How do I create a blank PDF from the command line?

convert, the ImageMagick utility used in Ketan's answer, also allows you to write something like

convert xc:none -page Letter a.pdf

or

convert xc:none -page A4 a.pdf

or (for horizontal A4 paper)

convert xc:none -page 842x595 a.pdf

etc., without creating an empty text file. @chbrown noticed that this creates a smaller pdf file.

"xc:" means "X Constant Image" but could really be thought of as "x canvas". It's a way to specify a single block of a color, in this case none. More info at http://imagemagick.org/Usage/canvas/#solid which is the "de facto" manual for ImageMagick. [supplemented with information from pipe] (Things like pdf:a can be used to explicitly declare the format of a file. label:'some text', gradient:, rose: and logo: seem to be other examples of special file formats.)

Anko suggested posting this modification as a separate answer, so I am doing it.

Like the smallest possible GIF, the smallest possible blank-page PDF needs to be worked out by hand, because it's so small that unnecessary-but-harmless bits of metadata become a significant part of the file size, and compression actually makes things bigger. It also requires careful attention to the rules in the PDF specification about what bits of the file structure are and are not required. (Did you know that page objects must contain a /Resources dictionary, even if it's empty, but are not required to include a /Contents stream?)

If you don't use PDF 1.5 object and cross-reference streams (which has the advantage that the file can be completely printable ASCII) I believe the best you can do is 317 bytes. If copying and pasting, take note that there needs to be a trailing space on all four of the cross-reference table entries (the lines between 0 4 and trailer<<...), and that there is not supposed to be a final newline after the %%EOF.

%PDF-1.4
1 0 obj<</Type/Catalog/Pages 2 0 R>>endobj
2 0 obj<</Type/Pages/Count 1/Kids[3 0 R]>>endobj
3 0 obj<</Type/Page/MediaBox[0 0 612 792]/Parent 2 0 R/Resources<<>>>>endobj
xref
0 4
0000000000 65535 f 
0000000009 00000 n 
0000000052 00000 n 
0000000101 00000 n 
trailer<</Size 4/Root 1 0 R>>
startxref
178
%%EOF

Replacing the cross-reference table with a manually crafted v1.5 cross-reference stream does make the file slightly smaller, at the price of its no longer being printable ASCII: 294 bytes. (For the sake of readability, not to mention being able to type it in at all, the xref stream below has been hexdumped, but this is not reflected in its stream dictionary. To recover a valid PDF you must either replace the hexdump with the corresponding raw binary bytes, or change /Length 15 to /Length 30/Filter/ASCIIHexDecode and accept a file that is 328 bytes long.)

%PDF-1.5
1 0 obj<</Type/Catalog/Pages 2 0 R>>endobj
2 0 obj<</Type/Pages/Count 1/Kids[3 0 R]>>endobj
3 0 obj<</Type/Page/MediaBox[0 0 612 792]/Parent 2 0 R/Resources<<>>>>endobj
4 0 obj<</Type/XRef/Size 5/W[1 1 1]/Root 1 0 R/Length 15>>stream
0000ff01090001340001650001b200endstream endobj
startxref
178
%%EOF

I also experimented with wrapping objects 1 through 3 into an object stream, but this adds back more overhead than it saves, even when the stream is compressed.

A possible alternative formulation of the xref stream is

4 0 obj<</Type/XRef/Size 4/W[0 1 0]/Index[1 4]/Root 1 0 R/Length 4>>stream
091365b2endstream endobj

Sadly, despite the substantial savings in the length of the actual stream data, the additional /Index[1 4] eats up all but one byte of the savings. Also, it is unclear to me whether you're allowed to leave object 0 completely out of the file. (It's also unclear to me whether object 0 must have generation number -1. If that's not required, you actually save more bytes with

4 0 obj<</Type/XRef/Size 5/W[1 1 0]/Root 1 0 R/Length 10>>stream
000001090134016501b2endstream endobj

.)

To change the paper size, replace 612 792 with the appropriate width and height, expressed in PostScript points (72 PostScript points = 1 U.S. inch or 25.4 millimeters). For instance, 595 842 for A4. You could embed this in a shell script that spits out a blank PDF of whatever paper size is desired; the only tricky part would be making sure that the startxref offset remained accurate even if the size of object 3 changed.

If you have convert (an ImageMagick utility) installed, you could do this:

touch a.txt && convert a.txt -page Letter a.pdf