Numerical Parameters¶
In this section, some purely numerical parameters for the phasefield, concentration, flow and stress solver are defined.
PhaseField Solver¶
Time stepping¶
The time step for the phasefield solver has to be defined. It is recommended to use the automatic
or automatic_limited
time step options. In the latter case, minimum and maximum values for the time step as well as optional correction factors (default 0.9) can be further specified which apply to the "phasefield" and "segregation" time stepping criteria. Not all of the 4 parameters must be given. Especially, the first value, the minimum time step, can be very useful to improve performance (for more details please use the MICRESS® forum at www.micress.de). The parameters for automatic_limited
can also be read as timedependent from a file.
For manual definition of the phasefield time step, only the corresponding value in seconds has to be given. As a hint for finding reasonable values, the timesteps for each phase pair according to the automatic criteria is given as screen output during initialization (see Figure 1.
The section ends with the definition of the number of iterations for initialisation of the interfaces. During these types of iterations, the sharp interfaces produced by the initial grain structure setting are smoothed by applying the phasefield solver without any chemical driving force. Due to the fact that the stabilisation term, which enforces the interface profile, and the curvature are not separated in the phasefield model, curved initial grains will shrink and finally vanish if the number of iterations is too high. For single round particles, an attempt has been made to compensate for that, but in the general case this cannot be avoided. In most cases, there will be no problem, if a small number (ca. 20) of initial iterations is used. On the other hand, the nonconserved phasefield variable gives the opportunity to finetune the initial grain fraction during initialisation, e.g. if the "sharp" grain consists of only a few grid cells.
Example 1¶
(Limited) Automatic time stepping with 10 iterations for initialisation
... # Phase field solver #  # Timestep ? # Options: fix ...[s] automatic automatic_limited automatic_limited # Options: constant from_file constant # Limits: (real) min./s, [max./s], [time step factor], [segregation factor] 1.E6 1.E2 #Coefficient for phasefield criterion 1.00 #Coefficient for segregation criterion 0.900 # Number of steps to adjust profiles of initially sharp interfaces [exclude_inactive]? 10 ...
Figure 1¶
Screen output for time step initialisation
... Beginning of initialisation =========================== ... Routine init calls routine initTwidth tWidth_max( LIQUID: FCC_A1) = 2.0552885E05 s tWidth_max( LIQUID: ALCU_THETA) = 1.0687500E05 s Maximal value for tWidth = 1.0687500E05 s for phasefield solver Maximal value for tWidth = 1.0687500E06 s for concfield solver Automatic time stepping (phasefield solver): decreased value for tWidth = 1.01531E06 s Initial value for tWidth = 1.01531E06 s for automatic time stepping (phasefield solver) ...
Type of potential and FDcorrection¶
The user can select either double_obstacle
or multi_obstacle
.
The double_obstacle
functional is of parabolic type: g(\phi)=\phi(1\phi), limited by discontinuities at \phi=0 and \phi=1. The multi_obstacle
functiona [^{1}] represents its generalized formulation for multiphase systems comprising contributions g_{\alpha\beta}(\phi)=\phi_\alpha \phi_\beta for all pairwise phase interactions. In the special case of a two phase systems, it exactly recovers the double_obstacle
functional. We recommend to use multi_obstacle
as a standard for all applications.
If double_obstacle
is selected for simulation of a multiphase problem, the antisymmetric approximation[^{1}] will be used. In this simplifying treatment, pairwise interactions between phases \alpha and \beta are directly derived from the corresponding two phase free energy functionals f_{\alpha\beta} instead of from the common free energy functional, thus neglecting higher order contributions \alpha\beta\gamma. As a consequence all dihedral angles at a junction will evolve as equal (i.e. a triple junction in 2D will always exhibit three 120° angles) independent of the energies of the adjacent interfaces. Therefore, this approximation can only be used in cases in which all boundaries energies are equal or in which dihedral angles are of no importance. For a correct reproduction of dihedral angles in multiphase problems select multi_obstacle
. Keep in mind that strongly differing interfacial energies may lead to high junction forces and extreme angles.
The numerical solution of the phasefield equation based on finite differences (FD) inherently implies a discretization error which may strongly bias the simulation results. Usually, it results in too low interface velocities. The error is inversely correlated to the number of numerical cells resolving the diffuse interface profile. However, high numbers of interface cells are detrimental for computational performance and also favor undesired numerical interface trapping. The problem can be overcome by an improved numerical solution of the phasefield equation, which takes advantage of the a priori knowledge of the phasefield profile, determined by the choice of the potential function[^{2}] [^{3}]. To activate this improved numerical solution, select fd_correction
and choose an interface width of 34 numerical cells. If you want to rerun a diffusioncontrolled simulation which has been run before without the fd_correction option, it might be necessary to recalibrate the interface mobility. The option fd_correction
is recommended to be used as a standard for all applications and may become default in coming versions. It is however not yet compatible with temperature coupling.
The option multi_obstacle
especially comprises corrections for triple junction terms and is important for correct wetting characteristics.
Phase Minimum and Interface Thickness¶
Next, a value for the phase minimum is required. A cell with a phase or grain fraction below this value is not considered to be in the interface any more but in a bulk region (liquid or solid). In most cases, a value of 1.E4 can be recommended.
As optional parameter in the same line, a hysteresis factor can be defined which has a default value of 2, For more information, please consult the MICRESS® forum.
The interface thickness (in cells) also has to be input in this section. Generally, with increasing interface thickness, the curvature evaluation of the phasefield profile is improved. On the other hand, a higher resolution is required to resolve highly curved structures, and numerical artefacts related to the finite interface thickness like artificial solute trapping" is increased.
Note that in case of using extremely small values for the interface thickness (less than 3 cells), curvature evaluation is poor and nucleation at the interface is not working correctly anymore!
PhaseField Data Structure¶
In this section, the initial dimensions for the internal fields iFace
and nTupel
must be specified. During runtime, the size of these fields is determined automatically, so in most cases the given values are of minor importance
The values have to be given relative to the size of the simulation domain, a value of 1.0 for iFace
for example would assume that the whole calculation domain could be covered by (twograin or twophase) interfaces without exceeding the given initial list size. The same holds for nTupel
and the coverage of the domain with triple or higher junctions. The actual usage for both fields can be found in the TabL output.
The arrays iFace
and nTupel
are fully dynamic, so the influence of the initial values is quite limited. In extreme cases, the specification of excessively high initial values can lead to a memory overflow during the program startup. Too low values can lead to an unnecessarily high number of reallocation steps which slow down the initialisation process e. g. in the case of grain growth simulation with a high number of initial grains. Initial values of 0.1 for both parameters are recommended and work in practically all cases.
Additional optional parameters can be specified for the minimal
and the target
usage. These parameters allow a finetuning of the reallocation criteria during runtime what can be helpful if a temporal memory overflow must be avoided. More details can be found or requested in the MICRESS® forum.
Example 2¶
iFace and nTupel
... # Phase field solver #  ... # Coefficient for initial dimension of field iFace # [minimum usage] [target usage] 0.10 # Coefficient for initial dimension of field nTupel # [minimum usage] [target usage] 0.10 ...
Concentration Solver¶
The maximum number of concentration solving per phasefield iteration has only to be defined, if concentration coupling is used in combination with a fixed phasefield time step. An error message is given if the number is too small to be consistent with the timestep criterion of the explicit diffusion solver. This input is not required when automatic timestepping is used.
The global cushion factor influences the diffusion time stepping, i.e. the lower the factor the smaller are the diffusion time steps. In MICRESS version lower than 7.0, this factor holds a default value of 0.5 to be on the safe side. However, reducing the diffusion time stepping by 5% is sufficient for the most applications and gives a performance boost for the diffusion solver part.
Example 3¶
Diffusion time stepping factor (cushion)
... # Concentration solver #  # Factor for diffusion time stepping? (0.0 < factor < 1.0) 0.95000 ...
Stoichiometry and more¶
The first type of required information is a list of all the stoichiometric components. Stoichiometric components are those elements in a phase for which no explicit slopes of the phase diagram lines are specified or can be specified. A typical case is an element which has no solubility range in a given phase (line compound), which also includes elements with zero solubility. In such cases, the composition of the element in this phase cannot be used as a thermodynamic condition which is important information if coupling to thermodynamic data is used. This type of stoichiometric condition is automatically detected and needs not to be (but may be) specified by the user.
The user may also decide to define an element with large slope in the phase diagram for numerical reasons as stoichiometric, other reasons could be opposite signs of the solidus and solvus slopes retrieved from the database which also can create numerical problems (pseudobinary "demixing").
The definition of the stoichiometric components is available in a compact terse
mode: In each line, a phase number and a list of the component numbers which are to be defined as stoichiometric in this phase are specified. The order of specification is arbitrary, different elements of the same phase can also be specified in extra lines. This input block is also used for further special options described below, for the definition of ternary interaction groups, and for setting concentration limits, and is closed by the keyword no_stoichio
or no_more_stoichio
.
In this section, a few expert options for the use of stoichiometric conditions are available, which are partly not documented in the standard input file (.in file). They allow a certain "fine tuning" of the stoichiometric option in case of severe numerical problems with TQ coupling which may occur in complex alloy systems. They should not be used except if absolutely necessary and after contacting MICRESS® experts.

solubility_on
/solubility_off
: This flag is helpful if components have a finite solubility range, but have to be defined as stoichiometric because of demixing problems., In this case the solubility can be taken into account for the quasiequilibrium calculation, but not during solute redistribution. If the keywordsolubility_on
is specified (as extra line), all stoichiometric phases defined after this line are assumed to have a solubility range. 
stoich_enhanced_on
/stoich_enhanced_off
: If a stoichiometric component has a solubility range and has been marked as such (using the keywordsolubility_on
), it is possible to use a more effective extrapolation scheme for redistribution, which assumes a "solidification path" approximation similar as in the growth restriction theory for nucleation. Although this approximation is not necessarily correct, it can be better than just assuming a constant composition, especially during solidification.
The use of stoichiometric conditions, especially when applied to avoid pseudobinary "demixing", can easily come to a limit: It is generally not recommended to set a stoichiometric condition for the same element in two phases between which a phase interaction is defined (see section Phase Interactions). With this limitation it may be almost impossible to avoid the effects of pseudobinary "demixing" in materials like stainless steels by use of the stoichiometric condition.
Therefore, from MICRESS® version 6.1, a "diagonal" extrapolation scheme has been included which only uses the diagonal elements of the redistribution matrix and which avoids demixing at the costs of a slightly lower extrapolation accuracy (which can be compensated for by more frequent updating of quasiequilibrium). It is invoked by specifying the keyword diagonal
followed by a list of elements. If diagonal
is used without further parameters, this extrapolation is applied for all dissolved components.
Furthermore, ternary element interactions can be included for selected element triples. Ternary extrapolation gives the user the possibility to account for some of the interactions between the (dissolved) elements which are neglected in case of normal (pseudobinary or diagonal) extrapolation, and thus to increase extrapolation accuracy. Using interaction
followed by two component numbers in the same line, a ternary interaction group can be established between two elements and the matrix component. Depending on the total component number, several such ternary interaction groups can be defined in extra lines, but each (dissolved) component can be used only once. The definition of ternary interaction groups applies to all phases and phase interactions.
If one of the keywords interaction
or diagonal
has been used, two interaction matrices per phase interaction (one for each direction) are written to the .log file (together with the linearization data, at the time of initialization of the phase interaction) which allow estimating the level of chemical higher order interactions between the dissolved elements and can serve as a guide for using interaction
in an efficient way. In case of ideal solution, all diagonal matrix elements are 1.0 and the others 0.0. More information about the different extrapolation schemes available in MICRESS® can be found in [^{4}].
Further phaserelated options which can be defined in this section are:

limits
: This option is followed by a phase and component number in the same line and restricts the allowed composition range of an element in a phase. The minimum and maximum value in at% (always!) is then specified in extra lines. This option is especially helpful if the same phase in the database exists as several composition sets (e.g. FCC_A1 and FCC_A1#2) andswitching
between the two phases has to be avoided. Using this option also triggers apenalty
driving force which helps to avoid conditions which lead to compositions outside the user limit. However, this penalty term can lead to strange effects or wrong results if thelimits
keyword is used in wrong way. Therefore, this option should be used carefully and only if necessary. 
criterion_higher
/criterion_lower
: These options can be regarded as alternatives forlimits
as they allow for setting a relative limit to phase compositions. The keyword is followed by three integer values for phase 1, phase 2 and a component number. The options allow to require that the specified component has a higher or lower composition in phase 1 than in phase 2. This option is helpful e.g. for the interaction of γ and γ' in Nibase superalloys to ensure that e.g. the Al composition is always lower in γ than in γ' without necessity to specify a discrete composition limit. 
switch_diff_data
: Very specific expert option which allows using diffusion data from a different phase. Further parameters are the phase number and the reference phase where the diffusion data should be taken from. 
diff_comp_sets
/vol_com_sets
/enth_comp_sets
: Allows adding composition sets to singlephase equilibrium calculations which are necessary when retrieving diffusion, volume or enthalpy data. Each keyword is followed by a phase list for which existing composition sets (those with the same base name) are added. Only those composition sets are added which are defined as extra phases in MICRESS® and which already have been initialized. The option helps to deal with problems when working close to miscibility gaps so that the phase at the average composition may be unstable.
If the volume_change
mode is activated, components can in this section be defined interstitial
by using this keyword followed by a phase number and component number(s). By default, components are defined as substitutional. In contrast to substitutional diffusion, diffusion of interstitial components is related to a local change in number of moles. Note that this is a new feature and the respective input will most probably still be moved to another section in future.
Example 4¶
Expert options for stoichiometric phases
... # Concentration solver #  ... # List of phases and components which are stoichiometric: # phase and component(s) numbers # List of concentration limits (at%): # <limits>, phase number and component number # List of penalty conditions: # <penalty>, phase 1, phase2, component number # List for ternary extrapolation (2 elements + main comp.): # <interaction>, component 1, component 2 # Switches: <stoich_enhanced_{onoff}> <solubility_{onoff}> # List of relative criteria on phase composition # <criterion_higher  criterion_lower>, phase No 1, phase No 2, component No # List of sublattice order conditions: # <ordereddisordered>, phase , sublattice 1, sublattice 2 # List of source changes for diffusion data # <switch_diff_data>, PhaseNo., reference phase # Switch: Add composition sets for calculation of diffusion/volume/enthalpy data # <diff_comp_sets  vol_comp_sets  enth_comp_sets>, phase list # End with 'no_more_stoichio' or 'no_stoichio' solubility_on interaction 2 1 Limits 0 1 # Minimal allowed value [at%] 0. # Maximal allowed value [at%] 20.0 no_more_stoichio ...
VolumeExpansion Solver¶
In case that the volume_change
has been specified, tow values have to be specified which control the expansionrelaxation solver: first, the maximal number of and second, a convergence value. The task of the expansionrelaxation solver is to homogenize local expansion by compensating material fluxes. The solver runs iteratively and stops when either the specified number of calls per times step is reached or when the mean difference in local expansion falls below the specified convergence value. The effectivity of the solver can be checked by monitoring the degree of homogenization in the expansion output file expa
and the consumed CPUtime listed in the TabP
file (row: expansion time).
Temperature Solver¶
The keyword 3d_correction
switches on a correction of the amount of latent heat which may be wrongly predicted in some types of 2D simulations. An additional phase number input allows specifying a reference phase for this correction, otherwise phase 0 is assumed. If lat_heat_3d
is chosen, the user can activate or deactivate the 2D/3D correction for each further phase by the keywords pseudo_3D
/ no_pseudo_3D
followed by the critical fraction of the matrix phase for limiting the correction (see section Topics  Temperature).
Example 5¶
3D correction
... # Temperature solver #  # 3D correction for latent heat in a 2D simulation ? # Options: no_3d_correction 3d_correction <matrix phase> 3d_correction 0 # Simulation with release of pseudo3D latent heat of phase 1 (FCC_A1)? # Options: pseudo_3d [crit. matrix fraction] no_pseudo_3d pseudo_3D 0.75 # Simulation with release of pseudo3D latent heat of phase 2 (ALCU_THETA)? # Options: pseudo_3d [crit. matrix fraction] no_pseudo_3d no_pseudo_3D # Interval for updating enthalpy data [s] 1.00000E02 ...
Stress Solver¶
The following section applies only for stress
and stress_coupled
simulations. The first input here is the phase number of the matrix phase. The strain and the calculation of its eigenvalue will be referenced to this phase.
The convergence criteria for the iterative BiCGstab matrix solver as well as an upper limit (maximum number) of iterations have to be defined. In the input line for the maximum number of iterations, the input of a second integer is also possible. It is set to 20 by default and specifies the maximum number of iterations allowed for achieving a good solution for quasistatic equilibrium. In most cases, the default settings are sufficient.
Flow Solver¶
For flow coupling a set of numerical parameters are queried in this section: the solver scheme, convergence criteria, and iteration limits. Section Flow Solver  Numerical Parameters describes the necessary parameters and ways to arrive at suitable values for a specific simulation.
Parallel Execution¶
The parallel variants of MICRESS® require one additional parameter for the number of threads to be used for parallel execution of the diffusion, stress, and temperature solvers.
These solvers are parallelized using the OpenMP (see www.openmp.org) paradigm for parallel programming which is intended for shared memory, thread level based parallelization.
It is possible to reduce the wallclock time of a simulation if the computational load is put heavily on the parallelized solvers. Check the TabP
output for getting an impression in which cases you can benefit from using the parallel variants of MICRESS®, see Figure 2. Each line of the text output shows the accumulated time values for the current simulation time step. Except the Simulation Time and CPU Time, all values show wallclock time. The third column, headed with Wallclock Time, shows the overall execution time and reflects the real time gone for the simulation.
Figure 2¶
Performance Output (TabP)
Two threads
Eight threads
Recommended choices of MICRESS® variants:

Serial versions: If there is only little load on the parallelized solvers, it is recommended to use MICRESS® standard serial variants. The parallelization itself introduces some overhead which results in a slow down of the parallel simulation.

Parallel versions: Simulation with a fair load on the parallel solvers will benefit from parallel execution with thread counts above 1. The performance will depend on the problem size and the available CPUs on your system when you choose high thread counts. Usually, there is an optimal thread count and the performance degrades beyond.
Example 6¶
Thread count for accelerating the execution of diffusion, stress, and temperature solvers in parallel MICRESS variants
# Numerical parameters # ==================== # ... # Parallel Execution #  # Number of parallel threads ? 5 ...

I. Steinbach, F. Pezzolla, B. Nestler, M. Seeßelberg, R. Prieler, G. J. Schmitz, and J. L. L. Rezende. A phase field concept for multiphase systems. Physica D: Nonlinear Phenomena, 943:135–147, jul 1996. doi:10.1016/0167278995002987. ↩↩

J. Eiken, B. Böttger, and I. Steinbach. Multiphasefield approach for multicomponent alloys with extrapolation scheme for numerical application. Physical Review, 2006. doi:10.1103/physreve.73.066122. ↩

Janin Eiken. Numerical solution of the phasefield equation with minimized discretization error. IOP Conference Series: Materials Science and Engineering, 33:012105, jul 2012. doi:10.1088/1757899x/33/1/012105. ↩

B. Böttger, J. Eiken, and M. Apel. Multiternary extrapolation scheme for efficient coupling of thermodynamic data to a multiphasefield model. Computational Materials Science, 108:283–292, oct 2015. doi:10.1016/j.commatsci.2015.03.003. ↩