A mathematical preprint ties exact sampling in a particular function space to the existence of a Hadamard matrix, the matrix structure that permits an N-node exact formula in this problem. The space is generated by the first N Rademacher functions, a structured family of mathematical functions used in the calculation. For the unrestricted problem, the minimum number of nodes needed to discretize its L2 norm exactly is N when a Hadamard matrix of order N exists, and N+1 when it does not.
The paper keeps two quantities separate: the minimum node count for exact discretization with unrestricted weights, and the minimum under a nonnegative-weight requirement. That distinction matters because the two restrictions ask different questions about how an exact formula can be built.
The calculation starts with a precise test
For this space, exactness has a precise algebraic test. The weighted product of any two distinct Rademacher functions must cancel across the selected nodes, and all the weights together must sum to 1. Those requirements are equivalent to exact discretization of the L2 norm here, so the problem can be studied through a finite set of equations.
After columns associated with zero weights are removed, the equations take the form of a weighted matrix equation. A rank argument then supplies the first hard boundary: any solution of the reduced equation needs at least N nodes.
When N nodes are enough
Reaching that boundary is exactly where Hadamard matrices enter. If a Hadamard matrix of order N exists, the selected matrix in the discretization is Hadamard as well, and the N weights are equal. Conversely, an N-node exact discretization would produce that matrix structure. The paper therefore identifies Hadamard-matrix existence as the precise condition for using only N nodes.
When the required Hadamard matrix does not exist, N nodes are unavailable. The preprint supplies an explicit construction with N+1 points, using dyadic intervals to build the needed arrangement. Combined with the rank lower bound, that construction fixes the unrestricted minimum at N+1 in those cases.
Positive weights create a separate barrier
The count changes under a stricter rule on the weights. For N congruent to 1 modulo 4 with N at least 5, or N congruent to 2 modulo 4 with N at least 6, strictly positive weights require more nodes than the unrestricted minimum. In both stated cases, that unrestricted minimum is N+1, so the strictly positive formula cannot attain the count available to the unrestricted one.
The positive-weight impossibility proof works directly with the matrix. The argument extends the weighted matrix by adding a unit row vector orthogonal to all existing rows. That added row creates the structure used to show that a strictly positive solution cannot exist at the unrestricted node count in the stated cases.
The paper also records what a minimum representation can look like when negative weights are permitted. When the unrestricted minimum is N+1, it gives an example with one negative weight and equal weights on all the remaining nodes. The example makes the distinction concrete: requiring positivity can increase the number of nodes.
A broader bound depends on a conjecture
How much positivity can cost more generally is tied to a separate conjecture. Conditional on the Hadamard conjecture, the paper states a uniform upper bound of two nodes for the gap between the positive-weight and unrestricted minima. The condition matters: this is a conditional bound, not an unconditional resolution of the matrix-existence problem.
The result remains a statement about one tightly defined mathematical setting: the subspace generated by the first N Rademacher functions and exact discretization of its L2 norm. The positive-weight conclusion reported here is likewise limited to the two congruence classes and thresholds specified above.
The document is an arXiv version 1 preprint dated 25 August 2026. No funding statement is reported in the supplied document or metadata.
Paper data and sources
Original title: On exact discretization of the $L_2$-norm in the space spanned by the first $N$ Rademacher functions
Authors: Anna Kazakova
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-25
DOI: Not available
Original paper · Full text