Tesla Valve

How a Tesla valve works

Picture a straight pipe. Now, every few centimeters, split off a curved side-loop that arcs away and rejoins the main channel — but make it rejoin pointing back upstream, like a freeway on-ramp built backwards. String a dozen of these teardrop loops in a row. That's a Tesla valve, and the whole trick is in which way those loops aim.

Send fluid in the forward (easy) direction and it behaves like water that ignores a backwards on-ramp: it mostly stays in the straight channel, blows past the mouth of each loop, and the loops carry only a trickle. The pressure drop is barely more than an equivalent straight pipe.

Send fluid the reverse (hard) direction and the geometry turns hostile. Now the loops are aimed into the oncoming flow. Each loop scoops up part of the stream, whips it through a roughly 180° turn, and fires it back out as a jet that collides head-on with the main flow coming down the straight channel. The collision spawns turbulence and recirculating vortices that partially plug the passage. Multiply that across a dozen segments in series and the reverse flow has to bull through a dozen self-inflicted collisions. The pipe resists itself.

The key idea: there is no flap, ball, spring, or membrane. Nothing seals, nothing wears, nothing can fatigue or stick. The one-way behavior is baked permanently into the fixed shape of the channel. That's what Tesla meant by a "valvular conduit" — a valve made of plumbing, not parts.

The physics: why the asymmetry exists

A straight, symmetric channel cannot rectify flow — reverse it and by symmetry you get exactly the same resistance. The Tesla valve breaks that symmetry with shape, but the symmetry-breaking only does anything when inertia dominates. The governing dimensionless number is the Reynolds number:

Re = ρ·V·D_h / μ

  ρ   = fluid density        (kg/m³)
  V   = mean velocity        (m/s)
  D_h = hydraulic diameter   (m)  = 4·A / P  (area / wetted perimeter)
  μ   = dynamic viscosity    (Pa·s)

At low Re (slow, viscous, "creeping" flow) inertia is negligible — the fluid follows the streamlines smoothly and reversibly in both directions, so the device barely rectifies (Di → 1). At high Re the captured jets carry real momentum, collide, separate, and shed vortices — and the reverse path lights up with loss. So a Tesla valve is fundamentally an inertial device.

The performance number is diodicity, the ratio of pressure drops at equal flow rate:

Di = ΔP_reverse / ΔP_forward      (both measured at the same Q)

  Di = 1      → no rectification (a straight pipe)
  Di > 1      → blocks reverse harder than forward
  Di → ∞      → an ideal one-way valve

Each direction's loss can be written with a minor-loss coefficient K:

  ΔP = K · ½·ρ·V²       so      Di = K_reverse / K_forward

Because the loss scales with ½·ρ·V² (dynamic pressure), the absolute blocking force grows with the square of velocity, and the diodicity itself climbs with Re until vortex shedding saturates. Staging matters too: because the reverse penalties of N segments roughly stack in series, multistage valves multiply the single-element effect — which is exactly why Tesla drew a long cascade in his patent rather than one loop.

Worked example: pressure penalty in a microchannel

Take a water micropump channel, the kind etched into a chip, running near the regime where a Tesla valve starts to work:

Fluid:           water    ρ = 1000 kg/m³,  μ = 1.0e-3 Pa·s
Hydraulic dia:   D_h = 200 µm = 2.0e-4 m
Mean velocity:   V   = 3.0 m/s
Reynolds number: Re  = ρ V D_h / µ
                     = 1000 · 3.0 · 2.0e-4 / 1.0e-3 = 600   (inertial — good)

Dynamic pressure:  q = ½ ρ V² = 0.5 · 1000 · 3.0² = 4500 Pa

Suppose a 15-element valve measures (typical lab numbers):
  forward minor-loss   K_fwd = 12
  reverse minor-loss   K_rev = 22

  ΔP_forward = K_fwd · q = 12 · 4500 = 54 kPa
  ΔP_reverse = K_rev · q = 22 · 4500 = 99 kPa

  Diodicity  Di = ΔP_rev / ΔP_fwd = 99 / 54 ≈ 1.83

So at 3 m/s the valve fights reverse flow about 1.8× harder than forward — a useful bias, but a far cry from a check valve. Now drop the velocity to 0.3 m/s (Re ≈ 60, viscous): the colliding jets never form, K_rev collapses toward K_fwd, and Di slumps toward ~1.1. The same device that rectifies at 3 m/s is nearly useless at 0.3 m/s — the headline trade-off of every Tesla valve.

Tesla valve vs other one-way devices

	Tesla valve	Ball check valve	Reed / flapper valve	Duckbill valve	Diaphragm pump valve
Moving parts	None	Ball + spring/seat	Flexing reed	Elastomer lips	Membrane + seats
Diodicity (effective)	~1.5–2	100–10,000+	50–1,000+	50–1,000+	100–1,000+
Forward pressure drop	Moderate (always on)	Low when open	Low when open	Low–moderate	Low when open
Wear / fatigue	None (monolithic)	Seat wear, sticking	Reed fatigue cracks	Lip set / tearing	Membrane fatigue
Min. feature size	Tens of µm (etchable)	~mm minimum	Sub-mm (hard)	~mm minimum	~mm minimum
Particle / bubble tolerance	Excellent	Poor (jams)	Moderate	Good	Poor
Static sealing (zero flow)	None — leaks both ways	Seals tight	Seals	Seals	Seals
Works at low Reynolds number	No (needs inertia)	Yes	Yes	Yes	Yes
Temperature / chemistry limit	Material of the channel only	Spring + elastomer limited	Reed material limited	Elastomer limited	Elastomer limited

The pattern is clear: the Tesla valve loses badly on rectification and can't hold a static seal, but it sweeps the durability, miniaturization, and fouling-tolerance columns because there is literally nothing inside it to break.

Design parameters and how to tune diodicity

Modern Tesla-valve design is a CFD-optimization problem. The levers engineers turn:

• Loop turn angle. The diverting loop must rejoin the main channel at a steep angle (Tesla's original is roughly 90° back into the flow) so the reverse jet collides hard. Shallower reentry angles bleed the jet downstream and waste the blocking.
• Number of segments (N). Reverse penalties stack roughly in series, so more stages mean higher total diodicity — at the cost of a longer device and higher forward drop too. Patent and lab designs commonly run 10–40 segments.
• Gap / island ratio. The split between the straight channel and the loop sets how much reverse flow gets captured. Too small a loop captures nothing; too large a loop bleeds forward flow and hurts the easy direction.
• Operating Reynolds number. Not a geometry knob but the dominant one: diodicity rises with Re until vortex shedding saturates, typically peaking somewhere in Re ≈ 1,000–10,000 for macro valves. Below ~Re 200 the device is essentially inert.
• Aspect ratio and corners. In etched microchannels the depth is fixed by the etch, so the 2D footprint does all the work; sharp inner corners promote the separation that helps the reverse direction.

Reported single-element diodicities cluster around 1.1–1.5; well-staged classical geometries reach ~1.8–2.0; and CFD-tuned modern shapes (e.g., the GMF — "geometry-modified" — and looped variants studied since the 2010s) have pushed measured diodicity above 2, with some simulations claiming higher still at elevated Re. Either way, you are buying robustness, not a perfect diode.

Where Tesla valves are actually used

Application	Why a Tesla valve	Notes
Piezoelectric / membrane micropumps	Rectify an oscillating membrane into net flow with no wearing check valves	Two valvular conduits replace two ball checks; classic MEMS pump topology
Lab-on-a-chip / microfluidics	Etched in the same lithography step as the channels; tolerant of cells and beads	No sub-mm moving part needed; whole device is one monolithic layer
Fuel-cell & electrolyzer flow fields	Bias gas/liquid flow and shed bubbles without a valve that two-phase flow would jam	Bubble-tolerance is the selling point over mechanical checks
Static mixers / reactors	The "bad" reverse turbulence is repurposed as deliberate mixing	The failure mode of a valve becomes the design intent of a mixer
Valveless pulsejet	Periodic combustion biased to expel exhaust out the tailpipe with no flapper to burn out	Tesla's own pitch; flapper-valve pulsejets famously cook their reed valves
Microelectronics & battery cooling	Passive flow direction control in tiny cooling loops, including studies of two-phase boiling flow	No moving part to fail inside a sealed cold plate

Common misconceptions and pitfalls

• "It only lets flow go one way." No — fluid flows both ways; reverse flow is merely harder. With a diodicity of ~2, reverse flow is only resisted twice as much as forward. It biases flow; it does not block it. Treat it as a soft, leaky diode, never a shutoff.
• "It works as a seal." Pitfall. At zero or near-zero flow there is no inertia, no jets, no blocking — it leaks freely in both directions. If you need a tight static seal, you need a moving check valve.
• "More loops always means much better." Diminishing returns. Stages add forward pressure drop too, lengthen the device, and the per-stage gain shrinks. Past a few dozen segments you're mostly buying pumping cost.
• "It'll work at any flow rate." The single most common design error. Run it below its inertial regime (low Re) and the diodicity collapses to ~1. Always compute the operating Reynolds number first and confirm the geometry will actually bite at that speed.
• "Direction doesn't matter much." It is the whole point. Install the cascade backwards and you've made your easy direction the hard one. The loops have a definite orientation — get it right.
• "It's a forgotten curiosity." It was, for ~70 years — until cheap microfabrication and CFD revived it. Today it's an active research and product area precisely because etched, monolithic, fatigue-free flow control is hard to get any other way.