OpenAI preprint, September 2026 · Section 4. Finite-dimensional weights and sparse block operators
Lemma4.1:For
and
,
dimension
suffices up to size
Unit vectors with controlled subfamily synthesis.
Hover over a blue or green term in the text, or a node in the graph, to find it in the other. Select it to read its
declaration both in Lean and in words.
The Setting
The finite-dimensional question is how many unit vectors can be placed in a complex Hilbert space while keeping synthesis norms bounded on small subfamilies.
Let be an index set and a finite coordinate set, and let carry its standard Euclidean Hilbert norm. For a family in and a finite subset , the synthesis operator is the bounded complex-linear map given by (1): (1)✦
A family in has subfamily synthesis bound up to size if (2) holds, with the empty subfamily included: (2)✦
For integers and , the subfamily size and frame dimension are the parameters in (3), where is the natural logarithm: (3)
The Claim
Theorem 1. Let and be integers, with and as in (3). There is a family in satisfying (4): (4)✦For any finite set of cardinality , there is a family in satisfying (5): (5)✦
What Is Checked
The claims concern finite families in complex Euclidean spaces, with subfamilies measured by their cardinality and synthesis operators measured in operator norm.