Prerequisites: compactness and uniform continuity, subsequences and compact images, and relative-domain limits.
An injective function lets us recover an input from an exact output. Continuity of the inverse asks whether nearby outputs recover nearby inputs. These are different statements. Compactness supplies a useful bridge: a continuous injection on a compact subset of the real line has a continuous inverse on its image, and that inverse is uniformly continuous. The domain need not be an interval.
Revision roadmap
Distinguish injectivity from inverse continuity. Prove the theorem by subsequences, then use paired subsequences for a positive separation bound. Compare a disconnected compact example with the noncompact counterexample.
Definitions and theorem checklist
Let be nonempty and compact, let be continuous and injective, and put . The inverse exists because is bijective onto its image. Continuity statements on and are relative to those sets; neither is assumed to be an interval.
Definition 6.1.1
(Separated pairs and the separation number)
For , the set of -separated input pairs is
If , its separation number is
It is the smallest output distance that two inputs at distance at least can have. If , all input pairs are already closer than .
Uniform continuity of asks for one threshold for each tolerance on all of at once. A Lipschitz inverse would need the stronger estimate with a single finite constant . Two facts used throughout: every sequence in compact has a subsequence converging to a point of , and a continuous image of a compact set is compact.
An inverse exists before its continuity is proved
Theorem 6.1.2
(Continuity and uniform continuity of the inverse)
If is nonempty and compact and is continuous and injective, then is continuous, and in fact uniformly continuous.
Proof
(Contradiction by a subsequence in the compact domain)
Injectivity gives a unique preimage for every point of . This defines , but says nothing yet about a limit in its domain. Fix . Suppose were not continuous there. The negation of relative continuity gives and, for each positive integer , a point with
Set . Compactness extracts a subsequence . Continuity of gives
Injectivity forces , contradicting the fixed separation of the subsequence. Thus is continuous at every point of . Since is a continuous image of compact , it is compact, and the Heine–Cantor theorem gives uniform continuity of .
The compactness belongs to the input set of the original function, not merely to its output set. A noncompact input sequence might approach a missing point or escape to infinity. The subsequence argument would then have no permitted limit in the original domain.
A separation estimate gives a direct uniform proof
The same idea, applied to pairs, produces an explicit threshold.
Proposition 6.1.3
(Separation bound)
Let and be as above, and let with . Then is attained by some pair in and . Consequently, for ,
If , every positive works for the tolerance .
Proof
(Paired minimizing sequence and compactness)
Write . The output distances over are nonnegative and finite, so the infimum exists. Choose paired points with
First extract a convergent subsequence of . Then extract a further convergent subsequence of the matching . The corresponding subsequence still converges. Write the two limits as .
Continuity of the absolute difference and of gives and . Thus the infimum is attained by a pair in . If , injectivity would give , contradicting . Hence .
Now take with . If , then , so , a contradiction. Therefore , which is uniform continuity with .
Proof X-ray
Where each hypothesis acts in the separation proof
Read the proof as four moves, each with one hypothesis behind it.
- Compactness of is used twice: once for and once, on the already thinned index set, for . Both extractions must keep the same indices, because the two inequalities and speak about the pair . Extracting the two subsequences independently would destroy the pairing and with it the information that the output distances approach .
- Continuity of and of carries both inequalities to the limit pair. The set is defined with a non-strict inequality, so the limit pair stays in , and the infimum is attained there.
- Injectivity is used exactly once, to turn into . This is the step that fails for on : the pair lies in and has output distance , so .
- The proof closes by contraposition: an input pair that is far apart lies in , so its outputs are at least apart. Nothing about intervals or connectedness was needed.
Because the threshold is an infimum over a compact set of pairs, it depends only on and , not on the points .
A compact sequence set with an explicit inverse
Worked example 6.1.4
(An isolated-point compact set)
Let and define on . This set is closed and bounded: its only accumulation point is zero, which it contains. Hence it is compact. The rational formula is continuous on a neighborhood of , and it is injective because
Its image is . Solving the equation gives on ; for example and . The theorem applies even though is not an interval and its nonzero points are isolated.
Here we can obtain an explicit bound. Since ,
This is a valid Lipschitz estimate with constant . The compact inverse theorem guarantees uniform continuity, not necessarily a linear modulus. For example, the cube-root function on is uniformly continuous but not Lipschitz at zero.
Quick check
(For on the set above, if two outputs satisfy , what bound on follows?)
Use the Lipschitz estimate with constant .
Solution · Answer
Since , we have , so .
Separate the hypotheses
Compactness does not replace continuity of , and compactness of the codomain does not replace compactness of the domain. The next two items show the typical slips.
Common mistake
(A bijection between compact sets is not automatically continuous)
Take as in the previous example and define by , and for every other . It is a bijection of a compact set onto itself. But for while , so is not continuous at . The inverse theorem needs to be continuous. For continuous bijections of onto itself, the theorem makes them homeomorphisms, and a homeomorphism must send the unique accumulation point to itself.
Conversely, lack of compactness does not force every inverse to fail: the identity map on has a uniformly continuous inverse. Compactness only supplies the general guarantee, and it cannot simply be dropped.
Counterexample mode
A continuous injection on the whole line with a non-uniform inverse
Claim: if is a continuous injection, then its inverse on is uniformly continuous. This is false. Take , which is continuous and strictly increasing, with image and inverse . For the inputs and ,
For this is . The outputs get arbitrarily close while for every , so no single works for . The inverse is continuous but not uniformly continuous. The repair is the compact domain of the theorem: on the same has , and the threshold shrinks to only as .
Quick check
(For on the non-injective domain , what is , and which hypothesis of the separation bound fails?)
Look for two inputs one unit apart with the same square.
Solution · Answer
The inputs and are one unit apart, and both have square , so the output distance is . Hence . Injectivity fails, and the separation bound is no longer positive.
Error in an output becomes error in a recovered input
Suppose an exact output is and an approximate output still belongs to . The uniform inverse estimate controls from . The condition matters: the inverse has only been defined on the actual image, so no arbitrary noisy value outside that set can be substituted without an additional extension rule.
Common mistake
(Substituting a noisy output outside the image)
For the set and the map of the previous section, the output is not in . The formula returns , which is not an element of and recovers no input. The Lipschitz estimate was proved only for , so it says nothing about such a .
An existence proof for gives a threshold without necessarily giving a simple formula. The cube shows a case where it can be computed.
Worked example 6.1.5
(The separation number of the cube)
Let on and . For ordered inputs , put and , so that and . Then
which uses no differentiation. For a pair in we have , so . Equality occurs at and , which lie in because . Hence . For , no two points of are apart, and is empty.
As a numerical check, take . Then . If two outputs in differ by less than , their cube roots differ by less than . The bound is sharp: the inputs have outputs , which are exactly apart, while the inputs are exactly apart.
Quick check
(For on , find .)
For write .
Solution · Answer
Let with . Then , with equality exactly when and . The pair lies in and , so .