Notes Viscosity

! Thursday: class will begin late, 11:30am
! In reality, nearby molecules travelling at
different velocities occasionally bump into
each other, transferring energy
• Differences in velocity reduced (damping)
• Measure this by strain rate (time derivative of
strain, or how far velocity field is from rigid
motion)
• Add terms to our constitutive law

cs533d-term1-2005 1 cs533d-term1-2005 2

Strain rate Viscous stress

! At any instant in time, measure how fast chunk ! As with linear elasticity, end up with two parameters if we
of material is deforming from its current state want isotropy:
• Not from its original state ! ijviscous = 2µ"˙ij + #"˙kk$ij
• So we"re looking at infinitesimal, incremental strain • µ and ! are coefficients of viscosity (first and second)
updates • These are not the Lame coefficients! Just use the same symbols
• Can use linear Cauchy strain! • ! damps only compression/expansion
! So the strain rate tensor is " Usually !!-2/3µ (exact for monatomic gases)
" So end up with
# "u "u &
!˙ij = 12 %% i + j (( % " u "u 2 "u (
$ "x j "x i ' ! ijviscous = µ'' i + j # k
$ij **
& " x j "x i 3 "x k )
cs533d-term1-2005 3 cs533d-term1-2005 4
 Navier-Stokes Nondimensionalization

! Navier-Stokes equations include the viscous ! Actually go even further

stress
! Select a characteristic length L
! Incompressible version:
ut + u ! "u + #1 "p = g + #1 " ! µ("u + "uT ) • e.g. the width of the domain,
! And a typical velocity U
"!u=0 • e.g. the speed of the incoming flow
! Often (but not always) viscosity µ is constant,
and this reduces to ! Rescale terms
ut + u ! "u + #1 "p = g + µ# " 2 u • x"=x/L, u"=u/U, t"=tU/L, p"=p/#U2
so they all are dimensionless
• Call $=µ/# the “kinematic viscosity” Lg # 2
u't +u'!"u'+"p'= + " u'
cs533d-term1-2005 5
U 2 UL cs533d-term1-2005 6

Nondimensional parameters Why nondimensionalize?

! Re=UL/$ is the Reynold"s number ! Think of it as a user-interface issue

• The smaller it is, the more viscosity plays a ! It lets you focus on what parameters matter
role in the flow • If you scale your problem so nondimensional
parameters stay constant, solution scales
• High Reynold"s numbers are hard to simulate ! Code rot --- you may start off with code which
! Fr= U gL is the Froude number has true dimensions, but as you hack around
they lose meaning
• The smaller it is, the more gravity plays a role • If you"re nondimensionalized, there should be only
in the flow one or two parameters to play with
• Note: often can ignore gravity (pressure ! Not always the way to go --- you can look up
gradient cancels it out) material constants, but not e.g. Reynolds
numbers
(0,#1,0) 1 2
ut + u ! "u + "p = + " u
Fr 2 Re cs533d-term1-2005 7 cs533d-term1-2005 8
 Other quantities Boundary conditions

! We may want to carry around auxiliary ! Inviscid flow:

quantities • Solid wall: u•n=0
• Free surface: p=0 (or atmospheric pressure)
• E.g. temperature, the type of fluid (if we have • Moving solid wall: u•n=uwall•n
a mix), concentration of smoke, etc. ! Also enforced in-flow/out-flow
! Use material derivative as before • Between two fluids: u1•n=u2•n and p1=p2+%&
! Viscous flow:
! E.g. if quantity doesn"t change, just is
• No-slip wall: u=0
transported (“advected”) around: • Other boundaries can involve coupling tangential
Dq components of stress tensor…
= qt + u
!"! "q
# =0 ! Pressure/velocity coupling at boundary:
Dt advection • u•n modified by #p/#n

cs533d-term1-2005 9 cs533d-term1-2005 10

What now? Vorticity

! Can numerically solve the full equations ! How do we measure rotation?

• Will do this later
• Not so simple, could be expensive (3D)
! Or make assumptions and simplify them,
then solve numerically
• Simplify flow (e.g. irrotational)
• Simplify dimensionality (e.g. go to 2D)
• Vorticity of a vector field (velocity)!is:= " # u
• Proportional (but not equal) to angular velocity
of a rigid body - off by a factor of 2
! Vorticity is what makes smoke look
interesting
• Turbulence

cs533d-term1-2005 11 cs533d-term1-2005 12
 Vorticity equation Vorticity equation continued

! Start with N-S, constant viscosity and density ! Simplify with more vector identities, and assume
ut + u ! "u + #1 "p = g + $" 2 u incompressible to get:
! Take curl of whole equation D!
= ! " #u + $# 2!
! " ut + ! " ( u # !u) + ! " $1 !p = ! " g + ! " (%! 2 u) Dt
! Lots of terms are zero: ! Important result: Kelvin Circulation Theorem
• g is constant (or the potential of some field) • Roughly speaking: if '=0 initially, and there"s no
• With constant density, pressure term too viscosisty, '=0 forever after (following a chunk of
fluid)
! " ut + ! " ( u # !u) = $! " ! 2 u
! Then use vector identities to simplify… " If fluid starts off irrotational, it will stay that way
(in many circumstances)
! " ut + ! " ((! " u) " u + 12 !u 2 ) = #! 2 (! " u)
$ t + ! " ($ " u) = #! 2$
cs533d-term1-2005 13 cs533d-term1-2005 14

Potential flow Potential in time

! If velocity is irrotational: ! What if we have a free surface?

!"u=0 ! Use vector identity u•(u=(()u))u+(|u|2/2
• Which it often is in simple laminar flow " Assume
! Then there must be a stream function (potential) • incompressible ((•u=0), inviscid, irrotational (()u=0)
such that: •
u = !" •
constant density
thus potential flow (u=(*, (2*=0)
! Solve for incompressibility: " Then momentum equation simplifies
(using G=-gy for gravitational potential)
! " !# = 0 ut + (! " u) " u + 12 ! u + #1 !p = g
2

! If u•n is known at boundary, we can solve it

!$ t + 12 ! u + #1 !p = %!G
2

cs533d-term1-2005 15 cs533d-term1-2005 16
 Bernoulli’s equation Water waves

! Every term in the simplified momentum ! For small waves (no breaking), can reduce
equation is a gradient: integrate to get geometry of water to 2D heightfield
p Can reduce the physics to 2D also
! t + 12 u 2 + = #G
!

" • How do surface waves propagate?

• (Remember Bernoulli"s law for pressure?) ! Plan of attack
• Start with potential flow, Bernoulli"s equation
! This tells us how the potential should • Write down boundary conditions at water surface
evolve in time • Simplify 3D structure to 2D

cs533d-term1-2005 17 cs533d-term1-2005 18

Set up Boundaries

! We"ll take y=0 as the height of the water at ! At sea floor (y=-H), v=0 !y = 0
rest
! H is the depth (y=-H is the sea bottom)
! At sea surface (y=h!0), v=ht
• Note again - assuming very small horizontal motion
! h is the current height of the water at (x,z)
! y = ht
! Simplification: velocities very small (small
amplitude waves) ! At sea surface (y=h!0), p=0
• Or atmospheric pressure, but we only care about
pressure differences
• Use Bernoulli"s equation, throw out small velocity
squared term, use p=0,

cs533d-term1-2005 19
! t = "gh cs533d-term1-2005 20
 Finding a wave solution Dispersion relation

! Plug in *=f(y)sin(K•(x,z)-'t) ! So the wave speed c is

• In other words, do a Fourier analysis in ! g
c= = tanh kH
horizontal component, assume nothing much k k
happens in vertical ! Notice that waves of different wave-
• Solving (2*=0 with boundary conditions on *y numbers k have different speeds
gives " cosh( K (y + H)) • Separate or disperse in time
!=A sin(K # (x,z) $ "t )
K sinh( K H )
! For deep water (H big, k reasonable -- not
tsunamis) tanh(kH)!1
• Pressure boundary condition then gives (with
k=|K|, the magnitude of K) g
! = gk tanh kH c=
k
cs533d-term1-2005 21 cs533d-term1-2005 22

Simulating the ocean Energy spectrum

! Far from land, a reasonable thing to do is ! Fourier decomposition of height field:

• Do Fourier decomposition of initial surface h(x,z,t) = # hˆ (i, j,t )e !1( i, j )" ( x,z )

height i, j
2
• Evolve each wave according to given wave ! “Energy” in K=(i,j) is S(K) = hˆ (K)
speed (dispersion relation)
! Update phase, use FFT to evaluate
! Oceanographic measurements have found
! How do we get the initial spectrum? models for expected value of S(K)
• Measure it! (oceanography) (statistical description)

cs533d-term1-2005 23 cs533d-term1-2005 24
 Phillips Spectrum Fourier synthesis

! For a “fully developed” sea ! From the prescribed S(K), generate actual
• wind has been blowing a long time over a large area, Fourier coefficients
( )
statistical distribution of spectrum has stabilized
hˆ (K,0) = 1 X + X !1 S (K )
! The Phillips spectrum is: [Tessendorf…] 2 1 2

" !1 %" K ( W % 2 • Xi is a random number with mean 0, standard

1
2 ! ( kl) '$
deviation 1 (Gaussian)
S(K) = A 4 exp$$ '
2
'
k # ( kL) &# K W & • Uniform numbers from unit circles aren"t terrible either
! Want real-valued h, so must have
• A is an arbitrary amplitude
• L=|W|2/g is largest size of waves due to wind velocity hˆ (K) = hˆ (!K)"
W and gravity g • Or give only half the coefficients to FFT routine and
• Little l is the smallest length scale you want to model specify you want real output

cs533d-term1-2005 25 cs533d-term1-2005 26

Time evolution Picking parameters

! Dispersion relation gives us '(K) ! Need to fix grid for Fourier synthesis
At time t, want h(x,t) = $ hˆ (K,0)e !1( K " x!#t ) (e.g. 1024x1024 height field grid)
"
! Grid spacing shouldn"t be less than e.g. 2cm (smaller
K =(i, j ) than that - surface tension, nonlinear wave terms, etc.
= $ hˆ (K,0)e ! !1#t
e !1K " x take over)
• Take little l (cut-off) a few times larger
K =(i, j )
! Total grid size should be greater than but still
" So then coefficients at time t are comparable to L in Phillips spectrum (depends on wind
• For j$0: speed and gravity)
• Others: figure out from conjugacy condition (or leave ! Amplitude A shouldn"t be too large
it up to real-valued FFT to fill them in) • Assumed waves weren"t very steep

hˆ (i, j,t ) = hˆ (i, j,0)e! !1"t

cs533d-term1-2005 27 cs533d-term1-2005 28
 Note on FFT output Choppy waves

! FFT takes grid of coefficients, outputs grid of ! See Tessendorf for more explanation
heights
! Nonlinearities cause real waves to have
! It"s up to you to map that grid sharper peaks and flatter troughs than
(0…n-1, 0…n-1) to world-space coordinates
linear Fourier synthesis gives
! In practice: scale by something like L/n
• Adjust scale factor, amplitude, etc. until it looks nice ! Can manipulate height field to give this
! Alternatively: look up exactly what your FFT effect
routines computes, figure out the “true” scale • Distort grid with (x,z) -> (x,z)+!D(x,z,t)
factor to get world-space coordinates K ˆ
D(x,t) = " ! !1 h (K,t )e !1K # x

K
K

cs533d-term1-2005 29 cs533d-term1-2005 30

Choppiness problems

! The distorted grid can actually tangle up

(Jacobian has negative determinant - not
1-1 anymore)
• Can detect this, do stuff (add particles for
foam, spray?)

cs533d-term1-2005 31