Math Matters: Why Do I Need To Know This? Bruce Kessler, Department of Mathematics Western Kentucky University Episode Thirteen 1 Matrix Multiplication – Complicated probabilities Objective: To illustrate how matrix multiplication can turn extremely complicated real-world problems into very simple ones. We introduce the concepts of stochastic matrices and using tree diagrams to calculate probabilities. Hello and welcome to “Math Matters: Why Do I Need To Know This?” It’s a TV show where we take a look at the entry-level mathematics being taught in our general ed. courses and try to apply that to everyday life. I’ve got some cool stuff to show you today, and I want to apologize to the 109 students, the general math students, because this is a more algebraic show. Last time we started talking about matrix multiplication and all the things you could do with that, and I gave one example, and I thought “Man, I could do so much more with matrices and matrix multiplication,” so I’m going to revisit that topic today for a couple of segments. I’d like to talk about how we can use matrix multiplication to calculate probabilities. This is a slightly different idea, and I have to go back and recap a little bit of this introduction of what a matrix is. It’s just a rectangular set of values, its something along these lines. I call that a matrix. And we consider the number of rows, the first dimension of this, the number of columns the second dimension, and we would call that a 3 by 4 matrix. (Figure 1) We multiply matrices in this fashion. The middle dimension here has to match up. And it will generate in this case a 3 by 3 matrix in this fashion – it’s messy but I’ll explain it. But here’s what’s happening, you get this first entry in the first row, first column. I look at the first row of the first matrix and the first column of the second matrix and I take those entries, multiply them and add them up. So 3 times 1 plus 1 times 2 gives me that entry. (Figure 2) And to get the next one right here I take the first row, second column and do the same stunt. (Figure 3) To get this one, I take the second row, first column and add those things up. (Figure 4) And you just go through all nine of those entries and that’s how you get the answer with matrix multiplication. (Figure 5) Now, we showed that in the last episode, but I wanted to recap that because I’m getting ready to use that trick again. And what I’d like to do is take a look at some very, very complicated probability situations, and show you how all that stuff built into matrix multiplication takes care of all that stuff for you. Now, I’m going to throw a situation at you that’s fairly contrived, but the way I’m going to solve it is to arrange probabilities in a matrix. So each one of these would be probabilities and the first entry is going to be the probability that I go from one state back into that state. Ok, I’m going to think about different places that I could be in my problem. The second entry here will be from the first state to the second state and the first state to the other states. And then from the second state to the first one and so forth. I don’t mean the United States of America – I mean different places I could be in my problem. (Figure 6) 1 Figure 1, Segment 1 Figure 2, Segment 1 2 Figure 3, Segment 1 Figure 4, Segment 1 3 Figure 5, Segment 1 Figure 6, Segment 1 4 I’ll give you an example, it’s a really contrived example, but it’s an easy example. And that’s what I want to show you first, how you could use this in an easy example and then you can adapt that if you’d like into tougher situations. Let’s take a look at this idea, I’ve got three people, Amy, Brad, and Carol throwing a Frisbee to each other in the backyard. Now, Amy shares the Frisbee equally with Brad and Carol, but Brad is kind of sweet on Carol, so he throws her the Frisbee 90% of the time. And then Carol and Amy are buds, and Carol doesn’t think too much of Brad, so she will typically throw the Frisbee to Amy about 75% of the time. So let me ask a very complicated sort of question: if Brad is the first guy to hold the Frisbee and they toss this thing around 12 times, what is the probability that each of these people will have the Frisbee after those 12 tosses? (Figure 7) Figure 7, Segment 1 Now there’s a hard way, and there’s an easy way. The hard way would be, and I’m just going to look at 2 tosses, the hard way would be to list out all the different ways someone could end up with the Frisbee. And I’ve done this for two tosses. Brad starts and he’s going to throw it to either Amy or Carol, and then Amy is going to throw it to either Brad or Carol, and Carol’s going to either throw it to Amy or Brad. You can put the probabilities on this tree diagram, that’s what it’s called. For example, Brad kind of likes Carol, so there’s a 9 1 chance that he’ll throw the Frisbee to her, that means there’s a 10 chance that he’ll throw 10 the Frisbee to Amy, those have to add up to 1. And then with Amy, she shares the Frisbee equally so that’s 12 and 12 . And then Carol 34 of the time will throw it to Amy, and that means 1 of the time she will throw it to Brad. (Figure 8) 4 So you can look then and calculate the probabilities for each of these paths. So Brad, 1 Amy, Brad would be the product of those probabilities – so 20 , and the probability that it ends 5 Figure 8, Segment 1 1 up in Carol’s hands like this is 20 . And the probability that it ends up in Amy’s hands in 9 27 . So this fashion is 40 , and that it ends up in Brads hands going through Carol would be 40 27 the probability that it ends up in Amy’s hands after two tosses is just this one, so its 40 , the 1 probability that Carol has it is just this one, so its 20 . And then Brad, you’ve got to take . Okay, I’m sorry I didn’t mean both of these into account and add them together to get 11 40 to flash that away. So, that would be the answer for two tosses. To do this for twelve tosses, you’d have to make your table that wide. You’d have to make a huge table. You’d have to have all these probabilities listed, you’d have to multiply them all out. You’d have all these things, its horrible. (Figure 8) Now watch what I can do with a matrix. We create what’s called a transition matrix, where I put the probabilities in. The probability that Amy throws to Amy is 0, she’s not going to throw to herself, but she’s going to divide it equally between Brad and Carol, so I’ll put those probabilities in – notice that the sum of those entries is one. Brad will not throw the Frisbee to himself, there’s a 9 in 10 chance he’ll throw to carol and a 1 in 10 chance he’ll throw to Amy. Notice again the sum of those entries in the column are one. Carol will not throw the Frisbee to herself, there’s a 43 chance she’ll throw to Amy and a 14 chance she’ll throw to Brad, the sum of those entries is 1. We call this a stochastic matrix where all the entries are probabilities between 0 and 1, they could be 0 or 1, and then the sum of the columns is 1. (Figure 9) So to work the problem, after the first toss, what I would do, I would start with Brad holding the Frisbee, that means a 100% chance that Brad has it and just multiply. That 1 9 gives me just really the middle column, there’s a 10 chance that Amy has the Frisbee, a 10 6 Figure 9, Segment 1 chance that Carol has it. After two tosses, well, I do this again, and really what I could do to get this matrix, what I could do is, I multiply this by this. So I could square that one and do it that way, but I’ll just go through, do the matrix multiplication, and that gives me the probabilities. The easy way lies right here: to do this is so many tosses, I want to raise this matrix to a power. (Figure 10) So to do this for twelve tosses what I would do I would put this in my graphing calculator, I wouldn’t do it by hand – that’s horrible. But these graphing calculators can do this kind of stuff for you, and I would raise that matrix to a power, which is fairly messy thing, you can see it here, but then multiply by that original thing. Brad had the Frisbee to start out with, so here are my probabilities, this middle column: 33% chance that Amy has the Frisbee, 27% chance Brad has it, 40% chance Carol has it. And the cool thing that you’ll notice is that each of these columns is pretty much the same, which indicates that after a period of time, it really doesn’t matter who started with the Frisbee, there’s just going to be a probability that you have the Frisbee in your hand. (Figure 11) That would be a horrible problem without the matrices and you can adapt the situation to business situations and things like that. I’ve got a couple pages that summarize what I just said, I’m going to flash those up really quick and get ready for the next segment. 7 Figure 10, Segment 1 Figure 11, Segment 1 8 Summary page 1, Segment 1 Summary page 2, Segment 1 9 2 Matrix multiplication – Tracking populations by age group Objective: To illustrate how matrix multiplication can turn extremely complicated real-world problems into very simple ones. We introduce the concepts of Leslie matrices and using matrices to track populations by age group. I want to stay with this theme of matrix multiplication and show you another cool application for matrices and multiplying them and raising them to powers, which would be a horrible thing to do by hand, but we are armed with technology today, boys and girls, we’ve got these crazy things. And, using these things, we can solve some very complicated problems. And the application I’d like to show you now is how to track populations through age groups. Let’s say that you’re raising cattle and your herd of cattle, you’ve paid attention to the age distribution throughout the years and you know the following facts about your group of cattle. You know their birthrates going from one age group to another. Cattle of this age do not actually have any cows of their own, but when they grow up, they’ll grow at a rate of fifty percent, so however many are in this group, two years later, I’m doing things on a two year cycle, half of that number will reappear in this group as being born into that group. So you see as they get older they’re reproducing more and then as they get, and when they get very old, their reproductive cycle kind of drops off. Also we have studied the survival rates, as younglings there’s a slightly higher chance that they won’t survive, that means a lower rate that they do survive, that’s the 0.7. And then about 0.9, there’s about a 100% chance that they’ll survive to the next stage, there’s always sickness illnesses and these kind of things. But then as they get older, they start to die of old age, so the survival rate goes down. And then I’m assuming here that nothing will live past 12 years. There’s a 0% probability that they will live past 12 years. (Figure 1) So here’s my question: if I start with 40 head, and they’re all the same age, I go to the stock yards and I buy 40 head of cattle, they’re all 2 years old, what would be the age distribution of my herd in 30 years, okay? This could be important. Maybe I want to take a certain set of those just like I had at the stock yards and sell them off. Maybe there’s different things you can do with your animals at different age groups. So were just going to try and track them, but were going to track them by groups, and were going to do this on 2 year cycles. Well, let’s think about how you do this year to year, and I should say every 2 years to 2 years in my example. If I have a general kind of age distribution, so many in one group, so many in another and so forth, and I think about how you get into that first age group – those are the young ones, so they have to be born into that age group. So to get the total in the next group, in the next cycle that are in that youngest age group, I look at the birth rates for each of the groups and I multiply it by how many are in there. So 0 for the first one, 0.5 for the second, 1.1 and so on. And that’s a fairly messy calculation but I can clean it up a little bit using matrices, watch this. I could say, just take your birth rates and put them in a row in a matrix, take my age distribution, put that in a column and then all of a sudden, that’s matrix multiplication. (Figure 2) If I worry about how many survive to get into that second age group, well that’s the way 10 Figure 1, Segment 2 Figure 2, Segment 2 11 you get there, you’re not born into that second age group, you live to make it into that second age group. And I know the survival rates, I know that 70% of those in the first age group will then end up in the second age group after a 2 year cycle. And then I can phrase this, now this looks messier then this, but I have a method to my madness. I’m going to take my survival rate, fill in the rest with zeros and then I’ll do my matrix multiplication and that gives me that result. (Figure 3) And then the third age group would be a similar thing, 0.9 of those survive from the second age group into the third age group, which I could phrase as matrix multiplication. (Figure 4) Figure 3, Segment 2 So the way I would do this in general would be that I would construct a 6 by 6 matrix, if you have more age groups you’d have a bigger matrix, I have six age groups. And I’ll put those birth rates across the top, and then when I multiply by my age distribution that will give me that number in the first age group in the next cycle. And then 0.7 is the, now these are survival rates so I take 70% of the first group, fill in with zeroes. To get the second one I put 0’s here. You’ll notice that these are falling right below the main diagonal, okay? So I can just fill in the rest of my matrix like that. We call this a Leslie matrix after the mathematician who kind of developed this idea, you put the birth rates along the top and then along that main diagonal you still have zeroes, you have zeroes everywhere else except along that diagonal right below the main diagonal, you put in then the survival rates. And then I’ll use that to show you how you can plot the number of animals in each age group. (Figure 5) If I do this after one 2 year cycle what I’ll do is I’ll multiply this by my original distribution, which is right here. I have 40 in my second age group. And all that’s going to do is sort of peel this part off right here and multiply by 40, right? Zero times this and all those is going 12 Figure 4, Segment 2 Figure 5, Segment 2 13 to be zero. But 0.5 times 40 will give me 20 and .9 times 40 will give me 36, so that’s how many I would have in the age groups after the first 2 year cycle. (Figure 6) Then if I wanted to do this after 4 years, that would be the second 2 year cycle. So again I would multiply by this Leslie matrix, let me show you this times my original result – do the arithmetic. This is just a little matrix multiplication. I’m not showing you the zero entries, the zero products. And then, I’m starting to get fractional parts of cows, which is not always a good thing, but you realize we’re going to have to do a little bit of rounding when were done. The quick way to do this is not to do the multiplication 15 times, but to simply raise that Leslie matrix, if you notice here, to get the answer, D1 , I got that by multiplying L times the original. So I could raise L to the second power. And just like the problem before where I was dealing with the probabilities and the transition matrices, that’s the quick way to do it. (Figure 7) Figure 6, Segment 2 So to get the age distribution after 30 years, okay, that’s 15 two-year cycles, I would just raise that matrix to the 15th power. And again that’s something we do on a graphing calculator, we could use a computer algebra system to do that. And it’s fairly messy, but what do I care? The calculators doing it, not me. Multiply that out and it gives me the age distribution for each of the age groups. And as you can see here, things have grown very very fast, and in fact I’ve got over a thousand cows now. That could be an issue. I may not have enough grass and hay and corn and things to feed this many cows. So what you’d have to do is go back and say, well, probably, I’m going to start selling some off and you’d reflect that sell off in your survival rates. It’s not necessarily survival “Are they alive?”, it’s survival “Do they make it to the next group?” And at some point there, you’d start selling them off and that would lower the so-called “survival rate.” (Figure 8) 14 Figure 7, Segment 2 Figure 8, Segment 2 15 Okay, again I have a couple slides to show you kind of what we talked about, and we’ll get ready for the next segment. Summary page 1, Segment 2 16 Summary page 2, Segment 2 17 3 Population models – Unrestricted and restricted growth models Objective: To illustrate how we can use functions to model population growth, and then use these models to predict future population sizes. We show examples using the unrestricted exponential growth model, and the generally more realistic restricted-growth logistic population model. The last thing I’d like to talk to you today is exponential growth and how we model that with a continuous model. What we’ve been using in the past two examples is really a discrete model. We’re taking snippets of what happens at the end of a year cycle, a 2-year cycle something like that. We can use functions and then a graphing calculator to draw the answer and then realize that that’s an approximation of what’s going on. And what I’d like to talk to you about is, okay, how do you really do that. If you’ve got some kind of population growth, how do you do that, okay? And I want to talk about two situations, unrestricted and restricted models. What we just looked at with that Leslie matrix was an unrestricted growth model and it was actually exponential growth because the rate of growth depended on how many cows or whatever it was I had. And that’s what we call exponential growth. That’s what we have when the rate of growth of an amount is proportional to how many you have. Now that kind of situation, we’ve talked about before, can be modeled with an exponential function, A0 ekt , where k is the rate of growth and A0 is the initial amount that you start with. In fact, in a previous episode, (Episode 1, Segment 3) we talked about how when your step gets smaller and smaller and smaller you get this function ekt . So we have kind of talked about this in the past, but the problem with this and I discussed it in the last example is that you’re assuming here that they can grow without bounds, that I have unlimited resource for sustaining growth. (Figure 1) The model in Figure 1, which is stated without proof in many algebra texts, is actually the solution of the differential equation mentioned in Figure 1: dA = kA, A(0) = A0 . dt The differential equation is separable, so we may integrate with respect to both variables: Z Z dA = k dt + C A ln |A| = kt + C |A| = ekt+C = eC ekt . By letting K = sign(A)eC , we may remove the absolute value signs, and our model becomes A(t) = Kekt . By letting t = 0 and using our initial condition A(0) = A0 , we see that K = A0 and the given model is derived. 18 Figure 1, Segment 3 Now I just want to take a second and prove to you, show you that this model does what I say it does. If I let the initial amount be 1 and the growth rate be 20% or 0.2, here is a graph of that curve. You’ll notice that the initial amount is 1 and if I take little snippets, if I take, the rate of change can be construed, and we’ve talked about this previously, as the slope of a tangent line. So if I took the slope of the tangent line at this point, and I’m just approximating it really, and then I took the slope of this, the slope at about this stage is 0.2, and you can get this by taking 2 points on a curve and calculating the slope. And then if I look at the function value, right there and I multiply it by this rate of growth, this k, well the function value right there is one so sure enough these things are equal and that’s what I said they should be. (Figure 2) If I take the tangent line a little bit higher up, let’s say at when t = 6, the slope of that line is almost 23 , 0.664 or so, and if I look at the function values right there and I say 0.2 times that function value, sure enough I get 0.664. So they are matching up, that’s what’s supposed to happen. Now, that’s a good financial model, we’ve used this when talking about money and, boy, you like to think your money can grow without bounds. But it is a bad population model as illustrated in the previous example. You can’t just assume that things can grow, grow, grow, grow, grow, especially living things because they have to eat, they have to breath, they have to drink, they have to have water, and they have to have medicine sometimes. Things like this. So we need to look at a model where the resources are capped. You have some kind of capacity for things living in an environment. (Figure 3) So that’s the next example. That is, restricted exponential growth, that is where the rate of growth of an amount is proportional to the product, you take them and multiply, its 19 Figure 2, Segment 3 Figure 3, Segment 3 20 proportional to the product of the amount and the amount less then the carrying capacity, and I’ll use a capital K for that, because I didn’t want to use C, I guess. But there’s some high number and I’ll say that’s the most you can have, that’s it. And after that things start to starve to death and business like that, or die off in someway, die of disease or whatever it is. Now the model for doing this is messy. Actually deriving that requires some calculus, so I’m not going to get into the derivation of it, but it’s this kind of, there is this pattern to what’s going on here. This is called the logistic equation, and this is something you’d be given to use as a model in some type of situation. We just have to fill in the blanks. A0 is our initial population, K is the carrying capacity, and this k is not the growth rate anymore but it is some type of constant that we would need to solve for. (Figure 4) Figure 4, Segment 3 The model in Figure 4, which is stated without proof in many algebra texts, is actually the solution of the differential equation mentioned in Figure 4: dA = kA(K − A), A(0) = A0 , dt with K > 0 and 0 ≤ A ≤ K. The differential equation is separable, so we may integrate with respect to both variables: Z Z dA = k dt + C A(K − A) Z Z dA dA + = kt + C A K −A 21 ln A − ln(K − A) = kt + C ln A = kt + C K −A A = ekt+C = eC ekt . K −A By letting M = eC , we may solve for A to get A(t) = KM ekt KM = . M ekt + 1 M + e−kt By letting t = 0 and using our initial condition A(0) = A0 , we see that M = K A(t) = A0 K−A0 A0 K−A0 A0 K−A0 and so + e−kt = KA0 . A0 + (K − A0 )e−kt So let me just convince you quickly that it does what I say it does. I’ll let the initial amount be 1, k be 0.5, and the highest amount be 10. If you graph this, here’s 10, you’ll notice that as it gets closer to this 10, it’s leveling off. I’ve filled in the blanks here, and sure enough that curve is doing what I said it would do. In fact if I check the tangent lines, the slopes here is about .45, this is when time is zero, and if I take the amounts, here’s the amount, and here’s 10 minus the amount, that would be 1 times 9, solve for n is about .05. (Figure 5) Now if I check at a different location, this is when time equals 6, the slope of this line is a little over 1, if I say .05 times the amount of that curve, the y value of that curve, and 10 minus the y value of that curve when time is equal to 6, sure enough I get the same amount. So its doing what I said it would do, but what I’d really like it to do is show you how that can give you a realistic model of populations. (Figure 6) So let me give you an example. Let’s say you have 40 cows like in our previous example, I’m not going to worry about tracking them by age group any more. I’m simply going to say, alright I’ve got 40 cows, I have the capacity to feed and sustain 100, okay? That’s all the pasture I’ve got, that’s all the feed I’ve got. And then I’d like to know, okay, I do want it to grow, I do want my numbers to grow, so say if at the end of the year I have 44 cows, that’s going to give me an indication of my growth rate, when can I expect that herd to grow to 80 in number? So we’re going to use this logistic equation to model the situation, and what I’m going to do is I’m going to plug in the values, my initial population is 40, so 40 would go there and there, etc. My carrying capacity is 100, so I’ll put 100 there and 100 there, and I don’t know my growth rate, k, yet. So this actually works out to this, I’ve simplified it a little bit. (Figure 7) What you have to do is use that bit of information, okay, I had 44 at the end of the year, to solve for k. And this is quickly just some solving of exponential equations, we do this in college algebra, to kind of clean it up a bit. You can actually switch those out. Clean that up and subtract the two from both sides. Go through then and divide by the 3. And then, to solve for the k, you take the natural log of both sides, so you get − k equals this, and then k would equal a negative logarithm, which just turns it upside down. That’s about 0.164303. 22 Figure 5, Segment 3 Figure 6, Segment 3 23 Figure 7, Segment 3 (Figure 8) And then you use that in your model and set you model equal to 80, and I know I’m doing this very quickly, but you’ve seen this before. It’s the same step, you just take things, swap them out, subtract 2 from both sides. you get this down to an exponential thing, take the log of both sides, and you get time, t, to be about, this kind of messy thing, but its about 10.9 years. And I’m relying on my calculator to get that. (Figure 9) And very quickly, I’ll just show you the graph of things to convince you very quickly that that is the correct answer. Here’s the curve that I graphed, here’s 80 and it does hit about at 10.9, right there. (Figure 10) Okay, a good example of how we can get a realistic model using exponential models. We’ve got a few summary pages and well come back and wrap things up. Closing I hope you’ve enjoyed the example that I’ve shown you today. I have to admit, I went off the deep end on algebra, and I realize that. Maybe I can make it up to the general math students next time. But I’m really impressed by the power that you can access when you’re using matrices and matrix multiplication. There’s a lot of powerful stuff that you can do with algebra in general, and hopefully I’ve illustrated a few of those things. I did things very quickly so if you want to download episodes of this, you can go to our web page and we’ll flash that up in a minute. I’m done, thanks, and I’ll see you next week. 24 Figure 8, Segment 3 Figure 9, Segment 3 25 Figure 10, Segment 3 Summary page 1, Segment 3 26 Summary page 2, Segment 3 27