On the quantum separability of qubit registers

Abstract

I show that the bipartite separability of a pure qubit state hinges critically on the combinatorial structure of its computational-basis support. Boolean cube geometry is used to introduce a taxonomy that distinguishes support-guaranteed separability from cases in which entanglement depends on probability amplitudes. I provide closed-form support counts, identify forbidden configurations that enforce multipartite entanglement, and show how these results can enable fast entanglement diagnostics in quantum circuits. This framework offers immediate utility in classical simulation, entanglement-aware circuit design, and quantum error-correcting code analysis. This establishes support geometry as a practical and scalable tool for understanding entanglement in quantum information processing.

1 Introduction

Quantum entanglement is typically introduced through the physical localization of the quantum states in terms of quantum systems being in or having those states. However, this phenomenon, which is indeed fundamentally nonlocal (Bell, 1964; Aspect et al., 1982), is not even required to demonstrate nonlocality (Wang et al., 2025). Fundamental quantum mechanics have been constructed directly outside classical physics and even outside general classical thinking (Mugur-Schachter, 2008), and the mathematical concepts of quantum state entanglement and separability can be examined separately from the system’s physical attributes. This bypasses the cybernetic problem (Gershenson, 2025) of defining a system and its boundaries, which is the root of various quantum paradoxes, of which Schrödinger’s cat is perhaps the most prominent example, as it requires a box in which a cat is localized.

This study examines the bipartite separability conditions of pure qubit quantum registers as rays in Hilbert spaces devoid of any spatial boundaries. Although a quantum state does not need to be described by qubits, any -level quantum state (pure or mixed) for can be represented as a state on qubits via a substitution.

I consider the states separable only across certain bipartitions, as well as those separable only for certain values of their probability amplitudes, and I provide their distributions with respect to their support sizes. The results can be applied in quantum computational applications.

2 Methods

We can consider Boolean space, , as a complete graph constructed upon -cube, where a distinct index and a distinct address can be assigned to each vertex (Łukaszyk, 2025a).

Consider a pure quantum register containing qubits in the computational basiswhere are the probability amplitudes (PA; we sometimes write them as ), are the basis components (kets in Dirac notation), and are their addresses. While the standard computational basis, with kets corresponding to the vertices of -cube (Duff, 2013), applies naturally to quantum circuits, quantum error correction codes, and so forth, it can also be considered a physical aspect of nature (Łukaszyk, 2025b).

The support of the state (Equation 1) is the set of basis kets for which , and the support size is the cardinality of this set. The maximum support sizes for one to four qubits are shown in Figure 1. There are distinct supports for qubits starting from one-ket states (single vertices of an -cube) and ending on a state containing all kets (all vertices of the -cube).

FIGURE 1

A pure quantum state is called separable if and only if the state can be written as a tensor productof at least two quantum states, which can be represented in the computational basis as similar qubit registersconsisting respectively of and qubits, where and .

Otherwise, is an entangled state, and the degree of its entanglement can be measured using various methods. One of them employs the entanglement (von Neumann) entropy (here in bits)where , are the eigenvalues of the reduced density matrixobtained by tracing out a specified set of, respectively, or qubits from the density matrix of the state . If the state is separable, . Otherwise, if , the state is inseparable, and for it is maximally entangled. The eigenvalues of the density matrix (Equation 4) can also be used to express the state using the Schmidt decomposition aswhere and are orthonormal bases of states and . Hence, the state (Equation 5) is separable if for some and then , . The entanglement between the states and is invariant under a unitary operation acting on the individual states and ; entanglement entropy (Equation 3) remains constant after such an operation.

States that can be written as tensor products of other pure states are called -separable (Dür and Cirac, 2000). However, in this study, we focus on the bipartite separability condition (Equation 2) of pure -qubit quantum registers across at least one bipartition and classify states according to the number and type of bipartitions across which they are separable, considering either equal or arbitrary, normalized PAs.

We explicitly exclude mixed states from the analysis as they lack unique computational-basis support and define entanglement via convexity rather than support geometry, thereby requiring methods designed for mixed states, such as the Peres–Horodecki criterion or entanglement witnesses. Furthermore, mixed states depend on the notion of classical probability.

3 Results

Proof. The total number of subsets of a set containing qubits is , including the empty set and this set itself. We have to exclude these two inseparable cases and also consider the symmetry of partitioning. Hence, the upper bound on the number of unique bipartitions for is given by . An entanglement across one bipartition reduces the cardinality of the set to qubits that can be similarly partitioned across bipartitions. Finally, an entanglement across all bipartitions can be thought of as a set containing only one element and is thus inseparable.

For example, the register of four qubits can be separable at most across bipartitions as , , , , , , and , where “” denotes the bipartition. However, an entanglement across one bipartition, say , reduces the cardinality of this set to three , , and .

In the following definition, we can give meaning to the free parameter we introduced in Lemma 1.

Geometrically, a support is an -dimensional coordinate face of the Boolean hypercube, which under the Segre embedding (Cirici et al., 2021) corresponds to a coordinate-fixed face, where coordinates are constant.

States that are separable across all bipartitions for all PAs are called fully separable (Dür et al., 1999). We redefine them in the context of bipartitions.

For example, the following state with supportis an APS state as it is separable across all seven bipartitions for all PAs , . As an APS state admits a superposition of, at most, one of its qubits, only states spanned over the vertices and 1-edges of cube are APS states. If two or more qubits vary across the support, then some bipartition encounters differing values on both sides, forcing amplitude-dependent constraints. As -cube has faces, a quantum register hasAPS states (vertices and blue edges in Figure 1).

The APS states introduce another definition.

A state with a support is separable across bipartitions for all PAs. Therefore, a state with a support is inseparable for all PAs.

Proof. Consider a set containing all distinct binary strings of length . bitstrings from this set can serve as a support of an -qubit quantum register. All one-element supports (vertices) are the APS states. The remaining supports can have common bit(s) in the same position(s). There are ways to choose a subset of positions from positions and ways to assign bits to these positions.

The remaining positions must ensure that, in none of them, do all strings agree. The strings are identical in the positions, so choosing them is equivalent to selecting distinct vectors from Boolean space such that no coordinate in these vectors is constant for all vectors. The total number of such subsets is , but we need to subtract cases where at least one coordinate is constant, which can be done using the inclusion–exclusion principle. Hence, we sum over , the number of coordinates forced to be constant. There are ways to choose a subset of coordinates from coordinates and ways to assign bits to these coordinates. The effective space size becomes , so the count for those is , with sign . The inclusion–exclusion provides the sum in Formula 6, where for if the binomial coefficient is defined in terms of a falling factorial. Formula 6 counts the number of ways bitstrings can be selected from the set of all bitstrings of length , so that each of these strings has only bit(s) in common in the same position(s), completing the proof.

For example, for (cube), (all possible edges) and (only the face diagonals), Formula 6 takes the formsumming all six edges on each face of a cube (including face diagonals), subtracting four blue 1-edges having two common bits in the same position, and multiplying the result by six faces of the cube to count the states , etc. (green edges in Figure 1c).

The distributions of states are listed in Table 1 as functions of the support size and summed in Table 2 along with the values given by Formula 6 for and . They were numerically cross-validated for by calculating the eigenvalues of the reduced density matrices (Equation 4) for each of quantum states corresponding to distinct supports for each of possible bipartitions, assuming equal or arbitrary PAs. The Schmidt decomposition (Equation 5) certifies that a given state is separable along a given bipartition if for some . For example, the statehas the following eigenvalues of the reduced density matrix (Equation 4)and is thus separable only across the partition , while for the remaining two partitions, it has a fractional, weak entanglement entropy (Equation 3) .

Proof. The number of such supports is given by summing all factors given by Formula 6 over . Thussimplifies to the formula of OEIS sequence A131288 for .

For example, for , the set has power set , and there are seven ways to choose such a collection of subsets of with full union and empty intersection corresponding to the states inseparable across all bipartitions for all PA values (Table 3).

Proof. Since out of bits are the same in all basis kets of the support, the remaining bits must be diversified in all the kets, and the maximum number of such sequences is . supports correspond to Hamming-weight-constrained subcubes.

Proof. This follows from substituting into Equation 6. For example, there are supports having such a maximum support size, defined by -dimensional facets of cube, as they have the largest support size for that can be partitioned across the same bipartition.

For example, the maximum support size for the state of three qubits is , and there are six states of the formin this case separable only across the bipartition for all normalized PAs.

Certain states are separable only for specific PAs. Therefore, we introduce the last definition.

The PAs’ matrix is an outer product of one column and one row matrices of PAs of the states and of the tensor product (Equation 2)

States and supports of Definitions 1-4 are listed in Figure 2. Unlike APS and states, the separability of the is not a property of the support alone: the support merely permits separability, which is realized only when the PA matrix factors as a rank-1 outer product.

FIGURE 2

Proof. The smallest composite number is 4, so two qubits achieve the minimum support size of an ADS state, . Larger support sizes must be either even to provide separability across at least one bipartition or composite numbers satisfying the inequalities and for .

For example, even though , : the support size of two qubits is at most 4, while five kets are required for the second state in the product . Such states correspond to genuinely multipartite entangled (Palazuelos and Vicente, 2022) or fully inseparable (Dür et al., 1999) states, such as Bell, Greenberger–Horne–Zeilinger (GHZ), or W-states.

Proof. We have to exclude states separable across bipartitions for all PAs (if any, i.e., for ), from the larger set containing also the states separable across bipartitions both for all and for specific PAs to find PAs’ specific bipartitions. In other words, indexes the depth of amplitude-dependent separability.

By Lemma 7, and APS states are mutually exclusive. A support of a state has an inherent classification, defining its separability for arbitrary PAs. That same state may also become separable across additional partitions (that were previously entangled across) if its PAs are fine-tuned.

It is always possible to adjust the PAs of the matrix of an state so that it has rank 1. In particular, equal PAs make the matrix constant, which factors as an outer product of two all-1 vectors and scaled by , and any such outer-product matrix is rank 1.

Possible support sizes of ADS states for are shown in Figure 3 along with forbidden sizes violating the inequality (Equation 9). Figure 3 also shows support sizes providing separability across different numbers of bipartitions. For example, a four-qubit state with (gray zone) can be a , , state separable across one bipartition or a , state separable across three bipartitions.

FIGURE 3

For example, the PAs of the statecan be written as an outer productas it is an state separable across the bipartition asif and only ifthat is, when the columns of the matrix (Equation 10) are linearly dependent or, equivalently, if its rank is 1. Similarly, the PAs of the statecan be written as an outer productand it is also the state separable only across the bipartition asagain, if all the rows and columns of the matrix (Equation 11) are linearly independent (its rank is 1). The state (omitting kets here for clarity)is a support separable across the bipartition , and it is also a state separable across two bipartitions and if the matrixhas a rank of 1; that is if

Any support having the maximum support size has basis kets with bits differing only in the same u : positions (i.e., spanning vertices of -cube). In this case, we can define as the PA corresponding to the ket being the bitwise complement of the relevant positions of the ket associated with a PA , and the separability condition for the PAs and isIn other words, the products of the PAs associated with the vertices defining all the longest diagonals of the -cube must be equal. In particular,

For example, the statehas two common bits in the same positions 1 and 2, so it is a support separable across bipartitions , , or for all PAs. But ifit is also an state separable across four bipartitions , , , and , separable asFor the same reasons, the state (Equation 7) is separable across all three bipartitions after swapping PAs with .

For example, the following state supported on four basis kets with two qubits spanned over four vertices of the 2-cubehas a support separable across three bipartitions , , and . However, if , it is also an state separable across four bipartitions , , , and .

A three-qubit register (we omit kets here for clarity)can be an state separable across one bipartition if the matrixis rank 1, which is equivalent toacross one bipartition if the matrixis rank 1, and across one bipartition if the matrixis rank 1. It can also be the state separable across all bipartitions if —that is, if the products for all four main 3-cube diagonals are equal. The four-qubit registercan be an state separable across one of four possible bipartitions ( is shown below) if the matrix is factorable asan state separable across one of three bipartitions ( is shown below) if the matrix is factorable asand an state separable across all seven bipartitions if all PAs are equal.

A 2-cube (square) is the smallest -cube that provides support size for the state. Even though one qubit (1-cube, segment) is always separable, the separability condition (Equation 12) can be extended to the case where it implies the equality of two PAs (Equation 13). This is a specific case of the condition of equal superposition of a qubit: the statewith a vanishing relative phase between the basis kets. On the complex plane, represents the same point on the circle of radius , while represents all points of this circle.

4 Applications

In general, determining across which bipartition (if any) a state is separable is computationally exponential. The computational complexity for finding all eigenvalues of a Hermitian matrix of size is —in our case, for the maximum number of qubits of the state or in the tensor product (Equation 2), yielding . Calculating the reduced density matrix also requires operations. This yields an exponential complexity of for all bipartitions to check. On the other hand, determining the distribution of bits in the same positions across basis kets of an -qubit state required to classify a support of a state as has a computational complexity of only . The speedup thus grows exponentially with for all as a complexity ratio. It is particularly useful, however, in the sparse-support regime , where the support geometry is informative, as is typical for many structured, post-selected, or shallow-circuit states.

The support structure alone provides immediate separability bounds, allowing one to bypass expensive density-matrix computations. In particular, is a direct parameter of an entanglement confinement, as the state always has the form ofcontaining classical (i.e., non-superposed) qubits up to relabeling.

Furthermore, states having support sizes of Lemma 6 cannot be realized by any separable bipartition even if PAs are equal. As a consequence, a uniform superposition over any 15 computational basis states in four qubits, for example, must be entangled across every possible bipartition—a genuinely multipartite entangled state (Palazuelos and Vicente, 2022). This provides an extremely simple way to construct states with guaranteed full multipartite entanglement by choosing any support of a forbidden size (e.g., with equal PAs).

Hence, the support taxonomy does not replace Schmidt ranks; rather, it restricts their necessity to ADS states.

Further exemplary applications of the introduced taxonomy are provided below.

4.1 Tracking entanglement spreading in quantum circuits

Consider the standard textbook circuit that creates an -qubit GHZ state which starts with the APS state, applies the Hadamard gate to the first qubit to obtain the support (still APS state), and then applies a chain of CNOT gates to arrive at the final state, as shown below.

Each operation reduces the number of common bits by exactly one and delocalizes the entanglement to one additional qubit. The taxonomy proposed in this study, therefore, gives an exact, single-integer metric that tracks how entanglement spreads through the CNOT chain—something that would otherwise require expensive Schmidt decompositions across all bipartitions. This is particularly useful for analyzing fault-tolerance thresholds in GHZ preparation or distribution: if an error occurs when is still large, the entanglement is still localized to a small block, so the error may be easier to correct locally. This is illustrated in Equation 15, where the fourth qubit which erroneously flipped after the first CNOT gate is corrected before the second one.

4.2 Classical simulation speedup for localized-entanglement states

The check for is vastly cheaper than computing entanglement entropies across all bipartitions, making it a practical pre-filter for tensor-network or exact simulators: “if some threshold, reduce to qubits”.

In summary, the taxonomy turns the hypercube support into a powerful diagnostic tool that is both theoretically clean and practically cheap, especially for tracking entanglement dynamics in circuits, thus enabling aggressive classical simulation when entanglement stays localized and constructing fully guaranteed inseparable states. The taxonomy enables both the detection of entanglement properties in existing states and the design of states with desired entanglement characteristics.

5 Conclusion

Classifying quantum states using -cube separability structures provides a fast ( vs. ) structural way to assess entanglement without full-state tomography or diagonalization. Other potential applications in quantum computing include quantum machine learning and data encoding, quantum circuit optimization, quantum communication and network protocols, and quantum-to-classical boundary studies.

References

• 1

  AspectA.GrangierP.RogerG. (1982). Experimental realization of Einstein-podolsky-rosen-bohm gedankenexperiment: a new violation of bell’s inequalities. Phys. Rev. Lett.49 (2), 91–94. 10.1103/PhysRevLett.49.91

  • CrossRef
  • Google Scholar
  • View reference in article

• 2

  BellJ. S. (1964). On the Einstein podolsky rosen paradox. Phys. Phys. Fiz.1 (3), 195–200. 10.1103/PhysicsPhysiqueFizika.1.195

  • CrossRef
  • Google Scholar
  • View reference in article

• 3

  CiriciJ.SalvadóJ.TaronJ. (2021). Characterization of quantum entanglement via a hypercube of Segre embeddings. Quantum Inf. Process.20 (7), 252. 10.1007/s11128-021-03186-x

  • CrossRef
  • Google Scholar
  • View reference in article

• 4

  DuffM. J. (2013). “Black holes and qubits,” in What is known and unexpected at LHC. erice-sicily, Italy (World Scientific), 57–66. Available online at: https://www.worldscientific.com/doi/abs/10.1142/9789814522489_0003 (Accessed March 12, 2016).

  • Google Scholar
  • View reference in article

• 5

  DürW.CiracJ. I. (2000). Classification of multiqubit mixed states: separability and distillability properties. Phys. Rev. A61, 042314. 10.1103/PhysRevA.61.042314

  • CrossRef
  • Google Scholar
  • View reference in article

• 6

  DürW.CiracJ. I.TarrachR. (1999). Separability and distillability of multiparticle quantum systems. Phys. Rev. Lett.83 (17), 3562–3565. 10.1103/PhysRevLett.83.3562

  • CrossRef
  • Google Scholar
  • View reference in article

• 7

  GershensonC. (2025). Self-organizing systems: what, how, and why?Complexity2 (1), 10. 10.1038/s44260-025-00031-5

  • CrossRef
  • Google Scholar
  • View reference in article

• 8

  ŁukaszykS. (2025a). Four cubes: on the necessity of four-dimensional perception. IPI Lett., 3 (3), 18–33. 10.59973/ipil.242

  • CrossRef
  • Google Scholar
  • View reference in article

• 9

  ŁukaszykS. (2025b). Black hole merger as an event converting two qubits into one. Front. Quantum Sci. Technol.4, 1656200. 10.3389/frqst.2025.1656200

  • CrossRef
  • Google Scholar
  • View reference in article

• 10

  Mugur-SchachterM. (2008). On a crucial problem in probabilities and solution. arXiv. 10.48550/arXiv.0801.2654

  • CrossRef
  • Google Scholar
  • View reference in article

• 11

  PalazuelosC.VicenteJ. I. D. (2022). Genuine multipartite entanglement of quantum states in the multiple-copy scenario. Quantum6, 735. 10.22331/q-2022-06-13-735

  • CrossRef
  • Google Scholar
  • View reference in article

• 12

  WangK.HouZ.QianK.ChenL.KrennM.AspelmeyerM.et al (2025). Violation of bell inequality with unentangled photons. Sci. Adv.11 (31), eadr1794. 10.1126/sciadv.adr1794

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article