<정년퇴임 기념강연> 작용소대수와 양자정보이론

&lt;정년퇴임 기념강연&gt; 작용소대수와 양자정보이론

4910
강연자 계승혁
소속 서울대학교

작용수대수에서 순서구조가 중요한 역할을 한다. C*-대수의 시작이라 할 수 있는 Gelfand-Naimark-Segal 표현정리는 양선형범함수로부터 *-준동형을 만들어내는데,
그 표현정리 이후 여러 가지 종류의 양사상에 대한 연구가 이루어졌다. 최근 활발하게 연구되고 있는 양자정보이론의 핵심 개념은 꼬임(enatnglement)인데, 선형사상과 텐서곱의 쌍대를 이용하면 이를 적절한 행렬대수 사이의 양사상들의 쌍대 개념으로 이해할 수 있다.이번 강의에서는 여러가지 양사상들 사이의 포함관계를 비롯하여 최근 연구자들의 관심을 끌고 있는 몇몇 문제와 관련된 경험을 소개한다.

<학부생을 위한 ɛ 강연> Exploring Variational Problems over Graphons with a Tailored Neural Architecture and LLMs

강연자 : 양홍석 | 소속 : 고등과학원

In this talk, I will describe our neural framework for exploring variational problems over graphons—symmetric measurable functions from [0,1]^2 to [0,1] that serve as limit objects for dense graph sequences. These problems arise in two areas of mathematics: extremal graph theory, which studies the optimisation of graph parameters under constraints, often in the limit as the number of vertices grows; and large-deviation theory for dense random graphs, which uses constrained graphon optimisation to study the probabilities and typical structures of rare events. We represent graphons with implicit neural networks and optimise graphon objectives by gradient descent. Our approach combines three ingredients: a multiscale sinusoidal residual architecture designed to capture sharp, step-like graphons; an embedded solver that enforces a single empirical density constraint, with gradients computed by implicit differentiation; and symmetry-aware Monte Carlo estimators. For generalised Turán problems whose known solutions required substantial human effort, our framework recovers known optimal graphons without human intervention. Applied to open instances, it produces candidate optima, including a previously unreported family of candidate extremal structures. This led us to a Goodman-type conjecture for odd-cycle densities, which we subsequently proved and formalised in Lean with the help of LLMs. For variational problems arising in large-deviation theory for Erdős–Rényi random graphs, the framework produces new candidate optimal graphons in both upper- and lower-tail regimes for several pattern graphs. In the upper-tail regime, these candidates achieve better objective values than the best-known benchmark construction due to Lubetzky and Zhao.