Lets deconstruct your question:
"mean weight of participants who consume caffeine on a daily basis and those who don’t" > Sounds like you might want to use a T-test so points to a TWOSAMPLEMEANS Proc power.
"detect a 10-pound difference between the groups" looks like a MEANDIFF
"standard deviation of weight will be 15 pounds" looks like a STDDEV, both groups with same StdDev will have same variance so a test of differences would be Test=Diff
So the last bit is what is the actual question? Do you want to determine the overall sample size or the number that should be in each group (seems likely)
Perhaps:
Proc power;
twosamplemeans
test=diff
stddev=15
meandiff=10
power=.85
npergroup=.
;
run;
Using defaults, such as alpha. The deconstruction is sort of a rough map of using Proc Power. The first is to determine the time to statistical test(s) the data may be used for. That will give you the main options in the documentation and match them up to the values you have (or estimate, or assume). Note that the value you look for , such a N or Npergroup or Power or whatever must be included and indicated as missing. ONE thing missing. Each type of sample has a variety of what may be needed as a minimum to work with. There are methods to examine multiple values such as Power=.8 .85 .95
