Twenty of the twenty-one

Squeeze a quartz crystal and a voltage appears across it. Apply a voltage and it changes shape. The effect is piezoelectricity, found by the Curie brothers in 1880, and it is the reason quartz keeps time in a watch and lead zirconate titanate moves the head of an inkjet printer.

The property is a rank-three tensor: it relates a stress, which is a symmetric matrix, to a polarisation, which is a vector. Eighteen independent components before any symmetry is imposed.

Eleven of the thirty-two classes forbid it immediately, because they contain the inversion and an odd-rank polar tensor cannot survive one. That leaves twenty-one, and twenty of the twenty-one permit it.

Why the inversion is fatal #

The argument is one line and it is worth doing, because it is the template for every parity argument in the field.

A rank-three tensor transforms with three factors of the operation matrix. Under the inversion, M = −I, so every component picks up (−1)³ = −1. Neumann’s principle requires the tensor to be unchanged, so every component must equal its own negative, so every component is zero.

The same argument kills every odd-rank polar property in every centrosymmetric class: the pyroelectric vector at rank one, the piezoelectric moduli at rank three, and the third-order optical susceptibilities beyond. Even-rank properties are untouched — the elastic constants pick up (−1)⁴ = +1 and do not notice the inversion at all, which is why the elastic counts are the same for a class and for the class with an inversion added.

So the first cut is clean and needs no computation: eleven classes out, twenty-one remain.

The twenty-first #

The remaining class is 432. It has twenty-four operations — the rotations of a cube, with no reflections and no inversion — and it permits no piezoelectricity at all.

No single operation of 432 forbids the effect. Every subgroup of it permits piezoelectricity — 23 permits one modulus, 422 permits one, 4 permits four — and adding the last few operations takes the count to zero. It is the combination that cancels.

That is unusual and it is why this entry is the most interesting one in the table. Every other zero in the piezoelectric column has a one-line reason: the class contains the inversion, and that is that. This one has no reason shorter than the sum.

What makes 432 different from 4̅3m #

The comparison that sharpens it is with the other cubic class of the same order.

432 and 4̅3m both have twenty-four operations. Neither contains the inversion. Both have four three-fold axes and three axes of order four along the cube edges. They differ in that 432’s four-fold axes carry proper rotations while 4̅3m’s carry rotoinversions, and that 4̅3m has six mirrors where 432 has six two-folds.

4̅3m permits one piezoelectric modulus. 432 permits none.

The class with mirrors permits the effect and the class with only rotations forbids it — which is exactly backwards from the intuition that improper operations are the restrictive ones. Zinc blende and gallium arsenide are 4̅3m and are piezoelectric, and the single modulus they have is the reason.

The twenty are not a uniform group #

Sorting the twenty by how many moduli they permit shows a range from eighteen down to one, and the spread says something about what piezoelectricity requires.

The ten polar classes are all in the twenty, and they carry between three and eighteen moduli. That is not a coincidence: a polar class has a direction along which a polarisation may sit, and piezoelectricity is a mechanism for producing a polarisation, so having somewhere to put one helps.

The other ten are non-polar and non-centrosymmetric, and they carry between one and three. 422, 622 and 23 permit exactly one; 4̅3m permits one; 32 permits two, which is quartz’s.

So a polar class permits more piezoelectric freedom than a non-polar one, on average by a wide margin, and the reason is the parity argument again read at rank three rather than rank one: a class that cancels the rank-one polar tensor cancels much of the rank-three one too, and a class that does not cancel the rank-one one has less machinery for cancelling anything odd.

The zero is not a bug, and here is why that had to be checked #

A property count of zero on a class with no inversion is exactly the kind of result that should be suspected before it is believed. The site’s habit is that a surprising output gets a second route, and this one got two.

The character sum was checked against the projected components. The shape calculation builds the averaging projector explicitly in a Cartesian frame and applies it to each of the eighteen components; every one comes back zero. That calculation shares no arithmetic with the character sum — one is a trace of integer matrices, the other a sum of outer products in floating point — and the two must agree or the build fails.

And the result is in the literature, where it has been since Voigt. 432 is the standard example of a non-centrosymmetric class that is not piezoelectric, and every table of piezoelectric classes lists twenty rather than twenty-one for this reason.

That agreement is a check on the machinery rather than a source for it. The number twenty is computed here and the literature is what it is compared against, in the same arrangement lib/space.js uses for the arithmetic-class counts — a number from a book appears in the code once, is read by nothing that produces an answer, and exists to be disagreed with.

The shape of the cancellation #

It is worth looking at what the sum is actually doing, because “the whole group cancels it” is unsatisfying as an explanation.

The piezoelectric representation of a group is a rank-three object, and the question is whether it contains a copy of the trivial representation. For most non-centrosymmetric classes it does, because the group is small enough that eighteen dimensions leave room.

432 is the largest group with no improper operations at all — the rotation group of the cube, order twenty-four — and the rotation groups are precisely the ones with the fewest one-dimensional representations to hide a trivial component in. Its subgroup 23 has order twelve and permits one modulus; doubling the group to 432 removes it.

So the answer is that 432 is the biggest chiral crystallographic group, and eighteen dimensions is not enough to survive it. That is a real explanation and it is still not a one-line one, which is the honest position.

Where the cancellation comes from, in one character #

There is a shorter reason for the difference between 432 and 4̅3m than the eighteen-component projection, and it fits in a sentence once the right question is asked.

The piezoelectric tensor is a vector index times a symmetric pair of indices, so as a representation it is V ⊗ Sym²V, and the number of independent moduli is how many times the trivial representation appears in that product. Decomposing it needs only which representation the vector belongs to.

In 432 — the rotation group of the cube — the vector is T₁. In 4̅3m it is T₂. That single difference is the whole answer: Sym²V comes out as A₁ + E + T₂ in both cases, and multiplying it by T₁ produces T₁ + (T₁ + T₂) + (A₂ + E + T₁ + T₂), in which no A₁ occurs — while multiplying it by T₂ produces T₂ + (T₁ + T₂) + (A₁ + E + T₁ + T₂), in which exactly one does.

So the cancellation is not a coincidence of twenty-four terms adding to nothing; it is a statement about which irreducible representation carries a polar vector, and the two cubic groups of order twenty-four differ in exactly that. Adding mirrors changed which representation the vector sits in, and that is the sense in which the class with mirrors is the less restrictive one — an answer the parity intuition cannot reach because parity is about inversion and neither group has one.

The same question at rank four, where nothing is forbidden #

Setting the piezoelectric count beside the elastic one makes the parity argument’s scope visible, and it is narrower than it looks.

The elastic tensor is rank four, so it transforms with four factors of the operation matrix and an inversion multiplies it by (−1)⁴ = 1. Every class permits elasticity, centrosymmetric or not, and the classification does nothing but reduce the count of independent constants — twenty-one in the triclinic class down to three in the cubic ones, 432 included.

That is the general rule the twenty are a special case of: an odd-rank polar property is killed by a centre and an even-rank one is not. Piezoelectricity at rank three is forbidden in eleven classes; the polar vector at rank one is forbidden in twenty-two; the elastic tensor at rank four is forbidden nowhere. And 432 is outside that pattern entirely — it has no centre, the parity argument says nothing, and what forbids the effect there is a fact about a representation rather than about a sign.

What the twenty are good for #

The classification does its usual job: it rules things out with certainty and permits things with none.

A material showing piezoelectricity is in one of the twenty, and that is a hard constraint on any proposed structure. A structure refined in a centrosymmetric space group cannot account for a measured piezoelectric response, and the standard resolution is that the refinement is wrong — this is one of the more common ways a published structure gets corrected.

A material in one of the twenty may show nothing measurable. The twenty is a permission and permitted is not present. The strongest piezoelectrics in use are ferroelectric ceramics in polar classes, poled to align their domains, and their coupling is two orders of magnitude above quartz’s — which is in class 32, permits two moduli, and is used for stability rather than for strength.

What the Curies actually did #

The 1880 discovery is worth a paragraph because it was a prediction rather than an observation, which is rare in this subject and is the reason the brothers went looking.

Pierre and Jacques Curie knew about pyroelectricity, and they knew — this is the part that matters — that pyroelectric crystals are those with a polar axis. Their idea was that if heating a crystal along a polar direction produces charge, then compressing it along the same direction ought to as well, since both are ways of changing the separation of charges along that axis. They looked in tourmaline, quartz, topaz and Rochelle salt, and found it.

The reasoning was symmetry reasoning before the tensor formalism existed to state it. What they had was the observation that the polar direction is the special one, and the guess that mechanical and thermal excitations of it would behave alike. Both are correct, and the modern statement is that the pyroelectric vector and the piezoelectric tensor are the rank-one and rank-three members of the same family of odd-rank polar properties, forbidden by the same operation for the same reason.

What their reasoning could not have found is the twenty classes that are not polar and are piezoelectric anyway — 222, 422, 32, 4̅3m and the rest. Those have no polar axis at all, so the compress-the-special-direction argument does not apply, and quartz in class 32 is one of them. The Curies found the effect in quartz because they tried it, not because their argument predicted it.

Where the exactness stops #

The count is of independent moduli, not of magnitudes. Class 1 permits all eighteen and a triclinic crystal may have eighteen negligible ones.

Domains cancel. An unpoled ferroelectric ceramic is a mosaic of domains whose piezoelectric responses point in different permitted directions, and the bulk response is near zero even though every domain has one. Poling — applying a field to align them — is what makes such a material useful, and it is a processing step rather than a symmetry fact.

And the effect measured is not always the effect named. Electrostriction, which is a quadratic response to a field, occurs in every class including the centrosymmetric ones, and a measurement that does not distinguish the linear from the quadratic response can report a piezoelectric effect where none is permitted. The symmetry argument constrains the linear term exactly and says nothing about the quadratic one.

The converse effect, and why symmetry gives it for free #

Applying a field to a piezoelectric crystal deforms it, and the coefficients of that converse effect are the same eighteen numbers in the same arrangement. That is thermodynamics rather than symmetry — the two effects are the two partial derivatives of one free energy, so the matrix of one is the transpose of the other — but the symmetry argument covers both at once for a reason worth naming.

Neumann’s principle constrains the tensor, not the experiment. A component that is zero is zero whichever direction the physics is read in, so a class permitting no direct piezoelectricity permits no converse effect either. 432 cannot be made to change shape in an electric field, linearly, by any arrangement.

This is a small point and it is the kind that a permission-based table makes easy to get right. Having computed which components of one tensor may be non-zero, every physical effect that tensor describes is covered, and there is no separate calculation for each direction the arrow points.

The same observation applies across the whole table: the dielectric tensor’s shape is simultaneously the shape of thermal expansion, of electrical conductivity and of thermal conductivity, because all four are symmetric rank-2 polar tensors and the principle does not care which physics produced them. Six characters cover several dozen named effects, which is the real economy of doing this by rank rather than by phenomenon.

Where the twenty are used, and the one that dominates #

Naming the materials makes the gap between permission and practice concrete, because the classes that matter industrially are not the ones with the largest counts.

Quartz, class 32, two moduli. Used in oscillators, filters and pressure sensors. Its coupling is modest and its temperature stability is exceptional, and it is the stability that put it in every watch and radio.

Lithium niobate, class 3m, four moduli. Surface-acoustic-wave devices and optical modulators; strong coupling and a high Curie temperature.

Zinc oxide and aluminium nitride, class 6mm, three moduli. Thin-film resonators, where being depositable as an oriented film matters more than the size of the effect.

Lead zirconate titanate, class 4mm in its tetragonal phase, three moduli. The transducer material, with a coupling two orders of magnitude above quartz — and it is a ferroelectric ceramic, poled to align its domains, so what is being used is not a single crystal at all.

Two observations follow. The counts of the classes involved are two, four, three and three — near the bottom of the range, not the top — because the useful materials are ones with enough symmetry to be stable and reproducible. And the strongest of them is in a polar class and works by domain alignment, which is a mechanism symmetry permits and does not describe.

The twenty is the list of places to look. It is not a ranking, and the ranking it most resembles runs the other way.

It is worth recording what 432 looks like in practice, since a class that forbids piezoelectricity while having no centre of symmetry sounds like it should be rare and is not especially.

Beta-manganese is in it. So are several intermetallics and a number of high-pressure phases. None of them is piezoelectric, and none of them is piezoelectric for a reason that has nothing to do with their chemistry: the arrangement of rotations in a cubic group with no improper operations cancels the rank-three coupling exactly, whatever is sitting at the lattice points.

That is the cleanest kind of statement this field makes — a prediction about every material in a class, made before any of them is examined, and one that no measurement will ever contradict.

The zero also sits inside the general caution this field carries. Twenty classes permit the effect and permission is not presence; one class forbids it outright, and that is the direction in which the principle says something no measurement will contradict.

Where this goes #

432 forbids piezoelectricity and permits something else. It is one of the eleven enantiomorphic classes, it can hold a single-handed structure, and it permits optical activity — while 4̅3m, which is piezoelectric, does not. Two cubic classes of the same order, each permitting exactly what the other forbids, is the cleanest illustration in the field of why “amount of symmetry” is not a quantity.