Non-Linear Infinite Resistor Ladder

  • Context: Undergrad 
  • Thread starter Thread starter bob012345
  • Start date Start date
  • Featured
Join the discussion
Registration is free. Ask a follow-up in this thread, or start your own.
53 replies · 4K views
lrg said:
I don't have an immediate answer, but I suspect that this is one of those computations where the initial condition matters a lot, if not most, and so only the 'right' numerical value can give the exact final value, but I'm just speculating...
Using spice to solve the 10 stage ladder.
With a shorted tail = 1.099990
With an open tail = 1.900991
The actual result must lie between those bounds.
Your result, √26 - 4 = 1.099019513592785 is on the wrong side, and slightly outside the bounds.
The tail will look like a straight wire, with resistors to ground of 1e10, 1e11, 1e12, ... = 0.900e10.
You seem to have chosen the wrong tail to the ladder.
 
Physics news on Phys.org
bob012345 said:
Doing that derives the limiting term as ##0.9⋅10^n## discussed in post #8.
I think an obstacle
Baluncore said:
@lrg
Welcome to PF.


1.099019513592785, is very different from
1.9009910801792881, the numerical result evaluated over more than 100 terms.

Your result is for the CF = 1+1/10+1/10+1/10+1/10+1/10+1/10+1/10+ 1/. . .

The CF we need is = [1, 1, 9, 9, 1, 99, 99, 1, 999, 999, 1, 9999, 9999, 1, ...
According to Google, this number is an example of an irrational number that is extremely well approximated by rational fractions at specific truncation steps.
Is that a Roth- Thue- Siegel result?
 
WWGD said:
Is that a Roth- Thue- Siegel result?
No idea. I took the numerical solution and then converted it to a CF.
There looked to be a pattern, so I used Google search to confirm it.
If the CF pattern had been cyclic, repeating the same coefficients, then the solution would have been a square root.
 
Reply
  • Like
Likes   Reactions: WWGD
lrg said:
Hello everyone,

I think the exact value should be $$R_{\infty} = \sqrt{26} - 4 = 1.099019513592785,$$ that is the solution of the equation $$ x = 1+ \frac{1}{\frac{1}{1}+\frac{10}{x}},$$ or $$x^2+8x-10 = 0,$$ since the equivalent resistance at the right of the first vertical resistor is equal to that at the input divided by 10 (it is an infinite ladder ...). Do you agree?
Welcome to PF!

I believe you are solving a different ladder that goes like this:

IMG_6331.webp

The ladder being studied in this thread has the same top resistances but the shunts increase by ten for each stage. In your case where each stage changes both resistors by the same factor, it can be solved exactly and gives the answer you gave.
 
bob012345 said:
Welcome to PF!

I believe you are solving a different ladder that goes like this: ...
Yes, you are right, very stupid mistake on my side. When I realized it, I decided to simulate the circuit (100 stages) with a Matlab/Octave script, changing the resistance at infinity (far right), and I got this graph:
fig1.webp

It shows that the input resistance can go from about 1.1 (short) to about 1.9 (open), as reported also by @Baluncore.
Then, just for fun, I tried to obtain my value (1.099), and I got it, with the help of a 'slightly negative' resistance (~-1.2e-3)! Look at this other graph:
fig2.webp

where I extended the range of the resistance at infinity to negative values (here the horizontal scale is linear, not logarithmic). It is even possible to get a short (0 value) as equivalent resistance ...
I think there is something interesting here, namely the infinite ladder cannot be said to have any equivalent input resistance, UNLESS we specify also the behavior at infinity, with a resistance at the far right (kind of boundary condition). Changing that value, we get different equivalent resistances. Of course, negative values are only for fun, but any value between 0 and ##+\infty## (short/open) should be considered 'reasonable', without preferences.
 
Reply
  • Like
Likes   Reactions: OmCheeto
lrg said:
Yes, you are right, very stupid mistake on my side. When I realized it, I decided to simulate the circuit (100 stages) with a Matlab/Octave script, changing the resistance at infinity (far right), and I got this graph:

Which circuit are you simulating with 100 stages? Our original ladder in post #1 or your ladder? Not sure what you mean by “changing the resistance at infinity”? It seems to me simulating a 100 stage ladder, it will make no difference what resistors you add or how you add them at the final stage.
 
bob012345 said:
It seems to me simulating a 100 stage ladder, it will make no difference what resistors you add or how you add them at the final stage.
It makes a difference for a zero ohm termination = tail.
Changing the tail, the starting value for evaluation, of a 100 term ladder.
1.900991080179288 ; Tail = 1e+201 = infinite, open
1.900991080179288 ; Tail = 0.8e+101 = good approximation
1.900991080179287 ; Tail = 1P = 1e15
1.900991080178494 ; Tail = 1T = 1e12
1.900991079385424 ; Tail = 1G = 1e9
1.900990286316557 ; Tail = 1M = 1e6
1.900198002672590 ; Tail = 1k = 1e3
1.502282916827386 ; Tail = 1 ohm
1.099990001000000 ; Tail = 1e-201 = zero, shorted
 
Last edited:
Reply
  • Like
Likes   Reactions: bob012345
Yes @Baluncore, that is what I found, and I put in the figure above. In fact, the continued fraction that computes the equivalent resistance of the original @bob012345's ladder should be:
$$R = 1 + \cfrac{1}{\frac{1}{1} + \cfrac{1}{b + \cfrac{1}{b + \cfrac{1}{b^2 + \cfrac{1}{b^2 + \cfrac{1}{b^3 + \cfrac{1}{b^3 + \dots}}}}}}}$$
with ##b=1/10##.
With that CF, the Seidel-Stern Theorem (search with Google) FAILS: it states that, to have a convergent continued fraction, the sum of the coefficients ##1, b, b, b^2, b^2 \dots## must diverge. Here the sum is 11/9 (when ##b=1/10##), so it cannot converge. This is consistent with the numerical result that shows that the truncated ladder fraction (here with 4 stages):
$$R = 1 + \cfrac{1}{\frac{1}{1} + \cfrac{1}{b + \cfrac{1}{b + \cfrac{1}{b^2 + \cfrac{1}{b^2 + \cfrac{1}{b^3 + \cfrac{1}{b^3+ \cfrac{1}{r_{\infty}}}}}}}}},$$
where ##r_{\infty}## is the terminating resistance at the right, is a variable function of ##r_{\infty}## even when the number of stages grows without limit. I understand @bob012345 when he writes in the post #36:

It will make no difference what resistors you add or how you add them at the final stage
because I used to think the same thing. But now I have realized that, when the continued fraction of the ladder does not converge, that's wrong: the terminating resistance matters, even when the ladder is infinite!
If you want a 'positive' example, take the 'golden ladder', the one with all the resistors equal to 1. The equivalent resistance is the golden ratio ##\frac{1+\sqrt{5}}{2}##, and the continued fraction converges to that value because the sum ##1+1+1+\dots## diverges. In that case, if you simulate the ladder by changing the terminating resistance, you always get the same result (the golden ratio) regardless of the value of that resistance.
It seems paradoxical, but we must remember that, physically, infinite ladders don't exist, so the problem we are discussing is purely mathematical, and in that case the terminating resistance is a kind of boundary condition, a 'hidden' parameter of the infinite ladder that is as important as the parameter ##b=1/10## that defines it.

A very interesting problem indeed.
 
Last edited:
Reply
  • Like
Likes   Reactions: bob012345
Baluncore said:
It makes a difference for a zero ohm termination = tail.
Ok, but then changing the tail to low values is changing the problem significantly. It no longer the ‘tail’ but a different ladder.
 
I was wrong to say changing the tail doesn’t change the ladder but like I said, it literally changes the ladder to be a different problem if it has no relationship to the actual infinite tail. In the original analysis above our tail value of ##0.9 x10^n## for the ##n^{th}## stage does and was derived from a recursion relation. All it did was make convergence faster. So our original convergence is not wrong.
 
Reply
  • Agree
Likes   Reactions: Baluncore
I like this problem. It overturns every assumption we would normally make.
 
Reply
  • Like
Likes   Reactions: bob012345
lrg said:
If you want a 'positive' example, take the 'golden ladder', the one with all the resistors equal to 1. The equivalent resistance is the golden ratio ##\frac{1+\sqrt{5}}{2}##, and the continued fraction converges to that value because the sum ##1+1+1+\dots## diverges. In that case, if you simulate the ladder by changing the terminating resistance, you always get the same result (the golden ratio) regardless of the value of that resistance.
It seems paradoxical, but we must remember that, physically, infinite ladders don't exist, so the problem we are discussing is purely mathematical, and in that case the terminating resistance is a kind of boundary condition, a 'hidden' parameter of the infinite ladder that is as important as the parameter ##b=1/10## that defines it.

A very interesting problem indeed.
If one takes the Golden Ladder equation ##R^2 -R -1=0## and instead of solving it exactly, write it as ##R = \sqrt{1+R}## then start with any terminating resistance under the radical and iterate you will get the Golden Ratio ##\phi##. Then the number of iterations you do for a specific desired precision is the number of stages you would need of a finite ladder to converge to.
 
bob012345 said:
Then the number of iterations you do for a specific desired precision is the number of stages you would need of a finite ladder to converge to.
In post #17, I gave an accelerator that works well on an alternating series like the golden ratio. The accelerator takes three successive alternating approximations, then gives a new best guess, and so doubles the number of digits per application.
 
Reply
  • Like
Likes   Reactions: bob012345
Baluncore said:
Simply fixed value ladders always do converge to a recurring continued fraction, and since an RCF can be evaluated with a quadratic equation, they will contain the square root of a non-square number, which is irrational.
In this case the ladder is exponentially converging, and so I expect it cannot be written as an RCF, but will almost certainly evaluate to an irrational number.

This code evaluates from right to left. A solution accurate to n digits requires n terms.
Code:
Dim As Integer i, n = 20
Dim As Double term, r = 10^(n+1)
For i = n To 0 Step -1
    term = 10^(-i)      ' this term
    r = term + 1 / r    ' add the term as conductance in parallel
    r = term + 1 / r    ' add the term as resistance in series
Next i
That gives an upper bound.
Since the horizontal resistances are so small beyond n=20, how about taking them as zero, giving an initial ##r_n=9*10^n## to get a lower bound?
 
haruspex said:
Since the horizontal resistances are so small beyond n=20, how about taking them as zero, giving an rn
= 9 * 10n initial to get a lower bound?
A lower bound for n ?
Using rn = 1.0 * 10100, r is correct to 16 digits for n > 13.
Using rn = 10. * 10n, r is correct to 16 digits for n > 12.
Using rn = 9.5 * 10n, r is correct to 16 digits for n > 12.
Using rn = 9.1 * 10n, r is correct to 16 digits for n > 11.
Using rn = 9.001 * 10n, r is correct to 16 digits for n > 9.
Using rn = 9.000001 * 10n, r is correct to 16 digits for n > 6.
Using rn = 9.0 * 10n, r is correct to 16 digits for n > 3.
Using rn = 8.999999 * 10n, r is correct to 16 digits for n > 8.
Using rn = 8.999 * 10n, r is correct to 16 digits for n > 11.
Using rn = 8.5 * 10n, r is correct to 16 digits for n > 14.
Using rn = 8.0 * 10n, r is correct to 16 digits for n > 15.
 
Baluncore said:
A lower bound for n ?
A lower bound for the resistance in the infinite case.
The OP's original way of starting with a finite ladder, taking all vertical resistances to the right to be open circuit, gave an upper bound for the resistance of the infinite ladder.
 
haruspex said:
A lower bound for the resistance in the infinite case.
An initial Rn = zero, is not possible because it short-circuits the ladder.
That requires an immediate division by zero.
The lowest computed was 1.099990001000000 when Rn = 1e-200.

You will see a list of lower bounds by going back to my post #37.
 
Baluncore said:
An initial Rn = zero, is not possible because it short-circuits the ladder.
That requires an immediate division by zero.
The lowest computed was 1.099990001000000 when Rn = 1e-200.

You will see a list of lower bounds by going back to my post #37.
Take the limit of ##\frac{R_n R}{R_n+R}## as ##R_n→0## which gives zero. So if one shorts the first stage the minimum resistance is 1 unit.
 
Baluncore said:
An initial Rn = zero, is not possible because it short-circuits the ladder.
That’s not what I suggested.
If you take all the "horizontal" resistors (I am taking the ladder to be horizontal, as in post #1, so the rungs are vertical) to be zero after n, the vertical resistors thereafter are all in parallel and can be evaluated to ##9*10^n##. That gives a lower bound for the infinite case.
As you observed in post #45, this converges very quickly.
Baluncore said:
The lowest computed was 1.099990001000000 when Rn = 1e-200.
Why is that interesting? When developing lower bounds you want the highest, not the lowest.
Baluncore said:
You will see a list of lower bounds by going back to my post #37.
I did not understand the first one:
Baluncore said:
900991080179288 ; Tail = 1e+201 = infinite, open
Is that supposed to be 1e+101? Either way, it would be an upper bound, not a lower one.
We can find a somewhat faster converging upper bound too (faster than taking Rn as infinite). Consider an infinite ladder with horizontals X, 10X, 100X … and verticals Y, 10Y, 100Y… The first vertical is before the first horizontal.
Say this has resistance S. Removing the X and Y leaves a ladder which is the same but with every resistance ten times as large, so it has resistance 10S. Plugging the X and Y back in we have ##1/S=1/Y+1/(X+10S)##
##10S^2-(9Y-X)S-XY=0##.
With ##X=10^{-n-1}, Y+10^{n+1}##, the resulting S is an upper bound for Rn:
##10S^2-(-10^{-n-1}+9\cdot 10^{n+1})S-1=0##
 
Last edited:
haruspex said:
If you take all the "horizontal" resistors (I am taking the ladder to be horizontal, as in post #1, so the rungs are vertical) to be zero after n, the vertical resistors thereafter are all in parallel and can be evaluated to 9 * 10n. That gives a lower bound for the infinite case.
That was pointed out by bob012345 in post #8
The fastest convergence is when the evaluation starts with 9 * 10n.
Notice from my post #45:
8.999999 * 10n, and
9.000001 * 10n, both take longer to converge than
9.000000 * 10n.
9 * 10n is on target. So to what is it a lower bound ?

You know exactly what you are thinking. I can see too many alternative interpretations to your questions.

haruspex said:
Is that supposed to be 1e+101?
No, 1e+201 is a big double precision value, that is equivalent to an open circuit.
1e-201 is an extreme short circuit value, and does not throw a division-by-zero exception.
For initial seeds, those give the upper and lower bounds of possible evaluations for n = 100.

I'm more interested in the rapid convergence to the same value, of this truncated continued fraction.
r = 1+1/(1+1/(9+1/(9+1/(1+1/(99+1/(99+1/(1+1/(999+1/999))))))))
when compared to
r = 1+1/(1+1/(0.1+1/(0.1+1/(0.01+1/(0.01+1/(0.001+1/(0.001+1/(0.0001))))))))
 
@Baluncore, I think that evaluating the ladder from left to right, instead of right to left, can be better (for ##\vert b \vert<1##). A forward continued fraction like that associated with the input impedance can be evaluated in the forward direction with the Euler continuants, but there is an equivalent way of thinking which is more 'circuital' and easy to grasp, at least for me: transmission matrices.
Consider the n-th ##(n=0 \dots N)## ladder L-stage, built with a series resistance ##r_n=b^n## and a shunt conductance ##g_n=b^n##. This two-port network is described by
$$\begin{pmatrix} V_1 \\ I_1 \end{pmatrix} = T_n \begin{pmatrix} V_2 \\ I_2 \end{pmatrix}$$
with
$$T_{n} = \begin{pmatrix} 1+r_n g_n & r_n \\ g_n & 1 \end{pmatrix}$$
The T matrix of the whole N-stage ladder is
$$T = T_0 \cdot T_1 \cdot T_2 \cdots T_N$$
The key point is that the product is associative, so it can be evaluated left to right. Moreover ##T_n##
converges (for ##\vert b \vert<1##) to the identity matrix ##Id = \begin{pmatrix} 1 & 0 \\ 0 & 1 \end{pmatrix}## for ##n \to \infty##. This is nice since, when the arithmetic (possibly multiple) precision limit is reached, it is useless to continue multiplying: we have reached the limit value of N for that precision.
Once the ladder T matrix is computed, it is easy the find the input impedance (and other useful things, like Thevenin and Norton equivalents at the right port) as a function of the load impedance we can put to the right:
$$Z_{in} = \frac{A Z_{out}+B}{C Z_{out}+D}$$
this is one mode of getting the figure I posted in post #35.
The original @bob012345 question has the answer
$$Z_{in}(Z_{out} = \infty) = \frac{A}{C}$$
The weird thing is, the four coefficients A, B, C and D, as (complex) functions of the (complex) variable b, converge for ##\vert b \vert<1##, and diverge for ##\vert b \vert \geq 1##. Nonetheless, ##Z_{in} = \frac{A(b)}{C(b)}## does converge for any value of b!
For ##b=1/10##, I found, with 51 stages and 50 decimal places:
$$\begin{aligned}
A &= 2.1335712819576269021587791038159481189622017101529 \\
B &= 1.2235692617434733525281261949318481910338494096970 \\
C &= 1.1223468138295548210750991909766361996746916300643 \\
D &= 1.1123458037174224855980959026243117080347869560167
\end{aligned}$$
and
$$\begin{aligned}
Z_{\infty} = Z_{in}(Z_{out} = \infty) &= 1.9009910801792873151547772738317474368474495864595 \\
Z_0 = Z_{in}(Z_{out} = 0) &= 1.0999900009999999900009999000099999999000199970002 \\
Z_1 = Z_{in}(Z_{out} = 1) &= 1.5022829168273864624216950620116492911165292953097
\end{aligned}$$
An interesting fact is that ##Z_{in}(Z_{out})## can be written as:
$$Z_{in} = \frac{Z_{\infty}(Z_1-Z_0) Z_{out}+Z_0 (Z_{\infty}-Z_1)}{(Z_1-Z_0)Z_{out}+(Z_{\infty}-Z_1)}$$
 
Reply
  • Like
Likes   Reactions: bob012345 and Baluncore
Baluncore said:
That was pointed out by bob012345 in post #8
No, post #8 observed that it gave the fastest convergence seen. My posts on this thread are all primarily concerned with bounds. These are not the same thing. Unless proved otherwise, it could be that the sequence generated by starting with 9 * 10n oscillates around the limit. I showed that it does not, whereas starting with 9.000000001 * 10n might.
Baluncore said:
The fastest convergence is when the evaluation starts with 9 * 10n.
The fastest observed convergence occurs there. Has anyone proved it is fastest?
Baluncore said:
No, 1e+201 is a big double precision value, that is equivalent to an open circuit.
Ok, but as I wrote, it produces an upper bound, not a lower one. And if you want a converging sequence of upper bounds, I claim the formula I gave at the end of post #49 should be faster.
 
bob012345 said:
As n→∞, the horizontal resistors become vanishingly small and the rest of the ladder collapses to 0.9 ⋅ 10n.

haruspex said:
If you take all the "horizontal" resistors (I am taking the ladder to be horizontal, as in post #1, so the rungs are vertical) to be zero after n, the vertical resistors thereafter are all in parallel and can be evaluated to 9 * 10n.
The rung and rail ladder analysis was by bob012345 in post #8, you began it again, 41 posts later, in post #49. His obvious typo did not invalidate his analysis. I numerically verified his analysis and used the result in my numerical search of the real space between a short and an open circuit.

haruspex said:
The fastest observed convergence occurs there. Has anyone proved it is fastest?
I was working numerically, I did not claim a mathematical proof.
The fastest numerical convergence I found was centred in the range ( 9.0 ± 1e-9 ) * 10n.
No one has shown how it might converge faster elsewhere.
 
Baluncore said:
The rung and rail ladder analysis was by bob012345 in post #8, you began it again, 41 posts later, in post #49.
I read post #8 before I posted on the thread. The point of post #49 is to show that that starting value produces a lower bound. Post #8 did not show that. This, in principle, has nothing to do with rates of convergence, whether merely observed or proven.
If a mathematician wants to claim the answer lies in some small range, as is often the case, it is not enough to find a sequence that seems to converge quickly. Commonly, the approach is to find a sequence of upper bounds and a sequence of lower bounds. Calculating these further and further down the sequence (monotonic, hopefully) then produces tighter bounds.
Baluncore said:
No one has shown how it might converge faster elsewhere.
I am not suggesting the formula at the end of post #49 would converge faster than what you have found - in all probability it would not. But for the purpose I described above, you want both a reasonably fast converging sequence of lower bounds AND a reasonably fast converging sequence of upper bounds. The formula is an attempt at the latter.
If you have no interest in proving bounds, fine, but I find it interesting so I posted my finding.