Why NaN even though all elements considered

I am comparing every column with other columns but getting the answer in R(4,2) and R(3,1)

for i=1:nrow-1
        a=data(1,i);
        c=data(2,i);
        e=data(3,i);
        g=data(4,i);
 for j=[1:i-1,i+1:nrow]
        b=a-data(1,j);
        d=c-data(2,j);
        f=e-data(3,j);
        h=g-data(4,j);
        kU=b'*b;
        kL=d'*d;
        pU=f'*f;
        pL=h'*h;
   S=(kU'*kU+kL'*kL)/(kU+kL)+(pU'*pU+pL'*pL)/(pU+pL);
   R(i,j) = S;
   R(j,i) = S;
   end
end
end

trial(A)
ans =
         0    1.8299       NaN    1.8299
    1.8299         0    1.8299       NaN
       NaN    1.8299         0    1.8299
    1.8299       NaN    1.8299         0

when the input

 A = 2.0000    2.2000    2.0000    2.2000
    3.0000    3.1000    3.0000    3.1000
    2.0000    2.2000    2.0000    2.2000
    3.0000    3.1000    3.0000    3.1000

[Stephen23]

Your algorithm divides zero by zero. This happens for these pairs of (i,j): (1,3), (2,4), (3,1). What do you expect division of zero by zero to do?

[Roger Stafford]

Actually it is zero divided by zero that is producing the NaNs here. A non-zero divided by zero will produce inf or -inf,