🔢 mathematics

Using topology to reveal all mutually unbiased bases in dimension six

By leveraging the topological connectedness of complex Hadamard matrices and interval arithmetic, this paper classifies all mutually unbiased bases in dimension six, proving that no set of four such bases exists.

Original author: Mateo Cárdenes Wuttig, Joseph Tindall

Published 2026-11-05

📖 4 min read🧠 Deep dive

CC BY 4.0

Original author: Mateo Cárdenes Wuttig, Joseph Tindall

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the author. For technical accuracy, refer to the original paper. Read full disclaimer

In the quantum world, the way we measure a system is just as fundamental as the system itself. Imagine trying to describe a spinning coin. If you look at it from the top, you see a clear head or tail. If you look from the side, you see a blur of motion. In quantum physics, these are like different "bases" of measurement. A special kind of measurement setup, called a mutually unbiased basis, ensures that if you are absolutely certain about the outcome of one measurement, you have zero information about the outcome of another. It is a state of perfect ignorance between perspectives. For systems with a number of states that is a prime number or a power of a prime, physicists know exactly how many of these completely different perspectives can exist. But for a system with six states, the answer has remained a stubborn mystery for over thirty years.

For decades, researchers knew how to find three such perspectives in a six-state system, but they could not prove whether a fourth one could ever exist. This question sits at the heart of understanding the limits of quantum information, with implications for how we secure data and test the fundamental laws of nature. Recently, a team of researchers used a new approach combining topology—the study of shapes and spaces—with rigorous computer verification to settle the matter. They mapped the entire landscape of possible mathematical structures that could support these measurements and found that the space is far more constrained than anyone imagined. Their work proves that in a six-state quantum system, it is impossible to have more than three mutually unbiased bases.

The researchers began by translating the problem of finding these measurement perspectives into the language of complex Hadamard matrices. These are special grids of numbers where every entry has a size of one, and the columns are perfectly balanced against each other. Finding a new measurement perspective is equivalent to finding a new matrix that fits perfectly with the existing ones. For thirty years, scientists had searched for these matrices, finding many families of them, but they could never be sure they had found every single possibility. Some matrices seemed to stand alone, while others formed continuous families that could be smoothly transformed into one another. The question was whether any of these could be extended to form a set of four.

To solve this, the team treated the collection of all possible matrices as a single, continuous shape. They discovered that almost all of these matrices, except for one very specific and isolated type, form a connected space. This means you can travel from any one matrix to any other within this group by making small, continuous changes, without ever hitting a dead end or a gap. This connectedness was the key. It allowed the researchers to break the vast, infinite space of possibilities into a manageable set of overlapping regions, like covering a globe with a finite number of map patches. They created 2,775 such patches, each centered on a specific mathematical seed, and used interval arithmetic—a method that tracks every possible rounding error in a calculation—to verify that these patches cover the entire space without missing a single point.

Inside each of these patches, the team performed a detailed search to see if a fourth measurement perspective could exist. They looked for the specific mathematical conditions required to add a fourth basis to a set of three. In every single patch they examined, the math showed that no such fourth basis could be constructed. The search was exhaustive and verified by independent computer programs to ensure absolute certainty. They found that every possible set of three bases belongs to one of two known types: a continuous family that can be described by two changing parameters, or a single, isolated set built from a standard Fourier pattern. Crucially, neither of these types can be extended to include a fourth basis.

The result is a complete classification of all possible mutually unbiased bases in dimension six. The researchers proved that the maximum number is exactly three. This confirms a long-standing conjecture that had puzzled the field for decades. It also reveals a surprising rigidity in the quantum world: while a pair of complementary measurements offers a great deal of flexibility, allowing one to be smoothly deformed into any other, adding a third measurement locks the system into a very specific structure. Adding a fourth is simply impossible. This finding closes a major chapter in quantum information theory, establishing dimension six as the first non-prime-power dimension where the full set of these measurement limits is known. It shows that the universe, even in its smallest quantum details, has strict boundaries on how many completely different ways we can look at it.

Technical Summary: Using Topology to Reveal All Mutually Unbiased Bases in Dimension Six

Problem Statement
The paper addresses the long-standing open problem of determining the maximum number of Mutually Unbiased Bases (MUBs), denoted $N(d)$ , in a $d$ -dimensional Hilbert space. While $N(d) = d+1$ is known to exist when $d$ is a prime or a prime power, the case for composite non-prime-power dimensions remains difficult. Dimension six ( $d=6$ ) is the smallest such dimension. It is known that three MUBs exist in $\mathbb{C}^6$ , but whether a fourth exists has been unresolved for over three decades. Previous attempts have ruled out specific families of MUB triples or used numerical searches, but a complete classification of all possible MUB sets in dimension six remained elusive. A recent computer-assisted proof by OpenAI suggested $N(6)=3$ , but it relied on a nine-dimensional cover of Hadamard matrices and did not fully classify the structure of the triples.

Methodology
The authors employ a topological approach centered on the space of complex Hadamard matrices, which describe the transitions between MUBs. The methodology proceeds in three main stages:

Minimal Four-Phase Reconstruction and Connectedness:
The authors simplify a previous reconstruction algorithm to show that every order-six complex Hadamard matrix, except for Tao's isolated matrix, can be constructed from a "strict seed" defined by four phases $\theta = (\theta_1, \theta_2, \theta_3, \theta_4)$ . They define a reconstruction invariant $R(\theta)$ , a trigonometric polynomial in these phases.
- They prove that the region $\Omega = \{\theta : R(\theta) > 0\}$ is topologically connected. This is achieved by identifying 18 positive local maxima of $R(\theta)$ and constructing explicit paths connecting them while maintaining $R(\theta) > 0$ .
- They demonstrate that the space of equivalence classes of non-Tao Hadamard matrices, denoted $X$ , is connected. This space is formed by the continuous image of the strict seeds (which covers the Karlsson family and approaches the Fourier matrix $F_6$ as a limit) plus the isolated Fourier class. Tao's matrix is shown to be an isolated point distinct from this connected space.
Phase-Ball Coverage:
Leveraging the connectedness of $X$ , the authors cover the entire space (excluding Tao's matrix) with a finite union of 2,775 overlapping closed "seed balls" in the four-dimensional phase space, plus a separate chart around the Fourier matrix $F_6$ .
- Using interval arithmetic, they certify that every point within these balls corresponds to a valid Hadamard matrix with a specific orientation (matched completion).
- They verify that the union of these balls and their symmetric images (conjugates, transposes) is open and closed, thereby covering the entire non-Tao space $X$ .
Local Classification and Exclusion:
Inside each certified ball, the authors perform rigorous local tests to classify MUB triples and exclude quartets:
- Triple Classification: For every pair of bases $\{I_6, H/\sqrt{6}\}$ where $H$ lies in a ball, they search for a third basis. They prove that any existing triple must possess an order-three symmetry (a unitary $T$ with $T^3=I_6$ ). Solving the constraints for this symmetry reveals that all such triples belong to either the known continuous two-parameter family or the isolated Fourier triple.
- Quartet Exclusion: Using a graph-search algorithm (similar to OpenAI's approach but applied locally within the four-dimensional balls), they search for four mutually orthogonal candidate vectors that could form a fourth basis. They prove that no such pattern exists in any of the certified regions.

Key Contributions and Results

Complete Classification of MUB Sets in Dimension Six: The paper proves that, up to equivalence, every non-empty set of MUBs in $\mathbb{C}^6$ is one of the following:
1. A single basis.
2. A pair consisting of the identity and a Hadamard matrix (Tao's matrix, the Fourier matrix $F_6$ , or a matrix from the continuous family).
3. A triple belonging to the continuous two-parameter family or the isolated Fourier triple.
- Crucially, no set contains four or more bases.
Proof of $N(6)=3$ : By showing that all possible triples are of known types (which cannot be extended to a quartet) and directly excluding quartets in the phase-ball cover, the authors provide a rigorous proof that $N(6)=3$ . This resolves Zauner's conjecture for dimension six.
Topological Characterization of Hadamard Matrices: The work establishes that the space of order-six complex Hadamard matrices (excluding Tao's) is connected and can be navigated by varying just four phases. This reduces the global classification problem to a finite number of local checks.
Verification of the Fourier-Vanishing Conjecture: The reconstruction identities derived in the paper provide an analytical proof of the Fourier-vanishing conjecture (that certain row/column moments vanish for non-Tao matrices), a result previously only supported by computer-assisted proofs.
Resolution of Constellation Problems: The classification implies that the constellation $\{5, 5, 5, 1\}_6$ (four groups of vectors with specific unbiasedness properties) is empty, resolving a specific open problem regarding mutually unbiased constellations.

Significance and Claims
The authors claim that dimension six is now the first non-prime-power dimension in which all sets of MUBs are classified. The significance of the work lies in:

Rigorous Completeness: Unlike previous numerical searches that could miss isolated solutions, this topological proof covers the entire space of possibilities (except the isolated Tao matrix, which is handled separately) using a finite, verified cover.
Independence from Prior AI Proofs: While the authors utilized AI assistance for development, their proof is logically independent of the recent OpenAI proof. It relies on a new topological result (connectedness) and a more efficient cover (2,775 balls in 4D vs. 249,878 balls in 9D).
Formal Verification: The results are supported by formal proofs in Lean and independent interval-arithmetic checkers (Arb and Julia), ensuring the mathematical validity of the numerical bounds and exclusion tests.
Methodological Advancement: The paper demonstrates that topological connectedness combined with certified local analysis is a viable strategy for solving high-dimensional classification problems in quantum information theory, potentially applicable to higher dimensions like ten or twelve.

The paper concludes that the freedom for complementary measurements in dimension six shrinks as more measurements are added, and a fourth mutually unbiased measurement is mathematically impossible, even under ideal conditions.

More like this