Chapter V: Shape Analysis and Energy Stability of Conductive Liquids
5.4 Geometric Variations of Surface Area and Volume
As a sanity check, we test the second variation ofC on a spherical capacitor occupying the region between two concentric spheres of inner radius ra and outer radius rb. The exact solution to the electric field within the capacitor is given by
ψ= ra
ra−rb 1−rb r
, ψ(ra) = 1, ψ(rb) = 0, (5.97) where r is the radial distance and the outward normal vector n = −gr on the inner surface. We consider a configuration mapχ(R, )such that the inner radiusrain spatial coordinates is mapped fromRain material frame,
ra=χ(Ra, ) =Ra+. (5.98) The rate of spatial deformation vector v = gr in this case is purely normal to the inner surface. The exact solution to the self-capacitance C[ra] and its first two total derivatives in for any radiusra can be easily obtained,
C[ra] = 4πrarb rb−ra
, dC[ra]
d = 4πrb2
(rb−ra)2, d2C[ra]
d2 = 8πr2b
(rb−ra)3. (5.99) The auxiliary problem of the Eulerian derivative ψ0 also admits the analytic solution,
ψ0 = rb (rb−ra)2
rb
r −1, ψ0(ra) =−v·dψ= rb/ra
rb−ra, ψ0(rb) = 0. (5.100) We also compute the necessary ingredients needed in the integral formula of the first and the second derivatives,
n·dψ|ra = rb/ra rb−ra
, n·dψ0|ra = r2b/r2a
(rb−ra)2, H= 1 ra
, vn=−1. (5.101) It can be easily verified that, with quantities from expression (5.101) substituted, the surface integrals (5.49) and (5.92) are indeed evaluated to the identical results from the exact solution (5.99).
First and second derivatives of area differential
Variational calculations of area element is much easier with convective Lagrangian co- ordinates. We start with an equivalent expression of a spatial area element dγ,
dγ =n·ndγ. (5.102)
Substituting Nanson’s relation (5.21) into (5.102) and taking total derivative with re- spect to yield
ddγ
d =n·d(ndγ) d +dn
d ·ndγ
=n·∂JF−>
∂ NdΓ +1 2d|n|2
d dγ Note |n|= 1
=n·hJ(divv)F−>−Jl>F−>iNdΓ
=n·h(divv)−l>indγ = divv−nl>ndγ. (5.103) Using the adapted curvilinear coordinates for surface γ, we arrive at a convenient ex- pression for the first total derivative (or deformation rate) of spatial area element,
d(dγ)
d = divγvdγ = divγvγ−2hvndγ. (5.104) Note from the calculation of (5.103) that, the following identity
dn
ddγ = d(ndγ)
d −nddγ d
= divv−l>ndγ−n divv−nl>ndγ =h nl>nn−l>nidγ (5.105) must hold for all area differentials dγ which is only possible if
dn
d = nl>nn−l>n. (5.106) Equation (5.106) is the first variation of surface normal vectornas boundaryγ deforms.
We then apply a similar strategy to derive the second total derivative of spatial area element which yields
d2n·ndγ d2
=n·d2(ndγ)
d2 +d2n
d2 ·ndγ+ 2dn
d ·d(ndγ)
d note d2|n|2 d2 = 0
=n·∂2JF−>
∂2 NdΓ −dn d ·dn
d dγ+dn
d ·d(ndγ) d +dn
d ·d(ndγ) d
=n·∂2JF−>
∂2 NdΓ +dn
d ·nddγ
d + nl>nn−l>n·(divv)−l>ndγ
=n·∂2JF−>
∂2 NdΓ +nl l>ndγ−(nl>n)2dγ, (5.107)
where∂2JF−>/∂2 can be explicitly expanded in aid of (5.28) and (5.30),
∂2JF−>
∂2 = 2∂J
∂
∂F−>
∂ +∂2J
∂2F−>+J∂2F−>
∂2
=−2J(divv)l>F−>+Jh(divv)2+ div( ˙v)−tr(l l)iF−>
+Jl>l>F−>−J(dv˙>−l>l>)F−>
=−2(divv)l>+ (divv)2−tr(l l) + 2l>l>+ div( ˙v)−dv˙>JF−>. (5.108) Applying identity (5.108) to (5.107) finally leads to the spatial form of the second derivative of area differential,
d2dγ
d2 =−2(divv)nl>n+ (divv)2−tr(l l) + 2nl>l>n
+ div ˙v−ndv˙>n+nl l>n−(nl>n)2dγ. (5.109) In surface-adapted curvilinear coordinates, equation (5.109) simplifies to
d2dγ
d2 =h(divγv)2−(nl>n)2−tr(l l) + 2nl>l>n+nl l>n−(nl>n)2+ divγv˙idγ.
(5.110) Using index notation, we can simply various terms in expression (5.110) and group them into contributions intrinsic and extrinsic to the surfaceγ,
2nl>l>n−tr(l l)−(nl>n)2 = 2(∇3vi)(∇iv3)−(∇jvi)(∇ivj)−(∇3v3)2
= 2(∇3v3)(∇3v3) + 2(∇3vα)(∇αv3)−(∇3v3)(∇3v3)
−(∇αv3)(∇3vα)−(∇βvα)(∇αvβ)−(∇3vα)(∇αv3)
−(∇3v3)2
=−(∇βvα)(∇αvβ) =−( ˆ∇βvα−iiαβv3)( ˆ∇αvβ−iiβαv3)
=−( ˆ∇βvα)( ˆ∇αvβ) + 2iiαβv3( ˆ∇αvβ)−iiβαiiαβ(v3)2, (5.111) nl l>n−(nl>n)2 = (∇jv3)(∇jv3)−(∇3v3)2
= (∇αv3)gαβ(∇βv3)
= ( ˆ∇αv3+iiαα0vα0)gαβ( ˆ∇βv3+iiββ0vβ0)
= ( ˆ∇αv3)( ˆ∇βv3) +iiβα0vα0( ˆ∇βv3)
+( ˆ∇αv3)iiαβ0vβ0+iiβα0iiββ0vα0vβ0, (5.112) (divγv)2 = (divγvγ)2−4h(divγvγ)vn+ 4h2vn2, (5.113) where∇ˆα is the low-dimensional covariant derivative restricted to surfaceγ, for which
∇βvα = ˆ∇βvα−iiαβ on γ (5.114)
holds (recall expansions (4.29) of the first kind Christoffel symbol). Substituting ex- pressions (5.111), (5.112) and (5.113) into second total derivative (5.110) of dγ and exploiting properties of the second fundamental formii we arrive at
d2|dγ|
d2 =hdivγv˙ + (divγv)2−(∇βvα)(∇αvβ) + (∇αv3)(∇αv3)idγ
=hdivγv˙ + (divγvγ)2−4h(divγvγ)vn+ 4h2vn2
−( ˆ∇βvα)( ˆ∇αvβ) + 2iiαβv3( ˆ∇αvβ)−iiβαiiαβ(v3)2 +( ˆ∇αv3)gαβ( ˆ∇βv3) + 2iiβα0vα0( ˆ∇βv3) +iiβα0iiββ0vα0vβ0
idγ
=hdivγv˙ + (divγvγ)2−4h(divγvγ)vn+ 4h2vn2
−( ˆ∇βvα)( ˆ∇αvβ) + 2iiαβ(v3∇ˆαvβ +vβ∇ˆαv3)−(4h2−2k)(v3)2 +( ˆ∇αv3)( ˆ∇αv3) +iiβα0iiββ0vα0vβ0
idγ
=hdivγv˙ + (divγvγ)2−( ˆ∇βvα)( ˆ∇αvβ) + ( ˆ∇αvn)( ˆ∇αvn) + 2kv2n
−4h(divγvγ)vn+ 2iiαβ∇ˆα(vnvβ) +iiβα0iiββ0vα0vβ0
idγ (5.115)
where∇ˆαis the low-dimensional covariant differential operation of surfaceγ. A cleanup of the algebras in expression (5.115) leads to the final result
d2dγ
d2 =ndivγv˙
+ 2hvγiivγ−kvγ·vγ
+dγvn·dγvn+ 2kv2n
+ (divγvγ)2−( ˆ∇βvα)( ˆ∇αvβ)
−4h(divγvγ)vn+ 2iiαβ∇ˆα(vnvβ)odγ.
(5.116)
The first line in (5.116) is due to spatial acceleration v˙ which is expected to have the identical form with the first derivative d(dγ)/d in (5.104) with v replaced by v.˙ The bilinear form in the second line in (5.116) is known as the third fundamental form tensor iiiαβ = 2hiiαβ −kgαβ. Integrands from the third and the fourth lines are the intrinsic contributions from tangential vector field vγ and from scalar (normal) field vn, respectively. We interpret the geometric meaning of these intrinsic contributions by recognizing
(divγvγ)2−( ˆ∇βvα)( ˆ∇αvβ) = (divγvγ)2−tr (dγvγ)(dγvγ)= 2 det( ˆ∇βvα). (5.117) Expression (5.117) exactly agrees with the second derivative of Jacobian∂2J/∂2 from (5.30) in the case of a flat two-dimensional manifold with tangential deformation vγ only. The last line in (5.116) represents the three-way coupling between vγ, vn and the extrinsic (mean) curvature of surfaceγ. If the second derivative ofdγ in (5.117) is
restricted to normal variations only (i.e. v = vnn), we then arrive at the well known formula in shape analysis,
d2dγ
d2 =divγv˙ +|dγvn|2+ 2kvn2dγ when vγ =0. (5.118) In the theory of minimal surfaces, integration of expression (5.118) over the entire surface of a critical shape γ leads to a quadratic shape functional whose convexity directly determines whether its total surface area is locally minimal or maximal. The expression of the first area variation (5.104) and the second variation (5.118) restricted to normal displacement can be found in standard textbooks on differential geometry (Stoker, 1988; Kreyszig, 1991). However, the full expression (5.116) of second area variation d2dγ/d2 especially including the tangential variations of the surface is less commonly known or documented, at least to the best of the author’s knowledge. We find it coincides with the expression derived in Capovilla and Guven (2004).
First and second derivatives of total volume
Variations of volume differential element are straightforward since they mostly deal with the relative Jacobian J. The first derivative of spatial volume differentialdω is directly given by ∂J/∂in (5.29),
ddω d = ∂J
∂ dΩ= divvdω. (5.119)
The first derivative of the total volume can be expressed as a surface integral through divergence theorem,
d d
Z
ωdω=Z
Ω
∂J
∂ dΩ=Z
γ
v·ndγ =Z
γ
vndγ. (5.120)
The second derivative of spatial volume differential dω is directly given by ∂2J/∂2 in (5.30),
d2|dω| d2 = ∂2J
∂2 dΩ=div ˙v+ (divv)2−tr(l l)dω. (5.121) In order to convert the second derivative of the total volume
d2 d2
Z
ωdω=Z
Ω
∂2J
∂2 dΩ=Z
ωdiv ˙v+ (divv)2−tr(l l) dω (5.122) to a surface integral only, we first recall an identity similar to the one we encounter in (5.76),
div(divv)v−lv=∇i(∇jvj)vi−(∇jvi)vj
= (∇jvj)(∇ivi)−(∇jvi)(∇ivj) + (∇i∇jvj)vi−(∇i∇jvi)vj
= (divv)2−tr(l l). Flat (5.123)
With identity (5.123) substituted in (5.122), the second derivative of the total volume now becomes a surface integral,
d2 d2
Z
ωdω=Z
γ
n·v˙ + (divv)(n·v)−nlvdγ. (5.124) We can gain more insights into the second derivative (5.124) by expanding it using index notation of surface-adapted curvilinear coordinates,
(divv)(n·v)−nlv= (∇ivi)v3−(∇iv3)vi = (∇αvα)v3−(∇αv3)vα
= ( ˆ∇αvα−iiααv3)v3−( ˆ∇αv3+iiαβvβ)vα
=v3∇ˆαvα−vα∇ˆαv3−2h(v3)2−iiαβvβvα. (5.125) Finally with identity (5.125) we arrive at
d2 d2
Z
ωdω=Z
γv˙n+vn(divγvγ)−vγ·dγvn−2hv2n−vγiivγdγ. (5.126) As usual, the first term in the second derivative of the total volume (5.126) is due to spatial accelerationv. The next two terms are the intrinsic coupling between tangential˙ vector fieldvγ and scalar normal fieldvn, The last two terms represent how vector field vγ and scalar field vn individually interact with extrinsic curvatures of surfaceγ. Both first (5.120) and second (5.126) derivatives of the total volume are surface integral and only involve information of vector fieldvdefined strictly intrinsic to surfaceγ (e.g., n·dv is considered extrinsic to surface γ). This is expected because the total volume of region ω can be expressed as a surface integral via divergence theorem,
Z
ωdω=Z
ω
1
3divxdω= 1 3
Z
γn·xdγ. (5.127)
Finally we would like to highlight the necessity of spatial acceleration v˙ in volume conservation. Consider a configuration mapχ(·, )which is constrained to conserve the total volume for all valid. The total derivative of Rωdω with respect to at any order must be zero over the entire range ofas well. As implied by (5.120), a spatial velocity field v = vγ always tangent to surface γ would guarantee that the first derivative of R
ωdωvanishes but not necessarily the second derivative! To see this claim, let’s consider a unit sphere to be the material frameΩwithχ(X, )being a rigid rotation that rotates Ω to a new spatial frameω about some axis (in this case is the rotation angle). We expect the total volume of spatial domain ω be conserved for all . Since v derived from χ(X, ) is everywhere tangent to the surface γ of the sphere, the first derivative of Rωdω is zero according to formula (5.120). Second fundamental formii in this case equals to the negative of the identity tensor everywhere on the surface of a unit sphere which renders the second derivative of the total volume
d2 d2
Z
ωdω=Z
γv˙n+vγ·vγdγ (5.128)
through the formula (5.126). It’s now evident that, integral (5.128) would always be non-negative without the help of spatial acceleration v˙ and hence violate volume con- servation.