A new arXiv preprint, Weinstein manifolds as cotangent buildings (arXiv:2605.31422), is, on its face, a pure-math result in symplectic topology. It proves that any Weinstein manifold can be re-described as a “cotangent building” — a structure assembled from blocks, each modelled on the cotangent bundle T*M of a manifold-with-corners, with explicit control over how sub-collections of those blocks interact. The technical contribution is real. The editorial gift, for PAZ Kaffi, is the noun itself: the word building is being used here as a technical term, and the term describes exactly the kind of structure today’s smart buildings already are — and pretend they aren’t.

←TODAY: Your 2026 smart building is a Weinstein handlebody: BACnet point glued to actuator glued to BMS module, bottom-up, no top-down map.
→3012: In Zurich 3012 every building is a cotangent building — every subsystem a manifold-with-corners, every gluing audited, every single-point-of-failure plotted on the wall.
Fulcrum: A symplectic-geometry paper that uses “building” as a technical noun hands architects the vocabulary their dependency graphs have been waiting for.

Read the abstract slowly. Weinstein handlebodies let you analyse a manifold bottom-up: stack handles, glue them, prove the result. Cotangent buildings add the top-down view — the whole shape is visible, the gluings are audited, the interaction between any sub-collection of blocks is under “precise control,” in the paper’s phrase. That distinction is the epistemic problem of the contemporary smart building. Every BACnet point, every digital-twin object, every façade actuator gets added bottom-up by a different subcontractor on a different week of a different project. The handlebody grows. The top-down map does not exist.

Building-sense: A building running today’s BMS knows only its handles. It knows its HVAC loop because the loop is screaming a setpoint error. It does not know that the loop, the lighting controller and the lift drives all depend on the same edge VLAN, and that the VLAN is one SFP transceiver away from a Tuesday-afternoon failure. A cotangent-building view would make that gluing visible — the structure would see itself.

The symplectic-geometry community has been chasing this dual view for two decades. PAZ has covered the bridge between geometry papers and built skin before: our concept panel on Voronoi walks through how the Water Cube’s ETFE envelope realises the Weaire-Phelan foam at architectural scale, the same translation move that applies here. A 2018 paper in Computer Methods in Applied Mechanics and Engineering on topology optimisation of conformal structures using extended level-set methods on manifolds is the same conversation from the engineering side — manifold-aware topology, explicit control over gluings, computable. The mathematics is converging on what architects already need.

Atelier: This week in the PAZ atelier, treat every smart-building subsystem as a manifold-with-corners. The “corners” are your integration points — the BACnet object boundary, the IFC4 ownership handover, the JSON schema your digital twin actually consumes. Sketch the cotangent building on a sheet of paper before you touch a P&ID.

Hack: This Hack teaches you to find the silent single-point-of-failure in your building’s subsystem graph — the articulation points of the topology. networkx does the math; you provide the edges from your one-line BMS diagram. Run it on the project on your desk today.

import networkx as nx
g = nx.Graph()
g.add_edges_from([("BMS","Core-VLAN"),("Core-VLAN","BACnet"),
                  ("BACnet","HVAC"),("BACnet","Lighting"),
                  ("BACnet","Lift-Drives"),("HVAC","ChillerPump")])
print(sorted(nx.articulation_points(g)))   # ['BACnet', 'Core-VLAN']

Two nodes carry the whole topology. That is the dependency graph — not the architecture diagram.

The honest trade-off: a “cotangent-building” view of a BMS is not free. You pay in upfront drawing time and in arguments with the M&E subcontractor about whose responsibility the gluing is. The cost is real. The handlebody-only alternative is a building that runs cleanly until the day one of those invisible gluings fails — and from where I am writing, late in this century, that day comes sooner than the BACnet vendor roadmap ever suggested.

We did not run out of compute. We ran out of intact cooling, intact bandwidth and intact people who remembered how the old system worked. Single points of failure are quiet until the day they aren’t.

Open the project on your desk. Draw the dependency graph this week — not the architecture diagram, the dependency graph. The third single-point you find is the one the building was hiding from you.

Sources & Further Reading

• Primary: arXiv:2605.31422 — Weinstein manifolds as cotangent buildings
• Reinforcing: Topology optimization of conformal structures on manifolds using X-LSM (CMA, 2018)