6

The ratio of the unsigned coefficients for the discriminants of $x^n+bx+c$ for $n=2$ to $5$ follow a simple pattern:

$$\left (\frac{2^2}{1^1},\frac{3^3}{2^2},\frac{4^4}{3^3},\frac{5^5}{4^4} \right )=\left ( \frac{4}{1},\frac{27}{4},\frac{256}{27},\frac{3125}{256} \right )$$

corresponding to the discriminants

$$(b^2-4c, -4b^3-27c^2,-27b^4+256c^3,256b^5+3125c^4).$$

Does the pattern for the ratios extend to higher orders? (An online reference would be appreciated.)

Tom Copeland
  • 2,587
  • The general formula for the discriminant curve is given on page 248 of Forsyth's Theory of Differential Equations Part II (Cambridge, 1900) as $\left ( \frac{c}{n-1} \right )^{n-1}-\left ( -\frac{b}{n} \right )^{n}=0$. – Tom Copeland Jul 01 '12 at 23:03
  • These results (and book about discriminants by Gelfand, Kapranov and Zelevinsky) could interest you too. – Raymond Manzoni Feb 24 '14 at 23:41
  • 1
    See also section 6 of "Discriminating deltas, depressed equations, and generalized Catalan numbers" (http://tcjpn.wordpress.com/2012/06/13/depressed-equations-and-generalized-catalan-numbers/) to relate the tangents of the discriminant curve to the equation $x^n + b x + c = 0$. – Tom Copeland Aug 01 '16 at 23:33
  • See also p. 775 of "Function series, Catalan numbers and random walks on trees" by Bajunaid, Cohen, Colonna, and Singman – Tom Copeland Dec 25 '19 at 16:50

2 Answers2

4

Yes. Sketch: $b$ is a symmetric polynomial of degree $n-1$ in the roots and $c$ is a symmetric polynomial of degree $n$, whereas the entire discriminant is a symmetric polynomial of degree $n(n-1)$. It follows that the discriminant is a linear combination of $b^n$ and $c^{n-1}$, and the coefficients can be determined by setting $b = 0, c = -1$ and then $b = -1, c = 0$ and reducing to the computation of the discriminant of $x^n - 1$.

Qiaochu Yuan
  • 468,795
  • This is perhaps the easier way to see where the factor $n^n$ comes from if you are familiar with the DFT matrix, $S_n$, given in https://ccrma.stanford.edu/~jos/mdft/Matrix_Formulation_DFT.html . $det (S^{*}_{n}S_n)=n^n$ – Tom Copeland Jun 22 '12 at 08:54
  • Qiaochu, I noticed you deleted your allusion to the Vandermonde matrix, but I think the corresponding Wikipedia reference helps to make the connection between the discriminant and the DFT matrix (as well as other connections). – Tom Copeland Jun 22 '12 at 09:38
  • @Tom: well, it wasn't quite the right thing to say. It was a specific Vandermonde matrix rather than a general one. – Qiaochu Yuan Jun 22 '12 at 10:46
  • See also http://math.stackexchange.com/questions/103491/using-vietas-theorem-for-cubic-equations-to-derive-the-cubic-discriminant?rq=1 and http://math.stackexchange.com/questions/18119/combinatorial-proof-of-big-prod-limits-0-leq-i-j-n-zetaj-zetai – Tom Copeland Jun 30 '12 at 12:20
  • Why does the discriminant need to be a linear combination of $b^n$ and $c^{n-1}$, given that they are symmetric polynomials in the roots of those particular degrees? – PrimeRibeyeDeal Jan 27 '15 at 23:17
  • @Prime: by the fundamental theorem of symmetric functions, any symmetric polynomial in the roots must be a polynomial in the coefficients. There are only two coefficients here, so any symmetric polynomial in the roots must be a polynomial in $b$ and $c$, and those are the only two polynomials in $b$ and $c$ of the right degree (exercise). – Qiaochu Yuan Jan 27 '15 at 23:25
  • @QiaochuYuan oh that makes sense. Thanks! – PrimeRibeyeDeal Jan 30 '15 at 22:34
3

Use the relation between the disciminant of $f$ and the resultant of $f$ and $f'$. The resultant is easy to calculate since $f'$ is so simple.

Gerry Myerson
  • 185,413
  • Thanks, clearly the Sylvester matrix for the resultant has a sub-diagonal related to $n^n$, but then I'm left with understanding where the matrix itself comes from. – Tom Copeland Jun 22 '12 at 09:19
  • The idea of the resultant of $f$ and $g$ is that it should be zero if and only if $f$ and $g$ have a common (non-constant) factor. That happens if and only if there are polynomials $a$ and $b$ with degree $a$ less than degree $g$, and degree $b$ less than degree $f$, such that $af+bg=0$. And by considering this as a system of linear equations in the coefficients, this happens if and only if the Sylvester matrix is singular. – Gerry Myerson Jun 22 '12 at 12:32