You probably want to use some tolerance from zero
x = randn(500000,1);
indices = find(abs(x-0)<eps);
x(indices) = x(indices) - 0.00006;
The tolerance is assuming that you have a vector of floating point numbers here. I've used eps with no argument, it may be better here to use eps(0), which is a smaller number, than eps, but it's hard to say not knowing your application. It may be sufficient for you to just use some small number bigger than eps like 1e-6 for example.
