24.02.2014 Views

Partial Differential Equations - Modelling and ... - ResearchGate

Partial Differential Equations - Modelling and ... - ResearchGate

Partial Differential Equations - Modelling and ... - ResearchGate

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Reduced-Order <strong>Modelling</strong> of Dispersion 249<br />

But the distance function can be non-uniform with anisotropy (the unit<br />

spheres being ellipsoids). In a chosen metric M the distance between A <strong>and</strong><br />

B is given by<br />

d M (AB) =<br />

∫ 1<br />

0<br />

(<br />

t −→ ABM(A + t<br />

−→ AB)<br />

−→ AB<br />

) 1/2<br />

dt,<br />

where M is positive definite <strong>and</strong> symmetric in symmetric geometries. With<br />

M = I, one recovers the Euclidean geometry <strong>and</strong> variable M permits to account<br />

for anisotropy <strong>and</strong> non-uniformity of the distance function. We have<br />

widely used this approach for mesh adaptation for steady <strong>and</strong> unsteady phenomenon<br />

[AGFM02, HM97, BGM97] linking the metric to the Hessian of the<br />

solution. This definition of the metric permits to equi-distribute the interpolation<br />

error over a given mesh <strong>and</strong>, therefore, monitor the quality of the<br />

solution.<br />

Consider now the following distance function definition:<br />

Definition 1. If A is upwind with respect to B then<br />

d(B,A) =∞ <strong>and</strong> d(A, B) =<br />

∫ B<br />

⊥<br />

A<br />

ds/u = T,<br />

where T is the migration time from A to B ⊥ along the characteristic passing<br />

by A.<br />

u is the local velocity along this characteristic <strong>and</strong> is, by definition, tangent<br />

to the characteristic. B ⊥ denotes the projection of B over this characteristic<br />

in the Euclidean metric. One supposes that this characteristic is unique,<br />

hence avoiding sources <strong>and</strong> attraction points in the flow field. In case of nonuniqueness<br />

of this projection, one chooses the direction of the projection which<br />

satisfies best the constraint u ·∇c g =0inB.<br />

Generalized Plume Solution<br />

Once this distance built, we assume the distribution of a passive scalar transported<br />

by a flow u can be written as:<br />

c g = c c (d)f(d ⊥ E,δ(d)). (4)<br />

Here the subscript g reads for global <strong>and</strong> mentions long distance transport.<br />

d ⊥ E is the Euclidean distance in the normal direction local to the characteristic<br />

at B ⊥ (i.e. along direction BB ⊥ ).<br />

Flow Field<br />

One should keep in mind that in realistic configurations, one has very little<br />

information on the details of the atmospheric flow compared to the accuracy

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!