一、5 效用最大化与支出最小化问题的关系 及若干其它问题

5.2：大部分已经讲过，此外，Varian 8.4节 Comparative statics using FOC's 自己复习

4.4 支出函数与希克斯需求的关系 （MWG: 3.E, JR: 1.4.2, V: 7）
5.3 斯勒茨基方程，收入效应与替代效应 （MWG: 3.G, JR: 1.5.2, V: 8）
5.4 替代矩阵的性质 （MWG: 2.F, 3.G, JR: 1.5.2, V: 8）
3.5 间接效用函数与马歇尔需求的关系 （MWG: 3.G, JR: 1.4.1, V: 7）
大纲外：Duality between direct and indirect utility （MWG: Exercise 3.H.5, JR: 2.1.3,
V: 8.6）
(大纲外：MWG 3.I Welfare evaluation of economic changes自己学习）
6.3 显示偏好强（或一般）公理与消费者选择的理性化 5.1：对偶性 （MWG：3.F）

3.G Relationships between Demand, Indirect Utility, and Expenditure Functions

1. h(p, u) = ∇p e(p, u) (i.e., hℓ (p, u) = ∂e(p, u)/∂pℓ , ℓ = 1, · · · , L) (Proposition 3.G.1)

▶ Proof 2: (F.O.C. argument) Focus on interior case where h(p, u) ≫ 0.

By the chain rule, we have

∇p e(p, u) = ∇p [p · h(p, u)] = h(p, u) + [p · Dp h(p, u)]T

Using the F.O.C. for interior solution to the EMP, p = λ∇u(h(p, u)), we get

∇p e(p, u) = ∇p [p · h(p, u)]

= h(p, u) + λ[∇u(h(p, u)) · Dp h(p, u)]T .

Because the constraint u(h(p, u)) = u holds for all p in the EMP, we know that ∇u(h(p, u)) ·
Dp h(p, u) = 0, so we have the result.

▶ Proof 3: (Envelope Theorem Argument).

• Recall the Envelope Theorem (Mathematical Appendix L):

Consider the maximization problem

max f (x; q)
x∈RN
s.t. g1 (x; q) = b1
..
.

gm (x; q) = bM

1
 We can write the Lagrangian function as

X
M


L = f (x; q) + λm bm − gm (x; q)
m=1

Let x(q) be the maximizer function. Let v(q) be the value function, that is, the maximized value
of the objective function f (x; q) when the parameter vector is q, f (x(q), q).

Suppose v(q) is differentiable at q and x(q) is the maximizer solution at q. The Envelope Theorem
helps calculate the marginal effect of changes in q on the value v(q) :

∂f (x(q); q) X
M
∂v(q) ∂gm (x(q); q)
= − λm ,
∂qs ∂qs m=1
∂qs

∂v(q) ∂L(x(q); q)
or simply =
∂qs ∂qs

where L(x(q); q) is the Lagrangian evaluated at the optimal solution x(q) :

X
M


L(x(q); q) = f (x(q); q) + λm bm − gm (x(q); q) .
m=1

▷ The important thing to note is that all derivatives are partial derivatives, corresponding to
the direct effect to be discussed below.

▷ As long as the set of binding constraints does not change near q, the above result also applies
to maximization problems with inequality constraints.

▷ It is also apparent that the arguments also work for minimization problems.

• Now, we apply this theorem to the EMP. Recall that the Lagrangian function for the EMP is

L = p · x + λ · (u − u(x))

The parameters pℓ only appear in the objective function.

By the Envelope Theorem, we have

∂e(p, u) ∂L(h(p, u); p, u)

= = hℓ (p, u).
∂pℓ ∂pℓ

▶ The idea behind the two proofs above is the same and can be seen most clearly in Proof 2.

Recall the following expression from Proof 2:

∇p e(p, u) = ∇p [p · h(p, u)] = h(p, u) + [p · Dp h(p, u)]T

| {z } | {z }
direct effect indirect effect

2
 The total effect on expenditure of a marginal price change can be broken into two parts:

◦ direct effect on expenditure from changing prices holding demand fixed

◦ indirect effect on expenditure from the induced changes in demand holding prices fixed

Because we are already at an expenditure minimizing bundle, the first-order conditions for
the EMP imply that the latter effect is zero. That is, the induced changes in demand have
‘no first-order effect’ on the consumer’s expenditure.

▶ Given that h(p, u) = ∇p e(p, u), we therefore have the relation Dp h(p, u) = Dp2 e(p, u). (Proposition
3.G.2.(i))

• By the concavity of e(p, u), the L-by-L Hessian matrix Dp2 e(p, u) and hence the Jacobian
matrix of h(p, u), Dp h(p, u), are negative semidefinite (NSD) and symmetric. (Proposition
3.G.2.(ii),(iii))

▷ NSD of Dp h(p, u) is the differential analog of the compensated law of demand: dp·dh(p, u) ≤ 0
and dh(p, u) = Dp h(p, u)dp ⇒ dp · Dp h(p, u)dp ≤ 0 ⇒ Dp h(p, u) is NSD

▷ Samuelson (1947): symmetry of Dp h(p, u) is a property just beyond what one would derive
without the help of mathematics.

▷ In fact, it is closely related to the transitivity aspect of rational preferences. After we study
Slutsky equation below, we will see that the symmetry property is also an evidence of the fact
that preference-based approach imposes more structure on consumer demand than choice-
based approach (c.f. Proposition 2.F.2).

▶ By Euler’s Formula, HD0 of h(p, u) ⇒ Dp h(p, u)p = 0 (Proposition 3.G.2.(iv))

• That is, the matrix Dp h(p, u) does not have full rank.

• some definitions: two goods ℓ and k are

▷ substitutes at (p, u) if ∂hℓ (p, u)/∂pk ≥ 0

▷ complements at (p, u) if ∂hℓ (p, u)/∂pk ≤ 0

▷ gross substitutes at (p, w) if ∂xℓ (p, w)/∂pk ≥ 0

▷ gross complements at (p, w) if ∂xℓ (p, w)/∂pk ≤ 0

• The ith equation in Dp h(p, u)p = 0 writes

∂hi (p, u) ∂hi (p, u) ∂hi (p, u)

p1 + · · · + pi + · · · + pL = 0.
∂p1 ∂pi ∂pL

Since ∂hi (p, u)/∂pi ≤ 0 by the negative semidefiniteness of Dp h(p, u) and since all prices are

3
 positive, at least one of the other price derivatives must be non-negative, that is, ∂hi (p, u)/∂pj ≥
0 for at least one j. Therefore, each good has at least one Hicksian substitute.

2. Slutsky equation

▶ This equation relates the price derivative of the Hicksian demand function to that of the Walrasian
demand function. That is, it is about the relation between the slopes of these demand functions.

▶ It computes the derivative of the unobservable hℓ (p, v(p, u)) with respect to p (i.e., the slope of
the compensated demand curve) from information about the observable xℓ (p, w) (Figure 3.G.1)

▶ Recall that
hℓ (p, u) = xℓ (p, e(p, u)).

Differentiating with respect to pk and evaluating at (p, u) give

∂hℓ (p, u) ∂xℓ (p, e(p, u)) ∂xℓ (p, e(p, u)) ∂e(p, u)
= +
∂pk ∂pk ∂w ∂pk

• (Varian book, page 120:) Imagine multiplying both sides by dpk and think carefully about the
meaning of the resulting expression:

▷ The LHS is how the compensated demand hℓ (p, u) changes when pk changes, starting from
the point (p, u).

▷ The RHS says that this change is equal to the changes in demand holding expenditure fixed
at e(p, u) plus the changes in demand when income changes times how much income has to
change to keep utility constant.

Since ∂e(p, u)/∂pk = hk (p, u), we have

∂hℓ (p, u) ∂xℓ (p, e(p, u)) ∂xℓ (p, e(p, u))
= + hk (p, u)
∂pk ∂pk ∂w

Because the wealth level in Walrasian demand must be w = e(p, u) and hk (p, u) = xk (p, e(p, u))
= xk (p, w), we can rewrite the equation above as

∂hℓ (p, u) ∂xℓ (p, w) ∂xℓ (p, w)

= + xk (p, w)
∂pk ∂pk ∂w

This is the Slutsky equation.

▶ (Varian book, pages 120 to 121) The Slutsky equation decomposes the demand change induced
by a price change ∆pk into two separate effects: the substitution effect and the income effect:

∂xℓ (p, w) ∂hℓ (p, u) ∂xℓ (p, w)

∆xℓ ≈ ∆pk = ∆pk − xk (p, w)∆pk .
∂pk ∂pk | ∂w {z }
| {z }
substitution effect income effect

4
We can also consider the effects of all prices changing at the same time:

Dp x(p, w) = Dp h(p, u) − Dw x(p, w)xT .

In the two-good case, for example, these Slutsky equations look like this:

 ∂x (p, w) ∂x1 (p, w)   ∂h1 (p, u) ∂h1 (p, u)   ∂x1 (p, w) 

1
 ∂p ∂p2   ∂p1 ∂p2   ∂w 
 ∂x (p,1 w) −
∂x2 (p, w)   ∂h2 (p, u)  ∂x2 (p, w)  1
= [ x (p, w) x2 (p, w) ]
2 ∂h2 (p, u)
∂p1 ∂p2 ∂p1 ∂p2 ∂w

where u = v(p, w).

Expanding the last term gives

   
∂x1 (p, w) ∂x1 (p, w) ∂x1 (p, w)
   x1 (p, w) x2 (p, w) 
 ∂x ∂w
(p, w)  [ x1 (p, w) x2 (p, w) ] =  ∂w
∂x2 (p, w)
∂w
∂x2 (p, w) 
2
x1 (p, w) x2 (p, w)
∂w ∂w ∂w

Suppose we consider a price change ∆p = (∆p1 , ∆p2 ) and we are interested in the approximate
change in demand ∆x = (∆x1 , ∆x2 ). According to the Slutsky equation, we can calculate this
change as
 ∂h ∂h1    
  1  ∂x1 ∂x1
x2   ∆p 
∆x1  ∂p1
=  ∂h ∂p2  ∆p1 −  ∂w x1 ∂w 1
∆x2 2 ∂h2  ∆p2  ∂x
2 ∂x2  ∆p2
x1 x2
∂p1 ∂p2 ∂w ∂w

   
∆xs1 ∆xI1
= −
∆xs2 ∆xI2

• The first vector is the substitution effect and indicates how the Hicksian demands change. Since
changes in Hicksian demand keep utility constant, the vector (∆xs1 , ∆xs2 ) will be tangent to the
indifference curve.

• The second vector is the income effect. The price change has cuased ‘purchasing power’ to
change by x1 ∆p1 + x2 ∆p2 and the vector (∆xI1 , ∆xI2 ) measures the impact of this change on
demand, with prices held constant at the initial level. This vector therefore lies on the income
expansion path.

5
 • This Slutsky decomposition is illustrated in Figure 8.4 of Varian (1993):

Here, prices change from p to p′ , and demand changes from x to x′ . To construct the Hicks
decomposition, first pivot the budget line around the indifference curve to find the optimal
bundle at prices p′ with utility fixed at the original level. Then, shift the budget line out to x′
to find the income effect. The total effect is the sum of these two movements.

▶ Recall that the Slutsky substitution matrix S(p, w) = Dp x(p, w) + Dw x(p, w)x(p, w)T derived
under the choice-based approach in MWG Chapter 2 has the same expression as the right-hand
side of the Slutsky equation just derived under the preference-based approach. That is,

S(p, w) = Dp x(p, w) + Dw x(p, w)x(p, w)T .

By Proposition 3.G.2, the matrix Dp h(p, u) is symmetric and negative semidefinite and also
satisfies Dp h(p, u)p = 0. Therefore, the Slutsky substitution matrix S(p, w), if derived under the
preference-based approach, also has these properties.

This shows that the preference-based approach studied in this chapter imposes more restriction
on demand behaviour than the choice-based approach studied in Chapter 2:

• If x(p, w) merely satisfies the WARP (and HD0 and Walras’ Law), S(p, w) need not be symmetric
when L > 2. However, if x(p, w) is generated by preference maximization, S(p, w) is always
symmetric.

▶ Figure 3.G.1 shows xℓ (p, w) and hℓ (p, u).

• Which demand function has a larger slope (in absolute value)?

6
 ▷ Suppose only pℓ changes from pℓ and other prices are fixed.

▷ At pℓ = pℓ , xℓ (p, w) = hℓ (p, u).

▷ When pℓ increases above pℓ , we must increase the consumer’s wealth to keep him at the same
level of utility.

▷ If good ℓ is a normal (an inferior) good, its demand falls by more (less) in the absence of
wealth compensation.

▷ Thus, the Hicksian demand function is steeper (flatter) than the Walrasian demand function
for normal (inferior) goods.

◦ Here, being ‘steeper’ is relative to the horizontal, quantity axis. If you try to reconcile
the figure with the mathematical expression of the Slutsky equation, don’t forget that the
slopes ∂hℓ (p, u)/∂pk and ∂xℓ (p, w)/∂pk are measured relative to the vertical, price axis.

• Can you re-draw Figure 3.G.1 for a Giffen good?

▶ We know that Hicks wealth compensation and Slutsky wealth compensation are different. Then,
why do the Hicks compensated demand function and the Slutsky compensated demand function
have the same derivatives according to the Slutsky equation?

• Suppose the initial position is (p, w), with utility level u. Then prices change to p′ .

∆wSlutsky = p′ · x(p, w) − w, ∆wHicks = e(p′ , u) − w.

• ∆wHicks ≤ ∆wSlutsky . In general the inequality is strict for discrete price changes. Figure 3.G.2
illustrates.

• However, for marginal changes in price, these two wealth compensations are the same because

∇p e(p, u) = h(p, u) = x(p, w).

In words, the total effect of a marginal price change on expenditure is simply the direct effect of
such price changes, which assumes that the consumption bundle x(p, w) does not change. But
this is exactly the calculation done for Slutsky wealth compensation.

lim ∆wSlutsky = lim [p′ · x(p, w) − p · x(p, w)] = x(p, w)dp,

dp→0 dp→0 | {z }
w
′
lim ∆wHicks = lim [e(p , u) − e(p, u)] = ∇p e(p, u)dp = h(p, u)dp.
dp→0 dp→0 | {z }
w

▶ (Varian book, page 121, slightly extended) Example: Slutsky equation associated with the Cobb-
Douglas utility function u(x1 , x2 ) = Axα 1−α
1 x2 , where A = 1/[αα (1 − α)1−α ].

7
 From the results derived before, we can get
−(1−α)
v(p1 , p2 , w) = p−α1 p2 w e(p1 , p2 , u) = pα 1−α
1 p2 u
αw (1−α)w
x1 (p1 , p2 , w) = p1 x2 (p1 , p2 , w) = p2
−α
h1 (p1 , p2 , u) = αpα−1
1 p1−α
2 u h2 (p1 , p2 , u) = (1 − α)pα1 p2 u

Thus, focusing on commodity 1, we have

∂x1 (p, w) αw ∂x1 (p, w) ∂x1 (p, w) α

=− 2 =0 =
∂p1 p1 ∂p2 ∂w p1

∂h1 (p, u) ∂h1 (p, v(p, w)) −(1−α)

= α(α − 1)pα−2
1 p1−α
2 u ⇒ = α(α − 1)pα−2
1 p1−α
2 [p−α
1 p2 w]
∂p1 ∂p1
= α(α − 1)p−2
1 w

∂h1 (p, u) ∂h1 (p, v(p, w)) −α −(1−α)

= α(1 − α)pα−1
1 p−α
2 u ⇒ = α(1 − α)pα−1
1 p−α
2 [p1 p2 w]
∂p2 ∂p2
= α(1 − α)p−1 −1
1 p2 w

Plugging these into the Slutsky equation gives

∂h1 (p, v(p, w)) ∂x1 (p, w) α(α − 1)w α αw

− x1 (p, w) = 2 −
∂p1 ∂w p1 p1 p1
[α(α − 1) − α ]w
2
=
p21
−αw ∂x1 (p, w)
= 2 =
p1 ∂p1

∂h1 (p, v(p, w)) ∂x1 (p, w) α(1 − α)w α (1 − α)w

− x2 (p, w) = −
∂p2 ∂w p1 p2 p1 p2
∂x1 (p, w)
=0= (∗)
∂p2

We see from equation (∗) that the positive substitution effect of a change in p2 on the (Hicksian)
demand for good 1 is exactly offset by its negative income effect, producing a zero total effect on
(Walrasian) demand. This explains why the Walrasian demand functions are independent of the
price of the other good. (Draw a figure to illustrate the vertical/horizontal price expansion paths
and the Slutsky decomposition.)

▶ (1) When utility function is strictly increasing and quasiconcave and is defined on two goods only,
they must be net substitutes. When there are more than two goods, two goods can either be net
substitutes for each other or be net complements of each other.

8
 (2) However, even with two goods, it is possible that one good is a gross substitute for the other,
while the second good is a gross complement to the first one.

Those who are interested in the second point above may want to read the following articles: Kris
De Jaegher. ‘Asymmetric substitutability: theory and some applications.’ Tjalling C. Koopmans
Research Institute Discussion Paper Series no. 08-02; Jensen, Robert, and Nolan Miller. ‘Giffen
Behavior and Subsistence Consumption.’ American Economic Review 98, no. 4 (2008): 1553-77.
The first shows that, among other possible causes, if good 1 in a two-good model is a Giffen good,
then good 1 is a gross substitute for good 2 while good 2 is a gross complement of good 1. The
second paper gives the first real-world empirical evidence of the existence of a Giffen good.

3. Roy’s identity

▶ We know that e(p, u) is the value function of the EMP and that v(p, w) is the value function
of the UMP. Given that, in the EMP, ∇p e(p, u) = h(p, u), do we also have that, in the UMP,
−∇p v(p, w) = x(p, w)?

• No. Walrasian demand x(p, w) is an ordinal concept and is invariant to increasing transforma-
tions of u(x), but v(p, w) is not invariant to increasing transformations of u(x) :

• For example, suppose that the solution to maxx u(x) s.t. p · x ≤ w is x(p, w) and that the value
function is v(p, w). Let f (·) be strictly increasing. Then the solution to maxx f [u(x)] s.t. p·x ≤ w
is still x(p, w) but the value function becomes f [v(p, w)]. Usually the derivatives of v(p, w) with
respect to p are not the same as those of f [v(p, w)].

• But with a small correction to the wrong equation ‘−∇p v(p, w) = x(p, w)’, we will get the right
one.

▷ The correction is to normalize the derivative of v(p, w) with respect to p by the marginal
utility of wealth.

▶ Roy’s Identity
∇p v(p, w)
x(p, w) = −
∇w v(p, w)

• Proof 1: We know from Proposition 3.E.1 that, for all p,

v(p, e(p, u)) = u

Differentiating with respect to p and evaluating at p = p lead to

∂v(p, e(p, u))

∇p v(p, e(p, u)) + ∇p e(p, u) = 0
∂w

9
 By Proposition 3.G.1, ∇p e(p, u) = h(p, u). So

∂v(p, e(p, u))

∇p v(p, e(p, u)) + h(p, u) = 0
∂w

Since w = e(p, u) and h(p, u) = x(p, w), we have

∂v(p, w)
∇p v(p, w) + x(p, w) = 0
∂w

Hence the result.

• Proof 2 (first-order condition argument): Assume that x(p, w) is differentiable and x(p, w) ≫ 0.
By the chain rule, we have

∂v(p, w) X ∂u(x(p, w)) ∂xk (p, w)

L
=
∂pℓ ∂xk ∂pℓ
k=1

Substituting for ∂u(x(p, w))/∂xk using the first-order condition for the UMP gives

∂v(p, w) X
L
∂xk (p, w)
= λpk
∂pℓ ∂pℓ
k=1

PL
By Proposition 2.E.2, k=1 pk ∂xk∂p
(p,w)
ℓ
= −xℓ (p, w). So we have

∂v(p, w)
= −λxℓ (p, w).
∂pℓ

In section 3.D, we already showed that λ = ∂v(p, w)/∂w. Using this, we get the result.

• Proof 3 (Envelope Theorem argument):

L = u(x) + λ(w − p · x)

By the Envelope Theorem, we have

∂v(p, w)
= −λxℓ (p, w).
∂pℓ

Again by the Envelope Theorem, we have

∂v(p, w)
= λ.
∂w

So we have the result.

• The essence of the last two proofs is explained in the textbook in the paragraph between proof
1 and proof 2: Because we are already at an optimum, the demand responses to a price change
can be ignored when considering the effect of marginal changes in price on the value function.

10
 4. Some Examples

4.1 Suppose an individual consumes two commodities and has an indirect utility function

w2
v(p1 , p2 , w) =
p1 p2

(a) Verify that this indirect utility function has all the properties listed in Proposition 3.D.3.

(Do this part by yourself.)

(b) Find the individual’s utility function.

▶ We will discuss two methods below. The first one is quicker but relies on experience and luck to
some extent. The second one is slower but always works.

▶ Here is the first method. Using Roy’s Identity, we can get

2

∂v(p1 , p2 , w)/∂p1 − wp2

p21
1
w
x1 (p1 , p2 , w) = − =− w = .
∂v(p1 , p2 , w)/∂w 2 p1 p2 2p1

Similarly,
w
x2 (p1 , p2 , w) = .
2p2
Thus, we have
v(p1 , p2 , w) = 4x1 (p1 , p2 , w)x2 (p1 , p2 , w).

Since
u(x1 (p1 , p2 , w), x2 (p1 , p2 , w)) = v(p1 , p2 , w),

we have
u(x1 , x2 ) = 4x1 x2 .

• If we were given v(p, w) = w2 /(p1 p2 ) + 2, then u(x1 , x2 ) = 4x1 x2 + 2.

▶ Section 8.6, V, subsection 2.1.3, JR and Exercise 3.H.5, MWG describe the second method.

• Here, it is convenient to normalize the indirect utility function first. We know that v(p, w) is
homogeneous of degree zero, so we have
 
p̃
v(p̃, w) = v , 1 ≡ v(p).
w

This normalized indirect utility function is the value function of the following maximization
problem:
v(p) = max u(x) s.t. p · x = 1
x

11
 It turns out that if we are given the indirect utility function v(p), we can find the direct utility
function by solving the following minimization problem:

u(x) = min v(p) s.t. p · x = 1

p

• The reason is as follows. Let x be the demanded bundle at prices p. Then by definition
v(p) = u(x). Let p′ be any other price vector that satisfies the budget constraint p′ · x = 1. That
is, the optimal bundle x under p is just affordable under p′ (and the normalized wealth level of
1). Then, the new utility-maximizing choice under p′ must bring utility at least as great as the
utility yielded by x; that is, v(p′ ) ≥ u(x) = v(p). Hence, the minimum of the indirect utility
function over all p that satisfy the budget constraint gives us the utility of x. This is illustrated
by Figure 8.5 of Varian:

• Example: Given v(p1 , p2 ) = −a ln p1 − b ln p2 , find the direct utility function.

Set up the minimization problem:

min −a ln p1 − b ln p2
{p1 ,p2 }

s.t. p1 x1 + p2 x2 = 1

The first-order conditions are

a
− = λx1
p1
b
− = λx2
p2
p 1 x 1 + p2 x 2 = 1

12
 From these, we can solve for the optimal prices
a b
p1 = p2 =
(a + b)x1 (a + b)x2
The direct utility function is the value function of the minimization problem above. So, to get
it, we substitute the optimal prices into the objective function. Then, we have
 
a b a b
u(x1 , x2 ) = −a ln − b ln = a ln x1 + b ln x2 − a ln + b ln
(a + b)x1 (a + b)x2 (a + b) (a + b)
This is a Cobb-Douglas utility function after an increasing transformation.

4.2 We can also find u(·) and v(·) if what we know is the expenditure function e(p, u). Suppose
√
e(p1 , p2 , u) = p1 p2 u. Find the utility function.

There are two ways to go about this. With the first, we try to find v(p1 , p2 , w) first, and then use
one of the two methods in the previous example to find u(x1 , x2 ). In the second way, we calculate
Hicksian demand first and see if it contains enough information to allows us to derive the utility
function.

(a) Find the indirect utility function first.

▶ By Proposition 3.E.1 (Equation 3.E.1), we can get the indirect utility function by directly inverting
the expenditure function. From the expenditure function, we have
e2
u=
p1 p2
Rewriting, we have
w2
v(p1 , p2 , w) = .
p1 p 2
(b) Find Hicksian demand functions first.

▶ By Proposition 3.G.1, we have

r
∂e(p1 , p2 , u) 1 p2 u
h1 (p1 , p2 , u) = =
∂p1 2 p1
r
∂e(p1 , p2 , u) 1 p1 u
h2 (p1 , p2 , u) = =
∂p2 2 p2
Therefore,
u
h1 (p1 , p2 , u)h2 (p1 , p2 , u) =
4
So
v(p1 , p2 , w)
h1 (p1 , p2 , v(p1 , p2 , w))h2 (p1 , p2 , v(p1 , p2 , w)) =
4
Thus,
u(x1 , x2 ) = 4x1 x2 .

Recommended Exercises: 3.G: 1, 2, 3, 4(c), 4(d), 6(c), 6(d), 6(e), 7, 8, 9, 10∼17

13