Could the correspondence fail only for specific spacetime dimensions or boundary CFT structures rather than universally?

Your question asks something subtle and important: **might AdS/CFT and related holographic dualities fail *selectively* — only in certain dimensions or for specific boundary structures — rather than breaking down universally?**

The preprints suggest the answer is yes, and illustrate several ways this selectivity can arise.

Dimension-Dependent Failures

The most direct evidence comes from cosmological holography. In de Sitter spacetime (relevant to our universe), the bulk doesn't obey full conformal invariance — only scale invariance — in modified gravity theories. Crucially, this breakdown depends on *what* gravity theory you have. Einstein-Aether theory, which breaks bulk boost symmetry, is conjectured to have only scale-invariant (not fully conformal) dual descriptions [AG-2026.02-334]. This means the correspondence changes character depending on the gravitational assumptions, not just the dimension.

More strikingly, the de Sitter/CFT correspondence itself appears dimension-sensitive. Recent work on "yes boundaries" formulates holographic duals specifically in three bulk dimensions, where explicit path integral calculations become tractable [AG-2026.04-745]. The authors compute beyond-general-relativity quantum effects that "contribute strongly," suggesting the correspondence's structure depends critically on whether you can actually solve the theory — implying it may look different in 4D or higher.

Boundary-Structure Failures

The correspondence is also *not universal* across different boundary topologies. Two-dimensional BCFTs (boundary conformal field theories) and XCFTs (crosscap CFTs) have holographic duals with qualitatively different structures from ordinary CFTs [AG-2025.01-691]. When you glue a boundary or crosscap into the spacetime, the correlator algebra changes — there's no single universal prescription.

Similarly, when you impose different asymptotic boundary conditions (different "fall-off indices" near infinity), the bulk-to-boundary correlators take on different functional forms. For instance, scalar correlators are unique up to normalization, but fermionic ones split into branches with different power-law exponents [AG-2026.01-916]. This means the correspondence selects which operator is allowed depending on how it behaves at infinity.

Why This Matters

These results suggest AdS/CFT isn't a monolithic principle that either works or doesn't. Instead, it's a family of correspondences — each pairing a specific bulk geometry, boundary topology, and asymptotic structure to a dual field theory. Failures are architectural: they occur when you mismatch pieces (e.g., assuming full conformal invariance in a theory that only has scale invariance, or trying to build a correlator that violates the boundary conditions).

The preprints provided don't directly address whether the correspondence fails in, say, de Sitter in all dimensions or only some — but the pattern suggests it's a question worth asking, because the correspondence's validity is already known to hinge on very specific structural choices.