Frequently Asked Questions
Is PDT guaranteed to increase experiment performance?
No. There are several reasons why PDT cannot offer such a comforting
guarantee, among which:
Experimental noise may wash out any structure available for
optimization. Experimental noise should be comfortably beneath 10%
for PDT to do its job. Experimental noise may be reduced by
performing replicates, and averaging response of replicates; the
noise goes down as √N, where N is the number of replicates.
The response surface may be too complex for the experimental
bandwidth available. Very complex response surfaces require more
How many experimental parameters should I have?
This depends on your experimental setup. You want to try and include
all parameters that are important, in the sense that they could
seriously affect the experimental response. PDT can handle 4-20
parameters, depending on the roughness of the response landscape and
your experimental bandwidth. If you have fewer than 4 parameters, you
might try simpler experimental design approaches, e.g. sweeping