Holographic Quantum Error Correction — When Physics Discovered Coding Theory
A point in the interior of anti-de Sitter space can be reconstructed from operators on the left half of the boundary, or the right half, or the top, or the bottom — but never from any single small patch. In 2015, three physicists noticed this redundancy pattern was identical to a Shor code. Spacetime, it turned out, was running quantum error correction the whole time.
The 2015 realization
Ahmed Almheiri, Xi Dong, and Daniel Harlow wrote down a paper — "Bulk Locality and Quantum Error Correction in AdS/CFT" — pointing out a strangeness in the AdS/CFT dictionary. A bulk operator can be expressed as a boundary operator in multiple, non-equivalent ways. The interior field is not stored in any one place on the boundary; it is smeared across the whole thing, recoverable from any sufficiently large region and from none that are too small.
That is, exactly, the definition of an error-correcting code. The "logical" data lives in the bulk. The "physical" qubits live on the boundary CFT. Erase a small piece of the boundary and you lose nothing; erase more than half and the logical bit is gone. The threshold separating the two regimes is the minimal Ryu-Takayanagi surface — the same surface whose area gives the entanglement entropy in the RT formula. The geometry was the code distance.
The HaPPY code
Later that year, Pastawski, Yoshida, Harlow, and Preskill ("HaPPY") built an explicit toy model: a tensor network tiling the hyperbolic plane with pentagons, each pentagon a "perfect tensor" (any bipartition of its indices yields a maximally entangled state). The network is simultaneously:
- a quantum error-correcting code,
- a satisfier of the Ryu-Takayanagi formula on the nose,
- a model in which the same bulk operator has multiple inequivalent boundary representations.
The construction proved that RT and QEC are not two coincident features of holography. They are the same feature written in two notations. The code distance — how many physical qubits an adversary must corrupt to destroy a logical qubit — equals the area of the minimal surface, measured in Planck units.
What this implies about spacetime
- Locality is emergent. A "local" operator in the bulk is a wildly non-local operator on the boundary. The boundary has no notion of bulk position. Bulk locality is a property of the code, not of the underlying degrees of freedom.
- Horizons are code boundaries. A black hole horizon is the RT surface separating interior from exterior. The interior is encoded in the exterior CFT, but protected: no small slice of Hawking radiation tells you what fell in. You need access to more than half the boundary to decode. This is why individual Hawking quanta look thermal, while the full ensemble must carry information.
- The island formula slots in cleanly. The Penington / Almheiri et al. resolution of the black hole information paradox (2019-2020) reads, in this language, as a code-subspace transition: at the Page time, the region of the bulk reconstructible from the radiation suddenly includes the interior. The "island" is the bulk region that joined the code.
Harlow's 2016 theorem closed the loop: exact RT is equivalent to exact QEC; the quantum-corrected FLM formula is equivalent to approximate QEC. The two ideas are not analogies. They are the same theorem.
What's contested
- Is the universe a code? AdS/CFT is on a spacetime with a negative cosmological constant. Ours has a positive one. Whether de Sitter space admits a holographic-QEC description — with the cosmic horizon as the codeword surface — is open. Almheiri, Susskind, and others have speculated yes; nobody has constructed it.
- What is the boundary theory in de Sitter? No agreed candidate exists. Without one, the "our universe is a code" claim is suggestive, not demonstrated.
- Are perfect tensors physical? The HaPPY code uses idealized tensors that real CFTs do not realize. Hyperinvariant tensor networks (Evenbly, 2017; refined 2023-2025) reproduce correct boundary two-point functions, but the full mapping from a real interacting CFT to a tensor-network code is incomplete.
- The code-subspace puzzle. The bulk Hilbert space that the code protects is small — far smaller than the full boundary Hilbert space. Which bulk states belong to the code subspace, and how that subspace shifts dynamically (as in the island story), is still being worked out.
Why this has to do with other realms
The same redundancy logic shows up in working quantum hardware. Holographic codes are now studied as practical fault-tolerant codes, not just toys: hyperinvariant networks have tunable rates and competitive distance scaling, and the LEGO_HQEC formalism (2024) automates their construction. A 2025 result demonstrated universal fault-tolerant logic with heterogeneous holographic codes. Traffic flows both ways — the geometry of concept ads cft correspondence is shaping the layout of error-correcting layers in real quantum computers, while improvements in concept quantum error correction feed back into how physicists think about emergent spacetime. The clearest sign that information theory and gravity are talking to each other is that the same lemma proves theorems in both.
An open question
If the cosmic horizon is a code surface, what is the logical Hilbert space it protects — and is there an observer-independent way to describe it, or is "the code" itself frame-dependent?
Key sources
- Almheiri, Dong, Harlow (2015), "Bulk Locality and Quantum Error Correction in AdS/CFT," JHEP — the founding paper.
- Pastawski, Yoshida, Harlow, Preskill (2015), "Holographic quantum error-correcting codes," JHEP — the HaPPY code construction.
- Harlow (2016), "The Ryu-Takayanagi formula from quantum error correction," Commun. Math. Phys. — proves the RT ↔ QEC equivalence.
- Penington (2019); Almheiri, Engelhardt, Marolf, Maxfield (2019) — the island formula and its code-subspace reading.
- To verify: Evenbly (2017) on hyperinvariant tensor networks and the 2023-2025 follow-ups extending them to correct CFT correlators.
- To verify: LEGO_HQEC software paper (Cao et al., ~2024) and the 2025 universal-fault-tolerance demonstration.
Further reading
- Quantum Computation and Quantum Information by Nielsen and Chuang — for the coding-theory side that the physics borrowed wholesale.
- Daniel Harlow's TASI lectures on quantum gravity (arxiv) — the cleanest introduction to the QEC-AdS/CFT dictionary.
- John Preskill's blog Quantum Frontiers — running commentary on where the field is, by one of HaPPY's authors.
- Patrick Hayden and Geoffrey Penington's lectures on entanglement and gravity (PSI / PiTP recordings) — for the island story told as code-subspace dynamics.
See Also
- concept holographic principle — the entropy bound that makes a code structure possible at all.
- concept ads cft correspondence — the duality of which holographic QEC is one feature.
- concept black hole information paradox — what the code structure was forced to resolve.
- concept spacetime from entanglement — bulk geometry as a pattern of boundary entanglement; the logical layer of the code.
- concept quantum error correction — the computing-realm cousin that lent its mathematics to gravity.
- concept distributed cognition — another system where redundant encoding survives node failure, in a very different substrate.