1 Introduction
Partial differential equations on networks arise in various applications including traffic flow, gas or water supply networks, and elastic multi-structures. We refer to GaravelloPiccoli ; LagneseLeugeringSchmidt ; Mugnolo
for mathematical background, further applications, and references. In this paper, we study scalar conservation laws on one dimensional network structures describing, e.g., the transport of a chemical substance in a flow through a network of pipes. A linear advection equation is used to model the transport within the pipes and appropriate coupling conditions are formulated to describe the mixing of flows and the conservation of mass at network junctions. For the semi-discretization in space, we consider a hybrid discontinuous Galerkin method which turns out to be particularly well-suited for dealing with the hyperbolic nature of the problem and the coupling conditions at network junctions. Stability and conservation of the semi-discrete scheme as well as order optimal error estimates are established.
The rest of the paper is structured as follows: In section 2, we introduce the basic notation and then give a complete formulation of the considered problem. A particular choice is made for the coupling conditions which allows us to prove conservation of mass and stability of the overall system. In section 3, we introduce the discretization and prove conservation, stability, and the error estimates. Some numerical tests are presented in section 4 for illustration of our results.
2 Notation and problem formulation
Following the notation of EggerKugler , the topology of the pipe network is described by a finite, directed, and connected graph with vertex set and set of edges . For any vertex , we denote by the set of edges having as a vertex, and we distinguish between inner vertices (junctions) and boundary vertices . For any edge , we define and to indicate the start and the end point of the edge, and we set if . We further identify with the interval of positive length and denote by the space of square integrable functions on the edge , and by
the corresponding space on the network. Here and below, denotes the restriction of a function defined over the network to a single edge . We use
to denote the natural norm and scalar product of and define by
the broken Sobolev spaces which are equipped with the canonical norms
Let us note that and for the functions are continuous along edges but may be discontinuous across junctions .
On every edge of the network, the transport is described by
(1) | |||||
(2) |
Here is the concentration of the substance on pipe , represents the cross sectional area of , and is the given volume flow rate.
Assumption 1
We have with and constant on every edge. Moreover, we require flow conservation at junctions, i.e.,
(C) |
The conditions on characterize an incompressible background flow. Using the above assumption, we can associate a unique flow direction to every edge and define for every vertex the sets and of edges producing flow into or out of the vertex. We also decompose into the sets of inflow and outflow vertices and . The local transport problems (1)–(2) are then complemented by coupling and boundary conditions
(3) |
with auxiliary values defined for by the relations
(4) |
for inflow vertices. On the remaining vertices , we set
(5) |
From Assumption 1, we deduce that , so that is well-defined for all and can be eliminated using (4) and (5). Furthermore, one can see that is a convex combination, i.e., a mixture, of the concentrations in the flows entering the junction . Using condition (C), one can further see that the mass at inner vertices is conserved, more precisely
(6) |
The transport problem on networks is now fully described by (1)–(5). It turns out that the number and type of coupling and boundary conditions is appropriate to guarantee stability and well-posedness of the problem and to ensure conservation of mass across network junctions.
Theorem 2.1
Proof
The energy identity can be derived directly from (1)–(5) and establishes stability of the evolution problem. Existence of a unique solution then follows from the Lumer-Phillips theorem and semigroup theory Engel ; Pazy ; a detailed proof can be found in Philippi . Related results can also be found in Dorn ; EggerKugler ; Kramar ; Mugnolo .
Remark 1
For junctions with more than two inflow pipes, the last term in the energy estimated does in general not vanish and represents dissipation, i.e., loss of information, due to mixing.
3 A hybrid discontinuous Galerkin method
We now formulate a discontinuous Galerkin method for the semi-discretization of problem (1)–(5); see Ern ; Johnson for a general introduction. Hybridization introduces additional unknowns at the grid points of the mesh which play a similar role as the auxiliary mixing values in the coupling conditions (3). The spatial grid is defined by
with local and global mesh size denoted by and . As approximation spaces for the concentration field, we choose
i.e., spaces of piecewise polynomials of degree . Functions in may formally take multiple values at grid points. We introduce grid dependent scalar products
where , and corresponding norms and . The broken Sobolev spaces over the mesh are denoted by
We further introduce the spaces of hybrid variables
with denoting the total number of grid points. Note that grid points associated to the same junction are identified. For the numerical approximation of (1)–(5), we then consider the following semi-discrete scheme.
Problem 1
Find and such that for all and such that for all and
(7) |
holds for all for all and and all , with bilinear form
(8) |
and upwind value in flow direction.
As noted in EggerSchoeberl , the hybrid variable can be eliminated from the system resulting in a standard discontinuous Galerkin discretization with upwind fluxes. At network junctions , the hybrid variable is determined by a discrete version of the coupling condition (5), which can be verified by appropriate testing. Let us start with summarizing some basic properties of the hybrid discontinuous Galerkin scheme.
Lemma 1
The bilinear form is semi-elliptic on the discrete spaces, i.e.,
As a consequence, Problem 1 is uniquely solvable. Moreover, the solution satisfies
for all , as well as the discrete energy identity
Finally, let be a sufficiently regular solution of (1)–(5) and set at grid points in the interior of the edges. Then
for all , and all , i.e., the method is consistent.
Proof
The semi-ellipticity of follows by standard arguments; see e.g. Ern ; EggerKugler . As a consequence of this identity and Assumption 1,
can be eliminated algebraically and the discrete problem can be turned into an linear ordinary differential equation. Existence of a unique solution then follows by the Picard-Lindelöf theorem. The conservation property and the energy identity follow by appropriate testing.
Remark 2
The discretization inherits most of the properties from the continuous problem. The dissipation terms in the energy estimate are partly due to possible jumps across network junctions, which are present also on the continuous level, and partly due to jumps at interior vertices, which are caused by numerical dissipation due to the upwind mechanism in the discontinuous Galerkin method.
We are now in the position to establish order optimal a-priori error estimates.
Theorem 3.1
Proof
As usual, the proof is based on an error splitting
into projection error and discrete error . Similar to Thomee , we use a particular projection defined for element by
Here is the outflow point of the element , i.e., if and otherwise. By standard estimates for this projection, we obtain
Further define for vertices of the network and for interior grid points on edge . We abbreviate , , and denote by the upwind value as in the definition of the method. Note that and by construction. Using consistency of the discrete problem, we get
for all , , and . Testing with and yields
Let us note that the constant only depends on the bound for and the polynomial degree . Integrating the remaining terms in time and applying Gronwall’s lemma then allows to bound the discrete error by the projection error.
Remark 3
Using the semi-ellipticity of the discrete bilinear form, it is possible to obtain similar bounds also for the error at the grid points. A sub-sequent time discretization, e.g., by implicit Runge-Kutta methods, can also be analyzed with standard arguments; see Ern ; Thomee . Since the problem is one-dimensional, the computational overhead of an implicit time integration scheme is negligible.
4 Numerical tests
For our numerical tests, we consider the following network topology.
We set and for all edges, and choose , , , and which satisfies condition (C). We further choose as initial conditions and as inflow boundary conditions, which fulfill the compatibility condition . The solution for this problem can be computed analytically and one can verify that . From the estimates of Theorem 3.1, we therefore expect second order convergence when discretizing with piecewise polynomials of order . For time integration, we utilize an implicit Euler method with sufficiently small step size , and we use
as measure for the error, where
denotes the element-wise linear interpolation.
As expected, the convergence rates observed in our numerical tests coincide with predictions from Theorem 3.1. The solution plot in Figure 1 clearly illustrates the discontinuity of the analytical solution at network junctions.
Acknowledgements.
This work was supported by the German Research Foundation (DFG) via grants TRR 154 C4 and the “Center for Computational Engineering” and at TU Darmstadt.Bibliography
- (1) Di Pietro, D. A., Ern, A.: Mathematical aspects of discontinuous Galerkin methods. Springer Science & Business Media (2011)
- (2) Dorn, B.: Semigroups for flows on infinite networks. M.Sc. thesis, Eberhard Karls Universität Tübingen (2005)
- (3) Egger, H., Kugler, T.: Damped wave systems on networks: Exponential stability and uniform approximations. Numer. Math. 138, 839–867 (2018)
- (4) Egger, H., Schöberl, J.: A hybrid mixed discontinuous Galerkin finite element method for convection-diffusion problems. IMA J. Num. Anal. 30, 1206–1234 (2009)
- (5) Engel, K. J., Nagel, R.: One-Parameter Semigroups for Linear Evolution Equations, 1 edn. Springer-Verlag New York (2000)
- (6) Garavello, M., Piccoli, B.: Traffic flow on networks, AIMS Series on Applied Mathematics, vol. 1. American Institute of Mathematical Sciences (AIMS), Springfield, MO (2006)
- (7) Johnson, C.: Numerical Solution of Partial Differential Equations by the Finite Element Method. Dover Publications (2009)
- (8) Kramar, M., Sikolya, E.: Spectral properties and asymptotic periodicity of flows in networks. Mathematische Zeitschrift 249, 139–162 (2005)
- (9) Lagnese, L. E., Leugering, G., Schmidt, E. J. P. G.: Modeling, Analysis and Control of Dynamic Elastic Multi-Link Structures. Systems & Control: Foundations & Applications. Springer Science+Business Media, New York (1994)
- (10) Mugnolo, D.: Semigroup methods for evolution equations on networks. Springer (2014)
- (11) Pazy, A.: Semigroups of Linear Operators and Applications to Partial Differential Equations, 1 edn. Springer-Verlag New York (1983)
- (12) Philippi, N.: Analysis and numerical approximation of transport equations on networks. M.Sc. thesis, TU Darmstadt (2019)
- (13) Thomée, V.: Galerkin finite element methods for parabolic problems. Springer (1984)
Comments
There are no comments yet.