This action might not be possible to undo. Are you sure you want to continue?

Welcome to Scribd! Start your free trial and access books, documents and more.Find out more

Vocabulary: Differential equation; solutions; ordinary; order; general solution, particular solution, initial value; direction field; integral curve; separable equation. Technique: separation of variables. [1] Welcome to 18.03.

**I hope you've picked up an information sheet and syllabus and a problem
**

set when you came in. Read the information sheet. It contains a lot

of important information about how this course will work this term.

In addition to the textbooks (Quantum Books), you should pick up

two course packets from Graphic Arts in the basement of Building 11.

We'll use EP (5th or 4th ed) but not the freely bundled Polking.

You should have been assigned to a recitation, and have gone to it

yesterday (or this morning). We had to cancel some recitations,

by the way, and the registrar messed up this process. If you got

letters from the registrar and from the UMO, the UMO is right.

I also hope you went to recitation on Tuesday.

where these yellow sheets were handed out

I hope you've now manufactured little booklets of them.

We'll use them for a primitive but effective form of communication

between us. It's private; only I can see the numbers you put up, pretty

much.

I will not use them every lecture, but I will on Friday.

If you need to change recitation, go to the UMO. This is also

where

you hand in homework, due on Wednesdays or Fridays. The current PS is

due

a week from today, Feb 15, at 1:00.

Any questions?

[2] Notice the "Ten Essential Skills" at the back of the Information Sheet. This is a kind of plot summary. There'll be one problem on the each of these skills on the final exam. Here's a list of some of the larger courses listing 18.03 as a prerequisite or co-requisite. Teachers of these courses know the list of skills. They expect you will know how to do these things. 2.001 Mechanics and Materials I 2.003 Dynamics and Vibrations 2.005 Thermal-Fluids Engineering 2.016 Hydrodynamics

3.23

Electrical, Optical, and Magnetic Properties of Materials

6.002 Circuits and Electronics 6.021 Quantitative Physiology: Cells and Tissues 6.630 Electromagnetics 8.04 Quantum Physics I Fluid Mechanics Fluid Dynamics of the Atmosphere and Ocean

10.301 12.800

16.01 Unified Engineering 18.100 18.330 18.353J Analysis I Introduction to Numerical Analysis Nonlinear Dynamics I: Chaos

[3] A DIFFERENTIAL EQUATION is a relation between a function and its derivatives. Differential equations form the language in which the basic laws of science are expressed. The science tells us how the system at hand changes "from one instant to the next." The challenge addressed by the theory of differential equations is to take this short-term information and obtain information about long-term overall behavior. So the art and practice of differential equations involves the following sequence of steps: one "models" a system (physical, chemical, biological, economic, or even mathematical) by means of a differential equation; one then attempts to gain information about solutions of this equation; and one then translates this mathematical information back into the scientific context. 'Solving' _______________\ /

Differential Equation: Short term information /\ \ \ Model \ \ \

Behavior over time

/ / / Interpretation / / \/ Physical World

A basic example is given by Newton's law, F = ma. a = acceleration, the second derivative of x = position. Forces don't effect x directly,

but only through its derivatives. This is a second order ODE, and we will study second order ODEs extensively later in the course. [4] In this first Unit we will study ODEs involving only the first

derivative:

first order: y' = F(x,y) .

Example 1: Example 2: y' = 2x y' = ky. Solution by integrating: Solution: y = Ce^{kt} . y = x^2 + c. MEMORIZE THIS

It's easy to check; a nice feature of differential equations in general. [5] Today: Graphical approach

In Example 1 the graphs are nested parabolas: vertical translates of each other. In Example 2, graphs of solutions are no longer merely vertical translates; they form a spray, and include the solution x = 0. The constant of integration is multiplicative, here: C. We have written down the "general solutions." Their graphs fill up the plane. A "particular solution" arises from choosing a specific value for the constant of integration. Often it occurs by specifying a point (a,b) you want the solution curve to pass through. This is an "initial value." The particular solution to y' = ky with This is a good DEFINITION of e^{kx} . y(0) = 1 is y = e^{kx} .

The ODE y' = F(x,y) specifies a derivative - that is, a slope - at every point in the plane. This is a DIRECTION FIELD or SLOPE FIELD. An INTEGRAL CURVE is a curve in the plane that has the given slope at every point it passes through. A SOLUTION is a function whose graph lies on an integral curve. Example 3: y' = y^2 - x. This equation does not admit solutions in elementary functions. Nevertheless we can say interesting things about its solutions. To draw the direction field, find where This is an ISOCLINE. Eg m = 0 : x = y^2. F(x,y) is constant, say m .

I drew in the direction field.

m = 1 : x = y^2 - 1 m = -1 : x = y^2 + 1 .

I invoked the Mathlet Isoclines and showed the example. I drew some solution curves. Many get trapped along the bottom branch of the parabola. Can we explain this? I cleared the solutions and called attention to the fact that everywhere, the direction on the null-cline points into the region between m = 0 and m = -1 , and everywhere to the right of some point (actually it's (5/4,-1/2) ) on m = -1 the direction field also points into the region. So solutions in that region stay in that region: they are trapped between those two parabolas, which are asymptotic as x ---> infinity. All these solutions become very close to the function - sqrt(x) for large x . This is an ideal situation! - we know completely about the long term behavior of these solutions, and the answer doesn't depend on initial conditions (as long as you are in this range). This is "stability." [6] We have seen in action the

**EXISTENCE AND UNIQUENESS THEOREM FOR ODEs:
**

y' = F(x,y) has exactly one solution such that y(a) = b ,

for any (a,b) in the region where F is defined.

(You actually have to put some technical conditions on F -- see EP.)

Example 3: y' = - x/y .

**Take a point (x,y) . The slope of the line from (0,0) to it is y/x .
**

-x/y is the slope of the perpendicular. I drew the direction field.

You can visualize the solutions.

Everyone knows that the slope of a perpendicular

is given by negative reciprocal. So the slope field now goes around the

origin, and the solutions look to be concentric circles.

The E and U theorem says that there is just one integral curve through

each point: EVERY POINT LIES ON AN INTEGRAL CURVE, and NO CROSSING

ALLOWED.

Direction fields let you visualize this, but we also want to be able to

solve ODEs "analytically," that is, using formulas.

[7] METHOD: Separation of variables (from recitation):

Step 1: put all the x's on one side, y's on the other: dy/dx = - x/y ====> y dy = - x dx .

(If this can't be done, the equation isn't separable and this method doesn't work.) Step 2: Integrate both sides: y^2/2 + c1 = - y^2/2 + c2

Clean up by combining constants of integration: x^2 + y^2 = c (where c = 2(c2 - c1) )

**Yes, we got circles. This is an IMPLICIT SOLUTION. Separation of
**

variables

usually leads to an implicit solution - equations for integral curves,

satisfied by solutions, rather than an explicit expression for y as a

function of x . We can solve here:

y = sqrt(c - x^2) or y = - sqrt(c - x^2).

**Each integral curve contains TWO solution functions: one above, one
**

below.

Your initial condition tells you which you are on. There is NO solution

with initial condition (x,0) , since the slope would be infinite.

You may be looking for a PARTICULAR SOLUTION to the equation,

specified by giving an INITIAL CONDITION: when x = value

of the solution to be 2, say. We are looking for an through

the point (1,2). We can get this by computing what it

happen: c = 5 works, and we find the solution y = differential

1 , we want the

integral curve

c must be to make

sqrt(5 - x^2) .

**A point to note here: solutions may not extend for ever.
**

This one exists only for x between -sqrt(5) and + sqrt(5).

But how do you find that in fact e = 2. Numerical methods Even if we can solve symbolically. Analytic. take the first order ODE y' = x .5 + (. February 10.718282828459045.25 . t .5 .y) with initial condition y(0) = 1. ? The answer is: numerical methods.) ..5 replaces [2] This is a general method for computing y(b) from y(a) and the direction field.y^2 = F(x.03 Class 2. But what about y(1) ? Here's an approach: use the tangent line approximation! Since F(0. the question of computing values remains.5.5) = . symbolic methods . Question: what is y(1) ? I revealed a picture of the direction field with this solution sketched. The number e is the value y(1) of the solution to y' = y with y(0) = 1. This is what we had on Wednesday but upside down: then we considered y' = y^2 . As an example. At this point I invoked the Mathlet Euler's Method. it approximates y(x) for all x between a and b. the graph of y(x) is close to the graph of \sqrt (x) . This solution seems to be one of those trapped in the funnel. the straight line best approximating the integral curve at (0. we know that the integral curve is NOT straight. What to do? Approximate it by a polygon! So use the tangent line approximation to go half way...(.5. There x . isoclines .5) = .50) is a vertex on this polygon.1) = -1 ..1) : so this gives the estimate y(1) is approximately 0 .5 (. We follow the line segment from there with this slope for a run of to get to the point with x = 1 and y = . Quantitative methods (direction fields. so for large x . Well. The direction field there has slope F(. . and then check the direction field again: y(. 2006 Numerical Methods [1] The study of differential equations rests on three legs: .1) has slope -1 .25)(.5 + (slope)(run) = .625 . exact. In fact.. and goes through the point (0. The funnel is at the top this time.5)^2 = .5) = .5) is approximately 1 + (slope)(run) = 1 + (-1)(.x ....18.

On the other hand making the stepsize small keeps the polygonal approximation close to the actual solution. .0 1 .y0). The y cooridinates you compute. Then y1 = y0 + h A0 y2 = y1 + h A1 Write In the line In the line and in general y(k+1) = yk + h A2 This is "Euler's method. Do this in an organized way: The x coordinates are easy: x0 = a. n = 2 . x2 = a + 2h. At places where the direction field is changing rapidly. n = 1 . . We'll use n steps of size h to get from a to b . and so on.yk^2 h Ak ________________________________________________________________________ __ 0 1 2 0 . xn = a + nh. and the direction field at its vertices differs more and more from the direction field under the actual solution.5 .. There is a tension here. from round off.5 . Much of numerical analysis is understanding and bounding potential error." [3] Keep this calculation organized in a talbe. Let's identify some sources for error. xN = b. A0 = F(x0. Usually more important than this is: (2) The Euler polygon is not an exact solution. Euler's method is rarely exact. .. We'll do our example one more time: take n = 2. A1 = F(x1.. It should probably be so that b = a + nh for some positive whole number n. So you want to use the fewest number of steps possible..25 -.5 1.y1) .125 I showed some more Euler polynomials for this equation. Each computation involves a small error.Pick a step size h (1 or 1/2 above). x1 = a + h. (1) Round off accumulation.625 -1 . . and these errors accumulate. Successively compute the vertices of the polygon. using the direction field. so h = 1/2 k xk yk Ak = xk .

Not sure y(-1. the Euler approximation is too low. Heun polls twice per step. In this example.10). y(0) = 1.this produces very bad approximations quickly. we can at least predict whether the Euler method will give an answer that is too large or too small. [This was a poorly chosen example. F(t." It is interesting to compare these methods. at least. so the actual answer is TOO BIG. Here's a comparison using the ODE y' = y. and we study y(1) = e.it takes time.y) = -ty with the initial condition QUESTION: is the Euler approx for 1. and RK4 polls 4 times. For short intervals.81 x 10^{-6} 7. Too low Blank. and use that information to give a better derivative for the next strut.99 x 10^{-15} Each evaluation of the direction field costs money . [4] The way to address the the problem of variability of the direction field is to poll the values of the direction field in the area that the next strut in the Euler polygon plans to traverse.35 x 10^{-3} 1. you get the "improved Euler method" (also known as Heun's method). so the cost of . the direction field turns up secretly while the polygon is running. I invoked the example (-2. If it's bending down.] If the integral curve is bending up. Equal cost: Method RK1 = Euler RK2 = Heun RK4 Steps 1000 500 250 Evaluations 1000 1000 1000 Error 1. Too high 2. the Euler approximation is too high. the solution is y = e^x. These polling methods can be very clever. This is a general thing.2) A click shows: too small. If you poll four times sequentially (in a certain very clever way) you get "RK4. If you poll once at the midway point and simply take the average.0.2.

in the event RK4 is even more accurate. but they involve more overhead as well and experience has shown that RK4 is a good compromise. The moral is that for good accuracy Euler is essentially useless.5. even though obviously the solutions are all asymptotically zero. There are still higher order methods. The Euler polygon goes WILD. Go back to the applet. For Heun the theory predicts error proportional to h^2. There.t example..y) = -ty and initial condition (2. h = 2 x 10^{- 3} here. .Euler with 1000 steps is about the same as the cost of Heun with 500 steps or RK4 with 250 steps. Follow the Euler polygon for h = 1 a little further. the actual solution enters the funnel. in the event we do better. use smaller stepsizes. and RK4 will always win. In fact. the actual solutions that go to infinity get there in finite time.10). For RK4 it predicts error proportional to h^4. if you start at (-1. One trick: when the direction field is steep. say. but these polygons never do that since the slope is finite everywhere. [5] There is a third source of error to keep in mind: (3) Catastrophic overshoot. -3) . The error for the Euler method is around 1/1000. so h^2 = 4 x 10^{-6} . Another type of overshoot can be seen in the y' = y^2 . which is 16 x 10^{-12} here. Beware! ODE solvers are tricky and avoid things like this. but Euler with h as small as 1/4 will go off to infinity.2. with F(t. even using 1000 steps of stepsize 1/1000. This reflects a general theorem: the expected error of Euler's method is proportional to the stepsize h.

Today we look at models giving first order linear equations. These days I is typically about 2% = 0. system.x(t) --------------------Delta t Now is the moment to let the interest period Delta t tend to zero: . mixing. q(t) might be negative too.. Then there is my deposit and withdrawal. In these notes I'll continue to write x' however. The bank is a system. It pays me for the money I deposit! This is called interest. So over a month (assuming I can use q(t) as the average over the month): x(t + Delta t ) = x(t) + I x(t) Delta t + q(t) Delta t Now subtract x(t) and divide by = I x + q Delta t : x(t + Delta t) ." They will occupy most of this course.. We will measure these as a RATE as well. the notation x-dot used. signal.03 Class 3. is often [2] Model 1. In the old days a bank would pay interest monthly: Then Delta t = 1/12 and x(t + Delta t ) = x(t) + I x(t) Delta t [ + .02 . cooling. Definition: A "linear ODE" is one that can be put in the "standard form" ___________________________ | | | x' + p(t)x = q(t) | |___________________________| When t = time is the independent variable. Feb 13. Models: banks. So I might contribute $100 every month: or $1200 per year.. general story deferred to Class 4. growth and decay. In general. x is a function of time. Bank account: I have a bank account. [1] If I had to name the most important general class of differential equations it would be "linear equations. Solution in case the equation is separable. ] I has units (year)^{-1} . say I deposit at the rate of q(t) per year. 2006 First order linear equations: Models Vocabulary: Coupling constant.18. from time to time: these are withdrawals. It has x dollars in it. system response. You don't get 2% each month! you get 1/12 of that. I can add money to the bank and make withdrawals.

T'(t) > 0 (assuming k > 0 )." Here's a picture: initial condition | | | V ______________ | | -----> | System | ----|______________| x(t). e. When Te(t) > T(t). We are also assuming that is independent of other variables. T(t) = root beer temperature at time t ." A "signal" is just a function of time. of heat. q = q(t) . but if I does depend upon x as well as t . the faster T(t) changes. We get a linear equation: . then the equation we are looking at is no longer linear. This situation is typical: the actual ODE controlling a bank account is nonlinear.I x = q The left hand side represents the SYSTEM: the bank." the "input signal. ----- q(t) input x(t) -----> output [4] Model 2: Diffusion. Neither feature affects this derivation.g. [3] We can put the ODE into standard form: x' . I put my root beer in a cooler but it still gets warm. The right hand side represents an outside influence on the system: it's a "signal. but it is well approximated by a linear one when the variables are restricted in size.T(t)) where Te(t) is the "external" temperature. Let's model it by an ODE. We'll make that assumption now: so I = I(t) .x' = I x + q Note: q(t) can certainly vary in time. The greater the temperature difference between inside and outside. Simplest ("linear") model of this: T'(t) = k (Te(t) . The system responds to the input signal and yields the function the "output signal. The interest rate can too. In fact the interest rate might depend upon x as well: a larger account will probably earn a better interest rate.

This happens often: a the input signal is a product and one of the factors is p (which is k here).k T = k Te k could depend upon t but let's imagine it as constant. k is a "coupling constant. [5] There is a theory of linear equations.integral p(t) dt .1: k large means 1. The other factor then has the same units as the output signal. Don't know. or Constant temperature of zero outside. "Homogeneous": note pronunciation. None 2. Answer: (b) and (c) are linear: 6 [6] Important case: Null input signal: x' + p(t) x = 0 k times the This reflects the system as it is in isolation. don't know. Question 3. All but (a) 7. All but (c) Blank.2. good insulation 2. All 6. All but (b) 8. Question 3." The system here is the cooler. It's zero when the insulation is perfect. complete with an algorithm for solving them. It's important to recognize them when you see them. A homogeneous linear equation is separable: dx ---x = . Which of the following are linear ODE's? (a) $\dot x+x^2=t$ (b) $\dot x=(t^2+1)(x-1)$ (c) $\dot x+x=t^2$ 1. (a) only 3. (c) only 5. without outside influence: No deposit or withdrawal. (b) only 4.p(t) dt [constant of integration is ln|x| = . bad insulation Blank.T' . large when it is bad. The input signal is external temperature. k is small when the insulation is good.

incorporated into the indef integral] Write P(t) for any primitive: P'(t) = p(t) . so integral p(t) dt = P(t) + c ln|x| = . x = C e^{-P(t)} where P'(t) = p(t) .e^c e^{-P} x = 0 is a good solution So we get to +.P + c |x| = e^c e^{-P} x = +. and too. . but was eliminated when we divided by x .e^c can be any number except zero.

the (heat) diffusion equation. [2] Method: "variation of parameter. x' + p(t) x = 0 . The associated homogeneous equation is x' + (1/3) x = 0. Example: x' + kT = kT_ext. This isn't separable: it's something new." or "trial solution": (1) First solve the "associated homogeneous equation" x' + p(t) x Write xh = 0 (*)_h for a nonzero solution to it.) Suppose the temperature outside is rising at a constant rate: T_ext = 60 + 6t (in hours after 10:00) x(0) = 32 . don't forget to substitute back in to get x . Feb 15.03 Class 4." Let's take it to be 1/3. This is separable. [1] Definition: A "linear ODE" is one that can be put in the "standard form" _____________________________ | | | x' + p(t)x = q(t) | |_____________________________| On Monday we looked at the Homogeneous case. Let's see how this works in our example. say xh = C e^{. and solve for u .95 at Target. 2006 First order linear equations: solutions. x(0) = 32 .18. and the solution is Now for the general case. k is the "coupling constant. (2) Then make the substitution (3) Finally. (This cooler cost $16.integral p(t) dt} (*) q(t) = 0 : and we need an initial condition: So the equation is x' + (1/3) x = 20 + 2t . x = xh u . We'll describe a method which works for ANY first order liner ODE. which has nonzero solution xh = e^{-t/3} Write x = e^{-t/3} u and plug into the differential equation: x' = (-1/3) e^{-t/3} u + e^{-t/3} u' .

We can solve this for u by integrating: u = int e^{t/3} (20 + 2t) dt = 3 . the cooler is 18 degrees cooler than the outside.(1/3) x = ( 1/3) e^{-t/3} u __________________________________________ 20 + 2t = e^{-t/3} u' This cancellation is what makes the method work. all solutions resemble 42 + 6t as t gets large. Since T_{ext}(t) = 60 + 6t .18) e^{t/3} = (42 + 6t) e{t/3} Are we done? Not quite: x = e^{-t/3} u = 42 + 6t There! Want to check? 2t ! [3] x' = 6 . w = 3 e^{t/3} [another place where we can take c = 0!] int 2t e^{t/3} dt = 2t 3 e^{t/3} . This always happens for first order linear equations. . 20 e^{t/3} + ?? int v dw = v w . so x' + (1/3) x = 6 + 14 + 2t = 20 + Wait! Where's the constant of integration? Answer: u had an additive constant attached: u = (42 + 6t) e^{t/3} + c x = e^{-t/3} u = 42 + 6t + c e^{-t/3} That's the general solution.18) e^{t/3} u = int e^{t/3} (20 + 2t) dt = 3 . We wanted x(0) = 32 : so initial value problem is c = -10 . [4] Let's work out the general case. this is saying that for large t.int w dv [sorry. and the solution to the x = 42 + 6t . "u" is used] This is parts: v = 2t . The term c e^{-t/3} is a "transient": it dies away as t gets large. dw = e^{t/3} dt dv = 2 dt . I sketched some solutions. and you can see why from this method of solution. 20 e^{t/3} + (6t .10 e^{-t/3} Notice the structure of the set of solutions: the constant of integration occurs as a coefficient of the homogeneous solution.int 3 e^{t/3} 2 dt = (6t .

x y' + y This one has an easier way: the left hand side is so we are solving (xy)' = 1/x .ln x)/x^2 . Also notice the obvious fact that x = 0 is a solution.c/x + (ln x)/x + c/x = 1/x . so xh' + p(t) xh = 0 Make the substitution x' = x = xh u (*)_h and solve for u: xh' u + xh u' p x = p u xh ____________________ q But xh = ( xh' + p xh ) u satisfied (*)_h or . + xh u' so the parenthetical quantity is zero! u' = xh^{-1} q q = u' xh which you can solve by integrating: u = integral xh^{-1} q dt And this determines x . so we have a solution. Integrate: so xy y = = ln x + c (ln x)/x + c/x (xy)' . Write xh for a nonzero solution.c/x^2 (1 . in fact it is never zero. One more example: x y' + y = 1/x (x > 0) y/x = /1x^2 Standard form: yh y = = y' + e^{-int dx/x} = 1/x 1/x [ int x/x^2 dx + c ] y' = = = (ln x)/x + c/x Check: (x/x . x = xh integral xh^{-1} q dt [5] Note about homogeneous solutions: the constant of integration occurs as a factor: all solutions are constant multiples of any nonzero solution.ln x)/x .x' + p(t) x = q(t) (*) Write xh for a nonzero solution of the associated homogeneous equation. Notice that not only is it not always zero.

[6] The method of "integrating factors" finds a function that you can multiply both sides and get into this position. It turns out that the right thing to multiply through by is xh^{-1} = e^{ integral p(t) dt} This is called an "integrating factor." Let's see how it works out in an example Solve x' + tx = 2t e^{ integral t dt } = e^{ t^2 / 2 }: The integrating factor is Let's check: ( e^{ t^2 / 2 } x )' = = as claimed. x = 2 x = = e^{ t^2 / 2 } 2t 2 e^{ t^2 / 2 } + c 2 + c e^{ . So we have e^{ t^2 / 2 } x' + t e^{ t^2 / 2 } x e^{ t^2 / 2 } ( x' + t x ) ( e^{ t^2 / 2 } x )' = Integrate both sides: e^{ t^2 / 2 } x so In particular.t^2 / 2 } is a solution .as you can easily check! .

The complex numbers fill out a plane. or -3 miles. It rotates the number around the origin by 90 degrees clockwise." Question 1. Maybe complex numbers seem obscure because you are used to imagining numbers by giving them units: 5 cars. 2.2 = -1 + 3i. Every complex number can be written as a b = = Re(a+bi) Im(a+bi) a + bi the real part the imaginary part: NB this is a real number. Complex numbers do not accept units. 3. or at least 2006. 4.18. 5. Multiplication by on a complex number. The point up one unit from 0 is written i . i has the following effect 1. None of the above. which is rotated by 90 degrees counterclockwise. no "<. Also. with a and b real. Feb 17. complex exponential Today. 6. It reflects the number across the real axis. It rotates the number around the origin by 90 degrees counterclockwise. 2006 Complex Numbers. there is no ordering on complex numbers. The new thing is i^2 = -1 . The usual rules of algebra apply. They had already been in use for several hundred years. but they were kept fairly secret and were regarded as perhaps not entirely real. Addition and multiplication by real numbers is as vectors. [1] Complex Algebra We think of the real numbers as filling out a line. For example FOIL: (1 + i)(1 + 2i) = 1 + 2i + i .03 Class 5. is the 200th anniversary of the birth of complex numbers. It reflects the number across the imaginary axis. In 1806 papers by Abb\'e Bul\'ee and by Jean-Robert Argand established the planar representation of complex numbers. It takes a number to the number pointing in the opposite direction with the same distance from the origin. Compute: i(a + bi) = -b + ai . [2] Complex conjugation Division occurs by "rationalizing the denominator: .

The angle up from the positive real axis is Arg(z) = "argument" = "angle" of z. it's only well defined up to adding multiples of . z = 0 5.(b+d)i (a+bi)(c+di) = (ac-bd) + (ad+bc)i (a-bi)(c-di) = (ac-bd) .. z is purely imaginary 2. z lies on the unit circle 4. proof: It's not pleasant to compute absolute values .(bi)^2 = a^2 + b^2 (*) (1-2i)/(1+4) = (1-2i)/5.but it's easy to compute squares: . As usual. z is real 3.Magnitudes Multiply : |wz| = |w||z| . ____ (*) encourages us to define the "complex conjugate" a+bi = a . 2 pi..z Proofs: (a+bi) + (c+di) = (a+c) + (b+d)i has conjugate (a+c) .bi _ and in these terms it reads: z.1/(1+2i) Now general = (1/(1+2i)) ((1-2i)/(1-2i)) (a+bi)(a-bi) so we can continue . complex numbers have polar descriptions. None of the above [3] Polar multiplication Being points in the plane. ___ _ _ __ _ _ Conjugation satisfies w+z = w + z .they involve square roots .(ad+bc)i Question 2: If _ z = . then is conjugate to 1.z = |z|^2 _ Divide by |z|^2 and z to see 1/z = z / |z|^2 . wz = w. The distance of z from zero is |z| = "absolute value" = "modulus" = "magnitude" of z.z .(b+d)i ______ ______ which coincides with (a+bi) + (c+di) = (a-bi) + (c-di) = (a+c) . = = a^2 .

1 4 . None of the above You went straight to the answer. 5.____ __ _ _ |wz|^2 = (wz)(wz) = wzwz = wwzz = |w|^2|z|^2 = (|w||z|)^2 It follows that . This checks with our question about multiplication by Question 3. 2. and if you understand complex numbers you'll never have to memorize those formulas again. (1+i)^1 = 1+i (1+i)^2 (1+i)^3 (1+i)^4 (z^0 = 1 has modulus 2 as long as z isn't zero) pi/4 sqrt(2) and angle pi/2 : 2i has modulus has modulus has modulus and angle 2 sqrt(2) 4 and angle pi : 3 pi/4 : 2(-1+i) and angle -4 The powers all lie on a spiral emanating from the origin. . Then: I'll check this in case (cos a + i sin a)(cos b + i sin b) = ((cos a)(cos b) . In fact multiplication of complex numbers contains in it the angle addition formulas for sin and cos . 1. z(t) = An expression a(t) + i b(t) parametrizes a curve in the plane.sqrt(2) 4 i (1+i)^4 = i above. 3.(sin a)(sin b)) + i ((cos a)(sin b) + (sin a)(cos b)) = cos(a+b) + i sin(a+b) using the angle addition formulas for cos and sin .4 . Arg(wz) = Arg(w) + Arg(z) w and z are both on the unit circle. [4] Complex exponential. For example . 4.Angles Add: |wz| = |w||z| since both sides are positive. but let's tabulate some powers: (1+i)^0 = 1 . 6.

The derivative is computed for each component. Here's an ODE we can try to solve: z' = iz . x' = kx. for any complex number a+bi to z' = (a+bi)z . This is a circle. so the solution is a curve such that the velocity vector is always perpendicular to the radius vector. compute z' iz = = . See also the Supplementary Notes or the Notes.sin t and they agree." you can compute that the solution In fact. So we will write the solution to (**) as e^{it} . On the other hand we found out that multiplication by i is rotation by 90 degrees. and gives you the velocity vector. z(0) = 1 . . z(0) = 1. "Euler's formula.sin t + i cos t i cos t .z = 1 + it parametrizes a line running vertically through 1 . (**) x(0) = In lecture 1 we saw that e^{kt} is the solution to 1. Here this is i : vertical. Thus: e^{it} = cos t + i sin t. and if we add the initial condition it is the unit circle: z = cos t + i sin t To check. is e^{(a+bi)t} = e^{at} (cos(bt) + i sin(bt)) With this we can compute the general exponential rule e^{wt} e^{zt} = e^{(w+z)t} .

: Negative real numbers have square roots in Any quadratic polynomial with real coefficients has a root in by the quadratic formula x^2 + bx + c In fact: = 0 has roots (-b +. In general. 2 itself is one. Since i^2 = -1 .2i and +.. The cube roots of unity start with 1 and divide the unit circle evenly into 3 parts. n = 3 : Angles Add. This gives ( -1 + sqrt 3 i ) / 2 Or it could be 4 pi / 3 . not quite: it could be 2 pi / 3 .03 Class 6. (+. or 2 pi. or 2 pi. C .. 2006 Roots of Unity. no..sqrt 3 i ) / 2 That's it. . then |z^n| = 1 .2 . Special case: n = 2 : z^n = 1 : z = +. there's no other way for it to happen.. so if z^3 = 1 then the argument of z is 0 .4c))/2 "Fundamental Theorem of Algebra": Any polynomial has a root in C (unless it is a constant function). if z^n = 1 . How about nth roots of unity divide the circle into n equal z^4 = 16 ? Now the magnitude must be a positive real fourth root of 16.. or . or -2 pi.a C.. the parts. It's better to think of the argument of 1 as a choice: 0. or . The others have argument such that 4 times the argument is 0.sqrt(b^2 .i sqrt(a))^2 = . Euler's formula. Feb 21. namely. which gives ( -1 . Sinusoidal functions [1] Let Roots of unity a > 0 .18... or 4 pi. but Magnitudes Multiply. 2: all the 4th roots of 16 lie on the circle of radius 2..: so you get +. so |z| = 1 : roots of unity lie on the unit circle. . since three times that is 2 pi.1 "n-th roots of unity" In general.

sin(t) + i cos(t) = i (cos(t) + i sin(t)) e^{it} . General fact about complex numbers: _ _ z + z z . -sqrt(3) . and sqrt(3) . z(0) = 1 . From Euler's formula. so one argument of z would be pi/2 : 2i is a cube root of -8i . magnitude must be 2 again. and the "general fact" at the start. Im(z) ______ Apply this to z = e^{it}.= 2 2 i Proof by diagram. We agreed to write this complex-valued function as This is "Euler's formula": e^{it} = cos(t) + i sin(t) . or by using the uniqueness theorem for solutions to ODEs.i . The argument of -8i is 3pi/2.z ----. In fact the same easy check shows that for any complex number the solution of z' = (a+bi) z with z(0) = 1 is z = e^{at} (cos(bt) + i sin(bt)) a+bi so we also agree to define e^{(a+ib)t} = e^{at} (cos(bt) + i sin(bt)) e^{(a+bi)t} e^{(a+bi)t} is is e^{at} bt (*) That is. the magnitude of and the argument of This definition e^{(z+w)t} (*) = satisfies the expected exponential rule: e^{wt} e^{zt} You can see this using the usual rule for real exponentials together with the angle addition formulas.i . is. we find . with verticies at 2i . The others will differ from that by 2 pi / 3 or 4 pi / 3 .How about z^3 = -8i ? Well. has solution cos(t) + i sin(t) We saw this geometrically but you can also just check it: z' = . [2] We saw that z = z' = iz . I will need to know what e^{it} Reflecting across the real axis reverses the angle: so ______ e^{it} = e^{-it} . (**). See the Supplementary Notes. You get a peace symbol.= Re(z) ----.

[3] Sinusoids A "sinusoidal function" (co)sine wave. f(t) = A cos( ??? ) P A For the moment let's suppose t = 0 gives a maximum for f(t) . We need to relate t to theta. 0 =< t_0 The offset from the standard picture = "time lag.e^{-it} ---------------2 i Sometimes these are also called Euler's Formulas. I drew the graph of sinusoid. For example sin(omega t) = cos(omega t . f(t) is one whose graph is shaped like a I drew a large general sinusoidal function.phi) where phi = omega t_0 phi is the "phase lag." It's measured in radians. or parameters: The height of its maxima = depth of its minima = "amplitude. cos(theta) . The "frequency" is simply 1/P .t_0)) There's another way to express the lag behind the cosine: = A cos(omega t . I drew t and theta axes. This is the time at which f(t_0) = A ." The elapsed time till it repeats = the "period" or (if spatial) the "wavelength" lambda Now. omega. . 2pi/P is the number of radians per unit time. It is called the the "circular" or "angular" frequency of f(t) ." t_0 . the general formula for a A cos(omega (t . It has a special symbol. t_0 .cos(t) = e^{it} + e^{-it} ---------------2 sin(t) = e^{it} . In terms of the parameters sinuosidal function is f(t) = A . Anything you want to know about sines and cosines can be obtained from properties of the (complex) exponential function. Usually you make sure < P . this is our model example of a A sinusoidal function is entirely determined by just three measurements. When P units of time have elapsed.pi/2) . omega. and saw that the relationship is theta = (2pi/P) t . so have 2 pi radians.

Or: Differenting a complex valued function is done on the real and imaginary parts. Same for integrating. Expand everything out: 1/(2+i) = = (2-i)/5 e^{2t} (cos(t) + i sin(t)) = (1/(2+i)) e^{(2+i)t} + c e^{(2+i)t} so the real part of the product is (1/5) e^{2t} (2 cos(t) + sin(t)) + c More direct than the high school method! [2] Sinusoidal signals: Solve x' + 2x = cos(t) : toy model for a cooler responding to oscillating temperature. Homogeneous solution is e^{-2t} and solve for u: 2 x = 2 e^{-2t} u so substitute x = e^{-2t} u x' = -2 e^{-2t} u + e^{-2t} u' _________________________ cos(t) = e^{-2t} u' so u = int e^{2t} cos(t) .03 Class 7. e^{2t} cos(t) = Re e^{(2+i)t} = so int e^{2t} cos(t) dt Re int e^{(2+i)t} dt and we can integrate exponentials because we know how to differentiate them! int e^{(2+i)t} dt We need the real part. Use Variation of Parameter.18. 2006 Applications of C: Exponential and Sinusoidal input and output: Euler: Re e^{(a+bi)t} Im e^{(a+bi)t} [1] Integration e^{2t} cos(t) ? = = e^{at} cos(bt) e^{at} sin(bt) Remember how to integrate Use parts twice. Feb 22.

Replacing sinusoidal signals with exponential ones Let'e go back to the original ODE x' + 2x = cos(t) This equation is the real part of a complex valued ODE: z' + 2z = e^{it} z(t) . is one solution. x = (1/5) (2 cos(t) + sin(t)) + ce^{-2t} This is a general thing: a linear ODE with p constant and sinusoidal input signal has a sinusoidal solution. maybe the output signal will be too: TRY x = A e^{3t} . [3] Linear constant coefficient ODEs with exponential input signal x' + 2x = 4 e^{3t} Let's try We could use variation of parameter or integrating factor.e^{rt} dt = r e^{rt} x Since the right hand side is an exponential. but: x' 2 x = = A 3 e^{3t} 2 A e^{3t} ----------------- 4 e^{3t} = A (3+2) e^{3t} which is OK as long as A = 4/5: x = (4/5) e^{3t} The general solution is this plus a transient: x [4] = (. We just saw how to get an exponential solution: z_p = 1/(i+2) e^{it} k = 2 . The inspiration here is based on the fact that differentiation reproduces exponentials: d -. This is a different ODE.8) e^{3t} + c e^{-2t} . If p > 0. r = i : To get a solution to the original equation we should take the real part of this! Expand each factor in real and imaginary parts: . and I use a different variable name.This is exactly what we integrated above: u = (1/5) e^{2t} (2 cos(t) + sin(t)) + c Going back. but instead let's use the method of optimism. or the inspired guess. I don't know what A is yet. any solution differs from this one by transients which die out as t ---> infty.

To see this I'll start with the general expression f(t) = A cos(omega t . phi . (*) as long as r + k [6] Rectangular expressions for sinusoidal functions. and such that b = A sin(phi) So I should try to find a = A cos(phi) There's a name for such A .b) = (1. phi = pi/3 (a.b) in the plane. Expand this and try to find out what I should take for using the cosine difference formula: f(t) = A cos(phi) cos(omega t) + A sin(phi) sin(omega t) A. I claim that a cos(omega t) + b sin(omega t) is sinusoidal.sqrt(3)) . phi: they are the polar coordinates of the point (a. try x = A e^{rt} : x' = A r e^{rt} kx = k A e^{rt} ______________________ B e^{rt} A = = A (r + k) e^{rt} B / (r + k) so: The Exponential Response Formula (for first order linear ODEs) The general solution to x' + kx is x = = B e^{rt} (B/(r+k)) e^{rt} + c e^{-kt} is not 0 .z_p = ((2-i)/5) ( cos(t) + i sin(t) ) x_p = Re(z_p) = (1/5) (2 cos(t) + sin(t) ) as before! [5] This exponential method is so useful that I'd like to do the general case: to solve x' + kx = B e^{rt} .phi) A and phi. They do exist! Example: cos(2t) + sqrt(3) sin(2t): which has polar cooridinates A = 2 .

.so cos(2t) + sin(2t) = 2 cos( 2t . I used the Mathlet Remarkable fact: Any sum of sinusoidal functions with given period is again sinusoidal (with the same circular frequency).pi/3 ) Trigonometric Id to illustrate this.

To see the rest of the direction field.18. Autonomous means conditions are constant in time. depending on the current population but NOT ON TIME. plot the graph of g(y). They are also graphs of solutions.y) today. 2006 Autonomous equations I'll use (t. y' = k0 (1 . Suppose that when y is small the growth rate is approximately k0 . constant solutions or "equilibria. so k(0) = k0 . meeting the horizontal . g(y) = 0 when y = 0 and when y = p . Eg [Logistic equation] Variable growth rate k(y). The horizontal lines with these values of y form the nullcline. and k(p) = 0 . It is a parabola opening downward. isoclines first. y' = F(t. When y > p . so y' = k(y) y. the population declines back to the maximal sustainable population." Isoclines are collections of horizontal lines. but that there is a maximal sustainable population p . Simplest version: A graph of k(y) against y . are called "critical points" for the equation y' = g(y).(y/p)) y = g(y) .(y/p)). Sketch direction field. k0 > 0 means the populuation (if positive) is growing.y) is the general first order equation y' = g(y) . The Logistic Equation is This is more realistic than Nat Growth for large populations. straight line with vertical intercept k0 and horizontal intercept p : k(y) = k0 (1 .03 Class 8. These are thus the only constant solutions. In the logistic case. Values of y such that g(y) = 0. and as y gets near to p the growth rate decreases to zero. though they may depend on the current value of y . decaying. the growth rate becomes negative. It is nonlinear. Autonomous equations are always separable.. but we aim for a qualitative grasp of solutions. k0 < 0 means it is falling. Autonomous ODE: Eg [Natural growth/decay] Constant growth rate: so y' = k0 y . Feb 24.

This says that the range of populations which are stable is declining: the maximum sustainable population decreases: [ it's the larger root of y^2 . Mark on it the equilibria. at a rate of equation is y' = (1-y)y . In terms of the graph of g(y). In suitable units (megafish and years. This is the "Logistic" or "S" curve. where g(y) = 0. and if the population falls below that it crashes to zero. the stable equilibria occur when Now.axis at y = 0 and y = p ." the zero population is "unstable." showing the penetration of technological innovations." Question 8. If the population exceeds the maximal stable population. its essent content can be compressed. y < 0 0 < y < p y > p the slopes are negative the slopes are positive the slopes are positive.y + a = 0 ] . the graph of g(y) moves down. looking like g(y) = y^3 . the equilibria move closer together. is the rightmost critical point stable or unstable? This can be made clear by sketching the phase line. Draw a vertical line. g'(p) < 0." Since the direction field is constant horizontally. Beyond that there are no equilibria. The new As a increases.y . If the harvest rate is pushed still higher. where g(y) has a graph which I sketch. It's called the "phase line. .1. suppose we model a fish population by means of a logistic equation. draw an upward pointing arrow if g(y) > 0 and a downward pointing arrow if g(y) < 0 . and the population is declining on both sides of it. Now fishing is allowed. In the autonomous equation y' = g(y) . a minimal sustainable population also appears. it falls back towards it. This simple diagram tells you roughly how the system behaves.a I invoked the Mathlet <Phase Lines> to visualize what happens. This says that for for for I drew some isoclines and some solutions. perhaps) the equation is y' = (1-y)y : the limiting population is p = 1. The max stable population is a "stable equilibrium. this is "semistable": the graph of g(y) is tangent to the horizontal axis. unstable when g'(p) > 0. a megafish per year. In between them. the two collide. I showed a graph from the article "Thwarted Innovation.

g'(c) = 0 (constant solutions) or case. i. Any horizontal (time) translate of a solution is another solution. y' = k_0y has three "fundamental solutions. the C in Ce^t comes from e^{t-c} = e^{-c} e^t with a sign or zero put in as well. So it's true.e. So (2) is the y' = g(y) y" = g'(y) y' by the chain rule. True False Extreme points occur where y' = 0 .) 1. Direction fields are constant in the horizontal direction. Question 1. In the autonomous case. Solutions of autonomous equations can never have strict local maxima or minima. Ex. the conditions represented by the ODE are constant in time. 2. 2. 3. and watched what happened as a increased. Question 2. As we know. .e^{k_0 t} y = e^{k_0 t} ." y = 0 . and they don't have strict maxima or minima. Nonconstant solutions of the autonomous ODE have inflection points at y for which: 1. solutions can't have strict extrema. Solutions given by the indefinite integral of f(t): F(t) + c. where g(y) = 0. Any vertical translate of a solution is a solution. (A strict local maximum for f(t) is a time t = a such that f(a) > f(t) for all t near but not equal to a . g(y)=0 g'(y)=0 g''(y)=0 So if y" = 0 at y = c then either g'(c) = 0 there. Let's compare autonomous equations with the calculus case: y' = f(t) . These are the constant solutions. Direction field constant in the vertical direction. y = .I returned a to zero and opened the Bifurcation Diagram. and any solution is a horizontal translate of one of them. You can see it on the S curve. y' = g(y) .

and take the real part of the solution. Try x = e^{-2t} u . or if q = 0 ) . it dies away and leaves x_p. .Is the signal exponential? q(t) = B e^{rt} . the solutions always are of the form x = x_p + c x_h where x_p is SOME solution ("particular solution") and x_h is a nonzero solution of the homogeneous equation.If no: Is the "coefficient" p constant?: "constant coefficient" e^{-pt} . .Otherwise. c x_h deserves to be called a "transient". General comment: for first order LINEAR equations. replace with z' + p z = B e^{i omega t} .Is the signal sinusoidal? especially B cos(omega t) If so. constant coefficient but signal neither exponential nor sinusoidal. 2006 Review: Linear v Nonlinear [1] review of linear methods [2] Comment on special features of solutions of linear first order ODEs not shared by nonlinear equations. . Feb 27. r constant A. solution to homogeneous equation is . [1] First Order Linear: x' + p(t) x = q(t) system. If so.If yes.18. solve that. try x = A e^{rt} and solve for . input signal. output signal = system response. So: VP: Homogeneous solution: e^{-2t} . If p > 0 . they are just less efficient. Note: VP and IF do work in general.If yes. Decision tree for solving first order linear equations Separable? ( p and q are both constant.03 Class 9. [2] Examples: x' + 2 x = e^{-2t} ( t + 1 ) Not separable. use either Variation of parameter or Integrating factor. then solve by separation of variables.

phi): = = Expand both factors in polar 1 + i so 2 / ( 1 + i ) sqrt(2) e^{pi i/4} (2/sqrt(2)) e^{-pi i/4} . called the "integrating factor": t + 1 so and [3] eg = e^{2t} ( x' + 2x ) t^2/2 + t + c x = = = ( e^{2t} x )' e^{2t} x e^{-2t} ( t^2/2 _ t + c ) Review of exponential replacement: x' + 2 x = 4 cos(2t) z' + 2 z = 4 e^{2it} <---| ---Re Try z = z' A e^{2it} = A 2i e^{2it} 2 z = 2 A e^{2it} ---------------------4 e^{2it} = A ( 2 + 2i ) e^{2it} so A z_p = = 4 / ( 2 + 2i ) = 2 / (1 + i ) ( 2 / ( 1 + i ) ) e^{2it} There are two ways to get the real part out of this.x' = ." (2) Solution as form: A cos(omega t .2 e^{-2t} u + e^{-2t} u' 2 x = 2 e^{2t} -------------------------------e^{-2t} ( t + 1 ) or and u' = t + 1 x = so = u = t^2/2 + t + c e^{-2t} u' e^{-2t} ( t^2/2 + t + c ) IF: Multiply through by the inverse of the homogeneous solution. (1) Solution as a cos(omega t) + b sin(omega t) : Expand both factors into a + bi : z_p so x_p = = ( 1 . Which to use depends upon what you want.i ) ( cos(2t) + i sin(2t) ) cos(2t) + sin(2t) This gives the sinusoidal response in "rectangular form.

and The real part is z_p = sqrt(2) e^{i(2t . 1 = 1/c so c = 1 and x = 1 / ( 1 . Properly speaking.pi/4)} x_p The amplitude is = sqrt(2) cos (2t . the solution ENDS. sqrt(2) [4] e. x^{-2} dx .x^{-1} x^{-1} x = = = = dt . one for t < 0 . solutions of differential equations are required to have connected graphs. This behavior leads to some danger.t This reaches infinity at t = 1 ! This behavior does not happen for linear equations: if p(t) and q(t) are well behaved (eg don't zip off to infinity themselves) then all solutions exist and stay finite for all time. Once we've gone up to infinity on this solution.t 1 / (c . t + c c . Comparison with solutions of nonlinear equations: x' = x^2 : separable.t ) actually describes TWO solutions: one for t > 0 . 1 / ( 1 .pi/4) and the phase lag is pi/4 . If I start with ) x(0) = 1 . This kind of explosion actually happens in the case of Newton's laws: Jeff Xia showed that a certain 5-planet system moves off to infinity in finite time! .t) The constant of integration is in a different place.g. There's no reasonable way to say which branch you might come back on when t > 0 .

. and an external force: mx" = F_spr + F_dash + F_ext The spring force is characterized by depending only on position: write F_spr(x). The dashpot force is frictional." So the equation is .03 Class 11. It acts against the velocity: If If If x' > 0 ." "Hooke's This is another example of a linearizing approxmimation. Write F_dash(x'). F_spr(x) < 0 F_spr(x) = 0 F_spr(x) > 0 I sketched a graph of F_spr(x) as a function of x . F_dash(x') < 0 F_dash(x') = 0 F_dash(x') > 0 The simplest way to model this behavior (and one which is valid in general for small x'. mass Take a spring attached to a wall. This means that it depends only on the velocity. 2006 Second order equations: Physical model. the "dashpot" (which is a way to make friction explicit). initial conditions [1] F = ma spring is the basic example. by the tangent line approximation) is F_dash(x) = -bx b > 0 the "damping constant. The simplest way to model this behavior (and one which is valid in general for small x .18. by the tangent line approximation) is Law" F_spr(x) = -kx k > 0 the "spring constant. structure of solutions. dashpot | || || |-------> F_ext || || | || || _____________ ___|___ || || _____ | | | ||---VVVVVVV---| |------|_____| |-------|| || |_______| _____________| || || O | O || || | || |-------> | x Set up the coordinate system so that at x = 0 the spring is relaxed. x' = 0 . March 3. . characteristic polynomial. x' < 0 . real roots. The cart is influenced by three forces: the spring. If If If x > 0 x = 0 x < 0 .

[3] Today we'll find some solutions in the constant coefficient homogeneous case. (This makes sure the x" is really there. The quantities m . without outside interference. It makes sense to try for exponential solutions of (*): c e^{rt} for some as yet undetermined constants c . But the input signal q(t) can certainly vary.mx" + bx' + kx = F_ext The left hand side represents the SYSTEM. so you should expect two constants of integration in the general solution. I asked whether all solutions to systems like this bounce. So: In the case of second order equations. Physically. an external force at work. b . The right hand side represents the INPUT SIGNAL. the spring/mass/dashpot system. [2] Solutions. but you also have to say what velocity you impart to it: x'(t0) is needed as part of the initial condition. you can release the spring at t = t0 from x(t0) . and the fuel burns so m decreases. There is no concept of direction field. I displayed a rubber band with a weight on it. This means F_ext = 0 : the system is allowed to evolve on its own." In general they may depend upon time: maybe the force is actually a rocket. solutions can cross. Maybe the spring gets softer as it ages. which has as solutoin x = e^{-kt} (and more generally multiples of this). instead of being multiplied by zero!) x" + b(t) x' + k(t) x = q(t) and k(t) Most of the time we will assume that the "coefficients" b(t) are CONSTANT. I drew some potential graphs of solutions through a single point. initial conditions (x(t0). Still. Maybe the honey in the dashpot gets stiffer with time. Constant coeffient. x'(t0)) determine the solution. Most people thought no. This is another simplifying approximation. It bounced. homogeneous: mx" + bx' + kx = 0 (*) This is a lot like x' + kx = 0 . The "standard form" of a second order linear ODE is gotten by dividing by mass. k are the "coefficients. To get from x" to x we must integrate twice. Let's see.

m] are flags indicating that I should multiply the corresonding line by this number. This is because you can differentiate the two solutions separately. This depends strongly on LINEARITY and HOMOGENEITY but not on CONSTANT COEFF. = ms^2 + bs + k = 0 x" + 5x' + 4x The characteristic polynomial s^2 + 5s + 4 . This one factors as (s + 1)(s + 4) so the roots are r = -1 and r = -4 . every vector in the plane is a linear combination of them. k ] b ] m ] 0 = x = c e^{rt} x' = c r e^{rt} x" = c r^2 e^{rt} ___________________ mx" + bx' + kx = c ( mr^2 + br + k ) e^{rt} is An exponential is never zero. x2 . b] . the general solution to x = c1 x1 + c2 x2 "linear combination" Just two solutions determine all solutions. One reason I wanted to write out the polynomial was to remember that you can find roots by factoring it. [4] Here's the appearance of the general solution of a second order homogeneous linear ODE x" + b(t) x' + k(t) x = 0 (*) neither of which is a constant (*) is For any pair of solutions x1 . It's also true that the SUM c1 e^{-t} + c2 e^{-4t} is a soluion as well. Organize the calculation: the k] . so we can cancel to see that c e^{rt} a solution to (*) for any c exactly when r is a root of the "characteristic polynomial" p(s) Example A. . The corresponding exponential solutions are c1 e^{-t} and c2 e^{-4t} .and r. We want the roots... plug x = e^{rt} into (*). This is like saying that for any two vectors in the plane such that neither is a multiple of the other. multiple of the other.. To see which r might work.

for example: 3 and then x = = -3 c2 c1 = 1 .4 c2 and we have to solve a pair if linear equations.So the general solution in our example is x [5] and = c1 e^{-t} + c2 e^{-4t} t = 0 with x(0) = 1 Initial conditions.4 c2 e^{-4t} x(0) = x'(0) = c1 + c2 .c1 e^{-t} . here by adding the equations.c2 = so 2: c2 = . Suppose we start at x'(0) = 2 .c1 .1 2 e^{-t} . Elimitate the unknowns. x' 1 2 = = = .e^{-4t} .

b. and k are real. k ] x = u + iv b ] x' = u' + iv' m ] x" = u" + iv" ___________________ 0 = (mu" + bu' + ku) + i(mv" + bv' + kv) Both things in parentheses are real. x" + 5x' + 4x = 0 .03 Class 12. so the only way this can happen is for . The corresponding exponential solutions are e^{-t} and e^{-4t} . then the real and imaginary parts of x are also solutions. We found that (*) has an exponential solution r is a root of the "characteristic polynomial" p(s) = s^2 + bs + k e^{rt} exactly when Example A. x" + 4x' + 5x = 0 The characteristic polynomial s^2 + 4s + 5 has roots r = -2 +." Example B. All solutions go to zero: no oscillation here. For this we have: Theorem: If x is a complex-valued solution to mx" + bx' + kx = 0. Proof: Write x = u + iv and build the table. where m. damping criteria.18. March 6. The general solution is a linear combination of these: x = c1 e^{-t} + c2 e^{-4t} . We did this: The characteristic polynomial s^2 + 5s + 4 factors as (s + 1)(s + 4) so the roots are r = -1 and r = -4 . When the roots are real and not equal to each other the system is called "Overdamped. and we have exponential solutions e^{(-2-i)t} e^{(-2+i)t} . 2006 Homogeneous constant coefficient linear equations: complex or repeated roots.i Our old friend i = sqrt(-1) appears.sqrt(4-5) = -2 +. dashpot. spring system without external forcing term. [1] We are studying equations of the form x" + b x' + k x = 0 (*) which model a mass. I guess we were expecting REAL valued solutions.

and is written omega_n. [2] are Remark on roots and coefficients: If the roots of r1 and r2 then s^2 + bs + k so = (s .sqrt(k) k > 0 : no damping : "Harmonic Oscillator. Taking linear combinations.e^{-2t} sin(t) The collection of linear combinations of these two functions is the same as the set of linear combinations of the original two. To recap: x" + omega_n^2 x cos(omega_n t) = and 0 has independent real solutions sin(omega_n t) and so has general solution x = = a cos(omega_n t) + b sin(omega_n t) A cos(omega_n t .phi) This is a "damped sinusoid. I would have gotten real and imaginary parts e^{-2t} cos(t) and e^{-2t} sin(-t} = . you end up with the same set of solutions." with "circular pseudofrequency" omega = 1. e^{(-2+i)t} has real part and imaginary part e^{-2t} cos(t) e^{-2t} sin(t) so we have those two solutions.phi) You can check easily and directly that this is a solution. If I had used the other. So in our situation." Note: I used only ONE of the two exponential solutions here. e^{(-2-i)t} = e^{-2t} e^{-it} .both of them to be zero. +." and and e^{-i sqrt(k) t} sin(sqrt(k) t) give e^{i sqrt(k) t} cos(sqrt(k) t) with real and imaginary parts sqrt(k) is called the "natural circular frequency" of the harmonic oscillator. It makes no difference which exponential function you use. Example C: Roots are x" + kx = 0 .r1)(s . When the roots are not real the system is called "Underdamped. we get the general solution x = = e^{-2t} ( a cos(t) + b sin(t) ) A e^{-2t} cos(t .r2) = s^2 + bs + k = 0 s^2 -(r1+r2)s + k .

die away as t --> infinity 3. so they die away. r = -b/2 . I'll take m = 1 Name* solutions b.product of the roots is Question 1: If k and b are both positive. are damped sinusoids 5. Generally solutions are built up out of exponentials. roots Exp. are exponential (though perhaps complex) 4. then the solutions are e^{-bt/2} times a sinusoid. One of three things must happen. So the correct answer is 2. The fact is that we can write down another solution. 2: This is tricky: If the roots are not real.b .-. The general solution is (at+b) e^{-bt/2} too. This is transitional. namely . If they are real. k = (b/2)^2 so the equation is x" + bx' + (b/2)^2 x = 0 . -.b : k so the average of the roots is -b/2 . There is just one characteristic root. so that sign is negative. Thus all these solutions die off too. then all solutions of x" + bx' + kx = 0 1. 3 is not right. oscillate 2. sol's Basic real . none of the above 1 and 4 are violated by Ex A. e^{-t} and e^{-4t} are both exponential functions. e^{-bt/2}. t e^{-bt/2} and it dies away in that case Here is a summary table of unforced system responses.sum of the roots is . Then r1 + r2 = . [3] We have not yet considered the critically damped case. in Ex A . You can check that this is a solution. but they are not all exponential. then the solutions are combinations of e^{r1 t} and e^{r2 t} Notice that (r1)(r2) = k implies that r1 and r2 are of the same sign. and so just one exponential solution.k relation Char. but their sum is not.

k) b^2 b^2 b^2 >= 4k = 4k < 4k so are -. but it doesn't have a good interpretation in terms of a mechanical system.sqrt(b^2 . e^{r2 t} (see * The name here is appropriate under the assumption that b and k both non-negative. The rest of the table makes sense in general.the roots are real if repeated if non-real complex conjugate if b/2] [and the real part is - . e^{r2 t} e^{rt} same e^{rt}.sqrt((b/2)^2 .b +. b^2/4 < k Non-real roots e^{r1 t}. The quadratic formula is r = = (.4k)) / 2 -b/2 +.Overdamped Critically te^{rt} damped Underdamped below) b^2/4 > k b^2/4 = k Two diff. real Repeated root e^{r1 t}.

" . If b > 0 and k > 0 .i omega_d The basic solutions are e^{-bt/2} cos(omega_d t) .k relation b^2/4 > k b^2/4 = k b^2/4 < k Char.(b/2)^2). te^{rt} e^{at} cos(ct). the pseudofrequency decreases. March 8. e^{at} sin(ct) are * The name here is appropriate under the assumption that b and k both non-negative." In the underdamped case. real r1. This is subtle but visible on "Damped Vibrations. 2006 Summary of solutions to homogeneous second order LTI equations. One of three things must happen to solutions of x" + bx' + kx = 0 . r2 Repeated root r = -b/2 Non-real roots a +. At that instant.ci Basic real solutions e^{r1 t}. The rest of the table makes sense in general. slowly at first. Name* Overdamped Critically damped Underdamped b. and in "polar form" the general solution is x = A e^{-bt/2) cos(omega_d t .03 Class 13. [1] We saw on Monday how to solve x" + bx' + kx = 0. the roots are -b/2 +. but faster as the damping approaches critical damping. but it doesn't have a good interpretation in terms of a mechanical system.18.omega_d where omega_d = sqrt(k . Introduction to inhomogneneous equations. The imaginary part of the roots is +.phi) (*) Some people prefer to call omega_d the "pseudofrequency" of (*) . e^{-bt/2} sin(omega_d t) . Notice that as you increase damping. since unless b = 0 this is not a periodic function and so properly speaking doesn't have a frequency. e^{r2 t} e^{rt}. then all solutions die off.(b/2)^2) is the "damped circular frequency. Here is a summary table of unforced system responses. The are "transients." and the real part of the roots is the "growth rate" -b/2 : -b/2 +.i sqrt(k . roots Two diff. the pseudoperiod becomes infinite and you don't get solutions which cross the axis infinitely often.

1. the roots look like this: | | | | | _______________________________________ | | -b/2 . necessarily [2] :-The roots of s^s + bs + k . -b/2 +. with B = 0.In the complex plane.i omega_d | * | | | -b/2 + i omega_d * I showed Poles and Vibrations. None of the above. The solutions decay to zero faster 4.k ). None of the above. The solutions decay to zero faster 4. necessarily Question 2: If I move the roots towards the real axis. are . The amplitude of solutions decreases 3. Question 1: If I move the roots the the left.sqrt( (b/2)^2 . The pseudofrequency of the solutions decreases 3. The amplitude of the solutions decreases 2. 1. The pseudoperiod increases 2.

oscillating k | oscillating . 3.. not oscillating not oscillating... real < 0 . | | | | . opposite sign ------------> ^ | |______ Here is a summary table of unforced system responses.. The system can be modeled by the constant coefficient equation x" + bx' + omega_n^2 x = 0 where 0 < b < 2\omega_n.. | . = b^2/4 ^ |<----. -----------... This system is linear but the coefficients are not constant in time. .k . Re < 0 | Re > 0 . | ... This system is nonlinear.. 2. . unstable..<----. One of three things must happen. | .b at least one zero . . k . root real.. . . .. I'll take m = 1 . .repeated if | . . .. ------------ real > 0 .<----. . . stable .b ^ | most solutions grow |______ some nonzero constant solutions Question 3: You observe an unforced system oscillating. . | unstable. | . . From this you can conclude: 1. . | .... and notice that the time between maxima spreads out as time goes on.. ^ |<-----purely imaginary complex roots . 2/4 k = b^ .. . . . ------------> .sinusoidal solutions | . | ..t e^{rt} too if stable.

then xp + xh Proof: Plug x into xp is any solution to (*) (*). Either 2 or 3 holds but we can't say which. Exp times sinusoidal Polynomial Exp times other (eg polynomial) Sums of these General periodic functions (via Fourier series) The general strategy in finding solutions is: Superposition II: If solution to (*)_h. and xh is a is again a solution to (*): k) x = xp + xh b) x' = xp' + xh' m) x" = xp" + xh" __________________ mx" + bx' + kx = (m xp" + b xp' + k xp) + (m xh" + b xh' + k xh) = F_ext + 0 as we wanted. mx" + bx' + kx = F_ext (*) Also important will be the "associated homogeneous equation" mx" + bx' + kx = 0 (*)_h Input signals we will study: Constant Sinusoidal Exponential. if xh is the general solution to the general solution to (*). This is to be compared with Superposition I: If x1 and x2 are solutions of a homogeneous linear equation. (*)_h then xp + xh is . In fact. then so is any linear combination c1 x1 + c2 x2 . [3] INHOMOGENEOUS EQUATIONS I drew the spring/mass/dashpot system and added a force to it: the little blue guy comes back into play.4.

omega^2| Imagine the natural frequency of the oscillator fixed. when the gain falls back towards zero. of course. The output amplitude is a multiple of the input amplitude. When Also: omega > omega_n . it passes "resonance. and we slowly increase the frequency of the input signal. and when omega is larger the response is exactly anti-phase. omega_n^2) plug this into (**) : xp = B cos(omega t) xp" = . and then no such sinusoidal solution exists. a "particular solution.B omega^2 cos(omega t) ----------------------------------A cos(omega t) = xp" + omega_n^2 xp = B(omega_n^2 . Why? And what's this resonance? It looks like perhaps there is a solution of the form xp = B cos(omega t) To see what B must be. (which we have worked on for a [4] First case: harmonic sinusoidal response." where the response amplitude is large. What happens with the weight and rubber band is that the nonlinear character of the spring asserts itself for large amplitude." and then (2) find the general solution of (*)_h while). This is RESONANCE.omega^2) cos(omega t) This works out if we take B = A / (omega_n^2 . the denominator is positive.omega^2) . and the omega < omega_n . Drive a harmonic oscillator by a sinusoidal signal: x" + omega_n^2 x = A cos(omega t) (**) There are two frequencies here: the natural frequency of the system and the frequency omega of the input signal.Superposition II splits the problem of finding the general solution to (*) into two parts: (1) find SOME solution to (*). I showed what happens with a weight on a rubber band: for small omega the weight follows the motion of my hand. and the ratio is the GAIN: H = B/A = 1/|omega_n^2 . The graph of the gain starts when omega = 0 at H = 1/omega_n^2 and then increases to a vertical asymptote at omega = omega_n . There are solutions. and we will come back to this case later.

On Friday we'll add in damping. When omega > omega_n the denominator is negative. and the output signal is a negative multiple of the input: this is PHASE REVERSAL. .output is a positive multiple of the input.

p(3) = 3^2 + 2(3) + 2 = 17 so p(s) = s^2 + bs + k x" + 2x' + 2x = 4 e^{3t} xp = ( 4 / 17 ) e^{3t} .18. . 2006 Exponential signals.(7/85) cos(3t) + (6/85) sin(3t) + e^{-t} ( a cos t + b sin t ). Try for a solution of the form k] b] xp xp' = = B e^{rt} xp = B e^{rt} : B r e^{rt} xp" = B r^2 e^{rt} _____________________ A e^{rt} or where Eg xp = = B ( r^2 + br + k ) e^{rt} ( A / p(r)) e^{rt} is the characteristic polynomial.6i) / 85 ) ( cos(3t) + i sin(3t) ) and find the imagnary part of the product: yp = . The general solution is given by x = (4/17) e^{3t} + e^{-t} ( a cos(t) + b sin(t) ) (*) for sinusoidal signals as well: Of course this will let us solve Eg y" + 2y' + 2y = sin(3t) This is the imaginary part of so z" + 2z' + 2z zp = = e^{3it} p(3i) = (3i)^2 + 2(3i) + 2 = -7 + 6i ( 1 / (-7 + 6i) ) e^{it} We want the imaginary part. Lets do it by writing out real and imaginary parts: zp = ( (-7 . operators [1] Exponential signals x" + bx' + kx = A e^{rt} (*) We want to find some solution.(7/85) cos(3t) + (6/85) sin(3t) The general solution is y = .03 Class 14. higher order equations. March 10.

Most systems are more general than the simple spring/dashpot/mass system we have been looking at. and we simply ignore both. xp = ( A / p(r) ) e^{rt} [2] Higher order equations A better model for the weight at the end of the rubber band is: My hand ----> ----------------------------------------| | > | < | y > | < V | | --------| m |--------------------------------- | | | x | V y = 0 x = 0 y is the position of the plunger.The work shows: The Exponential Response Formula: {rt} is given by a solution to provided x" + bx' + kx = A e^ p(r) is not zero. Arrange it so that if y = x = 0. ----------------------------------------------- | | > | k1 < | y(t) . The "relaxed" position of the rubber band just cancels the force of gravity. Thus the net downward force on the mass is given by m x" m x" + k x = = k ( y . relative some choice of zero point. A constant upward force on the mass exerted by the spring cancels gravity. the mass is at rest.x ) k y(t) It's time to think bigger. For example.

they are controlled by linear ODEs with constant coefficients." p(s) = an s^n + . the numbers ak are the "coefficients.. + a1 x + a0 give rise to exponential solutions to the = 0 (*)h xp . and the roots of p(s) homogeneous equation an x^{(n)} + .) aren't too big.x2 ) If you differentiate the second equation twice and plug in the first.> | < V | | --------- m_1 |------------------------- | | --------| | > | x_1 k2 < | > V < | --------- m_2 |------------------------ | --------| | | x_2 | V Now m1 x1" = k1 ( y(t) .x1 ) .. . + a1 x' + a0 x = q(t) (*) The theory of such systems is just like the theory we have developed for first and second order linear constant coefficient equations. you get a fourth order equation for x2 ." The system is called a "Linear Time Invariant" or LTI system.m2 k2] x2" + (k1 k2) x2 (t) = (k1 k2) y This is a general thing: more complicated systems are described by higher order equations.. The left hand side represents the system.. Such an equation has the form an x^{(n)} + . It has a "characteristic polynomial. + a1 x' + a0 x The general solution of (*) is gotten by finding some solution. and adding to it the general solution xh of (*)h .x1 ) m2 x2" = k2 ( x1 .k2 ( x2 . When the parameters (x1. When I do this I get: m1 m2 x2^(4) + [m1 k2 + m2 k1 .. .. x2...

.." It's the "differntiation operator.. and x' + kx = e^{rt} : a solution is given by xp = ( A / (r+k) ) e^{rt} -. + a1 x' + a0 x Our work shows the Exponential Shift Formula: by A solution of p(D) x = A e^{rt} is given = p(D) x xp = ( A / p(r) ) e^{rt} provided that Example: p(r) is not zero. x' + kx = (D + kI) x. p(s) = s + k ." We can iterate: D^2 = x" .a result we saw when we were looking at first order liner equations. Ix = x There's also the "identity operator": And we can add: (D^2 + 2D + 2I) x = x" + 2x' + 2x .[3] Operators Here's a neat piece of notation: Dx = x' D takes a funtion as input and returns a new function back as output: It is an "operator. p(s) = s^2 + 2s + 2 . and The characteristic polynomial here is it's irresistable to write D^2 + 2D + 2I so x" + 2x' + 2x = = p(D) p(D) x This works just as well with higher order LTI operators: an x^{(n)} + .

. + a_0) e^{rt} p(D) e^{rt} = p(r) e^{rt} try x_p = B e^{rt} . 2004 Operators: Exponential shift law Undetermined coefficients [1] Operators.I ) u = 1 so want . + a_0 I) e^{rt} = (a_n r^n + .. D e^{rt} so and or D^n e^{rt} = = The ERF is based on the following calculation: r e^{rt} = rI e^{rt} r^n I e^{rt} (a_n D^n + ..03 Class 15. What if p(r) = 0? eg x" .I ) u ( (D-I)^2 . (*) The key to solving this problem is the behavior of D on products: (d/dt) (xy) = x' y + x y' In terms of operators: D(vu) Especially: = v Du + u Dv e^{rt} Du + u r e^{rt} e^{rt} ( Du + ru ) e^{rt} ( D + rI ) u D(e^{rt} u) = = = Apply D again: D^2 (e^{rt} u) = = = D( e^{rt} (D+rI)u ) e^{rt} (D+rI)(D+rI) u e^{rt} (D+rI)^2 u (*): Use: let's try a variation of parameters approach to solving Try for Then -1] x = = e^{-t} u e^{-t} (D-I)^2 u D^2 x x = e^{-t} I u ------------------------e^{-t} = e^{-t} ( (D-I)^2 . So to solve p(D) x = A e^{rt} . p(D) (B e^{rt}) = B p(D) e^{rt} = B p(r) e^{rt} so we should take B = A/p(r) : x_p = e^{rt}/p(r) . March 13..x = e^{-t} .18.

. + b_1 t + b_0 . Proof by example: x" + 2x' + 3x = t^2 + 1 3 is not 0 : there is a solution of the The theorem applies since form x = at^2 + bt + c To find a . - Putting this together we get the "Exponential Shift Law": p(D) ( e^{rt} u ) and using it we find: ERF2: If p(r) = 0 = then a solution of p(D) x = A e^{rt} provided and p'(r) is given by = e^{rt} p(D+rI) u x_p (a/p'(r)) t e^{rt} is not zero. This is described in more detail in the Notes. c = (1/3)(1-2b-4a) = 11/27. Then u = -t/2 and x_p = .2D ) u = 1 i. With a constant right hand side. Finally..or ( D^2 . [2] Polynomial signals: Undetermined coefficients.e. plug in: 3) x = at^2 + bt + c 2) x' = 2at + b 1) x" = 2a _________________________ t^2 + 1 = 3at^2 + (3b+4a)t + (3c+2b+4a) The coefficients must be equal. u" . provided that p(0) = a_0 is not zero. Since 3 is not zero. you get a constant solution (unless the coefficient of v is zero): v = 1/2..1 so p'(s) = 2s and you recover the solution we worked out. we can divide by it to find a = 1/3. So .2v = 1 . p(D)x = q(t) has exactly one solution which is polynomial of degree less than or equal to k . c . b . + a_1 s + a_0 then p(0) = a_0. Theorem (Undetermined coefficients) Take q(t) = b_k t^k + . so we have v' ..t e^{-t}/2 . p'(-1) = -2. Notice that if p(s) = a_n s^n + a_(n-1) s^{n-1} + . In our case. Then b = -(1/3)4a = -4/9 . p(s) = s^2 .2u' = 1 and this we can do by "reduction of order": say v = u'.

b = -1 . u = t .1 [check it!] .t + c .xp If = (1/3)t^2 .(4/9)t + (11/27) we can use "reduction of order": = t so u' + u = t a_0 = 0 x" + x' Substitute u = x' u = at + b u' = a ---------------t = at + (a+b) a = 1 . x = t^2/2 . .

We solved this by luckily trying B . Solutions are arbitrary sinusoids with circular frequency the "natural frequency" of the system.phi = -pi for omega > omega_n . cos(omega t) is the "physical signal." as opposed to the force. x" + omega_n^2 x = omega_n^2 A cos(omega t) I am driving the system through the spring.phi rather than phi: this graph is constant zero for omega < omega_n and then switches discontinuously to . [2] Bode Plots. The traditional thing to graph is . cos(omega t . and then falls towards zero when omega > omega_n .phi) then and phi = 0 phi = . with a plunger moving sinusoidally with amplitude 1 . We regard the plunger position as the system input.pi for for omega < omega_n omega > omega_n . gain(omega) = |omega_n^2/(omega_n^2 .omega^2)) cos(omega t) in which case the system is in This is ok unless omega = omega_n .omega ^2)) e^{i omega t} No damping ===> denominator is real and so x_p = A (omega_n^2 / (omega_n^2 . increases to infinity when omega = omega_n.omega^2)| This has graph which starts with gain(0) = 1 . . The "gain" is the ratio of the output amplitude to the physical signal amplitude. There's a phase transition too: if we write the solution as x_p = gain . resonance with the signal. or "complete signal" omega_n^2 cos(omega t) . Let's do it using ERF: z" + omega_n^2 z z_p = = x_p = B cos(omega t) and solving for omega_n^2 A e^{i omega t} A (omega_n^2 / (omega_n^2 .18.03 Class 16. 2006 Frequency response [1] Frequency response: without damping x" + omega_n^2 x = 0 : First recall the Harmonic Oscillator: The spring constant is k = omega_n^2 . In this case. March 15. Drive it sinusoidally: omega_n .

[3] Damped systems:

Frequency response

Drive this system sinusoidally, through the spring: x" + bx' + kx = k A cos(omega t) and call omega_n the "natural

We continue to write k = omega_n^2 circular frequency" of the system.

**Physical input cos(omega t) has amplitude A . amplitude
**

of the sinusoidal output divided by A.

The gain is the

**I displayed "Amplitude and Phase, Second Order" and set b = .5 . In it, B = 1 .
**

ERF: z" + bz' + kz = k B e^{rt} z_p W(r) = W(r) B e^{rt} , W(r) = k / p(r) .

k = 3.24

and

is the "transfer function." W(i omega) =

Sinusoidal input means

r = i omega:

omega_n^2 / ((omega_n^2 - omega^2) + i b omega)

This is the "complex gain." In the expression x_p I claim that - phi gain Proof: Then z_p = = |W(i omega)| e^{- i phi} e^{i omega t} |W(i omega)| e^{i(omega t - phi)} = = Arg(W(i omega)) |W(i omega)| |W(i omega)| e^{- i phi} = gain . cos(omega t - phi)

W(i omega) =

which has real part x_p = |W(i omega)| cos(omega t - phi)

Compare this with the undamped case: there is an imaginary part to the denominator. This causes two effects: (1) The magnitude of the denominator is increased, causing the gain to decrease. Especially: the denominator can never be zero anymore, no matter what omega is, since it has a nonzero imaginary part. Thus you never encounter true resonance with a sinusoidal signal, if there is any damping.

(2) A phase lag appears. Since b > 0 , of the denominator is positive, so Im W(i omega) < 0 which says that 0 < phi < pi . unless

the imaginary part omega = 0

[5] More explicitly, W(i omega) When omega = 0 = k / ((k - omega^2) + i b omega) 1 : gain 1 , phase lag 0.

this is

As omega increases, plane. Gain: |W(i omega)|

W(i omega)

sweeps out a curve in the complex

=

k / sqrt{(k - omega^2)^2 + b^2 omega^2}

**If b is small, the gain is large (though not infinite)
**

when omega is near to the natural frequency of the system, since the

first term in the denominator is small. This is NEAR RESONANCE.

When omega gets very large,

the denominator is roughly omega^2 , k/omega^2.

so the gain tails off like

**As b grows larger, the second term dominates and for modest values of
**

omega

|W(i omega)| ~ k / (b omega)

**This doesn't have maxima anymore; for large b there is no near-
**

resonant peak.

Eg in the tool, when k = 1 the resonant peak vanishes when b = sqrt

(2) . Phase lag: Since k = omega_n^2 , W(i omega) is purely imaginary when omega = omega_n : that's when the phase lag of the solution is 90 degrees. |W(i omega_n)| = omega_n / b

When b = .5 and k = 3.24, omega_n = 1.8 and indeed when omega = 1.8 , the phase lag is exactly 90 degrees and the gain is 1.8/.5 = 3.60.

The gain won't be maximal then (think of the case of should expect it to be relatively large. b large), but you

18.03 Class 17, March 17, 2006 Application of second order frequency response to AM radio reception with guest appearance by EECS Professor Jeff Lang. [1] The AM radio frequency spectrum is divided into narrow segments which individual stations are required to broadcast in. The challenge of a receiver is to filter out the signals at frequences other than the target frequency. This can be done using a very simple RLC system which has a sharp peak in the gain curve at a given frequency (which can be tuned by changing the strength of one of the components of the system). Electronic circuits have mechanical analogues. Mechanical engineers often think of their systems in electronic terms, and vice versa. The mechanical system analogous to the radio receiver is this: | b m k |/

| |----------______ |/

|---| |=======------|______|----/\/\/\------|/

| |----------| |/

| | |/

|-------> |-------->

y(t) x(t) We are driving the system by motion of the far end of the dashpot, while keeping the far end of the spring constant. [2] Let's see the equation of motion for this system. Arrange the position parameter x so that x = 0 when the spring is relaxed. The spring exerts a force -kx .

The dashpot exerts a force proportional to the speed at which the piston is moving through the cylinder. This speed is (y-x)' . When y' > x' , the force is positive, so the dashpot force is b(y-x)' mx" = -kx + b(y-x)'

Putting the system terms on the left, mx" + bx' + kx = by'

If y is constant, we have a homogeneous equation; its solutions are transients. A transient for the system was displayed; it oscillated, so we are in the underdamped situation. We could measure the damped circular frequency omega_d and the "decrement," the ratio of the height of one peak to the preceding one; and using them we could determine b/m and k/m = omega_n^2 .

[3] Now let's drive the system with a sinusoidal signal. The radio waves are built up from them.

So set y = B cos(omega t) . The equation is then

mx" + bx' + kx = - b omega B sin(omega t)

We know that there will be a sinusoidal system response; and that that is the response we'll see very quickly, since the transients damp out. We also know that we should try to express the sinusoidal system response in terms of a gain and a phase lag with respect to the physical input signal. Despite the appearance of the right hand side of the equation, it's clear that we should take as physical input signal the function y = B cos(omega t), so we look to find gain and phase lag phi such that x_p = gain . B cos(omega t - phi)

**We also know that gain and phi are computed by finding the "complex
**

gain" W(i omega),

and then writing it out in polar terms:

W(i omega) so that and gain - phi = = = gain . e^{- i phi} |W(i omega)| Arg(W(i omega)) .

The handout "Driving through the dashpot" computes that W(i omega) = = b i omega / p(i omega) b i omega / (m(omega_n^2 - omega^2) + b i omega)

[4] The system was subjected to several input frequencies. One odd thing appeared: for small frequency,the system response is *ahead* of the system input: < 0 in that case. Also maximal gain seems to happen when the phase lag is zero. We can analyze this mathematically, by dividing numerator and denominator in the complex gain by b i omega: )) W(i omega) = 1 / ( 1 - (im/b) ( (omega_n^2 - omega^2) / omega^2

**As omega varies, this sweeps out a curve in the complex plane. To see
**

what

that curve is, look first at the denominator. Its real part is always 1

.

When omega = omega_n there is no imaginary part: W(i omega_n) = 1 .

When omega < omega_n , the imaginary part is negative,

when

omega > omega_n , it is positive: the direction of movement is

upward. Now if z is on this line, then its reciprocal is on the circle of

radius 1/2 and center 1/2 .

The angle gets reversed. So W(i omega) moves clockwise along that

circle.

Let's read off the gain and phase lag curves:

The gain starts small, grows to a maximal value of omega_n , and then falls as omega --> infty.

1 at omega =

**The angle, which is - phi , starts at pi/2, falls through 0 at
**

omega = omega_n , and then comes to rest near - pi/2 as omega --->

infty.

So we find mathematically that phi < 0 for small omega.

[5] These curves were then reproduced experimentally by subjecting the RLC circuit to a series of different frequencies. Then we tried to input a signal from an antenna, and we watched the system respond to some and not to others: 1030 khz was quite loud.

03 Class 18. which gives z_p = t e^{2it}/p'(2i) = -(it/4) e^{2it} so x_p = (t/4) sin(2t) . v_p u_p z_p x_p = = = = b = -a/4i = 1/16 . But there is a t there. In the real example I drive it: The complex equation is x" + 4x = t cos(2t) . We should then use "Variation of Parameters": Look for solutions of the form z = e^{2it} u for u an Unknown function. -it/4 + 1/16 -it^2/8 + t/16 ( -it^2/8 + t/16 ) e^{2it} (t^2/8) sin(2t) + (t/16) cos(2t) x_p + the homogeneous solution. v = at + b Reduction of order: v = u' . z" + 4z = t e^{2it} . If it weren't for the t we could try to apply ERF: p(s) = s^2 + 4. x" + 4x = 0 PLEASE KNOW the solution to the homogeneous harmonic oscillator x" + omega^2 x = 0 are sinusoids of circular frequency omega ! Here. and Frequency Response [1] Example. 2006 Review of constant coefficient linear equations: Big example. March 20. Use undetermined coefficients: 4i] v = at + b v' = a ---------------t = 4iat + (a + 4ib) so a = 1/(4i) = -i/4 . superposition. so it doesn't apply. The general solution is then [3] Superposition: putting special cases together.18. Suppose a bank is giving I percent per year interest: . though. a cos(2t) + b sin(2t). p(2i) = -4 + 4 = 0 . we do have the resonance response formula. 4] z = e^{2it} u 0] z' = e^{2it} ( u' + 2i u ) 1] z" = e^{2it} ( u" + 2i u' + 2i u' + (2i)^2 u ) -----------------------------------------------------e^{2it} t = e^{2it} ( u" + 4i u' + (4-4) u ) so u" + 4i u' = t v' + 4i v = t .

It is the essence of linearity.e^{-it}) / 2i is a possible signal or solution. Here we mean *complex* linear combinations and *complex* exponentials. It lets you break up the input signal into constituent parts.x' . Is this any different than opening ONE bank account and saving at the rate q_1(t) + q_2(t) ? Say the solutions with savings rates q1 and q2 are Is x1 + x2 a solution with savings rate q1 + q2 ? x1' I x1 = q1 x2' I x2 = q2 -------------------------- (x1 + x2)' .Ix = q(t) q_1 Suppose that I open TWO bank accounts and proceed to save at rates (t) and q_2(t) in them. Our work has shown a general result: Theorem: If q(t) is any linear combination of products of polynomials and exponential functions. In fact this is true for nonconstant coefficient linear equations too. One example is when q2 = 0 : then x2 is a solution to the homogeneous equation. then all solutions to p(D)x = q(t) are again linear combinations of products of polynomials and exponential functions. and then put the results back together. solve for them separately. and we find again that adding such a function to a solution of p(D)x = q gives another solution.I (x1 + x2) = q1 + q2 since differentiation respects sums (and multiplying by In general if then p(D) x1 = q1 and p(D) x2 = q2 p(D) (c1 x1 + c2 x2) = c1 q1 + c2 q2 I does too). This is why it isn't so bad that we spent all that time studying very special input signals. x1 and x2 . [4] Frequency response z = = = |z| e^{i Arg(z)} sqrt( a^2 + b^2 ) arctan(b/a) Polar form of a complex number: |a+bi| Arg(a+bi) . so for example sin(t) = (e^{it} . and it's the most general form of the superposition principle.

pi/4) w e^{i omega t} Amplitude of phi = x_p x_p 3/2(1+i) = z_p x_p Lesson: then if = Re z_p z_p |w| .i pi /4} (3/(2 sqrt 2)) e^{.Arg(1/p(i omega) = Arg(p(i omega)) There's no particular advantage in writing out a more explicit formula for this. it reaches a modest "near resonance" peak at omega = 1 . It is based on the following method of finding a sinusoidal system response in "polar" (amplitude/phase lag) form: Example: + 2i x" + 2x' + 3x z" + 2z' + 3z z_p Now write 3/2(1+i) = = = 3 cos(t) 3 e^{it} p(s) p(i) = = s^2 + 2s + 3 -1 + 2i + 3 = 2 (3 / 2(1 + i)) e^{it} Do the denominator first: in polar form . Good luck! .i pi /4} e^{it} (3/(2 sqrt 2)) cos(t . The phase lag is phi = . and when omega is large it falls off like 3/omega^2 .Arg(w) = = = = = phase lag of Suppose now that I let the input frequency be anything: x" + 2x' + 3x z" + 2z' + 3z p(i omega) = = = = z_p = 3 cos(omega t) 3 e^{i omega t} -omega^2 + 2i omega + 3 (3 . 1+i = (sqrt 2) e^{i pi / 4} (3/(2 sqrt 2)) e^{. In this case.Frequency response is about the amplitude and phase lag of a sinusoidal (steady state) response of a system to a sinusoidal signal of some frequency.omega^2) + 2i omega (3 / p(i\omega)) e^{i omega t} So the amplitude of the sinusoidal response is |1/p(i omega)| = 3 / sqrt((3-omega^2)^2 + 4 omega^2) This takes value 1 at omega = 0 .

(*) \ / \ ------.. or the sound of a violin. cos(0t) = 1 (and sin(0t) = 0 ).P/2] ." P > 0 such that f(t+P) = So strictly speaking the examples given are not periodic. You can choose the window as convenient. Otherwise. We'll use the basic window [-pi. A function f(t) is "periodic" if there is f(t) for every t ..sine series ------ / . which is often called THE period. For this reason a natural period to start with is P = 2\pi . 2006 Periodic signals. and g(t) are periodic of period P then so is So we can form linear combinations. ..pi] .03 Class 20. but rather they coincide with periodic functions for some period of time.. as usual. There is a standard notation for the coefficients: f(t) = a0/2 + a1 cos(t) + a2 cos(2t) + . Any "window" (interval) of length P determines the function. We'll often use the window [-P/2. or innumerable electronic signals. Also. Our methods will accept this approximation. I showed an example of violin and flute. n = 1 . The constant function is periodic of every period.18. [2] Sine and cosines are basic periodic functions... March 24. P is a "period. 3. and yield results which merely approximate real life behavior. + b1 sin(t) + b2 sin(2t) + .cosine series -------------.. t cos(t) is NOT periodic. Fourier series [1] Periodic functions: for example the heartbeat. all the periodic functions we'll encounter have a minimal period. af(t) + bg These are "harmonics" of the "fundamental" sinusoids with If f(t) (t) . 2pi ? Question: what other sines and cosines have period Answer: cos(nt) and sin(nt) for n = 2..

So the terms are badly chosen. and to illustrate the process of adding functions. A function odd if f(-t) = -f(t).' Linear combinations of evens are even. The average of a function of period Ave(f) = (1/2pi) integral_{-pi}^pi f(t) dt 2pi is Ave(f(t)+g(t)) = Ave(f(t)) + Ave(g(t)) Ave(cos(nt)) = 0 so applying Ave to Ave(f(t)) a0 = = for (*) : or n > 0 and Ave(sin(nt)) = 0 . f(t) is There are integrals for computing these coefficients. Show the Mathlet FourierCoefficients to see other examples." The Coefficients. cos(nt) is even. 2pi Theorem. The *definition* of the Fourier coefficients of a function this: they are the coefficients that make (*) true. sin(nt) is odd. but it's also (sine series) . but using the definition is usually easier. The only function which is both even and odd is the zero function. then f(t) . of odds are odd.(cosine series) is a linear combination of evens and hence even. a0/2 (1/pi) integral_{-pi}^pi f(t) dt. Even . odd is even . Any reasonable [piecewise continuous] function of period has exactly one expression as a Fourier series. even . f(t) is "even" if f(-t) = f(t) . Do the square wave case. even is even . For f(t) = f(-t) and f(t) = -f(-t) together imply that f(t) = -f(t) . odd . odd is odd. they behave more like 'positive' and 'negative. Some simple observations: [3] Average." We'll see why the odd choice of an and bn are the "Fourier a0/2 for the constant term shortly. If a periodic function f(t) is even. [4] Parity.This is a "Fourier Series.

since the product is odd and the interval you are integrating over is symmetric. Exactly the same method shows: bn = (1/pi) integral_{-pi}^pi f(t) sin(nt) dt .and so odd. integral_{-pi}^pi am sin(mt) cos(nt) dt Only one of all these terms is nonzero: and since then am = an . Application: (for n > 0) Substitute (*) into integral_{-pi}^pi f(t) cos(nt) dt Compute this integral term by term: integral_{-pi}^pi (a0/2) cos(nt) dt Then we have a bunch of cosines. The of them gives: = 0 m = n . The others require some trig identity which you can find in Edwards and Penney. an = 0 [The same argument shows that if a polynomial is even then it's a sum of even powers of t . but since cos(0t) = 1 formula is true for n = 0 (by our comment about averages above). so it's zero. or (1/pi) integral_{-pi}^pi f(t) cos(nt) dt the We did this calculation assuming n > 0 . ] [5] Integral expression: This will use the trigonometric integrals integral_{-pi}^pi cos(mt) sin(nt) dt = 0 integral_{-pi}^pi cos(mt) cos(nt) dt = 2pi = pi = 0 integral_{-pi}^pi sin(mt) sin(nt) dt = = pi 0 if if if if if m = n = 0 m = n > 0 m is not equal to m = n m is not equal to n n The first of these is easy. we discover integral_{-pi}^pi f(t) cos(nt) dt an = the cosine term with = am pi . so: The Fourier series of an even function is a cosine series: The Fourier series of an odd function is a sine series: bn = 0 . if it's odd then it's a sum of odd powers of t . The = 0 (since n > 0) m-th = = one gives: am pi 0 if if m = n m is not equal integral_{-pi}^pi am cos(mt) cos(nt) dt to n m-th And then a bunch of sines.

. sin(t) + (1/3) sin(3t) + (1/5) sin(5t) + . ] This is the Fourier series for the standard squarewave. Thus: sq(t) = (4/pi) [ sin(t) + (1/3) sin(3t) + (1/5) sin(5t) + . and observed the fit." which I denote by sq(t) : it has period 2pi and sq(t) = = 1 -1 for for 0 < t < pi -pi < t < 0 This is a standard building block for all sorts of "on/off" periodic signals. so the integral is twice the integral from 0 to pi: bn = (2/pi) integral_0^pi f(t) sin(nt) dt f(t) is even then Similarly.. The integrand f(t) sin(nt) is even.. if an = (2/pi) integral_0^pi f(t) cos(nt) dt sq(t) itself needs We have: In our case this is particularly convenient. we can simplify the integral for bn a little bit. bn = = = (2/pi) integral_0^pi sin(nt) dt (2/pi) [ . I used the Mathlet FourierCoefficients to illustrate this. It's odd.cos(n pi) -(-1) ] n : 1 . so we know right off that an = 0 for all n ..cos(nt) / n ]_0^pi (2/pi n) [ . If f(t) is any odd function of period 2pi.[6] Example: A basic example is given by the "standard squarewave.. . I built up the function (pi/4) sq(t) (**) = Actually. since different definitions depending on the sign of t.cos(n pi) 2 0 2 bn = = 0 4pi/n for for n n even odd and This depends upon n 1 2 3 cos(n pi) -1 1 -1 and so on.

Let's review that: Any odd function has a sine series -. April 3 Fun with Fourier series [1] If f(t) is any decent periodic of period 2pi.18. [2] Example: the "standard squarewave" sq(t) = 1 for 0 < t < pi. it has exactly one expression as (*) f(t) = (a0/2) + a1 cos(t) + a2 cos(2t) + ... so b_n = = (4/n pi) 0 if if n n is odd. there is a single list of coefficients such that this is true for every value of t = a for which f(t) is continuous at a.a_n = 0 for all n -and the b_n's can be computed using the simpler integral b_n = (2/pi) int_0^pi f(t) sin(nt) dt sq(t) = 1 .. The coefficients can be computed by the integral formulas a_n = (1/pi) integral_{-pi}^pi f(t) cos(nt) dt b_n = (1/pi) integral_{-pi}^pi f(t) sin(nt) dt but one can often discover them without evaluating these integrals.1 ] In that range int_0^pi sin(nt) dt shows that n | cos(n pi) | 1 .. ] (4/pi) sum_{n odd} (1/n) sin(nt) . .03 Class 21. .. sq(t) = = (4/pi) [ sin(t) + (1/3) sin(3t) + (1/5) sin(5t) + ..cos(n pi) -----------------------------------------0 | 1 | 0 1 | -1 | 2 2 | 1 | 0 .. is even..(1/n) cos(nt) |_0^pi . To be precise. so we must compute = = Now the graph of cos(t) . + b1 sin(t) + b2 sin(2t) + ... -1 for -pi < 0 < 0 has Fourier series sq(t) = (4/pi) sum_{n odd} (sin(nt))/n as we saw by calculating the integrals.(1/n) [ cos(n pi) ...

) General period. the Laplace transform. Example: This has period 2L . 2pi.We can see this on the applet. a1 = b1 = sqrt(2) Example [shifts and stretches]: f(t) = 4 for 0 < t < pi/2 . This means and all the others are zero. which records (pi/4) sq(t) = sin(t) + (1/3) sin(3t) + . f(t) = 0 for pi/2 < t < pi . and even periodic of period 2pi.b) identity we can use: Example [trig id]: How to write it like a cos(t) + b sin(t) = A cos(t . g(x) = = 1 -1 for for 0 < x < L . which record the harmonics above the fundamental tone...(1/3) cos(3t) + (1/5) cos(5t) . It represents a "transform" of the function. b = sin(phi) : For us. there's a trig if (a.cos(theta) so sin(theta + pi/2) = cos(theta) .pi/2) = .pi/4) = sqrt(2) (cos(t) + sin(t)) . [3] Any way to get an expression (*) will give the same answer! 2 cos(t .. The sequence of Fourier coefficients encodes the information of the function.... That's it: that's the Fourier series. f(t) = = 2 + 2 sq(t + pi/2) 2 + (8/pi) (sin(t + pi/2) + (1/3) sin(3(t + pi/2)) + . (*) ? Well. not I want to express this in terms of a function f(t) of a different variable..phi) (A. so has polar coord's a = b = sqrt(2) and cos(t . phi = pi/4. but one of period | t | / / . ) sin(theta .phi) a = cos(phi). This is quite amazing: the entire function is recovered from a sequence of slider settings.pi/4) . f(t) [4] = 2 + (8/pi) (cos(t) . We'll see another example of a transform later. -L < x < 0 2pi.

. discontinuous? What happens at points where f(t) is Definition: A function f(t) is "piecewise continuous" if it is continuous except at some sequence of points $c_n$.) Notation: lim_{t-->c from above} f(x) lim_{t-->c from below} f(x) = = f(c+) f(c-) To find out. More surprising is that the Fourier approximation has visible oscillation about the constant values of sq(t).. I got Matlab to sum the first 10 nonzero terms of the Fourier series for (pi/4) sq(t) . t = a . sin(t) + (1/3) sin(3t) + . + b2 sin(pi x / L) + b2 sin(2 pi x / L) + . ) ) = (4/pi) (sin(pi x / L) + (1/3) sin(3 pi x / L) + ..pi |---/ | /| | / | |/ | -------------| L = f(t) t = (pi/L)x x so we have g(x) = (4/pi) (sin(t) + (1/3) sin(3t) + ... an and bn can of course be translated The integral formulas for into the variable x: an bn = = (1/L) integral_{-L}^L g(x) cos(pi n x / L) dx (1/L) integral_{-L}^L g(x) sin(pi n x / L) dx [5] Gibbs effect. + (1/19) sin(19t) One thing's clear: the value of the Fourier series at This is a general fact: If f(t) isn't continuous at converges to (f(a+) + f(a-))/2 ... that is. 0 is 0 . (If the two limits are equal then the function is continuous there. and for each $n$ both one-sided limits exist at $c_n$.. then the sum at t = a . Then general appearance of the Fourier series for a function of period 2L is g(x) = a0/2 + a1 cos(pi x / L) + a2 cos(2 pi x / L) + ...

Matlab computed the first 100 nonzero terms..and seems to step back before it launches itself across the gap at the discontinuities of sq(t) . but still shows a sharp overshoot near the discontinuities in sq(t) . A. (I believe that he built a mechanical device for the purpose.. but his work was forgotten by everyone till after Gibb's publication.) He communicated his puzzlement to mathematicians. What he found is now called the "Gibbs effect": If f(t) is discontinuous at t = a then for any n there is near t = a such that the sum of the first n Fourier terms at differs from f(b) by at least 8% of the gap. Maybe we just need more terms. The experimental physicist A. and Gibbs took up the challenge and published an explanation in 1899. Michaelson computed the Fourier series to great accuracy for some discontinuous functions and discovered this overshoot in 1898. sin(t) + (1/3) sin(3t) + . The explanation of this strange irrational constant can be found in the Supplementary Notes.. + (1/199) sin(199t) The graph is even flatter. The Gibbs effect was first discovered theoetically by a British mathematician named Wilbraham in 1848..0894898722360836.) times the gap. . t = b t = b The actual limiting overshoot is given by (0.

) dt (4/pi) (cos(t) + (1/9) cos(3t) + (1/25) cos(25 t) + ..18. = pi/4 This is an alternating series.. April 5 Fourier series and harmonic response [1] My muddy point from the last lecture: I claimed that the Fourier series for f(t) converges wherever $f$ is continuous. it is no additional information to say that when -pi < t < 0 . We could calculate the coefficients.... [2] You can differentiate and integrate Fourier series. For example we could take sin(pi/2) = 1 . so 1 . a0/2 is the average value.(1/3) + (1/5) .(4/pi) integral (sin(t) + (1/3) sin(3t) + .pi/4 Then sin(5pi/2) = 1 . For a start. Or we could realize that f(t) + c = = f'(t) = . sin(t) + (1/3) sin(3t) + . = pi/4 (*) This says that when sin(t) + (1/3) sin(3t) + .sq(t) and integrate: . so we know it converges.... sin(3pi/2) = -1 . which is not a whole number multiple of 0 < t < pi ..t between 0 and pi. 2pi = ..03 Class 22..) f To find the constant term.. Example: Consider the function f(t) which is periodic of period and is given by f(t) = (pi/2) . pi.. .(1/7) + .. using the fact that f(t) is even and integration by parts. (pi/4) sq(t) for any value of t = sin(t) + (1/3) sin(3t) + (1/5) sin(5t) + . remember that it's the average value of . Did you know that it converges to pi/4? And so on: there are infinitely many summations like this contained in (*) .. What does this really say? For example. which is pi/2. Since all these sines are odd functions... t = pi/2 ..

e. you'd better have integral_0^{2pi} f(u) du = integral_0^0 f(u) du = 0 .(t). which is f(t) 0: = (4/pi) (cos(t) + (1/9) cos(3t) + (1/25) cos(25 t) + ..] [3] Now we come to the relationship with differential equations: We have a complicated wave. but even before that we optimistically tried x so x' = = B sin(omega t) . what does this mean. the point is that the integral of cos(nt) is periodic unless n = 0 and the integral of sin(nt) is always periodic. ) = (pi^2)/8 1 + 1/9 + 1/25 + .. [Example: sum_{n=1}^infinity 1/n^2 = pi^2 / 6 ... that's the Fourier series for Again. a periodic function f(t). But this IS the case if the average value of the function is zero. think of integrating the constant function 1 for example. This is due to Euler and is one of many analogous formulas. That's it. Suppose we drive an undamped spring system through the spring with it: x" + kx = k f(t) omega_n with omega_n^2 = k : The natural frequency of the system is x" + omega_n^2 x = omega_n^2 f(t) (*) What is the system response? Recall how we solved x" + omega_n^2 x = omega_n^2 A sin(omega t) We could use ERF..) f(t). at (pi/2) or t = 0 ? = (4/pi) (1 + 1/9 + 1/25 + .B omega cos(omega t) .can you see how?] [NB: it is not true in general that the integral of a periodic function is periodic. perhaps a square wave. This can be obtained from the sum of odd reciprocal squares using the geometric series .. If you think of this one term at a time. After all.g.

The circular frequency of the input signal is 1 so perhaps we should expect to see resonance when omega_n = 1... f(t ) has period 2pi and circular frequency Then we will have a particular solution x_p = (4/pi) omega_n^2 (sin(t)/(omega_n^2 . the sin(kt)/(omega_n^2 . periodic.B omega^2 sin(omega t) omega_n^2 x = omega_n^2 B sin(omega t) -----------------------------------------omega_n A cos(omega t) = B(omega_n^2 ..k^2) . By superposition and Fourier series we can now handle ANY periodic input signal. We can use Fourier series to analyze the system response more closely: When term omega_n is very near to k^2 . with sine input. For example. In fact. The system is detecting information about the timbre of the input signal here. but less than k^2 ..1) + sin(3t)/(omega_n^2 .. suppose that f(t) = sq(t) = (4/pi) (sin(t) + (1/3) sin(3t) + .omega^2) When the denominator vanishes we have resonance and no periodic solution. 6 .9) + . and discussed what happens as we change the natural frequency of the system. I showed the Harmonic Frequency Response applet..omega^2)] A x_p = A omega_n^2 cos(omega t)/(omega_n^2 . and all solutions will converge to it.omega^2) cos(omega t) so and B = [ omega_n^2 / (omega_n^2 . k odd. . But there is NOT resonance when omega_n = 2. I showed the Harmonic Frequency Response Applet.) which is a periodic solution. there is resonance when omega_n = 1. and the spring system detects them by responding vigorously when it is tuned to those frequencies. . by changing a capacitor setting or changing the spring constant.. So this is the most significant system response. There are hidden frequencies present in the signal.. in (*). harmonics of the fundamental. 4. 3. ) omega = 1 . there will be a similar particular solution. If there is any tiny damping in the system.x" = . 5.

using this system for the past 40 minutes: this is how the the cochlea. They have different natural frequencies. Then when omega_n passes k^2 the dominant term flips sign and becomes a large positive multiple of sin(kt) . Various more intensely in response to various different as a Fourier analyzer. This appears on the applet. springs.is a large negative multiple of sin(kt) . there is a row of hairs of different lengths. You have been ear works: in They act like hairs vibrate frequencies. Your ear acts along the cochlea. The omega_n axis is the axis .

it takes a small amount of time. What you do care about is the average value nearby that point. we don't care about these details.03 Class 23. The unit step function is a useful building block:-u(t-a) turns on at t = a Q1: What is the equation for the function which agrees with f(t) between a and b ( a < b ) and is zero outside this window? (1) (u(t-b) . At the longer time scale. The basic model is the Heaviside unit step function u(t) = 0 1 for for t < 0 t > 0 Of course a light doesn't reach its steady state instantaneously. u(0-) = 0 . April 7 Step and delta. These are captured by f(a-) f(a+) = = lim_{t-->a lim_{t-->a from below} f(t) from above} f(t) For example. One of the irrelevant details is the light output at exactly t = 0. then it is light.u(t-a)) f(t) . A good class of functions to work with is the "piecewise continuous" functions.Step functions [Heaviside] . you can see what happens. So u(t) is piecewise continuous but 1/t is not. It might move up smoothly. it might move up in fits and starts as different elements come on line.18. you care about lim_{t-->a} f(t) The function is continuous if that limit IS the value at t=a. it might overshoot. which are continuous except at a scattering of points and such that all the one-sided limits exist. . Two additions to your mathematical modeling toolkit. or. u(0+) = 1. If we use a finer time scale. you rarely care about the value of a function at any single point. first it is dark. A function is continuous at t=a if f(a) = f(a-) = f(a+) . t=a.Delta functions [Dirac] [1] Model of on/off process: a light turns on. You will also often care about the values just to the left of or just to the right. more precisely. In fact as a matter of realism. Modeling the process by u(t) lets us just ignore those details.

since u(t) is approximated by differentiable functions.(2) (3) (4) (5) (u(t-a) .u(t-b)) f(t-a) (u(t-a) . Q'(t) and Q(t) = = q(t) integral q(t) = c + t At t = 1 I won $K30 at the race track! I deposit this into the account. so delta(t) does too.u(t-b) f(t-b) none of these Ans: (3). We can also clip and drag: f_a(t) := u(t-a)f(t-a) = f(t-a) = 0 [3] for for t > a t < a From bank accounts to delta functions. t = 1 with the number 30 next . I can model the cumulative total deposit using the step function: Q(t) = c + t + 30 u(t-1) What about the rate? For this we would need to be able to talk about the derivative of u(t) . They all have area 1 under the graph. Q(t) (from some starting time). Using this we can write down a formula for the new rate of savings: q(t) = 1 + 30 delta(t-1) We can graph this using a "harpoon" at to it. I drew some graphs of functions approximating delta(t). But there is nothing to prevent us from using a symbol for a "rate" that plays this role: the "Dirac delta function. Bank account equation: x' + Ix = q(t) x = x(t) = balance (K$) I = interest rate ((yr)^{-1}) q(t) = rate of savings (K$/yr) I am happily saving at K$1/yr. in the usual sense. in such a way that its integral recovers u(t). and the details don't matter. Maybe the bank adds one dollar per millisecond: I don't care.u(t-b)) f(t) u(t-a) f(t-a) ." delta(t) = u'(t) delta is not a function but it is approximated by functions. Just as u(t) has many approximating functions. The concept of rate can be clarified by thinking about the cumulative total. Of course there is no such function. and the nonzero values are concentrated around t=0.

Dirac. whose profound researches into electro-magnetic waves have penetrated further than anyone yet understands. Whenever you have a gap in the graph of f(t) ..A. for the relativistic theory of the electron.. so that f(a+) is different from f(a-) .." A combination of a regular function and a combination of delta functions is a "generalized function": f(t) = f_r(t) + f_s(t) For example.] Paul A. at t = 2.'' up to a . 1980 Quotes: ``I consider that I understand an equation when I can predict the properties of its solutions without actually solving it." [5] Oliver Heaviside. P. Stephen Hawking. Swiss/British theoretical physicist. 1850--1925. 1669 . M. or by a harpoon pointing down with +40 to it.'' He was the one who wrote down Maxwell's equations in the compact vector form you see now on ``Let there be light'' T-shirts. called "singularity functions. Lucasian Chair. next [4] We'll call piecewise continuous functions "regular. It cost K$40: q(t) = 1 + 30 delta(t-1) .. I bought my BMW. The negative multiple of delta can be represented using a harpoon pointing up with -40 next to it.f(a-)) delta(t-a) Keeping these terms in the derivative lets us reconstruct f(t) constant. 1932 . 1664 Isaac Newton. the derivative will have a delta contribution: (f(a+) . 1902--1984. With the singular terms in place this is called the "generalized derivative..M.40 delta(t-2) It makes sense to say that Q'(t) = q(t) . Cambridge University: Isaac Barrow. Not too long after this. q_r(t) q_s(t) = = 1 30 delta(t-1) . British mathematical engineer ``. Nobel prize 1933.the area under the harpoon is 30. Dirac.40 delta(t-2) .." We can now add in combinations of delta functions.

then it falls at constant slope (the acceleration of gravity) till the instant when it hits the ground. . you exert a very large force on the bullet over a very short period of time.(Quoted in Frank Wilczek and Betsy Devine. If we integrate F = ma = mx" we see that a large force over a short time creates a sudden change in the momentum.''] [6] When you fire a gun. The derivative is zero for t < 0 . "Longing for the Harmonies") ``God used beautiful mathematics in creating the world. mx' . when it returns abruptly to zero. Q2: What does the graph of the generalized derivative of like? (1) ^ ^ | | v_0 | | v_0 | | ----------| |---------------- | | |________| ^ | v_0 | | ----------| --------------- | | |________| | | v_0 | v ^ | v_0 | | ----------| -------------- | | |_________| v(t) look (2) (3) (4) ---------- ------------- | | |_________| Ans: (1)." The graph of the elevation of the bullet. starts at zero. then abruptly rises in an inverted parabola. plotted against t. and then when it hits the ground it stops again. then it rises abruptly to v_0. the start is MIT and the end is CalTech. This is called an "impulse. Of course.

m(t) . So f determines a new "function. For example. One answer is that it is a symbol. what this instrument actually measures is M(f. over a period of time. If the light profile is f(t) . M(f. and reports an integrated value." sending each This is what is called a "distribution. representing a certain approximation to reality and obeying certain rules. which rises to a peak and then falls again.m) = integral f(t) m(t) dt The most we can ever know about the function f(t) is the collection of all these measurements. Each measuring device has a sensitivity profile. for example." I will make the assumption that the function (or better). This equipment gathers light. The time interval may be short but it is not of width zero. There are other answers. itself is continuous m(t) There are other ways to assign a number to each measuring device. That is what the "delta function" does: integral delta(t) m(t) dt = m(0) .m) as m varies over all measuring devices.Did not get to say: [7] People often want to know what the delta function REALLY IS. One is this: one measures the value of a function by means of a piece of equipment of some sort. m to a number. we can send m to m(0).

or the damping constant. and solve from this starting point. allow us to predict the system response to other signals. despite the fact that the signal is not. 2006 Unit impulse and step responses [1] In real life one often encounters a system with unknown system parameters. the easier it will be to interpret the system response and get information about the system parameters. The simpler the input signal. For a start. in turn. we should be sure that the system is at rest before we do anything to it. v(t) = = (1/k)(1 . NB: the solution is continuous. (**) x' + kx The solution to xp x = = (**) : xh = c e^{-kt} . April 10. is zero: x(t) = 0 for t < 0 . I'll denote the unit step response by v(t) today.18.e^{-kt}) The solution to (*) called the "unit step response" of the system. But we can watch how it responds to various input signals. t = 0 and watch what Let's just put one kg of plutonium in at . x(t) . (*) (*) is the same as = 1 with x(0) = 0 . So we'll start our experiment at t = 0 . [2] Unit step response. and assume that before it the output signal. Solving ODEs increases the degree of regularity. or the mass.03 Class 24. which will. For a simple model." Then we apply some input signal. This is "rest initial conditions. (1/k)(1 . rest initial conditions. solve x' + kx For t > 0 = u(t) .e^{-kt}) 0 for for t > 0 t < 0 . Suppose we start loading a reactor at a constant rate. [3] Unit impulse response. If it's a spring/mass/dashpot system you may not know the spring constant. so 1/k .

since u' = delta. e. and we'll see next how the unit impulse response can be used to reconstruct the system response to ANY signal.happens. The solution is discontinuous at [4] Two implications of time invariance. and the system response gives a very clean view of the system itself. delta(t) is the rate corresponding to the sudden addition of 1 to the cumulative total. In particular.g. but at least it is regular. (**) x' + kx The solution to The solution to (*) is the "unit impulse response" or the "weight function" and it's written w(t): w(t) = = e^{-kt} 0 for for t > 0 t < 0 t = 0 . x' + kx = delta(t) . p(D). the solution to x = = (b) If e^{-k(t-a)}) 0 = for for x' + kx = delta(t-a) t > a t < a p(D) x' = q'(t) . x = e^{-kt} . For t > 0 (*) is the same as = 0 (**) with is x(0) = 1 . v' = w : the derivative of the unit step response is the unit impulse response. For example. is p(D)x q(t) .e^{-kt}) = e^{-kt} . (*) Remember. This process will work for p(D) of any order. We'll work with a system modeled by an LTI operator The time invariance has two important consequences: (a) It doesn't really matter when you start the clock. The unit step and unit impulse functions are very simple signals. then This is because p(D) D = D p(D) (because the coefficients are constant) so p(d) x' = p(D) Dx = D p(D) x = D q = q' . D (1/k)(1 . rest initial conditions. if the system you are looking at is time-invariant. They determine the system (assuming it is LTI). [5] : Second order impulse response: Drive a spring/mass/dashpot system .

the weight function is a solution to the homogeneous equation p(D)x = 0 for t > 0 .2i 2(s^2 + 2s + 5) e^{-t} (a cos(2t) + b sin(2t)) . Strike a system and watch it ring.contains complete data about the LTI operator p(D) (and so about the system it represents). Take large enough to increase the momentum modeled by mx" + bx' + kx (*) For t > 0 = delta(t) . e. but its derivative jumps at . mx'(0) = 1 is mx" + bx' + kx (**) which we solve using the usual methods.g.the solution to p(D)x = delta(t) with rest initial conditions --. IN GENERAL. 2x" + 4x' + 10x p(s) so We want x = = = delta(t) has roots 1 +.sin(2t)) b (2) so b = 1/4 for for t > 0 t < 0 . The system is Suppose rest initial conditions. That gives you enough information to predict the system response to ANY input signal! In fact there is a formula which gives the system response (with rest initial conditions) to any input signal q(t) as a kind of "product" of w(t) with q(t) .mx" + bx' + kx = F_ext F_ext to be a hammer blow. More about this on Wednesday! . so a = 0 and x = b e^{-t} sin(2t) x(0) = 0: x ' = b e^{-t} (2 cos(2t) . rest initial conditions this is equivalent to = 0 . [6] Convolution. initial conditions x(0) = 0. t = 0 from 0 to 1/m 1/2 w(t) = = = (1/4) e^{-t} sin(2t) 0 It is continuous. I claim that the weight function w(t) --. This solution (times u(t)) the "unit impulse response" or "weight function" of mD^2 + bD + kI . mx' by one unit.

Fix a time t . If c pounds go in at time tau. Suppose phosphates from a farm run off fields into a lake.18. We'll need another letter to denote that changing time. the weight function tells you! w(t . at a rate q(t) which varies with the seasons. 2006 Convolution [1] We learn about a system by studying it responses to various input signals. Delta tau q(t) by a discrete (1 second maybe) . tau. x(0) = 0 The weight function for this system is w(t) = = e^{-at} 0 for for t > 0 t < 0 .the solution to p(D)x = delta(t) with rest initial conditions --. Divide time into very small intervals.tau) q(tau) Delta tau t? . then w(t-tau) c is the amount left at time t > tau. I claim that the weight function w(t) --. April 12. tau . This tells you how much of each pound is left after t units of time have elapsed. the phosphate decays: it's carried out of the stream at a rate proportional to the amount in the lake: x' + ax = q(t) .03 Class 25. In fact there is a formula which gives the system response (with rest initial conditions) to any input signal q(t) as a kind of "product" of w(t) with q(t) . We'll replace the continuous input represented by input. For definiteness let's say q(t) = 1 + cos(bt) Once in the lake. We'll think about how x(t) gets built up from the contributions made during the time between 0 and t . During the Delta tau time interval around time the quantity of phosphate entering the lake is q(tau) Delta tau How much of that drop remains at time Well.contains complete data about the LTI operator p(D) (and so about the system it represents).

and Time Invariant. The interesting thing is that just one single system response suffices to determine the system response to any signal at all. I used the "ConvolutionForward" tool with signal and weight function (written g(t) in the tool) Comment on solution of the ODE By superposition we can solve and add the results. Then the solution of p(D)x = q(t) satisfying rest initial conditions is x(t) = integral_0^t w(t-tau) q(tau) d tau This works for any order. and for systems of any order. Here and in the next few weeks all functions will interest for t > 0. in fact from the graph we could deduce the value of a and b that the programmmer chose. You learn about a system by studying how it responds to input signals. f(t)*g(t) = integral_0^t f(t-tau) g(tau) d tau is the "convolution product. This is based on two major assumptions: Linear. x(t) = integral_0^t e^{-a(t-tau)} (1 + cos(b tau)) d tau but (#) works in general. For us." This takes Each value of the two only be of two functions of t and returns another function of t. each producing a decaying ringing into the future. . In the second order case.phi) + c e^{-at} = 0 x(0) The graphs in the Mathlet show this effect. It is sometimes called the superposition integral. [2] The convolution product.phi) and x' + x = cos(bt) x_p x_p (1/a) + A cos(bt . x' + ax = 1 + cos(bt) : x' + ax = 1 = = 1/a A cos(bt . Restatement: Suppose w(t) is the weight function of p(D) . you think of the input signal as made up of many little blows. and this is the convolution integral. It's superposition of infinitesimals. of the convolution product depends upon the whole of each factors. not just first order systems. They get added up by superpostion. x' + ax x' + ax x = = = 1 : cos(bt) : f(t) = 1 + cos(bt) u(t)e^{-at} .Now we just have to add it all up: x(t) = integral_0^t w(t-tau) q(tau) d tau (#) That's the formula.

" For one thing.tau . In terms of an input/output diagram. (f(t)*g(t))*h(t) is also commutative: f(t)*g(t) = integral_0^t f(t-tau) g(tau) d tau tau = t . . This * is NOT just the product w(t)f(t) . [3] * Other properties. |____________| but also by definition the output is w(t) : w(t)*delta(t) = w(t). ____________ | | delta(t)---->| "w(t)" |---->w(t)*delta(t). Here's a proof using systems and signals: ____________ ____________ | | | | h(t)---->| "g(t)" |---->g(t)*h(t)---->| "f(t)" |---->f(t)*(g (t)*h(t)) |____________| |____________| What is the weight function of the composite system? ____________ ____________ | | | | delta(t)---->| "g(t)" |---->g(t)---->| "f(t)" |---->f(t)*g(t) |____________| |____________| Thus feeding h(t) into the composite system gives But we just saw that it gives f(t)*(g(t)*h(t)) .s .We have just learned that if w(t) is the solution to p(D)w = delta(t) (with rest initial conditions) then w(t)*f(t) is the solution to p(D)x = f(t) (with rest initial conditions). it is associative: (f(t)*g(t))*h(t) = f(t)*(g(t)*h(t)) The book carries out the integration manipulation you need to do to see this. ____________ | | f(t)---->| "w(t)" |---->w(t)*f(t) |____________| In particular. Nevertheless it deserves to be called a "product. ds = .d tau let s = t .

. The output signal is then v(t) . w(0) = 0 . has initial conditions zero: v^{{n-1)}(0) = 0 v'(0) = 0 ... .. with rest initial conditions. with rest initial conditions. That determines the unit step response. = w(t) By definition. Thus: v(t) = w(t)*1 = 1*w(t) = (and of course we integral_0^t 1 w(tau) d tau t = 0) of the unit impulse so we see again that the integral (from response is the unit step response. a_n not 0 a_n D^n + . of p(D)w = delta(t)).. w^{(n-1)}(0) = 0 ... ... Example: Suppose the input signal is f(t) = u(t) use rest initial conditions). Suppose we have a degree p(D) = n LTI operator . Example: f(t) = delta(t) .. + 0 = 1 so v^{(n)}(0+) = 1/a_n w(t) is the derivative of v(t): The unit impulse response w = v' . + a_1 D + a_0 I v(t) The unit step response v(0) = 0 ..e." Also!: and [4] w(t)*(f(t)+g(t)) w(t)*(cf(t)) = = w(t)*f(t) + w(t)*g(t) c w(t)*f(t) Step and impulse response for higher degree operators The convolution integral w(t)*q(t) gives the solution.. to p(D)x = q(t) for any LTI operator p(D) (where w(t) is the unit impulse response.. the solution. To make this useful we have to be able to compute w(t) .. We can calculate a_n v^{(n)}(0+) + 0 + .. i.. so w' = v" . = = = integral_t^0 f(s) g(t-s) (-ds) integral_0^t g(t-s) f(s) ds integral_0^t g(t-tau) f(tau) d tau = g(t)*f(t) . w(t)*delta(t) So the delta function serves as the "*-multiplicative unit. the unit step response.

w(t) is the solution to the homogeneous equation p(D) x = 0 satisfying these initial conditions. . We saw this in case n = 2 on Monday.and w^{(n-1)}(0+) = 1/a_n We have learned: for t > 0 .

03 Lecture 26. [1] The Laplace transform connects two worlds: -----------------------------------------------------------------------| The t domain | | | | t is real and positive | | | | functions f(t) are signals. s-shift law. with discontinuities | | and delta functions | | | | ODEs relating them | | | | convolution | | | | systems represented by their weight functions w(t) | | | -----------------------------------------------------------------------| ^ L | | L^{-1} v | -----------------------------------------------------------------------| The s domain | | | | s is complex | | | | beautiful functions F(s) . often rational = poly/poly | | | | and algebraic equations relating them | | .18. poles diagrams. April 14 Laplace Transform: basic properties. functions of a complex variable. perhaps nasty.

s = a + bi so To compute this limit.| | ordinary multiplication of functions | | | | systems represented by their transfer functions W(s) | | | -----------------------------------------------------------------------The use in ODEs will be to apply L to an ODE.1). Thus: F(s) = 1/s for Re(s) > 0 Re(s) < 0 . solve the resulting very simple algebraic equation in the s world. This goes to infinity with T if a < 0 and to zero if a > 0. there should exist a real number k such that for all large t . in that each value about ALL values of f(t). Example: F(s) f(t) = u(t) : = = = integral_0^infty e^{-st} dt lim_{T --> infty} e^{-sT}/(-s) |^T_0 (-1/s) (lim_{T --> infty} e^{-st} . Here it is. and diverges to the left. . provided that f(t) doesn't grow too fast. Technically. We continue to consider functions (possibly generalized) that f(t) = 0 for t < 0 . and the improper integral fails to converge for [3] This is typical behavior: the integral converges to the right of some vertical line in the complex plane C. write e^{-sT} = e^{-aT} (cos(-bT) + i sin(-bT)) The second factor lies on the unit circle. [2] The definition can be motivated but it is more efficient to simply give it and come to the motivation later. and then return to reality using the "inverse Laplace transform" L^{-1}. F(s) = integral_0^\infty e^{-st} f(t) dt f(t) such [to be emended] F(s) contains information This is like a hologram. so |e^{-sT}| = e^{-aT}.

e^{-i omega})/(2i) we find = s/(s^2 + omega^2) . We'll develop quite a few of these rules. L[e^{zt}f(t)] = F Here's the calculation: L[e^{zt}f(t)] = = = Using f(t) = 1 L[e^{zt}] [5] integral_0^infinity e^{zt} f(t) e^{-st} dt integral_0^infinity f(t) e^{-(s-z)t} dt F(s-z).i omega) + 1/(s + i omega))/2 = (e^{i omega t} + e^{-i omega t})/2 Cross multiplying. Especially. Using linearity and cos(omega t) we find L[cos(omega t)] = (1/(s . since z may be complex. we've computed This calculation (*) is more powerful than you may imagine at first. we can rewrite L[cos(omega t)] Using sin(omega t) = (e^{i omega} . Rule 1 (Linearity): L[af(t) + bg(t)] = aF(s) + bG(s). (*) L[e^{at}] for a real. and has the usual benefits.|f(t)| < e^{kt} In the definition we should add: "for Re(s) large." The expression obtained by means of the integration makes sense everywhere in C except for a few points . [4] This computation can be exploited using general properties of the Laplace Transform.like s = 0 here . and our calculation of its Laplace transform we find = 1/(s-z). If z is any complex number. Rule 2 (s-shift): (s-z). This is clear.and this is how we define the Laplace transform for values of s with small real part. and in fact normally you will not be using the integral definition to compute Laplace transforms.

L[sin(omega t)] [6] = omega/(s^2 + omega^2). = = = = integral_{0-}^infty f'(t) e^{-st} dt e^{-st} f'(t) dt du v = = -s e^{-st} dt f(t) e^{-st} f(t) |_{0-}^infty + s integral f(t) e^{-st} The evaluation of the first term at t = infty is zero. what is f'(t) ? If f(t) has discontinuities... There is one discontinuity in f(t) that we can't just wish away: f(0-) = 0 . Thus: .is zero because f(0-) = 0 . while we had better let f(0+) be whatever it wants to be. Thus.. The evaluation at t = 0. f(t) delta(t-b) = f(b) delta(t-b) . we must mean the generalized derivative. = = In particular. assuming that Re(s) is large enough. Just to keep the notation in bounds. Then f(t) is .. = s F(s) Now. L[delta(t)] = 1 integral_0^infty delta(t-b) e^{-st} dt e^{-bs} L[delta(t-b)] This example shows that actually we should write L[f(t)] = integral_{0-}^infty f(t) e^{-st} dt t = 0. to be sure to include any singularities at [7] The relationship with differential equations: Compute: L[f'(t)] u dv dt . Therefore whenever a < b < c . integral_a^c f(t) delta(t-b) dt = f(b): integrating against delta(t) picks out the value of f(t) at t = b . We have to expect a discontinuity at t = 0 . for b >= 0 . The delta function: Something new about delta(t): If f(t) is continuous at b . by our assumption about the growth of f(t). lets suppose that differentiable for t > 0.

f(0+) .(f')_r(t) (f')_s(t) is the ordinary derivative = f(0+) delta (t) Thus and the generalized derivative is the sum. = f(0+) + L[f'_r(t)] . L[f'(t)] and so L[f'_r(t)] = s F(s) .

. = s F(s) ... = = = = integral_{0-}^infty f'(t) e^{-st} dt e^{-st} f'(t) dt du v = = -s e^{-st} dt f(t) e^{-st} f(t) |_{0-}^infty + s integral f(t) e^{-st} We continue to assume that f(t) doesn't grow too fast with t (so that the integral defining F(s) converges for Re(s) sufficiently large).f(0+) Computations: L[1] = 1/s = 1/(s-a) = = = s/(s^2+omega^2) omega/(s^2+omega^2) L[e^{as}] L[cos(omega t)] L[sin(omega t)] L[delta(t-a)] e^{-as} [1] The t-derivative rule: Compute: L[f'(t)] u dv dt . s-shift: t-derivative: s F(s) . partial fractions: cover-up method. 2006 Laplace Transform II: inverse transform. Definition: F(s) = L[f(t)] = integral_{0-}^infty f(t) e^{-st} dt . Since we are always assuming rest initial conditions. s-derivative rule. t-derivative rule. the evaluation of the first term at infinity becomes zero. April 17. the evaluation at zero is also zero.. Re(s) >> 0 Rules: L is linear: L[af(t)+bg(t)] = aF(s) + bG(s) F(s) essentially determines L[ e^{at}f(t) L[f'(t)] = f(t) = F(s-a) if we omit the singularity of f'(t) at t = 0 .18. use in solving ODEs. This means that for s sufficiently large. Thus .03 Class 27.

if f(t) isn't required to be continuous. X(s) For this to work we have to recover information about f(t) from F(s). [2] In summary the use of Laplace transform in solving ODEs goes like this: L ---------> IVP for x(t) x(t) = . then f(a+) = g(a+). like integration.. since F(s) is defined by an integral. we must mean the generalized derivative. and this is typically what we will use. what one does is look for parts of F(s) in our table of computations. and (2) it does NOT compute the LT of f'(t). L^{-1} <--------- Alg equation for | | solve | v X(s) = .5) + 3X = 1/(s+1) . We can't expect to recover f(t) exactly. We have to expect a discontinuity at t = 0 . x(0+) = 5. which is left unchanged if we alter any individual value of f(t) . Example: Apply Solve L : x' + 3x = e^{-t}. f(a-) = g(a-) for every a . but rather of (f')_r(t) . So if [3] f(t) and g(t) are continuous at t = a. But remember.f(0+) . Then (f')_r(t) (f')_s(t) is the ordinary derivative = f(0+) delta (t) Thus and the generalized derivative is the sum. any occurances of delta functions are the same in f(t) as in g(t). There is one discontinuity in f(t) that we can't just wish away: f(0-) = 0 . What we have is: Theorem: If f(t) and g(t) are generalized functions with the same Laplace transform... = f(0+) + L[f'_r(t)] This is what the book tells us. . Just to keep the notation in bounds. There isn't a formula for L^{-1}. There is no free lunch.that is. L[f'(t)] and so L[f'_r(t)] = s F(s) .. using linearity.Now. while we had better let f(0+) be whatever it wants to be. It's an art. and the singular parts coincide: f_s(t) = g_s(t) -. let's suppose that f(t) is continuous for t > 0 (and is piecewise differentiable). then f(a) = g(a). (1) it is only good if f(t) is continuous for t > 0 . the Step 1: (sX . what is f'(t) ? If f(t) has discontinuities.

but rather to illustrate by a simple example how the method works. This problem is completely straightforward using our old methods: the Exponential Response Formula gives the particular solution xp = (1/2) e^{-t} . the basic homogeneous solution is e^{-3t}. (1/2)/(s+1) + (9/2)/(s+3) we can now recognize both terms: L^{-1}: = (1/2) e^{-t} + (9/2) e^{-3t} . It gives b too: 1/(-3+1) So Step 4: x Apply = X = 0 + b : b = -1/2. I don't show you this to advertise it as a good way to solve this sort of problem. all unwanted unknown coefficients. Two more rules: [4] The s-derivative rule : = = (d/ds) integral_{0-}^infinity e^{-st} f(t) dt integral_{0-}^infinity (-t e^{-st}) f(t) dt .table look-up for Step 2: so Solve for L[e^{at}] X: with a = -1 . but for excitement let me offer: The Cover-up Method: 1/(s+3) Step (ii) Set = Step (i) Multiply through by (s+1) : a + (s+1)(a/(s+3)) or s = -1 : a = 1/2 . and also s + 1 = 0 . This works fine. and the transient needed to produce the initial condition x (0) = 5 is (9/2) e^{-3t}. You have to be somewhat crazy to like this method. Here the method is: Partial Fractions: 1/((s+1)(s+3)) = a/(s+1) + b/(s+3) . Old method: cross multiply and identify coefficients.t f(t). and the t-derivative rule. Thus: F'(s) which is the Laplace transform of . = 1/(3-1) a + 0 : This process "covers up" occurances of the factor (s+1). (s+3)X = 5 + 1/(s+1) X = 5/(s+3) + 1/((s+1)(s+3)) Massage the result into a linear combination of recognizable Step 3: forms.

f_a(t) = 0 . f_a(t) = 0 .L[t f(t)] Sample use: = .F'(s) L[1] L[t] = = = 1/s = s^{-1} . d tau = dt . integral_{a-}^infinity f(t-a) e^{st} dt The lower limit is a because for t < a . so tau L[f_a(t)] = e^{-as} integral_{0-}^infinity f(tau) e^{-s tau} d This integral is precisely F(s) . t < a) . Thus: L[f_a(t)] = e^{-as} F(s) f(t) to be the step function u(t) . The method of wishful thinking suggests inventing a new letter for the quantity t .(d/ds) s^{-3} = (2 x 3) s^ L[t^n] = n! / s^{(n+1)} The general picture is [5] The t-shift rule : Notation: f_a(t) = u(t-a)f(t-a) = = f(t-a) 0 for for t > a t < a The graph of f_a(t) is the same as the graph of f(t) the right by a units. The LT is perfectly fine. a discontinuous one.(d/ds) s^{-1} = s^{-2} . so L[t^2] f(t) = t^2 so t f(t) = t^3 ---> . t = tau + a . . if we take L[u(t-a)] = e^{-as}/s We've found LT of a new signal. though. and e^{-as} is constant in the integral. .a : tau = t . with just a single part to its definition. one whose definition comes in two parts (t > a. so L[f_a(t)] = integral_{0-}^infinity f(tau) e^{-s(tau+a)} d tau By the exponential law e^{-s(tau+a)} = e^{-s tau} e^{-as} . a >= 0 L[f_a(t)] = but shifted to for us. while I want to remember the delta_a(t) in f_a(t) if there happened to be a delta(t) in f(t) . For t < a . we find For example.a .(d/ds) s^{-2} = 2 s^{-3} start with Now take and then {-4} f(t) = t. the choice of variable name inside the integral (u here instead of t) makes no difference to the value of the integral.

f(0+) g(t): Write down the L[g'(t)] So then t-derivative rule for = s G(s) .f(0+) s^2 F(s) .g(0+) . completing the square.03 Class 28. The application of this principle here is to write We'll assume that f(t) and f'(t) ignore singularities at t = 0 . We'll compute it by regarding f"(t) as the derivative of f'(t). 3.F'(s) s F(s) . n = 1. We'll employ a technique here that will get repeated several more times today: pick a new function symbol and use it to name some function that arises in the middle of a calculation.. Apr 21 Laplace Transform III: Second order equations.s f(0+) .f'(0+) s-shift: t-shift: s-derivative: t-derivative: L[e^{at}f(t)] L[f_a(t)] L[tf(t)] L[f'(t)] L[f"(t)] (ignoring singularities at t = 0 ) Computations: L[1] L[e^{as}] L[cos(omega t)] L[sin(omega t)] L[delta(t-a)] L[t^n] = = = = = = 1/s 1/(s-a) s/(s^2+omega^2) omega/(s^2+omega^2) e^{-as} n!/s^{n+1} . t > 0 and are continuous for s F(s) .. .18. 2. [1] To handle second degree equations we'll need to know the LT of f"(t). so that G(s) = L[g(t)] = g(t) = f'(t) . Rules: L is linear: L[af(t) + bg(t)] = aF(s) + bG(s) F(s) essentially determines f(t) = = = = = F(s-a) e^{-as} F(s) .

To apply it without losing your way. In the s-domain we really are finding the inverse of the operator.) Then write the whole expression using (s+1): (2s+7)/(s^2+2s+5) = ((2(s+1) + 5)/((s+1)^2 + 4) The s-shift rule will provide us with the (s+1)'s.2i or s = -1 +. Step 1: Step 2: [3] Apply LT: (s^2 X .this will always be the case.2s .2i . Solving the equation amounts to dividing by p(s). X X = (2s+7)/(s^2+2s+5) + 5/(s(s^2+2s+5)) . or the Key Formula) xp = 1/5 is a solution.3) + 2 (s X .s f(0+) - Example: x" + 2 x' + 5 x = 5 . So: p(s)X(s) = (data from initial conditions) + (LT of input signal) Application of the differential operator p(D) is represented in the s-domain by multiplication by p(s) . (Note that this gives you the roots of p(s): s+1 = +. [4] Back to our example. I recommend using a new function name: write F(s) = (2s+5)/(s^2+4) From the tables. Nevertheless we have some technique to show you in working it out using LT. x'(0+) = 3 .f(0+)) .f'(0+) = s^2 F(s) . the inverse LT so that (2s+7)/(s^2+2s+5) = F(s+1) . if we had used rest initial conditions this would have been zero. of F(s) is . we can look at the terms separately. Step 3: Massage into recognizable bits. which you choose to satisfy the initial conditions. To handle the quadratic denominator we use Method: Complete the square: p(s) = s^2 + 2s + 5 = (s+1)^2 + 4. x(0+) = 2. (2s + 7) is data from the initial conditions. 5/s is the LT of the signal. Look first at the first term. This is one of the attractions of the Laplace transform.2) + 5 X = 5/s X: (s^2 + 2 s + 5) X = (2s + 7) + 5/s Solve for Analysis of the form of this equation: (s^2 + 2 s + 5) is the characteristic polynomial p(s) ! .L[f"(t)] f'(0+) [2] = s (s F(s) . By linearity. Note that this is EASY to solve using our old linear methods: by inspection (or undetermined coefficients. the general solution is this plus a homogeneous solution.

You can find a. a = -1) gives: (*) The s-shift rule (with e^{-t}(2 cos(2t) + (5/2) sin(2t)). Rationalizing the denominator. We can pick either one. x" + 2 x' + 5 x = 5. which are supposed to be real. in the 1 + e^{-t}(cos(2t) + 2 sin(2t)) .( (s+1) + 4 ) / ( (s+1)^2 + 4 ) Step 4: Find numerator of x [5] = L^{-1}[X(s)] is now easy: (remember the omega L[sin(omega t)] ). b. You have to be crazy to like this method of solving x(0+) = 2. say s = -1 + 2i or s+1 = 2i. Roots: (s+1)^2 = -4 so s+1 = +-2i or s = -1 +.2i. in anticipation that I'll need things in that form when it comes time to appeal again to the s-shift rule to recognize things as Laplace transforms. c. or a = 1.f(t) = 2 cos(2t) + (5/2) sin(2t) . We get: 5/(-1+2i) = b(2i) + c. Now look at second term. but notice that we also complete the square: there are constants a.((s+1)+1)/((s+1)^2+4) We can either find L^{-1} of this and add it to what we did before. x'(0+) = 3. we can use the Method: "Complex Coverup": Multiply through by ((s+1)^2+4) and then set s equal to a root of this quadratic. 5/(s((s+1)^2+4) 1/s . (1-2i) = 5(1-2i)/(1+4) = = 2bi + c so b = -1 and c = -1. such that 5/(s((s+1)^2+4) = a/s + (b(s+1)+c)/((s+1)^2+4) Note that I've completed the square and written the numerator using (s+ 1). Notice how useful it was to have things expressed in terms of s+1 here. or (better) not have rushed to find L^{-1} before and assemble things now: X = 1/s . We'll use partial fractions for it. . by cross multiplying and equating coefficients. c. To find a multiply through by s and then set s = 0: 5/(1+4) = a To get b and c. b. using the s-shift rule again. Or you can use the coverup method. We can use this to solve for b and c.

I showed you this to illustrate LT technique. and then by w(t) = (1/2) e^{-t} sin(2t) . not to advertise it as a good way to solve such ODEs.After all. . so the general homogeneous solution is xh = e^{-t} (c1 cos(2t) + c2 sin(2t)) and x = 1 + xh for suitable choice of c1 and c2 . the roots of the characteristic polynomial are -1 +. we have W(s) = 1/(s^2 + 2s + 5) W(s) = Complete the square: Deal with 1/((s+1)^2 + 4) g(t) = (1/2) sin(2t) G(s) = 1/(s^2 + 4) : s-shift. which can be found by substituting in the initial conditions. And the unit step response v(t) is the solution to with rest initial conditions: apply LT: p(D)v = 1/s or V(s) = 1/(s p(s)) for the operator p(D)v = u(t) Example: to find the unit impulse response p(D) = s^2 + 2s + 5 .2i (a fact that we used in the complex coverup). On Monday I'll try to put this all together. xp = 1 ." It has the property that for any complex number r. and talk about what the Laplace transform is really good at. x = W(r)e^{rt} satisfies p(D)x = e^{rt} . The method of Laplace transform actually is quite good at solving ODEs if the initial conditions and the signal are as simple as possible: The weight function of p(D) is the solution to with rest initial conditions: apply LT: p(D)w = 1 = or 1/p(s) W(s) = 1/p(s) p(D)w = delta(t) L[w(t)] is the "transfer function.

] If 1/W(s) is not a polynomial. if you like. then so we discover. How? Well. that the mass is 1/2.2i .18. The roots don't quite determine the polynomial. the damping constant is 1. since you can always multiply through by a constant and get another polynomial with the same roots. [Of course we knew that. Laplace transform is the device for extracting the system parameters from the unit impulse response.03 Class 29. Apr 24 Laplace Transform IV: The pole diagram [1] I introduced the weight function = unit impulse response with the mantra that you know a system by how it responds. One of the virtues of the Laplace transform methodology is that it can be used to analyze systems whether or . If you normalize to s^2 + bs + k then b = .1 +. from w'(0) = 2 : the change in momentum is 1.(sum of roots) = 2 k = product of roots = 5 so up to a constant you get s^2 + 2s + 5 The constant is the mass. so if the change in velocity is to be 2. take the equation conditions) p(D) w = delta(t) (with rest initial F[f'(t)] = sF(s)): and apply LT to it (in the original form. and this can be derived too. If the unit impulse response is W(s) and 1/W(s) = (1/2)s^2 + s + (5/2) = (2)/((s+1)^2+4) e^{-t}sin(2t) for example. so p(s) W(s) = 1 so the Laplace transform of the weight function W(s) = 1/p(s) (*) w(t) is That is. and the spring constant is 2. the mass must be 1/2. so the roots of the characteristic polynomial are visible and must be . so if you let it respond to the simplest possible signal (with the simplest possible initial conditions) then you should be able to determine the system parameters. too: the impulse response is (for t > 0) a homogeneous system response. you have discovered that the system is not modeled by a differential operator.

for example. L[f(t)*g(t)] = F(s) G(s) L to This is consistent with other things we know." (Here we are supposing that the "physical input signal. [2] A few weeks ago we described the system response (with rest initial conditions) to a general input signal q(t) in terms of the unit impulse response w(t): p(D) x has solution = q(t) = with rest initial conditions w(t)*q(t) . apply f(t)*delta(t) F(s) 1 [3] = = f(t) F(s) and get check! Exponential signals The "transfer function" W(s) directly determines the system response to (almost) any exponential signal: Exponential response: x_p = p(D) x = e^{rt} has an exponential solution W(r) e^{rt} The transfer function is the Laplace transform of the weight function. LT to (*) we find (*) x(t) On the other hand.phi) |W(i omega)| Arg(W(i omega)) W(i omega) is the "complex gain.phi = = cos(omega t) e^{i omega t} W(i omega) e^{i omega t} Re[ W(i omega) e^{i omega t} ] gain(omega) cos(omega t . This can be used to find Sinusoidal response: p(D)x p(D)z z_p x_p = = = = = where gain(omega) ." with respect to which we should be measuring . if we apply p(s) X so X = = F(s) W(s)F(s) We have discovered an important principle (which can be proved by direct application of the definitions): Laplace transform converts convolution product of functions of t to ordinary product of functions of s .not they are controlled by a differential equation.

W(s) = a/(s-r1) + [5] Here's the vision that unifies most of what we have done in this course so far: You have a system (a black box. to get W(s) . This means really that you want to be able to predict its response to various input signals. with springs and masses and dashpots.1 . so its graph is a surface lying over the complex plane. Periodic signals decompose into sinusoidal signals. It sweeps up to infinity as s ---> 0 . The graph b/(s-r2) . for example) which you wish to understand.the gain and the phase lag.) [4] How can we understand the function 1/s as s varies? Just try to understand |1/s| . so it's enough just to study sinusoidal system responses. Look at W(s) = 1/p(s) . Maybe it's for this reason that we call s = 0 a pole of 1/s . You are especially interested in its periodic response to periodic signals. and delaying the input signal just results in delaying the system response. This is 1/|s| . with a tent post stuck in the ground at s = 0 . It falls off towards zero when s of |1/p(s)| is a tent with two poles. as are the roots. This will be a surface lying over the complex plane. So hit the system: feed it What comes out is Apply Graph L to w(t) w(t). We will only be able to analyze systems which are LINEAR and TIME INVARIANT: so superposition holds. It's like a tent. is just the input signal. 1/(distance from 0). In fact by partial fractions. This factors as p(s) = (1/2)((s-r1)(s-r2) where r1 r2 = = .2i s comes to be one of moves away. This is a function on the complex plane. W(s) then becomes infinite when r1 or r2 . put the argument aside for another day. by Fourier series. delta(t) as input signal. You'll be happy to understand the gain. There will be a gain and a phase lag involved. . |W(s)| . Suppose p(s) = (1/2)(s^2 + 2s + 5) above.1 + 2i . and leave th phase lag for another day. that is.

and eventually the two humps in the frequency response curve merge. and a more complicated amplitude response curve. you get more poles. Near resonance occurs because poles of W(s) . If you have a higher order system. . allowing negative omega). The intersection of the graph of W(s) with the vertical plane lying over the imaginary axis is the amplitude response curve (extended to an even function.Restrict s to purely imaginary values . the poles move deeper into negative real part space. i omega is getting near to one of the If you increase the damping. s = i omega. is needed to study sinusoidal input response: p(D) x = e^{i omega t} This is what has exponential solution x_p so |W(s)| = W(i omega) e^{i omega t} is the gain.

With two rabbit populations. x_1 x_2 = = e^{3t} e^{-3t} y . using +-3. The general case looks like x' y' = = ax + by cx + dy y to solve It seems to be impossible to solve.18. so rabbits from field 1 jump at the rate of 5 yr^{-1} . since you need to know for x and you need to know x to solve for y.9x = = 5x + 2(x' + 2x) 0 This is a SECOND ORDER ODE . Field 1 contains x(t) rabbits. Each gives a corresponding solution for y_1 y_2 = = 5e^{3t} -e^{-3t} y = x' + 2x .03 Class 31. 2006 First order systems: Introduction [1] There are two fields in which rabbits are breeding like rabbits.5x + y 3y . while rabbits from field 2 jump only at the rate of 1 yr^{-1}. each derivative is increased by virtue of the influx from the other field. though. The grass is greener in field 2. homogeneous.y + 5x = = -2x + y 5x + 2y (1) (2) The net growth rate of the field 1 population is -2 because of all the jumping. Note that the rabbits cancel. So the equations are x' y' = = 3x . which we can solve: The characteristic polynomial is s^2 = 9 . and the roots are We get two basic solutions. by a process called ELIMINATION: use (1) to express y in terms of x: y = x' + 2x and then plug this into (2): or x" + 2x' x" . April 28. In both fields the rabbits breed at a rate of 3 rabbits per rabbit per year. and the net growth rate in field 2 is 2. so the units are (year)^{-1} . On the other hand. Each of the four coefficients has a clear interpretation. This is a linear SYSTEM of equations. We can solve. we should clearly graph this on the x vs y plane: Notice that y_1 = 5x_1 : so this solution moves along the line through the . They can also leap over the hedge between the fields. field 2 contains y(t) rabbits.

Of course there are a lot of anti-rabbits hopping around here which are of interest to mathematicians but not to biologists." It does not show complete detail of the solutions. y ] = [ ax+by . We are studying x' = ax + by y' = cx + dy (*) We can represent linear equations using matrices. Each curve in the phase portrait is the trajectory of many different solutions. Matrix multiplication is set up so that | a | c or b | | x | d | | y | = | ax + by | | cx + dy | [ a b . The general solution is a combination of them: x y = = a e^{3t} + b e^{-3t} 5a e^{3t} . the population of field 2 tends to 5 times the population of field 1. cx+dy ] (*) can thus be written as The ODE . so that solution moves along the line These are two "trajectories" of the rabbit populations. it does not show the time at which the solution passes through a given point.origin of slope of slope -1. you can go through a point at any time you want. c d ][ x . c d ] There is another matrix in sight. the "column vector" | x | | y | or [ x . The matrix of coefficients of (*) is the array of numbers (enclosed by brackets) A = | a | c b | d | In these notes I will use Matlab notation and write this array as [ a b .b e^{-3t} In the long run. y ] with entries x and y . Said differently. y_2 = -x_2 . [2] Analysis of linear equations by matrices. For any given point there is a unique solution passing through it. 5 . This picture is a "phase portrait.

by solving the original equation x" . y' ] = [ a b .x' + 4x = 0 We can derive a first order linear system from this. This is the "phase space" picture of the solutions of the original . but is off phase. -4 1 ] is the "companion matrix" of the original second order equation. omega = sqrt(15)/2 ~ 1. The result is that the trajectory traced out by (x. Here's an important source of systems of equations. Companion matrices have top row [ 0 1 ] .94 The general solution is thus x = A e^{t/2} cos(omega t .4 x = -4 x + y = x' Together we have x' y' = = y -4 x + y This is a first order constant coefficient homogeneous linear system whose matrix A = [ 0 1 .phi) These oscillate under an exponentially growing envelope.y) is an expanding spiral. c d ][ x .i omega . y ] then u' = [ x' . We can see more precisely what the trajectories are in this case. Suppose we have a second order homogeneous linear equation.[ x' . by the trick of defining y so then y' = x" = x' . The derivative does the same. u for the column vector [ x . y ] If we write y' ] .x' + 4 x = 0 Its characteristic polynomial is p(s) = s^2 .s + 4 . say x" . You can find its roots using the quadratic formula: -(1/2) +. and u' = Au This compact expression is exactly equivalent to [3] Companion matrices: Anti-elimination (*) .

On the other hand. the change in his feelings towards Juliet has nothing to do with how he himself feels at the moment. he notices and his rate of increase of disinterest starts to ameliorate. The phase portrait of the companion system of a second degree equation shows the values of both the solution x and its derivative x' . if he seems to love her. He then starts to feel better towards her. If R denotes Romeo's love for Juliet. This causes him to continue to warm to her. . she does notice that he is fond of her.second order equation. So Romeo is fond of Juliet but she is neutral towards him. but now R has increased.0). We can see x' recorded vertically. She warms to him. However. We all know the sad outcome. you wind up at J = 0 again. and now both his attitude and hers cause her to feel progressively more well disposed towards him. it is completely dependent on how she feels about him. Following this around. Juliet is more complex. presently he stays away from her. just as he bottoms out. then R' J' = = J -R + 4J Romeo is a puppy dog. and J denotes Juliet's love for Romeo. This continues. She has a healthy self awareness. As she becomes more distant. if she loves him. and this very fact makes her more interested. but with each cycle the intensity increases. Eventually he is neutral and she really doesn't like him. He has little selfawareness. The MIT Humanities Department has analyzed the plot of Shakespeare's play and found the following. but still stays away. that very fact causes her to love him more. [5] It turns out that the same system models the relationship between Romeo and Juliet. This is a cyclical relationship. and this makes her somewhat hostile. Let's start the action at (1. Eventually she is neutral. she gets frightened and starts to love him less. his affection wanes.

We get a nonzero solution [x . This Greek letter is always used in this for some number context. A v = lambda v lambda. c d] and u = [x . so we have been thinking about A u = 0 as an equation in u . When is this product zero? One way is for x = y = 0. 0] : 0 is a linear combination of the two columns in a "trivial" way. y] .03 Class 32. [2] Solve u' = Au : for example with A = [1 2 . this is the ONLY way. May 1 Eigenvalues and eigenvectors [1] Prologue on Linear Algebra. If A is a larger *square* matrix the same theorem still holds. [3] This is a pure linear algebra problem: A is a square matrix. c d] [x . That is to say. c] and [b . we can find x and y so that the sum cancels.bc = 0. 2 1] . y] = x[a . c] + y[b . d] : Recall A matrix times a column vector is the linear combination of the columns of the matrix weighted by the entries in the column vector. c] and [b . It always has the "trivial" solution u = 0 = [0 . Let's find them first. y] exactly when the slopes of the vectors [a . d] coincide: c/a = d/b . If [a . and we are asking when it is a linear combination of them in a different. we are going to look for a solution of the form u(t) = r(t) v One thing for sure: the velocity vector u'(t) also points in the same (or reverse) direction as u(t). d] point in different directions. The "Linear Phase Portraits: Matrix Entry" Mathlet shows that some trajectories seem to be along straight lines. with the appropriate definition of the number det A . Write A = [a b . [a b .bc We have found: Theorem: Au = 0 has a nontrivial solution exactly when det A = 0 . and .18. This combination of the entries in A is so important it's called the "determinant" of the matrix: det(A) = ad . or ad . "nontrivial" way. But if they lie along a single line. with 0 coefficients. So for any vector v on this trajectory.

lambda I) v v . -1 ] = 0 All we are claiming is that . In order to get all the v's together. In our case. In our example. and there are two roots.lambda I ) A .lambda I) = 0 This is an equation in lambda . we will find lambda first. write the right hand side as lambda v = (lambda I) v where I is the identity matrix [1 0 . 1 .3 det ( A .lambda I = lambda from both diagonal entries [ 1 . and its roots are the "characteristic values" or "eigenvalues" A .we are looking for nonzero vectors v such that A v = lambda v for some number lambda. There is a nontrivial linear relation between the columns: A [ 1 .4 1 . There is one line for lambda_1 and another for lambda_2 . We have to find nonzero solution v to (A .4 lambda^2 . we are looking for a nonzero exact condition for such a solution: det(A .3) and lambda_2 = 3 . 2 2 ] lambda = lambda_1 = -1 .lambda = [ 2 2 . Then we can put this on the left: 0 = A v . 2 .lambda I) v eg with = 0 A .lambda I ) = = = This is the "characteristic polynomial" p_A(lambda) of of = det( A .2 lambda . lambda_1 = -1 [4] Now we can find those special directions. then. p_A(lambda) = (lambda + 1)(lambda . we subtract and then take the determanent: A .lambda ] (1-lambda)(1-lambda) . and lambda I is the matrix with lambda down the diagonal. 0 1] .lambda .2 lambda + lambda^2 .(lambda I) v = (A . 2 . We have just found an Don't forget. and then set about solving for v (knowing in advance only that there IS a nonzero solution).

We have found that there is a straight line solution of the form r(t) v where v = [1. 1] We can solve for example c1 and c2 using an initial condition: say for . -1] + c2 e^{3t} [1 .1] lambda = 3 . Back to the differential equation.1] [5] The general solution to u' = Au will be a linear combination of the two eigensolutions (as long as there are two distinct eigenvalues). We have r' v so (since r' = v = u' = A u = A rv = r A v = r lambda v is nonzero) lambda r and solving this goes straight back to Day One: r so for us solution: u = = c e^{lambda t} r = c e^{-t} and we have found our first straight line e^{-t} [1." General fact: the eigenvalue turns out to play a much more important role than it looked like it would: the straight line solutions are *exponential* solutions. [ -1 1 .-1] In fact we've found all solutions which occur along that line: u = c e^{-t} [1. A . 1 -1 ] v so [1. In our example. an "eigenvector" of A. -1 ] Any such v (even zero) is called and you can check this directly. -1 ] = . leads to [ -1 1 .1] is a nonzero eigenvector for the eigenvalue and there is another straight line solution e^{3t} [1.lambda I and = lambda_2 = 3 . The second eigenvalue. the general solution is u = c1 e^{-t} [1 . where lambda is an eigenvalue for the matrix and v is a nonzero eigenvector for this eigenvalue.-1] . e^{lambda t} v .-1] Any one of these solutions is called a "normal mode. 1 -1 ] = 0 has nonzero solution v = [1.A [ 1 .[ 1 .

Also. -10. Well. u(0) = c1 [1 . the two terms become comparable and the solution curves around. p_A(lambda) tr A = 2 = and det A = 3 . those for which b = c .b. -1] + e^{3t} [1 . When t is very negative.u(0) = [2 . Most real numbers we know about are eigenvalues of symmetric matrices . But for a symmetric matrix. for example.3 (2) Any multiple of an eigenvector is another eigenvector for the same eigenvalue. A . lambda^2 . As t gets near zero. they form a line. 10.(a+d) lambda . 1]. -1] + c2 [1 . 0]. that is a special feature of *symmetric* matrices. an "eigenline. The general solution is a combination of the two normal modes. say.the mass of an elementary particle.d] is p_A(lambda) = (a-lambda)(d-lambda) . [6] Comments: A = (1) The characteristic polynomial for the general 2x2 matrix [a. may be complex. 1] = [c1+c2 . roots of the characteristic polynomial.c.2 lambda . 1] . As t gets large. all the eigenvalues are real. the second term is very big and the first is tiny: the solution becomes asymptotic to the line through [1 . say. Both these facts hold in higher dimensions as well.bc = lambda^2 . -1]. . and .(ad-bc) The sum of the diagonal terms of a square matrix is the "trace" of tr A. generally the eigenvalues. -c1+c2] and for this to be [2 . so p_A(lambda) = lambda^2 .(tr A) lambda + (det A) In our example. not real." (3) The eigenlines for distinct eigenvalues are not genearally perpendicular to each other. the first term is very big and the second tiny: the solution is very near the line through [1 . 0] we must have c1 = c2 = 1: u(t) = e^{-t} [1 .

lambda_1 = 1 + 2i. or (A . (3) For each eigenvalue find a nonzero eigenvector --Av --say = lambda v v_2. which spirals and obviously has no such solution? Or. I showed the trajectories on Linear Phase Portraits: Matrix Entry. i ]. Then the "ray" solutions are multiples of e^{lambda_1 t} v_1 and e^{lambda_2 t} v_2 These are also called "normal modes.) We could abandon the effort at this point.lambda I) v = 0 v_1 .(tr A)lambda +(det lambda_2 v such that (2) Find its roots. -2 1 ] . but we had so much fun and success with complex numbers earlier that it seems we should carry on. tr(A) = 2 . easier. the eigenvalues lambda_1. so p_A(lambda) which has roots = lambda^2 . Let's apply the method and see what happens. 2 . 0 ] A . (As always for real polynomials. May 3 Complex or repeated eigenvalues [1] The method for solving u' = Au that we devised on Monday is this: (1) Write down the characteristic polynomial A) p_A(lambda) = det(A .2i . This is set up so the top entry in the product is 0 . We have a chance to check our work (mainly the calculation of the eigenvalues) by . But what about the Romeo and Juliet example. -2i ][ ? . ? ] Standard method: use the entries in the top row in reverse order with one sign changed: [ 2 ." The general solution is a linear combination of them.(1+2i)I : [ .2 lambda + 5 lambda_2 = 1 . what about A = [ 1 2 . [2] This makes you think there are always ray solutions.18. Find an eigenvector for lambda_1 = 1 + 2i : = [ 0 . 2i ] or. in this case.lambda I) = lambda^2 . the roots (if not real) come as complex conjugate pairs.2i. v_1 = [ 1 . -2 .03 Class 33. det(A) = 5.

1] ) and so e^t [cos(2t) . -i ] v_1: .1] . Both spiral around the origin.sin(2t) [0. Note: Since lambda_2 = conjugate of lambda_1. it's easy to find a and b from u(0): u(0) = a[1. perhaps.1] + sin(2t) [1. and doesn't represent a point on the plane. As in the case of second order equations.0] + i[0.0] .sin(2t)] Im(u) = e^t ( cos(2t) [0. 1 .0] . clockwise. They satisfy u_1(0) = [1.1] so a = x(0) and b = y(0) . cos(2t)] These are two independent real solutions.seeing that the bottom entry in the product is -2 . I showed their trajectories on the Mathlet Linear Phase Portraits: Matrix Entry. It's hard to visualize. . It is very hard to visualize the fact that all those spirals are linear combinations of any two of them. the real and imaginary parts of solutions are again solutions: So these are real solutions: e^{(1+2i)t} [ 1 . but we can still compute with it just fine.0] + b[0. In this case. u_2(0) = [0. a. i = 0 0 too: [ 1 . i ] = u_1 = = u_2 = = e^t ( cos(2t) + i sin(2t) ) ( [1.2i .0] ) e^t [sin(2t) . In general you have to solve a linear equation to get a and b . an eigenvector for lambda_2 is given by the conjugate of v_2 = [ 1 . while fleeing away from it exponentially. b real . so be it. So one normal mode is e^{(1+2i)t} [ 1 . OK. i ] [3] But we wanted real solutions.1] ) Re(u) = e^t ( cos(2t) [1. The general real solution is u = a u_1 + b u_2 . i ] is a vector with complex entries.

a "repeated eigenvalue": -1. So there is (up to multiples) only one normal mode: u_1 = e^{-t} [ 1 . -1 0 ] = lambda^2 + 2 lambda + 1 = (lambda + 1)^2 lambda_1 = lambda_2 = p_A(lambda) which has only one root. -i ] . ? ] has solution [0. Still. In our case: [ -1 1 .so another normal mode is e^{(1-2i)t} [ 1 . but they are complex and don't show up on our real phase plane. Here is how to find one. ALL eigenvctors for A lie on the line containing 0 and [ 1 . I won't go into details.(-1)I . So we discover that the possibility of complex eigenvalues really isn't a failure of the method at all.1] .lambda_1 I) w Then u_2 = e^{lambda_1 t} (t v_1 + w) = A . 1 ] .lambda_1 I v_1 . Summary: Nonreal eigenvalues lead to spiral solutions. Positive real parts lead to solutions going to infinity with t ("unstable") Negative real parts lead to solutions going to zero with t ("stable") Zero real parts lead to solutions parametrizing ellipses. find an eigenvector: A . but now find a vector w is a second solution. just give you the method. I moved the sliders on Linear Phase Portraits: Matrix Entry to show the phase plane. so its real and imaginary parts give the same solutions we had before (up to sign). 1 ] = [ -1 1 . [4] Second problem with our method: Illustrated by A = [ -2 1 . 0 ] : v_1 = [ 1 or any nonzero multiple. This is complex conjugate to the one we had before. 1 ]. -1 1 ][ ? . so = [ 1 . ? ] = [ 0 . 1 ] But we need another solution. which shows only one pair of opposed ray trajectories. -1 1 ] [ ? . Write down the same matrix such that (A . There are in fact ray solutions.

[5] A matrix with a repeated eigenvalue but only one lineful of eigenvectors is called "defective. 0 . .1] + c v_1 for any constant c." A matrix can have a repeated eigenvalue and not be defective: A = [ 2 0 .1] ) e^{-t} [ t .06 or 18/700. [0. 0 . A . and every solution is a normal mode. 0 ] [ 0 0 . lambda_1 ] For larger square matrices this becomes the story of Jordan form. This is called the "complete" case.1] . u_2(0) = [0. With this choice. if the eigenvalue is repeated you are in the defective case unless the matrix is precisely [ lambda_1 . For any vector v . It doesn't matter which one you pick.1] . ? ] Now ANY vector is an eigenvector! Instead of only one line you get the entire plane. 0 2 ] for example has characteristic polynomial p(lambda) so = lambda^2 . e^{2t} v is a solution. 0 0 ] [ ? .lambda_1 I : To find an eigenvector consider = [ 0 . Then you don't need the painful procedure described in In the 2x2 case. The general solution is u = a u_1 + b u_2 .1] + [0. u_1(0) = [1.2)^2 lambda_1 = lambda_2 = 2. t+1 ] does too [0.1] isn't the only vector that works here. To learn more about all this you should take 18.u_2 = = e^{-t} ( t [1.4 lambda + 4 = (lambda .

they determine the eigenvalues. b.03 Class 34. If the eigenvalues are real. tr(A) = lambda_1 + lambda_2 .2 are not real if are equal if 2/4 are real and different from each other if The boundary is the "critical parabola. so p_A(lambda) lambda_2) = = (lambda .bc. . Conversely.lambda_2) lambda^2 .18.2 = tr(A)/2 +. lambda1. this can be rewritten as A = [a b . Its roots are the eigenvalues. det(A) > tr(A)^2/4 det(A) = tr(A)^2/4 det(A) < tr(A)^ lambda1. tr(A) = a + d . as the roots: by the quadratic formula. d. det(A) = lambda_1 lambda_2 so the two numbers tr(A) and det(A) . are determined by the eigenvalues. May 5 Classification of Linear Phase Portraits The moral of today's lecture: Eigenvalues Rule (usually) A is [1] Recall that the characteristic polynomial of a square matrix p_A(lambda) In the 2x2 case p_A(lambda) where = det(A . c d ] = lambda^2 .det(A)).(tr A) lambda + (det A) det(A) = ad . extracted from the four numbers a.lambda I).(lambda_1 + lambda_2) lambda + (lambda_1 Comparing coefficients.sqrt(tr(A)^2/4 . Notice that if the eigenvalues are not real. c. they have the same sign exactly when their product is positive. their real part is tr (A)/2." where det(A) = tr(A)^2/4.lambda_1)(lambda . and that sign is positive if their sum is also positive.

real > 0 . one with positive eigenvalue and the other with negative. det = -3. The eigenvalues determine many general characteristics of the solutions.. the phase portrait is a "saddle. .<----. ------------> tr ^ | real. the spiral is moving counterclockwise. . 2 1 ] as we saw on Monday has phase portrait as shown on the Linear Phase Portraits: Cursor Entry. the e^{-t} gets . Two more comments on this: (1) The spirals move IN when Re(lambda) < 0 OUT when Re(lambda) > 0 "stable spiral" "unstable spiral" When Re(lambda) = 0 it turns out that the trajectories are ellipses. it is moving clockwise.. | . Normal modes along one move out.. | . with tr = 2 .Relationship between tr and det vs eigenvalues det ^ |<-----purely imaginary complex roots . A convenient one to pick is [1.. and along the other move in.. -----------. Re < 0 | Re > 0 . opposite sign |______ at least one zero e. | . s = 0. we get spirals.. If you take a solution which is not a normal mode. the one of slope -1 has eigenvalue -1. When the eigenvalues are real and of opposite sign. if negative. [2] Spirals." There are two eigenlines.repeated . For example A = [ 1 2 .v.0] : then u' = Au = (first column of A). | . .. The technical term for this type of phase portrait is "center. and theta = 0 . The one of slope +1 corresponds to eigenvalue 3. | ... when time gets large. | . The eigenlines are of slope +-1 . We have seen that when the eigenvalues are non-real.. . The general solution is a combination of these two. ." (2) We can tell which you get by thinking about what u' is at some point. . So if the bottom left entry is positive. [3] Saddles. real < 0 .

1 . nodes . there are many matrices with this same pair of eigenvalues. theta = pi/2) you can see that the eigenlines are again of slope +. -----------. s = 0. | . | . -1 ] = = [ -1 . when tr = 0 they are of equal size. | . On the Mathlet (with tr = -4. [4] Nodes: When the eigenvalues are real and of the same sign.. but distinct...1 : slope +1 with eigenvalue -1 and the other with eigenvalue -3." Eg 3) A = [ -2 . you have a "node. 1 .. . -2 ][ 1 . . and the e^{3t} is like the cube of 1/e^{-t}: so the solution converges to the -1 eigenspace more quickly than it does to the +3 eigenspace. The upper left box lets us explore them. spirals | spirals . 1 ] [ -2 . 1 .small. <--stable | unstable .. ------------> ^ | saddles |______ . . [5] The corresponding phase portraits exhibit the following behaviors: det ^ |<-----centers . stars or defective nodes tr degenerate cases . By moving the trace slider you change the relative size eigenvalues. One thing you can do is rotate the whole picture. stable . 1 . | . . but the one with eigenvalue -3 decays much faster: so the non-normal mode trajectories become tangent to the eigenline with smaller |lambda|. | .. The other thing you can do is change the angle between the eigenvectors.. -2 ][ 1 .. unstable nodes . -1 ] [ -3 . . You can check this: [ -2 . 1 . 1 . of the Returning to tr = 2 . | .. -2 ] = has = (lambda + 1)(lambda + p_A(lambda) lambda^2 + 4 lambda + 3 so the eigenvalues are -1 and -3. 3 ] Both normal modes decay to zero.. det = 3.

or [6] Stability: All linear systems fall into one of the following categories: Asymptotically stable: all solutions ---> 0 as t ---> infinity. These systems occupy the upper left quadrant. det = tr^2/4 . then the constant vector valued function u(t) = c v is a solution for any constant c: there is a line (at least) of constant solutions. otherwise. Several patterns are possible. Ellipses and stable combs are examples. Neutrally stable: all solutions stay bounded but most don't ---> 0 as t --> infinity. and they are illustrated in the Supplementary Notes. . so all the eigenvalues have negative real part. along the critical parabola: repeated real eigenvalues. det = 0 : "Degenerate. Unstable: most solutions ---> infinity as t ---> infinity Saddles and unstable nodes and spirals are examples. tr < 0 and det > 0. 0 lambda1 ] defective nodes . . If v is an eigenvector corresponding to this eigenvalue.There are also the special cases that happen along the curves separating these regions: ." At least one of the eigenvalues is zero. in the complete case [ lambda1 0 . The phase portraits are either stars.

it is the same as the characteristic polynomial of the original second order equation. and can have at most one critical point (zero slope).18. May 8 The companion matrix and its phase portrait.det) plane corresponds to Ans: The part of the upper left quadrant which is below the critical parabola. can cross the x = 0 axis at most once. then to find an eigenvector we look for v such that [ -lambda_1 . QUESTION 1: k > 0? What region in the (tr.e.. QUESTION 2: What region in the overdamping? (tr. -3/2 ] . This makes sense: has derivative x' = lambda_1 e^{lambda_1 t} . i. The companion system is obtained by setting x' y' = = y .det) plane corresponds to c > 0. and the characteristic polynomial of A is lambda^2 + b lambda + k. 1 . * ] v = 0 x = e^{lambda_1 t} v = [ 1 . det(A) = k.by whose matrix of coefficients is the "companion matrix" A = [ 0 1 .03 Class 35.kx . lambda_1 ] is a solution to the companion system. Ans: the upper left quadrant. * . If lambda_1 is an eigenvalue of a companion matrix. [1] We spent a lot of time studying the second order equation x" + bx' + kx = 0 and if b and k are nonnegative we interpreted them as the damping constant and spring constant (divided by the mass). -k -b ] Note that tr(A) = -b . The companion matrix is [ 0 . where there are stable nodes. The matrix exponential: initial value problems. lambda_1 ] does nicely. For example x" + (3/2)x' + (1/2)x = 0 is overdamped. so [ x . -1/2 . x' ] = e^{lambda t} [ 1 . 1 . Roots of the . We saw long ago that solutions to overdamped equations decay as time increases.

Final reminder: in the companion system case.characteristic polynomial. Why not make an analogous definition in the case of a system of equations? That is. then reaches a southerly extreme at C. well approximated by c e^{-t/2} for t >> 0. so its derivative is about . a stable spiral trajectory corresponds to a damped oscillation. e^{at} is DEFINED to be the solution to x' = ax with initial condition x(0) = 1 . and finally heads in towards zero as t ---> infty. -1/2 ] and e^{-t} [ 1 . [2] The Matrix Exponential x' = ax with initial condition x(0) has Recall from day one that solution x = x(0)e^{at} . What does the graph of the corresponding solution to x" + (3/2)x' + (1/2)x = 0 look like? Where are the points on it corresponding to A.e. B.c e^{-t} . Also improper nodes correspond to critically damped solutions. then hooks around. reaching an easterly most extreme point at B when it crosses the x axis. It then decays exponentially to zero. define the symbol e^{At} so that the solution to u' (**) = Au with initial condition u(0) u(0) is u(t) = e^{At}u(0). so the derivative is about (-c/2) e^{-t/2} . is a vector Note that the initial value valued is a vector. -1/2 ] and [ 1 . This is what led us (and Euler before us) to the expression e^{(a+bi)t} = e^{at} (cos(bt) + i sin(bt)) (*) remains true. There is a trajectory which is for very negative t roughly a line of slope -1 and large y intercept. This explains why for t << 0 the slope of the trajectory is about -1 . With this definition. -1 ] . it crosses the y axis at a point A . eigenvalues. Also. This explains why for t >> 0 the slope of the trajectory is about -1/2. x = 0 at A . and u(t) . i. are -1/2 and -1 . -1 ] respectively. The graph has a point of inflection where y = x' has a minimum (and so x" = 0) . x reaches a maximum at B and then falls. at C. (*) This conveniently expresses the solution to this ODE with arbitrary initial condition (at t = 0). and C? Answers: The solution is approximated by c e^{-t} for t << 0 . with eigenvectors [ 1 . so basic solutions e^{-t/2} [ 1 . the graph crosses the t axis .

Combining this with (**) we see: The first column of [1.b] .1]. The second column of [0. perhaps normal modes. If we write v_1 and v_2 for those two solutions. So the expression matrix valued function.0] . Call them u_1 and u_2. which we can write as [ u_1 where c_1 . u_2] is a "fundamental matrix" for written Phi(t) . as we have before). v_2 ] = then e^{At} [ a . The second column is also a solution : Phi(t) c_2 These two facts can be recorded using the matrix product e^{At} = Phi(t) [ c_1 c_2 ] The right hand matrix is there to get the initial conditions right. Evaluate at t = 0 : I = Phi(0) c c = Phi(0)^{-1} .b] . A square matrix has an inverse exactly This means that Reminder on inverse matrices: when .1] = [a. and at t = 0 we get a v_1(0) + b v_2(0) = a [1. b ] a v_1 + b v_2 This is a solution (because it's a linear combination of solutions). Let's check the claim (**) for any initial condition [a. Similarly. Note that at t = 0 we get the identity matrix.function. The general solution is a linear combination of u_1 and u_2. Begin by finding a linearly independent pair of solutions (basic solutions. e^{At} must denote a matrix. or rather a What could e^{At} be? What is its first column? Recall that the first column of any matrix B is the product B[1. A and is The matrix [u_1 . u_2 ] c_1 is a column vector.0] + b [0.0]. so that e^{At} = [ v_1 . e^{At} is the solution to u' = Au with u(0)= e^{At} is the solution to u' = Au with u(0) = This is the DEFINITION of e^{At} . [3] We need a method of computing the e^{At}.

-1/2 . 2 . [ e^{-t/2} . 1 . -1 .d] . and e^{-t/2} [ 1 . e^{-t} . e^{-t/2}+e^{-t} . Altogether: where Phi(t) e^{At} = Phi(t) Phi(0)^{-1} is any fundamental matrix for A . -e^{-t/2}+ 2e^t} ] .its determinant is nonzero. 2 . 1 .c. -e^{-t} ] [ 1 . c d ]^{-1} = A = [a. -(1/2)e^{-t/2} . We saw basic solutions Example: A = [ 0 . 2e^{-t/2}-2e^{-t} . -1/2 ] so Phi(t) Phi(0) = = e^{-t} [ 1 . -(1/2)e^{-t/2} . -3/2 ] . -e^{-t} ] [ 2 .b. -2 ] . -1 . -c a ] --. -1 ] . e^{At} ] = = Phi(t) Phi(0)^{-1} [ e^{-t/2} . and the off diagonal terms get their signs reversed. -1 ] -1/2 and so has determinant Phi(0)^{-1} = [ 2 .the diagonal terms get their positions reversed. -1/2 . (1/det A) [ d -b . -2 = [ 2e^{-t/2}-e^{-t} . In the 2x2 case [ a b . e^{-t} .

18.03 Class 36, May 10 Review of matrix exponential Inhomogeneous linear equations [1] Prelude on linear algebra: AB.

**If A and B are matrices such that the number of columns in A is
**

the same as the number of rows in B , then we can form the "product matrix" If the columns of B are

b_1 , ... , b_n then the columns of AB are

A b_1 , ... , A b_n [2] I know you don't want to hear more about Romeo and Juliet. This is about amorous armadillos, named Xena and Yan. x' y' = = - x + 3y

-3x - y

Matrix [ -1 , 3 ; -3 , -1 ] .

Characteristic poly: lambda^2 + 2 lambda + 10

Eigenvalues: -1 +- 3i

Stop! what do we want to know? We have a stable spiral, rotating

clockwise.

Do we want more? We already know that this romance will peter out into

dull acceptance. Maybe that's enough information about X and Y's love life

for us.

If not, we can go ahead and solve:

Eigenvector for -1 + 3i : [3 ; 3i] or just [1;i] .

Normal mode: e^{(-1+3i)t} [1;i]

Basic real solutions: e^{-t}[cos(3t) ; -sin(3t)]

e^{-t}[sin(3t) ; cos(3t)] Fundamental matrix (3t) ] Phi(t) = e^{-t} [ cos(3t) , sin(3t) ; -sin(3t) , cos Phi(t) [ a ; b ] . as a matrix-valued solution: (*)

Then the general solution is In fact you can think of Phi'(t) =

Phi(t)

A Phi(t) .

By the linear algebra prologue, if B is any 2 x 2 matrix then the columns of Phi(t) B are of this form and hence are solutions. The matrix exponential e^{At} is the fundamental matrix which is I when t = 0 . Its columns are the solutions which pass through [1;0] and

[0;1]

at t = 0 . It must be of the form Phi(t) B, and taking that Phi(0) B = I , or B = Phi(0)^{-1} and

e^{At} = Phi(t) Phi(0)^{-1}

t = 0

shows

For us, Phi(0) = I already! so we have found the matrix exponential. The solution u(t) with value [a;b] at t = 0 is e^{-t} [ cos(3t) , sin(3t) ; -sin(3t) , cos(3t) ] [ a ; b ] [3] The matrix exponential has various familiar properties, e^{As} e^{At} e^{A0} (e^{At})^{-1} = = = e^{A(s+t)} I e^{-At}

Evaluating it at t = other values gives other matrices with numbers (not functions) as entries; especially t = 1 , which gives a definition of e^A. Then the exponential law gives e^{nA} = (e^A)^n , n any integer.

Also, for those of you who love power series, e^{A} Warning: e^A e^B = I + A + (A^2)/2 + (A^3)/3! + .... e^{A+B}$.

is generally different from

[4] Now, a well known perfume manufacturer has taken an interest in Xeno and Yan. They produce a scent, Armamour, which exerts a certain influence on X and on Y : it increases X's rate of change of affection by a and Y's by b. Now we have: x' y' = = - x + 3 y + a -3 x y + b

This is an INHOMOGNENEOUS linear equation, which can be written u' where = A u + c c = [10;40] . then = u'_p = 0 , so

c = [a;b] . Specifically, u_p :

Expect a constant solution A u_p With = - c

or

u_p

- A^{-1} c

A = [-1,3;-3,-1] , u_p =

A^{-1} = (1/10)[-1,-3;3,-1] , so

- [-.1,-.3;.3,-.1] c

Specifically, then,

u_p

=

[.1,.3;-.3,.1] [10;40]

=

[13;1] .

This is the "equilibrium solution." The general solution is u_p + u_h where u_h is a solution of the homogeneous equation as above; it's a transient; all solutions spiral in to the single fixed point. [5] Now, in fact, even Armani's armamour fades with time -exponentially, of course. The truth is that q(t) = e^{-t}[10;40] . Abstractly, we may have a forcing term that varies with time: u' = Au + q(t)

We can solve this by "variation of parameters," as follows. Let Phi(t) be a fundamental matrix for A . Instead of looking for solutions Phi(t) c , which work if q(t) = 0 , let's look for solutions of the form u = Phi(t) v(t) , where v(t) is as yet unknown. (d/dt) Phi(t) v(t) = = Phi'(t) v(t) + Phi(t) v'(t)

A Phi(t) v(t) + q(t)

By (*), Phi'(t) = A Phi(t) , so those terms cancel and we find Phi(t) v'(t) v(t) u(t) = = = q(t) or so

integral Phi(t)^{-1} q(t) dt

Phi(t) integral Phi(t)^{-1} q(t) dt Phi(t)^{-1} . First, so

[6] In the Armani example, we have to work det Phi(t) =

e^{-2t}(cos^2(3t) + sin^2(3t)) = e^{-2t}

Phi(t)^{-t} = (3t) ] =

e^{2t} e^{-t} [ cos(3t) , -sin(3t) ; sin(3t) , cos e^t [ cos(3t) , -sin(3t) ; sin(3t) , cos(3t) ]

Phi(t)^{-1} q(t) = [ 10 cos(3t) - 40 sin(3t) ; 10 sin(3t) + 40 cos (3t) ] integral Phi(t)^{-1} q(t) dt = u(t) = = (10/3) [ -sin(3t) - 4 cos(3t) ; cos(3t) - 4 sin(3t)] + c Phi(t) integral Phi(t)^{-t} q(t) dt (10/3) e^{-t} [ -4 ; 1 ] + Phi(t) c

by an amazing cancellation which you should see for yourself.

The first term is the "particular solution," the second the homogeneous solution or transient. Despite Armani's best efforts, the relationship settles down to zip.

18.03 Class 37, May 12 Introduction to general nonlinear autonomous systems. [1] Recall that an ODE is "autonomous" if and not on t: x' = g(x) x' depends only on x

For example, I know an island in the St Lawrence River in upstate New York where there are a lot of deer. When there aren't many deer, they multiply with growth rate k ; x' = kx . Soon, though, they push up against the limitations of the island; the growth rate is a function of the population, and we might take it to be k(1-(x/a)) where a is the maximal sustainable population of deer on the island. So the equation is x' = k(1-(x/a))x , the "logistic equation." k = 3 and a = 3 , so x' = (3-x)x .

On this particular island,

**There are "critical points" at x = 0 and x = 3 . When 0 < x < 3 ,
**

x' > 0.

When x > 3 , x' < 0 , and, unrealistically, when x < 0 , x' < 0 too.

I drew some solutions, and then recalled the phase line:

------<-----*---->------*------<-------[2] One day, a wolf swims across from the neighboring island, pulls himself up the steep rocky shore, shakes the water off his fur, and sniffs the air. Two wolves, actually. Wolves eat deer, and this has a depressing effect on the growth rate of

deer.

Let's model it by

x' where y = (3-x-y) x

measures the population of wolves.

Now, wolves in isolation follow a logistic equation too, say y' = (1-y) y (no deer)

But the presence of deer increases their growth rate, say y' = (1-y+x) y

We have a nonlinear autonomous system x' = (3-x-y) x (*)

y) -. This happens when either x = 0 or is vertical along those two lines.y) g(x. the vector field is horizontal along those two lines.the system does not change over time.y)i^ + g(x. . in which f(x. both in direction and magnitude.y)j^ I showed a slide of the linear vector field F(u) = Au with A = [-1. Solutions are parametrized curves such that the velocity vector at the point v is given by F(v). It seemed to show some "equilibria. To find the equilibria. and how they thread their way through the vector field. [4] The fact is that normally a process spends most of its time near equilibrium. It is always in the act of returning to equilibrium after being jarred off it by something. that is (3-x-y) x = 0 .y' [3] = (1-y+x) y The general case would be x' y' = = f(x. It is horizontal where y' = 0 .-1] which modeled the effect of Armamour on Xena and Yan." points at which the vector field vanishes. but near to it. The vector field is both vertical and horizontal exactly when it vanishes.y) g(x.3. the homogeneous linear case. that is (1-y+x) y 3-x-y = 0 . This happens when either y = 0 or 1-y+x = 0 . Then I showed a slide of the deer/wolf vector field. and understanding how the system behaves near equilibrium is important.-3. Not exactly AT equilibrium. the vector field = 0 . first think about where the vector field is vertical: x' = 0 . We have been studying a special case of this. You can see some solutions.y) = = ax + by cx + dy Giving an autonomous equation is the same thing as giving a vector field: u'(t) = F(u) = f(x. So identifying equilibria is important.

which is at (1. so "to first order" x' y' = = 3x y The deer and wolf populations both expand by natural growth.There are four places where that happens: (0.y^2 Near the origin the quadratic terms in the vector field are insignificant. . we get exactly the phase line of the deer population without wolves.2) . These two phase lines sit inside the phase portrait of the deer/wolf system. x' y' = = f(x.0) . where y = 0 . (3. The eigenline for the smaller eigenvalue is the y axis. and along the y axis. where x = 0. This is a homogeneous linear system with matrix [ 3 0 . we get the phase line of the wolf population without deer.1) . 0 1 ] .0). so we have a node.0) .y) g(x. [5] We'll study the behavior near the critical point (0. and the place where both Notice that along the x axis.y) = = 3x . Expanding out.x^2 . The eigenvalues are 1 and 3 .xy y + xy . so this is the line to which all solutions (except for the other normal mode!) all solutions become tangent as the approach the origin. y = 3-x and y = 1+x . (0.

(It's surely zero when theta' = 0. Of course. The bob of a pendulum is attached to a rod.cL theta' . measured to the right.mg sin(theta) component of the force of gravity (and notice the sign!). so it can swing clear around the pivot.mg sin(theta) .03 Class 38. with straight down. May 15 Nonlinear systems: Jacobian matrices [1] The Nonlinear Pendulum.18. but controlling it if Write theta for need a dynamical variable. but for the moment let's suppose that we are restricting to small enough values of theta' so that the behavior is linear. This system is determined by three parameters: L m g length of pendulum mass of bob acceleration of gravity We will assume that the motion is restricted to a plane. To describe it we displacement. and also a frictional force which depends upon s' = L theta' Friction is very nonlinear. that angle. s = L theta s = 0 Newton's law says F = ms" = mL theta" The force includes the . We could use a horizontal it turns out to be easier to write down the equation you use the angle of displacement from straight down.) So: m L theta" = . Here is a force diagram: * |\ |theta | \ | \ mg | /\ (this is supposed to be a right angle!) | \/ | / | / mg sin(theta) | / |/ * Write s for arc length along the circle. in fact.

Let's compute the Jacobian: .k sin(theta) . g_x g_y ] (a. sin(x) = 0 .Divide through by theta" + where k = g/L mL and we get = 0 b theta' + k sin(theta) and b = c/m . near (0.y) Here ~ f(a.b) (y-b) f_x = partial f / partial x .0) by approximating it by a linear vector field.b) : f(x. and so the differential equation. This is a nonlinear second order equation.b) . say near an equilibrium at (a.b) v Near the critical point. Since (a. y = x' or y' = theta" = . We'll be particularly interested in approximating it near equilibria.b) .by This is an autonomous system.b) (x-a) + f_y(a. We can do that at any point. Let's study its phase portrait. the equation behaves v = [ bar-x . +-2pi .b theta' x' y' = = y .b) (y-b) g_x(a.b) (y-b) With bar-x = x-a and bar-y = y-b this system has matrix given by the "Jacobian matrix" J(x. the constant term vanishes and we have a homogeneous linear system x' y' = = f_x(a. [3] In the pendulum. .k sin(x) .b) (x-a) + f_y(a. J(a. +-pi . bar-y ] . Each coordinate in the vector field has a tangent line approximation near (a. It still has a "companion first order system.b) + f_x(a." obtained by setting x so = theta ...b) (x-a) + g_y(a. equilibria occur when also This occurs when x = 0 .y) evaluated at like v' where = = [ f_x f_y . [2] We studied the vector field for the deer/wolf population model.b) is a critical point.

+-4pi . I sketched these. Trajectories coming down from the left represent the pendulum swinging around in counterclockwise complete circles. It shows several features common to all companion systems: . J(x. Similarly. . -k -b ] When x = +-pi . -2 ] and the characteristic polynomial is lambda^2 + 2 lambda + 65 with roots -1 +.. Solutions are of the form x = A e^{-t} cos(8t . k = 65 . In the second case J(pi..46 . Let's take some particular numbers: In the first case J(0...0) = [ 0 . The period is 2pi/8 = pi/4 . 1 .lambda I has top row [ . k cos(x) = -1 and -b ] b = 2 . or about half.phi) and of course y = x' . With eigenvalue near 7.y) When = [ 0 .0) = [ 0 1 .the trajectories above the x axis perpendicularly.7 so the corresponding eigenvector is about [ 1 .sqrt{66} ~ 7.0) = [ 0 . +-2pi . -9 ] . those below move left. [4] Then I revealed the entire phase portrait. 65 . -b ] cos(x) = 1 and x = 0 . J(x. 2 ] and the characteristic polynomial is lambda^2 + 2 lambda .J(x. A . 1 . making a sharp V tilted somewhat to the right. moving counterclockwise. the eigenvector for eigenvalue -9 is about 1 ] [ 1 ..65 with roots . I sketched these at the appropriate critical points. 7 ] .. -9 We have a saddle.k cos(x) .0) = [ 0 1 . Trajectories coming up from . -65 . . +-3pi . .8i . the magnitude decreases by a factor of e^{-pi/4} ~ .1 +. The linear phase portrait is a spiral. The eigenlines are both pretty steep. 1 . In the course of one period. x axis move right..the trajectories cross the .

y . (0.0) .y) = [ 3 . and the amplitude of the swings decrease by about 50% with each swing.y) = = 3x . so we have a saddle. Let's do the critical point at J(1.2) To analyze these. with non-ray solutions becoming tangent to the y-axis as t ---> infty. they converge to it like e^{-9t} . J(3.y) g(x.0) = [ -3 -3 .2x . that is either x = 0 or y = 3 . In very exceptional cases.0) = [ 3 0 . expand: x' y' Thus = = f(x.2y ] We evaluate this at the critical points.xy y + xy . Eventually. J(0. at (0.y^2 J(x. For example.x and that the vector field is horizontal where y' = 0 . or equilibria. -7 ] respectively .1) . an unstable node. that is either y = 0 or y = x + 1 Consequently there are critical points.the right represent the pendulum swinging around in clockwise circles. 0 1 ] as we had before.2) -2 ] next: has characteristic polynomial lambda^2 + 3 lambda + 4 . The saddles represent the unstable equilibria which are straight up. but most likely miss and move away like e^{7t} . (1. 2 (1. The successive dips represent passing through the vertical position. The spiral has period approaching 8 . with eigenvectors [ 1 . the trajectory heads straight at the saddle equilibria. 0 4 ] For another example. (3. has eigenvalues -3 and 4 (since these are the diagonal entries in an upper triangular matrix . These snap nicely into place. With a student I animated the pendulum swinging around. We can do this with any of the critical points.2) = [ -1 -1 .x^2 .0) . the trajectory gets caught in a basin (actually it was always in that basin) and spiral in towards the attractor of that basin. 1 + x .y . -x . which is straight down. [5] Postscript on the deer/wolf population model x' y' = = (3-x-y) x (1-y+x) y (*) We found that the vector field is vertical where x' = 0 . 0 ] and [ 3 .

These joined up to give an overall picture of the behavior of this system. .and eigenvalues -(3/2) +." We have learned that the populations of deer and of wolves converge to these stable levels no matter what the initial populations are (as long as they are positive). rotating counterclockwise since the lower left entry is positive. or pseudoperiod 4 pi / sqrt(7) ~ 4. the phase portrait. they oscillate about them. We have a stable spiral. As the population levels approach this limiting value. I sketched this and snapped it into the nonlinear phase portrait. We can say very exactly how this happens: the solutions for the deer look like x-bar = e^{-(3/2)t} cos((sqrt(7)/2)t) with exponential decay constant -3/2 and pseudofrequency sqrt(7)/2 radians per unit time." with the equilibrium (1. We see that the entire upper right quadrant is a "basin.75 units of time.2) an "attractor.i sqrt(7)/2 .

May 17 Dangers of linearization. i. it is a degenerate system. 2x . They may be densely packed spirals.0) = = [ f_x . the linearization gives centers. But the actual trajectories of the nonlinear system may not be periodic. The actual phase plane near the equilibrium may be more like a spiral. if det = (tr/2)^2 > 0 . periodic trajectories. These spirals will rotate in the same direction as the predicted ellipse.e. x' y' = = 4xy x^2 .e. 4x . Similarly. has determinant zero. has at least one eigenvalue which is zero. ellipses. J(x.0). i. 0 0 ] = [ 4y . next steps in mathematics [1] The method we sketched on Monday works well "generically. but they may be either stable or unstable. The eigenvalues are both zero. Things get tricky if the trace and determinant of the Jacobian matrix lie on one of the dividing curves in the (tr. None of this is too bad. For example.y . It's as if the nonlinearity jostles the linear phase portrait off onto one of the regions bounded by the tr = 0 line.e. the linear phase plane is a star or a defective node. y' = 0 So the only critical point is at (0. almost all the time. g_y ] [ 0 0 .y^2 if x = +. unlike any linear portrait." i.18. If x = 0 for example then and a solution occurs as long as y' = -y^2 . every vector is an eigenvector.03 Class 39. or more like a node. Weird. -2y ] What is the phase portrait of this linear system? It just says u' = 0 so every point is a constant solution.y) so J(0. . limit cycles. This is when the Jacobian matrix is degenerate. g_x . I showed a Matlab plot of the phase portrait: it has six ray solutions.det) plane. chaos. On the other hand. but there is one case where the actual phase portrait may not resemble the linear one at all. These solutions behave very differently from ray solutions of linear systems. If tr = 0 and det > 0 . x' = 0 if either x = 0 or y = 0 . so there are other ray solutions along the vertical axis. f_y . We can still analyze this a bit.

This is a "limit cycle. so m = y'/x' = (1-m^2)/4m or m = +-1/sqrt(5)." The human heart (and I promise that this is the last time I'll mention it) is in fact controlled by an equation like this. [We can verify that there are ray solutions to this equation by finding the slope and substituting. the most important thing. This is why it returns to a normal periodic pattern after being disturbed.] [2] In studying a nonlinear autonomous system. If y = mx. they have a vertical asymptote. The companion system is x' y' = = y . This is two solutions. There are solutions along the lines through the origin with these slopes. For example. If J(a. racing down from infinity. is the possibility of periodic solutions. after equilibria. They approach the origin more slowly than any exponential decay (and the negative solutions begin their departure from the origin more slowly than any exponential growth). the "van der Pol" equation is x" + c (x^2-1) x' + x = 0 for fixed c .c (x^2-1) y This system turns out to continue to have periodic solutions.For example y' = -y^2 is separable: y = 1/(t-c). . When c = 0 this is just the harmonic oscillator. and all other nonzero solutions converge to it. with omega_n = 1 . it will look the same. all the nonzero solutions are periodic of period 2 pi. then x' = 4mx^2 and y' = (1-m^ 2)x^2.b) = 0 you have to assume that the phase portrait near the critical point (a.b) is nonzero. The lesson: If you find det J(a. When c > 0 the situation is in fact even better: there is ONLY ONE periodic trajectory. in order to keep the solution from wandering off the ray. This reflects the following behavior: Solutions for y < 0 start very slow but then speed up so quickly that they disappear to -infinity in finite time: graphed against time.x . The slope of the vector field must also be m . and then slow down as they near the origin.b) will look different from the linear phase portrait. depending upon whether t > c or t < c . Solutions for y > 0 appear at some time.

100 as prerequisite. Fourier series.440. See how many ways you can find it among the Ten Essential Skills: 1. 7. periodic solutions. <http://www. poly*exponential or sinusoidal signal. 0 < r < 30.html>. [5] Next steps in Mathematics 18.[3] It can be shown that limit cycles are typical for 2D systems. 4. Here's what happens for certain values of these parameters .112.04 (spring) Complex Variables with Applications (poles and such. They just wrap pretty crazily around the two nonzero unstable equilibria which exist provided that r > a/b . b .y xy . eigenvectors. Statistics) 18. 18. Second order LTI systems. confidence intervals) (Alternatively. r.443. But when you move to 3D things are much more complicated. Fourier analysis. Complex numbers and exponentials.and I showed the IDE tool "Lorenz Equations. 9. Delta functions. magical integral evaluations. who was modeling "convection rolls" in the upper atmosphere. The solutions don't ever settle down to a periodic orbit. and 18. 5.com/ide/idefiles/media/JavaTools/lnrzphsp. . linearization at equilibria. Euler's method. 2. a nonlinear autonomous system in 3 dimensions: x' y' z' = = = -ax + by -xz + rx . Nonlinear phase portraits. 18. [4] The unifying theme of the course has been the exponential function. transfer function and weight function. In 1963 he wrote down a fairly simple model. 3. amplitude gain and phase lag. The first such system was discovered right here at MIT by Edward Lorenz. First order models. 10. First order linear equations: Integrating factors or Variation of Parameter. 6. 8. but neither do they run off to infinity. Linear systems: eigenvalues. unit impulse response. conformal mappings) (Alternatively. convolution. Phase Plane.06 (fall and spring) Linear Algebra.aw-bc. Laplace transform. requiring 18. Probability.) 18. Variation of Parameters.05 (spring) Probability and Statistics (Random variables.bz for constants a . Linear phase portraits.

700. or 18..207: Nonlinear dynamics: Chaos. or 18.. 18.303: Linear PDEs.100C (spring): Analysis I.354J = 12.701) 18." 18..(Alternatively. 18. continuum systems. turbulence. . 18.100B (fall and spring).006 and 18. "pure" and "applied.152 or 18.353J = 12.. fluids.100A (fall and spring).

) We write a complex number z as z = x + iy (or x + yi). another solution of x2 + 1 = 0 is −i. Whereas the set of all real numbers is denoted by R. Im(−1 + 3i) = 3 Every real number x can be considered as a complex number x + i0. the set of all complex numbers is denoted by C. In other words. y ∈ R and the absolute value of z is |z| = (x2 + y 2 )1/2 and the conjugate of z is z = x − iy. We write Re z = x and Im z = y. Examples Re(−1 + 3i) = −1. ¯ Addition and multiplication in C are deﬁned by z1 + z2 = (x1 + iy1 ) + (x2 + iy2 ) = (x1 + x2 ) + i(y1 + y2 ) z1 z2 = (x1 + iy1 )(x2 + iy2 ) = x1 x2 + ix1 y2 + iy1 x2 + i2 y1 y2 = (x1 x2 − y1 y2 ) + i(x1 y2 + x2 y1 ) Clearly then z¯ = (x + iy)(x − iy) = x2 + y 2 = |z|2 z † This handout is an amended version of the original. where x and y are real numbers. the complex numbers z1 = x1 + iy1 and z2 = x2 + iy2 are equal if and only if x1 = x2 and y1 = y2 . Two complex numbers are said to be equal if they have the same real and imaginary parts. The number i is a solution of the equation x2 + 1 = 0 and has the property that i2 = −1. We call x the real part of z and y the imaginary part of z. magnitude of a complex number.MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey INTRODUCTION TO COMPLEX NUMBERS† Susanne C. Kaup Department of Mathematics and Computer Science Clarkson University Complex Arithmetic (Complex conjugation. Example 2 − 3i 6= 2 + 3i If z = x + iy ∈ C then x. J. division by complex numbers) Cartesian and Polar Forms Euler’s Formula De Moivre’s Formula Diﬀerentiation of Complex Functions One of the most important numbers in complex analysis is i. (Of course. a real number is just a complex number with vanishing imaginary part. Brenner and D. Complex Numbers Handout -1- Clarkson University . In other words.

The set R of real numbers is also a ﬁeld. (z1 + z2 ) + z3 = z1 + (z2 + z3 ). z · 1 = (x + iy) · (1 + i0) = x + iy = z (f) Each z has the unique additive inverse −z. z + (−z) = (x + iy) + (−x − iy) = 0 + i0 = 0 ¯ (g) Each nonzero z has the unique multiplicative inverse z −1 = 1/z = z /|z|2 . z ¯ zz −1 = z 2 = 1 |z| For example. the set C. ¯ ¯ z1 z2 = z1 z2 ¯¯ The operation of taking complex conjugates satisﬁes two basic algebraic rules: Note that if z = x + iy then z + z = x + iy + x − iy = 2x and z − z = x + iy − x + iy = 2iy. together with the two operations of addition and multiplication.MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey The Rules of Complex Arithmetic For z1 . z + 0 = (x + iy) + (0 + i0) = x + iy = z (e) Each z has the unique multiplicative identity 1. (a) Addition and multiplication are commutative. is a ﬁeld. z3 . z1 (z2 + z3 ) = z1 z2 + z1 z3 (d) Each z has the unique additive identity 0. Because addition and multiplication in C satisfy (a)-(g). z = x + iy ∈ C. (z1 z2 )z3 = z1 (z2 z3 ) (c) Multiplication distributes over addition. z2 . z1 z2 = z2 z1 (b) Addition and multiplication are associative. ¯ ¯ From these two equalities we have z+z ¯ z−z ¯ Re z = x = . Note that for z 6= 0. Im z = y = 2 2i Complex Numbers Handout -2Clarkson University . (g) says that z −1 = Example 1 z ¯ x − iy x −y = 2 = 2 = 2 +i 2 2 2 z |z| x +y x +y x + y2 Find the multiplicative inverse of 1 − 2i 1 1 1 + 2i 1 + 2i 1 2 = · = = +i 1 − 2i 1 − 2i 1 + 2i 1+4 5 5 z1 + z2 = z1 + z2 . (a) follows from the deﬁnition of addition in C and the commutative law for real numbers: z1 + z2 = (x1 + iy1 ) + (x2 + iy2 ) = (x1 + x2 ) + i(y1 + y2 ) = (x2 + x1 ) + i(y2 + y1 ) = (x2 + iy2 ) + (x1 + iy1 ) = z2 + z1 The other properties (b)-(g) may be similarly veriﬁed. z1 + z2 = z2 + z1 .

. with complex coeﬃcients c0 . We can therefore solve any polynomial equation completely by using complex numbers. . we deﬁne the modulus or absolute value of a complex number z = x+ iy √ as the nonnegative real number x2 + y 2 . . |z2 | |z|2 = x2 + y 2 = z z ¯ Why should we bother with complex numbers? Theorem (Fundamental Theorem of Algebra) Any polynomial equation cn z n + cn−1 z n−1 + · · · + c1 z1 + c0 = 0.MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey We can also ﬁnd the real and imaginary parts of z1 /z2 using complex conjugates. has n complex roots. Consider x2 + x + 1 = 0. ¯ ¯ ¯ z1 ¯ ¯ ¯ ¯ ¯ z2 = |z1 | (z2 6= 0). z1 1 z2 ¯ z1 z2 ¯ z1 z2 ¯ = z1 · = z1 · = = · 2 z2 z2 |z2 | z2 z2 ¯ z2 z2 ¯ x1 + iy1 x1 + iy1 x2 − iy2 (x1 + iy1 )(x2 − iy2 ) = = · = 2 x2 + iy2 x2 + iy2 x2 − iy2 x 2 + y2 2 Ã ! x 1 x 2 + y1 y2 x2 y1 − x1 y2 = +i 2 2 2 x2 + y2 x2 + y2 2 Example Simplify (3 − 2i)/(−1 + i) −5 − i 3 − 2i −1 − i −3 − 3i + 2i + 2i2 5 1 3 − 2i = = · = =− −i 2 −1 + i −1 + i −1 − i 1−i 2 2 2 Example Simplify (3i30 − i19 )/(2i − 1) 3i30 − i19 3(i2 )15 − (i2 )9 i 3(−1)15 − (−1)9 i −3 + i = = = 2i − 1 −1 + 2i −1 + 2i −1 + 2i = 5 + 5i −3 + i −1 − 2i 3 + 6i − i − 2i2 · = = =1+i 2 −1 + 2i −1 − 2i 1 − 4i 5 As mentioned above. c1 . cn and cn 6= 0. We write |z| = Example q x2 + y 2 | − 3 + 2i| = √ √ 9 + 4 = 13 The following modulus properties hold: |z1 z2 | = |z1 ||z2 |. The quadratic formula yields √ −1 ± 1 − 4 x= 2 Complex Numbers Handout -3Clarkson University . We can’t say the same thing for reals. If z2 6= 0. .

Im 6 x © © © r © © © © s z = x + iy = reiθ y - O © H © © H θ H H Re H s z = x − iy = re−iθ ¯ Polar Coordinates Let z = x + iy be a non-zero complex number. which is the (two dimensional) Cartesian plane. Examples‡ ¶ √ µ π π 1 + i = 2 cos + i sin . Hence. It is determined up to multiples of 2π. y) ∈ R2 . 1! = 1. Here. 5! = 5 · 4 · 3 · 2 · 1 = 120. 2! = 2 · 1 = 2. etc. n! (n factorial) is deﬁned by n! = n(n − 1)(n − 2) · · · 1 with 0! = 1. x = r cos θ. Therefore z = x + iy = (r cos θ + i sin θ) = r(cosθ + i sin θ) r(cos θ + i sin θ) is a way of expressing a complex number by using polar coordinates. y) has polar coordinates (r.” You can visualize z = x + iy as the point (x. 4! = 4 · 3 · 2 · 1 = 24. 4 4 µ ¶ √ 2π 2π + i sin −1 + i 3 = 2 cos 3 3 y = r sin θ Euler’s Formula We are all quite familiar with the Taylor series of the exponential function ea when the exponent a is real§ a2 a3 a4 a5 + + + + ··· ea = 1 + a + 2! 3! 4! 5! These relations are derived in an example on Page 6. The positive number r is just the modulus of z and the angle θ is called the argument of z.MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey which leads us to complex numbers. 3! = 3 · 2 · 1 = 6. § ‡ Complex Numbers Handout -4- Clarkson University . θ). The point (x. These numbers are also really not “imaginary. where r > 0.

In other words. Let θ be a real number and consider the exponential series when we replace a by θ. These are cos θ = 1 − θ2 θ4 θ6 + − + ··· 2! 4! 6! θ 3 θ5 θ 7 + − + ··· sin θ = θ − 3! 5! 7! Thus we have obtained the famous formula of Euler which is eiθ = cos θ + i sin θ A very important property of eiθ is that it has a modulus of unity: |e | = iθ q cos2 θ + sin2 θ = 1 for any real θ. We then have eiθ = 1 + iθ + (iθ)2 (iθ)3 (iθ)4 (iθ)5 + + + + ··· 2! 3! 4! 5! Now let’s reduce this series by using the relation i2 = −1 and collect the real terms together and the imaginary terms together. every complex number of the form eiθ lies on the unit circle x2 + y 2 = 1 in the xy-plane. Examples √ π π 1 3 e = cos + i sin = +i 6 6 2 2 π π 1 1 eiπ/4 = cos + i sin = √ + i √ 4 4 2 2 iπ/6 eiπ = −1 ein2π = 1 (n integer) ei(θ+n2π) = eiθ Complex Numbers Handout (θ ∈ R.MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey However. n integer) -5Clarkson University . we may also use the same Taylor series to deﬁne the exponential function of an imaginary exponent (and complex exponents as well). We obtain (after adding some more terms to make the ﬁnal result more obvious) e = 1 + iθ + i =1− iθ 2θ 2 2! +i 3θ 3 3! +i 4θ 4 4! +i 5θ 5 5! +i 6θ 6 6! +i 7θ 7 7! θ2 θ4 θ6 + − + ··· 4! 6! Ã 2! ! θ3 θ5 θ7 + − + ··· +i θ− 3! 5! 7! + ··· We now have that the real part of eiθ and the imaginary part of eiθ are each a familiar Taylor series.

This latter form will be called the polar form of the complex number z. So now the complex number z written in polar coordinates r(cos θ + i sin θ) can be written as reiθ .MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey The last result follows from the periodicity of cos θ and sin θ. From the rule for multiplication of complex numbers in polar form follows a rule for division. √ Example Convert z1 = 1 + i and z2 = −1 + i 3 into polar form and compute z1 z2 and z1 /z2 First compute the moduli r1 = |1 + i| = √ √ 1 + 1 = 2. it is very easy to compute their product. if n is an integer then cos(θ + n2π) = cos θ and sin(θ + n2π) = sin θ. Suppose that z1 = r1 eiθ1 = r1 (cos θ1 + i sin θ1 ) and z2 = r2 eiθ2 = r2 (cos θ2 + i sin θ2 ) are two non-zero complex numbers. to divide two nonzero complex numbers. divide their moduli and subtract their arguments. Note that if z = reiθ = r(cos θ + i sin θ). then z = r(cos θ − i sin θ) = r [cos(−θ) + i sin(−θ)] = re−iθ ¯ When two complex numbers are in polar form. We can easily see that 1 reiθ · e−iθ = ei(θ−θ) = e0 = 1 r Hence (reiθ )−1 = r−1 e−iθ We now turn our attention to dividing two complex numbers in polar form 1 r1 r1 eiθ1 = r1 eiθ1 · eiθ2 = ei(θ1 −θ2 ) iθ2 r2 e r2 r2 Therefore. √ √ r2 = | − 1 + i 3| = 1 + 3 = 2 Complex Numbers Handout -6- Clarkson University . We simply multiply their moduli and add their arguments.) So now we have a very easy formula for multiplying two complex numbers in polar form. We ﬁrst compute their product the hard way as follows: z1 z2 = (r1 r2 )eiθ1 eiθ2 = (r1 r2 )(cos θ1 + i sin θ1 )(cos θ2 + i sin θ2 ) = (r1 r2 ) [(cos θ1 cos θ2 − sin θ1 sin θ2 ) + i(cos θ1 sin θ2 + sin θ1 cos θ2 )] = (r1 r2 ) [cos(θ1 + θ2 ) + i sin(θ1 + θ2 )] = (r1 r2 )ei(θ1 +θ2 ) (We used trigonometric addition formulas to simplify in the second to last step.

since (a+b)3 = a3 +3a2 b+3ab2 +b3 .e. ±2. and the binomial expansion (a + b)n . ±2. ³ ´ ³ ´ sin 3θ = 3 cos2 θ sin θ − sin3 θ Clearly the real part gives cos 3θ in terms of cos θ and sin θ while the imaginary part gives the corresponding expression for sin 3θ. Therefore ¶ √ µ √ π π = 2eiπ/4 z1 = 1 + i = 2 cos + i sin 4 4 µ ¶ √ 2π 2π z2 = −1 + i 3 = 2 cos + i sin = 2ei2π/3 3 3 Using the rules for multiplication and division of complex numbers in polar form. or by plotting z1 and z2 in the complex plane. For example. Complex Numbers Handout -7Clarkson University . De Moivre’e Formula ³ ´n eiθ = einθ . we ﬁnd that θ1 is π/4 (plus any integer multiple of 2π) and θ2 is 2π/3 (plus any integer multiple of 2π). we have z1 z2 = ei11π/12 .MA 232 Hence 1+i= Diﬀerential Equations Fall 1999 Kevin Dempsey √ √ iθ 2e 1 = 2 (cos θ1 + i sin θ1 ) . z1 1 = √ e−i5π/12 z2 2 The following formula follows from the rules for multiplication and division of complex numbers in polar form. . . ±1. cos 3θ + i sin 3θ = (cos θ + i sin θ)3 = (cos θ)3 + 3(cos θ)2 (i sin θ) + 3(cos θ)(i sin θ)2 + (i sin θ)3 = cos3 θ + 3i cos2 θ sin θ − 3 cos θ sin2 θ − i sin3 θ = cos3 θ − 3 cos θ sin2 θ + i 3 cos2 θ sin θ − sin3 θ Equating real and imaginary parts gives cos 3θ = cos3 θ − 3 cos θ sin2 θ. ±1.√ . Ã ! √ −1 + i 3 = 2eiθ2 = 2 (cos θ2 + i sin θ2 ) √ ! 1 3 (cos θ2 . sin θ1 ) = 1 1 √ . . Thus if you know De Moivre’s formula. Then. let n = 3 in the above. i. . n = 0. . 2 2 Ã from which it is clear that (cos θ1 . you can always calculate any of the multiple angle formulas. Note that this gives us a very easy means for calculating the multiple angle formulas in trigonometry. 2 2 Either by looking at these points on the unit circle. sin θ2 ) = − . . (cos θ + i sin θ)n = cos nθ + i sin nθ. n = 0.

Equating moduli and arguments in the above equation gives π r3 = 2. Example Compute (1 + i)15 In polar form ¶ √ µ √ π π 2 cos + i sin = 2eiπ/4 4 4 and hence by the laws of exponents ¸ ³√ ´15 ³√ ´15 ∙ 15π 15π + i sin 2 ei15π/4 = 2 cos (1 + i)15 = 4 4 ³√ ´15 √ √ Now 2 = 215/2 = 27 2 = 128 2. We don’t know that the phase on the right-hand side is equal to the phase on the left-hand side.MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey Complex Exponential Functions One can also deﬁne a general complex exponential function. Also 15π/4 = 4π − π/4. gives z 3 = r3 ei3θ = 2i = 2eiπ/2 = 2ei(π/2+n2π) Note the n2π factor in the phase. start with z = reiθ . then. The argument of the exponential can be complex in general in which case ea+ib := ea eib = ea (cos b + i sin b) The complex exponential function has the following properties: ³ ´n ³ ´n ea+ib = ea eib = ean eibn = ean+ibn = e(a+ib)n ea1 +ib1 · ea2 +ib2 = ea1 eib1 · ea2 eib2 = (ea1 ea2 ) ei(b1 +b2 ) = ea1 +a2 ei(b1 +b2 ) = e(a1 +a2 )+i(b1 +b2 ) = e(a1 +ib1 )+(a2 +ib2 ) ea1 +ib1 = e(a1 +ib1 )−(a2 +ib2 ) a2 +ib2 e The details of the last equality are left to the reader. the phases must be the same. expressing 2i in polar form also. We do know that up to the factor of n2π. Thus the laws of exponents for real exponential functions also hold for complex exponential functions. 3θ = + n2π 2 Complex Numbers Handout -8Clarkson University . Therefore 1+i= (1 + i) 15 µ ¶¸ √ ∙ µ π¶ √ π 1 1 = 128 2 cos − + i sin − = 128 2 √ − i √ 4 4 2 2 Ã ! = 128 − 128i Example Solve the equation z 3 = 2i To solve this equation.

diﬀerentiating. The derivative of f with respect to t is given by f 0 (t) = u0 (t) + iv 0 (t) Example f (t) = t2 + i cos t. (Try it and see.26. complex exponential functions have the same kind of diﬀerentiation properties as real exponential functions.MA 232 so Diﬀerential Equations Fall 1999 Kevin Dempsey r = 21/3 . we will not obtain any more or diﬀerent roots. θ= π 2π π 4π +n = +n 6 3 6 6 3 1 +i 2 2 Ã √ ! √ − 3 1 3 = 2 +i 2 2 √ 3 = −i 2 Ã√ ! Plugging in n = 0. 1.) Diﬀerentiating a Complex Function A function f that depends on a real variable t can take on complex values. 2 we obtain the roots √ √ 3 3 z1 = 2eiπ/6 = 2 z2 = z3 = √ i5π/6 3 2e √ i3π/2 3 2e These roots are evenly spaced at 2π/3 = 120◦ intervals on a circle of radius 21/3 = 1. . 5. f 0 (t) = 2t − i sin t Let us see what happens if we diﬀerentiate the complex exponential function. We proceed by ﬁrst decomposing the complex function into real and imaginary parts. Note that if we continue and take n = 4. The individual steps are as follows: h e(a+ib)t = eat+ibt h i0 = eat (cos bt + i sin bt) ³ ³ ´0 = eat cos bt + ieat sin bt i0 = aeat cos bt + beat (− sin bt) + i aeat sin bt + beat cos bt = (a + ib)eat (cos bt + i sin bt) = (a + ib)e(a+ib)t ´0 ³ ´ Therefore. . and then ﬁnally reassembling the complex exponential. Such a complex function has the form f(t) = u(t) + iv(t) where u and v are real functions of t. . Complex Numbers Handout -9Clarkson University .

±i √ 4. (a) −32 − 32i. 4π. (c) 4ei2π/3 . dt dt it 4 2 Answers to Exercises on Page 11 1. (b) 2i. (a) 5. (a) ieit . θ = π/2. (c) 0 Complex Numbers Handout -10- Clarkson University . 11π/6} 9. (a) 1 ± 2i. (b) eit + tieit + 2ieit . (c) 1 + i. (b) −1/2 + i 3/2 8. (a) 3 + 2i. (b) {eiθ . (a) {eiθ . Compute L (eit ). 6π/5.MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey Let us now do an example directly related to ODE’s. (b) 2e−i5π/6 . (b) 5 + 17 3. Let’s take a complex function and see if it can be a solution of an ODE. 5. θ = 0. 2π/5. (a) f 0 (t) = cos t + i4t3 6. Example Let L = D4 + 2D2 + 1. (a) 2e−iπ/4 . (d) −1 + 4i √ √ 2. (d) 2e0 = 2 √ 5. 8π/5}. 7π/6. (b) ±1. (D4 + 2D2 + 1)(eit ) = D4 (eit ) + 2D2 (eit ) + (eit ) = (i4 + 2i2 + 1)(eit ) =0 dy dy So y(t) = e is a solution of constant-coeﬃcient ODE L(y) = 4 + 2 2 + y = 0.

Let L = D2 + D + 1. Solve completely (a) z 5 = 1 (b) z 3 = −i √ i13 + i 3) 7. Express the following in the form a + ib (a) i(2 − 3i) (b) (1 + i)2 (c) (3 + i)/(2 − i) (d) (−2 + 3i) + i(1 − i) 2.MA 232 Exercises Diﬀerential Equations Fall 1999 Kevin Dempsey 1. Compute (a) (1 − i)11 (b) h 1 (−1 2 6. Compute (a) L(eit ) it (b) L(te³ ) ´# " (c) L e − 1 +i 2 √ 3 2 t Complex Numbers Handout -11- Clarkson University . Find f 0 (t) if f (t) = sin t + it4 9. Show that (a) ¯ = z z (c) z is real if and only if z = z ¯ 8. Solve (a) z 2 − 2z + 5 = 0 (b) z 4 − 1 = 0 4. Express in the polar form reiθ (a) 1 − i √ (b) − 3 √ i − (c) (1 + i 3)2 (d) (1 + i)(1 − i) 5. Compute (a) |3 + 4i| (b) |i(2 + i)| + |1 + 4i| 3.

appear by themselves with exponent 1. Note that in (1) and (2) that the unknowns. J.MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey SOLVING LINEAR SYSTEMS K. The following systems are nonlinear: x2 + x2 = 1 1 x1 + x2 = 0 x1 x2 + x2 = −1 x1 − x2 = 0 2 (3) (4) We shall only concern ourselves with linear systems such as (1) and (2) that have the same number of equations as unknowns. Kaup Department of Mathematics and Computer Science Clarkson University 0 Linear Equations 2x1 + x2 = −1 x1 − x2 = 2 (1) Each equation of the linear system (1) represents a straight line. The system has: (a) no solution (the three planes are parallel or have no common intersection). or (c) an inﬁnite number of solutions (the two lines are coincident). Dempsey and D. 2x1 + 2x2 − 3x3 = 0 −x1 − x2 + x3 = 1 x1 + x2 − 2x3 = −1 (2) Each equation of the linear system (2) represents a plane. 1 2 × 2 Systems ax1 + bx2 = e cx1 + dx2 = f The general 2 × 2 (two equations and two unknowns) linear system is (5) Linear Systems Handout -1- Clarkson University . x2 and x3 . the coordinates x1 . The system has: (a) no solution (the two lines are parallel). M. or (c) an inﬁnite number of solutions (the three planes are coincident or intersect in a line). or (b) one unique solution (the two lines intersect). or (b) one unique solution (the three planes intersect at just one point).

then (7) and (11) become and 0x1 = de − bf 0x2 = af − ce (13) (14) If the right-hand side of either (13) or (14) is nonzero. The solution is unique and is given by de − bf x1 = ad − bc af − ce x2 = ad − bc If ad − bc = 0. Conclusion If ad − bc 6= 0. we would have the ridiculous equation zero=nonzero. i. -2- (15) Linear Systems Handout Clarkson University . to solve for x2 . then the system is in general not solvable. We ﬁrst note that equations (7) and (11) are always valid. we multiply (5a) by c and (5b) by a to obtain acx1 + bcx2 = ce acx1 + adx2 = af Subtracting (10a) from (10b) we get Under assumption (8). the 2 × 2 system ax1 + bx2 = e cx1 + dx2 = f is always solvable. If ad − bc = 0. we can solve for x2 : (ad − bc)x2 = af − ce x2 = (9) (10) (11) af − ce (12) ad − bc At this point you should realize that condition (8) is crucial to the success of solving system (5). Let us see what can happen if (8) does not hold.MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey To solve for x1 . which means that the system (5) is not solvable. ad − bc = 0.e. we multiply (5a) by d and (5b) by b to obtain adx1 + bdx2 = de bcx1 + bdx2 = bf Subtracting (6b) from (6a) we get Assuming that the solution for x1 is: (ad − bc)x1 = de − bf ad − bc 6= 0 x1 = (7) (8) (6) de − bf ad − bc Similarly.

we can also represent the solution by determinants. the important condition (8). for the solvability of ax1 + bx2 = e cx1 + dx2 = f is just ¯ ¯ ¯ ¯ ¯ ¯ a b ¯ ¯ ¯ = ad − bc c d ¯ (16) We can therefore state that a 2 × 2 system is always solvable if and only if the determinant of the coeﬃcient matrix is nonzero. Similarly. All we have to do is to recognize that ¯ ¯ ¯ ¯ ¯ a b ¯ ¯ ¯ 6= 0 c d ¯ (17) Then we can rewrite (15) as ¯ ¯ ¯ ¯ ¯ a e ¯ ¯ ¯ = af − ce c f ¯ de − bf = ad − bc ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ e b ¯ ¯ ¯ = ed − bf f d ¯ ¯ ¯ (18) x1 = e b ¯ ¯ ¯ f d ¯ a e c f a b ¯ ¯ ¯ c d ¯ a b c d ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ x2 = af − ce = ad − bc You may wonder how we Ã remember such formulas. ! e b a b can be obtained from the coeﬃcient matrix by Observe that the matrix f d c d Ã ! e replacing its ﬁrst column with the right-hand side of the system.MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey 2 Cramer’s Rule Ã We all know that the determinant of a 2 × 2 matrix ¯ ¯ ¯ ¯ ¯ ¯ a b c d ! is given by the formula Hence. Not only can we describe the solvability condition in terms of a determinant. the matrix f Linear Systems Handout -3Clarkson University ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ (19) . namely ad − bc 6= 0. It turns out that there is Ã easy ! can an way.

.MA 232 Ã ! Diﬀerential Equations Fall 1999 Kevin Dempsey a e can be obtained by replacing the second column of the coeﬃcient matrix with the c f right-hand side. is Ã 2 1 1 −1 !Ã x1 x2 ! = Ã −1 2 ! x1 = ¯ ¯ 2 1 ¯ ¯ ¯ 1 −1 x2 = ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ −1 1 2 −1 2 −1 ¯ ¯ ¯ 1 2 ¯ 2 1 1 −1 ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ = −1 1 = −3 3 = 5 5 =− −3 3 Now that we have a good understanding of 2 × 2 systems. n × n systems?” The answer is that Cramer’s rule works in the general case. Linear Systems Handout -4- Clarkson University . in matrix form. 4 × 4. 2. the natural question to ask is: “What about 3 × 3. we can rewrite the conclusion at the end of Section 1 in the following way. Cramer’s Rule for 2 × 2 Systems A 2 × 2 linear system is always solvable if and only if the determinant of the coeﬃcient matrix is nonzero. Therefore we can remember (19) by determinant of the matrix obtained by replacing the ﬁrst column of the coeﬁcient matrix with the right-hand side x1 = determinant of the coeﬃcient matrix determinant of the matrix obtained by replacing the second column of the coeﬁcient matrix with the right-hand side x2 = determinant of the coeﬃcient matrix (20) In summary. in which case the solution is unique and is given by determinant of the matrix obtained by replacing the kth column of the coeﬃcient matrix with the right-hand side xk = determinant of the coeﬃcient matrix for k = 1. . . . Example Solve the system 2x1 + x2 = −1 x1 − x2 = 2 Solution which.

The derivation of the general Cramer’s rule is nontrivial. . in which case the solution is unique and is given by determinant of the matrix obtained by replacing the kth column of the coeﬃcient matrix with the right-hand side xk = determinant of the coeﬃcient matrix for k = 1. (II) Multiply the terms in Step I alternately by + and −. In order to use the general Cramer’s rule we must know how to compute the determinant of n × n matrices.MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey Cramer’s Rule for n × n Systems An n × n linear system is always solvable if and only if the determinant of the coeﬃcient matrix is nonzero. Here we ask you to accept it on the basis of the 2 × 2 case. . . (III) Sum the terms obtained in Step III. Example Compute ¯ ¯ ¯ ¯ ¯ ¯ ¯ = ¯ ¯ ¯ a11 ¯ ¯ a22 a23 a32 a33 ¯ ¯ ¯ ¯ ¯ ¯ ¯ − a12 ¯ ¯ ¯ a21 a23 a31 a33 ¯ ¯ ¯ ¯ ¯ ¯ ¯ + a13 ¯ ¯ ¯ a21 a22 a31 a32 ¯ ¯ ¯ ¯ ¯ (21) ¯ ¯ ¯ ¯ ¯ ¯ ¯ 1 2 3 ¯ ¯ 4 5 6 ¯ ¯ ¯ 7 8 9 ¯ ¯ Linear Systems Handout -5- Clarkson University . You will see it in MA232/339. n. 3 Determinants of n × n Matrices The determinant of the 3 × 3 matrix a11 a12 a13 a21 a22 a23 a31 a32 a33 can be computed by the formula ¯ ¯ ¯ ¯ ¯ ¯ ¯ a11 a12 a13 a21 a22 a23 a31 a32 a33 Here is an easy way to remember (21): (I) Multiply each component (from left to right) on the ﬁrst row by the determinant of the matrix obtained by removing the row and column containing that component. This is dealt with in the next section. 2. . beginning with +.

In the every day world.MA 232 Solution Diﬀerential Equations Fall 1999 Kevin Dempsey ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ 1 2 3 ¯ ¯ ¯ 5 6 ¯ ¯ 4 6 ¯ ¯ 4 5 ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ 4 5 6 ¯ = 1¯ ¯ − 2¯ ¯ +3¯ ¯ ¯ ¯ 8 9 ¯ ¯ 7 9 ¯ ¯ 7 8 ¯ 7 8 9 ¯ = 1(5 · 9 − 8 · 6) − 2(4 · 9 − 7 · 6) + 3(4 · 8 − 7 · 5) = 1(−3) − 2(−6) + 3(−3) =0 ¯ What about the determinants of 4 × 4 matrices? We use the same three steps since we now know how to compute the determinants of 3 × 3 matrices. Linear Systems Handout -6- Clarkson University . Gaussian elimination is introduced here using matrices. We must point out that there are many properties of determinants that we have not discussed. For this reason. You will learn more about determinants in MA312/339. we can compute 5 × 5 determinants using Steps I-III above. and so on. But it is not an eﬃcient way to compute the value of the solution when n ≥ 3. 4 Gaussian Elimination Cramer’s rule is an important theoretical result because it enables us to represent the solution of an n × n linear system in terms of determinants. there are more eﬃcient ways of computing n × n determinants. The method of Gaussian elimination is more eﬃcient in these cases. Once we know how to handle 4 × 4 determinants. linear systems of equations are respesented in matrix form and are solved using matrix algebra. In particular. Example Compute ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ Solution ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ 1 −1 2 3 ¯ ¯ 0 1 3 2 ¯ ¯ ¯ −1 1 0 1 ¯ ¯ 0 2 1 0 ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ 1 −1 2 3 ¯ ¯ ¯ 1 ¯ 0 3 2 ¯ ¯ 0 1 2 ¯ ¯ 0 1 3 ¯ 3 2 ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ 0 1 3 2 ¯ ¯ 0 1 ¯ − (−1) ¯ −1 0 1 ¯ + 2 ¯ −1 1 1 ¯ − 3 ¯ −1 1 0 ¯ ¯ = 1¯ 1 ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ −1 1 0 1 ¯ ¯ ¯ 2 10 ¯ ¯ 0 1 0 ¯ ¯ 0 2 0 ¯ ¯ 0 2 1 ¯ ¯ 0 2 1 0 = 1(7) − (−1)(−2) + 2(−4) − 3(−5) = 12 There is nothing that can stop us now.

Deﬁnition A matrix is said to be in reduced row echelon form if: Linear Systems Handout -7- Clarkson University . the system will have a unique solution. Deﬁnition A matrix is said to be in row echelon form if (a) The ﬁrst nonzero entry in each row is 1. If the system is consistent and the nonzero rows of the the row echelon matrix form a triangular system. Deﬁnition In matrix algebra. Deﬁnition The process of using elementary row operations I. x1 and x2 .MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey The augmented matrix associated with the 2 × 2 linear system ax1 + bx2 = e cx1 + dx2 = f is Ã ! a b e c d f (22) Notice that the xi ’s. (b) If row k does not consist entirely of zeros. they are below the rows having nonzero entries. An invaluable aid to the latter is a check column each entry of which is simply the sum of all the terms of the corresponding row of the augmented matrix: Ã a b e c d f ! a+b+e c+d+f (23) The check column is shown here to the right of the augmented matrix. Otherwise. (III) Adding a multiple of one row to another. Note that elementary row operation II is necessary in order to scale the rows so that the leading coeﬃcients are all 1. and (c) If there are rows whose entries are all zero. in this case. Its usefulness will become clear as we proceed. II. and III to transform a linear system into one whose augmented matrix is in row echelon form is called Gaussian elimination. the number of leading zero entries in row k + 1 is greater than the number of leading zero entries in row k. are placeholders that can be omitted until the end of the elimination process. If the row echelon matrix contains a row of the form (00 · · · 0|1) the system is inconsistent. the system will be consistent. (II) Multiplying a row by a nonzero real number. an elementary row operation entails: (I) Interchanging two rows.

Gauss-Jordan reduction of a consistent 2 × 2 linear system proceeds as follows: Ã Ã Ã Ã ! ! ! x x x x x x x x x 0 x x x 0 x 0 x x 1 0 x1 0 1 x2 x x x x x x x x → → → ! Gauss-Jordan reduction of a consistent 3 × 3 linear system looks like: x x x x x → 0 x x x x 0 x x x x x x x x x → 0 x x x x 0 0 x x x x x 0 x x → 0 x 0 x x x 0 0 x x x 0 0 x x → 0 x 0 x x x 0 0 x x x x x x x x x x x x x x x x x The outside column is the check column.MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey (a) The matrix is in row echelon form. and (b) The ﬁrst nonzero entry in each row is the only nonzero entry in its column. after Gauss-Jordan reduction (I|x). Thus. Deﬁnition The process of using elementary row operations to transform a matrix into reduced row echelon form is called Gauss-Jordan reduction. for a consistent linear system Ax = b. the augmented matrix starts out as (A|b) and becomes. the solution vector is the last column of the reduced row echelon matrix. Linear Systems Handout -8Clarkson University 1 0 0 x1 x → 0 1 0 x2 x 0 0 1 x3 x . GaussJordan reduction transforms the coeﬃcient matrix into an identity matrix. For a consistent linear system. where I is the identity matrix. In other words.

x2 = − 3 3 Note that at each stage of the reduction.→ 3 0 1 0 3 −5 ! ¾ 4 −2 . In this case. we will use straightforward elimination which in this case proceeds as follows.→ R2 R1 1 −1 2 2 1 2 Ã 2 2 ! . for 2 × 2 linear systems.MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey 2 × 2 Example of Gauss-Jordan Reduction∗ 2x1 + x2 = −1 x1 − x2 = 2 Ã ! (24) 2 1 −1 1 −1 2 Ã 2 2 ! . ∗ Linear Systems Handout -9- Clarkson University .→ R1 −2R1 + R2 3R1 + R2 R2 R1 /3 R2 /3 ½ Ã Ã 1 −1 2 0 3 −5 ! 2 −2 . µ ¶ √ 5 1 + − = −1 2 3 3 µ ¶ µ ¶ √ 5 1 − − =2 3 3 µ ¶ Often. We observe that adding the two rows eliminates x2 and gives the equation 3x1 = 1 from which x1 = 1/3.→ The solution set for (24) is 1 0 1/3 0 1 −5/3 4/3 −2/3 1 5 x1 = . R1 and R2 refer to the preceding matrix. elsewhere in this class. Then from the second equation (either equation suﬃces) x2 = x1 − 2 = −5/3. Note also that you should always check the solution set by substituting it back into the original system of equations.

→ .→ R1 + 3R3 1 2 0 0 3 R2 + 2R3 0 1 0 1 2 R3 0 0 1 0 1 R1 − 2R2 1 0 0 −2 −1 R2 1 2 0 1 0 0 0 1 R3 0 1 {x1 = −2. Linear Systems Handout -10- Clarkson University . the solution set for (25) is . x2 = 1.→ R1 1 2 −3 0 0 R1 + R2 0 1 −2 1 0 R2 + R3 0 0 −1 0 −1 1 2 −3 0 0 R1 R2 0 1 −2 1 0 −R3 0 0 1 0 1 .→ Hence.MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey 3 × 3 Example of Gauss-Jordan Reduction x1 + 2x2 − 3x3 = 0 −x1 − x2 + x3 = 1 x1 + x2 − 2x3 = −1 (25) 1 2 −3 0 0 1 1 0 −1 −1 1 1 −2 −1 −1 . x3 = 0} .

→ 0 2 0 0 0 0 2 0 . x4 = −2} .→ R1 R2 R3 3R3 − 2R4 R1 − 2R4 R2 − R4 R3 − R4 R4 3 1 0 0 3 1 0 0 3 2 0 0 0 1 2 0 0 0 2 0 0 1 2 0 2 −1 5 4 1 1 1 0 3 1 1 5 .→ 1 0 0 0 1 0 0 0 2 0 0 0 0 0 1 0 .→ 1 0 0 0 0 1 0 0 0 −3 −2 0 2 3 0 1 2 −1 1 −2 0 −6 −4 0 4 6 0 2 4 −1 1 −2 The solution set for (26) is {x1 = −3. x2 = 2.→ R1 2R2 − R3 R3 R4 2R1 − 3R2 R2 R3 R4 R1 /2 R2 /2 R3 /2 R4 0 3 0 3 0 2 1 −2 0 3 0 4 0 2 1 −2 2 −1 5 1 1 4 1 0 3 1 −2 −1 7 5 4 −1 7 6 4 −1 .→ R1 R2 2R2 − R3 −R1 + R4 1 0 0 0 1 0 0 0 2 −1 5 1 1 4 2 5 1 3 10 0 3 1 0 0 0 1 2 3 . Linear Systems Handout -11Clarkson University . x3 = 1.MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey 4 × 4 Example of Gauss-Jordan Reduction x2 + x3 + x4 x1 + 3x2 + 2x4 x1 + 3x2 + 3x3 + 3x4 2x2 + x4 0 1 1 0 1 3 3 2 = = = = 1 −1 0 2 (26) 1 0 3 0 1 0 0 1 3 1 2 3 .→ R2 R1 R4 R3 1 1 4 5 2 −1 3 0 10 1 2 5 0 1 0 3 .

x3 = 1} 6. x1 + 3x2 = 0 2x1 + 4x2 = 6 Ans: {x1 = 9. x2 = 3. x2 = 0. x1 + 3x2 = 0 2x1 + 4x2 = 6 Ans: {x1 = 9. x3 = 1} 8. x3 = −1. x2 = −3} Ans: {x1 = 2. x3 = −3} −2 3 −1 −1 ½ ¾ 10. 2x1 + x2 + x3 = 8 4x1 + x2 = 11 −2x1 + 2x2 + x3 = 3 Solve the following systems using Gauss-Jordan reduction 7. Ans: 1 x1 = . x4 = 1 2 Linear Systems Handout -12- Clarkson University .MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey 5 ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ Exercises ¯ ¯ Compute the following determinants 1. x2 = −3} Ans: {x1 = 2. 1 −1 ¯ ¯ ¯ 3 2 ¯ Ans: 5 2. x2 = 3. 3. x2 = 0. 1 2 1 0 3 5 2 1 −1 1 −1 1 ¯ ¯ 2 0 1 ¯ ¯ ¯ −1 3 0 ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ Ans: 4 Ans: 6 ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ 4. 1 −1 0 1 0 2 −1 1 1 3 2 0 1 0 1 −1 Ans: −11 Solve the following systems using Cramer’s rule 5. Ans: {x1 = 3. 2x1 + x2 + x3 = 8 4x1 + x2 = 11 −2x1 + 2x2 + x3 = 3 2x1 − x2 = 6 −x1 + 2x2 − x3 = 0 −x2 + 2x3 = −6 3x2 + 3x3 + x4 2x1 + 4x2 + 2x4 2x1 + 7x2 + 9x3 + 7x4 6x3 + 5x4 = = = = 9.

then we found the second order LTI equation m¨ + bx + kx = ky . this equation is the real part of mz + bz + kz = biωBeiωt . Suppose that the position of the bottom of the dashpot is given by y(t). Spring. now arranged so that x = 0 when the spring is relaxed. x ˙ Since Re (ieiωt ) = − sin(ωt). and x(t) denotes the position of the mass. Amplitude and Phase: Second Order. x ˙ ˙ Suppose now that the motion is sinusoidal with circular frequency ω: y = B cos(ωt) . This can be rewritten m¨ + bx + kx = by . This time I’ll keep m around.03: Diﬀerential Equations. Here’s a frequency response analysis of this problem. and again the mass is at x(t). Then the force on the mass is given by m¨ = −kx + b x d (y − x) dt since the force exerted by a dashpot is supposed to be proportional to the speed of the piston moving through it. ¨ ˙ The Exponential Response Formula gives zp = where p(s) = ms2 + bs + k is the characteristic polynomial. arranged so that x = y when the spring is at rest. A new Mathlet. II. x ˙ Now suppose instead that we ﬁx the top of the spring and drive the system by moving the bottom of the dashpot instead. If y(t) denotes the displacement of the plunger at the top of the spring.18. 2006 Driving through the dashpot The Mathlet Amplitude and Phase: Second order considers a spring/mass/dashpot system driven through the spring. instead of setting it equal to 1 or dividing through by it. illustrates this system with m = 1. Then y = −ωB sin(ωt) so our equation is ˙ m¨ + bx + kx = −bωB sin(ωt) . Thus the “transfer function” is W (s) = bs p(s) biω Beiωt p(iω) .

divide numerator and denominator in the expression for W (iω) by biω: 2 i ωn − ω 2 W (iω) = 1 − b/m ω � �−1 . This is what we get with gain = |W (iω)| and −φ = Arg (W (iω)) . 1/z follows a circle of radius 1/2 and center 1/2. so I want to express the sinusoidal solution as xp = gain · B cos(ωt − φ) . through 0 when ω = ωn . 2 2 p(iω) = m(iω)2 + biω + mωn = m(ωn − ω 2 ) + biω .and the complex gain is W (iω) = so that zp = W (iω)Beiωt . Thus both the gain and the phase are displayed by the curve parametrized by the complex valued function W (iω). 2 As ω goes from 0 to ∞. moving upwards.” It shows that the gain starts small. To understand this curve. to near −π/2 when ω is large. For large ω. so the expression inside the brackets follows the vertical straight line in the complex plane with real part 1. where the resonant peak is not exactly at ωn and can be either very large or non existent depending on the strength of the damping). 2 m(ωn − ω 2 ) + biω We should certainly regard the “physical input signal” as B cos(ωt). which in turn corresponds to an abrupt phase transition from −φ near π/2 to −φ near −π/2. Thus a narrow resonant peak corresponds to a rapid sweep across the far edge of the circle. W (iω) is approximately −ib/mω. The Nyquist plot also shows that −φ = Arg (W (iω)) moves from near π/2 when ω is small. grows to a maximum value of 1 exactly when ω = ωn (in contrast to the springdriven situation. . And it shows that these two eﬀects are linked to each other. (ωn − ω 2 )/ω goes from +∞ to −∞. traversed clockwise (exercise!). As z follows this line. and then falls back to zero. p(iω) k/m . This circle is the “Nyquist plot. so the gain falls oﬀ like (b/m)ω −1 . Using the natural frequency ωn = � biω . It crosses the real axis when ω = ωn . so W (iω) = biω .

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

362 8:30—9:30 p.m.MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey Test I Wednesday. 360. September 29. 1999 SC 162. Name Student # Signature If your last name begins with letter You are in A-E SC 162 F-L SC 362 M-Z SC 360 Please do not detach pages Loose pages will NOT be graded Put a check mark next to your recitation section Section 11 12 13 14 15 16 17 18 19 √ 1 2 3 4 5 Total √ Day Period Time Room Leader Monday 2 9—9:50 SC 386 Scott Gregowske Monday 4 11—11:50 SC 386 Sovia Lau Monday 4 11—11:50 SC 342 Chiu Wah Cheng Monday 7 3—3:50 Rowley 150 Chaya Tuok Monday 8 4—4:50 SC 303 Phillip Allen Tuesday 1 8—8:50 SC 356 Elizabeth Baker Tuesday 5 1—1:50 SC 344 Tom Pearsall Tuesday 7 3—3:50 SC 344 Wesley Kent Tuesday 8 4—4:50 SC 305 Ed Landry Test I Clarkson University .

y2 ](x) of two functions y1 (x) and y2 (x) is identically zero on an F interval (a. b) then the two functions are linearly dependent on that interval (viii) If the two functions y1 (x) and y2 (x) are solutions of the second-order variable-coeﬃcient ordinary diﬀerential equation y 00 +p(x)y 0 +q(x)y = 0 on an interval (a.MA 232 1 (20 pts) Diﬀerential Equations Fall 1999 Kevin Dempsey Answer T (True) of F (False) to the following ten questions (i) The diﬀerential equation dy + x100 y = 0 is linear dx 2 T dy (ii) The diﬀerential equation + 4y = sin x is nonlinear dx 2 2 2 F (iii) ∂u ∂u − = 0 is a partial diﬀerential equation ∂t 2 ∂x 2 dy + xy = 2 and y(1) = 3. b) (ix) y(t) = t and y(t) = 2t + 1 are both solutions of dy y+1 = dt t+1 T (x) y 00 − 4y 0 + 13y − 9 = 0 is a homogeneous second-order diﬀerential equation F Test I Clarkson University . y2 } F is a fundamental solution set on (a. then y 0 (1) = 0 dx T (iv) If F (v) The diﬀerential equation dy 2 = ey sin2 y is autonomous dt T (vi) The line p = 1 dp is an isocline of the diﬀerential equation = p(1 − p) 2 dt T (vii) If the Wronskian W [y1 . b). then the set {y1 .

MA 232 2 (20 pts) Diﬀerential Equations Fall 1999 Kevin Dempsey (a)(10 pts) Solve the separable diﬀerentiable equation y dy = dt 1 + y2 Ans 1 + y2 dy = dt y ! Z Ã Z 1 + y dy = dt y y2 = t+C ln |y| + 2 (b)(10 pts) Evaluate the determinant ¯ ¯ ¯ ¯ ¯ ¯ ¯ 1 2 −3 ¯ ¯ 4 −1 2 ¯ ¯ ¯ 0 3 1 ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ −3¯ ¯ ¯ ¯ Ans = = 1(−1 − 6) − 2(4 − 0) − 3(12 − 0) = −7 − 8 − 36 = −51 ¯ ¯ 1¯ ¯ ¯ ¯ 4 2 −1 2 ¯ ¯ ¯ ¯ −2¯ ¯ 0 1 3 1 ¯ 4 −1 ¯ ¯ ¯ 0 3 ¯ ¯ Test I Clarkson University .

MA 232 3 (20 pts) Diﬀerential Equations Fall 1999 Kevin Dempsey Solve the initial value problem 3y dy + + 2 = 3x dx x y(1) = 1 Ans dy 3 + y = 3x − 2 dx x µ(x) = e = ( R (3/x)dx = e3 ln |x| = eln |x| = |x|3 x≥0 x<0 ¾ 3 Take µ(x) = x3 (since the sign cancels) y(x) = = = = y(1) = y(x) = 1 x3 (3x − 2) dx + C x3 ½Z ¾ 1 4 3 (3x − 2x ) dx + C x3 µ ¶ 1 3 5 1 4 x − x +C x3 5 2 3 2 x C x − + 3 5 2 x 3 1 C 9 − + → C= 5 2 1 10 9 3 2 x x − + x−3 5 2 10 ½Z x3 . −x3 . Test I Clarkson University .

→ . Ans 1 −1 1 −2 −1 3 −1 3 4 −1 1 −1 1 5 6 .→ . Be sure to state what your solution is.→ The solution set is ½ Test I Clarkson University . 3 1 x2 = .MA 232 4 (20 pts) Diﬀerential Equations Fall 1999 Kevin Dempsey Solve the linear system −x1 + 3x2 − x3 = 3 −x1 + x2 + 5x3 = 1 x1 − x2 + x3 = −2 using Gauss-Jordan reduction (it is imperative that you include the check column and refrain from working with fractions until the last step).→ 1 −1 1 −2 −1 R1 2 0 R1 + R2 0 1 3 R1 + R3 0 0 6 −1 5 6R1 − R3 6 −6 0 −11 −11 R2 0 2 0 1 3 R3 0 0 6 −1 5 6 0 0 −8 −2 R1 + 3R2 R2 0 2 0 1 3 0 0 6 −1 R3 5 R1 /6 1 0 0 −4/3 −1/3 R2 /2 0 1 0 1/2 3/2 R3 /6 0 0 1 −1/6 5/6 4 x1 = − . 2 x3 = − 1 6 ¾ .

on what interval(s)? Ans Let Ly = x2 y 00 − 4xy 0 + 6y L(x) = x2 (0) − 4x(1) + 6(x) = 2x 6= 0 √ L(x2 ) = x2 (2) − 4x(2x) + 6(x2 ) = 0 √ L(x3 ) = x2 (6x) − 4x(3x2 ) + 6(x3 ) = 0 L(x4 ) = x2 (12x2 ) − 4x(4x3 ) + 6(x4 ) = 2x4 6= 0 y1 (x) = x2 and y2 (x) = x3 are solutions of Ly = 0 The Wronskian W [y1 . x4 } contain a fundamental solution set for the diﬀerential equation x2 y 00 − 4xy 0 + 6y = 0 If it does.MA 232 5 (20 pts) Diﬀerential Equations Fall 1999 Kevin Dempsey Does the set {x. Test I Clarkson University . +∞). {x2 . x2 . 0) and (0. 0) and (0. y2 ](x) = 2x 3x2 ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ = 3x4 − 2x4 = x4 Since x4 6= 0 on (−∞. x3 . y2 ](x) of these two functions is x2 x3 W [y1 . +∞). x3 } is a fundamental solution set for the given ODE on the intervals (−∞.

Name Student # Signature If your last name begins with letter You are in A-E SC 162 F-L SC 362 M-Z SC 360 Please do not detach pages Loose pages will NOT be graded Put a check mark next to your recitation section Section 11 12 13 14 15 16 17 18 19 √ 1 2 3 4 5 Total √ Day Period Time Room Leader Monday 2 9—9:50 SC 386 Scott Gregowske Monday 4 11—11:50 SC 386 Sovia Lau Monday 4 11—11:50 SC 342 Chiu Wah Cheng Monday 7 3—3:50 Rowley 150 Chaya Tuok Monday 8 4—4:50 SC 303 Phillip Allen Tuesday 1 8—8:50 SC 356 Elizabeth Baker Tuesday 5 1—1:50 SC 344 Tom Pearsall Tuesday 7 3—3:50 SC 344 Wesley Kent Tuesday 8 4—4:50 SC 305 Ed Landry Test II Clarkson University .m. 1999 SC 162. October 27.MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey Test II Wednesday. 362 8:30—9:30 p. 360.

2. since |eiθ | = 1 for any angle θ Test II Clarkson University . . Ans Yes (−i)5 = (−i)2 (−i)2 (−i) = (−1)(−1)(−i) = −i −i 1 0 0 0 0 i 2 2 0 −i i i −i −i 1 −i −1 i 1 0 or (−i)5 = (ei3π/2 )5 = ei15π/2 = ei[6(2π)+3π/2] = ei3π/2 or z 5 + i = (z + i)(z 4 − iz 3 − z 2 + iz + 1) (d)(8 pts) Find the ﬁfth roots of −i (i. 10 10 10 10 10 3π 7π 11π 3π 19π . r=1 5θ = 3π + n2π 2 3π 2π θ= +n 10 5 3π 7π 11π 15π 19π = .e. . Completely solve z 5 + i = 0) Ans z5 + i = 0 z 5 = r5 ei5θ = −i = ei(3π/2+n2π) . . . = 10 10 10 2 10 (e)(3 pts) Do the roots in (d) all lie on the unit circle? Ans Yes.e. Is (−i)5 + i = 0?) Justify your answer. .) Ans | − i| = 1 so yes. −i does lie on the unit circle (b)(3 pts) What does the Fundamental Theorem of Algebra say about the complex polynomial z 5 + i = 0? Ans Since z 5 + i = 0 is ﬁfth-order polynomial.MA 232 1 (20 pts) Diﬀerential Equations Fall 1999 Kevin Dempsey (a)(3 pts) Does the complex number −i lie on the unit circle? (The unit circle is the circle of radius 1 centered at the origin. . . n = 0. . . it has ﬁve complex roots (c)(3 pts) Is −i a root of z 5 + i = 0? (i. 1.

y = c1 e−2x cos 3x + c2 e−2x sin 3x y 0 = (−2c1 e−2x cos 3x − 3c1 e−2x sin 3x) + (−2c2 e−2x sin 3x + 3c2 e−2x cos 3x) = (−2c1 + 3c2 )e−2x cos 3x − (3c1 − 2c2 )e−2x sin 3x y(0) = c1 = −1 y 0 (0) = −2c1 + 3c2 = 1 → c2 = − 1 3 1 y = −e−2x cos 3x − e−2x sin 3x 3 Test II Clarkson University . We have to ﬁnd the the values of c1 and c2 so that the solution passes through the point (0. e−2x sin 3x} and a general solution of the equation is therefore y = c1 e−2x cos 3x + c2 e−2x sin 3x where the constants c1 and c2 are arbitrary. −1) with a slope of 1.MA 232 2 (20 pts) Diﬀerential Equations Fall 1999 Kevin Dempsey Solve the initial value problem y 00 + 4y 0 + 13y = 0 y 0 (0) = 1 y(0) = −1 Ans Substituting y = erx gives the auxiliary equation r2 + 4r + 13 = 0 which has roots r = −2 ± 3i so the fundamental solution set for the given homogeneous diﬀerential equation is {e−2x cos 3x.

MA 232 3 (20 pts) Diﬀerential Equations Fall 1999 Kevin Dempsey Find a particular solution of y 00 + 2y 0 + y = x2 e−x + xe−x + e−x using the method of undetermined coeﬃcients (not just the form of the particular solution.1 on Page 197 of the text is Type (IV) so the particular solution form is yp = x2 (Ax2 + Bx + C)e−x Multiplying both sides by ex yp ex = Ax4 + Bx3 + Cx2 and then implicitly diﬀerentiating twice gives 0 yp ex + yp ex = 4Ax3 + 3Bx2 + 2Cx 00 0 0 (yp ex + yp ex ) + (yp ex + yp ex) = 12Ax2 + 6Bx + 2C 00 0 (yp + 2yp + yp )ex = 12Ax2 + 6Bx + 2C 00 yp + 2yp + yp = (12Ax2 + 6Bx2 + 2C)e−x 1 12 1 B= 6 1 C = 2 A= yp = x 2 µ 1 2 1 1 −x x + x+ e 12 6 2 ¶ Test II Clarkson University . xe−x }. Ans Substituting y = erx into the corresponding homogeneous diﬀerential equation y 00 +2y 0 +y = 0 gives the auxiliary equation r2 + 2r + 1 = (r + 1)2 = 0 which has repeated roots r = −1. you have to explicitly determine the constants). Now the RHS of the given nonhomogeneous equation is (x2 + x + 1)e−x which in Table 4. −1 so the fundamental solution set for the homogeneous equation is {e−x .

sin θ}. Ans Substituting y = erθ into the corresponding homogeneous equation y 00 + y = 0 gives the auxiliary equation r2 + 1 = 0 which has roots r = ±i so the fundamental solution set for the homogeneous equation is {y1 .MA 232 4 (20 pts) Diﬀerential Equations Fall 1999 Kevin Dempsey Find a general solution of the diﬀerential equation y 00 + y = sec θ using the method of variation of parameters. y2 } = {cos θ. The Wronskian of the fundamental solutions is ¯ ¯ ¯ cos θ sin θ ¯ ¯ ¯ ¯ = cos2 θ + sin2 θ = 1 W [y1 . y2 ](θ) = ¯ ¯ − sin θ cos θ ¯ Now the method of variation of parameters gives v1 = Z Z −gy2 Z (− sec θ)(sin θ) = dθ = − tan θ dθ = ln | cos θ| + C1 W 1 Z Z Z gy1 (sec θ)(cos θ) v2 = = dθ = 1 dθ = θ + C2 W 1 yp = v1 y1 + v2 y2 = cos θ ln | cos θ| + θ sin θ Hence a general solution of y 00 + y = sec θ is y = c1 cos θ + c2 sin θ + cos θ ln | cos θ| + θ sin θ Test II Clarkson University .

MA 232 5 (20 pts) Diﬀerential Equations Fall 1999 Kevin Dempsey Use the elimination method to ﬁnd a general solution of the ﬁrst-order linear syatem x0 + y = sin t −x + y 0 = sin t Ans In operator notation with D ≡ d the given system is dt D[x] + y = sin t −x + D[y] = sin t In order to eliminate y. sin t}. because the fundamental solution set for the corresponding homogeneous equation x00 + x = 0 is {cos t.1 the RHS of this equation is Type (III) and. the particular solution form is xp = tf where f = A cos t + B sin t Use the method of undetermined coeﬃcients to ﬁnd the constants A and B xp = tf x0p = f + tf 0 x00 = f 0 + f 0 + tf 00 = 2f 0 − tf = 2f 0 − xp p x00 + xp = 2f 0 = 2(−A sin t + B cos t) p A=B= xp = Hence a general solution for x is 1 x = c1 cos t + c2 sin t + t(cos t + sin t) 2 1 2 1 t(cos t + sin t) 2 Test II Clarkson University . operate on the ﬁrst equation by D to give D2 [x] + D[y] = cos t −x + D[y] = sin t Now subtract the second equation from the ﬁrst to give D2 [x] + x (= x00 + x) = cos t − sin t From Table 4.

MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey A general solution for y follows immediately from y = −x0 + sin t µ n ¶ µ ¶ = − −c1 sin t + c2 cos t + 1 (cos t + sin t) + 1 t(− sin t + cos t) + sin t 2 2 1 1 1 = c1 + sin t − c2 + cos t + t(sin t − cos t) 2 2 2 o If x is eliminated the general solution obtained for y is 1 y = a1 cos t + a2 sin t + t(sin t − cos t) 2 and the general solution for x that follows is x = y 0 − sin t µ 1 1 1 = a2 − cos t − a1 + sin t + t(cos t + sin t) 2 2 2 ¶ µ ¶ Test II Clarkson University .

m. 360. 362 8:30—9:30 p.MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey Test III Wednesday. Name Student # Signature If your last name begins with letter You are in A-E SC 162 F-L SC 362 M-Z SC 360 Please do not detach pages Loose pages will NOT be graded Put a check mark next to your recitation section Section 11 12 13 14 15 16 17 18 19 √ 1 2 3 4 5 Total √ Day Period Time Room Leader Monday 2 9—9:50 SC 386 Scott Gregowske Monday 4 11—11:50 SC 386 Sovia Lau Monday 4 11—11:50 SC 342 Chiu Wah Cheng Monday 7 3—3:50 Rowley 150 Chaya Tuok Monday 8 4—4:50 SC 303 Phillip Allen Tuesday 1 8—8:50 SC 356 Elizabeth Baker Tuesday 5 1—1:50 SC 344 Tom Pearsall Tuesday 7 3—3:50 SC 344 Wesley Kent Tuesday 8 4—4:50 SC 305 Ed Landry Test III Clarkson University . 1999 SC 162. December 1.

7 cos(7t − π/4) √ x0 (0) = 7. Ans x0 (t) = 7. x(t) is the displacement in meters of the mass above the equilibrium position t seconds after the motion begins. thereby stretching the spring a certain distance on coming to rest at equilibrium. Test III Clarkson University .1 sin(−π/4) = −1.1 sin 7t − 4 µ ¶ [This equation assumes that x(t) is positive upwards.444m/sec The spring was given an upwards velocity of 5.7/ 2 = 5. Hence the ﬁrst time that the mass passes downwards through the equilibrium position is given by t = 5π/28 = 0.444 m/sec (c)(6 pts) When will the mass ﬁrst pass through the equilibrium position in a downwards direction? Ans When the mass ﬁrst goes through the equilibrium position x(t) = 1. Ans √ x(0) = 1. The next time the mass goes through the equilibrium position x(t) = 1.778 m The mass was displaced 0. the equation of simple harmonic motion of the mass is π x(t) = 1.778 m downwards.MA 232 1a (10 pts) Diﬀerential Equations Fall 1999 Kevin Dempsey A mass is attached to a spring suspended from a high ceiling. (b)(2 pts) What was the initial velocity (direction and magnitude) of the mass? Explain your answer in words too.1 sin(7t − π/4) = 0 and 7t − π/4 = π at which time x0 (t) = −7.1 sin(7t − π/4) = 0 and 7t − π/4 = 0 but at this time x0 (t) = 7.1/ 2 = −0.561 seconds.7 m/sec (downwards).7 m/sec (upwards!).] (a)(2 pts) What was the initial displacement (direction and magnitude) of the mass? Explain your answer in words too. When the mass is given an initial displacement and an initial velocity.7 cos(−π/4) = 7.

57◦ ) θ1 = tan−1 (tan θ1 ) = tan−1 (−3) = −1. 5 1 sin θ2 = √ 5 θ2 = tan−1 (tan θ2 ) = tan−1 (1/2) = 0.57 (= 2 e−i0.45π = 261.MA 232 1b (10 pts) Diﬀerential Equations Fall 1999 Kevin Dempsey Express the complex number Ans 1 − 3i in polar form.45π ) 2+i 5e Ans √ 1 − 3i 1 − 3i 2 − i 2 − i − 6i + 3i2 −1 − 7i 1 7 = · = = = − − i = 2 eiθ 2 2+i 2+i 2−i 4−i 5 5 5 7 sin θ = − √ 5 2 1 cos θ = − √ .57 rad 1 − 3i √ i 4.87◦ ) Test III Clarkson University .40π = −71. 10 3 sin θ1 = − √ 10 (= −0.71i = 2 ei 4. 5 2 θ = tan−1 (tan θ) + π = tan−1 (7) + π = 1.25i √ = √ 0.46i = 2 e−1.57 √ i1.43 + π = 4. 2+i √ 10 eiθ1 √ 1 − 3i = √ iθ = 2 ei(θ1 −θ2 ) 2+i 5e 2 1 cos θ1 = √ .45π = 2e = 2e 2+i (= 1.55π = 2 ei1.57◦ ) √ √ √ √ 1 − 3i 10 e−1.46 rad (= 0.25 rad 2 cos θ2 = √ .15π = 26.

F (2) = Z ∞ 0 ¶ 1 e−3s e6−3s − + s s s−2 Z 3 0 (s > 2) e −2t f (t) dt = e −2t · 1 dt + Z ∞ 3 e −2t · e dt = 2t Z 3 0 e −2t dt + lim N→∞ 3 Z N 1 dt Since the limit does not converge. F is not deﬁned at s = 2. Ans F (s) = Z 3 0 ( Z ∞ 0 e−st f (t) dt of 1 e2t 0<t<3 3<t e−st · 1 dt + #3 0 Z ∞ 3 e−st e2t dt #N 3 e−st = −s + lim µ N→∞ " e−(s−2)t −(s − 2) e−3s 1 e−(s−2)N e−(s−2)3 − − = + lim N→∞ −(s − 2) −s −s −(s − 2) = Note that F (2) DNE. Test III Clarkson University .MA 232 2 (20 pts) Diﬀerential Equations Fall 1999 Kevin Dempsey Find the Laplace Transform F (s) = L{f (t)} (s) = f (t) = being careful to state the domain of F .

MA 232 3 (20 pts) Diﬀerential Equations Fall 1999 Kevin Dempsey Find the Laplace Transform of t sin 3t sin 5t Ans Let f (t) = sin 3t sin 5t 1 [cos(3t − 5t) − cos(3t + 5t)] 2 1 = (cos 2t − cos 8t) 2 µ ¶ 1 s s − F (s) = L{f (t)} = 2 s2 + 4 s2 + 64 = − d (1)(s2 + b2 ) − (s)(2s) s2 − b2 s =− = 2 ds s2 + b2 (s2 + b2 )2 (s + b2 )2 dF ds # L{t sin 3t sin 5t} = L{tf(t} (s) = − " s2 − 64 1 s2 − 4 = − 2 2 (s2 + 4)2 (s + 64)2 Test III Clarkson University .

MA 232 4 (20 pts) Diﬀerential Equations Fall 1999 Kevin Dempsey Find Y (s) = L{y(t)} (s) if y(t) satisﬁes the initial-value problem y 00 + 3y 0 + 5y = cos 3t y 0 (0) = 2 Ans h y(0) = −1 s2 Y (s) − sy(0) − y 0 (0) + 3 [sY (s) − y(0)] + 5Y (s) = h i s2 s2 s2 s2 s +9 s +9 s +9 s −s−1 +9 s2 Y (s) − s(−1) − 2 + 3 [sY (s) − (−1)] + 5Y (s) = (s2 + 3s + 5)Y (s) + s − 2 + 3 = (s2 + 3s + 5)Y (s) = i = s − (s + 1)(s2 + 9) s2 + 9 s − [s3 + 9s + s2 + 9] s2 + 9 = s3 + s2 + 8s + 9 Y (s) = − 2 (s + 3s + 5)(s2 + 9) Test III Clarkson University .

MA 232 5 (20 pts) Diﬀerential Equations Fall 1999 Kevin Dempsey Find the Inverse Laplace Transform of 2s2 − 3s + 13 (s2 − 2s + 5)(s + 3) Ans 2s2 − 3s + 13 A(s − 1) + B(2) C + = 2 − 2s + 5)(s + 3) 2 + 22 (s (s − 1) s+3 2s2 − 3s + 13 = [A(s − 1) + B(2)] (s + 3) + C (s − 1)2 + 22 s = −3 s=1 → → 18 + 9 + 13 = 0 + C (−4)2 + 4 h i h i → C =2 → B= 1 2 2 − 3 + 13 = [0 + B(2)] (4) + (2) [0 + 4] → 13 = [−A + 1] (3) + 2[1 + 4] → s=0 A=0 L −1 ( 2s2 − 3s + 13 1 2 2 = + 2 − 2s + 5)(s + 3) 2 + 22 (s 2 (s − 1) s+3 2s2 − 3s + 13 (s2 − 2s + 5)(s + 3) ) 1 2 = L−1 2 (s − 1)2 + 22 1 = et sin 2t + 2e−3t 2 ( ) +L −1 ½ 2 s+3 ¾ Test III Clarkson University .

Wherever possible. SC 360 & SC 362 Spring 1999 Please show your working. 1999 8:30-9:30 p. Three blank pages are provided for additional workspace. please DO NOT hand in loose pages.m. The last page is a formula sheet. February 17. check your answer. Problem 1 2 3 4 5 Total Score Name Student ID # Signature . and check to see that you have indeed answered the question. Feel free to use both sides of each page. they WILL NOT be graded. use correct notation.MA232 Differential Equations TEST I Wednesday.

the general solution for y not identically zero is y −1 + cos x = C Name Student ID # Signature . dy + y 2 sin x = 0 dx Solution (a)(5 pts) y ≡ 0 clearly satisfies the ODE and therefore is a solution. the variables x and y can be separated and the ODE can be written as the equality of two differentials as − dy = sin x dx y2 = ∫ sin x dx Integrating both sides of this equation gives ∫− y dy 2 1 = − cos x + C y Hence.1 (20 pts) Consider the first-order nonlinear separable ODE (a) What solution can you see immediately? (b) Solve the ODE for the general solution. (b)(15 pts) If y ≠ 0 .

and let x (t ) represent the amount of salt in the tank at time t . Now ∫ P(t ) dt = ∫ dt = ln (10 + t ) so the 10 + t 10 + t P ( t ) dt integrating factor µ (t ) = e ∫ = e ln (10+t ) = 10 + t . as the tank is about to overflow. and the general solution of x (t ) = 1 µ (t ) {∫ µ (t )Q(t ) dt + C}= 101+ t {∫ 40(10 + t ) dt + C} 1 {20(10 + t ) 2 + C} 10 + t C = 20(10 + t ) + 10 + t = Now at time zero there is no salt in the tank so x (0) = 200 + C = 0 → C = −2000 10 Thus the amount of salt in the tank at any time t is given by 2000 x (t ) = 20(10 + t ) − 10 + t After 10 minutes. the amount of salt in the tank is x (10) = 400 − 2000 = 400 − 100 = 300 g .2 (20 pts) A tank whose volume is 40 L initially contains 20 L of water. this equation is dt dx 1 + x = 40 K (*) { dt 10 + t Q(t ) 13 2 P(t ) wherein P(t ) = (*) is 1 1 and Q (t ) = 40 . A solution containing 10 g/L of salt flows into the tank at a rate of 4 L/min. 20 Alternatively. Then dx g L x (t ) g L x = 10 4 2 = 40 − − dt L min 20 + 2t L min 10 + t dx In the standard form + P(t ) x = Q . How much salt is in the tank just before the tank overflows? Solution Let t be the time in minutes [min] after the salt solution begins to flow. and the wellstirred mixture flows out at a rate of 2 L/min. one might do the integration above differently as per Name Student ID # Signature .

x (t ) = 1 µ (t ) {∫ µ (t )Q(t ) dt + C}= 101+ t {∫ 40(10 + t ) dt + C} 1 {400t + 20t 2 + C} 10 + t 1 {0 + 0 + C} = C = 0 → C = 0 x (0) = 10 + 0 10 2 400t + 20t x (t ) = 10 + t 4000 + 2000 x (10) = = 300 g 20 = Name Student ID # Signature .

Solution 3 − 2 2 1 − 2 1 2 −1 − 2 R2 → R1 → R2 − 3R1 R3 − 2 R1 R1 → R2 3R 2 − 4 R 3 R1 → R1 R3 1 − 2 1 3 − 2 2 2 −1 − 2 1 − 2 1 0 − 4 −1 0 3 − 4 1 − 2 1 0 4 −1 0 0 13 3x1 x1 2 x1 − 2 x2 − 2 x2 − x2 + 2 x3 + x3 − 2 x3 = 9 = 5 = −1 9 5 − 1 5 9 − 1 5 −6 − 11 5 − 6 26 5 − 6 2 3 − 4 2 3 − 1 2 1 − 1 2 1 − 2 1 R2 0 4 −1 R3 / 2 0 0 1 R1 − R3 R2 + R3 R3 R1 1 − 2 0 0 4 0 0 0 1 → → 1 − 2 0 R2 / 4 0 1 0 0 0 1 R3 R2 R3 1 0 0 0 1 0 0 0 1 R1 + 2 R2 → Name Student ID # Signature .3 (20 pts) Solve the linear system shown here on the right by performing elementary row operations on the augmented matrix.

in order to guarantee existence and uniqueness.e.4 (20 pts) (a) Determine the largest interval on which the initial value problem (t − 1) is certain to have a unique solution. q(t ) = . g (t ) = t −1 t −1 t −1 These functions are continuous on (−∞. but by Theorem 2. the largest interval on which the given initial value problem is guaranteed to have a unique solution is ( −∞. y ′( −2) = 1 is nonlinear (i. 2 dt dt y ( −2) = 2.+∞) .1) = {x : −∞ < x < 1} . (b) (7 pts) L( y1 + y 2 ) = ( y1 + y 2 )′′ + ( y1 + y 2 ) 2 2 ′ ′ = y1′ + y 2′ + y12 + 2 y1 y 2 + y 2 = L( y1 ) + L( y 2 ) + 2 y1 y 2 ≠ L( y1 ) + L( y 2 ) L(cy ) = ( cy )′′ + ( cy ) 2 = cy ′′ + c 2 y 2 ≠ cL( y ) Name Student ID # Signature .1) and (1. Solution (a)(13 pts) Writing the ODE in the standard form (Page 157) gives (for t away from 1) so by comparison d2y dy + p (t ) + q(t ) y = g (t ) of Theorem 2 2 dt dt d2y 3t dy 4 sin t − + y= 2 dt t − 1 dt t − 1 t −1 p( t ) = − 4 sin t 3t . Hence. (b) Show that the differential operator L defined by L( y ) = y ′′ + y 2 d2y dy − 3t + 4 y = sin t . Do not attempt to find the solution. the initial point x 0 = −2 has to be in the interval in question. not linear).

y1 ( x ) = e 2 x and y 2 ( x ) = e 3 x are linearly independent solutions of (*) on ( −∞. ∞) = {x : −∞ < x < ∞} . the { y1 . y 2 } is a fundamental solution set on ( −∞. y ′(0) = −4 Solution (a)(10 pts) Let L( y ) = d2y dy − 5 + 6 y . then 2 dt dt d 2 y1 dy L( y1 ) = − 5 1 + 6 y1 = ( 4e 2 x ) − 5( 2e 2 x ) + 6( e 2 x ) = 0 2 dt dt 2 d y2 dy L( y 2 ) = − 5 2 + 6 y 2 = (9e 3 x ) − 5(3e 3 x ) + 6( e 3 x ) = 0 2 dt dt so the two given functions are solutions to (*). is a general solution of (*). (b) Find a general solution to (*). ∞) and therefore it is immediately true that y ( x ) = c1e 2 x + c 2 e 3 x wherein c1 and c2 are arbitrary constants. (c)(5 pts) y ( x ) = c1e 2 x + c2 e 3 x y ′( x ) = 2c1e 2 x + 3c2 e 3 x y (0) = c1 + c2 = −1 y ′(0) = 2c1 + 3c2 = −4 c1 = 1 → c2 = −2 y ( x ) = e 2 x − 2e 3 x Name Student ID # Signature . ∞) . (c) Find the solution that satisfies the initial conditions y (0) = −1. ∞) .5 (20 pts) (a) Show that the two functions y1 ( x ) = e 2 x and y 2 ( x ) = e 3 x are linearly independent solutions of the linear second-order constant coefficient ODE d2y dy − 5 + 6y = 0 K (*) 2 dx dx on the whole real line ( −∞. (b)(5 pts) Also by virtue of the same Wronskian result. The associated Wronskian function is e2x W ( x) = 2 x 2e e3x = 3e 5 x − 2e 5 x = e 5 x 3x 3e Since W ( x ) ≠ 0 on ( −∞.

in SC 342 Tues 8-8:50 a.m. in SC 346 Tues 4-4:50 p. in SC 340 Mon 9-9:50 a. in SC 346 Instructor Tom Masser Wesley Kent Patrick Chan Lisa Meeker Sean Dorey Sean Dorey Kathy Christman Check One . in SC 344 Tues 11-11:50 a.m. in SC 344 Tues 10-10:50 a.m. The graders will assign credit for correct working. in SC 342 Mon 4-4:50 p. SC 360 & 362 Name Student ID# Signature Please show your work carefully and legibly.MA232 Diﬀerential Equations Spring 1999 Test II Wednesday. March 24.m.m.m. 1999 8:30—9:30 p.m. Problem Score 1 2 3 4 5 Total Period 1 2 8 1 3 4 8 Recitation MA232-11 MA232-12 MA232-13 MA232-21 MA232-22 MA232-23 MA232-24 Day-Time-Room Mon 8-8:50 a.m.

3. synthetic division for the factor (r − 3) gives 3 1 0 1 -5 3 -2 8 -6 2 -6 6 0 indicating that the cubic factorizes to (r − 3)(r2 − 2r + 2) = 0. Solution The auxiliary equation is r3 − 5r2 + 8r − 6 = 0. and a general solution of the third order diﬀerential equation is y(x) = c1 ex cos x + c2 ex sin x + c3 e3x . .Problem 1 (20 pts) Student ID # Use synthetic division to ﬁnd a general solution of the third order constant coeﬃcient differential equation y 000 − 5y 00 + 8y 0 − 6y = 0. so the roots of the auxiliary equation are r = 1 ± i.

determine the form of a particular solution yp (x) for the nonhomogeneous second order constant coeﬃcient diﬀerential equation y 00 − 4y 0 + 4y = g(x).ii) yp (x) = A cos 2x + B sin 2x ¯ ¯ ¯ ¯ ¯ 2x 2x ¯ e + x2e ¯ xe2x = e4x + 2xe4x − 2xe4x = e4x 6= 0 on (−∞. y2 on (−∞. so the fundamental solution set is {y1 = e2x . y2 = xe2x }. (b)(16 pts) Using Table 4. y2 ](x) = ¯ ¯ ¯ ¯ ¯ ¯ ¯ (ii) g(x) = x2 sin 2x. 2.iv) yp (x) = x2 ³ 1 2 x 12 − 1 2 ´ e2x . (iv) g(x) = x2 e2x − e2x . Solution (a) y 00 − 4y 0 + 4y = 0 has the auxiliary equation r2 − 4r + 4 = (r − 2)2 = 0 which has the repeated root r = 2.i) yp (x) = − 3 sin 2x 8 (b.ii) yp (x) = ³ 1 2 x 8 + 1 x cos 2x + − 1 x − 8 8 ´ ³ 3 32 ´ sin 2x (b. ∞) is demonstrated by noting that the Wronskian W [y1 .iii) yp (x) = e3x (b.Problem 2 (20 pts) Student ID # (a)(4 pts) Show that {e2x . ∞). actual particular solutions are (b. for (i) g(x) = 3 cos 2x.iv) yp (x) = x2 (Ax2 + Bx + C)e2x In fact.1 (see the Formula Sheet).i) (b.iii) yp (x) = Ae3x (b. xe2x } is a fundamental solution set for the homogeneous second order constant coeﬃcient diﬀerential equation y 00 − 4y 0 + 4y = 0. yp (x) = (Ax2 + Bx + C) cos 2x + (Dx2 + Ex + F ) sin 2x (b. The linear independence of y1 . (iii) g(x) = e3x . e2x 2e 2x (b.

Solution (a) yp (x) = Ax2 + Bx + C 0 yp (x) = 2Ax + B 00 yp (x) = 2A 00 0 yp − 2yp + yp = 2A − 2(2Ax + B) + Ax2 + Bx + C A B − 4A 2A − 2B + C yp (x) = = = = = Ax2 + (B − 4A)x + (2A − 2B + C) = x2 + 1 1 0 → B=4 1 → C =7 x2 + 4x + 7 (b) By linearity and the principle of superposition we need only ﬁnd a particular solution for y 00 − 2y 0 + y = x−2 ex using the method of variation of parameters and then add the result obtained in part (a). The Wronskian W [y1 .Problem 3 (20 pts) Student ID # (a)(10 pts) Find a particular solution yp (x) of the nonhomogeneous second order constant coeﬃcient diﬀerential equation y 00 − 2y 0 + y = x2 + 1 using the method of undetermined coeﬃcients. y2 } = {ex . xex }. Hence. (b)(10 pts) Find a particular solution yp (x) of the the nonhomogeneous second order constant coeﬃcient diﬀerential equation y 00 − 2y 0 + y = x−2 ex + x2 + 1 using the method of variation of parameters. a particular solution for x > 0 is yp (x) = −ex ln x + x2 + 4x + 7 while for x < 0 it is yp (x) = −ex ln(−x) + x2 + 4x + 7 . Thus Z −x−2 ex · xex dx = −x−1 dx = − ln |x| e2x Z −2 x x Z 1 x e ·e dx = x−2 dx = − v2 (x) = 2x e x 1 x x x yp (x) = (− ln |x|)e + (− )(xe ) = −e (1 + ln |x|) x yp (x) = −ex ln |x| v1 (x) = Z The last equation holds because ex satisﬁes the homogeneous equation. The homogeneous ODE y 00 − 2y 0 + y = 0 has the auxiliary equation r2 − 2r + 1 = (r − 1)2 = 0 so the fundamental solution set is {y1 . y2 ](x) = e2x as can be readily veriﬁed.

Problem 4

(20 pts)

Student ID #

Use the elimination method to ﬁnd a general solution for the ﬁrst order linear system dx − y = t2 , dt dy = 1. x+ dt . Solution In operator notation, the system is (with D ≡ d/dt) D[x] − y = t2 x + D[y] = 1 Operating on the ﬁrst equation by D gives D2 [x] − D[y] = 2t x + D[y] = 1 Adding the two equations eliminates y(t); D2 [x] + x = x00 + x = 2t + 1. The general solution of this equation is x(t) = c1 cos x + c2 sin x + 2t + 1. The ﬁrst equation of the original system gives then y(t) = dx − t2 = −c1 sin x + c2 cos x + 2 − t2 . dt

Problem 5

(20 pts)

Student ID #

A mass of 4 kg is attached to a spring hanging from a ceiling, thereby stretching the spring 9.8 cm on coming to rest at equilibrium. The mass is then lifted up 10 cm above the equilibrium point and given a downward velocity of 1 m/sec. Determine the equation for the simple harmonic motion of the mass. When will the mass ﬁrst reach its minimum height after being set in motion. Solution Hooke’s law gives the spring constant; mg = k` and the equation of motion is m¨ + kx = 4¨ + 400x = 4(¨ + 100x) = 0, x x x the general solution of which is x(t) = c1 cos 10t + c2 sin 10t = A sin(10t + φ) x(t) = 10A cos(10t + φ) ˙ 1 A sin φ = x(0) = − 10 x(0) ˙ 1 A cos φ = = 10 10 π φ = − 4 π x(t) = A sin(10t − ) 4 x(t) = A π → sin(10t − ) = 1 4 π π → (10t − ) = 4 2 3π → t = = 0.236 sec 40 → (4)(9.8) = k 9.8 100 → k = 400,

MA232

Diﬀerential Equations

Spring 1999

Test III

Wednesday, April 21, 1999 8:30—9:30 p.m. SC 360 & 362

Name Student ID# Signature

Please show your work carefully and legibly. The graders will assign credit for correct working.

Problem Score 1 2 3 4 5 Total

Period 1 2 8 1 3 4 8

Recitation MA232-11 MA232-12 MA232-13 MA232-21 MA232-22 MA232-23 MA232-24

Day-Time-Room Mon 8-8:50 a.m. in SC 340 Mon 9-9:50 a.m. in SC 342 Mon 4-4:50 p.m. in SC 342 Tues 8-8:50 a.m. in SC 344 Tues 10-10:50 a.m. in SC 344 Tues 11-11:50 a.m. in SC 346 Tues 4-4:50 p.m. in SC 346

Instructor Tom Masser Wesley Kent Patrick Chan Lisa Meeker Sean Dorey Sean Dorey Kathy Christman

Check One

Problem 1

(20 pts)

Student ID #

A linear homogeneous diﬀerential equation with constant coeﬃcients has the auxiliary equation (r + 1)(r − 3)2 (r + 2)3 (r2 + 2r + 5)2 r4 = 0. (a)(3 pts) What is the order of this equation?

14

(b)(3 pts) What does the fundamental theorem of algebra tell us about this auxiliary equation?

That there are 14 roots (counting multiplicities) which may be real or complex.

(c)(5 pts) List the roots of this auxilary equation? If a root has mutiplicity m, repeat it m times.

r = −1, 3, 3, −2, −2, −2, −1 ± 2i, −1 ± 2i, 0, 0, 0, 0 (d)(7 pts) Write down a general solution of the linear homogeneous diﬀerential equation whose auxiliary equation is shown above.

c1 e−x + (c2 + c3 x)e3x + (c4 + c5 x + c6x2 )e−2x + (c7 + c8 x)e−x cos 2x + (c9 + c10 x)e−x sin 2x + c11 + c12x + c13 x2 + c14 x3

(e)(2 pts) Should the number of arbitrary constants in your general solution agree with the number of roots (counting multiplicities) listed in Part (c)?

Yes

Problem 2

(20 pts)

Student ID #

**(a)(4 pts) Let f (t) be a function deﬁned on [0, +∞). The Laplace transform of f is the function F deﬁned by the integral
**

Z ∞

0

F (s) =

e−st f (t) dt

. . . (∗)

(b)(4 pts) Describe the domain of F (s)

All the values of s for which the integral in (*) exists

**(c)(4 pts) Explain mathematically what it means to say that the integral in (*) exists as an improper integral
**

Z ∞

0

e

−st

f (t) dt = lim

N →∞ 0

Z N

e−st f (t) dt

(d)(4 pts) Determine the Laplace transform of the function f(t) = cos bt, where b is a nonzero constant.

F (s) =

= =

**¯ e−st ¯ (−s cos bt + b sin bt¯ = lim ¯ N →∞ s2 + b2 0
**

N →∞

Z ∞

0

**e−st cos bt dt = lim
**

"

N →∞ 0

Z N

e−st cos bt dt

¯N #

lim

1 e−sN (−s cos bN + b sin bN ) − 2 (−s) 2 + b2 s s + b2 for s > 0

s

s2 + b2

(e)(4 pts) What is the domain of the Laplace transform determined in Part (d)? Explain why.

The domain is s > 0 since for values of s ≤ 0 the limit lim e−sN (−s cos bN + b sin bN ) and the integral in (*) N →∞ do not exist.

s > −5 and the domain common to all is s > 2 . and if f1 . in order. s > 0. The domain is s > 2 since the respective domains of the Laplace transforms in (b) are. what does the linearity of the Laplace transform operator L tell us about L {c1 f1 + c2 f2 }? c1L {f1 } + c2 L {f2} (b)(10 pts) Use the linearity of L to determine the Laplace transform n L 5 + 3e2t + 5t3 e2t + 7 cos 3t + 9e−5t cos 3t . c2 are arbitrary constants. s > 2. f2 are functions whose Laplace transforms exist for s > α. o L {5 + 3e2t + 5t3 e2t + 7 cos 3t + 9e−5t cos 3t} = 5L {1} + 3L {e2t } + 5L {t3 e2t } + 7L {cos 3t} + 9L {e−5t cos 3t} = 5 1 1 3! s s+5 +3 +5 +7 2 +9 s s−2 (s − 2)4 s + 32 (s + 5)2 + 32 (c)(5 pts) What is the domain of the Laplace transform in (b)? Explain why. s > 2. s > 0.Problem 3 (20 pts) Student ID # (a)(5 pts) If c1 .

Problem 4 (20 pts) Student ID # (a)(10 pts) Use an appropriate trigonometric identity to determine the Laplace transform L {sin 3t cos 7t} (s) From the identities on the formula sheet 1 [sin(A + B) + sin(A − B)] 2 1 sin 3t cos 7t = [sin 10t + sin(−4t)] 2Ã ! 1 10 −4 L {sin 3t cos 7t} (s) = + 2 2 s2 + 100 s + 16 5 2 = 2 − 2 s + 100 s + 16 sin A cos B = or 1 [sin(A + B) − sin(A − B)] 2 1 [sin 10t − sin 4t] cos 7t sin 3t = 2Ã ! 1 10 4 L {cos 7t sin 3t} (s) = − 2 2 s2 + 100 s + 16 5 2 = 2 − 2 s + 100 s + 16 cos A sin B = .

what is the Laplace transform of t2 t? 2 s3/2 √ 1 π L t (s) = F (s) = 2 s3/2√ n √ o 3 π dF (s) = − 5/2 L −t t (s) = ds 4√ s F n √ o d 15 π L t2 t (s) = (s) = 2 ds 8 s7/2 n√ o .Problem 4 (Cont’d) Student ID # √ √ √ 1 π (b)(10 pts) Given that the Laplace transform of t is .

y 0 (0) = 0. has the partial fraction expansion Y (s) = (s2 s−1 1 1 1 1 1 2 − + = . namely. where y(t) satisﬁes the initial value problem y 00 − 3y 0 + 7y = cos 3t. Q(s) Taking the Laplace transform of both sides [s2 Y (s) − sy(0) − y 0 (0)] − 3 [sY (s) − y(0)] + 7Y (s) = h s2 Y (s) − s(−2) − 1 − 3 [sY (s) − (−2)] + 7Y (s) = (s2 − 3s + 7) Y (s) = = = Y (s) = i s s2 + 32 s s2 + 9 s − 2s + 7 s2 + 9 s − (2s − 7)(s2 + 9) s2 + 9 s − (2s2 − 7s2 + 18s − 63) s2 + 9 −2s3 + 7s2 − 17s + 63 (s2 − 3s + 7)(s2 + 9) (b)(4 pts) If y(t) satisﬁes the initial value problem y 00 − 2y 0 + 7y = t2 et . 3 2 − 2s + 7 − 2s + 7)(s − 1) 18 s 18 s − 1 3 (s − 1)3 Find the solution of the initial value problem by inverting this Laplace transform? y(t) = √ 1 t 1 t 1 e cos 6t − e + t2 et 18 18 6 .Problem 5 (20 pts) Student ID # (a)(6 pts) Given that Y (s) = L {y} (s). y 0 (0) = 1. y(0) = 0. then the Laplace transform of y(t). Y (s) = L {y} (s). ﬁnd the rational function y(0) = −2. P (s) that Y (s) is equal to.

(s − 1)(s2 + 4s + 13) Y (s) = 4s2 + 13s + 19 (s − 1) [(s + 2)2 + 32] A B(s + 2) + 3C + s−1 (s + 2)2 + 32 = 4s2 + 13s + 19 = A [(s + 2)2 + 9] + [B(s + 2) + 3C] (s − 1) s = 1 → 4 + 13 + 19 = A(32 + 9) + 0 → 36 = 18A → A = 2 s = −2 → 16 − 26 + 19 = (2)(9) + (3C)(−3) → 9 = 18 − 9C → C = 1 s = 0 → 19 = (2)(4 + 9) + (2B + 3)(−1) → −4 = −2B → B = 2 Y (s) = 2 2(s + 2) + 3 + s−1 (s + 2)2 + 32 .Problem 5 (Cont’d) Student ID # (c)(10 pts) Determine the partial fraction expansion for the rational function Y (s) = 4s2 + 13s + 19 .

Problem 1 2 3 4 5 6 7 8 9 10 Bonus Total Score Period 1 2 8 1 3 4 8 Recitation MA232-11 MA232-12 MA232-13 MA232-21 MA232-22 MA232-23 MA232-24 Day-Time-Room Mon 8-8:50 a.MA232 Diﬀerential Equations Final Wednesday. in SC 344 Tues 10-10:50 a. May 5.m. Spring 1999 Name Student ID# Signature Please show your work carefully and legibly. 3:15 . in SC 340 Mon 9-9:50 a.m.m.m. in SC 346 Instructor Tom Masser Wesley Kent Patrick Chan Lisa Meeker Sean Dorey Sean Dorey Kathy Christman Check One . The graders will assign credit for correct working. in SC 346 Tues 4-4:50 p.m.m.6:15 p. in SC 344 Tues 11-11:50 a. in SC 342 Mon 4-4:50 p.m. in SC 342 Tues 8-8:50 a.m.

1 (20 pts) (a)(15 pts) Find a general solution of the separable ODE dy y cos x = dx 1 + 2y2 (b)(5 pts) Use your answer to Part (a) to solve the initial value problem y cos x dy . = dx 1 + 2y2 (a) 1 + 2y 2 dy = cos x dx y ln |y| + y 2 = sin x + C (b) 0+1 = 0+C ln |y| + y2 = sin x + 1 y(0) = 1 Student ID # .

and the well-stirred mixture ﬂows out at a rate of 3 L/min.2 (20 pts) Student ID # A 60 L tank initially contains 30 L of pure water. Then dx dt = (rate in) − (rate out) = (10)(6) − = 60 − 1 dx + x(t) = 60 dt 10 + t µ(t) = e x(t) = R dt 10+t x(t) ·3 30 + 3t x(t) 10 + t = eln(10+t) = 10 + t ∙Z ∙Z 1 10 + t 1 10 + t µ(t)Q(t) dt + C ¸ ¸ = (10 + t)60 dt + C = i 1 h 30(10 + t)2 + C 10 + t = 30(10 + t) + C 10 + t x(0) = 0 = 300 + C → C = −3000 10 3000 10 + t x(t) = 30(10 + t) − x(10) = 600 − 3000 = 600 − 150 = 450 g 20 . A solution containing 10 g/L of salt ﬂows into the tank at a rate of 6 L/min. How much salt is in the tank just before it overﬂows? Let x(t) denote the amount of salt in the tank at time t.

(iii) g(x) = ex cos 2x. determine the form of a particular solution yp (x) of the corresponding nonhomogeneous second order constant coeﬃcient diﬀerential equation y 00 − 2y 0 + 5y = g(x). (Be sure in each answer to specify the appropriate value of s. ex sin 2x} is a fundamental solution set. for each of the following four right-hand sides: (i) g(x) = x3 + 1.) (a) r2 − 2r + 5 = 0 (r − 1)2 + 4 = 0 r = 1 ± 2i So {ex cos 2x. (b)(16 pts) Using Table 4. (iv) g(x) = xex cos 2x. (b) (i) yp (x) = Ax3 + Bx2 + Cx + D (ii) yp (x) = A cos 2x + B sin 2x (iii) yp (x) = x(Aex cos 2x + Bex sin 2x) (iv) yp (x) = x{(Ax + B)ex cos 2x + (Cx + D)ex sin 2x} .3 (20 pts) (a)(4 pts) Show that yh (x) = c1 ex cos 2x + c2 ex sin 2x Student ID # is a general solution of the homogeneous second order constant coeﬃcient diﬀerential equation y 00 − 2y 0 + 5y = 0.1 (see the Test II Formula Sheet). (ii) g(x) = cos 2x.

W (x) = ¯ ¯ 3x ¯ e ¯ ¯ ¯ 3e3x Z Z ¯ ¯ ¯ ¯ = e6x + 3xe6x − 3xe6x = e6x 3x + x3e3x ¯ ¯ e xe3x v1 (x) = −g(x)y2 (x) dx = W (x) g(x)y1 (x) dx = W (x) Z −x−3 e3x · xe3x dx = W (x) Z −x−2 dx = x−1 v2 (x) = Z x−3 e3x e3x x−2 dx = e6x −2 1 1 yp (x) = v1 (x)y1 (x) + v2 (x)y2 (x) = x−1 e3x − x−2 xe3x = x−1 e3x 2 2 (c) 1 y(x) = c1 e3x + c2 xe3x + x−1 e3x 2 . ∞) as the following nonzero Wronskian attests. xe3x } is a fundamental solution set on (−∞.4 (20 pts) Student ID # (a)(4 pts) Find a general solution of the linear second order constant coeﬃcient homogeneous diﬀerential equation y 00 − 6y 0 + 9y = 0 (b)(12 pts) Find a particular solution yp (x) of the nonhomogeneous diﬀerential equation y 00 − 6y 0 + 9y = x−3 e3x using the method of variation of parameters. 3 yh (x) = c1 e3x + c2 xe3x (b){y1 (x). (c)(4 pts) From Parts (a) and (b). what is a general solution of the ODE in Part (b)? (a) r 2 − 6r + 9 = 0 (r − 3)2 = 0 r = 3. y2 (x)} = {e3x .

x+ dt D[x] + y = t2 x + D[y] = 1 D 2 [x] + D[y] = 2t x + D[y] = 1 x00 − x = 2t − 1 r2 − 1 = 0 → r = ±1 x(t) = c1 e−t + c2 et + 1 − 2t dx y(t) = − + t2 dt = c1 e−t − c2 et + 2 + t2 . dt dy = 1.5 (20 pts) Student ID # Use the elimination method to ﬁnd a general solution for the ﬁrst order linear system dx + y = t2 .

Determine the equation for the simple harmonic motion of the mass. The mass is then pulled down 10 cm below the equilibrium point and given an upward velocity of 1 m/sec. When will the mass ﬁrst reach its maximum height after being set in motion? 4x00 + 400x = 0 x00 + 100x = 0 r2 + 100 = mg = k` → (4)(9.8) = k 9.8 → k = 400 100 x(t) = c1 cos 10t + c2 sin 10t = A sin(10t + φ) x (t) = 10A cos(10t + φ) 1 A sin φ = x(0) = 10 x0 (0) 1 A cos φ = =− 10 10 3π φ = 4 x(t) = −A → sin(10t + φ) = −1 3π 10t + φ = 2 3π/2 − φ 3π t = = = 0.236 sec 10 40 0 → r = ±10i .8 cm on coming to rest at equilibrium.6 (20 pts) Student ID # A mass of 4 kg is attached to a spring hanging from a ceiling. thereby stretching the spring 9.

ﬁnd the rational function y(0) = −2. y 0 (0) = 1.) Y (s) = 1 1 −6(s − 1) + 1 − + s s+4 (s − 1)2 + 22 1 y(t) = 1 − e−4t − 6et cos 2t + et sin 2t 2 .7 (20 pts) Student ID # (a)(6 pts) Given that Y (s) = L {y(t)} (s). P (s) that Y (s) is equal to. (Fully expand the numerator polynomial P (s).) Q(s) h s2 Y (s) − sy(0) − y 0 (0) − 2 [sY (s) − y(0)] + 5Y (s) = s2 Y (s) − s(−2) − (1) − 2 [sY (s) − (−2)] + 5Y (s) = ³ i i s2 2 + 22 h 2 s2 + 4 s2 s2 2 +4 2 − 2s + 5 +4 s2 − 2s + 5 Y (s) + 2s − 1 − 4 = ³ ´ s2 − 2s + 5 Y (s) = Y (s) = ´ 2 − (2s − 5)(s2 + 4) (s2 − 2s + 5)(s2 + 4) −2s3 + 5s2 − 8s + 22 (s2 − 2s + 5)(s2 + 4) = (b)(4 pts) If the Laplace transform of y(t) is given by Y (s) = 1 1 7 − 6s − + 2 s s + 4 s − 2s + 5 what is y(t)? (Simply invert this Laplace transform. where y(t) satisﬁes the initial value problem y 00 − 2y 0 + 5y = sin 2t.

− 2s + 5) s(s2 3s + 5 − 2s + 5) = A B(s − 1) + C2 + s (s − 1)2 + 22 3s + 5 = A[(s − 1)2 + 4] + [B(s − 1) + 2C]s s = 0 → 5 = 5A → A = 1 s = 1 → 8 = (1)(4) + (0 + 2C)(1) → 2C = 4 → C = 2 s = −1 → 2 = (1)(8) + (−2B + 4)(−1) → 2B = −2 → B = −1 3s + 5 s(s2 − 2s + 5) = 1 −(s − 1) + 2(2) + s (s − 1)2 + 22 1 5−s + 2 s s − 2s + 5 = .7 (Cont’d) Student ID # (c)(10 pts) Determine the partial fraction expansion for the rational function Y (s) = s (s2 3s + 5 .

(b)(5 pts) Compute the matrix product shown below −1 −1 1 −2 1 3 x1 x3 7 2 −1 −1 −2 x2 = −5 4 1 −2 1 3 x 2 −1 −1 −2 y z R1 (A|I|b) = R3 R2 −1 2 R1 1 −2 1 3 1 0 0 7 −1 −1 0 0 1 1 −2 1 3 1 −2 0 1 0 −5 4 1 0 0 1 1 0 7 → 2R2 + R3 R1 → R2 R1 + R2 0 −1 0 0 −5 0 2 1 −6 3 1 1 0 0 1 1 0 7 2 1 −2 −1 0 R2 + R+3 4R1 + 3R3 → R3 0 4R2 + R3 0 −4 0 0 R1 − 2R2 R2 R3 R1 /4 0 −4 −4 1 3 1 −4 0 0 7 9 3 5 7 1 2 4 −8 16 −4 1 3 1 −4 0 0 −4 −3 −5 1 5 1 −5 4 −7 4 −3 4 7 3 1 4 −1 4 −1 4 4 8 4 0 → 1 0 0 1 −4 2 4 ³ ´ → −1 1 −2 1 3 x 2 −1 −1 −2 y z = −R2 /4 0 1 0 − 5 4 −R3 /4 0 0 1 −1 4 1 0 0 −3 4 −1 −1 = I|A |x 1 −x + y − 2z x − 2y + 3z 2x − y − z .8 (20 pts) (a)(15 pts) Student ID # Given the matrix equation Ax = b above. use augmented matrices and elementary row operations to simultaneously determine A−1 and x.

9 (20 pts) Rewrite the scalar equation Student ID # d4 w d3 w d2 w − 3 3 + 2 2 − w = et ln |t| dt4 dt dt as a ﬁrst order system in normal form. Express the system in matrix form x0 = Ax + f . x1 = w x2 = w 0 x3 = w00 x4 = w000 x0 = w 0 = x 2 1 x0 = w00 = x3 2 x0 = w 000 = x4 3 x0 = w iv = x1 − 2x3 + 3x4 + et ln |t| 4 x2 0 0 = x3 0 0 x1 0 0 1 0 1 0 0 x4 1 0 −2 3 + 1 0 0 0 0 et ln |t| .

10 (20 pts) Solve the initial value problem x0 = Student ID # 5 −1 3 1 x. ¯ ¯ −1 ¯ ¯ ¯ 1−r ¯ x(0) = 2 −1 |A − rI| = = = = ¯ ¯ ¯ 5−r ¯ ¯ ¯ 3 (5 − r)(1 − r) + 3 r 2 − 6r + 8 (r − 2)(r − 4) for r = 2. 4 r1 = 2 → (A − 2I)u1 = 3 −1 3 −1 1 −1 3 −3 1 3 1 1 =0 =0 r2 = 4 → (A − 4I)u2 = x(t) = c1 e2t c1 1 3 + c2 e4t 1 1 x(0) = 1 3 + c2 1 1 1 1 2 3 1 −1 → R1 3R1 − R2 2R1 − R2 R2 −3 2 7 2 = 1 1 3 1 c1 c2 = 2 −1 1 1 2 0 2 7 c1 c2 → = 2 0 −3 0 2 7 x(t) = 3 7 1 − e2t + e4t = 2 2 2 −9e2t + 7e4t 3 1 1 1 −3e2t + 7e4t .

s s>0 L {tf (t)} = − L t1/2 n o dF (s) ds d ³√ −1/2 ´ πs .Bonus (20 pts) (a)(15 pts) Use Z ∞ 0 Student ID # e −x2 dx = √ π to show that 2 L t n −1/2 o (s) = r π . s>0 (a) Letting t = x2 /s (which immediately implies that s > 0 since t ∈ [0. s>0 3/2 2s . x s s>0 s>0 = 2 √ s r Z ∞ 0 e−x dx. +∞)) gives L t−1/2 (s) = = n o N →∞ 0 lim Z N e−st t−1/2 dt −x2 N →∞ 0 lim Z √sN e 2 √ s 2x dx. ds s>0 = − = = 1 √ −3/2 πs . s>0 2 √ π . s s>0 (b)(5 pts) Use Part (a) to show that L t n 1/2 o (s) = √ π 2s3/2 . = (b) π .

tan x dx = − ln | cos x| + C.MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey MA232 EXAM FORMULAS Some Common Derivatives d n (x ) = nxn−1 . sin 2x = 2 sin x cos x sin(A + B) = sin A cos B + cos A sin B. the solution is given by 1 y(x) = µ(x) ½Z µ(x)Q(x) dx + C . Then. dx d x (x ) = xx (1 + ln x) dx d x d f (x) d f 0 (x) = ef (x) f 0 (x). for any choice of initial value y0 . Z csc x dx = ln | csc x − cot x| + C cot x dx = ln | sin x| + C Some Common Trigonometric Identities cos2 x + sin2 x = 1. Exam Formulas Clarkson University . f(x) Z Z cos x dx = sin x + C. sin x = cos x. (e ) = ex . there exists a unique solution y(x) on (a. tan x = sec2 x dx dx dx d d d sec x = sec x tan x. y(x0 ) = y0 . b) to the initial value problem dy + P (x)y = Q(x). e ln |f (x)| = dx dx dx f (x) d d d cos x = − sin x. n+1 Z Z ex dx = ex + C. dx In fact. cot x = − csc2 x dx dx dx Some Common Integrals Z xn dx = xn+1 + C. b) that contains the point x0 . cos(A + B) = cos A cos B − sin A sin B cos(A − B) = cos A cos B + sin A sin B (Existence and Uniqueness) Theorem 1 (Page 51) Suppose P (x) and Q(x) are continuous on the interval (a. csc x = − csc x cot x. cos 2x = cos2 x − sin2 x = 2 cos2 x − 1 = 1 − 2 sin2 x sin(A − B) = sin A cos B − cos A sin B. Z sin x dx = − cos x + C sec x dx = ln | sec x + tan x| + C. ¾ µ(x) = e R P (x) dx where the arbitrary constant C is determined by the initial condition y(x0 ) = y0 . Z Z f 0 (x)ef (x) dx = ef (x) + C Z f 0 (x) dx = ln |f (x)| + C.

W (x) = 1 0 0 y1 (x) y2 (x) 0 0 = y1 (x)y2 (x) − y1 (x)y2 (x) 6= 0 for all x in (a. b). ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ y (x) y2 (x) 3. b). † Pn (x) must include all its terms even if pn (x) has some terms that are zero. b). W (x0 ) 6= 0 for some x0 in (a. {y1 . y1 and y2 are linearly independent on (a. then the following are equivalent: 1. b) that contains the point x0 . where N = max(n. Fundamental Solution Sets.MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey (Existence and Uniqueness) Theorem 2 (Page 157) Suppose p(x). QN (x) = BN xN + · · · + B1 x + B0 and N = max(n. y2 } is a fundamental solution set on (a. m) xs {Aeαx cos βx + Beαx sin βx} xs eαx {PN (x) cos βx + QN (x) sin βx}. 4. Exam Formulas Clarkson University . b). 2. b) to the initial value problem d2 y dy + p(x) + q(x)y = g(x). and the Wronskian (Page 165) If y1 and y2 are solutions of y 00 + py 0 + qy = 0 on (a. Then. m) (VI) (VII) The nonegative integer s is chosen to be the smallest integer so that no term in the particular solution yp (x) is a solution of the corresponding homogeneous equation L[y](x) = 0. for any choice of the initial values y0 and y1 . y 0 (x0 ) = y1 .1 (Page 197) Method of Undetermined Coeﬃcients for L[y](x) = g(x) Type g(x) (I) pn (x) = an xn + · · · + a1 x + a0 (II) (III) (IV) (V) aeαx a cos βx + b sin βx pn (x)eαx pn (x) cos βx + qm (x) sin βx where qm (x) = bm xm + · · · + b1 x + b0 aeαx cos βx + beαx sin βx pn (x)eαx cos βx + qm (x)eαx sin βx yp(x) xs Pn (x) = xs {An xn + · · · + A1 x + A0 }† xs Aeαx xs {A cos βx + B sin βx} xs Pn (x)eαx xs {PN (x) cos βx + QN (x) sin βx}. there exists a unique solution y(x) on the interval (a. 2 dx dx y(x0 ) = y0 . q(x) and g(x) are continuous on an interval (a. Linear Independence. Table 4. b).

W [y1 .MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey Procedure for Solving Nonhomogeneous Equations y 00 + py 0 + qy = g (Page 191) (a) Determine a general solution. y2 ](x) ¯ ¯ ¯ ¯ ¯ v2 (x) = ¯ Z g(x)y1 (x) dx. y = c1 y1 + c2 y2 + yp . where v1 (x) = Here.1 A Brief Table of Laplace Transforms (Page 362) f (t) F (s) = L{f (t)} (s) 1 1 . to obtain a general solution of the given nonhomogeneous equation. then a particular solution yp of its nonhomogeneous counterpart y 00 + p(x)y 0 + q(x)y = g(x) is given by yp (x) = v1 (x)y1 (x) + v2 (x)y2 (x). s>a (s − a)2 + b2 Exam Formulas Clarkson University . . (b) Find a particular solution. s>0 2 + b2 s s cos bt . . s>0 2 + b2 s b eat sin bt . . a general solution of the nonhomogeneous diﬀerential equation is given by y(x) = c1 y1 (x) + c2 y2 (x) + yp (x) the ﬁrst two terms of which comprise the general solution of the homogeneous equation. Table 7. yp . Having determined yp . the Wronskian Z −g(x)y2 (x) dx. c1 y1 + c2 y2 . s>0 n+1 s n! eat tn . s>a (s − a)2 + b2 s−a eat cos bt . Variation of Parameters (Page 204) If {y1 . (c) Form the sum of the particular solution and a general solution to the homogeneous equation. 0 0 y1 y2 ¯ being nonzero is a property satisﬁed by fundamental solutions. of the given nonhomogeneous equation. 2. n = 1. n = 1. s>a s−a n! tn . y2 ](x) y y ¯ ¯ 0 0 W [y1 . y2 } is a fundamental solution set for the homogeneous second order not necessarily constant coeﬃcient diﬀerential equation y 00 + p(x)y 0 + q(x)y = 0. W [y1 . that is. s>a (s − a)n+1 b sin bt . s>0 s 1 eat . y2 ](x) = 1 2 ¯ = y1 y2 − y1 y2 . . of the corresponding homogeneous equation y 00 + py 0 + qy = 0. 2. . . . .

2 Properties of Laplace Transforms (Page 368) L{f + g} = L{f } + L{g} L{cf } = cL{f} n o for any constant c L eat f (t) (s) = F (s − a) where F (s) = L{f (t)} (s) L{f 0 } (s) = sF (s) − f (0) L{f 00 } (s) = s2 F (s) − sf(0) − f 0 (0) L f (n) s = sn F (s) − sn−1 f (0) − sn−2 f 0 (0) − · · · − f (n−1) (0) L{tn f(t)} (s) = (−1)n d F (s) ds n n n o Properties of the Inverse Laplace Transform (Pages 372 & 379) L−1 {F1 + F2 } = L−1 {F1 } + L−1 {F2 } L−1 {cF } = cL−1 {F } L Miscellaneous −1 ( n ) dF ds n (t) = (−t)n f (t) Z u dv = uv − x a d tan−1 = 2 dx a a + x2 Z 1 dx bx = tan−1 a2 + b2 x2 ab a Z ax e eax cos bx dx = 2 (a cos bx + b sin bx) a + b2 Z eax ax e sin bx dx = 2 (a sin bx − b cos bx) a + b2 Z v du (Integration by Parts) Exam Formulas Clarkson University .MA 232 Diﬀerential Equations Fall 1999 Kevin Dempsey Table 7.

- s8_partdifferentiation
- Ode
- L Transform
- 01 Introduction - Copy
- Inverse Laplace Transform of a Constant
- Notes3-5v2
- Variational Method
- Forms
- Derivative of Polinomials
- Mathematical Modeling - 03.pdf
- homework 4.pdf
- Trigonometry Calculator
- Euler Formula
- Complex Algebra
- Applications of Differentiation Handout
- f08w05 Ode Good Msc
- Vectors Tutorial
- week3
- Functional Equations Problems With Solutions
- Classical Monopole
- Spring 2013 MCA
- "Equilibrium Problems and Circuit Analysis"
- MATH219 Lecture 5
- Basic Numeric s 2
- Mark Rusi Bol
- Distance Notes
- 2008 Al-Khatib-Grysa-M-2-08
- Chapter2-Mathematics of Quantum Mechanics
- diffforms
- Differential Forms

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue listening from where you left off, or restart the preview.

scribd