All projects

Visual mathematics · Completed

three-square-balance

Exact record locations for the minimum spread of representations as three nonnegative squares.

At a glance

Evidence and scope

Repository measurements and recorded project results. Test and benchmark claims were inspected, not rerun during this catalog refresh. Physical line counts use the stated file scopes.

Published sequence
OEIS A395008Record locations for minimum spread in three-square representations.
Exhaustive range
n < 10^9Completeness is limited to the documented finite search range.
Verified records
65 locationsThe final retained record in range is 973,078,528 with spread 28,672.
Independent checks
4 pathsTwo differently ordered C searches plus PARI/GP and Python references.
Reviewed sources · 2026-10-15 · bf3b319a7e7c

27 tracked files · 7 documentation files · 0 workflow files. Inventory includes data and vendored or generated files; workflow presence does not establish a passing run.

Download inventory and line-count file scopes

Executive summary

three-square-balance asks how close the largest and smallest parts can be when an eligible integer is written as three nonnegative squares. It records the inputs where this minimum spread reaches a new high. The record-location sequence is published as OEIS A395008. The repository contains an optimized resumable search, an independently ordered validator, Python and PARI/GP references, exact data below one billion, and a browser visualization. The scaling law is proved; the finite-range completeness is exhaustively checked; the claimed pattern beyond that range is kept explicitly conjectural.

Problem and context

Not every integer is representable by three squares. For eligible values, the project minimizes x - z over ordered triples x ≥ y ≥ z ≥ 0. Zero spread occurs at three equal squares; larger values describe inputs forced farther away from that balanced shape.

Multiplication by four doubles every part, giving an exact scaling relation for the minimum spread. That makes scaled seed chains central to the observed record structure.

What was built

  • A fast OpenMP C engine with resumable long scans.
  • An independent ascending-search validator that does not reuse the production
  • search order.

  • Small Python and PARI/GP definitions for reference and sequence generation.
  • The 65 record locations and canonical triples below 10^9.
  • An interactive view of shape directions and record locations.

Evidence and validation

The repository retains exactly 65 record locations below one billion. Its final record in that range is 973,078,528 with a spread of 28,672. The primary and independent C paths use different enumeration orders, giving a stronger cross-check than running the same implementation twice. PARI/GP and Python paths provide additional small-scale reference behavior.

The record list in the verified range is reproduced by scaled terms from six primitive seeds, with one seed contributing only its initial term before another chain overtakes it.

The repository now also includes a LODA implementation (loda/A395008.asm) and a check against the retained b-file. It implements the observed recurrence; this additional implementation does not prove completeness beyond the scan bound.

Boundaries

The six-seed closure is exhaustive only below 10^9. Treating it as complete for all larger values is conjectural. The archived companion object records spread values rather than record locations and is not part of the published sequence focus.

Repository map and first run

cd src
make
make check
gp -q records.gp

Use sequences/spread-record-locations/ for the OEIS definition and b-file, data/ for retained records and triples, and viz/README.md for the visual mapping.