Xjyuk

The following example typesets five tabular environments. All five tabulars share the same definition,

begintabular p1.67cm 

However, the overall widths of the five tabulars differ considerably, depending on how various multicolumn statements are employed. Can somebody explain why this is happening?

documentclass[preview,border=100pt]standalone
setlengthtabcolsep0.2cm
begindocument

begintabular p1.67cm hline
multicolumn4c HEAD\hlinehline
multicolumn2cA& C & D\hline
A & B & C & D\hline
endtabular

bigskip
begintabular p1.67cm hline
multicolumn4c HEAD\hlinehline
multicolumn2cA& C & D\hline
multicolumn1cA & multicolumn1cB & multicolumn1cC & multicolumn1cD \hline
endtabular

bigskip
begintabular p1.67cm hline
multicolumn4c HEAD\hlinehline
multicolumn2cA& C & D\hline
endtabular

bigskip
begintabular p1.67cm hline
multicolumn4c HEAD\hlinehline
multicolumn3cA & D\hline
endtabular

bigskip
begintabular p1.67cm hline
multicolumn4c HEAD\hlinehline
multicolumn4cA\hline
endtabular
enddocument

If every cell is re-specified by a multicolumn then the original column specification in the tabular is not used at all. That is the case in all your examples, so the only effective column specifications are c which will use the natural width of the text of the cells in each column.

The key to understanding what's going on is to appreciate fully the meaning of the fact that multicolumn directives take three arguments:

• Arg 1: the number of columns to which the directive applies. This number can be as small as 1 and as large as n, where n is the total number of columns.




• Arg 2: The column type -- observe the singular form of the word "type" -- to be used for the combined column. It's important to realize that this column type completely overrides the underlying column type (or types) that is specified via begintabular....




• Arg 3: The contents of the combined column.

For the sake of completeness, I'll note that whereas the p column type has a fixed width, columns of type l, c and r have no ex ante fixed width. Instead, their widths will be that of the material they contain.

Let's apply this to each of the five tabular environments. First, a picture (slightly modified from your code, mainly to ease cross-referencing) to establish the five different widths:

• In the first tabular (with header HEAD1), there are cells in each of the four columns without multicolumn statements. The overall width of each cell is therefore governed by the width of the associated p column type (1.67cm for the first three columns, and 8cm for the final column).



• 
  

  In the second tabular, the material in columns 1 and 2 is never without a multicolumn statement. In both data rows, the c column type is employed by the multicolumn statements. Hence, LaTeX never gets to apply the "underlying" p column type for either of the first two columns. That's why the width of the second tabular is less than that of the first.

  
  
  

  Observe that the widths of the first two columns are given by the widths of the underlying letters (A and B, respectively) plus 2tabcolsep; the combined width of the first two columns is therefore the sum of the widths of the letters A and B plus 4tabcolsep plus arrayrulewidth.

  
  



• In the third tabular, there are only three "effective" columns, as there are no cells with separate material for the first two columns. The width of the effective first column is therefore given by the sum of the widths of the letters A and B plus 2tabcolsep.




• The fourth tabular contains effectively just two columns. The underlying p column types of the first three columns are never used. Only the fourth column's underlying "p" type gets any use.




• The fifth and final tabular effectively contains just one column, whose column type is c. Observe that the four underlying p column types are never used. That's why the fifth tabular is so much narrower than the first four are.

The thing to remember is that multicolumn is quite powerful -- and far more powerful than one might at first give it credit. In particular, it's important to realize that the consequences of the fact that the column type specified in the second argument of multicolumn overrides the underlying column types completely.

documentclass[preview,border=100pt]standalone
setlengthtabcolsep0.2cm
%%usepackagearray
begindocument

begintabular
hline
multicolumn4cHEAD1\
hlinehline
multicolumn2cAB & C & D\
hline
A & B & C & D\ % all four underlying column types get used in this row
hline
endtabular

bigskip
begintabular
hline
multicolumn4cHEAD2\
hlinehline
multicolumn2cAB & C & D\ % "p" column type is used only in columns 3 and 4
hline
multicolumn1cA & multicolumn1cB & 
multicolumn1cC & multicolumn1cD \
hline
endtabular

bigskip
begintabular
hline
multicolumn4cHEAD3\
hlinehline
multicolumn2cAB & C & D\
hline
endtabular

bigskip
begintabular
hline
multicolumn4cHEAD4\
hlinehline
multicolumn3cABC & D\ % "p" column type is used only in column 4

hline
endtabular

bigskip
begintabular
hline
multicolumn4cHEAD5\
hlinehline
multicolumn4cABCD\ % "p" column type isn't used anywhere
hline
endtabular
enddocument