Quantum information · Theory

Maximally Entangled Mutually Unbiased Bases in Bipartite Systems of Unequal Prime Dimensions

Bogdan S. Damski, Karol Życzkowski

arXiv:2611.18232v1 ↗PDF ↗Submitted 13 November 2026

Question
Two bases are mutually unbiased when every vector of one has squared overlap \(1/d\) with every vector of the other. For two systems of distinct prime dimensions \(p<q\), tensor products give \(p+1\) such bases, all made of product vectors. How many mutually unbiased bases can consist entirely of maximally entangled vectors? Lower bounds existed only for special pairs of dimensions, usually from case-by-case constructions.
Result
For every pair of distinct primes \(p<q\), an explicit formula gives \(p\) bases of maximally entangled vectors that, together with the computational basis, are mutually unbiased, so \(M(p,q)\ge p\). They come from applying one permutation, a controlled modular shift, to the standard product bases. Each new basis is a complex Hadamard matrix equivalent to the Fourier matrix \(F_{pq}\), and the matrices relating consecutive bases are given in closed form by quadratic Gauss sums. For \(p=2\), \(q=3\) the resulting three bases cannot be extended by a fourth.

The paper in detail

Mutually unbiased bases (MUBs) are bases in which a measurement in one reveals nothing about the outcome in another: every overlap between vectors of different bases has squared modulus \(1/d\). In prime and prime-power dimensions complete sets of \(d+1\) are known; for composite dimensions such as 6 the question is open, and Zauner conjectured that no more than three exist there. In \(\mathcal H_p\otimes\mathcal H_q\) one always gets \(\min(p,q)+1\) bases by tensoring local MUBs, but they consist of product vectors.

Entanglement enters through Lubkin's formula: averaged over a complete set, the purity of a subsystem's reduced state is \((d+d')/(1+dd')\). If a complete set in \(\mathcal H_2\otimes\mathcal H_3\) contained three product bases, the other four would need average purity \(1/2\), the minimum, so all their vectors would be maximally entangled. This motivates the question of how many bases made entirely of maximally entangled vectors can be mutually unbiased, a number written \(M(d,d')\).

The construction starts from the eigenbases of \(XZ^r\) in each factor, written so that the basis index \(t\in\mathbb Z_{pq}\) is shared through the Chinese Remainder Theorem, \(t\leftrightarrow(t\bmod p,t\bmod q)\). For \(r=1,\dots,p\) this gives \(p\) product bases that, with the computational basis, are mutually unbiased. The authors then apply the permutation \(\ket a\otimes\ket b\mapsto\ket a\otimes\ket{p^{-1}(b-a)\bmod q}\), which maps the computational basis to itself, so unbiasedness is preserved, and the new vectors take the single closed form of Eq. (33).

A bipartite pure state is maximally entangled exactly when its coefficient matrix satisfies \(CC^\dagger=I/p\). For the new vectors the off-diagonal entries reduce to a geometric sum over the second factor that vanishes because \(r(p+q)(a-a')\) is not divisible by \(q\) when \(r\le p<q\) and \(a\neq a'\). The matrices relating consecutive bases are quadratic Gauss sums over \(\mathbb Z_{pq}\), which the Chinese Remainder Theorem factors into Gauss sums over \(\mathbb Z_p\) and \(\mathbb Z_q\); this shows that they, and the bases themselves, are complex Hadamard matrices equivalent to the Fourier matrix.

For dimension 6 the three bases are unextendible, by an earlier proof that no triple of product MUBs in dimension 6 admits even one further unbiased vector, a property a common unitary preserves. For larger primes, whether the \(p+1\) bases extend, and whether prime powers or more than two parties admit similar closed forms, are left open.

Transferable methods

  1. A controlled modular shift that entangles while preserving unbiasedness

    Any unitary applied to all bases at once preserves mutual unbiasedness. A permutation of the computational basis keeps that basis in the set, and a controlled shift that mixes the two labels turns product vectors into entangled ones. Maximal entanglement then reduces to the vanishing of a linear exponential sum.

    With \(\eta_q(p)\) the inverse of \(p\) modulo \(q\) and \(\tau_d=-e^{i\pi/d}\), the permutation \(P_{pq}(\ket a\otimes\ket b)=\ket a\otimes\ket{\eta_q(p)(b-a)\bmod q}\) maps the product bases to \[\ket{\Psi(r,t)}=\frac1{\sqrt{pq}}\sum_{a=0}^{p-1}\sum_{b=0}^{q-1}\tau_{pq}^{(p+q)[r(a+pb)^2-2t(a+pb)]}\ket a\otimes\ket b,\qquad r=1,\dots,p.\] The proof writes \(j_q\equiv j_p+pb\pmod q\) and uses the congruences of the phase exponents modulo \(p\) and \(q\).

    For the coefficient matrix \(C_{ab}\) of \(\ket{\Psi(r,t)}\), the sum over \(b\) in \((CC^\dagger)_{aa'}\) is a Gauss sum with no quadratic term, \(G(0,r(p+q)(a-a'),q)\), which equals \(q\) if \(q\) divides \(r(p+q)(a-a')\) and 0 otherwise. Since \(p+q\equiv p\pmod q\) and \(0<|a-a'|<p<q\), \(1\le r\le p\), it vanishes unless \(a=a'\): \[(CC^\dagger)_{aa'}=\frac1p\,\delta_{aa'}.\]

    Editorial check: constructing the bases numerically for six pairs of primes from \((2,3)\) to \((5,7)\) confirms orthonormality, squared overlaps \(1/pq\) between all \(p+1\) bases, and \(CC^\dagger=I/p\) for every vector, to \(10^{-12}\). For \((2,3)\), Lubkin's average \(5/7\) is reproduced by three product bases of purity 1 and four of purity \(1/2\).

    How to reuse it

    To trade product structure for entanglement while keeping all overlaps, act on a whole set of bases with a permutation that fixes one basis and is controlled across subsystems; checking maximal entanglement then reduces to showing that a linear character sum vanishes, which is a divisibility condition.

  2. Gauss sums factored by the Chinese Remainder Theorem

    To show that the new bases are Fourier-type and to classify them, the authors compute the transition matrices between consecutive bases. Their entries are quadratic Gauss sums over \(\mathbb Z_{pq}\), which split into Gauss sums over the prime factors with known moduli and phases.

    After a change of summation variable, the transition matrix \(V=(H_r^{\mathrm{ent}})^\dagger H_{r+1}^{\mathrm{ent}}\) has entries \[\bra tV\ket s=\frac1{pq}\sum_{n=0}^{pq-1}\tau_{pq}^{(p+q)[n^2+2(s-t)n]}.\] For odd \(p,q\), \((p+q)/2\) is an integer, so the phase is \(\omega_{pq}^{\eta_{pq}(2)(p+q)[n^2+2(s-t)n]}\), and by the Chinese Remainder Theorem the sum factors as \[G\bigl(\eta_{pq}(2)(p+q),(p+q)(s-t),pq\bigr)=G\bigl(\eta_p(2),s-t,p\bigr)\,G\bigl(\eta_q(2),s-t,q\bigr).\]

    A quadratic Gauss sum over a prime with nonzero quadratic coefficient has modulus equal to the square root of the prime, so every entry of \(V\) has modulus \(1/\sqrt{pq}\): \(V\) is a complex Hadamard matrix. Explicitly \(V\sim\Delta F_{pq}^\dagger\Pi\Delta\) with a diagonal phase matrix \(\Delta\) and the permutation \(\Pi\ket j=\ket{j(p+q)\bmod pq}\), so \(V\) is equivalent to the Fourier matrix. For \(p=2\) the sum splits into even and odd \(n\), giving a two-term formula.

    Editorial check on the Butson classes: after removing a global phase, the entries of each \(H_r^{\mathrm{ent}}\) are \(pq\)-th roots of unity for odd \(p\), and for \(p=2\) they are \(4q\)-th roots when \(r=1\) and \(2q\)-th roots when \(r=2\); the transition matrices for \(p=2\) are \(4q\)-th roots. This matches the classes \(\mathrm{BH}(pq,pq)\), \(\mathrm{BH}(2q,4q)\) and \(\mathrm{BH}(2q,2q)\) derived in the proof.

    How to reuse it

    For constructions on \(\mathbb Z_N\) with squarefree \(N\), quadratic phases factor across the prime components, so Hadamard and unbiasedness properties can be checked prime by prime with the closed-form Gauss sums.

Source: Section III, Theorems 3 and 4 (Eqs. (31)–(52)). Based on the linked paper version. Derivations marked as editorial, and suggested uses in other problems, are editorial guidance rather than results of the paper.

Context & limitations

Why it matters. It is the first construction of maximally entangled unbiased bases valid for all pairs of unequal primes, using only Weyl–Heisenberg eigenvectors, the Chinese Remainder Theorem and Gauss sums, and it includes dimension 6, where whether a complete set of seven bases exists is a famous open problem. Unbiased bases with prescribed entanglement are used in tomography, key distribution and entanglement detection.

Limitations. The construction gives \(p+1\) bases, far from a complete set of \(pq+1\), and only \(p\) of them entangled; whether larger sets or extensions exist beyond \(p=2\), \(q=3\) is open. It covers prime local dimensions only. Editorial note: the statement of Theorem 4(ii) and (iii) gives the Butson order as \(1+\delta_{p,2}\delta_{r,1}\) and \(1+\delta_{p,2}\); the proof, and a direct computation, show these must be multiplied by \(d=pq\).

Description based on the full arXiv HTML of v1, read 2026-11-17: Sections I–IV, including the proofs of Theorems 1, 3 and 4. Appendix A (Gauss-sum identities) and Appendix B (explicit examples) were not read. Editorial checks: the bases of Eq. (33) were built for \((p,q)=(2,3),(2,5),(2,7),(3,5),(3,7),(5,7)\); all are orthonormal, all pairs among the \(p+1\) bases have squared overlaps \(1/pq\), and every coefficient matrix satisfies \(CC^\dagger=I/p\), to \(10^{-12}\). The entries are \(pq\)-th roots of unity up to a global phase for odd \(p\), and \(4q\)-th (\(r=1\)) or \(2q\)-th (\(r=2\)) roots for \(p=2\).