I always wondered if the fact that the quartic can be solved by a cubic can be generalized to other even degrees $n$, namely if there is an ordering of the roots $x_i$ of form $x_1x_2+x_3x_4+\dots+x_{n-1}x_{n} = y$ such that $y$ is reduced to an algebraic number of deg $n-1$. It seems it can be if its Galois group is of a certain kind.

Call the resolvent polynomial formed by $y$ as $P_{n}(y)$. (The labeling of the transitive groups below is from Kluener's database of number fields.)

The 3-deg $P_4(y)$ has a 3th deg factor for the general quartic.

The 15-deg $P_6(y)$ has a 5th deg factor if $6T12, 6T14$.

The 105-deg $P_8(y)$ has a 7th deg factor if $8T25, 8T36, 8T37, 8T48$ (the first two are solvable groups).

The 945-deg $P_{10}(y)$ has an 9th deg factor if (none?),

The 10395-deg $P_{12}(y)$ has an 11th deg factor if $12T179, \text{(and?)}$

For example, given,

x^12-2x^11+17x^10-28x^9+138x^8-157x^7+549x^6-382x^5+1099x^4-77x^3+1016x^2+9x+1912 = 0

(I've left it unformatted for easier copy-paste) which has group $L(2,11)$ and discriminant $D_{12} = 1831^6$. I use Mathematica, and its root ordering is,

$$x_1, x_2 = -0.822\mp0.744 i$$

$$x_3, x_4 = -0.801\mp2.308 i$$

$$x_5, x_6 = -0.452\mp2.271 i$$

$$x_7, x_8 = 0.800\mp1.760 i$$

$$x_9, x_{10} = 0.949\mp1.080 i$$

$$x_{11}, x_{12} = 1.324\mp2.123 i$$


$$y = x_1x_7+x_2x_8+x_3x_6+x_4x_5+x_9x_{10}+x_{11}x_{12}=15.60720071\dots$$

then $y$ is a root of the monic 11-deg,

$$14155013839 + 4895349263 y + 3076830095 y^2 + 520581160 y^3 - 43394407 y^4 - 3261079 y^5 + 376457 y^6 + 33595 y^7 - 42 y^8 - 162 y^9 - 17 y^{10} + y^{11} = 0$$

which has discriminant $D_{11} = 23^6(1831^4)$.


  1. What other transitive groups for 12-deg eqns will have a $P_{12}(y)$ that has an 11-deg factor?

  2. Does it follow that there are 14-deg eqns such that $P_{14}(y)$ has an 13-deg factor? (I suspect $14T30, 14T39$ which have groups $PSL(2,13), PGL(2,13)$, respectively.)

P.S. My thanks to the Spearman, Watanabe, Williams paper PSL(2,5) Sextic Fields with a Power Basis which was the clue.


There are no other examples of degree $12$ besides the one you found. Moreover, there are no examples of degrees $10$ or $14$, and in fact the only additional degrees up to $32$ in which there exist examples are $16$, $28$, and $32$, where the groups in the latter cases are precisely (in Klueners/GAP/MAGMA notation) 16T447, 16T777, 16T1079, 16T80, 16T1329, 16T1508, 16T1653, 16T1654, 16T1753, 16T1840, 16T1906, 28T165, 32T35272, 32T397065, 32T2795174. Of these, only the first three degree-$16$ groups and the first two degree-$32$ groups are solvable; those degree-$16$ groups are the unique solvable transitive degree-$16$ groups of orders $240$, $480$, and $960$, and the solvable degree-$32$ groups are the unique transitive degree-$32$ groups of orders $992$ and $4960$. The biggest nonsolvable degree-$16$ group on this list is $\text{AGL}_4(2)$, the degree-$28$ group is $\text{P}\Gamma\text{L}_2(8)$, and the nonsolvable degree-$32$ group is $\text{AGL}_5(2)$.

More generally, let $f(X)$ be a separable irreducible polynomial over a field $K$, and suppose that $n:=\text{deg}(f)$ is even. Suppose in addition that, if $\{C_1,C_2,\dots,C_{n/2}\}$ and $\{D_1,D_2,\dots,D_{n/2}\}$ are any two distinct partitions of the roots of $f(X)$ into $n/2$ two-element sets, then $$ \sum_{i=1}^{n/2} \prod_{x\in C_i} x \ne \sum_{i=1}^{n/2}\prod_{x\in D_i} x. $$ (Note that this condition will hold for any "randomly selected" polynomial having any prescribed Galois group.) Under these hypotheses, if $G$ denotes the Galois group of $f(X)$ over $K$, then the following are equivalent:

  1. there is an ordering $x_1,\dots,x_n$ of the roots of $f(X)$ such that $x_1x_2+x_3x_4+\dots+x_{n-1}x_n$ has degree $n-1$ over $K$.
  2. there is an index-$(n-1)$ subgroup $H$ of $G$, and a pair of distinct roots $(x_1,x_2)$ of $f(X)$, such that every element of $H$ maps $\{x_1,x_2\}$ to either itself or to a disjoint set of roots, but every subgroup of $G$ which properly contains $H$ must contain an element which maps $\{x_1,x_2\}$ to a set having exactly one element in common with $\{x_1,x_2\}$.

This can be used to test your condition for specific values $n$, and perhaps group-theoretic results can be used to prove theorems for infinitely many $n$. For instance, I will show later that $G=\text{AGL}_d(2)$ has the required property for $n=2^d$.

To see this equivalence, first assume that the first condition holds; then we can let $H$ be the set of elements in $G$ which fix $y:=x_1x_2+x_3x_4+\dots+x_{n-1}x_n$. This $H$ will be an index-$(n-1)$ subgroup of $G$, and every element of $H$ permutes the collection of 2-element sets $\{\{x_1,x_2\}, \{x_3,x_4\}, ..., \{x_{n-1},x_n\}\}$. Thus, each element of $H$ maps $\{x_1,x_2\}$ to some $\{x_i,x_{i+1}\}$ with $i$ odd. Moreover, since $f(X)$ is irreducible, $G$ acts transitively on $\{x_1,\dots,x_n\}$, so $[G:G_{x_1}]=n$; but since $[G:H]=n-1$ is coprime to $n$, we have $[G:G_{x_1}\cap H]=n(n-1)$ and thus $[H:G_{x_1}\cap H]=n$, so $H$ is transitive. Any subgroup $J$ of $G$ which properly contains $H$ must contain an element $j$ which does not fix $y$, so this element must not permute the collection $\{\{x_1,x_2\}, \{x_3,x_4\}, ..., \{x_{n-1},x_n\}\}$; by multiplying $j$ on both sides by suitable elements of $H$, we obtain an element of $J$ which maps $\{x_1,x_2\}$ to a set having one element in common with $\{x_1,x_2\}$.

Conversely, assume that the second condition holds. As above, $H$ is transitive. It follows that the images of $\{x_1,x_2\}$ under $H$ consist of $n/2$ pairwise disjoint two-element sets, which we can write as $\{x_1,x_2\}, \{x_3,x_4\}, ..., \{x_{n-1},x_n\}$. Then $y:=x_1x_2+x_3x_4+...+x_{n-1}x_n$ is fixed by $H$, but not by any larger subgroup of $G$, so $[K(y):K]=n-1$. This completes the proof of the equivalence.

Added later: here is a proof that $G:=\text{AGL}_d(2)$ has the required property for $n=2^d$. The group $G$ acts on the $\mathbf{F}_2$-vector space $V:=(\mathbf{F}_2)^d$, and $\text{GL}_d(2)$ consists of the elements fixing $0$. Pick some nonzero $c\in V$, and let $J$ be the stabilizer of $c$ in $G$. Let $H$ be the subgroup of $G$ generated by $J$ and the translations $x\mapsto x+u$. I claim that $H$ and $\{0,c\}$ have the required properties. Note that $\text{GL}_d(2)$ acts transitively on $V\setminus\{0\}$, so $[\text{GL}_d(2):J]=2^d-1$. Since the group of translations has order $2^d$, and is normalized by $\text{GL}_d(2)$ (and hence by $J$), we have $\#H=2^d\cdot\#J$ so $[G:H]=2^d-1$. Every translation maps $\{0,c\}$ to a set $\{b,b+c\}$, and any such set either equals $\{0,c\}$ or is disjoint from $\{0,c\}$. Finally, let $K$ be a subgroup of $G$ which properly contains $H$. Since $K$ contains all the translations, and every element of $G$ is a translation times an element of $\text{GL}_d(2)$, it follows that $K$ contains an element of $\text{GL}_d(2)$ which is not in $H$. Any such element maps $c$ to some $b\notin\{0,c\}$, and hence maps $\{0,c\}$ to the set $\{0,b\}$ which has exactly one element in common with $\{0,c\}$.

| cite | improve this answer | |
  • $\begingroup$ That is fascinating! I guess just like quartics can be solved as $x = \sqrt{u_1}+\sqrt{u_2}+\sqrt{u_3}$ where the $u_i$ are roots of a cubic, and some octics (such as this one) can be solved as $x = \sqrt{u_1}+\sqrt{u_2}+\dots+\sqrt{u_7}$ where the $u_i$ are the roots of a solvable 7th deg, then it seems there are 16 and 32-deg eqns that can be solved as $x = \sqrt{u_1}+\sqrt{u_2}+\dots+\sqrt{u_{15}}$ and $x = \sqrt{u_1}+\sqrt{u_2}+\dots+\sqrt{u_{31}}$, respectively. I would love to see such a 32-deg, but I guess that's out of reach for now. $\endgroup$ – Tito Piezas III Nov 19 '13 at 0:12

The common feature in all of your examples is that the nonabelian composition factors of the transitive permutation groups of degree $2n$ in question have permutation representations of degree $2n-1$. This is the case for 6T12 and 6T14 ($A_5$), 8T37 and 8T48 ($L(2,7)$), and 12T179 ($L(2,11)$).

There are no transitive groups of degree 10 with composition factors having a representation of degree 9, which might explain why you find no examples there (although $A_5$ and $A_6$ arise here, which have smaller degree representations).

Note that $L(2,13)$ has no permutation representation of degree 13, so I would not expect this to work in those cases.

| cite | improve this answer | |
  • $\begingroup$ Why did it work for 6T14 ($PGL(2,5)$), but not for 8T43 ($PGL(2,7)$)? (I tested it for the latter, and $P_8(y)$ only had a 14-deg factor.) $\endgroup$ – Tito Piezas III Nov 16 '13 at 21:50
  • $\begingroup$ I imagine that is because ${\rm PGL}(2,5) \cong S_5$ has a permutation representation of degree 5, but ${\rm PGL}(2,7)$ has none of degree 7 (but ${\rm PSL}(2,7)$ does). That suggests that it is not just the composition factor that matters, but the automorphisms that are present in the transitive group. Do you have a polynomial for 12T272 = $M_{11}$? According to my theory, that should work too. $\endgroup$ – Derek Holt Nov 17 '13 at 8:19
  • $\begingroup$ Although $M_{11}$ has a degree-$11$ permutation representation, still that group doesn't yield the required types of polynomials, since there are no degree-$11$ elements of the form $x_1x_2+x_3x_4+...+x_{11}x_{12}$. There are degree-$11$ elements given by other expressions in the $x_i$'s, but not by an expression of the required form. $\endgroup$ – Michael Zieve Nov 18 '13 at 4:49
  • $\begingroup$ @Michael Zieve: Yes I see that now! Thanks! $\endgroup$ – Derek Holt Nov 18 '13 at 10:59

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.