Fuglede's conjecture (Fuglede 1974) asserts that a measurable set
with finite positive Lebesgue
measure admits a spectrum iff
it is possible to tile
by a family of translates
of
. Fuglede proved the conjecture in the special case
that the tiling set or the spectrum are lattice subsets of
and Iosevich et al. (2001) proved that no smooth
symmetric convex body
with at least one point of nonvanishing Gaussian
curvature can admit an orthogonal basis of
exponentials.
Using complex Hadamard matrices of orders 6 and 12, Tao (2003) constructed counterexamples to
the conjecture in some small Abelian groups, and
lifted these to counterexamples in or
.
However, the conjecture has been proved in a great number of special cases (e.g., all planar convex bodies) and remains an open problem
in small dimensions. For example, it has been shown
in dimension 1 that a nice algebraic characterization
of finite sets tiling indeed implies one side of Fuglede's conjecture (Coven and
Meyerowitz 1999). Furthermore, it is sufficient to prove these conditions when the
tiling gives a factorization of a non-Hajós cyclic group (Amiot 2005).
The finite-group version uses restrictions of group characters in place of complex exponentials. Kadowaki (2026) proved that a nonempty
subset of
admits an orthogonal basis
of characters iff it tiles
by translations. Kadowaki (2026) also proved that every
16-element subset of a binary vector
space admitting such a basis tiles that vector space
by translations.
Here a subset is spectral when it admits the orthogonal basis of characters just specified. Ferguson
and Iosevich (2026a) extended the finite-field results to and
: a nonempty subset is spectral
iff it tiles the ambient vector
space by translations. In
this is also equivalent to having a tiling complement
that is a subspace. In
they proved the tiling-to-spectral implication, but the
converse remains unresolved for spectral sets of size 64.
Ferguson and Iosevich (2026a) credit ChatGPT with assistance writing the computational
verification programs.
Ferguson and Iosevich (2026b) gave a standalone treatment of the binary dimension-eight result and a simpler proof in dimension seven. The latter uses the five equivalence classes of order-16 Hadamard matrices in place of the order-32 classification. The dimension-eight theorem is the same result included in Ferguson and Iosevich (2026a).
For subsets of the real line, Bell (2026) proved that a spectral set which, up to a set of Lebesgue
measure zero, is a union of four bounded intervals
tiles
by translations. This establishes the spectral-to-tiling
implication for four intervals.
Allowing an exponential Riesz basis, instead of an orthonormal basis, gives the weaker notion of a Riesz spectrum.
Ferguson and Iosevich (2026c) distinguish translational tiling from having a linear-subspace complement in . A complement
to a set
means that each element of
has a unique expression
with
and
. Every nonempty translational
tile has a complement that is a linear subspace precisely
in dimensions
. Their explicit 64-point
tile in
has no such complement, but it does have a set spectrum.
It is therefore not a counterexample to the equivalence
between tiling and spectrality. The dimension-eight positive result uses an earlier
perfect-code classification.