I am currently reading "The Variational Principles of Mechanics - Cornelius Lanczos", in which the author talks about the variation of a function $F(q_1, q_2, \dots q_n)$ where $q_1, q_2, \dots q_n$ are the generalized coordinates

$$F=F(q_1, q_2, \dots q_n)$$

$$ \delta F=\frac{\partial F}{\partial q_1}\delta q_1+\frac{\partial F}{\partial q_n}\delta q_n+\dots+\frac{\partial F}{\partial q_n}\delta q_n \tag{1} $$ next he writes, $\delta q_1=\epsilon a_1, \delta q_2=\epsilon a_2, \dots ,\delta q_n=\epsilon a_n$

where $a_1, a_2 \dots , a_n$ are the direction cosines, and $\epsilon$ is a small variation. I find it wrong to use the same $\epsilon$ for all $\delta q_i $'s as it seems to be inconsistent with dimensions

For example if we are dealing with spherical coordinates $(r, \theta, \phi)$, according to the above the individual variations become

$\delta r=\epsilon \hat{r}$, So I expect $\epsilon$ to have a dimension of $r$ (unit vectors are dimensionless)

$ \delta \theta = \epsilon \hat{\theta},$ now $\epsilon$ has a dimension of $\theta$?

substituting these in $\text{eq}(1)$ makes it worse,

Further I like to think (I may be wrong) of the $\delta F$ as $\nabla F$, since both seem to have the same form, but as we know $\nabla$ is different for different coordinates, but $\text{eq}(1)$ seems to use $\delta$ as we use for Cartesian coordinates

Am I missing out something?


This still isn't a very great problem since the main goal is to find the stationary value which anyways leads to a conclusion $\frac{\partial F}{\partial q_k}=0$


1 Answer 1


I agree with you that his choice of notation is questionable.

Regarding $\epsilon$, I believe it can only make sense if it is chosen to be dimensionless; this is the only way for putting it in front of every variable to work.

Next regarding the gradient analogy, I believe you are misinterpreting.

When you have a function of multiple variables say $f(x,y,z)$, the total variation reads:

\begin{equation} df = \frac{\partial f}{\partial x} dx + \frac{\partial f}{\partial y} dy + \frac{\partial f}{\partial z} dz \end{equation}

This result is independent of how you define the gradient and holds whatever the variables mean.

To define the gradient from the above formula, you need to come up with a notion of length measure and a dot product in the space of variables you are considering and then define $\nabla f$ as the vector field such that:

\begin{equation} df = \nabla f \cdot d\vec{l} \end{equation}

where $d\vec{l}$ is an infinitesimal vector element that reads differently in different coordinate systems.

  • In cartesian coordinates it would read $d\vec{l} = dx \: \hat{u}_x + dy \: \hat{u}_y + dz \: \hat{u}_z$

  • In cylindrical coordinates it would read instead $d\vec{l} = dr \: \hat{u}_r + rd\theta \: \hat{u}_{\theta} + dz \: \hat{u}_z $

Thus leading to coordinate-system-dependent gradient expressions.

In any case, the variational formula you wrote is exactly analogous to the first variation formula I wrote and not to a gradient.


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.