arXiv subjects

Francesco Gentile

Publications and source records attributed to Francesco Gentile.

4 recordsLinked to original sources

arXiv2026

Entanglement entropy and conformal bounds for $d=5$ CFTs

The entanglement entropy of spacetime regions $A$ in odd-dimensional conformal field theories (CFTs) contains a universal constant term, $(-1)^{\frac{d-1}{2}}F(A)$. This quantity can be robustly defined by considering the mutual information of pairs of slightly deformed versions of $A$. In the case of general three-dimensional CFTs, $F(A)$ is positive definite and bounded below by the round disk result, $F(A)\geq F_0\equiv F(\partial A=\mathbb{S}^1)$. Additionally, strong evidence has been provided that for every region $A$, $F(A)/F_0$ is maximized, within the space of CFT$_3$'s, by the free scalar field result. In this paper we show that while $F(A)$ remains a local minimum around $F_0\equiv F(\partial A=\mathbb{S}^3)$ for small deformations of the spherical entangling surface, it can take values of arbitrarily large magnitude with either sign for more general regions, and hence it is neither upper- nor lower-bounded in general CFT$_5$'s. We argue that an analogous conjecture regarding the extremization of $F(A)/F_0$ for general regions within the space of theories fails in $d=5$. We instead analyze the viability of the weaker bound, $F_ε/F_0\leq \left[F_ε/F_0\right]_{\text{free scalar}}$, $\forall$CFT$_5$ for general small geometric deformations of the spherical entangling surface. This is equivalent to a general constraint involving the stress-tensor two-point function $C_T$ and the Euclidean partition function on the sphere, namely, $C_T/F_0\leq \left[C_T/F_0\right]_{\text{ free scalar}}\approx 0.314$, which we show to hold for all known CFT$_5$'s. We also comment on possible extensions of this result to higher dimensions.

hep-th↗

arXiv2026

From Weights to Concepts: Data-Free Interpretability of CLIP via Singular Vector Decomposition

As vision-language models are deployed at scale, understanding their internal mechanisms becomes increasingly critical. Existing interpretability methods predominantly rely on activations, making them dataset-dependent, vulnerable to data bias, and often restricted to coarse head-level explanations. We introduce SITH (Semantic Inspection of Transformer Heads), a fully data-free, training-free framework that directly analyzes CLIP's vision transformer in weight space. For each attention head, we decompose its value-output matrix into singular vectors and interpret each one via COMP (Coherent Orthogonal Matching Pursuit), a new algorithm that explains them as sparse, semantically coherent combinations of human-interpretable concepts. We show that SITH yields coherent, faithful intra-head explanations, validated through reconstruction fidelity and interpretability experiments. This allows us to use SITH for precise, interpretable weight-space model edits that amplify or suppress specific concepts, improving downstream performance without retraining. Furthermore, we use SITH to study model adaptation, showing how fine-tuning primarily reweights a stable semantic basis rather than learning entirely new features.

cs.CV↗

arXiv2025

Entanglement Hamiltonian of two disjoint blocks in the harmonic chain

We study the entanglement Hamiltonian of two disjoint blocks in the harmonic chain on the line and in its ground state. In the regime of large mass, the non vanishing terms are only the on-site and the nearest-neighbour ones. Analytic expressions are obtained for their profiles, which are written in terms of piecewise linear functions that can be discontinuous and display sharp transitions as the separation between the blocks changes. In the regime of vanishing mass, where the matrices characterising the entanglement Hamiltonian contain couplings at all distances, we explore the location of the subdominant terms and some combinations of matrix elements that are useful for the continuum limit, comparing the results with the corresponding ones for the free chiral current. The single-particle entanglement spectra of these entanglement Hamiltonians are also investigated.

cond-mat.stat-mech↗

arXiv2024

Holographic thermal entropy from geodesic bit threads

The holographic bit threads are an insightful tool to investigate the holographic entanglement entropy and other quantities related to the bipartite entanglement in AdS/CFT. We mainly explore the geodesic bit threads in various static backgrounds, for the bipartitions characterized by either a sphere or an infinite strip. In pure AdS and for the sphere, the geodesic bit threads provide a gravitational dual of the map implementing the geometric action of the modular conjugation in the dual CFT. In Schwarzschild AdS black brane and for the sphere, our numerical analysis shows that the flux of the geodesic bit threads through the horizon gives the holographic thermal entropy of the sphere. This feature is not observed when the subsystem is an infinite strip, whenever we can construct the corresponding bit threads. The bit threads are also determined by the global structure of the gravitational background; indeed, for instance, we show that the geodesic bit threads of an arc in the BTZ black hole cannot be constructed.

hep-th↗