M3's approach gave me some ideas, and I gave him thumbs up.
Let p(n, k) be the number of ways to write n as the sum of exactly k terms irrespective of order ( http://mathworld.wolfram.com/PartitionFunctionP.html , see after (54)). Usually it is uppercase, but the letter P is already used for the number of times the die is thrown.
The ways to have a sequence satisfying the desired condition (monotonic, all numbers represented) is 2 * p(P, 6). Take that as numerator. Then as denominator take 2 * ( sum_(i=2)^6 ( C(6, i) * p(P, i) ) ) + C(6, 1) * p(P, 1) = 2 * ( sum_(i=2)^6 ( C(6, i) * p(P, i) ) ) + 6, where C(n, k) is binomial coefficient. This quotient is the desired probability.
I often make mistakes with questions like these, so if anyone wants to verify my answer or point out a flaw, by all means do so.
Edit: Oh wait, we need order to matter... In this case we want compositions rather than partitions, and then I think it would work as desired. Would have to dig around to find a compositions function taking (n, k). (Further edit: Ah my memory is like Swiss cheese. Finding the number of compositions of n into exactly k pieces is easily handled by "stars and bars" technique, it's just C(P - 1, k - 1). I'd like to see how this compares with gianlino's solution, but I'm getting a little tired of looking at this problem and will set it aside for now.)
@M3, for your example, "p(5,3) = 2, since partitions of 5 of length 3 are {3 1 1} & {2 2 1}" -- this would correspond to die sequences involving three distinct face values, and this count would be multiplied by C(6, 3) specifying the face values, and further by 2 to distinguish increasing and decreasing. So, let's pretend for now I didn't mess up the order part -- then if we choose face-values {1, 2, 3}, the partition 5 = 3+1+1 would represent sequences 11123 and 33321. Of course for 11123 we also need 12223 and 12333, which is the order part that I missed. When we break into 6 parts (i.e., k = 6 instead of k = 3), we must have all 6 face values represented. For P = 5, this will result in 0.
Can't say I understand your solution though. You seem to have r1, ... , r6 appearing in your final solution, but it should be only in terms of P because we are not given r1, ... , r6. Wouldn't we have to sum over all possible values of r1, ... , r6? Maybe I just read it wrong. Thanks for the feedback.
Further edit: Okay, I checked my answer against gianlino's and they are almost the same. The only thing I don't understand is why gianlino changed the 6 to a 3. It seems it should stay a 6.
So this expression:
[((P+5) C 5) - 6 ((P+4) C 4) + 15 ((P+3) C 3) - 20 ((P+2) C 2) + 15 ((P+1) C 1) - 6] / [((P+5) C 5) - 3]
is equal to this expression:
2 * C(P-1, 5) / (2 * sum(i=2, 6, C(6, i) * C(P-1, i-1)) + 6)
Or, factoring out the two,
C(P-1, 5) / (sum(i=2, 6, C(6, i) * C(P-1, i-1)) + 3)
By the way, the modified expression of gianlino can be written this way as well:
sum(i=0, 5, (-1)^i * C(6, i) * C(P+5-i, 5-i)) / (C(P+5, 5) - 3)
Furthermore! The numerators and denominators are identical. That is,
C(P-1, 5) = sum(i=0, 5, (-1)^i * C(6, i) * C(P+5-i, 5-i))
and
sum(i=2, 6, C(6, i) * C(P-1, i-1)) + 3 = C(P+5, 5) - 3
So I don't think we'll get any simpler than this:
C(P-1, 5) / (C(P+5, 5) - 3)
:D