Linear and power fitting of original and transformed data not reproducible

I am fitting two variables that seem to have a power relationship (y=a*x^b). If I log-transform both variables, they should have a linear relationship (ln(y) = c*ln(x) + d. Notice that b=c and ln(a)=d. However, when I fit the power and linear equation to these, respectively, the estimated coefficients do not satisfy these relationships. Expectedly, R-square is also different. I tried changing the fit options and didn't manage to make the two fittings equivalent. Has anyone any idea why this happens?

PS. fitting in excel is satisfying these relationships and I get identical values

[Vladimir Sovkov]

Do you recompute the weights (or inaccuracies) of y into the weights (or inaccuracies) of ln(y)? If yes, the results must be equivalent, otherwise they must not. The fit of equally weighted y is far not the same as of equally weighted ln(y).

[Xen]

Thanks for the reply Vladimir. The weights of y are by default ones(n,1). How would you recompute the weights for ln(y)?

[Vladimir Sovkov]

The weights must be , however, it is better to check the documentation of the tool you are using.