Malkiel Shoshan

Geometric Measure Theory: 3 - 4,5

FIRST AND SECOND VARIATION: Let M ⊂ Rn+l be a C 1 , n dimensional submanifold, U open

subset, with U ∩ M 6= φ, H n (K ∩ M ) < ∞, for each compact K ⊂ U . Let {ϕt : U → U }−1≤t≤1 be a
1 parameter family of diffeomorphisms, such that for each compact K ⊂ U

2
ϕt (x) = ϕ(t, x) : (−1, 1) × U → U is C

ϕ0 (x) = x, ∀x ∈ U

ϕt (x) = x, ∀(t, x) ∈ (−1, 1) × U \ K


and let Mt = ϕt (M ∩ K) be the corresponding images.

So we are perturbing the manifold only in some compact subset of U , outside its fixed.
∂ 2 ϕ(t,x)
Let Xx = ∂ϕ(t,x)
∂t |t=0 , Zx = ∂t2 |t=0 , these are the initial velocity and acceleration vectors for ϕt .
So by Taylor
t2
ϕt (x) = x + tXx + Zx + O(t3 )
2
and X and Z are compactly supported in U .

d
We define the first variation as dt H n (Mt )|t=0 , i.e. the instantaneous rate of change of the ”area”i.e.
the Hausdorff measure of M ∩ K at time 0.

d2 n
The second variation dt2 H (Mt )|t=0 . We want to find a formula. From earlier, we have
Z
H n (ϕt (M ∩ K)) = Jψt dH n
M ∩K

where ψt = ϕt |M ∩ U . Things are smooth so we can differentiate under the integral, so we need to
∂ ∂2
find ∂t Jψt |t=0 and ∂t 2 Jψt |t=0 . We have for fixed x, and τ ∈ Tx M ,

t2
dψt (τ ) = Dτ ψt = τ + tDτ X + Dτ Z + O(t3 )
2
So relative to the natural orthonormal bases of Tx M and Rn+l , the matrix of the map dψt : Tx M →
Rn+l has ith column (because we are letting dψt act on the ith basis vector, in the τi basis it looks
like (0, ..., 1, 0, ..., 0))

t2
Dτi ψt = τi + tDτi X + Dτ Z + O(t3 )
2 i
So the matrix of dψt∗ ◦ dψt has entry bij

bij = Dτi ψt · Dτj ψt = δij + t(τi · Dτj X + τj · Dτi X)

1
+t2 (τi · Dτj Z + τj · Dτi Z) + (Dτi X) · (Dτj X) + O(t3 )

2
= δij + aij .
So using
1 1
det(I + A) = 1 + traceA + (traceA)2 − trace(A2 ) + O(|A|3 )
2 2
and
√ 1 1
1 + x = 1 + x − x2 + O(t3 )
2 8
n
t2 h X
Jψt = 1 + tdivM X + 2
divM Z + (divM X) + |(Dτi X)⊥ |2
2 i=1

1
 X i
− (τi · Dτj X)(τj · Dτi X) + O(t3 ).
i,j=1

So taking derivatives at t = 0 we get

Z
d n
H (Mt )|t=0 = divM XdH n
dt M
n
d2 n
Z X
H (M t )| t=0 = (div M Z + (div M X) 2
+ |(Dτi X)⊥ |2
dt2 M i=1
X
n
− (τi · Dτj X)(τj · Dτi X)dH .
i,j=1
n
We say M Ris stationary in U if H (M ∩ K) < ∞ for each compact K ⊂ U and if the first variation
is zero. So M divM XdH n = 0. Then by the integration by parts we have

Lemma: 1. If M is a C 2 submanifold of Rn+l , U as above with (n − 1) dimensional boundary ∂M ,

and let X be any vector field with compact support in U , then M is stationary if and only if H = 0
on M ∩ U and ∂M ∩ U = φ.

Since the boundary

R is not in U , and X is compactly supported in U , the boundary term is zero, so
we are left with M X · HdH n = 0, so H = 0.

We also talk about the case where M ⊂ N , and N is a C 2 , n + l dimensional submanifold of Rn+L ,
where l ≤ L (e.g. dimM = n = 1, l = 1, L = 2, dimN = 2), and U ⊂ Rn+L is open such that
(N \ N ) ∩ U = φ, and let {ϕt } be as before, with ϕt (U ∩ N ) ⊂ U ∩ N for compact K ⊂ U , then again
we say:

d
M is stationary in U ∩ N , if dt Hn (ϕt (M ∩ K)) = 0.

For any C 1 vector field X in U with compact support K, and Xx ∈ Tx N for x ∈ N , there is a 1
∂
parameter family ϕt as above with ∂t ϕ(x, t)|t=0 = Xx at each x ∈ N ∩ U , so M is stationary in U ∩ N
if and only if
Z
divM X = 0.
M
1 L
Now let v (z), · · · , v (z) be an orthonormal family (defined near y ∈ M ) of vector fields normal to
M , think normal bundle, with v 1 , · · · , v l tangent to N and v l+1 , ..., v L normal to N (PIC), then for
PL
any vector field X on M we can write X = X T + X ⊥ , where XzT ∈ Tz N and X ⊥ = j=l+1 (v j · X)v j .

Then for τ1 , ..., τn orthonormal basis for Ty M , we have

L
X
divM X = divM X T + (v j · X)divM v j
j=l+1

L
X n
X
= divM X T + (v j · X) τi · Dτi v j
j=l+1 i=1
n
X
= divM X T − X · By (τi , τi )
i=1
Pn
where By is the second fundamental form of N at y. We also let HM = i=1 By (τi , τi ) (not exactly
the mean curvature since By is the second fundamental form on N ).

2
So, after integrating, if M is stationary in U ∩ N , then by the above, divM X T = 0, we get
Z Z
divM X = − HM · X.
M M
RECTIFIABLE SETS:

We want to extend the class of sets that we consider to include rough surfaces with singular points.

Definition: A set M ⊂ Rn+l is countable n rectifiable if

M ⊂ M0 ∪ (∪∞ n
j=1 Fj (R ))

where H n (M0 ) = 0, Fj : Rn → Rn+l are Lipschitz.

This is equivalent to Aj ⊂ Rn by extension.

Lemma: M is countable n rectifiable if and only if M ⊂ ∪∞ n

j=0 Nj where H (N0 ) = 0 and each Nj is
1 n+l
an n dimensional embedded C submanifold of R .

Proof: The if part is trivial because the local coordinate maps are Lipschitz.

We recall the C 1 Approximation Theorem: Let f : Rn → R be Lipschitz. For ε > 0 there is a

g ∈ C 1 (Rn ) such that

Ln ({x : f (x) 6= g(x)}) < ε.

Thus, for each Fj we can choose C 1 functions gij : Rn → Rn+l such that H n ({x : Fj (x) 6= gij (x)}) <
1/i. Then letting Zj = Rn \ (∪∞ n
i=1 {x : Fj (x) = gij (x)}), we have H (Zj ) = 0, so for Ej = Fj (Zj )

ImFj ⊂ Ej ∪ ((∪∞ n
i=1 gij (R ))

i.e. contained in the set where f 6= g ∪ f = g. We have H n (Ej ) = 0 because Fj is Lipchitz. So

letting
N0 = ∪∞
j=1 Ej

we have H n (N0 ) = 0. So

M ⊂ M0 ∪ N0 ∪ (∪i,j gij (Rn )).

We recall that the set of Jgij = 0 has measure 0, so for x where J(x) 6= 0, gij (Bρ (x)) is an n dimen-
sional C 1 submanifold of Rn+l , so ∪ij gij (Rn ) can be written as the union of a set of measure 0 and
countable n dimensional C 1 submanifolds of Rn+l .

Remark: the submanifolds can be taken to be disjoint.

Definition: Let M ⊂ Rn+l be an H n measurable subset with H n (M ∩ K) < ∞ for compact K, then
the n dimensional subspace P ⊂ Rn+l is the approximate tangent space for M at x ∈ Rn+l if
Z Z
lim f (y)dH n (y) = f (y)dH n (y)
λ→0 ηx,λ (M ) P

for all f ∈ Cc0 (Rn+l ), ηx,λ : Rn+l → Rn+l defined by ηx,λ (y) = λ−1 (y − x). i.e. translates and then
rescales (stretches it out). If such a P exists we call it Tx M .

Example: The double parabola.

Remark: By choosing f : Rn+l → [0, 1] with f = 1 on B 1 (0) and f = 0 on Rn+l \ B1+ε (0), we see
H n (M ∩Bρ (x))
that Tx M exists implies that limρ→0 ωn ρn = 1 , because the RHS is the e.g. 3-d ball inter-
secting the plane in a circle so we get ωn ρn . On the LHS we get limλ→0 η(M )∩B = limρ→ 0 M ∩Bρ (x).

3
Given an l dimensional subspace π ⊂ Rn+l and an α ∈ (0, 1) let

Xα (π, x) = {y ∈ Rn+l : qπ (y − x) ≤ α|y − x|}

qπ is orthogonal projection of Rn+l onto π ⊥ .

Letting f approximate the indicator function on X 21 ((Tx M )⊥ , x) ∩ B1 (0) and 0 outside, then we note
that when λ → 0, M ∩ Bρ (x) ∩ X 21 ((Tx M )⊥ , x) is empty, so

H n (M ∩ Bρ (x) ∩ X 21 ((Tx M )⊥ , x))

lim = 0.
ρ→0 ωn ρn
Theorem: Suppose M is H n measurable with H n (M ∩ K) < ∞ for compact K ⊂ Rn+l . Then M
countable n rectifiable if and only if Tx M exists for H n a.e. x ∈ M

Proof: (⇒) We write M = ∪∞ n 1

j=0 Mj where H (M0 ) = 0, Mj ⊂ Nj and Nj are C embedded
submanifolds of dimensions n. For R > 0, f ∈ Cc0 (Rn+l ) with f = 0 on Rn+l \ BR (0), then for
x ∈ M j ⊂ Nj ,
Z Z Z Z
f dH n = f dH n − f dH n + f dH n .
ηx,λ (M ) ηx,λ (Nj ) ηx,λ (Nj \Mj ) ηx,λ (M \Mj )
1
but Nj is a C submanifold
Z Z
n
lim f dH = f dH n .
λ→0 ηx,λ (Nj ) Tx Nj

Also,

Z
f dH n ≤ sup |f |H n (BR (0) ∩ ηx,λ (M \ Mj )) = sup |f |λ−n H n (BλR (x) ∩ M \ Mj )
ηx,λ (M \Mj )

H n (B (x)∩M \Mj )
By the upper density thm lim supρ→0 λR
ωn λm = 0 for H n a.e. x not in Mj . Similarly, the
second term goes to zero. So Tx M = Tx Nj .

(⇐): Let µ = H n x(M ∩ BR (0)), µ is Borel regular with µ(Rn+l ) < ∞.

we define the distance

d(π, π 0 ) = sup |qπ (x) − qπ0 (x)|.
|x|=1

This is the | sin θ| of the angle between them, if n = l = 1.

By inner regularity, and since Tx M exists µ a.e., we can choose a closed set F ⊂ M such that
1
µ(Rn+l )
µ(Rn+l \ F ) ≤
4
such that for each x ∈ F , M has an approximate tangent space Px at x. So

µ(Bρ (x))
lim =1
ρ→0 ωn ρn
µ(Bρ (x) ∩ X 21 (πx , x))
lim =0
ρ→0 ωn ρn
where πx = (Tx M )⊥ . Define

µ(Bρ (x))
fk (x) = inf1
0<ρ< k ωn ρn

4
 µ(Bρ (x) ∩ X 21 (πx , x))
qk (x) = sup .
1
0<ρ< k ωn ρn
then lim fk (x) = 1, and lim qk (x) = 0 for all x ∈ F . We have previously shown these to be measurable,
so by Egoroff, we can choose a closed E ⊂ F with
1
µ(F \ E) ≤ µ(Rn+l )
4
and the limit will be uniform on E. Then for each ε > 0, there is a δ > 0 such that

µ(Bρ (x))
≥1−ε
ωn ρn
µ(Bρ (x) ∩ X 12 (πx , x))
≤ε
ωn ρn
for all x ∈ E, 0 < ρ < δ.

Now choose many l dimensional subspaces π1 , ..., πN of Rn+l N = N (n, l), choose enough such that
for each l dimensional subspace π ⊂ Rn+l , there is a 1 ≤ j ≤ N such that d(π, πj ) < 1/16 and let
E1 , ..., EN ⊂ E defined by
1
Ej = {x : d(πj , πx ) < }.
16

1
Then E = ∪N
j=1 Ej and Claim: Let ε = 16n and choose an appropriate δ then

X 14 (πj , x) ∩ Ej ∩ Bδ/2 (x) = {x}

for all x ∈ Ej , 1 ≤ j ≤ N , i.e. E won’t curve back into the cone. (Note in Simon’s notation, all balls
are closed balls.)

5
Indeed, by contradiction, suppose we have x ∈ Ej and y ∈ X 14 (πj , x) ∩ Ej ∩ ∂Bρ (x) from some
0 < ρ ≤ δ/2. But since x ∈ E and 2ρ ≤ δ we have

µ(X 21 (πx , x) ∩ B2ρ (x)) < εωn (2ρ)n

and using Bρ/8 (y) ⊂ X 21 (πj , x) ∩ B2ρ (x)) we have

µ(X 12 (πx , x) ∩ B2ρ (x)) ≥ µ(Bρ/8 (y)) ≥ ωn (ρ/8)n .

Contradiction. Therefore, we can write Ej as the graph of a Lipschitz function, by the extension
theorem, we can extend the function to all of Rn so that f : Rn → Rl , to be precise

Ej ∩ Bδ/4 (x0 ) ⊂ Q(graphf )

Where Q : Rn+l → Rn+l is an orthogonal transformation with Q(πj ) = {0} × Rl . We can do this for
each 1 ≤ j ≤ N , and all x ∈ Ej , so
N
E ⊂ Uj=1 Qj (graphfj ).
By our earlier estimates,
1
µ(Rn+l \ Uj=1
N
Qj (graphfj )) ≤ µ(Rn+l ).
2
Nk
But now iterate this argument inductively with µk = µk−1 x(Rn+l \ Uj=1 Qj (graphfj )) so that

Nk 1
µk (Rn+l \ Uj=1 Qj (graphfj )) ≤ ( )k µ1 (Rn+l ).
2
This implies we can find countable fj , Qj with

µ(Rn+l \ ∪∞
j=1 Qj (graphfj )) = 0.

Let Gj : Rn → Rn+l be defined by Gj (x) = (x, fj (x)) be the graph map, and Fj : Rn → Rn+l with
Fj = Qj ◦ Gj , then Fj are Lipschitz and

H n (M \ (∪∞ n
j=1 Fj (R ))) = 0. 

GRADIENTS, AREA
We can extend some of the geometric constructions from manifolds to countable n rectifiable sets.
So let M = ∪∞ n n
j=0 Mj disjoint union, where each Mj = Fj (R for Lipschitz fj , and H (M0 ) = 0, also
Mj ⊂ Nj as before.

Let f : U ⊂ Rn+l → R, be locally Lipschitz, U ⊃ M . We can define for H n a.e. y ∈ M ,

∇M f (y) = ∇Nj f (y)

for y ∈ Mj , because Ty M = Ty Nj where it’s defined.

For f : U → RP , P ≥ n, define dM fx : Tx M → RP by

dM fx (τ ) = Dτ f (y) = hτ, ∇M f (x)i

for τ ∈ Tx M . So we can define for H n a.e. x ∈ M
q p
JfM (x) = det(dM fx )∗ ◦ (dM fx ) = det B(x)
where Bij = Dτi · Dτj .

Claim: We have the following area formula

Z Z
M
Jf dH =n
H 0 (A ∩ f −1 (y))dH n (y)
A Rn+l

6
for H n measurable A ⊂ M .

Consider Fj (Rn ) = Mj ⊂ M . By C 1 approximation, we can, as earlier, remove a set of measure 0,

and have f |Mj = gj |Mj for some g ∈ C 1 .

Indeed, we can find C 1 functions gij : Rn → Rn+l with H n (Ei = {x : Fj (x) 6= gij (x)}) < 1i . Let

E = Rn \ ∪i {x : Fj (x) = gij (x)}.

So H n (E) = 0, and define gj : Rn \ E → Rn+l by gj (x) = gij (x) = Fj (x) for some i.
N
So, JfM (x) = Jgj j (x) for H n a.e. x ∈ Mj . Then we can use the area formula for the smooth case, so
Z Z
JfM dH n = H 0 (A ∩ Mj ∩ f −1 (y))dH n (y).
A∩Mj Rn+l

So sum over j.

VARIFOLDS:

Definition: A rectifiable n varifold is a pair (M, θ), where M ⊂ Rn+l is a countable n rectifiable set,
and θ : M → R is locally integrable, is called the multiplicity function.

We associate to a varifold the radon measure µV , called the weight measure,

Z
µV (A) = θdH n .
A∩M

By the results of chapter 3, M has an approximate tangent space with multiplicity θ(x) for µ a.e.
x ∈ Rn+l , which we call Tx V .

We define the mass, by

M(V ) = µV (Rn+l ).
We define the support of V by

sptV = sptµV = {x : ∃Bρ (x) µV (Bρ (x)) > 0}.

If A ⊂ Rn+l is H n measurable, we can define the restriction V xA to be the recifiable n varifold

V xA = (M ∩ A, θ|M ∩A ).
We say that Vk → V if µVk → µV , in the sense of Radon measures, i.e.
Z Z
f dµVk → f dµV
M M

for all f ∈ Cc0 (U ).

Now let M ⊂ U ⊂ Rn+l , W ⊂ Rn+l , U, W open. Suppose f : sptV ∩ U → W is proper (K ⊂ W

compact ⇒ f −1 (K) ∩ sptV is compact), 1-1, Lipschitz. Define (pushforward)

f# V = (f (M ), θ ◦ f −1 ).
We need to check that θ ◦ f −1 is locally integrable. But since K compact implies f −1 (K) compact,
so by the area formula
Z Z Z
θ ◦ f −1 dH n = θ ◦ f −1 dH n = JfM θdH n < ∞.
f (M )∩K f (M ∩f −1 (K)) M ∩f −1 (K)

Where JfM is as defined earlier

7
 q
JfM (x) = det Dτi (x) · Dτj (x).
If f is not 1-1, then we define

f# V = (f (M ), θ̃)
P
where θ̃ = x∈f −1 (y)∩M θ(x). Then we have

Z Z Z Z
M(f# V ) = θ̃(y)dH n (y) = θ(x) dH 0 (x) dH n (y) = JfM (x) θ(x) dH n (x).
f (M ) f (M ) f −1 (y)∩M M