Scribd
Upload
7 views2 pages

Effects of Kλ Operator on Functions

The document discusses composition operators of the form Kλ = eaxλ∂ and their effect on functions. It shows that: 1) Applying Kλ to the identity function x gives expressions involving factorials that indicate Kλ transforms x into x(1 - λaxλ)-1/λ. 2) This pattern holds more generally, with Kλ(f(x)) = f(x(1 - λaxλ)-1/λ), describing both shift and scale operators as special cases. 3) As λ gets very large, Kλ cuts off the values of f(x) to between -1 and 1, providing a sharpened sigmoid transformation.

Uploaded by

JFSP
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
7 views2 pages

Effects of Kλ Operator on Functions

The document discusses composition operators of the form Kλ = eaxλ∂ and their effect on functions. It shows that: 1) Applying Kλ to the identity function x gives expressions involving factorials that indicate Kλ transforms x into x(1 - λaxλ)-1/λ. 2) This pattern holds more generally, with Kλ(f(x)) = f(x(1 - λaxλ)-1/λ), describing both shift and scale operators as special cases. 3) As λ gets very large, Kλ cuts off the values of f(x) to between -1 and 1, providing a sharpened sigmoid transformation.

Uploaded by

JFSP
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

So far we have found expressions for composition operators parting from

their eect in an arbitrary function and from there deducing its form. We can
also go the other way, that is, start from a given exponential dierential operator
and see what kind of composition (if any) it produces.
A natural question to ask ourselves is what eect does the more general
K

= e
ax

operator have on a function, of which the shift operator T


h
= e
hx
0

and the scale operator S

= e
(log )x
1

can be regarded as special cases.


As our rst guess we can look for the action of this operator on the identity
function f(x) = x. If the operator is truly a consistent composition operator,
its action on the identity must give its dening function. For small values of ,
expanding the operator in a Taylor series we get:
K
2
(x) = x +ax
2
+a
2
x
3
+a
3
x
4
+... =
x
1 ax
K
3
(x) = x+ax
3
+
3
2
a
2
x
5
+
5
3
a
3
x
7
+... = x

n=0
(2n 1)!!
n!
a
n
x
2n+1
=
x

1 2ax
2
K
4
(x) = x+ax
4
+2a
2
x
7
+
5
3
a
3
x
1
0+... = x

n=0
(3n 2)!!!
n!
a
n
x
3n+1
=
x
(1 3ax
3
)
1
3
And so on (where n!! denotes the double factorial and n!!! the triple factorial
of n). We can start to see a pattern here.
One should note that there are issues of convergence: indeed, the radius
of convergence for K
2
(x) is
1
a
, and for K
3
(x) it is
1

2a
. However, analytically
continuing the function in the complex plane uniquely denes the function ev-
erywhere except at its singularities.
A more rigorous approach makes use of the ow method: we can model our
expression as a formal solution of a certain partial dierential equation in two
variables, and solving the equation gives the operators eect in the function
of initial conditions. More specically, we have the dierential equation of the
function in two variables g(x, t)
g
t
+D(g) = 0
where D is a dierential operator depending on x. The initial condition is
g(x, 0) = f(x)
The formal solution of this equation is g(t, x) = e
tD
f(x). Setting t = 0
and identifying D = ax

, we can nd the eect of K

in f just by solving
the equation.
We will solve it by the method of characteristics. Applying the chain rule
and separation of variables, we have
g
t
= ax

g
x
x
t
= ax

1
1
( 1)ax
1
= t +C
where C is a constant of integration. Setting = 1 for notational
simplicity, the equation of characteristics becomes
1
ax

t = C
This allows us to write g(x, t) in terms of a function h

1
ax

. We have
for t = 0:
h

1
ax

= f(x)
Solving for h(x) gives
h(x) = f

ax

Finally, setting t = 1:
h

1
ax

= f

a(
1
ax

1)

= f

1 ax

Our nal result is


K
+1
(f(x)) = f

1 ax

This remarkable result allows us to describe both the shift and scale operator
as particular cases of the K operator. To see that it gives the correct result, let
us compute the action of K
0
and K
1
explicitly:
K
1
(f(x)) = lim
0
+
f

1 ax

= lim
0
+
f(x(1 ax

) =
= lim
0
+
f(xe

(1ax

1)
) = lim
0
+
f(xe
ax

) = f(e
a
x)
K
0
(f(x)) = f

x
(1 (1)ax
1
)
1
1

= f

x
(1 +
a
x
)
1

= f(x +a)
In addition, let us compute the limit of K

for big values of :


K

(f(x)) = lim

x
(1 ax

)
1

f(1) if x < 1
f(x) if |x| < 1
f(1) if x > 1
If we plot the values of the argument of K

(f) with respect to x, we nd a


sharpened sigmoid graph that cuts all the values of x to a maximum absolute
value of 1).
2

You might also like