Imagine you're at a library with a suspicious librarian. You want to check out 3 books from a shelf of 10, but you don't want the librarian to know which 3 — and the librarian doesn't want you reading the other 7 by accident. Now add static to every conversation: some of your words get lost in transit. The question is: how fast can you complete this transaction given the noise? This paper gives the exact answer. The committed claim: for a binary erasure channel with erasure probability ε, the capacity of K-out-of-N oblivious transfer is min(Kε/(N−K), 1−ε). That's a clean, closed-form expression — not an upper bound, not an asymptotic approximation, but the exact rate. It generalizes two previously known special cases: Ahlswede-Csiszár's 1-out-of-2 result and the 1-out-of-N result for high erasure rates (ε ≥ (N−1)/N). The unification is the real contribution. The architecture is pure information-theoretic: converse bounds proved via entropy inequalities and Fano's inequality, achievability via random binning and channel coding over the BEC. No algorithms, no simulations, no hardware — this is a proof paper. The method leans on the structure of the erasure channel specifically: the server knows exactly which bits the client received (the erasure pattern is known to the sender in BEC), which is the cryptographic lever that makes OT possible without any computational assumptions. Everything is in the honest-but-curious model — both parties follow the protocol but try to learn extra information from their transcripts. The most operationally interesting result is the suboptimality finding. The naive approach to K-out-of-N OT is to run an optimal 1-out-of-N protocol K times. The authors prove this doesn't always achieve capacity. The gap isn't just theoretical hair-splitting; it means the interaction structure of multi-message retrieval is fundamentally richer than K independent single-message retrievals. However, they also identify specific regimes where a 1-out-of-M protocol (M < N) can be repeated to optimally build 1-out-of-N OT — so the reduction landscape has real structure, not just 'it's complicated.' Integrity is strong for what this is: a mathematical proof paper submitted to IEEE Transactions on Information Theory. The validation is the proof itself — no simulations to cherry-pick, no benchmarks to game. The comparison ladder is clean: they explicitly name the Ahlswede-Csiszár 1-out-of-2 capacity and the Naor-Pinkas / prior 1-out-of-N results, show their formula recovers those as special cases, and prove strict generalization. The extension to generalized erasure channels (GEC) is discussed but left partially open, which reads as honest scope management rather than cherry-picking. The milestone question for this subfield is bridging the gap from erasure channels to general discrete memoryless channels. The BEC is the information theorist's fruit fly — tractable because the erasure pattern is known. For general DMCs, the OT capacity is unknown even for 1-out-of-2 in most cases. Getting the full K-out-of-N capacity for a broader channel family (say, the binary symmetric channel) would be the next qualitative jump. That's likely years away; the BEC's algebraic structure is doing a lot of the heavy lifting here. The obvious experiment not run: extending the tight capacity result beyond the BEC to other specific channels like the BSC or Z-channel, even for K=1. The honest read is (a) — the proof techniques are BEC-specific (they exploit the sender's knowledge of the erasure pattern), and the extension is genuinely hard, not just computationally expensive. The GEC discussion in the paper is a partial step, but the authors are upfront that full characterization remains open.
The analogy
You're at a suspicious library: you want K books from a shelf of N, the librarian must not learn which K you picked, and you must not learn the other N−K — but every word you exchange has a chance of getting garbled. This paper proves the exact maximum checkout speed given the garble rate.
Library checkout
Requesting K specific books from N without the librarian learning your selection, over a noisy intercom
K-out-of-N OT capacity
Exact rate at which K of N messages can be privately retrieved over a binary erasure channel
Honest assessment
A clean, complete information-theoretic result that closes the K-out-of-N OT capacity question for the binary erasure channel. The suboptimality of naive reductions is the most surprising finding. This matters to the information-theoretic security community and channel-coding theorists; practitioners building real OT protocols over noisy links are a secondary audience since BEC is a model, not a deployment channel. The techniques are BEC-specific, so the real test of impact is whether the ideas seed progress on harder channels.
What Is CLAIMS?
70
/ 100
BedrockNews applies a structured analytical framework called CLAIMS to academic papers and technical research releases.
How CLAIMS worksClaim
4/5
Ladder
4/5
Architecture
4/5
Integrity
4/5
Milestone
2/5
Successor
3/5
What is this paper actually committing to?
Genuine first: closed-form K-out-of-N OT capacity for all (K,N,ε) over BEC. Unifies and strictly extends two separate prior results. Clean claim, cleanly stated.
The field fight
Can information-theoretic OT capacities be characterized in closed form beyond the simplest channel models?
Side A
BEC is a special case; general DMC capacity requires fundamentally new tools
Held by: Most information-theoretic security researchers
Side B
BEC results can be progressively extended to richer channel families via structural generalizations
Held by: This paper, GEC extension line
Milestone tracker
Unlocks: Information-theoretic OT capacity theory for realistic (non-erasure) channel models
Current
K,N,ε
full BEC param space
this paper
Next
K=1,N=2
BSC capacity
simplest non-erasure case
5+ years — requires fundamentally new converse techniques
Verified Facts (7)
K-out-of-N OT capacity over BEC with parameter ε is min(Kε/(N−K), 1−ε)
Generalizes known capacity results for 1-out-of-2 OT and for 1-out-of-N OT when ε ≥ (N−1)/N
Constructing K-out-of-N OT by repeating K times an optimal 1-out-of-N OT protocol can be suboptimal
Certain ranges of ε and M < N allow 1-out-of-N OT to be optimally realized via repetition of 1-out-of-M OT
Extensions to generalized erasure channel (GEC) are discussed
Submitted to IEEE Transactions on Information Theory
Model: honest-but-curious
Practical Implications (3)
Protocol designers for K-out-of-N OT over erasure-like channels now have an exact rate target — no need to guess whether their scheme is near-optimal
The suboptimality of K-fold repetition of 1-out-of-N means composite OT protocols must be designed holistically, not by naive composition
Identified regimes where 1-out-of-M can optimally build 1-out-of-N give designers modular building blocks with provable efficiency guarantees
More latest news
Read all