[sklearn] 'error, no feasible solution' when calling scipy's minimize function

I was using L-BFGS-B the bounded optimization method in scipy.optimize.minimize, but the returned OptimizeResult object kept telling me its failure. res.message notifies ‘error, no feasible solution’.

Thanks to this answer, I found the mistake is that I reverse the lower bound and the upper bound. Normally, we would think of learning rate from higher end to lower end. But the bounds for this function is strictly (lower,upper).