WorksMathematics
MPMICE: A hybrid MPM-CFD model for simulating coupled problems in porous media. Application to earthquake-induced submarine landslides
Abstract
In this paper, we describe a soil-fluid-structure interaction model that combines soil mechanics (saturated sediments), fluid mechanics (seawater or air), and solid mechanics (structures). The formulation combines the Material Point Method, which models large deformation of the porous media and the structure, with the Implicit Continuous-fluid Eulerian, which models complex fluid flows. We validate the model and simulate the whole process of earthquake-induced submarine landslides. We show that this model captures complex interactions between saturated sediment, seawater, and structure, so we can use the model to estimate the impact of potential submarine landslides on offshore structures.
Graphical Abstract

Figure 20. Seawater, air (Fluid Mechanics); Sediment (Soil Mechanics); Porous Mechanics; Fluid-Structure Interaction; Contact Mechanics; Structure (Solid Mechanics); soil-fluid-structure interaction; Pore water pressure; Velocity; Shear strain; Application to earthquake-induced submarine landslide.
Highlights
MPMICE: A hybrid MPM-CFD model for simulating coupled problems in porous media. Application to earthquake-induced submarine landslides
Quoc Anh Tran, Gustav Grimstad, Seyed Ali Ghoreishian Amiri
MPMICE is introduced for multiphase flow in porous media.
Material Point method allows to model large deformation of non-isothermal porous media.
ICE (compressible multi-material CFD formulation) allows to stablize pore water pressure and turbulent flow.
MPMICE is validated and apply to simulate the earthquake-induced submarine landslide.
Nomenclature
General variables
| Variable | Dimensions | Description |
| Representative volume | ||
| Porosity | ||
| Total stress tensor | ||
| Time increment | ||
| Body force | ||
| Constant volume specific heat | ||
| Drag forces in momentum exchange term | ||
| Internal forces | ||
| External forces | ||
| Heat exchange term | ||
| Weighting function | ||
| Gradient of weighting function |
Table 2.
Solid phase
| Variable | Dimensions | Description |
| Solid mass | ||
| Solid density | ||
| Solid volume fraction | ||
| Bulk Solid density | ||
| Solid Position vector | ||
| Solid Velocity vector | ||
| Solid Acceleration vector | ||
| Effective Stress tensor | ||
| Strain tensor | ||
| Solid Internal energy per unit mass | ||
| Solid Temperature | ||
| Solid Deformation gradient | ||
| Solid Volume |
Table 3.
Fluid phase
| Variable | Dimensions | Description |
| Fluid mass | ||
| Fluid density | ||
| Fluid volume fraction | ||
| Bulk Fluid density | ||
| Fluid Velocity vector | ||
| Fluid stress tensor | ||
| Fluid isotropic pressure | ||
| Fluid shear stress tensor | ||
| Fluid Internal energy per unit mass | ||
| Fluid Temperature | ||
| Fluid Specific volume | ||
| Thermal expansion | ||
| Fluid viscosity | ||
| Fluid Volume |
Table 4.
Superscript
| Variable | Dimensions | Description |
| Current time step | ||
| Lagrangian values | ||
| Next time step |
Table 5.
Subscript
| Variable | Dimensions | Description |
| Cell-centered quantity | ||
| Particle quantity | ||
| Node quantity | ||
| Face-centered quantity | ||
| Left and Right cell faces |
Table 6.
Introduction
Many geological natural processes and their interactions with man-made structures are influenced by soil-fluid-structure interactions. The prediction of these processes requires a tool that can capture complex interactions between soil, fluid, and structure, such as the process of submarine landslides. Indeed, The offshore infrastructure as well as coastal communities may be vulnerable to submarine landslides. Submarine landslides contain three stages: triggering, failure, and post-failure. Erosion or earthquakes can trigger slope failures in the first stage. Following the failure, sediments move quickly after the post-failure stage. In other words, solid-like sediments will behave like a fluid after failure. This phase transition makes the simulation of submarine landslides a challeging task.
Due to this phase transition, submarine landslide can be modeled by either the Computational Fluid Dynamics (CFD) or the particle-based methods. For simulating submarine slides, CFD methods solve governing equations in a full-Eulerian framework [1, 2, 3, 4] with interface capturing techniques. While CFD can handle complex flows (such as turbulent flows), it cannot account for the triggering mechanism of submarine landslides because it is not straightforward to consider ’soil constitutive laws’ of sediment materials in the Eulerian framework. In contrast, particle-based methods can overcome this problem by using the Lagrangian framework. These methods have been extensively used to simulate landslides, like Material Point Method (MPM) [5], Smooth Particle Hydro Dynamics [6], Particle Finite Element Method [7], or Coupled Eulerian Lagrangian Method [8]. For simplicity, these simulations adopt total stress analysis which neglects the pore pressure development which is key factor triggering slope failures.
Recent developments in particle-based methods model the coupling of fluid flows in porous media by sets of Lagrangian particles. For the MPM family, it is the double-point MPM ([9, 10, 11]) where fluid particles and solid particles are overlapped in a single computational grid. Even if fluid flows are considered, particle-based methods have numerical instability in modeling the fluid flow, which requires additional numerical treatments such as the B-bar method [9], null-space filter [12], or least square approximation [13, 14]. Indeed, CFD is a more optimal option for complex fluid flows especially dealing with large distortions of continuum fluid media. Therefore, it could be ideal to combine the CFD with particle-based methods. More than 50 particle-based methods have been developed to solve large deformations of solids over the last two decades [15], but the MPM appears to be the best candidate for coupling with the CFD. Because MPM incorporates a stationary mesh during computation, just like CFD. As such, both MPM and CFD can be coupled naturally in a unified computational mesh.
A numerical method for simulating soil-fluid-structure interaction (Figure 1 (PDF p. 9)) involving large deformations, is presented in this work in order to simulate the interaction between sediment (soil), seawater (fluid) and offshore structures (structure) namely MPMICE (Figure 2). In the MPMICE, the Material Point Method (MPM) is coupled with the Implicit Continuous Eulerian (ICE). The MPM method is a particle method that allows the porous soil to undergo arbitrary distortions. The ICE method, on the other hand, is a conservative finite volume technique with all state variables located at the cell center (temperature, velocity, mass, pressure). An initial technical report [16] at Los Alamos National Laboratory provided the theoretical and algorithmic foundation for the MPMICE, followed by the MPMICE development and implementation in the high-performance Uintah computational framework for simulating fluid-structure interactions [17]. This paper primarily contributes further to the development of the MPMICE for analyzing the soil-fluid-structure interaction, since sediment should be considered as a porous media (soil) and not as a solid to capture the evolution of the pore water pressure. Baumgarten et al. [18] made the first attempt at coupling the Finite Volume Method with the MPM for the simulation of soil-fluid interaction. In contrast to the mentioned work, we use implicit time integration for the multi phase flows instead of explicit time integration for the single-phase flow.

Figure 1 (PDF p. 9). Interaction between soil-fluid-structure

Figure 2. Coupling of soil-water-structure interaction using MPMICE
Theory and formulation
This section lay out the theoretical framework for the MPMICE model. We use the common notation of the continuum mechanics with vector and tensor denoted simply by using bold font and scalar denoted by using normal font. The notation are shown in Nomenclature.
Assumptions
The following assumptions are made for the MPMICE model.
Solid phases (MPM) are described in a Lagrangian formulation while fluid phases (ICE) are described in an Eulerian formulation in the framework of continuum mechanics and mixture theory.
Solid grains are incompressible while the fluid phases are compressible. Solid’s thermal expasion is negligible.
There is no mass exchange between solid and fluid phases.
Terzaghi’s effective stress is valid.
Governing equations
A representative element volume is decomposed by two domains: solid domains and fluid domains . Then, all domains are homogenized into two overlapping continua. Considering the volume fraction of solid and fluid with the true (or Eulerian) porosity of the representative element volume, the average density of solid and fluid phases are defined as:
The mass of solid and fluid phases are:
Reviewing the Terzaghi’s effective stress concept for the saturated porous media, the total stress is calculated by:
The balance equations are derived based on the mixture theory. The representative thermodynamic state of the fluid phases are given by the vector which are mass, velocity, internal energy, temperature, specific volume. The representative state of the solid phases are given by the vector which are mass, velocity, internal energy, temperature, effective stress and pore water pressure. The derivation is presented in detail in the Appendix.
Mass Conservation
The mass balance equations for both fluid (e.g., water, air) and solid phases are:
Solving the mass balance equation leads to:
Momentum Conservation
The momentum balance equation for the fluid phases (e.g., water, air) are:
The momentum balance equation for the solid phases are:
Energy Conservation
The internal energy balance equation for the fluid phases (e.g., water, air) are:
The internal energy balance equation for the solid phase is:
where is the specific heat at constant volume of the solid materials.
Closing the systems of equations, the following additional models are needed:
A constitutive equation to describe the stress - strain behaviour of solid phase (computing effective stress ).
Optional turbulent model to compute the viscous shear stress .
Frictional forces for the contact for soil-structure interaction between solid/porous materials with the friction coefficient .
Exchange momentum models (computing drag force ) for interaction between materials.
Energy exchange models (computing temperature exhange term ) for interaction between materials.
An equation of state to establish relations between thermodynamics variables of each fluid materials .
Four thermodynamic relations for the equation of states are:
Constitutive soil model
As a result of the explicit MPM formulation, we can derive the constitutive law in the updated Lagrangian framework of ”small strain - large deformation”. Therefore, the rotation of the particles (representative element volume) is manipulated by rotating the Cauchy stress tensor. First, the deformation gradient is decomposed into the polar rotation tensor and sketch tensor as
Then, before calling the constitutive model, the stress and strain rate tensor are rotated to the reference configuration as
Using the constitutive model with the input tensors , to compute the Cauchy stress tensor at the advanced time step then rotating it back to current configuration
In this paper, we adopt the hyper-elastic Neo Hooken model for the structure materials and additionally Mohr-Coulomb failure criteria for the soil (porous media) materials. The Cauchy stress of the hyper-elastic Neo Hookean model can be written as:
where and are bulk and shear modulus ad J is the determinant of the deformation gradient . And the yield function f and flow potentials g of the Mohr-Coulomb can be written as:
where the , and are cohesion and friction angle and dilation angle. and are maximum and minimum principal stress.
Turbulent model
The turbulent effect is modelled using a statistical approach namely large-eddy simulation. In this approach, the micro-scale turbulent influence in the dynamics of the macro-scale motion is computed through simple models like Smagorinsky model. In the Smagorinsky mode, the residual stress tensor is:
where the the strain rate tensor is given by
and the effective viscosity is sum of molecular viscosity and turbulent viscosity in which the turbulent viscosity is calculated by
where is the Smagorinsky constant and is the grid size that defines the subgrid length scale.
Frictional force for soil-structure interaction
MPMICE includes a contactlaw for the interaction between soil and structure using the first Coulomb friction contact for MPM presented by Bardenhagen et al. [19]. The magnitude of the friction force at the contact depends on the friction coefficient and the normal force computed from the projection of the contact force in the normal direction.
The contact determines whether the soil is sliding or sticking to the structure by comparing the friction force with the sticking force computed from the projection of the contact force in the tangent direction:
Frictional sliding between solid materials also generates dissipation and the work rate generated from the sliding can be calculated as:
where d is the sliding distance which can computed based on the sliding velocity between two materials.
Momentum and Energy exchange model
Currently, the energy exchange coefficient is assumed to be constant for the sake of simplicity. Then the energy exchange can be written as:
On the other hand, the drag force can be calculated as:
For the momentum exchange between fluid flows and porous media, we assume that the drag force depends on the average grain size of the grains , the porosity , the fluid viscosity , and is proportional to the relative velocities of soil grains and fluid . Based on recent investigation of CFD simulations of fluid flow around mono- and bi-disperse packing of spheres for and [20]. The drag force is given by:
where Reynolds number are computed as:
The function can be calculated as:
where the low Reynold coefficient is:
When validating the model with analytical solution, it requires to know the hydraulic conductivity. In such case, we convert the equation (25) to Kozeny-Carman formula by assuming , then the hydraulic conductivity will be expressed as .
Solving momentum and energy exchange with an implicit solver
The derivation of the implicit integration for the momentum exchange is presented in the Appendix’s section ’Momentum and energy exchange with an implicit solver’. The linear equations for multi phases i,j=1:N has the form:
where the intermediate velocity for fluid phases and for solid/porous phases can be calculated by
Also, the momentum exchange coefficient can be computed at every time step as and with the coefficient depending on the different type of interactions (see Figure 3) as for example:

Figure 3. Momentum exchange coefficient between materials
The drag force is set to zero in soil-structure interactions, and instead the frictional force is computed.
As a result of fluid-structure interaction, the momentum exchange coefficient should be extremely high when the solid material points are considered to be zero-porosity/zero-permeability.
In the case of soil-fluid interaction, the drag force is calculated using the equation (25). Considering that air has a much lower viscosity than water, its drag force is much lower than the drag force of water in a pore.
A momentum exchange coefficient of 1E5 is applied between multiphase flows. This value is far higher than reality [21], but it is necessary to have enough numerical stability to conduct simulations in the numerical example.
Similar approach applied for the energy exchange term leading to:
with is the energy exchange coefficient.
Equation of state for fluid plases
The equation of state establishes relations between thermodynamics variables . The choice of the equation of state depends on the types of the fluid materials. For example, for the air, it is possible to assume the equation of state for the perfect gas which obeys:
where is the gas constant. For the water, a simple linear equation of state is in the following form:
where reference pressure , reference temperature , reference density , the bulk modulus of water , and the water thermal expansion . Equation (31) matches well with the state of the water (see Figure 4).

Figure 4. Equation of state of water
Numerical implementation
The fluid phases are discretized in the grid with the state variables stored at the centroid of the cells while the solid phase is discretized in the particles with the state variables . In the Material Point Method, we use the generalized interpolation technique [22] using the weight function as a convolution of a grid shape function in a nodal domain and a characteristic function in a particle domain with the volume as follows:
where the volume of the material point p can be calculated as:
The charateristic function is the Heaviside function as if , otherwise 0 (see Figure 5). For the interpolation of the centroid of the cell, the linear basis function is used as:

Figure 5. GIMP weighting function (red as a convolution of the linear basis shape function (green) and the charateristic function (blue)
The time discretization are solved using the following steps:
Interpolation from Solid Particle to Grid
The nodal values of the solid state (mass, velocity, temperature, volume) are:
The nodal internal forces is calculated by
The nodal external forces and extra momentum from contact forces are computed here. The nodal velocity and nodal temperature are applied boundary conditions. Then we compute the solid cell variables as:
Compute equation of state for fluid phase
Considering the total fluid material volume of a cell is:
Then we need to find which allows each fluid materials obey their equation of states but also allow mass of all fluid materials to fill the entire the pore volume without ongoing compression or expansion following the condition:
Then, we can use the Newton-Raphson iteration to find the value of which satisfies the equation (38, 39) and each equation of states of each fluid materials.
Compute faced-centered velocity
Following the derivation in the Appendix: Advanced Fluid Pressure, we first compute the fluid face-centered velocity as
The equation (40) is discretized in three dimension (noted that ), for example the discretized equation in the x direction is
The face-centered solid velocity can be calculated as
The equation (42) is discretized in three dimension(noted that with ), for example the discretized equation in the x direction is
Computing the modified faced-centered velocity considering the momentum exchange (see the Appendix: Momentum exchange with an implicit solve)
Solving the linear equation below to obtain the increment of velocity with i,j = 1 : as:
Compute faced-centered temperature
Similar to the velocity, the faced temperature is computed as:
Computing the modified faced-centered temperature considering the energy exchange (see the Appendix: Momentum and energy exchange with an implicit solver)
Solving the linear equation below to obtain the increment of velocity with i,j = 1 : as:
Compute fluid pressure (implicit scheme)
For single phase flow, the increment of the fluid pressure can be computed as:
For multi-phase flows, the increment of the fluid pressure of the mixture can be computed as:
where . Then, the fluid pressure at cell center is:
Finally, the faced-centered advanced fluid pressure is
Compute viscous shear stress term of the fluid phase
This part compute the viscous shear stress for a single vicous compressible Newtonian fluid and optionally shear stress induced by the turbulent model.
Compute nodal internal temperature of the solid phase
The nodal internal temperature rate is computed based on the heat conduction model
where is the mechanical work rate computed from the constitutive model with is the plastic strain, is the work rate compted from the contact law due to the frictional sliding between solid materials. The heat flux is with being the thermal conductivity of the solid materials.
Compute and integrate acceleration of the solid phase
After interpolating from material points to the nodes, the nodal acceleration and velocity are calculate by
Compute Lagrangian value (mass, momentum and energy)
For the fluid phase, the linear momentum rate, the energy rate are
The Lagrangian value of the mass, linear momentum and energy of fluid phases without momentum exchange are
For the solid phase, the Lagrangian value of the linear momentum and energy of solid phase are
To consider the momentum exchange, the Lagrangian velocity is modified as
where the cell-centered intermediate velocity can be calculated by
And the increment of the velocity , can be computed by solving the linear equation with i,j = 1 : as:
To consider the energy exchange, the Lagrangian temperature is modified as
where the cell-centered intermediate temperature can be calculated by
And the increment of the velocity can be computed by solving the linear equation with as:
Finally, we obtain the cell-centered solid acceleration and temperature rate as
Compute Lagrangian specific volume of the fluid phase
To compute the Lagrangian value of the specific volume of the fluid phase, we need to compute the Lagrangian temperature rate as below
As such, the Lagrangian specific volume rate is:
where and . Finally, the Lagrangian specific volume is
Compute advection term and advance in time
The time advanced mass, linear momentum, energy and specific volume are:
Finally, the state variables of the fluid phases of the next time step are
Interpolate from cell to node of the solid phase
First we interpolate the acceleration, velocity and temperature rate to the node
Then the boundary condition and contact forces are applied to the nodal velocity and the acceleration is modified by
Update the particle variables
The state variables of the solid phase (velocity, position, velocity gradient, temperature, temperature gradient, deformation gradient, volume) are updated here
Finally, the effective stress is updated from the constitutive model and the pore water pressure is interpolated from the cell as:
Numerical examples
All input files and the analytical calculations in this section are provided in the Github repository for the reproduction of the numerical results.
Fluid Flow through isothermal porous media
Fluid flow through porous media is important in many engineering disciplines, like predicting water flow in soil. Fluid flow velocity in one dimension can be calculated from the porous media’s hydraulic conductivity as:
If the Carman-Kozeny formula is adopted , the hydraulic conductivity will be expressed as . Then, the analytical formula of average velocity in one dimension through the porous media is:
Our numerical model is validated by modeling fluid flow through a 1m long porous media. This fluid has water properties (bulk modulus is 2GPa, density is 998 kg/m3 at 5 degrees Celsius and 10325 Pa (1atm) pressure, dynamic viscosity is 1mPa s). The porous media is modeled by elastic material with Young’s modulus is 10 MPa, Poisson’s ratio is 0.3, and density is 2650 kg/m3. The volume fraction of porous media is [0.6, 0.62, 0.66, 0.68, 0.7] and the average grain diameter is 1mm. The model is discretized in 20 finite element and the porous media in 10 finite element with 1 material point per element. The pressure gradient is applied with three different value [0.25, 0.5, 1] atm. Figure 6 shows a good agreement of fluid flow prediction between the theory and the model.

Figure 6. (a) Discretization of the model (b) Comparision between analytical model and simulation. Figure 6: Numerical results of the fluid flow through isothermal porous media
Isothermal consolidation
A common benchmark fo a fully saturated porous media is the simulation of one-dimensional consolidation. Using the Carman-Kozeny formula, the time-dependent pressure can be caluated as:
where the consolidation rate , the consolidation coefficient and the Oedometer modulus . Our numerical model is validated by modeling the consolidation of a 1m column. This fluid has water properties (bulk modulus is 2GPa, density is 998 kg/m3 at 5 degrees Celsius and 101325 Pa (1atm) pressure, dynamic viscosity is 1mPa s). The porous media is modeled by elastic material with Young’s modulus is 10 MPa, Poisson’s ratio is 0.3, and density is 2650 kg/m3. The volume fraction of porous media is 0.7 which is equivalent to the porosity of 0.3 and the average grain diameter is 1mm. The model is discretized in 100 finite element with 1 material point per element. The external pressure applies to the top of the column is 10 kPa. Figure 7 shows a good agreement of fluid flow prediction between the theory and the model.

Figure 7. Compasion between analytical solution and numerical solution

Figure 8. Model schematic [23]
Thermal induced cavity flow
Another benchmark is the thermal induced cavity flow in porous media. Temperature and velocity distributions are calculated for a square non-deformable saturated porous media The top and bottom walls are insulated, and the left and right walls are at fixed temperatures differing by 1 degree. The fluid motion at stead state are cavity flow due to the temperature induced density variation.
The numerical is validated by comparing with the numerical solution of the finite element method. The fluid has water properties (bulk modulus is 2GPa, density is 998 kg/m3 at 5 degrees Celsius and 10325 Pa (1atm) pressure, dynamic viscosity is 1 mPa s). The porous media is modeled by non deformable material, and density is 2500 kg/m3. The specific heat capacity of the water and porous skeleton are 4181 J/kg.K and 835 J/kg.K respectively. The thermal conductivity of the water and porous skeleton are 0.598 W/m.K and 0.4 W/m.K. The volume fraction of porous media is 0.6 which is equivalent to the porosity of 0.4 and the average grain diameter is 1mm. The model is discretized in 20 x 20 finite element with 4 material point per element. Figure 9 shows a good agreement of numerical results of the model compared with the numerical solution of the finite element method.

Figure 9. (a) MPMICE model (b) FEM model [23]. Comparision between MPMICE model and FEM model
Underwater debris flow
The numerical example is validated by Rzadkiewicz et al.’s experiment on submarine debris flow [24]. During the experiment, sand in a triangular box is released and then slides along a rigid bed inclined 45 degrees under water, see Figure 10.

Figure 10. Model schematic
In the numerical model, the material properties are selected based on the experiment by Rzadkiewicz et al [24]. Sand has a saturated density of 1985 and yield stress of 200 Pa. Young’s modulus has little effect on debris flow run-out because of the extreme large deformation of the debris. Therefore, we select 50 MPa Young’s modulus with 0.25 Poisson’s ratio. The rigid bed is much stiffer with bulk modulus and shear modulus of Pa and Pa. Under gravity, the density of the water at the surface is 999.8 at the pressure of 1 atm. At the top boundary, the air has a density of 1.17 at the atmospheric pressure of 1 atm. At 5 Celcius degrees, air and water have viscosity of mPa s and 1 mPa s respectively. The numerical parameters used in this example are presented in Table 1 (PDF p. 33).
| Materials | Bulk modul (Pa) | Shear modul (Pa) | Density (kg/m3) | Temp (C) | Dynamic viscosity (Pa s) | Yield stress (Pa) |
| Water (at surface) | 2.15e9 | - | 999.8 | 5 | 855e-6 | - |
| Air (at top boundary) | - | - | 1.177 | 5 | 18.45e-6 | - |
| Sand (porous media) | 8.33e6 | 20e6 | 1985 | 5 | - | 200 |
| Rigid bed (solid) | 117e7 | 43.8e7 | 8900 | 5 | - | - |
Table 1 (PDF p. 33). Numerical parameters for the underwater submarine debris
On all boundary faces, the Dirichlet boundary condition is imposed for velocity ( m/s) and temperature ( Celcius degrees), while the Neuman boundary condition is imposed at the top boundary for pressure ( kPa) and density ( ). For the background mesh, there are
cells. In each cell of the debris flow and rigid bed, there are material points.
Figure 11a and Figure 11b show snapshots of the debris flow sliding in the plane at 0.4 s and 0.8 s. Our simulations match the computed results from Rzadkiewicz et al. [24]. The model also captures typical hydroplaning mechanism of the underwater debris flow (hydroplaning means the debris flow is lifted up and no longer in contact with the bottom layer). The elevation of the free surface at 0.4 s and 0.8 s is compared between our proposed method and other methods in Figure 12. Once again, our computed results were consistent with both the experiment and others computational results [7]. Unlike other computational models based on total stress analysis, the proposed model based on the effective stress analysis which allows to analyze the water pressure and temperature in the debris flow.

Figure 11. Simulation of underwater debris flow

(a) (a) 0.4 seconds

(b) (b) 0.8 seconds
Figure 12. Simulation of underwater debris flow
We also explore the difference between underwater debris flow and saturated debris flow in terms of interacting with obstacle. Figure 13 shows the snapshot of the simulations of underwater and saturated debris flow. The saturated debris flow (see Figure 13a) behaves like frictional flow as grain have contact forces with each other. On the other hand, the underwater debris flow (see Figure 13b) behaves like turbulent flow as grains are separated from each other and exhibit no contact forces between grains.

(a) (a) saturated debris flow using MPM

(b) (b) underwater debris flow using MPMICE
Figure 13. Simulation of underwater debris flow
Earthquake-induced submarine landslides

(a) acceleration

(b) velocity
Figure 15. Ground acceleration profile, frequency of 2Hz and magnitude of 1g
In the final example, we perform numerical analysis of the earthquake induced submarine landslides. A plane strain model with the slope under water is shown in Figure 14. A 20m high slope with slope gradient of 45 degrees is placed in a horizontal and vertical structure which was used to be a skaing table to apply earthquake loading. We simplify the earthquake loading by simulating the ground shaking for 20 seconds with the peak ground acceleration of 1g and the frequency of 2Hz (Figure 15a). The ground motion is applied in terms of velocity (Figure 15b). The earthquake of this magnitude can occured typically for the earthquake of magnitude of more than 6.

Figure 14. Numerical model of the earthquake-induced submarine landslide
A non-associated Mohr-Coulomb model is used for the soil. The soil grain has the density of 2650 , Young’s modulus of 10 kPa and Poisson’s ratio of 0.3 and zero cohesion. The mobilized friction angle is governed following the softening curve (see Figure 16) with the peak friction angle of 45 degrees and the residual friction angle of 10 degrees. The porosity is 0.3 and the average grain size of the soil is around 0.1 to mimic the undrained behavior. The mobilized dilatancy angle is calculated from the Row-stress dilatancy as follow:

Figure 16. Mobilized friction angle in Mohr Coulomb model
The solid plane is modeled as a rigid body acted as a shaking table. The contact between horizontal plane and the sand is the frictional contact with the friction coefficient of 0.1. No artificial damping is applied in the simulation. The contact between vertical plane and the sand is considered to be smooth with zero friction coefficient. Under gravity, the density of the water at the surface is 999.8 at the pressure of 1 atm. At the top boundary, the air has a density of 1.17 at the atmospheric pressure of 1 atm. At 5 Celcius degrees, air and water have viscosity of mPa s and 1 mPa s respectively. On all boundary faces, the symmetric boundary condition is imposed, while the Neuman boundary condition is imposed at the top boundary for pressure ( kPa) and density ( ). The mesh size is 0.25 x 0.25m with 300852 element cells and 142316 material points. The simulation takes a couple of hours to perform 60 seconds of the simulation using 4096 CPUs.
We demonstrate the entire process and the mechanism of the earthquake-induced submarine landslides by showing the shear strain (Figure 17), the pore water ressure in atm (Figure 18) and the velocity (Figure 19). The failure mechanism can be charaterized as the progressive failure mechanism. Here are some numerical observation:

Figure 17. Shear strain during the earthquake-induced submarine landslides

Figure 18. pore water pressure during the earthquake-induced submarine landslides

Figure 19. Velocity during the earthquake-induced submarine landslides
At the initial of the seismic event, the seismic loading triggers the first slide at 3 seconds. At 4 seconds, the debris start to move with the maximum speed of around 2-3 m/s with multiple shear band developed in the slope. The wave generated from the submarine slide is around 2-3m towards the slide direction.
When the onset of the shear band occurs in the slope (for example at 4 seconds and 20 seconds), the negative excess pore water pressure is developed along this shear band with pore water pressure is under 1atm. This is a typical dilatancy behavior when the soil is sheared rapidly in the undrained behavior.
When the seismic loading ends at 23 seconds, the last shear band is mobilized and the slope soon reaches to the final deposition. No more progressive failure developed in the slope. The turbulent flow developed as the interaction between debris flow and seawater.
Overall, we show the completed process of the earthquake-induced submarine landslides involving (1) earthquake triggering mechanism, (2) the onset of the shear band with the delvelopment of negative excess pore water pressure, (3) progressive failure mechanism, (4) submarine landslide induced wave to final deposition.
Conclusions
We have presented a numerical approach MPMICE for the simulation of large deformation soil-fluid-structure interaction, emphasizing the simulation of the earthquake-induced submarine landslides. The model uses (1) the Material Point Method for capturing the large deformation of iso-thermal porous media and solid structures and (2) Implicit Continuous Eulerian (compressible, conservative multi-material CFD formulation) for modeling the complex fluid flow including turbulence. This model is implemented in the high-performance Uintah computational framework and validated against analytical solution and experiment. We then demonstrate the capability of the model to simulate the entire process of the earthquake induced submarine landslides.
Acknowledgements
The authors gratefully acknowledge Dr. James Guilkey and Dr. Todd Harman from the University of Utah for sharing the insight on the theoretical framework of MPMICE which this work is based. This project has received funding from the European Union’s Horizon 2020 research and innovation program under the Marie Skłodowska-Curie Actions (MSCA) Individual Fellowship (Project SUBSLIDE “Submarine landslides and their impacts on offshore infrastructures”) grant agreement 101022007. The authors would also like to acknowledge the support from the Research Council of Norway through its Centers of Excellence funding scheme, project number 262644 Porelab. The computations were performed on High Performance Computing resources provided by UNINETT Sigma2 - the National Infrastructure for High Performance Computing and Data Storage in Norway.
Equation derivation
Before deriving the governing equation, we define the Lagrangian derivative for a state variable as:
we use some definition following [16] as below:
Then, the rate of volume with incompressible solid grains are calculated as below:
Evolution of porosity
Solving the solid mass balance equation (4) with the definition of solid mass in equation (2), it leads to the rate of porosity as below:
The soil grains are assumed to be incompressible, therefore, term 2 in the right hand side is zero.
Dividing all terms with with the equation , it leads to:
Momentum conservation
The linear momentum balance equation for the fluid phases based on mixture theory is:
On the right hand sand, the first term is the divergence of partial fluid phase stress, the third term is the body force, the fourth term is the drag force (momentum exchange) and the fifth term is the buoyant force described in [25] for the immiscible mixtures. The buoyant force is in the form:
As a result, the linear momentum balance equation for the fluid phases becomes:
The Reynolds stress component can be included in the term to consider the turbulent effects if needed. To derive the linear momentum balance equation for the solid phase, we begin with the linear momentum balance equation for the mixture as:
Combining Terzaghi’s equation (3) and subtracting both sides with equation (106), we obtain the linear momentum balance equation for the solid phase as:
Here the stems from the soil-structure interaction following the contact law between the soil/structure interfaces.
Energy conservation
We adopt the general form of the total energy balance equation for the porous media from [26], the total energy balance equations for the fluid phases are:
Applying the product rule , the left hand side of equation (109) becomes:
Combining equations (106), (109), (110), we obtain the final form of the internal energy balance equation for the fluid phases as:
On the right hand side, the terms include the average pressure-volume work, the average viscous dissipation, the thermal transport and the energy exchange between solid and fluid respectively. The heat flux is with being the thermal conductivity coefficient. To derive the internal energy balance equation for the solid phase, we introduce the rate of the internal energy for the thermoelastic materials as a function of elastic strain tensor and temperature :
is the specific heat at the constant volume of the solid materials. The total energy balance equation for the mixture based on [26] can be written as:
Subtracting equation (113), (112) to equations (109) and (108), we obtained the internal energy balance equation for solid phase as:
On the right hand side, the terms include the work rate from frictional sliding between solid materials , thermal transport and energy exchange between solid and fluid respectively. The heat flux is with being the thermal conductivity of the solid materials, the mechanical work rate computed from the constitutive model with is the plastic strain tensor, . By subtracting the term , we get the final form of the energy balance equation as:
Advanced Fluid Pressure
The discretization of the pressure equation begins with the Lagrangian face-centered velocity and the equation for the pressure
The divergence of the equation (116) with is
To solve this equation, we define the face-centered intermediate velocity as:
The divergence of the equation (119) is
Combining equations (117, 118, 120), it leads to
When the fluid is incompressible, approaches to zero and the equation (121) becomes the Poisson’s equation for the incompressible fluid flow.
Momentum and Energy exchange with an implicit solver
Considering the fluid momentum balance equation as
Assuming we get
As defined in the section ’Advanced Fluid Pressure’, the face-centered intermediate fluid velocity leading to
Considering the solid momentum balance equation as
We define the face-centered intermediate solid velocity as leading to
Combining equation (124) and (126) we get
Rearranging the equation (127), it leads to the linear system of equations
Solving this linear equations with and with K is the momentum exchange coefficient. Similar derivation can be performed to computed the cell-center velocity increment leading to
with and and the cell-centered intermediate velocity can be calculated by
For generalize multi materials i,j = 1:N, the linear equations is in the form:
Similar approach applied for the energy exchange term leading to:
with is the energy exchange coefficient.
References
- [1]D. Yuk, S. Yim, P.-F. Liu, Numerical modeling of submarine mass-movement generated waves using RANS model, Computers and Geosciences 32 (7) (2006) 927–935. doi:10.1016/j.cageo.2005.10.028. URL https://doi.org/10.1016/j.cageo.2005.10.028
- [2]S. Glimsdal, J.-S. L’Heureux, C. B. Harbitz, F. Løvholt, The 29th january 2014 submarine landslide at statland, norway—landslide dynamics, tsunami generation, and run-up, Landslides 13 (6) (2016) 1435–1444. doi:10.1007/s10346-016-0758-7. URL https://doi.org/10.1007/s10346-016-0758-7
- [3]S. Abadie, D. Morichon, S. Grilli, S. Glockner, Numerical simulation of waves generated by landslides using a multiple-fluid navier–stokes model, Coastal Engineering 57 (9) (2010) 779–794. doi:10.1016/j.coastaleng.2010.03.003. URL https://doi.org/10.1016/j.coastaleng.2010.03.003
- [4]M. Rauter, S. Viroulet, S. S. Gylfadóttir, W. Fellin, F. Løvholt, Granular porous landslide tsunami modelling – the 2014 lake askja flank collapse, Nature Communications 13 (1) (Feb. 2022). doi:10.1038/s41467-022-28296-7. URL https://doi.org/10.1038/s41467-022-28296-7
- [5]Q.-A. Tran, W. Solowski, Generalized interpolation material point method modelling of large deformation problems including strain-rate effects – application to penetration and progressive failure problems, Computers and Geotechnics 106 (2019) 249–265. doi:10.1016/j.compgeo.2018.10.020. URL https://doi.org/10.1016/j.compgeo.2018.10.020
- [6]T. Capone, A. Panizzo, J. J. Monaghan, SPH modelling of water waves generated by submarine landslides, Journal of Hydraulic Research 48 (sup1) (2010) 80–84. doi:10.1080/00221686.2010.9641248. URL https://doi.org/10.1080/00221686.2010.9641248
- [7]X. Zhang, E. Oñate, S. A. G. Torres, J. Bleyer, K. Krabbenhoft, A unified lagrangian formulation for solid and fluid dynamics and its possibility for modelling submarine landslides and their consequences, Computer Methods in Applied Mechanics and Engineering 343 (2019) 314–338. doi:10.1016/j.cma.2018.07.043. URL https://doi.org/10.1016/j.cma.2018.07.043
- [8]R. Dey, B. C. Hawlader, R. Phillips, K. Soga, Numerical modelling of submarine landslides with sensitive clay layers, Géotechnique 66 (6) (2016) 454–468. doi:10.1680/jgeot.15.p.111. URL https://doi.org/10.1680/jgeot.15.p.111
- [9]S. Bandara, K. Soga, Coupling of soil deformation and pore fluid flow using material point method, Computers and Geotechnics 63 (2015) 199–214. doi:10.1016/j.compgeo.2014.09.009. URL https://doi.org/10.1016/j.compgeo.2014.09.009
- [10]A. P. Tampubolon, T. Gast, G. Klár, C. Fu, J. Teran, C. Jiang, K. Museth, Multi-species simulation of porous sand and water mixtures, ACM Transactions on Graphics 36 (4) (2017) 1–11. doi:10.1145/3072959.3073651. URL https://doi.org/10.1145/3072959.3073651
- [11]A. S. Baumgarten, K. Kamrin, A general constitutive model for dense, fine-particle suspensions validated in many geometries, Proceedings of the National Academy of Sciences 116 (42) (2019) 20828–20836. doi:10.1073/pnas.1908065116. URL https://doi.org/10.1073/pnas.1908065116
- [12]Q.-A. Tran, W. Sołowski, Temporal and null-space filter for the material point method, International Journal for Numerical Methods in Engineering 120 (3) (2019) 328–360. doi:10.1002/nme.6138. URL https://doi.org/10.1002/nme.6138
- [13]X. Zheng, F. Pisanò, P. J. Vardon, M. A. Hicks, An explicit stabilised material point method for coupled hydromechanical problems in two-phase porous media, Computers and Geotechnics 135 (2021) 104112. doi:10.1016/j.compgeo.2021.104112. URL https://doi.org/10.1016/j.compgeo.2021.104112
- [14]Q. A. Tran, W. Sołowski, M. Berzins, J. Guilkey, A convected particle least square interpolation material point method, International Journal for Numerical Methods in Engineering 121 (6) (2019) 1068–1100. doi:10.1002/nme.6257. URL https://doi.org/10.1002/nme.6257
- [15]J.-S. Chen, M. Hillman, S.-W. Chi, Meshfree methods: Progress made after 20 years, Journal of Engineering Mechanics 143 (4) (Apr. 2017). doi:10.1061/(asce)em.1943-7889.0001176 URL https://doi.org/10.1061/(asce)em.1943-7889.0001176
- [16]B. A. Kashiwa, A multifield model and method for fluid-structure interaction dynamics, Tech. Rep. LA-UR-01-1136, Los Alamos National Laboratory (2001).
- [17]J. Guilkey, T. Harman, B. Banerjee, An eulerian–lagrangian approach for simulating explosions of energetic devices, Computers and Structures 85 (11) (2007) 660 – 674. doi:https://doi.org/10.1016/j.compstruc.2007.01.031. URL http://www.sciencedirect.com/science/article/pii/S0045794907000429
- [18]A. S. Baumgarten, B. L. Couchman, K. Kamrin, A coupled finite volume and material point method for two-phase simulation of liquid–sediment and gas–sediment flows, Computer Methods in Applied Mechanics and Engineering 384 (2021) 113940. doi:10.1016/j.cma.2021.113940. URL https://doi.org/10.1016/j.cma.2021.113940
- [19]S. Bardenhagen, J. Guilkey, K. Roessig, J. Brackbill, W. Witzel, J.C.Foster, An improved contact algorithm for the material point method and application to stress propagation in granular material, CMES-Computer Modeling in Engineering and Sciences 2 (2001) 509–522. doi:10.3970/cmes.2001.002.509.DOI
- [20]J. A. M. K. R. Beetstra, M. A. van der Hoef, Drag force of intermediate reynolds number flow past mono- and bidisperse arrays of spheres, Fluid Mechanics and Transport Phenomena 53 (2) (2007) 489 – 501. doi:https://doi.org/10.1002/aic.11065.
- [21]R. W. Stewart, The air-sea momentum exchange, Boundary-Layer Meteorology 6 (1-2) (1974) 151–167. doi:10.1007/bf00232481. URL https://doi.org/10.1007/bf00232481
- [22]S. G. B. amd E. M. Kober, The generalized interpolation material point method, Computer Modeling in Engineering and Sciences 5 (2004) 477–496. doi:10.1016/j.compgeo.2021.104199. URL https://doi.org/10.3970/cmes.2004.005.477
- [23]S. G. Amiri, E. Taheri, A. Lavासन, A hybrid finite element model for non-isothermal two-phase flow in deformable porous media, Computers and Geotechnics 135 (2021) 104199. doi:10.1016/j.compgeo.2021.104199. URL https://doi.org/10.1016/j.compgeo.2021.104199
- [24]S. A. R. C. Mariotti; P. Heinrich, Numerical simulation of submarine landslides and their hydraulic effects, Journal of Waterway, Port, Coastal, and Ocean Engineering 123 (4) (1997). doi:https://doi.org/10.1061/(ASCE)0733-950X(1997)123:4(149).
- [25]D. Drumheller, On theories for reacting immiscible mixtures, International Journal of Engineering Science 38 (3) (2000) 347 – 382. doi:https://doi.org/10.1016/S0020-7225(99)00047-6. URL http://www.sciencedirect.com/science/article/pii/S0020722599000476
- [26]S. Hassanizadeh, Derivation of basic equations of mass transport in porous media, part1: Macroscopic balance laws, Advances in Water Resources 9 (1986) 196–206. doi:https://doi.org/10.1016/0309-1708(86)90024-2.