{
  "slug": "trapping-set",
  "term": "Trapping set",
  "aliases": [
    "LDPC trapping set"
  ],
  "definition": "A trapping set is the nonempty set of all variable nodes that fail to become permanently correct for some input to a specified decoder.",
  "note": "The operational definition depends on the decoder and its input space, not only the graph. The (a,b) notation counts variables and odd-degree neighboring checks; it does not establish a decoder failure by itself. A wrong codeword can have b=0. [richardson2003]\n\nAbsorbing sets instead have graph-defined strict-majority conditions. A fully absorbing configuration is stable under the specified majority bit-flipping rule, not necessarily under every soft decoder. [dolecek2010]\n\nFor matrix rows 1100, 1010, 0110 and 1001, transmit 0000 and start hard decisions at 1110. Synchronous strict-majority flipping, with ties unchanged, gives 1110 -> 1111 -> 1111. The eventual failure support is all four bits, not the initial three-bit absorbing set. This exact toy calculation is not a NAND measurement. [example]\n\nFor conditionally independent bit-channel observations given the codeword, define L_i = ln(P(y_i|0)/P(y_i|1)). Choose illustrative LLRs (-1,-1,-1,+4), whose signs give the same 1110 hard decisions. Exhaustive enumeration finds only codewords 0000 and 1111. Their scores sum((1-2*x_i)*L_i) are +1 and -1; maximum-likelihood decoding selects 0000. The comparison changes both the decision rule and retained reliability information, so it isolates neither one's individual effect. [example]",
  "related": [
    "ldpc",
    "absorbing-set",
    "fully-absorbing-set"
  ],
  "citations": [
    {
      "id": "richardson2003",
      "title": "Tom Richardson. Error Floors of LDPC Codes. Allerton 2003, pages 1426-1435, sections 3.1 and 3.3.",
      "url": "https://web.stanford.edu/class/ee388/papers/ErrorFloors.pdf",
      "accessed": "2026-11-22"
    },
    {
      "id": "dolecek2010",
      "title": "Dolecek et al. Analysis of Absorbing Sets and Fully Absorbing Sets of Array-Based LDPC Codes (2010), section II.B, Definition 1. DOI 10.1109/TIT.2009.2034781.",
      "url": "https://people.eecs.berkeley.edu/~bora/Journals/2010/IT-10.pdf",
      "accessed": "2026-11-22"
    },
    {
      "id": "example",
      "title": "Hesela: executable synthetic parity-graph and decoder-dependent witness",
      "url": "https://hesela.dev/research/ldpc-absorbing.mjs",
      "accessed": "2026-11-22"
    }
  ],
  "provenance": [
    "The four-bit example compares an exactly specified hard-decision rule with exhaustive maximum-likelihood scoring on a toy code. It does not reproduce Richardson's algorithm, a NAND controller, a BER or FER curve.",
    "The fixed point proves this toy trajectory stays wrong; observing a general decoder for finitely many iterations alone would not prove its eventual behavior."
  ],
  "status": "research",
  "advantages": [],
  "limitations": [],
  "supersedes": [],
  "supersededBy": [],
  "firstSeen": null,
  "lastSeen": null,
  "updated": "2026-11-22"
}
