— From SU(2) and SO(3) to SO(10): a guide to Georgi’s “Lie Algebras in Particle Physics”
Particle-physics textbooks are lined with symbols that look like magic spells: SU(3)×SU(2)×U(1), SO(10), E8, and so on. But these names can in fact be read: the capital letters indicate “which quantity the collection of transformations preserves,” and the number is “the size of the matrix.” Once you have learned just this much about how to read them, the gauge group of the Standard Model, the accidental degeneracy of the hydrogen atom, and grand unified theories all become readable as sentences written in one and the same grammar.
This page is a map for working through Georgi’s “Lie Algebras in Particle Physics.” We leave the detailed proofs to the textbook and take just the backbone of the story in a single straight line. At the places where you will want to check things by hand, we point you to the paired simulator (liegroup-simulator-en.html) as we go.
Collect some transformations so that (i) a transformation performed after another is also a member (closure), (ii) the “do nothing” transformation is a member (the identity element), and (iii) every transformation has one that undoes it (the inverse) — a collection satisfying these conditions is a group. Two rotations in a row are again a rotation, and the rotation that undoes one is also a rotation. Put “symmetry” into the language of mathematics and groups appear automatically. And the names of the groups that appear in physics are given according to what those transformations preserve.
When an n-dimensional real vector is transformed as x → Ox, we require that the squared length xTx remain unchanged:
This is the orthogonal group O(n). Taking the determinant of both sides gives (det O)2 = 1, that is, det O = ±1. The det = −1 side contains reflections (flips). The collection of only the det = +1 side is SO(n) (S stands for “special,” meaning determinant 1) — the group of pure rotations, without any flipping.
Let us count the degrees of freedom (= the dimension of the group). The left-hand side of the condition OTO = 1 is a symmetric matrix, so there are n(n + 1)/2 independent conditions. Subtracting this from the n2 components gives
Intuitively, you can read it like this: there is one rotation for each choice of “the coordinate plane (xi, xj) in which to rotate,” and the number of such choices is n(n−1)/2 — which is exactly the number found above. SO(2) has 1 (one rotation angle), SO(3) has 3 (the Euler angles), SO(4) has 6, and SO(10) has 45. These numbers will appear later in physics as “the number of gauge particles” (§7).
A quantum-mechanical state is a complex vector ψ, and the inner product ψ†ψ is the total probability. The condition for a transformation ψ → Uψ that preserves it follows from exactly the same argument as for Eq. (1):
This is the unitary group U(n). The reason the symmetries of quantum mechanics are described by a unitary group rather than a real one is that what must be preserved is probability. As for the degrees of freedom: the complex n2 components amount to 2n2 real numbers, whereas U†U = 1 is an equality between Hermitian matrices and therefore imposes n2 real conditions. The net result is n2. Unitarity gives det U = eiφ (absolute value 1), so imposing det = 1 removes exactly one phase:
SU(2) has 3, SU(3) has 8, and SU(5) has 24. The 8 in “eight gluon colors” is precisely this 32 − 1.
The phase space (q1, …, qn, p1, …, pn) of classical mechanics has a third fundamental quantity, neither a length nor a complex inner product: the antisymmetric bilinear form ω(x, y) = xTJy — the sum of the “areas” projected onto the q-p planes. The group that preserves it is the symplectic group:
The dimension is n(2n + 1) (derived in §2). The relation to canonical transformations and Liouville’s theorem comes back in §7.
A matrix in U(n) can almost be decomposed into the phase factor eiθ multiplying the whole matrix (this is U(1), the 1×1 unitary group) and the det = 1 part SU(n). The word “almost” is needed because the matrices e2πik/n·1 (n of them) belong to both and are therefore counted twice; precisely, one has U(n) ≅ (SU(n)×U(1))/Zn. In physics it is convenient to treat the overall-phase U(1) independently, as a separate conserved quantity such as electric charge or hypercharge, and so the gauge group of the Standard Model, too, is written in the form SU(3)×SU(2)×U(1), “S-groups × U(1).”
| Symbol | Preserves | Defining condition | Real dimension | Examples |
|---|---|---|---|---|
| SO(n) | The length of a real vector xTx (and its orientation) | OTO = 1, det = 1 | n(n − 1)/2 | SO(3) = 3, SO(10) = 45 |
| SU(n) | The complex inner product ψ†ψ (probability) | U†U = 1, det = 1 | n2 − 1 | SU(2) = 3, SU(3) = 8 |
| Sp(2n) | The antisymmetric form (area in phase space) | MTJM = J | n(2n + 1) | Sp(2) = 3, Sp(4) = 10 |
| U(1) | The absolute value |z| of a complex number (the phase changes) | U = eiθ | 1 | The gauge group of electromagnetism |
Lie’s discovery can be summarized as follows — for a continuous group, examining just the immediate neighborhood of the identity tells you (almost) everything. Write an element of the group near the identity as O = 1 + εA (with ε infinitesimal) and substitute it into the defining condition. For SO(n):
In other words, the “derivative” of SO(n) is an antisymmetric matrix. An antisymmetric matrix has zeros on the diagonal and n(n − 1)/2 free components in the upper triangle — exactly matching the dimension of the group counted in Eq. (2). Naturally so: the dimension of the group = the dimension of the tangent space at the identity. Similarly for SU(n), pulling out an i in the physicists’ style and writing U = 1 + iεH, we get
The generators are Hermitian matrices. Furthermore, det(1 + iεH) = 1 + iε tr H + O(ε2), so the condition det = 1 becomes tr H = 0 — the generators of su(n) are Hermitian and traceless. As for the degrees of freedom: n real numbers on the diagonal and n(n − 1)/2 complex numbers off the diagonal, which makes n2 real numbers; subtracting 1 for tracelessness leaves n2 − 1. This too agrees with Eq. (4). Sp(2n) goes the same way: putting M = 1 + εS into Eq. (5) gives STJ + JS = 0, which is equivalent to “JS is a symmetric matrix,” so the degrees of freedom are the number of independent components of a symmetric matrix of order 2n, 2n(2n + 1)/2 = n(2n + 1). All the dimensions in the table of §1 come out in this way, simply by linearizing the defining condition.
Repeating an infinitesimal transformation N times gives a finite transformation:
Let us check this with a two-dimensional rotation. Apart from a scale factor, there is only one 2×2 antisymmetric matrix:
A2 = −1 is what allows the Taylor series to fold up into cos and sin, giving the familiar rotation matrix. The exp is the bridge “from the algebra to the group.” Whether the finite transformations built with this exp really obey the group condition can be checked by eye in tab ① “What does the group preserve?” of the simulator: in each of the three modes SO(3), SU(2), and Sp(2,R), as you move the transformation parameters, the invariants (the bond lengths and bond angles of a tetrahedral molecule / the norm ψ†ψ / the area in phase space) stay fixed as numbers. If you switch on the “Deliberately step outside the group” toggle and break the condition, those invariants start to drift away gradually.
So how can the group multiplication eXeY be written in the language of the algebra? As long as the matrices do not commute, it is not eX + Y. The correct answer is the Baker–Campbell–Hausdorff formula:
What matters is not the coefficients of the formula but that the right-hand side is written using commutators alone. In other words, once the commutation relations among the generators [X, Y] = XY − YX are given, the rule for multiplying group elements near the identity is completely determined. So “studying a Lie group” is almost replaced by “studying commutation relations.” This tangent space equipped with commutation relations is called the Lie algebra, and to distinguish it from the group it is written in lowercase letters, so(n), su(n), sp(2n).
In the case of rotations, the commutation relations of the Hermitian generators J1, J2, J3 (R = e−iθ n̂·J) are
This one line is the algebraic DNA of the concept of “rotation.” The reason that the whole theory of angular momentum in quantum mechanics starts from this commutation relation is that angular momentum is the generator of rotations (Noether’s theorem: generator of a symmetry = conserved quantity). §3–4 below explore how far one can get from this one line alone.
There are many sets of matrices that satisfy the commutation relations (11). The smallest is half of the Pauli matrices:
All of them are Hermitian and traceless — exactly the qualifications for su(2) that we derived in §2. And
On the other hand, the generators of three-dimensional rotations (3×3 antisymmetric matrices multiplied by i; in components, (Lk)ij = −iεkij) also satisfy exactly the same commutation relations. As Lie algebras, su(2) and so(3) are the same thing — as long as one looks only at the commutation relations, there is no way to tell them apart.
Lift to the groups, however, and a difference appears. Using (n̂·σ)2 = 1 (which follows from the anticommutation relations of the Pauli matrices), the same folding as in Eq. (9) gives
This is the rotation matrix for spin 1/2. Note that the angle appears as θ/2. Putting in θ = 2π (one full turn):
After one full turn it does not return to itself but picks up a factor −1. It returns only at 4π, that is, after two full turns. In SO(3), of course, R(2π) = 1, so SU(2) turns out to be a group that counts rotations “twice as finely.” In tab ② “SU(2) and SO(3)” of the simulator (liegroup-simulator-en.html), you can turn θ from 0 to 4π and see for yourself how the vector returns after one turn while the spinor takes two.
To state the relation between the two groups precisely, represent a three-dimensional vector x by the 2×2 Hermitian traceless matrix X = x·σ and sandwich it with U ∈ SU(2): UXU† is again Hermitian and traceless, so it must be the representation of some new vector x′:
Here det X = −|x|2 is invariant under the sandwiching, so |x′| = |x| — since the length is preserved, R(U) is a rotation. The map U ↦ R(U) thus gives SU(2) → SO(3), but, as Eq. (16) shows, U and −U give exactly the same R (the sign cancels because the sandwich has the matrix on both sides). That is, this correspondence is 2-to-1, with kernel {±1}: SO(3) ≅ SU(2)/{±1}.
An element of SU(2) can be written as , with |a|2 + |b|2 = 1 — that is, the sum of the squares of the four real components is 1. Hence SU(2) is the unit sphere S3 in four-dimensional space. SO(3) is the space obtained by gluing together the antipodal points U and −U (the real projective space RP3). Even though they are the same locally, their global shapes differ. On S3, every loop can be shrunk to a point (simply connected), but RP3 has “loops that cannot be shrunk,” which shrink only after going around twice — this is the topological origin of the “−1 for a 2π rotation,” which you can feel physically in the “Dirac belt trick,” where the twist in a belt is undone by two full turns.
The fact from §3 that the Pauli matrices (2×2) and the rotation generators Li (3×3) satisfy the same commutation relations is in fact the entrance to a much bigger story. Assigning a matrix D(g) to each element g of the group in such a way that the multiplication table of the group is preserved exactly is called a representation:
The same script (the group) can be performed on a two-dimensional stage or on a three-dimensional stage — the two-component spinor of spin 1/2 and the three-component vector of spin 1 are different performances of the same “rotation.” Every representation can be decomposed into blocks of irreducible representations, which cannot be split into smaller invariant subspaces, so listing all the irreducible representations completes the representation theory. For su(2), this can be done mechanically from the commutation relation (11) alone. Let us follow the derivation in Georgi’s style — the familiar calculation from physics.
Diagonalize Jz and label the states by its eigenvalue m. From the remaining two generators, form J± = Jx ± iJy and Eq. (11) transforms into the following form:
The first equation reads “multiplying by J± shifts the eigenvalue of Jz by ±1”: Jz(J±|m⟩) = (m±1)(J±|m⟩). That is, J± are the raising and lowering operators of a ladder. In a finite-dimensional representation the ladder must have a top rung. Write it as |j, j⟩, and require J+|j, j⟩ = 0 (the highest weight).
Here it is convenient to use the Casimir operator, which commutes with all the generators, J2 = Jz2 + (J+J− + J−J+)/2 and to rearrange with Eq. (17), which gives two equations that differ in the order of the operators: J−J+ = J2 − Jz2 − Jz and J+J− = J2 − Jz2 + Jz. For the highest state, J+|j, j⟩ = 0, so applying the former to |j, j⟩ makes the left-hand side vanish, giving J2 = j(j + 1). (The latter is the form used in the norm calculation of Eq. (18) below.) Within an irreducible representation J2 is a constant, so this value serves as the label of the whole representation. Next, we compute the norm of the state obtained by going down the ladder:
The norm cannot be negative. As we keep descending, this quantity must eventually reach 0 and stop the ladder, so the solution of j(j + 1) = m(m − 1) is m = −j (the other solution, j+1, lies above the upper limit and is therefore excluded). So the ladder runs from m = j down to −j in steps of 1, which is 2j steps. Since the number of steps is a non-negative integer,
Taking the square root of the norm (18) as the normalization factor fixes even the matrix elements completely:
From a single commutation relation, we have derived the allowed dimensions of representations (2j+1) and all the matrix elements. Writing out Eq. (20) for j = 1/2 reproduces half of the Pauli matrices, and for j = 1 the 3×3 Li (up to a change of basis). In tab ③ “Irreducible representations and multiplets” of the simulator, you can choose j and climb up and down the ladder with the raising and lowering operators to see how the coefficients of Eq. (20) change from step to step.
This representation theory is not only about rotations. Wherever the same commutation relations appear, the same conclusions can be used. In 1932, Heisenberg, noting that the masses of the proton and the neutron are remarkably close, regarded the two as “up and down in an abstract space called isospin.” If the strong interaction is invariant under SU(2) rotations in isospin space, particles should be organized into (2I+1)-plets:
| Multiplet | I | Dimension 2I+1 | Mass (MeV) |
|---|---|---|---|
| Nucleon (p, n) | 1/2 | 2 | 938.3 / 939.6 — differ by only 0.1% |
| Pion (π+, π0, π−) | 1 | 3 | 139.6 / 135.0 / 139.6 |
| Δ resonance (Δ−, Δ0, Δ+, Δ++) | 3/2 | 4 | all ≈ 1232 |
The masses within a multiplet are nearly degenerate because the symmetry is nearly exact (it is broken only by the electromagnetic interaction and the mass difference between the u and d quarks). “When you find a set of particles with nearly the same mass, suspect an irreducible representation of a hidden group” — this becomes the pattern of the story that continues in the following sections to su(3) and then to grand unification.
The state label of su(2) was just one: m. This is no accident. In su(2), only one mutually commuting generator can be chosen (any two of Jx, Jy, Jz fail to commute). The only quantities that can be determined simultaneously in quantum mechanics are mutually commuting operators — so the maximal number of simultaneously diagonalizable generators = the number of quantum-number labels that can be attached to a state. This maximal commuting part is called the Cartan subalgebra H1, …, Hr, and their number r is called the rank. The set of labels of a simultaneous eigenstate, μ = (μ1, …, μr) is the weight, which is a point in r-dimensional space. Rank = the dimension of the weight diagram. This is why the multiplets of su(2) (rank 1) were “2j+1 points on a line.”
su(3) has 8 generators (the Gell-Mann matrices λa/2). Of these, the diagonal ones are the two T3 = λ3/2 and T8 = λ8/2, so the rank is 2. In physics, instead of the second Cartan generator T8, one uses the flavor hypercharge Y = (2/√3) T8 and labels the states by the two numbers (I3, Y). Here Y = B + S (baryon number plus strangeness), and the electric charge is read off with the Gell-Mann–Nishijima relation Q = I3 + Y/2 (this Y is a different thing from the weak hypercharge of §7.3; see the caution in §7.3). That is why the multiplets of su(3) can be drawn as plane figures — the triangles and hexagons of the Eightfold Way are nothing but rank-2 weight diagrams. The weights of the fundamental representation 3 (the quarks u, d, s) are
— an inverted triangle. The antiquark 3 is its sign reversal and forms an upright triangle.
The remaining 6 generators can be recombined, exactly like J± of su(2), into raising and lowering operators Eα. Their defining condition is
It has the same form as the first equation of Eq. (17), and the reading is the same: multiplying by Eα translates the weight by the vector α. These displacement vectors α are called the roots. Eq. (21) is also the statement that “the eigenvalue of Eα itself, seen from the Cartan subalgebra, is α,” so a root is nothing other than a weight of the algebra acting on itself (the adjoint representation).
In su(3) there are three pairs of raising and lowering operators. In terms of quarks, they are “which quark is changed into which”:
| Operator | Action | Weight displacement (I3, Y) |
|---|---|---|
| I± (isospin) | d ↔ u | ±(1, 0) |
| V± | s ↔ u | ±(1/2, 1) |
| U± | s ↔ d | ±(−1/2, 1) |
In (T3, T8) coordinates, the 6 roots are ±(1, 0), ±(1/2, √3/2), ±(−1/2, √3/2) and they all have the same length and are spaced at 60°, forming a regular hexagon (when drawn with a Y axis the figure looks stretched vertically, because of the normalization Y = 2T8/√3).
So far we have looked at individual algebras, su(2) and su(3). At the end of the 19th century, Cartan and Killing posed the question in the opposite direction — what sets of commutation relations can make up a “consistent simple Lie algebra” in the first place? Remarkably, this question can be solved completely. The key is the geometry of the roots.
For any root α, the triple Eα, E−α, α·H forms an su(2) (it can be normalized so that it satisfies the same commutation relations as Eq. (17)). Viewing another root β from this “su(2) in the α direction,” β must have some Jz eigenvalue, and as shown in §4, the eigenvalue m of su(2) is quantized in half-integers. The consequence is
— the integer conditions. Multiplying the two together gives
(a value of 4 would mean parallel). That is, the angles between roots are quantized: between the simple roots chosen as a basis (their inner products are non-positive), there are only four possible angles: 90°, 120°, 135°, and 150°, and the length ratios are respectively arbitrary, 1, √2, and √3. The “blueprints of algebras,” which one would expect to come in a continuous infinity, have suddenly become a finite combinatorial problem.
Draw the simple roots as dots (○), and connect two dots according to the angle between them: a single line for 120°, a double line for 135°, a triple line for 150° (no line for 90°), and indicate a difference in length by an arrow (long → short) — this is the Dynkin diagram. Listing all the “consistently connected diagrams,” the answer comes down to four infinite series and five exceptional cases:
| Type | Dynkin diagram (abbreviated) | Compact group | Dimension | Rank |
|---|---|---|---|---|
| An | ○–○–⋯–○ | SU(n+1) | n(n + 2) | n (n ≥ 1) |
| Bn | ○–○–⋯–○⇒○ | SO(2n+1) | n(2n + 1) | n (n ≥ 2) |
| Cn | ○–○–⋯–○⇐○ | Sp(2n) (compact form USp(2n); see §7.6) | n(2n + 1) | n (n ≥ 3) |
| Dn | ○–○–⋯–○< (forked at the end) | SO(2n) | n(2n − 1) | n (n ≥ 4) |
| G2 | ○≣○ (triple line) | G2 | 14 | 2 |
| F4 | ○–○⇒○–○ | F4 | 52 | 4 |
| E6 | ○–○–○–○–○ (branch ○ at node 3) | E6 | 78 | 6 |
| E7 | ○–○–○–○–○–○ (branch ○ at node 3) | E7 | 133 | 7 |
| E8 | ○–○–○–○–○–○–○ (branch ○ at node 3) | E8 | 248 | 8 |
This is the periodic table of groups. Just as the periodic table of the elements says “these are all the atoms,” this table says “these are all the building blocks of continuous symmetry.” Whatever symmetry nature may use, it can only be chosen from this table — so the search for candidates for grand unified theories (§7) becomes, instead of a search for elements, a task of “finding the row in the table that satisfies the conditions.” The exceptional types stop at five, and E8 (dimension 248, rank 8) is the “superheavy element” at the very end of the table.
SO(odd) and SO(even) appear in separate rows for a reason that is deeper than it looks. In terms of root systems, so(2n+1) has, in addition to the long roots ±ei±ej the short roots ±ei (hence the Dynkin diagram has a double-line arrow), whereas so(2n) has just ±ei±ej, so all the roots have the same length (“simply laced”). They also appear differently in representation theory: the spinor representation is of one kind for so(2n+1) (dimension 2n), while for so(2n) it splits into two kinds, left and right (chiral and anti-chiral, each of 2n−1 dimensions). The “fork” at the end of the Dynkin diagram corresponds precisely to these two spinors. The 16 and 16 of so(10) = D5 — the leading roles of §7 — come from here.
| Rank | Isomorphism | Role in physics |
|---|---|---|
| 1 | su(2) ≅ so(3) ≅ sp(2) (A1 = B1 = C1) | Spin and rotations (§3) |
| 2 | so(4) ≅ su(2) ⊕ su(2) (D2 = A1⊕A1, not even simple) | Hydrogen atom (§7), Lorentz group (§7) |
| 2 | so(5) ≅ sp(4) (B2 = C2) | Both have dimension 10 |
| 3 | so(6) ≅ su(4) (D3 = A3, dimension 15) | Pati–Salam unification SU(4) |
The starting ranks “n ≥ 2, 3, 4” in the table are precisely the convention that avoids counting these overlaps twice. At low rank the periodic table is degenerate: algebras that should belong to different series turn out to be the same — only from rank 4 on do A, B, C, and D become completely different things. And this “degeneracy” is not a mathematical curiosity: the low-rank isomorphisms are used directly as mechanisms in physics. The so(4) of the hydrogen atom and the su(2)⊕su(2) of the Lorentz group both appear in the next section.
The tools are now in hand: “how to read names” (§1), “algebras and representations” (§2–4), “the geometry of labels” (§5), and “the complete list of what is in stock” (§6). From here on we take inventory of the groups that actually appear in physics, one by one.
Quarks have an internal degree of freedom called “color,” with three values, and belong to the fundamental representation 3. If we require that this SU(3) transformation can be performed independently at each point of spacetime (gauging it), a connection field — the gluons — becomes necessary, and their number is the dimension of the adjoint representation, 8 (= the dimension of the group, 32−1). “Why are there 8 kinds of gluons rather than 9?” The answer is that the 1 in 3⊗3 = 8⊕1 (the trace part = the component that mixes all colors equally) lies outside the gauge group. Color SU(3) is, as far as is known, a symmetry that holds exactly.
The same algebra, su(3), also appears in an entirely different guise. Because the mass differences among the three quarks u, d, and s are smaller than the QCD scale, the flavor SU(3), which interchanges the three, holds approximately. Hadrons are organized into irreducible representations of su(3) (octets and decuplets), and, as we saw in §5, their classification diagrams are rank-2 weight diagrams. In 1961, Gell-Mann and Ne’eman independently proposed this classification (the Eightfold Way), and in 1962 the existence and mass of the apex of the decuplet, then still vacant (the Ω−, strangeness −3), were predicted (about 1680 MeV from the equal-spacing rule). In 1964 the Ω− was discovered, with the predicted mass (1672 MeV). This is a landmark achievement: group theory predicted a new particle.
The weak interaction groups left-handed fermions into doublets of the weak isospin SU(2). The U(1) of electromagnetism is intertwined with this SU(2), and the electric charge Q is the sum of the third component of weak isospin, T3, and the weak hypercharge Y:
| Field (counted as left-handed) | SU(2) | T3 | Y | Q |
|---|---|---|---|---|
| Quark doublet Q = (uL, dL) | 2 | ±1/2 | 1/6 | 2/3, −1/3 |
| Lepton doublet L = (νL, eL) | 2 | ±1/2 | −1/2 | 0, −1 |
| uc (antiparticle of u) | 1 | 0 | −2/3 | −2/3 |
| dc (antiparticle of d) | 1 | 0 | 1/3 | 1/3 |
| ec (positron) | 1 | 0 | 1 | 1 |
Doesn’t this bother you? 1/6, −2/3, 1/3, −1/2, 1 — these oddly specific fractions are, within the Standard Model, nothing but “inputs fitted to experiment,” with no reason behind them. Why is the charge so neatly quantized? A candidate answer is the SU(5) of §7.7.
In the Coulomb problem H = p2/2m − k/r there is, besides the angular momentum L, the Runge–Lenz vector
which is also conserved (classically this corresponds to the fact that “the orientation of the elliptical orbit does not precess”). In the subspace of bound states (E < 0), normalizing to A′ = √(−m/2E) A gives a closed algebra generated by the six operators L and A′ — precisely the algebra of four-dimensional rotations, so(4) (dimension 6). And now the accidental isomorphism of §6 comes into play:
It splits into two independent copies of su(2). Furthermore, the constraint L·A′ = 0 makes the two Casimirs equal, j1 = j2 ≡ j. Setting n = 2j + 1, the degeneracy of the energy level is
— the familiar n2-fold degeneracy of the hydrogen atom (En = −mk2/2ℏ2n2 also follows through the Casimirs). The apparent symmetry SO(3) alone can explain only the (2l+1)-fold degeneracy within a given l (rotating m). The true nature of the “accidental degeneracy,” in which all of l = 0, 1, …, n−1 have the same energy, was a hidden SO(4) symmetry (check: ∑l=0n−1(2l+1) = n2).
The three-dimensional isotropic harmonic oscillator H = ℏω (a†·a + 3/2) is invariant under complex rotations of the creation and annihilation operators ai → Uijaj — the hidden symmetry is U(3). The degeneracy of the level N = nx + ny + nz is the number of ways to choose three non-negative integers whose sum is N:
This is exactly the dimension of the symmetric-tensor representation of su(3) (the representation whose weight diagram is a triangle of side N). The starting point for counting the magic numbers in the nuclear shell model is precisely this degeneracy. The hydrogen atom (so(4)) and the harmonic oscillator (u(3)) — the two great potentials that can be solved exactly in quantum mechanics — both turned out to have a hidden symmetry larger than the apparent SO(3).
Let us return to the Sp of §1. A canonical transformation in Hamiltonian mechanics, when linearized, is a matrix satisfying Eq. (5), that is, an element of Sp(2n,R). A symplectic matrix automatically has det = 1, so the volume of phase space is preserved — Liouville’s theorem is “one line in group-theoretic terms.” However, this group is non-compact: the squeeze transformation q → λq, p → p/λ can be continued without limit, so the parameter space does not close up. Because of a general theorem that a non-trivial finite-dimensional representation of a non-compact simple Lie group cannot be made unitary (this is why the squeeze operators of quantum optics are represented not by finite matrices but on the whole Fock space), the machinery of §4–6, which is built around finite-dimensional unitary representations, cannot be applied as it stands. The condition “simple” cannot be dropped: the additive group of real numbers, R, is also non-compact, but x ↦ eiax is a non-trivial one-dimensional unitary representation. Since Sp(2n,R) is simple, the theorem applies to this group as it stands. The Sp(2n) that appears in Georgi’s textbook is the compact form USp(2n) = U(2n) ∩ Sp(2n,C), which can be used with confidence (as the C series of §6). Even in the same row, Cn, the real forms differ — a small pitfall when reading names.
In 1974, Georgi and Glashow, seeking the smallest simple group that snugly contains the gauge group of the Standard Model, SU(3)×SU(2)×U(1), proposed SU(5). The idea is simple: line up the 3 color components and the 2 weak-isospin components vertically to make 5 components. The fundamental representation then decomposes as 5 = (3, 1) ⊕ (1, 2), and the 15 left-handed Weyl fields of one generation fit into exactly two irreducible representations:
(The 10 is the antisymmetric rank-2 tensor of the 5, of dimension 5·4/2 = 10: the quark doublet gives 3 colors × 2 = 6, uc gives 3, and ec gives 1, for a total of 10.) And here the magic happens. The electric charge Q is now one of the generators of SU(5). As we saw in §2, the generators of su(n) are traceless — so in any representation the sum of the charges must be zero. Applying this to the 5:
The charge of the d quark is, sign included, 1/3 that of the electron. The 3 in the denominator is the number of colors. The charges of quarks are multiples of 1/3 because there are three colors, and the hydrogen atom (p = uud plus e−) is exactly neutral because one and the same traceless generator determines the charges of quarks and leptons simultaneously — all the values of the weak hypercharge that we called “oddly specific” in §7.3 can be read off from Y = diag(−1/3, −1/3, −1/3, 1/2, 1/2) — a single matrix (whose trace is indeed 0). This matrix acts on the fundamental representation 5, and the values for 10 = Λ25 are read off as sums of two entries: −1/3 − 1/3 = −2/3 (uc), −1/3 + 1/2 = 1/6 (Q), and 1/2 + 1/2 = 1 (ec). The 5 of Eq. (29) is the conjugate representation, so the overall sign is reversed: Y = diag(1/3, 1/3, 1/3, −1/2, −1/2), and these are, in the table of §7.3, dc: 1/3 and the lepton doublet: −1/2. The price is that new gauge particles linking quarks and leptons (the X and Y bosons, 24 − 12 = 12 in all) appear, predicting proton decay. And it has not been found — the minimal SU(5) model has been experimentally excluded because this prediction does not show up in observations (Super-Kamiokande: τ(p → e+π0) > 2.4×1034 years). Even so, there is no other framework that explains the quantization of charge so beautifully.
There is one more step up. In SU(5), the 5 ⊕ 10 remained split into two representations. Going up to SO(10) (= D5 of the periodic table, dimension 45, rank 5), we can use the chiral spinor representation 16 foreshadowed in §6 (dimension 25−1 = 16):
The 15 particles of one generation plus one extra, 16 in all, fit into a single irreducible representation. The extra SU(5) singlet is none other than the right-handed neutrino νc — since the discovery of neutrino oscillations (mass), it has been rather welcome as a component of the seesaw mechanism.
The construction of the spinor 16 is surprisingly simple. Since so(10) has rank 5, the weights have 5 components. The weights of the spinor representation are
Of the 25 = 32 ways of choosing the signs, those with an even number of minus signs number 16: 5C0 + 5C2 + 5C4 = 1 + 10 + 5 = 16. Moreover, look at the breakdown: the 1 state with zero minus signs is the SU(5) 1 (the right-handed neutrino), the 10 states with two minus signs are the 10, and the 5 states with four minus signs are the 5 — the decomposition in Eq. (31) is visible directly as the combinatorics of signs. Electrons, quarks, and neutrinos alike are merely different “even/odd patterns of five ±1/2 labels” — this is the landscape that SO(10) grand unification presents. In tab ⑥ “Grand unification: SU(5) and SO(10)” of the simulator, you can see how the sign patterns of these 16 states are regrouped in three ways of bundling: the Standard Model, SU(5), and SO(10).
Finally, the symmetry of spacetime itself. The Lorentz group SO(3,1) is the group that “preserves the inner product of signature (+,−,−,−),” and its generators are the rotations Ji and the boosts Ki. The commutation relation [Ki, Kj] = −iεijkJk differs from that of so(4) only in the minus sign. The complex combinations
separate the algebra into two copies of su(2), and the representations can be classified by the pair of labels (j1, j2) with (0,0) the scalar, (1/2, 0) and (0, 1/2) the left-handed and right-handed Weyl spinors, (1/2, 1/2) the four-vector, and (1,0)⊕(0,1) the electromagnetic field tensor Fμν. And the story of SU(2)→SO(3) (§3) carries over completely: the double cover of the Lorentz group is SL(2,C) (complex 2×2 matrices of determinant 1), and when the four-vector is repacked into the 2×2 Hermitian matrix X = xμσμ, the quantity det X = t2 − |x|2 becomes the Minkowski inner product, and the same sandwiching as in Eq. (16) gives the 2-to-1 correspondence. Since a boost is a non-compact direction, in a finite-dimensional representation the generators of the boosts cannot be made Hermitian (the same situation as in §7.6; so(3,1) is simple, so that theorem applies as it stands) — and this is why, in quantum field theory, unitarity is realized on infinite-dimensional representations (the state space of particles).
| Symbol | Preserves | Dimension | Rank | Role in physics |
|---|---|---|---|---|
| U(1) | The absolute value of a complex number (the phase changes) | 1 | 1 | Electromagnetism, hypercharge |
| SU(2) | The inner product of complex 2-component vectors | 3 | 1 | Spin, isospin, the weak interaction |
| SO(3) | Length in three dimensions | 3 | 1 | Spatial rotations (= SU(2)/{±1}) |
| SU(3) | The inner product of complex 3-component vectors | 8 | 2 | Color (exact), flavor (approximate) |
| SO(4) | Length in four dimensions | 6 | 2 | Hidden symmetry of the hydrogen atom (n2-fold degeneracy) |
| U(3) | The inner product of complex 3-component vectors, and the phase | 9 | 3 | Degeneracy of the isotropic harmonic oscillator |
| Sp(2n) | The antisymmetric form of phase space | n(2n + 1) | n | Canonical transformations, Liouville’s theorem |
| SU(5) | The inner product of complex 5-component vectors | 24 | 4 | Grand unification (5⊕10 = one generation, charge quantization) |
| SO(10) | Length in ten dimensions | 45 | 5 | Grand unification (spinor 16 = one generation + νR) |
| SO(3,1) | The Minkowski inner product | 6 | 2 | Lorentz transformations (covered by SL(2,C)) |
| E8 | (exceptional type) | 248 | 8 | E8×E8 of heterotic string theory |
Let us look back over the road we have traveled.
Mochizuki Research Group, Tokyo University of Science / H. Georgi, Lie Algebras in Particle Physics (2nd ed., Westview Press) (Japanese translation: Butsurigaku ni okeru Rī Daisū, “Lie Algebras in Physics”). This page reorganizes the key points of the book’s first part (representation theory of su(2)), middle part (roots, weights, simple roots, and classification), and final part (SU(5) and SO(10) grand unification) into a single story, “how to read the names of groups.” The treatment of the hydrogen atom, the harmonic oscillator, and the symplectic group in §7.4–7.6 supplements what is in the textbook. If you would like to try out the generators, commutation relations, and weight diagrams by moving them yourself, go to the simulator (liegroup-simulator-en.html). If you would like to extend from the discrete symmetries of crystals (point groups and space groups) to continuous groups, please also see the documentation of CrystOD, which is developed in our laboratory. The equations are typeset in the browser-standard MathML (no external libraries or network communication are used).