I. Cloud Physics Application Open Questions II. Algorithm Open Issues III. Computer Science / Engineering Open issues 1
Part I. Cloud Physics Application Open Questions 2
Open mul)scale problems relevant to clouds R. Shaw Reynolds number dependence of cloud processes: how does physics change as increasing range of scales is added? Coupling of turbulent mixing, droplet phase changes, and gravita)onal sebling (decoupling) resolving across transi)on scale Coalescence of two droplets occurs in variable local environments (shear, acceleration, both Re dependent) what are the collision efficiencies over full size range, activated drops to raindrops Ice formation (and other threshold processes) strongly dependent on temperature field intermittent, Re dependent Radiative transfer strongly modulated by presence of condensed phase, which is randomly dispersed by turbulence; simultaneously, modulation of turbulence by radiative heating e.g., radiative cooling at cloud interface
Lian-Ping Wang The multiscale problem Global circulation model ~ 107 m Droplet-droplet interaction 10 5 ~ 10 4 m Numerical weather Prediction, 105 ~ 106 m Cloud-resolving LES Inertial-range scales ~ 10 ~ 10 1 m HDNS
Before we close the scale gap between LES and HDNS: How to properly apply the collision-kernel parameterization from HDNS in LES for large-scale dynamics? After the scale gap is closed: How to couple HDNS of cloud microphysics and LES of cloud dynamics? Inertial-range edddies 10 cm ~ 10m Cloud-resolving LES dx 1 m Actual HDNS domain 20 cm ~ 1 m
Particle dispersion in turbulent flows DNS of particle dispersion from the core of a turbulent vortex flow (Marshall, PF 2005) Traditional stochastic models disperse particles randomly in turbulent flows. Experiments and DNS show that particles cluster in high concentration sheets due to centrifugal action of turbulent eddies. Can dispersion models be designed to include particle clustering?
Part II. Algorithm Open Issues 7
Stokesian dynamics at higher flow Reynolds numbers Fluid flow is deflected by particle aggregates, influencing the fluid forces on the particles. Stokesian dynamics is a highly efficient approach for incorporating fluid - particle interaction into multiphase flows Current Stokesian dynamics methods require that the flow Reynolds number is small, and hence are restricted mainly to microfluidics problems. There are many multiphase flow problems in which the particle Reynolds number is small but the flow Reynolds number is large. Can Stokesian dynamics be extended for such problems?
Long-range hydrodynamic interaction If we embed particles in a periodic domain and assume a Stokes-flow disturbance, the cumulative effect is divergent. How can we resolve this issue? Should we alter the periodic domain? Should we alter the model for the embedded particles? Should we handle the problem artificially by truncating the interactions?
Lian-Ping Wang Particle-resolved DNS is desirable for finite droplet Grid spacing = 0.001 domain size Reynolds number. How to couple point-particle based hybrid DNS and particle-resolved DNS? 7th International Conference on Multiphase Flow, ICMF 2010, Tampa, FL, May 30 June 4, 2010 How to design particle-resolved DNS? Table 1: Particles parameters at release time. Case d ρp /ρf Np d/η d/λ φv φm τp /τk 1 2 3 4 8.0 8.0 11.0 2.56 5.0 2.56 0 6400 6400 2304 16.1 16.1 22.1 1.2 1.2 1.5 0 0.10 0.10 0.10 0 0.28 0.56 0.27 22.5 57.6 42.5 Inertial-range edddies 10 cm ~ 10m that the system has a zero net vertical mass flux. An important question for a random suspension is how the mean velocity varies with the particulate volume concentration. Figure 1 shows the mean sedimentation velocity normalized by V0, the terminal velocity of a single particle sedimenting in the same periodic domain. This is known as the hindered settling function and it is plotted as a function of particulate volume fraction. Also shown are results from Climent & Maxey (2003) at particle Reynolds numbers of 1, 5, and 10. Clearly, the particle average settling velocity is significantly reduced as its volume fraction is increased, and the larger the particle Reynolds number the smaller the settling velocity. The results are in good agreement with those of Climent & Maxey (2003), with our simulations predict a somewhat smaller settling velocity. The differences could be due to two reasons. First, there are statistical uncertainties in both studies. Second, the force coupling method is an approximate method in which the disturbance flow is not fully resolved. The relative vertical velocity fluctuations Vrms /Vmean are shown in Fig. 2. The overall trends are similar: the relative vertical velocity fluctuation increases with the volume fraction, but decreases with the particle Reynolds number. Quantitatively, our Actual HDNS domain 20 cm ~ 1 m Short-range lubrication force (a) (b) Particle-resolved DNS Difficult to cover full dissipation range and droplet size at the same time a << η
Discrete-element modeling for nanoparticles Accurate simulation of particle drift in a fluid requires that For nanoparticles St < 10-5, and this restriction limits applicability of DEM. If there are no particle collisions, we can deal with low Stokes numbers using the fast Euler approximation (Ferry & Balachandar, 2003) Nano-particles aligned on a micromachined surface during a spin-coating process. What can we do to speed up computations if there are collisions?