Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

The complex solid spherical harmonics can be defined as

$$ U_n^m(\boldsymbol{r}) = r^n P_n^m(\cos{\theta}) e^{im\phi}, $$

where $r,\theta,\phi$ are the usual spherical coordinates of $\boldsymbol{r}=(x,y,z)$. Note that $U_n^m(\boldsymbol{r})$ is a homogeneous polynomial in $x$, $y$, and $z$ of degree $n$, i.e. $U_n^m(\alpha\boldsymbol{r})=\alpha^n U_n^m(\boldsymbol{r})$. I want to work out the integrals

$$ I_n^m := \int_{-1}^1 dx \int_{-1}^1 dy \int_{-1}^1 dz \;U_n^m(\boldsymbol{r}) $$

but cannot get it (nor did I manage to have maple get it for me). Note that since the $U_n^m(\boldsymbol{r})$ are homogenous, we can reduce this to a surface integral:

$$ I_n^m = \frac{1}{n+3} \int \partial \Omega \;U_n^m(\boldsymbol{r}) $$

where the intregral is over the surface of the unit cube. Obviously, $I_n^m=0$ for any odd $n$ or $m$, but I need the full story. (I want to get the multipole moments for a uniform density within a cube).

share|improve this question
    
Out of curiosity, are you looking for (i) closed formulas, (ii) sums, (iii) recursion formulas, or something else? If sums are enough, is it not feasible to express the $U_n^m(r)$ in Cartesian form and integrate over the cube directly? –  user86418 Dec 22 '13 at 13:47
    
@user86418 (i) or (iii) would be okay. And yes, using the Cartesian form of the harmonics is perfectly okay. –  Walter Dec 25 '13 at 13:41
    
There are numerous notational conventions, and working out gritty details runs the risk of producing less-than-useful formulas. A sketch follows, using the physicists' convention for spherical coordinates ($x=r\cos\phi\sin\theta$, $y=r\sin\phi\sin\theta$, $z=r\cos\theta$). –  user86418 Dec 25 '13 at 15:24
    
Using the first two of these recursion relations gives \begin{align*} (n-m)U_n^m&=(2n-1)zU_{n-1}^m-(n+m-1)r^2U_{n-2}^m,\\ 2mzU_n^m&=-(x-iy)U_n^{m+1}-(x+iy)(n+m)(n-m+1)U_n^{m-1}. \end{align*} Writing $U_n^m=\sum_{i,j,k}a_{ijk}x^iy^jz^k$, recursion formulas for the $a_{ijk}$ (and thus for the desired integrals) are straightforward. –  user86418 Dec 25 '13 at 15:24
    
Yes, I know all of them. What I want is recursion (or otherwise) for the $I_n^m$ ... And of course, $U^m_n=\sum_{i,j,k}a^{nm}_{ijk}x^iy^jz^k$, i.e. the coefficients carry 5 indices, not 3. –  Walter Dec 26 '13 at 17:00

2 Answers 2

Here are details of the strategy outlined in the comments for recursively calculating the coefficients of $U_n^m(\mathbf{r})$. This doesn't fully answer the question, but perhaps it will be useful nonetheless.

To fix notation, let $(r, \phi, \theta)$ denote spherical coordinates with the physicists' convention: $$ x = r\cos\phi \sin\theta,\quad y = r\sin\phi \sin\theta,\quad z = r\cos\theta. $$ For non-negative integers $n$ and $m$, write $U_n^m(\mathbf{r}) = r^n P_n^m(\cos\theta) e^{im\phi}$ as $$ \sum_{i, j, k = 0}^\infty (a_n^m)_{i,j,k} x^i y^j z^k, $$ with the understanding that the coefficient $(a_n^m)_{i,j,k}$ is zero if any index is negative, or is larger than $n$. (Particularly, each sum is finite.)

The first recursion relation for the Legendre polynomials becomes, upon replacing "$x$" by $\cos\theta$, setting $\ell = n + 1$, $$ (n - m + 2) P_{n+2}^m(\cos\theta) = (2n + 3) \cos\theta P_{n+1}^m(\cos\theta) - (n + m + 1) P_n^m(\cos\theta). $$ Multiplying through by $r^{n+2} e^{\sqrt{-1}m\phi}$, $$ (n - m + 2) U_{n+2}^m(\mathbf{r}) = (2n + 3) z U_{n+1}^m(\mathbf{r}) - (n + m + 1) r^2 U_n^m(\mathbf{r}). $$ Writing each polynomial $U_n^m$ in terms of its coefficients, shifting indices, and equating coefficients of $x^i y^j z^k$ gives the coefficient relation \begin{align*} (a_{n+2}^m)_{i,j,k} &= \frac{2n + 3}{n - m + 2} (a_{n+1}^m)_{i,j,k-1} \\ &\quad- \frac{n + m + 1}{n - m + 2} \bigl((a_n^m)_{i-2,j,k} + (a_n^m)_{i,j-2,k} + (a_n^m)_{i,j,k-2}\bigr). \end{align*} Since $U_0^0(\mathbf{r}) = 1$, we have $(a_0^0)_{0,0,0} = 1$ and $(a_0^0)_{i,j,k} = 0$ for all other $i$, $j$, $k$.

Since $U_1^0(\mathbf{r}) = z$, we have $(a_1^0)_{0,0,1} = 1$ and $(a_1^0)_{i,j,k} = 0$ for all other $i$, $j$, $k$.

The preceding recurrence determines the coefficients $(a_n^0)_{i,j,k}$.

Similarly, $U_0^1(\mathbf{r}) = 0$, so $(a_0^1)_{i,j,k} = 0$ for all $i$, $j$, and $k$; and $U_1^1(\mathbf{r}) = -(x + \sqrt{-1} y)$, so $(a_1^1)_{1,0,0} = -1$, $(a_1^1)_{0,1,0} = -\sqrt{-1}$, while $(a_1^1)_{i,j,k} = 0$ for all other $i$, $j$, $k$.

The preceding recurrence determines the coefficients $(a_n^1)_{i,j,k}$.

The second recursion relation for the Legendre polynomials becomes, upon replacing "$x$" by $\cos\theta$ and $m$ by $m + 1$, and rearranging, $$ \sin\theta P_n^{m+2}(\cos\theta) = -2(m + 1) \cos\theta P_n^{m+1}(\cos\theta) - (n + m + 1)(n - m) \sin\theta P_n^m(\cos\theta). $$ Multiplying through by $r^{n+1} e^{\sqrt{-1}(m+1)\phi}$ and using $r\sin\theta e^{\pm \sqrt{-1}\phi} = x \pm \sqrt{-1} y$ gives $$ (x - \sqrt{-1} y) U_n^{m+2}(\mathbf{r}) = -2(m + 1) z U_n^{m+1}(\mathbf{r}) - (x + \sqrt{-1} y)(n + m + 1)(n - m) U_n^m(\mathbf{r}). $$ Writing each polynomial $U_n^m$ in terms of its coefficients, shifting indices, and equating coefficients of $x^i y^j z^k$ gives the coefficient relation \begin{align*} &(a_n^{m+2})_{i-1,j,k} - \sqrt{-1} (a_n^{m+2})_{i,j-1,k} \\ &\qquad= -2(m + 1) (a_n^{m+1})_{i,j,k-1} \\ &\qquad\quad- (n + m + 1)(n - m) \bigl((a_n^m)_{i-1,j,k} + \sqrt{-1}(a_n^m)_{i,j-1,k}\bigr). \end{align*}

Since the $(a_n^0)_{i,j,k}$ and $(a_n^1)_{i,j,k}$ are known for all $n \geq 0$, the coefficients $(a_n^m)_{i,j,k}$ may be found recursively, and the integrals $I_n^m$ computed as noted in the comments.

share|improve this answer
    
This is not what I was looking for. I had worked out a similarly tedious way myself (before asking), but was looking for a more compact formula. Thanks for your efforts anyway. –  Walter Dec 28 '13 at 19:06

I've solved the problem numerically using a C++ code and found that not only to odd n's vanish, all m's not divisible by 4 vanish - and to my surprise all quadrupole moments vanish. So the first non-zero term past n=0 is n=4 with m=-4,0,+4, then n=6 with m=-4,0,+4, then n=8 with m=-8,-4,0,+4,+8, etc.

Just curious - do you need this to do multipole calculations in a grid code?

EDIT - also, all imaginary parts vanish for all.

share|improve this answer
    
I wanted to know the multipole moments of constant density cube. So knowing which vanish is only part of the story. I also need the value for those which don't. Preferrably not the numerical value, but a closed formula. –  Walter Jun 29 at 15:28

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.