---
title: Horodecki Criterion and PPT Separability
url: https://www.emergentmind.com/topics/horodecki-criterion
type: topic
---

# Horodecki Criterion and PPT Separability

Across the cited literature, the expression “Horodecki criterion” denotes several distinct results associated with the Horodecki family. In its standard usage it is the separability criterion built from positivity under partial transpose and, in its most general form, from positivity under all positive maps. For a bipartite state \(\rho\in \mathcal L(\mathcal H_A\otimes\mathcal H_B)\), separability means
\[
\rho=\sum_k p_k\,\rho_k^A\otimes\rho_k^B,\qquad p_k\ge 0,\quad \sum_k p_k=1,
\]
while partial transposition on subsystem \(B\) is
\[
(\rho^{T_B})_{i\mu,j\nu}=\rho_{i\nu,j\mu}.
\]
The Peres criterion states that separability implies \(\rho^{T_B}\ge 0\). The Horodecki refinement shows that positivity under partial transpose (PPT) is also sufficient for separability in \(2\times 2\) and \(2\times 3\), but not in higher dimensions, where PPT entangled or bound entangled states exist [2001.08118]. In a broader formulation, a bipartite state is separable iff
\[
(I\otimes \Lambda)(\rho)\ge 0
\]
for all positive maps \(\Lambda\) [1607.03364].

## 1. Core statement and mathematical form

The standard separability problem begins with the distinction between convex mixtures of product states and genuinely entangled states. In matrix-element notation, partial transpose on the second subsystem acts by
\[
(M^{\mathrm{pt}})_{i\alpha,j\beta}=M_{i\beta,j\alpha},
\]
and the trace identity
\[
\operatorname{tr}(M\rho^{\mathrm{pt}})=\operatorname{tr}(M^{\mathrm{pt}}\rho)
\]
allows one to move the partial transpose from the state to the observable [1608.01177]. This identity underlies many operational reformulations of the criterion.

Within the standard hierarchy, the Peres test is a necessary condition for separability: if \(\rho\) is separable, then \(\rho^{T_B}\) must remain positive semidefinite. The Horodecki positive-map theorem is stronger. Rather than selecting the transpose map alone, it quantifies over all positive maps and thereby gives a necessary-and-sufficient characterization of bipartite separability in arbitrary finite dimension. In that sense, the usual PPT test is the most accessible special case of a more general positive-map framework [1607.03364].

This broader perspective also clarifies the status of partial transpose. The transpose map is positive but not completely positive, so its usefulness lies precisely in the fact that complete positivity is not imposed on the tested subsystem. That is why \(\rho^{T_B}\) can fail to be positive for entangled states even though \(\rho\) itself is a valid density operator.

## 2. Exact low-dimensional regime

The classical low-dimensional theorem is that PPT is necessary and sufficient for separability in \(2\times 2\) and \(2\times 3\). In the \(2\times 2\) case, this can be re-expressed on the map side by Størmer’s theorem: every positivity-preserving map on \(M_2^{sa}\) is decomposable, with
\[
\Phi(\rho)=\sum_j A_j\rho A_j^\dagger+\sum_k B_k\rho^T B_k^\dagger.
\]
By duality, this is equivalent to the statement that a state on \(\mathbb C^2\otimes\mathbb C^2\) is separable iff its partial transpose is positive [1512.03293]. The \(2\times 3\) case is attributed in the same source to Woronowicz.

For pure bipartite states, the separability problem simplifies further. If
\[
|\psi\rangle=\sum_{i=1}^n\sum_{J=1}^m C_{iJ}\,|i\rangle\otimes |J\rangle,
\]
then separability is equivalent to the coefficient matrix \(C\) having rank one. A necessary-and-sufficient test is that all \(2\times 2\) minors vanish:
\[
\det Q^{(s,t,u,v)}(C)=0 \quad \forall\, s<t,\;u<v.
\]
On that basis, PPT becomes not merely necessary but also sufficient for separability of pure bipartite states in arbitrary finite dimension [1304.5294].

These exact results explain why PPT occupies a privileged position in low-dimensional bipartite theory. In those settings it is not merely a witness-generating device or a useful relaxation; it is a complete decision criterion.

## 3. Breakdown beyond \(2\times 3\) and bound entanglement

For larger bipartite systems the PPT test ceases to characterize separability. The first symmetric case where this happens is \(3\times 3\), where entangled states with positive partial transpose exist. In the geometric picture used for two qutrits, state space decomposes into separable states (SEP), PPT-entangled states \(\delta_{PPTES}\), and negative-partial-transpose entangled states \(\rho_{NPT}\), with edge states on the SEP/PPTES and PPTES/NPT boundaries. The same source emphasizes that deciding PPT versus NPT is polynomial-time, whereas separability in general is NP-HARD, which helps explain both the power and the incompleteness of the criterion in higher dimension [2001.08118].

The canonical explicit example is the Horodecki two-qutrit state \(\rho_a\). It satisfies
\[
\rho_a^\Gamma \ge 0 \quad \text{for all } a\in[0,1],
\]
is separable for \(a=0\) and \(a=1\), and is entangled for \(a\in(0,1)\). The same family admits a nontrivial abelian covariance under
\[
G_0=\{\operatorname{diag}(e^{ix_1},e^{ix_2},e^{ix_1}) : x_1,x_2\in\mathbb R\},
\]
which yields invariant-subspace decompositions of both \(\rho_a\) and \(\rho_a^\Gamma\). In this representation, checking PPT reduces to positivity of two \(4\times 4\) leading principal blocks of the partial transpose, making transparent how a highly structured PPT-entangled state can arise already in \(3\otimes 3\) [1009.4385].

Recent work restates this limitation in especially sharp form: PPT bound entangled states are invisible not only to the Peres–Horodecki test itself but also to all criteria based solely on the partial-transpose spectrum. In that setting, realignment spectral features and chirality corrections are introduced as mathematically independent channels precisely because “all partial-transpose-based criteria are structurally blind to PPT states” [2605.14515].

The same boundary also motivates practical detours rather than theorem-strengthening. In two-qutrit tomography, automated learning models trained on local \(SU(3)\otimes SU(3)\) expectation values can distinguish SEP, PPTES, and NPT directly from tomographic data, but that program is explicitly presented as an empirical complement to the Horodecki criterion rather than a new necessary-and-sufficient separability theorem [2001.08118].

## 4. Reformulations and operational consequences

A large literature recasts the criterion into experimentally usable inequalities. One route starts from conserved quantities. For a spin-\(j\otimes j'\) system,
\[
\langle \mathbf S_a\cdot \mathbf S_b\rangle \in [-j'(j+1),\,jj'],
\]
and separability implies
\[
\left\langle (\mathbf S_a\cdot \mathbf S_b)^{\mathrm{pt}}\right\rangle \in [-j'(j+1),\,jj'].
\]
In the two-qubit specialization, this yields the bilinear witness
\[
W = I-\sigma_x\otimes\sigma_x+\sigma_y\otimes\sigma_y-\sigma_z\otimes\sigma_z,
\qquad \langle W\rangle\ge 0
\]
for separable states. A stronger nonlinear inequality follows from the generalized Schrödinger–Robertson uncertainty relation, and for Werner states it reproduces the exact PPT threshold \(x>1/3\) for entanglement [1608.01177].

Another route uses Hilbert–Schmidt decompositions and a partial-transpose-plus-local-unitary transformation (PTU). For qubit–qudit states of the form
\[
2d\,\rho=(I)^{2d}+S,
\]
PTU leads to the spectral relation
\[
\lambda(\rho(\mathrm{PTU}))=\frac{1}{d}-\lambda(\rho).
\]
Hence any eigenvalue of \(\rho\) larger than \(1/d\) certifies entanglement. For \(n\)-qubit maximally disordered subsystems (MDS), the corresponding sufficient criterion is an eigenvalue larger than \(1/2^{n-1}\). The same analysis shows that for odd \(n\) MDS states, PTU does not change the spectrum, so the Peres–Horodecki criterion becomes mute in that sector [1606.04304].

A third reformulation is combinatorial. For grid-labelled graph Laplacian states
\[
\rho(G_l^{a,b})=\frac{L(G_l^{a,b})}{2|E(G_l^{a,b})|},
\]
partial transpose becomes an operation on graph edges, and PPT is equivalent to preservation of the degree matrix:
\[
D(G_l^{a,b})=D(\Gamma(G_l^{a,b})).
\]
In \(2\times b\), this degree criterion is necessary and sufficient for separability; in \(3\times 3\), it is only necessary, and explicit PPT-entangled graph patterns such as the “cross-hatch” and “skew-mesh” show the same failure of sufficiency familiar from the ordinary matrix formulation [1605.03564].

These reformulations do not replace the original theorem. Rather, they show that the Horodecki criterion has a wide operational envelope: spectral, witness-based, uncertainty-based, Hilbert–Schmidt, and graph-theoretic.

## 5. Continuous-variable and phase-space version

For continuous-variable Gaussian states, the finite-dimensional matrix test is replaced by Simon’s phase-space form of the Peres–Horodecki criterion. In the Wigner representation, partial transpose becomes momentum reversal on one subsystem,
\[
W(\psi_{\mathbf M})(x_1,x_2,p_1,p_2)\rightarrow W(\psi_{\mathbf M})(x_1,x_2,p_1,-p_2),
\]
and on covariance matrices it acts as
\[
\widetilde{\boldsymbol{\Sigma}}=\boldsymbol{\Lambda}\boldsymbol{\Sigma}\boldsymbol{\Lambda}^T,
\qquad \boldsymbol{\Lambda}^T=\mathrm{diag}(1,1,1,-1).
\]
The separability condition is
\[
\widetilde{\boldsymbol{\Sigma}}+\frac{i\hbar}{2}\mathbf J\ge 0,
\]
or, equivalently, that the smallest symplectic eigenvalue satisfy
\[
\lambda_{\min}\ge \frac{\hbar}{2}.
\]
For bipartite Gaussian states this condition is necessary and sufficient [2412.04863].

In the specific family of squeezed coherent states generated from holomorphic Hermite functions, the product construction has diagonal covariance and remains separable, whereas the genuinely two-variable construction produces a non-factorized Gaussian with
\[
\lambda^{(1)}_{2,\alpha}=\frac{\hbar}{2\alpha},\qquad
\lambda^{(2)}_{2,\alpha}=\frac{\hbar\alpha}{2}.
\]
Since \(0<\alpha<1\), the minimal symplectic eigenvalue is \(\hbar\alpha/2<\hbar/2\), so the state is entangled. The same paper identifies
\[
\xi=-\frac{1}{2}\ln\alpha
\]
as the squeezing parameter and
\[
E_{\mathcal L}=\max(-\log \alpha,0)
\]
as the logarithmic negativity, making explicit how squeezing and entanglement are linked in the phase-space version of the criterion [2412.04863].

This continuous-variable form is not a different principle. It is the Gaussian-state transcription of the same PPT logic into covariance-matrix language.

## 6. Bell-nonlocality and other distinct usages

A separate and well-established usage of “Horodecki criterion” concerns Bell nonlocality rather than separability. For a two-qubit state with correlation matrix
\[
T_{ij}=\operatorname{Tr}\!\big[\rho\,(\sigma_i\otimes \sigma_j)\big],
\]
let \(s_1(T)\ge s_2(T)\ge s_3(T)\) be the singular values of \(T\). The CHSH-maximization criterion is
\[
H(T):=2\sqrt{s_1(T)^2+s_2(T)^2},
\]
and the state violates CHSH iff
\[
H(T)>2.
\]
This result is necessary and sufficient when the local measurements are unrestricted projective qubit observables [2109.09890].

That CHSH criterion has itself been generalized. For nonprojective two-outcome qubit observables
\[
X=B\,\mathbb 1+S\,\sigma\cdot x,\qquad S+|B|\le 1,
\]
exact fixed-strength criteria exist for unbiased measurements on arbitrary two-qubit states and for arbitrary measurements on \(T\)-states with maximally mixed marginals. In those regimes the Bell-violation condition becomes \(S_0>2\) or \(S_T>2\), where \(S_0\) and \(S_T\) are explicit functions of the state singular values, the measurement strengths, the biases, and the relative angles [2109.09890].

The same program has been pushed to a multi-setting two-outcome Bell family for two qudits of local dimension \(d=2^m\). There the Horodecki-like quantity is
\[
M_n(\rho)=
\max_{\{\mathcal S^{(n)}_l,\mathcal S_{l'}\}}
\sqrt{
\sum_{\tau_u\in \mathcal S^{(n)}_l}
\sum_{\tau'_v\in \mathcal S_{l'}}
\langle \tau_u\otimes \tau'_v\rangle_\rho^2},
\]
and
\[
M_n(\rho)>2^{1-n}(\mathcal I_n)_L
\]
is always sufficient for violation. It becomes necessary and sufficient for \(m\) copies of Bell diagonal states, for non-decomposable states whose correlation matrix is diagonalized by local unitaries, and for any arbitrary two-qubit state when \(n=3\) [2408.10350].

The same name also appears in the “Brandão–Horodecki” theorem on correlation decay and one-dimensional area laws. That result concerns entropy bounds of the form
\[
S(\rho_X)\le c' l_0 \exp\!\bigl(c\,\log(\xi)\,\xi\bigr)
\]
for pure states with exponential decay of correlations, and is explicitly distinct from the PPT separability criterion [1505.06468].

Taken together, these usages suggest a precise disambiguation. In quantum information, “Horodecki criterion” most often denotes the PPT-based separability criterion and its positive-map completion. In Bell theory it denotes the CHSH state criterion \(H(T)>2\). In one-dimensional many-body theory, “Brandão–Horodecki” refers to a different theorem linking correlation decay to area-law entanglement.

Source: https://www.emergentmind.com/topics/horodecki-criterion