Numerical Scheme to Modeling Open

Channels
Flores Anthony J.
Moreira Sergio D.
Tixi Jeffer F.
Zuñiga Carlos O.

Facultad en Ingenierı́a de Ciencias de la Tierra, Civil Engineering, Escuela Superior

Politécnica del Litoral, Guayaquil; km 30 21 perimeter
Email: anjoflor@espol.edu.ec
sedemore@espol.edu.ec
jefftixi@espol.edu.ec
caoczuni@espol.edu.ec

Auctions are mechanisms that formalise the rules with which automated trading
schemes can be conducted, and in this paper we model the interaction of bidder
and seller agents in sequential computerised auctions. We study the outcome of
strategies that a designated “special bidder” (SB) may follow in the presence
of a collection of other bidders in an English auction, under the assumption
that the SB can make bids based on its observation of the ongoing auction as
a collective system. In our model, bidding and sale events are continuous time
random processes with discrete state-space, where the state-space represents the
current value of the most recent bid. We obtain analytical solutions which allow
the evaluation of measures of interest to the SB such as the probability of winning,
the savings with respect to the maximum payable price in the event of a win, and
the expected waiting time to win. We examine the effects of the SB’s time to bid,
and study how its decisions may be selected so as to optimise the SB’s measures
of interest.

Keywords: Automated Auctions; Analytical Models; Autonomic Systems; Internet

Technologies; E-Commerce; Queueing Systems
Received 00 January 2009; revised 00 Month 2009

1. INTRODUCTION it is also neccesary to consider the computational cost

Science is constantly advancing, as well as water studies
and the phenomena that are related to them. Most of
the partial differential equation in fluid dynamics do
not have an analitic solution because of the complexity
of their nature. As it says (Abbott, 1999), societies
are now governed by the knowledge provided by
technological development, this is where a new concept
called hydroinformatics is presented, which links the
problems of the world of water and technology now.
Another term used for this is Computational Fluid
Dynamics (CFD) that maintains the same essence of
the previous concept.
(Johnson, Rannacher, Boman, 1995) states that
CFD is a necessary tool for engineering and that
numerical approximation schemes can be many. In
many cases, one can find precise solutions, while in
others you can not. It is relevant to develop and
evaluate an algorithm that is efficient and able to
control errors when solving a problem. Furthermore
and for all these reasons, researchers are constantly
investigating new algorithms or even more an exact
solution.
This article is going to compare several numerical
schemes in a semi-explicit method who combine explicit
and implicit numerical approximation in open channels.
2. LITERATURE REVIEW
2.1. Alternating Direction Implicit Method
Typically, there are two kind of numerics methods:
explicit and implicit. In the first one, the result of
the iteration is the variable that we are looking for
while in the second one the varible is split in two or
more variable, and for each one there are a iteration.
[R.B.KELLOGG]
It is know that the equation Ax = b, where A is a
square matrix, b is a know vector and x is a variable in
the Euclidean Space; does not have a efficent algorithm.
One of the Implicit Method to approximate the solution

Hydraulics, ??, ??, ????

2 Flores A., Moreira S., Tixi J., Zuñiga C.

develop by Peaceman and Rachford in 1955 consists in addition of these expansions gives
split A = H + V and consider a strictly decreasing
secuence of positive numbers {rn }n≥0 . Then, with a U (x + h) + U (x − h) = 2U (x) + h2 U 00 (x) + O(h4 )
initial solution x0 and the following iterations:
Finally, we obtain a discretization with the following
1
(rn I + H)xn+ 2 = (rn I − V )xn + b forms.
Central difference approximation:
1
(rn I + v)n+1 = (rn I + H)n+ 2 + b  
0 dU 1
1
The sequence xn+ 2 converges to a z while the U (x) = ≈ [U (x + h) − U (x − h)]
dx 2h
sequence xn convergs to a y with x = y + z.
Forward difference approximation:
2.2. Runge Kutta method
1
U 0 (x) ≈ [U (x + h) − U (x)]
This method is used to solve differential equations, h
these methods need a differential equation that has the
following structure: Backward difference approximation:

dy 1
= f (x, y) U 0 (x) ≈ [U (x) − U (x − h)]
dx h

Some advantages of the Runge-Kutta methods are If U is a function of variables x and t, U could be
that they allow the calculation and evaluation of the represented by the following grid:
derivatives of f (t, y). The first-order Runge-Kutta
method is Euler’s method, but the most popular
version of Runge Kutta method is the 4 stage one,
the information that is needed to use this method is
dy
the initial value yi and representative value of the dx
with solution in the interval (xi , xi +4x) (Szymkiewicz,
2012), where 4x is the step size.
The following expression is the common formula of
the fourth-order Runge-Kutta method:
4x
yi+1 = yi + (k1 + 2k2 + 2k3 + k4 )
6
where:

k1 = f (xi , yi ) Fig1. Spaced grid line for the function U (G.D. Smith,
2008)
 
4x k1
k2 = f xi + , yi + 4x
2 2
  2.3.1. Lax Diffusive Scheme
4x k2
k3 = f xi + , yi + 4x This scheme is a finite difference approximation method
2 2 where the plane is divided into a grid that has an
k4 = f (xi + 4x, yi + 4x · k3 ) interval on x-axis of 4x and t-axis of 4t , the point
i of x-axis will be called by i 4 x and the point (i + 1)
2.3. Finite Difference Explicit Method by (i+1)4x grid point. The k will be used for t-axis the
point (k + 1) by (k + 1) 4 x grid point. To solve Saint-
Partial differential equations are difficult to solve and Venant equations the finite different approximations are
find an analytical solution in the most of cases, so, these selected to use this method (Akbari and Firoozi, 2010).
equations could be solved by a numerical procedure,
the finite difference method could approximate the ∂f f k − fi−1
k

solutions. When a function U is continuous, the Taylors = i+1

∂x 24x
theorem (G. D. Smith, 2008):
∂f f k+1 − f ∗
1 1 = i
U (x+h) = U (x)+hU 0 (x)+ h2 U 00 (x)+ h3 U 000 (x)+. . . ∂t 4t
2 6
1 k
f∗ = k


and f + fi+1
2 i−1
1 1 1
U (x−h) = U (x)−hU 0 (x)+ h2 U 00 (x)− h3 U 00 (x)+. . . D∗ = k k


Di−1 + Di+1
2 6 2
Hydraulics, ??, ??, ????
 Numerical Scheme to Modeling Open Channels 3

1 k 
Sf∗ = Sfi−1 + Sfki+1 3. METHODOLOGY
2
The Lax Diffussive Scheme needs to transform the An open channel can be modelling by Saint venant
PDE into the next form: dynamic equations that can be solved by numerical
methods, the two equations describe the mass and
Ut + Fx + S = 0 momentum conservation and can be written as follows
(Moussa,2000):
1
1 4t
Uik+1 = k k k k
−S ∗ 4t


Ui+1 + Ui−1 − Fi+1 − Fi−1
2 2 4x ∂A ∂Q
−q+ =0 (1)
∂t ∂x
2.3.2. Courant Number
To obtain an efficient numerical solution without ∂V ∂V
stability problems, the relation of time and space +V + g · (Sr − S) = 0 (2)
∂t ∂x
discretization is semi-explicit, it means that only a
few terms of the equation are discretized implicitly, 3.1. Finite Difference explicit method applied
this relation is described by the Courant number to a opne channel
(Aldrighetti Elisa, 2007).
One dimensional water flow can be described by Saint
4x Venant equations, for the explicit method we use the
4t ≤ q
|U | + gA next discretization for a square channel, taken from
B
(Nils Reidar B. Olsen, 2012).
2.4. Euler Method ∂U Ui+1,j−1 − Ui−1,j−1
= (3)
The Taylor method with n = 1, is called Euler’s method ∂x 24x
and was perhaps the first numeric method generated
long before the existence of computers. This is the ∂y yi+1,j−1 − yi−1,j−1
simplest but least accurate of all methods, being of first = (4)
∂x 24x
order accuracy only. It is an explicit method, given by
the next expression: (Fenton J. D., 1992) ∂y yi,j − yi,j−1
= (5)
∂t 4t
2
xn+1 = xn + 4F (tn , yn ) + O(4 )
The result to replace these approximation into saint
Where the right side of the differential equation has venant equations, the solution of step i,j for mass
to be evaluated only once per time step. The Landau conservation is:
order symbol O(42 ) means that neglected terms vary
4t
like the second or higher powers of the time step A. yi,j = yi,j−1 − [Ui,j−1 (yi + 1, j − 1 − yi−1,j−1 ]
In the case of equation (1), equation (11) becomes the 24x
simple expression:
4t
− [yi,j−1 (Ui+1,j−1 − Ui−1,j−1 )]
24x
S(t + 4) = S(t) + 4{I(t) − Q[t, S(t)]} + O(42 ) (6)
The evaluation of the right side of equation (12)
to obtain S(t + 4) is a much simpler process than
the numerical solution of the transcendental equation For momentum equation at step i, j is:
(2). However, the accuracy of Euler’s method is not
particularly good. The results are in error by more (Ui+1,j−1 − Ui−1,j−1 4 t
than 10% at the peak. Also, greater accuracy can be Ui,j = Ui,j−1 − Ui,j−1
24x
obtained by choosing computational steps smaller than
those used to define the inflow hydrograph (when linear g 4 t(yi+1,j−1 − yi−1,j−1 )
interpolation might be used to give intermediate values − + g 4 t(I0 − If ) (7)
24x
of the hydrograph). (Gmez, 2011)
There is another advantage to using smaller time Where I0 is the bed slope and If is friction slope that
steps. Not only would it give rather more accurate was given by Manning equation (Francisco Jaime Meja,
results, but the peak of the hydrograph would be 2008):
identified rather more accurately. This is a rather better  2
nQ
method of determining the time and magnitude of the Sf = 2 (8)
AR 3
maximum would be to use, for example, quadratic or
spline interpolation. (Lawrence, 1998) Flowchart(Nils Reidar B. Olse,2012).

Hydraulics, ??, ??, ????

4 Flores A., Moreira S., Tixi J., Zuñiga C.

b−a
h= (11)
N

Where N is an integer and represents the uniform

space in the interval [a, b] which is defined by the
user according to (Burden Faires, 2001). Once the
initial conditions have been specified, the following is
to express the K factors for the fourth-order Runge-
Kutta expression applied to the problem in question as
follows:

4t
ηi+1 = ηi + (k1 + 2k2 + 2k3 + k4 ) (12)
6

where:

Fig 2. Flowchart of Finite Difference explicit method

3.2. Runge-Kutta method applied to a reser- k1 = f (ti , ηi )

voir
To apply the method, the following example is
presented: we have a tank or tank that shows the level
of the surface that varies with the inflow, determining
the outflow rate. The governing equations are granted
 
4t k1
by (Fenton, 1992): k2 = f ti + , ηi + 4t
2 2
dS
= I(t) − Q(η(t), t)) (9)
dt
Where S is the volume of water stored in the reservoir,
I is the input volume rate as a function of time and  
Q is the output volume rate that relates the output 4t k2
k3 = f ti + , ηi + 4t
flow to the volume of storage that in turn depends on 2 2
the height η. Equation (9) is a differential equation
where the volume of the reservoir can be estimated in
time, which can be estimated using the height η which
also depends on the time dη
dt which is as shown in the
following expression:
k4 = f (ti + 4t, ηi + 4t · k3 )
dη I(t) − Q(η, t)
= = f (t, η) (10)
dt A(η)

The fourth-order Runge-Kutta method can be used in What results from the estimation of the height of the
equation (10), which is a differential equation with the reservoir to then get the volume of water stored:
structure needed to estimate the height of the reservoir
over time (Chapra Canale, 2007). To approximate the
solution of the problem you have an initial value ηi in
the time interval [a, b] where η(a) = ηi . The size of step dS
h expressed in the following manner is also defined: = A(η) (13)
dη

Hydraulics, ??, ??, ????

 Numerical Scheme to Modeling Open Channels 5

3.4. Preissmann Scheme

Preissmann Scheme is a Finite Differences Scheme, with
the consideration of the term i + 12 between i and i + 1
in each step of x-direction in the case of Saint-Venant
1D equations.

The x-t solution plane used to generate the finite

difference equations

The space derivatives in Saint-Venant equations

using this method is:

∂Q Qj+1 − Qj+1 Qi − Qji

= θ i+1 i
+ (1 − θ) i+1 (18)
x 4xi+ 21 4xi+ 12

∂h hj+1 − hj+1 hi − hji

= θ i+1 i
+ (1 − θ) i+1 (19)
x 4xi+ 12 4xi+ 21
 j+1  j+1
Fig 3. Flowchart of Runge-Kutta method BQ2 2 BQ2 2
 
BQ2 2 −
∂ A A A
i+1 i
=θ
3.3. Euler method applied to a reservoir x 4xi+ 12
In this way, the equations of Euler for this problem  i  2 j
BQ2 2
A − BQ A
2
could be written as is below: i+1 i
+(1 − θ) (20)
4xi+ 12
I(ti ) − Q(ηi , ti )
ηi+1 = ηi + 4 + O(42 ) (14) And the time derivatives are:
A(ηi )
∂A Aj+1 − Aj Aj+1 − Aji
Notice that is necessary the equations of inflow = 0.5 i+1 j i+1 + 0.5 i (21)
hydrograph I(t) and flow Q(n, t), as well as the section ∂t 4t 4tj
area. Therefore, it considers a small detention reservoir,
square in plan, where it obtains the next expression: ∂Q Qj+1 − Qj Qj+1 − Qji
= 0.5 i+1 j i+1 + 0.5 i (22)
√ 3 ∂t 4t 4tj
Q(η) = 0.6 gbη 2 (15)
To get the constant terms, we have:
Where g is the acceleration, 9.8 sm2 , b is the sharp crested
j+1 j
weir and ηis the surface elevation relative to their crest. q = θqi+ 1 + (1 − θ)q (23)
2 1+ 1 2
And the equation for area is:
A(η) = (100 + 4η)2 (16) Sf = θ(Sf )j+1 + (1 − θ)(Sf )j1+ 1 (24)
i+ 1 2 2
By last it is possible obtain the equation for the inflow
hydrograph, that expression could be: Se = θ(Se )j+1
i+ 1
+ (1 − θ)(Se )j1+ 1 (25)
2 2
 5 j+1 j
t 1−t A = θAi + (1 − θ)Ai (26)
I(t) = Qmin + (Qmax − Qmin ) e max
T (17)
Tmax
where the bar(-) means the average of the nodes i
(Fenton J.,2010) and i + 1.

Hydraulics, ??, ??, ????

6 Flores A., Moreira S., Tixi J., Zuñiga C.

4. PERFORMANCE MEASURES OF IN- Concerning (3) the average difference between the
TEREST TO THE SB AND TO THE valuation v for the good, and the price at which the
SELLER auction concludes given that the SB makes the purchase,
is denoted by
Some measures of interest to the SB are:
Pv
1. whether the SB is actually able to purchase the (v − l)P (A(R, l))
φ(v) = l=1 Pv . (31)
item it is seeking, l=1 P (A(R, l))
2. how quickly it can purchase the item,
3. whether it is able to minimise the cost of its 4.1. Optimisation on the part of the SB
purchase or equivalently how much it saves with
All that the SB can do, without reverting to deceit, is to
respect to the maximum price that it is willing to
adjust its bidding rate β to the situation it is observing,
pay, and what is its savings per unit time with
including the bid rate it observes concerning other
respect to the maximum price v that it might have
bidders, so as to optimise the performance measures
paid.
that it is selfishly and legitimately interested in.
On the other hand, the seller’s interest may be to In order to minimise ψ(v) it would suffice to take
maximise its income from a sale, or to maximise its β >> nλ. Then the SB raises its bid to the valuation
income per unit time for a sequence of sales. v very quickly so that it is always the winner, and π(v)
Note that P (0) is the ratio of the average time tends to 1. However this means that the SB would
elapsing from when the auction starts until the first bid be buying the good at its maximum price, rather than
arrives, to the total average time τ an auction cycle lasts driving a good bargain.
(including the “rest time” of average valuation r−1 after Thus a reasonable approach would be to choose a
an auction ends). Since the system leaves state 0 only valuation of β which maximises the SB’s return on the
when the first bid in an auction is made, the average auction, such as γ(v), the average savings per unit time
time spent in this state is simply the inverse of the rate that the SB makes with respect to the maximum price
at which the first bid is made, i.e. [nλ + β], and that it would pay, or specifically
Average time in state 0 v
P (0) = (27) φ(v) X
τ γ(v) = =r (v − l)P (A(R, l)). (32)
ψ(v)
P (0)−1 l=1
τ=
nλ + β If v is replaced by the random variable V , the function
rδ + (r + δ)(nλ + β) of interest to the SB is
= .
rδ(nλ + β) ∞ X
v
X
When a sale is made, the expected income of the seller Γ = E[γ(V )] = r (v − l)p(v)P (A(R, l)), (33)
is Pv v=1 l=1
l[P (A(R, l)) + P (A(O, l))]
I = Pl=1v , (28) and with the previous analysis we have
l=1 [P (A(R, l)) + P (A(O, l))]
and the seller’s income per unit time is ∞ X
X v
Γ=δ (v − l)p(v)G(l)P (0). (34)
I
ι= . (29) v=1 l=1
τ
Concerning (1), the probability that the SB is the Hence the SB could choose a valuation of β that
bidder that makes the purchase at an auction, rather maximises Γ.
than one of the other bidders, which we denote by π, it
is given by 4.2. Numerical examples
Pv
l=1 P (A(R, l))
We will now provide some numerical examples that
π = Pv (30) illustrate the predictions of the model. In all the
l=1 [P (A(R, l)) + P (A(O, l))]
X v    numerical results that are shown, we provide curves
r −1 for the case when all bidders including the SB follow
= P (A(R, l)) · [P (0)]
nλ + β a symmetric bidding strategy, i.e. λ = β, and for the
l=1
X v    more interesting case when the SB varies its bidding
rδ + (nλ + β)(r + δ)
= P (A(R, l)) · . rate while the rest have a fixed bidding rate. The topic
δ(nλ + β) of mutual adaptation of all bidders to each other is yet
l=1

Hence regarding (2) the average time ψ that the SB another important subject which is not discussed in this
waits to win an auction is the inverse of its winning paper.
rate or Comparisons of the asymmetrical bidders case where
τ 1 λ is constant, against the case with identical bidders
ψ(v) = = Pv . with λ = β are also shown in Figures 1, 2, 3, 4.
π r l=1 P (A(R, l))

Hydraulics, ??, ??, ????

 Numerical Scheme to Modeling Open Channels
FIGURE 1. SB’s expected time to win with δ = 0.5, r = 1,
n = 10, V ∼ U (80, 100).
FIGURE 2. SB’s expected payoff with δ = 0.5, r = 1,
n = 10, V ∼ U (80, 100).
In Figure 1, as we would expect, we see that in the
asymmetric case it suffices for the SB to bid at a
sufficiently high rate (the x − axis) in order to reduce
its time until it can make a purchase (the y − axis).
In Figure 2 we study the quantity φ(v). It is
interesting to see that for fixed δ and λ, even if the
SB increases β to very high valuations (and hence wins
the bid), the “expected payoff” φ(v) does not tend to
zero and only drops slowly with β. However, if λ = β
and they increase, then the pay-off will tend to zero.
On the other hand, when buyers are interested in
purchasing multiple goods from the auction or when
they have a long term view of things, the expected
payoff per unit time γ(v) can be a good criterion
for decision making. Figure 3, with the y − axis in
logarithmic scale, shows that the expected pay-off per
unit time increases very rapidly with β, and furthermore
this effect is accentuated, and the pay-off is greater,
when the other bidders are relatively slower, i.e. have
smaller valuations of λ. Figure 3 shows that bidding
at a high rate increases the payoff rate for the bidder
and that it leads to diminishing returns of payoff per
time beyond some valuation of β. In other words,
when the SB’s actions do not impact the other bidders’
behaviour, it should bid quickly. This is contrary
to what we observe in online auctions such as eBay,
in which “sniping” is often used regardless of other
bidders’ strategies. Bidders wait until the last possible
moment before the auction expires to place their true
bids.
Indeed, it has been suggested that sniping is a good
strategy [?, ?, ?] if the information on the closing
time of the auction is made public by the seller. But
this strategy has its shortcomings: in balancing the
benefits of submitting the very last bid against the risk
of bid being rejected for arriving after the auction has
ended, the bidders can misjudge. Technical issues such
as communication delay can aggravate the problem,
causing the item to sell at a lower price than what it can
fetch. It is desirable that the time spent in waiting for
the auction to close be shortened, thus saving time for
both seller and bidders; this is especially so when the
seller has many items to sell and time is of the essence.
Interestingly enough, a variant of the auction protocol
that was until recently used by Amazon tackles sniping
behaviour by automatic deadline extension: if any bid
FIGURE 3. SB’s expected payoff per unit time with
δ = 0.5, r = 1, n = 10, V ∼ U (80, 100).
Hydraulics,
7
FIGURE 4. Expected income per unit time for the seller,
with δ = 0.5, r = 1, n = 10, V ∼ U (80, 100).
is submitted within the last 10 minutes of the scheduled
closing time, the deadline is automatically extended
for another 10 minutes. This process continues until
10 minutes have passed since the last received bid,
at which time the auction concludes. While it may
succeed in discouraging sniping [?], this approach is
not always time effective: the scheduled deadline is
the best-case time within which the seller can hope
to make a sale, and in general it is likely that it will
take longer. This may not be suitable if the seller is
pressed for time. Note also that Amazon has stopped
running auctions as indicated in their Changes to the
Participation Agreement of April 14 20081 .
Finally Figure 4 looks at things from the perspective
of the seller; for fixed λ we see that the SB’s bid rate
β affects the seller’s income per unit time, but only in
a moderate way. This is to be expected because after
the SB makes a bid, it must pause and the remaining
bidders then have a chance to bid. Since there are many
other bidders (in this example n = 10) they will have a
significant impact on the outcome, while the SB’s effect
remains limited.
5. WHEN THE SB TRIES TO KEEP UP
WITH OTHER BIDDERS
An interesting question arises if the SB adjusts its
bidding rate β in a manner proportional to the bidding
rate of all other bidders. From the state equations (??)
we can set a value µ representing the relative rate at
which both SBs and other bidders are bidding, with
respect to the other bidding and decision rates. Thus
the quantity µ illustrates the “similar” behaviour of SBs
and of the other bidders, and we have
β (n − 1)λ
µ≡ = .
nλ + δ (n − 1)λ + β + δ
Then, the outcome of the auction for the SB will be
equivalent to that for the other bidders taken together.
In fact if n is large enough, this simplifies to
0 = β 2 + β[nλ + δ] − nλ[nλ + δ], (35)
which yields
r 
nλ + δ nλ
β≈ 1+4 −1 , (36)
2 nλ + δ
or r 
1 nλ
µ≈ 1+4 −1 . (37)
2 nλ + δ
1 See http://www.amazon.co.uk/gp/help/customer/display.html
?ie=UTF8&nodeId=200239030 that we have accessed on 22-04-
2008
??, ??, ????
8 Flores A., Moreira S., Tixi J., Zuñiga C.

β(l − 1)
FIGURE 5. Expected time to win for SB when keeping P (R(l)) = P (O(l − 1)), 2 ≤ l ≤ v − 1,
nλ(l) + δ(l)
up with the other bidders for various λ and n. Other
parameters: δ = 0.5, r = 1. (n − 1)λ(l − 1)
P (O(l)) = P (O(l − 1))
δ(l)
nλ(l − 1)
FIGURE 6. Expected payoff per time for SB when keeping + P (R(l − 1)), l = v,
δ(l)
up with the other bidders for various λ and n. Other
parameters: δ = 0.5, r = 1. β(l − 1)
P (R(l)) = P (O(l − 1)), l = v,
δ(l)
δ(l)
P (A(O, l)) = P (O(l)), 1 ≤ l ≤ v,
Figure 5 shows our model’s predictions on the r
expected time to win for SB, while in figures 6 and δ(l)
7 we show the payoff and income rates, respectively, as P (A(R, l)) = P (R(l)), 1 ≤ l ≤ v,
r
functions of varying λ and δ, when SB follows this policy r X X v
in keeping up with the other bidders. In Figure 5, for P (0) = P (A(U, l)),
nλ(0) + β(0)
each case of n, there exists a minimum expected time U =O,R l=1
to win that occurs at some λ, and for increasing n this v
X X  
minimal point occurs at smaller λ. Likewise, the highest 1 = P (0) + P (U (l)) + P (A(U, l)) .
payoff per time is obtained at a distinct valuation U =0,R l=1

of λ and this decreases with n. These observations We will first give the general solutions for this system,
correspond to increasing competition with n, and the and then look at a plausible example of forms that the
penalty suffered by SB for an increase in λ is larger dependent functions λ, β and δ might assume. Suppose,
for systems with large n; the drop in payoff per time is following similar approach in (??), we let
increasingly steeper with n (see Figure 6). On the other
hand, the seller benefits from large n and its income per P (O(l)) = H(l)P (0), 1 ≤ l ≤ v,
time has higher peaks, as shown in Figure 7. P (R(l)) = G(l)P (0), 1 ≤ l ≤ v.

6. PRICE DEPENDENT BIDDING We can then express the second order recurrence
relations in H(l):
In many cases the current price attained by a good offers
H(l) = c1 (l)H(l − 1) + c2 (l)H(l − 2), 3 ≤ l ≤ v − 1,
useful information about its valuation, and about the
(39)
situation of other bidders. Thus a model with bidding
where the coefficients c1 and c2 are
rates dependent on price was analysed in [?]. Here we
extend this approach to the behaviour of both the SB (n − 1)λ(l − 1)
and the other bidders. c1 (l) = , (40)
(n − 1)λ(l) + β(l) + δ(l)
We use β(l) and λ(l) to denote the bidding rates when
nλ(l − 1)β(l − 2)
the price is at level l for the SB and the other bidders, c2 (l) =
((n − 1)λ(l) + β(l) + δ(l))
respectively. Likewise, δ(l) will be the seller’s decision
rate when price is at level l. By a simple extension of 1
× ,
the previous model, the steady state probabilities for (nλ(l − 1) + δ(l − 1))
the system satisfy and, the initial valuations will satisfy
nλ(0) nλ(0)
P (O(1)) = P (0), (38) H(1) = (41)
(n − 1)λ(1) + β(1) + δ(1) (n − 1)λ(1) + β(1) + δ(1)
β(0) 1
P (R(1)) = P (0), H(2) =
nλ(1) + δ(1) (n − 1)λ(2) + β(2) + δ(2)
 
(n − 1)λ(l − 1) n(n − 1)λ(0)λ(1) nλ(1)β(0)
P (O(l)) = P (O(l − 1)) × + .
(n − 1)λ(l) + β(l) + δ(l) (n − 1)λ(1) + β(1) + δ(1) nλ(1) + δ(1)
nλ(l − 1)
+ P (R(l − 1)), Clearly, the difference equations (39) are linear
(n − 1)λ(l) + β(l) + δ(l) homogeneous with variable coefficients (40), and, hence,
2 ≤ l ≤ v − 1, the solution for H(l) can be expressed in closed form,
purely in terms of the coefficients[?, ?, ?]. First, define
a matrix:
 
FIGURE 7. Expected income per time for the seller when c2 (l) c1 (l)
Ml ≡ .
SB keeps up with the other bidders for various δ and n. c1 (l + 1)c2 (l) c2 (l + 1) + c1 (l + 1)c1 (l)
Other parameters: λ = 1, r = 1. (42)

Hydraulics, ??, ??, ????

 Numerical Scheme to Modeling Open Channels 9

FIGURE 8. Payoff per unit time in the price dependent where β0 , λ0 and δ0 are fixed nominal rates. Although
bidding model, against nominal bid rate β0 for various we use the same κ for the SB and other bidders, it is easy
pressure coefficients. Here n = 10, λ0 = 1.0, r = 1, δ = 0.5, to relax this restriction. When κ = 1, we have the case
and σ = 0. of “harmonic discouragement”, and if κ = 0 the bidders
are insensitive to price, and consequently, the whole
system reduces to the previously solved model (??).
Then, the solution sequence {H(l) : 1 ≤ l ≤ v − 1}, can
Now, for functionals of form (48), the explicit
be represented as a product of matrices {Ml } and the
solutions for H(l) will follow (43) and (44), where the
initial valuations:
    coefficients c1 and c2 are
H(2j + 1) H(1)
= M2j+1 M2j−1 · · · M3 (43)  κ
H(2j + 2) H(2) l+1 (n − 1)λ0
c1 (l) = , (49)
j     l (n − 1)λ0 + β0 + lσ (l + 1)κ δ0
Y H(1) v−2 κ
= M2i+1 , 0≤j≤ . 
l+1 nλ0 β0
H(2) 2 c2 (l) =
i=1
l−1 ((n − 1)λ0 + β0 + lσ (l + 1)κ δ0 )
Solving the above equation yields a set of two H(l), 1
one corresponding to an odd l and another to an even, × ,
(nλ0 + lκ (l − 1)σ δ0 )
for every j. However, the solutions will not hold at
the boundary l = v, because it involves a different set and the initial valuations become
of coefficients as given in (38). Thus, the boundary
nλ0 2κ
solution will be distinct and dependent on the previous H(1) = , (50)
two valuations: (n − 1)λ0 + β0 + 2κ δ0
(n − 1)λ(v − 1) 3κ
H(v) = H(v − 1) H(2) =
δ(v) (n − 1)λ0 + β0 + 2σ 3κ δ0
(44) 
n(n − 1)λ0 2 nλ0 β0

nλ(v − 1)β(v − 2) × + .
+ H(v − 2). (n − 1)λ0 + β0 + 2κ δ0 nλ0 + 2κ δ0
δ(v)δ(v − 1)
Similarly, the solutions for G(l) will be For G(l), the solutions will follow the general forms (45)
  j   and (47), where the coefficient
G(2j + 2) Y H(1)
= N2j+2 M2i+1 , κ
G(2j + 3) H(2)

l+1 β0
 (45)
i=1 d(l) = , (51)

v−3 l nλ0 + lσ (l + 1)κ δ0
0≤j≤ ,
2 κ
and the initial valuation is G(1) = nλ20 +2
β0
κδ .
0
where the matrix and the coefficients are The examples in Figure 8 illustrate the effect of κ on
 
d(l) 0 β(l − 1) the expected payoff per unit time for the SB. We see
Nl ≡ , and d(l) = . (46)
0 d(l + 1) nλ(l) + δ(l) that the pressure coefficient does not make a difference
for relatively small bid rates β0 , and that a higher
Again, at the boundaries l = 1 and l = v, the solutions
coefficient fetches a better payoff rate at higher bid
will be different:
rates. Also, a small increase in κ from 0 to 0.2 yields
β(0) a bigger difference in payoff rates, than an equal-sized
G(1) = , (47)
nλ(1) + δ(1) increase from 0.8 to 1.0.
β(v − 1)
G(v) = H(v − 1).
δ(v) 7. CONCLUSIONS
The solutions above are general, and will hold for In this paper we have considered auctions in which
all price dependent functions. Suppose now, that the bidders make offers that are sequentially increasing in
dependencies are such that λ and β will decrease while value by a unit price in order to minimally surpass
δ will increase, with the price level l. Specifically, let the previous highest bid, and modelled them as
a pressure coefficient κ ≥ 0 [?] to represent the degree discrete state-space random processes in continuous
to which the attained price discourages bidding, while time. Analytical solutions are obtained and measures
σ ≥ 0 represents the effect of higher prices on the seller’s that are of interest to the SB are derived.
tendency to sell: The measures that can be computed in this way
β0 include the SB’s probability of winning the auction, its
β(l) = , l ≥ 0, (48) expected savings with respect to the maximum sum it is
(l + 1)κ
λ0 willing to pay, and the average time that the SB spends
λ(l) = , l ≥ 0, before it can make a purchase. An extension of the
(l + 1)κ
model that incorporates price-dependent behaviours of
δ(l) = lσ δ0 , l ≥ 1, the agents has also been presented.

Hydraulics, ??, ??, ????

10 Flores A., Moreira S., Tixi J., Zuñiga C.

The model allows us to quantitatively characterise
intuitive and useful trade-offs between improving the
SB’s chances of buying a good quickly, and the price
that it has to pay, in the presence of different levels of
competition from the other bidders.
There are interesting extensions and applications
of these models that can be considered, such as the
behaviour of bidders and sellers that may have time
constraints for making a purchase, and the possibility
of the SB’s moving among different auctions so as
to optimise measures which represent its self-interest.
Another interesting area of study may be to examine
bidders who are “rich” and are willing to drive away
rivals at any cost, and who may create different
auction environments for bidders that have significantly
different levels of wealth. Yet another area of interest
concerns auctions where items are sold in batches of
varying sizes, with prices which depend on the number
of items that are being bought.
[8] Chapra, S. C., Canale, R. P. (2007). Mtodos numricos
para ingenieros (Quinta ed.). Mxico: The Mc Graw-Hill
Interamericana.
[9] John D. Fenton (1992) Reservoir routing, Hy-
drological Sciences Journal, 37:3, 233-246, DOI:
10.1080/02626669209492584
[10] Fenton, J. (2010). Computational Hydraulics. Wien.
[11] Fenton, J. D. (1992). Reservoir routing. .
doi:10.1080/02626669209492584
[12] Gmez, A. R. (2011). A solution to a general system of
Euler Equations for a compressible fluid. ISSN 0124-
8170. Bogot.
[13] Lawrence, E. (1998). Partial differential equation. Vol
19. American Mathematical Society.

ACKNOWLEDGEMENTS

This research was undertaken as part of the ALADDIN

(Autonomous Learning Agents for Decentralised Data
and Information Networks) project and is jointly
funded by a BAE Systems and EPSRC (Engineering
and Physical Research Council) strategic partnership
(EP/C548051/1).

REFERENCES
[1] G. Akbari, B. Firoozi, (2010), Implicit and explicit
numerical solution of Saint-Venants equations for
simulating flood wave in natural rivers. In: NATIONAL
CONGRESS ON CIVIL ENGINEERING, 5., 2010,
Mashhad, Iran. Proceedings Mashhad: Ferdowsi
University of Mashhad, 2010.
[2] R. Moussa, C. Bocquillon. Approximation zones of
the Saint-Venant equations f flood routing with
overbankflow. Hydrology and Earth System Sciences
Discussions, European Geosciences Union, 2000, 4 (2),
pp.251-260. hal-00330834.
[3] Smith, G. D. (2008). Numerical solution of partial
differential equations: Finite difference methods.
Oxford: Clarendon Press, pp 6-9, ISBN: 9780198596509
0198596502.
[4] Francisco Jaime Meja. Relacin de las curvas de energa
especifica y pendiente de friccin con las zonas de flujo
libre en canales. Revista EIA, ISSN 1794-1237 Nmero
9, p. 69-75. Julio 2008.
[5] Aldrighetti, Elisa. (2007). Computational hydraulic
techniques for the Saint Venant Equations in arbitrarily
shaped geometry. Universit di Trento, 2007.
[6] Nils Reidar B. Olsen. (2012). Numerical Modelling and
Hydraulics. Department of Hydraulic and Environmen-
tal Engineering The Norwegian University of Science
and Technology, pp 33-38, ISBN: 8275980747
[7] Burden, R. L., Faires, J. D. (2001). Anlisis numrico
(Sptima ed.). Mxico: Thompson Learning.

Hydraulics, ??, ??, ????