---
title: Minimal Information-Carrying Subgraphs (MICS)
url: https://www.emergentmind.com/topics/minimal-information-carrying-subgraphs-mics
type: topic
---

# Minimal Information-Carrying Subgraphs (MICS)

Searching arXiv for the cited MICS-related papers to ground the article.
Minimal Information-Carrying Subgraphs (MICS) are inclusion-minimal connected subgraphs whose vertices collectively contain the complete information needed for reconstruction. In the formal framework of distributed secret storage introduced in "Robust secret storage in networks" [2606.30261], information is split into symbols and placed on vertices of a graph, and the secret is recoverable if some surviving connected subgraph still contains all symbols. MICS provide a reduced description of the reconstruction events relevant to survivability: instead of enumerating every connected subgraph that could carry the entire secret, it is enough to enumerate the minimal ones, because every larger information-carrying connected subgraph contains at least one MICS.

## 1. Definition and formal setting

In the robust secret-storage framework, a MICS is defined as “the connected subgraph that contains the whole information and that does not have proper subgraphs that also contain the whole information.” The appendix gives the equivalent formulation “inclusion-minimal connected subgraphs whose vertices collectively contain the complete information” [2606.30261].

The setting is a graph \(\Gamma\) whose vertices store symbols according to a placement \(\chi\). A connected subgraph \(\tilde{\Gamma}'\) is information-carrying if its vertices together contain all symbols of the secret. It is a MICS precisely when no proper connected subgraph of \(\tilde{\Gamma}'\) also contains the whole information. This makes a MICS the minimal reconstruction unit for survivability.

The minimality notion is inclusion-minimality among connected, complete-information subgraphs. Two MICS may overlap, but neither can be a proper subgraph of the other. This distinguishes MICS from arbitrary witness subgraphs: they are not merely sufficient for reconstruction, but minimal under connected-subgraph inclusion.

A direct consequence of the definition is the reduction principle used throughout the paper. If a MICS survives, the secret survives. Conversely, if any larger information-carrying connected subgraph survives, then some MICS inside it also survives. This is why MICS are sufficient to represent the full reconstruction event exactly.

## 2. Derivation and combinatorial characterization

The paper derives the MICS family by progressively checking connected subgraphs in increasing size [2606.30261]. The procedure is described as follows:

1. Start with subgraphs of size 1.  
2. Count those that already contain all information.  
3. Then, for larger subgraphs, ignore any that contain a smaller already-found complete-information subgraph.

This pruning defines the minimal family \(\tilde{\Gamma}'\). In the appendix, the set of all MICS is formalized as \(\widetilde{\mathcal G}\). For each MICS \(\widetilde{\Gamma}'\), the associated survival event is

\[
E_{\widetilde{\Gamma}'} = \left\{ V(\widetilde{\Gamma}')\subseteq V_S \right\},
\]

meaning that all vertices of that MICS survive the node-failure process.

This characterization is combinatorial rather than metric: MICS are defined by connectivity, information completeness, and exclusion of proper connected complete-information subgraphs. The resulting object is minimal with respect to the reconstruction property, not with respect to size alone. A MICS may therefore have more vertices than another MICS in the same graph, provided that each is minimal relative to its own information distribution and connectivity constraints.

This suggests that MICS capture a graph-dependent interaction between encoding and topology. The same symbol placement on a different graph, or a different placement on the same graph, can alter the set \(\widetilde{\mathcal G}\) even when the number of symbols is unchanged.

## 3. Exact representation of survivability

The central theorem-level role of MICS is that they give an exact representation of survivability in distributed secret storage [2606.30261]. Survivability \(\mathcal S(p,\chi,\Gamma)\) is the probability that after random node deletion there exists at least one surviving connected subgraph carrying the entire secret. Rather than summing over all such subgraphs, the paper proves that

\[
\mathcal S(p,\chi,\Gamma) = P\left( \bigcup_{\widetilde{\Gamma}'\in\widetilde{\mathcal G}} E_{\widetilde{\Gamma}'} \right).
\]

Thus the full event “the secret survives” is exactly the union of MICS survival events.

The inclusion-exclusion principle then yields an exact expansion:

\[
P \left( \bigcup_{\widetilde{\Gamma}'\in\widetilde{\mathcal G}} E_{\widetilde{\Gamma}'} \right) = \sum_{\emptyset\neq \mathcal A\subseteq \widetilde{\mathcal G}} (-1)^{|\mathcal A|+1} P \left( \bigcap_{\widetilde{\Gamma}'\in\mathcal A} E_{\widetilde{\Gamma}'} \right).
\]

For a selected family \(\mathcal A\),

\[
P \left( \bigcap_{\widetilde{\Gamma}'\in\mathcal A} E_{\widetilde{\Gamma}'} \right) = \bar p^{ \left| \bigcup_{\widetilde{\Gamma}'\in\mathcal A} V(\widetilde{\Gamma}') \right| }.
\]

Equivalently, survivability can be written as a polynomial-like expansion

\[
\mathcal S(p,\chi,\Gamma) = \sum_{r=1}^{|\Gamma|} a_r(\chi,\Gamma)\,\bar p^r,
\]

with coefficients

\[
a_r(\chi,\Gamma) = \sum_{\ell=1}^{M} (-1)^{\ell+1} \sum_{\substack{ 1\leq j_1<\cdots<j_\ell\leq M\\ |\tilde{\Gamma}'_{j_1}\cup\cdots\cup \tilde{\Gamma}'_{j_\ell}|=r }} 1.
\]

Here the \(\tilde{\Gamma}'_j\) are the MICS. The coefficients are determined entirely by cardinalities of unions of MICS, so the exact survivability polynomial is built from overlap structure among minimal reconstruction units.

The reduction is mathematically significant because it replaces a potentially much larger family of all reconstruction-capable connected subgraphs with a minimal generating family under union of survival events. A larger complete-information connected subgraph is redundant for survivability accounting once the MICS it contains are known.

## 4. Robustness functional and semi-local optimization

MICS enter the broader optimization problem through the survivability term in the robustness functional [2606.30261]. The paper defines

\[
\mathcal{F}(\alpha,\mathcal{S},\mathcal{H})=\alpha\mathcal{S}+(1-\alpha)(1-\mathcal{H}),
\]

and equivalently

\[
\mathcal{F}=P(I|\alpha,p,q,A,\chi),
\]

where \(I\) is the event that the information both survives failure and is not hacked. In this formulation, \(\mathcal S\) depends exactly on MICS, whereas \(\mathcal H\) captures resistance to adversarial compromise. MICS therefore determine the survivability side of the tradeoff exactly.

The paper uses the MICS representation to build semi-local approximations motivated by the observation that large random-failure components are typically small, so only MICS up to a limited size or within a finite-radius neighborhood are retained. This yields approximations such as

\[
\mathcal S_{R1}(p,\chi,\Gamma) = 1-\prod_{v\in V}\left(1-P_v^{(R)}\right),
\]

and the more refined

\[
\mathcal S_{R2}(p,\chi,\Gamma) = \sum_{\emptyset\neq \mathcal C\subseteq V} \sum_{\substack{ (\mathcal A_v)_{v\in\mathcal C}\\ \mathcal A_v\subseteq \widetilde{\mathcal G}_v^{(R)}  }} (-1)^{1+\sum_{v\in\mathcal C}|\mathcal A_v|} \, \bar p^{ \left| \bigcup_{v\in\mathcal C} \Gamma'_{\mathcal A_v} \right| }.
\]

Here \(\widetilde{\mathcal G}_v^{(R)}\) is the family of MICS rooted at \(v\) inside radius \(R\).

The resulting optimization strategy supports local or semi-local optimization, simulated annealing / max-sum message passing heuristics, and evaluation without global knowledge of the entire network. The paper explicitly states that MICS-based computations are useful because they allow optimization of robustness using only neighborhood information rather than the whole graph. This suggests that MICS function simultaneously as an exact combinatorial representation and as the basis of locality-preserving approximations.

The abstract further states that, in a limiting case, the robustness functional can be mapped naturally to an effective spin Hamiltonian. In that mapping, MICS contribute indirectly through the exact structure of \(\mathcal S\).

## 5. Examples, limiting cases, and relation to tropical subgraphs

The paper gives an explicit example in which the MICS are listed as

- \(\{2,6\}\)
- \(\{6,8\}\)
- \(\{1,2,3\}\)
- \(\{2,3,4\}\)
- \(\{5,6,9\}\)
- \(\{6,7,9\}\)

These are the minimal connected subgraphs that each contain all four symbols in the example configuration [2606.30261]. The exact survivability polynomial for that case is

\[
S=2p^2+3p^3-8p^4+5p^5-p^6.
\]

This example exhibits the intended role of MICS directly: the coefficients of the survivability polynomial are induced by the overlap structure of the listed minimal complete-information subgraphs.

The paper also connects MICS to the graph-theoretic notion of connected tropical subgraphs in vertex-colored graphs. If each vertex stores exactly one symbol, then symbols play the role of colors; a connected subgraph containing all colors is a tropical subgraph; and the MICS are the inclusion-minimal tropical connected sets. The storage model considered in the paper is more general because vertices may store multiple symbols, so MICS extend tropical connected sets to a multi-symbol storage framework.

An extreme limiting case clarifies the contrast between universal replication and minimal reconstruction. When all information is copied everywhere,

\[
\mathcal{S}(p,\chi_s)=1-p^{|\Gamma|}.
\]

The paper presents this as the opposite of minimality: the secret is everywhere, so every surviving vertex set containing at least one vertex may preserve it. By contrast, MICS isolate the smallest necessary connected carriers for exact reconstruction analysis.

A plausible implication is that MICS interpolate between two extremes of redundancy design: full replication, where minimality is largely absent, and highly distributed encodings, where survivability is concentrated in a structured family of minimal connected carriers.

## 6. Broader uses of the minimal-subgraph perspective

Although the term MICS is introduced in the context of robust secret storage, the supplied literature shows that the same minimal-subgraph perspective appears in several adjacent research areas. These uses should be distinguished carefully from the exact definition in distributed storage.

In "Minimal induced subgraphs of two classes of 2-connected non-Hamiltonian graphs" [2108.13558], the core objects are HC-obstructions: 2-connected graphs with no Hamiltonian cycle such that every induced subgraph is either the graph itself, not 2-connected, or has a Hamiltonian cycle. Within split graphs, the minimal obstructions are exactly the snare and the \(n\)-novae for \(n\ge 2\); within triangle-free graphs, they are exactly the thetas, triangle-free closed thetas, and triangle-free wheels. The paper explicitly frames these as minimal witnesses of non-Hamiltonicity under induced-subgraph inclusion. Viewed through the lens of MICS, this is a minimal-information-carrying perspective on obstruction structure, but the formal object there is an induced-subgraph obstruction rather than a reconstruction subgraph.

In "Minimal forbidden induced subgraphs of graphs of bounded clique-width and bounded linear clique-width" [1306.2114], the same perspective appears for width obstructions. The paper studies families such as \(Z_k\), \(S_k\), \(S_k^+\), \(M_2^{\pm}\), and \(M_{k,1,l}\), and shows that many of them remain minimal forbidden induced subgraphs in the class of all graphs. The recurring mechanism is that the whole graph has width at least \(k+2\) or at least \(4\), while every proper induced subgraph has smaller width. This is again a minimal witness phenomenon, but the relevant property is clique-width or linear clique-width rather than secret reconstructability.

A different extension appears in GraphRAG. "Retrieving Minimal and Sufficient Reasoning Subgraphs with Graph Foundation Models for Path-aware GraphRAG" [2603.07179] does not use the term MICS explicitly, but it introduces a query-conditioned subgraph selector optimized by an Information Bottleneck objective to identify a subgraph that is informationally sufficient and structurally minimal. The ideal objective is

\[
\mathcal{L}_{\text{IB}} = -I(y;\mathcal{G}_{\mathbf{q}}) + \beta I(\mathcal{G};\mathcal{G}_{\mathbf{q}}),
\]

with the operational label-free proxy

\[
\mathcal{L}_{\text{LIB}} = -I(\mathbf{q};\mathcal{G}_{\mathbf{q}}) + \beta I(\mathcal{G};\mathcal{G}_{\mathbf{q}}).
\]

The selected subgraph is explicitly described as containing informationally sufficient and structurally minimal golden evidence in a self-contained “core set.” This is closely aligned with the MICS idea, but it remains an approximate, learned, query-conditioned retrieval objective rather than the exact combinatorial object defined for secret storage.

Across these cases, the shared pattern is minimality relative to a target property: exact reconstruction, non-Hamiltonicity, width obstruction, or reasoning sufficiency. The precise mathematical order relation, however, changes from connected-subgraph inclusion in robust secret storage to induced-subgraph inclusion in obstruction theory and to learned bottleneck-based selection in GraphRAG.

## 7. Conceptual significance and common misunderstandings

The principal conceptual importance of MICS is that they convert a global survivability question into a reduced family of atomic reconstruction events [2606.30261]. The full event “some surviving connected subgraph carries the whole secret” is exact, but unwieldy if represented by all connected subgraphs. MICS show that the event can instead be represented by the union of minimal complete-information connected subgraphs and their overlaps.

A common misunderstanding is to identify MICS with smallest-cardinality subgraphs. The definition does not require global size minimality. It requires that a MICS have no proper connected subgraph that already contains the whole information. Minimality is therefore inclusion-based, not necessarily cardinality-based.

A second misunderstanding is to treat MICS as merely heuristic motifs. In the secret-storage framework they are exact objects: the paper proves that survivability is exactly the union probability over MICS events, and inclusion-exclusion over those events yields the exact survivability expansion.

A third misunderstanding is to assume that MICS eliminate overlap complexity. They do not. Two MICS may overlap, and the exact polynomial coefficients depend on unions of overlapping MICS. What MICS remove is redundancy from non-minimal information-carrying connected subgraphs, not the combinatorics of intersections among genuinely minimal ones.

Finally, the broader literature suggests a more general methodological lesson. Minimal subgraphs often act as compact witnesses of a property, but the exact semantics of “information-carrying” depend on context. In distributed secret storage, MICS are formal reconstruction units. In Hamiltonicity and clique-width papers, the comparable objects are minimal obstructions under induced-subgraph inclusion. In GraphRAG, the analogous object is a learned minimal sufficient reasoning subgraph. The shared intuition is strong, but the formal definitions are not interchangeable.

Source: https://www.emergentmind.com/topics/minimal-information-carrying-subgraphs-mics