问题
This seems like an incredibly simple and silly question to ask, but everything I've found about it has been too complex for me to understand.
I have these two very basic simultaneous equations:
X = 2x + 2z
Y = z - x
Given that I know both X and Y, how would I go about finding x and z? It's very easy to do it by hand, but I have no idea how this would be done in code.
回答1:
This seems like an incredibly simple and silly question to ask
Not at all. This is a very good question, and it has unfortunately a complex answer. Let's solve
a * x + b * y = u
c * x + d * y = v
I stick to the 2x2 case here. More complex cases will require you to use a library.
The first thing to note is that Cramer formulas are not good to use. When you compute the determinant
a * d - b * c
as soon as you have a * d ~ b * c
, then you have catastrophic cancellation. This case is typical, and you must guard against it.
The best tradeoff between simplicity / stability is partial pivoting. Suppose that |a| > |c|
. Then the system is equivalent to
a * c/a * x + bc/a * y = uc/a
c * x + d * y = v
which is
cx + bc/a * y = uc/a
cx + dy = v
and now, substracting the first to the second yields
cx + bc/a * y = uc/a
(d - bc/a) * y = v - uc/a
which is now straightforward to solve: y = (v - uc/a) / (d - bc/a)
and x = (uc/a - bc/a * y) / c
. Computing d - bc/a
is stabler than ad - bc
, because we divide by the biggest number (it is not very obvious, but it holds -- do the computation with very close coefficients, you'll see why it works).
Now, if |c| > |a|
, you just swap the rows and proceed similarly.
In code (please check the Python syntax):
def solve(a, b, c, d, u, v):
if abs(a) > abs(c):
f = u * c / a
g = b * c / a
y = (v - f) / (d - g)
return ((f - g * y) / c, y)
else
f = v * a / c
g = d * a / c
x = (u - f) / (b - g)
return (x, (f - g * x) / a)
You can use full pivoting (requires you to swap x and y so that the first division is always by the largest coefficient), but this is more cumbersome to write, and almost never required for the 2x2 case.
For the n x n case, all the pivoting stuff is encapsulated into the LU decomposition, and you should use a library for this.
回答2:
@Alexandre , you missed one condition.. Here is the final code
void SolveLinearEquations (float a,float b,float c,float d,float u,float v, float &x, float &y)
{
float f;
float g;
if (abs(a) > abs(c))
{
f = u * c / a;
g = b * c / a;
y = (v - f) / (d - g);
if(c != 0)
x = (f - g * y) / c;
else
x = (u - b * y)/a;
}
else
{
f = v * a / c;
g = d * a / c;
x = (u - f) / (b - g);
if (a != 0)
y = (f - g * x) / a ;
else
y = (v - d * x)/c;
}
}
回答3:
(1) ax + by = c
(2) dx + dy = f
(3)1*d adx + bdy = cd
(4)2*b abx + bdy = fb
(3)-(4) adx - abx = cd - fb
x(ad-ab) = cd - fb
x = (c*d - f*b)/(a*d-a*b) //use this equation for x
ax + by = c
by = c - ax
y = (c - a*x)/b //use this equation for y
回答4:
The following function might be useful for some:
function solve(s1,s2){ //only works if coefficients > 0
str=s1 + " " +s2
str=str.replace(/[^0123456789.-]/g, ' ') //eliminate letters
str=str.replace( /\s\s+/g, ' ' ) //no double spaces
var n=str.split(" "); //put into an array
var a=0,b=1,c=2,d=3,e=4,f=5 //see what were doing
var x = ( n[c]*n[e] -n[b]*n[f])/(n[a]*n[e] - n[b]*n[d])
var y= (n[c]-n[a]*x)/n[b]
return({x:x, y:y})
}
To use:
result=solve("12x + 2y =32", "9x -5y=55")
alert (result.x+" ----- "+result.y)
回答5:
Algorithm to sove: Ax + By = C, Dx + Ey = F
START
1) PRINT "Enter A"
2) INPUT A
3) PRINT "Enter B"
4) INPUT B
5) PRINT "Enter C"
6) INPUT C
7) PRINT "Enter D"
8) INPUT D
9) PRINT "Enter E"
10) INPUT E
11) PRINT "Enter F"
21) INPUT F
22) detS <-- AE - BC
22) detX <-- CE - BF
24) detY <-- AF - CD
25) x <-- detX / detS
26) y <-- detY / detS
27) PRINT x, y
END
来源:https://stackoverflow.com/questions/11609107/solving-a-simultaneous-equation-through-code