct line intersection gives

ρτ = [cos A(cos ε−1)+cos B cos(B−A)] / cos(B−A+τε) = 1+O(ε²),
tτ = −cos B sin(B−A)+ρτ sin(B−A+τε).

The endpoint-sensitive Taylor expansion is

t+si = ((b−a)(2a+1)/2)ε³+O(ε⁵),
si−t = 2ε+O(ε³),
si−t+ = ((b−a)(2a−1)/2)ε³+O(ε⁵),
t++si = 2ε+O(ε³).

Since a=3i≥3, all four margins are positive for small ε, so the earlier bar meets both later legs strictly inside.

Finally every narrow bar line passes through Xε=(cos ε,0). Its bar coordinate is sin(aε)(1−cos ε)=O(ε³), while the half-length is si=ε+O(ε³). Therefore two narrow bars meet in both interiors. Total: 4+2+1=7.

Narrow–wide pairs: eight crossings

Fix Ni,Wj, and abbreviate a=ai, L=Lj, D=Dj, μ=tan θj=2qj/(1−qj²), and ν=tan(θj/2)=qj.

The determinant between the wide low direction and the σ-leg of the narrow A is sin((L−a−σ)ε), and L−a−σ≥(4m+1)−(3m+1)=m>0. The two ray parameters satisfy

εrσ → 1/(L−a−σ),   εsσ → 1/(L−a−σ),

so both narrow legs meet the wide low leg properly.

For the other six incidences, use the chart (X,Y)=((x−1)/ε,y/ε) at C. The narrow legs tend to Y=±1 and the narrow bar tends to X=0, −1<Y<1. The wide high leg and bar tend to

Hj: Y−h = −μ(X+D),
Cj: Y−h = −ν(X+D),   X>−D.

The line Y=σ meets the wide bar with X+D=(h−σ)/ν>0, since h=3/2>1. Thus both narrow legs meet the wide bar in its interior. The wide high-ray apex escapes to infinity opposite the ray, so both finite intersections with the narrow legs lie inside the ray.

At X=0, the high leg and bar have heights h−Dμ=3/2−2/(1−qj²) and h−Dν=1/2. Since 0<qj≤1/4, the first lies in [−19/30,−1/2); both heights are strictly between −1 and 1. The narrow bar therefore meets the wide high leg and wide bar properly. Total: 2+6=8.

Wide–wide pairs: seven crossings

Let j<k and write μr=2qr/(1−qr²), νr=qr. Because qk≤qj/4 and qj≤1/4,

μj > νj > μk > νk > 0.

In the chart at C, every high leg or crossbar of Wr limits to a line through (−Dr,h) of slope −s, with s=μr or νr. For a j-line of slope −s and k-line of slope −t, s>t. With Δ=Dk−Dj>0, their displacements past their high endpoints are

Uj = tΔ/(s−t) > 0,   Uk = sΔ/(s−t) > 0.

These strict positive displacements give all four high/bar incidences: high–high, high–bar, bar–high, and bar–bar.

It remains to make the low leg of Wj meet all three components of Wk. In the chart (x,Y)=(x,(y+1)/ε), the low leg of Wr tends to Y=Lr(x−Ar), x>Ar. The k-th high leg tends locally to the half-line x=Ak,Y>0; its bar tends locally at the low endpoint to x=Bk,Y>κk, where

κk = Lk/sin θk = Lk(Dk+qk)/2.

Since Aj<Ak<Bk, the low j-leg hits the high k-leg forward. For the bar hit, Dj≤Dk/4 and Lk/Lj<5/4 give

Bk−Aj ≥ 7Dk/8,
κk/Lj < (5/4)(257/256)(Dk/2) < 7Dk/8.

Hence the low j-leg hits inside the k-bar. Finally the two low limiting lines meet at

x* = Ak + Lj(Ak−Aj)/ (Lk−Lj) > Ak > Aj,

so the low–low crossing is forward on both rays. These three low incidences plus the four high/bar incidences give 7.

05 · general position

Generic rational perturbation

The transparent base construction has intentional degeneracies; for example all narrow crossbar supporting lines concur at (cos ε,0). This is removed without losing any required crossing.

Choose ε small enough that all strict component-domain inequalities above hold simultaneously. They define an open neighborhood U in the parameter space of apices, radii, and leg directions. Use the rational chart for each unit direction

t ↦ ((1−t²)/(1+t²), 2t/(1+t²)).

After clearing denominators, every unwanted degeneracy—support-line parallelism, a foreign component through an apex or bar endpoint, or coincidence of two distinctly labelled finite crossings—is contained in the zero set of a polynomial in the rational chart parameters. Only finitely many labels occur. None of these polynomials vanishes identically on U, because a sufficiently small independent change of one participating apex, radius, or direction breaks the event. The bad locus is therefore a finite union of proper algebraic zero sets and has empty interior.

Rational points are dense in the chart, so U contains a rational point outside every bad set. This gives a generic rational arrangement with all strict required incidences preserved.

Completion

Take p=⌊n/2⌋ and q=⌈n/2⌉. There are pq=⌊n²/4⌋ cross-family pairs with eight crossings and all remaining pairs have seven. Thus

I ≥ 8pq + 7(C(p,2)+C(q,2))
= 7C(n,2) + ⌊n²/4⌋.

The upper bound gives equality, and F=I+2n+1 proves the theorem.

Every bipartition is realizable

The same argument works for arbitrary p,q≥1: there is a generic rational arrangement with eight crossings on every cross-family pair and exactly seven on every same-family pair. The exact region count is

7C(p+q,2) + pq + 2(p+q) + 1.

Colored extremal graphs also play a central role in the related published paper A Two-Graph Refinement of Paulsen's Lollipop Bounds.

06 · the 55-region witness

The exact n=4 construction

Four A's need 46 globally distinct proper crossings: four cross-part pairs contribute eight each and two within-part pairs contribute seven each. The region identity then gives 46+2·4+1=55.

Primary witness complicated rational coordinates · 46 crossings · 55 regions Download n04.json
Full 55-region arrangement produced from the complicated rational n=4 certificate
The full exact rational witness from n04.json. Horizontal scale is enlarged to separate its extremely compressed geometry. The colors distinguish the four A's and the face coloring makes the 55 complementary regions visible.
Numbered detail of the complicated rational n=4 witness showing its dense crossing region
Numbered detail of the same certificate. This view contains 45 of the 46 crossings; the region labels let the dense local structure be inspected rather than inferred from a compressed overview.

A low-height rational companion

A later rational simplification found a second witness with small, human-readable fractions. It proves the same 7/8 crossing pattern, but it is not the coordinate set used in the two primary figures above.

Three affine views of the simplified low-height rational n=4 witness
The low-height certificate in three affine views: the full arrangement, the main 43-crossing cluster, and a 30-crossing core. The anisotropic display transform only improves visibility; all intersection predicates are verified in the original Euclidean coordinates.

07 · exact finite checks

Rational certificates from n=4 through n=16

These thirteen files are independent reproducibility checks, not premises of the all-n proof. Every coordinate and direction parameter is rational. The verifier uses only Python's standard library and fractions.Fraction; it does not make floating-point or visual decisions.

What “exact” means here. The finite certificates can be checked mechanically without human geometric judgment: every sign, equality, and distinctness test is performed in rational arithmetic. The analytic all-n theorem is a mathematical proof and remains appropriate for ordinary expert review.
Show:
npartscrossingsa(n)exact certificate
42+24655n04.json
52+37687n05.json
63+3114127n06.json
73+4159174n07.json
84+4212229n08.json
94+5272291n09.json
105+5340361n10.json
115+6415438n11.json
126+6498523n12.json
136+7588615n13.json
147+7686715n14.json
157+8791822n15.json
168+8904937n16.json

a(0),…,a(16) 1, 3, 13, 30, 55, 87, 127, 174, 229, 291, 361, 438, 523, 615, 715, 822, 937.

What the exact verifier checks

  1. equal positive leg lengths, positive orientation, and positive radius for every A;
  2. all nine component-pair predicates for every pair, with strict ray and crossbar domains;
  3. the balanced seven/eight pair-crossing matrix;
  4. absence of cross-shape support-line parallelism;
  5. absence of every foreign apex or crossbar-endpoint incidence; and
  6. global distinctness of all proper rational crossings.

08 · research provenance

How the proof was reached

The path was a sequence of exact computational discoveries followed by an analytic compression. This record is included to make the model-assisted provenance explicit.

  1. Stage 1

    The n=4 breakthrough

    The 55-region discovery and its first proof were developed in the ChatGPT web UI with GPT-5.6 sol pro. This produced the complicated rational witness shown above and the exact target pattern: four 8-crossing pairs and two 7-crossing pairs.

  2. Stage 2

    Exact certificates through n=16

    Local Codex running GPT-5.6 sol xhigh, using the n=4 manuscript as its starting point, generated and checked rational equality certificates for every n from 5 through 16. Those experiments made the balanced complete bipartite 8-crossing graph impossible to miss.

  3. Stage 3

    The all-n construction

    A zip containing the certificates, proof directions, and discovery artifacts was supplied to the ChatGPT web UI running GPT-5.6 sol pro. It found the narrow/wide parametric construction, converted the finite pattern into the all-n analytic proof on this page, and closed the sequence.

  4. Stage 4

    Independent local audit

    Local Codex audited the proof, including an exact symbolic expansion of the endpoint-sensitive narrow–narrow calculation, an independent rational verifier for the finite certificates, and regression checks of the parametric base through balanced size 64. The models also assisted in preparing the manuscript and this HTML presentation.

Separation of roles. The certificates establish finite arrangements by exact arithmetic. The theorem for every n is established by the upper-bound argument, the explicit two-family construction, and the generic rational perturbation. Neither claim asks the reader to trust a plotted image or a floating-point optimizer.

09 · reproducibility

Paper, source, scripts, and archived outputs

Integrity

Package documentation

Original minimal-package notes and SHA-256 ledger.

README.md SHA256SUMS

Run the checks

Preserve the scripts/, certificates/, and verification/ sibling folders shown on this page, then run:

python3 scripts/verify_certificates.py \
  --output verification/exact_certificates_n4_n16.json

python3 scripts/verify_parametric_base.py \
  --max-n 64 \
  --output verification/parametric_base_n2_n64.json

Python 3.10 or later is recommended. No third-party package is used.

10 · sources

References

  1. David O. H. Cutler, Jonas Karlsson, and Neil J. A. Sloane, Cutting a Pancake with an Exotic Knife, arXiv:2511.15864. The two-object maximum and the obstruction to three pairwise eight-crossing A's are the local upper-bound inputs.
  2. OEIS Foundation Inc., A397182, “maximum number of regions formed by n constrained long-legged A's.”
  3. Willem Mantel, “Problem 28,” Wiskundige Opgaven 10 (1907), 60–61.
  4. Siddhartha Mahajan and Paras Chopra, A Two-Graph Refinement of Paulsen's Lollipop Bounds, arXiv:2606.06064 (2026).
  5. Siddhartha Mahajan and Paras Chopra, The Exact Region Formula for Arrangements of Constrained Long-Legged A's (23 July 2026).