Monotone Erasure Codes: Thresholds & Applications

• Monotone erasure codes are defined by increasing erasure sets that ensure adding more erasures never improves bit recovery, establishing a clear monotonicity property.
• They leverage symmetry via EXIT functions and sharp-threshold theorems to achieve capacity under MAP decoding for structured codes like Reed–Muller, affine-invariant, and BCH codes.
• Applications span distributed systems and hierarchical models, where linear monotone erasure codes (LMECs) and ordered erasure strategies ensure robust data recovery and efficient reconstruction.

In the cited literature, “monotone erasure codes” refers to settings in which recoverability or non-recoverability is monotone in an erasure pattern. On the binary erasure channel (BEC), the key object is the monotone set of erasure patterns that prevent maximum a posteriori (MAP) recovery of a bit; combined with transitivity, double transitivity, EXIT identities, and sharp-threshold theorems, this yields capacity results for deterministic linear codes such as Reed–Muller, affine-invariant, and BCH families (Kudekar et al., 2016). In distributed systems, the term also denotes erasure codes for arbitrary monotone access structures, including linear monotone erasure codes (LMECs) whose reconstruction condition is expressed as a rank constraint on node-assigned generator columns (Bammert et al., 21 May 2026). Related variants appear in hierarchical erasures over extension fields and in ordered deletion–erasure channels, where monotonicity is imposed by left-justified or ordered corruption patterns rather than by subsets of erased coordinates (Raviv et al., 2018, Ganesan, 2018).

1. Monotonicity on the binary erasure channel

The BEC has input alphabet $X=\{0,1\}$ and output alphabet $Y=\{0,1,*\}$; each transmitted bit is erased independently with probability $\varepsilon$ and otherwise received without error, and its capacity is $C(\varepsilon)=1-\varepsilon$ (Kudekar et al., 2016). For a binary linear code $C\subseteq\{0,1\}^n$ and a uniform prior on codewords, MAP decoding on the BEC reduces to unique recoverability from the linear equations induced by the unerased positions. Block-MAP seeks a unique codeword, whereas bit-MAP for coordinate $i$ seeks a unique value of $X_i$ from the extrinsic observation $Y_{\sim i}$.

The monotone structure is encoded by the failure set

$$\Omega_i \triangleq \Big\{ A \subseteq [N]\setminus\{i\} \mid \exists\, B\subseteq A,\; B\cup\{i\}\in C \Big\}.$$

Equivalently, in the formulation using erasure indicators $\omega\in\{0,1\}^{n-1}$,

$Y=\{0,1,*\}$0

If an erasure pattern prevents recovery of bit $Y=\{0,1,*\}$1, any superset of erasures also prevents recovery: adding erasures cannot help the decoder. Thus $Y=\{0,1,*\}$2 is an increasing set, and its indicator is a monotone Boolean function of the erasure pattern.

With the product measure $Y=\{0,1,*\}$3 on $Y=\{0,1,*\}$4, the bit-MAP EXIT function satisfies

$Y=\{0,1,*\}$5

On the BEC, this is exactly the probability that bit $Y=\{0,1,*\}$6 is not recoverable from the extrinsic observation. The bit erasure probabilities obey

$Y=\{0,1,*\}$7

because bit $Y=\{0,1,*\}$8 can remain unknown only when $Y=\{0,1,*\}$9 is itself erased and the remaining observations fail to determine it.

2. Symmetry, EXIT functions, and sharp thresholds

The threshold mechanism depends not only on monotonicity but also on code symmetry. For a linear code of length $\varepsilon$0, the vector EXIT function is

$\varepsilon$1

with scalar specialization $\varepsilon$2 at $\varepsilon$3 and average EXIT

$\varepsilon$4

If the permutation group of the code is transitive, then all bits have identical EXIT functions and

$\varepsilon$5

If the permutation group is doubly transitive, then for distinct $\varepsilon$6,

$\varepsilon$7

so the influences of the coordinates are equal at symmetric points (Kudekar et al., 2016).

The relevant analytic tool is the Margulis–Russo lemma:

$\varepsilon$8

where $\varepsilon$9 is the total influence. For $C(\varepsilon)=1-\varepsilon$0, this yields

$C(\varepsilon)=1-\varepsilon$1

Under equal influences, the symmetric sharp-threshold theorem gives a universal constant $C(\varepsilon)=1-\varepsilon$2 such that

$C(\varepsilon)=1-\varepsilon$3

and therefore, for any $C(\varepsilon)=1-\varepsilon$4,

$C(\varepsilon)=1-\varepsilon$5

For code sequences with doubly transitive permutation groups, $C(\varepsilon)=1-\varepsilon$6, so the EXIT transition width shrinks like $C(\varepsilon)=1-\varepsilon$7.

The area theorem fixes the transition point:

$C(\varepsilon)=1-\varepsilon$8

As the EXIT function becomes step-like, the area constraint forces the jump to occur at

$C(\varepsilon)=1-\varepsilon$9

In the formulation of Proposition 7, bit-MAP capacity, pointwise convergence of $C\subseteq\{0,1\}^n$0 to a step function, and vanishing threshold width are equivalent. The 2015 arXiv formulation already emphasized the same chain: monotone EXIT functions, 2-transitivity, and the Area Theorem pin the threshold of Reed–Muller codes at channel capacity under bit-MAP decoding (Kudekar et al., 2015).

3. Capacity-achieving code families on erasure channels

For Reed–Muller codes,

$C\subseteq\{0,1\}^n$1

with rate

$C\subseteq\{0,1\}^n$2

Choosing

$C\subseteq\{0,1\}^n$3

gives asymptotic rate $C\subseteq\{0,1\}^n$4 (Kudekar et al., 2016). Reed–Muller codes are affine-invariant; their automorphism group contains $C\subseteq\{0,1\}^n$5, which acts doubly transitively on coordinates. By Theorem 3, any sequence with $C\subseteq\{0,1\}^n$6 and rate $C\subseteq\{0,1\}^n$7 achieves BEC capacity under bit-MAP, with threshold at $C\subseteq\{0,1\}^n$8.

For block-MAP, the argument requires an additional conversion from vanishing bit erasure to vanishing block erasure. General inequalities are

$C\subseteq\{0,1\}^n$9

Theorem 4 uses the route $i$0, while Theorem 5 uses a stronger derivative lower bound on an interval $i$1 with $i$2, $i$3, and $i$4. For Reed–Muller codes, Theorem 4 does not directly apply because the minimum distance is sublinear; instead, stronger symmetry on $i$5 yields

$i$6

for $i$7, which implies block-MAP capacity via Theorem 5 (Kudekar et al., 2016).

The same symmetry-based framework extends to affine-invariant codes, hence to extended primitive narrow-sense BCH codes. The extended code $i$8 of length $i$9 is affine-invariant and has minimum distance at least $X_i$0; for the sequences considered in the paper,

$X_i$1

Therefore affine-invariance gives bit-MAP capacity, while the distance growth gives block-MAP capacity. Since BCH codes are cyclic, this yields sequences of primitive narrow-sense BCH codes that achieve capacity under both bit-MAP and block-MAP on the BEC, resolving affirmatively the existence of capacity-achieving sequences of binary cyclic codes (Kudekar et al., 2016).

A broader family is given by the dual Berman codes $X_i$2 and Berman codes $X_i$3 of length $X_i$4. Their parameters are

$X_i$5

and

$X_i$6

When $X_i$7, $X_i$8 is exactly $X_i$9. For $Y_{\sim i}$0, these codes are transitive but not doubly transitive; nevertheless, by using the Kumar–Calderbank–Pfister result on transitive codes whose stabilizer orbits grow to infinity, the paper proves BEC capacity under bit-MAP for sequences with target rate $Y_{\sim i}$1 (Natarajan et al., 2022). For odd $Y_{\sim i}$2, the same work identifies a larger abelian family of ideals in $Y_{\sim i}$3 that also achieves BEC capacity under bit-MAP.

4. Linear monotone erasure codes for arbitrary access structures

In the distributed-systems formulation, an access structure is a monotone family $Y_{\sim i}$4 of authorized node sets: if $Y_{\sim i}$5 and $Y_{\sim i}$6, then $Y_{\sim i}$7. A compact specification is a monotone Boolean formula $Y_{\sim i}$8 using only AND, OR, and threshold operators $Y_{\sim i}$9. A monotone erasure code for $$\Omega_i \triangleq \Big\{ A \subseteq [N]\setminus\{i\} \mid \exists\, B\subseteq A,\; B\cup\{i\}\in C \Big\}.$$0 consists of

• $$\Omega_i \triangleq \Big\{ A \subseteq [N]\setminus\{i\} \mid \exists\, B\subseteq A,\; B\cup\{i\}\in C \Big\}.$$1,
• $$\Omega_i \triangleq \Big\{ A \subseteq [N]\setminus\{i\} \mid \exists\, B\subseteq A,\; B\cup\{i\}\in C \Big\}.$$2,

with the correctness condition that for every authorized $$\Omega_i \triangleq \Big\{ A \subseteq [N]\setminus\{i\} \mid \exists\, B\subseteq A,\; B\cup\{i\}\in C \Big\}.$$3, decoding from the fragments in $$\Omega_i \triangleq \Big\{ A \subseteq [N]\setminus\{i\} \mid \exists\, B\subseteq A,\; B\cup\{i\}\in C \Big\}.$$4 recovers the file (Bammert et al., 21 May 2026). There is no privacy requirement; privacy is the domain of secret sharing rather than of MECs.

A basic construction follows the access tree. At an OR node, the whole input is assigned to each child. At a $$\Omega_i \triangleq \Big\{ A \subseteq [N]\setminus\{i\} \mid \exists\, B\subseteq A,\; B\cup\{i\}\in C \Big\}.$$5 node, the input is split into $$\Omega_i \triangleq \Big\{ A \subseteq [N]\setminus\{i\} \mid \exists\, B\subseteq A,\; B\cup\{i\}\in C \Big\}.$$6 subfragments so that any $$\Omega_i \triangleq \Big\{ A \subseteq [N]\setminus\{i\} \mid \exists\, B\subseteq A,\; B\cup\{i\}\in C \Big\}.$$7 suffice, for example by an $$\Omega_i \triangleq \Big\{ A \subseteq [N]\setminus\{i\} \mid \exists\, B\subseteq A,\; B\cup\{i\}\in C \Big\}.$$8 MDS code, and the recursion continues. At a leaf labeled by node $$\Omega_i \triangleq \Big\{ A \subseteq [N]\setminus\{i\} \mid \exists\, B\subseteq A,\; B\cup\{i\}\in C \Big\}.$$9, the assigned chunk is appended to node $\omega\in\{0,1\}^{n-1}$0’s fragment. This construction is complete for the access structure, though it does not necessarily minimize overhead.

The linear version fixes a finite field $\omega\in\{0,1\}^{n-1}$1, a full-rank generator matrix $\omega\in\{0,1\}^{n-1}$2, and an assignment $\omega\in\{0,1\}^{n-1}$3 from columns to nodes. For node $\omega\in\{0,1\}^{n-1}$4, the fragment is

$\omega\in\{0,1\}^{n-1}$5

where $\omega\in\{0,1\}^{n-1}$6 is the submatrix of columns assigned to $\omega\in\{0,1\}^{n-1}$7 and $\omega\in\{0,1\}^{n-1}$8 is the number of those columns. The total coded length is $\omega\in\{0,1\}^{n-1}$9, the overhead is

$Y=\{0,1,*\}$00

and the rate is $Y=\{0,1,*\}$01. The reconstruction theorem is purely algebraic: a subset $Y=\{0,1,*\}$02 can reconstruct if and only if

$Y=\{0,1,*\}$03

The paper gives three principal LMEC constructions. First, for a general access tree, it composes small Vandermonde blocks and Kronecker products. If $Y=\{0,1,*\}$04 is the maximum number of children of any nontrivial threshold node, choosing a prime power $Y=\{0,1,*\}$05 suffices. If $Y=\{0,1,*\}$06 is the set of internal vertices whose children are leaves, and

$Y=\{0,1,*\}$07

then the construction has

$Y=\{0,1,*\}$08

so

$Y=\{0,1,*\}$09

Second, when the minimal authorized sets $Y=\{0,1,*\}$10 are listed explicitly, optimal LMECs are obtained from an $Y=\{0,1,*\}$11 MDS base code by solving

$Y=\{0,1,*\}$12

From an optimal rational solution, one sets $Y=\{0,1,*\}$13 to the least common multiple of the denominators, chooses $Y=\{0,1,*\}$14, and labels $Y=\{0,1,*\}$15 columns of the base MDS code by node $Y=\{0,1,*\}$16. The overhead becomes

$Y=\{0,1,*\}$17

Third, for partitioned access structures, a dynamic program $Y=\{0,1,*\}$18 computes optimal parameters bottom-up. In the threshold special case $Y=\{0,1,*\}$19, the optimal LMEC is the standard $Y=\{0,1,*\}$20 MDS code with $Y=\{0,1,*\}$21 and one column per node; any $Y=\{0,1,*\}$22 nodes reconstruct by solving a $Y=\{0,1,*\}$23 linear system (Bammert et al., 21 May 2026).

5. Hierarchical and ordered erasure models

Hierarchical erasures arise for linear codes over extension fields $Y=\{0,1,*\}$24 when each symbol is revealed progressively in a fixed basis order. Writing

$Y=\{0,1,*\}$25

a hierarchical erasure pattern is a tuple $Y=\{0,1,*\}$26 with $Y=\{0,1,*\}$27 and $Y=\{0,1,*\}$28, where the first $Y=\{0,1,*\}$29 coordinates of symbol $Y=\{0,1,*\}$30 are erased. The induced sets

$Y=\{0,1,*\}$31

satisfy

$Y=\{0,1,*\}$32

so the erasures are left-justified and monotone in the hierarchy index. The central correctability criterion is

$Y=\{0,1,*\}$33

where $Y=\{0,1,*\}$34 is the $Y=\{0,1,*\}$35-subspace of components hidden by the pattern. Lemma 1 states that a code is $Y=\{0,1,*\}$36-correcting if and only if this condition holds for every $Y=\{0,1,*\}$37 (Raviv et al., 2018).

This framework yields several constructions. Universally Decodable Matrices (UDMs) give $Y=\{0,1,*\}$38-correcting codes of dimension $Y=\{0,1,*\}$39 through trace-based parity checks. Balanced and power erasure patterns are handled by recursive-basis parity constructions when $Y=\{0,1,*\}$40. Gabidulin codes provide a rank-metric route: for $Y=\{0,1,*\}$41 and $Y=\{0,1,*\}$42, the code

$Y=\{0,1,*\}$43

corrects

$Y=\{0,1,*\}$44

The paper places these constructions in applications such as distributed storage and flash memories (Raviv et al., 2018).

A different monotone model is the ordered deletion–erasure channel, where one deletion occurs before an optional erasure. For positions $Y=\{0,1,*\}$45, the map $Y=\{0,1,*\}$46 deletes $Y=\{0,1,*\}$47 and erases $Y=\{0,1,*\}$48, with the erasure position marked and the deletion position unknown. The construction augments the Varshamov–Tenengolts checksum with

$Y=\{0,1,*\}$49

defining

$Y=\{0,1,*\}$50

The redundancy bound is

$Y=\{0,1,*\}$51

while any code correcting all ordered deletion–erasure patterns must satisfy, for large $Y=\{0,1,*\}$52,

$Y=\{0,1,*\}$53

Decoding uses the discrepancy

$Y=\{0,1,*\}$54

and the modified checksum

$Y=\{0,1,*\}$55

scanning candidate deletion positions $Y=\{0,1,*\}$56 until the congruence modulo $Y=\{0,1,*\}$57 is satisfied. The order restriction $Y=\{0,1,*\}$58 is essential to the construction and proof (Ganesan, 2018).

6. Decoding regimes, applications, and open problems

The BEC capacity results are proved for bit-MAP and block-MAP decoding. The 2016 Reed–Muller paper emphasizes that MAP decoding is information-theoretically optimal but computationally expensive in general and does not claim low-complexity decoders achieving the same thresholds (Kudekar et al., 2016). In the BEC-specific linear setting, however, bit-MAP and block-MAP reduce to linear algebra over $Y=\{0,1,*\}$59 and can be implemented by Gaussian elimination, which is polynomial time though not typically near-linear for dense structured codes (Kumar et al., 2015). This distinction is important: the sharp-threshold arguments establish existence of threshold behavior under optimal decoding, not decoder practicality.

On the systems side, LMECs are used to generalize asynchronous verifiable information dispersal to arbitrary Byzantine quorum systems. The generalized AVID construction uses an LMEC whose access structure is the kernel system $Y=\{0,1,*\}$60 of the quorum system. The paper proves termination, agreement, availability, and correctness, and gives explicit costs: Disperse has message complexity $Y=\{0,1,*\}$61; with a vector of hashes its communication is

$Y=\{0,1,*\}$62

and with a Merkle root (GAVID-H) it becomes

$Y=\{0,1,*\}$63

Retrieve uses $Y=\{0,1,*\}$64 messages and

$Y=\{0,1,*\}$65

bits with Merkle authentication (Bammert et al., 21 May 2026).

Across the literature, several limitations remain explicit. For the BEC symmetry program, extending from erasures to general binary-input memoryless symmetric channels requires GEXIT-type analyses; the papers note that the natural functions are no longer Boolean or monotone, and extending the threshold-pinning argument remains open (Kudekar et al., 2016). The same work conjectures that weaker conditions than double transitivity—such as transitivity together with growth of both $Y=\{0,1,*\}$66 and dual $Y=\{0,1,*\}$67—might suffice, but leaves the question open. For Reed–Muller and related families, designing low-complexity decoders that preserve the MAP-level thresholds remains an open engineering challenge (Kudekar et al., 2015). For LMECs, finding optimal constructions from compact monotone Boolean formulas, adapting to dynamic access structures, and moving beyond Reed–Solomon-style MDS bases are identified as open problems (Bammert et al., 21 May 2026). In the hierarchical and ordered settings, the presented constructions do not claim unordered deletion–erasure correction, multiple deletions, or multiple ordered erasures, and broader extensions are left for future work (Raviv et al., 2018, Ganesan, 2018).

Taken together, these results make monotonicity a recurrent design principle rather than a single canonical model. On the BEC, monotone MAP failure events plus symmetry yield sharp EXIT thresholds and capacity. In access-structure coding, monotone authorization determines which node sets must satisfy a rank condition. In extension-field and synchronization settings, monotonicity is imposed on the pattern itself through left-justified or ordered corruption. The common feature is that erasure patterns are partially ordered, and the code or decoder is designed so that this order can be exploited algebraically or probabilistically.