SlideShare a Scribd company logo
1 of 8
Proceedings of ICEAE 2009
Email addresses:
bharat.agrawal@zeusnumerix.com (Agrawal, B.R.)
saurabh.pandey@zeusnumerix.com (Pandey, S.)
1Corresponding Authors
1. INTRODUCTION
Grid generation is a crucial first step in the whole process
of numerically simulating a physics problem. Currently
structured and unstructured meshes are widely used in the
industry applications. However, they have the limitation that
they cannot be automatically generated over complex
geometries thus requiring large number of man hours in
manually creating these grids. This arise a need for an
automatic grid generation process. [1][2] Hereby, software
has been developed which can automatically generate an oct-
tree based Cartesian mesh over geometries of any
complexity. The code has been tested on various geometries
and is capable of generating 10 million cells in 10 seconds
on a serial machine. To begin the process of writing a
complex solver on these meshes an incompressible Euler
solver has been written. The solver so far is tested for very
simple standard test cases. The results over these meshes are
also discussed in the later sections.
2. Background and Motivation
Computational Fluid Dynamics is the field of numerical
equations approximately requires the use of computers to
solve any engineering problem. The advancements in
computer technology since 1970 have made it possible to
use it to do large amount of computation in a small amount
of time [1].
But this was regarding solution of PDEs, grid generation
part in CFD still lagged behind requiring a lot of human
involvement, therefore it was a time consuming process.
This produced the need of a quick and automatic grid
generation algorithm with minimal human involvement.
Cartesian meshes were introduced to cater to this increasing
demand of quick mesh generation generator [3][4].
Cartesian Mesh was good for low fidelity simulation but
if the desired accuracy of the solution was high the number
of cells in the grid increased drastically as there is no
concept of clustering in such mesh. Hence Oct-tree based
Cartesian Grid was developed. These grids have the
advantage of clustering and that too adaptive. If the gradient
in a cell is going beyond a threshold the cell can be divided
into eight equal cells to capture the gradient correctly
[3][5][6].
Oct-tree based Cartesian grid is also very advantageous
when the geometry of the body is very complex. If there is a
sharp corner or irregularities in the surface, it is very
difficult to capture the geometry using fixed size cells. But
using adaptive oct-tree grids the cells can be refined to a
very high level i.e. very small volume at places where very
small size cells are required to capture the geometry. This
Revisiting Projection Methods over Automatic Oct-tree Meshes
Agrawal, Bharat R. a,1
, Pandey, Saurabha
a
Zeus Numerix Pvt. Ltd., Mumbai
Abstract: This work documents the ongoing effort on using projection methods over automatic
oct-tree meshes. The mesh generation algorithm uses a quick and automatic method to create
solver and geometry adaptive oct-tree mesh on a basic Cartesian grid. The mesh is a non-body
fitted type of mesh and therefore is independent of the topology of the geometry. The method
takes multiple water tight triangulated surface(s) as an input. The software then automatically
creates oct-tree based Cartesian grids both outside as well as inside the geometry. The
intersection of the cells with the triangles of the geometry is done using a very robust and
efficient polygon clipping algorithm allowing the capture of geometry with 100% accuracy.
Over this mesh an incompressible Euler solver based on projection method is being written. The
solver is tested working for some simple test cases.
Key words: Projection Method, Oct-tree, Cartesian Mesh, Incompressible Euler
ICEAE Paper GN-053
Proceedings of ICEAE 2009
improvement in capturing the geometry in turn improves the
accuracy of the solver [1].
These advantages of oct-tree grids over Cartesian grids
have a lot of promise in correctly simulating flow around
very complex geometries. As these grid can be generated in
no time over very complex geometries, if coupled with a
quick compressible or incompressible solver can used as a
design tool. Therefore oct-tree grid can be used to bring in
the utilization of CFD in designing phase [1][4].
3. Geometry Input
The software is aimed at reading geometries of any
complexities and which can be easily generated using CAD
software. There are various formats in which the CAD
software can save the geometry, however, triangulated
surfaces provides the best robustness required by a meshing
algorithm. The robustness parameter considered here is
primarily the water tightness of the geometry. The water
tight requirement arises from a function (further referred as
'inout function') which determines whether a point is outside
or inside the geometry. Further, the software can take
polygon surface also which is converted internally into a
triangulated surface by the software. The software can also
take multiple geometries as well which may or may not
intersect in the space.
Geometry is stored in a half edge data structure wherein
each edge is stored as a twin edge, each of these twin edges
are same edges with opposite direction. Each edge belongs
to two different adjacent triangles. This data structure
provides an excellent and efficient traversal over the edges
and vertices of all triangles and also their respective
neighbouring triangles.
4 Grid Data Structure
An oct-tree data structure has been used to store each cell
where each cell can have none or 8 children. The cells which
do not have children are the real existing cells of the grid
which will be used in the solver whereas the cells which
have children exist only for the completeness of the grid.
The data structure is designed in such a manner where any
neighbouring entity of a particular cell can be directly
accessed. Also, for a particular face the cells which share
that face can be directly accessed as right or left cell of that
face [1].
The data structure allows an object of class cell cut info,
face cut info and edge cut info to be associated with the cell,
face and edge respectively. Each of these cut info objects
stores the interaction of the cell with the geometry. Cell cut
info stores the surface polygons lying inside the cells, face
cut info stores the edges of these surface polygons which
completely lie in the faces and edge cut info stores the points
at which the edge is intersecting the geometry
5 Solver Requirements
5.1 Volume accuracy
First requirement is the accurate computation of the
volumes of all the cells in the computation domain. The
degree of accuracy in calculating the volumes of the cells
directly dictates the fidelity of the solution [3][5][6]. A large
number of cells in the oct-tree Cartesian mesh are regular
cuboids in shape. These regular cells have well defined plane
boundaries where the planes are x=const, y=const and
z=const. However, some of the cells which intersect with the
geometry can be chopped in any manner which creates
complicated polyhedron. Hence it is these cells whose
volume calculation requires computation effort and finally
dictating the solution fidelity.
5.2 Neighbour traversal
Second crucial requirement is the quick access to the
neighbouring entities of a cell such as neighbouring cells and
faces. This is taken care by the design of data structure
wherein each cell has pointers to its six faces and each face
has pointers to its left and right cells [1].
5.3 Level difference in adjacent cells
There is one constraint to ensure first order accuracy
while calculating gradients on the face. It is that the levels of
cells sharing a face, an edge or a vertex cannot differ by
more than one. In Fig. 1, the dotted lines are the refinement
happening because of this constraint [7].
5.4 Boundary carpet
Other constraint is to provide a layer of smooth equal
level cells over the whole geometry to ensure there is no
cells which have level less than the defined boundary level
(BL) very close to the boundary. This case arises when a cut
cell is very small in volume and since the adjacent cell is not
cutting the boundary it can remain at a lower level as
compared to the boundary level. This is done by making sure
each of the cells which are a candidate for being a cut cell
has all neighbours having level equal to that cell. The carpet
provides a smooth layer in which the boundary layer effects
Figure 1 Division of cells due to neighbours
Bharat Raj Agrawal, Saurabh Pandey
Proceedings of ICEAE 2009
Figure 2 Division of cell due to geometry surface
can be well captured. In Fig. 2, black region is the internal of
the geometry and the gray cells are candidates for being the
cut cells. Hence the cell to the right is divided because of the
constraint as shown by the dotted lines [7].
6 Inout function
To automatically generate cells and decide which cell
lies in which domain, a function which can tell whether a
point is inside or outside the geometry is required. The
'inout' function is a generalization of this requirement; it
determines the zone in which a point lies. The zones are
referred to space which is divided by various intersecting or
non intersecting water tight boundaries.
The function takes a point coordinate as input and then
fires a ray towards the positive x infinity. While doing so the
intersections with the various boundaries present in the space
is counted. If for a boundary, the ray crosses the boundary
odd number of times then the point lies inside that boundary
and if the ray crosses even number of times then the point
lies outside the boundary.
7 Grid geometry interactions
This is the most crucial part for any automatic meshing
algorithm where the level of accuracy of capturing the
geometry surface in the cut cells is decided. Since this level
directly dictates the fidelity of the solution, the present
algorithm achieves maximum possible accuracy in capturing
geometry surface inside the grid cells.
Each cell when created is populated with a list of
intersecting triangles. Here a geometry triangle may or may
not be associated with more than one grid cell depending
completely upon how many cells it intersects. Whenever a
bigger cell is divided to create new child cells, the list of
triangles of the parent cell gets distributed among the
children. Further, cut info stored in the parent cell is
discarded and the new cut info is found for newly generated
child cells. This is an important feature to allow AMR in the
region close to boundaries. Fig. 3 shows how a boundary
cell is divided into two parts by the geometry.
In order to find the exact grid geometry interactions, each
of the triangles associated with all the cells are clipped by
the respective cell boundaries. This is done using the
Sutherland-Hodgman polygon clipping algorithm. During
this process all triangles are clipped by all the cells they are
associated with. This gives a set of polygons in each
boundary cell which defines the cell cut info accurately to
the accuracy of the geometry input. Then the volume of each
cut cell is computed using divergence theorem on all the
faces of the cell and the clipped polygons present in that cell
[1][8].
8 Adaptive mesh refinement (AMR)
The basic grid which goes into the solver is generally
very coarse and meant to be subsequently refined as the
gradients develop across the coarse cells. The present
algorithm is capable of refining any particular cell or region
at any given time. This enables the solver algorithm to check
for high gradient cells in the computational domain and send
them for refinement after the solver iteration is complete.
Further, since the grid geometry interaction is dynamically
found as soon as a cell is generated there remains no
restriction on refining the grid at any point of time.
9 Parallelization
To run very complex solver algorithm on a large number
of cells parallelization is the only suitable method. For this
the mesh is divided into various zones where each of the
zones is solved by a single processor. The division of whole
mesh into zones is done using Hilbert-Peano space filling
curves which ensures minimum zone boundaries [9]. This is
required to minimise the intercommunication between
different processors.
In the zone division algorithm, each of the cells is given
a Hilbert value which arranges all the grid cells in a linear
manner. This Hilbert value is given using an algorithm given
by Hamilton [10]. Then depending on the total load of
computing cell cuts, the domain is partitioned such that each
processor gets equal load.
The parallelization has been tested on several nodes of
Synergy and PARAM YUVA clusters of Center for
Development of Advanced Computing (CDAC). Each of the
machines in the cluster has 64 GB RAM and 4 Xeon
Processors with each processor having 4 cores. On a single
node the code could generate 13 million leaf cells in 10
minutes.
Figure 3 Grid geometry interactions in a grid cell
ICEAE Paper GN-053
Proceedings of ICEAE 2009
10 Incompressible Euler Solver
The solver in progress on the oct-tree mesh is for a
constant density and inviscid fluid. The work done is based
on the incompressible solver developed by Popinet in 2003
[7]. The velocity and pressure are dependent on three spatial
as well as temporal variables. Given as follows
ͱ{˲ ˳ ˴ ˮ{ ˯{˲ ˳ ˴ ˮ{
 - ˰{˲ ˳ ˴ ˮ{ - ˱{˲ ˳ ˴ ˮ{˫
˜ ˜{˲ ˳ ˴ ˮ{
The Incompressible Euler equations used are
{ ͱ{ .˯{ ͱ{ . ˰{ ͱ{ . ˱{ ͱ{ . J
ͱ Ŵ
Apart from the standard inlet and outlet boundary
conditions the wall boundary is implemented with no
orthogonal flow component at the wall.
ͱ{˲ ˳ ˴ ˮ{ ; Ŵ ƒ– –Š‡ ™ƒŽŽ
where n is outward normal unit vector.
Projection method with fractional-step is used to solve
the above equations. In this method velocity U is known at
time t and pressure P at (t – ∆t/2). These values are stored at
the cell center.
As a first step in projection method U'' is computed using
the advection term.
ͱ ͱ
.V È$
,
In the above equation A is the advection term given by
[(U. )U].
In next step this U'' distribution is given to the projection
operator which gives the new velocity at (t+∆t). It also
updates the pressure field to the new pressure distribution at
(t+∆t/2).
11 Advection Term
The advection term is solved by applying divergence
theorem in each cell.
Vˮ- ˮÈŶ
ˤˢ
˕˥ˬˬ
{{ͱ {ͱ{ˮ- ˮÈŶ
ˤˢ
˕˥ˬˬ
{{ͱ {ͱ{ˮ- ˮÈŶ
ˤˢ
˕˥ˬˬ
ͱˮ- ˮÈŶ
˘II˥J
{ͱˮ- ˮÈŶ
;{ˤV
Rewriting the above equation for a cell in an oct-tree
mesh
Vˮ- ˮÈŶ
ŵ
ˢI˥ˬˬ
Ӝ ͱ˦II˥
ˮ- ˮÈŶ
ͱ˦II˥ ;
ˮ- ˮÈŶ
ӝ ˓˦II˥
˘II˥J
where Uface is the extrapolated velocity at the respective
face of the cell. Vcell is the volume of the cell and Aface is area
of the face in consideration. n is the outward normal unit
vector to the face.
The velocities are known at the center of the cell but to
calculate the advection term we need to extrapolate the
velocities at the faces. This extrapolation is performed using
Taylor series approximation. Hence the velocity at the face
can be calculated as follows
ͱˤˮ- ˮÈŶ ͱˮ
-
ˤ
Ŷ
ͱˮ
ˤ
-
ˮ
Ŷ
ͱˮ
ˮ
- ˛{ ˤŶ
ˮŶ
{
where d is the direction perpendicular to the plane of the
face, ∆d is the distance of the face from cell center. Now the
time derivative of velocity in the above equation can be
replaced by spatial derivatives using the Euler equations thus
giving
ͱ˲ˮ- ˮÈŶ ͱˮ
- @
˲
Ŷ
. ˯
ˮ
Ŷ
D
ͱˮ
˲
. ˰
ˮ
Ŷ
ͱˮ
˳
. ˱
ˮ
Ŷ
ͱˮ
˴
.
ˮ
Ŷ
˜ˮ
ͱ˳ˮ- ˮÈŶ ͱˮ
- @
˳
Ŷ
. ˰
ˮ
Ŷ
D
ͱˮ
˳
. ˯
ˮ
Ŷ
ͱˮ
˲
. ˱
ˮ
Ŷ
ͱˮ
˴
.
ˮ
Ŷ
˜ˮ
ͱ˴ˮ- ˮÈŶ ͱˮ
- @
˴
Ŷ
. ˱
ˮ
Ŷ
D
ͱˮ
˴
. ˯
ˮ
Ŷ
ͱˮ
˲
. ˰
ˮ
Ŷ
ͱˮ
˳
.
ˮ
Ŷ
˜ˮ
where Ux, Uy and Uz are velocities at the faces along x, y
and z directions respectively. Once the velocity at the face is
known from both sides of the cell simple upwinding is used
to decide the final velocity at the center of the face. But to
calculate the advection term we need normal velocity
component at the faces of a cell. Also for this method to
satisfy conservation equation the normal velocity should be
divergence free. Therefore to make the normal velocity
divergence free projection method is applied by solving the
following equation
˜ ˡ È$
where P is pressure at cell center, Ut+∆t/2
is velocity at the
face center. The resulting pressure field is then used to
correct the normal velocity component. Therefore the
corrected normal velocity is given by
˯ˤˮ- ˮÈŶ ˯ˤȊ . { ˜{ˤ
While correcting the normal velocities at the face center,
pressure gradient is also calculated at the cell center using
the gradient values at the faces of the cell. Calculation of
advection term also requires that the velocity is known at the
face center along with the normal component. It is done
using the same Taylor series approximation as mentioned
above. But this time the velocities for approximation are
Bharat Raj Agrawal, Saurabh Pandey
Proceedings of ICEAE 2009
now taken as the average of the normal corrected velocities
at the faces of the cell as opposed to the cell centered
velocity. The final velocity is then calculated using the
equation
ͱˤˮ- ˮÈŶ ˡˤ
Ȋ
. ˜
Here Ud' is face centerd velocity calculated by applying
the Taylor series approximation and upwinding at the face.
P is the interpolated value of gradient at the face. The
linear interpolation is performed between the cells sharing
the face.
12 Projection Method
The projection method due to Chorin [11] is used to
update the velocity. According to this method intermediate
velocity is calculated by solving the momentum equation
ignoring the pressure variation. Then this intermediate
velocity is used to compute pressure field such that the
corrected velocity is divergence free. The projection method
utilizes the Helmholtz-Hodge decomposition of a vector
field U', which can be written as follows
ͱȊ
ͱˮ- ˮÈŶ
- ˜ (1)
where
ͱ È$
Ŵ (2)
and the boundary condition is
ͱ ; Ŵ
In general, the Helmholtz-Hodge decomposition of a
vector field describes it in two components, one a divergence
free vector field and the other is the gradient of a scalar field.
Taking divergence of (1) and using the divergence free
result of (2) will lead to the Poisson equation
Ŷ
˜ ͱȊ
Hence the final divergence free velocity is given by
ͱˮ- ˮÈŶ
ͱȊ
. ˜
13 Numerical Validation
The first set of numerical validation is performed on the
Poisson Solver. A unit cubical domain is taken with origin
situated at the center of the cube. The cells in the cubical
domain are initialized with the following divergence
ͱ{˲ ˳ ˴{ . ${˫$
- ˬ$
- ˭${ •‹{ ˫˲{ •‹{ ˬ˳{ •‹{ ˭˴{
with k = l = m = 1. The exact solution of the Poisson
Equation with the above divergence as the source term is
given by
˜{˲ ˳ ˴{ •‹{ ˫˲{ •‹{ ˬ˳{ •‹{ ˭˴{ - ˕ (3)
where C is a constant. The domain is initialized with a
constant zero pressure throughout the domain. The above
problem is simulated with variable number of cells in the
domain. The error in pressure is measured with respect to the
pressure given by (3) and residue is the change in pressure
compared to the last iteration. The results are shown in Fig.
4 and 5.
The validation of Euler Solver is performed by following
the work of Minion [12] and Almgren et al. [13] where the
whole domain is initialized with the following velocity
˯{˲ ˳{ ŵ . Ŷ …‘•{Ŷ ˲{ •‹{Ŷ ˳{
˰{˲ ˳{ ŵ - Ŷ •‹{Ŷ ˲{ …‘•{Ŷ ˳{
The exact solution of the Euler Equation for these initial
conditions is given by
˯{˲ ˳ ˮ{ ŵ . Ŷ …‘• Ŷ {˲ . ˮ{ •‹ Ŷ {˳ . ˮ{
˰{˲ ˳ ˮ{ ŵ - Ŷ •‹ Ŷ {˲ . ˮ{ …‘• Ŷ {˳ . ˮ{
˜{˲ ˳ ˮ{ .Ŷ …‘• Ÿ {˲ . ˮ{ . …‘• Ÿ {˳ . ˮ{
Figure 5 Error plot for Poisson solver
Figure 4 Residue plot for Poisson solver
ICEAE Paper GN-053
Proceedings of ICEAE 2009
The domain is taken again as a unit cube and periodic
boundary condition is imposed. The variation of error and
residue with time is shown for two different cases of
refinement in Fig. 6 and 7. Since the above solution is
bounded the variation of upper bound of the solution is also
shown for different grid refinement cases in Fig. 8.
14 Results
The software has been tested on various geometries and
is tested capable of generating a million cells in few seconds
on a serial machine.
15 Mesh
Polygon clipping is evident in Fig. 9 of grid over a shark.
Clearly the boundary surface is very well captured even
when the cell size is very large as compared to the geometry
triangles' sizes. Fig. 10 shows the mesh over a mirage
geometry.
16 Heat Conduction Solver
A 3D heat conduction solver is also written. In Fig. 11,
there is a temperature gradient below the front part of the
shark. This temperature gradient is extending farther from
the surface as compared to other parts of surface. This is
because of the two fins of the shark, one going inside the
plane of paper and one extending outwards. These fins can
be seen in the grid result over shark. Thus this temperature
gradient clearly signifies the working of the software in 3
dimensional framework.
17 3D Incompressible Euler Solver
Fig. 12 shows pressure over the surface of a mirage.
18 Scalability
Using the PARAM YUVA cluster a scalability study is
also carried out. Result is in Fig. 13.
Figure 9 Mesh over man and shark at very low BL
Figure 8 Maximum velocity variation
Figure 7 Velocity residue plot
Figure 6 Velocity error plot
Bharat Raj Agrawal, Saurabh Pandey
Proceedings of ICEAE 2009
19 Future work
The present algorithm is targeted towards complex
geometries like full scale aircrafts, automobile, etc. To create
grid over such geometries in order to compute values such as
drag may require boundary cells of micron scale. This easily
translates into a billion of cells which is not possible to be
generated and simultaneously stored on a serial machine.
This arise a need for the parallelization of the grid generation
process. Currently an Euler incompressible solver is in
process of development which when fully developed will
require to be merged with the grid generation process. The
unified algorithm will also utilize the multigrid acceleration
which has been found to improve the convergence time by
an order of magnitude. The current grid lacks the advantage
provided by the conformal grids for computation of
boundary layer. For this purpose, adding the capability of
structured cells near the boundary to create a hybrid mesh is
also being formulated.
20 Conclusion
A very fast and robust automatic mesher has been created
for which the CFD industry has been eagerly waiting. Since
the mesher is automatic and quick, one of the potential uses
of this mesher can be directly seen in any aerodynamic
sensitive design.
21 Acknowledgement
We would like to thank CDAC for providing motivation
and platform to test our codes. Without their help this project
could not be possible. We would also like to express our
gratitude towards Basant K. Gupta, Sandeep S. and
Shubhayu C.
REFERENCES
1. Aftosmis, M.J., Melton, J.E., and Berger, M.J., Adaptive
Cartesian Mesh Generation. Chapter 22 in Handbook of
Grid Generation, Thompson, J, Weatherhill, N., and Soni, B.
eds. CRC Press 1998.
Figure 13 Scalability of solver code up to 1024 cores
Figure 12 Pressure variation over a mirage
Figure 11 Heat solution over shark after 500 iterations
Figure 10 Grid over mirage
Proceedings of ICEAE 2009
2. Berger, M.J., and Aftosmis, M.J., Aspects (and aspect ratios)
of Cartesian mesh methods. Proceedings of the 16
International Conference on Numerical Methods in Fluid
Mechanics, Arcachon, France, July 1998.
3. Aftosmis, M.J., “Solution adaptive Cartesian
for aerodynamic flows with complex geometries
Karman Institute for Fluid Dynamics, Lecture Series 199702,
Rhode-Saint-Genèse, Belgium, Mar. 3-7,
4. Aftosmis, M.J., Berger, M.J., Melton, J.E.,
efficient Cartesian mesh generation for component
geometry,” AIAA Paper 97-0196, Jan. 1997.
5. Berger, M.J., and Melton, J.E., “An Accuracy
Cartesian Grid Method for Steady
Geometries,” Proc. 8thrIntl. Conf. Hyp. Pro
Stonybrook, NY, Jun., 1995. also RIACS Report 95
6. De Zeeuw, D., and Powell, K., “An
Cartesian Mesh Solver for the Euler Equations,
91-1542, 1991.
7. Popinet, S. 2003. “Gerris: A tree based adaptive solver for the
incompressible Euler equations in complex geometries.
Comp. Phys. 190, 572–600.
8. Sutherland, I.E., and Hodgman, G.W.,
clipping,” Comm of the ACM, 17(1):32-42,
9. Aftosmis, M. J., Berger, M. J., and Adomovic
multigrid on Cartesian meshes with complex geometry
of the 8th Intl. Conf. on Parallel CFD.
(To appear Elsivier Press). Jun. 2000.
10. C. Hamilton. “Compact Hilbert indices.
CS-2006-07, Dalhousie University, Faculty of Computer
Science, Jul 2006.
11. Chorin, A. J., “Numerical Solution of the
Equations.” Mathematics of Computation
(Oct., 1968), pp. 745-762.
12. Minion, M. L., “A projection method for
grids.” J. Comp. Phys. 127 (1996), 158-178.
13. Almgren, A.S., Bell, J.B., Colella, P., Howell,
Welcome, M.L., A conservative adaptive projection method
for the variable density incompressible Navier
equations, J. Comp. Phys. 142 (1998) 1–46.
Bharat Raj Agrawal had the graduate
education at Department of Aerospace
Engineering, Indian Institute of Technology
Bombay and has been working as
Engineer at Zeus Numerix Pvt Ltd
current interest is in CFD
Saurabh Pandey had the graduate education
at Department of Aerospace Engineering,
Indian Institute of Technology Bombay and
has been working as CAE Engineer at Zeus
Numerix Pvt Ltd. His current interest is in
CFD
Berger, M.J., and Aftosmis, M.J., Aspects (and aspect ratios)
Proceedings of the 16th
International Conference on Numerical Methods in Fluid
France, July 1998.
Solution adaptive Cartesian grid methods
complex geometries”, von
, Lecture Series 199702,
7, 1997.
Aftosmis, M.J., Berger, M.J., Melton, J.E., “Robust and
generation for component based
Jan. 1997.
An Accuracy Test of a
flow in Complex
Conf. Hyp. Problems, Uppsala,
NY, Jun., 1995. also RIACS Report 95-02.
Adaptively Refined
the Euler Equations,” AIAA Paper
adaptive solver for the
equations in complex geometries.” J.
“Re-entrant polygon
42, 1974.
Adomovicius, G. “Parallel
Cartesian meshes with complex geometry” Proc.
Trondhiem Norway.
Compact Hilbert indices.” Technical Report
, Faculty of Computer
Numerical Solution of the Navier-Stokes
Computation, Vol. 22, No. 104
A projection method for locally refined
178.
Almgren, A.S., Bell, J.B., Colella, P., Howell, L.H.,
adaptive projection method
density incompressible Navier–Stokes
46.
had the graduate
Department of Aerospace
Engineering, Indian Institute of Technology
been working as CAE
Engineer at Zeus Numerix Pvt Ltd. His
had the graduate education
rtment of Aerospace Engineering,
Indian Institute of Technology Bombay and
has been working as CAE Engineer at Zeus
Numerix Pvt Ltd. His current interest is in
ICEAE Paper GN-053

More Related Content

What's hot

4 tracking objects of deformable shapes
4 tracking objects of deformable shapes4 tracking objects of deformable shapes
4 tracking objects of deformable shapes
prjpublications
 
Schematic model for analyzing mobility and detection of multiple
Schematic model for analyzing mobility and detection of multipleSchematic model for analyzing mobility and detection of multiple
Schematic model for analyzing mobility and detection of multiple
IAEME Publication
 
Brain tumor segmentation using asymmetry based histogram thresholding and k m...
Brain tumor segmentation using asymmetry based histogram thresholding and k m...Brain tumor segmentation using asymmetry based histogram thresholding and k m...
Brain tumor segmentation using asymmetry based histogram thresholding and k m...
eSAT Publishing House
 

What's hot (17)

4 tracking objects of deformable shapes
4 tracking objects of deformable shapes4 tracking objects of deformable shapes
4 tracking objects of deformable shapes
 
Microarray spot partitioning by autonomously organising maps through contour ...
Microarray spot partitioning by autonomously organising maps through contour ...Microarray spot partitioning by autonomously organising maps through contour ...
Microarray spot partitioning by autonomously organising maps through contour ...
 
Comparison of Phase Only Correlation and Neural Network for Iris Recognition
Comparison of Phase Only Correlation and Neural Network for Iris RecognitionComparison of Phase Only Correlation and Neural Network for Iris Recognition
Comparison of Phase Only Correlation and Neural Network for Iris Recognition
 
ALGORITHMIC AND ARCHITECTURAL OPTIMIZATION OF A 3D RECONSTRUCTION MEDICAL IMA...
ALGORITHMIC AND ARCHITECTURAL OPTIMIZATION OF A 3D RECONSTRUCTION MEDICAL IMA...ALGORITHMIC AND ARCHITECTURAL OPTIMIZATION OF A 3D RECONSTRUCTION MEDICAL IMA...
ALGORITHMIC AND ARCHITECTURAL OPTIMIZATION OF A 3D RECONSTRUCTION MEDICAL IMA...
 
Schematic model for analyzing mobility and detection of multiple
Schematic model for analyzing mobility and detection of multipleSchematic model for analyzing mobility and detection of multiple
Schematic model for analyzing mobility and detection of multiple
 
A Pattern Classification Based approach for Blur Classification
A Pattern Classification Based approach for Blur ClassificationA Pattern Classification Based approach for Blur Classification
A Pattern Classification Based approach for Blur Classification
 
IRJET- Significant Neural Networks for Classification of Product Images
IRJET- Significant Neural Networks for Classification of Product ImagesIRJET- Significant Neural Networks for Classification of Product Images
IRJET- Significant Neural Networks for Classification of Product Images
 
Automatic rectification of perspective distortion from a single image using p...
Automatic rectification of perspective distortion from a single image using p...Automatic rectification of perspective distortion from a single image using p...
Automatic rectification of perspective distortion from a single image using p...
 
DCT AND DFT BASED BIOMETRIC RECOGNITION AND MULTIMODAL BIOMETRIC SECURITY
DCT AND DFT BASED BIOMETRIC RECOGNITION AND MULTIMODAL BIOMETRIC SECURITYDCT AND DFT BASED BIOMETRIC RECOGNITION AND MULTIMODAL BIOMETRIC SECURITY
DCT AND DFT BASED BIOMETRIC RECOGNITION AND MULTIMODAL BIOMETRIC SECURITY
 
Detection of crossover & bifurcation points on a retinal fundus image by ...
Detection of crossover & bifurcation points on a retinal fundus image by ...Detection of crossover & bifurcation points on a retinal fundus image by ...
Detection of crossover & bifurcation points on a retinal fundus image by ...
 
Detection of crossover & bifurcation points on a retinal fundus image by anal...
Detection of crossover & bifurcation points on a retinal fundus image by anal...Detection of crossover & bifurcation points on a retinal fundus image by anal...
Detection of crossover & bifurcation points on a retinal fundus image by anal...
 
GEOMETRIC CORRECTION FOR BRAILLE DOCUMENT IMAGES
GEOMETRIC CORRECTION FOR BRAILLE DOCUMENT IMAGESGEOMETRIC CORRECTION FOR BRAILLE DOCUMENT IMAGES
GEOMETRIC CORRECTION FOR BRAILLE DOCUMENT IMAGES
 
Ok3425102516
Ok3425102516Ok3425102516
Ok3425102516
 
Zeeshan.ali.presentations
Zeeshan.ali.presentationsZeeshan.ali.presentations
Zeeshan.ali.presentations
 
Object Recognition Using Shape Context with Canberra Distance
Object Recognition Using Shape Context with Canberra DistanceObject Recognition Using Shape Context with Canberra Distance
Object Recognition Using Shape Context with Canberra Distance
 
VARIATIONAL MONTE-CARLO APPROACH FOR ARTICULATED OBJECT TRACKING
VARIATIONAL MONTE-CARLO APPROACH FOR ARTICULATED OBJECT TRACKINGVARIATIONAL MONTE-CARLO APPROACH FOR ARTICULATED OBJECT TRACKING
VARIATIONAL MONTE-CARLO APPROACH FOR ARTICULATED OBJECT TRACKING
 
Brain tumor segmentation using asymmetry based histogram thresholding and k m...
Brain tumor segmentation using asymmetry based histogram thresholding and k m...Brain tumor segmentation using asymmetry based histogram thresholding and k m...
Brain tumor segmentation using asymmetry based histogram thresholding and k m...
 

Similar to Revisiting Projection Methods over Automatic Oct-tree Meshes

Vertex Culling illustration at SBR07
Vertex Culling illustration at SBR07Vertex Culling illustration at SBR07
Vertex Culling illustration at SBR07
Syoyo Fujita
 
ModelingOfUnsegmentedCloudPointData-RP-SanjayShukla
ModelingOfUnsegmentedCloudPointData-RP-SanjayShuklaModelingOfUnsegmentedCloudPointData-RP-SanjayShukla
ModelingOfUnsegmentedCloudPointData-RP-SanjayShukla
Sanjay Shukla
 
DSP IEEE paper
DSP IEEE paperDSP IEEE paper
DSP IEEE paper
prreiya
 
ADVANCED ALGORITHMS FOR ETCHING SIMULATION OF 3D MEMS-TUNABLE LASERS
ADVANCED ALGORITHMS FOR ETCHING SIMULATION OF 3D MEMS-TUNABLE LASERSADVANCED ALGORITHMS FOR ETCHING SIMULATION OF 3D MEMS-TUNABLE LASERS
ADVANCED ALGORITHMS FOR ETCHING SIMULATION OF 3D MEMS-TUNABLE LASERS
ijctcm
 
Adaptive load balancing techniques in global scale grid environment
Adaptive load balancing techniques in global scale grid environmentAdaptive load balancing techniques in global scale grid environment
Adaptive load balancing techniques in global scale grid environment
iaemedu
 

Similar to Revisiting Projection Methods over Automatic Oct-tree Meshes (20)

M ESH S IMPLIFICATION V IA A V OLUME C OST M EASURE
M ESH S IMPLIFICATION V IA A V OLUME C OST M EASUREM ESH S IMPLIFICATION V IA A V OLUME C OST M EASURE
M ESH S IMPLIFICATION V IA A V OLUME C OST M EASURE
 
Design a 3D CAD Model of a Stealth Aircraft and Generate Mesh to Optimize Mes...
Design a 3D CAD Model of a Stealth Aircraft and Generate Mesh to Optimize Mes...Design a 3D CAD Model of a Stealth Aircraft and Generate Mesh to Optimize Mes...
Design a 3D CAD Model of a Stealth Aircraft and Generate Mesh to Optimize Mes...
 
Real-Time Cutting of Organs with Scissor
Real-Time Cutting of Organs with ScissorReal-Time Cutting of Organs with Scissor
Real-Time Cutting of Organs with Scissor
 
A Density Control Based Adaptive Hexahedral Mesh Generation Algorithm
A Density Control Based Adaptive Hexahedral Mesh Generation AlgorithmA Density Control Based Adaptive Hexahedral Mesh Generation Algorithm
A Density Control Based Adaptive Hexahedral Mesh Generation Algorithm
 
Vertex Culling illustration at SBR07
Vertex Culling illustration at SBR07Vertex Culling illustration at SBR07
Vertex Culling illustration at SBR07
 
grid.pptx
grid.pptxgrid.pptx
grid.pptx
 
ModelingOfUnsegmentedCloudPointData-RP-SanjayShukla
ModelingOfUnsegmentedCloudPointData-RP-SanjayShuklaModelingOfUnsegmentedCloudPointData-RP-SanjayShukla
ModelingOfUnsegmentedCloudPointData-RP-SanjayShukla
 
Garbage Classification Using Deep Learning Techniques
Garbage Classification Using Deep Learning TechniquesGarbage Classification Using Deep Learning Techniques
Garbage Classification Using Deep Learning Techniques
 
INVESTIGATIONS OF THE INFLUENCES OF A CNN’S RECEPTIVE FIELD ON SEGMENTATION O...
INVESTIGATIONS OF THE INFLUENCES OF A CNN’S RECEPTIVE FIELD ON SEGMENTATION O...INVESTIGATIONS OF THE INFLUENCES OF A CNN’S RECEPTIVE FIELD ON SEGMENTATION O...
INVESTIGATIONS OF THE INFLUENCES OF A CNN’S RECEPTIVE FIELD ON SEGMENTATION O...
 
3 article azojete vol 7 24 33
3 article azojete vol 7 24 333 article azojete vol 7 24 33
3 article azojete vol 7 24 33
 
Prediction of Deflection and Stresses of Laminated Composite Plate with Arti...
Prediction of Deflection and Stresses of Laminated Composite  Plate with Arti...Prediction of Deflection and Stresses of Laminated Composite  Plate with Arti...
Prediction of Deflection and Stresses of Laminated Composite Plate with Arti...
 
20320140503011
2032014050301120320140503011
20320140503011
 
DSP IEEE paper
DSP IEEE paperDSP IEEE paper
DSP IEEE paper
 
Two-dimensional Block of Spatial Convolution Algorithm and Simulation
Two-dimensional Block of Spatial Convolution Algorithm and SimulationTwo-dimensional Block of Spatial Convolution Algorithm and Simulation
Two-dimensional Block of Spatial Convolution Algorithm and Simulation
 
Application of Neural Network for Cell Formation in Group Technology
Application of Neural Network for Cell Formation in Group  TechnologyApplication of Neural Network for Cell Formation in Group  Technology
Application of Neural Network for Cell Formation in Group Technology
 
CONVOLUTIONAL NEURAL NETWORK BASED FEATURE EXTRACTION FOR IRIS RECOGNITION
CONVOLUTIONAL NEURAL NETWORK BASED FEATURE EXTRACTION FOR IRIS RECOGNITION CONVOLUTIONAL NEURAL NETWORK BASED FEATURE EXTRACTION FOR IRIS RECOGNITION
CONVOLUTIONAL NEURAL NETWORK BASED FEATURE EXTRACTION FOR IRIS RECOGNITION
 
CONVOLUTIONAL NEURAL NETWORK BASED FEATURE EXTRACTION FOR IRIS RECOGNITION
CONVOLUTIONAL NEURAL NETWORK BASED FEATURE EXTRACTION FOR IRIS RECOGNITION CONVOLUTIONAL NEURAL NETWORK BASED FEATURE EXTRACTION FOR IRIS RECOGNITION
CONVOLUTIONAL NEURAL NETWORK BASED FEATURE EXTRACTION FOR IRIS RECOGNITION
 
ADVANCED ALGORITHMS FOR ETCHING SIMULATION OF 3D MEMS-TUNABLE LASERS
ADVANCED ALGORITHMS FOR ETCHING SIMULATION OF 3D MEMS-TUNABLE LASERSADVANCED ALGORITHMS FOR ETCHING SIMULATION OF 3D MEMS-TUNABLE LASERS
ADVANCED ALGORITHMS FOR ETCHING SIMULATION OF 3D MEMS-TUNABLE LASERS
 
Medial Axis Transformation based Skeletonzation of Image Patterns using Image...
Medial Axis Transformation based Skeletonzation of Image Patterns using Image...Medial Axis Transformation based Skeletonzation of Image Patterns using Image...
Medial Axis Transformation based Skeletonzation of Image Patterns using Image...
 
Adaptive load balancing techniques in global scale grid environment
Adaptive load balancing techniques in global scale grid environmentAdaptive load balancing techniques in global scale grid environment
Adaptive load balancing techniques in global scale grid environment
 

More from Abhishek Jain

More from Abhishek Jain (20)

Unsteady Problems & Separation Studies @ Zeus Numerix
Unsteady Problems & Separation Studies @ Zeus NumerixUnsteady Problems & Separation Studies @ Zeus Numerix
Unsteady Problems & Separation Studies @ Zeus Numerix
 
Optimization in CFD and Case Studies
Optimization in CFD and Case StudiesOptimization in CFD and Case Studies
Optimization in CFD and Case Studies
 
Aero Acoustic Field & its Modeling @ Zeus Numerix
Aero Acoustic Field & its Modeling @ Zeus NumerixAero Acoustic Field & its Modeling @ Zeus Numerix
Aero Acoustic Field & its Modeling @ Zeus Numerix
 
CEM Workshop Lectures (11/11): CEMExpert Usage of Almond Geometry for RCS Cal...
CEM Workshop Lectures (11/11): CEMExpert Usage of Almond Geometry for RCS Cal...CEM Workshop Lectures (11/11): CEMExpert Usage of Almond Geometry for RCS Cal...
CEM Workshop Lectures (11/11): CEMExpert Usage of Almond Geometry for RCS Cal...
 
CEM Workshop Lectures (10/11): Numerical Modeling of Radar Absorbing Materials
CEM Workshop Lectures (10/11): Numerical Modeling of Radar Absorbing MaterialsCEM Workshop Lectures (10/11): Numerical Modeling of Radar Absorbing Materials
CEM Workshop Lectures (10/11): Numerical Modeling of Radar Absorbing Materials
 
CEM Workshop Lectures (9/11): Modelling Electromagnetics Field
CEM Workshop Lectures (9/11):  Modelling Electromagnetics FieldCEM Workshop Lectures (9/11):  Modelling Electromagnetics Field
CEM Workshop Lectures (9/11): Modelling Electromagnetics Field
 
CEM Workshop Lectures (8/11): Method of moments
CEM Workshop Lectures (8/11):  Method of momentsCEM Workshop Lectures (8/11):  Method of moments
CEM Workshop Lectures (8/11): Method of moments
 
CEM Workshop Lectures (7/11): PO/PTD Solver for RCS Computation
CEM Workshop Lectures (7/11): PO/PTD Solver for RCS ComputationCEM Workshop Lectures (7/11): PO/PTD Solver for RCS Computation
CEM Workshop Lectures (7/11): PO/PTD Solver for RCS Computation
 
CEM Workshop Lectures (6/11): FVTD Method in CEM
CEM Workshop Lectures (6/11): FVTD Method in CEMCEM Workshop Lectures (6/11): FVTD Method in CEM
CEM Workshop Lectures (6/11): FVTD Method in CEM
 
CEM Workshop Lectures (5/11): Best Practices in RCS Prediction
CEM Workshop Lectures (5/11): Best Practices in RCS PredictionCEM Workshop Lectures (5/11): Best Practices in RCS Prediction
CEM Workshop Lectures (5/11): Best Practices in RCS Prediction
 
CEM Workshop Lectures (4/11): CEM of High Frequency Methods
CEM Workshop Lectures (4/11): CEM of High Frequency MethodsCEM Workshop Lectures (4/11): CEM of High Frequency Methods
CEM Workshop Lectures (4/11): CEM of High Frequency Methods
 
CEM Workshop Lectures (3/11): Mesh Generation in CEM
CEM Workshop Lectures (3/11): Mesh Generation in CEMCEM Workshop Lectures (3/11): Mesh Generation in CEM
CEM Workshop Lectures (3/11): Mesh Generation in CEM
 
CEM Workshop Lectures (1/11): ABC of CEM and RCS
CEM Workshop Lectures (1/11): ABC of CEM and RCSCEM Workshop Lectures (1/11): ABC of CEM and RCS
CEM Workshop Lectures (1/11): ABC of CEM and RCS
 
CFD Lecture (8/8): CFD in Chemical Systems
CFD Lecture (8/8): CFD in Chemical SystemsCFD Lecture (8/8): CFD in Chemical Systems
CFD Lecture (8/8): CFD in Chemical Systems
 
CFD Lecture (7/8): Best Practices in CFD
CFD Lecture (7/8): Best Practices in CFDCFD Lecture (7/8): Best Practices in CFD
CFD Lecture (7/8): Best Practices in CFD
 
CFD Lecture (6/8): Solvers- Incompressible Flow
CFD Lecture (6/8): Solvers- Incompressible FlowCFD Lecture (6/8): Solvers- Incompressible Flow
CFD Lecture (6/8): Solvers- Incompressible Flow
 
CFD Lecture (5/8): Solvers- Compressible Flow
CFD Lecture (5/8): Solvers- Compressible FlowCFD Lecture (5/8): Solvers- Compressible Flow
CFD Lecture (5/8): Solvers- Compressible Flow
 
CFD Lecture (4/8): Compressible Flow- Basics
CFD Lecture (4/8): Compressible Flow- BasicsCFD Lecture (4/8): Compressible Flow- Basics
CFD Lecture (4/8): Compressible Flow- Basics
 
CFD Lecture (3/8): Mesh Generation in CFD
CFD Lecture (3/8): Mesh Generation in CFDCFD Lecture (3/8): Mesh Generation in CFD
CFD Lecture (3/8): Mesh Generation in CFD
 
CFD Lecture (2/8): Fluid Mechanics: CFD Perspective
CFD Lecture (2/8): Fluid Mechanics: CFD PerspectiveCFD Lecture (2/8): Fluid Mechanics: CFD Perspective
CFD Lecture (2/8): Fluid Mechanics: CFD Perspective
 

Recently uploaded

Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FME
Cloud Frontiers:  A Deep Dive into Serverless Spatial Data and FMECloud Frontiers:  A Deep Dive into Serverless Spatial Data and FME
Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FME
Safe Software
 
Modular Monolith - a Practical Alternative to Microservices @ Devoxx UK 2024
Modular Monolith - a Practical Alternative to Microservices @ Devoxx UK 2024Modular Monolith - a Practical Alternative to Microservices @ Devoxx UK 2024
Modular Monolith - a Practical Alternative to Microservices @ Devoxx UK 2024
Victor Rentea
 

Recently uploaded (20)

Artificial Intelligence Chap.5 : Uncertainty
Artificial Intelligence Chap.5 : UncertaintyArtificial Intelligence Chap.5 : Uncertainty
Artificial Intelligence Chap.5 : Uncertainty
 
presentation ICT roal in 21st century education
presentation ICT roal in 21st century educationpresentation ICT roal in 21st century education
presentation ICT roal in 21st century education
 
Exploring Multimodal Embeddings with Milvus
Exploring Multimodal Embeddings with MilvusExploring Multimodal Embeddings with Milvus
Exploring Multimodal Embeddings with Milvus
 
Navigating the Deluge_ Dubai Floods and the Resilience of Dubai International...
Navigating the Deluge_ Dubai Floods and the Resilience of Dubai International...Navigating the Deluge_ Dubai Floods and the Resilience of Dubai International...
Navigating the Deluge_ Dubai Floods and the Resilience of Dubai International...
 
Emergent Methods: Multi-lingual narrative tracking in the news - real-time ex...
Emergent Methods: Multi-lingual narrative tracking in the news - real-time ex...Emergent Methods: Multi-lingual narrative tracking in the news - real-time ex...
Emergent Methods: Multi-lingual narrative tracking in the news - real-time ex...
 
Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FME
Cloud Frontiers:  A Deep Dive into Serverless Spatial Data and FMECloud Frontiers:  A Deep Dive into Serverless Spatial Data and FME
Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FME
 
ProductAnonymous-April2024-WinProductDiscovery-MelissaKlemke
ProductAnonymous-April2024-WinProductDiscovery-MelissaKlemkeProductAnonymous-April2024-WinProductDiscovery-MelissaKlemke
ProductAnonymous-April2024-WinProductDiscovery-MelissaKlemke
 
Strategize a Smooth Tenant-to-tenant Migration and Copilot Takeoff
Strategize a Smooth Tenant-to-tenant Migration and Copilot TakeoffStrategize a Smooth Tenant-to-tenant Migration and Copilot Takeoff
Strategize a Smooth Tenant-to-tenant Migration and Copilot Takeoff
 
Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...
Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...
Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...
 
TrustArc Webinar - Unlock the Power of AI-Driven Data Discovery
TrustArc Webinar - Unlock the Power of AI-Driven Data DiscoveryTrustArc Webinar - Unlock the Power of AI-Driven Data Discovery
TrustArc Webinar - Unlock the Power of AI-Driven Data Discovery
 
"I see eyes in my soup": How Delivery Hero implemented the safety system for ...
"I see eyes in my soup": How Delivery Hero implemented the safety system for ..."I see eyes in my soup": How Delivery Hero implemented the safety system for ...
"I see eyes in my soup": How Delivery Hero implemented the safety system for ...
 
Exploring the Future Potential of AI-Enabled Smartphone Processors
Exploring the Future Potential of AI-Enabled Smartphone ProcessorsExploring the Future Potential of AI-Enabled Smartphone Processors
Exploring the Future Potential of AI-Enabled Smartphone Processors
 
How to Troubleshoot Apps for the Modern Connected Worker
How to Troubleshoot Apps for the Modern Connected WorkerHow to Troubleshoot Apps for the Modern Connected Worker
How to Troubleshoot Apps for the Modern Connected Worker
 
2024: Domino Containers - The Next Step. News from the Domino Container commu...
2024: Domino Containers - The Next Step. News from the Domino Container commu...2024: Domino Containers - The Next Step. News from the Domino Container commu...
2024: Domino Containers - The Next Step. News from the Domino Container commu...
 
Corporate and higher education May webinar.pptx
Corporate and higher education May webinar.pptxCorporate and higher education May webinar.pptx
Corporate and higher education May webinar.pptx
 
Modular Monolith - a Practical Alternative to Microservices @ Devoxx UK 2024
Modular Monolith - a Practical Alternative to Microservices @ Devoxx UK 2024Modular Monolith - a Practical Alternative to Microservices @ Devoxx UK 2024
Modular Monolith - a Practical Alternative to Microservices @ Devoxx UK 2024
 
[BuildWithAI] Introduction to Gemini.pdf
[BuildWithAI] Introduction to Gemini.pdf[BuildWithAI] Introduction to Gemini.pdf
[BuildWithAI] Introduction to Gemini.pdf
 
AWS Community Day CPH - Three problems of Terraform
AWS Community Day CPH - Three problems of TerraformAWS Community Day CPH - Three problems of Terraform
AWS Community Day CPH - Three problems of Terraform
 
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, AdobeApidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
 
EMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWER
EMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWEREMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWER
EMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWER
 

Revisiting Projection Methods over Automatic Oct-tree Meshes

  • 1. Proceedings of ICEAE 2009 Email addresses: bharat.agrawal@zeusnumerix.com (Agrawal, B.R.) saurabh.pandey@zeusnumerix.com (Pandey, S.) 1Corresponding Authors 1. INTRODUCTION Grid generation is a crucial first step in the whole process of numerically simulating a physics problem. Currently structured and unstructured meshes are widely used in the industry applications. However, they have the limitation that they cannot be automatically generated over complex geometries thus requiring large number of man hours in manually creating these grids. This arise a need for an automatic grid generation process. [1][2] Hereby, software has been developed which can automatically generate an oct- tree based Cartesian mesh over geometries of any complexity. The code has been tested on various geometries and is capable of generating 10 million cells in 10 seconds on a serial machine. To begin the process of writing a complex solver on these meshes an incompressible Euler solver has been written. The solver so far is tested for very simple standard test cases. The results over these meshes are also discussed in the later sections. 2. Background and Motivation Computational Fluid Dynamics is the field of numerical equations approximately requires the use of computers to solve any engineering problem. The advancements in computer technology since 1970 have made it possible to use it to do large amount of computation in a small amount of time [1]. But this was regarding solution of PDEs, grid generation part in CFD still lagged behind requiring a lot of human involvement, therefore it was a time consuming process. This produced the need of a quick and automatic grid generation algorithm with minimal human involvement. Cartesian meshes were introduced to cater to this increasing demand of quick mesh generation generator [3][4]. Cartesian Mesh was good for low fidelity simulation but if the desired accuracy of the solution was high the number of cells in the grid increased drastically as there is no concept of clustering in such mesh. Hence Oct-tree based Cartesian Grid was developed. These grids have the advantage of clustering and that too adaptive. If the gradient in a cell is going beyond a threshold the cell can be divided into eight equal cells to capture the gradient correctly [3][5][6]. Oct-tree based Cartesian grid is also very advantageous when the geometry of the body is very complex. If there is a sharp corner or irregularities in the surface, it is very difficult to capture the geometry using fixed size cells. But using adaptive oct-tree grids the cells can be refined to a very high level i.e. very small volume at places where very small size cells are required to capture the geometry. This Revisiting Projection Methods over Automatic Oct-tree Meshes Agrawal, Bharat R. a,1 , Pandey, Saurabha a Zeus Numerix Pvt. Ltd., Mumbai Abstract: This work documents the ongoing effort on using projection methods over automatic oct-tree meshes. The mesh generation algorithm uses a quick and automatic method to create solver and geometry adaptive oct-tree mesh on a basic Cartesian grid. The mesh is a non-body fitted type of mesh and therefore is independent of the topology of the geometry. The method takes multiple water tight triangulated surface(s) as an input. The software then automatically creates oct-tree based Cartesian grids both outside as well as inside the geometry. The intersection of the cells with the triangles of the geometry is done using a very robust and efficient polygon clipping algorithm allowing the capture of geometry with 100% accuracy. Over this mesh an incompressible Euler solver based on projection method is being written. The solver is tested working for some simple test cases. Key words: Projection Method, Oct-tree, Cartesian Mesh, Incompressible Euler
  • 2. ICEAE Paper GN-053 Proceedings of ICEAE 2009 improvement in capturing the geometry in turn improves the accuracy of the solver [1]. These advantages of oct-tree grids over Cartesian grids have a lot of promise in correctly simulating flow around very complex geometries. As these grid can be generated in no time over very complex geometries, if coupled with a quick compressible or incompressible solver can used as a design tool. Therefore oct-tree grid can be used to bring in the utilization of CFD in designing phase [1][4]. 3. Geometry Input The software is aimed at reading geometries of any complexities and which can be easily generated using CAD software. There are various formats in which the CAD software can save the geometry, however, triangulated surfaces provides the best robustness required by a meshing algorithm. The robustness parameter considered here is primarily the water tightness of the geometry. The water tight requirement arises from a function (further referred as 'inout function') which determines whether a point is outside or inside the geometry. Further, the software can take polygon surface also which is converted internally into a triangulated surface by the software. The software can also take multiple geometries as well which may or may not intersect in the space. Geometry is stored in a half edge data structure wherein each edge is stored as a twin edge, each of these twin edges are same edges with opposite direction. Each edge belongs to two different adjacent triangles. This data structure provides an excellent and efficient traversal over the edges and vertices of all triangles and also their respective neighbouring triangles. 4 Grid Data Structure An oct-tree data structure has been used to store each cell where each cell can have none or 8 children. The cells which do not have children are the real existing cells of the grid which will be used in the solver whereas the cells which have children exist only for the completeness of the grid. The data structure is designed in such a manner where any neighbouring entity of a particular cell can be directly accessed. Also, for a particular face the cells which share that face can be directly accessed as right or left cell of that face [1]. The data structure allows an object of class cell cut info, face cut info and edge cut info to be associated with the cell, face and edge respectively. Each of these cut info objects stores the interaction of the cell with the geometry. Cell cut info stores the surface polygons lying inside the cells, face cut info stores the edges of these surface polygons which completely lie in the faces and edge cut info stores the points at which the edge is intersecting the geometry 5 Solver Requirements 5.1 Volume accuracy First requirement is the accurate computation of the volumes of all the cells in the computation domain. The degree of accuracy in calculating the volumes of the cells directly dictates the fidelity of the solution [3][5][6]. A large number of cells in the oct-tree Cartesian mesh are regular cuboids in shape. These regular cells have well defined plane boundaries where the planes are x=const, y=const and z=const. However, some of the cells which intersect with the geometry can be chopped in any manner which creates complicated polyhedron. Hence it is these cells whose volume calculation requires computation effort and finally dictating the solution fidelity. 5.2 Neighbour traversal Second crucial requirement is the quick access to the neighbouring entities of a cell such as neighbouring cells and faces. This is taken care by the design of data structure wherein each cell has pointers to its six faces and each face has pointers to its left and right cells [1]. 5.3 Level difference in adjacent cells There is one constraint to ensure first order accuracy while calculating gradients on the face. It is that the levels of cells sharing a face, an edge or a vertex cannot differ by more than one. In Fig. 1, the dotted lines are the refinement happening because of this constraint [7]. 5.4 Boundary carpet Other constraint is to provide a layer of smooth equal level cells over the whole geometry to ensure there is no cells which have level less than the defined boundary level (BL) very close to the boundary. This case arises when a cut cell is very small in volume and since the adjacent cell is not cutting the boundary it can remain at a lower level as compared to the boundary level. This is done by making sure each of the cells which are a candidate for being a cut cell has all neighbours having level equal to that cell. The carpet provides a smooth layer in which the boundary layer effects Figure 1 Division of cells due to neighbours
  • 3. Bharat Raj Agrawal, Saurabh Pandey Proceedings of ICEAE 2009 Figure 2 Division of cell due to geometry surface can be well captured. In Fig. 2, black region is the internal of the geometry and the gray cells are candidates for being the cut cells. Hence the cell to the right is divided because of the constraint as shown by the dotted lines [7]. 6 Inout function To automatically generate cells and decide which cell lies in which domain, a function which can tell whether a point is inside or outside the geometry is required. The 'inout' function is a generalization of this requirement; it determines the zone in which a point lies. The zones are referred to space which is divided by various intersecting or non intersecting water tight boundaries. The function takes a point coordinate as input and then fires a ray towards the positive x infinity. While doing so the intersections with the various boundaries present in the space is counted. If for a boundary, the ray crosses the boundary odd number of times then the point lies inside that boundary and if the ray crosses even number of times then the point lies outside the boundary. 7 Grid geometry interactions This is the most crucial part for any automatic meshing algorithm where the level of accuracy of capturing the geometry surface in the cut cells is decided. Since this level directly dictates the fidelity of the solution, the present algorithm achieves maximum possible accuracy in capturing geometry surface inside the grid cells. Each cell when created is populated with a list of intersecting triangles. Here a geometry triangle may or may not be associated with more than one grid cell depending completely upon how many cells it intersects. Whenever a bigger cell is divided to create new child cells, the list of triangles of the parent cell gets distributed among the children. Further, cut info stored in the parent cell is discarded and the new cut info is found for newly generated child cells. This is an important feature to allow AMR in the region close to boundaries. Fig. 3 shows how a boundary cell is divided into two parts by the geometry. In order to find the exact grid geometry interactions, each of the triangles associated with all the cells are clipped by the respective cell boundaries. This is done using the Sutherland-Hodgman polygon clipping algorithm. During this process all triangles are clipped by all the cells they are associated with. This gives a set of polygons in each boundary cell which defines the cell cut info accurately to the accuracy of the geometry input. Then the volume of each cut cell is computed using divergence theorem on all the faces of the cell and the clipped polygons present in that cell [1][8]. 8 Adaptive mesh refinement (AMR) The basic grid which goes into the solver is generally very coarse and meant to be subsequently refined as the gradients develop across the coarse cells. The present algorithm is capable of refining any particular cell or region at any given time. This enables the solver algorithm to check for high gradient cells in the computational domain and send them for refinement after the solver iteration is complete. Further, since the grid geometry interaction is dynamically found as soon as a cell is generated there remains no restriction on refining the grid at any point of time. 9 Parallelization To run very complex solver algorithm on a large number of cells parallelization is the only suitable method. For this the mesh is divided into various zones where each of the zones is solved by a single processor. The division of whole mesh into zones is done using Hilbert-Peano space filling curves which ensures minimum zone boundaries [9]. This is required to minimise the intercommunication between different processors. In the zone division algorithm, each of the cells is given a Hilbert value which arranges all the grid cells in a linear manner. This Hilbert value is given using an algorithm given by Hamilton [10]. Then depending on the total load of computing cell cuts, the domain is partitioned such that each processor gets equal load. The parallelization has been tested on several nodes of Synergy and PARAM YUVA clusters of Center for Development of Advanced Computing (CDAC). Each of the machines in the cluster has 64 GB RAM and 4 Xeon Processors with each processor having 4 cores. On a single node the code could generate 13 million leaf cells in 10 minutes. Figure 3 Grid geometry interactions in a grid cell
  • 4. ICEAE Paper GN-053 Proceedings of ICEAE 2009 10 Incompressible Euler Solver The solver in progress on the oct-tree mesh is for a constant density and inviscid fluid. The work done is based on the incompressible solver developed by Popinet in 2003 [7]. The velocity and pressure are dependent on three spatial as well as temporal variables. Given as follows ͱ{˲ ˳ ˴ ˮ{ ˯{˲ ˳ ˴ ˮ{ - ˰{˲ ˳ ˴ ˮ{ - ˱{˲ ˳ ˴ ˮ{˫ ˜ ˜{˲ ˳ ˴ ˮ{ The Incompressible Euler equations used are { ͱ{ .˯{ ͱ{ . ˰{ ͱ{ . ˱{ ͱ{ . J ͱ Ŵ Apart from the standard inlet and outlet boundary conditions the wall boundary is implemented with no orthogonal flow component at the wall. ͱ{˲ ˳ ˴ ˮ{ ; Ŵ ƒ– –Š‡ ™ƒŽŽ where n is outward normal unit vector. Projection method with fractional-step is used to solve the above equations. In this method velocity U is known at time t and pressure P at (t – ∆t/2). These values are stored at the cell center. As a first step in projection method U'' is computed using the advection term. ͱ ͱ .V È$ , In the above equation A is the advection term given by [(U. )U]. In next step this U'' distribution is given to the projection operator which gives the new velocity at (t+∆t). It also updates the pressure field to the new pressure distribution at (t+∆t/2). 11 Advection Term The advection term is solved by applying divergence theorem in each cell. Vˮ- ˮÈŶ ˤˢ ˕˥ˬˬ {{ͱ {ͱ{ˮ- ˮÈŶ ˤˢ ˕˥ˬˬ {{ͱ {ͱ{ˮ- ˮÈŶ ˤˢ ˕˥ˬˬ ͱˮ- ˮÈŶ ˘II˥J {ͱˮ- ˮÈŶ ;{ˤV Rewriting the above equation for a cell in an oct-tree mesh Vˮ- ˮÈŶ ŵ ˢI˥ˬˬ Ӝ ͱ˦II˥ ˮ- ˮÈŶ ͱ˦II˥ ; ˮ- ˮÈŶ ӝ ˓˦II˥ ˘II˥J where Uface is the extrapolated velocity at the respective face of the cell. Vcell is the volume of the cell and Aface is area of the face in consideration. n is the outward normal unit vector to the face. The velocities are known at the center of the cell but to calculate the advection term we need to extrapolate the velocities at the faces. This extrapolation is performed using Taylor series approximation. Hence the velocity at the face can be calculated as follows ͱˤˮ- ˮÈŶ ͱˮ - ˤ Ŷ ͱˮ ˤ - ˮ Ŷ ͱˮ ˮ - ˛{ ˤŶ ˮŶ { where d is the direction perpendicular to the plane of the face, ∆d is the distance of the face from cell center. Now the time derivative of velocity in the above equation can be replaced by spatial derivatives using the Euler equations thus giving ͱ˲ˮ- ˮÈŶ ͱˮ - @ ˲ Ŷ . ˯ ˮ Ŷ D ͱˮ ˲ . ˰ ˮ Ŷ ͱˮ ˳ . ˱ ˮ Ŷ ͱˮ ˴ . ˮ Ŷ ˜ˮ ͱ˳ˮ- ˮÈŶ ͱˮ - @ ˳ Ŷ . ˰ ˮ Ŷ D ͱˮ ˳ . ˯ ˮ Ŷ ͱˮ ˲ . ˱ ˮ Ŷ ͱˮ ˴ . ˮ Ŷ ˜ˮ ͱ˴ˮ- ˮÈŶ ͱˮ - @ ˴ Ŷ . ˱ ˮ Ŷ D ͱˮ ˴ . ˯ ˮ Ŷ ͱˮ ˲ . ˰ ˮ Ŷ ͱˮ ˳ . ˮ Ŷ ˜ˮ where Ux, Uy and Uz are velocities at the faces along x, y and z directions respectively. Once the velocity at the face is known from both sides of the cell simple upwinding is used to decide the final velocity at the center of the face. But to calculate the advection term we need normal velocity component at the faces of a cell. Also for this method to satisfy conservation equation the normal velocity should be divergence free. Therefore to make the normal velocity divergence free projection method is applied by solving the following equation ˜ ˡ È$ where P is pressure at cell center, Ut+∆t/2 is velocity at the face center. The resulting pressure field is then used to correct the normal velocity component. Therefore the corrected normal velocity is given by ˯ˤˮ- ˮÈŶ ˯ˤȊ . { ˜{ˤ While correcting the normal velocities at the face center, pressure gradient is also calculated at the cell center using the gradient values at the faces of the cell. Calculation of advection term also requires that the velocity is known at the face center along with the normal component. It is done using the same Taylor series approximation as mentioned above. But this time the velocities for approximation are
  • 5. Bharat Raj Agrawal, Saurabh Pandey Proceedings of ICEAE 2009 now taken as the average of the normal corrected velocities at the faces of the cell as opposed to the cell centered velocity. The final velocity is then calculated using the equation ͱˤˮ- ˮÈŶ ˡˤ Ȋ . ˜ Here Ud' is face centerd velocity calculated by applying the Taylor series approximation and upwinding at the face. P is the interpolated value of gradient at the face. The linear interpolation is performed between the cells sharing the face. 12 Projection Method The projection method due to Chorin [11] is used to update the velocity. According to this method intermediate velocity is calculated by solving the momentum equation ignoring the pressure variation. Then this intermediate velocity is used to compute pressure field such that the corrected velocity is divergence free. The projection method utilizes the Helmholtz-Hodge decomposition of a vector field U', which can be written as follows ͱȊ ͱˮ- ˮÈŶ - ˜ (1) where ͱ È$ Ŵ (2) and the boundary condition is ͱ ; Ŵ In general, the Helmholtz-Hodge decomposition of a vector field describes it in two components, one a divergence free vector field and the other is the gradient of a scalar field. Taking divergence of (1) and using the divergence free result of (2) will lead to the Poisson equation Ŷ ˜ ͱȊ Hence the final divergence free velocity is given by ͱˮ- ˮÈŶ ͱȊ . ˜ 13 Numerical Validation The first set of numerical validation is performed on the Poisson Solver. A unit cubical domain is taken with origin situated at the center of the cube. The cells in the cubical domain are initialized with the following divergence ͱ{˲ ˳ ˴{ . ${˫$ - ˬ$ - ˭${ •‹{ ˫˲{ •‹{ ˬ˳{ •‹{ ˭˴{ with k = l = m = 1. The exact solution of the Poisson Equation with the above divergence as the source term is given by ˜{˲ ˳ ˴{ •‹{ ˫˲{ •‹{ ˬ˳{ •‹{ ˭˴{ - ˕ (3) where C is a constant. The domain is initialized with a constant zero pressure throughout the domain. The above problem is simulated with variable number of cells in the domain. The error in pressure is measured with respect to the pressure given by (3) and residue is the change in pressure compared to the last iteration. The results are shown in Fig. 4 and 5. The validation of Euler Solver is performed by following the work of Minion [12] and Almgren et al. [13] where the whole domain is initialized with the following velocity ˯{˲ ˳{ ŵ . Ŷ …‘•{Ŷ ˲{ •‹{Ŷ ˳{ ˰{˲ ˳{ ŵ - Ŷ •‹{Ŷ ˲{ …‘•{Ŷ ˳{ The exact solution of the Euler Equation for these initial conditions is given by ˯{˲ ˳ ˮ{ ŵ . Ŷ …‘• Ŷ {˲ . ˮ{ •‹ Ŷ {˳ . ˮ{ ˰{˲ ˳ ˮ{ ŵ - Ŷ •‹ Ŷ {˲ . ˮ{ …‘• Ŷ {˳ . ˮ{ ˜{˲ ˳ ˮ{ .Ŷ …‘• Ÿ {˲ . ˮ{ . …‘• Ÿ {˳ . ˮ{ Figure 5 Error plot for Poisson solver Figure 4 Residue plot for Poisson solver
  • 6. ICEAE Paper GN-053 Proceedings of ICEAE 2009 The domain is taken again as a unit cube and periodic boundary condition is imposed. The variation of error and residue with time is shown for two different cases of refinement in Fig. 6 and 7. Since the above solution is bounded the variation of upper bound of the solution is also shown for different grid refinement cases in Fig. 8. 14 Results The software has been tested on various geometries and is tested capable of generating a million cells in few seconds on a serial machine. 15 Mesh Polygon clipping is evident in Fig. 9 of grid over a shark. Clearly the boundary surface is very well captured even when the cell size is very large as compared to the geometry triangles' sizes. Fig. 10 shows the mesh over a mirage geometry. 16 Heat Conduction Solver A 3D heat conduction solver is also written. In Fig. 11, there is a temperature gradient below the front part of the shark. This temperature gradient is extending farther from the surface as compared to other parts of surface. This is because of the two fins of the shark, one going inside the plane of paper and one extending outwards. These fins can be seen in the grid result over shark. Thus this temperature gradient clearly signifies the working of the software in 3 dimensional framework. 17 3D Incompressible Euler Solver Fig. 12 shows pressure over the surface of a mirage. 18 Scalability Using the PARAM YUVA cluster a scalability study is also carried out. Result is in Fig. 13. Figure 9 Mesh over man and shark at very low BL Figure 8 Maximum velocity variation Figure 7 Velocity residue plot Figure 6 Velocity error plot
  • 7. Bharat Raj Agrawal, Saurabh Pandey Proceedings of ICEAE 2009 19 Future work The present algorithm is targeted towards complex geometries like full scale aircrafts, automobile, etc. To create grid over such geometries in order to compute values such as drag may require boundary cells of micron scale. This easily translates into a billion of cells which is not possible to be generated and simultaneously stored on a serial machine. This arise a need for the parallelization of the grid generation process. Currently an Euler incompressible solver is in process of development which when fully developed will require to be merged with the grid generation process. The unified algorithm will also utilize the multigrid acceleration which has been found to improve the convergence time by an order of magnitude. The current grid lacks the advantage provided by the conformal grids for computation of boundary layer. For this purpose, adding the capability of structured cells near the boundary to create a hybrid mesh is also being formulated. 20 Conclusion A very fast and robust automatic mesher has been created for which the CFD industry has been eagerly waiting. Since the mesher is automatic and quick, one of the potential uses of this mesher can be directly seen in any aerodynamic sensitive design. 21 Acknowledgement We would like to thank CDAC for providing motivation and platform to test our codes. Without their help this project could not be possible. We would also like to express our gratitude towards Basant K. Gupta, Sandeep S. and Shubhayu C. REFERENCES 1. Aftosmis, M.J., Melton, J.E., and Berger, M.J., Adaptive Cartesian Mesh Generation. Chapter 22 in Handbook of Grid Generation, Thompson, J, Weatherhill, N., and Soni, B. eds. CRC Press 1998. Figure 13 Scalability of solver code up to 1024 cores Figure 12 Pressure variation over a mirage Figure 11 Heat solution over shark after 500 iterations Figure 10 Grid over mirage
  • 8. Proceedings of ICEAE 2009 2. Berger, M.J., and Aftosmis, M.J., Aspects (and aspect ratios) of Cartesian mesh methods. Proceedings of the 16 International Conference on Numerical Methods in Fluid Mechanics, Arcachon, France, July 1998. 3. Aftosmis, M.J., “Solution adaptive Cartesian for aerodynamic flows with complex geometries Karman Institute for Fluid Dynamics, Lecture Series 199702, Rhode-Saint-Genèse, Belgium, Mar. 3-7, 4. Aftosmis, M.J., Berger, M.J., Melton, J.E., efficient Cartesian mesh generation for component geometry,” AIAA Paper 97-0196, Jan. 1997. 5. Berger, M.J., and Melton, J.E., “An Accuracy Cartesian Grid Method for Steady Geometries,” Proc. 8thrIntl. Conf. Hyp. Pro Stonybrook, NY, Jun., 1995. also RIACS Report 95 6. De Zeeuw, D., and Powell, K., “An Cartesian Mesh Solver for the Euler Equations, 91-1542, 1991. 7. Popinet, S. 2003. “Gerris: A tree based adaptive solver for the incompressible Euler equations in complex geometries. Comp. Phys. 190, 572–600. 8. Sutherland, I.E., and Hodgman, G.W., clipping,” Comm of the ACM, 17(1):32-42, 9. Aftosmis, M. J., Berger, M. J., and Adomovic multigrid on Cartesian meshes with complex geometry of the 8th Intl. Conf. on Parallel CFD. (To appear Elsivier Press). Jun. 2000. 10. C. Hamilton. “Compact Hilbert indices. CS-2006-07, Dalhousie University, Faculty of Computer Science, Jul 2006. 11. Chorin, A. J., “Numerical Solution of the Equations.” Mathematics of Computation (Oct., 1968), pp. 745-762. 12. Minion, M. L., “A projection method for grids.” J. Comp. Phys. 127 (1996), 158-178. 13. Almgren, A.S., Bell, J.B., Colella, P., Howell, Welcome, M.L., A conservative adaptive projection method for the variable density incompressible Navier equations, J. Comp. Phys. 142 (1998) 1–46. Bharat Raj Agrawal had the graduate education at Department of Aerospace Engineering, Indian Institute of Technology Bombay and has been working as Engineer at Zeus Numerix Pvt Ltd current interest is in CFD Saurabh Pandey had the graduate education at Department of Aerospace Engineering, Indian Institute of Technology Bombay and has been working as CAE Engineer at Zeus Numerix Pvt Ltd. His current interest is in CFD Berger, M.J., and Aftosmis, M.J., Aspects (and aspect ratios) Proceedings of the 16th International Conference on Numerical Methods in Fluid France, July 1998. Solution adaptive Cartesian grid methods complex geometries”, von , Lecture Series 199702, 7, 1997. Aftosmis, M.J., Berger, M.J., Melton, J.E., “Robust and generation for component based Jan. 1997. An Accuracy Test of a flow in Complex Conf. Hyp. Problems, Uppsala, NY, Jun., 1995. also RIACS Report 95-02. Adaptively Refined the Euler Equations,” AIAA Paper adaptive solver for the equations in complex geometries.” J. “Re-entrant polygon 42, 1974. Adomovicius, G. “Parallel Cartesian meshes with complex geometry” Proc. Trondhiem Norway. Compact Hilbert indices.” Technical Report , Faculty of Computer Numerical Solution of the Navier-Stokes Computation, Vol. 22, No. 104 A projection method for locally refined 178. Almgren, A.S., Bell, J.B., Colella, P., Howell, L.H., adaptive projection method density incompressible Navier–Stokes 46. had the graduate Department of Aerospace Engineering, Indian Institute of Technology been working as CAE Engineer at Zeus Numerix Pvt Ltd. His had the graduate education rtment of Aerospace Engineering, Indian Institute of Technology Bombay and has been working as CAE Engineer at Zeus Numerix Pvt Ltd. His current interest is in ICEAE Paper GN-053