Unizor - Double Integrals - Circular Base

Notes to a video lecture on http://www.unizor.com

Double Integrals -
Circular Base

Previous lecture introduced a concept of a double integral as a volume of a solid with a rectangular base.
In particular, we came to a conclusion that in this case we can integrate, first, by one of the argument and then by another or, equally well, in an opposite order. Rectangular base of a solid allowed such a freedom.

In this lecture we will consider a solid with a circular base and, which is important, the technique of calculating the volume of this solid would be valid for many other types of bases, not necessarily circular. Another important point is that the integration in one order cannot be performed in another order without modifications.

Consider the following problem.
Given a smooth function of two variables f(x, y) with non-negative values (that is, f(x, y) ≥ 0), defined on an area bounded by a circle of a radius R with a center at the origin of coordinates, and represented by a surface in (X,Y,Z) coordinate space.



Our task is to find the "volume of a solid" - the measure of a part of coordinate space bounded on the top by a surface representing this function, on the bottom - by X,Y-plane and sides forming a cylindrical surface based on a circle described above, rising on a height determined by the values of our function on the circle's border.

We will follow the same logic as in a case of a rectangular base.
The process of approximation starts with dividing our circular base into many small rectangles by lines x=x_i (where i∈[0,M], x₀=−R, x_M=R) and y=y_j (where j∈[0,N], y₀=−R, y_N=R) and constructing rectangular parallelepipeds on the base of these small rectangles with the height equaled to a value of function f(x, y) in the corner of each base rectangle with largest coordinates x and y ("top right corner").

If point (x_i, y_j) represents the coordinates of a "top right" corner of a small base rectangle and Δx_i=x_i−x_i−1, Δy_j=y_j−y_j−1are sides of this small rectangle, the volume of a constructed rectangular parallelepiped equals to
ΔV_i,j = f(x_i, y_j)·Δx_i·Δy_j

The combined volume of all these parallelepipeds approximates the volume of our solid. With increasing number of small base rectangles and, correspondingly, decreasing their dimensions we assume that there is a limit of the combined volume of all these parallelepipeds.
It can be proven that under certain conditions of smoothness of our function f(x, y) the limit does exist and is unique, regardless of the way how exactly we divide the domain of our function into small rectangles, as long as their dimensions are converging to zero (or, more precisely, the dimensions of the largest among them converges to zero).

Now let's consider how we can summarize the volumes of all parallelepipeds constructed on the basis of small rectangles. We cannot summarize all x_i and y_j, as we did in a case of a rectangular base, because many of them will go outside of a circle where the function is undefined.
Instead, let's use the following logic. We will choose one of the arguments, let's say x, as the "primary" and will change it from x₀=−R to x_M=R, that is through all values of x_i. For each chosen x=x_i we will choose only those values of y that fall within a circle of radius R, that is from −√R²−x² to √R²−x².

Therefore, while the limits of summation by "primary" argument x are all its values from x₀=−R to x_M=R, the limits of summation by "secondary" argument y depend on the value of "primary" x as specified above.
This dependence of the limits of summation (and, subsequently, integration) by secondary argument on the value of the primary one is the only difference of this case of circular base from the previously considered rectangular base.

Decreasing the intervals between division points by both arguments and going to a limit, we will have, firstly, an integral by a primary argument x in limits from −R to R and, for each value of x we will have to integrate by a secondary argument y in limits that depend on the value of primary argument x - from −√R²−x² to √R²−x².

The final formula would look like this:
∫_−R^R [∫_{−√R²−x²}^√R²−x² f(x, y) dy] dx

Let's use this formula for calculating the volume of a cylinder with f(x, y)=H - the height of a cylinder.
According to the rules of geometry, the volume of a cylinder equals to a product of an area of a base by its height, so we have to get the answer V=πR²H.
Let's check if we get it through integration.

∫_−R^R [∫_{−√R²−x²}^√R²−x² H dy] dx = ∫_−R^R [2H√R²−x²] dx
Substitute
x=R·sin(t),
where −π/2 ≤ t ≤ π/2
Then √R²−x² = R·cos(t),
dx=R·cos(t) dt,
getting
∫_−π/2^π/2[2HRcos(t)]R·cos(t)dt = 2HR²∫_−π/2^π/2cos²(t)dt

To simplify it further, use the trigonometric identity
cos(2t) = cos²(t)−sin²(t) =
= 2cos²(t) −1
from which we derive
cos²(t) = [cos(2t)+1]/2
Our integral now looks like this
2HR²∫_−π/2^π/2[cos(2t)+1]/2 dt =
= HR²∫_−π/2^π/2[cos(2t)+1] dt =
= HR²∫_−π/2^π/2cos(2t) dt + HR²∫_−π/2^π/2 dt
Indefinite integral (anti-derivative) of cos(2t) is (1/2)·sin(2t).
Therefore, the first integral in this sum equals to
∫_−π/2^π/2cos(2t) dt = (1/2)·[sin(2π/2)−sin(−2π/2)] = (1/2)·[sin(π)−sin(−π)] = 0

The second integral in a sum above equals to
∫_−π/2^π/2 dt = π/2 − (−π/2) = π
Therefore, the result of integration is
HR²·π = πR²H which corresponds to results from geometry.

Unizor - Double Integrals - Volume of Solid

Notes to a video lecture on http://www.unizor.com

Double Integrals -
Volume of a Solid

Consider the following problem.
Given a smooth function of two variables f(x, y) (we will always consider smooth functions in terms of continuity and sufficient differentiability), with non-negative values (that is, f(x, y) ≥ 0), defined on a closed rectangle a ≤ x ≤ b, c ≤ y ≤ d and represented by a surface in (X,Y,Z) coordinate space.



In the following we will use the word "area" in a sense of a two-dimensional part of a plane and as a quantitative measure of this part of plane. The context would clarify which one it is used in every case.

Our task is to find the "volume of a solid" - the measure of a part of coordinate space bounded on the top by a surface representing this function, on the bottom - the X,Y-plane and on the four sides by planes x = a, x = b, y = c and y = d.

There is no ready to use formula for such a volume. We do know how the volume of a rectangular parallelepiped is defined, it is a product of its three dimensions - length multiplied by width, multiplied by height, but not of such a complicated figure as the one we consider now.

We really have to define what the volume of this figure is and then attempt to calculate it based on values of function f(x, y) on a rectangle [a, b, c, d].
We did have a similar problem in Geometry with the volume of solids and approached it as a sequence of approximations of a complex figure with simple ones. Let's do the same now.

We will use certain intuitive considerations to define the volume of a solid and will prove that this definition is mathematically valid.

The process of approximation starts with dividing rectangle [a, b, c, d] into many small rectangles by lines x=x_i (where i∈[0,M], x₀=a, x_M=b) and y=y_j(where j∈[0,N], y₀=c, y_N=d) and constructing M·N rectangular parallelepipeds on the base of these small rectangles with the height equaled to a value of function f(x, y) in the corner of each base rectangle with largest coordinates x and y ("top right corner").

If point (x_i, y_j) represents the coordinates of a "top right" corner of a small base rectangle and Δx_i=x_i−x_i−1, Δy_j=y_j−y_j−1are sides of this small rectangle, the volume of a constructed rectangular parallelepiped equals to
ΔV_i,j = f(x_i, y_j)·Δx_i·Δy_j

The combined volume of all these parallelepipeds approximates the volume of our solid. With increasing number of small base rectangles and, correspondingly, decreasing their dimensions we assume that there is a limit of the combined volume of all these parallelepipeds.
It can be proven that under certain conditions of smoothness of our function f(x, y) the limit does exist and is unique, regardless of the way how exactly we divide the domain of our function into small rectangles, as long as their dimensions are converging to zero (or, more precisely, the =dimensions of the largest among them converges to zero).

Now let's consider how we can summarize the volumes of all parallelepipeds constructed on the basis of small rectangles. There are two simple methods.

1. We can go from one point y=y_j to another, varying index j from 1 to N and for each value of y=y_j summarize the volumes of all parallelepipeds based on all values of x=x_i, varying index i from 1 to M. So, for each index j we calculate the sum of all V_i,j, varying index i.
Then we summarize these sums by index j deriving with an approximate volume
V_M·N = Σ_j∈[1,N] Σ_i∈[1,M] V_i,j =
= Σ_j Σ_i f(x_i, y_j)·Δx_i·Δy_j =
= Σ_j [Σ_i f(x_i, y_j)·Δx_i]·Δy_j

2. On the other hand, we can go from one point x=x_i to another, varying index i from 1 to M and for each value of x=x_isummarize the volumes of all parallelepipeds based on all values of y=y_j, varying index j from 1 to N. So, for each index i we calculate the sum of all V_i,j, varying index j.
Then we summarize these sums by index i deriving with an approximate volume
V_M·N = Σ_i∈[1,M] Σ_j∈[1,N] V_i,j =
= Σ_i Σ_j f(x_i, y_j)·Δx_i·Δy_j =
= Σ_i [Σ_j f(x_i, y_j)·Δy_j]·Δx_i

Both approaches to summarizing the volumes of small parallelepipeds should result in the same value of a combined volume.

Let's examine now what happens with both our formulas for total volume when the rectangles are getting smaller and smaller.

Recall the definition of a definite integral of function φ(t) on a segment p ≤ t ≤ q as a limit of sums of areas of rectangles obtained by dividing segment [p,q] into small pieces:
Σ_k φ(t_k)·Δt_k → ∫_p^q φ(t) dt

We can now say that
Σ_j f(x_i, y_j)·Δy_j converges to
φ(x_i) = lim Σ_j f(x_i, y_j)·Δy_j =
= ∫_c^d f(x_i, y) dy
under condition of maximum among Δy_j converges to zero. Then we continue with the second summation getting
Σ_i [Σ_j f(x_i, y_j)·Δy_j]·Δx_i →
→ Σ_i [∫_c^d f(x_i, y) dy]·Δx_i →
→ ∫_a^b[∫_c^d f(x_i, y) dy] dx

On the other hand, we can similarly state that
Σ_i f(x_i, y_j)·Δx_i converges to
ψ(y_j) = lim Σ_i f(x_i, y_j)·Δx_i =
= ∫_a^b f(x, y_j) dx
under condition of maximum among Δx_i converges to zero. Then we continue with the second summation getting
Σ_j [Σ_i f(x_i, y_j)·Δx_i]·Δy_j →
→ Σ_j [∫_a^b f(x, y_j) dx]·Δy_j →
→ ∫_c^d[∫_a^b f(x, y) dx] dy

But these are supposed to be the same results since the difference is only in the order of summation. The sums are the same, therefore the limits are the same, therefore we can write this limit as a double integral:
∫_c^d∫_a^b f(x, y) dx dy
or
∫_a^b∫_c^d f(x, y) dy dx

Both are the same and constitute two integrations of function f(x, y) performed consecutively, either by x from a to b (keeping y as a constant) and then by y from c to d or, in reverse, first by y from c to d (keeping x as a constant) and then by x from a to b.

Unizor - Partial Differential Equations - Heat Solution

Notes to a video lecture on http://www.unizor.com

Partial Differential Equations
Solution to Heat Equation

Our purpose is to solve the heat equation that describes the dynamics of temperature distribution within a thin rod and looks like this:
∂T(t,x)/∂t = a²·∂²T(t,x)/∂x²
where
t - time,
x - distance of a point on a thin rod from its edge,
T(t,x) - temperature at time t of a point on a rod at distance x from its edge,
a - constant that depends on physical characteristics of a rod.

This is a partial differential equation for function T(t,x) describing the distribution of temperature T within a thin rod at time t at location with X-coordinate x, assuming it's positioned along the X-axis with the left edge at x=0.

First of all, let's state that this distribution of temperature depends not only on the equation itself, which reflects dynamics of heat movement within a rod, but also depends on the initial conditions - what was the temperature at different locations of a rod at time t=0, described by function T(x,0).

Another type of condition that might be necessary to take into account is so-called boundary condition. It plays an important role in cases when edges of a rod are maintained at certain (not necessarily fixed) temperature all the time, which can be described by functions T(0,t) and T(L,t) (here L is the length of a rod).

Now we are ready to solve the equation.
Let's find a solution to our partial differential equation in a form
T(t,x) = f(t)·g(x)
We don't know if we will succeed on this way, but it's worth to try, since this relatively simple approach might bring us to solution. If it does, great. If it does not, at least, we tried.

Then
∂T(t,x)/∂t = f'(t)·g(x)
∂²T(t,x)/∂x² = f(t)·g''(x)
Now our heat equation looks like
f'(t)·g(x) = a²·f(t)·g''(x)
or
f'(t)/f(t) = a²·g''(x)/g(x)
Now we have a peculiar situation, when a function of t is equal to a function of x. The only possibility of this might be if both functions are constants that are equal to each other.
Therefore,
f'(t)/f(t) = A and
a²·g''(x)/g(x) = A
where A - some unknown constant that might be determined using initial conditions.

These are two ordinary differential equations, both types were addressed in the previous lectures. Let's get to their solutions.

The equation for f(t) is solved as follows:
df/f = A·dt
Integrating both sides results in
ln|f| = A·t + C
(where C - any constant)
|f(t)| = e^A·t+C
or, considering C is any constant,
f(t) = C·e^A·t

It's easy to check that this is a solution since
f'(t) = C·A·e^A·t and
f'(t)/f(t) = A

The equation for g(x) is a linear equation of the second order, considered in a prior lecture about oscillation of a spring.
We can represent it as
g''(x) − (A/a²)·g(x) = 0
We will look for its solution in a form
g(x) = e^λ·x
where λ - any complex number.
Since
g''(x) = λ²·e^λ·x
we come to a simple quadratic equation for λ:
λ² − A/a² = 0
with solutions
λ = ±(1/a)·√A
This results in solutions for g(x):
g(x) = C₁·e^{(1/a)·√A·x} + C₂·e^{−(1/a)·√A·x}

General solution to a heat equation can now be represented as
T(t,x) = f(t)·g(x) =
= C·e^A·t·(C₁·e^{(1/a)·√A·x} + C₂·e^{−(1/a)·√A·x})
where C, C₁ and C₂ are unknown constants, that can be defined only if additional conditions (initial conditions or boundary conditions) are given.

What remains is analysis of our solution for a purpose to get only real component of it, ignoring imaginary part, and apply initial condition on distribution of temperatures at time t=0, like T(0,x)=u(x), and boundary conditions (if applicable) T(t,0)=T₀ and T(t,L)=T_L in order to determine unknown constants. These are purely technical issues and lie outside the scope of this course.

Unizor - Partial Differential Equations - Heat Equation

Notes to a video lecture on http://www.unizor.com

Partial Differential Equations
Overview - Heat Equation

Since we talked about ordinary differential equations, where a function of one argument together with its derivatives participates in the equation, we might as well touch on differential equations, where function of two (or more) arguments together with its partial derivatives participates in the equation. These equations are called partial differential equations.

In order to present this material in a more practical light and to emphasize the importance of differential equations (particularly, partial) in practical applications, let's discuss one particular physical process that naturally leads to partial differential equations.

Let's imagine an insulated thin metal rod heated on one end and examine how its temperature T at different distance from the edge changes with time.
We assume that the rod is stretched along the X-axis with its heated end at origin. The distance from this end to any point on a rod is X-coordinate of this point. Time t will be measured from the moment the heat is applied to the edge at x=0.

Considering our rod is insulated and thin, we can safely assume that the heat dissipates only along its length from point x=0 towards the other edge, and the rod's temperature is a function of two arguments:
distance x from the edge
and time t:
T = T(t,x)
The illustration below will be handy in our analysis.


Before embarking on calculations let us remind a few physical properties that affect the process of heat dissipation.

1. Heat Q - a form of energy manifested in vibration of molecules of an object.
It is measured in the same units as energy.

2. Temperature T - a measure of intensity of the molecular vibration.
It is measured in degrees of different scales.

3. Specific heat capacity C - an amount of energy (heat) needed to increase temperature of a unit of mass by a unit of temperature.
It is assumed to be a constant for any specific material within reasonable range of temperatures and precision.
If an object of mass m increased in temperature by ΔT, it consumed ΔQ=C·m·ΔT units of heat (energy).

4. Thermal conductivity k - a measure of how fast molecular vibration is transferred within an object.
It was experimentally established that the amount of heat transferred from one zone of an object to another per unit of time through a unit of area on the boundary between these zones is proportional to a rate of change of temperature at this boundary and depends on the qualities of the material. For every material it is a constant within a reasonable range of temperatures.
In one-dimensional case on an above picture of a thin rod that has area S of crosscut, when the heat is transformed through a point x and the rate of change of temperature at that point and at that time is ∂T(t,x)/∂x, the amount of heat that goes through during time Δt can be calculated as
ΔQ = −k·S·Δt·∂T(t,x)/∂x
Minus in this equality reflects that heat is transferred from hot to cold area and, therefore, the partial derivative of temperature T by x is negative, while energy must be positive.

Now we are ready to connect all the parameters mentioned above into one heat equation.

Consider a part of a thin rod from point x to point x+Δx.
As the heat dissipates from left to right, during the time Δt certain amount of heat enters this area through crossing at point x and certain amount of heat exits this area through crossing at point x+Δx.

The heat entering through crossing at point x measures
ΔQ_in = −k·S·Δt·∂T(t,x)/∂x
The heat exiting through crossing at point x+Δx measures
ΔQ_out = −k·S·Δt·∂T(t,x+Δx)/∂x

The difference between them is an amount of heat that contributed to a rise in temperature of the part of a rod from x to x+Δx. This difference equals to
(A) ΔQ₊ = k·S·Δt·[∂T(t,x+Δx)/∂x − ∂T(t,x)/∂x]

On the other hand, the heat ΔQ₊, consumed by a part of a rod of length Δx, area of crosscut S and specific heat capacity C should increase the temperature by ΔT related to this heat as
ΔQ₊ = C·m·ΔT
where m is a mass of this part of a rod, that can be calculated as
m = ρ·S·Δx
where ρ is a mass of a unit of volume of the material this rod is made of.
Therefore,
(B) ΔQ₊ = C·ρ·S·Δx·ΔT

From equations (A) and (B) we derive the following equality:
C·ρ·S·Δx·ΔT = k·S·Δt·[∂T(t,x+Δx)/∂x − ∂T(t,x)/∂x]
This can be transformed into
ΔT/Δt = [k/(C·ρ)]·[∂T(t,x+Δx)/∂x−∂T(t,x)/∂x]/Δx

When Δt→0 and Δx→0 our equation can be represented as
∂T(t,x)/∂t = [k/(C·ρ)]·∂²T(t,x)/∂x²

Traditionally, this heat equation is written as
∂T(t,x)/∂t = a²·∂²T(t,x)/∂x²
where a² = k/(C·ρ)

Unizor - Ordinary Differential Equations - Hooke's Law

Notes to a video lecture on http://www.unizor.com

Higher Order Ordinary
Differential Equations -
Hooke'sLaw

Our next subject is Hooke's Law.
This law describes the force of a stretched or compressed spring.

Let's assume that we have a weightless spring horizontally lying on the frictionless table along an imaginary X-axis and fixed at the left end. Its free right end is at coordinate x=0 and there is a point mass m attached to this free end of a spring.
Then we stretch this spring by pulling the right end from a neutral position by certain length x.

Obviously, the spring exerts a force to compress back to a neutral position. The Hooke's Law states that within certain reasonable boundaries (no over-stretching) this force is proportional to a difference in length between a stretched string and a string in a neutral position.
This is expressed by the formula
F = −k·x
where F is the force exerted by a spring, x is a displacement of the free end of a spring from a neutral position, k is a positive constant that characterizes a spring (called a spring constant) and the minus sign signifies that the direction of force is opposite to the direction of displacement because, if displacement is positive (stretching), the force is directed towards negative direction of the X-axis and, if displacement is negative (compression), the force is directed towards positive direction of the X-axis.

Now recall Newton's Second Law that related the force and acceleration
F = m·a
where F is the force, m is the mass of an object and a is its acceleration.

From these two laws we conclude that
m·a = −k·x

Since x is a distance along the X-axis and a is an acceleration along this axis, that is a second derivative from a distance by time, we came up with the following differential equation
m·x''(t) = −k·x or
m·x''(t) + k·x(t) = 0 or
x''(t) + (k/m)·x(t) = 0
This is a second order ordinary differential equation. It is a little more complex than we considered in a lecture about acceleration and Newton's Second Law.
Let's try to solve it.

First of all, let us mention that even a simple guessing in this and many other cases is a good choice. Recall that first derivative of sin() is cos() and the first derivative of cos(), that is the same as the second derivative of sin(), is −sin(). So, the equation x''(t)+x(t)=0 has a solution x=sin(t). This is very close to what we have. Adding a factor α to an argument might help to satisfy multipliers in our equation:
if x(t)=sin(α·t) then
x'(t)=α·cos(α·t) and
x''(t) = −α²·sin(α·t)
and, therefore,
x''(t)+α²·x(t) = 0
Now we can choose α to satisfy α²=k/m, and the solution to our equation is found.

Guessing is good, when we can guess (as in this case), but guessing might not be successful and, even if you managed to guess one solution, it's not a guarantee that all solutions are found. By the way, if we start with cos(), we will also find a solution.
So, let's have some theory.

Our differential equation belongs to a class of linear ordinary differential equations of second order with constant coefficients and can be generalized as
x''(t) + p·x'(t) + q·x(t) = 0

As we saw above, functions sin() and cos() might be involved in a solution. Analogous quality of derivative being similar to a function itself is possessed by exponential functions. Recall also that exponential functions with complex exponent is related to trigonometric function through famous Euler's formula
e^it = cos(t) + i·sin(t)
So, exponential functions, in some way, are more general than trigonometric, they encompass them.
Therefore, it's only natural to look for a solution in terms of exponential functions.
Let's try.

Assume, we are looking for a solution to our equation in the form x(t)=e^λ·t, where λ might be any (including complex to accommodate trigonometric functions) number.
Then derivatives of this function are:
x'(t) = λ·e^λ·t
x''(t) = λ²·e^λ·t
Putting this into our equations, we get
λ²·e^λ·t + λ·p·e^λ·t + q·e^λ·t = 0
Canceling e^λ·t, we get a simple quadratic equation for λ called a characteristic polynomial of a given differential equation:
λ² + p·λ + q = 0
Since this equation always has two solutions λ₁ and λ₂ among complex numbers, we will have two particular solutions to our differential equation:
e^λ₁·t and e^λ₂·t
Finally, any linear combination of these two particular solutions will also be a solution (since our differential equation is linear and a derivative of linear combination of functions is a linear combination of derivatives).
Therefore, we can state the general solution to our differential equation:
x(t) = C₁·e^λ₁·t+C₂·e^λ₂·t
which depends on two unknown complex constants C₁and C₂, their values can be defined only if some initial conditions of the movement are given.

Let's get back to a movement of a spring.
Our initial equation
x''(t) + (k/m)·x(t) = 0
has a characteristic polynomial
λ² + k/m = 0
(where both k and m are positive) with two solutions:
λ₁ = √(k/m)·i and
λ₂ = −√(k/m)·i
where i²=−1 is an imaginary unit in the field of complex numbers.

Let ω = √(k/m).
Now we can represent the general solution to a movement of a spring based on the Hooke's Law as follows:
x(t) = C₁·e^iωt + C₂·e^−iωt
This expression can be easily transformed using Euler's formula into
x(t) = C₁·cos(ωt)+i·C₁·sin(ωt)+
+C₂·cos(−ωt)+i·C₂·sin(−ωt)

Since C₁ can be represented as A₁+i·B₁ and C₂ can be represented as A₂+i·B₂, where A₁, B₁, A₂ and B₂ are undefined unknown real numbers, the whole expression can be represented as
D₁·cos(ωt) + D₂·sin(ωt) + i·Z
where coefficients D₁ and D₂are any real numbers and i·Z represents purely imaginary part.

Since we deal with physics, we should exclude all imaginary solutions and leave only those, where D₁ and D₂ are real numbers.
So, the general physical solution looks like
x(t) = D₁·cos(ωt) + D₂·sin(ωt)
where D₁ and D₂ are undefined unknown real numbers.

In our experiment we have stretched a string by some known distance from a neutral position and let it spring back. That means, we know the initial position x(0)=d and initial speed x'(0)=0.
These initial conditions are sufficient to determine two unknown constants in our equation of a motion:
x(0)=d ⇒
⇒ D₁cos(0)+D₂sin(0)=d
⇒ D₁ = d
x'(0)=0 ⇒
⇒ −ω·D₁sin(0)+ω·D₂cos(0)=0
⇒ D₂ = 0

The final form of an equation of motion is
x(t) = d·cos(ωt)
where ω = √(k/m), k is a spring constant, m is a point mass at its free end and d is the initial distance we have stretched a spring from its neutral position.
As seen from this equation of motion, a free end of a spring with a mass attached to it will indefinitely oscillate around the neutral point.
The end.

Unizor - Ordinary Differential Equations - Acceleration

Notes to a video lecture on http://www.unizor.com

Higher Order Ordinary
Differential Equations -
Acceleration

Differential equations can include derivatives of higher order - second derivative, third, etc.
Probably, most common equations of this type are those with the second order derivative.
These equations are very often occur in science, especially in Physics. Let's address these equations and approaches to solve them.

Our first subject is a concept of acceleration and Newton's Second Law.

Recall that speed measures how fast a distance from some starting point changes, that is, if this distance is represented as a function x(t) of time t, speedv(t) at any moment t is the first derivative of distance by time:
v(t) = x'(t) = dx/dt

But speed does not have to be constant, we can move faster, increasing our speed (accelerating), or slower, decreasing it (decelerating).
To measure how fast our speed changes with time, as usually, when we want to measure how fast anything changes with time, we use a derivative.
Differentiating speed (a function of time) by time we obtain this measure of change of speed at any moment. This derivative of speed by time is called acceleration a(t):
a(t) = v'(t) = dv/dt =
= x''(t) = d²x/dx²
That is, acceleration is the second derivative of distance by time.

Newton's Second Law states that the force F applied to an object and the acceleration a this object obtains as a result of this application of force are related as follows:
F = m·a
where m is the object's mass (presumed constant).

Assuming that our motion occurs along a straight line with coordinates and, therefore, the position of an object is defined by its X-coordinate x(t), Newton's Second Law is an ordinary differential equation of second order because acceleration is the second derivative of the X-coordinate of an object:
F(t) = m·x''(t)
Usually our task is to find where exactly our object is located (that is, its X-coordinate), if the force, as a function of time, is given.

Consider a case when there is no force applied to an object, that is F(t)=0.
Then, according to the Newton's Second Law,
0 = m·a(t)
from which we derive a(t)=0
Since a(t)=v'(t), we can find the speed:
v'(t)=0
⇒ v(t) = C
(where C is an unknown constant)
⇒ x'(t) = C
⇒ x(t) = C·t + D
(where D is another unknown constant)

That concludes the solution of our differential equation of the second order, and the solution includes two unknown constants that cannot be determined from the equation alone. It's understandable since we don't know initial position of an object on the coordinate axis x(0) and the initial speed it moved v(0). These two additional pieces of information (initial conditions) are needed to determine unknown constants participating in the solution.
If x(0)=x₀ and v(0)=v₀, we can easily determine
x₀ = D and
v₀ = C
which results in the final equation of motion of an object, to which no forces are applied (or, more generally, all forces applied to it are balancing each other).
x(t) = x₀ + v₀·t

By solving the above differential equation of the second order, we have mathematically derived the Newton's First Law as a consequence of the Second Law.
Newton's First Law (law of inertia) states that if the sum of all forces applied to an object is zero, then the object at rest will continue to stay at rest (its speed is and will be 0) and objects moving at some speed will continue to move with the same speed and direction (its speed is constant).

Now consider a case when the force applied to an object is not zero, but constant, that is F(t)=P (const). Let's attempt to solve our differential equation in this case to determine the coordinate of an object as a function of time x(t).
F(t)=P=m·a(t)=m·v'(t)
where P is a known constant.
This implies that acceleration a(t) must be a known constant and equals to P/m. Let's use symbol a instead of a(t) to signify this.
Since a=v'(t), we derive
v(t) = a·t + C
where C is an unknown constant.
Then
x'(t) = a·t + C
⇒ x(t) = a·t²/2 + C·t + D
where D is another unknown constant.
To determine two unknown constants we need additional information - initial conditions.
Assume that the original position of an object is x(0)=x₀. This allows to determine D=x₀.
If initial speed v(0)=v₀ is known, we can determine C=v₀.
So, the final equation of the motion, when a constant force is applied is
x(t) = a·t²/2 + v₀·t + x₀

In general, if the force is variable and/or the mass is variable, from Newton's Second Law we can construct a differential equation of the second order, where the second derivative is explicitly represented by a known function:
F(t) = m(t)·x''(t)
⇒ x''(t) = F(t)/m(t)
⇒ d/dt[x'(t)] = F(t)/m(t)
⇒ x'(t) = ∫[F(t)/m(t)]dt
⇒ x(t) = ∫{∫[F(t)/m(t)]dt}dt

As we see, Newton's Second Law presents the simplest kind of ordinary differential equation of the second order. It can be solved by double integration.
It should not be forgotten that in the process of each integration there will appear an unknown constant, to get its value an initial condition should be known and applied.

Unizor - Ordinary Differential Equations - Linear Equations

Notes to a video lecture on http://www.unizor.com

Linear Ordinary Differential Equations

Standard form of linear ordinary differential equations is
f(x)·y' + g(x)·y + h(x) = 0
As the first step, we can divide all members of this equation by f(x) (assuming it's not identically equal to 0), getting a simpler equation
y'+u(x)·y+v(x) = 0
The suggested solution lies in the substitution y(x)=p(x)·q(x), where p(x) and q(x) are unknown (for now) functions.
Express y'(x) in terms of p(x) and q(x):
y' = p·q'+q·p'
Substitute this into our equation:
p·q'+q·p'+u·p·q+v = 0
Let's simplify this
p(q'+u·q)+q·p'+v = 0
If there are such functions p(x) and q(x) that satisfy conditions
(1) q'+u·q = 0 and
(2) q·p'+v = 0
our job would be finished.
Let's try to find such functions.
From the equation (1) in our pair of equations we derive
q'/q = −u,
which can be converted into
dq(x)/q(x) = −u(x)·dx
that can be solved by integrating:
ln(q(x)) = −∫u(x)·dx
q(x) = e^{−∫u(x)·dx}
Once q(x) is found, we solve the equation (2) for p(x):
p'(x) = −v(x)/q(x),
which can be integrated to find
p(x) = −∫v(x)/q(x) dx
and, consequently, y(x)=p(x)·q(x) can be fully determined.
Let's consider a few examples.

Example 1

Solve the following linear differential equation
y' + y + x = 0

Let's look for a solution in a form
y(x)=p(x)·q(x)
Then
y'(x) = p'(x)·q(x)+p(x)·q'(x)
Our equation looks like this now
p'·q+p·q' + p·q + x = 0
Factor out p, getting
p·(q'+q) + (p'·q+x) = 0
We will try to find p(x) and q(x) to separately bring to zero q'+q and p'·q+x.
Let's look for a function q(x) that brings expression q'+q to zero:
q'+q = 0
dq/q = −dx
∫dq/q = −∫dx
ln(q) = −x + A
(where A is any constant)
q(x) = B·e^−x
(where B=e^A, so it represents any positive number)
Next, let's find a solution to
p'·q+x = 0
p'·B·e^−x+x = 0
p(x) = −(1/B)·∫x·e^xdx
This integral can be found using the "by parts" technique:
p(x) = −(1/B)·(x·e^x−∫e^xdx) =
= −(1/B)·(x·e^x−e^x−C) =
= −(1/B)·(x−1)·e^x+C/B
(where B is any positive constant and C is any constants)
Now let's find y(x)=p(x)·q(x):
y(x) = [−(1/B)·(x−1)·e^x+C/B]·[B·e^−x] =
= 1−x+C·e^-x

Checking:
y'(x) = −1 − C·e^-x
y'+y+x = −1 − C·e^-x+1−x+C·e^-x+x = 0
Solution was correct.

Example 2

Solve the following linear differential equation
y'·cos(x) + y·sin(x) − 1 = 0

First of all, let's normalize it by dividing by cos(x), noticing that sin(x)/cos(x)=tan(x) and 1/cos(x)=sec(x):
y' + y·tan(x) − sec(x) = 0
Let's look for a solution to this equation in a form
y(x)=p(x)·q(x)
Then
y'(x) = p'(x)·q(x)+p(x)·q'(x)
Our equation looks like this now:
p'·q+p·q'+p·q·tan(x)−sec(x) = 0
Factor out p, getting
p·(q'+q·tan(x))+p'·q−sec(x) = 0
First, let's find function q(x) such that
q' + q·tan(x) = 0
It can be solved using the technique of separation:
dq/q = −tan(x)dx
Since [ln(x)]' = 1/x and [cos(x)]' = −sin(x), the last equation can be transformed into
d(ln(q)) = d(cos(x))/cos(x)
d(ln(q)) = d(ln(cos(x)))
Now it's easy to integrate, the result is
ln(q) = ln(cos(x))+C
where C - any real number, from which, raising number e to both left and right sides, follows that
q = D·cos(x)
(new constant D=e^C represents any positive number)
Now let's find function p(x) such that
p'·q − sec(x) = 0
Substitute already found q(x) getting
p'·D·cos(x) − sec(x) = 0
p'(x) = (1/D)·(1/cos²(x))
p(x) = (1/D)·∫dx/cos²(x) =
= (1/D)·[tan(x) + E]
where D and E are constants (D - any positive, E - any real)
Let's determine y=p·q now.
y(x) = p(x)·q(x) =
= (1/D)·[tan(x) + E]·D·cos(x) =
= sin(x) + E·cos(x)

Let's check this result.
y'(x) = cos(x)−E·sin(x)
y'·cos(x) + y·sin(x) − 1 =
= cos²(x)−E·sin(x)·cos(x) + sin²(x)+E·cos(x)·sin(x)−1 =
= sin²(x) + cos²(x) −1 = 0
which proves the correctness of our answer.


Example 3

Solve the following differential equation
ln(x·y'+y) = ln(2x)+x²

It's not linear, but can be made linear if we raise e to a power defined by its left and right sides, getting
x·y'+y = 2x·e^x²
Let's normalize it by dividing by x:
y' + y/x = 2e^x²
Now it's a linear equation that we know how to solve.
Let's look for a solution in a form
y(x)=p(x)·q(x)
Then
y'(x) = p'(x)·q(x)+p(x)·q'(x)
Our equation looks like this now
p'·q+p·q' + p·q/x = 2e^x²
Factor out p, getting
p·(q'+q/x) + p'·q = 2e^x²
We will try to find p(x) and q(x) to separately
(a) bring to zero q'+q/x and
(b) equalize p'·q with 2e^x².
Let's solve equation (a) and look for a function q(x) that brings expression q'+q/x to zero:
q'+q/x = 0
dq/q = −dx/x
∫dq/q = −∫dx/x
ln(|q|) = −ln(|x|) + A
where A - any constant.
Raising e to both sides of this equation, we get
|q| = B/|x|
where B=e^A - any positive number.
Let's get rid of absolute values in the above equation by allowing B to be any non-zero real number, so
q = B/x
Substitute it to equation (b):
p'·B/x = 2e^x²
p(x) = (1/B)∫2x·e^x²·dx
Since derivative of x² is 2x,
p(x) = (1/B)∫e^x²·d(x²)
Now we can integrate directly:
p(x) = (1/B)e^x² + C
where C is any real number.
This allows to express the solution to our differential equation in the form
y(x) = p(x)·q(x)
where
p(x) = (1/B)e^x² + C and
q(x) = B/x
That produces
y = e^x²/x + C/x = (e^x²+C)/x
where C - any real number.

Checking:
y' = −(e^x²+C)/x² + e^x²·2x/x =
= e^x²·(2−1/x²) −C/x²
y/x = e^x²/x² + C/x²
y' + y/x = 2e^x²,
which corresponds to the original equation after multiplying both sides by x and taking logarithm.
The end.