- #1
- 753
- 140
I have problem with Lagrange's equations and their derivation, the way it is presented in Goldstein's "Classical Mechanics". I have never seen this problem mentioned anywhere, so I wonder if I am the only one who see this problem.
To see the problem, consider a simple case where the motion of a system can be described by just one generalized coordinate [itex]q[/itex], plus time explicitly. Let [itex]x[/itex] be a cartesian coordinate of one particle of the system. Then [itex]x=x(q,t)[/itex].
Then, [tex]\dot x=\frac{\partial x}{\partial q}\dot q+ \frac{\partial x}{\partial t}.[/tex]
(In Goldstein we have several generalized coordinates, but the reasoning is the same.)
Now, reasoning as Goldstein, we could without problem differentiate [itex]\dot x[/itex] wrt [itex]\dot q[/itex], obtaining [itex]\partial \dot x / \partial \dot q =\partial x/\partial q[/itex], and also plug in [itex]\dot q [/itex] in the expression for the total kinetic energy [itex]T[/itex] and the Lagrangian [itex]L=T-V[/itex], and then without problem differentiate [itex]L[/itex] wrt [itex]\dot q[/itex], obtaning Lagrange's Equations (one for each generalized coordinate):
[tex]\frac d {dt}\frac{\partial L}{\partial\dot q}-\frac{\partial L}{\partial q}=0. [/tex]
(Differentiating [itex]L[/itex] wrt [itex]q[/itex] seems to be no problem, either.)
But how can we differentiate [itex]\dot x[/itex] (and then [itex]L[/itex]) wrt [itex]\dot q[/itex]? In the formula we obatined for [itex]\dot x[/itex], we have only one indepentent variable: [itex]t[/itex], and therefore it shouldn't make sense to differentiate wrt any other variable.
To be able to differentiate wrt [itex]\dot q[/itex], we must actually view [itex]\dot x[/itex] (and similarly, [itex]L[/itex]) as a composite function, [itex]\dot x= f(q,\dot q,t) [/itex], where [itex]f(u_1,u_2,u_3)=g_1(u_1,u_3)u_2+g_2(u_1,u_3)[/itex] and [itex]g[/itex] is the function which gives the coordinate transformation: [itex]x=x(q,t)=g(q,t)[/itex], and [itex]f[/itex] and [itex]g[/itex] can be differentiated wrt any of their arguments. ([itex]g_1[/itex] and [itex]g_2[/itex] are the partial derivatives of [itex]g[/itex].)
The expression [itex]\partial \dot x /\partial \dot q[/itex] will then make sense if we recognize it as the same as [itex]f_2(q,\dot q,t)[/itex], and then it will be equal to [itex]\partial x/\partial q[/itex], just as it should be.
However, this will be consistent only if the choice of the function [itex]f[/itex] is unique. If there was another function [itex]h(u_1,u_2,u_3)\ne f(u_1,u_2,u_3)[/itex], which also satisfies [itex]\dot x = h(q,\dot q,t)[/itex], but for which [itex]h_2(q,\dot q,t)\ne f_2(q,\dot q,t)[/itex], then [itex]\partial \dot x/\partial \dot q[/itex] cannot be uniquely defined (and neither can [itex]\partial L/\partial \dot q[/itex]).
This problem is not discussed in Goldstein, and I have not seen it mentioned anywhere (although I did not search very much). Am I really the only person who see this problem?
Fortunately, the problem can be solved. It turns out that the function [itex]f[/itex] above must be unique. The reason is that the equation [itex]\dot x= f(q,\dot q,t) [/itex] must hold for all possible paths [itex]q(t)[/itex], and, for every triplet of values [itex](u_1,u_2.u_3)[/itex], it is always possible to find a path [itex]q(t)[/itex] such that [itex]q(t)=u_1[/itex] and [itex]q'(t)=u_2[/itex], for [itex]t=x_3[/itex]. This implies that [itex]f[/itex] is uniquely determined by the requirement [itex]\dot x= f(q,\dot q,t)[/itex], it must be the function given above. This argument can be generalized to any number of generalized coordinates, and it also follows that [itex]\partial L/\partial \dot q[/itex] will be meaningful and unique.
But it is not trivial to find and prove this. I was very confused when I read Goldstein the first time, and I wondered how on Earth one could differentiate wrt [itex]\dot q[/itex]. It took a while until I found the "proof" above. I think it is the author's job to do this, not the reader's.
To see the problem, consider a simple case where the motion of a system can be described by just one generalized coordinate [itex]q[/itex], plus time explicitly. Let [itex]x[/itex] be a cartesian coordinate of one particle of the system. Then [itex]x=x(q,t)[/itex].
Then, [tex]\dot x=\frac{\partial x}{\partial q}\dot q+ \frac{\partial x}{\partial t}.[/tex]
(In Goldstein we have several generalized coordinates, but the reasoning is the same.)
Now, reasoning as Goldstein, we could without problem differentiate [itex]\dot x[/itex] wrt [itex]\dot q[/itex], obtaining [itex]\partial \dot x / \partial \dot q =\partial x/\partial q[/itex], and also plug in [itex]\dot q [/itex] in the expression for the total kinetic energy [itex]T[/itex] and the Lagrangian [itex]L=T-V[/itex], and then without problem differentiate [itex]L[/itex] wrt [itex]\dot q[/itex], obtaning Lagrange's Equations (one for each generalized coordinate):
[tex]\frac d {dt}\frac{\partial L}{\partial\dot q}-\frac{\partial L}{\partial q}=0. [/tex]
(Differentiating [itex]L[/itex] wrt [itex]q[/itex] seems to be no problem, either.)
But how can we differentiate [itex]\dot x[/itex] (and then [itex]L[/itex]) wrt [itex]\dot q[/itex]? In the formula we obatined for [itex]\dot x[/itex], we have only one indepentent variable: [itex]t[/itex], and therefore it shouldn't make sense to differentiate wrt any other variable.
To be able to differentiate wrt [itex]\dot q[/itex], we must actually view [itex]\dot x[/itex] (and similarly, [itex]L[/itex]) as a composite function, [itex]\dot x= f(q,\dot q,t) [/itex], where [itex]f(u_1,u_2,u_3)=g_1(u_1,u_3)u_2+g_2(u_1,u_3)[/itex] and [itex]g[/itex] is the function which gives the coordinate transformation: [itex]x=x(q,t)=g(q,t)[/itex], and [itex]f[/itex] and [itex]g[/itex] can be differentiated wrt any of their arguments. ([itex]g_1[/itex] and [itex]g_2[/itex] are the partial derivatives of [itex]g[/itex].)
The expression [itex]\partial \dot x /\partial \dot q[/itex] will then make sense if we recognize it as the same as [itex]f_2(q,\dot q,t)[/itex], and then it will be equal to [itex]\partial x/\partial q[/itex], just as it should be.
However, this will be consistent only if the choice of the function [itex]f[/itex] is unique. If there was another function [itex]h(u_1,u_2,u_3)\ne f(u_1,u_2,u_3)[/itex], which also satisfies [itex]\dot x = h(q,\dot q,t)[/itex], but for which [itex]h_2(q,\dot q,t)\ne f_2(q,\dot q,t)[/itex], then [itex]\partial \dot x/\partial \dot q[/itex] cannot be uniquely defined (and neither can [itex]\partial L/\partial \dot q[/itex]).
This problem is not discussed in Goldstein, and I have not seen it mentioned anywhere (although I did not search very much). Am I really the only person who see this problem?
Fortunately, the problem can be solved. It turns out that the function [itex]f[/itex] above must be unique. The reason is that the equation [itex]\dot x= f(q,\dot q,t) [/itex] must hold for all possible paths [itex]q(t)[/itex], and, for every triplet of values [itex](u_1,u_2.u_3)[/itex], it is always possible to find a path [itex]q(t)[/itex] such that [itex]q(t)=u_1[/itex] and [itex]q'(t)=u_2[/itex], for [itex]t=x_3[/itex]. This implies that [itex]f[/itex] is uniquely determined by the requirement [itex]\dot x= f(q,\dot q,t)[/itex], it must be the function given above. This argument can be generalized to any number of generalized coordinates, and it also follows that [itex]\partial L/\partial \dot q[/itex] will be meaningful and unique.
But it is not trivial to find and prove this. I was very confused when I read Goldstein the first time, and I wondered how on Earth one could differentiate wrt [itex]\dot q[/itex]. It took a while until I found the "proof" above. I think it is the author's job to do this, not the reader's.
Last edited: