Assistant Professor (Jiangshi) by special appointment (Quanzhitepin), also a faculty postdoc (Shizi Bohou) · School of Mathematics & Information Science, Guangzhou University
I work in complex and algebraic geometry — the analytic side of it. My research centers on the Ohsawa–Takegoshi extension theorem and pluripotential theory, with separate interests in Nevanlinna theory and hyperbolicity. I also build things: small, sharp software systems that let me work the way I want to.
The Ohsawa–Takegoshi extension theorem is one of the central analytic tools of modern complex and algebraic geometry: it guarantees that a holomorphic function defined on a subvariety extends to the whole space, with explicit L² control. The size of the constant in that control is not a technicality — it ties directly to Bergman kernel estimates, the Suita conjecture, Skoda's division theorem, and the existence of sections in algebraic geometry.
Using the Berndtsson–Lempert method together with geodesics in the space of Kähler metrics and the Ross–Witt Nyström correspondence, I work toward understanding and computing this optimal constant — and toward turning that understanding into new proofs of Skoda's L² division theorem and new effective criteria around the existence of holomorphic sections of KX + L (the Fujita-conjecture circle of problems).
A separate line of work runs through value distribution theory: developing a Nevanlinna-theoretic notion of hyperbolicity, and asking whether it is equivalent to Kobayashi hyperbolicity.
Current work in preparation: optimal constants in the Ohsawa–Takegoshi theorem via Berndtsson–Lempert, and partial Bergman kernel estimates on compact Riemann surfaces.
Alongside the mathematics, I design and build software — usually small, sharp systems that get language models to do real, reliable work instead of demos. The instinct is the same one I bring to a proof: find the actual structure, then make it clean.
A document system that makes the dependency structure of mathematics explicit: every lemma, computation, or definition is a node in a directed acyclic graph, with machine-readable links to its context, prerequisites, and consequents. Numbered, individually-annotated proof steps let me tell an AI collaborator exactly which step is unclear — and any subtopic can be promoted to a self-contained subproject, anchored to its parent. The design borrows a principle from scheme theory: there is no absolute object, only an object over its base.
A local-first assistant that reads Chinese academic and administrative forms, reuses durable facts, and produces copy-paste fill sheets — turning a recurring tax of academic life into a few clicks.
An AI agent node on a home VM that collects messages from WeChat, WeCom, and email, then syncs them to my working machine for downstream processing.
A real-time feed that turns live match commentary into plain-English explanation — speech-to-text into a streaming language model.
A private network layer that routes my machines through my own infrastructure, plus a reverse-tunnel setup that turns a second laptop into an on-demand remote-compute node.
Everyone knows their sun sign. Fewer know that it — and the ascendant, the sign rising on the eastern horizon at the moment of birth — is fixed entirely by elementary geometry, and that the whole problem linearises under stereographic projection. Two great circles on the sphere become two circles in the complex plane, each satisfying r² − |c|² = 1; their intersection is a pair of antipodal points, v₁·conj(v₂) = −1; and a single inequality decides which of the two is rising.
What falls out is not folklore. Above latitude 66.56° the ascendant stops circling the ecliptic once a day and merely oscillates — a winding number dropping from 1 to 0 — and there the closed form used by standard astrology software returns the descending point about 21.5% of the time.
Eight sections, from the definition of stereographic projection to a real birth time and place — with the counterexamples that killed the wrong routes kept in rather than tidied away. The companion tool draws the same objects four ways: a draggable 3D Earth, the north and south projections, and the Mercator map (which is exactly the logarithm of the north projection). It prints every intermediate value from Julian day to rising sign, and runs two invariants live on screen, so nothing has to be taken on trust.
I travel map-first: one reliable map, and everything else follows from it. So before a trip I try to understand how a city grew — its walls, rivers, hills and treaty-port districts — and then draw the map myself, with walking routes computed on the real street network and a judgment attached to every place: worth it, only if there's time, or skip.
A self-drawn map of Fuzhou from OpenStreetMap data, with 2-, 3- and 4-day walking plans: the lanes and courtyards of the old city, a walk across the Min River to the treaty-port hillside of Yantaishan, and the stone villas of Guling. Around the map: how the city is layered, a verdict on each district, what Fuzhou food actually is, and which famous sights to leave out. Every place opens in Apple Maps or Amap in one tap.
A self-drawn map of Zhangzhou's old town, treated honestly as pleasant rather than spectacular: one 7 km loop through the arcaded streets, the two Ming stone archways and the Confucian temple, then across the Jiulong River on the old bridge at sunset — plus a short walk east to the canal quarter of Putou. Around the map: what the local food is (thick braised noodles, peanut-sauce noodles, satay beef), and what to skip. Shops open in Apple Maps or Amap by name search, since their positions are estimates.
Open to collaboration in complex/algebraic geometry, students interested in several complex variables, and conversations about AI-assisted research workflows.