1
$\begingroup$

If I have (finite) $k$ vectors, $u_1,...,u_k\in\mathbb{R}^N$ that are in general linearly dependent is it possible to take positive affine transformations of the form: $$u'_i=\alpha_i u_i +\beta_i\vec{1}_N$$ $$~\text{with}~~\alpha\in\mathbb{R}_{++} \text{ and } \beta\in\mathbb{R}$$

(The idea is each $u$ is an expected utility function and these transformations preserve the preference ordering on lotteries)

So that the collection of the transformed vectors are linearly independent?

I'm working on a project and this would be a useful condition to have. Any help is appreciated!

We can assume that there are 2 dimensions (call them $b$ and $w$) such $\forall u_i$ the $b^{th}$ entry of vector is the largest and the $w^{th}$ entry is the smallest.

It seems like I should be able to rescale such that for every $u$ has $u'(w)=-1$, which let's me focus on convex combinations rather than linear combinations. This seems useful though not quite enough.

$\endgroup$
2
  • 1
    $\begingroup$ But how is the addition of $u_i + \beta_i$ defined if $u_i$ is a vector and $\beta_i$ is a number? $\endgroup$ – daw Oct 9 '14 at 15:35
  • $\begingroup$ Right, so adding the same constant to every component. $\endgroup$ – Kevin Oct 9 '14 at 19:49
1
$\begingroup$

This can be only done if the following two conditions are satisfied: $$ \dim \mathrm{span}(v_1\dots v_{k}) = k-1 $$ and $$ e\not\in \mathrm{span}(v_1\dots v_{k}). $$ where $e=(1,\dots,1)^T$.

If $\dim \mathrm{span}(v_1\dots v_{k}) < k-1$, then $$ \dim \mathrm{span}(v_1\dots v_{k}, e) < k, $$ which implies that there are not $k$ linear independent vectors in that set.

The same can be said, if $e\in \mathrm{span}(v_1\dots v_{k})$.

Now let me argue that the conditions above are also sufficient: Let $i$ be an index such that $v_i \in \mathrm{span}( \{v_1\dots v_k\}\setminus v_i)$. Then $v_i+e$ is not in that span. Moreover, $$ \dim \mathrm{span}( \{v_1\dots v_{k}\}\setminus \{v_i\} \cap \{v_i+e\})=k, $$ hence these vectors are linearly independent. Thus the vectors $$ u'_j = u_j \quad j\ne i, \ u'_i = u_i + e $$ satisfy the requirements.

$\endgroup$

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.