Lec FV2/FE1

20 important questions on Lec FV2/FE1

What is the great advantage of choosing CV faces centered between nodes over centering nodes in CVs

Allows us to easily get most accurate flux calculations, because we can interpolate node values through the centre in the most accurate way.

Explain the integral form of the advection/diffusion equation from the FV perspectve

No longer solving it at one infinitely small point, but approach it as a control volume (cube).

It is a math theorem that says that if we want to find the change of the state variable within volume is kinda the integral of all the fluxes over the surfaces of the cube (volume). (Or find change of volume by fluxes that act on it.)

What is the difference between Q and q?

q: flux function (kg/m2/s) tells you how big the flux is, but not how much the flux contributes to the change of the state variable, bc for that we need to know the area that's involved.
Q: flux*area (kg/s). Thus the flux multiplied by the area over which it travels
  • Higher grades + faster learning
  • Never study anything twice
  • 100% sure, 100% understanding
Discover Study Smart

Explain how voronoi cells are made?

Nice thing about FV with help of voronoi cells, we can define our grid by making expanding circles and where circles meet becomes the interface of the control volumes. And like this we have CVs that belong to the node that is closest to them.

Why do we have an exact mass conservation in FV?

Every flux occurs twice in the model, it is a gained term for one CV and a lost term for the CV next to it.

Why do we need a different method to solve our equation instead of finding your perfect solution like on slide 9?

Blue function of your unknown values is unknown a priori, only a strating value to guess your solution  is present

How does Newton Raphson work if a solution of =0 should be found?

  1. Function to solve is drawn, this is the balance for your CVs, which should add up to 0 when solved
  2. Calculate balance and slope in balance (derivative)
  3. Where extrapolation of derivative intersects the x-axis (=0 line), the  balance is calculated in the next iteration. If solution is >0, derivative in negative direction is followed and vice versa.
  4. Final solution is approached by various iterations. Solution of =0 should be found, if not we have an error

How do we deal with transient equations in FV? Transient equations monitor how the balance

Balance eq for CV, we could also formulate the amount of balance that we have as smt that changes over time. So we want to track how it moves to the steady state solution over time. So weintroduce initial conditions


Initial conditions are introduced, eg H1=H2=H3 = 5 m So start flat, turn on the rain and then see what happens to states.

Why should we solve to =0 in numerical solvers?

Numerical solvers are root finding algorithms, so they look for 0/we always formulate =0, if term is nonzero, we have mistake in our balance.

Mention 1 general problem related to derivatives and 4 problems that could occur with the Newton Raphson method?

Often the derivative has to be calculated numerically for complex functions.

Problem with NR for complex (non)linear functions:
  1. send solver in the wrong direction
  2. no solution present
  3. local min/max could occur
  4. non-linear solution does not converge

How can you visually check if you have an error in Newton Raphson?

y-axis represents your entire function (before the =0, H2=x). We are testing our balance, if the value for H2 would be put in on the y-axis, we find it is non-zero, thus an error is present. A perfectly solved model would be exactly zero. If not (we have a value somewhere on the y-axis that is not the intersection with x-axis), we have an error.

Why do we generally use implicit methods to solve transient systems?

Because fluxes can be very non-linear and in that way we can prevent an unstable solutions.

How does FE differ from FV?

FV:
  • same value assigned to entire volume, where the node is in the middle.
  • You need a different shape for a different mesh
  • Voronoi cells

FE:
  • nodes are placed on edges/borders of triangles
    Y
  • You can always make 2 types of grids/meshes with the same nodes.
  • More smooth result  
  • Delauney triangularisation   
  • Works based on inverse distance



Voronoi and Delauney are each other's duals in mesh

How does FE work in Liekes lectures' example?

  • If Q would fall on node Xleft, I assign Xleft value 1.
  • If it falls on the neighbouring node (Xright), I assign 0 to node Xleft.
  • In the middle, we linearly interpolate.
  • If it would fall in the middle, 50/50 would be assigned to both nodes.
  • If it falls off the middle, the linear interpolation determines the fraction each node gets

Does external point flux move to nodes other than the two it falls within?

No, because we cannot create extra mass, or it would have to be a spatial flux

Why is FE more smooth?

FE is a lot smoother because it divides eg a point flux over two neighbouring nodes, whereas FV just assignes it to the volume that's closest.

! Note that even though sometimes if a point flux is closer to a node that is not a neighbour, it is still divided over the two neighbours. If you want it to be more accurate, more nodes should be added. Note that this only works if the resolution of your observations is finer than your nodal distance.

How do you use FE for spatial fluxes

Integrate your redistribution factor * spatial flux with your

With integration, we take an infetisemal small step and for each step we multiply the redistribution factor with the spatial flux at that step. We do this over the complete distance in between the nodes.

External point flux in 2D

Now we determine for any of these samples, how often this external influx is in the red square (=probability for A), etc. So we determine the relative probability of how often this external input is falling into a certain region.

Then we can determine the total probability redistributing this external factor to a certain point.

Why do you need two integrals for the spatial flux in 2D for FE?

Because we integrate over an area and the spatial flux varies over x and y, we need two integrals to integrate over both x and y. Also our redistribution factor has to be calculated for both x and y!

When does the advantage of FE really come in?

If there is more spatial variation occurring within your nodes, as then the extra calculations you're doing are worth it.

The question on the page originate from the summary of the following study material:

  • A unique study and practice tool
  • Never study anything twice again
  • Get the grades you hope for
  • 100% sure, 100% understanding
Remember faster, study better. Scientifically proven.
Trustpilot Logo