29.9. THE COAREA FORMULA 1043

det[(

Dy2gi (y)∗ Dy2gic (y)∗ )( Dy2gi (y)

Dy2gic (y)

)]1/2

= det(Dy2g(y)∗Dy2g(y)

)1/2

Therefore, 29.9.53 reduces to∫

Kij∩A det

(Df(x)Df(x)∗

)1/2 dx =

∫Rm

∫f−1(y1)∩Ki

j∩Adet(Dy2g(y)∗Dy2g(y)

)1/2 dy2dy1. (29.9.57)

Then z ∈ g(

y1, f−1 (y1)∩Kij ∩A

)if and only if

fi (z) =(

f(z)zic

)∈(

y1f−1 (y1)∩Ki

j ∩A

)if and only if z ∈ f−1 (y1) and zic ∈ f−1 (y1)∩Ki

j ∩ A. Letting ĝ be the function y2 →g(y1,y2) , this shows that z ∈ ĝ

(f−1 (y1)∩Ki

j ∩A)

if and only if y2 = zic ∈ f−1 (y1)∩Kij∩A

and so ĝ(

f−1 (y1)∩Kij ∩A

)= f−1 (y1)∩Ki

j ∩A. Of course ĝ actually depends on y1 butthis is suppressed here. Therefore,

g(

y1, f−1 (y1)∩Kij ∩A

)= f−1 (y1)∩Ki

j ∩A

By this observation and the area formula, the equations 29.9.53, 29.9.57 imply∫Ki

j∩Adet(Df(x)Df(x)∗

)1/2 dx =∫Rm

H n−m(

f−1 (y1)∩Kij ∩A

)dy1.

Using Lemmas 29.9.6 and 29.9.5, along with the inner regularity of Lebesgue measure, Kij

can be replaced with E ij. Therefore, summing the terms over all i and j,∫

Adet(Df(x)Df(x)∗

)1/2 dx =∫Rm

H n−m (f−1 (y)∩A)

dy

which verifies the coarea formula whenever A is a closed subset of Rn \{S∪N} .By Lemma 29.9.6 again, this formula is true for all A a closed subset of Rn \S. Using

the same two lemmas again, we see this coarea formula holds for all A a measurable subsetof Rn \S.

It remains to verify the formula for all measurable sets A, regardless of whether theyintersect S. Recall

S≡

{x : ∑

idet(

Dfi (x))2

= 0

}= {x : detU (x)≡ J∗ (x) = 0} .

Consider the case where A ⊆ S. Let A be compact so that by Lemma 29.9.5, y →H n−m

(A∩ f−1 (y)

)is Borel measurable. For ε > 0, define kε , p : Rn×Rm → Rm by

kε (x,z)≡ f(x)+ εz, p(x,z)≡ z.

29.9. THE COAREA FORMULA 1043; 1/2det ( Dyr8i(y)” Dy, Bi. (¥)” )( pet )| = det (Dy,8(y)" Dy,@(y))Therefore, 29.9.53 reduces to fyi,,, det (Df (x) Df (x)*) 2 Oy =Jdet (Dy,g(y)*D 1? dyydyy. 29.9.57hea besgoas (Dy,8(y)" Dy8(y)) © dyzdy1 ( )Then z€g (vif (yi) OK} nA) if and only ifij, ( f(z) yif@)= ( Zi. ) c ( f-! (vi)OKVNAif and only if z€f-!(y;) and x, € f! (y:) NKi A. Letting & be the function yz >£(y1,y2), this shows that z € & ( (y1) NK} nA) if and only if yo =2;, ef! (yi)Ki NAand so & ( (yi) NK} nA) =f! (y,) NK A. Of course § actually depends on y; butthis is suppressed here. Therefore,g (vit! (yi) Ki nA) =f! (y,)MKinABy this observation and the area formula, the equations 29.9.53, 29.9.57 implyfom KO"-™ (r (yi) nKi nA) dy.I det (Df (x) Df (x)") "7 dx =KinaUsing Lemmas 29.9.6 and 29.9.5, along with the inner regularity of Lebesgue measure, Kican be replaced with Ek. Therefore, summing the terms over all i and /,[eee(ot(o de(x)")' a= | KH" (f! (y) MA) dymwhich verifies the coarea formula whenever A is a closed subset of R” \ {SUN}.By Lemma 29.9.6 again, this formula is true for all A a closed subset of R” \ S. Usingthe same two lemmas again, we see this coarea formula holds for all A a measurable subsetof R"\S.It remains to verify the formula for all measurable sets A, regardless of whether theyintersect S. RecallS= . ; yi det (pf (x)) -0| = {x: detU (x) =J* (x) =0}.Consider the case where A C S. Let A be compact so that by Lemma 29.9.5, y >worm (A nf! (y)) is Borel measurable. For € > 0, define ke, p : R” x R” — R” byke (x,z) =f (x) + €z, p(x,z) =z.