Is there a rule (or maybe a rule of thumb) that provides guidance for choosing the layout convention (see Layout conventions, wiki) to use when dealing with matrix derivatives?

I found a hint in this answer here to a related question but I could hardly find any well justified answer.

  • $\begingroup$ Try working a few problems in each. Then pick the one you like best and stick with it. And if you read an article using the other convention, it isn't difficult to translate it to your preferred convention. $\endgroup$ – greg Nov 29 '18 at 21:21
  • $\begingroup$ @greg That's what I've been doing. But, I believe the layouts are not equivalently meaningful. Of course, any layout would give the derivatives but the arrangement seems important. This is explained in a section of Magnus and Neudecker's "Matrix Differential Calculus" (under First-order differentials and Jacobian matrices) that talks about why some notation (layout) is bad. The authors also give what they consider good notation and why. $\endgroup$ – Likely Dec 1 '18 at 4:02
  • $\begingroup$ Magnus and Neudecker make some good points. I would suggest that you read "Complex-Valued Matrix Derivatives" by Are Hjorungnes for another point of view. $\endgroup$ – greg Dec 1 '18 at 4:11
  • $\begingroup$ @greg Okay, thank you for the reference. $\endgroup$ – Likely Dec 2 '18 at 22:02

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.