---
title: Mutually Orthogonal Quantum Latin Squares
url: https://www.emergentmind.com/topics/mutually-orthogonal-quantum-latin-squares
type: topic
---

# Mutually Orthogonal Quantum Latin Squares

Searching arXiv for recent and foundational papers on mutually orthogonal quantum Latin squares.
Mutually orthogonal quantum Latin squares are quantum-combinatorial analogues of classical mutually orthogonal Latin squares in which symbols are replaced by vectors in Hilbert space and orthogonality is imposed through tensor-product basis conditions rather than symbol-pair uniqueness. A quantum Latin square of order \(n\) is an \(n\times n\) array of unit vectors in \(\mathbb C^n\) such that every row and every column forms an orthonormal basis of \(\mathbb C^n\); two such squares are orthogonal when the \(n^2\) vectors obtained by tensoring corresponding entries form an orthonormal basis of \(\mathbb C^n\otimes\mathbb C^n\) [1804.04042]. In this sense, mutually orthogonal quantum Latin squares (MOQLS) extend the classical theory of MOLS while retaining a stringent product-state geometry. Their study connects combinatorial design theory, unitary error bases, mutually unbiased maximally entangled bases, quantum orthogonal arrays, and multipartite entanglement [1804.04042], [2111.04055].

## 1. Definition and basic formalism

A quantum Latin square (QLS) of dimension \(n\) is an \(n\times n\) array
\[
\bigl\{\,|\Psi_{ij}\rangle \in \mathbb C^n \mid i,j\in [n]\,\bigr\},
\]
with \([n]=\{0,1,\dots,n-1\}\), such that each row and each column is an orthonormal basis of \(\mathbb C^n\) [1804.04042]. Equivalently, for each fixed row index \(i\), the set \(\{|\Psi_{ij}\rangle\}_j\) is an orthonormal basis, and for each fixed column index \(j\), the set \(\{|\Psi_{ij}\rangle\}_i\) is an orthonormal basis. Classical Latin squares embed into this framework by replacing each symbol with a computational-basis vector [1804.04042].

For two QLS \(\Phi\) and \(\Psi\) of dimension \(n\), the standard orthogonality notion is the tensor-basis criterion
\[
\left\{\,|\Phi_{ij}\rangle \otimes |\Psi_{ij}\rangle \;\middle|\; i,j\in [n]\,\right\}
\]
forming an orthonormal basis of \(\mathbb C^n\otimes \mathbb C^n\) [1804.04042]. Equivalently,
\[
\langle \Phi_{ij}|\Phi_{pq}\rangle\, \langle \Psi_{ij}|\Psi_{pq}\rangle = \delta_{ip}\delta_{jq}
\qquad \text{for all } i,j,p,q
\]
[1804.04042]. A family of \(m\) QLS is mutually orthogonal if it is pairwise orthogonal in this sense [1804.04042].

The literature also contains a distinct notion based on bipartite or multipartite array entries, where a “pair of orthogonal quantum Latin squares” is itself an array of bipartite states and row and column sums satisfy 1-uniformity conditions [1708.05946]. This entangled notion is not identical to the product-state MOQLS formalism above. The distinction is mathematically consequential, especially at order \(6\), where the entangled setting behaves differently from the non-entangled one [2603.02334].

## 2. Orthogonality notions and their unification

A central development in the theory was the simplification of orthogonality for QLS. Earlier work of Goyeneche, Raissi, Di Martino, and Życzkowski used a larger tensor expression with partial traces. A later formulation proved that the essential content reduces exactly to the tensor-basis condition above, and that the extra partial-trace conditions are redundant [1804.04042]. In particular, the simplified criterion is equivalent to the earlier GRMZ notion, but only one trace condition carries substantive information [1804.04042].

The same simplification extends from pairs to families. Pairwise orthogonality is sufficient; no genuinely higher-arity trace condition is needed for MOQLS families [1804.04042]. This aligns the subject with classical MOLS, where mutual orthogonality is likewise pairwise.

A separate line of work developed “weak orthogonality” for QLS in connection with mutually unbiased maximally entangled bases. There, for QLS \(\mathcal P,\mathcal Q\), weak orthogonality requires that for all \(i,j\), there exists a unique \(t\) such that
\[
\sum_{k=0}^{n-1} |k\rangle \langle Q_{ki}\mid P_{kj}\rangle = |t\rangle
\]
[1605.08919]. In the classical case this is equivalent to left orthogonality of Latin squares, and hence to ordinary orthogonality up to left conjugation [1605.08919]. This notion underlies a different MUB-oriented construction and should not be conflated with the product-state MOQLS definition used in the modern QLS literature [1605.08919], [1804.04042].

The distinction between product-state orthogonality and entangled orthogonality is equally important. In the entangled framework of quantum combinatorial designs, entries may be bipartite or multipartite states that do not factor into separate QLS, and orthogonality is encoded by global basis orthogonality together with 1-uniformity of line sums [1708.05946]. A plausible implication is that several statements in the literature that use the phrase “orthogonal quantum Latin squares” are only comparable after fixing which orthogonality notion is intended.

## 3. Structural results and bounds

The quantum theory preserves a key extremal feature of classical MOLS: any family of MOQLS of dimension \(n\) has size at most \(n-1\) [1804.04042]. The proof normalizes first rows via equivalence transformations and then extracts \(m+1\) linearly independent vectors in \(\mathbb C^n\), forcing \(m\le n-1\) [1804.04042]. Thus quantization does not enlarge the maximal family size beyond the classical bound.

This upper bound has stronger consequences in certain regimes. It is known that if equality holds, the family is classical [2603.02334]. The order-six nonexistence work further notes that if there exist \(n-2\) MOQLS\((n)\), then there exist \(n-2\) MOLS\((n)\), hence \(n-1\) MOLS\((n)\), and therefore a projective plane of order \(n\) [2603.02334]. This suggests that near-maximal MOQLS families are forced back into classical incidence geometry.

Equivalence transformations also play a structural role. Orthogonality is preserved under common row and column permutations together with individual unitaries and phase multipliers on each QLS [1804.04042]. In the non-classical existence theory, two QLS \(\Phi\) and \(\Psi\) are equivalent if there exist a unitary \(U\), phases \(c_{ij}\), and permutations \(\sigma,\tau\) such that
\[
|\Psi_{ij}\rangle = c_{ij} U |\Phi_{\sigma(i),\tau(j)}\rangle
\]
for all \(i,j\) [2507.20154]. A QLS equivalent to one arising from a classical Latin square is called classical; otherwise it is non-classical [2507.20154].

A useful diagnostic for non-classicality is that classical QLS have only inner products of modulus \(0\) or \(1\) among entries. Hence any QLS exhibiting an entry overlap of modulus strictly between \(0\) and \(1\) is genuinely quantum [2111.04055]. This criterion is used repeatedly in explicit constructions of non-classical orthogonal pairs [2111.04055].

## 4. Existence theory and explicit constructions

The first explicit pair of orthogonal QLS not equivalent to any pair of orthogonal classical Latin squares was constructed in dimension \(9\) [1804.04042]. That work also proved that the example is genuinely nonclassical by showing that one square cannot be transformed into a computational-basis Latin square via a single global unitary, phases, and permutations [1804.04042]. The same paper’s constructive novelty is at the level of pairs rather than larger nonclassical MOQLS families [1804.04042].

A more systematic existence theory was later developed for non-classical MOQLS using combinatorial design methods [2507.20154]. That work introduced idempotent QLS, self-orthogonal QLS, holey quantum Latin squares, and orthogonality on these variants, then applied PBD constructions and filling-in-holes constructions to obtain broad asymptotic existence theorems [2507.20154]. The principal results include:

| Structure | Existence claim | Exception range stated |
|---|---|---|
| non-classical \(2\)-idempotent MOQLS\((v)\) | exists for \(v\ge 6\) | except possibly \(\{6,7,8,9,10,11,12,14,15,18,19,23\}\) |
| non-classical \(2\)-MOQLS\((v)\) | exists for \(v\ge 4\) | except possibly \(\{4,5,6,7\}\) |
| non-classical \(3\)-MOQLS\((v)\) | exists for \(v\ge 4\) | except possibly \(\{4,5,\dots,15\}\) |
| non-classical SOQLS\((v)\) | exists for \(v\ge 13\) | none stated beyond threshold |

These results place MOQLS existence into the recursive-combinatorial tradition of classical MOLS, but with local non-computational bases injected into chosen subspaces to force non-classicality [2507.20154].

An earlier construction paper already provided direct product methods for MOQLS and lower bounds for the maximal number \(M(d)\) of mutually orthogonal non-classical QLS of dimension \(d\) [2111.04055]. If there exists a \(2\)-MOQLS\((d_1)\) and a \(2\)-MOQLS\((d_2)\), then there exists a \(2\)-MOQLS\((d_1d_2)\) by tensoring corresponding entries [2111.04055]. More generally, if there exists a \(t_j\)-MOQLS\((d_j)\) for each \(j\), then there exists a \(t\)-MOQLS\((d_1d_2\cdots d_l)\) with \(t=\min\{t_1,\dots,t_l\}\) [2111.04055].

The same work transferred classical MOLS existence into quantum existence. If \(d=d_1d_2\cdots d_l\), \(l\ge 2\), and \(m(d_j)\ge 2\) for all \(j\), then there exists \(t\)-MOQLS\((d)\) with \(t=\min\{m(d_1),\dots,m(d_l)\}\) [2111.04055]. It also proved lower bounds such as
\[
M(d)\ge \min\{p_i^{r_i}-1:1\le i\le s\}
\]
when \(d=p_1^{r_1}\cdots p_s^{r_s}\) with \(s\ge 2\), together with broad existence statements outside explicit exceptional sets \(E_2,E_3,E_4\) [2111.04055].

## 5. Incomplete, generalized, and entangled variants

Incomplete quantum Latin squares (IQLS) were introduced as quantum analogues of incomplete Latin squares with holes indexed by mutually orthogonal subspaces \(V_1,\dots,V_n\subset \mathbb C^d\) [2111.04055]. Orthogonality for two IQLS is defined by requiring their superimposed tensor states to form an orthonormal basis of
\[
(\mathbb C^d\otimes \mathbb C^d)\setminus \bigoplus_{i=1}^n (V_i\otimes V_i)
\]
[2111.04055]. These incomplete objects support hole-filling constructions that produce full QLS, MOQLS, and self-orthogonal QLS [2111.04055], [2507.20154].

Generalized mutually orthogonal quantum Latin squares (GMOQLS) enlarge the framework further by allowing each cell to contain a \(t\)-partite pure state
\[
|\psi_{i,j}\rangle \in (\mathbb C^d)^{\otimes t}
\]
rather than a product of \(t\) local vectors [2111.04055]. The defining conditions require orthogonality of all cells, row and column quantum-Latin conditions after tracing out \(t-1\) parties, and a global support condition after tracing out \(t-2\) parties [2111.04055]. If every entry is fully separable, GMOQLS reduce to ordinary MOQLS [2111.04055].

This generalized theory is equivalent to quantum orthogonal arrays of size \(d^2\). Specifically,
\[
QOA(d^2,t+2,d,2)
\quad\Longleftrightarrow\quad
t\text{-GMOQLS}(d)
\]
[2111.04055]. The corresponding multipartite state
\[
|\Phi\rangle = \sum_{i,j=0}^{d-1} |ij\rangle\otimes |\Phi_{i,j}\rangle
\]
is then 2-uniform [2111.04055]. This places MOQLS within a broader entanglement-theoretic architecture in which ordinary MOQLS appear as the fully separable corner of a larger family of quantum combinatorial designs.

A different but related entangled framework from quantum combinatorial designs gives, for every \(d\ge 2\), a triple of entangled MOQLS of size \(d\) derived from generalized Bell states
\[
|\phi_{i,j}\rangle= \sum_{l=0}^{d-1}\omega^{il}|l+j,l\rangle
\]
and equivalent to a \(QOA(d^2,3_C+2_Q,d,2)\) [1708.05946]. In that setting the resulting designs are “essentially quantum,” with entangled entries that need not decompose into separate QLS [1708.05946]. This suggests that the entangled and non-entangled theories should be read as parallel, not interchangeable, generalizations.

## 6. Order six, Euler’s problem, and current interpretation

Order \(6\) is the critical case in the subject because it sharpens the distinction between non-entangled MOQLS and entangled quantum Latin structures. A recent theorem proves:
\[
\text{There does not exist a pair of orthogonal quantum Latin squares of order six.}
\]
[2603.02334]. This is the product-state quantum analogue of the classical nonexistence of orthogonal Latin squares of order \(6\), and it establishes that “quantization without entanglement does not defeat Euler’s obstruction” [2603.02334].

The proof combines pattern analysis, local-unitary normalization, graph-theoretic reformulation via orthonormal representations of complements of Latin square graphs, classification of Latin squares of order six up to paratopy, and targeted computer-assisted elimination [2603.02334]. A key reduction shows that if two MOQLS\((6)\) existed, then one could assume one of them classical [2603.02334]. This reduction allows the remaining problem to be phrased in terms of orthonormal representations of complements of Latin square graphs in \(\mathbb C^6\) [2603.02334].

The same paper emphasizes the sharp contrast with entangled quantum Latin squares of order \(6\), which do exist and are equivalent to an \(\mathrm{AME}(4,6)\) state [2603.02334]. Thus the slogan that “the thirty-six quantum officers, if they exist, must be entangled” is literal in this context [2603.02334].

Single QLS of order \(6\) nevertheless remain relevant as potential test cases. Two explicit QLS of order \(6\) with cardinalities \(13\) and \(17\) were constructed in 2026 [2605.15540]. The first uses a direct-sum decomposition
\[
\mathbb C^6 = \mathbb C^4 \oplus \mathbb C^2,
\]
while the second is built from two-dimensional Hadamard pairs supported on coordinate planes [2605.15540]. The paper does not prove any orthogonality theorem, but it exposes structural features—support patterns, repeated vectors up to phase, and direct-sum or coordinate-plane geometry—that are directly relevant to MOQLS searches [2605.15540].

For the cardinality-\(13\) square, the repeated-use pattern is severe: only \(13\) phase-classes occur across \(36\) cells [2605.15540]. This implies that any orthogonal partner would face strong packing constraints, because repeated vectors in one square force orthogonality relations among the matching entries of the partner [2605.15540]. For the cardinality-\(17\) square, the geometry is local to coordinate \(2\)-planes, suggesting a different search strategy based on support partitions and local Hadamard blocks [2605.15540]. A plausible implication is that explicit low-cardinality QLS can serve more as obstruction witnesses and structured ansätze than as direct evidence toward existence of MOQLS at order \(6\).

## 7. Applications and related frameworks

Orthogonal QLS and MOQLS have several direct applications in quantum information. Orthogonal QLS yield quantum codes, and the same principle extends to quantum Latin isometry squares, from which one obtains error-detecting codes and a characterization of unitary error bases [1804.04042]. In that generalized setting, orthogonality of isometry squares leads to an encoding tensor
\[
T:=\sum_{i,j=0}^{n-1} |i\rangle \otimes q_{ij}k_{ij}^\dagger \otimes |j\rangle
\]
that detects a single error [1804.04042].

Weakly orthogonal QLS also support the construction of mutually unbiased maximally entangled bases in square dimension [1605.08919]. Given two families of Hadamards \(H_k\) and \(G_j\), and a pair of weak orthogonal QLS \(\mathcal P,\mathcal Q\), the bases \(B(\mathcal Q,H_k)\) and \(B(\mathcal P,G_j)\) are mutually unbiased [1605.08919]. The construction strictly generalizes the Beth–Wocjan construction because it allows quantum Latin squares rather than classical Latin squares and allows indexed families of Hadamards rather than a single fixed Hadamard [1605.08919].

The broader quantum-combinatorial design framework connects MOQLS, quantum orthogonal arrays, and \(k\)-uniform states. In particular, a family of \(m\) MOQLH gives a \(k\)-uniform state
\[
|\phi\rangle=\sum_{i_1,\dots,i_k}|i_1,\dots,i_k\rangle|\varphi_{i_1,\dots,i_k}\rangle
\]
on \(N=k+m\) parties [1708.05946]. For the square case \(k=2\), this relationship underlies the role of MOQLS and GMOQLS in constructing 2-uniform and absolutely maximally entangled states [1708.05946], [2111.04055].

A structural perspective from quantum magic squares is also relevant. Quantum Latin squares correspond to rank-one quantum magic squares, and the semiclassical quantum Latin squares are precisely those constructed from classical Latin squares [2209.10230]. This clarifies that the classical-to-quantum embedding is exact at the semiclassical level, while the full set of quantum Latin squares is strictly larger [2209.10230]. This suggests that any genuinely quantum MOQLS theory must extend beyond the semiclassical subclass.

## 8. Conceptual landscape and common misunderstandings

A common misconception is that any “quantum solution” of a Latin-square orthogonality problem automatically belongs to the MOQLS framework. Order \(6\) disproves this: entangled quantum Latin squares of order \(6\) exist, but non-entangled MOQLS of order \(6\) do not [2603.02334]. The resource that changes the answer is entanglement, not merely superposition.

Another misconception is that orthogonality of quantum Latin squares is intrinsically multipartite or requires higher-order trace constraints. In the non-entangled QLS framework, orthogonality is simply the pairwise tensor-basis condition, and the more complicated GRMZ-style family conditions collapse to ordinary pairwise orthogonality [1804.04042].

It is also inaccurate to treat all uses of “orthogonal quantum Latin squares” as equivalent across the literature. At least three notions coexist: the product-state MOQLS notion used in modern QLS theory [1804.04042], the weak-orthogonality notion developed for MUB constructions [1605.08919], and the entangled orthogonality notion for quantum Latin arrangements and quantum orthogonal arrays [1708.05946]. These notions agree only in restricted classical or separable settings.

Finally, the existence theory for non-classical MOQLS should not be read as implying arbitrary abundance in every order. There are strong positive results outside explicit exceptional sets [2507.20154], [2111.04055], but order \(6\) is now known to be a sharp nonexistence point for non-entangled pairs [2603.02334]. The subject therefore exhibits both asymptotic abundance and low-dimensional rigidity.

## 9. Outlook

The modern theory of mutually orthogonal quantum Latin squares has settled several foundational issues. The product-state orthogonality notion is now conceptually minimal and equivalent to earlier cumbersome formulations [1804.04042]. Explicit non-classical orthogonal pairs exist [1804.04042]. Broad recursive existence results for non-classical \(2\)- and \(3\)-MOQLS are available through PBD and filling-in-holes constructions [2507.20154]. The generalized entangled theory is tied precisely to quantum orthogonal arrays and 2-uniform states [2111.04055], and order \(6\) has been sharply separated into impossible non-entangled MOQLS versus possible entangled analogues [2603.02334].

Open structure remains. The order-\(7\) question “Are 2 MOQLS(7) classical?” is explicitly identified as unresolved [2603.02334]. More broadly, the theory still lacks a full classification of genuinely quantum MOQLS in low orders, and explicit constructions beyond pairs remain comparatively sparse in the non-entangled framework [1804.04042]. The recent appearance of highly structured QLS of order \(6\) with controlled cardinalities and transparent geometric support patterns suggests that future progress may rely on combining combinatorial recursion with fine-grained Hilbert-space geometry, especially support partitions, Hadamard rotations, and low-cardinality obstruction analysis [2605.15540].

Source: https://www.emergentmind.com/topics/mutually-orthogonal-quantum-latin-squares