An Exact and Grid-free Numerical Scheme for the Hybrid Two Phase Traffic Flow Model Based on the Lighthill-Whitham-Richards Model with Bounded Acceleration Thesis by Shanwen Qiu In Partial Fulfillment of the Requirements For the Degree of Masters of Science King Abdullah University of Science and Technology, Thuwal, Kingdom of Saudi Arabia July, 2012
67
Embed
An Exact and Grid-free Numerical Scheme ... - KAUST …repository.kaust.edu.sa/kaust/bitstream/10754/245871/1/Shanwen Qiu... · Lastly, I would like to thank the people at KAUST,
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
An Exact and Grid-free Numerical Scheme for the
Hybrid Two Phase Traffic Flow Model Based on
the Lighthill-Whitham-Richards Model with
Bounded Acceleration
Thesis by
Shanwen Qiu
In Partial Fulfillment of the Requirements
For the Degree of
Masters of Science
King Abdullah University of Science and Technology, Thuwal,
Kingdom of Saudi Arabia
July, 2012
2
The thesis of Shanwen Qiu is approved by the examination committee
Proof — The proof is similar to the initial condition case. �
Note that the structure of this modified component differs from the unmodified
solution component (3.13): it now contains a transition zone in which vehicles have
an uniform acceleration.
We illustrate kc
(l)modified
as well as the isolines of Nc
(l)modified
corresponding to an internal
condition in Figure 6.5.
43
Figure 6.5: Density map kc
(l)modified
corresponding to an internal condition, for a
triangular fundamental diagram.
44
Chapter 7
Implementation
7.1 Algorithm structure
Given the derivations of the modified component solutions associated with the ini-
tial, boundary and internal conditions, one can build a numerical scheme for solving
the hybrid two phase LWR-bounded acceleration model semi-analytically with a low
computational cost, as in [21].
The proposed numerical scheme is based on the inf-morphism property (3.15). It
is based on the minimization of analytic formulas, and is thus guaranteed to be exact.
45
Algorithm 1 Pseudo-code implementation for the Lax-Hopf based computation ofthe Moskowitz function and the associated density at a single point (x, t) prescribedby the user.
Input: x ∈ [x0, xn], t ∈ [0, tm], {input space domain, time domain}
N← +∞ {initialization of the Moskowitz function to infinity}{iteration on initial conditions}for j = jmin to jmax do
compute Nc
(i)ini
(x, t) using (6.2) or (6.4) {component induced by the initial
condition c(i)ini}
{if the current component contributes to the solution}if Nc
(i)ini
(x, t) < N then
N← Nc
(i)ini
(x, t) {update Moskowitz function}k ← k
c(i)ini
(x, t), computed using (6.3) or (6.5) {compute density}end if
end for{iteration on upstream boundary conditions}for j = 0 to jup do
compute Nc
(j)up
(x, t) using (6.6) {component induced by the upstream boundary
condition c(j)up}
{if the current component contributes to the solution}if Nc
(j)up
(x, t) < N then
N← Nc
(j)up
(x, t) {update Moskowitz function}k ← k
c(j)up
(x, t), computed using (6.7) {compute density}end if
end for{iteration on downstream boundary conditions}for j = 0 to jdown do
compute Nc
(j)down
(x, t) using (6.8 {component induced by the downstream
boundary condition c(j)down}{if the current component contributes to the solution}if N
c(j)down
(x, t) < N then
N← Nc
(j)down
(x, t) {update Moskowitz function}k ← k
c(j)down
(x, t), computed using (6.9) {compute density}end if
end for{iteration on internal conditions}for l = 0 to lintern do
compute Nc
(l)intern
(x, t) using (6.11) {component induced by the internal
condition c(l)intern}
{if the current component contributes to the solution}if Nc
(l)intern
(x, t) < N then
N← Nc
(l)intern
(x, t) {update Moskowitz function}k ← k
c(l)intern
(x, t), computed using (6.13) {compute density}end if
end forOutput: N, k
46
7.2 Numerical examples
We implemented 1 as a Matlab Toolbox, freely downloadable from XXX. All nu-
merical computations were performed using an Intel i3 CPU running at 3.0 GHz,
operated by Windows 7 (32 bit), with 4 GB of RAM.
We consider a triangular fundamental diagram defined by (6.1), where the pa-
rameters are defined as vf = 30 m/s, w = −5 m/s, κ = 0.1 veh/m and kc =
−w κvf−w = 0.014 veh/m. The acceleration of the vehicles during the bounded accel-
eration phases is set to a = 2m/s2, consistent with the capabilities of modern vehicles.
The computational domain consists in a 1000 m section, for a total duration of
50 s.
We first compute the density and trajectories (isolines of N) corresponding to a
given set of initial and boundary conditions, defined as follows:
initial condition:
xini = [0, 250, 500, 750, 1000];
kini = [10E − 3, 40E − 3, 5E − 3, 50E − 3];
upstream boundary condition
tup = [0, 20, 30, 50];
qup = [1, 0.3, 0.1];
Though the algorithm is gridless, we compute the exact solutions on a rectangular
grid (for visualization purposes) of 500, 000 points (1000×500). By construction, the
numerical solutions are exact up to machine accuracy, i.e. the numerical errors are
on the order of machine zeros.
The solution corresponding to the set of initial and boundary conditions outlined
above was computed in 1.14s, and is illustrated in Figure 7.1.
47
Figure 7.1: Density map and vehicle trajectories corresponding to a set of initialand upstream boundary conditions.
We now consider the same initial and boundary conditions, while also assuming
that a fixed (Vintern = 0) obstruction (for instance a red traffic light, or a car accident)
prevents traffic propagation (qintern = 0) at location x = 400m, between times 20 s
and 30 s. The solution was computed in 1.44 s, and is illustrated in Figure 7.2.
While internal conditions can model fixed obstructions, they can also model more
complex traffic flow scenarios, such as moving bottlenecks. To illustrate this, we
simulate the solution to the same initial and boundary conditions, but we now assume
that a slow bus moving at Vintern = 5 m/s is restricting the road capacity along its
path, allowing a maximal passing flow of qintern = 0.025 veh/s. The solution was
computed in 1.38 s, and is illustrated in Figure 7.3.
While we only show numerical computations involving one internal condition for
simplicity, an arbitrary number of fixed and moving bottlenecks can be integrated in
our proposed algorithm. As illustrated in the three above examples, incorporating
fixed or moving bottlenecks does not dramatically increase the computational time
much.
48
Figure 7.2: Density map and vehicle trajectories corresponding to a set of initial andupstream boundary conditions, with a fix bottleneck preventing traffic propagation.The fixed bottleneck is highlighted by a red dash.
Figure 7.3: Density map and vehicle trajectories corresponding to a set of initial andupstream boundary conditions, with a moving bottleneck. The moving bottleneck,represents a bus restricting the road capacity and allowing a passing flow of qintern =0.025 veh/s. We highlight the trajectory of the bus by a red dash.
49
7.3 Benefits of a grid-free and exact method
The fact that the proposed numerical scheme is both exact and gridless is very im-
portant for solving practical problems involving the kinematic effects of vehicles on
traffic flow propagation. For instance, dealing with the effects of traffic flow (pollu-
tion, energy consumption, noise...) often requires the coupling between traffic flow
propagation models and vehicle property models (for instance the energy consumption
of the vehicle as a function of its target path). Optimizing these effects (for instance
minimizing pollution of a bus driving through traffic) thus requires the computation
of an optimal single vehicle trajectory, which is itself a function of the surrounding
traffic. Most algorithms compute these trajectories iteratively, and thus any error
of the numerical traffic solver is propagated through the iterations, leading to poor
overall accuracy. Since the algorithm described in this article is exact, it does not
add any uncertainty to the results.
The fact that the algorithm is grid-free also allows for a faster search over all
possible vehicle trajectories for single vehicle trajectory optimization problems.
50
Chapter 8
Conclusion
This article presents a new semi analytical expression for the solutions to the Lighthill-
Whitham-Richards traffic flow model with bounded vehicle acceleration. Based on
this semi-analytical expression, we compute the analytical solution component blocks
associated with the triangular fundamental diagram. These analytical solution blocks
allow us to construct the solution to the modified traffic flow model as a finite min-
imization of functions. The resulting numerical scheme is both grid-free and exact,
which are very favorable characteristics when dealing with optimization problems in-
volving the coupling between the modified LWR model and vehicle models. We then
use the algorithm to compute the solutions to various problems involving fixed and
moving bottlenecks of increased complexity.
Future work will deal with the extension of this algorithm to networks, as bounded
accelerations frequently occur in junctions, whenever congestion occurs because of
capacity restriction. Another important avenue is the development of an hybrid
model based on the Lighthill-Whitham-Richards traffic flow model with both bounded
acceleration and deceleration, which is to date an open problem.
51
REFERENCES
[1] M. J. Lighthill and G. B. Whitham, “On kinematic waves. II. A theory of
traffic flow on long crowded roads,” Proceedings of the Royal Society of London,
vol. 229, no. 1178, pp. 317–345, 1956.
[2] P. I. Richards, “Shock waves on the highway,” Operations Research, vol. 4,
no. 1, pp. 42–51, 1956.
[3] G. F. Newell, “A simplified theory of kinematic waves in highway traffic, Part
(I), (II) and (III),” Transporation Research B, vol. 27B, no. 4, pp. 281–313, 1993.
[4] N. Geroliminis and C. F. Daganzo, “Existence of urban-scale macroscopic
fundamental diagrams: Some experimental findings,” Transportation Research
Part B: Methodological, vol. 42, no. 9, pp. 759–770, 2008.
[5] R. Ansorge, “What does the entropy condition mean in traffic flow theory?”
Transportation Research, vol. 24B, no. 2, pp. 133–143, 1990.
[6] M. Bardi and I. Capuzzo-Dolcetta, Viscosity Solutions of Hamilton-Jacobi-
Bellman Equations. Boston, MA: Birkhauser, 1997.
[7] L. Leclercq, “Bounded acceleration close to fixed and moving bottlenecks,”
Transportation research. Part B: methodological, vol. 41, no. 3, pp. 309–319,
2007.
[8] J. A. Laval and C. F. Daganzo, “Lane-changing in traffic streams,” Trans-
portation Research Part B: Methodological, vol. 40, no. 3, pp. 251–264, 2006.
52
[9] J. P. Lebacque, “Two-phase bounded-acceleration traffic flow model: analyti-
cal solutions and applications,” Transportation Research Record: Journal of the
Transportation Research Board, vol. 1852, no. -1, pp. 220–230, 2003.
[10] L. Leclercq, J. A. Laval, and N. Chiabaut, “Capacity drops at merges: An
endogenous model,” Transportation Research Part B: Methodological, 2011.
[11] F. Giorgi, L. Leclercq, and J. B. Lesort, “A traffic flow model for urban
traffic analysis: extensions of the lwr model for urban and environmental appli-
cations,” in Proceedings of the 15th International Symposium on Transportation
and Traffic Theory, 2002, pp. 393–416.
[12] A. Aw and M. Rascle, “Resurrection of “second order” models of traffic flow,”
SIAM Journal of Applied Mathematics, vol. 60, no. 3, pp. 916–938, 2000.
[13] S. Blandin, D. Work, P. Goatin, B. Piccoli, A. Bayen et al., “A general
phase transition model for vehicular traffic,” SIAM Journal on Applied Mathe-
matics, vol. 71, no. 1, pp. 107–127, 2011.
[14] J. P. Lebacque, “A two phase extension of the LWR model based on the bound-
edness of traffic acceleration,” in Transportation and Traffic Theory in the 21st
Century. Proceedings of the 15th International Symposium on Transportation
and Traffic Theory, 2002.
[15] S. K. Godunov, “A difference method for numerical calculation of discontinuous
solutions of the equations of hydrodynamics,” Math. Sbornik, vol. 47, pp. 271–
306, 1959.
[16] C. Daganzo, “The cell transmission model: a dynamic representation of high-
way traffic consistent with the hydrodynamic theory,” Transportation Research,
vol. 28B, no. 4, pp. 269–287, 1994.
[17] ——, “The cell transmission model, part II: network traffic,” Transportation
Research, vol. 29B, no. 2, pp. 79–93, 1995.
53
[18] V. Henn, “A wave-based resolution scheme for the hydrodynamic LWR traffic
flow model,” in Proceedings of the Workshop on Traffic and Granular Flow 03,
B. S. W. Hoogendoorn, Luding, Ed., Amsterdam, 2003, pp. 105–124.
[19] C. G. Claudel and A. M. Bayen, “Lax-Hopf based incorporation of in-
ternal boundary conditions into Hamilton-Jacobi equation. Part I: theory,”
IEEE Transactions on Automatic Control, vol. 55, no. 5, pp. 1142–1157, 2010,
doi:10.1109/TAC.2010.2041976.
[20] ——, “Lax-Hopf based incorporation of internal boundary conditions into
Hamilton-Jacobi equation. Part II: Computational methods,” IEEE Trans-
actions on Automatic Control, vol. 55, no. 5, pp. 1158–1174, 2010,
doi:10.1109/TAC.2010.2045439.
[21] P. E. Mazare, A. Dehwah, C. G. Claudel, and A. M. Bayen, “Analyti-
cal and grid-free solutions to the lighthill-whitham-richards traffic flow model,”
Transportation Research Part B: Analytical., vol. 45, no. 10, pp. 1727–1748, 2011.
[22] M. Garavello and B. Piccolli, Traffic flow on networks. Springfield, MO:
American Institute of Mathematical Sciences, 2006.
[23] B. D. Greenshields, “A study of traffic capacity,” Proc. Highway Res. Board,
vol. 14, pp. 448–477, 1935.
[24] C. F. Daganzo, “On the variational theory of traffic flow: well-posedness, du-
ality and applications,” Networks and heterogeneous media, vol. 1, pp. 601–619,
2006.
[25] K. Moskowitz, “Discussion of ‘freeway level of service as influenced by volume
and capacity characteristics’ by D.R. Drew and C. J. Keese,” Highway Research
Record, vol. 99, pp. 43–44, 1965.
[26] C. F. Daganzo, “A variational formulation of kinematic waves: basic theory
and complex boundary conditions,” Transporation Research B, vol. 39B, no. 2,
pp. 187–196, 2005.
54
[27] Y. Makigami, G. F. Newell, and R. Rothery, “Three-dimensional represen-
tation of traffic flow,” Transportation Science, vol. 5, no. 3, pp. 303–313, Aug
1971.
[28] L. C. Evans, Partial Differential Equations. Providence, RI: American Math-
ematical Society, 1998.
[29] S. Chanut. L. Leclercq and J. B. Lesort, “Moving Bottlenecks in Lighthill-
Whitham-Richards Model: A Unified Theory,” Transportation Research Record,
vol. 1883, pp. 3–13, 2004.
[30] J.-P. Aubin, A. M. Bayen, and P. Saint-Pierre, “Dirichlet problems for some
Hamilton-Jacobi equations with inequality constraints,” SIAM Journal on Con-
trol and Optimization, vol. 47, no. 5, pp. 2348–2380, 2008.
[31] J.-P. Aubin, Viability Theory, ser. Systems and Control: Foundations and Ap-
plications. Boston, MA: Birkhauser, 1991.
[32] I. S. Strub and A. M. Bayen, “Weak formulation of boundary conditions for
scalar conservation laws,” International Journal of Robust and Nonlinear Con-
trol, vol. 16, no. 16, pp. 733–748, 2006.
55
APPENDICES
A Derivation of the modified
initial condition component for an
uncongested initial condition
If 0 ≤ ki ≤ kc, the original solution component associated with the affine initial
condition (3.2) is expressed by:
Nc
(i)ini
(x, t) =
(i) ki(tvf − x) + bi : xi + tvf ≤ x ≤ xi+1 + tvf
(ii) kc(tvf − x) + bi + xi(kc − ki) : xi + tw ≤ x ≤ xi + tvf
(A.1)
Applying the first condition of (5.1) to equation (A.1)(i) yields: v(t, x) = vf , t = t
and x ≥ x. Thus,
Nc(i)(t, x) = ki(tvf − x) + bi ≥ ki(tvf − x) + bi
with equality when x = x. When x = x, the condition xi + tvf ≤ x ≤ xi+1 + tvf
56
yields xi + tvf ≤ x ≤ xi+1 + tvf , which implies:
inf Nc(i)(t, x) = ki(tvf − x) + bi for xi + tvf ≤ x ≤ xi+1 + tvf (A.2)
For x ≥ xi+1 + tvf , we plug x = xi+1 + tvf into equation (A.1)(i) to compute
inf Nc(i)(t, x):
inf Nc(i)(t, x) = −kixi+1 + bi for x ≥ xi+1 + tvf (A.3)
Applying the second condition of (5.1) to equation (A.1)(i) yields the following ex-
pression:
inf Nc(i)(t, x) =
(i) ki(tvf − x) + bi : xi + tvf ≤ x ≤ xi+1 + tvf
(ii) −xi+1ki + bi : x ≥ xi+1 + tvf
(A.4)
Similarly, applying (5.1) to equation (A.1)(ii) yields the following candidate values
for inf Nc(i)(t, x). Applying the first condition of (5.1) yields:
inf Nc(i)(t, x) =
(i) kc(tvf − x) + bi + xi(kc − ki) : xi + tw ≤ x ≤ xi + tvf
(ii) −xiki + bi : x ≥ xi + tvf
(A.5)
Applying the second condition of (5.1) yields:
inf Nc(i)(t, x) =
(i) kc(tvf − x) + bi + xi(kc − ki) : xi + t(w − vf ) + tvf ≤ x ≤ xi + tvf
(ii) −xiki + bi : x ≥ xi + tvf
(A.6)
Since t ≥ t and w ≤ 0, we have xi + t(w − vf ) + tvf = xi + tw + (t− t)vf ≥ xi + tw.
Therefore, the domain of equation (A.6)(i) can be extended to xi+tw ≤ x ≤ xi+1+tvf
57
and (A.6) becomes:
inf Nc(i)(t, x) =
(i) kc(tvf − x) + bi + xi(kc − ki) : xi + tw ≤ x ≤ xi + tvf
(ii) −xiki + bi : x ≥ xi + tvf
(A.7)
Using the definition of (5.1), we the minimum value among (A.2),(A.3),(A.4),(A.5)
and (??) in their corresponding domains, which yeilds the following expression for
Nc
(i)modified
when 0 ≤ ki ≤ kc:
Nc
(i)modified
(t, x) =
(i) kc(tvf − x) + bi + xi(kc − ki) : xi + tw ≤ x ≤ xi + tvf
(ii) ki(tvf − x) + bi : xi + tvf ≤ x ≤ xi+1 + tvf
(iii) −xi+1ki + bi : x ≥ xi+1 + tvf
(A.8)
kc
(i)modified
(x, t) = −∂N
c(i)modified
∂x(x, t) =
(i) kc : xi + tw ≤ x ≤ xi + tvf
(ii) ki : xi + tvf ≤ x ≤ xi+1 + tvf
(iii) 0 : x ≥ xi+1 + tvf
(A.9)
58
B Derivation of the modified
initial condition component for an
congested initial condition
If kc < ki ≤ κ, the initial condition imposes a congested state:
Nc
(i)ini
(x, t) =
(i) ki(tw − x)− κtw + bi : xi + tw ≤ x ≤ xi+1 + tw
(ii) kc(tw − x)− κtw + xi+1(kc − ki) + bi : xi+1 + tw ≤ x ≤ xi+1 + tvf
(B.1)
Applying the first condition of (5.1) to equation (B.1)(i) yields v(t, x) = v = w(1− κki
),
x ≥ x+ (t− t)v + a(t−t)2
2and 0 ≤ t− t ≤ τ , where τ =
vf−v(t,x)
a. Thus,
Nc
(i)ini
(x, t) = ki(tw − x)− κtw + bi ≥ ki(tw − x)− κtw + bi +kia(t− t)2
2
with equality when x = x+ (t− t)v + a(t−t)2
2. Thus:
inf Nc
(i)ini
(x, t) = inf(ki(tw − x)− κtw + bi +kia(t− t)2
2), for (t, x) satisfying:
59
(i) 0 ≤ t− t ≤ τ
(ii) x = x+ (t− t)v + a(t−t)2
2
(iii) x+ tw ≤ x ≤ xi+1 + tw
(iv) t ≥ 0
(B.2)
To compute the minimum value of Nc
(i)ini
(x, t), we first have to determine the range of
t− t under the constraint (B.2). Let T = t− t. By plugging (B.2)(ii) into (B.2)(iii),
we can rewrite (B.2) as:
(i) 0 ≤ T ≤ τ
(ii) −aT 2
2+ (w − v)T − (xi + tw − x) ≥ 0
(iii) −aT 2
2+ (w − v)T − (xi+1 + tw − x) ≤ 0
(iv) T ≤ t
(B.3)
Let us define the following auxiliary variables:
∆1 = (w − v)2 − 2a(xi + tw − x)
T1 =(w − v)−
√∆1
a
T2 =(w − v) +
√∆1
a
∆2 = (w − v)2 − 2a(xi+1 + tw − x)
T3 =(w − v)−
√∆2
a
T4 =(w − v) +
√∆2
a
where:
T = T1 and T = T2 are solutions to − aT 2
2+ (w− v)T − (xi + tw− x) = 0 (T1 ≤ T2)
60
T = T3 and T = T4 are solutions of − aT2
2+(w−v)T −(xi+1 + tw−x) = 0 (T3 ≤ T4)
We have the following cases:
1. If ∆1<0, then (B.3)(ii) has no real solution.
2. If ∆1 ≥ 0 and ∆2 ≤ 0, we have xi + tw− (w−v)2
2a≤ x ≤ xi+1 + tw− (w−v)2
2a. The
solution of (B.3)(ii) is T1 ≤ T ≤ T2, while the solution of (B.3)(iii) is T ∈ R.
Hence, (B.3) can be rewritten as:
(i) 0 ≤ T ≤ τ
(ii) T1 ≤ T ≤ T2
(iii) T ≤ t
From its definition, we have that T1 ≤ 0. Thus, Nc
(i)ini
(x, t) has its minimum
value at T = t− t = 0 if and only if T2 ≥ 0, that is, if x ≥ xi + tw. Therefore:
inf Nc
(i)ini
(x, t) = ki(tw − x)− κtw + bi
for: xi + tw ≤ x ≤ xi+1 + tw − (w − v)2
2a
(B.4)
3. If ∆1 ≥ 0 and ∆2 ≥ 0, (B.3) can be simplified to:
(i) 0 ≤ T ≤ τ
(ii) T1 ≤ T ≤ T2
(iii) T ≤ T3 or T ≥ T4
(iv) T ≤ t
(B.5)
From their definitions, we have that T1 ≤ 0, T2 ≤ 0 and T4 ≤ T2. If T2<0, the
(B.5) has no solution. So we need T2 ≥ 0, which implies x ≥ xi + tw. Using
61
this last inequality, (B.5) can be rewritten as:
(i) 0 ≤ T ≤ τ
(ii) T4 ≤ T ≤ T2
(iii) T ≤ t
(B.6)
There are two cases for which (B.6) has a solution in T :
� T4 ≤ 0 and T2 ≥ 0. In this case, xi + tw ≤ x ≤ xi+1 + tw and Nc
(i)ini
(x, t)
has its minimum value at T = t− t = 0. Therefore:
inf Nc
(i)ini
(x, t) = ki(tw − x)− κtw + bi for: xi + tw ≤ x ≤ xi+1 + tw
(B.7)
We can see that the case (B.4) is included in (B.7).
� 0 ≤ T4 ≤ τ and T4 ≤ t. In this case, xi+1 + tw ≤ x ≤ xi+1 + tw + τ(v +
12aτ − w) when t ≥ τ and xi+1 + tw ≤ x ≤ xi+1 + tw + (at−w+v)2−(w−v)2
2a
when t ≤ τ . Nc
(i)ini
(x, t) has its minimum value for T = t− t = T4:
inf Nc
(i)ini
(x, t) = ki(tw − x)− κtw + bi + kia2T 2
4
for: xi+1 + tw ≤ x ≤ xi+1 + tw + τ(v + 12aτ − w) when t ≥ τ or
xi+1 + tw ≤ x ≤ xi+1 + tw + (at−v+v)2−(w−v)2
2awhen t ≤ τ
(B.8)
For x ≥ xi+1 + tw+ τ(v+ 12aτ −w) when t ≥ τ , we should plug x = xi+1 + tw+ τ(v+
12aτ − w) into (B.8) to compute inf N
c(i)ini
(x, t):
inf Nc
(i)ini
(x, t) = κ(τ − t)w−kixi+1 + bi for: x ≥ xi+1 + tw+ τ(v+1
2aτ −w) and t ≥ τ
(B.9)
62
For x ≥ xi+1 + tw + (at−w+v)2−(w−v)2
2awhen t ≤ τ , we should plug x = xi+1 + tw +
(at−v+v)2−(w−v)2
2ainto (B.8) to compute inf N
c(i)ini
(x, t):
inf Nc
(i)ini
(x, t) = −kixi+1 + bi for: x ≥ xi+1 + tw+(at− v + v)2 − (w − v)2
2aand t ≤ τ
(B.10)
Applying the second condition yields: v(t, x) = v = w(1 − κki