MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Suppose I have distinct real numbers $a_i \in [-1,1]$, $i \in [k]$. I want to choose real numbers $b_j, j\in [k]$ such that the matrix $(\arccos(a_i b_j))_{i,j \in [k]}$ is nonsingular.

Is this always possible? Equivalently, does $\arccos$ not satisfy a functional equation of the form $\det(\arccos(a_i x_j)) = 0$ for indeterminates $x_j, j \in [k]$?

It seems like this should hold for any sufficiently ``non-algebraic'' function in place of $\arccos$. Is there a theory that handles such questions? Does functional analysis deal with these questions? I know very little functional analysis, so some general references may be helpful.

share|cite|improve this question
A quick numerical test indicates that it seems to hold for most polynomials, though certainly not for monomials. – Steve Huntsman May 1 '10 at 22:06
No it never holds for polynomials. The functions $f(a_ix)$ are linearly dependent if their number is greater than degree plus 1. – Sergei Ivanov May 2 '10 at 10:51
up vote 12 down vote accepted

A quick counter-example to the question as stated is $a_0=0$, $a_1=1$ $a_2=-1$. Since $2arccos(0)-arccos(b)-arccos(-b)=0$ for all $b$, we have $2M_1-M_2-M_3=0$ where $M_1, M_2, M_3$ are the rows of the matrix.

So it is better to assume that the $a_i$ are nonnegative. In this case, the answer is yes. More generally, consider this problem for a function $f$. You want to choose $n$ linearly independent vectors from the set $$ X := \{ (f(a_1x), f(a_2x),\dots, f(a_nx)) \mid x\in\mathbb R \} \subset \mathbb R^n $$ This is not possible if and only if $X$ lies in an $(n-1)$-dimensional subspace. This means that there are constants $c_1,\dots,c_n$ (not all zero) such that $c_1v_1+\dots+c_nv_n=0$ for all $v\in X$. Or, equivalently, $$ c_1 f(a_1x) + \dots c_n f(a_nx) = 0 $$ for all $x\in\mathbb R$ (such that all $a_ix$ belong to the domain of $f$). In other words, the functions $x\mapsto f(a_ix)$ are linearly dependent over $\mathbb R$.

This cannot happen if $f=arccos$. Indeed, assuming that $a_n$ is the maximum of $a_i$ such that $c_i\ne 0$, the above sum is well-defined and analytic on $[0,1/a_n)$ but its derivative goes to infinity as $x$ approaches $1/a_n$. Hence it is not constant on $[0,1/a_n)$, and hence non-constant in any neighborhood of 0.

UPDATE. A similar argument shows that the answer is the same for any non-polynomial function analytic near the origin (assuming $a_i>0$). Indeed, if $f(x)=\sum_{j=1}^\infty q_j x^{k_j}$ where $q_j\ne 0$, then Taylor expansion of the identity $$ c_1 f(a_1x) + \dots c_n f(a_nx) = 0 $$ implies that $\sum_i c_i a_i^{k_j}=0$ for all $j$. This cannot happen because, if $a_n$ is the maximum of $a_i$, the term $a_i^{k_j}$ grows faster (or decays slower) than all other terms as $k_j\to\infty$.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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