1

Suppose there is an $n$-dimensional affine subspace - let us call it $S$ of $\mathbb{R}^d$, $d>n$. Suppose we have $k>n$ hyperplanes in $\mathbb{R}^d$. We want to determine the dimensions of the intersection of these hyperplanes and $S$. It is given that the intersection is non-empty.

A very helpful answer here says, for $k>n$ the intersection can have any number of dimensions from $0$ to $n−1$. My question is: How do we determine the actual dimensions of the intersection? My hyperplanes are all of the form $x_{l} = x_{l'}$, i.e. they stipulate equality of pairs of coordinates of a point in $d$ dimensions.

My approach: The $k$ equalities are all independent. But it is also given that the intersection is non-empty. I'm at a loss trying to determine how this is possible.

Edit: An earlier version of the question did not metion $\mathbb{R}^d$ at all, reduced the space to $S$ and spoke as if we have a set of $k>n$ equalities, which create a subspace of $S$, which in turn implies that some of the $k$ equalities (when the space is reduced to S) have to be redundant (since it is given that the intersection is non-empty). I subsequently edited the question to add the details about $\mathbb{R}^d$ at the suggestion of user Zanxiong, who very helpfully provided the first answer.

Canine360
  • 1,481

1 Answers1

2

Express each hyperplane by its analytical equation $a_{i1}x_1 + a_{i2}x_2 + \cdots + a_{in}x_n = b_i, i = 1, \ldots, k$. Then the intersection is the solution to the linear system $Ax = b$, i.e., the set $S = \{x \in \mathbb{R}^n: Ax = b\}$, where \begin{align*} A = \begin{pmatrix} a_{11} & a_{12} & \cdots & a_{1n} \\ a_{21} & a_{22} & \cdots & a_{2n} \\ \vdots & \vdots & \ddots & \vdots \\ a_{k1} & a_{k2} & \cdots & a_{kn} \end{pmatrix}, \quad x = \begin{pmatrix} x_1 \\ x_2 \\ \vdots \\ x_n \end{pmatrix}, \quad b = \begin{pmatrix} b_1 \\ b_2 \\ \vdots \\ b_n \end{pmatrix}. \end{align*}

Now you can unleash the theory of linear system to address the problem. When $b \neq 0$, then $S \neq \emptyset$ if and only if $\text{rank}((A, b)) = \text{rank}(A)$. However, in general $S$ is not a subspace of $\mathbb{R}^n$, hence it's pointless to discuss the dimension of $S$.

When $b = 0$, then it is well known that $\dim(S) = n - \text{rank}(A)$. So to determine the dimension of $S$ boils down to determine the rank of the coefficient matrix $A$. Since $0 \leq \text{rank}(A) \leq \min(n, k) = n$, $0 \leq \dim(A) \leq n$.

Zhanxiong
  • 14,040
  • Actually in my case they are both part of a $d$-dimensional space where $d>k>n$. In the $d$-dimensional space, the $k$ equalities are all independent. Does that tell us anything about how its restriction to the $n$-dimensional subspace would behave, in terms of dimensionality? – Canine360 Sep 23 '20 at 18:10
  • In your question you stated the space is $n$-dimensional. It seems to me that one of $d$ and $n$ is redundant. Please edit the question if you mean something different. – Zhanxiong Sep 23 '20 at 19:16
  • Yes, the $n$ dimensional space is a subspace of the $d$-dimensional space. Basically the equalities in $A$ (per your notation) define another subspace of the $d$-dimensional space. I'm trying to determine the intersection between the two. I didn't talk about the $d$ dimensional space in which they are both contained, because I thought it was not important to the answer. I might be wrong, of course. Based on the above clarification - do you still suggest I edit the question? – Canine360 Sep 23 '20 at 19:41
  • Yes, I think you still need to edit. Your original question doesn't refer to the second $n$-dimensional space at all, which actually might invalidate my answer... – Zhanxiong Sep 23 '20 at 21:14
  • Thank you for your suggestions. I have thought more about the problem, and I think your answer is correct even with the details about $\mathbb{R}^d$. Nevertheless, I have added those details. Thanks again. – Canine360 Sep 29 '20 at 15:48
  • Also, I have had to ask another linear algebra related question here, which I think is very simple. I would request you to kindly have a look if you can. Thank you again. https://math.stackexchange.com/questions/3845038/extreme-points-of-polytopes – Canine360 Sep 29 '20 at 15:51