6
$\begingroup$

It is known that if $x_1, x_2, ..., x_n$ are all positive distinct real numbers, then the matrix

$$ \begin{pmatrix} x_1^{a_1} & x_1^{a_2} & \cdots & x_1^{a_n} \\ x_2^{a_1} & x_2^{a_2} & \cdots & x_2^{a_n} \\ x_3^{a_1} & x_3^{a_2} & \cdots & x_3^{a_n} \\ \vdots & \vdots & \vdots & \vdots \\ x_n^{a_1} & x_n^{a_2} & \cdots & x_n^{a_n} \end{pmatrix}$$

is of rank $n$, where $a_1, a_2, ... , a_n$ are distinct natural (zero included) numbers (Apart from the proof in the link, I think it is a consequence of Descartes' rule of signs.).

Therefore, assuming $x_1, x_2, ..., x_n$ are all positive distinct real numbers, if we divide the columns of the matrix $$\begin{pmatrix} 1 & x_1 &x_1^2 \cdots & x_1^{2n-2} \\ 1 & x_2 &x_2^2 \cdots & x_2^{2n-2} \\ 1 & x_3 &x_3^2 \cdots & x_3^{2n-2} \\ \vdots & \vdots & \vdots & \vdots \\ 1 & x_n &x_n^2 \cdots & x_n^{2n-2} \end{pmatrix},$$ into two subsets, then the vectors (or elements) of one of the subsets span the whole $\mathbb R^n$ since one of the two subsets contains at least $n$ vectors, which are independent.

Now let's assume non zero $x_1, x_2, ..., x_n$ are not necessarily positive but their absolute values are all distinct. If we divide the columns of the matrix $$\begin{pmatrix} 1 & x_1 & x_1^2 \cdots & x_1^{4n-4} \\ 1 & x_2 & x_2^2 \cdots & x_2^{4n-4} \\ 1 & x_3 & x_3^2 \cdots & x_3^{4n-4} \\ \vdots & \vdots & \vdots & \vdots \\ 1 & x_n &x_n^2 \cdots & x_n^{4n-4} \end{pmatrix},$$ into two subsets, then the vectors of one of the subsets span the whole $\mathbb R^n$ since we have $2n-1$ even powers and so one of the two subsets has at least $n$ vectors with even powers, which are independent (Just compare it with the initial fact.).

However, when working with some examples, it seems that, having $4n-3$ vectors (or columns), is not a sharp upper bound. For example when $n=2$, $2n=4$ vectors do the job. I think when $n=3$, $2n=6$ vectors do the job too. I investigated most of the possible cases for $n=3$ and couldn't find any counterexamples. So my goal is to figure out if the following statement is true or not:

Assume $x_1, x_2, ..., x_n$ are non zero real numbers with distinct absolute values. If we divide the columns of the matrix $$\begin{pmatrix} 1 & x_1 & x_1^2 \cdots & x_1^{2n-1} \\ 1 & x_2 & x_2^2 \cdots & x_2^{2n-1} \\ 1 & x_3 & x_3^2 \cdots & x_3^{2n-1} \\ \vdots & \vdots & \vdots & \vdots \\ 1 & x_n & x_n^2\cdots & x_n^{2n-1} \end{pmatrix},$$ into two subsets, at least one of them spans $\mathbb R^n.$

If this statement is not true, any sort of counterexample is a great help.


I know that this problem can be transformed into a problem involving polynomials and their real roots. For example, when $n=3$, assume that $\{1, x^2, x^3\}$ and $\{x,x^4,x^5\}$ are our subsets (we can identify the vectors with their powers). $\{1, x^2, x^3\}$ not being independent means there are $c_0, c_2, c_3$ such that $x_1, x_2, x_3$ are roots of $c_0+c_2X^2+c_3X^3$, which implies $\frac{1}{x_1}+\frac{1}{x_2}+\frac{1}{x_3}=0$ (compute the coefficient of $X$). On the other hand, $\{x, x^4, x^5\}$ not being independent means there are $d_1, d_4, d_5$ such that $x_1, x_2, x_3$ are roots of $d_1X+d_4X^4+d_5X^5$, which implies $x_1, x_2, x_3$ and some other non zero real number, say $x_4$, are roots of $d_1+d_4X^3+d_5X^4$, which implies $\frac{1}{x_1}+\frac{1}{x_2}+\frac{1}{x_3}+\frac{1}{x_4}=0$ (compute the coefficient of $X$), and this is a contradiction. Similarly, we can rule out the other cases.


EDIT: I included the "Number Theory" tag since any proof or counterexamples for integers could be also interesting.

$\endgroup$
0

1 Answer 1

6
+50
$\begingroup$

As stated, this is too good to be true. For example, you can take $n=6$, and $x_i$ being the roots of a generic polynomial of the form $a+bx+cx^2+dx^6+ex^7+fx^8$enter image description here like this one. Then you partition the columns to two groups, one of them contains the columns number 0,1,2,6,7,8, another group contains the columns 3,4,5,9,10,11, i.e., the first group shifted by 3. Both minors corresponding to these sets of columns are singular: the column vector $(a,b,c,d,e,f)^t$ annulates them.

$\endgroup$
4
  • $\begingroup$ Thank you for your answer. I think the $7$th power is missing in your example. And, how do you know that generic polynomial can have $6$ real roots? $\endgroup$ Commented Jul 18 at 20:31
  • 2
    $\begingroup$ Not every generic, but this example or its fluctuations works in the sense that absolute values are distinct. Zero coefficient is ok, but again, after peturbation you get all six guys $a, b, c, d, e, f$ non-zero $\endgroup$ Commented Jul 18 at 20:38
  • $\begingroup$ I see. Having $e=0$ means that those five remaining vectors are not independent, so neither of the subsets can span $\mathbb R^n.$ Right? $\endgroup$ Commented Jul 18 at 20:49
  • $\begingroup$ Right. So, there are other bad partitions too $\endgroup$ Commented Jul 19 at 3:05

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.