Zonoid Sparsification for ℓ1 Geometry

• Zonoid sparsification is the problem of approximating centrally symmetric convex bodies by zonotopes with few segments, preserving the ℓ1 support function within a (1±ε) factor.
• The method achieves a linear-size sparsifier using O(n/ε² log(1/ε)) segments, improving Talagrand’s earlier O(n/ε² log n) bound for efficient high-dimensional approximations.
• The approach combines convex geometry, reweighting of zonotope generators, and techniques like Minkowski subtraction and discrepancy methods to control approximation error.

Zonoid sparsification is the approximation problem in which a centrally symmetric convex body represented as a zonoid or zonotope is replaced by a zonotope with a controlled number of segments while preserving its support function, or equivalently preserving the body up to multiplicative set inclusion. In the formulation developed for $\ell_1$ geometry, if $Z\subseteq \mathbb R^n$ is a zonotope and $\varepsilon\in(0,1/2]$, one seeks a zonotope $Z'$ with few generators such that $$(1-\varepsilon)Z\subseteq Z'\subseteq (1+\varepsilon)Z$$. The strongest theorem in the supplied literature proves that every $n$-dimensional zonotope admits such an approximation with $$O\!\left(\frac{n}{\varepsilon^2}\log\frac1\varepsilon\right)$$ segments, improving the previous $O\!\left(\frac{n}{\varepsilon^2}\log n\right)$ bound of Talagrand (1990) (Reis et al., 26 Jun 2026). The same result is exactly equivalent to linear-size $\ell_1$ row sparsification, because the support function of the zonotope generated by the rows of a matrix $A$ is $Z\subseteq \mathbb R^n$0 (Reis et al., 26 Jun 2026).

1. Geometric and analytic formulation

Let $Z\subseteq \mathbb R^n$1. The zonotope generated by these vectors is

$Z\subseteq \mathbb R^n$2

If $Z\subseteq \mathbb R^n$3 has rows $Z\subseteq \mathbb R^n$4, then

$Z\subseteq \mathbb R^n$5

For any convex body $Z\subseteq \mathbb R^n$6, the support function is

$Z\subseteq \mathbb R^n$7

For the zonotope $Z\subseteq \mathbb R^n$8,

$Z\subseteq \mathbb R^n$9

Hence, when $\varepsilon\in(0,1/2]$0 has rows $\varepsilon\in(0,1/2]$1,

$\varepsilon\in(0,1/2]$2

This identifies the matrix and convex-geometric viewpoints. Rows of $\varepsilon\in(0,1/2]$3 correspond to zonotope generators, $\varepsilon\in(0,1/2]$4 corresponds to the support function $\varepsilon\in(0,1/2]$5, and diagonal reweighting

$\varepsilon\in(0,1/2]$6

corresponds to replacing each segment $\varepsilon\in(0,1/2]$7 by $\varepsilon\in(0,1/2]$8. Writing

$\varepsilon\in(0,1/2]$9

one has

$Z'$0

For centrally symmetric convex bodies, support-function domination is equivalent to set inclusion: $Z'$1 Therefore

$Z'$2

is equivalent to

$Z'$3

This equivalence is the basic dictionary of zonoid sparsification in the $Z'$4 setting (Reis et al., 26 Jun 2026).

2. Main sparsification theorem

The central theorem states that for every matrix $Z'$5 and every $Z'$6, there exists a diagonal matrix

$Z'$7

with at most

$Z'$8

nonzero diagonal entries such that

$Z'$9

Equivalently, one can select and reweight only

$$(1-\varepsilon)Z\subseteq Z'\subseteq (1+\varepsilon)Z$$0

rows of $$(1-\varepsilon)Z\subseteq Z'\subseteq (1+\varepsilon)Z$$1 while preserving the $$(1-\varepsilon)Z\subseteq Z'\subseteq (1+\varepsilon)Z$$2-norm of $$(1-\varepsilon)Z\subseteq Z'\subseteq (1+\varepsilon)Z$$3 for every $$(1-\varepsilon)Z\subseteq Z'\subseteq (1+\varepsilon)Z$$4 (Reis et al., 26 Jun 2026).

In geometric form, for any zonotope $$(1-\varepsilon)Z\subseteq Z'\subseteq (1+\varepsilon)Z$$5 and any $$(1-\varepsilon)Z\subseteq Z'\subseteq (1+\varepsilon)Z$$6, there exists a zonotope $$(1-\varepsilon)Z\subseteq Z'\subseteq (1+\varepsilon)Z$$7 generated by at most

$$(1-\varepsilon)Z\subseteq Z'\subseteq (1+\varepsilon)Z$$8

segments such that

$$(1-\varepsilon)Z\subseteq Z'\subseteq (1+\varepsilon)Z$$9

A more explicit weighted form keeps the original generator directions. If $n$0 is generated by the rows of $n$1, then there exists

$n$2

with

$n$3

such that

$n$4

where

$n$5

Thus sparsification is not a change of ambient dimension or a change of generator directions. Rows $n$6 are the original segment directions, the weights $n$7 or diagonal entries $n$8 are the new segment lengths, zero weights delete generators, and nonzero weights reweight the retained generators. The same theorem also has a Banach-space form: if $n$9 is an $$O\!\left(\frac{n}{\varepsilon^2}\log\frac1\varepsilon\right)$$0-dimensional subspace of $$O\!\left(\frac{n}{\varepsilon^2}\log\frac1\varepsilon\right)$$1, then there exists an $$O\!\left(\frac{n}{\varepsilon^2}\log\frac1\varepsilon\right)$$2-dimensional subspace $$O\!\left(\frac{n}{\varepsilon^2}\log\frac1\varepsilon\right)$$3 with

$$O\!\left(\frac{n}{\varepsilon^2}\log\frac1\varepsilon\right)$$4

such that

$$O\!\left(\frac{n}{\varepsilon^2}\log\frac1\varepsilon\right)$$5

(Reis et al., 26 Jun 2026).

3. Zonoids, zonotopes, and representation theory

A zonoid is typically a Hausdorff limit of zonotopes; equivalently, it is a centrally symmetric convex body whose support function has an integral representation

$$O\!\left(\frac{n}{\varepsilon^2}\log\frac1\varepsilon\right)$$6

for a finite even measure $$O\!\left(\frac{n}{\varepsilon^2}\log\frac1\varepsilon\right)$$7. The paper establishing the linear-size bound is explicit about scope: its theorem is stated and proved directly for zonotopes, not for arbitrary zonoids. For general zonoids, one obtains an existential consequence by standard approximation, because zonoids are Hausdorff limits of zonotopes; however, the paper does not state a separate formal theorem for arbitrary zonoids (Reis et al., 26 Jun 2026).

The broader zonoid literature gives several equivalent representations that make sparsification natural. For centered zonoids $$O\!\left(\frac{n}{\varepsilon^2}\log\frac1\varepsilon\right)$$8, one has the support-function formula

$$O\!\left(\frac{n}{\varepsilon^2}\log\frac1\varepsilon\right)$$9

with $O\!\left(\frac{n}{\varepsilon^2}\log n\right)$0 a unique even finite measure on the sphere (Mathis et al., 2022). For centrally symmetric zonoids, another representation is

$O\!\left(\frac{n}{\varepsilon^2}\log n\right)$1

where $O\!\left(\frac{n}{\varepsilon^2}\log n\right)$2 is an integrable random vector, and

$O\!\left(\frac{n}{\varepsilon^2}\log n\right)$3

If $O\!\left(\frac{n}{\varepsilon^2}\log n\right)$4 has finite support, the resulting body is a finite zonotope; conversely, finite zonotopes arise from finitely supported laws. Positive measures on projective space correspond to zonoids via the cosine transform, and on the positive cone this correspondence is a homeomorphism (Breiding et al., 2021).

These representations locate sparsification at the level of measure discretization or distribution discretization. A plausible implication is that zonoid sparsification can be viewed as replacing a continuous or complicated generating measure by an atomic one while controlling the support function. The literature also makes clear that symmetry is essential in the current theory: zonotopes of the form $O\!\left(\frac{n}{\varepsilon^2}\log n\right)$5 are symmetric, the support function is an $O\!\left(\frac{n}{\varepsilon^2}\log n\right)$6-type sum of absolute values, and the final approximation takes the form $O\!\left(\frac{n}{\varepsilon^2}\log n\right)$7. No non-symmetric analogue is developed in the theorem of (Reis et al., 26 Jun 2026).

4. Proof architecture and algorithmic status

The proof of the linear-size theorem is not based on Lewis weights, leverage scores, effective resistances, standard row-sampling arguments, or matrix Chernoff or matrix Bernstein concentration. Instead, it combines convex geometry, Minkowski subtraction, volume inequalities, separate convexity, a discrepancy or fractional-coloring step, and iteration (Reis et al., 26 Jun 2026).

For $O\!\left(\frac{n}{\varepsilon^2}\log n\right)$8, define

$O\!\left(\frac{n}{\varepsilon^2}\log n\right)$9

The proof studies vectors $\ell_1$0 through the modified zonotope $\ell_1$1. If $\ell_1$2, the $\ell_1$3-th segment disappears; if $\ell_1$4, that segment is shrunk or enlarged. The key volumetric tool is Minkowski subtraction,

$\ell_1$5

together with the identity

$\ell_1$6

A core lemma shows that for a zonotope $\ell_1$7, the function

$\ell_1$8

is separately convex in the coordinates $\ell_1$9. Jensen’s inequality for separately convex functions with independent coordinates then yields a random inclusion theorem: $A$0

When this is specialized to $A$1 and $A$2, one obtains

$A$3

This is the origin of the $A$4 factor. The proof then considers the convex feasible set

$A$5

shows that it contains $A$6, and proves that all coordinate sections have large relative volume. To obtain two-sided approximation, one needs the symmetrizer $A$7, and the main convex-geometric theorem shows that under the appropriate lower bound on $A$8, this symmetric feasible set still has exponentially large volume.

At that point a discrepancy-style fractional-coloring theorem applies: for every $A$9, there exists $Z\subseteq \mathbb R^n$00 such that if $Z\subseteq \mathbb R^n$01 is symmetric convex with $Z\subseteq \mathbb R^n$02, then there exists

$Z\subseteq \mathbb R^n$03

with at least $Z\subseteq \mathbb R^n$04 coordinates satisfying $Z\subseteq \mathbb R^n$05. After a sign choice, at least $Z\subseteq \mathbb R^n$06 coordinates equal $Z\subseteq \mathbb R^n$07, so an iterative update deletes at least a quarter of the remaining generators while incurring multiplicative distortion $Z\subseteq \mathbb R^n$08 in that round. Repetition with geometrically changing error parameters yields the final support bound and the total distortion bounds

$Z\subseteq \mathbb R^n$09

The result is existential rather than polynomial-time constructive. The diagonal matrix $Z\subseteq \mathbb R^n$10 can be computed in time

$Z\subseteq \mathbb R^n$11

equivalently the sparse zonotope can be computed in randomized time $Z\subseteq \mathbb R^n$12 times a polynomial factor. The bottleneck is membership or separation for the intermediate convex body $Z\subseteq \mathbb R^n$13, which amounts to repeated testing of zonotope inclusions $Z\subseteq \mathbb R^n$14. The paper notes that this inclusion problem is $Z\subseteq \mathbb R^n$15-complete in general when both zonotopes are given by generators (Reis et al., 26 Jun 2026).

5. Historical development and quantitative significance

The quantitative history recorded in the literature is a progression in the number of segments sufficient to approximate an $Z\subseteq \mathbb R^n$16-dimensional zonotope within factor $Z\subseteq \mathbb R^n$17 (Reis et al., 26 Jun 2026).

Work	Segment bound
Schechtman (1987)	$Z\subseteq \mathbb R^n$18
Bourgain–Lindenstrauss–Milman (1989)	$Z\subseteq \mathbb R^n$19
Talagrand (1990)	$Z\subseteq \mathbb R^n$20
"Linear-size $Z\subseteq \mathbb R^n$21 sparsifiers"	$Z\subseteq \mathbb R^n$22

The improvement over Talagrand replaces the logarithmic factor $Z\subseteq \mathbb R^n$23 by $Z\subseteq \mathbb R^n$24. If $Z\subseteq \mathbb R^n$25 is large and $Z\subseteq \mathbb R^n$26 is fixed, $Z\subseteq \mathbb R^n$27 is constant while $Z\subseteq \mathbb R^n$28 grows. The result is therefore stronger quantitatively, especially in high dimension with fixed accuracy. The paper also states that it answers a question of Schechtman from 1986/1987 affirmatively (Reis et al., 26 Jun 2026).

The advance is primarily quantitative. The asymptotic number of segments or nonzeros is reduced from

$Z\subseteq \mathbb R^n$29

For fixed $Z\subseteq \mathbb R^n$30, the paper notes that this is also a qualitative statement in the sense of obtaining truly linear-size $Z\subseteq \mathbb R^n$31 sparsification, up to a constant depending on $Z\subseteq \mathbb R^n$32. At the same time, the improvement is not accompanied by a polynomial-time algorithm for the $Z\subseteq \mathbb R^n$33 or zonotope case (Reis et al., 26 Jun 2026).

6. Related zonoid frameworks and broader context

Two related arXiv lines place zonoid sparsification in a wider framework without themselves proving a cardinality theorem. The paper "The zonoid algebra, generalized mixed volumes, and random determinants" develops a representation theory in which every centrally symmetric zonoid is an expectation of symmetric segments, zonotopes correspond to finitely supported data, and empirical finite zonotopes converge almost surely to the target zonoid in the Hausdorff sense. It also proves that multilinear maps on vector spaces induce continuous Minkowski-multilinear maps on zonoids, yielding the zonoid algebra and formulas such as

$Z\subseteq \mathbb R^n$34

These results do not give a sparsifier with complexity bounds, but they identify support functions, length, intrinsic volumes, mixed volumes, and determinant expectations as natural invariants under approximation (Breiding et al., 2021).

The paper "Expectation of a random submanifold: the zonoid section" introduces a pointwise zonoid

$Z\subseteq \mathbb R^n$35

attached to a random zero set. In that framework, the first intrinsic volume of $Z\subseteq \mathbb R^n$36 is the Kac–Rice density, its center computes the expected current, wedge products correspond to intersections, and pull-backs correspond to preimages. The paper is explicit that it does not supply cardinality bounds, algorithmic constructions, or deterministic or randomized sparsifiers with guarantees. However, it shows that zonoids can act as compressed sufficient statistics for large classes of geometric expectations. In a random level-set example,

$Z\subseteq \mathbb R^n$37

so the zonoid section is literally a single segment, i.e. a maximally sparse representation (Mathis et al., 2022).

Taken together, these works distinguish three levels of theory. First, zonoids admit integral, measure, and random-vector representations. Second, those representations support multilinear operations and continuous functionals relevant to convex geometry and probability. Third, the 2026 $Z\subseteq \mathbb R^n$38 theorem converts that structural background into a concrete sparsification statement with

$Z\subseteq \mathbb R^n$39

generators for zonotopes. A plausible implication is that future zonoid sparsification results will continue to combine measure discretization, support-function control, and multilinear stability, but the formal breakthrough currently established in the supplied literature is the zonotope theorem of (Reis et al., 26 Jun 2026).