Why surface tension decides this design
Surface tension is the pull that holds the surface of a liquid together. It is why a drop of water beads up and why a paper clip can rest on a glass of water. For most cooling fluids it is a minor property. For ours it is the centre of the design.
How the self-pumping idea works
Every FluxZero blend pairs a liquid that boils early with one that boils later and has a stronger surface pull. The idea is that where a chip runs hot, the early-boiling part leaves first, the surface pull rises at that spot, and liquid is drawn back toward the heat. If that works, the fluid pumps itself toward the hottest place on the chip with no moving parts. Whether it works depends on how large the difference in surface pull really is.
Where the number comes from
We have not mixed any of these blends. So the size of that difference comes from simulation: a computer model that tracks every molecule in a thin slab of liquid and measures the pull across its two surfaces. Before we put weight on what that model says about our own blends, we wanted to know how it does on fluids whose surface tension has already been measured and published.
The test we wrote down before the runs
We chose three commercial electronics-cooling fluids with published datasheets: a hydrofluoro-ether, a perfluorinated liquid, and a hydrofluoro-olefin. Each maker publishes a surface tension for its fluid. Those published values were the targets.
The pass band and the control
Before any run started, we wrote down what would count as a pass: the simulated value had to land within ±15% of the published one. We also wrote down which averaging method we would use, and we added a control. The control was pentane, a simple fluid our pipeline had simulated many times before. If the pentane result did not come back where it had before, the method itself would be in doubt, whatever the three commercial fluids did.
How the runs were done
Each fluid was run three times from different random starting points, and the three results were averaged. The graded runs are published, including every individual run and the control.
What came back
All three fluids came out high.
- The hydrofluoro-ether is published at 13.6 millinewtons per metre. Our simulation gave 18.5, which is 36% too high.
- The perfluorinated liquid is published at 10.0. We got 13.3, or 33% too high.
- The hydrofluoro-olefin is published at 13.0. We got 18.4, or 41% too high.
The control missed too
None of the three landed inside the band. The control failed too: under the averaging method we chose in advance, pentane did not come back where it had before. A second averaging method, recorded in the same file, brings the control back inside its bar, but it still leaves all three commercial fluids outside theirs.
The runs disagreed with each other
The three runs of each fluid also disagreed with each other more than we would like. For the perfluorinated liquid, one run gave 7.366 and another gave 20.859. With three runs and a spread like that, the average itself is uncertain by a few units, which the file records beside each average.
Why the numbers ran high
A surface-tension simulation only means something if the liquid in the box behaves like the real liquid. The most basic check is density: how much mass sits in each litre. If the simulated liquid is too thin, the molecules sit too far apart and the measured pull can be wrong.
What the density check found
After the runs we went back and checked density, which we had not planned to do. The simulated liquids were between 26% and 59% less dense than the published densities. For the hydrofluoro-ether, the three runs ended at 844.1, 743.6 and 801 kilograms per cubic metre, against a published 1520.
Why the liquids were too thin
The record says why that probably happened. Every box was started at 550 kilograms per cubic metre and given only a short time to settle. These fluorinated liquids are very heavy, so a short settling stage left them far too thin.
What we can and cannot conclude
That is a likely cause of the high surface tension. It is not a proven one. We found it after the fact, we have not yet fixed the settling stage, and we have not re-run. Until we do, the honest reading is that our surface-tension method is not yet reliable on fluorinated fluids, and that the control failure means it may not be reliable on simple ones either.
What this means for you
If you make or test cooling fluids, the useful part of this result is the method, not the misses. We fixed the band before the runs, kept a control, published every run, and checked a property we had not planned to check when the answers looked wrong. You can open the file and see each number we used.
If you are weighing our blends
If you are weighing our blends, read the surface-tension figures on this site as model outputs, not as properties of a fluid. The strongest surface-tension difference among the blends we simulated comes from the same kind of simulation, so it carries the same doubt until the method passes its own control. Read it together with the comparisons with published values on our research page.
The bench plan we have written down (not published) includes, second after a mixing check, a pendant-drop surface-tension measurement for exactly this reason: it would replace the simulated number with a measured one.
What this does not show
It does not show that the self-pumping effect fails. It shows that our current way of computing surface tension overshoots on these fluids, so it cannot yet tell us how strong the effect is.
Not yet shown either
It does not show that a fix to the settling stage would bring the numbers into the band. That is a hypothesis we have not tested.
It does not cover the fluids in our own blends. The three fluids here were chosen because their values are published, not because we plan to use them. No FluxZero fluid has been mixed or measured, and nothing in this post changes that.