What? I asked and you said it couldn't be done. Of course I'm not gonna just roll over and admit defeat that easily. I'm gonna try to prove you wrong. :-)
No, what I'm saying that you can't manipulate the equations to come up with the exact values to use that will do the effect. It's not possible (chuck it into an equation solver, and the result is dependent per pixel per channel), so don't try. It may be possible to do something that comes close, given the range of blend modes available (multiply blend not supporting alpha is annoying), so do try that.