10.2. INVARIANCE OF DOMAIN 219

Proof: It suffices to show that if p ∈ U then f(p) is an interior point of f(U). LetB(p,r) ⊆ U. By Lemma 10.2.3, f(U) ⊇ f

(B(p,r)

)⊇ B(f(p) ,δ ) so f(p) is indeed an

interior point of f(U).The inverse mapping theorem assumed quite a bit about the mapping. In particular it

assumed that the mapping had a continuous derivative. The following version of the inversefunction theorem seems very interesting because it only needs an invertible derivative at apoint.

Corollary 10.2.5 Let U be an open set in Rp and let f : U →Rp be one to one and contin-uous. Then, f−1 is also continuous on the open set f(U). If f is differentiable at x1 ∈U andif Df(x1)

−1 exists for x1 ∈U, then it follows that Df(f(x1)) = Df(x1)−1.

Proof: |·| will be a norm on Rp, whichever is desired. If you like, let it be the Euclideannorm. ∥·∥ will be the operator norm. The first part of the conclusion of this corollary isfrom invariance of domain. From the assumption that Df(x1) and Df(x1)

−1 exists,

y− f(x1) = f(f−1 (y)

)− f(x1) = Df(x1)

(f−1 (y)−x1

)+o(f−1 (y)−x1

)Since Df(x1)

−1 exists,

Df(x1)−1 (y− f(x1)) = f−1 (y)−x1 +o

(f−1 (y)−x1

)by continuity, if |y− f(x1)| is small enough, then

∣∣f−1 (y)−x1∣∣ is small enough that in the

above, ∣∣o(f−1 (y)−x1)∣∣< 1

2

∣∣f−1 (y)−x1∣∣

Hence, if |y− f(x1)| is sufficiently small, then from the triangle inequality of the form|p−q| ≥ ||p|− |q|| ,∥∥∥Df(x1)

−1∥∥∥ |(y− f(x1))| ≥

∣∣∣Df(x1)−1 (y− f(x1))

∣∣∣≥

∣∣f−1 (y)−x1∣∣− 1

2

∣∣f−1 (y)−x1∣∣

=12

∣∣f−1 (y)−x1∣∣

|y− f(x1)| ≥∥∥∥Df(x1)

−1∥∥∥−1 1

2

∣∣f−1 (y)−x1∣∣

It follows that for |y− f(x1)| small enough,∣∣∣∣∣o(f−1 (y)−x1

)y− f(x1)

∣∣∣∣∣≤∣∣∣∣∣o(f−1 (y)−x1

)f−1 (y)−x1

∣∣∣∣∣ 2∥∥∥Df(x1)−1∥∥∥−1

Then, using continuity of the inverse function again, it follows that if |y− f(x1)| is possiblystill smaller, then f−1 (y)−x1 is sufficiently small that the right side of the above inequalityis no larger than ε . Since ε is arbitrary, it follows

o(f−1 (y)−x1

)= o(y− f(x1))

10.2. INVARIANCE OF DOMAIN 219Proof: It suffices to show that if p € U then f(p) is an interior point of f(U). LetB(p,r) CU. By Lemma 10.2.3, f(U) D t(B(p.7)) > B(f(p),6) so f(p) is indeed aninterior point of f(U). flThe inverse mapping theorem assumed quite a bit about the mapping. In particular itassumed that the mapping had a continuous derivative. The following version of the inversefunction theorem seems very interesting because it only needs an invertible derivative at apoint.Corollary 10.2.5 Let U be an open set in R? and let f: U — R? be one to one and contin-uous. Then, f—! is also continuous on the open set f(U). If f is differentiable at x, € U andif Df(x,)~| exists for x, € U, then it follows that Df (f (x,)) = Df (x;)!.Proof: |-| will be a norm on R’, whichever is desired. If you like, let it be the Euclideannorm. ||-|| will be the operator norm. The first part of the conclusion of this corollary isfrom invariance of domain. From the assumption that Df (x;) and Df(x,)~' exists,y—f(x1) =f(f ' (y)) —f(x1) = Df (x1) (f(y) —x1) +0(f ! (y)—x1)Since Df (x,)~! exists,DE (x1) ' (y—f(x1)) =f! (y) —x1 +0 (fF! (y) x1)by continuity, if |y — f(x, )| is small enough, then |f- 'iy)-x, | is small enough that in theabove,lo(f"(y) —x1) |< 5 |) —™1]Hence, if |y—f(x;)| is sufficiently small, then from the triangle inequality of the formIp—4| = |lp|— lal,pr) "|| (y -£0x1))IV[pe xi) | (y—f(x1))]> |e y)—mf—5 Py) —m|= si 1) —xi|y—f(ai)] > ote) | 5 [et (9) —x)It follows that for |y — f(x, )| small enough,o(f-' (y) —x1) o(f-! (y) —x1)y—f(x1) f(y) —x2[orsThen, using continuity of the inverse function again, it follows that if |y — f(x,)| is possiblystill smaller, then f~' (y) — x; is sufficiently small that the right side of the above inequalityis no larger than €. Since € is arbitrary, it followso(f! (y) —x1) =0(y—f(x1))