Naturally linear Diophantine equations
Pyth, 92 bytes
I!%vzhK%2u?sm,ed-hd*ed/F<G2cG2@G1G+~Q,hQ_eQj9 2)J*L/vzhKtKeoSNm-VJ/RhK_*LdQsm+LdtM3/V*LhK_JQ
It's quite a monster.
Try it online: Demonstration. The input format is c\n[a,b] and the output format is [x,y].
In the case that no integer solution exists, I'll print nothing, and in the case that no natural integer solution exists, I'll simply print a random integer solution.
Explanation (Rough Overview)
At first I'll find an integer solution to the equation
ax + by = gcd(a,b)by using the Extended Euclidean algorithm.Then I'll modify the solution (my multiplying
aandbwithc/gcd(a,b)) to get an integer solution ofax + by = c. This works, ifc/gcd(a,b)is an integer. Otherwise there doesn't exist a solution.All the other integer solutions have the form
a(x+n*b/d) + b(y-n*a/d) = cwithd = gcd(a,b)for integern. Using the two inequalitiesx+n*b/d >= 0andy-n*a/d >= 0I can determine 6 possible values forn. I'll try all 6 of them and print the solution with the highest lowest coefficient.
Explanation (Detailed)
The first step is to find an integer solution to the equation ax' + by' = gcd(a,b). This can be done by using the extended Euclidean algorithm. You can get an idea on how it works at Wikipedia. The only difference is, that instead of using 3 columns (r_i s_i t_i) I'll use 6 columns (r_i-1 r_i s_i-1 s_i t_i-1 t_i). This way I don't have to keep the last two rows in memory, only the last one.
K%2u?sm,ed-hd*ed/F<G2cG2@G1G+~Q,hQ_eQj9 2) implicit: Q = [a,b] (from input)
j9 2 convert 9 to base 2: [1,0,0,1]
+ Q add to Q => [a,b,1,0,0,1]
this is the initial row
u ) start with G = ^ and update G repeatedly
by the following expression, until
the value of G doesn't change anymore
? @G1 if G[1] != 0:
cG2 split G into parts of 2
m map the parts d to:
, the pair
ed d[1]
-hd*ed/F<G2 d[0]-d[1]*G[0]/G[1]
s unfold
else:
G G (don't change it, stop criterion for u)
%2 take every second element
we get the list [gcd(a,b),x',y']
K store this list in K
~Q,hQ_eQ afterwards change Q to [Q[0],-Q[1]] = [a,-b]
This will be important for the other parts.
Now I want to find a solution to ax + by = c. This is possible only, when c mod gcd(a,b) == 0. If this equation is satisfied, I simply multiplying x',y' with c/gcd(a,b).
I!%vzhK...J*L/vzhKtK implicit: z = c in string format (from input)
%vzhK evaluated(z) mod K[0] (=gcd(a,b))
I! if not ^ than:
/vzhK c/K[0]
*L tK multipy ^ to each element in K[1:] (=[x',y'])
J and store the result in J, this is now [x,y]
We have an integer solution for ax + by = c. Notice, that x, y or both may be negative. So our goal is to transform these to non-negative.
The nice thing about Diophantine equations is, that we can describe all solution using only one initial solution. If (x,y) is a solution, that all other solutions are of the form (x-n*b/gcd(a,b),y+n*a/gcd(a,b)) for n integer.
Therefore we want to find a n, where x-n*b/gcd(a,b) >= 0 and y+n*a/gcd(a,b >= 0. After some transformation we end up with the two inequalities n >= -x*gcd(a,b)/b and n >= y*gcd(a,b)/a. Notice that the inequality symbol might look in the other direction due the division with a potential negative a or b. I don't care that much about it, I simply say that one number of -x*gcd(a,b)/b - 1, -x*gcd(a,b)/b, -x*gcd(a,b)/b + 1 definitly satisfies inequality 1, and one number of y*gcd(a,b)/a - 1, y*gcd(a,b)/a, y*gcd(a,b)/a + 1 satisfies inequality 2. It there is a n, that satisfies both inequalities, one of the 6 numbers also does.
Then I calculate the new solutions (x-n*b/gcd(a,b),y+n*a/gcd(a,b)) for all 6 possible values of n. And I print the solution with the highest lowest value.
eoSNm-VJ/RhK_*LdQsm+LdtM3/V*LhK_JQ
_J reverse J => [y,x]
*LhK multiply each value with K[0] => [y*gcd,x*gcd]
/V Q vectorized division => [y*gcd/a,-x*gcd/b]
m map each d of ^ to:
tM3 [-1,0,1]
+Ld add d to each ^
s unfold
these are the possible values for n
m map each d (actually n) of ^ to:
*LdQ multiply d to Q => [a*n,-b*n]
_ reverse => [-b*n,a*n]
/RhK divide by K[0] => [-b*n/gcd,a*n/gcd]
-VJ vectorized subtraction with J
=> [x+b*n/gcd,y-a*n/gcd]
oSN order the solutions by their sorted order
e print the last one
The sort by their sorted order thing works the following way. I'm using the example 2x + 3y = 11
I sort each of the 6 solutions (this are called keys), and sort the original solutions by their keys:
solutions: [1, 3], [4, 1], [7, -1], [-5, 7], [-2, 5], [1, 3]
keys: [1, 3], [1, 4], [-1, 7], [-5, 7], [-2, 5], [1, 3]
sort by key:
solutions: [-5, 7], [-2, 5], [7, -1], [1, 3], [1, 3], [4, 1]
keys: [-5, 7], [-2, 5], [-1, 7], [1, 3], [1, 3], [1, 4]
This sorts a complete non-negative solution to the end (if there is any).