An interface between a gas and liquid is often referred to as a free surface. The reason for the “free” designation arises from the large difference in the densities of the gas and liquid (e.g., the ratio of density for water to air is 1000). A low gas density means that its inertia can generally be ignored compared to that of the liquid. In this sense the liquid moves independently, or freely, with respect to the gas. The only influence of the gas is the pressure it exerts on the liquid surface. In other words, the gas-liquid surface is not constrained, but free.
In heat-transfer texts the term ‘Stephen Problem’ is often used to describe free boundary problems. In this case, however, the boundaries are phase boundaries, e.g., the boundary between ice and water that changes in response to the heat supplied from convective fluid currents.
Whatever the name, it should be obvious that the presence of a free or moving boundary introduces serious complications for any type of analysis. For all but the simplest of problems, it is necessary to resort to numerical solutions. Even then, free surfaces require the introduction of special methods to define their location, their movement, and their influence on a flow.
In the following discussion we will briefly review the types of numerical approaches that have been used to model free surfaces, indicating the advantages and disadvantages of each method. Regardless of the method employed, there are three essential features needed to properly model free surfaces:
- A scheme is needed to describe the shape and location of a surface.
- An algorithm is required to evolve the shape and location with time.
- Free-surface boundary conditions must be applied at the surface.
Lagrangian Grid Methods
Conceptually, the simplest means of defining and tracking a free surface is to construct a Lagrangian grid that is imbedded in and moves with the fluid. Many finite-element methods use this approach. Because the grid and fluid move together, the grid automatically tracks free surfaces.
At a surface it is necessary to modify the approximating equations to include the proper boundary conditions and to account for the fact that fluid exists only on one side of the boundary. If this is not done, asymmetries develop that eventually destroy the accuracy of a simulation.
The principal limitation of Lagrangian methods is that they cannot track surfaces that break apart or intersect. Even large amplitude surface motions can be difficult to track without introducing regridding techniques such as the Arbitrary-Lagrangian-Eulerian (ALE) method. References 1970 and 1974 may be consulted for early examples of these approaches.
The remaining free-surface methods discussed here use a fixed, Eulerian grid as the basis for computations so that more complicated surface motions may be treated.
Surface Height Method
Low amplitude sloshing, shallow water waves, and other free-surface motions in which the surface does not deviate too far from horizontal, can be described by the height, H, of the surface relative to some reference elevation. Time evolution of the height is governed by the kinematic equation, where (u,v,w) are fluid velocities in the (x,y,z) directions. This equation is a mathematical expression of the fact that the surface must move with the fluid:
Finite-difference approximations to this equation are easy to implement. Further, only the height values at a set of horizontal locations must be recorded so the memory requirements for a three-dimensional numerical solution are extremely small. Finally, the application of free-surface boundary conditions is also simplified by the condition on the surface that it remains nearly horizontal. Examples of this technique can be found in References 1971 and 1975.
Marker-and-Cell (MAC) Method
The earliest numerical method devised for time-dependent, free-surface, flow problems was the Marker-and-Cell (MAC) method (see Ref. 1965). This scheme is based on a fixed, Eulerian grid of control volumes. The location of fluid within the grid is determined by a set of marker particles that move with the fluid, but otherwise have no volume, mass or other properties.
Grid cells containing markers are considered occupied by fluid, while those without markers are empty (or void). A free surface is defined to exist in any grid cell that contains particles and that also has at least one neighboring grid cell that is void. The location and orientation of the surface within the cell was not part of the original MAC method.
Evolution of surfaces was computed by moving the markers with locally interpolated fluid velocities. Some special treatments were required to define the fluid properties in newly filled grid cells and to cancel values in cells that are emptied.
The application of free-surface boundary conditions consisted of assigning the gas pressure to all surface cells. Also, velocity components were assigned to all locations on or immediately outside the surface in such a way as to approximate conditions of incompressibility and zero-surface shear stress.
The extraordinary success of the MAC method in solving a wide range of complicated free-surface flow problems is well documented in numerous publications. One reason for this success is that the markers do not track surfaces directly, but instead track fluid volumes. Surfaces are simply the boundaries of the volumes, and in this sense surfaces may appear, merge or disappear as volumes break apart or coalesce.
A variety of improvements have contributed to an increase in the accuracy and applicability of the original MAC method. For example, applying gas pressures at interpolated surface locations within cells improves the accuracy in problems driven by hydrostatic forces, while the inclusion of surface tension forces extends the method to a wider class of problems (see Refs. 1969, 1975).
In spite of its successes, the MAC method has been used primarily for two-dimensional simulations because it requires considerable memory and CPU time to accommodate the necessary number of marker particles. Typically, an average of about 16 markers in each grid cell is needed to ensure an accurate tracking of surfaces undergoing large deformations.
Another limitation of marker particles is that they don’t do a very good job of following flow processes in regions involving converging/diverging flows. Markers are usually interpreted as tracking the centroids of small fluid elements. However, when those fluid elements get pulled into long convoluted strands, the markers may no longer be good indicators of the fluid configuration. This can be seen, for example, at flow stagnation points where markers pile up in one direction, but are drawn apart in a perpendicular direction. If they are pulled apart enough (i.e., further than one grid cell width) unphysical voids may develop in the flow.
Surface Marker Method
One way to limit the memory and CPU time consumption of markers is to keep marker particles only on surfaces and not in the interior of fluid regions. Of course, this removes the volume tracking property of the MAC method and requires additional logic to determine when and how surfaces break apart or coalesce.
In two dimensions the marker particles on a surface can be arranged in a linear order along the surface. This arrangement introduces several advantages, such as being able to maintain a uniform particle spacing and simplifying the computation of intersections between different surfaces. Surface markers also provide a convenient way to locate the surface within a grid cell for the application of boundary conditions.
Unfortunately, in three-dimensions there is no simple way to order particles on surfaces, and this leads to a major failing of the surface marker technique. Regions may exist where surfaces are expanding and no markers fill the space. Without markers the configuration of the surface is unknown, consequently there is no way to add markers.
Reference 1975 contains examples that show the advantages and limitations of this method.
Volume-of-Fluid (VOF) Method
The last method to be discussed is based on the concept of a fluid volume fraction. The idea for this approach originated as a way to have the powerful volume-tracking feature of the MAC method without its large memory and CPU costs.
Within each grid cell (control volume) it is customary to retain only one value for each flow quantity (e.g., pressure, velocity, temperature, etc.) For this reason it makes little sense to retain more information for locating a free surface. Following this reasoning, the use of a single quantity, the fluid volume fraction in each grid cell, is consistent with the resolution of the other flow quantities.
If we know the amount of fluid in each cell it is possible to locate surfaces, as well as determine surface slopes and surface curvatures. Surfaces are easy to locate because they lie in cells partially filled with fluid or between cells full of fluid and cells that have no fluid.
Slopes and curvatures are computed by using the fluid volume fractions in neighboring cells. It is essential to remember that the volume fraction should be a step function, i.e., having a value of either one or zero. Knowing this, the volume fractions in neighboring cells can then be used to locate the position of fluid (and its slope and curvature) within a particular cell.
Free-surface boundary conditions must be applied as in the MAC method, i.e., assigning the proper gas pressure (plus equivalent surface tension pressure) as well as determining what velocity components outside the surface should be used to satisfy a zero shear-stress condition at the surface. In practice, it is sometimes simpler to assign velocity gradients instead of velocity components at surfaces.
Finally, to compute the time evolution of surfaces, a technique is needed to move volume fractions through a grid in such a way that the step-function nature of the distribution is retained. The basic kinematic equation for fluid fractions is similar to that for the height-function method, where F is the fraction of fluid function:
A straightforward numerical approximation cannot be used to model this equation because numerical diffusion and dispersion errors destroy the sharp, step-function nature of the F distribution.
It is easy to accurately model the solution to this equation in one dimension such that the F distribution retains its zero or one values. Imagine fluid is filling a column of cells from bottom to top. At some instant the fluid interface is in the middle region of a cell whose neighbor below is filled and whose neighbor above is empty. The fluid orientation in the neighboring cells means the interface must be located above the bottom of the cell by an amount equal to the fluid fraction in the cell. Then the computation of how much fluid to move into the empty cell above can be modified to first allow the empty region of the surface-containing cell to fill before transmitting fluid on to the next cell.
In two or three dimensions a similar procedure of using information from neighboring cells can be used, but it is not possible to be as accurate as in the one-dimensional case. The problem with more than one dimension is that an exact determination of the shape and location of the surface cannot be made. Nevertheless, this technique can be made to work well as evidenced by the large number of successful applications that have been completed using the VOF method. References 1975, 1980, and 1981 should be consulted for the original work on this technique.
The VOF method has lived up to its goal of providing a method that is as powerful as the MAC method without the overhead of that method. Its use of volume tracking as opposed to surface-tracking function means that it is robust enough to handle the breakup and coalescence of fluid masses. Further, because it uses a continuous function it does not suffer from the lack of divisibility that discrete particles exhibit.
Variable-Density Approximation to the VOF Method
One feature of the VOF method that requires special treatment is the application of boundary conditions. As a surface moves through a grid, the cells containing fluid continually change, which means that the solution region is also changing. At the free boundaries of this changing region the proper free surface stress conditions must also be applied.
Updating the flow region and applying boundary conditions is not a trivial task. For this reason some approximations to the VOF method have been used in which flow is computed in both liquid and gas regions. Typically, this is done by treating the flow as a single fluid having a variable density. The F function is used to define the density. An argument is then made that because the flow equations are solved in both liquid and gas regions there is no need to set interfacial boundary conditions.
Unfortunately, this approach does not work very well in practice for two reasons. First, the sensitivity of a gas region to pressure changes is generally much greater than that in liquid regions. This makes it difficult to achieve convergence in the coupled pressure-velocity solution. Sometimes very large CPU times are required with this technique.
The second, and more significant, reason is associated with the possibility of a tangential velocity discontinuity at interfaces. Because of their different responses to pressure, gas and liquid velocities at an interface are usually quite different. In the Variable-Density model interfaces are moved with an average velocity, but this often leads to unrealistic movement of the interfaces.
Even though the Variable-Density method is sometimes referred to as a VOF method, because is uses a fraction-of-fluid function, this designation is incorrect. For accurately tracking sharp liquid-gas interfaces it is necessary to actually treat the interface as a discontinuity. This means it is necessary to have a technique to define an interface discontinuity, as well as a way to impose the proper boundary conditions at that interface. It is also necessary to use a special numerical method to track interface motions though a grid without destroying its character as a discontinuity.
A brief discussion of the various techniques used to numerically model free surfaces has been given here with some comments about their relative advantages and disadvantages. Readers should not be surprised to learn that there have been numerous variations of these basic techniques proposed over the years. Probably the most successful of the methods is the VOF technique because of its simplicity and robustness. It is this method, with some refinement, that is used in the FLOW-3D program.
Attempts to improve the VOF method have centered on better, more accurate, ways to move fluid fractions through a grid. Other developments have attempted to apply the method in connection with body-fitted grids and to employ more than one fluid fraction function in order to model more than one fluid component. A discussion of these developments is beyond the scope of this introduction.
1965 Harlow, F.H. and Welch, J.E., Numerical Calculation of Time-Dependent Viscous Incompressible Flow, Phys. Fluids 8, 2182.
1969 Daly, B.J., Numerical Study of the Effect of Surface Tension on Interface Instability, Phys. Fluids 12, 1340.
1970 Hirt, C.W., Cook, J.L. and Butler, T.D., A Lagrangian Method for Calculating the Dynamics of an Incompressible Fluid with Free Surface, J. Comp. Phys. 5, 103.
1971 Nichols, B.D. and Hirt, C.W.,Calculating Three-Dimensional Free Surface Flows in the Vicinity of Submerged and Exposed Structures, J. Comp. Phys. 12, 234.
1974 Hirt, C.W., Amsden, A.A., and Cook, J.L.,An Arbitrary Lagrangian-Eulerian Computing Method for all Flow Speeds, J. Comp. Phys., 14, 227.
1975 Nichols, B.D. and Hirt, C.W., Methods for Calculating Multidimensional, Transient Free Surface Flows Past Bodies, Proc. of the First International Conf. On Num. Ship Hydrodynamics, Gaithersburg, ML, Oct. 20-23.
1980 Nichols, B.D. and Hirt, C.W., Numerical Simulation of BWR Vent-Clearing Hydrodynamics, Nucl. Sci. Eng. 73, 196.
1981 Hirt, C.W. and Nichols, B.D., Volume of Fluid (VOF) Method for the Dynamics of Free Boundaries, J. Comp. Phys. 39, 201.