I am currently playing around with hailstone sequences (collatz numbers).

I would like to create a plot like this (from Wikipedia):

enter image description here

But I don't know how this kind of plot is called (it's neither a line plot, nor a scatter plot or a bar chart or a histogram, although it looks similar).

I would like to plot how many steps you need from a number n to get to 1. The x-axis should be numbers (from 1 to M, with M as high as possible) and the y-axis should be the number of steps.

As I didn't know how the kind of plot above is called (and I was not able to find a name) I used a scatterplot:

enter image description here

Sadly, this seems to consume a lot of memory. Plotting 4100 (n, steps from n to 1) pairs worked, but when I increased it to 4150 pairs, I got:

! TeX capacity exceeded, sorry [main memory size=3000000].
\endvarwidth ...\vbox \bgroup \unvcopy \@tempboxa 
                                                  \@tempdima -\maxdimen \let...
l.28 \end{document}

I think the following could eventually solve my problem:

  • switch to gnuplot (I have no idea how to read a CSV-file with gnuplot and I only used gnuplot+LaTeX once. Does anybody have a minimal working example with instructions how to compile it?)
  • switch diagram type to the diagram above (How is it called?)
  • don't actually write the numbers n in csv file, but get it from the line number (How can I get the line number from a CSV file in my code below for x?)
  • switch to another type of plot like here (I guess this will be more complicated and it seems as if I had to use gnuplot / R. Does anybody know how to apply this to my problem?)

All code I've written is on GitHub. This is the LaTeX code I've written so far:

\documentclass[varwidth=true, border=2pt]{standalone}
\usepackage[utf8]{inputenc} % this is needed for umlauts
\usepackage[ngerman]{babel} % this is needed for umlauts
\usepackage[T1]{fontenc}    % this is needed for correct output of umlauts in pdf
\usepackage[margin=2.5cm]{geometry} %layout


            axis x line=middle,
            axis y line=middle,
            enlarge y limits=true,
            %xmin=0, xmax=2150,
            %ymin=0, ymax=600,
            width=15cm, height=8cm,     % size of the image
            grid = major,
            grid style={dashed, gray!30},
            legend style={at={(0.1,-0.1)}, anchor=north}
          \addplot[scatter,only marks] table [x=n, y=steps, col sep=comma] {../steps.csv};

I would like to plot at least 100,000 elements (rather 10,000,000 if possible). How can I do it?

  • 1
    Is the first one like a comb plot (see the pgfplots manual)? May 16, 2013 at 9:27
  • Yes, xcomb (pgfplots.sourceforge.net/gallery.html ) looks like what I would like to get :-) I'll try it in two hours (currently I'm in a lecture). Do you know how to get rid of the dots at the end? May 16, 2013 at 9:42
  • Not sure if this will help much, but you can increase the capacity of TeX's memory. See for example this.
    – user10274
    May 16, 2013 at 9:51
  • Torbjørn T.: Thanks, it worked as expected (I couldn't resist to try it and leaved the lecture). @MarcvanDongen: This will not help, as I have 4GB in total and I already gave TeX 3GB. May 16, 2013 at 10:22

3 Answers 3


pgfplots, comb plot

Thanks to Torbjørn T., who has suggested to look at comb plots. This allowed me to plot 7150 (but not 7200) data points:

enter image description here



enter image description here

You have to create a file that tells gnuplot what to do. I've called it plot.gnuplot:

set terminal latex
set output "plot-tmp.tex"
set datafile separator "," 
set title "Collatz Number of steps"
plot '../steps.csv' every::1 

Then you have to run gnuplot: gnuplot plot.gnuplot This will produce a "plot-tmp.tex" that can be included into your LaTeX document:



\input plot-tmp

This allows me to plot 22300 data points (but not 22350).

comb plot

To get a comb plot, you have to add "with impulses to the plot command. So your plot.gnuplot file should look like this:

set terminal latex
set output "plot-tmp.tex"
set datafile separator "," 
set title "Collatz Number of steps"
plot '../steps.csv' every::1 with impulses

which produces:

enter image description here


You can (should?) plot large data to PDF-files with R. R could handle the 120 MB file with 10,000,000 data points.

To do so, you have to install R and ggplot2 (sudo apt-get install r-cran-ggplot2).

Now you can start R from command line with sudo R (it has to be a capital letter) to install hexbin: install.packages("hexbin", dependencies=TRUE); (source)

Now create a file called analyze.R and copy this:


mydata = read.csv("/home/moose/Downloads/algorithms/collatz/steps.csv")

# Prepare data
p<-ggplot(mydata, aes ( x=n,y=steps ))

# Plus means you add those options to the plot
p + geom_hex( bins=30 ) + opts(panel.background = theme_rect(fill='white', colour='white'))

Execute it: R -f analyze.R

And you get:

enter image description here


enter image description here

The formulation of the question looks a bit confusing. First it talks about the graph from Wikipedia link, which is an ordinary bar chart that represents a histogram: for every integer x (the number of steps to reach 1 from some number from 1..100000000) there is an integer y (the amount of numbers from that interval, that have the same number of steps x). Note, that the x range is not so big, just less than 600 individual points collect the statistics obtained from 100mln numbers. Next question is more simple and more difficult at the same time: to plot the number of steps for every number tested in a range. It is simpler, because there is no need to accumulate the amount of numbers with the same x steps, while the difficulty is... To plot such a graph with, say, a "humble number" of 100000 points, every point has to be plotted (otherwise it would be just some kind of useless random mess). Assuming that every bar is just one point wide, to print if at 1200dpi, the total width of the x-axis would be roughly two meters wide.

So, this Asymptote code (without any extra programs) is used to plot the same kind of the graph as that of Wikipedia link, a statistics obtained from nMax=100000000 (100mln) numbers. It took about 20 min to run in the background on a not-very-advanced laptop. The histogram data are accumulated in the integer array stepCount and printed to stdout. The output also includes:

  • n=100000000 - max number in a range tested,
  • highest=2185143829170100 - max number reached during the whole test,
  • maxStep=949 max number of steps to reach 1,
  • maxStepArg=63728127 the number, that needs maxStep steps to reach 1.

So, starting from 63728127 it should take 949 steps to get 1.


import graph;
int nMax=100000000;
int highest=1;
int maxStepArg=1;
int[] arStep=array(nMax,0); 

int stepMaxLimit=1000;
int[] stepCount=array(stepMaxLimit,0); 

int steps(int nn){
  int n=nn;
  int m=0;
  int nextn;
      assert(n<floor((intMax-1)/3),"too big n="+string(n));
    if(nextn<arStep.length && arStep[nextn-1]>0){
      return m;
  return m;

int maxStep=1;
int j;
for(int i=1;i<=nMax;++i){

+", highest="+string(highest)
+", maxStep = "+string(maxStep)
+", maxStepArg = "+string(maxStepArg)

int nj;
real t;

pen barPen=rgb(0,0,0.5608)+opacity(0.6)+0.8pt;
guide g=(0,0);
for(int i=1;i<=maxStep;++i){



I make the hailstone graph via Asymptote for this 10-year-old question. It's exciting that the Collatz conjecture is still open.

To draw, I use a struct named Collatz with 3 attributes: chain for the chain, len and max_value for its length (stopping time) and maximum value. In case one wants the bar graph, just replace dot((i,d),2pt+red); by draw((i,d)--(i,0),blue);

It takes a few second with n=10000 to compile online. I have not yet tested to run offline.

enter image description here

enter image description here

// Collatz: the hailstone graph
// Run on http://asymptote.ualberta.ca/ 
struct Collatz {
    int[] chain;      // the Collatz chain
    int len;          // the length of the Collatz chain
    int max_value;    // the maximum of the Collatz chain

Collatz collatz_calculate(int n) {
    Collatz c;   
    c.chain.push(n);  // Initialise the chain with the first node n
    c.len = 1;        // Initialise the length chain as 1
    c.max_value = n;  // Initialise the maximum value as n
    while (n > 1) {
        n = (n % 2 == 0) ? (n # 2) : (3 * n + 1);
        ++ c.len;
        c.max_value = max(n, c.max_value);
    return c;

import graph;
int n = 10000;
for (int i=2; i < n; ++i) { 
    int d = collatz_calculate(i).len;
    //draw((i,d)--(i,0),blue);    // the bar graph
    dot((i,d),2pt+red);           // the hailstone graph

yaxis("stopping time",LeftRight,RightTicks);
label("\bfseries Collatz conjecture: hailstone graph",point(N),2N);

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .