... so a better solution would be to switch into the s-domain (Laplace transform).
The same three equations then become:
vg = vC1 + vC2 + R1*vC2*C2*s
vg = vC1 + vC3 + R2*vC3*C3*s
vC1*C1*s = vC2*C2*s + vC3*C3*s
The Heaviside function is quite simple to write in the s-domain:
vg = 1/s
Solve for desired quantity and perform an inverse Laplace transform, so you get the output function in the time domain.
This is how one would solve the circuit analytically. The process is rather tricky and time consuming, but (contrary to what some in this thread claim) it can be done. Kirchoff's laws apply. Always. It is a good idea to help yourself solve problems like these with programs like Matlab.
We could simplify our life and approach the problem by slapping everything into SPICE and running the simulation. This is what you get, if you choose C1=2µF, C2=3µF and C3=1µF:
Interesting, huh?
Yes, so interesting I calculated the closed solution as learned in EE circuits class using a ti89 calculator.
I wrote current equations rather than voltage equations since the arithmetic is easier. Using KVL and KVC I get:
(I1 + I2)/sC1 + I2/sC2 + RI2 = 1/s
(I1 + I2)/sC1 + I3/sC3 + RI3 = 1/s
I1 = I2 + I3
Substituting:
I2(833333 + 1000s) + I3(500000) = 1
I2(500000) + I3(1500000 + 1000s) = 1
Solving for I2 and I3:
I2 = (s + 1000)/(1000s^2 + 2333333s + 9999999500)
I3 = (s + 333.333)/(100s^2 + 2333333s + 9999999500)
Expanding in partial fractions:
I2 = .000639/(s + 1767.59) + .000361/(s + 565.471)
I3 = .001193/(s + 1767.59) + .000193/(s + 565.471)
Finally, inverse Laplace transforms gives the solution:
i2 = .000639(e^(-1767.59t)) + .000361(e^(-565.471t))
i3 = .001193(e^(-1767.59t)) - .000193(e^(-565.471t))
Switching over to voltages and doing more of the same gives:
v1 = .51822(1 - e^(-1767.59t)) + .148478(1 - e^(-565.471t))
v2 = .120503(1 - e^(-1767.59t)) + .212695(1 - e^(-565.471t))
v3 = .67493(1 - e^(-1767.59t)) - .341146(1 - e^(-565.471t))
vn2 = .639e^(-1767.59t) + .361e^(-565.471t)
vn3 = 1.193e^(-1767.59t) - .193e^(-565.471t)
I attach the plot below to show that they agree with your plots and that the arithmetic is correct.
The good people at Mathworks want about $3000 for a Matlab license. One must have a good business using it in order to justify that kind of expense. Yes, LTSpice is an unbelievably good and free tool. And so easy to use. The above calculation is long and tedious, as you mentioned.
I took the circuits class a long (35 yrs) time ago. Back then it was one of the killer classes you had to pass to get your degree. Is it still so important today? It's old (goes back to Heaviside and about 1890), not very helpful in understanding the circuit and not used much, if any, in practice.