Journal of Mathematics and Statistics 6 (3): 265-270, 2010

ISSN 1549-3644
© 2010 Science Publications

Simple Improvement of Momentum Interpolation Equation for

Navier-Stoke Equation Solver on Unstructured Grid

Adek Tasri
Department of Mechanical Engineering, Andalas University,
Kampus Unand Limau Manis, Padang 25163, Indonesia

Abstract: Problem statement: Pressure and velocity decoupling have been source of problem in
solving Navier-Stokes and continuity equation particularly in complex collocated grid. The problem of
pressure velocity decoupling is usually reduced by using momentum interpolation to calculate mass
flux at face of control volume. Equation of momentum interpolation was derived by assumption that
the face of cell is equidistant from two neighbor cell centers and face of cell is collinear with two
neighbor cell centers. This assumption is not valid in many unstructured grid and cause significant
error. Approach: In this article a simple improvement of momentum interpolation for using in
unstructured grid was proposed. The improvement was done by added a correction term to original
equation. Results: The method was compared with original method in Kovasznay’s flow and Laminar
Poisseulli Flow. The method was found able to reduce error about 40% in both cases. Conclusion: The
correction added to original momentum interpolation is able to reduce error in Navier-Stokes equation
solver on unstructured grid.

Key words: Momentum interpolation, unstructured mesh, finite volume

INTRODUCTION only very limited number of study related to accuracy

of application of the Rhie and Chow (1983)
It is well known that numerical solving of Navier- interpolation method on unstructured grid have been
Stokes equation in primitive variable on collocated grid published. A popularly used method to implement Rhie
produce checkerboard oscillation if central difference is and Chow (1983) interpolation method on unstructured
used to approximate face velocity in discretised grid is based on the way proposed by Mathur and
continuity equation and pressure gradient in momentum Murthy (1997). Mathur and Murthy (1997) applied
equation (Yu and Kawaguchi, 2002). The oscillation is momentum interpolation to unstructured grid by
caused by decoupling of pressure and velocity in assuming the face is laid down inline and in the middle
discretised equation (Miller and Schmidt, 1988). Rhie between two neighbor cell centers where the
and Chow (1983) first solved this problem for interpolation was claimed formally second order. If the
structured collocated grid by modifying the interpolated face is not inline and in the middle of two neighbor cell
face velocities so as to recognize the pressure difference as the case of most unstructured grid the accuracy of
across the face. The momentum interpolation of Rhie interpolation will down to first order. Therefore in this
and Chow (1983) is popularly used to calculate face study a simple modification to interpolation of Mathur
velocity in discretised momentum equation for SIMPLE and Murthy (1997) in order to solve this problem is
based algorithm since. Later, it was found that the proposed.
original Rhie and Chow method had some drawback In this study, two examples are used as test case
related to dependence of calculation result on under and solve in unstructured grid:
relaxation factor and time step size influence on
• Kovasznay’s flow
convergence of the solution as reported by Majumdar
(1988) and Choi (1999). The way to solve this problem • Laminar Poisseulli flow
also been proposed by some author (Yu and MATERIAL AND METHODS
Kawaguchi, 2002; Nie et al., 2000).
Improvement of Rhie and Chow interpolation in Flow equation: The governing equation of 2D laminar
order to improve accuracy was reported by steady flow of incompressible Navier-Stokes fluids in
Papageorgakopoulos et al. (2000) and Yu et al. (2002). arbitrary domain is continuity and Navier-Stokes
But both methods are for used in structured grid. So far equation:
265
 J. Math. & Stat., 6 (3): 265-270, 2010
ur
divρV = 0 (1) to surface grid cell. For numerical solution, these
equations require discretisation so that they may be
ur ∂P applied to the finite volume cells of an unstructured
divρV1 V = div(µ∇V1 ) − (2) mesh overlaid on the solution domain. The equation, for
∂x
a finite volume grid cell with a finite number f of
ur ∂p identifiable plane faces, may be discretised as:
divρV2 V = div(µ∇V2 ) − (3)
∂y ur r r
∑ (ρφV) .nA = ∑ (µ∇φ) .nA
f
f f
f
f f +q φΩ (8)
Where:
V1 and V2 = Cartesian component of velocity vector where the subscript f denotes value of the variables for
ρ and µ = Density and viscosity respectively the face f. Af is the surface area of face f, whilst
ur
quantities such as φf and V f are interpreted as face
The momentum equation can then be written in
average values of the scalar φ and velocity vector
form of general transport equation for scalar φ: ur
V respectively.
ur φf can be approximated through reconstruction
divρφV = div(µ∇φ) + q φ (4)
from upstream cell value as following:
uur
where, φ is velocity component V1 and V2. qφ in this φf = φP + (∇φ) • rPf (9)
case, represent a gradient of pressure in momentum
uur
equation which known as source term. φP and rPf are φ at upstream cell centre and distance
Finite volume discretisation: According to SIMPLE vector from upstream cell centre to centre of face f
algorithm the continuity equation is discretised to form respectively as shown in Fig. 1.
pressure correction equation. For unstructured grid the The diffusive term is approximated according to
equation is (Mathur and Murthy, 1997): Mathur and Murthy (1997) as following:
ur ur
r (φ − φnb ) A • A
a p p ' = ∑ a nb p 'nb + b (5) (µ∇φ)f .nA f = µ f P ur uur +
nb ds A • es
ur ur (10)
ur ur  ur r A•A 
ρf (Ω p + Ω nb )A • A µ f  (∇φ)f • A − (∇φ)f • es ur uur 
=−  A • es 
a nb ur r 
(a p + a pnb )ds A • es (6)
a P = ∑ a nb Gradient of field variable on face centre, (∇φ)f, is
nb calculated as average of gradient at cell centre P and
cell neighbor, nb (Fig. 1). The cell centre gradient itself
ap and a pnb are point coefficient of dicretised is estimate as:
momentum equation (Eq. 13), Ωp and Ωnb are volume a ur
ur r
of cell P and cell neighbor nb. A and es are face area
∇φ = ∑
Ω f
φf A (11)
vector and unit vector from cell P to cell neighbor nb
respectively. a is a limiter used to avoid introducing Local extrema.
Finite volume discretisation of general transport
equation is done by integrating the equation using
Green-Gauss’s transformation yielding:
ur r r
∫ (ρφV).ndS = ∫ (µ∇φ).ndS + ∫ q dΩ
S S Ω
φ (7)

where the surface integrals are taken over the boundary

of a control region, which for numerical solution will be
a grid cell and the source term qφ is integrated over the
r
volume of this region. n in Eq. 7 is unit vector normal Fig. 1: Finite volume grid
266
 J. Math. & Stat., 6 (3): 265-270, 2010

The last part of discretised equation is a source ur ur

ur  V P + V nb  ρf (∆Ω p + ∆Ω nb )
term which can be estimated as: & f = ρA • 
m −
 2  a p + a pnb
  (15)
ur ur ur ur
q φΩ = ∑ p f A • i j  PP − Pnb r A•A
(12)  − ∇p f • es  r
f
 ds  A • es

Where: r
ur where, ds and es are distance and unit vector from cell
i j = Unit vector in Cartesian coordinate centre P to centre of neighbor cell. First part of
ur
i1 = Unit vector in X direction Mathur’s equation is linear interpolation estimation of
ur ur
i 2 = Unit vector in Y direction
face value velocity, V f in Rhie and Chow (1983)
equation, Eq. 14. Similar to the first part the second part
Pf = Pressure at face of control volume which can be is linear interpolation estimation of second part of Rhie
estimate using linear interpolation from pressure and Chow’s equation. Because of it was based on linear
of cell P and cell neighbor nb interpolation, to be second order accurate, the face
uur should be in the midway of line from cell centre P to
The face velocity vector Vf of convective term neighbor cell nb, the face also should be inline with line
ur r P-nb or gradient of line P-f and gradient of line nb-f are
(Eq. 8) or in form of mass flux m & f = (ρV)f .nA f is
same. But, It is not a case in most technical problem
calculated using momentum interpolation which will be where unstructured grid is used in order to
discussed in the momentum interpolation subsection. accommodate a complex geometry. So that accuracy of
Substitute Eq. 9, 10, 12 and 15 to Eq. 8 and after momentum interpolation will be down to first order.
some mathematical arrangement dicretised momentum
equation can be written as following form: Improvement of momentum interpolation: Mass flux
on face f of unstructured mesh (Fig. 2) can be
a p φ = ∑ a nb φnb + b (13) calculated using velocity on the face as following:
nb

& f = ρA nx u f + ρA ny v f
m (16)
Momentum interpolation:
Current momentum interpolation: It is well known where, Anx, Any, uf, vf are component of face area
that the combination of a collocated grid arrangement normal to x and y direction and face velocity in x and y
with the use of linear interpolation for estimation of direction respectively. uf and vf can be reconstructed
face velocity in continuity equation and estimation of form face velocity of imaginary face f’, mass flux then
pressure gradient in momentum equation (Eq. 9) cause can be written as:
checkerboard oscillation. Rhie and Chow (1983) first
solved this problem for structured collocated grids by r r
& f = ρA nx (u f ' + ∇u f ' • dr) + ρA ny (v f ' + ∇u f ' • dr)
m (17)
modifying the interpolated face velocities by added a
diffusive term. Rhie and Chow idea can be written as
following:

Ω   ∂p   ∂p  
Vf = Vf −   +  (14)
a p   ∂n  f  ∂n f 

Ω and a p represent average of cell volume and

centre coefficient of discretised momentum equation in
left and right of face f. Vf is average of velocity in left
and right face f. Mathur and Murthy (1997) applied the
Rhie and Chow (1983) idea to unstructured grid and
write the equation in form of mass flux instead of face
velocity: Fig. 2: Unstructured mesh
267
 J. Math. & Stat., 6 (3): 265-270, 2010

Equation 13 is rearranged to be:  y  y 

u = 4V1max  1 −  
 h  h 
& f = (ρA nx u f ' + ρA ny v f ' ) +
m v=0 (21)
r r (18)
(ρA nx u f ' • dr + ρA ny∇v f ' • dr) V1max
p = p 0 − 8vρ x
h2
r
In Eq. 18 dr is a vector from centre of imaginary
Where:
face f’ to centre of face f. ∇v f ' and ∇u f ' are gradient of
po = A constant
velocity on the imaginary face f’ which is calculated as h = High of channel
average of left and right cell value. The first term of V1max = Maximal x velocity which is selected to
Eq. 18 is flux on the face f’ which can be calculated YIELD Reynolds number of 200 based on
using Mathur’s equation which accurate to second high of channel
order. The flux then can be written as:
ur ur
ur  V P + V E  ρ (∆Ω + ∆Ω )
& f = ρA • 
m − f P E
 2  a p + a pnb
 
ur ur
 pP − pE s A•A
 − ∇p f • es  ur r + (19)
 ds  A • es
r r
(ρA nx u f ' • dr + ρA ny∇v f 'dr)

The first and second term of right hand side of

Eq. 19 is a original Mathur’s momentum interpolation
and the third is a correction term.

Test case:
Kovasznay flow: In order to test performance of
momentum interpolation, 2D Burgers equation in
domain −0.5 ≤ x ≤ 1, −0.5 ≤ y ≤ 1.5 was used as test case. Fig. 3: Geometry and mesh for Kovasznay flow test
Because of it has complexities and non linear effect, it case
is a good tool to test accuracy of Navier-Stokes solver.
The analytical solution of this flow was reported by
Kovasznay (1948):

u = 1 − e kx cos 2πy
v = λe kx sin 2y / 2π
p = (1 − e 2 λx ) / 2 (20)
Re Re
λ= − 4π
2 4

Re in this equation is Reynolds number.

The test is carried out on square domain which has
two different number of unstructured mesh i.e., 1500
and 1000 as shown in Fig. 3 at Re = 40. Fig. 4: Streamline of Kovasznay flow

Laminar Poisseulli flow: Laminar Poisseulli flow is

used to test accuracy of momentum interpolation in
distorted mesh (Fig. 5). The domain of 0.5 m high and 5
m long have rectangular mesh with distorted mesh in
the middle. The analytical solution of the flow is: Fig. 5: Geometry and mesh for Poisseulli flow test case
268
 J. Math. & Stat., 6 (3): 265-270, 2010

Solving of Kovasznay flow (Fig. 4) and Laminar Accuracy of both momentum interpolation in calculate
Poisseulli flow is done by using standard SIMPLE laminar Poisseulli flow is shown through contour of
algorithm of Patankar (1980) and Spalding (1980) velocity and velocity distribution along centre line of flow
which also used by Mathur and Murthy (1997). The channel as shown in Fig. 6 and 7 respectively.
solution procedure of SIMPLE algorithm is consisting
of five steps: DISCUSSION

(a) Assume initial value of pressure and velocity at all The present research confirms that there are
cell centers differences of calculation result using standard and
(b) Obtain temporary velocity field by solving modified momentum interpolation.
momentum equation, Eq. 8 In Kovasznay’s flow solution, face centre of grid
(c) Obtain pressure correction by solving pressure used in calculation is lied down almost in the middle
correction equation, Eq. 5 between two neighbor cells, so that error of face value
(d) Use pressure correction to update pressure and of flow variable calculated through linear interpolation
velocity field using following equation: as it in Equation 15 is quite low. Both of standard and
modified momentum interpolations produce quite
uuur ur
∑ P A oi
f
'
j
similar velocity field and the velocity fields quite close
to exact value. The differences of both velocity field,
p = p* + α p p ' ϕ = ϕ* − f
(22)
ap calculated using standard and modified momentum
interpolation, is visually indistinguishable. The
where p* and φ* are initial value of pressure and differences can be shown through L1 error of velocity
temporary velocity field respectively field. From Table 1 it can be shown that L1 error of
(e) Repeat steps (b)-(d) where corrected pressure from velocity field calculated using standard interpolation is
step (d) is used as new pressure guess until a almost two fold of it calculated using modified one.
converged solution is obtained
Table 1: Error in Kovasznay flow test (L1U and L1V are L1 error of u
RESULTS and v velocity)
Mathur’s Improved Mathur’s
Accuracy of standard momentum interpolation and Mesh size momentum intp. momentum intrp.
modified momentum interpolation in calculate 1500 L1u = 1.1e-2 L1u = 6.5e-3
L1v = 5.8e-3 L1v = 2.7e-3
Kovasznay flow can be compared through the value of 1000 L1u = 2.1e-2 L1u = 1.4e-2
L1 error of x and y velocities as shown in Table 1. L1v = 9.8e-3 L1v = 5.3e-3

(a)

Fig. 7: Velocity along horizontal centre line of channel.

(b) (A)Velocity distribution without distortion on
the mesh. (B) Velocity distribution calculated
Fig. 6: X-Velocity contour around distorted mesh. (a) using modification of Mathur’s momentum
Calculated using standard Mathur’s momentum interpolation. (C) Velocity distribution
interpolation; (b) Calculated using modification calculated using standard Mathur’s momentum
of momentum interpolation interpolation
269
 J. Math. & Stat., 6 (3): 265-270, 2010

Laminar Poisseulli flow was calculated using Mathur, S.R., J.Y. Murthy, 1997. A pressure based
distorted mesh in the middle of domain where face methods for unstructured meshes. Numer. Heat
centre almost do not lied down in the middle and inline Trans. Part B., 31: 195-215. DOI:
with both cell neighbor. Difference of two velocity 10.1080/10407799708915105
fields, calculated by using standard and modified Miller, T.F and W. Schmidt, 1988. Use of a pressure-
momentum interpolation, can be shown easily through weighted interpolation method for the solution of
velocity contour and velocity distribution along incompressible Navier-stokes equations on a non
horizontal centre-line of flow channel. Velocity contour staggered grids system: Numer. Heat Trans.,
calculated using standard momentum interpolation is 14: 213-233. DOI: 10.1080/10407788808913641
more distorted than velocity contour calculated using Nie, J.H., Z.Y. Li, Q.W. Wang and W.Q. Tao, 2000. A
modified one, Fig. 6. Whilst error in velocity method for viscous incompressible flows with
distribution calculates using standard momentum simplified collocated grid system. Proceeding of
interpolation is almost double of error in velocity the 21st Century Symposium on Energy
calculated using modified momentum interpolation as Engineering in the 21st Century, Jan. 9-13,
shown in Fig. 7. Institute of Thermal Engineering, Hong Kong,
Some improvements of accuracy have been pp: 177-183.
achieved by adding a correction in order to reduce error Papageorgakopoulos, J., G. Arampatzis, D. Assimacopoulos
caused by using linear interpolation in derivation of and N.C. Markatos, 2000. Enhancement of the
momentum-interpolation equation. But linear momentum interpolation method on non-staggered
interpolation is still used to calculate face value of flow grids. Int. J. Numer. Methods Fluids, 33: 1-2. DOI:
variable in equation of gradient of flow variable at cell 10.1002/(SICI)1097-0363(20000515)33:1<1::AID-
centre, Eq. 11. Further research to find a correction for FLD978>3.3.CO;2-S
this linear interpolation is very interesting in order to Patankar, S.V., 1980. Numerical Heat Transfer and
reduce error of calculation. Fluid Flow. 1st Edn., Taylor and Francis, New
York, ISBN 10: 0891165223, pp: 214.
CONCLUSION Rhie, C.M. and W.L. Chow, 1983. Numerical study of
the turbulent flow past an air foil with trailing edge
A simple improvement have been done to Mathur’s separation. AIAAJ, 21: 1525-1532. DOI:
momentum interpolation in order to reduce error of 10.2514/3.8284
Navier-Stokes Equation solver when face center do not Spalding, D.B., 1980. A mathematical modeling of
laid down inline and in the middle between two fluid dynamics, heat transfer and mass transfer
neighbor cell centers. The improvement which done in processes. Imperial College.
this work can reduce L1 error of finite volume Yu, B. and Y. Kawaguchi, 2002. Checkerboard
calculation almost 40% of its original value as shown in pressure prediction due to the under relaxation
two sample problems used in this study. factor and time step size for non staggered grid
with momentum interpolation method: Numerical
REFERENCES Heat Trans. Part B., 41: 85-94. DOI:
10.1080/104077902753385027
Choi, S.K., 1999. Note on the use of momentum Yu, B., W. Tao, J. Wei, Y. Kawaguchi and T.
interpolation method for unsteady flows. Numer. Tagawa et al., 2002. Discussion on momentum
Heat Trans. A., 36: 545-550. DOI: interpolation methods for collocated grids of
10.1080/104077899274679 incompressible flow. Numer. Heat Trans. Part B.,
Kovasznay, L.I.G., 1948. Laminar flow behind a two- 42: 141-166. DOI: 10.1080/10407790190053879
dimensional grid: Proc. the Cambridge Philosop.
Soc., 44: 58-62. DOI:
10.1017/S0305004100023999
Majumdar, S., 1988. Role of under-relaxation in
momentum interpolation for calculation of flow
with non staggered grids. Numer. Heat Trans.,
13: 125-132. DOI: 10.1080/10407788808913607

270