2
$\begingroup$

Trying to solve an inertial navigation problem in terms of quaternions, I ended up with the following indefinite integral $$ I = \int e^{t \ln q} \, p \, e^{-t \ln q} dt, $$ where $q \in \mathbb H$ is a unit quaternion, $p \in \mathbb H$ is a pure imaginary quaternion, and both $q$ and $p$ are constant with respect to the integration variable $t \in \mathbb R$.

Assuming $q \neq 1$, I integrated by parts: $$ I = \int \underbrace{\mathstrut{}e^{t \ln q} p}_u \, \underbrace{\mathstrut{}e^{-t \ln q}}_{v'} \, dt = e^{t \ln q} p e^{-t \ln q} (-\ln q)^{-1} - \int e^{t \ln q} (\ln q) \, p e^{-t\ln q} (-\ln q)^{-1} \, dt, \\ I = e^{t \ln q} p \, e^{-t \ln q} (-\ln q)^{-1} + (\ln q) \, I \, (\ln q)^{-1}, $$ and got the equation $$ (\ln q) \, I - I \, (\ln q) = e^{t \ln q} \, p \, e^{-t \ln q}. $$

From this question I learned that the general solution for the $ax + xb = c$ equation is $$ x= \left( |b|^2 + 2b_0a + a^2 \right)^{-1} \left( ac +c b^* \right), $$ but it did not work for my case because of zero quaternion inversion. I also tried to expand this equation into quaternion components, but came to a system of linear equations with a singular matrix. This looks quite strange to me, since I'm pretty sure this integral should exist.

Could please someone point me to a mistake in my computations and show how to calculate this antiderivative properly? Might there be a well known solution for the problem?

$\endgroup$
1
$\begingroup$

Note that conjugation preserves vectors, so the integrand and hence $\mathbf{I}$ are pure imaginary.

For pure quaternions $\mathbf{a}$ and $\mathbf{b}$,

$$ \mathbf{ab}=-\mathbf{a}\cdot\mathbf{b}+\mathbf{a}\times\mathbf{b}. $$

Thus, writing $``\ln q "=\mathbf{v}$ we have

$$ (\ln q)\mathbf{I}-\mathbf{I}(\ln q)=2\mathbf{v}\times \mathbf{I}=\exp(t\mathbf{v})\mathbf{p}\exp(-t\mathbf{v}). $$

Note that $\mathbf{v}\times\mathbf{I}$ deletes the $\mathbf{v}$-component of $\mathbf{I}$, so this equation fails to "see" this component. However, this actually isn't a problem: conjugating by $\exp(t\mathbf{v})$ fixes the $\mathbf{v}$-component of $\mathbf{p}$, call it $\mathbf{p}_{\|}$, within the integrand for $\mathbf{I}$, so we may conclude the $\mathbf{v}$-component for $\mathbf{I}$ must be $\mathbf{I}_{\|}=t\mathbf{p}_{\|}$.

Applying $\mathbf{v}\times$ to $2\mathbf{v}\times\mathbf{I}_{\perp}=\exp(t\mathbf{v})\mathbf{p}\exp(-t\mathbf{v})$ (with $v=\|\mathbf{v}\|$) yields

$$ -2v^2\mathbf{I}_{\perp}=\mathbf{v}\times\big(e^{t\mathbf{v}}\mathbf{p}e^{-t\mathbf{v}}\big) $$

from which we may conclude

$$ \mathbf{I}=t\big(\mathbf{p}\cdot\frac{\mathbf{v}}{\|\mathbf{v}\|}\big)\frac{\mathbf{v}}{\|\mathbf{v}\|}+\frac{1}{2}\big(e^{t\mathbf{v}}\mathbf{p}e^{-t\mathbf{v}}\big)\times\frac{\mathbf{v}}{\,\,\|\mathbf{v}\|^2}+\mathbf{C}. $$


Another route is to expand the exponentials $\exp(t\mathbf{v})$ using Euler's formula so you can multiply out and integrate directly. This works especially nice if, setting $\hat{\mathbf{v}}=\frac{\mathbf{v}}{\|\mathbf{v}\|}$ you write $\mathbf{p}$ with respect to an oriented orthonormal basis $\{\mathbf{u},\hat{\mathbf{v}},\mathbf{w}\}$.

| cite | improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.