Perspectives on spherical harmonics

Math

Spherical harmonics appear in lots of different places and have different interpretations that at first sight don’t seem to have anything to do with one another. In this post, I’ll try to connect three very common ones (namely as harmonic polynomials, as eigenfunctions of the Laplacian and as irreps of SO(3)\operatorname{SO}(3)).

We’re going to define spherical harmonics as homogeneous harmonic polynomials R3C\mathbb{R}^3 \to \mathbb{C}. Let’s break this down:

  • A polynomial of three variables is a finite sum of the form
    αaαxαxyαyzαz\sum_{\alpha} a_\alpha x^{\alpha_x}y^{\alpha_y}z^{\alpha_z}
    over multi-indices αN03\alpha \in \mathbb{N}_0^3. Some examples are x2y+2zx^2y + 2z or xyz+y2xyz + y^2.
  • The coefficients aαa_\alpha can be complex numbers, but we will only plug in real numbers for xx, yy and zz. That’s why we interpret polynomials as functions R3C\mathbb{R}^3 \to \mathbb{C}.
  • Homogeneous mean that αx+αy+αz\alpha_x + \alpha_y + \alpha_z is the same for all the α\alpha we sum over, so all the terms in the sum have the same degree. For example, x2+2yz+xzx^2 + 2yz + xz is homogeneous, while xy+zxy + z is not.
  • Harmonic means that the Laplacian of the polynomial vanishes: pp is harmonic if Δp=0\Delta p = 0. Here, Δ=2x2+2y2+2z2\Delta = \frac{\partial^2}{\partial x^2} + \frac{\partial^2}{\partial y^2} + \frac{\partial^2}{\partial z^2}.

We will write Hl\mathcal{H}_l for the space of all homogeneous harmonic polymonials of degree ll (meaning αx+αy+αz=l\alpha_x + \alpha_y + \alpha_z = l for all summands).

If you’ve seen spherical harmonics before (and you presumably have, if you’re reading this post), it’s probably been in the form of functions \(Y_l^m(\theta, \varphi)\) defined on the sphere. So why are we talking about these polynomials on R3\mathbb{R}^3 instead?

The answer is that every polynomial pHlp \in \mathcal{H}_l can be written in spherical coordinates as

p(x,y,z)=rlY(θ,φ)p(x, y, z) = r^l Y(\theta, \varphi)

for some function Y:S2CY: S^2 \to \mathbb{C}. To see why, write xx, yy and zz in spherical coordinates and plug them into the polynomial. They each have a factor of rr and then some factors depending on θ\theta and φ\varphi. So because pp is homogeneous, each summand consists of a factor rlr^l times something that depends only on θ\theta and φ\varphi. So we can think of homogeneous polynomials as polynomials defined on the sphere — their continuation to R3\mathbb{R}^3 is automatically determined by their degree ll. Therefore, we won’t really distinguish between homogeneous harmonic polynomials defined on R3\mathbb{R}^3 and their restrictions to S2S^2, we will refer to both as spherical harmonics.

This should also explain the name: spherical harmonics are harmonic polynomials living on the sphere.

The functions YlmY_l^m that you may have seen are just a particular choice of basis for the vector space of spherical harmonics. If you multiply them by rlr^l, you get polynomials in Hl\mathcal{H}_l, and

{rlYlmlml}\{r^l Y_l^m| -l \leq m \leq l\}

is a basis for Hl\mathcal{H}_l.

Eigenfunctions of the Laplacian

One of the reasons that spherical harmonics are so ubiquitous is that they are the eigenfunctions of the spherical Laplacian ΔS2\Delta_{S^2}. They key to that is the following fact (which is just a brief calculation): for a function Y:S2CY: S^2 \to \mathbb{C},

Δ(rlY)=rl2(l(l+1)Y+ΔS2Y).\Delta (r^l Y) = r^{l - 2}\left(l(l + 1)Y + \Delta_{S^2}Y\right)\thinspace.

So rlY(θ,φ)r^l Y(\theta, \varphi) is harmonic if and only if

ΔS2Y=l(l+1)Y.\Delta_{S^2}Y = -l(l + 1)Y\thinspace.

This already proves that spherical harmonics are eigenfunctions of the spherical Laplacian.

But we can say more than that: if we take any eigenfunction f:S2Cf: S^2 \to \mathbb{C} of the spherical Laplacian and multiply by rlr^l (with ll such that l(l+1)-l(l + 1) gives the eigenvalue1), then rlf(θ,φ)r^l f(\theta, \varphi) must be harmonic. So the eigenfunctions of the spherical Laplacian are in fact in 1-to-1 correspondence with harmonic homogeneous functions on R3\mathbb{R}^3. It then turns out — and this part is far from obvious — that all such functions are polynomials2! So the spherical harmonics aren’t just eigenfunctions of the spherical Laplacian, they make up all of its eigenfunctions.

Irreducible representations of SO(3)\operatorname{SO}(3)

Another famous role that spherical harmonics play is as the irreducible representations of SO(3)\operatorname{SO}(3) (more precisely: the (complex) irreducible representations of SO(3)\operatorname{SO}(3) are exactly the spaces Hl\mathcal{H}_l). This is connected to the fact that they are the eigenfunctions of the spherical Laplacian.

That the eigenspaces of the spherical Laplacian are representations of SO(3)\operatorname{SO}(3) follows directly from the fact that the Laplacian commutes with rotations: we have a representation GL2(S2,C)G \curvearrowright L^2(S^2, \mathbb{C}) via

(rf)(x):=f(r1x)(r \cdot f)(x) := f(r^{-1}x)

for any rotation rr and fL2(S2,C)f \in L^2(S^2, \mathbb{C}). For an eigenfunction of the Laplacian, we get

ΔS2(rf)=rΔS2f=rλf=λ(rf),\Delta_{S^2}(r \cdot f) = r \cdot \Delta_{S^2} f = r \cdot \lambda f = \lambda (r \cdot f)\thinspace,

so each eigenspace is invariant under the action of SO(3)\operatorname{SO}(3). Therefore, the representation on L2(S2)L^2(S^2) can be restricted to each eigenspace, so each Hl\mathcal{H}_l gives a representation of SO(3)\operatorname{SO}(3).

Showing that these representations are in fact irreducible is much more difficult (there’s a proof here for example, if you really want to dive into that). But if we just take that for granted, it’s again easy to show that every irreducible subrepresentation of L2(S2)L^2(S^2) is a space of spherical harmonics: because the Laplacian is an equivariant map on each such representation, Schur’s Lemma implies that it must be either the zero map (which it isn’t) or multipication by a constant λC\lambda \in \mathbb{C}. Therefore, each irreducible representation is contained in an eigenspace of the Laplacian. But these eigenspaces are themselves irreducible, so the representation in question must already be equal to the eigenspace.

Finally, it’s possible to show that all irreducible representations of SO(3)\operatorname{SO}(3) are subrepresentations of L2(S2)L^2(S^2). This is again much more difficult and is also a very special fact about SO(3)\operatorname{SO}(3) (for example, the Laplacian’s eigenspaces are still irreducible representations in higher dimensions, but they are not the only ones anymore). But combining this with our results from above, the spherical harmonics make up all the irreducible representations of SO(3)\operatorname{SO}(3).


  1. I’m skipping over some details here, see for example Claim 4.0.1 here
  2. See Corollary 4.0.6 in the same document for a proof