Given #{a,b,c} in [-L,L]# What is the probability that the roots of #a x^2+b x + c = 0# be real?

1 Answer
Nov 19, 2016

#(41+6ln(2))/72~~0.6272#

Explanation:

We will operate on the assumptions #L>0# and that the probability is equal that #(a, b, c) = (x_1, x_2, x_3)# for all #(x_1, x_2, x_3) in [-L, L]^3#

Initial observations:

  • #ax^2+bx+c = 0# has real roots if and only if the discriminant of the quadratic is nonnegative, that is, #b^2-4ac >= 0#, or #b^2>=4ac#.

  • We can partition the probability space into the two cases in which #"sgn"(a) = "sgn"(c)# or #"sgn"(a)!="sgn"(c)#, and
    #P("sgn"(a)="sgn"(c)) = P("sgn"(a)!="sgn"(c)) = 1/2#

  • If #"sgn"(a) != "sgn"(c)#, then #4ac <= 0#, meaning # b^2>=4ac#.

  • By symmetry:
    #P(b^2>=4ac|a, c<=0) = P(b^2>=4ac|a, c>=0)#

  • As #b^2=(-b)^2# for all #b#, we can restrict #b# to #[0, L]# without changing the probability of it falling within a certain range. Thus
    #P(b^2>=4ac|a, c>=0) = P(b^2>=4ac|a, b, c>=0)#

With these, we can reformulate the problem as follows:

#P(ax^2+bx+c" has real roots") = P(b^2>=4ac)#

#=1/2P(b^2>=4ac|"sgn"(a)!="sgn"(c))#

#+1/2P(b^2>=4ac|"sgn"(a)="sgn"(c))#

#=1/2(1)+1/2P(b^2>=4ac|a, c<0 or a, c >0)#

#=1/2+1/2P(b^2>=4ac|a, c>0)#

#=1/2+1/2P(b^2>=4ac|a, b, c>0)#

#=1/2+1/2P(b^2>=4ac|(a, b, c) in [0, L]^3)#

We will now calculate #P(b^2>=4ac|(a, b, c) in [0, L]^3)#


If we consider a coordinate system in 3-space, then #[0, L]^3# is equivalent to a cube with side length #L# resting in the first octant and having sides colinear with the axes. Let #S# be the solid bounded by this cube and above by the surface #y^2 = 4xz#, and #V_S# be the volume of this solid. Then the probability that #b^2>=4ac# given an arbitrary #(a, b, c) in [0, L]^3# is equal to the probability that #(a, b, c)# falls within #S# given that #(a, b, c)# falls within the cube, i.e. #V_S/V_"cube" = V_S/L^3#

To find #V_S#, we will first consider the area #A_b# of the slice of #S# found by fixing #y=b# for some #b in [0, L]#, and then we will add up all of these areas by integrating #A_b# as #b# goes from #0# to #L#.

If we fix #y=b# for some #b#, then #A_b# is the area bounded by the lines #x=0, x=L, z=0, z=L# and the curve #4xz=b^2#, which we can rewrite as #z=b^2/(4x)#

desmos.com

Notice that the curve intersects the square at the points where #x=L# or #z=L#, that is, at #(x, z) in{(L, b^2/(4L)), (b^2/(4L), L)}#

With that, we can now set up our integrals. The upper bound from #x=0# to #x=b^2/(4L)# is the line #z=L#. The upper bound from #x=b^2/(4L)# to #x=L# is the curve #z=b^2/(4x)#. Thus

#A_b = int_0^(b^2/(4L))Ldx + int_(b^2/(4L))^Lb^2/(4x)dx#

and

#V_S = int_0^LA_bdb#

Omitting the full process of integration to save space, we find the result

#V_S = (5+6ln(2))/36L^3#

Thus

#P(b^2>=4ac|a, b, c in [0, L]) = V_S/L^3= (5+6ln(2))/36#

Substituting this into our initial equation, we get our final result:

#P(ax^2+bx+c" has real roots")#

#=1/2+1/2P(b^2>=4ac|(a, b, c) in [0, L]^3)#

#=1/2+1/2((5+6ln(2))/36)#

#=(41+6ln(2))/72#

#~~0.6272#