Fluid Mechanics Current Methods The Heat Equation Back to Navier–Stokes Recap
A Near-Optimal Spectral Method for Simulating
Fluids in a Cylinder
David Darrow
Mentors:
Dr. Alex Townsend and Prof. Grady Wright
MIT Primes
May 20, 2017
David Darrow A Near-Optimal Spectral Method for Simulating Fluids in a Cylinder
Fluid Mechanics Current Methods The Heat Equation Back to Navier–Stokes Recap
The Navier–Stokes Equations
The Incompressible Navier–Stokes Equations model the motion of
incompressible fluids (e.g. liquid):
∂ 1 2
∇ · ~v = 0, + ~v · ∇ ~v = ∇ ~v − ∇p.
∂t Re
~v = fluid velocity
p = internal pressure
Re = “Reynolds number”
Low Re, high viscosity High Re, low viscosity
David Darrow A Near-Optimal Spectral Method for Simulating Fluids in a Cylinder
Fluid Mechanics Current Methods The Heat Equation Back to Navier–Stokes Recap
Solving in a Cylinder
We will focus on solving these equations in a cylinder.
David Darrow A Near-Optimal Spectral Method for Simulating Fluids in a Cylinder
Fluid Mechanics Current Methods The Heat Equation Back to Navier–Stokes Recap
Fluids Move in Mysterious Ways
Why are the Navier–Stokes equations a challenge?
I There is no analytical solution.
I Designing accurate/efficient numerical methods is non-trivial.
I They are highly nonlinear when Re 0.
I There is no explicit equation for the pressure.
David Darrow A Near-Optimal Spectral Method for Simulating Fluids in a Cylinder
Fluid Mechanics Current Methods The Heat Equation Back to Navier–Stokes Recap
Current Methods
There are three issues with current solution methods:
I They do not appropriately resolve the boundary of the domain.
I They have relatively low accuracy.
I Their computational cost is high, leading to long simulation times.
Some methods try to approximate the
boundary with a polygon.
Turbulent flow is more complicated and
important to model near the boundary.
David Darrow A Near-Optimal Spectral Method for Simulating Fluids in a Cylinder
Fluid Mechanics Current Methods The Heat Equation Back to Navier–Stokes Recap
Starting Off Simple
When Re ≈ 0, the Navier–Stokes equations resemble the heat equation:
∂ 2
− c∇ ~v = f (x, y, z, t), c > 0.
∂t
On the cylinder, though, this is still difficult to solve efficiently.
David Darrow A Near-Optimal Spectral Method for Simulating Fluids in a Cylinder
Fluid Mechanics Current Methods The Heat Equation Back to Navier–Stokes Recap
How Do We Solve the Heat Equation?
To approximate a solution to the heat equation, we only enforce it at
points on a discretization grid:
Our solution satisfies the heat equation at each black point on this cylinder.
By choosing more points, we get a more accurate solution for the whole
cylinder.
David Darrow A Near-Optimal Spectral Method for Simulating Fluids in a Cylinder
Fluid Mechanics Current Methods The Heat Equation Back to Navier–Stokes Recap
Doubling up the Discretization Grid
By “doubling up” the grid, we remove a fake boundary at the centerline
of the cylinder and spread points out more evenly.
Mapped onto rectangular coordinates
A colored cylinder
Doubled up
David Darrow A Near-Optimal Spectral Method for Simulating Fluids in a Cylinder
Fluid Mechanics Current Methods The Heat Equation Back to Navier–Stokes Recap
Increasing the Accuracy
We also employ new, fast, and accurate “spectral” methods for solving
differential equations, which means we do not require as many points in
the discretization grid to have high accuracy.
”Finite difference” Spectral methods Spectral methods,
methods doubled
David Darrow A Near-Optimal Spectral Method for Simulating Fluids in a Cylinder
Fluid Mechanics Current Methods The Heat Equation Back to Navier–Stokes Recap
Accuracy vs. Computation Time
David Darrow A Near-Optimal Spectral Method for Simulating Fluids in a Cylinder
Fluid Mechanics Current Methods The Heat Equation Back to Navier–Stokes Recap
We can now solve the heat equation with O(n3 log n) steps.
Animation
David Darrow A Near-Optimal Spectral Method for Simulating Fluids in a Cylinder
Fluid Mechanics Current Methods The Heat Equation Back to Navier–Stokes Recap
Ramping Up the Reynolds Number
The higher the Reynolds number, the harder a system is to model.
How does heat diffusion relate to the Navier–Stokes equations?
With some more machinery, they can be turned into heat equations.
David Darrow A Near-Optimal Spectral Method for Simulating Fluids in a Cylinder
Fluid Mechanics Current Methods The Heat Equation Back to Navier–Stokes Recap
The PT Decomposition
~ and any unit vector ẑ,
For any vector field A
~
∇ × [λa ẑ] + ∇ × ∇ × [γa ẑ] = ∇ × A.
+ =
Toroidal Field Poloidal Field Velocity Field
~ in O(n3 log n) steps.
We can compute the PT decomposition of A
David Darrow A Near-Optimal Spectral Method for Simulating Fluids in a Cylinder
Fluid Mechanics Current Methods The Heat Equation Back to Navier–Stokes Recap
Rewriting Navier–Stokes
We can decompose the Navier–Stokes equations into Poloidal and
Toroidal components:
∂ 1 2
+ ~v · ∇ ~v = ∇ ~v − ∇p (Velocity Field )
∂t Re
becomes
∂ 1 2
− ∇ λω = f1 (~
ω) (Toroidal Field )
∂t Re
and
∂ 1 2
− ∇ γω = f2 (~
ω ), (Poloidal Field )
∂t Re
which can be numerically solved as two scalar heat equations.
David Darrow A Near-Optimal Spectral Method for Simulating Fluids in a Cylinder
Fluid Mechanics Current Methods The Heat Equation Back to Navier–Stokes Recap
Recap
Current Methods Our method
I have relatively low accuracy. I has digits of accuracy
I either over-resolve the origin or proportional to n.
under-resolve the boundary. I resolves the boundary and
I have a high computational cost. selects discretization points
more evenly.
I only requires O(n3 log n)
operations.
David Darrow A Near-Optimal Spectral Method for Simulating Fluids in a Cylinder
Fluid Mechanics Current Methods The Heat Equation Back to Navier–Stokes Recap
Future Developments
Future Goals:
I Combining the pieces into a single Navier–Stokes code.
David Darrow A Near-Optimal Spectral Method for Simulating Fluids in a Cylinder
Fluid Mechanics Current Methods The Heat Equation Back to Navier–Stokes Recap
Acknowledgments
I Dr. Alex Townsend
I Prof. Grady Wright
I Jonasz Slomka
I Dr. Khovanova, Prof. Etingof, Dr. Gerovitch, and everyone else
involved in organizing the PRIMES-USA program
Figures:
I Artery image retrieved from mrsdallas.weebly.com
I Volcano image retrieved from researchgate.net
I Rocket image retrieved from wikimedia.org
I Honey image retrieved from styletips101.com
I Water image retrieved from nuocuongvihawa.com
I Turbulent flow modeled by the Argonne National Laboratory
David Darrow A Near-Optimal Spectral Method for Simulating Fluids in a Cylinder
