For any Assignment related queries, Call us at : - +1 678 648 4277

You can mail us at : - support@mathhomeworksolver.com or

reach us at : - https://www.mathhomeworksolver.com/

Maths Assignment Help

Problem 1:
Here are a few questions that you should be able to answer based
only on 18.06:
a) Suppose that B is a Hermitian positive-definite matrix. Show
that there is a unique matrix √ B which is Hermitian positive-
definite and has the property √( B)2 = B. (Hint: use the
diagonalization of B.)

(b) Suppose that A and B are Hermitian matrices and that B is

positive-definite.
(i) Show that B−1A is similar (in the 18.06 sense) to a Hermitian
matrix. (Hint: use your answer from above.)

(ii) What does this tell you about the eigenvalues λ of B− 1A , i.e.
the solutions of B−1Ax = λx?

(iii) Are the eigenvectors x orthogonal?

(iv) In Julia, make a random 5 × 5 real-symmetric matrix via
A=rand(5,5); A = A+A’ and a random 5 × 5 positive-definite
matrix via B = rand(5,5); B = B’*B ... then check that the
eigenvalues of B−1A match your expectations from above via
lambda,X = eigvals(B\A) (this will give an array lambda of the
eigenvalues and a matrix X whose columns are the eigenvectors).
(v) Using your Julia result, what happens if you compute C = XT
BX via C=X’*B*X? You should notice that the matrix C is very
special in some way. Show that the elements Cij of C are a kind
of “dot product” of the eigenvectors i and j, but with a factor of B
in the middle of the dot product.

(c) The solutions y(t) of the ODE y “− 2y ‘− cy = 0 are of the

form y(t) =
for some constants C1 and C2
determined by
the initial conditions. Suppose that A is a real-symmetric 4×4
matrix with eigenvalues 3, 8, 15, 24 and corresponding
eigenvectors x1, x2, . . . , x4, respectively.
(i) If x(t) solves the system of ODEs Ax with initial conditions x(0)

= a0 and x′ (0) = b0, write down the solution x(t) as a closed-

form expression (no matrix inverses or exponentials) in terms
of the eigenvectors x1, x2, . . . , x4 and a0 and b0. [Hint:
expand x(t) in the basis of the eigenvectors with unknown
coefficients c1(t), . . . , c4(t), then plug into the ODE and
solve for each coefficient using the fact that the eigenvectors
are _________.]

(ii) After a long time t ≫ 0, what do you expect the

approximate form of the solution to be?
 Problem 2:

In class, we considered the 1d Poisson equation

for the vector space of functions u(x) on x ∈ [0, L] with the
“Dirichlet” boundary conditions u(0) = u(L) = 0, and solved it in
terms of the eigenfunctions of
(giving a Fourier sine series). Here, we will consider a couple of
small variations on this:

a) Suppose that we we change the boundary conditions to the

periodic boundary condition u(0) = u(L).

(i) What are the eigenfunctions of now?

(ii) Will Poisson’s equation have unique solutions? Why or

why not?

(iii) Under what conditions (if any) on f(x) would a solution

exist? (You can restrict yourself to f with a convergent
Fourier series.)

(b) If we instead consider for functions v(x) with the boundar

conditions v(0) = dx2 v(x) v(L) + 1, do these functions form

a vector space? Why or why not?
(c) Explain how we can transform the v(x) problem of the previous
part back into the original
problem with u(0) = u(L), by writing u(x)
= v(x) + q(x)
and f(x) = g(x) + r(x) for some functions q and r. (Transforming
a new problem into an old, solved one is always a useful thing
to do!)

Problem 3:

For this question, you may find it helpful to refer to the notes
and reading from lecture 3. Consider a finite-difference
approximation of the form:

(a) Substituting the Taylor series for u(x+ Δx) etcetera (assuming
u is a smooth function with a convergent Taylor series, blah
blah), show that by an appropriate choice of the constants c and
d you can make this approximation fourth-order accurate: that is,
the errors are proportional to (Δx)4 for small Δx.
(b) Check your answer to the previous part by numerically
computing u ′ (1) for u(x) = sin(x), as a function of Δx, exactly as
in the handout from class (refer to the notebook posted in lecture
3 for the relevant Julia commands, and adapt them as needed).
Verify from your log-log plot of the |errors| versus Δx that you
obtained the expected fourth-order accuracy.
 Solutions
Problem 1:

(a) Since it is Hermitian, B can be diagonalized: B = QΛQ ∗,

where Q is the matrix whose columns are the eigenvectors
(chosen orthonormal so that Q−1 = Q∗ ) and Λ is the diagonal
matrix of eigenvalues. Define √ Λ as the diagonal matrix of the
(positive) square roots of the eigenvalues, which is possible
because the eigenvalues are > 0 (since B is positive-definite).
Then define √ B = Q √ ΛQ∗ , and by inspection we obtain (√
B)2 = B. By construction, √ B is positive-definite and Hermitian.

It is easy to see that this √ B is unique, even though the

eigenvectors X are not unique, because any acceptable
transformation of Q must commute with Λ and hence with √ Λ.
Consider for simplicity the case of distinct eigenvalues: in this
case, we can only scale the eigenvectors by (nonzero) constants,
corresponding to multiplying Q on the right by a diagonal
(nonsingular) matrix D. This gives the same B for any D, since
QDΛ(QD)−1 = QΛDD−1Q−1 = QΛQ−1 (diagonal matrices
commute), and for the same reason it gives the same √ B. For
repeated eigenvalues λ, D can have off-diagonal elements that mix
eigenvectors of the same eigenvalue, but D still commutes with Λ
because these off-diagonal elements only appear in blocks where
Λ is a multiple λI of the identity (which commutes with anything).
(b) Solutions:

(i) From 18.06, B−1A is similar to C = MB−1AM−1 for any

invertible M. Let M = B1/2 from above. Then C =
B−1/2AB−1/2, which is clearly Hermitian since A and B−1/2
are Hermitian. (Why is B−1/2 Hermitian? Because B1/2 is
Hermitian from above, and the inverse of a Hermitian matrix
is Hermitian.)

(ii) From 18.06, similarity means that B−1A has the same
eigenvalues as C, and since C is Hermitian these eigenvalues
are real.

(iii) No, they are not (in general) orthogonal. The eigenvectors Q
of C are (or can be chosen) to be orthonormal (Q∗Q = I), but
the eigenvectors of B−1A are X = M−1Q = B−1/2Q, and hence
X∗X = Q∗B−1Q = I unless B = I.

(iv) Note that there was a typo in the pset. The eigvals function
returns only the eigenvalues; you should use the eig function
instead to get both eigenvalues and eigenvectors, as explained
in the Julia handout.
The array lambda that you obtain in Julia should be purely real, as
expected. (You might notice that the eigenvalues are in somewhat
random order, e.g. I got -8.11,3.73,1.65,­1.502,0.443. This is a side
effect of how eigenvalues of non-symmetric matrices are
computed in standard linear-algebra libraries like LAPACK.) You
can check orthogonality by computing X∗X via X’*X, and the
result is not a diagonal matrix (or even close to one), hence the
vectors are not orthogonal.

(v) When you compute C = X∗BX via C=X’*B*X, you should

find that C is nearly diagonal: the off-diagonal entries are all very
close to zero (around 10−15 or less). They would be exactly zero
except for roundoff errors (as mentioned in class, computers keep
only around 15 significant digits). From the definition of matrix
multiplication, the entry Cij is given by the i-th row of X ∗
multiplied by B, multiplied by the j-th column of X. ∗ But the j-th
column X is the j-th eigenvector xj , and the i-th row of X∗ is x .
Hence i ∗ Cij = xi Bxj, which looks like a dot product but with B
in the middle. The fact that C
is diagonal means that which is a kind of
orthogonality relation.

In fact, if we define the inner product (x, y) = x∗By, this is a

perfectly good inner product (it satisifies all the inner-product
criteria because B is positive-definite), and we will see in the next
pset that B−1A is actually self-adjoint under this inner product.
Hence it is no surprise that we get real eigenvalues and orthogonal
eigenvectors with respect to this inner product.]
(c) Solutions:

(i) If we write x(t) then plugging it into the ODE and

using the
eigenvalue equation yields

Using the fact that the xn are necessarily orthogonal (they are
eigenvectors of a Hermitian matrix for distinct eigenvalues), we
can take the dot product of both sides with xm to find that c¨n −
2˙cn − λnc = 0 for each n, and hence

Again using orthogonality to pull out the n-th term, we find

(note that we were not given that xn were normalized to unit
length, and this is not automatic) and hence we can solve for αn
and βn to obtain:

(ii) After a long time, this expression will be dominated by the

fastest growing term, which √ (1+ 1+λn)t is the e term for λ4 =
24, hence:

Problem 2:

(a) Suppose that we we change the boundary conditions to the

periodic boundary condition u(0) = u(L).
As in class, the eigenfunctions are sines, cosines, and exponentials,
and it only remains to
apply the boundary conditions. sin(kx) is periodic if k =
n = 0 because we do not allow zero eigenfunctions and excluding
n < 0 because they are not linearly independent), and cos(kx) is
periodic if n = 0, 1, 2, . . . (excluding n < 0 since they are the same
functions). The eigenvalues are −k2 = −(2πn/L)2.
for n = 1, 2, . . . (excluding
is periodic only for imaginary k = but in this case we obtain

cos(2πnx/L) + isin(2πnx/L), which is not linearly

independent
of the sin and cos eigenfunctions above. Recall from 18.06 that
the eigenvectors for a given eigenvalue form a vector space (the
null space of A − λI), and when asked for eigenvectors we only
want a basis of this vector space. Alternatively, it is acceptable to
start with exponentials and call our eigenfunctions
for all integers n, in
which case we wouldn’t give sin and cos eigenfunctions
separately.
Similarly, sin(φ + 2πnx/L) is periodic for any φ, but this is not
linearly independent since sin(φ + 2πnx/L) = sin φ cos(2πnx/L) +
cos φ sin(2πnx/L).
[Several of you were tempted to also allow sin(mπx/L) for odd m
(not just the even m considered above). At first glance, this seems
like it satisfies the PDE and also has u(0) = u(L) (= 0). Consider,
for example, m = 1, i.e. sin(πx/L) solutions. This can’t be right,
however; e.g. it is not orthogonal to 1 = cos(0x), as required for
self-adjoint problems. The basic problem here is that if you
consider the periodic extension of sin(πx/L), then it doesn’t
actually satisfy the PDE, because it has a slope discontinuity at the
endpoints. Another way of thinking about it is that periodic
boundary conditions arise because we have a PDE defined on a
torus, e.g. diffusion around a circular tube, and in this case the
choice of endpoints is not unique—we can easily redefine our
endpoints so that x = 0 is in the “middle” of the domain, making it
clearer that we can’t have a kink there. (This is one of those cases
where to be completely rigorous we would need to be a bit more
careful about defining the domain of our operator.)]

(ii) No, any solution will not be unique, because we now have a
nonzero nullspace spanned by the constant function u(x) = 1
(which is periodic):

Equivalently, we have a 0 eigenvalue corresponding to cos(2πnx/L)

for n = 0 above.
(iii) As suggested, let us restrict ourselves to f(x) with a
convergent Fourier series. That is, as in class, we are expanding
f(x) in terms of the eigenfunctions:
(You could also write out the Fourier series in terms of sines and
cosines, but the complexexponential form is more compact so I
will use it here.) Here, the coefficients cn, by the usual
orthogonality properties of the Fourier series, or equivalently by
self-adjointness of

In order to solve as in class we would divide each term by its

eigenvalue
−(2πn/L)2, but we can only do this for n 0. Hence, we can
only solve the equation if the n = 0 term is absent, i.e. c0 = 0.
Appling the explicit formula for c0, the equation is
solvable (for f with a Fourier series) if and only if:
There are other ways to come to the same conclusion. For
example, we could expand u(x) in a Fourier series (i.e. in the
eigenfunction basis), apply d2/dx2, and ask what is the column
space of d2/dx2? Again, we would find that upon taking the
second derivative the n = 0 (constant) term vanishes, and so the
column space consist of Fourier series missing a constant term.

The same reasoning works if you write out the Fourier series in
terms of sin and cos sums separately, in which case you find that
f must be missing the n = 0 cosine term, giving the same result.

(b) No. For example, the function 0 (which must be in any

vector space) does not satisy those boundary conditions. (Also
adding functions doesn’t work, scaling them by constants,
etcetera.)

(c) We merely pick any twice-differentiable function q(x) with

q(L) − q(0) = −1, in which case u(L) − u(0) = [v(L) − v(0)] +
[q(L) − q(0)] = 1 − 1 = 0 and u is periodic. Then, plugging v =
u − q into we obtain
 which is the (periodic-u) Poisson equation for u with a (possibly)
modified right-hand side.

For example, the simplest such q is probably q(x) = x/L, in

which case d2q/dx2 = 0 and u solves the Poisson equation with
an unmodified right-hand side.

Problem 3:

We are using a difference approximation of the

form:

(a) First, we Taylor expand:

The numerator of the difference formula flips sign if Δx → −Δx,

which means that when you plug in the Taylor series all of the
even powers of Δx must cancel! To get 4th-order accuracy, the
Δx3 term in the numerator (which would give an error ∼ Δx2)
must cancel as well, and this determines our choice of c: the Δx3
term in the numerator is
and hence we must have
c = 23 = 8 The remaining terms in the numerato

are the Δx term and the Δx5 term:

Clearly, to get the correct u' (x) as Δx → 0, we

d = 12
must have
Hence, the error is approximately

which is ∼ Δx4 as desired.

Figure 1: Actual vs. predicted error for problem 1(b), using
fourth-order difference approximation for u' (x) with u(x) =
sin(x), at x = 1.
b) The Julia code is the same as in the handout, except now we
compute our difference approximation by the command: d = (-
sin(x+2*dx) + 8*sin(x+dx) - 8*sin(x+dx) + sin(x-2*dx)) ./ (12 *
dx); the result is plotted in Fig. 1. Note that the error falls as a
straight line (a power law), until it reaches ∼ 10−15, when it
starts becoming dominated by roundoff errors (and actually gets
worse). To verify the order of accuracy, it would be sufficient to
check the slope of the straight-line region, but it is more fun to
plot the actual predicted error from the previous part, where
Clearly the predicted error is almost exactly

right (until roundoff errors take over).