Metric completeness and closure from a basis
Introduced| English | Meaning |
|---|---|
| basis/ˈbeɪsɪs/ | A family of open sets such that every open set is a union of its members. |
| isometry/aɪˈsɒmətri/ | A map that preserves all distances between pairs of points. |
| complete/kəmˈpliːt/ | Describing a metric space in which every Cauchy sequence converges to a point of that space. |
A Cauchy sequence can run out of space
- A sensor records a real input x through the compressed scale $h(x)=\arctan x$. If distance means the gap between sensor readings, inputs $1,2,3,\ldots$ get arbitrarily close together in the tail, yet approach no real input.
- This lesson separates a metric from its induced topology. It also studies a basis 基 of open sets, where closure is decided by neighbourhoods rather than ordinary numerical closeness.
Pull distance back through a map
- For a map $h:X\to\mathbb R$, define $d(x,y)=|h(x)-h(y)|$. Distances are nonnegative and symmetric, and $d(x,x)=0$. These statements need no continuity or injectivity assumption on h.
- Applying the real triangle inequality to the three image values gives $d(x,z)\le d(x,y)+d(y,z)$. Thus d always has the triangle inequality. The remaining question for a metric is whether distinct inputs can have distance zero.
For any map $h:X\to\mathbb R$, the function $d(x,y)=|h(x)-h(y)|$ satisfies the triangle inequality, even if h is not injective.
Difficulty 1/5 · 100 coins + a small bonus on first completion today
True. The ordinary real triangle inequality applies to $h(x),h(y),h(z)$. Lack of injectivity can fail separation of distinct inputs, not the triangle inequality.
Injectivity makes it a metric and an isometry
- If h is injective, $d(x,y)=0$ forces $h(x)=h(y)$ and hence $x=y$. Conversely, a collision of distinct inputs violates this metric axiom. For $h(x)=x^2$ on $\mathbb R$, the distance between $1$ and $-1$ is zero: this is a pseudometric, not a metric.
- With injective h, the map is an isometry 等距映射 from $(X,d)$ onto its image with ordinary real distance. It is a bijection onto that image and preserves each distance by definition. It need not cover all of $\mathbb R$.
For $h(x) = x^2$, $d(x,y) = |x^2 - y^2|$ is a metric on $\mathbb R$.
Difficulty 1/5 · 100 coins + a small bonus on first completion today
False. The map $x\mapsto x^2$ is not injective on $\mathbb R$: $d(1,-1)=0$ even though the inputs differ. Nonnegativity, symmetry and the triangle inequality still hold.
Cauchy and convergence are different tests
- A sequence is Cauchy if, for every $\varepsilon>0$, some N makes $d(x_m,x_n)<\varepsilon$ for all $m,n\ge N$. A metric space is complete 完备的 when every such sequence has a limit belonging to that space.
- Convergence to p means $d(x_n,p)\to0$. Every convergent metric sequence is Cauchy, by the triangle inequality through p. The converse requires completeness; finding a limit outside the space does not meet the definition.
Transfer both tests to the image
- Since the distances are identical, $(x_n)$ is d-Cauchy exactly when $(h(x_n))$ is Cauchy in ordinary real distance. Likewise, $x_n\to x$ in d exactly when $h(x_n)\to h(x)$ in the image.
- If the image is complete, take its limit y and pull back to $h^{-1}(y)$. Conversely, pull any image Cauchy sequence back, use completeness of $(X,d)$, then map its limit forward. Thus, after checking injectivity, completeness is equivalent on the two spaces.
Closed in the real line is the relevant condition
- For $S\subseteq\mathbb R$ with ordinary distance, S is complete exactly when S is closed in $\mathbb R$. If S is closed, a Cauchy sequence has a real limit by completeness of $\mathbb R$, and closedness keeps that limit in S.
- Conversely, if a real limit point y is missing from S, choose $s_n\in S$ with $|s_n-y|<1/n$. This is a Cauchy sequence, but it cannot have an S-limit: uniqueness of real limits would force that limit to be y. So S is incomplete.
For injective h, knowing only that $h(X)$ is closed as a subset of itself proves that the pullback metric is complete.
Difficulty 1/5 · 100 coins + a small bonus on first completion today
False. Every space is closed in itself. The criterion for a real image requires closedness in $\mathbb R$ with ordinary distance; the arctangent image is a counterexample.
On $X=(0,\infty)$ define $d(x,y)=|\ln x-\ln y|$. Is this a complete metric? Is $x_n=1/n$ d-Cauchy? Justify both answers.
Check the domain, image, ambient space and quantified neighbourhood or tail condition against the model.
Prove the arctangent sequence is Cauchy
- For positive integers $m,n\ge N$, both arctangent values lie between $\arctan N$ and $\pi/2$. Put $\delta_N=\pi/2-\arctan N$. Their gap is less than $\delta_N=\arctan(1/N)<1/N$. Choosing $N>1/\varepsilon$ proves the Cauchy condition.
- But $\arctan n\to\pi/2$, outside the image $(-\pi/2,\pi/2)$. For any real candidate L, the distance tends to the positive number $\pi/2-\arctan L$, so no d-limit exists. Adding two new image endpoints gives the complete interval $[-\pi/2,\pi/2]$, a completion of this metric. The original figure also previews a divisor-closure example.
Under $d(x,y)=|\arctan x-\arctan y|$ on $\mathbb R$, the sequence $x_n=n$ for positive integers n is a ____ sequence with no d-limit in $\mathbb R$.
Difficulty 1/5 · 100 coins + a small bonus on first completion today
The arctangent images have tail gaps below $1/N$ for indices at least N, so the sequence is Cauchy. Their real limit $\pi/2$ is outside the image, so no real input is its d-limit.
The arctangent tail proof bounds every gap for $m,n\ge N$ by $1/N$. What is the smallest positive integer N satisfying $1/N<0.01$?
Difficulty 2/5 · 140 coins + a small bonus on first completion today
The strict inequality requires $N>100$. Thus the smallest integer is 101, not 100.
Work all four original completeness options
- On $X=\mathbb R$, $h(x)=2x+1$ has image $\mathbb R$, so its pullback metric is complete. The arctangent image is open and missing its endpoints. The injective map $x/(1+|x|)$ has image $(-1,1)$, also incomplete.
- For $h(x)=e^x$, the image is $(0,\infty)$ and misses zero. The input sequence $x_n=-n$ has image $e^{-n}\to0$, so it is d-Cauchy with no d-limit. Large positive inputs are not the witness for this missing endpoint.
On $X=\mathbb R$, which h gives a complete metric $d(x,y)=|h(x)-h(y)|$?
Difficulty 1/5 · 100 coins + a small bonus on first completion today
All four maps are injective. The affine map has image $\mathbb R$, closed in the real line. The others have images $(-\pi/2,\pi/2)$, $(0,\infty)$ and $(-1,1)$, each missing a real limit point.
For $d(x,y)=|e^x-e^y|$ on $\mathbb R$, which sequence witnesses incompleteness by approaching the missing image endpoint zero?
Difficulty 1/5 · 100 coins + a small bonus on first completion today
For $x_n=-n$, the images $e^{-n}$ tend to zero, a missing point of $(0,\infty)$. The input sequence is d-Cauchy but has no d-limit.
The same topology can have different completeness
- The maps $\arctan:\mathbb R\to(-\pi/2,\pi/2)$ and $x^3:\mathbb R\to\mathbb R$ are homeomorphisms for the usual topologies: each is continuous, bijective onto its image, and has a continuous inverse. Their pullback balls therefore generate the usual topology on the input line.
- Yet the cubic pullback metric is complete and the arctangent pullback metric is not. A homeomorphism preserves open sets but need not preserve which sequences are Cauchy. Completeness belongs to the chosen metric, not merely to its open sets.
The domain changes the image
- Restrict arctangent to $X=[0,1]$: its image is the closed interval $[0,\pi/4]$, so the pullback metric is complete. Restrict the exponential map to $X=[0,\infty)$: its image is $[1,\infty)$, also complete, although unbounded.
- Every space is closed as a subset of itself. That observation cannot replace closedness in the ambient real line: $[0,1)$ is closed in itself but incomplete with ordinary distance. Check the actual image and the ambient space used in the criterion.
Complete pullback metric on the real line?
For every item the domain is all real numbers and d(x,y)=|h(x)-h(y)|. Check injectivity before image closedness.
Which specified map and input domain give a complete pullback metric?
Difficulty 1/5 · 100 coins + a small bonus on first completion today
The restricted exponential map is injective with image $[1,\infty)$, closed in the real line. The first two images miss limit points; the last map is not injective.
On the same set $X=[0,1)$ compare ordinary distance with the pullback from $h(x)=x/(1-x)$. Decide completeness for each and test $x_n=1-1/n$.
Check the domain, image, ambient space and quantified neighbourhood or tail condition against the model.
Verify a family before calling it a basis
- A family of subsets is a basis for a topology if it covers X and has this refinement property: whenever x belongs to two family members, some family member contains x and lies inside their intersection. The open sets are unions of these members, including the empty union.
- A basic neighbourhood of x is a member containing x. Every open set containing x must contain one of these basic neighbourhoods. This basis is a family of sets, distinct from a vector-space basis in linear algebra.
Closure tests neighbourhoods of the candidate
- A point x belongs to $\overline A$ exactly when every basic neighbourhood of x meets A. Indeed, every open neighbourhood contains a basic one. Testing only neighbourhoods of points already in A answers a different question.
- To include x, justify all its basic neighbourhoods. To exclude x, one basic neighbourhood disjoint from A is enough. If a smallest neighbourhood exists, it can reduce this universal test to one set.
The divisor sets really form a basis
- Work on the infinite set $X=\{2,3,4,\ldots\}$. For each integer $k\ge2$, let $U_k=\{n\in X:n\mid k\}$. The family covers X because $x\in U_x$ for every x in X.
- If $g=\gcd(j,k)\ge2$, then $U_j\cap U_k=U_g$. If $g=1$, the intersection is empty: the divisor 1 is not in X. Thus every point in an intersection has a basic refinement, so this family defines a topology.
On $X=\{2,3,4,\ldots\}$ let $U_k=\{n\in X:n\mid k\}$. How many elements are in $U_{12}\cap U_{18}$?
Difficulty 2/5 · 140 coins + a small bonus on first completion today
The common divisors in X are $\{2,3,6\}=U_6$, since $\gcd(12,18)=6$. The divisor 1 is not in X.
Find the smallest neighbourhood and count it
- If $x\in U_k$, then $x\mid k$. Every divisor of x also divides k, so $U_x\subseteq U_k$. Since $U_x$ itself contains x, it is the smallest basic neighbourhood, and is contained in every open neighbourhood of x.
- For the original count, $U_{12}=\{2,3,4,6,12\}$ has five elements. Do not insert 1. Also $U_4=\{2,4\}$ and $U_{16}=\{2,4,8,16\}$; these are sets of divisors, not sets of multiples.
For $X=\{2,3,4,\ldots\}$ and $U_k=\{n\in X:n\mid k\}$, how many elements does $U_{12}$ have?
Difficulty 2/5 · 140 coins + a small bonus on first completion today
$U_{12}=\{2,3,4,6,12\}$ has five elements. The divisor 1 is excluded because it does not belong to X.
Derive the closure rule for any set
- The smallest-neighbourhood test gives $x\in\overline A$ exactly when $U_x\cap A\ne\varnothing$. Equivalently, some $a\in A$ divides x. Thus the closure of a singleton $\{a\}$ is all multiples of a that belong to X.
- In the original example, $16\in\overline{\{8\}}$ because $8\mid16$. But $4$ is excluded: $U_4$ misses 8. Every basic neighbourhood of 16 contains $U_{16}$ and hence contains 8, so the inclusion checks all neighbourhoods, not only one chosen set.
Work the original closure and extend it
- For $A=\{6\}$, closure means divisibility by 6. The original candidates $6,12,18$ are included, while $3,20$ are excluded. For $A=\{4,6\}$, a candidate belongs when it is divisible by 4 or by 6: 12 is included and 10 is excluded.
- For the infinite set of primes, every integer in X has a prime divisor, so its closure is all of X. For $A=\varnothing$, no divisor witness exists and the closure is empty. The rule does not depend on A being finite.
On $X=\{2,3,4,\ldots\}$ use the basis $U_k=\{n\in X:n\mid k\}$. Which listed candidates lie in the closure of $\{6\}$?
Difficulty 2/5 · 140 coins + a small bonus on first completion today
A candidate x lies in the closure exactly when its smallest neighbourhood $U_x$ contains 6, which means $6\mid x$. Thus the listed members are 6, 12 and 18.
Use the divisor topology on $X=\{2,3,4,\ldots\}$. For $A=\{9,10\}$, determine which candidates $3,9,18,20,30,45$ belong to the closure, and give a neighbourhood witness for one excluded candidate.
Check the domain, image, ambient space and quantified neighbourhood or tail condition against the model.
Interior and limits also depend on these neighbourhoods
- A point x is in the interior of A exactly when $U_x\subseteq A$. For the even integers, 8 is interior because $U_8=\{2,4,8\}$; 6 is not, because $3\in U_6$. In fact the interior consists of powers of 2, while the closure is the set of even integers.
- The constant sequence $x_n=2$ converges to every even candidate x: each neighbourhood of x contains $U_x$ and hence contains 2. It does not converge to 3, since $U_3=\{3\}$. Unlike a metric topology, this topology permits distinct limits for the same sequence.
Which neighbourhood test succeeds?
Use X={2,3,...} with basic divisor sets Uk. Every item names whether it asks about closure or interior.
In the divisor topology on $X=\{2,3,4,\ldots\}$, which listed point is interior to the set of even integers?
Difficulty 1/5 · 100 coins + a small bonus on first completion today
$U_8=\{2,4,8\}$ consists entirely of even integers. Each other listed candidate has an odd divisor greater than 1 in its smallest neighbourhood.
In the divisor topology on $X=\{2,3,4,\ldots\}$, find the interior of $A=\{2,3\}$. Does the constant sequence 2 converge to 4, 6 or 3? Can this topology arise from a metric?
Check the domain, image, ambient space and quantified neighbourhood or tail condition against the model.
Recap
- Check injectivity before calling a pullback distance a metric. Transfer Cauchy sequences and their possible limits to the image, then test whether that image is closed in the real line. A limit outside the image does not establish completeness.
- Check basis axioms before using basic neighbourhoods. For the divisor topology, the smallest neighbourhood of x is its divisor set: closure requires some member of A to divide x, while interior requires every member of that neighbourhood to lie in A.
- basis: A family of open sets such that every open set is a union of its members.
- isometry: A map that preserves all distances between pairs of points.
- complete: Describing a metric space in which every Cauchy sequence converges to a point of that space.