1. Basic Problem Varying Backside Case Missing Nodes Case
Finding the Shape of a Resistor Grid from
Boundary Measurements
Esther Chiew, University of Illinois
Vincent Selhorst-Jones, Pomona College
Rose-Hulman REU, Summer 2008
July 23, 2008
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
2. Basic Problem Varying Backside Case Missing Nodes Case
Basic Problem
Introduction
Fudamental Equations
Example of Propagation
Remarks on Propagation
Varying Backside Case
The Model
The Algorithm
Instability
Missing Nodes Case
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
3. Basic Problem Varying Backside Case Missing Nodes Case
Introduction
Basic Problem
Given (some of) the
boundary data on a grid of
uniformly 1 ohm resistors,
what can we determine
about the shape of the grid?
For example, might know
voltage and current on front
nodes while knowing the
sides are insulated and the
back nodes are at 0 volts.
What is the height of each
column of the grid?
Figure: A rectangular resistor grid.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
4. Basic Problem Varying Backside Case Missing Nodes Case
Introduction
Compare the below. Let sides be insulated. If we set same current
and voltage on the bottoms, we’ll get differing results on the tops.
Figure: An undamaged grid. Figure: One missing node.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
5. Basic Problem Varying Backside Case Missing Nodes Case
Fudamental Equations
Fundamental Equations
The resistor network is governed by two laws:
Ohm’s Law: V = IR,
Kirchoff’s Current Law: node i = 0.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
6. Basic Problem Varying Backside Case Missing Nodes Case
Fudamental Equations
Looking at any inner node (recall R = 1 uniformly), we have:
v1 − v = i1
v2 − v = i2
v3 − v = i3
v4 − v = i4
node i = 0.
Summing the center equations, this gives the relation:
v1 + v2 + v3 + v4
4
= v,
i.e., every inner node is the average of its surrounding nodes.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
7. Basic Problem Varying Backside Case Missing Nodes Case
Example of Propagation
Example of Propagation
The next few slides are an example of how data is propagated
inwards. For it, we will assume
Know voltage and current on front nodes,
Sides are insulated (notice that this is equivalent to simply
omitting the resistors sticking out of the sides).
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
8. Basic Problem Varying Backside Case Missing Nodes Case
Example of Propagation
From voltage and
current on front
nodes, can use
Ohm’s Law to obtain
voltage on next row.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
9. Basic Problem Varying Backside Case Missing Nodes Case
Example of Propagation
Can obtain new
voltage by using rule
that each node is the
average of its
surrounding nodes.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
10. Basic Problem Varying Backside Case Missing Nodes Case
Example of Propagation
Do this this for the
enitre row.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
11. Basic Problem Varying Backside Case Missing Nodes Case
Example of Propagation
Continue to
propagate this
information upwards.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
12. Basic Problem Varying Backside Case Missing Nodes Case
Example of Propagation
Information can be
pushed up arbitrarily
high by this method.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
13. Basic Problem Varying Backside Case Missing Nodes Case
Remarks on Propagation
The following things should be noted about how the propagation
works:
Each row is entirely determined by the two rows under it.
This process uses only linear operations, and thus can be
represented by a propagation matrix acting on a voltage
vector whose length is twice the width of the grid.
The process is highly unstable, i.e. small differences in initial
data will give wildly varying results. This comes from the fact
that the propagation matrix has eigenvalues greater than 1.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
14. Basic Problem Varying Backside Case Missing Nodes Case
The Model
Varying Backside Case
For recovering a varying backside, we use the following model:
We know the width of the grid.
Sides of grid are insulated.
Voltage and current known on
front nodes.
Every node above the “height” of a
column is at 0 volts.
Current can leave through 0 volt
nodes in either direction. (They
still affect adjacent inner nodes.)
Figure: Heights: 3, 3, 2, 3, 4.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
15. Basic Problem Varying Backside Case Missing Nodes Case
The Algorithm
Fundamental Theorem
To implement the algorithm we will use, we need the following:
Theorem
If the voltage of every node on the front is greater (less) than 0,
then the voltage of all nodes below the backside must be as well.
Proof
Furthermore, it’s reasonable to expect we have control over the
front voltages (say, because we’re the ones inducing them with a
voltage generator), so we can assume the front voltages are strictly
positive for purposes of the algorithm.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
16. Basic Problem Varying Backside Case Missing Nodes Case
The Algorithm
The Algorithm
1. Use the voltage and current for the front nodes to determine
the voltages on the first inner row.
2. Using the top two rows found so far and the fact that each
node is the average of its surrounding nodes, calculate the
voltages for the next row above.
3. If any of the nodes in the latest row are less than or equal to 0
(or some error term, say 5 · 10−3), set that node and all the
above nodes in that column to 0.
4. Repeat from Step 2 until you obtain a row that is all 0’s.
5. The first row a column contains a 0 in is the column’s height.
Example
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
17. Basic Problem Varying Backside Case Missing Nodes Case
Instability
Given starting data with no error and a computational method that
introduces no error, the algorithm works perfectly. However, since
we’re dealing with the real world, let’s look at what error does to
the algorithm:
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
18. Basic Problem Varying Backside Case Missing Nodes Case
Instability
The following is an example of a ten-wide grid with 10 volts put on
each front node and the following column heights:
9 10 9 8 6 5 6 7 8 9
Putting voltage into the system induces currents on the front.
Feed the voltage and currents into the algorithm, we get:
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
19. Basic Problem Varying Backside Case Missing Nodes Case
Instability
The following is an example of a ten-wide grid with 10 volts put on
each front node and the following column heights:
9 10 9 8 6 5 6 7 8 9
Putting voltage into the system induces currents on the front.
Feed the voltage and currents into the algorithm, we get:
9 ∞ 9 8 6 5 6 7 9 9
Not bad. But what if we introduce a tiny error of, say, 0.01% to
the reading of the voltage from the eighth node?
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
20. Basic Problem Varying Backside Case Missing Nodes Case
Instability
The following is an example of a ten-wide grid with 10 volts put on
each front node and the following column heights:
9 10 9 8 6 5 6 7 8 9
Putting voltage into the system induces currents on the front.
Feed the voltage and currents into the algorithm, we get:
9 ∞ 9 8 6 5 6 7 9 9
Not bad. But what if we introduce a tiny error of, say, 0.01% to
the reading of the voltage from the eighth node?
9 ∞ 8 ∞ 6 6 6 ∞ 6 ∞
Not good.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
21. Basic Problem Varying Backside Case Missing Nodes Case
Instability
Clearly, small errors can cause extreme damage to the algorithm.
With every new row that is generated, the initial errors become
considerably greater. So what can we do about this instability?
Shape: The more rows the data has to propagate through, the
worse the error gets. Thus, short, squat grids can be
recovered, but tall ones work much less well.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
22. Basic Problem Varying Backside Case Missing Nodes Case
Instability
Clearly, small errors can cause extreme damage to the algorithm.
With every new row that is generated, the initial errors become
considerably greater. So what can we do about this instability?
Shape: The more rows the data has to propagate through, the
worse the error gets. Thus, short, squat grids can be
recovered, but tall ones work much less well.
Regulation: The growth in error is caused by the large
eigenvalues of the propagation matrix. If we manually shrink
these eigenvalues, error propagation decreases.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
23. Basic Problem Varying Backside Case Missing Nodes Case
Instability
Clearly, small errors can cause extreme damage to the algorithm.
With every new row that is generated, the initial errors become
considerably greater. So what can we do about this instability?
Shape: The more rows the data has to propagate through, the
worse the error gets. Thus, short, squat grids can be
recovered, but tall ones work much less well.
Regulation: The growth in error is caused by the large
eigenvalues of the propagation matrix. If we manually shrink
these eigenvalues, error propagation decreases.
Voltage: Choosing the initial voltage well can allow for better
seeing. If we have an expected backside, we can use that to
generate a voltage pattern that might let us recover more.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
24. Basic Problem Varying Backside Case Missing Nodes Case
Instability
Regulation is far from perfect, though. Heavy regulation wipes out
large eigenvalues – Not a problem for perfectly rectangular grids,
but large eigenvalues are what primarily propagates the information
of a varied backside.
Nonetheless, the process can still be improved through the use of
light regulation. In addition, we can use the algorithm to obtain a
hypothetical backside, from which we can find a voltage vector
that gives a more “distinct” current. Using this new current, we
can repeat the whole process a few times and improve the
algorithm slightly.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
25. Basic Problem Varying Backside Case Missing Nodes Case
The End! Thanks for listening!
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
26. Proof of Theorem Example Use of Algorithm
Appendix Index
Proof of Backside Theorem
Example of Backside Algorithm
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
27. Proof of Theorem Example Use of Algorithm
Theorem
If the voltage of every node on the front is greater (less) than 0,
then the voltage of all nodes below the backside must be as well.
Proof.
Let the voltage of every front node be greater than 0. Let p denote
an interior node with the lowest voltage obtained by an interior
node. Assume p is non-positive. Then since p is the average of its
surrounding nodes and also has the lowest possible voltage, every
node surrounding p must have equal voltage. Moving down the
column p is contained in, we eventually hit the front, implying that
one of the front nodes is not strictly positive. ⇒⇐
Back Appendix Index
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
28. Proof of Theorem Example Use of Algorithm
Let’s look at an example
of using the backside
algorithm. To begin with,
recall that we know the
current and voltage on the
front nodes.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
29. Proof of Theorem Example Use of Algorithm
Using this information, we
can determine the voltages
in the second row as well.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
30. Proof of Theorem Example Use of Algorithm
As noted earlier, knowing
two rows gives the next
one above.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
31. Proof of Theorem Example Use of Algorithm
Continue to propagate in
this manner until we find a
node with a voltage of 0.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
32. Proof of Theorem Example Use of Algorithm
By theorem, we know that
node is the “top” of its
column. Thus we set all
above nodes to 0 as well.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
33. Proof of Theorem Example Use of Algorithm
Continue to propagate
upwards, using this
information.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
34. Proof of Theorem Example Use of Algorithm
When new 0’s are
encountered, repeat the
process of setting above
nodes to 0.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
35. Proof of Theorem Example Use of Algorithm
Repeat until the top of
each column is found.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
36. Proof of Theorem Example Use of Algorithm
A column’s height is the
first row in it that has a 0.
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones
37. Proof of Theorem Example Use of Algorithm
The algorithm is complete.
We now have the height of
each column.
Back
Appendix Index
Finding the Shape of a Resistor Grid from Boundary Measurements Chiew & Selhorst-Jones