Expanded Sufficiently Scattered Condition

• Expanded Sufficiently Scattered Condition (p-SSC) is a geometric criterion requiring the factor matrix rows to fill a latent simplex, thereby ensuring a large inscribed sphere.
• It bridges classical separability and standard SSC by quantifying data spread via the parameter p, leading to explicit, finite-noise recovery guarantees.
• The robustness bounds demonstrate that lower p values and well-conditioned bases improve stability and tolerance to perturbations in minimum-volume NMF.

The expanded sufficiently scattered condition (p-SSC) is a quantitative structural property introduced to precisely characterize the geometric spread of data in minimum-volume nonnegative matrix factorization (min-vol NMF). It has emerged as a central tool for establishing the noise robustness of min-vol NMF, bridging the gap between the classical notions of separability and the standard sufficiently scattered condition (SSC). The p-SSC formulates a new rigorously defined measure of how well the rows of the factor matrix fill out the latent simplex, parameterized by $p\in [1,\sqrt{r-1})$, and enables sharp finite-noise recovery guarantees for min-vol NMF in the presence of perturbations.

1. Model Formulation and Latent Simplex Structure

Consider a data matrix $X = [x_1, \dots, x_n] \in \mathbb{R}^{m \times n}$ modeled as

$$X \approx W H^\top,\qquad W\in\mathbb{R}^{m\times r},\quad H\in\mathbb{R}^{n\times r},\quad r \ll \min\{m,n\},$$

with a simplex-structured constraint

$$H e = e,\quad H \ge 0, \quad x_i = W H(i,:)^\top \quad\Longrightarrow\quad x_i\in\operatorname{Conv}\{W(:,1),\dots,W(:,r)\},$$

where $e$ is the all-ones vector in $\mathbb{R}^{r}$. Often, data columns are normalized so that

$$\Delta^r = \left\{h \in \mathbb{R}_{\geq 0}^r \mid e^\top h = 1\right\}$$

and thus $$\operatorname{Conv}(X) \subseteq \operatorname{Conv}(W)$$. This normalization ensures every data column lies in the standard probability simplex, rendering the geometric analysis of the factorization meaningful in terms of volume and scatter.

2. Definition of the Expanded Sufficiently Scattered Condition

For $p\geq1$, define the cone

$$C_p = \left\{x\in\mathbb{R}_{\ge 0}^r \mid e^\top x \geq p\|x\|\right\}.$$

On the affine hyperplane $X = [x_1, \dots, x_n] \in \mathbb{R}^{m \times n}$0, this corresponds to the set

$X = [x_1, \dots, x_n] \in \mathbb{R}^{m \times n}$1

with $X = [x_1, \dots, x_n] \in \mathbb{R}^{m \times n}$2.

A matrix $X = [x_1, \dots, x_n] \in \mathbb{R}^{m \times n}$3 is $X = [x_1, \dots, x_n] \in \mathbb{R}^{m \times n}$4-SSC if and only if

$X = [x_1, \dots, x_n] \in \mathbb{R}^{m \times n}$5

The $X = [x_1, \dots, x_n] \in \mathbb{R}^{m \times n}$6-SSC thus demands that the convex hull of the rows of $X = [x_1, \dots, x_n] \in \mathbb{R}^{m \times n}$7 contains a Euclidean ball of radius determined by $X = [x_1, \dots, x_n] \in \mathbb{R}^{m \times n}$8, centered at the barycenter of the simplex.

Key regimes:

$X = [x_1, \dots, x_n] \in \mathbb{R}^{m \times n}$9 value	Condition	Interpretation
$$X \approx W H^\top,\qquad W\in\mathbb{R}^{m\times r},\quad H\in\mathbb{R}^{n\times r},\quad r \ll \min\{m,n\},$$0	$$X \approx W H^\top,\qquad W\in\mathbb{R}^{m\times r},\quad H\in\mathbb{R}^{n\times r},\quad r \ll \min\{m,n\},$$1	Separability
$$X \approx W H^\top,\qquad W\in\mathbb{R}^{m\times r},\quad H\in\mathbb{R}^{n\times r},\quad r \ll \min\{m,n\},$$2	Standard sufficiently scattered condition (SSC)	Inscribed sphere in $$X \approx W H^\top,\qquad W\in\mathbb{R}^{m\times r},\quad H\in\mathbb{R}^{n\times r},\quad r \ll \min\{m,n\},$$3
$$X \approx W H^\top,\qquad W\in\mathbb{R}^{m\times r},\quad H\in\mathbb{R}^{n\times r},\quad r \ll \min\{m,n\},$$4	Expanded SSC (stronger than SSC)	Excludes near-flat configurations

3. Geometric Interpretation and Comparison to Classical SSC

Classical SSC ($$X \approx W H^\top,\qquad W\in\mathbb{R}^{m\times r},\quad H\in\mathbb{R}^{n\times r},\quad r \ll \min\{m,n\},$$5) requires the largest possible inscribed sphere of $$X \approx W H^\top,\qquad W\in\mathbb{R}^{m\times r},\quad H\in\mathbb{R}^{n\times r},\quad r \ll \min\{m,n\},$$6 to be contained in $$X \approx W H^\top,\qquad W\in\mathbb{R}^{m\times r},\quad H\in\mathbb{R}^{n\times r},\quad r \ll \min\{m,n\},$$7, forcing the $$X \approx W H^\top,\qquad W\in\mathbb{R}^{m\times r},\quad H\in\mathbb{R}^{n\times r},\quad r \ll \min\{m,n\},$$8 to occupy positions relatively far from simplex faces. In contrast, the p-SSC for $$X \approx W H^\top,\qquad W\in\mathbb{R}^{m\times r},\quad H\in\mathbb{R}^{n\times r},\quad r \ll \min\{m,n\},$$9 tightens this by requiring that a larger sphere $$H e = e,\quad H \ge 0, \quad x_i = W H(i,:)^\top \quad\Longrightarrow\quad x_i\in\operatorname{Conv}\{W(:,1),\dots,W(:,r)\},$$0 is contained. Geometrically, p-SSC rules out degenerate configurations where data points are nearly affine-dependent or clustered near faces of the simplex, which degrade identifiability.

As $$H e = e,\quad H \ge 0, \quad x_i = W H(i,:)^\top \quad\Longrightarrow\quad x_i\in\operatorname{Conv}\{W(:,1),\dots,W(:,r)\},$$1, the region allowed by the condition shrinks, and the tolerance for noise diminishes. As $$H e = e,\quad H \ge 0, \quad x_i = W H(i,:)^\top \quad\Longrightarrow\quad x_i\in\operatorname{Conv}\{W(:,1),\dots,W(:,r)\},$$2, one approaches separability, admitting maximal robustness.

4. Robustness Guarantees for Minimum-Volume NMF under p-SSC

Given the approximate min-vol NMF formulation: $$H e = e,\quad H \ge 0, \quad x_i = W H(i,:)^\top \quad\Longrightarrow\quad x_i\in\operatorname{Conv}\{W(:,1),\dots,W(:,r)\},$$3 where $$H e = e,\quad H \ge 0, \quad x_i = W H(i,:)^\top \quad\Longrightarrow\quad x_i\in\operatorname{Conv}\{W(:,1),\dots,W(:,r)\},$$4, and the generative model

$$H e = e,\quad H \ge 0, \quad x_i = W H(i,:)^\top \quad\Longrightarrow\quad x_i\in\operatorname{Conv}\{W(:,1),\dots,W(:,r)\},$$5

the main recovery theorem asserts—letting $$H e = e,\quad H \ge 0, \quad x_i = W H(i,:)^\top \quad\Longrightarrow\quad x_i\in\operatorname{Conv}\{W(:,1),\dots,W(:,r)\},$$6—that there exist absolute constants $$H e = e,\quad H \ge 0, \quad x_i = W H(i,:)^\top \quad\Longrightarrow\quad x_i\in\operatorname{Conv}\{W(:,1),\dots,W(:,r)\},$$7 such that if

$$H e = e,\quad H \ge 0, \quad x_i = W H(i,:)^\top \quad\Longrightarrow\quad x_i\in\operatorname{Conv}\{W(:,1),\dots,W(:,r)\},$$8

then for any optimizer $$H e = e,\quad H \ge 0, \quad x_i = W H(i,:)^\top \quad\Longrightarrow\quad x_i\in\operatorname{Conv}\{W(:,1),\dots,W(:,r)\},$$9, the factor recovery error obeys

$e$0

where $e$1 runs over all $e$2 permutation matrices, and $e$3 is the smallest nonzero singular value of $e$4. In the near-separable regime $e$5, this simplifies to

$e$6

The bounds underscore that robustness depends critically on $e$7, $e$8, the conditioning of $e$9, and the noise bound $\mathbb{R}^{r}$0. In particular, tightness of the bound on $\mathbb{R}^{r}$1 diminishes rapidly as $\mathbb{R}^{r}$2 approaches $\mathbb{R}^{r}$3 (i.e., standard SSC), and is strongest in the near-separable regime ($\mathbb{R}^{r}$4). The result establishes—quantitatively and for the first time—that the expanded sufficiently scattered condition yields provable stability for min-vol NMF under explicit noise models.

5. Principal Geometric Lemmas

A suite of geometric lemmas underpins the analysis of p-SSC:

• Dual cone: For a cone $\mathbb{R}^{r}$5, the dual is $\mathbb{R}^{r}$6.
• Geometry of $\mathbb{R}^{r}$7 and associated cones: $\mathbb{R}^{r}$8, and $\mathbb{R}^{r}$9, with $$\Delta^r = \left\{h \in \mathbb{R}_{\geq 0}^r \mid e^\top h = 1\right\}$$0 and $$\Delta^r = \left\{h \in \mathbb{R}_{\geq 0}^r \mid e^\top h = 1\right\}$$1.
• Auxiliary simplex and necessary convex hull: Any $$\Delta^r = \left\{h \in \mathbb{R}_{\geq 0}^r \mid e^\top h = 1\right\}$$2-SSC $$\Delta^r = \left\{h \in \mathbb{R}_{\geq 0}^r \mid e^\top h = 1\right\}$$3 must satisfy $$\Delta^r = \left\{h \in \mathbb{R}_{\geq 0}^r \mid e^\top h = 1\right\}$$4 for $$\Delta^r = \left\{h \in \mathbb{R}_{\geq 0}^r \mid e^\top h = 1\right\}$$5, with $$\Delta^r = \left\{h \in \mathbb{R}_{\geq 0}^r \mid e^\top h = 1\right\}$$6.
• Lower bound on determinant of linking map: There exists $$\Delta^r = \left\{h \in \mathbb{R}_{\geq 0}^r \mid e^\top h = 1\right\}$$7 with $$\Delta^r = \left\{h \in \mathbb{R}_{\geq 0}^r \mid e^\top h = 1\right\}$$8 and $$\Delta^r = \left\{h \in \mathbb{R}_{\geq 0}^r \mid e^\top h = 1\right\}$$9.
• Approximate orthogonality: Writing $$\operatorname{Conv}(X) \subseteq \operatorname{Conv}(W)$$0 for the $$\operatorname{Conv}(X) \subseteq \operatorname{Conv}(W)$$1th row of $$\operatorname{Conv}(X) \subseteq \operatorname{Conv}(W)$$2, $$\operatorname{Conv}(X) \subseteq \operatorname{Conv}(W)$$3, $$\operatorname{Conv}(X) \subseteq \operatorname{Conv}(W)$$4, $$\operatorname{Conv}(X) \subseteq \operatorname{Conv}(W)$$5 up to $$\operatorname{Conv}(X) \subseteq \operatorname{Conv}(W)$$6.
• Geometric localization: Each $$\operatorname{Conv}(X) \subseteq \operatorname{Conv}(W)$$7 lies in a small "ice-cream" sector around a canonical basis vector $$\operatorname{Conv}(X) \subseteq \operatorname{Conv}(W)$$8, with angular radius $$\operatorname{Conv}(X) \subseteq \operatorname{Conv}(W)$$9; distinct rows cannot coincide at the same vertex.

These elements facilitate the structural identification and stability analysis under the p-SSC framework.

6. Parametric Dependencies and Regime Analysis

The roles of the principal parameters are as follows:

• $p\geq1$0 (noise magnitude): Smaller $p\geq1$1 is required for successful recovery as $p\geq1$2 increases; robustness is sharply diminished for larger $p\geq1$3, i.e., as the sphere inscribed in the simplex shrinks.
• $p\geq1$4: Encodes the "well spread" property of ground-truth rows of $p\geq1$5; regimes interpolate between separability ($p\geq1$6), which is maximally robust, and the classical SSC ($p\geq1$7), which admits no fixed noise tolerance.
• $p\geq1$8: The “dual slack” parameter appearing in bounds; smaller $p\geq1$9 (i.e., $$C_p = \left\{x\in\mathbb{R}_{\ge 0}^r \mid e^\top x \geq p\|x\|\right\}.$$0 closer to $$C_p = \left\{x\in\mathbb{R}_{\ge 0}^r \mid e^\top x \geq p\|x\|\right\}.$$1) both tightens the allowable $$C_p = \left\{x\in\mathbb{R}_{\ge 0}^r \mid e^\top x \geq p\|x\|\right\}.$$2 and worsens the final error.
• $$C_p = \left\{x\in\mathbb{R}_{\ge 0}^r \mid e^\top x \geq p\|x\|\right\}.$$3, $$C_p = \left\{x\in\mathbb{R}_{\ge 0}^r \mid e^\top x \geq p\|x\|\right\}.$$4: Reflect the conditioning of the true basis matrix; poor conditioning directly increases sensitivity to noise.

A plausible implication is that, for practical applications requiring noise robustness in min-vol NMF, it is advantageous for data to admit representations with $$C_p = \left\{x\in\mathbb{R}_{\ge 0}^r \mid e^\top x \geq p\|x\|\right\}.$$5 as close to $$C_p = \left\{x\in\mathbb{R}_{\ge 0}^r \mid e^\top x \geq p\|x\|\right\}.$$6 as possible, and for $$C_p = \left\{x\in\mathbb{R}_{\ge 0}^r \mid e^\top x \geq p\|x\|\right\}.$$7 to be well-conditioned.

7. Summary and Implications

The expanded sufficiently scattered condition (p-SSC) provides a continuum of structural regimes interpolating between separability and the classical sufficiently scattered condition. By enforcing a suitably large inscribed sphere in the latent simplex, p-SSC enables explicit, quantitative guarantees for the identifiability and robustness of minimum-volume NMF under bounded perturbations. The derived bounds demonstrate that when data are sufficiently well-scattered (with smaller $$C_p = \left\{x\in\mathbb{R}_{\ge 0}^r \mid e^\top x \geq p\|x\|\right\}.$$8), min-vol NMF becomes provably stable, while for data satisfying only the classical SSC, robustness rapidly deteriorates. The introduction of p-SSC thus offers a principled geometric criterion for algorithmic design and analysis in applications of NMF subject to noise, enhancing the theoretical understanding and guiding practical data preconditioning.