# CHAPTER

1

First-Order Differential Equations

1.1

Dynamical Systems: Modeling

Constants of Proportionality 1. 3. 5.

dA = kA dt

(k < 0)

2. 4.

dA = kA dt

(k < 0)

dP = kP(20,000 − P ) dt

dA kA = dt t

dG kN = dt A
A Walking Model

6.

Because d = υt where d = distance traveled, υ = average velocity, and t = time elapsed, we have d the model for the time elapsed as simply the equation t = . Now, if we measure the distance

υ

traveled as 1 mile and the average velocity as 3 miles/hour, then our model predicts the time to be d 1 t = = hr , or 20 minutes. If it actually takes 20 minutes to walk to the store, the model is υ 3

perfectly accurate. This model is so simple we generally don’t even think of it as a model.
A Falling Model 7.

(a)

Galileo has given us the model for the distance s(t ) a ball falls in a vacuum as a function of time t: On the surface of the earth the acceleration of the ball is a constant, so d 2s = g , where g ≈ 32.2 ft sec2 . Integrating twice and using the conditions s(0) = 0 , dt 2 ds(0) = 0 , we find dt 1 1 s(t ) = gt 2 s(t ) = gt 2 . 2 2

1

2

CHAPTER 1

First-Order Differential Equations

(b)

We find the time it takes for the ball to fall 100 feet by solving for t the equation 1 100 = gt 2 = 16.1t 2 , which gives t = 2.49 seconds. (We use 3 significant digits in the 2 answer because g is also given to 3 significant digits.)

(c)

If the observed time it takes for a ball to fall 100 feet is 2.6 seconds, but the model predicts 2.49 seconds, the first thing that might come to mind is the fact that Galileo’s model assumes the ball is falling in a vacuum, so some of the difference might be due to air friction.

The Malthus Rate Constant k 8.

(a)

Replacing
e0.03 ≈ 1.03045

(b)

in Equation (3) gives
y = 0.9(103045)t , .

which increases roughly 3% per year.

y 10 9 8 7 6 5 4 3 2 1

Malthus

World population t

1800 1820 1840 1860 1880

(c)

Clearly, Malthus’ rate estimate was far too high. The world population indeed rises, as does the exponential function, but at a far slower rate. If y(t ) = 0.9ert , you might try solving y(200) = 0.9e200r = 6.0 for r. Then

200r = ln so r≈ which is less than 1%.
Population Update 9.

6 ≈ 1897 . 0.9

1897 . ≈ 0.0095, 200

(a)

If we assume the world’s population in billions is currently following the unrestricted growth curve at a rate of 1.7% and start with the UN figure for 2000, then
y0 e kt = 6.056e0.017 t ,

SECTION 1.1

Dynamical Systems: Modeling

3

and the population in the years 2010 (t = 10) , 2020 (t = 20) , and 2030 (t = 30) , would be, respectively, the values 6.056e 6.056e 6.056e
0.017 (10 )

= 7.176 ≈ 8.509 ≈ 10.083.

0.017 ( 20 ) 0.017 ( 30 )

These values increasingly exceed the United Nations predictions so the U.N. is assuming a growth rate less than 1.7%. (b) 2010: 6.056e10 r = 6.843 6.843 = 1.13 6.056 10r = ln(1.13) = 0.1222 e10 r = r = 1.2% 2020: 6843e10 r = 7568 7.578 = 1.107 6.843 10r = ln(1.107) = 0.102 e10 r = r = 1.0% 2030: 7.578e10 r = 8.199 8.199 = 1.082 7.578 10r = ln(1.082) = 0.079 e10 r = r = 0.8%

The Malthus Model 10.

(a)

Malthus thought the human population was increasing exponentially e kt , whereas the food supply increases arithmetically according to a linear function a + bt . This means the ekt , which although not a number of people per food supply would be in the ratio (a + bt ) pure exponential function, is concave up. This means that the rate of increase in the number of persons per the amount of food is increasing.

(b)

The model cannot last forever since its population approaches infinity; reality would produce some limitation. The exponential model does not take under consideration starvation, wars, diseases, and other influences that slow growth.

Student Project . A nonlinear model that approaches some finite upper limit would be more appropriate. e.9) = 0. y4 = (103)4 (0. Discrete-Time Malthus 11. (c) The discrete model will always give a value lower than the continuous model. y3 = (103)3(0.927 .9 billion).9) = 0.9 (0. 1801.956 . (a) Taking the 1798 population as y0 = 0.9) = 0. . when we study compound interest. 1800. Later. y2 = (103)2 (0. respectively y1 = 103(0. From the equation y ′ = y(k − cy) . but reaches a steady state ( y ′ = 0) when y = . hence y192 = (103)192 (0.9) ≈ 262 (262 billion).4 CHAPTER 1 First-Order Differential Equations (c) A linear growth model for food supply will increase supply without bound and fails to account for technological innovations. There is no reason to suspect the two values would be the same and so a model like this would seem to be promising if we only knew their values. . when you get sick a bacteria might multiply exponentially until your body’s defenses come into action or you receive appropriate medication. we have the population in the years 1799. (d) An exponential model is sometimes reasonable with simple populations over short periods of time. Verhulst Model 12. pesticides and genetic engineering. . such as mechanization. The constant k affects the initial growth of the population whereas the constant c dt controls the damping of the population for larger y.983 . dy = y(k − cy) .g. we see that for small y the population closely obeys k y ′ = ky . and 1802.. you will learn the exact relationship between discrete compounding (as in the discrete-time Malthus model) and continuous compounding (as described by y ′ = ky ).9) = 1023. c Suggested Journal Entry 13. (b) In 1990 we have t = 192 .

Substituting y = 3t + t 2 y ′ = 3 + 2t 1 into y ′ = y + t yields the identity t 3 + 2t ≡ 1 3t + t 2 + t .2 Solutions and Direction Fields Verification 1. using the product rule and the fundamental theorem of 2 calculus. 2. t a f 3.2 Solutions and Direction Fields 5 1. If y = 2 tan 2t . then. 2 which is 1 as the differential equation requires.SECTION 1. we have y ′ = e2t e−2t + 4te2t 2 2 2 z t 0 e−2 s ds = 1 + 4te2t 2 2 z t 0 e−2 s ds . Substituting y = t 2 ln t y ′ = 2t ln t + t into y ′ = 2 y + t yields the identity t 2t ln t + t ≡ 2 2 t ln t + t . t b g 4. 2 Substituting y ′ and y into y ′ − 4ty yields 1 + 4te2t 2 z t 0 e−2 s ds − 4te2t 2 2 z t 0 e−2 s ds . . Substituting y ′ and y into y ′ = y 2 + 4 yields a trigonometric identity 4 sec2 (2t ) ≡ 4 tan2 (2t ) + 4 . If y = z t −2 s2 − t 2 e 0 b gds = e2t 2 z t 0 e−2 s ds . then y ′ = 4 sec2 2t .

. IJ K which is equal to e−t as the differential equation requires.6 CHAPTER 1 First-Order Differential Equations IVPs 5. 2 2 d i 8. Here 1 −t e − e−3t 2 1 y ′ = − e−t + 3e−3t . If y(0) = 2 . we get the 2 2 2 identity 2ctet ≡ 2t cet . then we have ce0 ≡ c = 2 . We have y = et cos t + cet y ′ = et cos t − et sin t + cet and substituting y and y ′ into y ′ − y yields be cos t − e sin t + ce g − be cos t + ce g . then −1 = e0 cos 0 + ce0 yields c = −2 . and so the initial condition is also satisfied. 2 y= Substituting into the differential equation y ′ + 3y = e−t we get FG − 1 e H 2 −t + 3e−3t + 3 IJ FG 1 e K H2 −t − e −3 t . Another direct substitution Applying Initial Conditions 7. 2 6. It is also a simple matter to see that 1 y(0) = − . If y(0) = −1. t t t t t which is −et sin t . then we have y ′ = 2ctet and if we substitute y and y ′ into y ′ = 2ty . If y = cet .

SECTION 1.. we see it satisfies the DE y ′ = t − y because 1 ≡ t − (t − 1) . Linear Solution 12. y′ = t − y y 2 t –2 2 –2 Solutions are y = t − 1 + ce−t . Solutions are y = c − t 2 . . y′ = − t y y 2 t –2 2 –2 2 t –2 –2 Solutions are y = ce2t . It appears from the direction field that there is a straight-line solution passing through (0. y′ = 2 y y 2 10.2 Solutions and Direction Fields 7 Using the Direction Field 9. Computing y ′ = 1. –1) with slope 1. 11. i. the line y = t − 1.e.

y ′ = (−4)(−3) = 12 . y= − 1 . y = 1. y ′ = y ( y + 1) = 0 When y = 0. 2 y ′ = 3(4) = 12 y ′ = 1(2) = 2 1 ⎛ 1 ⎞⎛ 1 ⎞ y ′ = ⎜ − ⎟⎜ ⎟ = − 4 ⎝ 2 ⎠⎝ 2 ⎠ y ′ = (−2)(−1) = 2 y = −2. For y = 3. slopes are negative. slopes are positive. the direction field shows a stable equilibrium solution. For y > 1. y = 4. y′ = 1 − y = 0 When y = 1.8 CHAPTER 1 First-Order Differential Equations Stability 13. and when y = −1. a stable equilibrium solution exists. an unstable equilibrium solution exists. for y < 1. 14.

20. Because it is undefined when t = 0 and the directional field has slopes that are independent of y. All slopes become steeper as they are found further from the vertical axis. 21. (C) (D) (F) Because the slope is always the same Because the slope is always the value of y Because F is the only direction field that has vertical slopes when t = 0 and zero slopes when y = 0 Because it is the only direction field that has all zero slopes when t = 0 The slope is always positive and equal to the square of the distance from the origin. Match Game 16. (B) (E) (A) . y ′ = t 2 (1 − y 2 ) Two equilibrium solutions: y = 1 is stable y = −1 is unstable Between the equilibria the slopes (positive) are shallower as they are located further from the horizontal axis. 18.2 Solutions and Direction Fields 9 15. 17. with the same sign as that of t 19.SECTION 1. Outside the equilibria the slopes are negative and become steeper as they are found further from the horizontal axis.

2) Horizontal axis is locus of inflection points. Equilibrium solutions occur when y = 2 (unstable) or when y = −2 (stable). shaded regions are where solutions are concave down. and y ∈ (0.10 CHAPTER 1 First-Order Differential Equations Concavity 22. Solutions are concave up above the parabola of inflection points. concave down below. y′ = y + t 2 y ′′ = y ′ + 2t = y + t 2 + 2t = 0 When y = −t 2 − 2t . concave down for y < −2. 23. y′ = y 2 − 4 y ′′ = 2 yy′ = 2 y ( y + 2)( y − 2) When y = 0. so we have a locus of inflection points. y′′ = 0. Parabola is locus of inflection points. shaded regions are where solutions are concave down. we find inflection points for solutions. and y ∈ (−2.0) . Solutions are concave up for y > 2. .

if you look far enough. The locus of inflection points has two branches: Above the upper branch. and solutions will be horizontal translates. Asymptotes 25. and to the right of the lower branch. Below the upper branch but outside the lower branch. it’s a good idea to check how fast. y′ = y 2 Because y ′ depends only on y. If the y-axis extends high enough. . isoclines will be horizontal lines. shaded regions are where solutions are concave down. y′ = y 2 − t y ′′ = 2 yy′ − 1 = 2 y 3 − 2 yt − 1 = 0 When t= 2 y3 − 1 1 = y2 − . Bold curves are the locus of inflection points.SECTION 1. solutions are concave up. you may suspect (correctly) that undefined solutions will each have a (different) vertical asymptote. solutions are concave down. Slopes get steeper ever more quickly as distance from the x-axis increases.2 Solutions and Direction Fields 11 24. The direction field will give good intuition. Compare with y ′ = y for a case where the solutions do not have asymptotes. then y ′′ = 0 2y 2y and we have a locus of inflection points. When slopes are increasing quickly.

y′ = t 2 There are no asymptotes. as solutions approach or depart from the horizontal axis. Every solution has a vertical asymptote when it is close to the horizontal axis.12 CHAPTER 1 First-Order Differential Equations 26. so solutions do not cross either axis. 27. y′ = 1 ty The DE is undefined for t = 0 or y = 0. However. but they do not actually approach vertical for any finite value of t. No asymptote . they asymptotically approach a vertical slope. As t → ∞ (or t → −∞) slopes get steeper and steeper.

y = 0 is a solution to the DE. 1. and the other solutions do not cross the horizontal axis for t ≠ ±1. you can verify that this line satisfies the DE. The direction field shows that as y → 0 from either above or below.5. moving down then up on the right. 1 at t = . Oblique Asymptote 29. y ′ = −2ty + t Here we have a horizontal asymptote. The equation of this asymptote can be at least approximately read off the graphs as y = −2t − 2. y ′ = 2t + y Solutions to this DE have an oblique asymptote– they all curve away from it as t → ∞. y′ = ty t −1 2 At t = 1 and t = −1 the DE is undefined. In fact. However.) Vertical asymptotes for t → 1 or t → −1 .2 Solutions and Direction Fields 13 28.SECTION 1. so this asymptote is also a solution. (See Picard’s Theorem Sec. simply down on the left. 2 Horizontal asymptote 30. solutions asymptotically approach vertical slope.

In other words the isoclines where the slope is 4 are y = ±2 . ±2 shown in the figure. the isocline is the hyperbola y = − . t ≠ 0 or t hyperbolas in the ty plane. 34. y′ = t . t 1 For c = −1 . The isoclines for c = −1. 0. t –2 33. as follows for c = 0. The isoclines where the slope is c > 0 are the horizontal lines y = ± c ≥ 0 . y′ = − y . 2 y slopes 4 slopes 2 slopes 0 Here the slope of the solution is always ≥ 0. ±1. we see that the points where the c slope is c are along the curve y = − . 2. –2 t 2 –2 slopes 2 slopes 4 y ′ = −ty . The isoclines for c = 0. t –2 slopes –1 –2 2 slopes 1 y slopes –1 slopes 0 2 slopes 1 t .14 CHAPTER 1 First-Order Differential Equations Isoclines 31. 2 y slopes –1 slopes 0 –2 slopes 1 2 Here the slope of the solution is negative when y > 0 and positive for y < 0. The isoclines are vertical lines t = c . 1 are shown in the figure. –2 2 y t 2 –2 32. y′ = y2 . and 4 are shown in the figure. 1 For c = 1 . Setting −ty = c. the isocline is the hyperbola y = .

0. … −1 when y = π. y ′ = 2t − y .2 Solutions and Direction Fields 15 When t = 0 the slope is zero for any y. and y = 0 is in fact a solution. are shown from left to right (see figure). y ′ = y 2 − t . –2 y 2 slopes 2 slopes 0 slopes –2 t 2 –2 37. The isocline where y ′ = c is the straight line y = 2t − c . The isocline where y ′ = c is a parabola that opens to the right. … Additional observations: y ′ ≤ 1 for all y. when y = 0 the slope is zero for any t. Three isoclines. act as horizontal asymptotes. The isoclines with slopes c = −4 . 0. with slopes c = 2. –2. 4 are shown from left to right (see figure). this information produces a slope field in which the constant solutions. y ′ = cos y 0 when y = odd multiples of y′ = c = 1 when y = 0. 4π. at y = (2n + 1) π 2 . 35. See figure for the direction field for this equation with isoclines for c = 0. –2. 2π. –2 y 2 t 2 –2 36. . 3π. 2.SECTION 1. When y = π 2 π 4 . ±1.

do not cross. y ′ = cos( y − t ) 0 when y − t = − π π 3π . 2π.... … 3π π 3π 1 when t = − . … π 2 or y = t ± 2nπ −1 when y − t = −π. which you can verify as y = −cost. .5). .16 CHAPTER 1 First-Order Differential Equations 38... π. 2 2 2 π 3π −1 when t = − .. . 3π. with different y-intercepts. 2 2 2 or y = t ± (2n + 1) y′ = c = 1 when y − t = 0. .. . 39. π.. by uniqueness. . … or y = t ± (2n + 1)π All these isoclines (dashed) have slope 1. 2π.. See Section 1. The isoclines for solution slopes 1 are also solutions to the DE and act as oblique asymptotes for the other solutions between them (which. 2 2 The direction field indicates oscillatory periodic solutions. y ′ = sin t y′ = c = 0 when t = 0.

All solutions are periodic oscillations. then y′ = 1 . Zooming out y ′ = 2 − sin t If t = nπ. Although there is a periodic pattern to the direction field. See Figures.SECTION 1. the oscillations of the solutions look somewhat more regular. but are always moving upward.. . the solutions are quite irregular and not periodic. If you zoom out far enough. .. . 2 2 2 All slopes are between 1 and 3. 3π π 5π If t = − ... then y′ = 2. y ′ = cos10t ⎛π ⎞ 0 when 10t = ± (2n + 1) ⎜ ⎟ ⎝2⎠ 1 when 10t = ±2nπ y′ = c = −1 when 10t = ±(2n + 1)π y′ is always between +1 and −1.2 Solutions and Direction Fields 17 Periodicity 40. 10 Zooming in 41. with period 2π .

Zooming further out y ′ = − cos y . we observe that between every adjacent pair of constant solutions. and solutions between the constant solutions cannot cross them. 2 For y = ±2nπ. by uniqueness. To further check what happens in these cases we have added an isocline at y = ⎛π ⎞ y ′ = cos ⎜ ⎟ ≈ −0. the solutions are horizontal translates. Slope y′ is always between −1 and 1. y′ = −1 If y = ± (2n + 1) π For y = ±(2n + 1)π. y ′ = 1. then y ′ = 0 and these horizontal lines are equilibrium solutions. in the vertical direction with period 2π. . Furthermore. ⎝4⎠ π 4 . but there is a periodicity to the direction field.7. where Solutions are not periodic.18 CHAPTER 1 First-Order Differential Equations Zooming out 42.

Hence they are not strictly periodic.2 For 10t = ±(2n + 1) π 2 nπ 10 y′ = 0. . y ′ = cos10t + 0. but at the same time they move ever upward.157 ± For 10t = ±2nπ.314 ± Or 10t = ±(1.2. with rough sketch solution. Direction field and solutions over a larger scale.77 + 2nπ) The solutions oscillate in a periodic fashion.2. t ≈ 0.8. Direction field (augmented and improved in lower half). t ≈ 0. y′ = 1.SECTION 1.2 Solutions and Direction Fields 19 43. Compare with Problem 40. t ≈ ± For 10t = ±(2n + 1)π 2nπ 10 2nπ 10 To get y′ = 0 we must have cos 10t = −0.2 y′ = −0.

For y > 0 the solutions wiggle downward but never cross the horizontal axis—they get sent upward a bit first. The solutions are not periodic. The solutions are not periodic. 45. though there is a periodic (and diagonal) pattern to the overall direction field. Slopes are positive in the regions trapped by the two isoclines. despite the periodic function in the DE. you will see it goes downward a lot more of the time than upward. For y < 0 solutions eventually get out of the upward-flinging regions and go forever downward.20 CHAPTER 1 First-Order Differential Equations 44. y ′ = cos( y − t ) See Problem #39 for the direction field and sample solutions. If you try to sketch a solution through this configuration. y ′ = y (cos t − y ) Slopes are 0 whenever y = cos t or y = 0 Slopes are negative outside of both these isoclines. .

then y′ = 0 .. Solutions are periodic. slopes will not remain between ±1. There are additional isoclines of zero slope where sin 2t = −cos t. 2sin t cos t i.3 The figures on the next page are crucial to seeing what is going on... 4 4 7π 15π For t = .7. y′ ≈ 1 − 0. ..SECTION 1. 4 4 π 9π y′ ≈ 1 + 0..g. where sin t = − 1 and 2 t= − 5π π 7π 11π .7 = 0. Adding these isoclines and slopes shows there are more wiggles in the solutions. For t = 4 4 5π 13π For t = . ..7 = 1. 4 4 3π 11π . then y ′ = −1. then y ′ = 0. 2 If t = ±(2n + 1)π. .3 y′ ≈ −1 + 0. .. with period 2π.. From this information it seems likely that solutions will oscillate with period 2π. If t = ±(2n + 1) . .. π 2 .7 = −1. For t = . See figures on next page.e..7.. these are . But beware—this is not the whole story...2 Solutions and Direction Fields 21 46. y ′ = sin 2t + cos t If t = ±2nπ. π Isoclines are vertical lines. 6 6 6 6 There is a symmetry to the slope marks about every vertical line where t = ±(2n + 1) some of the isoclines of zero slope. y′ ≈ −1 − 0. e....− . rather like Problem 40.. and solutions are vertical translates. .7 = −0. For y ′ = sin 2t + cos t....

continued) Direction field. Direction field and solutions by computer. .22 CHAPTER 1 First-Order Differential Equations (46. sketched with ever increasing detail as you move down the graph.

The only symmetry visible in the direction field is point symmetry. but the resulting direction field does not have visual symmetry. but the resulting picture is not. y′ = y 2 Note that y′ depends only on y.2 Solutions and Direction Fields 23 Symmetry 47. . Positive and negative values of t give the same slope. so isoclines are horizontal lines. The figures are given with Problem 25 solutions. y′ = t 2 Note that y′ depends only on t. 48. so isoclines are vertical lines. so the slope values are repeated symmetrically across the vertical axis. Positive and negative values of y give the same slopes. Hence the slope values are symmetric about the horizontal axis. about the origin (or any point on the t-axis).SECTION 1. The only symmetry visible in the direction field is point symmetry through the origin (or any point on the y-axis).

slopes are negative. slopes are positive. For t > 0. the direction field is reflected across the horizontal axis. For y < 0. The result is pictorial symmetry of the vector field about the vertical axis. . y ′ = −t Note that y′ depends only on t. For y > 0. y′ = − y Note that y′ depends only on y. so isoclines are vertical lines. slopes are positive.24 CHAPTER 1 First-Order Differential Equations 49. As a result. 50. For t < 0. slopes are negative. so isoclines are horizontal lines.

y2 . where the DE is not defined. Note: Across the horizontal axis. 9 The direction field has point symmetry through the point (−1. repeat them about the t-axis. Slopes are always positive. It is sufficient therefore to calculate slopes for the first quadrant only. t 4 If y = ±2 y′ = . slope is . If y = 0. 1 If y = ±1.2 Solutions and Direction Fields 25 51. t about the vertical axis. this fact does not give symmetry to the direction field or solutions. the result is a pictorial symmetry However because the sign of t gives the sign of the slope. so they will be repeated.SECTION 1. y′ = y2 t y2 . that is. so you can plot them for a single t Positive and negative values for y give the same slopes. If t = 0 or −2. 4 1 If t = 2 or −4. or any point on the line t = −1. slope is . 0). y′ = 1 (t + 1) 2 Note that y ′ depends only on t. 1 If t = 1 or −3. 52. t . y′ = 0. not reflected. slope is 1. across t = −1. y ′ = . See figures on the next page. reflect them about the y-axis. positive y-value and then repeat them for the negative of that y-value. so isoclines will be vertical lines.

are undefined. Indeed. This observation is confirmed by the sign of y ′′ = 1 + y ′ = 1 + t + y.26 CHAPTER 1 First-Order Differential Equations Second-Order Equations 53. y′ = t + y (a) (b) (c) There are no constant solutions. those below are concave down. which is not constant. There are no points where the DE. y = −t − 1 satisfies the DE. Solving these equations. so y = −e−2t + 3e−t . or its solutions. In shaded region. . gives A = −1. We see one straight line solution that appears to have slope m = −1 and y-intercept b = −1. and y ′′ into the differential equation reduces it to an identity. (a) (b) (c) (d) Direct substitution of y. Long-Term Behavior 54. (d) All solutions above y = −t − 1 are concave up. solutions are concave down. zero slope requires y = −t. y ′ . B = 3. Direct computation Direct computation Substituting y(t ) = Ae2t + Be−t y ′(t ) = 2 Ae2t − Be−t into the initial conditions gives y(0) = A + B = 2 y ′(0) = 2 A − B = −5.

so that y+t In shaded region. There are no periodic solutions. solutions above y = −t − 1 approach ∞. all solutions approach y = −t. y′ = y −t y+t (a) (b) (c) (d) There are no constant solutions. so solutions are concave up for y < −t Hence y ′′ is (e) (f) (g) As t → ∞. The DE is undefined along y = −t. (t 2 + y 2 ) . all solutions are seen to emanate from ∞.SECTION 1. All solutions become more vertical (at both ends) as they approach y = −t. those below approach −∞. ( y + t )( y ′ − 1) − ( y − t )( y ′ + 1) y ′′ = ( y + t )2 −2t y −t − y −t Simplify using y ′ − 1 = y+t 2y and y ′ + 1 = y ′′ = −2 y −t + y +t . which is oblique. but solutions will have zero slope along y = t. 55. solutions are concave down. ( y + t )3 Never zero < 0 for y + t > 0. appears if we go backward in time—then all solutions are ever closer to y = −t − 1.2 Solutions and Direction Fields 27 (e) (f) (g) As t → ∞. As t → −∞. As t → −∞. going backward in time. There are no straight line solutions. we see that all solutions emanate from y = −t. . There are no periodic solutions. The only asymptote. so solutions are concave down for y > −t > 0 for y + t < 0.

nor even any point with zero slope. y′ = (a) (b) (c) . y′ = 1 ty (a) There are no constant solutions. 1 From y ′ = . ( ) (e) (f) (g) so there are no inflection points. solutions are concave down. The DE is undefined along y = t. There appears to be one straight line solution with slope 1 and y-intercept −1. solutions asymptotically approach vertical slopes as y → 0. t y (b) (c) (d) In shaded region. those in lower quadrant emanate from −∞. y′ = 1 when y = t − 1. As t → ∞. so solutions will not cross either axis. In the left and right half plane. we get ty −1 1 y ′′ = 2 y′ − 2 . concave up below the t-axis. Straight line solution In shaded region.28 CHAPTER 1 First-Order Differential Equations 56. ty t y This simplifies to 1 y ′′ = − 2 3 1 + y 2 . we see that solutions in upper quadrant emanate from +∞. because is never zero. There are no straight line solutions. ty The DE is undefined for t = 0 or for y = 0. indeed y = t − 1 satisfies the DE. which is never zero. Solutions will be concave down above the t-axis. solutions are concave down. 57. solutions in upper quadrant →∞ solutions in the lower quadrant →−∞ As t → −∞. or even 1 zero slopes. 1 t−y There are no constant solutions. There are no periodic solutions.

so we calculate (2t − y ′) y ′′ = − 2 = 0 when y ′ = 2t (t − y ) 2 1 y′ = 2 = 2t From DE when t −y 1 y = t 2 − . solutions are increasing. (e) (f) (g) As t → ∞. slopes → 0 and solutions → horizontal asymptotes. drawn as a thicker dashed 2t curve with two branches. 58. y ′ > 0 . As solutions approach y = t2. so solutions will not cross this locus. (f) As t → −∞. (g) In backwards time the line y = t − 1 is an oblique asymptote.SECTION 1. concave down for solutions below the left branch of y ′′ = 0 .2 Solutions and Direction Fields 29 (d) y ′′ = − (1 − y ′) y − (t − 1) = (t − y ) 2 (t − y ) 3 y ′′ > 0 when y > t − 1 and y< t y ′′ < 0 when y < t − 1 and y > t ⎫ ⎬ y > t − 1 and y > t ⎭ (e) As t → ∞. Solutions concave up Solutions concave down solutions below y = t − 1 approach ∞. solutions are concave down. In shaded region. . y′ = 1 t −y 2 (a) (b) (c) (d) There are no constant solutions. but locus of inflection points There are no periodic solutions. and concave down below the right branch of y ′′ = 0. We see no straight line solutions. solutions are seen to emanate from horizontal asymptotes. solutions below y = t emanate from −∞. The DE is undefined along the parabola y = t2. The DE is undefined on the boundary of the parabola. their slopes approach vertical. Outside the parabola y < t 2 . solutions above y = t − 1 approach y = t ever more vertically. There are no periodic solutions. solutions above y = t emanate from ∞. The dark curves are not solutions. We see inflection points and changes in concavity. As t → −∞. so y ′ < 0 and solutions are decreasing. Inside the parabola y > t 2 .

However in the first quadrant solutions above the parabola where y ′′ = 0 fly up toward +∞. The horizontal axis is not a solution. solutions do not cross the y-axis. In the right half plane. (e) As t → ∞. where t = 0. above the lower half of the parabola where y ′′ = 0 . In the left half plane solutions emanate from ∞. solutions seem to emanate from the upper y-intercept of the parabola. The DE is not defined for t = 0. In shaded region. But the DE is not defined there. . solutions are concave down. so there is no straight line solution. and in backward time for solutions in the lower right half plane. 2 t t that is. below the parabola they emanate from −∞. (f) (g) The negative y-axis seems to be an asymptote for solutions in the left-half-plane.30 CHAPTER 1 First-Order Differential Equations 59. There are no periodic solutions. The only straight path in the direction field is along the y-axis. most solutions approach −∞. The parabola is composed of two solutions that act as a separator for behaviors of all the other solutions. (d) Concavity changes when 2 yy ′t − y 2 y y ′′ = = 2 (2 y 2 − y − 2t ) = 0. when y = 0 or along the parabola 2 1⎞ ⎛ 1⎞ ⎛ t− ⎟=⎜y− ⎟ ⎜ 4⎠ ⎝ 16 ⎠ ⎝ (obtained by solving the second factor of y ′′ for t and completing the square). y′ = y2 −1 t (a) (b) (c) There are no constant solutions. just a locus of inflection points.

hence the constant solutions are y(t ) ≡ 0. Notice from the direction field or from the sign of the derivative that solutions starting at 0 or 1 remain at those values. We find the constant solutions by setting y ′ = 0 and solving for y. (i) y′ = y2 y 2 –2 semistable equilibrium t 2 –2 .2 Solutions and Direction Fields 31 Logistic Population Model 60. and solutions starting between 0 and 1 increase asymptotically to 1. (a) Autonomous: #9 #13 #14 #16 #17 #32 #33 #37 The others are nonautonomous. solutions starting larger than 1 decrease to 1 asymptotically.SECTION 1. The following figure shows the direction field of y ′ = y(1 − y) and some sample solutions. This gives ky(1 − y) = 0 . Autonomy y 2 1 stable equilibrium 0 0 unstable equilibrium 3 t Logistic model 61. 1. Comparison 62. y′ = 2 y y′ = 1 − y y ′ = y ( y + 1) y′ = 1 y′ = y y′ = − y y′ = y 2 y ′ = cos y (b) Isoclines for autonomous equations consist of horizontal lines.

. 1− t which blows up at t = 1. y(0) = 1. (b) Because y(t ) = 1 is a solution of the initial-value problem y ′ = y 2 . all with y(0) = 1. We then know that the solution of y ′ = y 2 + 1. y ′ = y 2 + 1. we will find out where the solution blows up. y(0) = 1 will blow up (approach infinity) somewhere between 0 and 1. –2 2 –2 (a) For y > 0 we have y 2 < y 2 + 1 < ( y + 1)2 . the solution of y ′ = ( y + 1)2 will be the largest and the solution of y ′ = y 2 will be the smallest. but these equilibria occur at different levels. For the three equations y ′ = y 2 . Equation (c) t has no equilibrium at all.32 CHAPTER 1 First-Order Differential Equations (ii) y ′ = ( y + 1)2 y 2 –2 t 2 semistable equilibrium –2 (iii) y′ = y2 + 1 y 2 Equations (a) and (b) each have a constant solution that is unstable for higher values and stable for lower y values. When we solve this problem later using the method of separation of variables. and y ′ = ( y + 1)2 . All three DEs are autonomous. so within each graph solutions from left to right are always horizontal translates.

0) . 1. –1 2 t . and y(t ) ≡ 0 has a basin attraction of the single value {0}.2 Solutions and Direction Fields 33 Coloring Basins 63. –3 t 3 We can therefore. 2) . we conclude the constant solution y(t ) ≡ 1 has a basin of attraction of (0.SECTION 1. 2) . giving y(t ) ≡ 0. For y > 2. 2 have the following basins of attraction: y(t ) ≡ 0 has the single point {0} basin of attraction. The constant solutions are found by y 2 setting y ′ = 0. we have y ′ > 0. for −2 < y < 2 we have y ′ < 0. the solutions approach –∞. hence solutions with initial conditions in this range decrease. –1 t 4 y ′ = y 2 − 4. conclude that the constant solution y = 2 has a basin of attraction of the single value {2}. whereas the constant solution y = −2 has the basin of attraction of (−∞. (2. (0. we have y ′ > 0. 3 derivative in each of the intervals (−∞. y ′ = y(1 − y) . When the solutions have negative initial conditions. conclude solutions with initial conditions greater than 2 increase. we conclude that the constant solutions y(t ) ≡ 0. hence solutions with initial conditions in this interval increase. therefore. 1. ∞) . (1. 2) y ′ = y( y − 1)( y − 2) . The constant solutions are the (real) y 3 roots of y 2 − 4 = 0 . and for y < 0. 64. 1) . y(t ) ≡ 1 has the basin of attraction (0. ∞) . and y(t ) ≡ 2 has the single value {2} basin of attraction. We. or y = ±2 . Analyzing the sign of the y 65. Either by looking at the direction field or by analyzing the sign of the derivative.

0 2 t Computer or Calculator The student can refer to Problems 69–73 as examples when working Problems 67. There are no constant or periodic solutions to this equation. In backwards time most solutions approach –2 t –2 2 the top part of this parabola. Student Project 2 68. y ′ = 2 y + t . y ′ = y 2 + t . y ′ = (1 − y )2 . we conclude the constant solution y(t ) ≡ 1 has basin of attraction the interval (−∞. This estimate is fairly accurate be1 cause y(t ) = sin 2t + c has period π. 2 t –2 2 –2 . The direction field indicates that the y 2 equation has periodic solutions with the period roughly 3. y′ = y . –2 unstable equilibrium 2 t –2 70. which is unstable (see figure). 71. In backwards time most solutions approach the top part of this parabola. 69. and for positive t solutions move away from zero slope. For negative t solutions approach zero slope. Because the derivative y ′ is always 2 zero or positive. 68. We see that eventually all solutions y 2 approach plus infinity. The direction field shows one constant 2 solution y(t ) ≡ 0 . and 74. 1 . You might also note that the isocline y 2 + t = 0 is a parabola sitting on its side for t < 0 .34 CHAPTER 1 First-Order Differential Equations y 66. 67. Student Project y y ′ = ty . y ′ = cos2t .

± π . … are constant solutions with 0. ± 2 π . Student Project . y ′ = sin(ty) . The solutions between the equilibria have positive or negative slopes depending on the y interval. We can see from the direction field that y = 0. y ′ = − sin y . ± 2 π . We have a constant solution y(t ) ≡ 0 2 and there is a symmetry between solutions above and below the t-axis. From left to right these solutions are horizontal translates.SECTION 1. ± 3π . … being stable and stable equilibrium 5 unstable equilibrium stable equilibrium –5 unstable equilibrium stable equilibrium–5 y ± π . ± 4 π . 74. … unstable.2 Solutions and Direction Fields y 35 72. Student Project Suggested Journal Entry II 76. –4 4 t –2 73. t 5 y ′ = 2 y + t . Note: This equation does not have a closed form solution. Student Project Suggested Journal Entry I 75.

y′ = t2 . no constant solution. Separating variables. we get y dy = t 2dt .3 Separation of Variables: Quantitative Analysis Separable or Not 1. y′ = y +1 + y . no constant solutions. Not separable. Not separable. y ′ = sin(t + y) . no constant solutions. 6. 9. Separating variables. y ′ = et e y . Not separable. y +1 y ′ = t ln y 2t + t 2 = t 2 (2 ln y + 1) . = dt t . 4. 1+ y dy = dt .36 CHAPTER 1 First-Order Differential Equations 1. Separable. Separable. 2 ln y + 1 y t + . 10. . Integrating each side gives the implicit solution y 1 2 1 3 y = t + c. Separable. Separable. 3. y ′ = ln(ty) . Separable. 1 + y2 t Solving by Separation 11. e− y ( y + 1)dy = et dt . constant solutions y(t ) ≡ 0. e− y dy = et dt . Separable. 7. ty y′ = et e y . Solving for y gives the explicit solution y = sin(ln t + c) . y ′ = 1 + y . 8. The equilibrium solutions are y = ±1 . Not separable. 12. ± 1 . constant solution y(t ) ≡ e−1 2 . dy = dt . no constant solutions. y′ = b g dy = t 2dt . t y y′ = 1 + y2 dy . 5. constant solution y ≡ −1. 2. no constant solutions. we get dy 1 − y2 = dt . 2 3 Solving for y yields branches so we leave the solution in implicit form. t Integrating gives the implicit solution sin −1 y = ln t + c . no constant solutions. y − y3 y ′ = y − y 3 . no constant solutions. ty ′ = 1 − y 2 .

where C = ±ec1 . ty ′ = 4 y . Separating variables we get the equation y′ = 4 y − 4 y3 by Integrating gives the implicit solution 4 − 4 y 3 dy = t 2 + 7 dt . 15. 4t dy = (y2 + ty2)dt y(1) = 1 ∫4 y = ∫ 2 dy 1+ t 1 dt = + 1 dt t t ∫ −4y−1 = ln t + t + C For y(1) = 1. we obtain C = −5. g b g 1 5 1 y − y 4 = t 3 + 7t + c . Solving for y gives the explicit solution y = Ct 4 where C is an arbitrary constant. 16. The equilibrium solution is y = 0. y t Integrating gives the implicit solution ln y = 4 ln t + c . dy = y cos t dt y = 0 is an equilibrium solution. dy = cos t dt y For y ≠ 0. t2 + 7 .3 Separation of Variables: Quantitative Analysis 37 13.SECTION 1. Separating variables we get dy dt =4 . so that y = Cesin t . so that y= −4 ln t + t − 5 . ∫ ∫ ln y = sin t + c1 e ln y = esin t ec1 . 14. 5 3 We cannot find an explicit solution for y.

y(1) = −2 . the implicit solution is y−2 = ±ece4t = ke4t y+2 where k is an arbitrary constant. N 4( y − 2) 4( y + 2) Q ln y − 2 − ln y + 2 = 4t + c Integrating we get or y−2 = ec e4 t . Solving for y. Hence. Solving for y we get y(t ) = − −2t 2 + 2t + 4 .38 CHAPTER 1 First-Order Differential Equations 17. y′ = 1 − 2t . y ′ = y 2 − 4 . 1 − ke4t b g Substituting in the initial condition y(0) = 0 gives k = −1. 18. y(0) = 0 . 2 Substituting in the initial condition y(1) = −2 gives c = 2 . the implicit solution is given by y 2 = 2t − 2t 2 + 4 . y −4 2 Rewriting this expression as a partial fraction decomposition (see Appendix PF). . we get the general solution y(t ) = 2 1 + ke4t . y+2 Hence. we get LM 1 1 − 1 OP dy = dt . Separating variables gives dy = dt . Separating variables gives y y dy = (1 − 2t )dt . Note that we take the negative square root so the initial condition is satisfied. Integrating gives the implicit solution 1 2 y = t − t2 + c.

π 4 . sec2 y dy = ln t dt + c z z . y′ = 2t . −1 + 1 + 4 t 2 − 4 . Integrating gives the implicit solution y + y2 = t 2 + c . K π 2 y ′ = cos2 y ln t . Solving for y the preceding quadratic equation in y we get y= 20. Substituting in the initial condition y(0) = −1 gives c = tan−1(−1) = − y = tan − tan−1 t − Integration by Parts 21.SECTION 1. y(0) = −1. we find z z dy = ln t dt + c cos2 y tan y = t ln t − t + c y = tan −1(t ln t − t + c) . Substituting in the initial condition y(2) = 0 gives c = −4 .3 Separation of Variables: Quantitative Analysis 39 19. =− 1 + y2 1+ t2 Integrating gives tan−1 y = − tan−1 t + c. Integrating. Solving for y gives F H π 4 I. cos2 y b g . 2 a f y′ = − 1 + y2 . The equilibrium solutions are y = (2n + 1) Separating variables we get dy = ln tdt . we get the equation 1+ t2 dy dt . Separating variables 1+ 2y (1 + 2 y)dy = 2t dt . y(2) = 0 . Separating variables.

2 4 z b g LM b g N 2 OP Q Solving for y. (B) 27. Separating variables we get dy = t 2 − 5 cos 2t dt . we find y= b g b g = t 2 cos 2t dt − 5 cos2t dt + c 1 1 5 = 2t 2 − 1 sin 2t + t cos 2t − sin 2t + c . y ′ = t 2 − 5 cos 2t . Separating variables we get dy = te−t dt .40 CHAPTER 1 First-Order Differential Equations 22. 4 2 2 23. (C) 26. y Integrating. (A) 30. LM 1 bt − t ge N2 2t 1 − e2 t − c . we find z z dy = te−t dt + c y y = Qe−(t +1)e . −t ln y = −te−t − e−t + c Equilibria and Direction Fields 25. 4 OP Q y ′ = t ye− t . zb z t 2 − 5 cos 2t dt + c g b g z y ′ = t 2e y + 2t . (F) 29. we get y = − ln 24. Separating variables we get dy 2 2t = t e dt . ey Integrating. (D) . (E) 28. The equilibrium solution is y = 0. we find z e− y dy = t 2e2t dt + c −e − y = e− y 1 2 1 t − t e2 t + e2 t + c 2 4 1 2 1 =− t − t e 2 t + e2 t + c . Integrating.

we have dy dy dy = + = dt . 2 are equilibrium solutions. we divide by 2 y − y 2 and rewrite the equation in differential form as dy = dt . ke2t − 1 . Simplifying. dy dy dy = + = dt . y(2 − y) By a partial fraction decomposition (see Appendix PF). we find 1 1 ln y − ln 2 − y = t + c 2 2 . we find y= 32. we get − ln 1 − y + ln 1 + y = 2t + 2c ln 1+ y = 2t + 2c 1− y (1 + y) = ke2t (1 − y ) where k is any nonzero real constant. (1 − y)(1 + y) 2(1 − y) 2(1 + y) Integrating. To find the nonconstant solutions. y(2 − y) 2 y 2(2 − y) Integrating. we find 1 1 − ln 1 − y + ln 1 + y = t + c 2 2 where c is any constant. To find the nonconstant solutions we divide by 1 − y 2 and rewrite the equation in differential form as dy = dt . ke2t + 1 y′ = 2 y − y2 We note first that y = 0. If we now solve for y. y′ = 1 − y2 We note first that y = ±1 are equilibrium solutions. 1 − y2 By a partial fraction decomposition (see Appendix PF).3 Separation of Variables: Quantitative Analysis 41 Finding the Nonequilibrium Solutions 31.SECTION 1.

y( y − 1)( y + 1) By finding a partial fraction decomposition. we get ln y − ln 2 − y = 2t + 2c ln y = 2t + 2c 2− y y ( 2 − y ) = Ce 2t where C is any positive constant. y2 Multiplying each side of the above equation by y 2 gives a quadratic equation in y. 1 + ke2t f Initial conditions will tell which branch of this solution would be used. getting y=± a 1 .42 CHAPTER 1 First-Order Differential Equations where c is any real constant. ±1 are equilibrium solutions. which can be solved. Simplifying. we get y= 2 ke2t . To find the nonconstant solutions. y = ke 2t 2− y where k is any nonzero real constant. y ′ = y( y − 1)( y + 1) We note first that y = 0. If we solve for y. (see Appendix PF) dy dy dy dy =− + + = dt . y( y − 1)( y + 1) y 2( y − 1) 2( y + 1) Integrating. 1 + ke2t 33. we divide by y( y − 1)( y + 1) and rewrite the equation in differential form as dy = dt . . we find 1 1 − ln y + ln y − 1 + y + 1 = t + c 2 2 −2 ln y + ln y − 1 + ln y + 1 = 2t + 2c or ln ( y − 1)( y + 1) = 2t + 2c y2 ( y − 1)( y + 1) = ke2t .

y ′ = cos t . y(−1) = −1 is y = −et +1. The solutions are shown in the figure. y(1) = 1. y(−1) = −1 The solution of the initial-value problem y 2 y ′ = cos t . These solutions are shown in the figure. y 3 t –2 2 –3 36. y(1) = 1 is y = et −1 . Seeking nonconstant solutions. y(1) = 1 is y(t ) = sin t + 1 − sin(1) . y ′ = ( y − 1)2 We note that y = 1 is a constant solution.3 Separation of Variables: Quantitative Analysis 43 34. y(−1) = −1 The solution of y ′ = y . y(1) = 1. −t + c Help from Technology 35. y(−1) = −1 is y = sin t − 1 + sin(−1) . we divide by ( y − 1)2 dy = dt . y ′ = y .SECTION 1. The solution of –6 6 t y ′ = cos t . The solution of y ′ = y . This can be integrated to get getting ( y − 1)2 1 − =t+c y −1 1 y −1= −t + c 1 y = 1+ . –2 .

we find y dy = 2t dt y +1 or –2 2 t z y dy = 2t dt + c . we find the implicit solution ln y = sin t + c . we find c = − sin(1) . we get 1 − ln 2 = 1 + c or c = − ln 2 . y(−1) = −1 y y 2 Separating variables and assuming y ≠ −1. we find c = are shown in the figure. we find the implicit solution y − ln y + 1 = t 2 + c . For y(1) = 1. y(1) = 1. 3 Subsituting y(1) = 1. y′ = 2t ( y + 1) . 38. y(1) = 1.44 CHAPTER 1 First-Order Differential Equations y 37. dy t = . These two solutions are plotted on the direction field (see figure). y(−1) = −1 2 dt y 1 + t 2 Separating variables and integrating we find the implicit solution 2 t z or y 2dy = z t 1 + t2 –2 2 dt + c –2 1 3 y = 1 + t2 + c. With y(−1) = −1. y′ = t y2 1 + t 2 1 1 − 2 . 39. –8 With y(1) = 1. For y(−1) = −1 we find c = − − 2 . y(1) = 1. y –6 2 t 6 Integrating. we find c = sin(1) . y +1 z –2 y′ = 2t ( y + 1) y Integrating. y(−1) = −1 Separating variables we get dy = cos t dt . For y(−1) = −1 we can see even more easily that y ≡ −1 is the solution. These two curves 3 3 y y ′ = y cos t . Note that . These two implicit solution curves are shown imposed on the direction field (see figure).

y 2 t –6 6 –2 Making Equations Separable 41. The initial condition y(1) = 1 lies on the upper branch. and the solution through that point does not cross the t-axis. Hence. However. t . t t dv y = 1 + v . t Integrating gives v = ln t + c and y = t ln t + ct . 40.3 Separation of Variables: Quantitative Analysis 45 the implicit solution involves branching. we can draw its direction field along with the requested solutions (see figure). y(1) = 1. we write t y′ = y2 + t 2 y t 1 = + =v+ . Given y′ = we let v = y+t y = 1+ . y ′ = sin(ty) . yt t y v But y = tv so y ′ = v + tv′ . 42.SECTION 1. Separating variables gives and get the separable equation v + t dt t dt = dv . we have v + tv′ = v + or tv′ = 1 v 1 v or v dv = dt . Letting v = y . y(−1) = −1 This equation is not separable and has no closed form solution.

Given y′ = with the new variable v = 2 y4 + t 4 2 y t3 1 = + 3 = 2v + 3 . 3 t ty y v y . 43. we get t dv dt = . we have the explicit solution y = t tan(ln t + c) . Solving for v gives v = tan ( ln t + c ) . v +1 t v +1 3 v Integrating gives the solution 1 ln t = ln v 4 + 1 + c 4 or ln t = 44. . Hence. so t y = ±t 2 ln t + c . But v = y . y(t ) = −t 2 ln t + 4 . ( ) 1 ln 4 LMF y I + 1OP + c . NH t K Q 4 Given y′ = with the new variable v = y 2 + ty + t 2 y 2 y = 2 + + 1 = v2 + v + 1 t t2 t y . Using y ′ = v + tv′ and separating variables gives t dt dv dv = 4 = v3 4 . The initial condition y(1) = −2 requires the negative square root and gives c = 4. v +1 t 2 Integrating gives the implicit solution ln t = tan −1 v + c .46 CHAPTER 1 First-Order Differential Equations Integrating gives the implicit solution 1 2 v = ln t + c 2 or v = ± 2ln t + c . Hence. Using y′ = v + tv′ and separating variables.

and y = 1 − t − ln c − t . or ∫e −u du = dt .3 Separation of Variables: Quantitative Analysis 47 Another Conversion to Separable Equations 45. so ∫ t + e−t−y+1 = c. so +1 ∫ 46.. if you follow along a horizontal line y = k in the ty plane. Then du dy = + 1 = u 2 + 1. we have dx dy =− dx ∂f ∂x ∂f ∂y . Orthogonal Families 48. dy = et + y −1 − 1 dt Let u = t + y − 1. Thus. These slopes are the slopes of the tangent lines. and dt dt tan −1 u = t + c u = tan(t + c) y + t = tan(t + c) so y = tan(t + c) − t ∫u 2 du = dt . Autonomous Equations 47. Another way to say this is that solutions for y(t ) through any y all have the same slope. the slopes of the line elements do not change). y) = c . −t − y + 1 = ln c − t . (b) y′ = 1 + y y′ = y − y 3 y′ = y 2 − 4 The isoclines of an autonomous equation are horizontal lines (i. 2 and 18 are autonomous: #1 #2 #18 All the others are nonautonomous. (a) Starting with f ( x. and dt dt −e−u + c = t . y′ = ( y + t )2 Let u = y + t. Then du dy =1+ = 1 + eu − 1.SECTION 1. we differentiate implicitly getting the equation ∂f ∂f dx + dy = 0 ∂x ∂y Solving for y ′ = dy .e. . (a) Problems 1.

dx f x x which is a separable equation having solution y = kx . the orthogonal curves satisfy dy = dx ∂f ∂y ∂f ∂x . we have ∂f ∂f = 2x . For the family y = cx 2 we have f ( x. we have 1 y2 = − x2 + K 2 or x2 + 2 y2 = C . Graphs of the orthogonal families are shown in the figure. = 2 y and ∂y ∂x so our equation is dy y = . this last equation gives a family of ellipses that are all orthogonal to members of the family y = cx 2 . (c) Given f ( x. Integrating. Orthogonal trajectories Hence. fy = 2 . Hence. from part (b) the orthogonal trajectories satisfy the dx x dy f y y = = . .48 CHAPTER 1 First-Order Differential Equations (b) Taking the negative reciprocal of the slopes of the tangents. differential equation More Orthogonal Trajectories 49. y) = x 2 + y 2 . x x and the orthogonal trajectories satisfy x dy f y = =− 2y dx f x or y so x2 y 3 2 1 –3 –2 –1 –1 –2 –3 1 2 3 x 2y dy = − x dx . y) = 2y 1 fx = − 3 .

For the family y = c we have f ( x. we have the solution y2 − x2 = C . y 3 2 1 –3 –2 –1 –1 –2 –3 1 2 3 x Orthogonal trajectories . Graphs of the two orthogonal families of hyperbolas are shown. in differential form. y ) = x 2 y so x2 x2 y 3 2 1 f x = 2 xy . Integrating. xy = c . the preceding family of hyperbolas are orthogonal to the hyperbolas xy = c . we have 1 y 2 = x 2 + C or 2 y 2 − x 2 = K . y) = xy so f x = y . the preceding equations give a family of hyperbolas that are orthogonal to the original c family of hyperbolas y = 2 . The orthogonal trajectories satisfy dy f y x = = dx f x y or. Hence. in differential form. f y = and the orthogonal trajectories satisfy x dy f y = = dx f x 2 y or. 2y dy = x dx . Integrating.SECTION 1. y dy = x dx . Here f ( x. x 51. Graphs of the orthogonal families are shown. Calculator or Computer –5 y 5 5 t –5 Orthogonal hyperbolas 52. 2 –3 –2 –1 –1 –2 –3 1 2 3 x Orthogonal trajectories Hence. f y = x . We know the orthogonal trajectories of this family of horizontal lines is the family of vertical lines x = C (see figure). y = c .3 Separation of Variables: Quantitative Analysis 49 50.

50

CHAPTER 1

First-Order Differential Equations

53.

4 x 2 + y 2 = c . Here f ( x, y) = 4 x 2 + y 2 and f x = 8 x , f y = 2 y , so the orthogonal trajectories satisfy dy f y 2 y y = = = dx f x 8 x 4 x 4dy dx or = , which has the implicit solution the y x family y 4 = Cx where C is any constant different from zero. These orthogonal families are shown in the figure. x 2 = 4cy 3 . Here f ( x, y) = and f x = x2 4 y3
–2 –1 –2 –1

y 2

1 x

1 –1 –2

2

Orthogonal trajectories

54.

y 2

1 x

3x 2 x , f y = − 4 . The differential equa2 y3 4y dy f y −3x = = 2y dx f x

1 –1 –2

2

tion of the orthogonal family is

Orthogonal trajectories or 2 y dy = −3x dx , which has the general solution 2 y 2 + 3x 2 = C , where C is any real constant. These orthogonal families are shown in the figure.
55.

x 2 + y 2 = cy . Here f ( x, y) =

x2 + y2 , so y fx = y2 − x2 2x , fy = . y y2

The differential equations of the orthogonal family are
dy = = dx f x fy
y2 − x2 y2 2x y

=

y2 − x2 . 2 xy

We are unable to solve this equation analytically, so we use a different approach inspired by looking at the graph of the original family, which consists of circles passing through the origin with centers on the y-axis.

SECTION 1.3

Separation of Variables: Quantitative Analysis

51

x2 + y −

F H

Completing the square of the original equation, we can write x 2 + y 2 = cy as c 2

I K

2

=

c2 c , which confirms the description and locates the centers at 0, . 2 4

F I H K

We propose that the orthogonal family to the original family consists of another set of C 2 C2 C + y2 = centered at , 0 and passing through the origin. circles, x − 2 4 2

F H

I K

F H

I K

To verify this conjecture we rewrite this equation for the second family of circles as x2 + y2 x 2 + y 2 = Cx , which gives C = g( x, y ) = x x2 − y2 2y or gx = , gy = . Hence the proposed x2 x second family satisfies the equation 2 xy dy , = = 2 dx gx x − y 2 gy

y

2

–2 –2

2

x

Orthogonal circles which indeed shows that the slopes are perpendicular to those of the original family derived above. Hence the original family of circles (centered on the y-axis) and the second family of circles (centered on the x-axis) are indeed orthogonal. These families are shown in the figure.
The Sine Function 56.

The general equation is
y 2 + ( y ′)2 = 1

or

dy = ± 1 − y2 . dx
Separating variables and integrating, we get

± sin−1 y = x + c or y = sin ± x + c = ± sin x + c . This is the most general solution. Note that cos x is included because cos x = sin x −

c b gh

b g

F H

π

I. 2K

52

CHAPTER 1

First-Order Differential Equations

Disappearing Mothball 57.

(a)

dV = − kA , where V is the volume, t is time, A is the surface area, and k is a dt 4 positive constant. Because V = π r 3 and A = 4π r 2 , the differential equation becomes 3 We have 4π r 2 or dr = −k . dt 1 1 1 Integrating, we find r (t ) = − kt + c . At t = 0, r = ; hence c = . At t = 6, r = ; hence 2 2 4 1 , and the solution is k= 24 r (t ) = − 1 1 t+ , 24 2 dr = −4 kπ r 2 dt

where t is measured in months and r in inches. Because we can’t have a negative radius or time, 0 ≤ t ≤ 12 . (b) Solving − 1 1 t + = 0 gives t = 12 months or one year. 24 2

Four-Bug Problem 58.

(a)

According to the hint, the distance between the bugs is shrinking at the rate of 1 inch per second, and the hint provides an adequate explanation why this is so. Because the bugs are L inches apart, they will collide in L seconds. Because their motion is constantly towards each other and they start off in symmetric positions, they must collide at a point equidistant from all the bugs (i.e., the center of the carpet).

(b) (c)

The bugs travel at 1 inch per second for L seconds, hence the bugs travel L inches each.
Q dr Ardθ B r + dr P r, θ 0

This sketch of text Figure 1.3.8(b) shows a typical bug at P = (r , θ ) and its subsequent position Ax (r + dr , θ + dθ ) as it heads toward the next bug at Q = r , θ +

r

F H

π

2

I . Note K

that dr is negative, and consider that dθ is a very small angle, exaggerated in the drawing.

and the other bugs’ paths simply shift θ by π e for each successive bug. we can write T4 dT = − kdt . which is a right isosceles triangle. Hence our bug is following the path r = e−θ . F I H K M +T T = 4kM 3t + C . For small dθ : • • • angle BAP is approximately a right angle. Radiant Energy 59. Solving this separable DE gives r = ce−θ . π 4 . 2 −M T + M2 U V W we find the implicit solution M −T T 1 1 ln − tan −1 = −2kM 2t + c M 2M M + T M or in the more convenient form ln Suggested Journal Entry 60. Hence triangle ABP is similar to triangle OQP. Separating variables.3 Separation of Variables: Quantitative Analysis 53 Consider the small shaded triangle ABP. + 2 arctan M −T M F I H K Student Project . so −dr ≈ rdθ . angle APB = angle OQP = side BP lies along QP. We then write − M4 T4 Integrating 1 1 1 1 . and the initial condition r (0) = 1 gives c = 1. = 2 = − 4 2 T2 − M 2 2 T2 − M 2 2 T2 + M 2 2M 2M −M T +M b gb g b g b g R ST T 2 1 1 − 2 dT = −2 kM 2dt .SECTION 1.

54

CHAPTER 1

First-Order Differential Equations

1.4

Euler’s Method: Numerical Analysis

Easy by Calculator y ′ = 1.

t , y (0) = 1 y

(a)

Using step size 0.1 we enter t0 and y0 , then calculate row by row to fill in the following table:
Euler’s Method ( h = 0.1) n

tn = tn −1 + h
0
0.1 0.2 0.3 approximations at

′ yn = yn −1 + hyn −1
1
1

′ yn =

tn yn

0
1 2 3 The requested

1.01 1.0298

0 =0 1 0.1 = 0.1 1 0.2 = 0.1980 1.01 0.3 = 0.2913 1.0298

t = 0.2

and

t = 0.3

are

y3 ( 0.3) ≈ 1.0298 . (b)

y2 ( 0.2 ) ≈ 1.01 ,

Using step size 0.05, we recalculate as in (a), but we now need twice as many steps. We get the following results.
Euler’s Method ( h = 0.05 )

n 0 1 2 3 4 5 6

tn
0

yn
1

′ yn
0 0.05 0.0998 0.1489 0.1971 0.2440 0.2893

0.05 0.1 0.15
0.2

1 1.0025 1.0075
1.0149

0.25
0.3

1.0248
1.03698

The approximations at t = 0.2 and t = 0.3 are now y4 ( 0.2 ) ≈ 1.0149 , y6 ( 0.3) ≈ 1.037 .

SECTION 1.4

Euler’s Method: Numerical Analysis

55

(c)

Solving the IVP y ′ = Integration gives

t , y ( 0 ) = 1 by separation of variables, we get y dy = t dt . y

1 2 1 2 y = t +c. 2 2 The initial condition y ( 0 ) = 1 gives c = for y gives the explicit solution
y (t ) = 1 + t 2 .

1 and the implicit solution y 2 − t 2 = 1 . Solving 2

To four decimal place accuracy, the exact solutions are
y ( 0.3) = 1.0440. Hence, the errors in Euler approximation are
h = 0.1: h = 0.05 : error = error = error = error =

y ( 0.2 ) = 1.0198 and

y ( 0.2 ) − y2 ( 0.2 ) = 1.0198 − 1.0100 = 0.0098, y ( 0.3) − y3 ( 0.3) = 1.0440 − 1.0298 = 0.0142, y ( 0.2 ) − y4 ( 0.2 ) = 1.0198 − 1.0149 = 0.0050, y ( 0.3) − y6 ( 0.3) = 1.0440 − 1.0370 = 0.007

Euler approximations are both high, but the smaller stepsize gives smaller error.

Calculator Again y ′ = ty , y ( 0 ) = 1 2.
(a) For each value of h we calculate a table as in Problem 1, with y ′ = ty . The results are summarized as follows.

Euler’s Method Comparison of Step Sizes

h=1
t y≈ t

h = 0.5
y≈ t

h = 0.25
y≈ t

h = 0.125
y≈

0

1

0 0.5

1 1

0 0.25 0.50 0.75

1 1 1.062 1.195

0 0.125 0.250 0.375 0.50 0.625 0.750 0.875

1 1 1.0156 1.0474 1.0965 1.1650 1.2560 1.3737

1

1

1

1.25

1

1.419

1

1.5240

56

CHAPTER 1

First-Order Differential Equations

(b)

Solve the IVP y ′ = ty , y ( 0 ) = 1 by separating variables to get

dy = tdt . Integration yields y

ln y =

2 t2 + c , or y = Cet 2 . Using the initial condition y ( 0 ) = 1 gives the exact solution 2 2 y ( t ) = et / 2 , so y (1) = e1 2 ≈ 1.6487 . Comparing with the Euler approximations gives

h = 1: h = 0.5 : h = 0.25 : h = 0.125 :

error = 1.6487 − 1 error = 1.6487 − 1.25 error = 1.6487 − 1.419 error = 1.6487 − 1.524

= 0.6487 = 0.3987 = 0.2297 = 0.1247

y ′ = 3t 2 − y , y ( 0 ) = 1 ; [ 0, 1] . Using a spreadsheet and Euler’s method we obtain the following values:

A 1 2 n 0 = A2 + 1 B tn 0 = B 2 + .1 C ′ yn = yn −1 + hyn −1 1 = C 2 + .1 ∗ D 2 D
2 3tn − yn

= 3 ∗ t ∗ B2 ^ 2 − C

3

Using step size h = 0.1 and Euler’s method we obtain the following results.

Euler’s Method ( h = 0.1)
t y≈ t y≈

0
0.1 0.2 0.3 0.4 0.5

1
0.9 0.813 0.7437 0.6963 0.6747

0.6 0.7 0.8 0.9

0.6822 0.7220 0.7968 0.9091

1.0

1.0612

Smaller steps give higher approximate values yn ( tn ) . The DE is not separable so we have no exact solution for comparison.

The DE is not separable so we have no exact solution for comparison.5696 10. y ′ = t 2 + e − y .3525 3. [ 0.3568 0.0 3.9858 1.) 5.4736 Smaller steps give higher approximate values yn ( tn ) .8 1.1855 0.2 0.01 and Euler’s method we obtain the following results.8792 8. y (1) = 1 .8099 3.01) t y t y 1 1.4 Euler’s Method: Numerical Analysis 57 4.4 0.01) t y≈ t y≈ 0 0.4 1.5355 0.8078 2.) Euler’s Method ( h = 0.8 1.5 2 2.4203 5 12.0 0 0.01 .2 1.6 0.6 1.9942 5.2915 1.1521 2. [1. . The DE is not separable so we have no exact solution for comparison. and Euler’s method we obtain the following results.5 4 4. y ′ = t + y . y ( 0 ) = 0 .4283 Smaller steps give higher yn ( tn ) .7453 2.7395 0. (Table shows only selected values.6740 2.5 6. 5] Using step size h = 0. (Table shows only selected values.SECTION 1.5 3 1 1. Euler’s Method ( h = 0. 2] Using step size h = 0.

1630 1.0755 Smaller steps give higher yn ( tn ) .6783 2. 5] Using step size h = 0.1204 1.6435 1.8143 3.2211 1.8075 1.8 0.) Euler’s Method ( h = 0.2962 0.58 CHAPTER 1 First-Order Differential Equations 6. y ′ = t 2 − y 2 .2615 3 3.5 2 2.5 1 0.05 and Euler’s method we obtain the following results. The DE is not separable so we have no exact solution for comparison.8682 4.9 1. . y ( 0 ) = 1 .7463 1. [ 0.3 0.3843 5 4.) Euler’s Method ( h = 0.5 2 1.5 1 1.8967 Smaller steps give higher approximate values yn ( tn ) . (Table shows only selected values.01 and Euler’s method we obtain following results.6 0. y′ = t − y .1 0.2 0.5 2.6992 0.5 4 4.3464 3. The DE is not separable so we have no exact solution for comparison. (Table shows only selected values. 7.7 0.1171 1.05 ) t y≈ t y≈ 0 0. y ( 0 ) = 2 Using step size h = 0.3903 1.5053 1.01) t y t y 0 0.4 0.0916 1 1.

9 0.2 2. that the solution to this IVP does not exist for t > 1.4 0.1 0.5 Smaller stepsize predicts lower value.1 and Euler’s method we obtain the following results. y (0) = 1 y Using step size h = 0. Hence.3421 3 1.3 0. the absolute error at t = 1 is 0.8405 0.9 1.0827 1.3755 .1) t y≈ t y≈ 0 0.5601 1 0.SECTION 1.6 0.7 2.2 0.3079 1.0418 1.) Euler’s Method ( h = 0.7 0.2366 1. 9.3 2. y ( 2) = 1 t Using step size h = 0.6 2.7691 0. 1 1.1 2.1226 1. y′ = sin y .0000 0.4 Euler’s Method: Numerical Analysis 59 8. however.8963 0. y′ = − t .9900 0. whose value at t = 1 is y (1) = 0 .4 2.6781 0.8 2. (Note.3994.5 1 1.9389 0.9698 0.05 ) t y≈ t y≈ 2 2.05 and Euler’s method we obtain the following results.3994 The analytical solution of the initial-value problem is y (t ) = 1 − t 2 .1616 1. (Table shows only selected values.2727 1.1995 2. ) You can experiment yourself to see how this error is diminished by decreasing the step size or by using a more accurate method like the Runge-Kutta method.8 0. Euler’s Method ( h = 0.

8375 0.0021 . y ′ = −ty .7 0.5 1 0. y0 = 1 Using step size h = 0. .01 and Euler’s method we obtain the following results.9 0. the absolute error at t = 1 is error = 0.1 0.9249 0.4 0.7284 0.3 0.01) t y≈ t y≈ 0 0.9574 0. Hence.6065 − 0.9955 0. (Table shows only selected values.6086 = 0.8845 0.) Euler’s Method ( h = 0. The analytical solution of the initial-value problem is y ( t ) = e −t 2 2 whose exact value at t = 1 is y (1) = 0.60 CHAPTER 1 First-Order Differential Equations 10.8 0.9812 0.2 0.6692 1 0.6 0.6065 .7850 0.6086 Smaller step size predicts lower value.

90 4.0000 1.0005 3.4 Euler’s Method: Numerical Analysis 61 Stefan’s Law Again 11.60 0.25 0.70 0.0001 (b) The graph shows that the larger step approximation (black dots) overshoots the mark but recovers.0000 3.00 2. T ( 0 ) = 4 . 1 0 1 t .50 0.30 0.80 0.0237 3.75 4.10 0.00 T 5 3. (a) dT = 0.00 0.1 tn Tn 0 1 2 3 4 0. which is confirmed both by the direction dT field and the slope .20 0.0002 1.8125 2.1250 3.00 0. 3 (c) There is an equilibrium solution at T = 3 . dt ( ) Euler’s Method h = 0.0480 0 1 2 3 4 5 6 7 8 9 10 0.0049 3.0107 3. while the smaller step approximation (white dots) avoids that problem.SECTION 1.6901 3.0023 3.50 0.40 0.0010 3.05 34 − T 4 .9810 1.0532 3. This is an exact dt solution that both Euler approximations get very close to by the time t = 1 .25 n tn Tn n h = 0.

25 2.6917 5. So Euler’s 1− t method gives an approximation far too small.25 0.0025 0.3525 1.0135 0. we have estimated the solution of this IVP at t = 1 .0034 0. y ( 0 ) = 1 Using Euler’s method with h = 0.25 1.6533 2. Euler’s Method ( h = 0. The true value of y = et for t = 1 is e ≈ 2.001 .025 0. y ( 0 ) = 1 Using Euler’s method with different step sizes h.1245 0.6850 2.0013 0.25 ) t y≈ y′ = y 2 0 0.7169 0.05 0. we can see that the solution blows up at 1.7115 2.0332 0.005 0.7182818… .00 3. h Euler’s Method y (1) ≈ e − y (1) 2.6517. or from the direction field.5937 2.6406 2.75 1 1.25 we obtain the following values.0650 0. y′ = y . whereas from the analytical solution 1 y (t ) = .50 0. y′ = y 2 . Approximating e 13.6517 Euler’s method estimates the solution at t = 1 to be 3.01 0.5 0.7048 2.5625 2.4683 0.1 0.62 CHAPTER 1 First-Order Differential Equations Nasty Surprise 12.7149 2.3135 1 1.0068 0.

9913 1. getting the following values.718281828 0.5213 2.21 × 10−5 0.2029 0.5454 0.5 2 2.5 3. y ( 0 ) = 0 (a) The solution starting at the initial point y ( 0 ) = 0 never gets off the ground (i.5 1 1.00093 0.025 0. y ( 0 ) = 0.7918 3.87 × 10−8 0.1 0.13 × 10−6 0.718279744 2. which is better than Euler’s method with stepsize h = 0.5 3 0.22 × 10−11 Note that even with a large step size of h = 0.4 Euler’s Method: Numerical Analysis 63 We now use the fourth-order Runge-Kutta method with the same values of h.718281693 2.717346191 2.1241 2.001.01 0.5 the Runge-Kutta method gives y (1) correct to within 0.9424 6 7. it returns all zero values for yn ).9134 .001 .9134 .3005 5. y = y1 3 . yn ( 6 ) = 0 .01 .5 4 4.01 ( h = 0.1336 6. (b) Euler’s Method y ′ = y1 3 .5187 4.05 0. Double Trouble or Worse 14.718281820 2.SECTION 1. We have given a few values in the following table and see that yn ( 6 ) ≈ 7.5 0.. h Runge-Kutta Method e − y (1) y(1) 2.e.0151 6. the solution increases.1 ) t y t y 0 0. For this IVP.5 5 5. Starting with y ( 0 ) = 0.01 0.

64 CHAPTER 1 First-Order Differential Equations (c) The direction field of y ′ = y1 3 for 0 ≤ t ≤ 6 . the accumulate roundoff error grows at an exponential rate. .165. Hence. If a roundoff error of ε occurs in the initial condition.026ε t = 20 ⇒ difference = ε e 20 = 485. y ( 0 ) = A + ε is y ( t ) = ( A + ε ) et = Aet + ε et . because near y = −2 the perturbed solution will move towards the stable solution –2. On the other hand. then the solution of the new IVP y ′ = y . Think Before You Compute 16. a roundoff error in computations starting near y = −2 is not as serious as near y = 2 . 0 ≤ y ≤ 10 confirms the values found in (b). y 10 0 0 6 t y′ = 3 y Roundoff Problems 15. The difference between this perturbed solution and Aet is ε et . any initial conditions starting at these values should remain there.195ε . Because y = 2 and y = −2 are constant solutions. This difference at various intervals of time will be t = 1 ⇒ difference = ε e t = 10 ⇒ difference = ε e10 ≈ 22.

y(0) = 0. y(1) = −2 + e ≈ 0. (c) If then y(t) = −t − 1 + et. y ′ = t + y .5 ⎣2⎦ 2 By 4th order Runge Kutta. y1 = y0 + h(t0 + y0 ) = 0 By 2nd order Runge Kutta y1 = y0 + hk02.646 6⎝ ⎝2⎠ ⎝4⎠ ⎠ 6 (b) Second-order Runge Kutta is much better than Euler for a single step approximation.375 2 ⎠ ⎝ ⎝ 2 ⎠⎝ 4 ⎠ y`1 = 0 + ⎞ 1 1⎛ ⎛1⎞ ⎛3⎞ ⎜ 0 + 2 ⎜ ⎟ + 2 ⎜ ⎟ + 1.SECTION 1. h = 1 (a) By Euler’s method.375 ⎟ = (3. y1 = y0 + h ( k01 + 2k02 + 2k03 + k04 ) 6 k01 = t0 + y0 = 0 h⎞ ⎛ h ⎞ 1 ⎛ k02 = ⎜ t0 + ⎟ + ⎜ y0 + k01 ⎟ = 2⎠ ⎝ 2 ⎠ 2 ⎝ h⎞ ⎛ h ⎛ ⎞ 1 1⎛1⎞ 3 k03 = ⎜ t0 + ⎟ + ⎜ y0 + k02 ⎟ = + ⎜ ⎟ = 2⎠ ⎝ 2 ⎝ ⎠ 2 2⎝ 2⎠ 4 h ⎛ ⎞ ⎛ 1 ⎞⎛ 3 ⎞ k04 = (t0 + h) + ⎜ y0 + k03 ⎟ = 1 + ⎜ ⎟⎜ ⎟ = 1. .875) ≈ 0. k01 = t0 + y0 = 0 h⎞ ⎛ h ⎞ ⎛ k02 = ⎜ t0 + ⎟ + ⎜ y0 + k01 ⎟ 2⎠ ⎝ 2 ⎠ ⎝ = 1 +0 2 ⎡1⎤ 1 y1 = 0 + ⎢ ⎥ = = 0. but fourth-order RK is almost right on (slightly low).4 Euler’s Method: Numerical Analysis 65 Runge-Kutta Method 17.718.

5 ⎝ 2⎠ By 4th order Runge Kutta.875 2 ⎠ 8 ⎝ ⎝ 2 ⎠⎝ 4 ⎠ 1⎛ 7⎞ 1 ⎛ 1⎞ ⎛ 1⎞ y`1 = 0 + − ⎜ 0 + 2 ⎜ − ⎟ + 2 ⎜ − ⎟ + − ⎟ = − (−2.66 CHAPTER 1 First-Order Differential Equations 18. y1 = y0 + h ( k01 + 2k02 + 2k03 + k04 ) 6 k01 = t0 + y0 = 0 h⎞ ⎛ h ⎞ 1 ⎛ k02 = ⎜ t0 + ⎟ + ⎜ y0 + k01 ⎟ = − = − 0. . Fourth order R-K is very close.368. k01 = t0 + y0 = 0 h⎞ ⎛ h ⎞ 1 ⎛ k02 = ⎜ t0 + ⎟ + ⎜ y0 + k01 ⎟ = − 2 2⎠ ⎝ 2 ⎠ ⎝ ⎛ 1⎞ y1 = y0 − 1⎜ − ⎟ = 0. y(0) = 0.396 6 6⎝ 8⎠ ⎝ 2⎠ ⎝ 4⎠ (b) Second-order Runge Kutta is high though closer than Euler. y(−1) = e−1 ≈ 0.375) ≈ 0. y ′ = t + y . y1 = y0 + h(t0 + y0 ) = 0 By 2nd order Runge Kutta y1 = y0 + hk02. h = −1 (a) By Euler’s method.25 2⎠ ⎝ 2 2 ⎝ 2 ⎠⎝ 2 ⎠ 4 ⎝ ⎠ h 7 ⎛ ⎞ ⎛ 1 ⎞⎛ 1 ⎞ k04 = (t0 + h) + ⎜ y0 + k03 ⎟ = −1 + ⎜ − ⎟⎜ − ⎟ = − = − 0. (c) If then y(t) = −t − 1 + et.5 2⎠ ⎝ 2 ⎠ 2 ⎝ h⎞ ⎛ h 1 ⎛ 1 ⎞⎛ 1 ⎞ 1 ⎛ ⎞ k03 = ⎜ t0 + ⎟ + ⎜ y0 + k02 ⎟ = − + ⎜ − ⎟⎜ − ⎟ = − = − 0.

y ′ = 3t 2 − y . y′ = t − y .148 1.8145 1.1036 We compare this with #7 where Euler’s method gives y (1) ≈ 1.1 0. 1] Using the fourth-order Runge-Kutta method and h = 0.0 0.3 0.8 0.7359 0.7870 0. y ( 0 ) = 1 .7 0.0612 for h = 0. y ( 0 ) = 2 Using the fourth-order Runge-Kutta method and h = 0.9 1. y ( 0 ) = 1 t y t y 0 0. y ′ = t − y . y (1) ≈ 1.9058 0.6562 1. y ′ = 3t 2 − y .1197 1. Runge-Kutta Method.1 we arrive at the following table of values.SECTION 1.5 y 2 1. Euler 19.046 for step h = 0.3 0.4 0.9972 1.1 we arrive at the following table of values. .7659 0.1606 We compare this with #3 where Euler’s method gave y (1) ≈ 1.5 1 0.6 0.4 0.8734 0. Exact solution by separation of variables is not possible.2 0. Exact solution by separation of variables is not possible. y ( 0 ) = 2 t 0 0.7 0. 20.4 Euler’s Method: Numerical Analysis 67 Runge-Kutta vs.2 0.1 0.3196 t 0. Runge-Kutta Method.7284 0.05 .1 .1 .07545 for step h = 0.9 1.6 0. [0.8 0.4110 1.8263 0.1898 1.0 y 1.5225 1.7173 0.2464 1.

1 0. and the exact solution y ( t ) = 1 − t 2 gave y (1) = 0 .7141 0.9798 0.9 1 0.6086 .9560 0.6065 We compare this with #10 where Euler’s method for step h = 0.7 0.5 exact solution y ( t ) = e −t within given accuracy. y ′ = −ty .7 0.9165 0. The Runge-Kutta approximate solution is exact .3 0. The Runge-Kutta approximate solution is much closer to the exact solution.9231 0. y′ = − t .4358 0.9802 0. (Table shows only selected values. 1 0. y ( 0 ) = 1 t y t y 0 0. Runge-Kutta Method.8 0.6670 0.0 0.5 1 0.7261 0.04880 We compare this with #8 where Euler’s method for step h = 0.8 0.01 to arrive at the following table.1 gave y (1) ≈ 0. y ′ = − ty .3 0.8353 0.6000 0.9 1. y ( 0) = 1 y y 0 0.68 CHAPTER 1 First-Order Differential Equations 21.8825 0.1 gave y (1) ≈ 0.6065 . y (0) = 1 y Using the fourth-order Runge-Kutta method and h = 0.6 0.4 0.1 0.3994 . and the 2 2 gave y (1) = 0.9539 0. y ( 0 ) = 1 22. y ′ = − t Y t t .9950 0.9950 0.8660 0.6 0.1 we arrive at the following table of values.4 0. Using the 4th-order Runge Kutta method and h = 0.7827 0.2 0.2 0.) Runge-Kutta Method.8000 0.

if h ≤ M desired condition en ≤ E . (b) The expression y ( tn + h ) = y ( tn ) + y ′ ( tn ) h + 1 * y ′′ tn h 2 2 ( ) is simply a statement of Taylor series to first degree. (a) Differentiating y ′ = f ( t . which in this case is 3.4 Euler’s Method: Numerical Analysis 69 Euler’s Errors 23. so y ′′ is continuous as well. we have the new rule 1 yn +1 = yn + hf ( tn . For the equation y ′ = f ( t . f y and y ′ = f are continuous. yn ) f ( tn . y ) = (c) t 1 t we have ft ( t . y ) . ⎦ 2 ⎣ (b) The local discretization error has order of the highest power of h in the remainder for the approximation of yn +1 . Three-Term Taylor Series 24. with remainder. yn ) ⎤ . we have the second derivative of y ′′ on the interval [tn . Hence. y ) f ( t . we get y ′′ = ft ( t . y ) y ′ = ft ( t . yn ) + f y ( tn . where M is the maximum of the 2 2E . Hence. (a) Starting with y ′ = f ( t . y ) gives y ′′ = ft + f y y ′ = ft + f y f .SECTION 1. 2 (d) We can make the local discretization error en in Taylor’s method less than a preassigned value E by choosing h so it satisfies en ≤ Mh 2 ≤ E . tn +1 ] . y ) = . ⎝ yn ⎠ 2 ⎣ yn yn ⎦ Using this formula and a spreadsheet we get the following results. y ) + f y ( t . yn ) + h 2 ⎡ f t ( tn . y ) + f y ( t . (c) Direct computation gives en +1 ≤ M h2 . Here we assume ft . . f y ( t . y ) . y ) = − 2 and so the y y y preceding three-term Taylor series becomes ⎛ t ⎞ 1 ⎡ 1 t2 ⎤ yn +1 = yn + h ⎜ n ⎟ + h 2 ⎢ − n3 ⎥ . and differentiating with respect to t.

3749 1.6444 ≈ 0.4151 ≈ 0.7 0.70 CHAPTER 1 First-Order Differential Equations Taylor’s Three-Term Series t Approximation of y ′ = .8 0. y ) = ty we have ft ( t .6 0. y ) = t .4142… . Hence the error e − 1. which has an error of 2 − 1.5 1 1. ⎦ 2 ⎣ Using this formula and a spreadsheet.3 0.0458 1.9 1.0443 1. so y (1) = e ≈ 1.9 1. y ( 0 ) = 1 is y ( t ) = 1 + t 2 .4151.005 1.3262 1.2 0. y ( 0 ) = 1 y t y t y 0 0. (d) For the differential equation y ′ = f ( t . we arrive at the following results.2213 1. so the Euler three-term approximation becomes yn +1 = yn + htn yn + 1 2 2 h ⎡ yn − t n y n ⎤ .5 1 1.1962 1.1667 1.005 1. Taylor’s Three-Term Series Approximation of y ′ = ty .4 0.0009 . y ( 0 ) = 1 t Y t y 0 0.4 0.3 0.7 0.6444 The solution of y ′ = ty .1 0.4962 1.649… .0043 . y ) = y .1 0.6 0. .0201 1.2761 1.8 0.1325 2 0.0 1.0442 1.4151 t .2314 1.0199 1.1083 1. f y ( t . y ( 0 ) = 1 is y ( t ) = et at t = 1 using Taylor’s three-term method is 2 .2 0. Taylor’s three-term method gave the value 1.0 1. so we y The exact solution of the initial-value problem y ′ = have y (1) = 2 ≈ 1.1185 0.

yR t ∗ = ∗ ∗ ( ) y t∗ .SECTION 1.2 ) . h 1. Our calculations are listed in the following table (on the next page). 25.2155 ( ) y t∗ . y ( 0 ) = 1 .1 = 1.0025 1. Our calculations are listed in the following table.1 1.1025 1.0408 27.1 0.2 ) . Note that we use yR ( 0.0176 1. y ′ = ty .01505 1. h) − y ( t . Note that we use yR ( 0. h ( ) ( ) 2 y ( t . This is legitimate with no further argument if y ′ is positive and monotone increasing.1052 e0.1) as .2183 ( ) ( ) 2 y ( t . One-step EulerTwo-step Euler t∗ y t∗ .04 = 1.4 Euler’s Method: Numerical Analysis 71 Richardson’s Extrapolation Sharp eyes may have detected the elimination of absolute value signs when equation (7) is rewritten as equation (9). as is the case in the suggested exercises.1050 1.2 = 1. y ( 0 ) = 1 .0101 e0.005 1. Note that we use yR ( 0.02005 e0.01 = 1. y′ = y 2 .1 0. initial condition for computing yR ( 0.2211 e0. y ( 0 ) = 1 . h) Exact solution y = et 2 0.2 1.2 ) . h) Exact solution y = et 0. h Richardson approx.1) as initial condition for computing yR ( 0.2 1. y′ = y . One-step EulerTwo-step Euler t∗ Richardson approx.0 1. h) − y ( t . yR t ∗ = ∗ ∗ y t∗ . Our calculations are listed in the following table.1) as initial condition for computing yR ( 0.2214 26. h 1.

y ( s )) ds + c . the integral equation is equivalent to IVP. starting with the initial-value problem y ′ = f ( t . h) Exact solution y = 1 (1 − t ) 1.2 ) .1102 1.0150 ( ) y t∗ . Note that we use yR ( 0. gives the constant c = y0 .0050 1. Our calculations are listed in the following table.1 1. y ′ = sin ( ty ) . We also have y ( t0 ) = y0 . h 1.1 0.2405 ( ) ( ) 2 y ( t .2335 ( ) y t∗ . h 1. yR t ∗ = ∗ ∗ t∗ 0.0201 1. yR t ∗ = ∗ ∗ t∗ 0.2476 28. y ( t ) ) .0025 1. h) − y ( t . h) Exact solution no formula 1. Hence. y ( s ) ) ds t t0 we differentiate respect to t. h 1.2 y t∗ . Conversely.1051 1. y ( t ) ) . getting y ′ = f ( t . h) − y ( t .1 0.1) as initial condition for computing yR ( 0. .0176 ( ) ( ) 2 y ( t .2 y t∗ . t t0 Using the initial condition y ( t0 ) = y0 . y ( t0 ) = y0 we integrate getting the solution y (t ) = ∫ f ( s.1 1. (a) Starting with y ( t ) = y0 + ∫ f ( s.2500 1. One-step Euler Two-step Euler Richardson approx. h 1.02013 by Runge-Kutta Integral Equation 29.72 CHAPTER 1 First-Order Differential Equations One-step Euler Two-step Euler Richardson approx.1111 1. y ( 0 ) = 1 .

1 .5840 10. (c) The Riemann sum only holds for integrals of the form ∫ f ( t ) dt . ⎣ ⎦ If we. getting y (T ) = y0 + ∫ f ( s ) ds T 0 ≈ y0 + h ⎡ f ( 0 ) + f ( h ) + … + f (T − h ) ⎤ .3618 3.5 2 2. Fourth Order Runge-Kutta Method t Y t y 1 1. however. We solve the IVP of Problem 5 y′ = t + y .8910 8. t 0 To find the approximate value of the solution at t = T . y (1) = 1 by several different methods using step size h = 0. we evaluate the preceding integral at t = T using the Riemann sum with left endpoints.0010 5.8100 2. Sample study of different numerical methods. The table shows a printout for selected values of y using one non-Euler method.4373 12. y ′ = f ( t ) .8144 4.4 Euler’s Method: Numerical Analysis 73 (b) The initial-value problem. is transformed into the integral equation y ( t ) = y0 + ∫ f ( s ) ds .SECTION 1.5 4 4. b a Computer Lab: Other Methods 30. y ( 0 ) = y0 . write the expression as y (T ) = y0 + h ⎡ f ( 0 ) + f ( h ) + … + f (T − h ) ⎤ ⎣ ⎦ = y1 + hf ( h ) + … + hf (T − h ) = y2 + hf ( 2h ) + … + hf (T − h ) = y3 + hf ( 3h ) + … + hf (T − h ) + yn −1 + hf (T − h ) … … = yn −1 + h (T − h ) = yn .5 3 1 1.5 5 6. we get the desired conclusion.4480 .

2519 y ( 5 ) ≈ 12. Suggested Journal Entry I 31. but you might use step h = 0.1 to approximate y ( 5 ) by other methods (for example Adams-Bashforth method or Dormand-Prince method) then explain which method seems most accurate. A graph of the direction field could give insight.4283 y ( 5 ) ≈ 12.4480 h = 0.01 h = 0.74 CHAPTER 1 First-Order Differential Equations We can now compare the following approximations for Problem 5: Euler’s method (answer in text) Euler’s method (solution in manual) Runge-Kutta method (above) We have no exact solution for Problem 5. h = 0. Student Project .1 y ( 5 ) ≈ 12.1 Student Project Suggested Journal Entry II 32.

(b) (c) 2. f y = − . The functions f and f y are continuous for t ≠ 0 . it isn’t defined. which we know exists. It would also be useful to draw the direction field of this equation and see the big picture. y′ = 2− y .5 Picard’s Theorem: Theoretical Analysis 75 1. y0 ) with t0 = 0 . (c) If we think of DEs as models for physical phenomena. (b) Uniqueness/existence in either the right half plane t > 0 or the left half plane t < 0 . When Here f ( t . Hence. so t t there is a unique solution passing through any initial point y ( t0 ) = y0 with t0 ≠ 0 . No solution of this DE . The direction field is shown in the figure. y ) = 1 − ty . Picard’s Theorem: Theoretical Analysis Picard’s Conditions (a) y ′ = f ( t . y ) = passes through points ( t0 .the answer to part (a) is positive. y (0) = 1 t (a) 2− y 1 . Not applicable . In particular the DE with IC y ( 0 ) = 1 does not make sense. y ( 0 ) = 0 Hence f y = −t . The fact that f is y 3 continuous for all t tells us a solution exists passing through each point in the ty plane. The direction field is shown in the figure. there is a unique solution of this equation passing through y ( 0 ) = 0 . The further fact that the derivative f y is also continuous for all t and y tells us that the solution is unique.SECTION 1. we might be tempted to replace t0 in the IC by a small number and examine the unique solution. –2 3 t y 6 t0 = 0 the derivative y′ is not only discontinuous. –3 t –3 3 Picard’s conditions hold in entire ty plane.5 1. any rectangle that does not include t = 0 will satisfy Picard’s Theorem.

there Picard’s conditions hold in entire ty plane. . so there is really no need to resort to Picard’s theorem to conclude there is no solution passing through such points. y ) = y4 3 4 f y = y1 3 . which we know to be y ( t ) ≡ 0 . The directions field of the equation is shown in the figure. y ) = t−y t+y 2t fy = − ( t + y )2 t –4 4 –4 are continuous for t and y except when y = −t . so by Picard’s theorem we conclude that the will be a unique solution passing through y ( 0 ) = 0 . y′ = t−y . Not applicable . (b) (c) 4. y ( 0 ) = −1 t+y y 4 (a) Here both f (t. DE has a unique solution through any initial condition y ( t0 ) = y0 . In particular. (a) y′ = y 4 3 . so any rectangle that does not include any part of y = −t satisfies Picard’s Theorem. When y = −t the derivative y′ is not only discontinuous but also not even defined.the answer to part (a) is positive. y ( 0 ) = 0 Here f (t. (b). (c) Picard’s conditions hold for the entire ty plane except the line y = −t . Hence.76 CHAPTER 1 First-Order Differential Equations 3. there is a unique solution passing through any initial condition y ( t0 ) = y0 as long as y0 ≠ −t0 . 3 –4 y 4 t 4 –4 Here f and f y are continuous for all t and y.

± . 6.… . Hence. there exists a unique solution passing through y ( t0 ) = y0 except when y=± The IVP problem passing through π 2 .5 Picard’s Theorem: Theoretical Analysis 77 5.… . The direction field of the equation illustrates these ideas (see figure). It would be useful to look at the direction field to get an idea of the behavior of solutions for nearby initial points. there is a unique solution passing through any initial point y ( t0 ) = y0 except y ( 0 ) = 0 . 2 π 2 does not have a solution. y ) = 2 t –2 2 ( ) 2 –2 are continuous for all t and y except at the point y = t = 0 . so the IVP does not make sense. 2 2 π –3π /2 Hence.SECTION 1. (b) Picard’s Theorem gives existence/uniqueness for any rectangle that does not include the origin. In this case f does not exist. The direction field of the equation shows that where Picard’s Theorem does not work the slope has become vertical (see figure). (a) y ′ = tan y . y ) = tan y f y = sec 2 y t –2 2 are both continuous except at the points 3π y=± . (c) It may be useful to replace the initial condition y ( 0 ) = 0 by y ( 0 ) = y0 with small but nonzero y0 .± 3π . y ( 0) = 0 t + y2 2 y 2 Here both 1 t + y2 2y f y (t. y ( 0 ) = Here π 2 3π /2 y f ( t . y ) = − 2 t + y2 f (t. (a) y′ = 1 . .

2⎠ ⎝ y ′ = ln y − 1 .78 CHAPTER 1 First-Order Differential Equations (b) Existence/uniqueness conditions are satisfied over any rectangle with y-values between two successive odd multiples of π 2 . ⎟ . 8. See figure for the direction field of the equation. –4 where neither is defined. Hence. y ) = y y −t t t –4 4 fy = − ( y − t )2 –4 are continuous for all t and y except when y ≠ t where neither function exists. we can be assured there is a unique solution passing through y ( t0 ) = y0 except when not make sense. there is a unique solution passing through any initial point y ( t0 ) = y0 with y0 ≠ 1 . y ) = ln y − 1 fy = 1 y −1 t –4 4 are both continuous for all t and y as long as y ≠1. y ( 0 ) = 2 Here y 4 7. t0 = y0 . so it holds for any rectangle that does not include any part of y = t. When t0 = y0 the derivative isn’t defined. (b). so IVP problems with these IC does . In particular. Hence the IVP with y (1) = 1 is not defined. The direction field of the equation illustrates these ideas (see figure). (a) y′ = Here y . (b) The Picard Theorem holds for the entire ty plane except the line y = t . (a) f ( t . (c) ⎛ π⎞ There are no solutions going forward in time from any points near ⎜ 0. Hence. there is a unique solution passing through y ( 0 ) = 2 . y (1) = 1 y −t y 4 f (t. (c) The Picard Theorem holds for entire ty plane except the line y = 1 .

no unique solution passes through C where the derivative is not uniquely defined. A unique solution will pass through each of the points B.5 Picard’s Theorem: Theoretical Analysis 79 (c) It may be useful to replace the initial condition y (1) = 1 by y (1) = 1 + ε . B. 14. y ) = q (t ) − p (t ) y f y (t. 13. if we assume p ( t ) and q ( t ) are continuous. where finite slope does not exist. we can write y ′ = q ( t ) − p ( t ) y and so f (t. B. C. and D. C. C. 16. 12. B. 11. C. A unique solution will pass through each of the points A. and D. possibly unique solutions at D where t = y = 0 . A unique solution will pass through each of the points A. where there is no unique slope. Solutions exist only for t > t A or t < t A because all solutions appear to leave from or go toward A. a unique solution passes through D for positive t. ε < 0 will send solution toward zero. B. Nonunique solutions at A. and D and the solutions appear to exist for all t. A unique solution will pass through each point A. then Picard’s theorem holds at any point y ( t0 ) = y0 . and D. Linear Equations 9. you should note that the direction field shows that ε > 0 will send solution toward ∞. Unique solutions exist passing through points B and C on intervals until the solution curve reaches the t-axis. Solutions appear to exist for all t. y′ + p ( t ) y = q ( t ) For the first-order linear equation. Unique solutions will pass through each of the points A. A unique solution passes through A and B defined for negative t. C. 15. Solutions appear to exist for all t. Eyeballing the Flows For the following problems it appears from the figures given in the text that: 10. Hence.SECTION 1. and D. y ) = − p (t ). . Solutions appear to exist for all t. However.

(d) Because Picard’s theorem holds for all t. To find the size of the interval of existence. B. we get − t+c 1 . Integrating gives the result − y −1 = t + c and 1 . the solution may not exist over the entire plane. y ( 0 ) = 1 important weakness of Picard’s Theorem: For any particular initial condition. C. y0 Hence. B. t < 1 . f y = 2 y . Solutions appear to exist for all t. y0 ) . C. However. (c) The separated equation is y −2 dy = dt . y ) = y 2 . y. gives c = −1 . the interval over which this solution is defined cannot pass through t = t0 + which implies an interval of ⎛ 1 ⎞ ⎜ −∞. 1 . and the solution with IC y ( 0 ) = 1 exists on the interval ( −∞. solving for y. 18. Hence the existence and uniqueness conditions hold for any initial –3 t –3 3 (b) y 3 t =1 condition in the entire ty plane. Solutions appear to exist for all t. y so by Picard’s theorem there is a unique solution passing through any point t. and D. (a) f (t. and D. y > 0 .80 CHAPTER 1 First-Order Differential Equations 17. getting y (t ) = − 1 t − t0 − 1 y0 . Substituting the initial condition y ( 0 ) = 1 . we must solve the IVP. The interval over which this solution is Hence. A unique solution will pass through each of the points A. 1) . A unique solution will pass through each of the points A. y we conclude there exists a unique solution to y ′ = y 2 . y (0) = 1 are both continuous for all t. we have y ( t ) = 1− t defined cannot pass through t = 1 . y ( t0 ) = y0 for any ( t0 . t0 + ⎟ y0 ⎠ ⎝ for positive y0 and . Local Conclusions 19. In the given IVP the solution exists only for t < 1 . this example exhibits an Solution of y′ = y 2 .

SECTION 1. However. y ( 0 ) = 0 Because f = y1 3 is continuous for all ( t . y ) . Nonuniqueness 20. we have found one solution of the initial-value problem as ⎛2⎞ y (t ) = ± ⎜ ⎟ ⎝3⎠ But clearly. y 3 c=0 c = –1 c = –2 y=0 4 t c = –2 t< c t≥ c –3 c=0 c = –1 for any c ≤ 0 . we find c = 0 .5 Picard’s Theorem: Theoretical Analysis 81 ⎛ ⎞ 1 . and integrating gives 3 23 y =t+c. y ′ = y1 3 . 2 Picking the initial condition y ( 0 ) = 0 . getting an infinite number of solutions to the initial-value problem as ⎧0 ⎪ y ( t ) = ⎨ ⎛ 2 ⎞3 2 32 ⎪± ⎜ ⎟ ( t + c ) ⎩ ⎝3⎠ 1 2 3 32 t3 2 . getting y −1 3 dy = dt . using any c < 0 . f y = 1 −2 3 y is not continuous when y = 0 so Picard’s 3 theorem does not guarantee a unique solution through any point where y = 0 . In fact. We first separate variables. Picard’s theorem says that there exists a solution through any point y ( t0 ) = y0 . Nonuniqueness of solutions through y ( 0) = 0 . In fact we can find an infinite number of solutions passing through the origin. ∞⎟ ⎜ t0 − ⎜ ⎟ y0 ⎝ ⎠ for negative y0 . Futhermore. y ( t ) ≡ 0 is another solution. we can also paste together y = 0 with infinitely many additional solutions. we can paste these solutions together at t = 0 . Hence. A few of these solutions are plotted (see figure).

Consequently the hypothesis of Picard’s Theorem is not fulfilled by the DE in any region containing points on the x-axis. y ( 0 ) = 0 . Because the line y = −1 passes through every point with y-coordinate = 1. the DE becomes y ′ = y. the solution is y ( t ) ≡ 0 . ∂f = 3t2 is continuous for all t and y ∂y By Picard’s Theorem. Converse of Picard’s Theorem Fails 23. No. the DE becomes y ′ = − y. so they agree. Note that the only solution that satisfies the IVP occurs when C = 0. so that f (t . we have y = At t = t0 the left-hand derivative of 1 ( t − t0 ) 2 . 4 y ( t ) ≡ 0 is 0. y) is continuous for all t and y. no other solution can merge with y = −1 and can only approach y = −1 asymptotically. Consider y ′ = 3t 2 (1 + y ) = f (t . 2. which has the general solution y = Ce−t. y ). Note that y = −1 is an equilibrium solution. For t > t0 . y ) = ⎨ dt ⎩y ∂f ⎧ −1 =⎨ ∂y ⎩1 y<0 y≥0 has a partial derivative y<0 y > 0. so that is a unique solution. We observe: 1. 4 Seeing vs. we know there is a unique solution through any initial point. which is precisely the function y ≡ 0. (b) Note that y ≡ 0 is a solution of the IVP dy = y dt y(0) = 0 When y < 0. y ′ = y . t0 > 0 For t < t0 . y ) . the solution does not “merge” with y = −1. . (a) Note that ⎧− y dy = y = f (t . which has general solution y = Cet.82 CHAPTER 1 First-Order Differential Equations More Nonuniqueness 21. and the right-hand derivative of y (t ) = 1 ( t − t0 )2 is 0. Believing 22. that is not continuous at y = 0. f(t. When y ≥ 0.

⎝ 4π ⎠ 23 . where A is the surface area of the snowball and k > 0 is the dt rate at which the snowball decreases in volume. h ) = −k h . given by A = 4π r . h (t ) h0 (b) Let us assume the bucket becomes empty at t = T < t0 .5 Picard’s Theorem: Theoretical Analysis 83 Hubbard’s Leaky Bucket 24. we cannot be sure of unique solutions passing ∂h through any points where h ( t ) = 0 . k Hence the time to empty the bucket is T= The Melted Snowball 25. we find an infinite number of 1 ( kT − kt )2 4 h (t ) = 0 h (t ) = for t < T for t > T . We show a few such solutions for T < t0 . Because ∂f k =− ∂h 2 h ∂f is not continuous at h = 0 . and between the surface 3 2 area of the snowball and its radius. Solving the IVP with solutions. (a) We are given dV = − kA . which is V = π r 3 . various T values t0 t Each one of these functions describes the bucket emptying. we can relate A and V by ⎛ 3 ⎞ 23 23 3 A = 4π ⎜ ⎟ V = 36π V . Hence. Given the relationships between the 4 volume of the snowball and its radius r. 4 2 h0 . then (b) gives h (0) = 1 2 2 k T = h0 . (c) If we start with a full bucket when t = 0 . h (T ) = 0 . dh = −k h dt (a) f ( t .SECTION 1. we don’t know when the bucket became empty.

dt for K = k 3 and t < T . Hence. we cannot conclude that the IVP dV = − kV 2 3 . The function f ( t .84 CHAPTER 1 First-Order Differential Equations (b) Here f ( t . we find 3V 1 3 = −kt + c . 10 Integrating. But we know V ( t ) ≡ 0 is also is a solution of this initial-value problem. V ( t0 ) = 0 dt has a unique solution. V ) = −kV 2 3 ∂f 2 = − kV −1 3 . so we can piece together the nonzero solutions with the zero solution and get for T < t0 the infinite family of solutions ⎧ ⎛ t − T ⎞3 ⎪− K V (t ) = ⎨ ⎜ 3 ⎟ ⎝ ⎠ ⎪0 ⎩ (d) t <T t ≥T. dt y 20 V − 2 3 dV = − kdt . (c) Separating we have dV = − kV 2 3 where k > 0 . V ) = V 2 3 does not satisfy the uniqueness condition of Picard’s theorem when V = 0 . . we cannot tell when the snowball melted. the backwards solution is not unique. Let T < t0 be the time the snowball melted. Then using the initial condition V (T ) = 0 we find ⎛ t −T ⎞ V (t ) = −K ⎜ ⎟ ⎝ 3 ⎠ 3 t0 –3 various T values 2 t dV = − kV 2 3 . ∂V 3 Because the uniqueness condition for Picard’s theorem does not hold when V = 0 . Solutions with y ( t0 ) = 0 .

so we can piece together the nonzero solutions with the zero solution. V ( t ) ≡ 0 is also a solution of this initial-value problem. We substitute into dt We are given 4 V = π r 3 . dt (b) Separating variables in the above DE. we have 23 V − 2 3 dV = kdt . Using the initial condition V ( t0 ) = 0 . But clearly. we get the relation c = − kt0 . (a) dV = kA . to get the infinite family of solutions ⎧0 ⎪ V ( t ) = ⎨ ⎛ t − t0 ⎞ 3 ⎪K ⎜ ⎟ ⎩ ⎝ 3 ⎠ t < t0 t ≥ t0 3 . A = 4π r 2 3 for the volume V and area A of a raindrop in terms of its radius r. ⎝ 4π ⎠ Hence dV = kV 2 3 . and hence ⎛ t − t0 ⎞ V (t ) = K ⎜ ⎟ ⎝ 3 ⎠ where K = k 3 .5 Picard’s Theorem: Theoretical Analysis 85 The Accumulating Raindrop 26. where A is the surface area of the raindrop and k > 0 is the rate dt dV = kA the relationships at which the raindrop increases in volume. getting ⎛ 3 ⎞ 23 23 3 A = 4π ⎜ ⎟ V = 36π V .SECTION 1. Integrating. we find 3V 1 3 = kt + c .

(c) For (a).86 CHAPTER 1 First-Order Differential Equations Different Translations 27. (a) y ′ = y has an infinite family of solution of the form y = Cet. solutions y = Cet gradually approach zero as t → −∞. Note that for any real number a. 2 ⎩(t − a) for t ≥ a . with s′ = 2 s . for t < a ⎧0 For (b). with y ′ = y . solutions y = ⎨ go to zero at t → a. y = et − a = Cet is a solution for every a ∈ R. (To check: y ′ = (Cet )′ = Cet = y. (b) t<a ⎧0 Differentiating s(t) = ⎨ 2 ⎩(t − a) t ≥ a we obtain a continuous derivative t<a ⎧0 s′(t ) = ⎨ ⎩2(t − a) t ≥ a Note that s′ = 2 s for both parts of the curve.

SECTION 1. y0 ( t ) = 1 y1 ( t ) = 1 + y2 ( t ) = 1 + y3 ( t ) = 1 + 1 ∫ ( s − y ) ds = 1 + ∫ ( s − 1) ds = 1 + 2 t t t 2 0 0 0 −t ∫ ∫ 1 3 2 ⎛ 1 2 ⎞⎫ ⎨ s − ⎜ 1 + s − s ⎟ ⎬ ds = − t + t − t + 1 0 6 ⎠⎭ ⎩ ⎝ 2 t⎧ t⎧ 0 1 4 1 3 2 ⎡ 1 3 2 ⎤⎫ ⎨ s − ⎢ − s + s − s + 1⎥ ⎬ ds = t − t + t − t + 1 24 3 ⎦⎭ ⎩ ⎣ 6 29. y0 ( t ) = 1 + t y1 ( t ) = 1 + y2 ( t ) = 1 + 3 ∫ ( s − y ) ds = 1 + ∫ t 0 0 t 0 t 2 t 0 2 ⎡ s − (1 + s ) ⎤ ds = 1 + − ds = 1 − t ⎣ ⎦ ∫ 0 t ∫ ( s − (1 − s ) ) ds = t − t + 1 1 y ( t ) = 1 + ∫ ⎡ s − ( s − s + 1) ⎤ ds = − t ⎣ ⎦ 3 0 3 + t2 − t +1 . y0 ( t ) = e −t y1 ( t ) = 1 + y2 ( t ) = 1 + y3 ( t ) = 1 + −s −t ∫0 ( s − y0 ) ds = 1 + ∫0 ( s − e ) ds = e + t t 1 2 t 2 ∫ ∫ t⎡ 0 1 3 1 2 ⎛ −s 1 2 ⎞⎤ −t ⎢ s − ⎜ e + 2 s ⎟ ⎥ ds = e − 6 t + 2 t ⎠⎦ ⎣ ⎝ 1 4 1 3 1 2 ⎛ −s 1 3 1 2 ⎞⎤ −t ⎢ s − ⎜ e − 6 s + 2 s ⎟ ⎥ ds = e + 24 t − 6 t + 2 t 0 ⎠⎦ ⎣ ⎝ t⎡ 31.5 Picard’s Theorem: Theoretical Analysis 87 Picard Approximations 28. y0 ( t ) = t − 1 y1 ( t ) = 1 + y2 ( t ) = 1 + y3 ( t ) = 1 + ∫ ( s − y ) ds = 1 + ∫ t 0 0 t 0 ⎡ s − ( s − 1) ⎤ ds = t + 1 ⎣ ⎦ ∫ ∫ t 0 t 0 ⎡ s − ( s + 1) ⎤ ds = −t + 1 ⎣ ⎦ ⎡ s − (1 − s ) ⎤ ds = t 2 − t + 1 ⎣ ⎦ 30.

t ) . t ) .88 CHAPTER 1 First-Order Differential Equations Computer Lab 32. We leave for the reader the other starting functions for #28. because f ( t . = 1 + int ( t − y4 . = 1 + int ( t − y3 . = 1 + int ( t − y5 . y ( s ) ) ds = 1 + ∫ ( s − y ( s ) ) ds t t 0 n 0 n If you then hit the enter key you will see displayed y0 y1 y2 y3 y4 y5 y6 = t −1 = t +1 = −t + 1 = t2 − t +1 1 = − t3 + t2 − t + 1 3 1 1 = t 4 − t3 + t 2 − t + 1 12 3 1 5 1 4 1 3 2 = − t + t − t + t − t +1 60 12 3 Of course you can find more iterates in the same way. t ) . 30. y ( 0 ) = 1 with starting function y0 ( t ) = t − 1 . then hit Enter. t ) . t).. E. (a) We show how the computer algebra system Maple can be used to estimate the solution of #29 y ′ = t − y . y ( 0 ) = 1 . In Maple open a new window and type the int() command. ∫ f ( s. t ) . = 1 + int ( t − y0 . = 1 + int ( t − y2 . and 31. you will see y7 = − 1 6 1 5 1 4 1 3 2 t − t + t − t + t − t +1 360 60 12 3 . y ) = t − y . = 1 + int ( t − y1 . In this problem. we can find the sequence of approximations yn +1 ( t ) = y0 + by typing y0 y1 y2 y3 y4 y5 y6 = t − 1. if you type y7 = 1 + int(t − y6. t ) .g.

y 2 34.. y ) = y1 4 ∂f 1 −3 4 = y ∂y 4 Note the direction field is only defined when y ≥ 0 . Calculator or Computer 33.4.8 1. however.2 0 1 Picard Approximation t 2 3 4 Picard’s sixth approximation to y′ = t − y . y 2 1.4 0. y ′ = f ( t . y ( 0 ) = 1 (b) If you recall the Maclaurin series e −2t ≈ 1 − 2t + 1 1 ( −2t )2 + ( −2t )3 +… and carry out a 2 3! little algebra.5 Picard’s Theorem: Theoretical Analysis 89 To get a plot of y6 and the solution y ( t ) = 2e −t + t − 1 (see part (b)) as shown (see figure).2 1 0. through any point y ( t0 ) = y0 . Picard’s theorem guarantees existence for points y ( t0 ) = y0 when y0 = 0 .8 0.. t=0. y ) = sin ( ty ) ∂f = t cos ( ty ) ∂y Picard’s theorem guarantees both existence and uniqueness for any point ( t0 .SECTION 1.6 0. you will convince yourself that these Picard’s approximations are converging to the analytical solution y ( t ) = 2e −t + t − 1 . such a nice identification is not possible. The direction field shown illustrates these ideas.2).6 1. –5 t 5 –2 . The direction field shown also indicates these ideas. y=0. type the Maple command plot({2*exp(–t)+t–1. y ′ = f ( t .4 1. DE does not exist for y < 0 . For most initial-value problems. but not uniqueness t –3 3 y 2 y ′ = y1 4 . y6}. y0 ) .

the DE has a solution through every point ( t0 . See figure for this direction field.90 CHAPTER 1 First-Order Differential Equations y 2 35. y0 ) but Picard’s theorem does not guarantee uniqueness through points for which y0 = t0 . y ) = ( y − t ) ∂f 1 −2 3 = ( y − t) 3 ∂y 13 y 4 The function f is continuous for all ( t . y ) = y 5 3 ∂f 5 2 3 = y ∂y 3 Picard’s theorem guarantees existence and uniqueness for all initial conditions y ( t0 ) = y0 . y′ = f ( t . y′ = f ( t . y ) = ( ty ) 13 y 3 ∂f 1 1 −2 3 = ( ty ) t = t1/ 3 y −2 / 3 3 ∂y 3 The function f is continuous for all ( t . –3 t 3 37. y ′ = f ( t . See figure for this direction field. Hence. we are not guaranteed uniqueness through points ( t0 . y ) . but f y is not continuous when y = t (it doesn’t exist). The direction field shown also illustrates this fact. but f y is not continuous when y = 0 . y ) . t –2 2 –2 36. Hence. y0 ) when y0 is zero. t –4 4 –4 .

y0 ) .SECTION 1. y ) = 6t 2 − ∂f 3 =− ∂y t 3 y t y 1 The function f is continuous except when t = 0 .5 Picard’s Theorem: Theoretical Analysis 91 38. The direction field of the equation as shown indicates that no solutions pass through initial points of the form ( 0. Also f y is continuous except when t = 0 . and so the DE has a unique solution for all initial conditions except possibly when t0 = 0 . y0 ) except possibly when t0 = 0 . Student Project . hence there exists a solution through all points ( t0 . t –1 1 –1 Suggested Journal Entry 39. y ′ = f ( t .