"TL;DR: A hand-wavy treatment of viscous term in N-S Equation. Refer to my future notes for a more rigorous one."
Let's consider the famous Navier Stokes' Equation:
DtD(ρv)=−∇⋅P+(∇⋅τ)T+ρfother
where DtD is the total time derivative, ρ is the density of fluid, P is the pressure, and f is some other forces on the fluid. ⋅ denotes inner product.
Here, for Cartesian coordinate (x,y,z)∈R3,
τ=τxxτyxτzxτxyτyyτzyτxzτyzτzz
is a viscous tensor of rank two.
The reason we use a tensor to express forces is due to the fact each face of a fluid parcel suffers forces from left/right (along x-axis), front/back (along y-axis), and up/down (along z-axis). THIS IS AN IMPORTANT FACT.
The first letter in the subscript of τ denotes the normal direction of the face, while the later one denotes the direction the force is pointing. This is demonstrated in the following diagram.
— Figure 1. Free body diagram of an infinitesimal fluid parcel. Image from Viscous Fluid by Frank White.
Viscous Tensor as Forces per Unit Area
To see that (∇⋅τ)T gives us the correct description of viscous forces per unit volume fviscous on the fluid parcel, we consider, without loss of generality, the component of this force in the x direction.
It is not hard to see from Fig. 1 that the force in the x direction on the rightmost face is just the sum of each τix that points to the right multiplied by the area of that face:
Great! I think at this point it is clear why we use a tensor to express the viscous force.
Viscous Tensor is Symmetric
The reasoning behind this is that we don't want the fluid parcel to rotate. This is a physical assumption. If you think about water we drink being a giant cloud of rotating cubes, that would be really weird.
Expressing this fact in physical terms, we need the net angular momentum Tpar of the fluid parcel to be 0.
Hence, we proved that the tensor is symmetric since τij=τji.
The Final N-S Equation
— Figure 2. A plot of stress versus speed change with height change in a fluid parcel determined by experiment.
Here we consider only Newtonian fluid (μ is scalar constant). For some other fluid, where thermal effects come into play, you can refer to the full treatment given in the fluid mechanics notes by Dr. Joseph M. Powers at University of Notre Dame.
We see from Fig. 2 that we have the following equality:
μ=∂y∂vxτyx
It is not hard to see that without loss of generality:
μ=∂xj∂viτji
assuming an isotropic fluid, i.e. μ is the same for each face.
In a more compact notation, we write the above as: