Closure, Interior, and the Test for Nowhere Denseness
A set can be small in one sense and large in another. The integers are spread through the real line but leave open intervals between their points; the rationals meet every open interval, even though they do not contain any open interval. To classify such sets, we must inspect not just a set’s interior, but the interior of its closure.
Throughout, \(\mathbb{R}\) has its usual topology, whose open neighborhoods contain open intervals around each of their points. For \(A\subseteq\mathbb{R}\), \(\overline{A}\) denotes its closure and \(\operatorname{int}A\) its interior. Recall that \(x\in\overline{A}\) exactly when every open neighborhood of \(x\) meets \(A\), and that \(\operatorname{int}A\) is the union of all open sets contained in \(A\).
This definition immediately gives a useful distinction. If \(A\) is closed, then \(\overline{A}=A\), so \(A\) is nowhere dense exactly when \(\operatorname{int}A=\varnothing\). If \(A\) is not closed, its closure may have a nonempty interior even when \(A\) itself has none.
Singletons and Finite Sets
Proof. Let \(F\subseteq\mathbb{R}\) be finite. If \(F=\varnothing\), it is closed and has empty interior, so it is nowhere dense. Now suppose \(F\ne\varnothing\), and write \(m\) for its number of elements. To show \(F\) is closed, take any \(y\in\mathbb{R}\setminus F\). For each \(f\in F\), the distance \(|y-f|\) is positive. The minimum \(d=\min_{f\in F}|y-f|\) exists because \(F\) is finite, and \(d>0\). The interval \((y-d/2,y+d/2)\) contains \(y\) and misses \(F\), since every \(f\in F\) is at distance at least \(d\) from \(y\). Thus every point of \(\mathbb{R}\setminus F\) has an open neighborhood in \(\mathbb{R}\setminus F\), so the complement is open and \(F\) is closed. Hence \(\overline{F}=F\).
To show that \(F\) has empty interior, suppose an open interval \((x-r,x+r)\), with \(r>0\), were contained in \(F\). The \(m+1\) numbers \(x+r/(j+2)\), for \(j=1,\ldots,m+1\), are distinct and all lie in \((x,x+r)\subseteq(x-r,x+r)\). A set with \(m\) elements cannot contain \(m+1\) distinct numbers, a contradiction. No nonempty open interval is contained in \(F\), so \(\operatorname{int}F=\varnothing\). Because \(F\) is closed, \(\operatorname{int}\overline{F}=\operatorname{int}F=\varnothing\). Therefore \(F\) is nowhere dense. \(\square\)
Worked Example: A Singleton
Take \(A=\{3\}\). The finite-set result shows that \(A\) is closed, so \(\overline{A}=\{3\}\). Its interior is empty: every open neighborhood of \(3\) contains points other than \(3\), and therefore cannot be a subset of \(\{3\}\). It follows that \(\operatorname{int}\overline{A}=\varnothing\), so \(\{3\}\) is nowhere dense.
Check: the interval \((2,4)\) is a neighborhood of \(3\), and it contains \(5/2\), which is not in \(\{3\}\). This confirms that \(\{3\}\) contains no neighborhood of its point. Its closure is still exactly \(\{3\}\) because the set is closed.
Worked Example: A Finite Set with Several Points
Let \(F=\{-2,0,5\}\). The finite-set result gives \(\overline{F}=F\) and \(\operatorname{int}\overline{F}=\varnothing\), so \(F\) is nowhere dense. For a direct check that it is closed, take \(y=1\). Its distances to the three points are \(|1-(-2)|=3\), \(|1-0|=1\), and \(|1-5|=4\). The minimum is \(1\), so \((1-1/2,1+1/2)=(1/2,3/2)\) contains \(1\) and misses \(F\).
For a direct check that the interior is empty, \((-1,1)\) contains \(1/2\notin F\), and \((4,6)\) contains \(9/2\notin F\). More generally, the proposition proves that no open interval can be contained in this three-point set. Thus both parts of the nowhere-dense test hold: the closure is \(F\), and that closure contains no open interval.
The Integers: Closed but with Empty Interior
Proof. First we show that \(\mathbb{Z}\) is closed. If \(y\notin\mathbb{Z}\), there is an integer \(k\) with \(k<y<k+1\). Both \(y-k\) and \(k+1-y\) are positive, so let \(r=\frac12\min(y-k,k+1-y)>0\). Then \((y-r,y+r)\subset(k,k+1)\), which contains no integers. Thus every point outside \(\mathbb{Z}\) has a neighborhood disjoint from \(\mathbb{Z}\), and \(\mathbb{R}\setminus\mathbb{Z}\) is open. Hence \(\mathbb{Z}\) is closed and \(\overline{\mathbb{Z}}=\mathbb{Z}\).
Next, no nonempty open interval is contained in \(\mathbb{Z}\). Indeed, any open interval contains a bounded open subinterval, which contains infinitely many distinct real numbers. Only finitely many integers can lie in a bounded interval: if the interval is contained in \((a,b)\), its integers must lie between two fixed integers that bound \(a\) and \(b\). Therefore the subinterval contains a point not in \(\mathbb{Z}\). This proves \(\operatorname{int}\mathbb{Z}=\varnothing\). Since \(\mathbb{Z}\) is closed, \(\operatorname{int}\overline{\mathbb{Z}}=\varnothing\), as required. \(\square\)
Worked Example: Testing an Integer and a Noninteger
Consider the interval \((2.8,3.2)\) around the integer \(3\). It contains no integer other than \(3\): the neighboring integers are \(2\) and \(4\), and \(2<2.8\) while \(4>3.2\). This kind of gap does not remove \(3\) from the closure of \(\mathbb{Z}\), because \(3\) itself belongs to \(\mathbb{Z}\). The set is closed, so its closure contains exactly its integer points.
Now take the noninteger \(y=3.4\). The adjacent integers are \(3\) and \(4\), with distances \(3.4-3=0.4\) and \(4-3.4=0.6\). The proof’s choice gives \(r=\frac12\min(0.4,0.6)=0.2\), so the neighborhood \((3.2,3.6)\) contains \(3.4\) and no integer. This confirms directly that \(3.4\notin\overline{\mathbb{Z}}\).
Check: every integer has arbitrarily small neighborhoods, but none of those neighborhoods is a subset of \(\mathbb{Z}\); for example, \((3-0.1,3+0.1)\) contains \(3.05\), which is not an integer. The closure is \(\mathbb{Z}\), whose interior is empty, so the set is nowhere dense.
The Rationals: Dense but Not Nowhere Dense
The rationals provide the contrasting case. We will verify both relevant facts: every nonempty open interval contains a rational, so the closure is all of \(\mathbb{R}\); but every nonempty open interval also contains an irrational, so the rationals themselves have empty interior.
Proof. Let \((a,b)\) be any nonempty open interval, so \(a<b\). Choose a positive integer \(n\) with \(n(b-a)>2\), using the Archimedean property of \(\mathbb{R}\). Let \(m=\lfloor na\rfloor+1\), where \(\lfloor na\rfloor\) is the greatest integer not exceeding \(na\). Then \(na<m\le na+1<nb\), because \(nb-na>2\). Dividing by \(n>0\) gives \(a<m/n<b\). Thus every nonempty open interval contains a rational number. Every neighborhood of every real number contains such an interval, so every neighborhood meets \(\mathbb{Q}\). By the neighborhood test for closure, \(\overline{\mathbb{Q}}=\mathbb{R}\).
To show \(\operatorname{int}\mathbb{Q}=\varnothing\), take any nonempty open interval \((a,b)\). By the result just proved, choose \(q\in\mathbb{Q}\cap(a,b)\). Choose a positive integer \(n\) large enough that \(\sqrt{2}/n<b-q\). Then \(q+\sqrt{2}/n\) is still in \((a,b)\), since it is greater than \(q>a\) and less than \(b\). It is irrational: if it were rational, subtracting the rational number \(q\) and multiplying by the integer \(n\) would make \(\sqrt{2}\) rational. To recall why this is impossible, if \(\sqrt{2}=p/s\) in lowest terms with positive integers \(p,s\), then \(p^2=2s^2\), so \(p\) is even; writing \(p=2t\) gives \(s^2=2t^2\), so \(s\) is even too, contradicting that \(p/s\) was in lowest terms. Thus every nonempty open interval contains a point outside \(\mathbb{Q}\), and \(\operatorname{int}\mathbb{Q}=\varnothing\).
Finally, \(\operatorname{int}\overline{\mathbb{Q}}=\operatorname{int}\mathbb{R}=\mathbb{R}\), which is not empty. By definition, \(\mathbb{Q}\) is therefore not nowhere dense. \(\square\)
Worked Example: Rationals and Irrationals in an Interval
Consider \((1/3,2/3)\). The rational \(1/2\) lies in this interval because \(1/3<1/2<2/3\). The number \(1/2+\sqrt{2}/100\) lies there too: it is greater than \(1/2>1/3\), and \(\sqrt{2}<2\) gives \(1/2+\sqrt{2}/100<1/2+2/100=52/100<2/3\). It is irrational, since a rational value for \(1/2+\sqrt{2}/100\) would imply a rational value for \(\sqrt{2}\).
This interval illustrates the two different claims. It contains a rational, as every nonempty open interval does, and it also contains an irrational, so it is not a subset of \(\mathbb{Q}\). In fact, the rationals meet every neighborhood in \(\mathbb{R}\), making their closure all of \(\mathbb{R}\). The closure has nonempty interior, which is why \(\mathbb{Q}\) fails the nowhere-dense test.
Check: the inequalities above place \(1/2\) and \(1/2+\sqrt{2}/100\) strictly between the endpoints. The first belongs to \(\mathbb{Q}\); the second does not. Thus an empty interior for \(\mathbb{Q}\) does not imply that its closure has empty interior.
A Common Pitfall: Checking the Wrong Set
A frequent error is to reason, “\(\operatorname{int}\mathbb{Q}=\varnothing\), so \(\mathbb{Q}\) is nowhere dense.” The first statement is correct, but it checks only whether \(\mathbb{Q}\) contains an open interval. The definition asks whether \(\overline{\mathbb{Q}}\) contains an open interval. Since \(\overline{\mathbb{Q}}=\mathbb{R}\), its interior is \(\mathbb{R}\), not the empty set. Therefore the rationals are not nowhere dense.
The comparison can be summarized as follows. The closure of a finite set is itself and has empty interior; the closure of \(\mathbb{Z}\) is itself and has empty interior; but the closure of \(\mathbb{Q}\) is the whole real line. The closure step is essential, especially when the set under consideration is not closed.
Check Your Understanding
For each question, use the closure-and-interior definition rather than checking only the set’s own interior.
- Is \(\{0,1,2\}\) nowhere dense in \(\mathbb{R}\)? State its closure and the interior of that closure.
- Explain why the empty set is nowhere dense using the definition.
- Why does the fact that \(\operatorname{int}\mathbb{Q}=\varnothing\) not make \(\mathbb{Q}\) nowhere dense?
- Give a neighborhood of \(7/2\) that misses \(\mathbb{Z}\), and explain what this shows about the closure of \(\mathbb{Z}\).
- Which part of the nowhere-dense test changes when a set is not closed?