Kuramoto-Sivashinsky equation¤
In 1D:
In higher dimensions:
Uses the combustion format via the gradient norm that easily scales to higher dimensions.
exponax.stepper.KuramotoSivashinsky
¤
Bases: BaseStepper
Source code in exponax/stepper/_kuramoto_sivashinsky.py
8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 |
|
__init__
¤
__init__(
num_spatial_dims: int,
domain_extent: float,
num_points: int,
dt: float,
*,
gradient_norm_scale: float = 1.0,
second_order_diffusivity: float = 1.0,
fourth_order_diffusivity: float = 1.0,
dealiasing_fraction: float = 2 / 3,
order: int = 2,
num_circle_points: int = 16,
circle_radius: float = 1.0
)
Timestepper for the d-dimensional (d ∈ {1, 2, 3}
) Kuramoto-Sivashinsky
equation on periodic boundary conditions. Uses the combustion format
(or non-conservative format). Most deep learning papers in 1d considered
the conservative format available as
KuramotoSivashinskyConservative
.
In 1d, the KS equation is given by
uₜ + b₂ 1/2 (uₓ)² + ν uₓₓ + μ uₓₓₓₓ = 0
with b₂
the gradient-norm coefficient, ν
the diffusivity and μ
the
hyper viscosity. Note that both viscosity terms are on the left-hand
side. As such for ν, μ > 0
, the second-order term acts destabilizing
(increases the energy of the system) and the fourth-order term acts
stabilizing (decreases the energy of the system). A common configuration
is b₂ = ν = μ = 1
and the dynamics are only adapted using the
domain_extent
. For this, we espect the KS equation to experience
spatio-temporal chaos roughly once L > 60
.
In this combustion (=non-conservative) format, the number of channels does not grow with the spatial dimension. A 2d KS still only has a single channel. In higher dimensions, the equation reads
uₜ + b₂ 1/2 ‖ ∇u ‖₂² + ν (∇ ⋅ ∇) u + μ ((∇ ⊗ ∇) ⋅ (∇ ⊗ ∇))u = 0
with ‖ ∇u ‖₂
the gradient norm, ∇ ⋅ ∇
effectively is the Laplace
operator Δ
. The fourth-order term generalizes to ((∇ ⊗ ∇) ⋅ (∇ ⊗ ∇))
which is not the same as ΔΔ = Δ²
since the latter would mix
spatially.
Arguments:
num_spatial_dims
: The number of spatial dimensionsd
.domain_extent
: The size of the domainL
; in higher dimensions the domain is assumed to be a scaled hypercubeΩ = (0, L)ᵈ
.num_points
: The number of pointsN
used to discretize the domain. This includes the left boundary point and excludes the right boundary point. In higher dimensions; the number of points in each dimension is the same. Hence, the total number of degrees of freedom isNᵈ
.dt
: The timestep sizeΔt
between two consecutive states.gradient_norm_scale
: The gradient-norm coefficientb₂
. Note that the gradient norm is already scaled by 1/2. This factor allows for further modification. Default: 1.0.second_order_diffusivity
: The diffusivityν
in the KS equation. The sign of this coefficient is interpreted as if the term was on the left-hand side. Hence it should have a positive value to act destabilizing. Default: 1.0.fourth_order_diffusivity
: The hyper viscosityμ
in the KS equation. The sign of this coefficient is interpreted as if the term was on the left-hand side. Hence it should have a positive value to act stabilizing. Default: 1.0.order
: The order of the Exponential Time Differencing Runge Kutta method. Must be one of {0, 1, 2, 3, 4}. The option0
only solves the linear part of the equation. Use higher values for higher accuracy and stability. The default choice of2
is a good compromise for single precision floats.dealiasing_fraction
: The fraction of the wavenumbers to keep before evaluating the nonlinearity. The default 2/3 corresponds to Orszag's 2/3 rule. To fully eliminate aliasing, use 1/2. Default: 2/3.num_circle_points
: How many points to use in the complex contour integral method to compute the coefficients of the exponential time differencing Runge Kutta method. Default: 16.circle_radius
: The radius of the contour used to compute the coefficients of the exponential time differencing Runge Kutta method. Default: 1.0.
Notes:
- The KS equation enters a chaotic state if the domain extent is chosen large enough. In this chaotic attractor it can run indefinitely. It is in balancing state of the second-order term producing new energy, the nonlinearity transporting it into higher modes where the fourth-order term dissipates it.
- If the domain extent is chosen large enough to eventually enter a chaotic state, the initial condition does not really matter. Since the KS "produces its own energy", the energy spectrum for the chaotic attractor is independent of the initial condition.
- However, since the KS develops a certain spectrum based on the
domain length, make sure to use enough discretization point to
capture the highes occuring mode. For a domain extent of 60, this
requires at least roughly 100
num_points
in single precision floats. - For domain lengths smaller than the threshold to enter chaos, the KS equation, exhibits various other patterns like propagating waves, etc.
- For higher dimensions (i.e.,
num_spatial_dims > 1
), a chaotic state is already entered for smaller domain extents. For more details and the kind of dynamics that can occur see: https://royalsocietypublishing.org/doi/10.1098/rspa.2014.0932
Good Values:
- For a simple spatio-temporal chaos in 1d, set
num_spatial_dims=1
,domain_extent=60
,num_points=100
,dt=0.1
. The initial condition can be anything, important is that it is mean zero. The first 200-500 steps of the trajectory will be the transitional phase, after that the chaotic attractor is reached.
Source code in exponax/stepper/_kuramoto_sivashinsky.py
14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 |
|
__call__
¤
__call__(
u: Float[Array, "C ... N"]
) -> Float[Array, "C ... N"]
Performs a check
Source code in exponax/_base_stepper.py
231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 |
|