Generating Non-Decomposable Maps with Differentiable Semidefinite Programming
Angela Rosy Morgillo, Davide Poderini, Fabio Anselmi, Fabio Benatti, Massimiliano F. Sacchi, Chiara Macchiavello · 2026
Read Generating Non-Decomposable Maps with Differentiable Semidefinite Programming by Angela Rosy Morgillo, Davide Poderini, Fabio Anselmi, Fabio Benatti, Massimiliano F. Sacchi, Chiara Macchiavello (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-14644.txt.
Introduction
Morgillo and team build a new way to make positive non-decomposable maps. Such maps are a key tool for spotting bound entangled states, states that stay entangled yet pass a common test. The method uses differentiable semidefinite programming, a form of optimisation with matrix constraints. Certificates of positivity and non-decomposability join gradient search over maps of varied input and output size, under flexible limits on Choi matrices. The search yields new number examples, a parametrised family from masked Choi matrices, and real non-decomposable maps.
Catalog record
| CREATOR | Angela Rosy Morgillo, Davide Poderini, Fabio Anselmi, Fabio Benatti, Massimiliano F. Sacchi, Chiara Macchiavello |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 35892 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.14644
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Generating Non-Decomposable Maps with Differentiable Semidefinite Programming Angela Rosy Morgillo, Davide Poderini, Fabio Anselmi, Fabio Benatti, Massimiliano F. Sacchi, Chiara Macchiavello Abstract Positive maps that are not decomposable are a key resource in entanglement theory because they can detect bound entangled states, yet systematic methods for constructing them remain limited. We introduce an optimization framework based on differentiable semidefinite programming (SDP) for generating positive non-decomposable maps under flexible structural constraints on their Choi matrices. The method combines SDP-based certificates of non-decomposability and positivity with gradient-based optimization, enabling a systematic search over maps with different input and output dimensions. Within this framework, we generate previously unknown numerical examples, identify a parametrized family of maps arising from masked Choi matrices, and construct real non-decomposable maps. We further show that the same approach can be adapted to explore open questions in quantum information theory, including the PPT square conjecture and recently proposed eigenvalue bounds for 2-positive trace-preserving maps. I Introduction Positive maps that are not completely positive are fundamental tools in quantum information theory because they provide an operational way to detect entanglement 7 ; 29 ; 22 . A linear map \Phi is positive if it preserves positive semidefinite operators. Although complete positivity is required for physically realizable quantum channels, positivity alone is sufficient for entanglement detection: if \mathbbm{1}\otimes\Phi maps a bipartite state \rho to a non-positive operator, then \rho must be entangled. The best-known example is the Peres–Horodecki criterion based on partial transposition 46 ; 28 . Among positive maps, non-decomposable ones are especially important because they can detect entangled states with positive partial transpose (PPT), which are invisible to decomposable witnesses 37 . A decomposable map can be written as the sum of a completely positive map and a completely copositive one, whereas a non-decomposable map cannot. Despite their central role in entanglement theory, explicit constructions are still relatively scarce, especially in higher dimensions, and no broadly applicable framework for their systematic generation is currently available 43 ; 52 ; 38 ; 58 ; 12 ; 5 ; 51 ; 61 ; 4 ; 49 ; 25 . Since Choi’s original example 14 , only a limited number of explicit families of non-decomposable positive maps have been identified 54 ; 34 ; 16 ; 30 ; 40 ; 32 ; 44 ; 6 ; 36 ; 8 ; 60 ; 23 . Most subsequent developments generalize or reparametrize known constructions 45 ; 24 ; 11 ; 17 ; 31 , while comparatively few works address systematic generation strategies 57 ; 35 . Developing flexible numerical methods for constructing such maps is therefore of both mathematical and physical interest. A natural approach is to formulate the search as an optimization problem. In this setting, semidefinite programming has become a standard tool in quantum information because it provides efficient convex formulations for optimization over positive semidefinite operators 55 ; 50 ; 41 . SDP-based methods are particularly effective in problems involving quantum channels, separability tests, and entanglement certification 3 ; 19 ; 20 ; 59 . More recently, SDP techniques have also been applied to the study of k -positive maps and related conjectures, including the PPT square conjecture 57 ; 10 ; 39 . In this work, we develop a differentiable SDP framework for generating positive non-decomposable maps. The key idea is to treat SDP-based tests of non-decomposability and positivity as differentiable layers inside a gradient-based optimization loop acting directly on a parametrized Choi matrix. This yields a systematic search procedure that can incorporate additional requirements such as trace preservation, reality constraints, or masked sparsity patterns. The framework is flexible across dimensions and can be adapted to problem-specific objectives. Our main contributions are as follows. First, we introduce an optimization procedure that combines differentiable SDP layers with standard gradient-based training to generate positive non-decomposable maps. Then, we use this procedure to obtain new numerical examples in dimensions between 2 and 4 and to identify parametrized families arising from masked Choi matrices. Finally, we show how the same framework can be adapted to construct real non-decomposable maps and to explore open questions in quantum information, such as the PPT square conjecture and an eigenvalue bound for 2-positive trace-preserving maps. Throughout, positivity is certified through positivity on PPT k -symmetrically extendable states 19 ; 20 , which yields a practical sufficient condition amenable to SDP. The paper is organized as follows. In Section II , we review the relevant background on positive maps, decomposability, the Choi–Jamiołkowski isomorphism, and semidefinite programming. In Section III , we present the SDP formulations and the differentiable optimization framework. In Section IV , we report the maps generated by the method, present parametrized families obtained from masked Choi matrices, and illustrate customized loss functions for additional tasks. Finally, in Section V , we summarize the main results and outline possible directions for future work. II Theory In this section, we review the concepts needed to describe and analyze positive maps in quantum information theory. We recall the notions of positivity, complete positivity, and decomposability, as well as the Choi–Jamiołkowski isomorphism. We also briefly introduce semidefinite programming, which underlies the optimization framework developed in this work. II.1 Positive and non-decomposable maps Let \mathcal{B}(\mathbb{C}^{n}) denote the set of n\times n complex matrices. A linear map \Phi:\mathcal{B}(\mathbb{C}^{d})\to\mathcal{B}(\mathbb{C}^{d^{\prime}}) is positive if it sends positive semidefinite operators to positive semidefinite operators, namely \Phi(X)\succeq 0\,,\quad\forall\,X\succeq 0\,. (1) A stronger notion is k -positivity : \Phi is k -positive if its extension \mathbbm{1}_{k}\otimes\Phi preserves positivity on operators acting on \mathbb{C}^{k}\otimes\mathbb{C}^{d} , that is, (\mathbbm{1}_{k}\otimes\Phi)(X)\succeq 0\,,\quad\forall\,X\succeq 0\,. (2) Here, \mathbbm{1}_{k} denotes the identity map on \mathbb{C}^{k} . A map is completely positive (CP) if it is k -positive for all k\geq 1 . A map is trace-preserving (TP) if \Tr[\Phi(X)]=\Tr[X]\,,\quad\forall\,X\,. (3) Completely positive trace-preserving maps are quantum channels. CP maps admit a Kraus representation 13 , \Phi(\cdot)=\sum_{l}E_{l}\,(\cdot)\,E_{l}^{\dagger}\,, (4) and the TP condition is equivalent to \sum_{l}E_{l}^{\dagger}E_{l}=I_{d} , where I_{d} is the d\times d identity matrix. Positive maps display a substantially richer structure. A positive map \Phi is called decomposable if it can be written as \Phi=\Phi_{1}+\Phi_{2}\,, (5) where \Phi_{1} is CP and \Phi_{2} is completely copositive, meaning that \Phi_{2}\circ T is CP, with T denoting transposition. Maps that do not admit such a decomposition are called non-decomposable . These maps are of particular interest because they detect PPT entangled states that cannot be detected by decomposable maps 7 . Moreover, for dd^{\prime}\leq 6 all positive maps are decomposable 46 ; 28 ; non-decomposable maps therefore occur only in sufficiently high dimensions. A linear map can be represented by its Choi matrix C_{\Phi} , which establishes a one-to-one correspondence between linear maps and bipartite operators via the Choi–Jamiołkowski isomorphism 13 . It is defined by C_{\Phi}=(\mathbbm{1}_{d}\otimes\Phi)\outerproduct{\psi}{\psi}\,, (6) where \ket{\psi}=\sum_{i=1}^{d}\ket{i}\otimes\ket{i} (7) is the maximally entangled vector on \mathbb{C}^{d}\otimes\mathbb{C}^{d} . The CJ isomorphism allows one to study linear maps through operators. In this representation, \Phi is completely positive if and only if C_{\Phi} is positive semidefinite, whereas mere positivity is equivalent to block positivity: \bra{v }\otimes\bra{w}C_{\Phi}\ket{v}\otimes\ket{w}\geq 0\,,\quad\forall\,v\in\mathbb{C}^{d}\,,\ \forall\,w\in\mathbb{C}^{d^{\prime}}\,. (8) Accordingly, negative eigenvalues of the Choi matrix indicate that the map cannot be completely positive, although it may still be positive. II.2 Differentiable semidefinite programming Semidefinite programming (SDP) is a