OPTIMAL CONTROL APPLICATIONS & METHODS, VOL.

2, 1-21 (1981)

OPTIMAL EXPLORATION AND CONSUMPTION OF A

NATURAL RESOURCE-DETERMINISTIC CASE*

N. A. DERZKO
Department of Mathematics, University of Toronto, Toronto M5S I A I , Canada

AND

S. P. SETHI
Faculty of Management Studies, University of Toronto M5S I V4, Canada

SUMMARY
An exploration and consumption model for exhaustible natural resources is used to formulate a deterministic
optimal control problem over an infinite time horizon. A dynamic programming approach is employed
throughout. The non-linear partial differential equation which results from an application of the necessary
conditions is solved analytically by the method of characteristics. Several properties of the solution are
discussed.
KEY WORDS Optimal exploration and consumption Exhaustible natural resources Economic
models Dynamic programming

1. INTRODUCTION
The Hotelling model’ predicts that the price of a natural resource rises at the market rate of
interest. If there are extraction costs, then the price would be the sum of the marginal extraction
cost and the rent on scarce resources, and the rent rises at the rate of interest.
In the case of many minerals, particularly oil from 1950 to 1970, the Hotelling result is not
observed. Not only did prices or rents not rise at the rate ofinterest, but they even declined in some
cases. The Hotelling model and many subsequent exhaustible resource models, it should be noted,
have no mechanism which can lower the price from one period to the next. Arrow’ has attempted
to incorporate such a mechanism by introducing uncertainty into the exhaustible resource models
in order to explain the discrepancy in the behaviour of prices as noted above. Hence, new
discoveries yielding upward revision of estimated reserves change the basis of calculation for the
Hotelling rents. The predicted rise at the rate of interest is offset by repeated downward revision of
the initial price in response to changing supply estimates.
Arrow assumes that the process of discovering resource mines is given by a Poisson process.
Arrow has partially analysed this very interesting but difficult model. Much of his analysis remain
conjectural, however. We intend to conduct an extensive analysis of Arrow’s model in a follow-up
paper. In order to do so, we must first attempt to solve the certainty analogue of the Arrow model
and obtain some useful insights. The certainty model is an interesting model in its own right. Its
mathematics is elegant as well as fairly complicated. Most importantly, the solution procedure

* This research is supported by the Natural Resources Management Program and Grant A4619 from the Natural Science
and Engineering Research Council of Canada.
0143-2087/81/010001-21$01.~ Received 19 December 1979
@ 1981 by John Wiley & Sons, Ltd. Revised 29 August 1980
2 N. A. DERZKO AND S. P. SETHI

developed here promises to be extremely helpful in conducting a thorough analysis of the difficult
stochastic model.
In Section 2, we develop the certainty model as a deterministic optimal control problem over an
infinite horizon. Section 3 contains the dynamic programming formulation of the model. The
solution of the model is carried out in Sections 4 and 5. The procedure involves a Picard
formulation of the partial differential equations arising in the paper and their solution by the
method ofcharacteristics. We use this method and not the maximum principle because the present
method can be easily extended to the stochastic model. The solution procedure also requires
manipulating a (missing) boundary condition. The economic interpretations of the various
solutions are also provided in these sections. Discussion of results along with ideas on further
extensions of the certainty model in Section 7 concludes the paper.

2. THE CERTAINTY MODEL

We introduce the following notation:
R(t)= the known reserves of the natural resource at time t; R(t)>O,
cqt) = the rate of consumption of the resource at time t; c(t)>O,
U ( c )= the utility function with U ( c ) 2 0 ,U’(c)>O, U’(0)= m, U”(c)<O,
p = the constant discount rate, p>O,
A ( t )= the area of unexplored region at time t ; A ( t ) a O ,
i. = the amount of natural resource per unit area of the unexplored region;
O<i.< x ,
L . ( I ) = the rate of exploration at time t; O < u(f)< ti (a given positive
constant),
P = cost of exploration per unit area (in terms of goods-in-general2);
O<P<;C.
The deterministic optimal control problem can now be stated:

subject to the state equations

A=-r, A(0) = A , (2)
A = - c + i . ~ , R(0)= R, (3)
with the state variable inequality constraints

and the control variable constraints

and

The solution of the problem (1H6)is carried out using the method of dynamic programming
formulated in the next section.*

* Note that the utility function assumed in this paper does not explicitly depend on time. This simplying assumption leads
to a non-increasing consumption over time. The fact that this did not happen in reality, say for oil from 1950 to 1970. can
be explained by such factors as population growth, technological development, etc. To include these factors would
require the utility function of the form U ( c , r )making the model unnecessarily complex for our immediate purpose.
 OPTIMAL EXPLORATION AND CONSUMPTION 3

3. THE DYNAMIC PROGRAMMING FORMULATION

To apply the method of dynamic programming to our problem, we let V(A,R), the current-value

r
f ~ n c t i o n be
.~
V(A,R) = max e - P ‘ [ U ( c ) - P u ] d t , A(t) = A , R(t) = R
c, v

Note that the above holds for any t; in particular, it holds for t = 0. We use the Bellman optimality
principle to state that

c, u
+
V(A,R) = max [{ V(c)- Pu} dt e - p d f V[A(t+dt), R(t +d t ) ] ] (7)
where the maximization in (7)is subject to constraints (4H6).By Taylor’s expansion? and the use of
(2) and (3), we can write
V [ A ( t+dt), R(t +dt)] = V(A,R)- VA u dt + VR( - C + L ~ ) d t (8)
Substituting (8) in (7), noting that e-pdf= 1 -pdt, cancelling V(A,R) on both sides, dividing
throughout by dt, and taking the limit as dt + 0 gives the Hamilton-Jacobi-Bellman equation
p V = max[U(c)-VRc+u(IZVR-VA-P)], V(0,O)= O
f, v
(9)
which the value function must satisfy subject to (4H6).
Since the maximization in (9) is subject to constraints (4H6),the complete specification of
V(A,R) on the (A, Rtquadrant would require us to analyse (9)in three mutually exclusive regions
of the (A, @-quadrant:
Case 1: A > 0, R > 0 (interior),
Case 2: A = 0, R > 0 (R-axis),
Case 3: A > 0, R = 0 (A-axis).
Before we begin the case-wise analysis of (9,we must observe that the optimal consumption will
always be an interior solution, i.e.
c*(t)>O (10)
on account of the assumption that U‘(0) = co.This will allow us to obtain c* by simply equating
the c-derivative of appropriate expression to zero. Furthermore, we shall henceforth assume
U(C)= Jc (11)
in order to obtain a closed-form solution for V(A,R). This is done simply for convenience in
exposition as the entire development in this paper can be carried out for any general utility function
having properties specified in Section 2.

Case 1: A>O, R>O

In the interior of the (A, Rtquadrant, the maximization in (9) need only be subject to the control
constraints ( 5 ) and (6). In view of (lo), it is easy to see that U’(c*) = V,. Thus,
1
c* = 7
4VR

for the square-root utility function (11). To maximize (9) with respect to u, we observe that (9) is
t This assumes Vto be continuously differentiable. This assumption is not unduly restrictive because the solution we find
satisfies this requirement.
4 N. A. DERZKO AND S. P. SETH1

linear in u. Consequently, the optimal exploration rate u* is given by

if W<O

if W>O
where
w = iV,-V,-P (14)
The function W is known as the switching function for obvious reasons. W <Odefines the region of
no exploration, and W > 0 defines the exploration region.
Notice that when W = 0, the maximum operation in (9)does not give any information about u*.
This situation is referred to as the singular control ~ i t u a t i o n . ~While
. its resolution usually

+["
requires some further analysis, it will not be found to be so in the present case.
With (12) and ( I 3), we can rewrite (9)in the interior as the following Hamilton-Jacobi equation:
1
in the no-exploration region ( W<0) (15a)

1
-+(iVR-V,-P)C in the exploration region (W > 0) (15b)
4VR
We also observe that, for W = 0, (15) reduces to p V = 1/4V,. Thus, the double-value definition in
(15) does not imply discontinuities. We will make additional remarks on this singular situation
whenever necessary.

Cuse 2: A = 0, R > 0
On the R-axis, there is no area to beexplored. To satisfy A 2 0 of(4), we must set u* = Oin (9).The
form of c* is the same as in (12). Thus,
1
PJ'(O, R ) = 4Vd0, R)' V(0,O)= 0
and the solution becomes
V(0,R) = (R/2p)+, C * = 2pR (16)
it is possible to obtain (16) directly by solving the simple optimal control problem of maximizing
j g e - P , / ~ d t subject to I? = - c and constraints R(t)>O and c(t)>O for all t .

Case 3: A > O , R = 0
Ifthere is to be any consumption (see (10))in this case, there must be some exploration to support
it. More precisely, we must satisfy (4) by requiring

R =-c+lv20 (17)
Of course, the path would enter the interior (i.e. Case 1) if c<Au. While this possibility can be
mathematically ruled out, we prefer to use an economic argument to conclude that

c =Iu (18)
Equation (18) implies that once the path enters the A-axis it will stay on the A-axis.
Consider the point in time t > O when the optimal path enters the A-axis. The optimal
consumption rate c*(t),<iV from (6). Thus, we have three possible optimal control scenarios: (i)
 OPTIMAL EXPLORATION AND CONSUMPTION 5

c*(t) = IF, u*(t) = q(ii)c*(t)<AtF,u*(t) = c*(t)/A;(iii)c*(t)<Av;u*(t)>c*(t)/A.Condition(18)holds

for (i) and (ii).To rule out (iii),we note that the positive discount rate in our model implies that c*(t)
cannot increase over time. This can also be rigorously proved. Thus, setting u*(z) = c*(z)/Afor z 2 t
is feasible and increases the value of the objective function by postponing the exploration of some
areas. This is a contradiction which implies that (iii) cannot be an optimal scenario.
With (18), we can reduce problems (1H6)into a simple problem along the A-axis, i.e. a problem
ofmaximizingJ,"e-P'[,/c-Pc/I]dt subject toA = -c/IandconstraintsA2OandO<c(t)<IF. It
is then possible to derive, either directly or from (9), the followingequation for V(A,O)(or Vo(A)for
convenience):
pv0 = max [Jc -( P
O<c<rid
+ vO, c / I ] , ~ ' ( 0 )= V(O,0) = o (19)
By carrying out the maximization operation, we obtain

c* = min I61

and

PV = 1
N
\
-,4
h
a
N
\
e
v

I1
h

4
0
v
P

-
c
V(A,O), the solution of (21) A
Figure 1. Characterization of V(A,R )
6 N. A. DERZKO AND S. P. SETHI

Equation (21)will be solved in the next section to obtain V o ( A )or V(A,O).The complete solution
will be obtained by solving the partial differential equation in the interior to satisfy the boundary
conditions V ( 0 ,R ) given in (16) and V(A,O)given by the solution of (21) in the next section. The
previous statement is depicted in Figure 1. It should be intuitively obvious that, in case the interior
contains both exploration and no-exploration regions, the exploration region must be below their
interface.

4. PICARD FORMULATION AND SOLUTION OF V(A,O)

In order to apply standard theory to (21a) and (21b), we reformulate these conditions to read
V/oA= $(VO)
which we call the Picard formulation.
By solving both possibilities in (21a, b), one immediately finds that

depending on whether the respective conditions hold. Further, we check then when V o< J(i$/2p
both formulas yield ;.'/4(VO, + P)' < j.6. We conclude that VO, = V o )in this case because the
alternative violates the condition of (21b). When V o >J(jL$/2p, we conclude VO, = dZ(Vo)by an
, that continuity of 4 is assured.
analogous argument. When V o = ,/(i6)/2p, $l(Vo) = 4 2 ( V o ) so
Thus, we can now rewrite (21) in Picard formulation as follows:

For small V o ,and for that matter for small A since V o ( A )increases with A , we can solve (22a) to
obtain'

It is easily shown that (23) implicitly specifies a unique increasing function V o ( A )= V(A,O)in the
range [0, i./4pP). A sketch of V ( A ,0) is shown in Figure 2.
Furthermore, the optimal exploration rate

is shown in Figure 3.
It should be clear that if
1.
GqP2 (25)

then L:* in Figure 3 is always feasible, and (23)specifies V ( A ,0) over the entire A-axis. If, on the other
 OPTIMAL EXPLORATION AND CONSUMPTION 7

x
-
4PP

A
.
*

Figure 2. Graph of VA,0)

hand,

then we can find A for which u*(a= fi, i.e.

A 1 &
A= 2pP
This is also indicated in Figure 3. Moreover, it is easily shown that (23) yields
V(A,O)= J(Ao3PP (28)
In the case of (26), we must solve (22b), that is,

The solution of (29) turns out to be

V(A,O)= 2P +P [ J(”(i)- [
2 P ] 1 - exp [$(A- A)]

Once again, V(A,0) for A > A is an increasing function of A, and is sketched in Figure 4. In
Figure 4, we have also included the V(A,0) of Figure 2 which is applicable to the case of sufficiently
large 6 given by (25).
8 N. A. DERZKO AND S. P. SETHI

v’

x
-
4P

-
V

\
- *
A A
Figure 3. Graph of u*

We have now obtained V ( A ,0) for both cases, i.e. when V2i./4P2 and V<A/4P2. This provides us
with the value of V(A,O)on the A-axis required for the complete solution of the problem as
indicated in Figure 1. We synthesize the complete solution in the next section using the
method of characteristics.

5. SYNTHESIS OF THE COMPLETE SOLUTION

The complete solution or synthesis of V ( A ,R) in the entire ( A ,R)-quadrant will be carried out under
two separate cases: (i) 172
i.j4P2 and (ii) 6< E.j4P2.

Cuse 1: d>i./4P2
When dki./4P2, we only need (22a)or (21a) to compute V(A,O).Equation (2la)was derived from
( 19).It
is also possible to derive (21a) from the original equation (9).In this full setting, the optimal
exploration u* given by (24) is singular which implies that W = AVR- V,-P = 0, or
V”+P
VR = - (31)
I.

With W = 0, the optimal consumption from (9) and (31) is

1 R
c* = = 4(V”+P)2‘
 OPTIMAL EXPLORATION AND CONSUMPTION 9

I
-A
I

A
.
d

Figure 4. Graphs of V o from (23) and (30)

which is precisely the consumption obtained from (21a). We, therefore, conclude that the solution
of (22a) corresponds with W = 0 in the full setting.
It is now possible to attach a no-exploration (or, W< 0) solution to the A-axis solution V(A,0).
Solving p i / = 1/4VR with the i/(A,O) of (23) yields
V(A,R) = [V(A,0)' +R/2p]* (33)
It is easily shown in the Appendix that this value function solution implies W<0 in the interior.
Hence, it provides us with the optimal solution. We summarize this solution in Figure 5.

Case 2: 6<L/4Pz
In this case, we can attach the no-exploration solution given by (15a) or (33) to the part of the A-
axis defined by A < A. It can be shown that attaching the no-exploration to the part A > A of the A-
axis leads to W>O in the interior and thus leads to a contradiction. This should be expected
because the V(A,0) solution for A >A corresponds to the W >0 solution in the full setting. To see
this, we must try to derive equation (21b)from the original equation (9); note that it was obtained
from (19)in the previous section. Since u* = 6,it corresponds with W >0 in (9). On the A-axis, this
means c* = 16in order to maximize (9) subject to the constraint R 20.
The attachment of the exploration solution satisfying (15b) to the part A > A of the A-axis is
carried out next. It is convenient to summarize our task in Figure 6.
10 N. A. DERZKO AND S . P. SETHI

V(A,R) = [V(A,O)
2
+ V(0,R)
2
114
= [V(A,O)
2
+ R/2pl
112

2
c * = 1/4VR , v* = 0

(no-exploration region)

0
I1
Y

A
V(A,O) g i v e n by ( 2 3 )
n

Figure 5. Optimal solution for V> 1./4P2

The solution of V ( A ,R ) in the region A > A will be carried out by the method of characteristics7
which applies to general first-order partial differential equations of the type
F(A, R, V, V,, V,J = 0 (34)
If we let
a = V,, b = V,, a =Fa, p = F,, y =F, (35)
the characteristic curves satisfy

’(
-
dt ; ) = [ ! b / i ’ )a+ay
-a

-b P+bY
 OPTIMAL EXPLORATION AND CONSUMPTION 11

1
3v =-, w 0
4vR

no-exploration region

V(A,R) of (33)

exploration region

-
V(A,O) of (23) A V(A,O) of (30)
Figure 6. Summary of results for synthesis of V(A,R)

In our case, we have

where
w = AvR-vA-P>o
If we let 7 denote the parameter of the autonomous system, we have
-v

(39)

- Pb
where i denotes dzldz; in fact, the parameter 7 denotes the forward time with z = 0 corresponding
12 N. A. DERZKO AND S. P. SETH1

to the time t at which the optimal path enters the A-axis. The solution of (39) is given by
a = aoepr (40)

-
U 1
V = I/o+-(-ao+ibO)(eP'- l)+--a(e-p*-l). (44)
P 4Pb
To obtain the various initial conditions in (40H44),we calculate
W = i.b-a-P = ( P + Wo)eP'-P (45)
where
W o= j,bo-ao-P (46)
We have already concluded that W o> 0. It is also easily seen that W ( t )decreases as 7 decreases
from 0. We would be interested in finding the points in ( A ,R)-space where W = 0 as is noted in
Figure 6.
We now obtain the inital conditions in equations (40H44), i.e. the conditions corresponding to
the A-axis. We know that R' = 0, and V o is given by (30). We can then compute

To obtain bo, we set

1
pVo = -a+(Abo-ao-P)ij
46
Note that (48) is the exploration-region equation (15b) which corresponds to (22b) in the full
setting. It yields
1
bo=211(lv) (49)

If we now make use of equation (49, we can find I; i.e. the value of parameter 7 , for which W = 0:

T=--1n
P ( "p")
I+-

Using (50)in (42)and (43),we can now obtain the value A' which corresponds to A and R values on
the singular manifold W = 0. Thus,

A=Ao+-ln
P ( Yo)
I+-

and

Since Wocan be easily written in terms of A', equations (51) and (52) are parametric equations for
the singular manifold or exploration/no-exploration interface. Let us denote
R = Rs((A) and V = VS(A) (53)
 OPTIMAL EXPLORATION AND CONSUMPTION 13

as the equation of the singular interface and the value function, respectively, along it. Then, we can
express the complete solution as is shown in Figure 7.

R d b I
I
I
I
I -
V(AyR) = I
I
V(A,R) =
[
V"(A)2 +
2P

E(A,0)2 + I
2P I
1
of (23)
I
I w < o
hl
I
\
4 I w = o

V ( A y R ) given by
( 4 4 ) Y (47)Y (49) and (30)

V(A,O) of (23)
V(A,O) of (30) W > 0
Figure 7. Complete specification of V(A,R)

It must be mentioned that the singular manifold has zero slope at the point (A, 0). Furthermore,
Rs(A) is an increasing function of A with R(A)= 0 and

To obtain (54), we rewrite Wof (45) with T = Tand set it to zero. This yields

and implies that

Using (56) in (52) gives (54).

14 N. A. DERZKO A N D S. P. SETHI

6. PROPERTIES OF THE SOLUTION

Having found the value function and corresponding optimal solution, we are ready to consider
some subsidiary questions.
It sometimes happens that optimal time trajectories follow a singular interface. We show this is
not the case here. Along any path we must have
-cdt = RdA+dR
(57)
-vdt=dA
It follows that
c
-
dR
= j.+-
V dA
and from (12)

Noting that in (51) and (52) both R and A are parameterized by Wo,we write
dR dR/dWo
dA = dA/d Wa
Furthermore
dR
--=-
dW
;.F( I+--
P
yo
and

>-( P P l+WO/P
1 )
whence

Using this result, we find from (58) that

1
u>
4m31 + WO/P)2
Using (41), (49), and (50),we obtain

and substitute into (59) to obtain

1

This shows that it is not feasible to coast along the singular interface.
 OPTIMAL EXPLORATION AND CONSUMPTION 15

Next, we consider the time behaviour of VR and V, along optimal trajectories to verify the
Hotelling result' on shadow prices. Now,

d
s V , 4 = PV,.

The proof is of necessity by regions in (A, R) space.

No-exploration region
Equations (3) and (12) give

Since optimal trajectories in the region satisfy A = 0, we have

From equation (15), it follows that

VZ,+ VVRR = O
or
1-
_V VRR-+T/;,
- -72--
VR

Combining (64) and (65) produces (61).

The proof of (62) is further subdivided into Case I (A c A) and Case I1 (A > A, R > Rs((A)).In Case
I, we proceed as follows. Equation (33) gives

V(A,R)2 = V(A,0)2 -
2P
+R
and, after differentiating, we have

V(A,R ) aA V(A,R ) = V(A,0) B,V(A, 0)

where 8, denotes a partial derivative with respect to A. As R changes with time

16 N. A. DERZKO AND S. P. SETH1

Therefore, from (63) and (64),we have

R - Rs(A)
V ( A ,R)2 = V ( A ,R5(A))2+
2P
R
= ;(A) +-
2P
which defines [ ( A ) . Partial differentiation with respect to A yields
V ( A ,R ) ? , V ( A ,R ) = +('(A) (67)
Since the right side of (67) depends on A alone, the steps following (66) are valid here, and (62)
follows as in Case I.

Exploration region ( A > A, R < Rs(A))

When R>O equations (61) and (62) are implicit in this case in equations (35),(40), (41), and the
observation on t following equation (39). For R = 0 (62) follows from (2), (22) and (241, while (61)
does not hold in the normal sense.
Finally, we discuss the optimal trajectories c*(t) and u*(t) as functions of time. Consider a path
through ( A , R ) space as shown in Figure 8. The path goes through four regions as labelled.
Discussion of the relevant formulas by region follows.
I. L'* = 0

VR = Vie@

At time t , the junction a is reached.

11. Upon crossing the junction a, u* jumps to v* = 17,and V, remains continuous. V, = Vie@,
and c* = l/4% still holds here. At time t , (junction j),V, = 1/2J(LG) from equation (49) so
that
c* = j.6

111. t'* remains at value 6,while c* = A6 from t , until t , is reached at the junction y.
IV. In this region,

From (20) and (21), and

c * = c*/E.
 OPTIMAL EXPLORATION AND CONSUMPTION 17

IV Y 111 B A
Figure 8. Optimal path in the (A,Rtquadrant

Also, equation (23) enables us to calculate Vo(A)so that c* and, v* can be written as functions of
A. Since dA/dt = - u*(A), this equation enables us to solve for A(t)which, after substitution, gives
c*(t) and v*(t). One final comment is in order. We show that c*(t) and v * ( t ) remain positive as
t + 00. This is done as follows. Analysis of (23) for small v yields

Therefore,
1
-AxVZ,
4P
and it follows that
dA -
_ 1
- -v*= -4p2-A =-plA
dt 4P
Hence, A >0 as t -,00, and the desired conclusionfollows. Based on the above consideration, we
can sketch Figure 9 for the case when A> 1.
18 N. A. DERZKO AND S. P. SETHI

c*,v*

C*

no closed formula
for c*
C* = hv (see discussion)

- - - - --
c -_---._--
------- %
---- - _
-- -.. --- L-
>
I tl I1 t2 I11 IV t
Figure 9. Graphs of c*(t) and u*(t)

7. DISCUSSION AND EXTENSIONS

We have now completed the solution of the exploration and consumption model under the
assumption of certainty. We restate the following highlights of the solution.
(1) The value function along the A-axis has an upper bound. This upper bound increases with I
and decreases with P and p. When Cis low enough to be restrictive, we note that this upper bound
also decreases as C decreases.
The upper bound comes about because the exploration is expensive. This is easily seen by noting
that V(0,R = co) = 00 while V(A = 00,O)cco even though the infinite unexplored area contains
an infinite amount of resource.
(2) Because U’(0)= co,we note that all the area will eventually be explored. That is, the amount
of unexplored area goes to zero as t --t co.
(3) The exploration rate never exceeds A/4P2. This is true even if there is no upper bound on u.
Parenthetically, we note that when there is no upper bound on u, it is possible to impulse-explore
any finite amount of area instantaneously. It is trivial to conclude that there will be no exploration
in this case as long as R > 0. This, of course, is a very special case of our model when 62 A/4Pz.
Note further that the maximum exploration rate increases proportionally with I but decreases
inversely with the square of unit exploration cost P. This comes about because of the square root
utility of consumption with linear exploration costs.
(4) When Vis small, postponing exploration until R = 0 when A > A would mean a subsequent
lower consumption limited by 6even though there is plenty of resource, i.e. more than IA, under
 OPTIMAL EXPLORATION AND CONSUMPTION 19

ground. Therefore, exploration starts at some time before the optimal path hits the A-axis. For
A <A,there is not enough resource under ground, and Gprovides no restriction on what would be
consumed once R has dropped to zero. Furthermore, c* and u* both decrease over time as is shown
in Figure 9.
(5) The derivatives, VR and V, of unexplored land, increase over time at the same rate as the
discount rate. This means that the shadow price of the resource, which is V, in the interior,
increases at the rate p. On the A-boundary, the shadow price of the resource is (P+ VJ/A;P/A is the
marginal extraction cost, and VJA is the rent. Only the rent part of this price increases at the rate p.
This is precisely the Hotelling results for 6 2 il/(4PZ),when VAA,0 +) = ( P + VAA,O))/A.
For IF< A/(4Pz),we have V ( A ,0 +) = ( P + VA(A,0) + Wo)/A.Thus, the shadow price undergoes a
downward jump of Wo/Aat time t , (see Figure 9) when R becomes zero. This lowering of the
shadow price occurs due to the constraint u,< V; which restricts the optimal consumption along
R = 0 in the interval ( t z ,t3).
In our next paper, we will deal with a stochastic version of this model developed by Arrow.' As
indicated earlier, this model assumes a Poisson process for resource discovery. Because of the
uncertainty, we shall find that exploration starts before the path hits the A-axis even though 17is
sufficiently large or even when G = co.

APPENDIX: THE SIGN O F W

For economy in notation, we let V(A,R) and V o denote V(A,O). From (33), we have
I/' = Vo2+R/2p (68)
This implies
2VVR = 1/2p (69)
for R > 0 so that by continuity this will hold for the one-sided derivative VO,when R = 0. Also, (22a)
can be written as
O=-PvO-VOVo,+-
A
4P
= -PvO-vOI/o,+vOVO,

= VO wo
where we have used (69) and (14). We conclude that W o= 0 for O < A < A .
If we compute the A-partial derivative of (68), we deduce

Then,

=-PV-VOVo,+- ;1
4P

because V(A,R) > V ( A ,0) for R >0.

We conclude that W<0 for A <A and R > 0.
20 N. A. DERZKO AND S. P. SETHI

Next, we deal with the case A > A and R = 0. Equation (49) tells us V i = 1/2,/(A3 which
together with (14) and (22b) yields

=.(V“-&)>o
0
2p

Since the exploration region (part of A > A, R > 0) is defined by the condition W >0 (see (45) and
associated arguments), there is nothing to prove in this region.
We proceed to the no-exploration part of the region A > A , R>O, and begin by showing that
both V, and V, are continuous across R = R”(A).In fact, only the continuity of V, and the following
lemma need to be proved.

Lemma
If I/ is continuous in the whole quadrant and continuously differentiable on either side
of the curve R = R’(A), Rs’(A,) exists, and VR(AO, R , +) = VR(Ao,R , -), then,
b5(&,R , ) = VA(&, Ro - ).

Proof: We choose ( A , + A A , R , + A R ) on R = R”(A) and close to (Ao,Ro). Continuous

differentiability gives

and

V ( A o+ AA, R , ) + V(Ao,R,) - V(A0,R, + AR) - kfA0 + AA, R,)

+ A R-
= lim
AA-0 AA

= CVR(Ao,R,+)- V d 4 0 7 & - ) l R”(A0)

from which the result follows. This completes the proof.
To prove continuity of V,, we simply obtain the limiting governing differential equations above
and below R = R”(A)
 OPTIMAL EXPLORATION AND CONSUMPTION 21

and

Since W = 0 along R = Rs(A)and V(A,R +) = V ( A ,R -), it follows that VAA,R +) = VAA,R -).
In the no-exploration region, we have for ($4) approximately defined

V(A,R)’ = ((A) +-2PR

and, therefore
2V(A,R ) V A ( 4 R ) = (’(4
In particular, when R , = Rs(A,) we conclude 2V(A0,R , +) VA(Ao,R , +) = (’(A,) so that
w,,
R) VA(A0,R ) = w,,
R , +) V,(A,, R, + 1
for R > Rs(A,).
Finally, continuity of VRand V, across R = Rs((A)gives W(A,, R, +) = 0, and an adaptation of
the argument following (71) yields W(A,, R ) < 0 for R > RS(A,).

REFERENCES
1. Hotelling, H., ‘The economics of exhaustible resources’, Journal of Political Economy, 39, 137-175 (1931).
2. Arrow, K., ‘Optimal pricing, use and exploration of uncertain natural resource stocks’, presented at the Conference on
Natural Resource Pricing, Trail Lake, Wyoming, August 1977.
3. Arrow, K. and M. Kua,Public Investment, The Rate of Return, and Optimal Fiscal Policy, The Johns Hopkins Press,
Baltimore, Maryland, 1970.
4. Bell, D. J. and D. H. Jacobsen, Singular Optimal Control Problems, Academic Press, New York, 1975.
5. Sethi, S. P. and G. L. Thompson, Applied Optimal Control and Management Science, Martinus Nijhoff, Boston, 1981.
6. Spiegel, M., Mathematical Handbook, McGraw-Hill, New York, N.Y., 1968.
I. Webster, A. G., Partial Differential Equations of Mathematical Physics, Dover, New York, 1955.