Affine Spectral-Independence in Discrepancy Theory

• Affine spectral-independence is a covariance domination condition that extends spectral-independence to entire families of affine combinations, controlling update variances.
• It is employed in SDP-guided discrete Brownian motion frameworks to achieve improved discrepancy bounds in Beck-Fiala and Komlós problems.
• The technique decouples the evolution of multiple discrepancy rows, fostering near pairwise-independence essential for efficient algorithmic rounding.

Affine spectral-independence is a covariance domination condition for random updates that extends the usual spectral-independence or sub-isotropy constraint from single linear forms to entire families of affine combinations. In the discrepancy-theoretic formulation introduced in "Decoupling via Affine Spectral-Independence: Beck-Fiala and Komlós Bounds Beyond Banaszczyk," a zero-mean random vector $u\in\mathbb{R}^h$ with covariance $U$ is $\alpha$-affine spectrally-independent with respect to a fixed matrix $E\in\mathbb{R}^{r\times h}$ if

$EUE^\top \preceq \alpha\,\mathrm{diag}(EUE^\top).$

The paper uses this condition inside an SDP-guided discrete Brownian motion to obtain efficient polynomial-time algorithms, resolve the Beck-Fiala Conjecture for $k\geq \log^2 n$, and prove a $\widetilde{O}(\log^{1/4}n)$ discrepancy bound for the Komlós problem, improving on the $O(\sqrt{\log n})$ bound due to Banaszczyk (Bansal et al., 5 Aug 2025).

1. Definition and relation to spectral-independence

Let $h\in\mathbb{N}$, let $U\in\mathbb{R}^{h\times h}$ be positive semidefinite, let $U$0 denote the diagonal matrix with the same diagonal as $U$1, and let $U$2 be a fixed “affine” matrix. A zero-mean random vector $U$3 with covariance $U$4 is $U$5-affine spectrally-independent with respect to $U$6 if

$U$7

Equivalently, for every $U$8,

$U$9

In this formulation, $\alpha$0 is the vector of affine-combinations of the coordinates of $\alpha$1, and $\alpha$2 in the applications described in the paper, with $\alpha$3 a fixed constant. The condition is therefore a structured decoupling statement: it bounds the covariance matrix of a whole family of discrepancy updates by its diagonal part, rather than controlling each coordinate in isolation.

The paper presents affine spectral-independence as a simple extension of the usual “$\alpha$4-spectral-independence” or “sub-isotropy” constraint. In the standard form,

$\alpha$5

which is equivalent to

$\alpha$6

The new condition lifts this from single linear forms to matrices $\alpha$7 chosen to encode families of row-discrepancy updates (Bansal et al., 5 Aug 2025).

2. SDP-guided Brownian motion

The algorithmic framework is a discrete Brownian motion guided by a semidefinite program. At time $\alpha$8, the process maintains a PSD matrix $\alpha$9, where $E\in\mathbb{R}^{r\times h}$0 is the number of “alive” variables, and a fractional coloring $E\in\mathbb{R}^{r\times h}$1. The SDP chooses $E\in\mathbb{R}^{r\times h}$2 by maximizing $E\in\mathbb{R}^{r\times h}$3, subject to coordinate-spread, blocking-subspace, spectral-independence, affine spectral-independence, and PSD constraints.

Constraint	Role
$E\in\mathbb{R}^{r\times h}$4	Coordinate-spread
$E\in\mathbb{R}^{r\times h}$5 for $E\in\mathbb{R}^{r\times h}$6	Blocking subspace
$E\in\mathbb{R}^{r\times h}$7	Spectral-independence
$E\in\mathbb{R}^{r\times h}$8	Affine spectral-independence
$E\in\mathbb{R}^{r\times h}$9	PSD feasibility

Here $EUE^\top \preceq \alpha\,\mathrm{diag}(EUE^\top).$0 are small constants satisfying

$EUE^\top \preceq \alpha\,\mathrm{diag}(EUE^\top).$1

$EUE^\top \preceq \alpha\,\mathrm{diag}(EUE^\top).$2 is the span of all currently “large” or blocked rows, the current fractional coloring $EUE^\top \preceq \alpha\,\mathrm{diag}(EUE^\top).$3, and certain “dangerous” affine rows $EUE^\top \preceq \alpha\,\mathrm{diag}(EUE^\top).$4, and $EUE^\top \preceq \alpha\,\mathrm{diag}(EUE^\top).$5 captures how many rows share a given scale. Feasibility is guaranteed whenever

$EUE^\top \preceq \alpha\,\mathrm{diag}(EUE^\top).$6

After solving the SDP, the algorithm diagonalizes $EUE^\top \preceq \alpha\,\mathrm{diag}(EUE^\top).$7, draws a random Rademacher vector $EUE^\top \preceq \alpha\,\mathrm{diag}(EUE^\top).$8, and sets

$EUE^\top \preceq \alpha\,\mathrm{diag}(EUE^\top).$9

Then

$k\geq \log^2 n$0

and the process updates

$k\geq \log^2 n$1

with $k\geq \log^2 n$2, freezing variables when $k\geq \log^2 n$3 reaches $k\geq \log^2 n$4. The key property is that the SDP constraints guarantee that the process is blocked along large or dangerous directions, that every linear combination $k\geq \log^2 n$5 has controlled variance, and that every affine combination induced by $k\geq \log^2 n$6 has controlled variance (Bansal et al., 5 Aug 2025).

3. Decoupling mechanism

The central effect of affine spectral-independence is decoupling. The paper states that the extra affine constraints effectively decouple the evolution of discrepancy across different rows and allow better control over how many rows accumulate large discrepancy at any point during the process. This is the mechanism used to improve on the $k\geq \log^2 n$7 and $k\geq \log^2 n$8 bounds associated with Banaszczyk in the Beck-Fiala and Komlós settings.

For each row $k\geq \log^2 n$9, the analysis introduces a regularized discrepancy

$\widetilde{O}(\log^{1/4}n)$0

and a row is called dangerous whenever $\widetilde{O}(\log^{1/4}n)$1, at which point it is blocked or upgraded to the next level. The increments $\widetilde{O}(\log^{1/4}n)$2 are shown to satisfy a negative drift and an almost pairwise-independence property, with the affine constraints playing the decisive role when multiple rows must be controlled jointly.

The paper then invokes a Freedman-type tail bound for processes with $\widetilde{O}(\log^{1/4}n)$3-pairwise independence. This implies that in each column at most $\widetilde{O}(\log^{1/4}n)$4 rows ever become dangerous, and that total discrepancy is $\widetilde{O}(\log^{1/4}n)$5. The stated interpretation is that any collection of rows represented by the rows of $\widetilde{O}(\log^{1/4}n)$6 sees nearly pairwise-independent updates, so their high-discrepancy events almost behave as if independent, thereby avoiding the union-bound blow up. A plausible implication is that affine spectral-independence functions as a covariance-level substitute for genuine independence in settings where many discrepancy events must be controlled simultaneously (Bansal et al., 5 Aug 2025).

4. Beck-Fiala and Komlós bounds

The primary results are discrepancy bounds beyond Banaszczyk. For the Beck-Fiala problem, the paper states that if $\widetilde{O}(\log^{1/4}n)$7 has column-sparsity $\widetilde{O}(\log^{1/4}n)$8, then

$\widetilde{O}(\log^{1/4}n)$9

This is presented as resolving the Beck-Fiala Conjecture for $O(\sqrt{\log n})$0.

A more general statement gives, for all $O(\sqrt{\log n})$1,

$O(\sqrt{\log n})$2

The abstract also states an $O(\sqrt{\log n})$3 bound for $O(\sqrt{\log n})$4, where $O(\sqrt{\log n})$5 hides $O(\sqrt{\log n})$6 factors.

For the Komlós problem, if $O(\sqrt{\log n})$7 has $O(\sqrt{\log n})$8 for each column, then

$O(\sqrt{\log n})$9

The paper presents this as improving upon the previous $h\in\mathbb{N}$0 bound. All of these results also admit efficient polynomial-time algorithms. The paper further emphasizes that the technique of “decoupling via affine spectral-independence” is quite general and may be of independent interest (Bansal et al., 5 Aug 2025).

5. Extensions to prefix discrepancy and Steinitz-type problems

A subsequent paper applies the same technique to $h\in\mathbb{N}$1 prefix discrepancy and Steinitz problems. In "Near-Optimal Constructive Bounds for $h\in\mathbb{N}$2 Prefix Discrepancy and Steinitz Problems via Affine Spectral Independence," the framework again uses a discrete Brownian motion guided by an SDP, but now the affine spectral-independence constraint is imposed on prefix-discrepancy updates. For a fixed prefix $h\in\mathbb{N}$3, with $h\in\mathbb{N}$4 denoting row $h\in\mathbb{N}$5 restricted to $h\in\mathbb{N}$6, the increment vector $h\in\mathbb{N}$7 is said to be $h\in\mathbb{N}$8-affinely spectrally independent if

$h\in\mathbb{N}$9

Equivalently,

$U\in\mathbb{R}^{h\times h}$0

The key lemma states that if each increment $U\in\mathbb{R}^{h\times h}$1 satisfies this covariance condition, then for any $U\in\mathbb{R}^{h\times h}$2, with probability at least $U\in\mathbb{R}^{h\times h}$3,

$U\in\mathbb{R}^{h\times h}$4

where $U\in\mathbb{R}^{h\times h}$5 and $U\in\mathbb{R}^{h\times h}$6. The paper combines this with a “Global Interval Tree” data structure that simultaneously controls the deviations for all prefixes while enforcing only $U\in\mathbb{R}^{h\times h}$7 affine spectral-independence constraints at once.

The main theorem in that work states that if $U\in\mathbb{R}^{h\times h}$8 with $U\in\mathbb{R}^{h\times h}$9 and $U$00, then one can in $U$01 time find $U$02 so that for every prefix $U$03,

$U$04

Under $U$05, the same assumptions yield a permutation $U$06 such that

$U$07

This suggests that affine spectral-independence is not limited to row discrepancy in sparse matrices; it can also be adapted to control entire families of prefix processes in constructive discrepancy minimization (Dutta et al., 14 Apr 2026).

6. Scope, interpretation, and distinct usage in random-matrix theory

In discrepancy theory, affine spectral-independence is a PSD-order covariance constraint used to design random walks and to decouple the evolution of many discrepancy quantities. It is therefore not a claim of literal independence of random variables; rather, it is a structured domination of covariance by diagonal variance that yields “almost pairwise-independence” behavior in the analysis.

The phrase also appears in a different sense in an expository narrative for "Spectral independence of almost fully correlated random matrices." There, two coupled random matrices are written as

$U$08

with arbitrary non-negative linear filters $U$09 and deterministic shifts $U$10. Under the decorrelation condition $U$11, the paper proves that the fluctuations of individual eigenvalues in the bulk are still asymptotically independent, and the narrative describes this as an “affine” robustness phenomenon because the decoupling survives arbitrary affine transformations of the underlying Wigner-type cores. In particular, for fixed bulk energies $U$12, the joint $U$13-point correlation function factorizes against compactly supported smooth test functions up to an $U$14 error, and Example 2.9 shows that the threshold $U$15 is necessary (Kolupaiev, 27 Mar 2025).

A common source of confusion is therefore terminological rather than mathematical. In the discrepancy papers, affine spectral-independence is an explicit matrix inequality enforced as an SDP constraint. In the random-matrix setting, the phrase refers to independence of local spectral statistics that persists under deterministic linear maps and affine shifts. The two usages share a decoupling theme, but they operate in different problem classes, with different objects and proofs.

7. Significance and possible generalizations

The discussion accompanying the discrepancy results emphasizes three features. First, the decoupling effect: any collection of rows represented by the rows of $U$16 sees nearly pairwise-independent updates, so high-discrepancy events almost behave as if independent. Second, modularity: one may choose different families $U$17, for example corresponding to different scales of row sizes or blocks of rows sharing structure, and enforce their combined evolution to remain decoupled. Third, potential generalizations: the paper explicitly points to other discrepancy problems where one must simultaneously control many linear forms, including hypergraph discrepancy and the Beck-Sós problem; to combinatorial rounding and randomized rounding in packing LPs, where affine spectral-independence could yield better tail bounds for sums of dependent rounding decisions; and to MCMC sampling and correlation decay, where the “affine” variant may help prove faster mixing for more complex spin systems (Bansal et al., 5 Aug 2025).

Within the discrepancy literature, the main conceptual contribution is that a small number of extra SDP constraints can be used to control not only variances of individual row updates but also the covariance structure of entire families of affine combinations. The paper’s stated conclusion is that this lets one decouple the fate of many discrepancy rows and break long-standing log-factor barriers in the classical Beck-Fiala and Komlós problems.