Every maker who has fought a 45-degree overhang knows the rule: gravity wins unless you give it something to rest on. A team of physicists at the University of Amsterdam has found a way around that rule, at least for one material. In a paper published September 3 in PNAS and announced through the UvA Institute of Physics, Menno Demmenie, Stefan Kooij and Daniel Bonn describe printing free-standing ice pillars that lean as low as 14 degrees from the surface, with no support material, no freezer and no cryogenic plumbing. The trick is not cold. It is a vacuum.

Sweat, but for a print head

The physics is the same thing that happens on your skin on a hot day. "When you sweat, the water on your skin evaporates and the skin cools down," the UvA release explains. Every molecule that leaves the liquid phase takes a chunk of heat with it. Drop the ambient pressure far enough and that evaporation becomes violent enough to pull the remaining liquid well below zero.

The Amsterdam rig exploits that directly. A water jet just 16 micrometers across is fed into a chamber pumped down to roughly 2 millibar. The jet loses mass to evaporation on the way in, and the water that survives arrives supercooled, sitting below 0 degrees Celsius while still liquid. Supercooled water is metastable: it wants to freeze and only needs a nucleation site. Existing ice is the ideal one. So when the jet hits a previously printed ice surface, it solidifies on contact.

That contact-freeze behavior is what makes this printing rather than spraying. Material only accretes where ice already exists, so the growing structure defines its own build surface. The chamber stays at room temperature: no chilled bed, no liquid nitrogen, no Peltier stack. The paper's title says it plainly: "Three-Dimensional Printing of Ice Structures via Evaporative Cooling in Vacuum."

Speed sets the angle

The part that should interest anyone who has ever tuned a slicer is how the team controls geometry. On a conventional FDM machine, overhang angle is a property of the model and the slicer's job is to compensate with supports or bridging tricks. Here, the angle is a property of the toolpath speed.

The UvA release describes it in one line: the method "can also be used to print small pillars of ice under an angle, simply by varying the speed of the 3D printer." The preprint adds that by changing the print velocity while keeping the jet velocity the same, "curved, angled, and straight ice pillars can be formed." In other words, one knob, the speed of the head, sets whether a pillar stands vertical, leans, or curves. By dialing that speed, the researchers set pillar growth angles down to a floor of 14 degrees measured from the surface. That is not a 45-degree overhang; that is a structure closer to horizontal than vertical, standing on its own in near-vacuum.

To show the technique handles more than isolated pillars, the group printed a human face in profile, a shape full of the nose-and-chin overhangs that give FDM printers trouble. No supports were used. An earlier arXiv preprint from December 2025, cheekily titled "An Ice Christmas Tree," also showed cones, vertical pillars and free-standing zigzags, and reported that the tree structure took about 26 minutes to build. The preprint's framing is the same as the journal paper's: all of it done "without cryogenic infrastructure, supporting materials, or external refrigeration."

The cleanup step is the point

Ice is an odd choice for a build material until you think about what happens when you are done with it. As the UvA release puts it: "when you turn off the vacuum pump, everything melts neatly back into clean water." No solvent bath, no snapping off support trees, no filament scraps. The structure simply stops existing, and what remains is water.

That property is why the researchers are pointing at sacrificial molds as the first real application. Print a network of ice channels, cast something around them, then let the ice melt out. The resulting hollow paths are microfluidic channels, and because the ice printer lays down shallow angled branches without supports, the channel geometry can be more complex than a supported process would leave behind cleanly. The same logic applies to tissue-engineering scaffolds, where a temporary lattice defines the voids that cells will fill and residual support material would be a contaminant.

The researchers also float a more speculative case: construction on Mars, where a thin, cold atmosphere and local water roughly match the conditions the vacuum chamber recreates on Earth. That is a looking-ahead idea for now, but it shows the process depends on a pressure regime, not a particular piece of lab hardware.

What It Means for Makers

Let's be clear about what this is not. It is not a desktop machine, and nothing about a 2-millibar chamber and a 16-micrometer jet suggests one is coming. The demonstrations are lab-scale, and the preprint is six pages of physics, not a bill of materials.

What it is, is a genuinely different answer to the overhang problem. Every support strategy in the hobbyist world, from tree supports to soluble interface layers to bridging-speed tuning, treats gravity as the enemy and adds material to fight it. The Amsterdam approach instead makes deposition self-selecting: the jet freezes only where ice already exists, so the structure is its own support and toolpath speed does the work a slicer's support generator would otherwise do. That is worth noting even if you never touch a vacuum pump.

The sacrificial-mold angle is the one most likely to reach a workshop, indirectly. A mold that turns into clean water on command is a cleaner primitive than a dissolvable core that needs a wash, and if a compact vacuum-based ice printer ever becomes a lab commodity, makers who cast silicone or resin around printed cores would be natural early adopters.

Bottom Line

Three physicists in Amsterdam built a printer that uses evaporation instead of refrigeration to freeze water on contact, tilts its output by changing head speed, holds overhangs down to 14 degrees without supports, and erases its own prints by switching off a pump. The PNAS paper (DOI 10.1073/pnas.2608173123) is the formal write-up; the December preprint is the more readable introduction. For makers, it is a lab result to watch, not a machine to buy, but it is one of the more elegant reframings of the support problem in a while.

Sources