Localizing the Gardam unit: the support geometry of units in F_2[P] and its non-unique-product relatives

Abstract

Gardam's counterexample to the Kaplansky unit conjecture is a unit of $\mathbb{F}_2[P]$, $P$ the Promislow group, whose support has $21$ elements, and Gardam has asked whether $21$ is least possible. Over $\mathbb{F}_2$ the unit equation is a parity condition on a pair of supports, and for amenable groups it is necessary and sufficient, hence decidable by search over word balls. The picture at small radii is due to Dietrich-Lee-Nies-Vinyals, and the radius-$3$ statement is a special case of a theorem of Craven-Pappas. We re-derive these facts independently and upgrade them to certified statements: the $B(3)$ non-existence carries a DRAT proof checked by an independent verifier, and the threshold $m_u(P;4)=42$ -- every nontrivial unit with both supports in $B(4)$ has total support at least $42$ -- is certified by a $90$-way size-split DRAT sweep, confirming Gardam's expectation in ball-limited form. Alongside $P$ we sweep the torsion-free non-unique-product groups $H_4$ and the Nielsen-Soelberg groups $G_1,G_2,G_3$: none admits a unit in the searched balls ($H_4$, $G_1$, $G_3$ to radius $3$; $G_2$ to radius $1$). For $G_3$ this is sharp: the universal Nielsen-Soelberg group is exactly the amalgam $S$ of Gardam's second counterexample, and we localize the unit of his Theorem B into the radius-$4$ ball, so $r_u(G_3)=4=r_u(P)$ -- a second group whose unit localization radius is determined. Further, $H_4$ resists the twisted-unitary ansatz through radius $5$, evidence for the proposal of Dietrich-Lee-Nies-Vinyals that $H_4$ may satisfy the trivial units property while failing the unique product property. Finally we prove an effective localization principle: in $\mathbb{F}_2[P]$, if a nontrivial unit of total support $n$ exists, one exists inside an explicitly bounded ball, so the minimal support size is computable in principle; the new ingredient is a parity coarsening lemma.