Open-Closed Virasoro TQFT Overview

• Open-Closed Virasoro TQFT is a conceptual framework in mathematical physics aiming to integrate open and closed sectors with Virasoro symmetry.
• It focuses on key constructs such as moduli spaces, cobordisms, and boundary operations, which remain underexplored in some modern sources.
• Recent research in related fields, like multi-scale attention models in ML, highlights a gap in documented work on open-closed TQFT.

Searching arXiv for papers on Open-Closed Virasoro TQFT and related terms. Open-Closed Virasoro TQFT is not described in the supplied source material. The only cited source is "Hierarchical Kernel Transformer: Multi-Scale Attention with an Information-Theoretic Approximation Analysis" (Cirrincione, 10 Apr 2026), which concerns a multi-scale attention architecture, kernel-theoretic properties of hierarchical score matrices, expressivity relative to self-attention and causal convolution, and empirical results on ListOps, sequential CIFAR-10, and IMDB character-level sentiment. No definition, construction, theorem, or historical account concerning open-closed Virasoro topological quantum field theory appears in the provided data.

1. Scope of the supplied source

The supplied source is a machine learning paper rather than a work in mathematical physics or topological field theory. It defines the Hierarchical Kernel Transformer (HKT) as a multi-scale attention mechanism operating over $L$ resolution levels with trainable causal downsampling, level-specific bilinear score matrices, and learned convex fusion across scales (Cirrincione, 10 Apr 2026).

Within that scope, the source develops architectural formulas for hierarchical scoring, computational complexity bounds, kernel-theoretic propositions, a symmetric–antisymmetric decomposition of score matrices, an expressivity comparison with standard attention and causal convolution, and an approximation error analysis. These topics are specific to sequence modeling and do not supply the conceptual or technical ingredients ordinarily required for an encyclopedia treatment of Open-Closed Virasoro TQFT.

2. Terminological mismatch

The terminology in the source belongs to deep learning. Its central objects include token embeddings $X^{(l)}$, depthwise separable convolution, causal masking, bilinear score matrices $S^{(l)}$, convex weights $\lambda_l$, and softmax attention (Cirrincione, 10 Apr 2026). Theoretical statements are framed in terms of positive semidefinite kernels, RKHS approximation, multiple correlation, kurtosis, and computational overhead.

By contrast, the requested topic invokes the vocabulary of open-closed theories, Virasoro structures, and TQFT. No such terms occur in the supplied data. This suggests that any substantive exposition of moduli, cobordisms, operadic or categorical structure, state spaces, boundary sectors, or Virasoro constraints would require sources not present in the record supplied here.

3. Mathematical content actually present

The source provides four named theoretical results. First, Proposition 3.1 gives a sufficient condition under which the hierarchical score matrix defines a positive semidefinite kernel via the symmetrised bilinear form

$$M^{(l)}_{\mathrm{sym}} =\frac{W_Q^{(l)\top}W_K^{(l)}+W_K^{(l)\top}W_Q^{(l)}}{2\sqrt{d_k^{(l)}}}.$$

Second, Propositions 3.5–3.6 establish a unique decomposition of the asymmetric score matrix into symmetric and antisymmetric parts, interpreted respectively as reciprocal and directional attention. Third, Proposition 3.4 states that HKT strictly subsumes single-head standard attention and causal convolution for the stated parameter regime. Fourth, Theorem 4.3 and Proposition 4.4 decompose approximation error into hierarchical approximation, quantisation, and optimisation terms, with a geometric decay bound in the number of levels (Cirrincione, 10 Apr 2026).

None of these results concern topological quantum field theory. They do not define a functorial field theory, open or closed state sectors, sewing axioms, or Virasoro operators. Accordingly, they cannot be repurposed into an account of Open-Closed Virasoro TQFT without introducing material absent from the source.

4. Architectural and empirical material in the source

The source describes HKT in three stages: causal downsampling, level-specific bilinear scoring, and fusion across scales. Causal downsampling is implemented by

$$X^{(l)}=\phi_l\bigl(X^{(l-1)}\bigr) =\mathrm{GELU}\bigl(\mathrm{LayerNorm}(\mathrm{Conv}_{\text{dw}^{(l)}}(X^{(l-1)}))\bigr),$$

where the convolution is 1D, depthwise separable, kernel-size $3$, stride $s$, and uses causal left padding. The fused hierarchical score is

$$S^{\mathrm{hier}}_{ij} =\sum_{l=0}^{L-1}\lambda_l\, \tilde S^{(l)}_{\lfloor i/s^l\rfloor,\lfloor j/s^l\rfloor}.$$

The total computational cost is bounded by $\tfrac{4}{3}$ times that of standard attention, reaching $X^{(l)}$0 for $X^{(l)}$1 (Cirrincione, 10 Apr 2026).

The empirical study compares HKT-Small against a retrained standard MHA baseline over 3 random seeds. Reported gains are $X^{(l)}$2 percentage points on synthetic ListOps, $X^{(l)}$3 percentage points on sequential CIFAR-10, and $X^{(l)}$4 percentage points on IMDB character-level sentiment, all at $X^{(l)}$5 cost. These experiments establish claims about sequence-model performance rather than any statement relevant to Open-Closed Virasoro TQFT.

5. Information absent for the requested topic

The supplied material does not contain the elements needed for an encyclopedia article on Open-Closed Virasoro TQFT. Missing items include a definition of the theory, its mathematical setting, the role of open and closed sectors, the meaning of “Virasoro” in the construction, foundational theorems, examples, relations to conformal or topological field theory, and a research history.

It also does not mention authors, research groups, or papers associated with open-closed TQFT or Virasoro constraints beyond the single machine learning paper (Cirrincione, 10 Apr 2026). A plausible implication is that the requested topic and the supplied source were mismatched at the level of domain and bibliography.

6. Closest faithful characterization supported by the source

The most precise characterization supported by the supplied data is that no source-backed account of Open-Closed Virasoro TQFT can be extracted from the record provided. The data support an encyclopedia entry on the Hierarchical Kernel Transformer, including its multi-scale attention construction, PSD surrogate interpretation, reciprocity–directionality decomposition, strict subsumption of attention and convolution, and information-theoretic approximation analysis (Cirrincione, 10 Apr 2026).

Any attempt to describe Open-Closed Virasoro TQFT beyond noting its absence from the source would go beyond the evidentiary basis supplied. Under a strict factual-fidelity criterion, the topic therefore remains undocumented in the provided material.