OK, I produced an absolutely crazy "solution" myself. This is mostly to make clearer what the problems are; I wouldn't suggest using the (very long) code below. This "solution" only provides italic correction for single letters `A`

to `Z`

and `a`

to `z`

, and it works by making `$`

active. (I could also have used `\(`

and `\)`

, but I don't like those.) Moreover, everything is adjusted "by hand" for 10pt CM fonts, so it won't work for other fonts (but should approximately work for other font sizes). Here's the output:

In the 1st line you see the result of `If $U$ or $V$ ...`

without any correction; in the 2nd line my correction is applied, and in the 3rd line the `$`

s are omitted, i.e., the usual italic font is used. I'm not claiming that the 2nd line is good on all counts, e.g. the space between "f" and "from" is rather small. What I wanted to achieve is that the spacing is just as with the "normal" italic font, that is, in the second and third lines the spacing is (almost) the same. (The 2nd line is slightly longer since the math letters are wider.)

Note in particular that the spacing before punctuation in the 2nd line is different from the 1st line. (I'm not sure which version is the better one.) Clearly, the positioning of "U", "V" and "Y" in the 1st line is not good (I would say horrible); in the 2nd line it's a lot better.

Of course one could change all these numbers in my code to try and further improve the spacing. But I only wanted to point out something else: If you look at the numbers, then you see that it would be very hard indeed to have this correction "automatically" and without changing the font metrics.

```
\documentclass{article}
\makeatletter
\let\mydollar=$
\catcode`\$=\active
\def\my@testtoken{\my@testtoken}
\def$#1${\ifx\my@testtoken#1\my@testtoken
\mydollar\mydollar
\else
\test@single@character#1\my@testtoken
\fi
}
\def\test@single@character#1#2\my@testtoken{%
\def\math@format##1{\mydollar##1\mydollar}%
\ifx\mytesttoken#2\mytesttoken
\ifcat#1a%
\ifdim\fontdimen\@ne\font>\z@
\def\math@format##1{\mydollar\xdef\currentfont{\the\textfont1}\mydollar
{\corrected{##1}}%\currentfont##1}%
}%
\fi
\fi
\fi
\math@format{#1#2}%
}
\def\corrected#1{\csname @correct@#1\endcsname}
\def\correct#1#2,#3,{\expandafter\def\csname @correct@#1\endcsname{\mydollar\mskip#2mu#1\mskip-#3mu\mydollar}}
\makeatother
\correct A0.15,0, %1st number is the correction before the letter,
\correct B0.3,1.5, %2nd number is (minus) the correction after it.
\correct C1.75,2.2,
\correct D0.25,1.4,
\correct E0.3,1.7,
\correct F0.3,1.95,
\correct G1.8,1.15,
\correct H0.25,2.6,
\correct I0.3,2.6,
\correct J0.1,2.2,
\correct K0.3,2.4,
\correct L0.25,0.6,
\correct M0.3,2.6,
\correct N0.3,2.6,
\correct O1.75,1.3,
\correct P0.2,1.5,
\correct Q1.75,1.3,
\correct R0.2,0.2,
\correct S0.4,1.8,
\correct T2.7,1.9,
\correct U2.4,2.6,
\correct V2.4,2.95,
\correct W2.4,2.9,
\correct X0.4,2.5,
\correct Y2.6,3.1,
\correct Z0.4,2.2,
\correct a1.2,1,
\correct b1.2,0.3,
\correct c1.2,0.2,
\correct d1.2,0.8,
\correct e1.2,0.85,
\correct f-1.5,3.5,
\correct g0.7,1.2,
\correct h0.4,1,
\correct i1,1.4,
\correct j-0.5,2.2,
\correct k0.4,1.5,
\correct l0.9,1.8,
\correct m1,0.95,
\correct n1,0.95,
\correct o1.2,0.3,
\correct p1,0.3,
\correct q1.2,1.2,
\correct r1,2,
\correct s0.5,1,
\correct t1,1.25,
\correct u1,0.95,
\correct v1,1.55,
\correct w1,1.5,
\correct x0.4,1.75,
\correct y1,1.2,
\correct z0.4,1.75,
\newcommand\test[1]{%
{\let$\mydollar #1} \par
#1 \par
\let$\relax #1
}
\begin{document}
\it
\test{If $U$ or $V$ and $X$, and $f$ from $j$. Let $T$ be $S$ if $Y$.}
\end{document}
```

--- edit ---

Here's a version of the above correction table for use with newtxtext and newtxmath.

```
\correct A-1.3,-0.2,
\correct B-0.6,0.3,
\correct C0.2,0.9,
\correct D-0.6,0.4,
\correct E-0.55,1.4,
\correct F-2.6,1.1,
\correct G0.0,0.5,
\correct H-0.4,1.0,
\correct I-0.75,1.3,
\correct J-1.0,1.15,
\correct K-0.25,1.3,
\correct L-0.9,0.95,
\correct M-1.0,1.35,
\correct N-2.55,1.55,
\correct O0.25,0.5,
\correct P-0.65,0.45,
\correct Q0.2,0.5,
\correct R-0.8,0.2,
\correct S-0.4,0.6,
\correct T0.7,1.85,
\correct U0.5,1.05,
\correct V0.5,1.65,
\correct W0.5,1.5,
\correct X-1.0,1.45,
\correct Y1.0,1.95,
\correct Z-1.0,1.45,
\correct a-0.3,0.25,
\correct b-0.5,-0.3,
\correct c-0.3,0.35,
\correct d-0.5,0.9,
\correct e-0.25,0.35,
\correct f-3.15,3.15,
\correct g-0.05,0.55,
\correct h-0.6,0.35,
\correct i0.35,0.7,
\correct j-2.0,1.3,
\correct k-0.65,1.3,
\correct l-0.1,1.1,
\correct m-0.2,0.15,
\correct n-0.2,0.0,
\correct o-0.25,0.05,
\correct p-0.75,0.0,
\correct q-0.2,0.4,
\correct r0.2,1.3,
\correct s-0.55,0.45,
\correct t0.15,1.2,
\correct u0.1,0.25,
\correct v-0.1,0.75,
\correct w-0.25,0.6,
\correct x-0.95,0.85,
\correct y-1.0,0.7,
\correct z-0.5,0.55,
```

## Best Answer

With version 2.5 of the`microtype`

package, available on CTAN since 13 March 2013, the simple answer to my question is "use`microtype`

".Thanks, Herbert and Joseph, for telling me that this can be done with the

`microtype`

package. I did know that`microtype`

can do margin kerning, but I had always thought that this is only about punctuation. In reality, one can adjust the margin kerning for every single letter.What I had been wondering: Why is the output in Herbert's and Joseph's answer a bit better, but still far from good? The answer, of course, is that the margin kerning table provided by

`microtype`

just isn't good enough. So I went ahead and went through all the letters ... The output now looks as follows, and I'm quite happy with it:Here's the code, which essentially consists of the adjusted margin kerning table

(that has now been added to the`microtype`

package):