Sturm-Liouville Problem/Unit Weight Function

From ProofWiki
Jump to navigation Jump to search


This article needs to be linked to other articles.
You can help $\mathsf{Pr} \infty \mathsf{fWiki}$ by adding these links.
To discuss this page in more detail, feel free to use the talk page.
When this work has been completed, you may remove this instance of {{MissingLinks}} from the code.



This article needs proofreading.
Please check it for mathematical errors.
If you believe there are none, please remove {{Proofread}} from the code.
To discuss this page in more detail, feel free to use the talk page.
When this work has been completed, you may remove this instance of {{Proofread}} from the code.



This page has been identified as a candidate for refactoring of basic complexity.
In particular: Lemmata into their own pages
Until this has been finished, please leave {{Refactor}} in the code.

New contributors: Refactoring is a task which is expected to be undertaken by experienced editors only.

Because of the underlying complexity of the work needed, it is recommended that you do not embark on a refactoring task until you have become familiar with the structural nature of pages of $\mathsf{Pr} \infty \mathsf{fWiki}$.
To discuss this page in more detail, feel free to use the talk page.
When this work has been completed, you may remove this instance of {{Refactor}} from the code.


Theorem

Let $P, Q: \R \to \R$ be real mappings such that $P$ is smooth and positive, while $Q$ is continuous:

$\map P x \in C^\infty$
$\map P x > 0$
$\map Q x \in C^0$

Let the Sturm-Liouville equation, with $\map w x = 1$, be of the form:

$-\paren {P y'}' + Q y = \lambda y$

where $\lambda \in \R$.

Let it satisfy the following boundary conditions:

$\map y a = \map y b = 0$


Then all solutions of the Sturm-Liouville equation, together with their eigenvalues, form infinite sequences $\sequence {y^{\paren n} }$ and $\sequence {\lambda^{\paren n} }$.

Furthermore, each $\lambda^{\paren n}$ corresponds to an eigenfunction $y^{\paren n}$, unique up to a constant factor.


Proof

Outline

Firstly, an equivalence between the Sturm-Liouville equation and minimisation of functional $\ds J \sqbrk y = \int_a^b \paren {P y'^2 + Q y^2} \rd x$ problems is established.

Then, the lower bound of the functional $J$ is found, thus allowing $J$ to have a finite minimisation.

Afterwards, a trial minimizing sequence is chosen, and $J$ becomes a function of expansion coefficients. The more coefficients, (possibly) the lower value.

Sequences of trial mappings $\sequence {y_n^{\paren 1} }$ and values $\sequence {\lambda_n^{\paren 1} }$ of $J$ are introduced.

Convergence of $\sequence {\lambda_n^{\paren 1} }$ is shown at once.

As for $\sequence {y_n^{\paren 1} }$, convergence to $y^{\paren 1}$ is proved for its subsequence.

The rest of arguments rest upon this weaker result.


Furthermore, $\lambda^{\paren 1}$ and $y^{\paren 1}$ are observed to satisfy Sturm-Liouville equation.

Convergence of the original sequence $\sequence {y_n^{\paren 1} }$ is secured.

Finally, construction of the rest of eigenfunctions and eigenvalues is described.


This needs considerable tedious hard slog to complete it.
In particular: Maybe add a diagrammatical presentation of the whole proof with Lemmas and somehow enumerated intermediate passages
To discuss this page in more detail, feel free to use the talk page.
When this work has been completed, you may remove this instance of {{Finish}} from the code.
If you would welcome a second opinion as to whether your work is correct, add a call to {{Proofread}} the page.



Lemma 1

The given Sturm-Liouville equation is an Euler equation of the following functional:

$\ds J \sqbrk y = \int_a^b \paren {P y'^2 + Qy^2} \rd x$

constrained by a subsidiary condition:

$\ds \int_a^b y^2 \rd x = 1$


Proof

According to Simplest Variational Problem with Subsidiary Conditions, the following equation must hold:

$F_y - \dfrac \rd {\rd x} F_{y'} + \lambda \paren {G_y - \dfrac \rd {\rd x} G_{y'} } = 0$

where:

$F = P y'^2 + Q y^2$
$G = y^2$

Then the Euler equation reads:

$2 Q y - 2 \paren {P y'}' + 2 \lambda y = 0$

Division by $2$ and rearrangement of terms yields the desired result.

$\Box$


By Necessary Condition for Integral Functional to have Extremum for given function, if $y$ is an extremum of $J$, it is also a solution of the Sturm-Liouville equation.


Lemma 2

$J$ is bounded from below.


Proof

Since $Q$ is continuous on an interval, it is bounded.

Since $P > 0$, it holds that:

$\ds \int_a^b \paren {P y'^2 + Q y^2} \rd x > \int_a^b Q y^2 \rd x \ge M \int_a^b y^2 \rd x = M$

where:

$\ds M = \min_{a \mathop \le x \mathop \le b} \map Q x$

Therefore, $J$ is bounded from below.

$\Box$


Introduce a new variable:

$t := \pi \dfrac {x - a} {b - a}$

Then the interval of consideration $\closedint a b$ is mapped onto $\closedint 0 \pi$.

Choose Ritz sequence $\sequence {\map {\phi_n} t} = \sequence {\sin n t}$, where $n \in \N$.


Lemma 3

The elements of the sequence $\sequence {\sin n t}$ are orthogonal on the interval $\closedint 0 \pi$:

$\ds \int_0^\pi \map \sin {k t} \map \sin {l t} \rd x = \frac \pi 2 \delta_{k l}$


Proof

The product involves two elements of the sequence $\sequence {\map \sin {n t} }$.

Their indices either match each other or not.

Suppose $k = l$.

Then:

\(\ds \int_0^\pi \map {\sin^2} {k t} \rd t\) \(=\) \(\ds \int_0^\pi \frac {1 - \map \cos {2 k t} } 2 \rd t\) Square of Sine
\(\ds \) \(=\) \(\ds \intlimits {\frac \pi 2 - \frac {\map \sin {2 k t} } {4 k} } 0 \pi\)
\(\ds \) \(=\) \(\ds \frac \pi 2\) Sine of Multiple of Pi

Suppose $k \ne l$.

Then:

\(\ds \int_0^\pi \map \sin {l t} \map \sin {k t} \rd t\) \(=\) \(\ds \int_0^\pi \frac {\map \cos {\paren {l - k} t} - \map \cos {\paren {l + k} t} } 2 \rd t\) Werner Formula for Sine by Sine
\(\ds \) \(=\) \(\ds \intlimits {\frac 1 2 \paren {\frac {\map \sin {\paren {l - k} t} } {l - k} - \frac {\map \sin {\paren {l + k} t} } {l + k} } } 0 \pi\)
\(\ds \) \(=\) \(\ds 0\) Sine of Multiple of Pi

By Proof by Cases, the statement is proved.

$\Box$


Let the trial solution be of the following form:

$\ds \map y x = \sum_{k \mathop = 1}^n \alpha_k \map \sin {k \map t x}$

Trial solution has to satisfy boundary and subsidiary conditions.

Boundary conditions are satisfied without further requirements.

Subsidiary condition results into an additional constraint on coefficients $\alpha_k$:

\(\ds \int_a^b \paren {\sum_{k \mathop = 1}^n \alpha_k \map \sin {k \map t x} }^2 \rd x\) \(=\) \(\ds \frac \pi {b - a} \int_0^\pi \paren {\sum_{k \mathop = 1}^n \alpha_k \sin {k t} }^2 \rd t\)
\(\ds \) \(=\) \(\ds \frac \pi {b - a} \int_0^\pi \paren {\sum_{k \mathop = 1}^n \alpha_k \sin {k t} } \paren {\sum_{l \mathop = 1}^n \alpha_l \sin l x} \rd t\) Second summation label $k$ relabelled as $l$
\(\ds \) \(=\) \(\ds \frac \pi {b - a} \sum_{k, l \mathop = 1}^n \alpha_k \alpha_l \int_0^\pi \map \sin {k t} \map \sin {l t} \rd t\)
\(\ds \) \(=\) \(\ds \frac \pi {b - a} \sum_{k, l \mathop = 1}^n \alpha_k \alpha_l \frac \pi 2 \delta_{k l}\) as $\ds \int_0^\pi \map \sin {k t} \map \sin {l t} \rd x = \frac \pi 2 \delta_{k l}$
\(\ds \) \(=\) \(\ds \frac \pi {b - a} \frac \pi 2 \sum_{k \mathop = 1}^n \alpha_k^2\)
\(\ds \) \(=\) \(\ds 1\)


All the points $\boldsymbol \alpha$ constitute a set $\sigma_n$ which is a surface of an $n$-dimensional sphere, defined by the subsidiary condition.

For the assumed trial mapping the functional $\map {J_n} {\boldsymbol \alpha}$ reads as:

$\ds \map {J_n} {\boldsymbol \alpha} = \frac \pi {b - a} \int_0^\pi \sqbrk {\map P {\sum_{k \mathop = 1}^n \alpha_k \sin {k t} }'^2 + \map Q {\sum_{k \mathop = 1}^n \alpha_k \sin {k t} }^2} \rd t$

The integrand is a second order polynomial with respect to the components of $\boldsymbol \alpha$.

Hence, $J$ is continuous with respect to the components of $\boldsymbol \alpha$.

The components of $\boldsymbol \alpha$ constitute a closed and bounded set.

By definition, $\sigma_n$ is a compact set.

Thus, $\map {J_n} {\boldsymbol \alpha}$ is continuous on $\sigma_n$.

By Continuous Function on Compact Subspace of Euclidean Space is Bounded, $\map {J_n} {\boldsymbol \alpha}$ has a minimum on $\sigma_n$.

Let $\map {y_n^{\paren 1} } x$ be defined as:

$\ds \map {y_n^{\paren 1} } x = \sum_{k \mathop = 1}^n \alpha_k^{\paren 1} \sin {k \map t x}$

for which $\map {J_n} {\boldsymbol \alpha}$ achieves the minimum $\lambda_n^{\paren 1}$, unrelated to $\lambda$.

Then the $n$-th element of the sequence $\sequence {y_n^{\paren 1} }$ corresponds to the $n$-th element of the sequence of minima $\sequence {\lambda_n^{\paren 1} }$ of $\map {J_n} {\boldsymbol \alpha}$.

Since $\sigma_n \subset \sigma_{n + 1}$, where $\sigma_n$ has $\alpha_{n + 1} = 0$, it holds that:

$\map {J_n} {\alpha_1, \ldots, \alpha_n} = \map {J_{n + 1} } {\alpha_1, \ldots, \alpha_n, 0}$

By Ritz Method implies Not Worse Approximation with Increased Number of Functions:

$\lambda_{n + 1}^{\paren 1} \le \lambda_n^{\paren 1}$

Therefore, by increasing the domain of definition of $y_n^{\paren 1}$ through additional summands, the minima of $\map {J_n} {\boldsymbol \alpha}$ cannot increase.


This article, or a section of it, needs explaining.
In particular: Define "domain of definition"
You can help $\mathsf{Pr} \infty \mathsf{fWiki}$ by explaining it.
To discuss this page in more detail, feel free to use the talk page.
When this work has been completed, you may remove this instance of {{Explain}} from the code.


From the last inequality and $J$ being bounded from below it follows, that the following limit exists:

$\ds \lambda^{\paren 1} = \lim_{n \mathop \to \infty} \lambda_n^{\paren 1}$


Lemma 4

The sequence $\sequence {y_n^{\paren 1} }$ contains a uniformly convergent subsequence.


Proof

The sequence

$\ds \lambda_n^{\paren 1} = \frac \pi {b - a} \int_0^\pi \paren {P {y_n^{\paren 1} }'^2 + Q {y_n^{\paren 1} }^2} \rd t$

is convergent with its limit being $\lambda^{\paren 1}$.

Hence, it is bounded:

$\ds \frac \pi {b - a} \int_0^\pi \paren {P {y_n^{\paren 1} }'^2 + Q {y_n^{\paren 1} }^2} \rd t \le M $

Furthermore:

\(\ds \frac \pi {b - a} \int_0^\pi P {y_n^{\paren 1} }'^2 \rd t\) \(\le\) \(\ds M - \frac \pi {b - a} \int_0^\pi Q {y_n^{\paren 1} }^2 \rd t\)
\(\ds \) \(\le\) \(\ds M + \frac \pi {b - a} \size {\int_0^\pi Q {y_n^{\paren 1} }^2 \rd t}\)
\(\ds \) \(\le\) \(\ds M + \max_{a \mathop \le x \mathop \le b} \size {\map Q x} \size {\int_a^b {y_n^{\paren 1} }^2 \rd x}\)
\(\ds \) \(=\) \(\ds M + \max_{a \mathop \le x \mathop \le b} \size {\map Q x}\) as $\ds \int_a^b y^2 \rd x = 1$, where $y = y _n^{\paren 1}$ minimizes $\map {J_n} {\boldsymbol \alpha}$
\(\ds \) \(=\) \(\ds M_1\)

Consequently:

$\ds \frac \pi {b - a} \min_{a \mathop \le x \mathop \le b} \map P x \int_0^\pi \map { {y_n^{\paren 1} }'^2} t \rd t \le \frac \pi {b - a} \int_0^\pi P \map { {y_n^{\paren 1} }'^2} t \rd t \le M_1$

For positive $P$, division by $P$ does not affect the direction of inequality.

It follows that:

$\ds \int_0^\pi \map { {y_n^{\paren 1} }'^2} t \rd t \le \frac {b - a} \pi \frac {M_1} {\min_{a \mathop \le x \mathop \le b} } \map P x = M_2$

Consider squared absolute value of $y_n^{\paren 1}$.

Then, for $0 \le t \le \pi$:

\(\ds \size {\map {y_n^{\paren 1} } t}^2\) \(=\) \(\ds \size {\map {y_n^{\paren 1} } t - 0}^2\) as $\map {y_n^{\paren 1} } {t = 0} = 0$
\(\ds \) \(=\) \(\ds \size {\map {y_n^{\paren 1} } t - \map {y_n^{\paren 1} } 0}^2\)
\(\ds \) \(=\) \(\ds \size {\int_0^t \map { {y_n^{\paren 1} }'} \zeta \rd \zeta}^2\)
\(\ds \) \(\le\) \(\ds \int_0^t \map { {y_n^{\paren 1} }'^2} \zeta \rd \zeta \int_0^t \rd \zeta\) Cauchy-Bunyakovsky-Schwarz Inequality for Definite Integrals:
\(\ds \) \(\le\) \(\ds \int_0^\pi \map { {y_n^{\paren 1} }'^2} \zeta \rd \zeta \int_0^\pi \rd \zeta\) as all integrands are non-negative
\(\ds \) \(\le\) \(\ds M_2 \pi\)

In other words:

$\forall t \in \closedint 0 \pi, n \in \N : \size {\map {y_n^{\paren 1} } t - \map {y_n^{\paren 1} } 0} \le \sqrt {M_2 \pi}$

Thus $\sequence {y_n^{\paren 1} }$ is uniformly bounded.

In addition to this, for $0 \le t_1, t_2 \le \pi$:

\(\ds \size {\map {y_n^{\paren 1} } {t_2} - \map {y_n^{\paren 1} } {t_1} }^2\) \(=\) \(\ds \size {\int_{t_1}^{t_2} \map { {y_n^{\paren 1} }'} t \rd t}^2\) Fundamental Theorem of Calculus
\(\ds \) \(\le\) \(\ds \int_{t_1}^{t_2} {y_n^{\paren 1} }'^2 \rd x \size {\int_{t_1}^{t_2} \rd t}\) Cauchy-Bunyakovsky-Schwarz Inequality for Definite Integrals
\(\ds \) \(\le\) \(\ds \int_0^\pi {y_n^{\paren 1} }'^2 \rd x \size {\int_{t_1}^{t_2} \rd t}\)
\(\ds \) \(\le\) \(\ds M_2 \size {t_2 - t_1}\)

Let $\epsilon$ be any strictly positive real number such that $\epsilon = \sqrt {M_2 \delta}$, where $\delta$ is a strictly positive real number.

Suppose $\delta$ is such that $\size {t_2 - t_1} < \delta$.

Then:

\(\ds \size {\map {y_n^{\paren 1} } {t_2} - \map {y_n^{\paren 1} } {t_1} }\) \(\le\) \(\ds \sqrt {M_2 \size {t_2 - t_1} }\)
\(\ds \) \(<\) \(\ds \sqrt {M_2 \delta}\)
\(\ds \) \(=\) \(\ds \epsilon\)

In other words:

$\forall \epsilon \in \R_{>0}: \exists \delta \in \R_{>0}: \forall n \in \N: \forall t_1, t_2 \in \closedint 0 \pi: \size {t_2 - t_1} < \delta \implies \size {\map {y_n^{\paren 1} } {t_2} - \map {y_n^{\paren 1} } {t_1} } < \epsilon$

where metric is induced by norm.

Thus $\sequence {y_n^{\paren 1} }$ is uniformly equicontinuous.

By Arzela's Theorem, there exists a uniformly convergent subsequence $\sequence {y_{n_m}^{\paren 1} }$ from $\sequence {y_n^{\paren 1} }$.


$\Box$

Denote:

$\ds \map {y^{\paren 1} } x = \lim_{m \mathop \to \infty} \map {y_{n_m}^{\paren 1} } x$

Now a proposition is established, needed for the upcoming Lemma 6.


Lemma 5

Let $\map y t$ be continuous in $\closedint 0 \pi$.

Suppose:

$\ds \forall h \in C^2 \openint 0 \pi : \map h 0 = \map h \pi = \map {h'} 0 = \map {h'} \pi = 0 : \int_0^\pi \sqbrk {-\paren {P h'}' + Q_1 h} y \rd t = 0$


Then $\map y t \in C^2 \openint 0 \pi$ and:

$-\paren {P y'}' + Q_1 y = 0$


Proof

By Integration by parts, Product Rule for Derivatives, boundary conditions for $h$, and noticing that:

\(\ds \map \d {\int_0^t \map {f'} \xi \rd \xi}\) \(=\) \(\ds \map \d {\map f t - \map f 0}\) Fundamental Theorem of Calculus
\(\ds \) \(=\) \(\ds \map {f'} t \rd t\)
\(\ds \map \d {\int_0^t \paren {\int_0^\xi \map g \zeta \rd \zeta} \rd \xi}\) \(=\) \(\ds \paren {\int_0^t \map g \zeta \rd \zeta} \rd t\)
\(\ds \) \(=\) \(\ds \paren {\int_0^t \map g \xi \rd \xi} \rd t\)

the previous integral can be rewritten as:

\(\ds \int_0^\pi \paren {-\paren {P h'}' + Q_1 h} y \rd t\) \(=\) \(\ds -\int_0^\pi P h y \rd t - \int_0^\pi P' h' y \rd t + \int_0^\pi Q_1 h y \rd t\)
\(\ds \) \(=\) \(\ds -\int_0^\pi P h y \rd t - \int_0^\pi h' \rd \paren {\int_0^t P' y \rd \xi} + \int_0^\pi h \rd \paren {\int_0^t Q_1 y \rd \xi}\)
\(\ds \) \(=\) \(\ds -\int_0^\pi P h y \rd t - \paren {\intlimits {h' \int_0^t P' y \rd \xi} {t = 0} {t = \pi} - \int_0^\pi h \paren {\int_0^t P' y \rd \xi} \rd t} + \paren {\intlimits {h \int_0^t Q_1 y \rd \xi} {t \mathop = 0} {t \mathop = \pi} - \int_0^\pi h' \paren {\int_0^t Q_1 y \rd \xi} \rd t}\)
\(\ds \) \(=\) \(\ds -\int_0^\pi P h y \rd t + \int_0^\pi h \paren {\int_0^t P' y \rd \xi} \rd t - \int_0^\pi h' \paren {\int_0^t Q_1 y \rd \xi } \rd t\)
\(\ds \) \(=\) \(\ds -\int_0^\pi P h y \rd t + \int_0^\pi h \paren {\int_0^t P' y \rd \xi} \rd t - \paren {\int_0^\pi h' \rd \paren {\int_0^t \paren {\int_0^\xi Q_1 y \rd \zeta} \rd \xi} }\)
\(\ds \) \(=\) \(\ds -\int_0^\pi P h y \rd t + \int_0^\pi h \int_0^t P' y \rd \xi - \paren {\intlimits {h' \int_0^t \paren {\int_0^\xi Q_1 y \rd \zeta} \rd \xi} {t \mathop = 0} {t \mathop = \pi} - \int_0^\pi h \paren {\int_0^t \paren {\int_0^\xi Q_1 y \rd \zeta} \rd \xi} \rd t}\)
\(\ds \) \(=\) \(\ds -\int_0^\pi P h y \rd t + \int_0^\pi h \int_0^t P' y \rd \xi + \int_0^\pi h \paren {\int_0^t \paren {\int_0^\xi Q_1 y \rd \zeta} \rd \xi} \rd t\)
\(\ds \) \(=\) \(\ds \int_0^\pi \paren {- P y + \int_0^t P' y \rd \xi + \int_0^t \paren {\int_0^\xi Q_1 y \rd \zeta} \rd \xi} h \rd t\)
\(\ds \) \(=\) \(\ds 0\)

From the lemma:

$\ds -P y + \int_0^t P' y \rd \zeta + \int_0^x \paren {\int_0^t Q_1 y \rd t} \rd \zeta = c_0 + c_1 t$

The right hand side as well as the second and third terms on the left hand side are differentiable with respect to $t$.

Thus $\paren {P y}'$ exists.

Differentiation with respect to $t$ leads to:

$\ds -\paren {P y}' + P' y + \int_0^t Q_1 y \rd \zeta = c_1$

or:

$\ds -P y' + \int_0^t Q_1 y \rd \zeta = c_1$

The right hand side, the second term on the left hand side and $P$ are continuous and differentiable with respect to $t$, while $P$ is also positive.

Therefore, $y'$ exists and is continuous.

Hence, $\paren {P y'}'$ exists and:

$-\paren {P y'}' + Q_1 y = 0$

Furthermore, $P$ is continuous and differentiable, while $Q_1$ is continuous.

Then $y$ exists and is continuous.

$\Box$


Lemma 6

$y^{\paren 1}$ together with $\lambda^{\paren 1}$ satisfy the Sturm-Liouville equation, where $\map w x = 1$:

$-\paren {P {y^{\paren 1} }' }' + Q y^{\paren 1} = \lambda^{\paren 1} y^{\paren 1}$


Proof

Let $J_n$, together with its subsidiary condition $\ds \int_a^b y^2 \rd x = 1$, achieve a minimum for $\boldsymbol \alpha = \boldsymbol \alpha^{\paren 1}$.

Then the necessary condition for its minimum is:

\(\ds \map {\dfrac \partial {\partial \alpha_r} } {\map {J_n} {\boldsymbol \alpha} - \lambda_n^{\paren 1} \int_a^b y^2 \rd x}\) \(=\) \(\ds \map {\dfrac \partial {\partial \alpha_r} } {\map {J_n} {\boldsymbol \alpha} - \lambda_n^{\paren 1} \frac \pi {b - a} \int_0^\pi \paren {\sum_{k \mathop = 1}^n \alpha_k \sin {k x} }^2 \rd x}\)
\(\ds \) \(=\) \(\ds 0\)

Notice, that:

\(\ds \dfrac \partial {\partial \alpha_r} \paren {\sum_{k \mathop = 1}^n \alpha_k \sin {k x} }^2\) \(=\) \(\ds \dfrac {\partial \paren {\sum_{k \mathop = 1}^n \alpha_k \sin {k x} }^2} {\partial \sum_{k = 1}^n \alpha_k \sin {k x} } \dfrac {\partial \sum_{j \mathop = 1}^n \alpha_j \sin {j x} } {\partial \alpha_r}\) Derivative of Composite Function
\(\ds \) \(=\) \(\ds 2 \sqbrk {\sum_{k \mathop = 1}^n \alpha_k \sin {k x} } \sum_{j = 1}^n \delta_j^r \sin {j x}\) as $\dfrac {\partial \alpha_j} {\partial \alpha_r} = \delta_j^r$
\(\ds \) \(=\) \(\ds 2 \sqbrk {\sum_{k \mathop = 1}^n \alpha_k \sin {k x} } \sin {r x}\)
\(\ds \dfrac {\partial J_n} {\partial \alpha_r}\) \(=\) \(\ds \frac {\partial} {\partial \alpha_r} \frac \pi {b - a} \int_0^\pi \sqbrk {\map P {\sum_{k \mathop = 1}^n \alpha_k \sin {k t} }'^2 + Q \paren {\sum_{k \mathop = 1}^n \alpha_k \sin {k t} }^2} \rd t\)
\(\ds \) \(=\) \(\ds \frac \pi {b - a} \int_0^\pi \sqbrk {P \dfrac {\partial \paren {\sum_{k \mathop = 1}^n \alpha_k \sin {k t} }'^2} {\partial \paren {\sum_{j \mathop = 1}^n \alpha_j \sin {j t} }'} \dfrac {\partial \paren {\sum_{j \mathop = 1}^n \alpha_j \sin {j t} }'} {\partial \alpha_r} + 2 Q \sqbrk {\sum_{k \mathop = 1}^n \alpha_k \sin {k t} } \sin {r t} } \rd t\) Derivative of Composite Function
\(\ds \) \(=\) \(\ds \frac {2 \pi} {b - a} \int_0^\pi \sqbrk {\map P {\sum_{k \mathop = 1}^n \alpha_k \sin {k t} }' \dfrac {\partial \paren {\sum_{j \mathop = 1}^n \alpha_j j \cos {j t} } }{\partial \alpha_r} + Q \sqbrk {\sum_{k \mathop = 1}^n \alpha_k \sin {k t} } \sin {r t} } \rd t\)
\(\ds \) \(=\) \(\ds \frac {2 \pi} {b - a} \int_0^\pi \sqbrk {\map P {\sum_{k \mathop = 1}^n \alpha_k \sin {k t} }' \paren {\sum_{j \mathop = 1}^n \delta_j^r j \cos {j t} } + Q \sqbrk {\sum_{k \mathop = 1}^n \alpha_k \sin {k t} } \sin {r t} } \rd t\)
\(\ds \) \(=\) \(\ds \frac {2 \pi} {b - a} \int_0^\pi \sqbrk {\map P {\sum_{k \mathop = 1}^n \alpha_k \sin {k t} }' \paren {r \cos {r t} } + Q \sqbrk {\sum_{k \mathop = 1}^n \alpha_k \sin {k t} } \sin {r t} }\rd t\)
\(\ds \) \(=\) \(\ds \frac {2 \pi} {b - a} \int_0^\pi \sqbrk {\map P {\sum_{k \mathop = 1}^n \alpha_k \sin {k t} }' \paren {\sin {r t} }' + Q \sqbrk {\sum_{k \mathop = 1}^n \alpha_k \sin {k t} } \sin {r t} } \rd t\)

This leads to a system of equations:

$\ds \int_0^\pi \paren {\map P t \sqbrk {\sum_{k \mathop = 1}^n \alpha_k^{\paren 1} \paren {\sin {k t} }'} \paren {\sin {r x} }' + \sqbrk {Q - \lambda_n^{\paren 1} } \sqbrk {\sum_{k \mathop = 1}^n \alpha_k^{\paren 1} \sin {k t} } \sin {r t} } \rd t = 0$

Multiplying each equation by an arbitrary constant $C_r^{\paren n}$ and summing over $r$ results in:

$\ds \int_0^\pi \sqbrk {P y_n' h_n' + \paren {Q - \lambda_n^{\paren 1} y_n h_n} } \rd t = 0$

where:

$\ds \map {h_n} t = \sum_{r \mathop = 1}^n C_r^{\paren n} \sin {r t}$
$\ds y_n = \sum_{k \mathop = 1}^n \alpha_k \sin {k t}$

By Integration by parts:

\(\ds \int_0^\pi \sqbrk {P y_n' h_n' + \paren {Q - \lambda_n^{\paren 1} y_n h_n} } \rd t\) \(=\) \(\ds \int_0^\pi P h_n' \map {\rd y_n} t + \int_0^\pi \sqbrk {\paren {Q - \lambda_n^{\paren 1} y_n h_n} } \rd t\)
\(\ds \) \(=\) \(\ds P h_n' y_n \big \vert_0^\pi - \int_0^\pi y_n \rd P \map {h_n'} t + \int_0^\pi \sqbrk {\paren {Q - \lambda_n^{\paren 1} y_n h_n} } \rd t\)
\(\ds \) \(=\) \(\ds \int_0^\pi \sqbrk { - \paren {P h_n'}' + \paren {Q - \lambda_n^{\paren 1} y_n h_n} } y_n \rd t\)

Consider all real mappings $h$ such that:

$\map h x \in C^2 \openint 0 \pi$

and satisfying the boundary conditions.

Then $C_r^{\paren n}$ can be chosen, such that:

$\ds \lim_{n \mathop \to \infty} \int_0^\pi \size {\map {h_n} x - \map h x}^2 \rd x = 0$
$\ds \lim_{n \mathop \to \infty} \int_0^\pi \size {\map {h_n'} x - \map {h'} x}^2 \rd x = 0$
$\ds \lim_{n \mathop \to \infty} \int_0^\pi \size {\map {h_n} x - \map {h} x}^2 \rd x = 0$


This needs considerable tedious hard slog to complete it.
In particular: proof of the $C_r^{\paren n}$ segment
To discuss this page in more detail, feel free to use the talk page.
When this work has been completed, you may remove this instance of {{Finish}} from the code.
If you would welcome a second opinion as to whether your work is correct, add a call to {{Proofread}} the page.


Due to the existence of uniformly convergent subsequence, $y_n^{\paren 1}$ converges to $y^{\paren 1}$ uniformly on $\closedint 0 \pi$:

$\ds \lim_{m \mathop \to \infty} \int_0^\pi \paren {-\paren {P h_{n_m}'}' + \paren {Q - \lambda_{n_m}^{\paren 1} } h_{n_m} } y_{n_m}^{\paren 1} \rd x = \int_0^\pi \paren { -\paren {P h'}' + \paren {Q - \lambda^{\paren 1} } h} y^{\paren 1} \rd x = 0$


This needs considerable tedious hard slog to complete it.
In particular: prove
To discuss this page in more detail, feel free to use the talk page.
When this work has been completed, you may remove this instance of {{Finish}} from the code.
If you would welcome a second opinion as to whether your work is correct, add a call to {{Proofread}} the page.


By Lemma 5, where $Q_1 = Q - \lambda^{\paren 1}$, $y^{\paren 1} \in C^2 \closedint 0 \pi$ and satisfies Sturm-Liouville equation with $w = 1$.


$\Box$


Lemma 7

$\sequence {\map {y_n^{\paren 1} } x}$ pointwise converges to $\map {y^{\paren 1} } x$.


Proof

By Existence and Uniqueness of Solution for Linear Second Order ODE with two Initial Conditions, where $\map R x = 0$, the Sturm-Liouville equation

$-\paren {P y'}' + Q y = \lambda y$

satisfying the boundary conditions:

$\map y 0 = \map y \pi = 0$

and the subsidiary condition:

$\ds \int_0^\pi \map {y^2} t = 1$

is unique up to the sign of $y$.

Let $\map {y^{\paren 1} } t$ be a solution corresponding to $\lambda = \lambda^{\paren 1}$

Due to the subsidiary condition, the condition $\map {y^{\paren 1} } t = 0$ cannot hold in the entire closed interval $\closedint 0 \pi$.

Hence, the set of roots to this condition is countable.

Then:

$\exists t_0 \in \closedint 0 \pi: \map {y^{\paren 1} } {t_0} \ne 0$

Choose the sign so that $\map {y^{\paren 1} } {t_0} > 0$

Similarly, let $\map {y_n^{\paren 1} } t$ be a solution corresponding to $\lambda = \lambda_n^{\paren 1}$

Choose the signs so that:

$\forall n \in \N : \map {y_n^{\paren 1} } {t_0} \ge 0$

Suppose $\map {y_n^{\paren 1} } t$ does not pointwise converge to $\map {y^{\paren 1} } t$.

By Arzela's Theorem there exists another subsequence from $\sequence {\map {y_n^{\paren 1} } t}$, converging to another solution $\overline y^{\paren 1}$, where $\lambda = \lambda^{\paren 1}$.

Because of the uniqueness of solutions, except for the sign, both solutions may differ only in their signs:

$\map {\overline y^{\paren 1} } x = - \map {y^{\paren 1} } t$

Therefore:

$\map {\overline y^{\paren 1} } {t_0} < 0$

This is impossible, since:

$\forall n \in N : \map {y_n^{\paren 1} } {t_0} \ge 0$

Therefore $\map {y_n^{\paren 1} } t$ pointwise converges to $\map {y^{\paren 1} } t$, provided $\map {y_n^{\paren 1} } t$ is chosen with the correct sign.


This needs considerable tedious hard slog to complete it.
In particular: In fact, convergence is uniform; establish this result by a proof
To discuss this page in more detail, feel free to use the talk page.
When this work has been completed, you may remove this instance of {{Finish}} from the code.
If you would welcome a second opinion as to whether your work is correct, add a call to {{Proofread}} the page.



$\Box$


This needs considerable tedious hard slog to complete it.
In particular: Check how properly insert Limit of Subsequence equals Limit of Sequence as an alternative
To discuss this page in more detail, feel free to use the talk page.
When this work has been completed, you may remove this instance of {{Finish}} from the code.
If you would welcome a second opinion as to whether your work is correct, add a call to {{Proofread}} the page.



Lemma 8

Sequences $\sequence {y^{\paren n} }$ and $\sequence {\lambda^{\paren n} }$ are infinite.


Proof

Suppose, $y^{\paren r}$ and $\lambda^{\paren r}$ are known.

The next eigenfunction $y^{\paren {r + 1} }$ and the corresponding eigenvalue $\lambda^{\paren {r + 1} }$ can be found by minimising

$\ds J \sqbrk y = \int_0^\pi \paren {P y'^2 + Q y^2} \rd x $

where boundary and subsidiary conditions are supplied with orthogonality conditions:

$\forall m \in \N : {1 \le m \le r} : \ds \int_0^\pi \map {y^{\paren m} } t \map {y^{\paren {r + 1} } } t \rd t = 0$

The new solution of the form:

$\ds \map {y_n^{\paren {r + 1} } } t = \sum_{k \mathop = 1}^n \alpha_k^{\paren {r + 1} } \sin {k t}$

is now also orthogonal to mappings:

$\ds \map {y_n^{\paren m} } t = \sum_{k \mathop = 1}^n \alpha_k^{\paren m} \sin {k t}$

This results into:

$\ds \sum_{k \mathop = 1}^n \alpha_k^{\paren {r + 1} } \int_0^\pi \sin {k t} \paren {\sum_{l \mathop = 1}^n \alpha_l^{\paren m} \sin {l t} } \rd t = \frac \pi 2 \sum_{k \mathop = 1}^n \alpha_k^{\paren {r + 1} } \alpha_k^{\paren m} = 0$

These equations describe $r$ distinct $\paren {n - 1}$-dimensional hyperplanes, passing through the origin of coordinates in $n$ dimensions.

These hyperplanes intersect the sphere $\sigma_n$, resulting in an $\paren {n - r}$-dimensional sphere $\hat \sigma_{n - r}$.

By definition, it is a compact set.

By Continuous Function on Compact Subspace of Euclidean Space is Bounded, $\map {J_n} {\boldsymbol \alpha}$ has a minimum on $\hat {\sigma}_{n - r}$.

Denote it as $\lambda_n^{\paren {r + 1} }$.

By Ritz Method implies Not Worse Approximation with Increased Number of Functions:

$\lambda_{n + 1}^{\paren {r + 1} } \le \lambda_n^{\paren {r + 1} }$

This, together with $J$ being bounded from below, implies:

$\ds \lambda^{\paren {r + 1} } = \lim_{n \mathop \to \infty} \lambda_n^{\paren {r + 1} }$

Additional constraints may or may not affect the new minimum:

$\lambda^{\paren r} \le \lambda^{\paren {r + 1} }$

Let:

$\ds \map {y_n^{\paren {r + 1} } } t = \sum_{k \mathop = 1}^n \alpha_k^{\paren {r + 1} } \sin {k t}$

$y^{\paren {r + 1} }$ satisfies Sturm-Liouville equation together with boundary, subsidiary and orthogonality conditions.

By Lemma 7, which is not affected by additional constraints, $\sequence {y_n^{\paren {r + 1} } }$ uniformly converges to $y^{\paren {r + 1} }$.

Thus, $y^{\paren {r + 1} }$ is an eigenfunction of Sturm-Liouville equation with an eigenvalue $\lambda^{\paren {r + 1} }$.

Orthogonal mappings are linearly independent.

Each eigenvalue corresponds only to one eigenfunction, unique up to a constant factor.

Thus:

$\lambda^{\paren r} < \lambda^{\paren {r + 1} }$

$\blacksquare$


Sources

Retrieved from "https://proofwiki.org/w/index.php?title=Sturm-Liouville_Problem/Unit_Weight_Function&oldid=680478"