gmx tcaf computes tranverse current autocorrelations. These are used to estimate the shear viscosity, η. For details see: Palmer, Phys. Rev. E 49 (1994) pp 359-366.
Transverse currents are calculated using the k-vectors (1,0,0) and (2,0,0) each also in the y- and z-direction, (1,1,0) and (1,-1,0) each also in the 2 other planes (these vectors are not independent) and (1,1,1) and the 3 other box diagonals (also not independent). For each k-vector the sine and cosine are used, in combination with the velocity in 2 perpendicular directions. This gives a total of 16*2*2=64 transverse currents. One autocorrelation is calculated fitted for each k-vector, which gives 16 TCAFs. Each of these TCAFs is fitted to f(t) = exp(-v)(cosh(Wv) + 1/W sinh(Wv)), v = -t/(2 τ), W = sqrt(1 - 4 τ η/ρ k^2), which gives 16 values of τ and η. The fit weights decay exponentially with time constant w (given with -wt) as exp(-t/w), and the TCAF and fit are calculated up to time 5*w. The η values should be fitted to 1 - a η(k) k^2, from which one can estimate the shear viscosity at k=0.
When the box is cubic, one can use the option -oc, which averages the TCAFs over all k-vectors with the same length. This results in more accurate TCAFs. Both the cubic TCAFs and fits are written to -oc The cubic η estimates are also written to -ov.
With option -mol, the transverse current is determined of molecules instead of atoms. In this case, the index group should consist of molecule numbers instead of atom numbers.
The k-dependent viscosities in the -ov file should be fitted to η(k) = η_0 (1 - a k^2) to obtain the viscosity at infinite wavelength.
Note: make sure you write coordinates and velocities often enough. The initial, non-exponential, part of the autocorrelation function is very important for obtaining a good fit.