Absolutely Maximally Entangled States at New Parameter Points

Absolutely Maximally Entangled States at New Parameter Points

Introduction: symmetry to build, exact math to prove

If you’ve ever wondered whether nature can host “perfectly balanced” quantum correlations across many parties, Absolutely Maximally Entangled (AME) states are the clean mathematical answer. They’re the kind of states where every sufficiently small group looks maximally mixed—no matter which parties you pick—so the entanglement is spread as evenly as it possibly can be.

New research from Bevins and Bidav pushes the AME story forward by showing how to construct specific AME states using symmetry-guided code design, and then—crucially—how to certify them with fully exact computation. In particular, they build Hermitian self-dual maximum-distance-separable (MDS) codes that translate (via the nonbinary stabilizer construction) into AME quantum states at parameters that were previously open in the stabilizer sector.

The headline results: they explicitly construct AME states at
- AME(12,5)
- AME(18,11) and AME(18,13)
and then use a clean one-party projection trick to obtain
- AME(17,11) and AME(17,13).

What makes this paper feel “new” rather than just another existence proof is the verification style: not just clever theory, but exact field arithmetic plus complete minor enumeration and additional graph cut-rank checks, down to every single-vertex deletion in the 18-party graphs.

Why this matters: perfect entanglement isn’t just a curiosity anymore

This research is significant right now because the boundary between “quantum information theory” and “quantum engineering” is getting thinner. Systems that try to implement distributed quantum protocols—think threshold secret sharing, fault-tolerant subspace encoding, and teleportation-like primitives—need resources whose correlations are as uniform as possible. AME states sit exactly at that sweet spot: the entanglement is distributed so evenly that many error patterns and information-splitting requirements become more structured.

A concrete “today” scenario: imagine you’re designing a quantum secret-sharing protocol where no small coalition learns anything, but a sufficiently large set can reconstruct. The AME condition is tightly linked to this threshold behavior. So having explicit AME(18,11) and AME(18,13) states gives you more room to build protocols with different local dimensions (qudits), which can matter in hardware where dimension choice and noise structure aren’t arbitrary.

And the timing vs. AI-era research? The paper’s construction pipeline is partly a story about how search becomes tractable. Earlier work often leaned on classical combinatorics routes (orthogonal arrays, graph-state families, multiunitary matrices). Here, they combine symmetry with finite-field code structure so the search space collapses. That’s a similar philosophical move to what’s happening across AI research: reduce complexity by extracting the right invariances. The difference is that they don’t just use symmetry as a heuristic—they use it to generate structured templates, then prove the result with independent “stand-alone” verifiers. That’s a strong antidote to the “looks plausible” failure mode.

Hermitian self-dual MDS codes: the blueprint behind AME

At the heart of the paper is a coding-theory-to-quantum-state pipeline. Let me translate the main ingredients into an intuitive picture.

AME states as “quantum MDS codes”

An AME state AME(n, q) (local dimension q) is defined by being maximally mixed on every reduction to t = floor(n/2) parties. In code language, this is equivalent to a pure quantum MDS code of the form
[[n, 0, floor(n/2)+1]]_q.

When n is even, these are exactly the codes that saturate the quantum Singleton bound—the “best possible” distance for that dimension.

Hermitian self-duality: the algebraic “staying power”

The paper focuses on codes over F_{q^2} (a finite field extension) with a Hermitian form. A linear code C is Hermitian self-dual when C = C^{⊥h}.

In systematic form, the generator matrix looks like:
G = [I | A]
(with length 2k), and Hermitian self-duality becomes a neat matrix equation:
A A^{T} = -I.

That equation is powerful because it’s checkable entry-by-entry and because it interacts cleanly with the MDS requirement.

The MDS condition becomes “every square submatrix is nonsingular”

To guarantee MDS-ness, they use a criterion sometimes described as systematic superregularity: for a systematic generator block A, the code is QMDS if and only if every nonempty square submatrix of A has nonzero determinant.

In the paper’s exact wording: the family counts and certificates depend on how many square minors exist, and they explicitly enumerate them:
- For the 6×6 case (yielding [12,6,7]_{25}), there are 923 nonempty square minors.
- For the 9×9 case (yielding [18,9,10]_{121} and [18,9,10]_{169}), there are 48,619 nonempty square minors.

Those numbers matter because they’re not “random checks”—they’re the full brute-force guarantee that the MDS property holds.

Pipeline snapshot: what codes give what AME states?

The authors translate three classical code constructions into AME states using the nonbinary stabilizer construction. The mapping they realize is summarized below.

Then they apply one-party projection (more on that next).

How they certify AME: minors, graph cuts, and “delete every vertex”

Building a candidate is one thing. Proving it’s correct—without relying on floating-point numerics—is another. This paper is heavy on certification.

Certification type 1: complete minor enumeration for MDS

For each systematic block A, they do the straightforward-but-expensive check: enumerate all possible square row/column subsets, compute determinants with exact finite-field arithmetic, and verify all are nonzero.

The results are not just “yes/no”; the paper reports the determinant-count distributions by size, for example:
- For k=6 (the A in the 12-party code), minor sizes yield a distribution summing to 923.
- For k=9, minor size distributions sum to 48,619.

This makes the certificate extremely robust: if you change field conventions or accidentally swap a basis encoding, the verifiers would fail.

Certification type 2: graph-state cut-rank tests

Even though the AME property can be shown from code MDS/self-duality, they also convert the stabilizer description into a graph state representative. For prime local dimensions p, a qudit graph state is AME across balanced cuts if the corresponding cut-rank condition holds.

Concretely, for each subset S with |S| ≤ |Sc|, define B = Θ[S, Sc]. Then maximal mixing is equivalent to:
- rank_p(B) = |S|.

The paper counts and checks the full set of balanced cuts:
- For n=12 and q=5, there are 462 balanced cut blocks (they check every one).
- For n=18, the balanced cut count is 24,310 per field.

Certification type 3: one-party deletion checks for AME(17,q)

The AME→projection step says: if AME(2m,q) exists, then AME(2m-1,q) exists. But they don’t stop at trusting the lemma—they verify the “odd party” cases directly through graph checks.

They take each length-18 graph, delete every single vertex (so 18 deletions), and for each deletion they check all balanced cut ranks of the resulting 17-vertex graph.

They report:
- For each deletion, the number of relevant cut matrices is 437,580 rank tests over each field.

That is a huge amount of deterministic verification, and it directly supports the existence of:
- AME(17,11)
- AME(17,13)

Why this multi-layer certification is a big deal

You might ask: “Why both minors and graph cuts?” Because they test the property in two different mathematical presentations:
- coding/MDS conditions on A blocks
- entanglement conditions on graph cut ranks derived via local Clifford/graph reduction

If both routes agree, you’re extremely unlikely to have made a hidden convention error.

Symmetry-guided construction: shrinking the search from impossible to manageable

The other “secret sauce” in this work is how they find the codes without drowning in combinatorics.

Stage I: length 12 search without prescribed symmetry

For AME(12,5) they do a systematic search in systematic form:
G = [I6 | A] over F_{25}
with constraints enforcing Hermitian self-duality A A^{T} = -I6 and incrementally enforcing the nonsingularity of square minors.

They emphasize this important point: the [12,6,7]25 code was found without prescribing coordinate symmetry. Once found, its automorphism group reveals a hidden structure (more below). The search stopped at the first solution—so it’s not a classification of all such codes, just an explicit constructive proof.

Stage II: length 18 search using symmetry transfer

Once they know the automorphism group of the length-12 code, they use its coordinate action to guide a much bigger length-18 search.

Here the idea is: instead of searching over a completely general 9×9 Hermitian constraint matrix A, impose that A respects a regular action of H = Z3 × Z3 on two nine-coordinate halves.

This forces the 9×9 block into a group-algebra (circulant-like) structure:
A_{x,y} = a(y−x)
so only nine field elements (the kernel a) define the full 9×9 matrix.

That’s the conceptual leap: from “choose every entry” to “choose a tiny kernel, everything else follows.”

The role of the -q pairing and character decomposition

The paper then diagonalizes the action using character blocks of the group algebra. When you do this, the Hermitian constraint splits into independent equations on blocks.

This matters because it turns one big coupled constraint into several smaller constraint problems.

They also find a parity obstruction phenomenon: for certain blocks, if the block dimension multiplicity is odd, the self-dual constraint has no solution. In their invariant-template family, they use exactly this logic to predict when a symmetry template can even work.

Field-specific simplification at q=13 using norm-one scaling

For q=13, they add another symmetry quotient: coordinate scaling by norm-one elements of F_{q^2}. This reduces the number of distinct solutions to consider by dividing out an extra factor of q+1, which at q=13 is 14.

In other words, there’s a “gauge freedom” they mod out, and they keep the classification counts consistent with that.

The exact constructions: the explicit kernels behind AME(12,5), AME(18,11), AME(18,13)

The paper gives concrete generator matrices (through systematic blocks) defined over specific finite fields with explicit polynomial-basis conventions. That’s important: it’s part of the reproducibility certificate.

The [12,6,7]_{25} code defining AME(12,5)

They work in F_{25} = F_5(α) with α^2 + α + 2 = 0, then encode field elements as integers with a + 5b ↔ a + bα.

The code generator is:
- G5 = [I6 | A5]
where A5 is an explicit 6×6 matrix (printed in the paper). They verify:
- Hermitian self-duality: A5 A5^{T} = -I6
- MDS: all 923 nonempty square minors of A5 are nonzero

This yields:
- C5 = [12,6,7]25
- hence AME(12,5).

They also confirm that this [12,6,7]25 code is not monomially equivalent to a generalized Reed–Solomon code, using a Schur-square dimension argument: the Schur-square dimension over F_{25} is 12, which is too large for a generalized Reed–Solomon structure.

A shared symmetry-driven Z3×Z3 template at q=11 and q=13

For both q=11 and q=13, they use the same symmetry template group H = Z3×Z3, ordered by pairs (u,v).

The 18-party generator matrices are:
- Gq = [I9 | Aq]
and the full 9×9 matrix is determined by a kernel aq(u,v) via:
- Aq_{x,y} = aq(y−x).

They implement two different field setups:
- q=11: F_{121} = F_{11}(γ) with γ^2 = -1
- q=13: F_{169} = F_{13}(β) with β^2 + β + 2 = 0

For each field, they verify:
- Hermitian self-duality: Aq Aq^{T} = -I9
- MDS property: all 48,619 square minors are nonzero

So they get exactly:
- AME(18,11) from [[18,0,10]]_{11}
- AME(18,13) from [[18,0,10]]_{13}

One-party projection gives AME(17,11) and AME(17,13)

By Lemma 3 (one-party projection), projecting one vertex of an AME(18,q) graph state gives an AME(17,q) state.

But again: they don’t just cite it—they check all single-vertex deletions, confirming the cut-rank rank conditions on the 17-vertex graphs for both q=11 and q=13.

Key Takeaways

Key Takeaways

• New explicit AME states are constructed and certified at:
  • AME(12,5)
  • AME(18,11), AME(18,13)
  • and by projection: AME(17,11), AME(17,13).
• The core method builds Hermitian self-dual MDS codes C=[n,k,d] that translate into AME via nonbinary stabilizer machinery.
• Certification isn’t hand-wavy: they use exact finite-field arithmetic, complete square-minor enumeration (923 minors for the 12-party case and 48,619 for the 18-party cases), and independent graph-state cut-rank checks on all balanced cuts.
• Symmetry is used constructively:
  • they find the 12-party code without preset symmetry,
  • then read off an automorphism action (a regular Z2^3 orbit structure),
  • and impose a Z3×Z3 invariant template to shrink the 18-party search to a 9-parameter kernel.
• For practical use (e.g., distributed quantum protocols and secret sharing), this adds explicit high-party, high-dimensional entanglement resources with verifiable correctness—not just theoretical “might exist” claims.

If you want, I can also rewrite the symmetry transfer story as a step-by-step “search reduction” narrative (what changes from the length-12 stage to the length-18 stage), or explain how the one-party projection shows up directly in the graph-state polynomial picture.