ERDŐS/DAILY

← back to the ledger

ERDőS #1083 · PARTIAL

Erdős problem #1083 — wave 8d report

Date of audit: 2026-07-28 (UTC)

This report gives two concrete outputs. First, it proves an elementary,

dimension-uniform finite regime:

\[ f_d(n)=1\quad(2\leq n\leq d+1),\qquad f_d(n)=2\quad\left(d+2\leq n\leq {d+1\choose2}\right) \tag{1} \]

for every \(d\geq3\). Second, published few-distance classifications,

combined with exact-arithmetic witnesses checked here from scratch, give

all the values of \(f_3(n)\) through \(n=21\) and of \(f_4(n)\) through

\(n=25\). These are finite results and do not settle the fixed-\(d\),

\(n\to\infty\) question.

The labels used throughout are:

external theorem;

named and linked;

proposed route that is not a theorem;

with its scope stated explicitly.

0. Mandatory live-page audit

I fetched both the live page, its

LaTeX view, and its

discussion thread

through a Bright Data browser session on 2026-07-28. Direct page status

and metadata below are live-page observations, not mathematical

inferences.

Verbatim current statement

> Let $d\geq 3$, and let $f_d(n)$ be the minimal $m$ such that every set of $n$ points in $\mathbb{R}^d$ determines at least $m$ distinct distances. Estimate $f_d(n)$ - in particular, is it true that\[f_d(n)=n^{\frac{2}{d}-o(1)}?\]

Everything currently listed on the page

\[ n^{1/d}\ll_d f_d(n)\ll_d n^{2/d}, \]

with the upper bound supplied by lattice points.

\(f_3(n)\gg n^{1/2}\).

\[ f_d(n)\gg n^{1/(d-90/77)-o(1)} \]

for \(d\geq3\), including exponent \(0.546\) for \(d=3\).

\[ f_d(n)\gg_d n^{2/d-c/d^2}\qquad(d\geq4) \]

for some \(c>0\). The page explicitly says that its displayed

three-dimensional consequence combines their recursion with the

Guth–Katz planar result and is slightly stronger than the result

printed in their paper.

\(g_d(n)>m\) iff \(f_d(m)

#1089, and emphasizes fixed \(d\) with \(n\to\infty\).

collaborating”, “Currently working on this problem”, both difficulty

votes, and both formalisation-work markers are all None.

“Slight typo, Gubias should be spelled Guibas.”

Thus the requested collision/claimed-proof stop rule did not trigger.

1. Primary-source literature audit

Verified asymptotic sources

(b) Solymosi and Vu, *Near optimal bounds for the Erdős distinct

distances problem in high dimensions*, Combinatorica 28 (2008),

113–125 (author PDF,

DOI), really does state

\[ \Omega\!\left(n^{\,2/d-2/(d(d+2))}\right)\quad(d\geq3) \]

in its abstract and Corollary 1.4 for \(d\geq4\). This verifies the

paper and the \(2/d-O(d^{-2})\) form cited by the live page; the live

page's \(d=3\) update is recorded above exactly as the page states it.

(b) Bardwell-Evans and Sheffer, *A Reduction for the Distinct

Distances Problem in \(\mathbb R^d\)*,

arXiv:1705.10963v2, J. Combin.

Theory Ser. A 166 (2019), 171–225

(DOI), supplies the

sharpest concrete reduction I found. Its exact bottleneck is recorded

in Section 4 below.

Auditing papers that claim the full exponent

There are two easy-to-find preprints whose abstracts appear to close

the problem; neither provides a usable solution.

arXiv:2002.01248, claimed

\(\Omega(n^{2/d})\) for \(d\geq3\), but the authoritative arXiv record

is withdrawn and says: “the proof of Theorem 1.2 is not correct.”

It is therefore not evidence that #1083 is solved.

arXiv:2002.00502v10, last revised

6 May 2026, is not withdrawn and claims

\(\gg n^{2/k-o(1)}\). The claimed conclusion does not follow from

its proof for elementary quantifier reasons:

1. Problem #1083 asks for a lower bound for every \(n\)-point

configuration. The proofs of Theorems 3.1 and 3.2 instead begin

“carefully choose \(n\) points” and conclude only for that chosen

construction. An existential construction can upper-bound

\(f_d(n)\); it cannot supply the required lower bound.

2. In the proof of Theorem 3.2, the summation is restricted by

\(d_i\ne d_j\) before the number of distinct gaps has been proved.

This assumes the distinctness that the argument is meant to

establish.

3. The unit-distance proof at one point takes the cardinality of a

set of numerical distances all constrained to equal \(1\). Such a

set has cardinality at most one; it is not the number of ordered

or unordered point pairs at unit distance.

Any one of (1)–(3) blocks the claimed application. This is a direct

logical audit of the current v10 text, not an appeal to reputation or

citation counts.

(c) I searched exact-title, exponent, incidence, and higher-

dimensional-distinct-distance queries through July 2026. I found

special-configuration results (curves, surfaces, other norms), the two

claims audited above, and the 2019 reduction, but no verified

unconditional improvement beyond the live page's displayed

Solymosi–Vu/Guth–Katz consequences for arbitrary Euclidean point sets.

This is an honest search miss, not a theorem that no such paper exists.

It agrees with the live page's current OPEN status and displayed best

bounds.

2. Exact finite progress

For \(s\geq0\), define

\[ M_d(s)=\max\{|X|:X\subset\mathbb R^d\text{ has at most }s \text{ nonzero pairwise distances}\}. \]

The bridge from few-distance sets to \(f_d\)

Lemma (a). If \(M_d(s-1)

\(Y\subset\mathbb R^d\), \(N\geq n\), having at most \(s\) distances,

then \(f_d(n)=s\).

Proof. An \(n\)-point set with at most \(s-1\) distances would

contradict \(n>M_d(s-1)\), so every \(n\)-point set has at least \(s\)

distances. Any \(n\)-point subset of \(Y\) has at most \(s\)

distances. The two inequalities give equality. \(\square\)

A dimension-uniform elementary interval

(a) At most \(d+1\) points in \(\mathbb R^d\) can be equidistant.

Indeed, after choosing one point as the origin, the \(r\) difference

vectors of an equilateral \((r+1)\)-point set have Gram matrix with

diagonal \(\lambda^2\) and off-diagonal \(\lambda^2/2\). Its

eigenvalues are positive, so it has rank \(r\), whence \(r\leq d\).

The \(d+1\) standard basis vectors in their \(d\)-dimensional affine

hull attain the bound. Consequently,

\[ M_d(0)=1,\qquad M_d(1)=d+1. \]

(a) Now take the Johnson configuration

\[ J(d+1,2)=\{e_i+e_j:0\leq iIt has \({d+1\choose2}\) points in the affine hyperplane

\(\sum x_i=2\). Its affine dimension is \(d\): relative to

\(e_0+e_1\), the vectors

\[ e_j-e_1\ (2\leq j\leq d),\qquad e_2-e_0 \]

are \(d\) independent differences. Two distinct two-element supports

intersect in zero or one element, so their squared distance is,

respectively, \(4\) or \(2\). Both cases occur because \(d+1\geq4\).

Applying the lemma proves (1) for every \(d\geq3\).

Named classification inputs

The following are theorem inputs, not conclusions of the supplied

checker.

| input | exact value | source and scope |

|---|---:|---|

| \(M_3(2)\) | 6 | (b) Nozaki–Shinohara's table of known optimal two-distance sets, arXiv:0906.0199, citing the Croft and Einhorn–Schoenberg classifications |

| \(M_4(2)\) | 10 | (b) the same table, citing the four-dimensional classification and Lisoněk's construction |

| \(M_3(3)\) | 12 | (b) Shinohara, arXiv:1309.2047, and the independent verification in Szöllősi–Östergård, Theorem 4.5 |

| \(M_3(4)\) | 13 | (b) Szöllősi–Östergård, Theorem 4.3 |

| \(M_4(3)\) | 16 | (b) Szöllősi–Östergård, Theorem 4.4 |

| \(M_3(5)\) | 20 | (b) Nozaki–Shinohara, Theorem 1.2, arXiv:2009.13111 |

The Szöllősi–Östergård source is *Constructions of maximum

few-distance sets in Euclidean spaces*,

arXiv:1804.06040, Electronic

Journal of Combinatorics 27(1) (2020), P1.23

(DOI). Its classification proofs use

isomorph-free graph generation and Gröbner bases. I treat those

published classification theorems as named inputs (b); I do not

pretend that the small verifier below reproduces their exhaustive

searches. Since the maxima for fewer distances in each row are

strictly smaller, the cited “exactly \(s\)-distance” classifications

give the displayed “at most \(s\)” values.

Explicit witnesses and exact distance checks

Put \(\phi=(1+\sqrt5)/2\). The three-dimensional coordinate families

used are:

\[ \begin{aligned} T&=\{(1,1,1),(1,-1,-1),(-1,1,-1),(-1,-1,1)\},\\ O&=\{\pm e_1,\pm e_2,\pm e_3\},\\ I&=\{(0,\pm1,\pm\phi),(\pm1,\pm\phi,0), (\pm\phi,0,\pm1)\},\\ D&=\{(\pm1,\pm1,\pm1)\}\\ &\quad{}\cup\{(0,\pm\phi^{-1},\pm\phi), (\pm\phi^{-1},\pm\phi,0), (\pm\phi,0,\pm\phi^{-1})\}. \end{aligned} \]

All signs in a displayed family vary independently. In dimension

four, the witnesses are the five standard basis vectors in their

affine hull, \(J(5,2)\), the rational squared-distance matrix

\(G_{16}(1,2,3)\) printed in the verifier, and the 24-cell

\[ C_{24}=\{\text{all permutations of }(\pm1,\pm1,0,0)\}. \]

The complete exact profiles are:

| ambient dimension | witness | size | distinct squared distances |

|---:|---|---:|---|

| 3 | \(T\) | 4 | \(\{8\}\) |

| 3 | \(O\) | 6 | \(\{2,4\}\) |

| 3 | \(I\) | 12 | \(\{4,6+2\sqrt5,10+2\sqrt5\}\) |

| 3 | \(I\cup\{0\}\) | 13 | previous set plus \(\{(5+\sqrt5)/2\}\) |

| 3 | \(D\) | 20 | \(\{4,6-2\sqrt5,6+2\sqrt5,8,12\}\) |

| 3 | \(D\cup\{0\}\) | 21 | previous set plus \(\{3\}\) |

| 4 | regular 4-simplex | 5 | \(\{2\}\) |

| 4 | \(J(5,2)\) | 10 | \(\{2,4\}\) |

| 4 | \(G_{16}(1,2,3)\) | 16 | \(\{1,2,3\}\) |

| 4 | \(C_{24}\cup\{0\}\) | 25 | \(\{2,4,6,8\}\) |

The coordinate identities in this table are (a): they are direct

algebra in \(\mathbb Q(\sqrt5)\). The \(G_{16}\) realization is

additionally checked (d) as follows. Anchor its sixteenth point at

zero and form twice the \(15\times15\) Gram matrix

\[ C_{ij}=D_{i,16}+D_{j,16}-D_{ij}. \]

Exact rational symmetric Schur-complement elimination finds \(C\)

positive semidefinite of rank \(4\), and reconstructs every entry of

the prescribed distance matrix. Thus it is an exact Euclidean

distance matrix for 16 distinct points in \(\mathbb R^4\), with no

floating-point tolerance.

Resulting exact table

Combining the bridge lemma, the named maximum theorems, and arbitrary

subsets of the witnesses gives:

| dimension | exact values |

|---:|---|

| 3 | (b) \(f_3(n)=1\) for \(2\leq n\leq4\); \(=2\) for \(5\leq n\leq6\); \(=3\) for \(7\leq n\leq12\); \(=4\) for \(n=13\); \(=5\) for \(14\leq n\leq20\); \(=6\) for \(n=21\) |

| 4 | (b) \(f_4(n)=1\) for \(2\leq n\leq5\); \(=2\) for \(6\leq n\leq10\); \(=3\) for \(11\leq n\leq16\); \(=4\) for \(17\leq n\leq25\) |

For \(n=1\), the natural convention gives \(f_d(1)=0\) (a).

3. Standalone reproducibility

The verifier is

runs/erdos1083_wave8d_reverify.py.

It uses only the Python standard library. It performs exact arithmetic

in \(\mathbb Q(\sqrt5)\) or \(\mathbb Q\), checks distinctness, all

pairwise squared-distance multiplicities, affine dimensions, the

\(G_{16}\) Gram certificate, the interval bookkeeping, and a regression

sample of the dimension-uniform Johnson construction. It explicitly

prints the classification values as theorem inputs rather than

silently assuming that it proved them.

Run:

python runs/erdos1083_wave8d_reverify.py

The 2026-07-28 run ended:

PASS G16 distance matrix: 16 distinct abstract points, squared distances {1,2,3}, exact Gram PSD rank 4
...
ALL EXACT-ARITHMETIC CHECKS PASSED

The fact that this particular execution passed is (d). Every

coordinate identity checked by it can also be read as the finite exact

algebra described above; no randomized search or numerical solver is

used.

4. What remains asymptotically: a precise incidence wall

Here \(N\) denotes the number of flats, to avoid confusing it with a

point-set size elsewhere.

(b) Bardwell-Evans–Sheffer Theorem 1.2 reduces the desired

\(\Omega(n^{2/d})\) lower bound to this rich-point problem. Given \(N\)

distinct \((d-1)\)-flats in \(\mathbb R^{2d-1}\), assume:

1. every two flats meet in at most one point;

2. every point is incident to \(O(\sqrt N)\) flats; and

3. every hyperplane contains \(O(\sqrt N)\) of the flats.

For \(2\leq k=O(N^{1/d+\varepsilon})\), proving

\[ \#\{\text{\(k\)-rich points}\} =O\!\left(\frac{N^{(2d-1)/d}}{k^{2+\varepsilon}}\right) \tag{2} \]

for the specially structured flats produced by their reduction would

give the conjectured \(\Omega(n^{2/d})\) distinct-distance bound.

The word “specially” is essential. The paper explicitly says that

(2) is false for arbitrary flats even under the three displayed

conditions; further restrictions inherited from its

\(\operatorname{Spun}(d)\)/Lie-group construction must be used.

Polynomial partitioning handles the analogous \((d-1)\)-flat problem

in \(\mathbb R^{2d-2}\), while \(\mathbb R^{2d-1}\) is described there

as just beyond the method's capabilities.

For \(d=3\), the paper isolates an even more concrete possible missing

lemma: a sufficiently strong distinct-distance theorem for points on

an arbitrary constant-degree surface in \(\mathbb R^3\). At the time

of that paper, the requisite form was available for planes, spheres,

and two-sheeted hyperboloids, but not arbitrary constant-degree

surfaces. Thus a legitimate asymptotic advance must either:

  • (c) prove (2) using additional algebraic restrictions actually

satisfied by the \(\operatorname{Spun}(d)\) flats; or

  • (c) replace this reduction with one that avoids the

codimension-one incidence loss (and, for \(d=3\), the general-surface

obstruction).

Merely proving an incidence statement for arbitrary flats cannot work,

because the cited paper gives counterexamples to that formulation.

This is the exact missing structural lemma exposed by the strongest

verified reduction found in this audit.

5. The next finite wall and its cost

(c) Extending the three-dimensional table by the same method would

require either a 22-point six-distance witness or a proof/classification

of \(M_3(6)\). I did not locate such a classification in the audited

few-distance literature.

(a) A deliberately naive labeled search for a 22-point

six-distance set starts with six colors on the

\({22\choose2}=231\) edges:

\[ 6^{231}=10^{179.753\ldots} \]

assignments. Even at the unrealistically generous rate of \(10^9\)

assignments per core-second, this is about

\(10^{167.197}\) core-hours (a). At an assumed \(10^5\) exact

Gram/rank checks per core-second (c), it would be about

\(10^{171.197}\) core-hours (a). The verifier independently prints

both orders of magnitude (d). These figures are only the raw-space

cost, not a complexity lower bound: isomorph-free generation,

distance-color permutations, forbidden minors, and incremental

rank/Gröbner constraints can prune enormously. A credible next

computation would have to implement all of those ideas; blind

enumeration is not a few-CPU-minute experiment.

Finally, no finite list of exact values supplies the uniform

\(n\to\infty\) step. The results in Section 2 are genuine exact

progress, but the incidence estimate (2), or an equally strong

replacement, remains necessary for the question on the live page.

PARTIAL: Exact values (modulo named published few-distance classifications) are established for \(f_3(n)\) through \(n=21\) and \(f_4(n)\) through \(n=25\), with an elementary all-\(d\) two-distance interval and exact-arithmetic witnesses; the asymptotic conjecture remains at the structured rich-flat incidence bottleneck.

This is the AI working report, labelled by outcome — not an independently verified claim unless marked PROVED. ← ledger