Gerrit Lohmann date: April 19, 2021
based on Chapter 1 from Holton and Hakim (2013)
Conservation of mass \[ \nabla_d \cdot \mathbf{u_d} = 0 \] Conservation of momentum
\[ \frac{\partial}{\partial t_d } \mathbf{u_d} + ( \mathbf{u_d} \cdot \nabla_d) \mathbf{u_d} = - \nabla_d p_d + \frac{1}{Re} \nabla_d^2 \mathbf{u_d} \]
The dimensionless parameter \[ Re=UL/ \nu \]
is the Reynolds number and the only parameter left!
For large Reynolds numbers, the flow is turbulent. \( Re \sim (10^4-10^8) \). Flow develops steep gradients locally.
trailer Cellules de Bénard (1min), Rayleigh–Bénard convection: cooking oil and small aluminium particles (5 min), Was haben Benard-Zellen mit Kochen zu tun? (3 min, German)
Rayleigh Benard Thermal
Convection with LBM (5 min),
Rayleigh Benard Thermal
Convection 3D Simulation (2 min)
sketch
Bifurcation youtube,
Bifurcation Khan academy,
Bifurcation K
Max and Moritz, not lazy,
sawing secretly a gap in the bridge.
When now this act is over, you suddenly hear a scream:
“Hey, out! You billy goat!
Tailor, tailor, bitch, bitch, bitch!”
And there he is on the bridge,
Cracks! The bridge is falling apart;
Schneider Böck has a fast stream flowing in front of his house. M & M taunt him by making goat noises, until he runs outside. The bridge breaks; the tailor is swept away and nearly drowns (but for two geese, which he grabs a hold of).
A typical example could be the consumer-producer problem where the consumption is proportional to the (quantity of) resource.
For example:
\[ \frac{dx}{dt}= \underbrace{r x \cdot \left( 1- x \right)}_{growth} \quad - \quad \underbrace{ \, p \quad x \, }_{consumption} \]
where \[ rx(1-x) \]
is the logistic equation of resource growth:
Rate of reproduction proportional to the population, and available resources
\[ x_{E 1} = 0 \qquad \]
\[ x_{E 2 } = 1 - \frac{p}{r} \]
Biological population with size N:
\[ \frac{dN}{dt}=r N \cdot \left( 1- \frac{N}{K} \right) \]
r growth rate and K carrying capacity.
the early, unimpeded growth rate is modeled by the first term \( r N. \)
As the population grows, \( -r N^2/K \) becomes large as some members interfere with each other by competing for some critical resource (food, living space). The competition diminishes the combined growth rate, until the value of N ceases to grow (maturity of the population).
\[ N(t) = \frac{K N_0 e^{rt}}{K + N_0 \left( e^{rt} - 1\right)} = \frac{K }{K/N_0 e^{-rt} + 1- e^{-rt} } \quad \rightarrow_{t\to \infty } K \]
In climate, the logistic equation is also important for Lorenz's forecast error.
Explosion of human population over the last 10,000 years along with some relevant historical events.
Think about the ways that each of these events might have affected birth and death rates of the human population.
N is the number of cases,
r infection rate,
p cure rate,
K final epidemic size.
dN/dt linearly decreases with the number of cases.
\[ \frac{dN}{dt}=r N \cdot \left( 1- \frac{N}{K} \right) - p N \]
When is the growth rate peak?
How many infections ?
When do we need places in hospitals?
Using \[ x = N/K \]
\[ \frac{dx}{dt}= f(x) = \underbrace{r x \cdot \left( 1- x \right)}_{growth} \quad - \quad \underbrace{ p x }_{cure} \]
Without any medicine: \[ r = 3 p \]
\[ x_{E 1} = 0, \mbox{ and } x_{E 2 } = 1 - \frac{p}{r} = 2/3 \]
\[ f'(x) = r - 2 r x -p \]
\[ f'(x_{E 1})= r-p > 0 \quad \mbox{instability} \]
\[ f'(x_{E 2})= -r+p < 0 \quad \mbox{stability} \]
r=2000/150000
Bev=85000000
K=1
dt=0.01
N=150000/Bev
vN=c(0); vNp=c(0); vt=c(0)
vN[1]=N; vNp[1]=0; vt[1]=0
for(i in 2:100000){
N1=N+r*N*(1-N/K)*dt
vNp[i]=r*N*(1-N/K)
vN[i]=N1
vt[i]=i*dt
N=N1}
plot(vt,vN,type="l",xlab="time [days]",ylab="N(t)")
plot(vt,vNp*Bev/100,type="l",xlab="time [days]",ylab="dN(t)/dt * 1/100 ")
max(vNp[]*Bev/100)
[1] 2833.333
r=5000/150000
Bev=85000000
K=1
dt=0.01
N=10000/Bev
vN=c(0); vNp=c(0); vt=c(0)
vN[1]=N; vNp[1]=0; vt[1]=0
for(i in 2:100000){
N1=N+r*N*(1-N/K)*dt
vNp[i]=r*N*(1-N/K)
vN[i]=N1
vt[i]=i*dt
N=N1}
plot(vt,vN,type="l",xlab="time [days]",ylab="N(t)")
plot(vt,vNp*Bev/100,type="l",xlab="time [days]",ylab="dN(t)/dt * 1/100 ")
max(vNp[]*Bev/100)
[1] 7083.333
Lecture: April 19. (Monday), 14:00 Prof. Dr. Gerrit Lohmann
Tutorial: April 19. (Monday), Justus Contzen and Lars Ackermann
Time required for Sheet 2: 8-9 h
Content in the script: Rayleigh-Bénard convection, Lorenz system,
nonlinear dynamics, bifurcations, multiple
equilibria
Stability Theory (Chapter 2) ](http://paleodyn.uni-bremen.de/study/Dyn2/dyn2script_chapter2.pdf) Reading/learning (the sections with a star are voluntary). It might take 90-120 min.
Consider the continuous dynamical system described by
\[ \dot x=f(x,\lambda)\quad \]
A bifurcation occurs at \[ (x_E,\lambda_0) \]
if the Jacobian matrix
\[ \textrm{d}f/dx (x_E,\lambda_0) \]
has an Eigenvalue with zero real part.
Consider the continuous dynamical system described by \[ \dot x=f(x,\lambda)\quad \] A bifurcation occurs at \[ (x_E,\lambda_0) \] if the Jacobian matrix \[ \textrm{d}f/dx (x_E,\lambda_0) \] has an Eigenvalue with zero real part.
a fixed point interchanges its stability with another fixed point as the control parameter is varied. Bifurcation at \( r=0 \).
\[ \frac{dx}{dt}=rx (1-x) \, \]
The two fixed points are 0 and 1. When r is negative, the fixed point at 0 is stable and 1 is unstable. But for \( r>0 \), 0 is unstable and 1 is stable.
The normal form: \[ \frac{dx}{dt}=b+x^2 \]
\[ b<0: \mbox{ a stable equilibrium point at } -\sqrt{-b} \mbox{ and an unstable one at } \sqrt{-b} \qquad \qquad \]
\( b=0 \): exactly one equilibrium point, saddle-node fixed point.
\( b>0 \): no equilibrium points. Saddle-node bifurcations may be associated with hysteresis loops.
\[ f(x) = b + x^2 = 0 \]
\[ x_{E1,E2} = \pm \sqrt{-b} \mbox{ for } b \le 0 \]
\[ f'(x) = 2x \]
\[ f'(x_{E1}) = 2\sqrt{-b} > 0 \quad \mbox{unstable} \]
\[ f'(x_{E2}) = -2\sqrt{-b} < 0 \quad \mbox{stable} \]
For b=0: equilibrium point \( x_E=0 \) which is indifferent (not stable/unstable).
\[ \frac{dx}{dt}=b+x^2 = - \frac{d}{dx} \left( - b x - \frac{x^3}{3} \right) = - \frac{d}{dx} V (x) \]
Global analysis including basins of attraction for \( x_{E2}: (-\infty,3) \)
\[ \frac{dx}{dt}=b+x^2 \]
Rayleigh (1916) temperature difference between the upper- and lower-surfaces \[ T(x, y, z=H) = \, T_0 \] \[ T(x, y, z=0) \, = \, T_0 + \Delta T \]
Furthermore \[ \rho = \rho_0 = const. \] except in the buoyancy term, where:
\[ \varrho = \varrho_0 (1 - \alpha(T-T_0)) \mbox{ with } \alpha > 0 \quad . \]
common feature of geophysical flows
Diffusion: Temperature varies linearly with depth:
\[ T_{eq} = T_0 + \left(1 - \frac{z}{H}\right) \Delta T \]
No movement of particles:
\[ u = w= 0 \]
When this solution becomes unstable, convection should develop.
\[ D_t u = - \frac{1}{\rho_0} \partial_x p + \nu \nabla^2 u \label{eqref:einse} \qquad \qquad \qquad \qquad \qquad \qquad (1) \]
\[ D_t w = - \frac{1}{\rho_0} \partial_z p + \nu \nabla^2 w + g (1- \alpha (T-T_0)) \qquad \qquad (2) \]
\[ \partial_x u + \partial_z w = 0 \qquad \qquad \qquad \qquad \qquad \qquad \qquad (3) \]
\[ D_t T = \kappa \nabla^2 T \qquad \qquad \qquad \qquad \qquad \qquad \qquad (4) \]
\[ D_t \left( \nabla^2 \Psi\right) = \nu \nabla^4 \Psi - g \alpha \frac{\partial \Theta}{\partial x} \]
\[ T_{eq} = T_0 + \left(1 - \frac{z}{H}\right) \Delta T \]
\[ \mbox{with } \quad T = T_{eq} + \Theta \quad , \quad \mbox{where } \quad \Theta \quad \mbox{is the anomaly} \]
In vorticity equation:
\[ \quad T \frac{\partial }{\partial x} g (1- \alpha (T_{eq} + \Theta -T_0)) = - g \alpha \frac{\partial }{\partial x} \Theta \]
\[ T_{eq} = T_0 + \left(1 - \frac{z}{H}\right) \Delta T \]
\[ \mbox{with } \quad T = T_{eq} + \Theta \quad , \quad \mbox{where } \quad \Theta \quad \mbox{is the anomaly} \]
In vorticity equation:
\[ \frac{\partial }{\partial x} g (1- \alpha (T_{eq} + \Theta -T_0)) = - g \alpha \frac{\partial }{\partial x} \Theta \]
Temperature equation:
\[ D_t T = D_t T_{eq} + D_t \Theta = w \cdot \frac{- \Delta T}{H} + D_t \Theta = - \frac{\Delta T}{H} \frac{\partial \Psi}{\partial x} + D_t \Theta \]
\[ D_t \Theta \quad = \frac{\Delta T}{H} \frac{\partial \Psi}{\partial x} + \kappa \nabla^2 \Theta \quad \]
\[ \frac{1}{T} \frac{1}{L^2} \frac{L^2}{T} D_t \left( \nabla_d^2 \Psi_d\right) = \nu \frac{1}{L^4} \frac{L^2}{T} \nabla_d^4 \Psi_d - g \alpha \frac{ \Delta T}{L} \frac{\partial \Theta_d}{\partial x_d} \]
\[ \frac{\Delta T}{T} D_t \Theta_d \quad = \frac{\Delta T}{H} \frac{L^2}{ T L} \frac{\partial \Psi_d}{\partial x_d} + \kappa \frac{\Delta T}{L^2} \nabla_d^2 \Theta_d \quad \]
\[ \frac{1}{T} \frac{1}{L^2} \frac{L^2}{T} D_t \left( \nabla_d^2 \Psi_d\right) = \nu \frac{1}{L^4} \frac{L^2}{T} \nabla_d^4 \Psi_d - g \alpha \frac{ \Delta T}{L} \frac{\partial \Theta_d}{\partial x_d} \]
\[ \frac{\Delta T}{T} D_t \Theta_d \quad = \frac{\Delta T}{H} \frac{L^2}{ T L} \frac{\partial \Psi_d}{\partial x_d} + \kappa \frac{\Delta T}{L^2} \nabla_d^2 \Theta_d \quad \]
\[ \mbox{Inserting} \quad T= H^2/\kappa \] \[ \mbox{Rayleigh number} \quad R_a = \frac{g \alpha H^3 \Delta T}{\nu \kappa} \] \[ \mbox{Prandtl number} \quad \sigma = \frac{ \nu}{ \kappa} \]
\[ \frac{1}{T} \frac{1}{L^2} \frac{L^2}{T} D_t \left( \nabla_d^2 \Psi_d\right) = \nu \frac{1}{L^4} \frac{L^2}{T} \nabla_d^4 \Psi_d - g \alpha \frac{ \Delta T}{L} \frac{\partial \Theta_d}{\partial x_d} \]
\[ \frac{\Delta T}{T} D_t \Theta_d \quad = \frac{\Delta T}{H} \frac{L^2}{ T L} \frac{\partial \Psi_d}{\partial x_d} + \kappa \frac{\Delta T}{L^2} \nabla_d^2 \Theta_d \quad \]
\[ \mbox{Inserting} \quad T= H^2/\kappa \] \[ \mbox{Rayleigh number} \quad R_a = \frac{g \alpha H^3 \Delta T}{\nu \kappa} \] \[ \mbox{Prandtl number} \quad \sigma = \frac{ \nu}{ \kappa} \]
\[ D_t \left( \nabla_d^2 \Psi_d\right) = \sigma \nabla_d^4 \Psi_d - R_a \sigma \frac{\partial \Theta_d}{\partial x_d} \]
\[ D_t \Theta_d \quad = \frac{\partial \Psi_d}{\partial x_d} + \nabla_d^2 \Theta_d \quad \]
\[ \mbox{ Saltzman (1962): Expand } \Psi, \Theta \mbox{ in double Fourier series in x and z: } \]
\[ \Psi (x,z,t) \, = \, \sum_{k=1}^\infty \sum_{l=1}^\infty \Psi_{k,l} (t) \, \, \sin \left(\frac{k \pi a}{H} x \right) \, \times \, \sin \left(\frac{ l \pi}{H} z \right) \] \[ \Theta (x,z,t) \, = \, \sum_{k=1}^\infty \sum_{l=1}^\infty \Theta_{k,l} (t) \cos \left(\frac{k \pi a}{H} x \right) \, \times \, \sin \left( \frac{l \pi}{H} z \right) \]
Approximation: Just 3 Modes X(t), Y(t), Z(t)
\[ \frac{a}{1+a^2} \, \kappa \, \Psi = X \sqrt{2} \sin\left(\frac{\pi a}{H} x \right) \sin\left(\frac{\pi}{H} z \right) \]
\[ \pi \frac{R_a}{R_c} \frac{1}{\Delta T} \, \Theta = Y \sqrt{2} \cos\left(\frac{\pi a}{H} x\right) \sin\left(\frac{\pi}{H} z \right) - Z \sin\left(2 \frac{\pi}{H} z \right) \]
\[ \mbox{Motion develops if } \quad R_a = \frac{g \alpha H^3 \Delta T}{\nu \kappa} \quad \mbox{exceeds } \quad R_c = \pi^4 \frac{(1+a^2)^3}{a^2} \]
\[ \mbox{The minimum value of $R_c = 657.51$ occurs when $a^2 = 1/2$. } \]
\[ \mbox{When } R_a < R_c,\mbox{ heat transfer is due to conduction} \]
\[ \mbox{When } R_a > R_c, \mbox{ heat transfer is due to convection.} \]
Bifurcation at \[ r = R_a/R_c = 1 \]
Geometry constant \[ b = 4(1+a^2)^{-1} \]
Famous low-order model:
\[ \dot X = -\sigma X + \sigma Y \]
\[ \dot Y = r X - Y - X Z \]
\[ \dot Z = -b Z + X Y \]
\[ \mbox{dimensionless time } \quad t_d = \pi^2 H^{-2} (1+a^2) \kappa t, \]
\[ \mbox{ Prandtl number } \quad \sigma = \nu \kappa^{-1}, \]
r=24
s=10
b=8/3
dt=0.01
x=0.1
y=0.1
z=0.1
vx<-c(0)
vy<-c(0)
vz<-c(0)
for(i in 1:10000){
x1=x+s*(y-x)*dt
y1=y+(r*x-y-x*z)*dt
z1=z+(x*y-b*z)*dt
vx[i]=x1
vy[i]=y1
vz[i]=z1
x=x1
y=y1
z=z1}
plot(vx,vy,type="l",xlab="x",ylab="y")
plot(vy,vz,type="l",xlab="y",ylab="z")
r=0.9
s=10
b=8/3
dt=0.01
x=1.1
y=0.1
z=11.1
vx<-c(0)
vy<-c(0)
vz<-c(0)
for(i in 1:100){
x1=x+s*(y-x)*dt
y1=y+(r*x-y-x*z)*dt
z1=z+(x*y-b*z)*dt
vx[i]=x1
vy[i]=y1
vz[i]=z1
x=x1
y=y1
z=z1}
plot(vx,type="l",xlab="time",ylab="x")
plot(vy,type="l",xlab="time",ylab="y")
r=3.5
s=10
b=8/3
dt=0.01
x=1.1
y=0.1
z=11.1
vx<-c(0)
vy<-c(0)
vz<-c(0)
for(i in 1:1000){
x1=x+s*(y-x)*dt
y1=y+(r*x-y-x*z)*dt
z1=z+(x*y-b*z)*dt
vx[i]=x1
vy[i]=y1
vz[i]=z1
x=x1
y=y1
z=z1}
plot(vx,type="l",xlab="time",ylab="x")
plot(vy,type="l",xlab="time",ylab="y")