WorksMathematics
An isolated complex Hadamard matrix with nonzero defect
Abstract
We exhibit a complex Hadamard matrix of nonzero defect which is isolated among complex Hadamard matrices, up to Hadamard equivalence. This counterexample is realized by a Butson-type Hadamard matrix of size , with entries roots of order of unity. In the language of commuting squares, this shows that the span condition, previously introduced by the second author as a sufficient condition for isolation, is not a necessary condition for isolation.
Introduction
A complex Hadamard matrix is a matrix whose entries have absolute value 1 and whose rows are mutually orthogonal. Equivalently, is unitary and all its entries have absolute value . If is the algebra of diagonal matrices and , the Hadamard condition is equivalent to the commuting square condition for
Here the commuting square condition means that for , .
The span condition was introduced in [8] as a sufficient condition for a commuting square to be isolated, modulo isomorphisms. In the case (1), it asks that
The defect of a complex Hadamard matrix, introduced and studied by Tadej and Życzkowski [10, 11], measures the dimension of the dephased first order deformation space. These two notions are related by
Thus defect zero implies isolation. The converse asks whether every isolated Hadamard matrix has defect zero.
This converse was formulated explicitly by Banica, Özteke and Pittau [1 Conjecture 1.8]: an isolated partial Hadamard matrix should have minimal undephased defect. For a square matrix of order , their undephased defect is , and its minimal value is . Thus their conjecture asserts that isolation implies . It was still recorded as open in the 2026 review of McNulty and Weigert [5 Appendix B, following Lemma B.1], where they write: “It is presently unknown if isolation is equivalent to the span condition”.
There is also an earlier claim which must be distinguished from this open question. In his 2014 dissertation, Motta [6 Section 2.4.3, Proposition 16, p. 118] asserted that the order 9 matrix , whose defect is 2, is isolated. The section is entitled “An Isolated Matrix with Positive Defect.” Karlsson subsequently placed in a non-affine two-parameter family [2 Section 10]. The later paper of Motta and Shipman [MS, Section 4] explicitly records that “the defect arises from a non-affine 2-dimensional family” containing this matrix. Thus the order 9 example does not disprove the converse.
In this paper we give an example which does. In the following we will use the notation for the Hadamard matrices of order whose entries are -th roots of unity.
Theorem 1.1. There exists a complex Hadamard matrix which is isolated up to equivalence and satisfies . For , its spin model commuting square is isolated up to commuting square isomorphism, and
The matrix is given explicitly in Section 4. It is already present in the dataset of Lampio, Östergård and Szöllősi [4], accompanying their classification paper [3]. Our result concerns the local isolation of that matrix in the moduli space of non-equivalent complex Hadamard matrices, thus providing a counterexample to [1 Conjecture 1.8].
Our proof follows the method of sequential deformations used in [8, 9]. Assuming that is not isolated, we take a sequence of inequivalent Hadamard matrices converging to , and then a limit of the normalized differences of the corresponding unitaries. A relation between commutators gives a second order restriction on this direction of convergence. This restriction is the one in [9 Corollary 3.3(ii)] for a general commuting square. We include in this paper a direct proof for commuting squares based on Hadamard matrices. For the matrix in Theorem 1.1, the first order space is one-dimensional and its generator violates the restriction. This gives sequential isolation, without any differentiability assumptions.
AI disclosure. The AI model Astra was used to search for possible candidates for a counterexample among known databases of complex Hadamard matrices, and for verifying finite computations given by exact integer and cyclotomic identities.
Preliminaries
We denote by the trace on , and by the normalized trace. We use the notation for the norm 2 on matrices. For matrices of the same size, denotes their entrywise product (also called Hadamard product).
Two Hadamard matrices are equivalent if one can be obtained from the other by permutations of rows and columns and multiplication of rows and columns by scalars of absolute value 1. A matrix is dephased if its first row and first column consist of 1’s. Every Hadamard matrix is equivalent to a dephased one. More precisely, the map
is continuous on matrices with unimodular entries, takes values in the dephased matrices, and fixes every dephased matrix.
A Hadamard matrix is isolated up to equivalence if some neighborhood contains no inequivalent Hadamard matrix. For a dephased matrix, this is equivalent to being isolated among dephased Hadamard matrices: a fixed equivalence class has only finitely many dephased representatives, obtained by dephasing its row and column permutations.
Fix a dephased Hadamard matrix of order . Put
The dephased defect is . This is the dimension of the solution space of the linearized Hadamard equations; we do not assume that a vector in this space is the derivative of a deformation. For any real matrix , write
Then , and the equations in the second line of (5) are equivalent to .
Let , , , and . For define
A direct computation gives
Thus every gives a relation between commutators.
For completeness, we show that these formulas also prove (3). The real linear map has image , whose real dimension is : this real image consists of skew-adjoint matrices, and its complexification is . Its kernel is the undephased solution space of the first order equations. The matrices
form a -dimensional subspace of this kernel. Subtracting from gives its unique dephased representative. Hence the kernel is the direct sum of this subspace and , and rank-nullity gives (3).
Since and are abelian, is orthogonal to with respect to the trace inner product. Also , by the commuting square condition, so . Therefore (2) is equivalent to , the span condition for (1).
Directions of convergence and a second order condition
We first recall the restriction on a direction of convergence that we will use. It is the scalar-corner case of [9 Corollary 3.3(ii)]. For completeness, we reprove that result here in the case of spin model (i.e., Hadamard-based) commuting squares.
Lemma 3.1. Let be unital -subalgebras satisfying for , . Suppose are unitaries converging to such that
and, with ,
Then , . Moreover, every relation
satisfies
Proof. Put . Unitarity gives
Dividing by and using (9), we obtain . The equality follows from the definition of .
Let . The commutator relation gives , and hence . By the two commuting square relations and the trace property,
Divide this by and let tend to infinity. Since , the resulting equality is (10).
We now give a new interpretation to the restriction from the previous lemma. For , define
This is a real homogeneous cubic on .
Proposition 3.2. If a dephased Hadamard matrix is not isolated up to equivalence, there exists such that .
Proof. Choose Hadamard matrices which are not equivalent to . Applying (4), we may assume that each is dephased. Set
Then is unitary, , and . The sequence has norm 1. By finite dimensionality, after passing to a subsequence we may assume that it converges to a matrix as in Lemma 3.1.
Write . Since , we have
The equality gives
Taking limits shows that is real. The first row and column of vanish, so (13) also gives . Thus and . Since , we have ; since , we have .
Take and as in (7). By (8), and . The Hadamard condition on says precisely that and satisfy the hypothesis of Lemma 3.1. Consequently,
Now and , so
After dividing by and summing over , (14) becomes .
:::
Corollary 3.3. Suppose with . If , then is isolated up to equivalence.
Proof. Otherwise Proposition 3.2 gives a nonzero with . Write , where . Then , a contradiction.
Remark 3.4. The argument above excludes arbitrary convergent sequences of inequivalent Hadamard matrices. In particular, it excludes differential deformations whose first derivative vanishes at the initial matrix.
An isolated matrix of order thirteen
Let and define , where
This is the matrix on line 394 of matrices/BH-13-6.txt in [4]. Our argument uses this matrix, but does not depend on the completeness of the classification in [3].
Consider the real matrix
Explicit calculations
Proposition 4.1. The matrices (15) and (16) satisfy
In particular,
Proof. All computations below use exact arithmetic. In we use and . The six powers of are
Using these expressions in the row inner products gives .
To determine the defect, set
For , these are the values of and , respectively. Let be the integer matrix obtained from the equations in (5) as follows. Its columns are indexed lexicographically by , . Order the pairs by
For each pair include first the row and then the row, where
The real kernel of is precisely the space of cores of matrices in . Direct integer multiplication gives .
Delete column 144 and rows 132, 145, 146, , 156 from , with indices starting at 1, and keep the remaining rows and columns in their original order. Gaussian elimination over on this matrix gives
Thus as an integer, and . Since is in the kernel, we conclude that and .
Finally, put . Multiplication in gives
and summing the cubes of the entries of (16) gives . Since and , it follows that
□
Proof of isolation
Proof of Theorem 1.1. Proposition 4.1 gives and . Corollary 3.3 proves isolation of . Equation (3) gives .
We also verify the assertion about the commuting square. We first note that it is nondegenerate: if , then
so the products form a basis of , with .
By the local reduction in [9 Lemma 2.5], failure of isolation of would give unitaries such that the commuting squares , obtained by replacing with , are non-isomorphic to . Then are Hadamard matrices converging to . If were equivalent to , the corresponding squares would be isomorphic: right monomial multiplication leaves unchanged, and left monomial multiplication conjugates the square by a unitary normalizing . This contradicts the isolation already proved. □
For the explicit commutator relation, take and from (7). Equations (8) and (19) give
Thus the relation between commutators that makes the span condition fail is incompatible with the second order condition on any nonzero element satisfying the first order condition.
References
References
- [1]T. Banica, D. Özteke and L. Pittau, Isolated partial Hadamard matrices and related topics, Open Syst. Inf. Dyn. 25 (2018), no. 2, 1850009, 27 pp. doi:10.1142/S1230161218500099. See also arXiv:1706.00986v2, Conjecture 1.8.DOI
- [2]B. R. Karlsson, BCCB complex Hadamard matrices of order 9, and MUBs, Linear Algebra Appl. 504 (2016), 309–324. doi:10.1016/j.laa.2016.04.012.DOI
- [3]P. H. J. Lampio, P. R. J. Östergård and F. Szöllősi, Orderly generation of Butson Hadamard matrices, Math. Comp. 89 (2020), 313–331. doi:10.1090/mcom/3453.DOI
- [4]P. Lampio, P. Östergård and F. Szöllősi, Dataset for Orderly Generation of Butson Hadamard Matrices, Zenodo, March 13, 2019. doi:10.5281/zenodo.2585766. File matrices/BH-13–6.txt, line 394. License: CC BY 4.0.arxiv.org/abs/1707.02287
- [5]D. McNulty and S. Weigert, Mutually unbiased bases in composite dimensions—a review, Quantum 10 (2026), 2051. doi:10.22331/q-2026-04-01-2051. Appendix B, following Lemma B.1; see also arXiv:2410.23997v2.DOI
- [6]F. C. Motta, Discrete-time topological dynamics, complex Hadamard matrices, and oblique-incidence ion bombardment, Ph.D. dissertation, Colorado State University, 2014. Section 2.4.3, pp. 114–118, especially Proposition 16, p. 118. hdl:10217/83793.
- [8]R. Nicoară, A finiteness result for commuting squares of matrix algebras, J. Operator Theory 55 (2006), 295–310. Journal article.DOI
- [9]R. Nicoară, Limit points of commuting squares, Indiana Univ. Math. J. 60 (2011), 847–858. doi:10.1512/iumj.2011.60.4294.DOI
- [10]W. Tadej and K. Życzkowski, A concise guide to complex Hadamard matrices, Open Syst. Inf. Dyn. 13 (2006), 133–177. doi:10.1007/s11080-006-8220-2.DOI
- [11]W. Tadej and K. Życzkowski, Defect of a unitary matrix, Linear Algebra Appl. 429 (2008), 447–481. doi:10.1016/j.laa.2008.02.036.DOI