Ioana's LaTeX squeezing tips

This is a set of notes I drafted in early 2010 for my students of that time. Current students asked that I blog them here, so here it goes:

It is a timeless truth that papers may need squeezing. Below, find a list of basic tricks, which are not subtle but very helpful, and which every successful author in the field knows.

As part of getting a PhD, you are supposed to learn these  and apply them independently and appropriately.

1. Removing lines
This works tremendously well. The idea is that if a paragraph has a short last line, it is very often possible to rephrase it so that it is one line less. This helps a lot, because the box computed by LaTeX to surround the paragraph become smaller, then LateX  may find a way to pull it from one page to another, from one column to another etc.
It is always possible to shorten the first version of a text (after 3 or 4 rounds of reformulation it starts becoming a bit cryptic 😉

Of course this assumes that (1)  you understand sufficiently the paragraph (2) you have sufficiently good English skils, in order to be able to propose a truly equivalent rephrasing. If the paper is complex and several authors helped with several sections, it is best that each author does this to “his/her sections” in order to have the best quality of reformulations.

2. Removing nice formatting and rephrasing in plain text

Item lists, enumerations, and descriptions take space not only because of the spaces put by LaTeX before and after them, but also because the length of the line is shortened wrt normal text, which may lead an item to occupy more lines than if it had been put in plain text. The solution is  not to use them. They can be simulated / replaced in the following way:

begin{itemize}
item bla1
item bla2
item bla3
end{itemize}

or

begin{enumerate}
item bla1
item bla2
item bla3
end{enumerate}

become:

($i$)~bla1; ($ii$)~bla2; and ($iii$)~bla 3

Moreover

begin{description}
item[bla1] blabla1
item[bla2] blabla2
end{description}

becomes

noindenttextbf{Bla1} blabla1

noindenttextbf{Bla2) blabla2

3. Eating space around headers

One can typically eat  2mm  above and 0.5-1mm below each section header. Thus

begin{section}

becomes

vspace{-2mm}
begin{section}
vspace{-0.5mm}

However this can only be done UNIFORMLY FOR ALL SECTIONS, otherwise it is visible and reviewers will complain: “there is too much material in this paper, it is hard to read, the authors should re-write it and re-submit”.
I HAVE SEEN THIS COMMENT HUNDREDS OF TIMES, it is sometimes made when the reviewer has submitted a competing paper at the same time and (s)he prefers that her (his) paper gets published first and your paper never (because (s)he got there first).

If you want to avoid such comments, make sure there is no reason to make them.

A simple solution is to re-define the command at the beginning of the paper e.g.

newcommand{mysection}{vspace{-2mm} begin{section} vspace{-0.5mm}}

This has the advantage that you modify only the command, in order to uniformly change the spacing of *** all *** section headers.

Similarly for sub-section (at least 0.5mm above and 0.5 mm below can typically be eaten and it typically helps, when you sum it over the whole paper).

4. Carefully vspace figures or tables

Often there seems to be too much space in and around a figure. Two cases occur:

(a) There is space  between the figure and the caption. This may get really bad if the *content* of the figure is centered but the *caption* is not, which leads to the usual problem of LaTeX leaving some space empty between centered text, and justified text. The solution here is:
– make sure the figure content and the figure caption are both centered
only after  eat 1-2 mm between them with a vspace.

(b) There is space *** between the caption and the text after the figure ***. This is tricky, because a figure is a float (in LaTeX terms) and the text which appears after the figure in .dvi or .pdf is not the text whcih appears after the figure in .tex.  (LaTeX places the float, e.g. the image, where it considers best according to all the specifications of the figure).

The only thing that makes sense here, is to vspace before the end of the environment, i.e.

like this [CORRECT]:
begin{figure}

caption{…}
vspace{-2mm}
end{figure}

not like that [WRONG]:
begin{figure}

caption{…}
end{figure}
vspace{-2mm}

The former [CORRECT] above means: “in the bounding box of this figure float, please eat 2mm at the end (make that bounding box smaller, or rather: allow the next bounding box *** whatever it is *** to start 2mm higher than it would have normally started).
DONE THIS WAY (CORRECTLY), even if the figure moves across the paper, it will always carry its space saving.

Is there a real risk that the figure may move across the paper (i.e. be placed in another column or page?)
YES, A HUGE ONE. THAT’S WHAT PAPER SQUEEZING IS ABOUT: writing things so that LaTeX will place the boxes differently. So, paper squeezing begs LaTeX to reconsider its box (thus, float) placement.

DONE THE WRONG WAY, a text paragraph which used to be placed by LaTeX after your float (figure or table) will suddenly jump into its previous text paragraph or even worse, overlap with it.
THIS IS AWFUL because it is very visible.
On the regularity of a page with the same amount of space between the lines, even 1mm missing shows, and it is VERY ugly.

5. Squeeze the bibliography

– put the bib in small font and squeeze the lines into each other (when you put *just* small font, LaTeX will still leave more spaces between lines than necessary)
– use the abbreviated bib style (I. Manolescu instead of Ioana Manolescu)

do not give page numbers, country, city, month, ISBN or book editors. Title, year, venue, and authors are the minimum necessary.
– use abbreviated venue names if they are sufficiently well-known in the place where you submit your papers. I.e. write booktitle=”SIGMOD” and never “Proceedings of the ACM SIGMOD Conference” (obviously! 🙂
– some people are comfortable with cutting authors from a reference and leaving “et al.” instead. E.g.

“S. Abiteboul, I. Manolescu, N. Polyzotis, N. Preda and C. Sun”

becomes

“S. Abiteboul et al.”

Depending on whom among the authors of this reference gets to review *your* submission, this may be diversely appreciated ;D
So, I don’t do such things (unless there are more than 6-7 authors and it is really becoming too space-consuming to cite them all). It is a desperate measure of last resort, and it only saves a bib line. It may help, though.

Assuming you always have more interesting things to say in the paper than enumerate the references, BY DEFAULT ALL YOUR PAPERS SHOULD HAVE A BIB SECTION APPLYING ALL OF THE ABOVE AND OF THIS FORM:

renewcommand{baselinestretch}{0.8}
bibliographystyle{abbrv}
small
bibliography{paper}
renewcommand{baselinestretch}{1}

6. Minimal affiliation information

The only things required are the authors’ last name, given name initial, institution, country, and e-mail (of course unless otherwise specified).
The rest is noise and competes with space for presenting your ideas.

This is not an insignificant saving: before cutting a few words to save a line and make a paragraph shorter (and its idea more obscure), cut the crap (the long list of affiliations and other consortia to whom we owe politeness).

Use simpler names whenever possible, i.e. “INRIA” is better than “Inria Saclay–^Ile-de-France and LRI, Universit’e de Paris Sud-11”. Update March 2014: For the INRIA affiliation, simply use “INRIA” (no Saclay).

7. Acknowledgments

Acknowledgements of institutions, grants, and anything administrative *** only go in the camera-ready *** (again, you are not going to lose precious submission space for such stuff).

Acknowledgements of people who discussed with you, coded, gave you code, or otherwise *** scientifically helped *** should go in the submission.
They should go under this form:

noindenttextbf{Acknowledgments} We are grateful to X (or We thank X) for Y.

8. Bundle content together in the same figure

If you have two figures, then you will have two captions and space lost twice around the placement of two floats.
If you have 1 figure, then there will be only one caption, and space lost only once.

So, whenever two figures can be grouped under a common caption, e.g. two experiments on the impact of X on Y in various settings, put the two graphs in the same figure. You will then refer to “the top/left graph in Figure~ref{fig:bla}”
or
“the graph at the top / at left in Figure~ref{fig:bla}”

*Caution*
– don’t do this if the figures have nothing to do with each other
– adding “top/left” to the reference to the figure in the text makes it a bit more tiresome to read.

9. Scaling down boxes

Just learned from Konstantinos 🙂

scalebox{0.80}{…}
will reduce to scale the bounding box corresponding to “…”.

10. Change data presentation

If there are less than, say, 12 points to show, you may consider using a table with the numbers, than a figure which may be too expensive in terms of space. It has less impact, but you can comment on the table just as you would have commented on the figure (“we see that this saves 60% of …”)

NONE OF THE ABOVE REQUIRES REMOVING CONTENT.
Therefore, they should all be applied before removing any of the actual content.

Even if  you will also remove content, it makes sense to do all the above, before cutting content, because doing these will allow you to understand how much you need to really cut.

So basically the are  “space issue” only if the content still doesn’t fit, after applying any and all of such techniques.

Happy writing 🙂

Comments are closed.