SPE-180110-MS

An Application of the Isogeometric Analysis Method to Reservoir

Simulation
Eric A. Lynd, John T. Foster, and Quoc P. Nguyen, The University of Texas at Austin

Copyright 2016, Society of Petroleum Engineers

This paper was prepared for presentation at the SPE Europec featured at 78th EAGE Conference and Exhibition held in Vienna, Austria, 30 May – 2 June 2016.

This paper was selected for presentation by an SPE program committee following review of information contained in an abstract submitted by the author(s). Contents
of the paper have not been reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect
any position of the Society of Petroleum Engineers, its officers, or members. Electronic reproduction, distribution, or storage of any part of this paper without the written
consent of the Society of Petroleum Engineers is prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300 words; illustrations may
not be copied. The abstract must contain conspicuous acknowledgment of SPE copyright.

Abstract
The isogeometric analysis method (IGA) has been an emerging technique for exact geometrical discret-
ization and efficient high order approximations in the field of computational engineering. While its
potential has been demonstrated in a variety of disciplines, it has yet to be applied extensively to the field
of reservoir simulation. This work shows IGA’s potential as a useful tool for next generation reservoir
simulation, highlighting its ability to exactly capture complex-shaped geometrical features in the reservoir
and, consequentially, accurately model the pressure fields. First, an IGA code is validated against the
analytic solution for a quarter five-spot pattern as well as a reference solution for a straight line source
problem with pressure enforced at the corners of the domain. Next, the numerical efficiency of IGA is
compared against a finite element method solution of single phase flow in a 2D representative reservoir
containing a centralized ⬙S-shaped⬙ line source. Results suggest a clear advantage of the IGA method over
the classical finite element method, which has lower convergence rates in terms of number of degrees of
freedom.
Introduction
Reservoir fluid transport in many formations is dominated by localized phenomena that require accurate
geometric descriptions and/or high-order approximations of the field equations for simulation codes to be
predictive. As an example, consider either natural or hydraulically stimulated fractures (Berkowitz, 2002;
McCord et al., 1992). The standard approaches for modeling fractures in commercial reservoir simulators
are dual-porosity models, which suffer shortcomings in the presence of fractures that are not well-
connected, distributed non-uniformly, or which have a few large but connected fractures that dominate the
flow (Gerke and Genuchten, 1993; Zimmerman et al., 1993). Additionally, they use ad hoc transfer
functions to account for fluid exchange between the porous matrix and the fractures (Sahimi, 2011).
Discrete fracture models have also been used which model the fractures explicitly, but require conforming
finite element discretizations (Karimi-Fard et al., 2003; Matthai et al., 2005). If piecewise constants (i.e.,
finite volumes) or the typical linear basis functions are used for the approximation of the solution over the
elements, then a highly refined region near the fractures is required for accurate resolution of both steep
pressure gradients and fracture geometry. These highly refined regions lead to impracticable computation
times for models at reservoir scales. Besides fractures, ⬙snake wells⬙ are another example where complex
2 SPE-180110-MS

geometry manifests in reservoir engineering and traditional simulators are challenged to be predictive
(Johan et al., 2004).
There has been a rising popularity of non-uniform rational B-spline (NURBS) enhanced methods
throughout multiple engineering fields (Sevilla et al., 2006; Tan et al., 2015). NURBS have the ability to
exactly represent all conic sections and are the standard geometric interpolant used in computer aided
design (CAD) software. Recently, isogeometric analysis (IGA) has been established as an alternative to
finite-element analysis (FEA) and is novel for its use of NURBS basis functions in both geometry
description and analysis (Hughes et al., 2005). Such treatment remedies the time cost typically associated
with meshing, and allows for exact representation of complex geometries. In addition, IGA possesses a
k-refinement strategy in which order and continuity of basis functions across elements can be increased
while adding fewer degrees of freedom compared to a standard FEA p-refinement. More recently
developed features that have become available to IGA include local refinement with T-splines (Bazilevs
et al., 2010), efficient use of Gauss quadrature points for high continuity basis functions (Adam et al.,
2015), and a Bezier extraction operator that allows closer resemblance of an IGA code to finite element
analysis (Borden et al., 2011).
Applications of IGA in the modeling of flow through fractured and other complex media represent a
possible shift in an industry dominated by more traditional methods (Irzal et al., 2014; Shahrokhabadi et
al., 2014). The present study explores IGA’s ability to discretize complex geometrical features within the
reservoir by simulating pressure distribution around a S-curved shaped line source, which may represent
an idealized fracture, deviated well trajectory, or other reservoir feature. IGA’s convergence with regards
to separate h, p, and k refinement schemes are compared against that of an h-refined, 1st order standard
finite element method (SFEM) formulation which, by its polynomian basis construction, cannot exactly
discretize the curvilinear geometry. Since the SFEM is expected to converge more efficiently than the
current industry standard finite difference methods, this study suggests IGA’s effectiveness over both
finite element and finite difference methods for reservoir simulation. Additionally, the study also
demonstrates the degree to which geometrical discretization error plays a role in the determination of
pressure fields. Broadly, the goal of the paper is to introduce the petroleum engineering community to IGA
and demonstrate its efficacy on reservoir problems where complex geometries are present.
In the following sections, the basic governing equations are summarized. Then, several fundamental
concepts of IGA are reviewed. To verify the IGA code, numerical experiments are carried out using an
analytic five-spot well solution and a simple line source well-described by a SFEM formulation.
Ultimately, IGA is compared against SFEM for discretization of the S-shaped line source, where the true
utility is realized.

Governing Equations
The problems of concern in the later sections of this report involve the calculations of pressure fields
across 2-D, isotropic, horizontal porous media. Furthermore, flow is idealized to be single phase (water),
Newtonian, steady state, and incompressible. Line sources of constant flux are used to simulate the effect
of wells or fractures on the pressure field of a reservoir. Below are summarized the equations involved
in the development of the numerical method studied in this work. The approach is observed to be similar
to that of the standard finite element case.
Darcy’s Law and Mass Balance Darcy’s Law for 2-D flow is given as:
(1)

where k is the permeability, u is viscosity, is the velocity vector, ⵜ is the gradient operator, and P
is the pressure field. Assuming incompressible, steady state flow, the mass balance equation is given as
SPE-180110-MS 3

(2)

where f * is the distributed mass source/sink term and ␳ is the density.

Substituting (1) into (2) gives the PDE governing the pressure distribution in the 2-D physical domain
⍀, which is essentially the Poisson equation for an unknown pressure
(3)

Dividing both sides by p converts the source term from an expression of mass to one of volume
(4)

where f is the distributed volumetric source/sink term. Lastly, Dirichlet and Nuemann boundary
conditions may be expressed as:
(5)

(6)

where and are prescribed values for pressure and velocity, ⌫D is the Dirichlet boundary, and ⌫N
is the Neumann boundary.
Discretization of Pressure Field Equation A Ritz-Galerkin weak form for (4) is derived using an
approach identical to that for the standard FEM case (Becker et al., 1981). The resulting discretization for
the IGA pressure field is given in the below equations, with the concept of nodes and Lagrange
polynomial basis functions being replaced by control points and NURBS basis functions (see the IGA
section for a more thorough discussion on NURBS basis functions and control points):
(7)

(8)

(9)

where ␬ is , ␴ is the constant flux along a line source, ⌫ls is the line source interface, n is the unit
vector normal to the Neumann boundary, dA is the incremental change in area across ⍀, Kij is the (i, j)
entry in the stiffness matrix K, Fi is the ith entry in the load vector F, Ri is the ith NURBS basis function,
Pi is the pressure affiliated with control point i, and ds is the incremental change in arc-length along ⌫ls
or ⌫N. For all problems discussed in future sections, velocity is zero along the domain boundaries. The
expression for load vector terms in (9) then simplifies to:
(10)

Isogeometric Analysis
Below, several fundamental concepts of IGA are addressed. The topics include knot vectors, basis
functions, and control nets, which work together to determine the mesh characteristics in subdomains
referred to as patches. A thorough introductory discussion of IGA can be found in Cottrell et al. (2009),
while useful algorithms for mesh refinement may be found in Piegl and Tiller (1997).
Knot vectors Knot vectors are comprised of a collection of points called knots which control the order,
continuity, and number of basis functions across elements in each patch. Knot vectors have the form:
4 SPE-180110-MS

(11)

where ⌶ is the knot vector on patch i in the parametric ␰ direction and each ␰j is a particular knot. The
n subscript refers to the number of basis functions in the ␰ parametric direction while the p subscript refers
to the order of smoothness of the basis functions. Within each knot vector, each knot is greater than zero
and less than or equal to the following knot. The first knot is taken to be 0, while the last knot typically
has a value of 1 but can take on any positive value. ⬙Open⬙ knot vectors refer to the case where bases are
C0 continuous at patch boundaries, which occurs when the first p ⫹ 1 knots are 0 and the last p ⫹ 1 knots
are the maximum knot value within the knot vector. The multiplicity of interior knots, i.e. knots
inequivalent to the first or last knot, determine the continuity of the basis functions across the patch. For
uniform knot vectors, in which the multiplicity of all interior knots is equal to some positive integer m less
than or equal to p, bases will have p – m continuity within the patch. Each knot span, or jump between
unequal knots, represents an element in the mesh. Each element will contain a number of bases with
support equivalent to p ⫹ 1, with the convention that linear functions have order 1, quadratics have order
2, etc.
B-splines and NURBS Once a knot vector is defined based on the desired element quantity, continuity
between elements, and polynomial order, the Cox-de-Boor recursive algorithm is used to determine the
B-spline functions:
(12)

(13)

where Ni is the ith B-spline basis function evaluated at the parametric coordinate ␰ and p goes from 0
to the polynomial order. It should also be noted that in (13), 0/0 is defined as 0. NURBS basis functions
are then derived from the B-spline basis functions by projecting the B-spline function to one lower
dimensional space by a weighting factor w. As an example, consider the NURBS basis functions for a
parameterized domain in principal coordinates of ␰ and ␩:
(14)

where q is the polynomial order of the bases in the ␩ direction, m is the number of bases in the ␩
direction, Mj is the jth B-spline function in ␩ (determined from (13) using the ␩ knot vector H), and is
the NURBS basis function affiliated with Ni and Mj having polynomial order p and q in the ␰ and ␩
directions, respectively. With the NURBS basis functions, the parameterized geometry is produced by
taking the dot product of the basis functions with a set of control points designated by:
(15)

where is the ith row in the matrix of control points.

Control points Control points are somewhat analogous to nodes in FEA, but rather than laying along
the meshed geometry instead provide a lattice within which the mesh is plotted. Within the code
architecture, control points are organized into a matrix referred to as the control net. The jth column of the
control net references the jth physical coordinate of the ith control point in the ith row. The last column is
reserved as a place holder for the weight wi of the ith control point. The total number of control points in
a given mesh will be equivalent to the total number of basis functions.
SPE-180110-MS 5

Numerical Verification
Five-spot Well Numerical verification against a known analytic solution is necessary for establishing
whether the IGA code is working properly. Here, the quarter five-spot well pattern is selected as a
test-case for functionality since the boundary conditions and domain simplify quite nicely. First, a
five-spot well pattern is defined in 2-D by an infinite set of adjacent squares with injectors at their corners
and a producer at their center. Flow is assumed to be single phase, incompressible, steady-state and
Newtonian and the porous medium to be isotropic. Next, it is noted that the analytic solution for the full
five-spot well pattern is given as (Katiyar et al., 2014):
(16)

where P is the pressure, ␮ is the viscosity, k is permeability, qi is the volumetric flow rate at well i,
and (xi, yi) is the location of well i.Analysis of (16) would reveal that no flow boundaries exist at the edges
of the quarter five-spot in Fig. 1a. This, in turn, causes (10) to simplify to:
(17)

Figure 1—Five-spot diagram and convergence plot. The injector in (a) is located at the origin. The analytic and IGA solution in (b) is
plotted along the dashed red line in (a).

where fi is the point source term at either the injector or producer. Additionally, given that the pressure
becomes zero at the two corners not affiliated with an injector or producer, the Dirichlet boundary
conditions become:
(18)

where L refers to the width of the square quarter five-spot.

In the five-spot verification study, reservoir parameters were chosen such that ␮ ⫽ 10-3[Pa · s], k ⫽
10 [m2], qi⫽ ⫾10-3[m3/s], and that the injector and producer depicted in Fig. 1a are distanced by 400[m]
-13

in the x and y directions. Fig. 1b shows the IGA solution overlays the analytic solution along the radial
distance between the injector and producer under the prescribed boundary conditions and parameters. By
virtue of the successful reproduction of the analytic solution, the IGA code is shown to work in a medium
free of geometrical inclusions. To demonstrate IGA’s ability to handle a wider class of problems, an
additional verification test with a line source included in the domain was carried out.
Simple Line Source A line source representing an idealized horizontal inclusion in a reservoir is now
introduced (Fig. 2a). The IGA mesh was discretized such that the center most patches shared a boundary
with the idealized interface. Parameters are such that k ⫽ 10-13 [m2] and ␮ ⫽ 10-3 [Pa · s]. The domain
is square with area 640,000[m2], and the the centralized line source has a strength of 10-6 [m3/(m2 · s)].
6 SPE-180110-MS

For boundary conditions, pressure is set to zero at the four corners and volumetric flux is set to zero along
the domain edges. This causes (10) to reduce to:
(19)

Figure 2—Vertical line source diagram and reference solution. Bold black lines define patch boundaries in (b), while fainter black lines
outline element boundaries.

Since no analytic solution was available for this particular problem, a reference solution to the pressure
field was obtained in Fig. 2b using a highly refined IGA mesh.
Two convergence studies were carried out to compare the IGA solution against the SFEM solution. The
first study, depicted in Fig. 3a, started converging from the coarsest mesh possible that exactly described
the geometry (Fig. 3b). The coarsest mesh consisted of six patches, each with a single element and 1st
order bases. It is observable in Fig. 3a that the SFEM solution overlaps with the h-refined IGA solution.
This is to be expected, as the two numerical methods share an equivalent basis function structure for the
1st order polynomial case.

Figure 3—Vertical line source convergence plot and coarsest mesh for 1st convergence study. Bold black lines define patch
boundaries in (b).

The second convergence study in Fig. 4a treats the initial discretization similarly to the problem in the
complex line source section. Fig. 4b depicts a mesh in which the center most patches use 2nd order bases
in the parametric direction parallel to the line source. The basis functions for all other patches, as well as
for the parametric direction in the center patches that is perpendicular to the line source, are 1st order. The
convergence plot in Fig. 4a is similar to that in Fig. 3a, except the IGA curves have been shifted slightly
to the right.
SPE-180110-MS 7

Figure 4 —Vertical line source convergence plot and coarsest mesh for 2nd convergence study. Bold black lines define patch
boundaries in (b).

Comparing the plots in Fig. 3a and Fig. 4a, the local application of 2nd order bases has made the IGA
convergence overall less efficient. It can be inferred from Fig. 2b that the approximation is largely
determined by the resolution at the corners, where the pressure gradients are steepest-as indicated by the
dense packing of contour lines-and therefore require high h or p refinement to model. It follows that
adding additional degrees of freedom in patches not containing the corners will lead to negligible
reductions in error, which accounts for the performance discrepancy between the two convergence studies.
Although global polynomial refinement increases the approximation accuracy, the local refinement
depicted here is counterproductive and unnecessary.
The results from the 1st convergence study suggest that the IGA code functions correctly when
including a line source into the discretization, since the IGA approximation expectedly overlaps with the
SFEM approximation (which itself can correctly approximate the pressure fields given the simple
geometries involved). In addition, because the structure of the stiffness matrix and load vector have not
changed, the underlying physics of the problem have been preserved from the five-spot example. Given
the correct matching of the IGA solution to the five-spot analytic solution and correct convergence
behavior when subjected to a simple line source, the versatility of IGA can be further investigated by
applying a line source for which no exact discretization can be made using the traditional finite-element
approach.

Complex Line Source (S-curve)

The geometry from Fig. 2a was modified such that the resulting fracture could not be exactly discretized
by finite elements. Fig. 5a shows the new shape of the fracture, an S-curve, which was selected based on
the known location of control points in IGA for the generation of circular arcs (Cottrell et al., 2009), i.e.
there was no need to actually use CAD geometry in this case even though that is one of the benefits of
IGA. Boundary conditions are the same as in the vertical line source problem. Unlike the vertical line
source problem, however, the S-curve inclusion required the use of four elements and 2nd order bases for
exact representation. A reference solution was obtained similarly as before using a highly refined IGA
mesh, and can be found in Fig.5b.
8 SPE-180110-MS

Figure 5—S-curve line source diagram and reference solution. Bold black lines define patch boundaries in (b), while fainter black lines
outline element boundaries.

A convergence comparison between SFEM and a h,p,and k refined IGA mesh may be found in Fig. 6a.
IGA is seen to outperform SFEM on a per degree of freedom basis for all refinement strategies. The
efficiency of the k-refinement method is also demonstrated, as the convergence slope is steeper than both
the h and p refinement curves. From the coarse SFEM and IGA meshes in Fig. 6b and Fig. 6c,
respectively, the geometric discretization capabilities of IGA are immediately apparent. For only four
additional degrees of freedom, IGA has exactly captured the complex geometry of the line source whereas
the SFEM mesh remains fairly crude. In addition, comparing Fig. 5b and Fig. 6c, exact geometrical
discretization is preserved from the coarsest level. From the discussion in the vertical line source section,
it can also be inferred that the majority of improvement in the convergence rate comes from the geometry
discretization and not the use of higher order bases along the line source.

Figure 6 —Convergence plot and coarsest meshes for S-curve problem. Thin black lines define SFEM element boundaries in (b), while
fainter black lines and bold black lines define IGA element and patch boundaries in (c), respectively.

Even for this base case, where only one line curve is considered and pressure considerations vastly
reduced, significant improvements in computational efficiency were made by implementing IGA over the
1st order SFEM method. The number of degrees of freedom expected for SFEM to resolve features would
only be exacerbated under realistic conditions where, for example, networks of fractures exist on multiple
scales or where multiple wells are drilled to produce from a reservoir. It is therefore worthwhile to further
explore IGA as a means of efficient reservoir simulation.
Conclusions
The method of isogeometric analysis was applied to the modeling of pressure fields around complex
geometries. Even without the use of a Bézier extraction operation that allows for more efficient
implementation of IGA into existing finite element formulations, the discretization and analysis process
of the two methods still contained many similarities. This likeness, at minimum, makes IGA relatively
SPE-180110-MS 9

easy to learn. Additionally, IGA’s performance with respect to the 1st order finite element method means
it could act as a replacement to the finite difference method when complicated geometrical inclusions
factor heavily into the approximation of a reservoir’s various fields. The result of the k-refinement was
particularly meaningful, in that reservoir phenomena that may require higher continuity between elements
to model effectively can be done so and with the added benefit of requiring fewer degrees of freedom than
if h or p refinement schemes are used. Overall, the avenues of research into IGA’s application to reservoir
simulation are vast, and it’s outlook as a successor to finite difference and SFEM promising.

Acknowledgments
The authors would like to thank the CMG Reservoir Simulation Foundation (FCMG) and the University
of Texas at Austin for their support of this project.

References
Adam, C., T. Hughes, S. Bouabdallah, M. Zarroug, and H. Maitournam (2015). Selective and reduced numerical
integrations for nurbs-based isogeometric analysis. Computer Methods in Applied Mechanics and Engineering 284,
732–761.
Bazilevs, Y., V. M. Calo, J. A. Cottrell, J. A. Evans, T. Hughes, S. Lipton, M. Scott, and T. Sederberg (2010). Isogeometric
analysis using t-splines. Computer Methods in Applied Mechanics and Engineering 199(5), 229 –263.
Becker, E. B., G. F. Carey, and J. T. Oden (1981). Finite elements: an introduction. Englewood Cliffs N.J.: Prentice-Hall.
Berkowitz, B. (2002). Characterizing flow and transport in fractured geological media: A review. Advances in water
resources 25(8), 861–884.
Borden, M. J., M. A. Scott, J. A. Evans, and T. J. Hughes (2011). Isogeometric finite element data structures based on
bezier extraction of nurbs. International Journal for Numerical Methods in Engineering 87(1-5), 15–47.
Cottrell, J. A., T. J. Hughes, and Y. Bazilevs (2009). Isogeometric analysis: toward integration of CAD and FEA. John
Wiley & Sons.
Gerke, H. and M. v. Genuchten (1993). A dual-porosity model for simulating the preferential movement of water and
solutes in structured porous media. Water Resources Research 29(2), 305–319.
Hughes, T. J., J. A. Cottrell, and Y. Bazilevs (2005). Isogeometric analysis: Cad, finite elements, nurbs, exact geometry
and mesh refinement. Computer methods in applied mechanics and engineering 194(39), 4135–4195.
Irzal, F., J. Remmers, C. Verhoosel, and R. Borst (2014). An isogeometric analysis bezier interface element for mechanical
and porome-chanical fracture problems. International Journal for Numerical Methods in Engineering 97(8), 608 –628.
Johan, A. H., K. Schrader, et al. (2004). Combination of snake well design & smart completions: Key enablers for
champion west development. In SPE Asia Pacific Oil and Gas Conference and Exhibition. Society of Petroleum
Engineers.
Karimi-Fard, M., L. Durlofsky, K. Aziz, et al. (2003). An efficient discrete fracture model applicable for general purpose
reservoir simulators. In SPE Reservoir Simulation Symposium. Society of Petroleum Engineers.
Katiyar, A., J. T. Foster, H. Ouchi, and M. M. Sharma (2014). A peridynamic formulation of pressure driven convective
fluid transport in porous media. Journal of Computational Physics 261, 209 –229.
Matthai, S., A. Mezentsev, M. Belayneh, et al. (2005). Control-volume finite-element two-phase flow experiments with
fractured rock represented by unstructured 3d hybrid meshes. In SPE Reservoir Simulation Symposium. Society of
Petroleum Engineers.
McCord, J., M. Reiter, and F. Phillips (1992). Heat-flow data suggest large ground-water fluxes through fruitland coals
of the northern san juan basin, colorado-new mexico. Geology 20(5), 419 –422.
Piegl, L. and W. Tiller (1997). The nurbs book. Monographs in Visual Communication.
Sahimi, M. (2011). Flow and transport in porous media and fractured rock: from classical methods to modern approaches.
John Wiley & Sons.
Sevilla, R., S. Femandez-Mendez, and A. Huerta (2006). Nurbs-enhanced finite element method. In ECCOMAS CFD
2006: Proceedings of the European Conference on Computational Fluid Dynamics, Egmond aan Zee, The Nether-
lands, September 5-8, 2006. Delft University of Technology; European Community on Computational Methods in
Applied Sciences (ECCOMAS).
Shahrokhabadi, S., F. Vahedifard, andE. Ghazanfari (2014). Modeling flow regime in shale using isogeometric analysis.
In Shale Energy Engineering 2014* sTechnical Challenges, Environmental Issues, and Public Policy, pp. 239 –245.
ASCE.
10 SPE-180110-MS

Tan, M. H., M. Safdari, A. R. Najafi, and P. H. Geubelle (2015). A nurbs-based interface-enriched generalized finite
element scheme for the thermal analysis and design of microvascular composites. Computer Methods in Applied
Mechanics and Engineering 283,1382–1400.
Zimmerman, R. W., G. Chen, T. Hadgu, and G. S. Bodvarsson (1993). A numerical dual-porosity model with semiana-
lytical treatment of fracture/matrix flow. Water resources research 29(7), 2127–2137.