Why water is bent

Water’s bond angle is 104.5 degrees. The tetrahedral angle is 109.47. The gap is about five degrees, the standard explanation accounts for its sign, and it does not account for its size.

The standard account #

Oxygen has four regions of electron density round it: two O–H bonds and two lone pairs. Four regions arrange tetrahedrally, as the minimisation shows, so the arrangement is roughly tetrahedral and the H–O–H angle should be roughly 109.5.

It is smaller because lone pairs are held closer to the nucleus than bonding pairs — a bonding pair is shared between two nuclei and pulled away — so a lone pair occupies more angular space and squeezes the bonding pairs together.

That is a good argument. It predicts the right sign, it predicts correctly that ammonia’s angle (107) is between water’s and the tetrahedral value, and it generalises to a large family of molecules.

The arrangement that argument starts from is not looked up either. VSEPR, computed minimises the repulsion between four points on a sphere and gets 109.47 degrees, and water’s two bonds occupy two of those directions with its two lone pairs in the other two.

Where it stops being quantitative #

The account contains the phrase “more angular space”, and nothing in it says how much more.

That is the whole difficulty. The rule predicts smaller than tetrahedral, and every observed angle in the family is smaller than tetrahedral, so the rule is never wrong and never sharp. Ammonia at 107 and water at 104.5 are both consistent with it; so would be 108 and 100.

A rule that cannot be wrong is not making a prediction. It is organising a set of facts, which is useful and is a different thing.

What a sharper rule looks like #

Henry Bent proposed one in 1961, and it does considerably better.

Bent’s rule: atomic s character concentrates in orbitals directed toward electropositive substituents, and p character concentrates toward electronegative ones.

The reasoning is that an s orbital has density at the nucleus and is lower in energy than a p; an atom will therefore keep s character for itself where it can. A bond to an electronegative partner has its electrons drawn away, so the central atom invests less s character there — and a lone pair, which is not shared at all, gets the most s character of anything.

More s character means a larger angle, because pure s is spherical and pure p orbitals are at 90 degrees to one another. So lone pairs, being s-rich, spread apart and push the s-poor bonding orbitals together.

Why it is sharper #

Because electronegativity is a number, Bent’s rule is quantitative in the right direction and it predicts a trend rather than a single case.

The oxygen family bears it out. Water is 104.5. Replace the hydrogens with more electronegative fluorines and the angle should fall: OF₂ is 103.3. Replace them with less electronegative substituents and it should rise: in dimethyl ether the C–O–C angle is about 111, above tetrahedral.

That is a prediction with a direction and a rough magnitude, tested across a series, which the lone-pair argument on its own does not supply.

The heavier hydrides, where both accounts struggle #

Now the case that shows the limits of everything above.

Water is 104.5. Hydrogen sulfide is 92.1. Hydrogen selenide is 91. Hydrogen telluride is about 90. The trend is large, systematic, and not a small correction to a tetrahedral starting point.

Neither VSEPR nor Bent’s rule predicts that on its own, and the usual explanation is different in kind: the heavier central atoms barely hybridise at all. Their valence s and p orbitals differ enough in energy and size that mixing them costs more than it gains, so the bonds use nearly pure p orbitals — and two p orbitals are at 90 degrees.

On that account water is the anomaly rather than H₂S. Oxygen is small enough for s–p mixing to be favourable, so its angle is pushed up from 90 toward tetrahedral; sulfur is not, so it stays near 90.

The inversion is worth dwelling on, because it changes which fact needs explaining. Taught in the usual order, the family reads as tetrahedral-with-corrections and the corrections grow embarrassingly large by tellurium — nineteen and a half degrees is not a correction. Taught in the other order it reads as ninety-degrees-with-one-exception, and the exception has a reason attached: the 2s and 2p orbitals of a first-row atom are close in energy and similar in size, and nothing further down the group is.

The same reordering repairs a second oddity. On the tetrahedral-first account, the lone pairs of H₂S must be squeezing its bonds by seventeen degrees while water’s squeeze by five, and no version of “lone pairs take more space” explains why sulfur’s should take three times as much. On the ninety-degrees-first account there is nothing to explain: the bonds are close to pure p, two p orbitals are perpendicular, and the lone pairs are barely involved.

Which explanation to believe #

An honest answer: all three are describing the same energy surface in different vocabularies, and none of them is the calculation.

The shape is whatever minimises the total electronic energy. That is a many-electron problem, computable to high accuracy by modern methods, and it does not decompose cleanly into “lone pair repulsion” or “s character” or anything else. The rules are heuristics for what the calculation will produce.

What separates them is how much they predict. VSEPR gives a family of shapes and a sign; Bent’s rule gives a trend with substituent electronegativity; the hybridisation-cost argument gives the heavy-hydride series. Each earns its place by covering cases the others do not, and treating any of them as the mechanism is a mistake the hybridisation essay takes up in general.

The sp³ hybrid directions point at 109.471 degrees — hybrids are a basis constructs them — which is the geometry all three accounts start from. Water’s actual angle is five degrees below it, and every explanation above is an account of the five degrees rather than of the 109.

The thing water’s angle is often used to prove #

One further caution, because water’s geometry is pressed into service for a claim it does not support.

It is common to see the 104.5 degrees offered as evidence that oxygen is “sp³ hybridised”. It is not evidence of that, and it could not be: hybridisation is a choice of basis, and a basis choice makes no prediction about a bond angle. The angle is an observable and the hybridisation is a description; one cannot confirm the other.

What the angle does support is that the arrangement is closer to tetrahedral than to 90 degrees, which is a statement about the molecule and a useful one.

What symmetry settles without any of this #

There is one statement about water’s electronic structure that requires no repulsion argument, no electronegativity and no hybridisation, and it is worth separating out because it is the only one in this essay that cannot be wrong.

Water is bent, so its point group is C₂ᵥ: a twofold axis and two mirror planes, four operations in all, recovered from its coordinates. Whatever the bonding turns out to be, the two hydrogen 1s orbitals must span a representation of that group, and there is only one thing they can span.

That result forbids the picture the standard account implies. Two equivalent O–H bond orbitals would be degenerate; a₁ and b₂ are not, and water’s photoelectron spectrum accordingly shows them at different energies. The same argument applies to the lone pairs: the two “rabbit ears” of the localised picture are equivalent by construction, while the canonical orbitals are one pure p perpendicular to the molecular plane and one in-plane orbital with s character, at clearly different energies.

Neither picture is wrong — they are related by the same kind of transformation methane’s are — but only one of them is what a measurement of orbital energies couples to, and symmetry says which before any measurement is taken.

The same reduction applied to water’s motions gives a count rather than a rule of thumb: nine Cartesian displacements, less three translations and three rotations, leaves three vibrations — two of a₁ symmetry and one of b₂, and every one active in both the infrared and the Raman spectrum because water has no centre of inversion. What an absence proves does that reduction. A subtraction returning anything other than 3N − 6 would mean the characters were wrong.

Three modes is a count, not a rule of thumb, and the counting is the check: a subtraction that produced anything other than $3N-6$ would mean the characters were wrong.

What it costs to have a rule that can be wrong #

The three accounts in this essay differ in cost in a way that tracks exactly how much they risk.

VSEPR costs nothing. Count the regions, look up the arrangement, adjust downward for lone pairs. It cannot be wrong about water because it predicts only a direction, and a rule that predicts only a direction is free in both senses — free to apply and free of consequences.

Bent’s rule costs an electronegativity, and that is a much larger bill than it appears. There is no single number called the electronegativity of an atom; there are four scales in common use, they are defined from four unrelated things, and — as the dipole essay measures — they order some ordinary elements differently. Bent’s rule inherits that ambiguity, and the inheritance is invisible because nobody states which scale was used.

The hybridisation-cost argument costs a comparison of orbital energies, which is a real quantity from atomic spectra, and it correspondingly predicts the one thing the others cannot: the size of the drop from oxygen to sulfur.

The symmetry statement costs a search over the coordinates, takes milliseconds, and predicts something none of the three others attempts — how many distinct orbital energies there are, and how many vibrational bands. It cannot say what any of them is.

Setting them out that way makes the shape of the subject visible. Each account buys a different kind of prediction with a different currency, and the temptation to be resisted is the one this site names elsewhere: promoting whichever of them is cheapest into a mechanism.

What is measured, and how #

The number itself deserves a word, since it is quoted to a tenth of a degree.

Water’s geometry comes from rotational spectroscopy — the microwave spectrum gives moments of inertia, and the moments give bond lengths and angles. The equilibrium angle is 104.48 degrees and the O–H length 0.9578 ångström.

Those are equilibrium values, meaning the bottom of the potential well. The molecule is never at rest there: even at absolute zero it has vibrational zero-point motion, and the average angle over that motion is slightly different. A quoted geometry always belongs to one of several definitions, and the differences are of the order of the effect this essay is about.

The comparison across the family #

Setting the related molecules beside one another is what makes the trends legible.

Take the lone pairs away entirely and there is nothing left to squeeze, which is the control the sequence needs at its top end.

Those three are the sequence the rule was built on, and one more molecule is enough to show what the sequence is a sequence in.

Methane, ammonia, water: 109.47, 107, 104.5, with zero, one and two lone pairs. The ordering is exactly what the rule says, the spacing is roughly even, and neither fact is derived from anything — the rule was constructed to fit them. Sulfur dioxide is the reminder that the “smaller than tetrahedral” the sequence suggests is not the rule’s content at all: the rule says smaller than the reference for that many regions, and which reference applies is decided by a count that has to be made before the rule can be applied. Get the count wrong and the rule is confidently wrong; get it right and the rule was never at risk.

That is not a criticism so much as a description of what kind of thing the rule is. It organises a family correctly and predicts the next member’s ordering, which is genuinely useful, and it does not compute an angle. Bent’s rule does better because it has a number in it, and even that predicts a trend rather than a value.

The calculation nobody in this essay is doing #

It is worth being explicit about what would settle the question, because the honest answer is that it is not far out of reach and it is deliberately not done here.

The angle is whatever minimises the total electronic energy. Getting that number requires solving for the electronic structure at a series of angles and finding the minimum — a self-consistent field calculation in a small Gaussian basis, which is a few hundred lines and entirely feasible in advance. The two-electron integrals over Gaussian functions have closed forms; the iteration is standard; the answer for water in a modest basis comes out within a degree or two of the measurement, and improving the basis improves it.

That calculation is held back on purpose, and the reason is the discipline the whole site runs on. A wrong self-consistent field that produces plausible numbers is exactly the failure this site exists to prevent. An orbital contour that encloses the wrong fraction can be caught by re-running the integral. A node count can be caught by walking the function. A converged but wrong energy has no such check available from inside itself — it looks like a result, it has the right units and roughly the right magnitude, and the only way to know is to validate it against published values calculation by calculation.

So the boundary this essay draws is drawn deliberately. Everything computed here — the tetrahedral angle from minimising repulsion, the point group from the coordinates, the representations the hydrogens span, the count of vibrations — is exact within a stated model and checkable against something outside itself. The five degrees are not, and rather than a number with no test attached, this page offers three heuristics with their costs marked.

What would settle it experimentally #

One more thing is worth saying, because the essay has so far weighed explanations against one another rather than against measurements.

The three accounts differ in what they predict, so each is testable in a different place.

VSEPR predicts a sign, and every hydride in the family obliges. It is untestable in the only sense that matters: no observed angle in this family could contradict it.

Bent’s rule predicts a trend with substituent electronegativity, and that is testable across a series at one central atom. OF₂ at 103.3 against water at 104.5 against dimethyl ether at about 111 is the test, and the rule passes it. What that series cannot test is the mechanism, since a rule predicting the same ordering for a different reason would pass identically.

The hybridisation-cost account predicts the heavy-hydride series, and it is the only one of the three that predicts a large effect rather than a small one. Water at 104.5 down to H₂Te at about 90 is nineteen degrees, and an account that got the direction right and the magnitude wrong by a factor of two would be visibly refuted.

Ranking them that way puts the most-taught explanation last. VSEPR is the one that cannot fail and correspondingly the one that establishes least, and it survives because it is the only one that can be applied without looking anything up.

Which angle is 104.5 degrees #

The tests above compare measured angles with each other, and that comparison carries an assumption the essay has not made explicit: that a molecule has an angle.

It does not, quite. A bending coordinate is the softest one a triatomic has, water’s bend costs about 1595 cm⁻¹, and even in its lowest vibrational state the molecule is sampling a range of angles rather than sitting at one. Several different quantities are therefore all called “the bond angle”, and they are not equal:

The equilibrium angle is the minimum of the potential — the geometry no molecule is ever at, and the one every calculation reports.

The ground-state average is the expectation value over the zero-point motion. Because the bending potential is not symmetric about its minimum, that average is displaced from it, and the displacement does not vanish as the temperature falls.

The angle a diffraction experiment reports is derived from averaged interatomic distances, and averaging a distance is not the same operation as averaging an angle.

For water these differ by a few tenths of a degree — small against the five-degree gap this essay is about, and not small against the comparison Bent’s rule is tested on. OF₂ at 103.3 against water at 104.5 is a difference of 1.2°, so the test has a margin of perhaps four or five times the ambiguity in what is being compared. It passes, and it passes with less room than the two figures suggest.

The dimethyl ether comparison is the safe one: six and a half degrees is far outside any convention. A rule tested on a large effect is tested; the same rule tested on a degree needs the definitions stated.

The minimisation, the basis, and five sites #

The minimisation the argument starts from is VSEPR computed.

The status of hybridisation as a description rather than a cause is a basis is not a thing, with the sharper version in hybridisation does not explain.

And the case where an arrangement’s inequivalent sites do something visible is five sites.

What the pictures here cannot show. A drawn molecule is at one geometry, and a real one is vibrating through a range of them. Nothing on this page computes an energy, so none of these figures can adjudicate between the explanations — they show the geometries the explanations are about.