Genetic Algorithms, Help?

I'm doing a program which uses genetic algorithms at its core. It is supposed to find an exponential fit of a rather special form: a^(k^(bx)).
My generations have 100 individuals and I use 20 points. Which would be the best way to recombine my population? My teacher told me that I should try to use, say, 20% from the top and 20% bottom. Now, I'm not sure about two things. Hope you guys can help me.

Is the new generation forcefully comprised of completely new individuals? How should I recombine the numbers (a and b, since k is a constant)?
The numbers are doubles all around and each curve has a "fitness" associated to it. 

Bonus joke:
A Chinaman and a Jew are in a racist argument when the Jew rears back and punches the Chinaman in the nose. The Chinaman says, "What was that for?"
The Jew responds, "That was for Pearl Harbor you son of a bitch."
The Chinaman looks confused and says, "Pearl Harbor was bombed by the Japanese. I'm Chinese!"
So the Jew says, "Japanese. Chinese. What's the difference?!" Then the Chinaman rears back and punches the Jew in the nose.
The Jew says, "What was that for?"
The Chinaman responds, "That was for the Titanic!"
The Jew looks confused and says, "The Titanic?? The Titanic was sunk by an iceberg!"
So the Chinaman says, "Iceberg. Goldberg. What's the difference?!"


$$$$$$$$$$$$$$$$$L   .$$$$$$$$$$$$$$$u
$$$$$$$$$$$$$$$$$$N.@$$$$$$$$$$$$$$$$ *
$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ '>.n=L
$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$RR$$$$ 'b"  9
$$$$$$$$$$$$$$$$$$$$$$$$$$$R#"  .$$$$ @   .*
$$$$$$$$$$$$$$$$$$$$$$$$`   .e$$$$$$$P   e"
$$$$$$$$$$$$$$$$$$$$$R#    o$$$$$$$$P   @
$$$$$$$$$$$$$$$$$$$P" .e> 4$" '$$$$F  .F
$$$$$$$$$$$$$$$$$R  .$$$& '$   $$$$  .#>
$$$$$$$$$$$$$$$$$b.o$$$$$  $N  "$$" ."'>
$$$$$$$$$$$$$$$$$$$$$$$$$  $$N  "` .$ '>
$$$$$$$$$$$$$$ "$$$$$$$$R  $$$&    $$ '>
$$$$$$$$$$$$$$  E"$$P`9$E  $$$$   8$$ '>
$$$$$$$$$$$$$$  E  "  9$F  $$$$k .$$$ '>
$$$$$$$$$$$$$$  E     9$N  $$$$$$$$$$ '>
$$$$$$$$$$$$$$  E     9$$.u$$$$$$$$$$ '>
$$$$$$$$$$$$$$ o"     9$$$$$$$$$$$$$$ d
**************#       ***************"




$$$$$$$$$$$$$$$$$L   .$$$$$$$$$$$$$$$u
$$$$$$$$$$$$$$$$$$N.@$$$$$$$$$$$$$$$$ *
$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ '>.n=L
$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$RR$$$$ 'b"  9
$$$$$$$$$$$$$$$$$$$$$$$$$$$R#"  .$$$$ @   .*
$$$$$$$$$$$$$$$$$$$$$$$$`   .e$$$$$$$P   e"
$$$$$$$$$$$$$$$$$$$$$R#    o$$$$$$$$P   @
$$$$$$$$$$$$$$$$$$$P" .e> 4$" '$$$$F  .F
$$$$$$$$$$$$$$$$$R  .$$$& '$   $$$$  .#>
$$$$$$$$$$$$$$$$$b.o$$$$$  $N  "$$" ."'>
$$$$$$$$$$$$$$$$$$$$$$$$$  $$N  "` .$ '>
$$$$$$$$$$$$$$ "$$$$$$$$R  $$$&    $$ '>
$$$$$$$$$$$$$$  E"$$P`9$E  $$$$   8$$ '>
$$$$$$$$$$$$$$  E  "  9$F  $$$$k .$$$ '>
$$$$$$$$$$$$$$  E     9$N  $$$$$$$$$$ '>
$$$$$$$$$$$$$$  E     9$$.u$$$$$$$$$$ '>
$$$$$$$$$$$$$$ o"     9$$$$$$$$$$$$$$ d
**************#       ***************"




$$$$$$$$$$$$$$$$$L   .$$$$$$$$$$$$$$$u
$$$$$$$$$$$$$$$$$$N.@$$$$$$$$$$$$$$$$ *
$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ '>.n=L
$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$RR$$$$ 'b"  9
$$$$$$$$$$$$$$$$$$$$$$$$$$$R#"  .$$$$ @   .*
$$$$$$$$$$$$$$$$$$$$$$$$`   .e$$$$$$$P   e"
$$$$$$$$$$$$$$$$$$$$$R#    o$$$$$$$$P   @
$$$$$$$$$$$$$$$$$$$P" .e> 4$" '$$$$F  .F
$$$$$$$$$$$$$$$$$R  .$$$& '$   $$$$  .#>
$$$$$$$$$$$$$$$$$b.o$$$$$  $N  "$$" ."'>
$$$$$$$$$$$$$$$$$$$$$$$$$  $$N  "` .$ '>
$$$$$$$$$$$$$$ "$$$$$$$$R  $$$&    $$ '>
$$$$$$$$$$$$$$  E"$$P`9$E  $$$$   8$$ '>
$$$$$$$$$$$$$$  E  "  9$F  $$$$k .$$$ '>
$$$$$$$$$$$$$$  E     9$N  $$$$$$$$$$ '>
$$$$$$$$$$$$$$  E     9$$.u$$$$$$$$$$ '>
$$$$$$$$$$$$$$ o"     9$$$$$$$$$$$$$$ d
**************#       ***************"




$$$$$$$$$$$$$$$$$L   .$$$$$$$$$$$$$$$u
$$$$$$$$$$$$$$$$$$N.@$$$$$$$$$$$$$$$$ *
$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ '>.n=L
$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$RR$$$$ 'b"  9
$$$$$$$$$$$$$$$$$$$$$$$$$$$R#"  .$$$$ @   .*
$$$$$$$$$$$$$$$$$$$$$$$$`   .e$$$$$$$P   e"
$$$$$$$$$$$$$$$$$$$$$R#    o$$$$$$$$P   @
$$$$$$$$$$$$$$$$$$$P" .e> 4$" '$$$$F  .F
$$$$$$$$$$$$$$$$$R  .$$$& '$   $$$$  .#>
$$$$$$$$$$$$$$$$$b.o$$$$$  $N  "$$" ."'>
$$$$$$$$$$$$$$$$$$$$$$$$$  $$N  "` .$ '>
$$$$$$$$$$$$$$ "$$$$$$$$R  $$$&    $$ '>
$$$$$$$$$$$$$$  E"$$P`9$E  $$$$   8$$ '>
$$$$$$$$$$$$$$  E  "  9$F  $$$$k .$$$ '>
$$$$$$$$$$$$$$  E     9$N  $$$$$$$$$$ '>
$$$$$$$$$$$$$$  E     9$$.u$$$$$$$$$$ '>
$$$$$$$$$$$$$$ o"     9$$$$$$$$$$$$$$ d
**************#       ***************"

https://plus.google.com/114096099683887759200/posts/9AFqWmGL1qM

Shalom!~

>>3
Heil!

>>4
Shalom, Baruch Hyatt!

>>5
SIDF plz go

Vorsicht vor Dipl.-Psych. Erika Triller in Remscheid

>>1

First if your problem has dimension 2 and your fitness function is cheap you'd better use simple hill-climbing from a lot random starting points.

Is the new generation forcefully comprised of completely new individuals? How should I recombine the numbers (a and b, since k is a constant)?

You'd better have "elitism" and  keep between 0-50% of your current population at each generation.
New members will be "children" of existing solutions. The more a solution is successful, the more it should have children (google roulette wheel selection).

I suggest to recombine the genes with a RNG: either father's A, either mother's A (same for B). Do not blend factors because it may break a good solution.

This seems like an incorrect usage of GA, try a neural net instead.

>>8
What do you mean by father's A? Just a chromosome?

>>10
yeah.