Icalp related work

ICALP Related Work

At ICALP, related work is where you prove your result is new and better, not where you list adjacent papers. A theory referee arrives already knowing the best bound for your problem; your job is to state exactly how your result compares and to credit the lineage correctly. Vague or missing positioning is a top rejection cause even for correct proofs.

The comparison is the point

• Name the best known prior bound for your exact problem and setting, and state your improvement in the same sentence. "Prior work achieved X under assumption A; we achieve Y, removing A" is the contract.
• If your result is incomparable to prior work (different regime, different parameter), say so explicitly and give the reader the axis of comparison — do not let them assume you missed a stronger result.
• Distinguish unconditional from conditional results (and which assumption: SETH, 3SUM, APSP, etc.). A conditional lower bound compared against an unconditional one is a category error referees catch instantly.

Cover the right literature lanes

For a Track A paper, the lanes are typically: the algorithmic line (best upper bounds), the hardness line (matching or conditional lower bounds), and the model/parameter line (dynamic, parallel, streaming, parameterized). For a Track B paper: the expressiveness line, the decidability/complexity line, and the prior characterizations of the class. Miss a lane and a referee who works in it reads you as unaware.

Credit the right venues

TCS results scatter across venues and journals; attribute them correctly:

• Conference origins: STOC, FOCS, SODA, ICALP, LICS, CCC, SoCG, CONCUR, ... — cite the venue and year that established the result you compare against.
• Journal versions matter. Many TCS results have a definitive journal version (JACM, SICOMP, TALG, TCS, LMCS) with the full proof and the tightest bound; cite it when it is the version of record for the bound you claim to beat.
• Do not misattribute a famous result to ICALP because it "feels" like a Track A/B result — check dblp for the actual venue (see resources/exemplars/library.md for sibling-venue traps like the PCP theorem at FOCS, or AKS in the Annals).

Lightweight double-blind self-citation

Under ICALP's lightweight double-blind, your own prior work must be cited in the third person:

• Not "in our previous paper [12] we showed" but "it was shown in [12] that".
• Keep the citation in the bibliography; do not remove your own relevant work (that itself signals authorship and weakens positioning) — just phrase it neutrally.
• Do not cite "our full version on arXiv" in a way that de-anonymizes; refer to "the full version" without the identifying handle during review.

Positioning patterns that work

[Delta-first]     "The best known bound for P is B [ref]. We improve this to B', and show B' is
                  optimal under <assumption> (Thm 2)."
[Barrier-named]   "Prior approaches were limited by <obstruction>; we bypass it via <technique>."
[Regime-split]    "For dense inputs [ref] is best; our result is the first improvement in the sparse
                  regime."
[Characterization] "Earlier work classified <subclass>; we complete the picture with a full
                  dichotomy over the whole family."

Anti-patterns

• A related-work paragraph that only lists titles with no comparison of bounds.
• Omitting the matching lower bound literature so the reader cannot judge tightness.
• Claiming novelty against a weaker prior bound while a stronger one exists (the referee knows it).
• First-person self-citation that breaks the blind.
• Attributing a result to the wrong venue because a coauthor or technique is familiar.

Output format

[Best prior bound] <named result, venue/journal, the exact quantity>
[Our delta] <improvement, conditional/unconditional, regime>
[Lanes covered] algorithmic / hardness / model — or expressiveness / complexity / characterization
[Attribution check] venues verified on dblp? journal versions cited where they are the record?
[Anonymity] self-citations third-person? full version referenced without de-anonymizing?
[Fix queue] <ordered positioning edits>