The joint numerical range of three hermitian $4\times 4$ matrices

This paper investigates the joint numerical range of three $4\times 4$ Hermitian matrices by classifying fifteen non-generic boundary structures with explicit examples, establishing that intersections of distinct one-dimensional faces form corner points, and comparing the standard range with a separable subset defined via a tensor product structure to study quantum entanglement.

The Gist

Imagine you have a magical 3D sculpture made of clay. This sculpture represents all the possible outcomes you could get if you measured three different properties of a tiny quantum system (specifically, a system made of two "qubits," which are the basic units of quantum information).

In the world of mathematics and physics, this sculpture is called the Joint Numerical Range. The paper you are asking about is a deep dive into the shape of this sculpture when it is made from 4x4 matrices (a specific size of mathematical grid).

Here is a breakdown of what the authors discovered, using simple analogies:

1. The Smooth vs. The Bumpy

Usually, if you pick three random matrices, your 3D sculpture will be perfectly smooth, like a polished marble ball or an egg. It has no flat spots and no sharp corners. The authors call this the "generic" case.

However, the paper is interested in the non-generic cases—when the sculpture gets weird. Sometimes, the smooth surface gets flattened into a flat face (like the side of a box), or it gets a sharp corner (like the tip of a pyramid).

2. The Four Shapes of "Flat Spots"

The authors found that when these flat spots appear, they aren't just random shapes. They fall into exactly four distinct categories, which they named based on what they look like:

• The Oval (Type 0): A smooth, flat oval patch, like a coin lying on a table.
• The Loaf (Type 1): A shape that looks like a loaf of bread. It has a flat side, but the ends are rounded.
• The Droplet (Type 2): A shape that looks like a teardrop or a raindrop. It has a flat side and a sharp point at the bottom.
• The Triangle (Type 3): A flat triangular face, like the side of a pyramid.

3. The Great Classification (The 15 Rules)

The main achievement of the paper is a "rulebook" for how these shapes can mix and match.

Imagine you are building a 3D object using these four types of flat faces. The authors asked: "How many of each shape can we have on one object?"

They discovered there are exactly 15 possible combinations (or "classes") of these shapes that can exist together on the surface of this 3D sculpture.

• You could have just one "Oval."
• You could have two "Loaves."
• You could have a "Droplet" and a "Triangle."
• You could even have a full Tetrahedron (a pyramid with four triangular faces).

For every single one of these 15 combinations, the authors didn't just say "it's possible"; they actually built a specific mathematical example (a set of three matrices) that creates exactly that shape. It's like they provided the recipe for every possible flavor of ice cream cone in existence.

4. The "Corner" Rule

The paper also found a strict rule about corners.
If you look at the sculpture and see three different flat lines (edges) meeting at a single point, that point must be a sharp corner. You cannot have three edges meeting at a point unless that point is a "corner" of the shape.

• Analogy: Think of a tent. If three poles meet at the top, that top point is the peak (a corner). You can't have three poles meeting at a point on the side of the tent without it being a peak.
• The Catch: The authors proved this rule works perfectly for their 4x4 system. However, if you tried to do this with a bigger system (5x5 matrices), the rule breaks! You could have three edges meeting at a point that isn't a sharp corner. This highlights that the 4x4 system is special and unique.

5. The "Separable" vs. "Entangled" Split

The paper also introduces a second, smaller sculpture inside the big one.

• The Big Sculpture (W): Represents all possible quantum states, including "entangled" states (where two particles are linked in a spooky, non-classical way).
• The Small Sculpture (Separable Range): Represents only the "classical" or "separable" states (where the particles act independently).

The authors compared the surfaces of these two sculptures. They found that the "classical" (small) sculpture often has ruled patches — picture a tight rubber sheet stretched over a convex blob that has several protrusions: where the sheet is pulled taut between the bumps, flat triangular and ruled regions naturally appear. The "all-inclusive" (big) sculpture, by contrast, tends to be smoother. Crucially, the small sculpture sits INSIDE the big one, and every flat face of the big sculpture lies tangent to the small one — the big set's flat faces always come right up against the surface of the small set, which is how the two are linked geometrically.

Summary

In short, this paper is a geometric census of a specific type of 3D quantum shape.

1. It identified 4 types of flat faces (Oval, Loaf, Droplet, Triangle).
2. It listed 15 specific ways these faces can combine to form a 3D object.
3. It proved a strict rule about how corners form when edges meet.
4. It compared the "full quantum" shape with the "classical" shape to see how they differ.

The authors didn't just guess these shapes; they provided the exact mathematical "ingredients" (matrices) to build every single one of the 15 possibilities.

Technical Summary

Technical Summary: Joint Numerical Ranges of Three Hermitian 4 × 4 Matrices

Problem Statement
This paper investigates the geometry of the joint numerical range (JNR), denoted $W(A_1, A_2, A_3)$, for three Hermitian matrices of order $n=4$. The JNR is defined as the set of expectation values of these matrices over the set of density matrices:
$$W := \{(\text{tr}(\rho A_1), \text{tr}(\rho A_2), \text{tr}(\rho A_3))^\top : \rho \in M_4, \rho \ge 0, \text{tr}(\rho) = 1\}.$$
While the JNR for $n=3$ matrices was previously classified into 10 boundary structures, the case $n=4$ is distinct because 4 is a composite number ($4 = 2 \times 2$). This allows for the introduction of a tensor product structure ($\mathbb{C}^4 = \mathbb{C}^2 \otimes \mathbb{C}^2$), corresponding to a two-qubit system. This structure enables the definition of a "separable numerical range," a subset of $W$ relevant to quantum entanglement. The primary goal is to classify the non-generic structures of the boundary $\partial W$, specifically focusing on non-elliptic flat portions (faces), and to analyze the relationship between the standard JNR and the separable JNR.

Methodology
The authors employ a combination of algebraic geometry, convex analysis, and matrix theory:

1. Algebraic Framework: The study utilizes Bonsall and Duncan's notion of the joint numerical range with respect to a $*$-algebra. The faces of the JNR are analyzed via "access sequences" of orthogonal projections. A face $F$ corresponds to a unique projection $P$ such that $F$ is the image of the density matrices of the subalgebra $P M_4 P$.
2. Reduction to Lower Dimensions: Using unitary similarities and affine transformations, the authors reduce the analysis of a rank-$r$ face of the $4 \times 4$ JNR to the joint numerical range of $r \times r$ matrices. Specifically, non-elliptic faces (2D faces that are not elliptic disks) are shown to be rank-3 faces, which are isometric to the JNR of three $3 \times 3$ Hermitian matrices.
3. Kippenhahn's Classification: The shapes of these reduced $3 \times 3$ numerical ranges are classified using Kippenhahn's curve theory. The paper identifies four types of non-elliptic faces based on the Kippenhahn curve of the associated $3 \times 3$ matrix:
   • Type 0: Oval (degree-6 curve).
   • Type 1: Loaf (degree-4 curve).
   • Type 2: Droplet (ellipse and a point outside).
   • Type 3: Triangle (three affinely independent points).
4. Combinatorial Classification: By analyzing the intersection properties of these faces (specifically that any two distinct non-elliptic faces intersect along a segment, and three mutually distinct ones intersect at a corner point), the authors derive a combinatorial classification of all possible configurations of non-elliptic faces.
5. Separable Range Analysis: For the separable numerical range $W^{\text{sep}}$, the authors utilize the Peres-Horodecki (PPT) criterion. They employ semidefinite optimization to numerically approximate the boundary of $W^{\text{sep}}$ and compare it with the standard JNR.

Key Contributions and Results

• Classification of Non-Elliptic Faces: The paper identifies 15 distinct classes of joint numerical ranges based on the number and types of non-elliptic faces on their boundaries. These classes are defined by the tuple $(a_0, a_1, a_2, a_3)$, representing the count of Type 0, 1, 2, and 3 faces, respectively. The authors provide explicit examples of Hermitian $4 \times 4$ matrices for each of the 15 classes.
• Corner Point Theorem: A significant theoretical result (Theorem 5.10) states that if three mutually distinct one-dimensional faces of $W$ have a non-empty intersection, that intersection is a corner point of $W$. This property is specific to $n=4$; the paper provides a counterexample for $n=5$ where such an intersection is not a corner point.
• Structure of the Boundary: The authors confirm that in the generic case, the boundary of $W$ is smooth and strictly convex. Non-generic cases involve flat portions. The paper establishes that the number of non-elliptic faces is constrained; for instance, a JNR cannot have exactly three triangular faces (it must be a tetrahedron with four).
• Separable vs. Standard Numerical Range: The paper compares $W$ with $W^{\text{sep}}$. It proves that every rank-2 or rank-3 face of the standard JNR intersects the separable numerical range. Numerical experiments suggest that the boundary of the separable numerical range often contains singular features (cusps, flat faces, or ruled surfaces) even for generic matrices, contrasting with the smooth boundaries often found in the standard JNR.
• Elliptic Faces: The paper discusses elliptic faces (rank-2 faces). Citing Ottem et al., it notes that for generic real symmetric matrices, the number of elliptic faces is even and at most 10. However, the authors demonstrate that non-transversal spectrahedra can yield odd numbers of elliptic faces (e.g., 1, 3, or 5) if some of these faces are actually rank-3.

Significance
The paper provides a complete combinatorial classification of the possible boundary structures for the joint numerical range of three $4 \times 4$ Hermitian matrices, a step beyond the previously solved $n=3$ case. This classification is directly motivated by quantum information theory, as the $4 \times 4$ case corresponds to two-qubit systems. By distinguishing between the standard JNR and the separable JNR, the work offers a geometric framework for understanding the constraints imposed by entanglement on the set of achievable expectation values. The identification of corner points formed by the intersection of one-dimensional faces provides a specific geometric signature for certain quantum states. The results serve as a foundation for further studies on the geometry of quantum states and the detection of entanglement via numerical ranges.