Imagine you're a building inspector with a checklist. Your checklist has 10 items, then 100, then 1,000 — each longer list catches more code violations. But what if someone proved that no matter how long you make the checklist, there's always a building that passes every item yet is structurally unsound? That's what Chaturvedi and Pawłowski just did to the NPA hierarchy, the most widely used computational tool for certifying whether correlations could be quantum. The committed claim: for every finite level k of the NPA hierarchy, there exist correlations that pass all level-k constraints yet violate information causality — a physical principle that quantum mechanics obeys. The violation isn't marginal. It grows without bound as the number of measurement settings increases, using just one binary-outcome correlation and one classical bit of communication. The NPA hierarchy, introduced in 2007 by Navascués, Pironio, and Acín, is a sequence of semidefinite programming relaxations that outer-approximate the set of quantum correlations. Level 1 is loose; higher levels tighten. The field has long treated convergence of the full infinite hierarchy as sufficient and finite truncations as 'good enough' for practical certification. This paper says: not for information causality, and the gap is not small. Information causality, proposed by Pawłowski et al. in 2009, bounds how much information a receiver can extract about a sender's database when only m classical bits are communicated. Quantum correlations obey this bound. The key result here is that at any fixed NPA level, the optimal information gain about independent database bits grows logarithmically with the number of measurement settings — while information causality demands it stay bounded by one bit. So the gap is O(log n) vs O(1), and that log diverges. The construction is surgical. At each NPA level k, the authors exhibit explicit correlations — binary-outcome, with all positive outcome probabilities — that thread every semidefinite constraint at level k while flagrantly violating the one-bit information causality bound. The positivity of all probabilities matters: it rules out the cheap escape of saying violations only occur at boundary or degenerate points. Even more striking, for any fixed number of measurement settings on each side, every choice of settings admits a local hidden-variable model. The nonlocality only emerges in the information-theoretic aggregate. What does this mean architecturally for the field? The NPA hierarchy belongs to the family of semidefinite programming (SDP) outer approximations — you relax the quantum constraint into a hierarchy of moment conditions and solve each level as an SDP. This paper shows that the relaxation never closes the gap to information causality at any finite level. The implication is structural: information causality is not in the SDP closure of any finite NPA level. If you want to certify information causality compliance, you need either the full infinite hierarchy (impractical) or a fundamentally different tool. Integrity here is strong by the standards of mathematical physics. The result is a proof, not a simulation. The construction is explicit and the bound is analytic (logarithmic growth). The paper is 34 pages with supplementary information, suggesting the argument is worked out in detail rather than sketched. The comparison target — the NPA hierarchy — is the community-standard tool, not a straw man. There are no benchmarks to cherry-pick because the claim is a separation theorem, not an empirical improvement. The successor question is the interesting one. The authors show that information causality defeats every finite NPA level, but they do not construct a computable outer approximation that does respect information causality. That's the obvious next move — and the honest read is that it's genuinely hard, not that they're sandbagging. Building a tractable hierarchy that enforces information causality (or a principle equally strong) is likely a multi-paper program. The field should also ask: does macroscopic locality, or any other known physical principle, also defeat every finite NPA level? If information causality is not alone, the NPA hierarchy has a deeper structural problem.
The analogy
Imagine a building inspector whose checklist gets longer every year — 10 items, 100, 1,000. No matter how long the checklist gets, there's always a building that passes every item yet is structurally unsound. The NPA hierarchy is the checklist; information causality is structural soundness. This paper proves the checklist never catches everything.
Building checklist
A finite-length code-compliance checklist that always misses some structurally unsound buildings no matter how many items you add
NPA hierarchy
A finite-level SDP relaxation that always admits correlations violating information causality no matter how high the level
Honest assessment
This is a clean, significant impossibility result in quantum foundations. It settles a structural question — can any finite NPA level enforce information causality? — with a definitive no. The construction is explicit, the gap is unbounded, and the target is the community's main tool. What it does NOT do is offer a replacement: the field now knows the standard tool leaks, but has no IC-compliant alternative. Practitioners doing device-independent protocols at fixed NPA levels should note the theoretical gap, but won't feel operational pain until someone constructs correlations that exploit it in a protocol-relevant setting. Foundations researchers and anyone working on the quantum correlation boundary should read this carefully.
What Is CLAIMS?
77
/ 100
BedrockNews applies a structured analytical framework called CLAIMS to academic papers and technical research releases.
How CLAIMS worksClaim
5/5
Ladder
4/5
Architecture
4/5
Integrity
5/5
Milestone
2/5
Successor
3/5
What is this paper actually committing to?
Genuine separation theorem: every finite NPA level admits IC-violating correlations with unbounded gap. Not incremental — a structural impossibility result about the field's main computational tool.
The field fight
Are finite NPA levels sufficient proxies for physically realizable (quantum) correlations?
Side A
Finite NPA levels are practically sufficient for device-independent certification and most applications
Held by: Applied DI-QKD community, computational quantum info practitioners
Side B
Finite NPA levels leave structurally important gaps; physical principles like information causality expose failures no finite level can fix
Held by: Foundations community, Pawłowski, Chaturvedi, this paper
Milestone tracker
Unlocks: Tight computable certification of quantum correlations that respects known physical principles
Current
0
IC-compliant hierarchies
No computable tool respects IC
Next
1
IC-compliant hierarchy
First tractable alternative
Unknown — multi-year research program, possibly requires new mathematical machinery
Verified Facts (6)
For every finite level of the NPA hierarchy, correlations exist that satisfy all its constraints but violate information causality
One use of a binary-outcome correlation and one classical bit suffice for a violation that grows without bound as the number of measurement settings increases
The optimal information gain at any fixed NPA level grows logarithmically with the number of settings
The violation remains unbounded even when all outcome probabilities are positive
For any fixed number of measurement settings on each side, every choice of those settings admits a local model
34 pages, 1 figure; includes Supplementary Information
Practical Implications (3)
Device-independent QKD security proofs relying on finite NPA levels have a theoretical gap — information causality violations can hide behind any finite certification level
Researchers building self-testing or certification protocols should not treat finite NPA levels as proxies for physical realizability without additional constraints
The result motivates a new research program: constructing computable relaxations that respect information causality or equivalent physical principles
More latest news
Read all