In my most recent, nonreported, second problem (the first one was heavily reported twofour years ago which discouraged me from making problems), there was a change made to it by the staff which got me thinking (they added that the die were unbiased), what if the die were biased but equally, can Alex get a higher expected score (Xela uses a copy of the die Alex uses so the probabilities of both die are equal), turns out it’s true, he can get a higher expected score but since I do not know how to do multivariable calculus, I couldn’t find the upper bound
Defining some terminology: $p_i$ = Probability of i appearing on the dice, $i ∈ {1,2,3,4,5,6}$
Chance of getting out $= C = \displaystyle \sum_{i=1}^6 p_i^2$
Average score per round $= A = \frac{\displaystyle 2 × \sum_{j=2}^6 j× p_j \sum_{i=1}^{j1} p_i}{\displaystyle 1C}$
Expected score $= X = A × (\frac{1}{C} 1) = \frac{\displaystyle 2 × \sum_{j=2}^6 j× p_j \sum_{i=1}^{j1} p_i}{\displaystyle C}$
Here’s a picture to help you see how I got these formulae
Using this program I was able to find somewhat where the answer should lie
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 

After more than 1 billion random checks, the highest expected score I reached was 23.752142249471124 with $p_1=$ 0.1487492948991272, $p_2=$ 0.14899405525187628, $p_3=$ 0.154656034331149, $p_4=$ 0.16515216949050074, $p_5=$ 0.1808399982323611, $p_6=$ 0.20160844779498574, which is not very far from the 23.33… obtainable from an unbiased die
If there are any mistakes in the formulae I made please tell.
If there are any improvements I can make to the program to make it faster or if u got a higher expected score than me, please put it down with the probability distribution
Note: If you want to know what’s up with the number of cosines and sines in the program, it’s to get a fair distribution of all possible probability distributions of the die, the probability distribution represents points on a six dimensional cube with its diagonals as the axes(points are taken from only one face as I require only positive numbers), and the list “a” contains the angles to reach that point in a circular coordinate system.
Easy Math Editor
This discussion board is a place to discuss our Daily Challenges and the math and science related to those challenges. Explanations are more than just a solution — they should explain the steps and thinking strategies that you used to obtain the solution. Comments should further the discussion of math and science.
When posting on Brilliant:
*italics*
or_italics_
**bold**
or__bold__
paragraph 1
paragraph 2
[example link](https://brilliant.org)
> This is a quote
\(
...\)
or\[
...\]
to ensure proper formatting.2 \times 3
2^{34}
a_{i1}
\frac{2}{3}
\sqrt{2}
\sum_{i=1}^3
\sin \theta
\boxed{123}
Comments
Sort by:
Top NewestI am not very sure but I think the diagonals of the cube are the axes but can’t verify as I can’t as of yet think in six dimensions,five dimensions or four dimensions, sometimes even three dimensions.
Log in to reply
i just used values from 1 to 6 and got as result 6 x 140/36 (23.xxx, similar to your result)
(the output shows: on average a round has 5 rolls)
this is the code is used:
Log in to reply
Nice code, I like that y>x part, I would have used two ifs, could you try the general case of biased die, you could use the value of X given above
Log in to reply
And you can also bias your probability distribution so that it’s around the probability distribution I gave [0.14852376005678114, 0.14916858930276797, 0.15466163703137473, 0.1655010257912824, 0.18074644440113202, 0.2013985434166617] ( most of the other better expected scores had about 0.02 difference from the probability distribution given (0.01 majorly but 0.02 came some times))
Log in to reply
I am currently thinking on how to get a program to actively reduce the range of the probability distributions it should check from, using the information on where the best expected scores are( I guess it’s Machine Learning then and I have no experience with this)
Log in to reply
ok. it might take some time.
i try to understand that part about the biased die.
and the use of that sin and cos.
Log in to reply
Just replace the word six to three in the note and program(comments have been added)
The factor required is just the 1/sum of those three coordinates, I extrapolated everything from three dimensions(as I can’t think in six dimensions, also it took me so much time to think of this in three dimensions itself)
Log in to reply
does it mean the die prefers one side?
Log in to reply
Log in to reply
$C$ and $X$ formulae are the only completely useful stuff, just ignore line 8 to 12 in my code and redefine the “p” list ( which contains the probabilities of 1,2,3,4,5,6 appearing on the die ) to be a set of random probabilities adding to one.
If you are not able to understand this six dimensional stuff no worries, it was just for the programming part (minor part in the programming too, I just wanted some equal probability distributions and this is the best method to do so, all other probability distributions made will have some centre value which will come up more often than others), the aboveLog in to reply
i might check it bcs it sounds interesting, but if we want a biased die, we could just say what are the probabilities for the different numbers and then calculate the outcome.
Log in to reply
Log in to reply
Log in to reply
thank you for that info. now i understand the idea.
Log in to reply
Log in to reply
Log in to reply
I thought that I can’t partially differentiate $X$ with respect to $p_i$ because of the $\displaystyle \sum_{i=1}^{6} p_i$ constraint, turns out I was wrong and I got six equations ( $X$ was partially differentiated with respect to $p_1,p_2, p_3,p_4, p_5,p_6$ and equated to zero
$\displaystyle \bigg(\sum_{i=1}^6 p_i^2 \bigg) \bigg(x\sum_{j=1}^{x1}p_j + \sum_{j=x+1}^6jp_j \bigg) = 2p_x \sum_{j=2}^6jp_j \sum_{i=1}^{j1}p_i , \forall x \in \big\{1,2,3,4,5,6\big\}$
Log in to reply
If I solve these six equations I should get the answer I require, but since these are non linear and there are no non linear equation solvers ( 6 equations type ) when I searched, this has to be also programmatically done (I do feel multiple probabilities will come from these six equations)
Log in to reply
The equations almost satisfy the highest expected value I got, deviating only about 0.4% at the highest (some went as low as 0.07%)
Log in to reply
Also when I added these 6 equations and reduced it, I found another formula for $X$ which is only satisfied at the maximas and minimas
$X=\displaystyle \sum_{i=1}^6\bigg(i\sum_{j=1}^{i1}p_j + \sum_{j=i+1}^6jp_j\bigg)$ which simplifies to $X=20p_1+20p_2+21p_3+23p_4+26p_5+30p_6$
Log in to reply
Rearranging the equation in the before comment gives $\displaystyle \bigg(x\sum_{j=1}^{x1}p_j + \sum_{j=x+1}^6jp_j \bigg) = X × p_x, \forall x \in \big\{1,2,3,4,5,6\big\}$and using $\displaystyle \sum_{i=1}^{6}p_i=1$ we can reduce it to $\displaystyle xx × p_x+\sum_{j=x+1}^{6}(6j)×p_j = X × p_x, \forall x \in \big\{1,2,3,4,5,6\big\}$ which can be further reduced to $\displaystyle p_x= \dfrac{x+ \displaystyle \sum_{j=x+1}^{6}(6j)×p_j}{X+x}, \forall x \in \big\{1,2,3,4,5,6\big\}$ Notice that $p_x$ is now represented in terms of $X$ and its higher terms but not lower terms using this property we can now find out $p_6$ in terms of $X$ as it does not have any higher terms, and using $p_6$ we can find out $p_5$ and so on.
The end result is that $p_6=\frac6{X+6}$ $p_5=\frac{11}{X+5}\frac{6}{X+6}$ $p_4=\frac{18}{X+4}\frac{11}{X+5}\frac{3}{X+6}$ $p_3=\frac{\frac{55}2}{X+3}\frac{18}{X+4}\frac{\frac{11}2}{X+5}\frac1{X+6}$ $p_2=\frac{\frac{481}{12}}{X+2}\frac{\frac{55}2}{X+3}\frac{9}{X+4}\frac{\frac{11}6}{X+5}+\frac{\frac14}{X+6}$ $p_1=\frac{\frac{6771}{120}}{X+1}\frac{\frac{481}{12}}{X+2}\frac{\frac{55}4}{X+3}\frac{3}{X+4}+\frac{\frac{11}{24}}{X+5}+\frac{\frac{19}{20}}{X+6}$
Putting these values into this equation $X=20p_1+20p_2+21p_3+23p_4+26p_5+30p_6$ $\text(or)$ $p_1+p_2+p_3+p_4+p_5+p_6=1$
We get a six degree equation for $X$, so the root is mostly unsolvable, using desmos I was able to approximate $X$ to $23.7521552886$, the other roots were negative
Log in to reply
I did notice a pattern coming up while I was doing the partial fractions, the numerators of the fractions for a given denominator first remain the same then in the next iteration is divided by $2$ in the next divided by $3$, divided by $4$ then divided by $\frac{5}{19}$
Log in to reply
I wonder why the pattern was so obvious in the beginning and then started to churn out weird terms, any explanations
Log in to reply
Chris Lewis showed me how for a three faced die $p_1=p_2$, this extends to the six faced die as well due to symmetry of the scores made by $1$ and $2$, maybe using this information we can reduce the calculation of X by a big factor
Log in to reply
Using this information, the equation for X reduces to a five degree equation, so it still may be unsolvable but now is more likely to be solvable in radicals
Log in to reply
@Sarah Alam See this is how I can call you
Log in to reply
Oh, cool I got a notification. That’s useful. But it’s weird how you know my ‘mentioning ID’ and I don’t know what it is.
Log in to reply
Just do this @ me here then type my name like this and you will get the mention id ( I typed it and shows my mention Id )@Jason Gomez mention[4821179:Jason Gomez]
Log in to reply
Log in to reply
Log in to reply
@Sarah Alam WAS IT SUPPOSED TO DO SOMETHING DIFFERENT?!?!?!
Log in to reply
Log in to reply
Log in to reply
@Sarah Alam $\text{mention[11184426:Sarah Alam]}$
What I type is somewhat like thisLog in to reply
Log in to reply
@Sarah Alam <Your name again>
Log in to reply
Log in to reply
Log in to reply
Log in to reply
Log in to reply
link from Frisk Dreemurr
Check thisLog in to reply
Log in to reply
Log in to reply
Log in to reply
@Sarah Alam, just some text here, mention[11184426:Sarah Alam] , some more text here , mention[11184426:Sarah Alam] , text not needed mention[11184426:Sarah Alam]
Yeah infact you can do thisLog in to reply
Log in to reply
@Sarah Alammention[11184426:Sarah Alam]
Log in to reply
Log in to reply
@Sarah Alam, mention[11184426:Sarah Alam], to do this i just wrote 'at' your name and then repeated you name (without 'at'). here are some details:
to find the mention id: type @name as reply in a post from him/her and get the autocomplete, then write ", " and again the name but without @ (spelling has to be correct including caps)
Log in to reply
Log in to reply
Log in to reply
if someone has a page open where this comment is not shown as deleted then he/she can even answer it (but i would be the only one (besides the staff) who can read that answer).
sry for deleting the comments, i just tried to redo getting an id by memorizing how to do it (finally i had to look at my notes to figure how to do it)
and i deleted these tries.
my main mistake was to ignore the caps.
Log in to reply
Log in to reply
Log in to reply
Now using my mention id put it anywhere and an @ before it and now you can summon me wherever
Log in to reply
The anatidaephobia is … marvellous to think about (how do you get scared thinking a duck is watching you), anyways your comment was worded with your emotion perfectly in it so it was easy to figure it out
Log in to reply
This looks interesting  can I make sure I've understood correctly?
Alex's score starts at zero. Each round, two dice are rolled. If they match, the game ends. If not, Alex's score increases by the larger number on the two dice, and another round starts.
The dice are identical, but are biased.
Is that right?
If so, a couple of points to note: first, the "neatest" way to include the constraint on the sum of the probabilities being $1$ is to use Lagrange multipliers. This lets you combine the constraint and the function you want to optimise into a single function. The "neat" thing about it is that there's no "special" face on the dice: you don't have to choose, say, face six, and write $p_6=1p_1p_2p_3p_4p_5$. This is helpful because you preserve symmetry.
Secondly, intuitively, Alex wants a high score to be likely, but not so likely that they keep getting matching scores. The best result for Alex is to roll a six on one die, and not a six on the other; so he wants to maximise the probability of this happening. But...I'm not sure how it'll interact with other high scores. Better do some maths on that ;).
Log in to reply
I have completed the problem actually, I don’t know about langrange multipliers or langragian mechanics, have tried learning langragian mechanics but just keep getting stunned
Log in to reply
The solution is in the comments, took me a whole week (and the first maths problem I actually relied on a computer so heavily)
Log in to reply
If possible could you show me the proof through langrange multipliers or atleast a part of it
Log in to reply
And yes, all you have understood is correct
Log in to reply
Yipes. This is getting messy (but interesting). What probabilities did you get in the end? I'm finding something that might be worth investigating with the differences between $p_i$ and $p_{i+1}$. If there's any sort of more obvious pattern that might be an easier way in.
Log in to reply
Log in to reply
$p_6 = 0.201666062233$ $p_5=0.180913952711$ $p_4=0.165185227219$ $p_3=0.154455268174$ $p_2=0.148889744828$ $p_1=0.148889744827$
I think the last digit alone is inaccurate but it maybe the last two as well (highly unlikely), the probabilities add to 0.999999999993, so they are right
Log in to reply
$p_3p_2=0.00556552334673$ $p_4p_3=0.0107299590452$ $p_5p_4=0.0157287254919$ $p_6p_5=0.0207521095217$
Log in to reply
Log in to reply