eldorado@EN.ECN.PURDUE.EDU (David D Jansen) (02/08/90)
Keywords: mandelbrot
Could someone with fractal experience tell me why I get weird data. It gets
the basic shape but not quite right. A friend of mine (a fractal "guru")
can't make out what the problem. Here is my type declaration:
type
imaginary = record {an imaginary number i.e. a + bi}
a : real;
b : real;
end;
region = record {a rectangle on the imaginary plane}
amin : real;
bmin : real;
amax : real;
bmax : real;
end;
screen = record {the data that describes a screen}
where : region;
bitmap : packed array [1..hor,1..ver] of boolean;
end;
procedure plotset (var pages:screen;var fraccalc:boolean);
{a procedure to put the fractal from the sent region into the variable page}
var
dval,z,val:imaginary;
h,v,n:integer;
begin
dval.a:=(pages.where.amax - pages.where.amin)/(hor - 1);
dval.b:=(pages.where.bmax - pages.where.bmin)/(ver - 1);
for v:=1 to ver do
begin
for h:=1 to hor do
begin
val.a:=pages.where.amin + (h - 1) * dval.a; {val is the value of the}
val.b:=pages.where.bmin + (v - 1) * dval.b; {current point on the screen}
n:=0;
z.a:=0;
z.b:=0;
while ((z.a*z.a) + (z.b*z.b) <= 4) and (n < maxiter) do
begin
z.a:=(z.a * z.a) - (z.b * z.b) + val.a;
z.b:=2 * z.a * z.b + val.b;
n:=n + 1;
end;
if (((z.a * z.a) + (z.b * z.b)) <= 4) then
pages.bitmap[h,v]:=true
end;
writeln (v/ver*100:5:2,'% done');
end;
fraccalc:=true;
end;
Someone last year uploaded a mandelbrot generator and I modified his code to
test in my program. No go, it was worse than my attempt. I don't think I
tested my code in his program. That was a mistake on my part. I have had
this problem for more than a year. There could be a reward for the correct
answer.
While I am at it, what is Applied Engineering's phone number? Did anyone
download the Prodos Hyper C that was uploaded? Did it work?
Dave Jansen (The Gilded One)
eldorado@en.ecn.purdue.edu
"We have to make all our programs idiot-proof. Idiots are very intelligent!"BRL102@psuvm.psu.edu (Ben Liblit) (02/08/90)
Newsgroups: comp.sys.apple Subject: Re: AE, Mandelbrot generator References: <9002071850.AA29665@en.ecn.purdue.edu> Organization: Penn State University Date: Wednesday, 7 Feb 1990 18:28:42 EST From: Ben Liblit <BRL102@psuvm.psu.edu> Message-ID: <90038.182842BRL102@PSUVM.BITNET> In article <9002071850.AA29665@en.ecn.purdue.edu>, eldorado@EN.ECN.PURDUE.EDU (David D Jansen) says: > >Could someone with fractal experience tell me why I get weird data. It gets >the basic shape but not quite right. >type >region = record {a rectangle on the imaginary plane} > amin : real; > bmin : real; > amax : real; > bmax : real; > end; This declaration isn't the error source, but makes me cringe. It seems that a nicer-looking declaration, that better reflects what you're representing, would be: TYPE region = RECORD min, max : imaginary END; Of course, my format is different, but that doesn't matter -- it's the idea that counts. As for the actual source of your errors, your real variables are probably being rounded off. After a few hundred iterations, that can add up to some rather large inaccuracies. There are a few things you can try: If your compiler has other, higher precision floating point types, use them. Look for something called double or extended. The tradeoff is reduction in speed. You could also junk the floating point altogether, and switch to straight integer math. It's blindingly fast, and I *think* it's supposed to be more accurate. Someone posted the core routines for Mandelbrot set iterating in C a short while back. Anyone remember where this appeared? Major tradoff of going all-integer: you have to rewrite the multiplication and division routines. Ben Liblit BRL102 @ psuvm.bitnet -- BRL102 @ psuvm.psu.edu "Fais que tes reves soient plus longs que la nuit."
gwyn@smoke.BRL.MIL (Doug Gwyn) (02/08/90)
In article <9002071850.AA29665@en.ecn.purdue.edu> eldorado@EN.ECN.PURDUE.EDU (David D Jansen) writes: >Could someone with fractal experience tell me why I get weird data. It gets >the basic shape but not quite right. That's easy -- in computing z.b you're using the NEW value of z.a but you should be using the OLD value. Use a temporary variable to keep from clobbering the old data until you're done with it.
reeder@reed.UUCP (Doug Reeder) (02/09/90)
In article <9002071850.AA29665@en.ecn.purdue.edu> eldorado@EN.ECN.PURDUE.EDU (David D Jansen) writes: >Could someone with fractal experience tell me why I get weird data. It gets >the basic shape but not quite right. A friend of mine (a fractal "guru") >can't make out what the problem. > if (((z.a * z.a) + (z.b * z.b)) <= 4) then > pages.bitmap[h,v]:=true Nowhere do you initialize the bitmap to false, and Pascal does not initialize your variables for you, so the value of any given point is undefined when you start. The simplest way to fix this is to add: else pages.bitmap[h,v]:=false > end; > writeln (v/ver*100:5:2,'% done'); etc. Some pascal compilers will initialize your variables for you, but don't depend on it. -- Doug Reeder USENET: ...!tektronix!reed!reeder from ARPA: tektronix!reed!reeder@berkeley.EDU BITNET: reeder@reed.BITNET the Little Mermaid on materialism: I just don't see how a world that makes such wonderful things ... could be bad!