Consider the representation of $\textrm{SO}(4)$ on $\textrm{Sym}^2(\wedge^2\mathbb{R}^4)$ induced by the standard representation of $\textrm{SO}(4)$ on $\mathbb{R}^4$. I am interested in the ring of invariants of this representation, i.e. the ring of all polynomial functions on $\textrm{Sym}^2(\wedge^2\mathbb{R}^4)$ that are invariant.

Here is a way of constructing some of these invariants. The space of $\textrm{SO}(4)$-linear maps $\wedge^2\mathbb{R}^4\to\wedge^2\mathbb{R}^4$ is $2$-dimensional: We have the identity $I$ and the Hodge star operator $\star$. Thus the coefficients of the polynomial $$\det(x\cdot I+y\cdot \star+A)\in\mathbb{R}[x,y]$$are polynomials in the entries of $A\in\textrm{Sym}^2(\wedge^2\mathbb{R}^4)$ which are $\textrm{SO}(4)$-invariant. My question is: Do these polynomials already generate the ring of invariants?


The answer is 'no', though I don't know an easy way to see this without doing an explicit calculation. Here is where to look though, if you want to do the calculation yourself:

Things work out a bit better if one uses indeterminates $z = x+y$ and $w = x-y$. Then one has an expansion $$ \det\bigl(x{\cdot}I + y{\cdot}\star + A\bigr) = \sum_{0\le i,j\le 3} P_{ij}(A)\,z^{3-i}w^{3-j} $$ where $P_{ij}(A)$ is a polynomial in $A$ of degree $i{+}j$.

One finds that $P_{00}=1$. Meanwhile, the polynomials $P_{10}$, $P_{01}$, $P_{20}$, $P_{11}$, $P_{02}$, $P_{30}$, $P_{21}$, $P_{12}$, and $P_{03}$ are independent and generate all of the polynomial invariants of degree $3$ or less. Moreover, the three quartic polynomials $P_{3,1}$, $P_{2,2}$, and $P_{1,3}$ are algebraically independent over the ring generated by the lower degree $P_{ij}$.

However, the $P_{ij}$ with $i+j\le 4$ do not generate all of the invariant polynomials of degree $4$. There is one further quartic polynomial invariant, say $Q$, that is not a polynomial in the $P_{ij}$ with $i+j\le 4$. Since the remaining three coefficients $P_{3,2}$, $P_{2,3}$ and $P_{3,3}$ are all of degree 5 or higher, $Q$ does not lie in the ring generated by the $P_{ij}$.

Added detail: $\mathrm{SO}(4)$ preserves the splitting $\Lambda^2(\mathbb{R}^4)= \Lambda^2_+(\mathbb{R}^4)\oplus\Lambda^2_-(\mathbb{R}^4)$ into self-dual and anti-selfdual forms, i.e., the split into the eigenspaces of $\star$, which has trace-zero and square $\star^2 = I$. It follows that we can write block decompositions $$ \star = \begin{pmatrix} I_3&0_3\\0_3&-I_3\end{pmatrix} \quad\text{and}\quad A = \begin{pmatrix} a&b\\b^T&c\end{pmatrix}, $$ where $a=a^T$, $b$ and $c=c^T$ are $3$-by-$3$ matrices. The action of $\mathrm{SO}(4)$ on $A$ can be described as $$ g\cdot A = \begin{pmatrix} g_+ag_+^T&g_+bg_-^T\\g_-b^Tg_+^T&g_-cg_-^T\end{pmatrix} $$ where $g_\pm:\mathrm{SO}(4)\to\mathrm{SO}(3)$ are homomorphisms such that $(g_+,g_-):\mathrm{SO}(4)\to\mathrm{SO}(3)\times\mathrm{SO}(3)$ is surjective, with kernel $\{\pm I_4\}$. We can write $$ \det(x{\cdot}I_6 + y{\cdot}\star + A) = \det\begin{pmatrix} zI_3+a & b\\ b^T& wI_3 + c\end{pmatrix} = \sum_{0\le i,j\le 3} P_{ij}(A)\,z^{3-i}w^{3-j}, $$ and we can refine our notation by writing $P_{ij}(A) = P_{ij}(a,b,c)$.

Now, because of the way $a$, $b$, and $c$ transform under the action of $\mathrm{SO}(4)$, we see, for example, that $\mathrm{tr}(a^k)$ and $\mathrm{tr}(c^l)$ are invariant polynomials, as is $\mathrm{tr}(bb^T)$ (though $\mathrm{tr}(b)$ is not), and, indeed, we see, by inspection, that, for example, $$ P_{10} = \mathrm{tr}(a)\quad\text{and}\quad P_{01} = \mathrm{tr}(c), $$ and a little reflection shows that $$ P_{20} = \tfrac12\bigl(\mathrm{tr}(a)^2-\mathrm{tr}(a^2)\bigr) \quad\text{and}\quad P_{02} = \tfrac12\bigl(\mathrm{tr}(c)^2-\mathrm{tr}(c^2)\bigr) $$ while it's not hard to see that $$ P_{11} = \mathrm{tr}(a)\mathrm{tr}(c) - \mathrm{tr}(bb^T). $$ To go further, imagine that we have an alphabet of 4 letters, $a$, $b$, $b^T$, and $c$, and we declare a valid `word' to be string of these letters subject to the following rules: $a$ can be followed only by $a$ or $b$, $b$ can only be followed by $c$ or $b^T$, $c$ can only be followed by $c$ or $b^T$, $b^T$ can only be followed by $a$ or $b$, and finally, that the number of $b$s in the word is the same as the number of $b^T$s. Thus, for example, $a^4$, $abb^T$, and $abcb^T$ are valid words, but $abc$ is not valid. What one sees that is a vaild word $w$, when interpreted as a product of matrices, transforms under $\mathrm{SO}(4)$ into $g_\pm w g_\pm^T$, and hence $\mathrm{tr}(w)$ is an invariant polynomial for any valid word.

Now we can go on: For example $P_{30} = \det(a)$ which is a weighted homogeneous polynomial in $\mathrm{tr}(a)$, $\mathrm{tr}(a^2)$, and $\mathrm{tr}(a^3)$, and similarly with $P_{03}= \det(c)$. Moreover, one finds that $$ P_{21} = P_{20}P_{01}-P_{10}(P_{10}P_{01}-P_{11}) +\mathrm{tr}(abb^T) $$ with a similar formula for $P_{12}$. Thus, all of the polynomials $P_{ij}$ for $0<i+j\le 3$ are expressed as polynomials in the 9 (independent) invariants $$ \mathrm{tr}(a),\mathrm{tr}(c),\ \mathrm{tr}(a^2),\mathrm{tr}(c^2),\mathrm{tr}(bb^T),\ \mathrm{tr}(a^3),\mathrm{tr}(abb^T),\mathrm{tr}(cb^Tb), \mathrm{tr}(c^3), $$ and conversely. These are all of the invariants of degree $3$ or less. (Of course, $\mathrm{tr}(a^4)$ is a polynomial in these. Not all valid words yield an invariant that cannot be expressed as a polynomial in invariants of lower degree.)

But now, one sees where the discrepancy comes in. The quartic polynomials $P_{31}$, $P_{22}$, $P_{13}$ can be expressed in terms of the 'word-trace' invariants, but there are ony three of them, whereas there are four new quartic word-trace invariants: $$ \mathrm{tr}(aabb^T),\ \mathrm{tr}(bb^Tbb^T),\ \mathrm{tr}(ccb^Tb),\ \mathrm{tr}(abcb^T) $$ and it is not hard to see that no nontrivial linear combination of them can be written as a polynomial in the word-trace invariants of lower degree.

In fact, the situation gets worse: There are only two $P_{ij}$ of degree 5, but there are four new word-trace invariants of degree 5, and $P_{33}$ is the only one of degree 6 while there are two new word-trace invariants of degree 6.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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