30
$\begingroup$

Suppose I have a function f : A → B between two sets A and B. (The same question applies to group homomorphisms, continuous maps between topological spaces, etc. But for simpicity let's restrict ourselves to the case of arbitrary maps between sets.)

If I have subset A' ⊆ A of the domain A, then I can restrict f to A'. There is the function f|A' : A' → B which is given by f|A'(a) = f(a) for all a ∈ A'.

Suppose now that I have subset B' ⊆ B of the codomain B that contains the image of the map f. Similarly I can restrict f to B', meaning that there is a function g : A → B' which is given by g(a) = f(a) for all a ∈ A.

In general, it might be useful to consider such functions g and have a name for them, for example if I have a function B' → C that I want to apply afterwards. Unfortunately, I haven't seen a name or symbol for this function g in literature.

Is there notation for the restriction g of f to a subset of the codomain similar to the notation f|A' for a restriction to a subset of the domain?

$\endgroup$
17
  • 8
    $\begingroup$ @Willie Wong: the target of your function is still $B$. @noname: Let $\iota: B'\to B$ be the inclusion; then the function you are referring to is sometimes called $\iota^*f$, which as the notation suggests is a special case of the pullback. $\endgroup$ Jun 29, 2010 at 13:21
  • 17
    $\begingroup$ I'd be tempted to invent the notation ${}_{B'}|f$ for this. $\endgroup$ Jun 29, 2010 at 13:58
  • 26
    $\begingroup$ Is the inclusion map {0}->{0,1} the same as the identity map {0}->{0}? According to one well-established "usual convention", they are, since a function is a set of ordered pairs. According to another, they are not. You need the latter convention in order to meaningfully ask questions like "is this function surjective?" $\endgroup$ Jun 29, 2010 at 14:12
  • 33
    $\begingroup$ @Harald: Alternatively, for ease of typesetting: $f|^{B'}$. $\endgroup$
    – Ryan Reich
    Jun 29, 2010 at 15:49
  • 13
    $\begingroup$ Yes, but the original poster is obviously in a situation where it matters. $\endgroup$ Jun 29, 2010 at 18:00

7 Answers 7

15
$\begingroup$

Concerning the name for the notion in question, but not the notation, Exposé 2 by A. Andreotti in the Séminaire A. Grothendieck 1957, available at www.numdam.org, suggests the following:

Consider a morphism $f:A\rightarrow B$ in some category, subobjects $i:U\rightarrow A$ and $j:V\rightarrow B$ of $A$ and $B$, respectively, and quotient objects $p:A\rightarrow P$ and $q:B\rightarrow Q$ of $A$ and $B$, respectively.

Then, $f\circ i$ is the restriction of $f$ to $U$. Dually, $q\circ f$ is the corestriction of $f$ to $Q$. (In particular, with the usual usage of the prefix "co", corestriction is not suitable for the notion in question.)

Moreover, if there is a morphism $g:P\rightarrow B$ with $g\circ p=f$, then $g$ is the astriction of $f$ to $P$. Dually, if there is a morphism $h:A\rightarrow V$ with $j\circ h=f$, then $h$ is the coastriction of $f$ to $V$. (Of course one can argue whether one should swap the terms astriction and coastriction (as suggested by Gerald Edgar).)

$\endgroup$
1
13
$\begingroup$

I would call it simply a pullback (along $B'$). Thus, you may denote it by $f \times_B B'$.

If we pullback a map $f : A \to B$ along a subset $B' \subseteq B$, we get the map $f^{-1}(B') \to B', x \mapsto f(x)$. If the image of $f$ happens to be a subset of $B'$, then $f^{-1}(B')=A$.

$\endgroup$
3
  • 2
    $\begingroup$ I agree. This is the most sensible thing to do. $\endgroup$ May 24, 2011 at 4:14
  • 1
    $\begingroup$ I agree with this (and Daniel Litt's comment). $\endgroup$
    – S. Carnahan
    May 24, 2011 at 4:36
  • 1
    $\begingroup$ I think the preimage (mathworld.wolfram.com/Preimage.html) notation $f^{-1}(B')$ is very sensible, and therefore I would denote the restricted function with ${f}|_{f^{-1}(B')}$ $\endgroup$
    – mareoraft
    Aug 3, 2015 at 19:23
12
$\begingroup$

If I wanted a name, I might use "corestriction."

$\endgroup$
5
$\begingroup$

It's called a range restriction. There's no established notation, but you might as well use the Z notation which is f  ▷ B'. (f \rhd B' in LaTeX plus amsfonts)

$\endgroup$
3
$\begingroup$

The terminology I have seen for this is "astriction".

$\endgroup$
1
  • 20
    $\begingroup$ Well, yet another example that not all existing terminology should be perpetuated! :) $\endgroup$ May 24, 2011 at 4:03
3
$\begingroup$

Here's some notation.

Given a morphism $f : A \rightarrow B$ and a subobject $b': B' \rightarrow B$, the astriction of $f$ to $B' \subseteq B$ can be denoted $b'\backslash\, f : A \rightarrow B'$, pronounced "$b'$ under $f$." It satisfies $b'\circ (b'\backslash\, f).$ For example, if we have a function $f : A \rightarrow B$ and subsets $X$ of $A$ and $Y$ of $B$ such that $f(X)\subseteq Y$, then the relevant map $X \rightarrow Y$ can be denoted $Y\,\backslash\, (f \circ X).$

Similar notation can be used in linear algebra to denote solution sets. For example; to denote the set of all vectors $x$ satisfying a linear equation $Ax=b,$ I tend to write $A \backslash b.$ This has an intuitive look about it: $$Ax = b \iff x \in A \backslash b.$$

The connection basically is that in the category of matrices, $Ax=b$ can be seen as a factorization of $b$ through the tail of $A$ (because $A$ is on the left). So, the general principle is that:

In any category, $g \backslash \,f$ is good notation for the set of all factorizations of $f$ through the tail of $g$, or for the unique such factorization just in case a unique factorization exists.

Dually, if we have a function $f : A \rightarrow B$ and a quotient object $a' : A \rightarrow A'$, we can write $f/a'$ for the relevant morphism $A' \rightarrow B$ when it exists, or else use this notation to denote the set of all such morphisms.

$\endgroup$
1
2
$\begingroup$

This would fit as a response to Harald Hanche-Olsen's remark, but I have not enough points for this.

Anyway, Mizar mathematial library chose exactly this notation; actually, it is introduced in the article on basic relations RELAT_1.MIZ, Def. 12, so it fits any relation and any set.

In a Mizar article you have just ASCII, so no subscripts, but you can always avoid ambiguities like f|X versus X|f by typing the two objects appearing in the notation as Relation and set respectively. I find interesting that what is expressible on paper by varying font size is emulated by typing in a proof checker.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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