We establish global well-posedness and asymptotic behavior of BV solutions to a system of balance laws modeling traffic flow with nonconcave fundamental diagram. We prove the results by finding a convex entropy-entropy flux pair and verifying the Kawashima condition, the sub-characteristic condition, and the partial dissipative inequality in the framework of Dafermos. This problem is of specific interest since nonconcave fundamental diagrams arise naturally in traffic flow.
Citation: Tong Li, Nitesh Mathur. Global well-posedness and asymptotic behavior of BV solutions to a system of balance laws arising in traffic flow[J]. Networks and Heterogeneous Media, 2023, 18(2): 581-600. doi: 10.3934/nhm.2023025
[1] | Felisia Angela Chiarello, Harold Deivi Contreras, Luis Miguel Villada . Nonlocal reaction traffic flow model with on-off ramps. Networks and Heterogeneous Media, 2022, 17(2): 203-226. doi: 10.3934/nhm.2022003 |
[2] | Maria Laura Delle Monache, Paola Goatin . Stability estimates for scalar conservation laws with moving flux constraints. Networks and Heterogeneous Media, 2017, 12(2): 245-258. doi: 10.3934/nhm.2017010 |
[3] | Felisia Angela Chiarello, Giuseppe Maria Coclite . Nonlocal scalar conservation laws with discontinuous flux. Networks and Heterogeneous Media, 2023, 18(1): 380-398. doi: 10.3934/nhm.2023015 |
[4] | Shyam Sundar Ghoshal, John D. Towers, Ganesh Vaidya . BV regularity of the adapted entropy solutions for conservation laws with infinitely many spatial discontinuities. Networks and Heterogeneous Media, 2024, 19(1): 196-213. doi: 10.3934/nhm.2024009 |
[5] | Raimund Bürger, Kenneth H. Karlsen, John D. Towers . On some difference schemes and entropy conditions for a class of multi-species kinematic flow models with discontinuous flux. Networks and Heterogeneous Media, 2010, 5(3): 461-485. doi: 10.3934/nhm.2010.5.461 |
[6] | Jan Friedrich, Oliver Kolb, Simone Göttlich . A Godunov type scheme for a class of LWR traffic flow models with non-local flux. Networks and Heterogeneous Media, 2018, 13(4): 531-547. doi: 10.3934/nhm.2018024 |
[7] | Paola Goatin, Sheila Scialanga . Well-posedness and finite volume approximations of the LWR traffic flow model with non-local velocity. Networks and Heterogeneous Media, 2016, 11(1): 107-121. doi: 10.3934/nhm.2016.11.107 |
[8] | Helge Holden, Nils Henrik Risebro . Follow-the-Leader models can be viewed as a numerical approximation to the Lighthill-Whitham-Richards model for traffic flow. Networks and Heterogeneous Media, 2018, 13(3): 409-421. doi: 10.3934/nhm.2018018 |
[9] | Mauro Garavello, Roberto Natalini, Benedetto Piccoli, Andrea Terracina . Conservation laws with discontinuous flux. Networks and Heterogeneous Media, 2007, 2(1): 159-179. doi: 10.3934/nhm.2007.2.159 |
[10] | Raimund Bürger, Harold Deivi Contreras, Luis Miguel Villada . A Hilliges-Weidlich-type scheme for a one-dimensional scalar conservation law with nonlocal flux. Networks and Heterogeneous Media, 2023, 18(2): 664-693. doi: 10.3934/nhm.2023029 |
We establish global well-posedness and asymptotic behavior of BV solutions to a system of balance laws modeling traffic flow with nonconcave fundamental diagram. We prove the results by finding a convex entropy-entropy flux pair and verifying the Kawashima condition, the sub-characteristic condition, and the partial dissipative inequality in the framework of Dafermos. This problem is of specific interest since nonconcave fundamental diagrams arise naturally in traffic flow.
This paper studies the following system of balance law arising in traffic flow
ρt+(ρv)x=0,vt+(12v2+g(ρ))x+v−ve(ρ)τ=0 | (1.1) |
with initial data
(ρ(x,0),v(x,0))=(ρ0(x),v0(x)), | (1.2) |
where x∈R,t>0, ρ is the density, v is the velocity, ve(ρ) is the equilibrium velocity, and ρ0,v0 are initial data. We assume that the relaxation time, τ>0,
ρ0(x)≥δ1>0, | (1.3) |
where δ1>0 is a constant. In Eq (1.1), g is the anticipation factor which satisfies
g′(ρ)=ρ(v′e(ρ)θ)2, | (1.4) |
where
0<θ<1. | (1.5) |
We study the global existence and asymptotic behavior of bounded variation (BV) solutions to Eq (1.1) in the framework of Dafermos [6,7]. Our goal is to verify conditions in [6,7] that are needed to find admissible BV solutions to the Cauchy problem (1.1) and (1.2). In particular, we will find a convex entropy-entropy flux pair and verify the Kawashima condition, the sub-characteristic condition, and the partial dissipative inequality. Previously, constructing global BV solutions have been studied in [1,2,3,4,5,6,22,24,25].
A traffic system can exhibit complicated behavior since it is based on interactions between roadways, vehicles, and drivers. Factors that need to be considered in analyzing such a system include nonlinear dynamics and human behavior. Microscopic [8], mesoscopic [27] and macroscopic models [4,10,13,14,15,16,17,18,19,20,21,23,26,28,31] have been utilized to deal with this phenomenon. Constructing global solutions and finding zero relaxation limits of traffic flow models have been a recent focus of study [10,16,17,18,19,21]. This paper is concerned with a specific macroscopic model (1.1).
The following macroscopic models have been important in the study of traffic flow: Lighthill-Whitham-Richards (LWR) model [23,28], Payne-Whitham (PW) model [26,30], viscous models by Kuhne, Li [14,15,22], and Aw-Rascle [4] and Zhang's higher continuum models [31] (ARZ).
When the state is in equilibrium, v=ve(ρ), the model (1.1) reduces to the LWR model
ρt+(ρve(ρ))x=0, | (1.6) |
where x∈R,t>0, and ve(ρ) is a decreasing function of ρ. The fundamental diagram is defined as
R(ρ)=ρve(ρ). | (1.7) |
In the current paper, we study Eqs (1.1) and (1.2) with nonconcave R(ρ). Nonconcave flux arises naturally from traffic flow. We will solve the important problem of studying global BV solutions to nonconcave fundamental diagrams as suggested from traffic experiment data [12,13]. In systems with nonconcave flux functions, the characteristic fields are neither linearly degenerate nor genuinely nonlinear [29].
In our model (1.1), g(ρ) is a pseudo-pressure function accounting for drivers' anticipation of downstream density changes with 0<θ<1 from Eq (1.5), whereas θ=1 in the ARZ model [4,31]. While ARZ model adopted a relative wave propagating speed to the car speed at equilibrium [18], we adopt a larger relative speed. Larger relative speed implies quicker reaction time, which leads to safer and smoother traffic conditions on highways.
The plan of the paper is as follows. We first display preliminaries in Section 2. In Section 3, we introduce symmetrizable system of balance laws and entropy-entropy flux pair. Then we find a convex entropy-entropy flux pair in Section 3 and Section 4. In Section 5, we will verify that the conditions from [6,7] indeed hold true. We then transform our system into an equivalent form in Section 6 as required in [6,7]. Lastly, we prove an a priori estimate in Section 7. We will then present our main result in Section 8 and elaborate on the implications in the conclusion, Section 9.
In this section, we present the preliminaries.
For this study, we take the equilibrium velocity ve(ρ) in Eq (1.1) satisfying
ve(0)=b, | (2.1) |
v′e(0)=−a, | (2.2) |
ve(1)=0, | (2.3) |
where a,b>0 and ve(ρ) is a decreasing function i.e.
v′e(ρ)<0. | (2.4) |
The equilibrium characteristic speed is the characteristic speed of Eq (1.6)
λ∗(ρ)=R′(ρ)=ve(ρ)+ρv′e(ρ). | (2.5) |
We consider general inhomogeneous, strictly hyperbolic system of balance laws as in [7] (Chpt. 16, Eq (16.6.1))
Ut+F(U)x+P(U)=0 | (2.6) |
with initial data
U(x,0)=U0(x), | (2.7) |
where x∈R,t>0, U is a vector in R2,U0 is initial data, and F(U),P(U) are vector fields in R2. Let O be an open subset of R2 containing the origin. In [7], it was assumed that
P(0)=0 | (2.8) |
so that U≡0 is an equilibrium solution. In order to satisfy Eq (2.8), we make the following change of variables in Eq (1.1)
U=(ρ,v−b)T=(ρ,u)T, | (2.9) |
where b is from Eq (2.1). Under Eq (2.9), system (1.1) is reduced to
ρt+(ρ(u+b))x=0,ut+(12(u+b)2+g(ρ))x+u+b−ve(ρ)τ=0, | (2.10) |
where x∈R,t>0. Due to Eq (2.9), Eq (2.10) satisfies Eq (2.8). The initial data of Eq (1.2) is reduced to
(ρ(x,0),u(x,0))=(ρ0(x),u0(x))=(ρ0(x),v0(x)−b) | (2.11) |
after change of variables (2.9).
We identify F and P in Eq (2.10) as
F(U)=(ρ(u+b),12u2+ub+g(ρ))T, | (2.12) |
P(U)=(0,u+b−ve(ρ)τ)T. | (2.13) |
From Eq (2.12), we compute the Jacobian of F as
DF(U)=[u+bρg′(ρ)u+b]. | (2.14) |
The eigenvalues of Eq (2.14) are
λ1,2=u+b±ρv′e(ρ)θ. | (2.15) |
Hence, the system is strictly hyperbolic for ρ>0 since λ1≠λ2. The corresponding right eigenvectors are
r1,2=(±θv′e(ρ),1)T. | (2.16) |
The ith characteristic field is said to be genuinely nonlinear (GNL) if ∇λi⋅ri≠0 for i=1,2. We calculate ∇λi⋅ri as follows
∇λi⋅ri=2v′e(ρ)+ρv″e(ρ)v′e(ρ)=R″(ρ)v′e(ρ),i=1,2, | (2.17) |
where nonconcave R is defined in Eq (1.7). Since R″(ρ) changes signs, the characteristic fields are not genuinely nonlinear.
First we introduce symmetrizable system of balance laws and entropy entropy flux pairs from Dafermos [7] (Chpts. 1 and 3). We will then apply the theory to Eq (2.10) to find an entropy-entropy flux pair.
Now we present the definition of entropy-entropy flux pair from Dafermos [7] (Chpts. 1 and 3).
Let K be an open subset of Rk,k>0. A system of balance laws is given by
div G(U(X),X)=Π(U(X),X), | (3.1) |
where G and Π are given smooth functions defined on O×K taking values in Mn×k and Rn respectively.
Finding entropy-entropy flux pairs is important in the study of balance laws. Our goal is find a smooth entropy-entropy flux pair (η,q)(U) with η convex, normalized by η(0)=0,Dη(0)=0. Admissible solutions U satisfy the entropy inequality [7]
∂tη(U(x,t))+∂xq(U(x,t))+Dη(U(x,t))P(U(x,t))≤0. | (3.2) |
One way to derive entropy-entropy flux pairs is by considering the companion of G. A smooth function Q, defined on O×K taking values in M1×k is called a companion of G if there is a smooth function B, defined on O×K and taking values in Rn such that for all U∈O and X∈K,
DQα(U,X)=B(U,X)TDGα(U,X), α=1,...,k, | (3.3) |
where D=[∂/∂U1,...,∂/∂Un] and Gα(U,X) denotes the α-the column vector of the matrix G(U,X). The significance of companion balance laws is that any classical solution U of Eq (3.1) is automatically a classical solution of the companion balance law
div Q(U(X),X)=H(U(X),X), | (3.4) |
where
H(U,X)=B(U,X)TΠ(U,X)+∇⋅Q(U,X)−B(U,X)T∇⋅G(U,X). | (3.5) |
Eq (3.1) is called symmetric when the n×n matrices DGα(U,X),α=1,...,k, are symmetric, for U∈O and X∈K. In [7], it was proved that a system of balance laws is endowed with nontrivial companion balance laws if and only if it is symmetrizable. When a system of balance laws (3.1) is endowed with a companion balance law (3.4), we can find an entropy-entropy flux pair.
Now we apply the theory from Subsection 3.1 to Eq (2.10) to find an entropy-entropy flux pair. For Eq (2.10), where n=k=2, we have
η(U)=Q1(U),q(U)=Q2(U), | (3.6) |
where η is called the entropy for the system and q is the entropy flux associated with η. Now we solve for an entropy-entropy flux pair (η,q) for Eq (2.10). Evaluating Eq (3.3) at Eq (2.10), we get
DQ1(U)=B(U)TDG1(U),DQ2(U)=B(U)TDG2(U), | (3.7) |
where G1=U,G2=F(U) from Eqs (2.9) and (2.12), and D=[∂/∂U1,∂/∂U2]. Then we have
G1(ρ,u)=[ρu],G2(ρ,u)=[ρ(u+b)12u2+ub+g(ρ)]. | (3.8) |
Then, DG1 and DG2 are as follows
DG1=[1001],DG2=[u+bρg′(ρ)u+b]. | (3.9) |
From Eqs (3.7) and (3.9), we have
DQ1(U)=B(U)TDG1(U)[Q1,ρ,Q1,u]=[B1,B2][1001]. |
Therefore
Q1,ρ=B1,Q1,u=B2. | (3.10) |
Next we plug Eq (3.10) into the second equation of Eq (3.7) to get
[Q2,ρ,Q2,u]=[B1,B2][u+bρg′(ρ)u+b]=[Q1,ρ,Q1,u][u+bρg′(ρ)u+b]=[Q1,ρ(u+b)+Q1,ug′(ρ),Q1,ρρ+Q1,u(u+b)]. |
Hence, we have the two equations for Q2
Q2,ρ=Q1,ρ(u+b)+Q1,ug′(ρ), | (3.11) |
Q2,u=Q1,ρρ+Q1,u(u+b). | (3.12) |
Taking partial derivatives of Eq (3.11) with respect to u and partial derivatives of Eq (3.12) with respect to ρ and subtracting the latter from the former, we get
0=g′(ρ)Q1,uu−ρQ1,ρρ. | (3.13) |
By Eq (1.4), we can rewrite Eq (3.13) as
Q1,uu−s2(ρ)Q1,ρρ=0, | (3.14) |
where we define
s(ρ)=−θv′e(ρ)>0 | (3.15) |
due to Eq (1.5) and Eq (2.4).
Since Eq (3.14) is a second order hyperbolic partial differential equations in two variables, we follow Evans' method in Partial differential equations (Chpt. 7.2.5) [9]. Consider
2∑i,j=1aijuxixj+2∑i=1biuxi+cu=0, | (3.16) |
a11a22−(a12)2<0, | (3.17) |
where the coefficients aij,bi,c (i,j=1,2) with aij=aji and the unknown u are functions for x1 and x2 in some region in D⊂R2. In order to solve Eq (3.14) in Evans' framework, define
˜Q1(u,ρ)=Q1(ρ,u). | (3.18) |
Letting u=~Q1, Eq (2.6) becomes
a11˜Q1,uu+a12˜Q1,uρ+a21˜Q1,ρu+a22˜Q1,ρρ+b1˜Q1,u+b2˜Q1,ρ+c˜Q1=0. | (3.19) |
In Eq (3.19), if we choose
a11=1,a12=a21=0,a22=−s2(ρ),b1=b2=c=0, | (3.20) |
then we arrive at Eq (3.14). In particular, Eq (3.17) is satisfied since
a11a22−(a12)2=−s2(ρ)<0 in D, | (3.21) |
where s(ρ) is defined in Eq (3.15).
Next, we will do the following transformation. Set
{y1=Φ1(u,ρ)y2=Φ2(u,ρ). | (3.22) |
Let
˜Q1(u,ρ)=h(y1,y2) | (3.23) |
for some smooth function h(y1,y2). With the change of variables (3.22), we have
˜Q1,uu=hy1y1(Φ1u)2+hy2y2(Φ2u)2+2hy1y2Φ1uΦ2u+hy1Φ1uu+hy2Φ2uu, | (3.24) |
˜Q1,ρρ=hy1y1(Φ1ρ)2+hy2y2(Φ2ρ)2+2hy1y2Φ1ρΦ2ρ+hy1Φ1ρρ+hy2Φ2ρρ. | (3.25) |
Now we substitute Eqs (3.24) and (3.25) in Eq (3.14) to get
0=hy1y1((Φ1u)2−s2(ρ)(Φ1ρ)2)+hy2y2((Φ2u)2−s2(ρ)(Φ2ρ)2)+hy1(Φ1uu−s2(ρ)Φ1ρρ)+hy2(Φ2uu−s2(ρ)Φ2ρρ)+2hy1y2(Φ1uΦ2u−s2(ρ)Φ1ρΦ2ρ). | (3.26) |
In order to simplify the first two terms in Eq (3.26), we choose (Φ1,Φ2) satisfying
(Φ1u)2−s2(ρ)(Φ1ρ)2=0, | (3.27) |
(Φ2u)2−s2(ρ)(Φ2ρ)2=0. | (3.28) |
This will only be possible if Φ1 and Φ2 solve the following
(wu)2−s2(ρ)(wρ)2=0 in D, | (3.29) |
where w is a solution of Eq (3.29). Note that Eq (3.29) is a product of two linear first-order PDE, namely
wu+s(ρ)wρ=0 in D, | (3.30) |
wu−s(ρ)wρ=0 in D. | (3.31) |
Now we calculate the third and fourth terms in Eq (3.26)
∂2uw−s(ρ)∂2ρw−s′(ρ)s(ρ)∂ρw=(∂u+s(ρ)∂ρ)(∂u−s(ρ)∂ρ)w=0 | (3.32) |
by Eqs (3.30) and (3.31). From Eqs (3.27), (3.28), and (3.32), we calculate
{(Φ1)uu−s2(ρ)(Φ1)ρρ=s(ρ)s′(ρ)Φ1ρ,(Φ2)uu−s2(ρ)(Φ2)ρρ=s(ρ)s′(ρ)Φ2ρ, | (3.33) |
where s(ρ) is defined in Eq (3.15). Then by Eqs (3.27), (3.28), and (3.33), Eq (3.26) becomes
2hy1y2(Φ1uΦ2u−s2(ρ)Φ1ρΦ2ρ)+hy1(s(ρ)s′(ρ)Φ1ρ)+hy2(s(ρ)s′(ρ)Φ2ρ)=0. | (3.34) |
We now solve for Φ1 and Φ2 from Eqs (3.30) and (3.31) respectively. We want a smooth solution Φ1 of Eq (3.30) satisfying ∇Φ1≠0. Φ1 is constant along trajectories x=(x1,x2) of Eq (3.30)
{˙x1=1˙x2=s(ρ). | (3.35) |
Then we have
∇Φ1⊥(1s(ρ)). |
Thus,
∇Φ1||(−s(ρ)1). |
Hence, for some α1(u,ρ), we have
∇Φ1=α1(u,ρ)(−s(ρ)1). | (3.36) |
Indeed, we find an exact solution for Eq (3.36),
Φ1=n1(u+b+ve(ρ)θ), | (3.37) |
where n1>0 and b>0 from Eq (2.1). From Eqs (3.36) and (3.37), we see that α1(u,ρ)=−n1s(ρ).
Similarly, we want a smooth solution Φ2 of Eq (3.31) satisfying ∇Φ2≠0. Φ2 is constant along trajectories x=(x1,x2) of Eq (3.31)
{˙x1=1˙x2=−s(ρ). | (3.38) |
Then we have
∇Φ2⊥(1−s(ρ)),∇Φ2||(s(ρ)1). |
Hence,
∇Φ2=α2(u,ρ)(−s(ρ)−1) | (3.39) |
for some α2(u,ρ). From Eq (3.39), we find an exact solution
Φ2=n2(u+b−ve(ρ)θ), | (3.40) |
where n2>0 and b>0 from Eq (2.1). From Eqs (3.39) and (3.40), we see that α2(u,ρ)=−n2s(ρ).
In this section, we solve for Q1 by integrating factor method. Then we solve for a special case, estimate the solution of Q1 by perturbation analysis, prove convexity, and finalize the entropy-entropy flux pair for Eq (2.10).
In this subsection, we solve for Q1(ρ,u) by integrating factor method.
Plugging Eqs (3.37) and (3.40) into Eq (3.34), then Eq (3.34) reduces to
4n1n2hy1y2−n1s′(ρ)hy1+n2s′(ρ)hy2=0, | (4.1) |
where y1,y2 are defined in Eq (3.22) and h(y1,y2) is defined in Eq (3.23). Since 4n1n2>0, (4.1) becomes
hy1y2−s′(ρ)4n2hy1+s′(ρ)4n1hy2=0. | (4.2) |
Solving Eq (4.2) by integrating factor method and from Eq (3.18), we get
Q1(ρ,u)=~Q1(u,ρ)=h(y1,y2)=μ−12(∫μ2(μ−11f(y1)) dy1+G(y2)), | (4.3) |
where f,G are C2 functions and
μ1=e∫−s′(ρ)4n2 dy2, | (4.4) |
μ2=e∫s′(ρ)4n1 dy1, | (4.5) |
under the assumption
s″(ρ)s(ρ)+(s′(ρ)2)2=0. | (4.6) |
In this subsection, we consider the special case
s′(ρ)=0, | (4.7) |
where s(ρ) is defined in Eq (3.15). Eq (4.7) is equivalent to v″e(ρ)=0. Indeed, we take
ˆve(ρ)=−aρ+b, | (4.8) |
where a,b>0 from Eqs (2.1) and (2.2). In particular, ˆve(ρ) satisfies Eqs (2.1) and (2.2). Then, the fundamental diagram is
ˆR(ρ)=ρˆve(ρ). | (4.9) |
Note that ˆR″(ρ)<0. ˆR(ρ) in Eq (4.9) is an actual fundamental diagram observed in traffic flow, see [11].
Denote ˜ˆQ1 as a solution to Eq (3.14)
˜ˆQ1(u,ρ)=ˆh(ˆΦ1(u,ρ),ˆΦ2(u,ρ)), | (4.10) |
where ˆh is a smooth function. Due to Eqs (4.7) and (4.8), Eq (4.1) reduces to
4n1n2ˆhˆy1ˆy2=0. | (4.11) |
From Eqs (3.37) and (3.40), 4n1n2>0. Then, Eq (4.11) reduces to
ˆhˆy1ˆy2=0. | (4.12) |
Solving Eq (4.12), we get
ˆh(ˆy1,ˆy2)=ˆβ(ˆy1)+ˆG(ˆy2), | (4.13) |
where ˆβ and ˆG are arbitrary C2 functions and ˆy1,ˆy2 are defined in Eq (3.22).
Therefore, from Eqs (3.18), (4.7), (4.8) and (4.10), the solution to Eq (3.14) is
ˆQ1(ρ,u)=˜ˆQ1(u,ρ)=ˆβ(ˆΦ1(u,ρ))+ˆG(ˆΦ2(u,ρ)), | (4.14) |
where
ˆΦ1=n1(u+b+ˆve(ρ)θ), | (4.15) |
ˆΦ2=n2(u+b−ˆve(ρ)θ), | (4.16) |
and b>0 as defined in Eq (2.1).
In this subsection, we estimate Q1(ρ,u), the solution of Eq (3.14), by ˆQ1(ρ,u). From Eq (3.15), we have
s′(ρ)=θv″e(ρ)(v′e(ρ))2. | (4.17) |
Assume that
|s′(ρ)|≤γ<<1, | (4.18) |
where γ>0 is a small constant.
From Eqs (4.17) and (4.18), we derive that
|v″e(ρ)|≤Cγ, | (4.19) |
where C>0 is a universal constant. Eq (4.19) means that the equilibrium velocity ve(ρ) is close to ˆve(ρ) defined in Eq (4.8). Hence, the fundamental diagram R(ρ) in Eq (1.7) is close to ˆR(ρ) defined in Eq (4.9).
By [29] (Chpt. 19, Sec. B), ||(ρ,u)||L∞≤|(ρ,u)(∞,t)|+|(ρ,u)(−∞,t)|+T.V.(ρ,u). Since we are looking for solutions to Eq (2.10) in bounded variation space, we assume a priori that ||(ρ,u)||L∞ is bounded.
We first estimate
Φ1−ˆΦ1=n1θ(ve(ρ)−^ve(ρ)), | (4.20) |
Φ2−ˆΦ2=−n2θ(ve(ρ)−^ve(ρ)), | (4.21) |
where ^ve(ρ) is as in Eq (4.8) and Φ1,Φ2,ˆΦ1,ˆΦ2 are from Eqs (3.37), (3.40), (4.15), and (4.16) respectively.
We now estimate the right hand side of Eqs (4.20) and (4.21). By Taylor expansion and Eq (4.19), there is a ξ∈(0,ρ) such that
|ve(ρ)−ˆve(ρ)|=|ve(0)−ˆve(0)+(v′e(0)−ˆv′e(0))ρ+12(v″e(ξ)−ˆv″e(ξ))ρ2|=12|v″e(ξ)|ρ2≤Cγ, | (4.22) |
where ρ is bounded, ve(0)=b=ˆve(0), v′e(0)=−a=ˆv′e(0) are from Eqs (2.1) and (2.2), and ˆv″e(ρ)=0 due to Eq (4.8).
Plugging Eq (4.22) into Eqs (4.20) and (4.21), we conclude that
|Φ1−ˆΦ1|≤Cγ, | (4.23) |
|Φ2−ˆΦ2|≤Cγ. | (4.24) |
Next, we estimate Q1. From Eq (4.3) and under condition Eq (4.18), we derive
Q1(ρ,u)=β(Φ1(u,ρ))+G(Φ2(u,ρ))+O(γ), | (4.25) |
where γ>0 is small and β(y1)=∫f(y1) dy1, G are C2 functions. In Eq (4.25), we choose β=ˆβ,G=ˆG as defined in Eq (4.14).
Now we estimate Q1−^Q1. From Eqs (4.14), (4.25), and under the condition (4.18), by the mean value theorem, there is a ξ1∈(0,ρ) such that
Q1(ρ,u)−^Q1(ρ,u)=ˆβ′(ξ1)(Φ1−ˆΦ1)+ˆG′(ξ1)(Φ2−ˆΦ2)+O(γ), | (4.26) |
where γ>0 is small. Then from Eqs (4.23) and (4.24) under the condition Eq (4.18), we can estimate Eq (4.26) as
|Q1(ρ,u)−ˆQ1(ρ,u)|≤Cγ, | (4.27) |
where γ>0 is small and C is a universal constant.
In this subsection, we will prove convexity for ˆQ1(ρ,u) and Q1(ρ,u). In order for Q1 to be convex, we need
Q1,ρρ>0, | (4.28) |
D=det[Q1,ρρQ1,ρuQ1,uρQ1,uu]>0. | (4.29) |
From Eq (4.14), we calculate ˆQ1,ρρ as
ˆQ1,ρρ=(ˆv′e(ρ)θ)2(ˆβ″n21+ˆG″n22)+ˆv″e(ρ)θ(ˆβ′n1−ˆG′n2)=(ˆv′e(ρ)θ)2(ˆβ″n21+ˆG″n22), | (4.30) |
where we used ˆv″e(ρ)=0 due to Eq (4.8). Next, we calculate the determinant ˆD defined in Eq (4.29) for ˆQ1
ˆD=4(ˆv′e(ρ)θ)2ˆβ″ˆG″n21n22+ˆv″e(ρ)θ(n31ˆβ′ˆβ″−n32ˆG′ˆG″+n1n22ˆβ′ˆG″−n21n2ˆβ″ˆG′)=4(ˆv′e(ρ)θ)2ˆβ″ˆG″n21n22, | (4.31) |
where we used ˆv″e(ρ)=0 due to Eq (4.8). For ˆQ1 to satisfy Eqs (4.28) and (4.29), we require
ˆβ″,ˆG″≥m>0, | (4.32) |
where m>0 is a constant. Under condition Eq (4.32), we derive from Eq (4.30) that there is m1>0 constant such that
ˆQ1,ρρ≥m1>0. | (4.33) |
Similarly, we derive from Eq (4.31) that
ˆD≥m1>0. | (4.34) |
This proves the convexity conditions Eqs (4.28) and (4.29) for ˆQ1.
Now we prove Eq (4.28) for Q1(ρ,u). Under conditions Eq (4.18), from Eqs (4.27) and (4.33) we have
Q1,ρρ≥ˆQ1,ρρ−Cγ≥m1−Cγ>0 | (4.35) |
by choosing γ>0 small enough.
Similarly, we prove Eq (4.29) for Q1(ρ,u). Under conditions Eq (4.18), from Eqs (4.27) and (4.34) we have
D≥ˆD−Cγ≥m1−Cγ>0 | (4.36) |
by choosing γ>0 small enough. Hence, we proved the convexity for ˆQ1(ρ,u)andQ1(ρ,u) under conditions Eqs (4.18) and (4.32).
In this subsection, we finalize the entropy-entropy flux pair for Eq (2.10).
We choose ˆβ(ˆΦ1)=(ˆΦ1)2,ˆG(ˆΦ2)=(ˆΦ2)2 in ˆQ1 defined in Eq (4.14) which satisfy Eq (4.32) with m=2. Then by Eq (3.6)
ˆη(ρ,u)=ˆQ1(ρ,u)=(ˆΦ1(ρ,u))2+(ˆΦ2(ρ,u))2. | (4.37) |
From Eqs (4.15) and (4.16), we have
ˆη(ρ,u)=(n1(u+ˆve(ρ)θ+b))2+(n2(u−ˆve(ρ)θ+b))2. | (4.38) |
Under Eq (4.8), if we integrate Eq (3.11) with respect to ρ, we get
ˆq(ρ,u)=ˆQ2(ρ,u)=ˆQ1(ρ,u)(u+b)+∫ˆQ1,u ρ(ˆv′e(ρ)θ)2 dρ+j(u), | (4.39) |
where ˆQ1 from Eq (4.38) and j(u) is a smooth function of u. We can solve for j(u) by using Eq (3.12).
Now we derive Q1(ρ,u) and Q2(ρ,u). By Eqs (3.6), (4.18), and (4.27), we have the following entropy-entropy flux pair
Q1(ρ,u)=(n1(u+ve(ρ)θ+b))2+(n2(u−ve(ρ)θ+b))2+O(γ), | (4.40) |
Q2(ρ,u)=Q1(ρ,u)(u+b)+∫Q1,u ρ(v′e(ρ)θ)2 dρ+j(u)+O(γ), | (4.41) |
where γ>0 is small.
In this section, we verify the partial dissipative inequality, the Kawashima condition, and the sub-characteristic condition.
We want to show that the partial dissipative inequality is satisfied for Q1.
According to [6,7], assume that P in Eq (2.13) is dissipative semidefinite relative to η, i.e.,
Dη(U)P(U)≥α|P(U)|2, U∈O | (5.1) |
with α>0.
For Eq (5.1) to be satisfied by ˆQ1, we need to show that
[∂ˆη∂ρ∂ˆη∂u][0u+b−ˆve(ρ)τ]≥α(u+b−ˆve(ρ)τ)2, | (5.2) |
where ˆη is defined in Eq (4.37). From Eq (4.38), we calculate
∂ˆη∂u=2(n21+n22)(u+b)+2(n21−n22)ˆve(ρ)θ. | (5.3) |
Since 0<θ<1 from Eq (1.5), we can choose n22>n21 such that
n21+n22=1θ(n22−n21)>0. | (5.4) |
Define
Γ=n21+n22. | (5.5) |
Then, Eq (5.3) becomes
∂ˆη∂u=2Γ(u+b−ˆve(ρ)). | (5.6) |
Plugging Eq (5.6) into Eq (5.2), we get
2Γ(u+b−ˆve(ρ))2τ≥α(u+b−ˆve(ρ)τ)2. |
Hence in order to satisfy Eq (5.2), we require
0<α≤2Γτ, | (5.7) |
where Γ is as in Eq (5.5). Therefore partial dissipative condition Eq (5.1) is satisfied by ˆQ1.
Now we prove Eq (5.1) for η=Q1(ρ,u) defined in Eq (4.40). We need to show
[∂η∂ρ∂η∂u][0u+b−ve(ρ)τ]≥α(u+b−ve(ρ)τ)2. | (5.8) |
For u+b−ve(ρ)=0, Eq (5.8) is satisfied. Now assume u+b−ve(ρ)≠0.
Under condition Eq (4.18), we derive from Eqs (4.40) and (5.5)
∂η∂u−2Γ(u+b−ve(ρ))≥−Cγ, | (5.9) |
where γ>0 is small and C is a universal constant. To satisfy Eq (5.8), by utilizing Eq (5.9), we require
2Γ(u+b−ve(ρ))2τ−Cγ(u+b−ve(ρ)τ)≥α(u+b−ve(ρ)τ)2. | (5.10) |
By choosing γ>0 small enough in Eq (4.18), we have
Γ(u+b−ve(ρ))2τ−Cγu+b−ve(ρ)τ>0. | (5.11) |
By using Eq (5.11), we require
Γ(u+b−ve(ρ))2τ≥α(u+b−ve(ρ)τ)2 | (5.12) |
for Eq (5.10) to be satisfied. We require
0<α≤Γτ, | (5.13) |
so Eq (5.12) is satisfied. Hence, partial dissipative condition (5.1) for Q1(ρ,u) is satisfied under conditions (4.18) and (5.13).
Therefore, if γ>0 small enough in Eq (4.18), then the estimates for perturbation analysis, convexity, and partial dissipative inequality in Subsections 4.3, 4.4, and 5.1 are established respectively.
Now we verify that the Kawashima condition is satisfied for Eq (2.10).
The Kawashima condition guarantees that the system resulting from linearizing Eq (2.10) does not admit solutions representing undamped traveling waves [6]. From [6,7], the Kawashima condition is given by
DP(0)ri(0)≠0, i=1,...,n. | (5.14) |
For Eq (5.14) to be satisfied, from Eqs (2.2), (2.13) and (2.16), we need
DP(0)ri(0)=[00aτ1τ][±θv′e(0)1]=[0∓θτ+1τ]≠[00] | (5.15) |
since 0<θ<1 from Eq (1.5). Hence, the Kawashima condition (5.14) is satisfied.
To get linear stability, we check if sub-characteristic inequality [7] is satisfied for Eq (2.10) i.e.,
λ1≤λ∗≤λ2, | (5.16) |
where λ∗ is given in Eq (2.5). Recall that u=ve(ρ)−b. Then for v=ve(ρ), from Eqs (2.5) and (2.15) the subcharacteristic condition is satisfied since
ve(ρ)+ρv′e(ρ)θ<ve(ρ)+ρv′e(ρ)<ve(ρ)−ρv′e(ρ)θ. | (5.17) |
Eq (5.17) is true since 0<θ<1 and v′e(ρ)<0 due to Eqs (1.5) and (2.4) respectively.
In this section, we convert Eq (2.10) to the general form according to [7] (Chpt. 16, Eq. (16.6.10)).
∂tV+∂xG(V,W)=0,∂tW+∂xH(V,W)+C(V,W)W=0, | (6.1) |
where x∈R,t>0 and
ηWW(0,0)C(0,0)>0. | (6.2) |
From Eq (2.13), we calculate
DP(0)=1τ[00a1]=[10−a1][0001τ][10a1]=S−1ΓS. |
We are going to do the following change of variables.
ˆU=SU=[10a1][ρu]=[ρaρ+u]. |
Let V and W be
V=ρ, | (6.3) |
W=aρ+u. | (6.4) |
Under Eqs (6.3) and (6.4), system (2.10) is transformed to
Vt+[V(W−aV+b)]x=0, | (6.5) |
Wt+[12(W2−a2V2)+bW+g(V)]x+C(V,W)W=0. | (6.6) |
The systems (6.5) and (6.6) is in the form of Eq (6.1) if we let
G(V,W)=V(W−aV+b)=VW−aV2+bV, | (6.7) |
H(V,W)=12(W2−a2V2)+bW+g(V), | (6.8) |
C(V,W)=1W(W−aV+b−ve(V)τ). | (6.9) |
Since ve(0)=b and τ>0, C(0,W)=1W(W+b−ve(0)τ)=1τ>0. Consequently, C(0,0)=1τ>0. Since we have a convex entropy η (4.40), we have that ηWWC(0,0)>0.
Let Z=(V,W)=(ρ,aρ+u). From Eq (2.11), the corresponding initial conditions for (6.5), (6.6) are
Z0=(V0,W0)=(ρ0,aρ0+u0). | (6.10) |
In this section, we show that ρ≥12δ1>0 for all t under condition (1.3).
We first derive estimates for V and W. From Eq (6.6)
dWdt+1τW=0, t>0, | (7.1) |
where ddt is the derivative along trajectory of Eq (6.6). Therefore,
|W|=|W0|e−1τt, t>0, | (7.2) |
along trajectory of Eq (6.6).
If W=0, (6.5) is a scalar conservation law. Thus, by Theorem 16.1 in [29], we have
V0,inf≤V≤V0,sup, t>0, | (7.3) |
where V0,inf and V0,sup are infimum and supremum of V0 respectively. Since Z0 is of bounded variation, V0,inf,V0,sup, and ||V||L∞ are bounded.
In the general case, W is present in Eq (6.5) and |W| decays exponentially with respect to t, see Eq (7.2). By modifying the proof of Lemma 16.2 in [29] and using Eq (7.2), we obtain
V0,inf−K||W0||L∞≤V≤V0,sup+K||W0||L∞, t>0 | (7.4) |
for some K>0.
By Eqs (1.3) and (6.3), we have V0=ρ0≥δ1>0. Hence,
V0,inf≥δ1>0. | (7.5) |
Choose ||W0||L∞ small such as
||W0||L∞≤12Kδ1. | (7.6) |
Then we have from Eqs (6.3), (7.4), and (7.6) that
ρ=V≥V0,inf−K||W0||L∞≥δ1−K||W0||L∞≥12δ1>0, t>0. | (7.7) |
We verified all assumptions in Dafermos theory [6,7] for Eq (2.10). Now we present the main result of this paper. We show the global existence and asymptotic behavior of BV solutions to the Cauchy problem (6.5), (6.6) and (6.10) for a nonconcave fundamental diagram in traffic flow.
Theorem 8.1 (Admissible BV Solution to the Cauchy Problem). Under the conditions (1.3), (1.5), (4.6), (4.18), (5.13) and (7.6), the system of balance laws (6.5) and (6.6) is endowed with a convex entropy-entropy flux pair (η,q) (4.40) and (4.41), satisfies the partial dissipative inequality (5.1), and satisfies the Kawashima condition (5.15). Consider the Cauchy problem (6.5), (6.6) and (6.10). For δ0,σ0>0, suppose that Z0 decays, as |x|→∞, sufficiently fast to render the integral
∫∞−∞(x2+1)|Z0(x)|2 dx=σ2<σ20, | (8.1) |
with bounded variation
TV(−∞,∞)Z0(⋅)=δ<δ0, | (8.2) |
and
∫∞−∞V0(x) dx=0, | (8.3) |
then there exist positive constants c0,c1,c2,ν>0 so that the Cauchy problem (6.5), (6.6) and (6.10) possesses a unique admissible BV solution Z defined on (−∞,∞)×[0,∞) and
∫∞−∞|Z(x,t)| dx≤c0σ, 0≤t<∞, | (8.4) |
TV(−∞,∞)Z(⋅,t)≤c1σ+c2δe−νt, 0≤t<∞, | (8.5) |
∫∞−∞|Z(x,t)| dx→0, ast→∞, | (8.6) |
TV(−∞,∞)Z(⋅,t)→0, as t→∞. | (8.7) |
The above results are in Z=(V,W) defined in Eqs (6.5) and (6.6). Using Eqs (2.9), (6.3) and (6.4), we can transform our theorems in the original unknowns (ρ,v) satisfying system (1.1) and (1.2).
In the framework of Dafermos, we proved the existence and asymptotic behavior of global BV solutions to the Cauchy problem for a traffic flow model with nonconcave fundamental diagram. The nonconcave fundmental diagram in Eq (1.7) is close to a concave fundamental diagram in Eq (4.9). We derived the partial dissipative inequality, the sub-characteristic condition, the Kawashima condition, and a convex entropy-entropy flux pair to prove our Theorem 8.1.
We adopted the model (1.1) and (1.5) with larger anticipation factors than the ARZ model. Anticipation factor describes the effect of drivers reacting to conditions downstream. Due to higher pressure from the traffic, the driver's anticipation increases, which causes traffic flow to be more regular. Larger anticipation factors lead to safer and smoother traffic conditions on highways.
The authors would like to thank the editors and reviewers for their kind consideration and careful reading.
The authors declare there is no conflict of interest.
[1] |
D. Amadori, A. Corli, Global existence of BV solutions and relaxation limit for a model of multiphase reactive flow, Nonlinear Anal., 72 (2010), 2527–2541. https://10.1016/j.na.2009.10.048 doi: 10.1016/j.na.2009.10.048
![]() |
[2] |
D. Amadori, G. Guerra, Global BV solutions and relaxation limit for a system of conservation laws, Proc. Roy. Soc. Edinburgh Sect. A, 131 (2001), 1–26. https://10.1017/S0308210500000767 doi: 10.1017/S0308210500000767
![]() |
[3] |
F. Ancona, L. Caravenna, A. Marson, On the structure of BV entropy solutions for hyperbolic systems of balance laws with general flux function, PJ. Hyperbolic Differ. Equ., 16 (2019), 333–378. https://10.1142/S0219891619500139 doi: 10.1142/S0219891619500139
![]() |
[4] |
A. Aw, M. Rascle, Resurrection of "second order" models of traffic flow, SIAM J. Appl. Math., 60 (2000), 916–938. https://10.1137/S0036139997332099 doi: 10.1137/S0036139997332099
![]() |
[5] |
G. Q. Chen, D. C. Levermore, T. P. Liu, Hyperbolic conservation laws with stiff relaxation terms and entropy, Comm. Pure Appl. Math., 47 (1994), 787–830. https://10.1002/cpa.3160470602 doi: 10.1002/cpa.3160470602
![]() |
[6] |
C. M. Dafermos, Asymptotic behavior of BV solutions to hyperbolic systems of balance laws with relaxation, J. Hyperbolic Differ. Equ., 12 (2015), 277–292. https://10.1142/S0219891615500083 doi: 10.1142/S0219891615500083
![]() |
[7] | C. M. Dafermos, Hyperbolic Conservation Laws in Continuum Physics, Springer-Verlag, Berlin, 2016. |
[8] | D. C. Gazis, R. Herman, R. W. Rothery, Nonlinear follow-theleader models of traffic flow, Operat. Res., 9 (1961), 545–567. |
[9] | L. C. Evans, Partial differential equations, American Mathematical Society, Providence, 2010. |
[10] |
P. Goatin, N. Laurent-Brouty, The zero relaxation limit for the Aw-Rascle-Zhang traffic flow model, Z. Angew. Math. Phys., 70 (2019). https://10.1007/s00033-018-1071-1 doi: 10.1007/s00033-018-1071-1
![]() |
[11] | B. Greenshields, A study of traffic capacity, Highway Research Board Proceedings, 14 (1933), 448–477. |
[12] | D. Helbing, A. Hennecke, V. Shvetsov, M. Treiber, MASTER: macroscopic traffic simulation based on a gas-kinetic, non-local traffic model, Transportation Res. Part B: Methodol., 35 (2001), 183–211. |
[13] | A. Klar, R. Wegener, Kinetic derivation of macroscopic anticipation models for vehicular traffic, SIAM J. Appl. Math, 60 (2000), 1749–1766. |
[14] | R. D. Kühne, Macroscopic Freeway Model for dense traffic-stop-start waves and incident detection, 9th Int. Symp. on Transp. and Traffic Theory, VNU Science Press, Delft, (1984), 21–42. |
[15] | R. D. Kühne, Freeway control and incident detection using a stochastic continuum theory of traffic flow, Proc. 1st Int. Conf. on Applied Advanced Technology in Transportation, Engineering, San Diego, (1989), 287–292. |
[16] |
C. Lattanzio, P. Marcati, The zero relaxation limit for the hydrodynamic Whitham traffic flow model, J. Differ Equ, 141 (1997), 150–178. https://10.1006/jdeq.1997.3311 doi: 10.1006/jdeq.1997.3311
![]() |
[17] |
Y. Lee, Thresholds for shock formation in traffic flow models with nonlocal-concave-convex flux, J. Differ Equ, 266 (2019), 580–599. https://10.1016/j.jde.2018.07.048 doi: 10.1016/j.jde.2018.07.048
![]() |
[18] |
T. Li, Global solutions and zero relaxation limit for a traffic flow model, SIAM J. Appl. Math., 61 (2000), 1042–1061. https://10.1137/S0036139999356788 doi: 10.1137/S0036139999356788
![]() |
[19] | T. Li, L1 stability of conservation laws for a traffic flow model, Electron. J. Differ Equ, (2001), 14–18. |
[20] |
T. Li, H.M. Zhang, The mathematical theory of an enhanced nonequilibrium traffic flow model, Netw. Spat. Econ., 1 (2001), 167–177. https://10.1023/A:1011585212670 doi: 10.1023/A:1011585212670
![]() |
[21] | T. Li, Global solutions of nonconcave hyperbolic conservation laws with relaxation arising from traffic flow, J. Differ Equ, 190 (2003), 131–149. |
[22] | T. Li, Stability of traveling waves in quasi-linear hyperbolic systems with relaxation and diffusion, SIAM J. Mat. Anal., 40 (2008), 1058–1075. |
[23] |
M. J. Lighthill, G. B. Whitham, On kinematic waves. Ⅱ. A theory of traffic flow on long crowded roads, Proc. Roy. Soc. London Ser. A, 229 (1955), 317–345. https://10.1098/rspa.1955.0089 doi: 10.1098/rspa.1955.0089
![]() |
[24] |
T. Luo, R. Natalini, T. Yang, Global BV solutions to a p-system with relaxation, J. Differ Equ, 162 (2000), 174–198. https://10.1006/jdeq.1999.3697 doi: 10.1006/jdeq.1999.3697
![]() |
[25] | R. Natalini, Convergence to equilibrium for the relaxation approximations of conservation laws, Comm. Pure Appl. Math., 49 (1996), 795–823. |
[26] | H. J. Payne, Models of Freeway Traffic and Control, in Simulation Councils Proc. Ser.: Mathematical Models of Public Systems, Simulation Councils, La Jolla, (1971), 51–60. |
[27] | I. Prigogine, R. Herman, Kinetic Theory of Vehicular Traffic, American Elsevier Publishing Company Inc., New York, 1971. |
[28] |
P. I. Richards, Shock waves on the highway, Operations Res., 4 (1956), 42–51. https://10.1287/opre.4.1.42 doi: 10.1287/opre.4.1.42
![]() |
[29] | J. Smoller, Shock Waves and Reaction-Diffusion Equations, Springer-Verlag, New York, 1994. https://10.1007/978-1-4612-0873-0 |
[30] | G. B. Whitham, Linear and nonlinear waves, John Wiley & Sons, New York, 1974. |
[31] |
H. Zhang, New Perspectives on continuum traffic flow models, Netw. Spat. Econ., 1 (2001), 9–33. https://10.1023/A:1011539112438 doi: 10.1023/A:1011539112438
![]() |
1. | Tong Li, Nitesh Mathur, 2024, Chapter 23, 978-3-031-55259-5, 307, 10.1007/978-3-031-55260-1_23 |