Minimum Entanglement Drop

• Minimum Entanglement Drop is a multipartite entanglement measure that captures the reduction in global one-to-group entanglement when a single qubit is traced out.
• It employs tangle and negativity formulations to ensure monotonicity and detect sensitivity to particle loss, especially distinguishing biseparable states.
• Applications include analysis of W and graph states under noise, and it provides a structural probe for experimental diagnostics via randomized measurements.

Searching arXiv for recent and foundational papers on “Minimum Entanglement Drop” and closely related formulations. Minimum Entanglement Drop (MED) is a multipartite entanglement quantity defined by the reduction in global one-to-group entanglement when a single particle is traced out, followed by a minimization over which particle is removed. In its explicit modern formulation, one fixes a focus qubit $q_1$, compares the entanglement across the bipartition $q_1\mid(q_2\cdots q_n)$ before and after tracing out one other qubit, and takes the smallest resulting decrease. This construction is intended to quantify sensitivity to particle loss and thereby probe genuine multipartite structure (Dong et al., 20 Jun 2026). The phrase has also appeared in adjacent literatures with different meanings: as worst-case entanglement generation by a quantum gate rather than loss (Chen et al., 2012), as the minimum noise cost required to erase entanglement completely (Berta et al., 2017), or as suppression of avoidable scattering-induced entanglement (Nicolaou et al., 2019). In the strict terminological sense, however, MED refers to the loss-sensitivity monotone introduced for multipartite qubit systems (Dong et al., 20 Jun 2026).

1. Formal definition

For an $n$-qubit state with focus qubit $q_1$, the relevant baseline quantity is the global one-to-group entanglement across the bipartition $q_1\mid(q_2\cdots q_n)$. The paper first introduces a monogamy-motivated precursor, the minimum residual entanglement,

$$\mathcal{R}_{q_1} = \min_{i\in\{2,\dots,n\}} \sqrt{\tau_{q_1(q_2\dots q_n)} - \tau_{q_1q_i} - \tau_{q_1(q_2\dots \hat{q}_i \dots q_n)}}.$$

It then defines the minimum tangle drop

$$\mathcal{D}_{q_1} = \min_{i\in\{2,\dots,n\}} \sqrt{\tau_{q_1(q_2\dots q_n)} - \tau_{q_1(q_2\dots \hat{q}_i \dots q_n)}},$$

and the computationally preferred minimum negativity drop

$$\mathcal{D}_{\mathcal{N},q_1} = \min_{i \in \{2, \dots, n\}} \left( \mathcal{N}_{q_1|\mathrm{rest}} - \mathcal{N}_{q_1|\mathrm{rest}\setminus q_i} \right), \qquad \mathcal{N}_{A|B} = \|\rho^{T_A}\|_1 - 1.$$

For mixed states, the tangle and negativity versions are defined via convex roof extension (Dong et al., 20 Jun 2026).

The physical interpretation is direct. If tracing out qubit $q_i$ produces a large decrease in $q_1$'s one-to-group entanglement, then $q_1\mid(q_2\cdots q_n)$0 is structurally important for the entanglement of the focus qubit. Taking the minimum asks for the least damaging single-particle loss. This immediately explains why biseparable states tend to yield zero: if one removable particle lies outside the entangled block containing $q_1\mid(q_2\cdots q_n)$1, its removal causes no drop (Dong et al., 20 Jun 2026).

A plausible implication is that MED belongs to a family of loss-based entanglement diagnostics rather than to residual-entanglement measures based on subtracting pairwise terms. The distinction matters because earlier residual constructions can vanish on states such as the $q_1\mid(q_2\cdots q_n)$2 state, whereas the minimum-drop construction is designed to avoid that specific failure mode.

2. Monotonicity, vanishing properties, and scope

The MED construction is proved to be an entanglement monotone under local operations and classical communication. For the tangle-based quantity, nonnegativity follows from the generalized one-focus monogamy inequality

$q_1\mid(q_2\cdots q_n)$3

which in particular implies

$q_1\mid(q_2\cdots q_n)$4

The convex-roof monotonicity proof uses a flagged-superposition criterion together with concavity of one-to-group tangle and convexity of the subtractive mixed-state terms. The negativity-drop proof proceeds analogously, using the fact that negativity already has the required degree-2 homogeneity (Dong et al., 20 Jun 2026).

A central structural property is vanishing on biseparable states. If a pure state factorizes across some partition $q_1\mid(q_2\cdots q_n)$5 with $q_1\mid(q_2\cdots q_n)$6, tracing out a qubit in $q_1\mid(q_2\cdots q_n)$7 does not change the entanglement between $q_1\mid(q_2\cdots q_n)$8 and the remaining system. Hence one drop term is zero, so the minimum is zero. For negativity this is expressed through $q_1\mid(q_2\cdots q_n)$9 and $n$0 (Dong et al., 20 Jun 2026).

The scope is more limited than a universally faithful multipartite entanglement measure. The paper states that the tangle-drop version satisfies four of the five usual criteria for a genuine multipartite entanglement measure, but may still fail strict positivity on some genuinely multipartite entangled states. The negativity-drop version is stronger operationally, yet beyond three qubits it also has explicit blind spots (Dong et al., 20 Jun 2026).

This suggests that MED is best viewed as a monotone with a specific structural semantics: sensitivity to single-particle loss. That semantics is narrower than generic multipartite entanglement quantification.

3. Tripartite regime and relation to earlier measures

In the three-qubit setting, MED admits especially transparent formulas. For a focus qubit $n$1,

$n$2

Using the Coffman–Kundu–Wootters relation

$n$3

one obtains

$n$4

After minimizing over the choice of focus qubit, this is physically equivalent to the previously studied minimum pairwise concurrence,

$n$5

so in the tripartite regime the minimum tangle drop reproduces an established faithful genuine-multipartite quantity (Dong et al., 20 Jun 2026).

The negativity version is even sharper in this regime. For three-qubit pure states,

$n$6

while for mixed two-qubit states,

$n$7

If $n$8, then for some $n$9 one has $q_1$0, hence $q_1$1. Combined with CKW, this forces vanishing minimum pairwise concurrence and therefore biseparability. The paper accordingly proves that the minimum negativity drop is faithful for three-qubit states (Dong et al., 20 Jun 2026).

This tripartite equivalence places MED in a clear lineage relative to earlier residual-tangle programs. By contrast, the original three-tangle vanishes on $q_1$2-class states, and the monogamy-inspired precursor $q_1$3 retains that deficiency. The minimum-drop reformulation is specifically introduced to remedy that behavior (Dong et al., 20 Jun 2026).

4. State families, exact formulas, and failure cases

The pure $q_1$4-qubit $q_1$5 state,

$q_1$6

provides the canonical exact example. Symmetry implies that the drop does not depend on the choice of focus qubit or the removed qubit. The one-to-group tangle is

$q_1$7

while after tracing out one nonfocus qubit it becomes

$q_1$8

Hence

$q_1$9

The value is positive for every $q_1\mid(q_2\cdots q_n)$0, but decays as $q_1\mid(q_2\cdots q_n)$1, showing that larger $q_1\mid(q_2\cdots q_n)$2 states are increasingly robust to single-particle loss (Dong et al., 20 Jun 2026).

The same family remains exactly solvable under the noise model

$q_1\mid(q_2\cdots q_n)$3

For the pure superpositions

$q_1\mid(q_2\cdots q_n)$4

the paper finds

$q_1\mid(q_2\cdots q_n)$5

independent of $q_1\mid(q_2\cdots q_n)$6, and by the rank-2 convex-roof argument concludes that the mixed state itself obeys

$q_1\mid(q_2\cdots q_n)$7

This gives an exact scaling law jointly in system size and noise strength (Dong et al., 20 Jun 2026).

Graph states illustrate both the diagnostic power and the limitations of MED. For connected three-qubit graph states, the method identifies a single global cluster $q_1\mid(q_2\cdots q_n)$8. For connected four-qubit graph states, it distinguishes different local Clifford orbits and, within one orbit, can identify disjoint two-qubit clusters such as $q_1\mid(q_2\cdots q_n)$9 and $$\mathcal{R}_{q_1} = \min_{i\in\{2,\dots,n\}} \sqrt{\tau_{q_1(q_2\dots q_n)} - \tau_{q_1q_i} - \tau_{q_1(q_2\dots \hat{q}_i \dots q_n)}}.$$0, or diagonal pairings $$\mathcal{R}_{q_1} = \min_{i\in\{2,\dots,n\}} \sqrt{\tau_{q_1(q_2\dots q_n)} - \tau_{q_1q_i} - \tau_{q_1(q_2\dots \hat{q}_i \dots q_n)}}.$$1 and $$\mathcal{R}_{q_1} = \min_{i\in\{2,\dots,n\}} \sqrt{\tau_{q_1(q_2\dots q_n)} - \tau_{q_1q_i} - \tau_{q_1(q_2\dots \hat{q}_i \dots q_n)}}.$$2 for the square ring graph. For the star and fully connected four-qubit graphs, it identifies a single global cluster $$\mathcal{R}_{q_1} = \min_{i\in\{2,\dots,n\}} \sqrt{\tau_{q_1(q_2\dots q_n)} - \tau_{q_1q_i} - \tau_{q_1(q_2\dots \hat{q}_i \dots q_n)}}.$$3 (Dong et al., 20 Jun 2026).

However, the quantity is not faithful for arbitrary multipartite states. An explicit counterexample is the genuinely multipartite entangled chain-type state

$$\mathcal{R}_{q_1} = \min_{i\in\{2,\dots,n\}} \sqrt{\tau_{q_1(q_2\dots q_n)} - \tau_{q_1q_i} - \tau_{q_1(q_2\dots \hat{q}_i \dots q_n)}}.$$4

for which the minimum entanglement drop vanishes. The paper also notes that the diagnostic sensitivity strictly vanishes for the 5-qubit error-correcting code because every subsystem of fewer than 3 qubits is maximally mixed and every two-qubit reduced state factorizes, so tracing out one qubit does not affect the focus qubit’s one-to-group profile (Dong et al., 20 Jun 2026).

5. Structural probe and experimental estimation

MED is presented არა only as a scalar monotone but as a structural probe. If tracing out qubit $$\mathcal{R}_{q_1} = \min_{i\in\{2,\dots,n\}} \sqrt{\tau_{q_1(q_2\dots q_n)} - \tau_{q_1q_i} - \tau_{q_1(q_2\dots \hat{q}_i \dots q_n)}}.$$5 causes a positive decrease,

$$\mathcal{R}_{q_1} = \min_{i\in\{2,\dots,n\}} \sqrt{\tau_{q_1(q_2\dots q_n)} - \tau_{q_1q_i} - \tau_{q_1(q_2\dots \hat{q}_i \dots q_n)}}.$$6

then $$\mathcal{R}_{q_1} = \min_{i\in\{2,\dots,n\}} \sqrt{\tau_{q_1(q_2\dots q_n)} - \tau_{q_1q_i} - \tau_{q_1(q_2\dots \hat{q}_i \dots q_n)}}.$$7 and $$\mathcal{R}_{q_1} = \min_{i\in\{2,\dots,n\}} \sqrt{\tau_{q_1(q_2\dots q_n)} - \tau_{q_1q_i} - \tau_{q_1(q_2\dots \hat{q}_i \dots q_n)}}.$$8 must belong to a mutually inseparable subset. This leads to the notion of a directly entangled cluster $$\mathcal{R}_{q_1} = \min_{i\in\{2,\dots,n\}} \sqrt{\tau_{q_1(q_2\dots q_n)} - \tau_{q_1q_i} - \tau_{q_1(q_2\dots \hat{q}_i \dots q_n)}}.$$9, defined as the set containing the focus qubit and all particles whose removal yields a positive drop. The entanglement within that identified cluster is then quantified by

$$\mathcal{D}_{q_1} = \min_{i\in\{2,\dots,n\}} \sqrt{\tau_{q_1(q_2\dots q_n)} - \tau_{q_1(q_2\dots \hat{q}_i \dots q_n)}},$$0

The full qubit-resolved pattern $$\mathcal{D}_{q_1} = \min_{i\in\{2,\dots,n\}} \sqrt{\tau_{q_1(q_2\dots q_n)} - \tau_{q_1(q_2\dots \hat{q}_i \dots q_n)}},$$1 is interpreted as a connectivity fingerprint (Dong et al., 20 Jun 2026).

This loss-sensitivity picture underlies the graph-state applications. States within the same local Clifford equivalence class can still exhibit different MED fingerprints because the method is probing local vulnerability to particle loss rather than only abstract entanglement class. A plausible implication is that MED can encode architecture-level information relevant to experimental connectivity even when standard multipartite measures cannot separate two states.

The proposed experimental route uses randomized measurements and classical shadows. For globally pure states, the vanishing of negativity drop is equivalent to factorization of the two-qubit reduced state,

$$\mathcal{D}_{q_1} = \min_{i\in\{2,\dots,n\}} \sqrt{\tau_{q_1(q_2\dots q_n)} - \tau_{q_1(q_2\dots \hat{q}_i \dots q_n)}},$$2

which can be reformulated as the purity condition

$$\mathcal{D}_{q_1} = \min_{i\in\{2,\dots,n\}} \sqrt{\tau_{q_1(q_2\dots q_n)} - \tau_{q_1(q_2\dots \hat{q}_i \dots q_n)}},$$3

Thus MED-style cluster identification reduces to estimating one- and two-qubit purities. The paper argues that this can reduce classical post-processing from an exhaustive search over $$\mathcal{D}_{q_1} = \min_{i\in\{2,\dots,n\}} \sqrt{\tau_{q_1(q_2\dots q_n)} - \tau_{q_1(q_2\dots \hat{q}_i \dots q_n)}},$$4 bipartitions to $$\mathcal{D}_{q_1} = \min_{i\in\{2,\dots,n\}} \sqrt{\tau_{q_1(q_2\dots q_n)} - \tau_{q_1(q_2\dots \hat{q}_i \dots q_n)}},$$5, where $$\mathcal{D}_{q_1} = \min_{i\in\{2,\dots,n\}} \sqrt{\tau_{q_1(q_2\dots q_n)} - \tau_{q_1(q_2\dots \hat{q}_i \dots q_n)}},$$6 is the number of identified macroscopic clusters, and in highly structured cases such as GHZ and $$\mathcal{D}_{q_1} = \min_{i\in\{2,\dots,n\}} \sqrt{\tau_{q_1(q_2\dots q_n)} - \tau_{q_1(q_2\dots \hat{q}_i \dots q_n)}},$$7 states even to $$\mathcal{D}_{q_1} = \min_{i\in\{2,\dots,n\}} \sqrt{\tau_{q_1(q_2\dots q_n)} - \tau_{q_1(q_2\dots \hat{q}_i \dots q_n)}},$$8 (Dong et al., 20 Jun 2026).

6. Related meanings of the phrase in earlier literature

Before MED became a formal multipartite monotone, closely related expressions referred to different operational questions. In random-gate theory, the nearest antecedent is the minimum entangling power of a bipartite unitary. For a gate $$\mathcal{D}_{q_1} = \min_{i\in\{2,\dots,n\}} \sqrt{\tau_{q_1(q_2\dots q_n)} - \tau_{q_1(q_2\dots \hat{q}_i \dots q_n)}},$$9 and an entanglement measure $$\mathcal{D}_{\mathcal{N},q_1} = \min_{i \in \{2, \dots, n\}} \left( \mathcal{N}_{q_1|\mathrm{rest}} - \mathcal{N}_{q_1|\mathrm{rest}\setminus q_i} \right), \qquad \mathcal{N}_{A|B} = \|\rho^{T_A}\|_1 - 1.$$0,

$$\mathcal{D}_{\mathcal{N},q_1} = \min_{i \in \{2, \dots, n\}} \left( \mathcal{N}_{q_1|\mathrm{rest}} - \mathcal{N}_{q_1|\mathrm{rest}\setminus q_i} \right), \qquad \mathcal{N}_{A|B} = \|\rho^{T_A}\|_1 - 1.$$1

with the minimization over product inputs. For Haar-random bipartite unitaries, even this worst-case output entanglement is generically close to maximal; for Schmidt rank one has the exact almost-sure value

$$\mathcal{D}_{\mathcal{N},q_1} = \min_{i \in \{2, \dots, n\}} \left( \mathcal{N}_{q_1|\mathrm{rest}} - \mathcal{N}_{q_1|\mathrm{rest}\setminus q_i} \right), \qquad \mathcal{N}_{A|B} = \|\rho^{T_A}\|_1 - 1.$$2

and for entropy the minimum entangling power concentrates near $$\mathcal{D}_{\mathcal{N},q_1} = \min_{i \in \{2, \dots, n\}} \left( \mathcal{N}_{q_1|\mathrm{rest}} - \mathcal{N}_{q_1|\mathrm{rest}\setminus q_i} \right), \qquad \mathcal{N}_{A|B} = \|\rho^{T_A}\|_1 - 1.$$3 (Chen et al., 2012). This is not entanglement drop under particle loss, but it is a minimum-over-worst-case construction with similar terminology.

A second operationally distinct use concerns entanglement erasure. In the catalytic randomization model, the asymptotic noise cost to render a bipartite state separable is

$$\mathcal{D}_{\mathcal{N},q_1} = \min_{i \in \{2, \dots, n\}} \left( \mathcal{N}_{q_1|\mathrm{rest}} - \mathcal{N}_{q_1|\mathrm{rest}\setminus q_i} \right), \qquad \mathcal{N}_{A|B} = \|\rho^{T_A}\|_1 - 1.$$4

the regularized relative entropy of entanglement. Here the optimization concerns the minimum rate of injected randomness needed to drive entanglement all the way to zero, not the sensitivity of multipartite entanglement to subsystem loss (Berta et al., 2017).

Other adjacent notions include pre-managing how far entanglement may decay before a prescribed time $$\mathcal{D}_{\mathcal{N},q_1} = \min_{i \in \{2, \dots, n\}} \left( \mathcal{N}_{q_1|\mathrm{rest}} - \mathcal{N}_{q_1|\mathrm{rest}\setminus q_i} \right), \qquad \mathcal{N}_{A|B} = \|\rho^{T_A}\|_1 - 1.$$5 under amplitude, phase, or depolarizing noise via threshold disentanglement (Qian, 2012), and minimizing scattering-induced entanglement by imposing the mass–width condition

$$\mathcal{D}_{\mathcal{N},q_1} = \min_{i \in \{2, \dots, n\}} \left( \mathcal{N}_{q_1|\mathrm{rest}} - \mathcal{N}_{q_1|\mathrm{rest}\setminus q_i} \right), \qquad \mathcal{N}_{A|B} = \|\rho^{T_A}\|_1 - 1.$$6

for one-dimensional Gaussian wave-packet collisions (Nicolaou et al., 2019). These problems concern entanglement preservation or suppression, but not MED in the specific multipartite monotone sense.

The modern MED formulation is therefore best understood as a distinct construct: the minimum loss of one-to-group entanglement under tracing out a single constituent, used both as a monotone and as a structural probe of multipartite entanglement organization (Dong et al., 20 Jun 2026).