30
$\begingroup$

As defined in Convex Optimization written by Stephen Boyd & Lieven Vandenberghe, both interior and relative interior seems to describe a same thing: a set that peels away its boundary points. So, what on earth is the difference between these two concepts ?

Here are the definitions of these two concepts from Convex Optimization:

interior: The set of all points interior to $C$ is called the interior of $C$

relative interior: its interior relative to the affine hull of $C$

$\endgroup$
1
  • 6
    $\begingroup$ To make your question useful for everybody, you should copy the actual definitions of those two concepts given in the book. $\endgroup$ Feb 12 '15 at 7:37
37
$\begingroup$

The relative interior of a set is the interior of the set when it is viewed as a subset of the affine space it spans.

For example, the interior of the segment connecting $(0,0)$ to $(1,1)$ in the plane is empty, but the relative interior is the open segment with those endpoints.

$\endgroup$
4
  • $\begingroup$ For any subset of $\mathbb R^n$ (or maybe more generally of a topological vector space), not just for a convex set. $\endgroup$ Feb 12 '15 at 7:42
  • $\begingroup$ Indeed. (You can edit my answers, you know :-) ) $\endgroup$ Feb 12 '15 at 7:43
  • $\begingroup$ I downvoted since it is not enough elaborated $\endgroup$ Jan 8 at 11:06
  • $\begingroup$ @RémyHosseinkhanBoucher, you are of course free to downvote it — I wish you could do it more than once! — but honestly the question is «what is the difference between interior and relative interior?» and how much more elaborated can you get apart explaining what is the difference? There is even an example! (I'd say that The Point's answer does not make a lot of sense, for "undesirable" is quite a badly defined criterion!) $\endgroup$ Jan 10 at 7:28
27
$\begingroup$

I came across this question whilst studying the same textbook, but I found that Mariano Suárez-Álvarez's gracious response, whilst, of course, appreciated, to be lacking the detail I desired. Fortunately, Justin Rising posted a more elaborate answer to the same question on Quora. I have decided to copy and paste it here for the benefit of others.

Justin Rising's response to the same question on Quora:

Consider $\mathbb{R}^3$ with the standard Euclidean metric, and look at the set of points $(x, y, 0)$ such that $x^2 + y^2 \leq 1$. What's the interior of this set?

A point is in the interior of a set if you can draw a small open ball around it which is itself contained in the set. But for any point in our set, any open ball around any point in it will contain points outside the $x-y$ plane, and so it has empty interior.

That's technically correct, but it's undesirable behavior. The relative interior of a set is obtained by looking at the affine hull (http://en.wikipedia.org/wiki/Affine_hull) of the set, which in this case is $\mathbb{R}^2$, and using the appropriate metric there. So while our set has empty interior, its relative interior is the interior of the unit disc, which is what it "really should be".

Relative interiors are a big deal in convex optimization, where you'd really be restricting the set of problems you can solve for no good reason if you insisted on using interiors rather than relative interiors.

It came up in the comments, and it's worth adding to the answer, that there are very good reasons to talk about both interiors and a relative interiors. The relative interior is only defined in vector spaces; in particular, you have notions of addition and scalar multiplication to define an affine hull. If I were to define a topology on the words in the dictionary, I could talk about the interior of a set immediately, but I'd have to do quite a bit of legwork to be able to talk about relative interiors.

$\endgroup$
2
  • 2
    $\begingroup$ According to the definition of affine hull, the affine hull of the set (x,y,0) s.t. x^2+y^2<=1 should be a a circle area on R^2 but not the entire R^2, right? $\endgroup$
    – czxttkl
    Jun 4 '20 at 0:08
  • $\begingroup$ @czxttkl no dont think so. Since it includes all combinations of points thus vectors spanning entire R2. What confuses me though is what will be the relative interior, will it be outside the circle or inside the circle. $\endgroup$
    – DuttaA
    Sep 20 '20 at 17:15
0
$\begingroup$

Put it simple and intuitive:

The relative interior of C is viewed from the affine space $\mathrm{aff(C)}$, and the interior of C is viewed from the whole space where C lies in, which is sometimes "bigger" than $\mathrm{aff(C)}$. E.g. in Boyd's book example 2.2: The relative of C is viewed from $\mathrm{aff(C)=R^2}$ while the interior of C is viewed from $\mathrm{R^3}$.

Inspired by the answer from Michal Adamaszek.

$\endgroup$

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.