Fluids

The straw that holds its water upside down

Put a finger over the top of a drinking straw full of water and lift it out: the water stays in, with nothing under it but air. A glass of water turned upside down does not, unless a card is held over its mouth. Air pressure is the same under both and could hold up ten metres of water in either. What it cannot do is stop the surface from rippling, and a ripple on the underside of water grows, because the heavy fluid is on top. Surface tension stops short ripples, a sealed tube allows only ripples that fit inside it, and a straw is too narrow to fit one that grows.

Assumes: The skin that is not a skin · The thread that cannot stay a thread

Dip a drinking straw into a glass of water, put a finger over its top, and lift it out. The water stays in the straw. Take the finger away and it falls out at once. The usual explanation is that the air pressure below the water holds it up, and that removing the finger lets air in above, so the pressures balance and gravity wins. All of that is true, and it explains nothing about the more interesting comparison, which is with the glass the straw came out of.

Turn a full glass upside down and the water falls out, whatever is done to the top of the glass. The base of a glass seals it far better than a fingertip seals a straw. The air pressure below is the same, a hundred kilopascals, enough to hold up a column of water ten metres high, and a glass holds only ten centimetres. By the air-pressure argument an inverted glass should hold its water as easily as a straw does, and it does not. Hold a postcard over the mouth before turning it, the old party trick, and then it does. What the card supplies is not pressure. It is flatness.

Why a flat underside will not stay flat

Water resting on air is the wrong way up. A surface between two fluids is stable when the heavier one is underneath: push a bump up into the lighter fluid and gravity pulls it back. Put the heavier fluid on top and the same bump is pushed further, because the water above a dip weighs more than the air it displaced, so dips deepen and bulges rise. This is the Rayleigh–Taylor instability, and it is what empties a glass. The water does not fall as a block. The surface buckles, a tongue of water falls on one side while a bubble of air rises on the other, and the column drains through its own disturbance, which is why an inverted glass empties with a gulping sound and a bottle glugs.

It is worth being clear first about what air pressure does do. In a straw ten centimetres long, sealed by a finger, the water at the bottom is at atmospheric pressure, held there by the air outside, and the water just under the finger is lower by the weight of the column, ρgh\rho g h, about one per cent of an atmosphere. The finger feels that as a slight suction. A column ten metres long would bring the top to zero absolute pressure, and a longer one would have to go into tension, which water can sustain, as the column that is pulled, not pushed found in the xylem of trees, but which a column in a straw never needs. So the straw’s weight is no problem for the air to carry, and neither is the glass’s.

Air pressure cannot help with the instability, because it acts equally on the whole surface. It can hold up the column’s weight, but only if the surface stays flat; once the surface is rippled, the pressure in the water at the crest of a bulge and at the bottom of a dip differs by the weight of the water between them, and nothing balances that. What resists a ripple is surface tension. The skin that is not a skin described it as an energy per unit area, and a rippled surface has more area than a flat one, so making a ripple costs energy, more for short ripples than for long ones.

For a ripple of wavenumber kk, wavelength 2π/k2\pi/k, the two effects can be written in one line. Gravity drives the growth in proportion to Δρ gk\Delta\rho\, g k and surface tension σ\sigma resists in proportion to σk3\sigma k^3, so the growth rate ss obeys

s2=Δρ g k−σk3ρwater+ρair.s^2 = \frac{\Delta\rho\, g\, k - \sigma k^3}{\rho_{\text{water}} + \rho_{\text{air}}}.

When s2s^2 is positive the ripple grows exponentially; when it is negative the ripple oscillates and goes nowhere. The two terms balance at k=1/ℓck = 1/\ell_c, with

ℓc=σΔρ g,\ell_c = \sqrt{\frac{\sigma}{\Delta\rho\, g}},

the capillary length, 2.73 millimetres for water. How high water will climb met the same length as the scale at which surface tension and gravity are equally strong, setting how far water rises up a glass wall. Ripples longer than 2πℓc2\pi\ell_c, seventeen millimetres, grow. Shorter ones are stiffened out of existence.

Which ripples on an upside-down water surface grow. The growth rate of a ripple on the underside of a water column against its wavenumber, drawn negative where the ripple only oscillates: gravity drives every ripple, surface tension resists short ones, and the crossing is at 1/ℓc, ℓc = 2.73 mm. A tube admits only certain ripples; the dots mark each tube's slowest volume-conserving one with the contact line free to slide and, for the wider two, pinned at the rim (the straw's pinned ripple lies far off to the right). A 6 mm straw admits nothing that grows; a 70 mm glass admits ripples that grow either way; a 14 mm test tube is stable only if its rim pins the edge.
Fig. 1 The growth rate squared of a ripple on the underside of a column of water, against its wavenumber. Positive means it grows and the water falls; negative means it oscillates and the water stays. The crossing is at 1/ℓc1/\ell_c, 0.367 per millimetre. Each tube’s pair of dots marks the slowest ripple it allows: left with the edge of the surface free to slide on the wall, right with it pinned to the rim. A straw’s ripples are all on the stable side; a 70 mm glass’s are deep in the unstable side; a 14 mm test tube is stable only if its rim pins the edge.

The same relation, with the sign of gravity reversed, is the dispersion relation of ripples on a pond, which the speed below which nothing on water leaves a wake used to find a slowest wave of 23 centimetres per second. The wavenumber at which that slowest wave travels is the same 1/ℓc1/\ell_c. On a pond it is where gravity hands over to surface tension as the restoring force; turned upside down, it is where gravity stops being a restoring force at all and starts tearing the surface apart. One crossing serves both, and the upside-down version is the more dramatic, because on one side of it the surface survives and on the other it does not.

The ripples a tube will admit

In open water every wavenumber is available, so an unbounded inverted surface always has ripples long enough to grow, and nothing can hold it. A tube changes that. The surface of the water in a tube is a disc, and only certain shapes of ripple fit on a disc: the ones that satisfy the condition at its edge. They are the normal modes of the disc, like the modes of a drumhead, and their shapes are Bessel functions. Each has a definite wavenumber, set by the tube’s radius RR, and the longest available wavelength is a few times the radius. If that longest available ripple is shorter than 2πℓc2\pi\ell_c, every ripple the tube admits is stable, and the column stays.

Two further conditions decide which mode is the longest. The first is that the column is sealed. A finger on the straw, or the base of the glass, means that no water can leave or enter except by air bubbling up through it, so the surface cannot move up or down as a whole; that would change the volume of the column. The simplest disturbance, a uniform sag, is forbidden. Every allowed ripple must lift as much water on one side as it lowers on the other.

The first ripple a sealed tube allows. The shape of the slowest ripple that can grow on the underside of a liquid column in a sealed tube, across a diameter. Liquid cannot leave or enter, so the surface cannot simply sag; the first allowed motion is a slosh, up on one side and down on the other. If the contact line can slide, the surface meets the wall level, and the ripple's wavenumber is 1.841 over the radius; if it is pinned to the rim it must meet the wall at zero height, and the wavenumber is 3.832 over the radius. The pinned ripple is twice as short and twice as hard for gravity to drive.
Fig. 2 The slowest volume-conserving ripple of a liquid column in a sealed tube, across a diameter: a slosh, up on one side and down on the other. With the edge free to slide the surface meets the wall level and the wavenumber is 1.841/R1.841/R; pinned to the rim it meets the wall at zero height and the wavenumber is 3.832/R3.832/R.

The slowest such ripple is a slosh: the surface tips, rising on one side of the tube and falling on the other, with a node across the middle. Its exact shape depends on the second condition, what happens at the wall. If the edge of the water surface is free to slide up and down the wall, as it is when the meniscus sits inside a clean glass tube, the surface meets the wall at a fixed angle and its slope there does not change; the slosh then has wavenumber 1.841/R1.841/R, the first zero of the derivative of the Bessel function J1J_1. If the edge is pinned, as it is when water fills the tube right to a sharp rim and the contact line catches on the edge, the surface cannot move at the wall at all; the slosh then has to fit between two fixed points and has wavenumber 3.832/R3.832/R, the first zero of J1J_1 itself. The angle a liquid makes explained why a contact line catches on a sharp edge: the contact angle can take any value across the corner without the line moving, so a rim holds the edge in place over a wide range of tilts. The same pinning is what lets a drying drop of coffee keep its footprint and pile its grounds at the edge, which the ring the drop leaves behind traced to an edge that cannot move. Here a pinned edge does something more useful: it doubles the width of tube that can hold water.

Setting the slosh’s wavenumber equal to 1/ℓc1/\ell_c gives the widest tube that holds:

Dfree=2×1.841 ℓc=3.68 ℓc,Dpinned=2×3.832 ℓc=7.66 ℓc.D_{\text{free}} = 2 \times 1.841\,\ell_c = 3.68\,\ell_c, \qquad D_{\text{pinned}} = 2\times3.832\,\ell_c = 7.66\,\ell_c.

For water that is 10.0 millimetres with a sliding edge and 20.9 millimetres with a pinned one. A drinking straw, six millimetres across, is comfortably inside both. A wide glass, seventy millimetres across, is far outside both and admits dozens of growing ripples. A test tube of fourteen millimetres falls between: it holds when the water is filled to the brim and the surface hangs from the rim, and spills when the meniscus is drawn up inside the tube and can slide. Both outcomes are observed, which has made the inverted test tube a demonstration that works for some people and not for others.

Six liquids, one rule

The rule contains the liquid only through its capillary length, which is the square root of surface tension over density. Nothing else about the liquid enters, not its viscosity, not how well it wets the wall except through whether the edge slides or sticks.

How wide a tube each liquid can be held upside down in. The largest tube diameter in which a column of each liquid stays put upside down with the top sealed: the lower end of each bar with the contact line free to slide, 3.68ℓc, and the upper end with it pinned to the rim, 7.66ℓc. Water: 10.0 to 20.9 mm. Mercury, thirteen times as dense, holds in nearly the same tubes because its surface tension is nearly seven times as strong. Ethanol and liquid nitrogen, with weak surface tension, need narrow tubes.
Fig. 3 The widest tube that holds each liquid upside down, sealed at the top: each bar runs from the free-edge limit, 3.68 ℓc3.68\,\ell_c, to the pinned-edge limit, 7.66 ℓc7.66\,\ell_c. Water 10.0–20.9 mm; mercury, thirteen times denser but nearly seven times stronger in surface tension, holds in nearly the same tubes; liquid nitrogen needs tubes under 4 mm.

The surprising entry is mercury. It is thirteen and a half times as dense as water, so gravity drives its ripples thirteen times harder. Its surface tension is also nearly seven times water’s, the highest of any liquid at room temperature, and the capillary length depends on the ratio. The two almost cancel, and mercury is held upside down in tubes nearly as wide as water is. That is consistent with the old experience of Torricelli’s barometer, in which a column of mercury stands in a tube inverted into a dish; when such a tube is lifted with a finger over the open end, the mercury stays, provided the tube is narrow. Mercury does not wet glass, which shows that adhesion to the wall is not what holds the column. What holds it is the stiffness of its whole surface against rippling.

Ethanol and liquid nitrogen are the opposite case. Their surface tensions are a third and an eighth of water’s, their densities not much lower, and their capillary lengths are short, so only narrow tubes hold them. A pipette of alcohol drips where the same pipette of water would not, which every chemist has seen without necessarily connecting it to the shape of a Bessel function.

Gravity sets the scale

The capillary length goes as one over the square root of gg. Weaken gravity and the critical diameter grows; strengthen it and it shrinks.

The weaker the gravity, the wider the tube that holds. The widest tube that holds water upside down, against the strength of gravity: the diameter scales as one over the square root of g, because the capillary length does. With the edge free to slide it is 10 mm on the Earth, 25 mm on the Moon and 6.3 mm at Jupiter's gravity; with the rim pinned, about twice as much. In orbit, with no gravity, any vessel holds.
Fig. 4 The widest tube that holds water upside down, against the strength of gravity, scaling as g−1/2g^{-1/2}: with the edge free to slide, 10 mm on the Earth, 25 mm on the Moon, 6.3 mm at Jupiter’s gravity; with the rim pinned, about twice as much.

The structure of the argument, a destabilising force that wins at long wavelengths against a restoring force that wins at short ones, with a confining size that removes the long wavelengths, is the structure of gravitational collapse as well. The disturbance that grows instead of travelling found that a self-gravitating gas is stable to compressions shorter than the Jeans length and unstable to longer ones; a cloud smaller than that length cannot collapse, as a tube narrower than its critical diameter cannot spill. In both the critical size is the square root of a ratio of a restoring stiffness to a driving force.

On the Moon a tube two and a half centimetres wide holds water upside down, and with a pinned rim one of five centimetres does, the width of a small jar. In a centrifuge spinning at a thousand times gravity, only a capillary a third of a millimetre across holds. In orbit, where the water and the air around it are in free fall together, there is no driving term at all, and liquid stays wherever surface tension puts it. Fuel tanks on spacecraft are designed around exactly that: in weightlessness the propellant clings to walls and vanes by surface tension and has to be guided to the outlet by the shape of the tank, because nothing pulls it there.

How quickly the water lets go

A tube that is too wide does not empty instantly. The ripples that can grow start from whatever tiny disturbance is present, and grow exponentially at their rate ss.

How fast an unheld column lets go. The e-folding time of the fastest-growing ripple the tube allows, against its diameter, for water. Just above the critical diameter the only ripple the tube allows grows slowly, so a tube near the limit may hesitate before it empties; well above it the time falls to 27 ms for a 70 mm glass, and an inverted glass of water empties in a few tens of milliseconds, by bubbles rising through it as the surface buckles.
Fig. 5 The e-folding time of the fastest-growing ripple a tube admits, against its diameter, for water. Just above the critical diameter only one slow slosh is unstable and the time is long; well above it the fastest ripple grows e-fold in 27 ms for a 70 mm glass.

Far above the critical size, a tube admits so many ripples that one near the fastest-growing wavelength of open water, about thirty millimetres, is always available, and the growth time settles to a few tens of milliseconds. How long the surface hangs depends on how large the ripple is to begin with. Even a perfectly still surface ripples with thermal motion: the surface whose thickness depends on the view found that capillary waves excited by heat roughen water by a few tenths of a nanometre. Growing from there to a millimetre takes fifteen e-foldings, about four tenths of a second. Any real handling disturbs the surface far more, by micrometres at least, from which a millimetre is seven e-foldings away, a fifth of a second. That is roughly how long an inverted glass of water hangs before it empties, which is too short for the eye to register as a pause but long enough for a fast camera.

Near the critical size, only the slowest sloshing mode is unstable, and only just. Its growth rate falls to zero at the critical diameter and the e-folding time rises without limit. A tube just slightly too wide therefore does not dump its water but hesitates, wobbles, lets a bubble in, and drips. That intermediate behaviour is the inverted test tube’s unreliability seen from the other side: in a tube whose critical diameter is close to its actual diameter, the outcome depends on small things, the way the tube was lifted, a fleck of dirt on the rim, a slight tilt, any of which can tip the balance.

The card, the sieve and the thread

The postcard over a glass works by removing the long ripples. Pressed against the water surface, it is a flat, rigid boundary; the water surface cannot ripple at all under it, and air pressure then holds up the column exactly as the straw argument says. If the card is replaced by a fine sieve or a piece of gauze, the glass still holds. The mesh divides the surface into thousands of small surfaces, each pinned at the wires around its hole. Each hole can admit only ripples that fit inside it, and a millimetre-sized hole admits nothing long enough to grow. Water can move from one hole to another, so a hole’s surface can sag as long as another bulges, which relaxes the volume constraint for each hole separately, but the critical size of a pinned round hole for such a sag is still about 2.4ℓc2.4\ell_c in radius, more than a centimetre across. Every kitchen sieve is far below it.

The instability whose cure this is has a sibling in which surface tension plays the opposite part. The thread that cannot stay a thread found that a cylindrical jet of water breaks into drops because surface tension drives the ripples that reduce its area, with gravity irrelevant. Upside-down water is the reverse: gravity drives the ripples and surface tension resists them. In the jet, long ripples grow and short ones are stable because short ripples increase the area; in the inverted column, long ripples grow and short ones are stable because short ripples cost too much area. Both are cured by confinement only in the inverted case, because a jet has no wall to restrict its wavelengths, while a column in a tube does.

What the stability analysis leaves out

The calculation is linear. It decides whether a small ripple grows and says nothing about what happens once it is large, or about whether a large disturbance can overturn a column that is stable to small ones. A straw shaken hard enough loses its water; tapping a narrow inverted tube can start it draining. Linear stability is a statement about gentle handling.

It treats the column as deep and the air below as unbounded, and the wall as either perfectly sliding or perfectly pinned. Real contact lines do neither: they stick until the contact angle reaches a threshold and then slide, so the true condition at the edge lies between the two limits, and which limit applies depends on the cleanliness of the wall and the shape of the rim. The figures therefore give a band for each liquid rather than a number, and a tube’s real critical diameter lies somewhere in it.

The static surface is taken as flat. In a narrow tube it is a curved meniscus, concave or convex depending on the contact angle, and the pressure jump across it shifts the balance slightly. And the column is taken as sealed with nothing compressible inside. A finger over a straw usually traps a small pocket of air, and air can compress, so a uniform sag becomes possible after all, resisted by the stiffness of the trapped pocket. A large air pocket can let a column in a narrow tube sag and drip from the bottom even though no ripple can grow. The more air is trapped, the less the column is held.

Still open: where a real edge sits between the limits

Contact-line pinning is one of the least settled parts of the physics of wetting. The angle at which a line on a real surface begins to move, and the way that angle depends on the speed of motion, the surface’s roughness and its chemical heterogeneity, can be measured for a given surface but not predicted from first principles. So whether a particular glass tube, filled a particular way, behaves closer to the sliding or the pinned limit is a question for experiment, and a careful experiment on the inverted test tube would be a clean measurement of how well a rim pins. The theory is classical; what is still open is how it meets the dirty, hysteretic boundary that every real tube has.

The habit worth carrying away is to ask what a boundary allows. An inverted water surface is torn apart by every ripple longer than 2πℓc2\pi\ell_c, seventeen millimetres, and air pressure, acting equally everywhere, does nothing to stop them; a sealed tube admits only ripples that conserve its volume, the slowest a slosh of wavenumber 1.841/R1.841/R with the edge sliding and 3.832/R3.832/R with it pinned, so water stays upside down in any tube narrower than 10 mm and in a pinned one narrower than 21 mm. A straw holds its water because it is too narrow to contain a ripple that grows. A glass is not.

Part 14 of 14

This essay is one argument about Surface tension. The others:

The objects named here

The third axis, after the field and the reading path: the things themselves, and every essay that touches each one.

Bessel functionCapillary lengthContact angleNormal modesRayleigh taylor instabilitySurface tension