Suppose $f_1, f_2,...$ are a set of functions $\mathbb{R} \rightarrow \mathbb{R}$ so that each is the power of some non-constant function h. So $ f_ i=h^{n_i}$ for some natural number n.

Is it possible for such a set to be linearly dependent?

What additional conditions could ensure independence?

I was thinking about this because I need to show that a set in linearly independent. And I was not convinced that having the functions be powered lhjlj would be sufficient.

  • $\begingroup$ Do you mean $f_i=h^{n_i}$ for some natural number $n_i$? $\endgroup$ – André Nicolas Jul 5 '11 at 0:14
  • $\begingroup$ Yes. I was not thinking of n as fixed, but I think that would make it more clear. $\endgroup$ – Noteventhetutorknows Jul 5 '11 at 0:16
  • 1
    $\begingroup$ Sorry to be pesky, but I want people on this site to be able to help. You have $N$ in there, I assume you mean $n_i$. And what does "set of functions in R2" mean? What is R2? What is the domain of these functions? Is it $\mathbb{R}^2$? Are these functions to the reals? To $\mathbb{R}^2$? $\endgroup$ – André Nicolas Jul 5 '11 at 0:26

The following applies over an arbitrary field, with scalar valued functions on an arbitrary set (and was written before these were specified). I will make the simplifying assumption that in fact $f_i = h^i$, which results in no loss of generality.

The set $\{h,h^2,h^3,\ldots\}$ is linearly independent if and only if $h$ takes on infinitely many values. (In particular, with the assumption that $h:\mathbb R\to\mathbb R$, it would suffice for $h$ to be nonconstant and continuous.)

First suppose that $h$ takes on infinitely many values, and let $n$ be a positive integer. To see that $\{h,h^2,\ldots,h^n\}$ is linearly independent, suppose that $a_1,a_2,\ldots,a_n$ are scalars such that $a_1h+a_2h^2+\cdots a_nh^n=0$ identically. Let $x_1,\ldots, x_n$ be points in the domain where $h$ takes distinct nonzero values, say $y_i=h(x_i)$, with $y_i\neq y_j$ if $i\neq j$. Then the equations you get from plugging the $x_i$s into the identity $a_1h+a_2h^2+\cdots a_nh^n=0$ can be organized into the matrix equation $$\begin{bmatrix} y_1 & y_1^2 & \cdots & y_1^n\\ y_2 & y_2^2 & \cdots & y_2^n\\ \vdots & \vdots & \ddots & \vdots\\ y_n & y_n^2 & \cdots & y_n^n \end{bmatrix} \begin{bmatrix} a_1 \\a_2\\ \vdots \\ a_n\end{bmatrix}=0.$$

If $A$ is the square matrix in this equation, note that $A$ has the same determinant as the invertible Vandermonde matrix $\begin{bmatrix} 1&\vec 0\\ \vec1& A\end{bmatrix}$, where $\vec{1}$ is the column vector of all $1$s and $\vec{0}$ is the row vector of all $0$s. Thus, $A$ is invertible, which implies that $a_1=a_2=\cdots=a_n=0$.

Now suppose that $h$ takes on only finitely many values, say $k$ of them, and that the sets where $h$ takes on distinct values are $E_1,E_2,\ldots,E_k$. The vector space of functions that are constant on each $E_i$ is finite dimensional with dimension $k$, and each $h^i$ lies in this space. Therefore $\{h,h^2,\ldots,h^{k+1}\}$ is linearly dependent.

  • $\begingroup$ Another way to see that $A$ is invertible using Vandermonde matrices is to note that $$A=\begin{bmatrix} y_1 & 0 & \cdots & 0\\ 0 & y_2 & \cdots & 0\\ \vdots & \vdots & \ddots & \vdots\\ 0 & 0 & \cdots & y_n \end{bmatrix} \begin{bmatrix} 1 & y_1 & \cdots & y_1^{n-1}\\ 1 & y_2 & \cdots & y_2^{n-1}\\ \vdots & \vdots & \ddots & \vdots\\ 1 & y_n & \cdots & y_n^{n-1} \end{bmatrix}.$$ $\endgroup$ – Jonas Meyer Jul 5 '11 at 1:06

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.