Setup Guide: Unsteady Flows and DES

Draft. These settings are drawn from SU2 source, tutorials and regression cases. Recommendations marked (to be reviewed) still need an experienced user’s review.

Use this guide to choose settings and understand their consequences. Follow the linked Tutorials for worked examples and consult V&V for numerical and physical comparisons. Example settings are starting points, not evidence that a case is converged or validated.

The documentation guide explains the source tags used below: (code), (test case), (tutorial), (docs), (advice) and (to be reviewed).

• When to use this guide
• Start from
• Settings that matter
• Check convergence and results
• Common problems
• Related tutorials and test cases

When to use this guide

Use this guide when the flow changes in time and you want to follow that change. Typical cases are:

• vortex shedding behind a bluff body or a stalled airfoil,
• buffet, large separated regions, wakes,
• bodies that move, such as a pitching airfoil.

The guide covers two approaches:

• URANS (unsteady RANS): the RANS equations solved in time. The turbulence model still models all turbulent eddies. URANS captures large, regular unsteadiness such as vortex shedding (advice).
• DES (detached eddy simulation), a hybrid RANS/LES method: RANS near the walls, large-eddy simulation (LES) in separated regions. In LES, the mesh and the time step resolve the large turbulent eddies and the model handles only the small ones (advice).

A steady RANS run that does not converge can be a sign that the flow is unsteady (advice). For steady settings, see Setup Guide: Steady RANS for External Aerodynamics.

Start from

URANS

This config comes from the unsteady NACA0012 tutorial: compressible_flow/Unsteady_NACA0012/unsteady_naca0012.cfg in the Tutorials repository. We changed RESTART_SOL to NO and removed RESTART_ITER, so the run starts from the free stream. The tutorial restarts at time step 499 to skip the start-up phase. We also removed the adjoint and file name options. All other values are unchanged.

% ------------- Problem ------------------------------------------------%
SOLVER= RANS
KIND_TURB_MODEL= SA
REF_DIMENSIONALIZATION= DIMENSIONAL
MATH_PROBLEM= DIRECT
RESTART_SOL= NO
%
% ------------- Free stream --------------------------------------------%
MACH_NUMBER= 0.3
AOA= 17.0
FREESTREAM_TEMPERATURE= 293.0
FREESTREAM_PRESSURE= 101325.0
REYNOLDS_NUMBER= 1000.0
REYNOLDS_LENGTH= 1.0
REF_ORIGIN_MOMENT_X= 0.25
REF_ORIGIN_MOMENT_Y= 0.00
REF_ORIGIN_MOMENT_Z= 0.00
REF_LENGTH= 1.0
REF_AREA= 1.0
%
% ------------- Boundary conditions ------------------------------------%
MARKER_HEATFLUX= ( airfoil, 0.0 )
MARKER_FAR= ( farfield )
MARKER_PLOTTING= ( airfoil )
MARKER_MONITORING= ( airfoil )
%
% ------------- Time integration (dual time stepping) ------------------%
TIME_DOMAIN= YES
TIME_MARCHING= DUAL_TIME_STEPPING-2ND_ORDER
TIME_STEP= 5e-4
TIME_ITER= 2000
INNER_ITER= 10
%
% ------------- Numerics -----------------------------------------------%
NUM_METHOD_GRAD= WEIGHTED_LEAST_SQUARES
CONV_NUM_METHOD_FLOW= JST
JST_SENSOR_COEFF= ( 0.5, 0.005 )
CONV_NUM_METHOD_TURB= SCALAR_UPWIND
MUSCL_TURB= NO
TIME_DISCRE_FLOW= EULER_IMPLICIT
TIME_DISCRE_TURB= EULER_IMPLICIT
CFL_NUMBER= 1e12
CFL_ADAPT= NO
LINEAR_SOLVER= FGMRES
LINEAR_SOLVER_ERROR= 0.1
LINEAR_SOLVER_ITER= 10
%
% ------------- Convergence inside each time step ----------------------%
CONV_FIELD= REL_RMS_DENSITY
CONV_RESIDUAL_MINVAL= -3
CONV_STARTITER= 0
%
% ------------- Time averages and convergence in time ------------------%
WINDOW_CAUCHY_CRIT= YES
WINDOW_START_ITER= 500
WINDOW_FUNCTION= HANN_SQUARE
CONV_WINDOW_FIELD= ( TAVG_DRAG, TAVG_LIFT )
CONV_WINDOW_STARTITER= 0
CONV_WINDOW_CAUCHY_EPS= 1E-4
CONV_WINDOW_CAUCHY_ELEMS= 10
%
% ------------- Input and output ---------------------------------------%
MESH_FILENAME= unsteady_naca0012_mesh.su2
MESH_FORMAT= SU2
HISTORY_WRT_FREQ_INNER= 0
SCREEN_WRT_FREQ_INNER= 100
SCREEN_OUTPUT= ( TIME_ITER, INNER_ITER, RMS_DENSITY, REL_RMS_DENSITY, DRAG, LIFT, CAUCHY_TAVG_DRAG, CAUCHY_TAVG_LIFT )
HISTORY_OUTPUT= ( TIME_ITER, INNER_ITER, REL_RMS_RES, RMS_RES, AERO_COEFF, TAVG_AERO_COEFF, CAUCHY )
OUTPUT_FILES= ( RESTART, PARAVIEW )
OUTPUT_WRT_FREQ= ( 1, 1 )

The tutorial page Unsteady NACA0012 explains dual time stepping and the windowed averages.

DES

This config comes from the DDES regression test TestCases/ddes/flatplate/ddes_flatplate.cfg. We removed the adjoint and file name options, and options that have no effect with these settings. All other values are unchanged. This is a 2D flat plate. It checks that the DES options run. It is not a realistic DES: real DES runs are 3D (see Mesh for DES).

% ------------- Problem ------------------------------------------------%
SOLVER= RANS
KIND_TURB_MODEL= SA
HYBRID_RANSLES= SA_EDDES
MATH_PROBLEM= DIRECT
RESTART_SOL= NO
%
% ------------- Free stream --------------------------------------------%
MACH_NUMBER= 0.2
AOA= 0.0
SIDESLIP_ANGLE= 0.0
FREESTREAM_TEMPERATURE= 300.0
REYNOLDS_NUMBER= 5000000.0
REYNOLDS_LENGTH= 1.0
REF_ORIGIN_MOMENT_X= 0.25
REF_ORIGIN_MOMENT_Y= 0.00
REF_ORIGIN_MOMENT_Z= 0.00
REF_LENGTH= 1.0
REF_AREA= 2.0
%
% ------------- Time integration ---------------------------------------%
TIME_DOMAIN= YES
TIME_MARCHING= DUAL_TIME_STEPPING-2ND_ORDER
TIME_STEP= 0.000288
MAX_TIME= 10
UNST_CFL_NUMBER= 0.0
INNER_ITER= 20
TIME_ITER= 500
%
% ------------- Boundary conditions ------------------------------------%
MARKER_HEATFLUX= ( wall, 0.0 )
MARKER_INLET= ( inlet, 302.4, 118309.784, 1.0, 0.0, 0.0 )
MARKER_OUTLET= ( outlet, 115056.0, farfield, 115056.0 )
MARKER_SYM= ( symmetry )
MARKER_PLOTTING= ( wall )
MARKER_MONITORING= ( wall )
%
% ------------- Numerics -----------------------------------------------%
NUM_METHOD_GRAD= GREEN_GAUSS
CFL_NUMBER= 10.0
CFL_ADAPT= NO
CONV_NUM_METHOD_FLOW= SLAU2
MUSCL_FLOW= YES
SLOPE_LIMITER_FLOW= NONE
TIME_DISCRE_FLOW= EULER_IMPLICIT
CONV_NUM_METHOD_TURB= SCALAR_UPWIND
MUSCL_TURB= NO
TIME_DISCRE_TURB= EULER_IMPLICIT
CONV_RESIDUAL_MINVAL= -15
CONV_STARTITER= 0
%
% ------------- Input and output ---------------------------------------%
MESH_FILENAME= mesh_flatplate_turb_137x97.su2
MESH_FORMAT= SU2
OUTPUT_WRT_FREQ= 1000
SCREEN_OUTPUT= (TIME_ITER, INNER_ITER, RMS_DENSITY, RMS_NU_TILDE, LIFT, DRAG, TOTAL_HEATFLUX)

For a real DES run, also write the DES and time-average fields (code):

VOLUME_OUTPUT= ( COORDINATES, SOLUTION, PRIMITIVE, DDES, VORTEX_IDENTIFICATION, TIME_AVERAGE )

The mesh files of the test cases are in the TestCases repository.

Settings that matter

Dual time stepping

Dual time stepping splits the run into physical time steps. Inside each time step, SU2 runs “inner iterations” to find the new solution, the same way it converges a steady problem (tutorial, docs).

• TIME_DOMAIN= YES turns on the time-accurate solver (docs). TIME_MARCHING= DUAL_TIME_STEPPING-2ND_ORDER is used by the URANS test cases and by the flat plate and backscatter DES tests (test case). The short ONERA M6 DES regression tests of PR #2150 use DUAL_TIME_STEPPING-1ST_ORDER (test case). DUAL_TIME_STEPPING-1ST_ORDER and TIME_STEPPING (one update per time step, no inner iterations) also exist (code).
• TIME_STEP is the physical time step. It is always in seconds, also when the equations are non-dimensional: SU2 divides it by the reference time (code).
• TIME_ITER is the number of time steps. The default is 1 (code).
• MAX_TIME is the final physical time in seconds. The default is 1.0. SU2 stops at MAX_TIME or at TIME_ITER, whichever comes first (code, docs). Set MAX_TIME larger than TIME_STEP × TIME_ITER if you want TIME_ITER steps.
• INNER_ITER is the maximum number of inner iterations per time step. The default is 1 (code). Always set it.
• Do not use ITER in an unsteady run. SU2 stops with an error and asks for TIME_ITER and INNER_ITER (code).
• With UNST_CFL_NUMBER other than 0.0, SU2 computes the time step from this CFL number and ignores TIME_STEP, as the comments in config_template.cfg say. The default is 0.0 (code).

See Solver Setup.

Time step size

• The time step must resolve the fastest change you care about (advice). For vortex shedding, estimate the period from the Strouhal number St = f D / U. St is about 0.2 for many bluff bodies (advice, to be reviewed).
• Use at least 50 to 100 time steps per shedding period for URANS (advice, to be reviewed).
• Check the result by halving the time step: the period and the averaged forces should not change much (advice).
• Examples: the NACA0012 tutorial uses TIME_STEP= 5e-4. The square cylinder test uses TIME_STEP= 0.0015 with MAX_TIME= 3.75. The von Karman cylinder tutorial uses TIME_STEP= 0.005 with 20 inner iterations (tutorial, test case).

Convergence inside each time step

Each time step needs converged inner iterations. If not, the time history is wrong even if the run looks stable (tutorial).

• The inner loop uses the same criteria as a steady run: CONV_FIELD, CONV_RESIDUAL_MINVAL, CONV_STARTITER (docs). It stops at the criterion or at INNER_ITER.
• REL_RMS_DENSITY is the residual relative to its value at the first inner iteration of the time step (code). CONV_FIELD= REL_RMS_DENSITY with CONV_RESIDUAL_MINVAL= -3 asks for a drop of three orders of magnitude in each time step (tutorial).
• Set CONV_STARTITER= 0 so the check starts at the first inner iteration (tutorial). The default is 5 (code).
• The inner iterations use CFL_NUMBER as in a steady run. The NACA0012 tutorial uses CFL_NUMBER= 1e12 with JST (tutorial). The DES test uses 10 with SLAU2, the square cylinder test 2.0 with ROE (test case).
• Watch INNER_ITER on the screen. If most time steps end at the maximum without reaching the criterion, raise INNER_ITER or the CFL, or lower the time step (advice).
• In a time-domain run, SU2 writes the history file once per time step by default. Set HISTORY_WRT_FREQ_INNER to also write inner iterations (code).

Restarting an unsteady run

SU2 writes one restart file per saved time step, with the step number in the name, for example restart_flow_00499.dat (tutorial, docs).

• OUTPUT_WRT_FREQ sets how often (in time steps) SU2 writes files (docs). With second-order time stepping, SU2 also writes the step before, because a restart needs two old solutions (code).
• To restart, set RESTART_SOL= YES and RESTART_ITER to the first new time step. With second order, SU2 reads the files of steps RESTART_ITER−2 and RESTART_ITER−1. With first order, it reads step RESTART_ITER−1 (code, docs).
• SU2 reads the files named after SOLUTION_FILENAME, so rename or copy the restart files first (docs).
• TIME_ITER counts from the start of the whole simulation, not from the restart. SU2 runs TIME_ITER − RESTART_ITER more steps (code). TIME_ITER must be larger than RESTART_ITER (code).
• To start URANS from a converged steady solution, copy the steady restart file to the two file names that SU2 reads for RESTART_ITER= 2 (for example solution_flow_00000.dat and solution_flow_00001.dat) (advice, to be reviewed). The square cylinder test restarts with RESTART_ITER= 2 (test case).

See Restart File.

Time averages and convergence in time

History averages. In a time-domain run, SU2 adds a time-averaged version of every coefficient, with the prefix TAVG_, for example TAVG_DRAG (code, tutorial).

• The average starts at time step WINDOW_START_ITER. If you do not set it, it starts at RESTART_ITER. It must not be smaller than RESTART_ITER, or SU2 stops with an error (code).
• Start the average after the start-up phase, when the signal is periodic or statistically steady (tutorial).
• WINDOW_FUNCTION sets the weight of each sample: SQUARE (equal weights, the default), HANN, HANN_SQUARE or BUMP (code). For a periodic signal, the windowed averages converge faster than SQUARE (tutorial).

Stop when the average stops changing. WINDOW_CAUCHY_CRIT= YES turns on a Cauchy criterion on the averages (docs). This criterion looks at how much the average changes from step to step.

• CONV_WINDOW_FIELD lists the averaged fields, for example ( TAVG_DRAG, TAVG_LIFT ) (tutorial).
• CONV_WINDOW_CAUCHY_EPS (default 1E-3), CONV_WINDOW_CAUCHY_ELEMS (default 100) and CONV_WINDOW_STARTITER (default 15, counted from WINDOW_START_ITER) play the same role as CONV_CAUCHY_EPS, CONV_CAUCHY_ELEMS and CONV_STARTITER in a steady run (code, docs).
• The tutorial uses CONV_WINDOW_CAUCHY_EPS= 1E-4 with CONV_WINDOW_CAUCHY_ELEMS= 10 (tutorial).

Volume averages. Add TIME_AVERAGE to VOLUME_OUTPUT (code). SU2 then writes:

• mean values: MEAN_DENSITY, MEAN_VELOCITY-X, MEAN_VELOCITY-Y, MEAN_VELOCITY-Z, MEAN_PRESSURE, MEAN_SKIN_FRICTION-X, MEAN_SKIN_FRICTION-Y, MEAN_SKIN_FRICTION-Z;
• mean squares and products: RMS_U, RMS_V, RMS_W, RMS_UV, RMS_UW, RMS_VW, RMS_P;
• Reynolds stresses and pressure fluctuations: UUPRIME, VVPRIME, WWPRIME, UVPRIME, UWPRIME, VWPRIME, PPRIME (code).

How these averages work (code):

• They start at WINDOW_START_ITER and use equal weights. WINDOW_FUNCTION does not apply to them.
• SU2 only computes them if TIME_AVERAGE is in VOLUME_OUTPUT.
• SU2 keeps them in memory. They are not read back from the restart files, so after a restart they start again from RESTART_ITER.

So average over one continuous run, without restarts (advice).

Hybrid RANS/LES: which DES variant

The SA_* variants build on the SA model, the SST_* variants on the SST model. Set KIND_TURB_MODEL and pick a matching variant with HYBRID_RANSLES (code). The SST variants come with PR #2150 and are in develop once it is merged.

In original SA_DES, the length scale is min(d, C_DES Δ), where d is wall distance and Δ the grid measure. The delayed variants use d − f_d·max(0, d − C_DES Δ): f_d near zero protects attached boundary layers, and f_d near one recovers the DES limit. The SST variants use the same idea with the RANS length √k / (β* ω) in place of d. The grid measure also changes with the variant. Thus distance from the wall alone does not determine the RANS/LES mode (code, reference). See Hybrid RANS/LES.

Advice:

• SA_DES can switch to LES inside an attached boundary layer if the mesh is fine along the wall. The mesh, not the flow, can then cause separation. SA_DDES and SA_EDDES add shielding to reduce this risk; the result still needs mesh and sensor checks (advice).
• SA_ZDES in SU2 is DDES with the ZDES cell size. It is not the full zonal method of the reference, where the user picks RANS and LES zones (code, to be reviewed).
• SU2 stops with an error if the variant does not match KIND_TURB_MODEL, or if TIME_DOMAIN= NO (code).
• With an SST_* variant, SU2 also stops if SST_OPTIONS contains COMPRESSIBILITY-WILCOX or COMPRESSIBILITY-SARKAR. How these corrections should act in the LES region is not validated yet (code). See Hybrid RANS/LES.

Related options (code):

• DES_CONST: the DES constant. The default is 0.65.
• LES_FILTER_WIDTH: a fixed LES filter width. The default -1.0 means SU2 uses the local cell size (the largest distance from a point to its neighbours).
• ENFORCE_LES= YES: LES mode everywhere. It needs HYBRID_RANSLES other than NONE.
• These three options act only on the SA_* variants. With an SST_* variant, SU2 stops with an error if you set DES_CONST, give a positive LES_FILTER_WIDTH, or set ENFORCE_LES= YES (code). The SST variants use their own constant, C_DES = 0.78·F1 + 0.61·(1 − F1). The screen shows the constant in use at the start of the run (code).
• STOCHASTIC_BACKSCATTER= YES and the SBS_ options: a stochastic backscatter model, which feeds energy from the modelled to the resolved scales. It works only in 3D and only with an SA_* variant. SU2 stops with an error otherwise (code). The test case backscatter/backward_step/backwardStep.cfg shows all its options (code, test case).

Mesh for DES

General points for the DES mesh (advice, to be reviewed):

• DES resolves turbulent eddies, which are three-dimensional. Run DES in 3D. 2D DES runs do not give physical turbulence.
• Near the walls, the mesh must be fine enough for RANS: first-cell y+ about 1.
• In the regions you want in LES mode (wakes, separated zones), use cells of similar size in all directions. Original DES/DDES use the longest edge from a node to its neighbours; ZDES/EDDES use the variant-specific measures described in Hybrid RANS/LES (code). Stretched cells can limit LES resolution; inspect the chosen grid measure and the RANS/LES sensor rather than assuming that refinement in one direction is sufficient.
• Keep the cell size changes smooth between the RANS and LES regions.
• For a body with a spanwise direction, the spanwise extent must hold the largest eddies of interest.

Time step for DES

• The time step must resolve the eddies the mesh can hold. A common target is a convective CFL number of about 1 in the LES region, that is, a time step of about cell size / local velocity (advice, to be reviewed).
• Use DUAL_TIME_STEPPING-2ND_ORDER (test case).
• The flat plate DES test and the backscatter test use INNER_ITER= 20 (test case). The short ONERA M6 DES regression tests use INNER_ITER= 2 and are not a setup example. Check the inner convergence as for URANS.

Numerical dissipation

Reduce numerical dissipation where you can. Upwind schemes and limiters add it, and in the LES region it damps the resolved eddies (advice).

Compressible solver. ROE_LOW_DISSIPATION lowers the upwind part of the flux where the flow is resolved (code, docs). Details from the code (code):

• It works with ROE, SLAU and SLAU2.
• FD: uses the DDES shielding function, computed from the wall distance.
• NTS: the blending of Travin and Shur. It uses DES_CONST and the cell size. With an SST_* variant you cannot set DES_CONST, so it uses the default, 0.65.
• FD_DUCROS and NTS_DUCROS: as above, plus the Ducros sensor, which keeps the full dissipation at shocks (code, docs).
• SU2 keeps at least 5 % of the upwind dissipation for stability.
• Write the ROE_DISSIPATION output group to see where the dissipation is reduced.

No test case in develop uses ROE_LOW_DISSIPATION (test case).

Incompressible solver. The LD2 central scheme (low dissipation, low dispersion) is available for constant density flows only. With the energy equation on, SU2 uses JST for the energy equation (code). The backscatter test uses CONV_NUM_METHOD_FLOW= LD2 with JST_SENSOR_COEFF= ( 0.0, 0.002 ) and MUSCL_FLOW= NO (test case).

Limiters. The DES flat plate test uses MUSCL without a limiter (SLOPE_LIMITER_FLOW= NONE) (test case). A limiter adds dissipation where it is active (advice).

Averaging a DES run

• Wait until the start-up phase has left the domain before you start averaging. A common rule is a few flow-through times (advice, to be reviewed).
• Then average over many characteristic times, so that the mean and the fluctuations settle (advice).
• Use WINDOW_START_ITER and the TIME_AVERAGE volume group described above (code).
• To record time signals at points, the history output can include probes defined with CUSTOM_OUTPUTS. The von Karman cylinder tutorial shows how (tutorial).

Check convergence and results

• Inner convergence. REL_RMS_DENSITY (or your CONV_FIELD) reaches the criterion in each time step before INNER_ITER (tutorial).
• Start-up phase over. Plot LIFT and DRAG against time. The signal should be periodic or statistically steady before WINDOW_START_ITER (tutorial).
• Averages settled. Compare averages from separate sampling blocks and extend the sampling interval. A slowly changing cumulative average alone does not establish statistical convergence, especially for DES. Monitor TAVG_LIFT and TAVG_DRAG. With WINDOW_CAUCHY_CRIT= YES, the CAUCHY_TAVG_ values are below CONV_WINDOW_CAUCHY_EPS (tutorial).
• Time step and mesh. Halve the time step and refine the mesh. The frequency and the averages should change little (advice).
• DES regions. Write the DDES group: DES_LENGTHSCALE, WALL_DISTANCE and LES_SENSOR. LES_SENSOR is 0 where the model acts as RANS. In LES mode it is 1 for SA_DES, the value of the shielding function f_d for the SA DDES variants, SST_DDES and SST_EDDES, and 1 − f̃_d for SST_IDDES and SST_SIDDES (code). Attached boundary layers should stay at 0 (advice). See Hybrid RANS/LES for an extra check with SST.
• Resolved eddies. Write the VORTEX_IDENTIFICATION group: VORTICITY (2D) or VORTICITY_X, VORTICITY_Y, VORTICITY_Z (3D), and Q_CRITERION (code). Iso-surfaces of Q_CRITERION show the resolved eddies (advice).

See History and Solution Output and the Output Fields Reference.

Common problems

Related tutorials and test cases

Tutorials:

• Unsteady NACA0012: URANS, dual time stepping, windowed averages, restart after the start-up phase.
• Unsteady von Karman vortex shedding: laminar incompressible cylinder; time step, inner iterations, probes.
• Unsteady shape optimization NACA0012: unsteady adjoint with windowed objective functions.

Test cases (develop branch):

• unsteady/square_cylinder/turb_square.cfg: URANS with SST and ROE, restart from step 2.
• unsteady/pitching_naca64a010_rans/turb_NACA64A010.cfg: URANS of a pitching airfoil with SA and rigid mesh motion.
• ddes/flatplate/ddes_flatplate.cfg: SA_EDDES with SLAU2.
• ddes/oneram6/ddes_oneram6_*.cfg (PR #2150): short regression tests of SST_DDES, SST_IDDES, SST_SIDDES, SST_EDDES and SA_EDDES on the ONERA M6 wing. See Hybrid RANS/LES.
• backscatter/backward_step/backwardStep.cfg: incompressible SA_DDES with stochastic backscatter, LD2 scheme and time-average output.

User guide pages:

• Solver Setup, Restart File, History and Solution Output, Convective Schemes, Physical Definition.
• All options: Option Reference. All output fields: Output Fields Reference.

Source review: 22 October 2026, against SU2 docs_site at 41ea6a219a. The DES statements were updated on 23 October 2026 against PR #2150 at 1bcd4a7425. See the review scope and outstanding validation. Setup recommendations remain draft where indicated.