Usually, you can reduce the correlations between the coefficients by transforming the components to pseudo-components. Pseudo-components, in effect, rescale the constrained data area so the minimum amount allowed (the lower bound) of each component is zero.
The following table shows two components expressed in amounts, proportions, and pseudo-components. Suppose the total mixture is 50 ml. Let X1 and X2 be the amount scale. Thus X1 + X2 = 50. Suppose X1 has a lower bound of 20 (this means that the upper bound of X2 is 50 minus 20, or 30). The following are some points on the three scales: