See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/352643016

CFD Course: The Fractional Time-Step Method for incompressible flows

Technical Report · June 2021

DOI: 10.13140/RG.2.2.25472.76805

CITATIONS READS

0 4

1 author:

Filippo Maria Denaro

Università degli Studi della Campania "Luigi Vanvitelli
57 PUBLICATIONS   415 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

New modelling in the integral-based LES View project

CFD in diastole View project

All content following this page was uploaded by Filippo Maria Denaro on 22 June 2021.

The user has requested enhancement of the downloaded file.

CFD Course F.M. Denaro

The Fractional Time-Step Method for incompressible flows

Filippo Maria Denaro

Dept. Engineering – Università della Campania “Luigi Vanvitelli”

Summary
The numerical integration of the Navier-Stokes equations for incompressible flows demands
efficient and accurate solution algorithms for pressure-velocity splitting. Such decoupling was
traditionally performed by adopting the Fractional Time-Step Method that is based on a formal
separation between convective-diffusive momentum terms from the pressure gradient term. This
idea is strictly related to the fundamental theorem on the Helmholtz-Hodge orthogonal
decomposition of a vector field in a finite domain, from which the name projection methods
originates. This note briefly addresses the steps of the method and highlights the issue.
Any notification of error in the note is welcome.

1. Introduction
Consider, in a finite domain , the Navier-Stokes equations for isothermal, incompressible,
Newtonian, viscous flows, written in non-dimensional divergence form (the integral form is not
illustrated here but can be deduced):
 v = 0 (1)

v  1 
+   (vv ) +  p =    v  , (2)
t  Re 
having posed p' = p / 0U r2 and Re = U r Lr / the Reynolds number. The problem consists in
finding the vector fields v and  p ' satisfying (1-2) in ' =   (t0 , t ) for given initial v(x, t0)= v0(x)
and boundary conditions v ( x, t ) = v ( t ) , x  , assigned along the frontier .
v
Let us define, in the interval (t0 , t ) , both the Eulerian acceleration =: a ( t ) and the vector
t
 1 
field R  v ( t )  := −  ( vv ) +    v  such that the momentum equation (2) can be recast as
 Re 
a +  p = R . (3)
Numerical solutions of the equations (1), (2) would require heavy computational efforts caused by
the velocity-pressure coupling, the pressure being a Lagrangian multiplier, not a thermodynamic
state variable. The splitting methodology, specifically devoted to the Navier-Stokes equations,

1
CFD Course F.M. Denaro

which is here illustrated, is based on the Helmholtz-Hodge orthogonal decomposition theorem in a

finite domain [1, 2].
Let start by integrating the momentum equation (2) in a time interval t n , t n+1 = t n + t ( )
t n+1 (4)
v n +1 = v n +  R dt − t  p ' ,
tn

t n+1
having introduced the time-averaged pressure variable   p ' dt = t  p ' = t  p ' .
tn
Now introduce some type of numerical discretization for the term
t n+1 (5)
ˆ + O ( t k +1 ) ,
 R dt = t R
tn

(
ˆ vn+1 , v n , v n−1 ,
where R ) stands for a term discretized at k-th accuracy order by implicit and/or
explicit time integration. This way the Eq.(4) can be rewritten as

ˆ − t p ' + O ( t k +1 )
vn+1 = vn + t R
Avn+1 = q ( vn , vn−1 , ) − t p ' + O ( t k +1 ) (6)
wherein the operator A takes into account for a possible adoption of an implicit time integration in
R̂ (often adopted for the diffusive terms).

2. Step 1: The computation of the intermediate velocity field

The first step of the fractional step method consists in splitting the solution of the velocity field
from that of the pressure field. In this note the pressure-free projection method [3, 4] is illustrated.
That is, let us solve Eq.(6) by disregarding totally the pressure term (and the truncation error)

Av* = q ( vn , vn−1 , ), (7)

being v* an intermediate velocity field that is not divergence-free. Let us highlight that this
intermediate velocity is defined as a discrete field, defined for the time interval under consideration.
Note that the physical boundary conditions (normal and tangential velocity components) are
inserted in q but specific boundary conditions v* ( x ) = v* , x  would be also required for
closing the problem.
In fact, two possible situations are possible:
a) An explicit time integration method, where in (7) one has A= I and the equation is solved for all
interior points, without requiring numerical boundary conditions v* .
b) An implicit time integration method, where some assumption on the boundary conditions v* are
required to solve (7) numerically.

2.1: An expression for the intermediate boundary conditions

In case b), such boundary conditions have to be determined from the knowledge of physical
boundary conditions (that is the physical velocity) and a well-established procedure is based on a

2
CFD Course F.M. Denaro

proposal illustrated in [3]. Indeed, one could be tempted to simply set for (7) the boundary condition
v* = v n+1 but, however, this assumption has consequences in terms of final accuracy [2, 4].

According to the proposal in [3], the intermediate velocity is supposed to be a time-

continuous function v* = v* ( t ) , defined in the interval of the time integration. By using a second

v*
order accurate Taylor expansion at the time tn, considering that := R  v ( t )  and v*n = vn one
t
writes

n n
v*  v 
v *n +1
= v + t
*n
+ O ( t 2 ) = v n + t  +  p   + O ( t 2 )
t  t 

= vn+1 + t pn + O ( t 2 ) (8)

and approximate boundary conditions for the intermediate velocity are deduced according to the
final expression

v*  v n +1 + t  pn . (9)

However, this kind of boundary condition shows some incongruence when the order of the time
integration scheme is equal or greater than k=2. Indeed, the Taylor expansion (8) should be

( )
congruently extended up to order O t k +1 in each single time step.

Furthermore, let us see the Eq.(4) in the framework of the Helmholtz-Hodge decomposition
vn+1 = v* − t  p ' . If we compare this latter to the (9), one deduces  p ' =  pn on the

boundary. It is expected that the local truncation error will degrade to first order magnitude near the
boundaries since during the successive projection steps (detailed in the next section) the constraint
 n p 'n+1 =  n p 'n =  n p '0 on  will be implicitly enforced and a numerical boundary layer [2, 4]

is generated. This fact can deteriorate the solution up to the interior, although the convergence of
discrete method was proved.
 k v*  k −1R
Furthermore, if the expansion (8) is extended to higher order terms, the derivatives = k −1
t k t
have to be carefully expressed since R is a function of v ( t ) not of v* ( t ) . Remember that only at

the initial time one has v*n = vn . This process is then concluded and for the computation of the
successive time step, the intermediate velocity is reset as sketched in Fig.1.
3
CFD Course F.M. Denaro

Figure 1. Sketch of the time integration process for the intermediate velocity.

3. Step 2: The computation of pressure equation.

The next step of the fractional method consists in performing the projection, that is in computing
the gradient of a scalar function  that ensures the divergence-free constraint.
First, we show the relation existing between the pressure p’ and the computed scalar field  . To

this aim, let us assume that the divergence-free velocity vn+1 = v* − t  is substituted substitutes
in Eq.(6) according to

A ( v* − t  ) = q ( vn , vn−1 , ) − t p ' + O ( t k +1 ) ,

A =  p ' + O ( t k ) , (10)

having exploited Eq.(7). The relation (10) shows that in case of an explicit integration (A= I) one
gets directly  =  p ' + O ( t k ) but if an implicit integration is adopted the relation is more

4
CFD Course F.M. Denaro

complex. This fact must be taken into account when considering the boundary conditions (9) where
 pn is required.

After this clarification, the background of the projection step is based on the Helmholtz-Hodge
decomposition [2] in a divergence-free and a curl-free vector fields

v* = vn+1 + t  . (11)

Then, the continuity equation is introduced to project the intermediate velocity onto the subspace of
divergence-free vector functions according to  vn+1 =  ( v* − t ) = 0 . Therefore, one solves

the projection step constituted by the Poisson problem:

1
  =   v* , x  (12)
t

n  ( v* − v n+1 )
1
n  = , x  (13)
t

that admits a unique solution (apart from a constant), being the compatibility condition verified by
the Neumann boundary conditions (13) with the integral continuity constraint. This kind of
boundary condition accomplishes the fact that neither pressure nor intermediate velocity values are
required on the boundary, but only the knowledge of the exact normal velocity is necessary. The
final result expressed by Eq.s(12)-(13) can also be seen, equivalently, as a Poisson problem with
homogeneous Neumann boundary conditions n  = 0 and a modified source term  v* ,
obtained when the divergence operator is defined onto the subspace of vectors with normal
component on the boundary equal to the normal velocity component; let us remark that this
equivalence does not imply that the computed pressure field has a vanishing normal derivative on
the boundary.

4. Step 3: The correction of the intermediate velocity.

Now, the Eq.(11) is used to produce the correction step according to
vn+1 = v* − t  , x  . (14)

5
CFD Course F.M. Denaro

As a consequence, the projection step will ensure that v n+1 is a divergence-free field having the
correct normal component of the velocity on  provided by the boundary condition (13).
Let us highlight one of the issues of the fractional method that appears in the tangential component
of the velocity. Indeed, the (14) produces a tangential velocity component on the boundary
t  ( v* − t  ) , x  that does not necessarily satisfy the exact (physical) velocity component

( ) ( )
t  v n+1 . In other words, one would have t  vn+1 = t  v* − t  + O t k , x  .

In the fractional method, this error is numerically “cancelled” on the boundary because only the
exact tangential component is prescribed for the next time step. But in the region of fluid close to
the boundary, this error can be present, leading to the “numerical boundary layer” addressed in
many papers (for example in [5]).

6
 CFD Course F.M. Denaro

References

1 (2). A. J. Chorin, Numerical Solution of the Navier-Stokes Equations, Math. Comp., 22, 745-762,
1968.
2. (24) F.M. Denaro, On the application of the Helmholtz-Hodge Decomposition in projection
methods for incompressible flows with general boundary conditions, Int. J. Num. Methods in
Fluids, submitted, 2002.
3. (6). J. Kim, P. Moin, Application of A Fractional-Step Method to Incompressible Navier-Stokes
Equations, J. Comp. Phys., 59, 308-323, 1985.
4. (21). D. L. Brown, R. Cortez, M. L. Minion. Accurate Projection Methods for the Incompressible
Navier-Stokes Equations, J. Comp. Physics, 168, 2001.
5. (11). W. E, J.G. Liu, Projection Method I: Convergence and Numerical Boundary Layers, SIAM
J. Num. An., 32, 4, 1017-1057, 1995.
6. F.H. Harlow, J.E. Welch, Numerical Calculation of Time‐Dependent Viscous Incompressible
Flow of Fluid with Free Surface, The Physics of Fluids 8, 2182 (1965)

View publication stats