QABE Lecture 16
Multivariable Calculus: The Partial Derivative
UNSW Economics
c
UNSW Economics
Agenda
1 Functions of two-variables – the third dimension;
2 Analysing slope in 3D – the partial derivative;
3 Applications of the partial derivative;
4 Further analysis and methods.
c
UNSW Economics
The third-dimension...
• So far, we have dealt (in calculus) with functions of the form,
f ( x) , f ′ ( x) , f ′′ (x)
• Common to them all is the single independent variable (x);
• What if we have a function of the form,
???
f (x1 , x2 ) f ′ (x1 , x2 )
• ... What does this kind of function look like?
• ... Is there such a thing as the derivative?
• (We need to be careful... what does f ′ (·) actually mean in this
context??)
c
UNSW Economics
Functions of two variables: an example
z
• Suppose we have some (x0 , y0 , f (x0 , y0 ))
function, z = f (x, y)
• Then a point will be
(x0 , y0 , z0 ) =
(x0 , y0 , f (x0 , y0 )),
• Or, a plane will be b
constructed by fixing
one axis value:
1 XZ plane: fix y = y0 ; x
2 XY plane: fix z = z0 ; y
3 YZ plane: fix x = x0 ;
c
UNSW Economics
Functions of two variables: an example
z
• Suppose we have some
function, z = f (x, y)
• Then a point will be
(x0 , y0 , z0 ) =
la ne
(x0 , y0 , f (x0 , y0 )), xz p
• Or, a plane will be b
constructed by fixing
one axis value:
1 XZ plane: fix y = y0 ; x
2 XY plane: fix z = z0 ; y
3 YZ plane: fix x = x0 ;
c
UNSW Economics
Functions of two variables: an example
z
• Suppose we have some
function, z = f (x, y)
• Then a point will be
(x0 , y0 , z0 ) =
(x0 , y0 , f (x0 , y0 )),
• Or, a plane will be b
la ne
constructed by fixing xy p
one axis value:
1 XZ plane: fix y = y0 ; x
2 XY plane: fix z = z0 ; y
3 YZ plane: fix x = x0 ;
c
UNSW Economics
Functions of two variables: an example
z
• Suppose we have some
function, z = f (x, y)
• Then a point will be yz
(x0 , y0 , z0 ) = pl
(x0 , y0 , f (x0 , y0 )), a ne
• Or, a plane will be b
constructed by fixing
one axis value:
1 XZ plane: fix y = y0 ; x
2 XY plane: fix z = z0 ; y
3 YZ plane: fix x = x0 ;
c
UNSW Economics
What is the slope???
• Can we get some kind of expression for slope in our new function
of two variable?
• What would it look like?
• Recall: slope is normally expressed as,
df
d...
... that is, we differentiate with respect to a variable – we are
calculating the change in the function value, with respect to a
change in an input variable...
• In our new framework, we do just the same thing...
c
UNSW Economics
Definition: Partial Derivative
For some function f (x, y), the PARTIAL DERIVATIVE of f with respect to
x, expressed as,
∂f
= fx (1)
∂x
and for the point (x0 , y0 ) is given by the limit,
f (x0 + h, y0 ) − f (x0 , y0 )
fx (x0 , y0 ) = lim . (2)
h→0 h
The partial derivative with respect to y is given similarly
∂f
= fy (3)
∂y
and found in the same way.
c
UNSW Economics
Back to the plot
• Consider our function z
again, z = f (x, y),
• The partial derivative of
f with respect to x at the
point (x0 , y0 ) will be the
, y0 )
slope of f in the xz fx (x0
b
plane.
• Similarly, the partial
derivative of f with x
respect to y at the point
y
(x0 , y0 ) will be the slope
of f in the yz plane.
c
UNSW Economics
Back to the plot
• Consider our function z
again, z = f (x, y),
• The partial derivative of
f with respect to x at the
fy (
point (x0 , y0 ) will be the x
0, y
slope of f in the xz 0)
b
plane.
• Similarly, the partial
derivative of f with x
respect to y at the point
y
(x0 , y0 ) will be the slope
of f in the yz plane.
c
UNSW Economics
What does fx tell us?
• The partial derivative, say fx , tells us the answer to the question,
By how much does f vary, for a given change in x all else remaining
the same.
• That is, we are analysing the slope of f in one dimension only,
keeping other dimensions fixed.
• This helps to untangle the influences of the many variables on the
function f .
c
UNSW Economics
Example
Given y = f (x1 , x2 ) = 3x21 + x1 x2 + 4x22 , find fx1 and fx2 .
To find fx1 , we remember to treat x2 as constant,
∂y
= fx1 = 6x1 + x2
∂x1
and then to find fx2 we remember to treat x1 as constant,
∂y
= fx2 = x1 + 8x2
∂x2
So the partial derivatives of f are fx1 = 6x1 + x2 and fx2 = x1 + 8x2 .
c
UNSW Economics
Example
Find fx and fy given that f (x, y) = (2x + 4)(y − 3).
Using the product rule, treating y as a constant, we have:
fx = (2x + 4)(0) + (2)(y − 3)
= 2(y − 3)
And for fy we have,
fy = (2x + 4)(1) + (0)(y − 3)
= 2x + 4
c
UNSW Economics
Higher order partial derivatives?
• Recall that in differentiation, with one variable, we could
calculate,
df d2 f d3 f
... 2
... ...
dx dx dx3
• ... Which was useful to determine the ‘rate of change of the
derivative of f ’ – in other words, the ‘acceleration’ or
‘deceleration’ of the change in f ;
• Can we do a similar thing with partial derivatives? What would it
mean?
c
UNSW Economics
• Suppose we took the partial derivatives of a function f (x, y),
fx and fy
∂(fx ) δ(fx ) ∂(fy ) δ(fy )
∂x ∂y and ∂y ∂x
• Now suppose we took the partial derivatives of each of these.
Caution!
We get FOUR different partial derivatives:
fxx fxy fyy fyx
c
UNSW Economics
• What do each of the FOUR partial derivatives mean??
• What does fxx tell us?
• fxx tells us the rate of change of fx with respect to x; or
• fxx tells us by how much the gradient (with respect to x) of f
changes, as x is varied.
• What about fxy ? (a CROSS PARTIAL DERIVATIVE or MIXED PARTIAL
DERIVATIVE)
• fxy tells us the rate of change of fx with respect to y; or
• fxy tells us by how much the gradient (with respect to x) of f
changes, as y is varied.
c
UNSW Economics
Definition: The Jacobian Matrix
For some function f (x, y), the J ACOBIAN M ATRIX, J, is a matrix of all
the first-order partial derivatives. That is:
h i
J = ∂f (∂xx,y) ∂f (∂yx,y)
More generally, the J ACOBIAN MATRIX for any function,
f (x1 , x2 , · · · , xn ), is given by:
h i
J = ∂f∂x(·) ∂f∂x(·) · · · ∂f∂x(·)
1 2 n
c
UNSW Economics
Definition: The Hessian Matrix
For some function f (x, y), the H ESSIAN M ATRIX, H, is a matrix of all
the second-order partial derivatives
2
∂ f (x,y) ∂2 f (x,y)
H= ∂x2 ∂x∂y
∂f (x,y) ∂f (x,y)
∂y∂x ∂y2
More generally, the H ESSIAN MATRIX for any function,
f (x1 , x2 , · · · , xn ), is given by:
2
∂2 f (·) ∂2 f (·)
∂ f (·)
···
2∂x21 ∂x1 ∂x2 ∂x1 ∂xn
∂ f (·) ∂2 f (·) ∂2 f (·)
∂x2 ∂x1 ∂x22
··· ∂x2 ∂xn
H=
.. .. ..
..
. . . .
∂2 f (·) ∂2 f (·) ∂2 f (·)
∂xn ∂x1 ∂xn ∂x2 ··· ∂x2n
c
UNSW Economics
Higher derivatives: fyx graphically
z
• We now know about fy –
the partial derivative of
f , keeping x constant;
• As we take the partial
derivative w.r.t. x of fy , b
we get fyx : it asks, how
does the gradient of f
with respect to y change x
as x is varied? (green
y
arrow)
c
UNSW Economics
Higher derivatives: fyx graphically
z
• We now know about fy –
the partial derivative of
f , keeping x constant;
• As we take the partial
b
derivative w.r.t. x of fy , b
we get fyx : it asks, how
does the gradient of f
with respect to y change x
as x is varied? (green
y
arrow)
c
UNSW Economics
Higher derivatives: fyx graphically
z
• We now know about fy –
the partial derivative of
f , keeping x constant;
• As we take the partial
b
derivative w.r.t. x of fy , b
b
we get fyx : it asks, how
does the gradient of f
with respect to y change x
as x is varied? (green
y
arrow)
c
UNSW Economics
Example
In the graphical example above, the function is f (x, y) = 2 − (1 − x)2 − (1 − y)2 . Show
that fxy = fyx = 0. That is, that the partial derivatives fx and fy do not change as the
‘other’ variable is changed.
We have,
fx = (−1)(2)(−1)(1 − x) = 2(1 − x) ,
fy = (−1)(2)(−1)(1 − y) = 2(1 − y) ,
Taking the cross-partial derivatives,
∂fx
= 0
∂y
∂fy
= 0
∂x
... as required.
c
UNSW Economics
Example
Find the second-order partial derivatives of the function z = f (x, y) = x3 e−2y .
The first-order partial derivatives,
fx = 3x2 e−2y
fy = −2x3 e−2y
And the second-order derivatives:
fxx = 6xe−2y
fyy = (−2)(−2x3 )e−2y = 4x3 e−2y
fxy = (−2)(3x2 )e−2y = −6x2 e−2y
fyx = (3)(−2x2 )e−2y = −6x2 e−2y
Note: fxy = fyx (again!).
c
UNSW Economics
Caution!
Don’t be surprised if you keep finding that
fxy = fyx
... a theorem (Young’s theorem) tells us that so long as the two
cross-partial derivatives are continuous, the order in which we
calculate the cross-partial derivative doesn’t matter. ... In this course,
we’ll only deal with this kind of function.
c
UNSW Economics
Further methods in partial differentiation
• Suppose that instead of the situation we have been dealing with
z = f (x, y):
f x
z
y
f
• We have the following:
f g
x
z t
y
f h
where z = f (x, y) but also, x = g(t) and y = h(t);
• Here, z is affected by t indirectly, but can we get an expression for
∂z
∂t ?
c
UNSW Economics
Definition: Chain-rule
Given a function z = f (x, y) where x and y are both functions of
another variable (or variables), e.g. x = g(t), and y = h(t), then if the
functions f , g and h have continuous partial derivatives, then we have,
dz ∂z ∂x ∂z ∂y
= + (4)
dt ∂x ∂t ∂y ∂t
or,
dz ∂x ∂y
= fx + fy . (5)
dt ∂t ∂t
c
UNSW Economics
Example (Chain rule)
dz x +y
Find dt given that z(x, y) = 2x2
and x(t) = 2t2 + 1 and y(t) = 3t.
So we have,
dz dx dy
= fx + fy
dt dt dt
the partial derivatives are:
−(x + 2y) 1
fx = fy =
2x3 2x2
and the secondary functions give,
dx dy
= 4t =3
dt dt
so we have,
dz −(x + 2y) 1
= (4t) + 2 (3)
dt 2x3 2x
c
UNSW Economics
Getting all the change together...
• In practice, a special application of the chain rule gives us a useful piece of
information;
• We ask the question, ‘by how much does z change, given an infinitesimal
change in x or y?’ (given that x and y are a function of something else)
• We can see that (by the chain rule) we have
dz ∂z dx ∂z dy
= +
dt ∂x dt ∂y dt
as a first approximation, but we notice that we can take out the common
1
dt factor, to be left with,
dz = fx dx + fy dy
• This is called the TOTAL DIFFERENTIAL of the function z and the process
by which we got there is TOTAL DIFFERENTIATION. (note, the ‘other
variable’, here t has disappeared)
c
UNSW Economics
Definition: The Total Differential
Given a two variable function z = f (x, y), then the total change in z for
infinitesimal changes in x and y is given by,
dz = fx dx + fy dy , (6)
and is known as the TOTAL DIFFERENTIAL. In general for any number
of variables z = f (x1 , . . . , xn ) the total derivative is given by,
dz = fx1 dx1 + fx2 dx2 + · · · + fxn dxn . (7)
c
UNSW Economics
Example (Total differentiation)
Find the total differential of z = 3x2 + xy − 2y3 .
First we find the partial derivatives of z:
fx = 6x + y
fy = x − 6y2
then we can write down (by the chain rule) the total differential:
dz = (6x + y) dx + (x − 6y2 ) dy
c
UNSW Economics
Total differential: reprise
• So what can we see in the total differential:
dz = fx dx + fy dy
• We have a summation of the changes to z that occur for a given
infinitesimal change in x or y;
• And fx and fy are acting as ‘converters’, to make sure that the tiny
changes in x and y will affect z in the correct quantities.
c
UNSW Economics