464 CHAPTER 17. THE AREA FORMULA

Proof: By assumption that det(Df (x)Df (x)∗

)= ∑i∈Λ(n,m) det

(Df i (x)

)2> 0, we

can let Ai ≡ ∩j ̸=i

{x : Df j (x) = 0

}.■

Maybe some of these Ai are /0 but this will not matter.Suppose f i is one to one on a Borel set Ei ⊆ Rn which has positive measure and that

its inverse, denoted as gi is also Lipschitz on f i(Ei)

and Df i is invertible on Ei. Thus,for x ∈ Ei∩A,(

y1y2

)= y= f i (x) =

(f (x)xic

), giic

(f i (x)

)= y2 =xic , y1 = f

(gi (y)

)(17.18)

Differentiate y1 = f(gi (y)

)with respect to y2 to obtain

0 = Dxif(gi (y)

)Dy2g

ii (y)+Dxic

f(gi (y)

)Dy2g

iic (y)

= Dxif(gi (y)

)Dy2g

ii (y)+Dxic

f(gi (y)

). (17.19)

Also,Df i

(gi (y)

)Dgi (y) = I,

∣∣det(Dgi (y)

)∣∣= ∣∣detDf i(gi (y)

)∣∣−1

Say y = (y1,y2)T and suppose z = (zi,zic)

T ∈ f−1 (y1)∩Ei∩A. Then

f i

(zizic

)=

(y1y2

)=

(f (z)zic

), so

(zizic

)= gi

(y1y2

)Thus,

gi(f−1 (y1)∩Ei∩A

)= f−1 (y1)∩Ei∩A (17.20)

so if we fix y1, then y2→ gi (y1,y2) gives a parametrization for the Borel set f−1 (y1)∩Ei∩A and Dy2g

iic(y) = I.

Now from the area formula,∫Ei∩A

det(Df i (x)Df i (x)∗

)1/2dx = (17.21)

=∫fi(Ei∩A)

det(

Df i(gi (y)

)Df i

(gi (y)

)∗)1/2∣∣detDf i(gi (y)

)∣∣−1dy (17.22)

Letting y ≡ (y1,y2) , and using what was just shown about y2 → gi (y1,y2) being aparametrization, the above integral can be expressed as the following iterated integral:∫

Rm

∫f−1(y1)∩Ei∩A

det(

Dfxi

(gi)

Dfxi

(gi)∗)1/2∣∣detDfxi

(gi)∣∣−1

(y1,y2)dy2dy1,

(17.23)Therefore, the inner integral is measurable and the integrand is

det[(

Dxif(gi (y)

)Dxic

f(gi (y)

))( Dxif(gi (y)

)∗Dxic

f(gi (y)

)∗ )]1/2 ∣∣detDfxi

(gi (y)

)∣∣−1.

(17.24)Let A≡ Dxi

f(gi (y)

)so A is m×m, B≡ Dy2g

ii (y) an m× (n−m) . Using 17.19, 17.24

is of the form

det[(

A −AB)( A∗

−B∗A∗

)]1/2

|detA|−1

= det [AA∗+ABB∗A∗]1/2 |detA|−1

= det [A(I +BB∗)A∗]1/2 |detA|−1 = det(I +BB∗)1/2

464 CHAPTER 17. THE AREA FORMULAProof: By assumption that det (Df (a) Df (@)*) = Dica(nm) det (Df* (x))” > 0, wecan let A; = jx; {a : Df) (x) =0}.8Maybe some of these A; are 9 but this will not matter.Suppose f* is one to one on a Borel set E* C R” which has positive measure and thatits inverse, denoted as g* is also Lipschitz on f* (E*) and Df? is invertible on E*. Thus,forx Ee E*NA,(% )-u-Fa=( re) ) ah @) == 0. n =F (9'w)) (17.18)Yo Ti,Differentiate y, = f (g* (y)) with respect to y> to obtain0 = Dz, f (9*(y)) Dy gi (y) + De, Ff (9° (Y)) Pug. (Y)= Dz,f (9° (y)) Dyrgi (y) +Da,, f (9° (y)). (17.19)Also,Df? (9! (y)) Dg* (y) =1, |det (Dg* (y)) | = |detD Fé (9! (y))|Say y =(y1,y>)/ and suppose z = (23,24,)’ € f-! (y;) NE*NA. Then(2 )-G)-C2) 9 (eG)9 (fF '(yi)NE*NA) =f '(yy)NE*NA (17.20)so if we fix y,, then yy + g*(y),y>) gives a parametrization for the Borel set f~' (y,)NE* A and Dy, gj, (y) =I.Now from the area formula,I, det (Df? (x) Df* (x)*) "dx = (17.21)E*nAThus,= | eingy (PF (9!) PF (9 w)") er #*(9*(w)) ay 07.22Letting y = (y;,y>), and using what was just shown about y, + g*(y),y2) being aparametrization, the above integral can be expressed as the following iterated integral:. ge \ 1/2 |det (Df,. (g*) Df». (g' detDf,,. (g’ Yo) dyrdy,Lode reg ynwira et (Pfs (9*) Pha: (9)") Jeet Fo, (9°) (ita) dado(17.23)Therefore, the inner integral is measurable and the integrand isbye \ 71/2 :det|(Da.f (9'(u)) Pa f(a'(w))) ( pei). \| aero, (a'w))|° (17.24)Let A = Dz, f (g* (y)) so A is mx m, B = Dy, gi (y) an m x (n—m). Using 17.19, 17.24is of the formdet|( A -aB )( A’ )] aera—B*A*det [AA* + ABB*A*]!? |detA|7!= det[A(1+BB*)A*]'/? |detA|~! = det (J + BB*)!/?