mailRe: [bug #9259] Reduced spectral density mapping yielding bad values

Others Months | Index by Date | Thread Index
>>   [Date Prev] [Date Next] [Thread Prev] [Thread Next]



Posted by Edward d'Auvergne on June 19, 2007 - 12:03:

I have thoroughly checked all the units of the physical constants,
spectral densities, rotational correlation times, and relaxation rates
and have a few important corrections about the units which are used in
relax and elsewhere.  Please read below for the details and a long
story about SI vs. CGS units, frequency vs. angular frequency, and how
it all relates to angular momentum.

On 6/15/07, Sebastien Morin <sebastien.morin.1@xxxxxxxxx> wrote:

Here are the different values I obtain for a residue with R1 = 1.1336 +-
0.0851 ; R2 = 12.9336 +- 0.9649 ; and NOE = 0.463921 +- 0.045

    J(0)                      J(wN)                    J(wH)
    =====================     =====================    =====================

Here are the results with Leo Spyracopoulos's Mathematica notebook.

    4.704231413115747e-9      2.664085520910741e-10    9.485555428699657e-12

Here are the results with relax without multiplying frequencies in Hz by
2 pi.

    4.6978912534878238e-09    2.6603551824374712e-10

    0.9986522432526923179     0.9985997677461966745

Here are the results with relax when multiplying frequencies in Hz by 2 pi.

    3.4634030539343071e-09    1.9612804482358541e-10

    0.7362314371436068543     0.7361927508863804185

The very small discrepancy we get (ratio 0.999...) is due to several
small differences in the definitions of constants (e.g. gn = -2.7126 in
relax and -2.7108 in Leo's notebook). This is not important. However,
there is a non negligible difference with the use of either frequencies
in Hz or frequencies transformed to rad/s.

Now, what do we do with that ?

The spectral densities are in units of rad / s = rad s^-1. (these are
not SI units, however)

The units of radians per second, or radian Hertz, is the angular
frequency (omega) rather than the frequency (nu), yet both are
nevertheless SI units.  For example see or related sites.
These SI units are the same in the CGS unit system.  The reason that
we use the angular frequency rather than frequency is because we are
talking about angular momentum.

The rates (R1, R2 and sigmaNOE) are in units of s^-1.

The constant 'd' is in units of s^-2.

This is not correct.  The correct units are rad^2.s^-2.  This can be
found by doing a unit analysis on the SI dipolar constant with the
mu0/4pi component (or alternatively doing the analysis in the CGS
system).  The reason for the radians being part of the equation is
because we are using Dirac's constant (h_bar) rather than Planck's
constant (h).  While Planck's constant has the units of Joules per
Hertz (or Joule seconds) because Dirac's constant is divided by 2pi
its units are Joules per Hertz per radian.  However in most cases the
radian units of Dirac's constant are not stated as we are usually
talking about angular momentum (the angular part means we use radians
hence it is indirectly implied).  The other reason is because the
gyromagnetic ratio also has units of radian, and again this is usually
not reported (for example see Table 1.1 of Cavanagh in which the
gyromagnetic ratios are in fact in rad.s^-1.T^-1, although they are
not reported as such).  All of this is still in SI units - the radian
components are independent of the SI or CGS systems.

The relevant units in the SI system for the dipolar constant, defined as

d = (mu0/4pi)^2 . (gH.gX.h_bar/<r>^3)^2,


mu0 -> kg.m.s^-2.A^-2,
h_bar -> J.s.rad^-1,
gx -> rad.s^-1.T^-1,
r -> m,

where tesla (T) is equal to the units kg.A^-1.s^-2.

The SI units for the CSA constant, defined as

c = (omegaX . csa / 3)^2,


omegaX -> rad.s^-1,
csa -> unitless.

Hence both constants have the units of rad^2.s^-2.

The constant 'c' is in units of s^-2 also (or rad^2 s^-2 if we use
frequencies in rad s^-1, which are not SI units).

The constant c is defined by the angular frequency (omega) rather than
the frequency (nu).  Hence the units of this constant in the
relaxation equations, in both SI and CGS units, is always rad^2.s^-2
as well.

I now hesitate between 3 views.

1. Should the spectral densities be in SI units (i.e. in s, as the
frequencies are in s^-1 and the rates in s^-1) ?

relax currently reports the spectral densities in the SI units for
angular frequency of radian Hertz.   Although often reported as Hertz,
the correct unit is radian Hertz.  This is again because radians are
implied, but this time because we are talking about rotations.  The
reason is as follows:

The spectral density function for isotropic diffusion is

J(w) = tm / (1 + (^2).

The units for w (or omega) is rad.s^-1.  Because the product is
unitless the units for tm are in reality s.rad^-1.  Hence the units of
the spectral density function J(w) is also s.rad^-1.

Although not reported in text books such as Cavanagh, you can still
see remnants of the radians.  For example in that book tc (or tm) is
described as being "approximately the average time for the molecule to
rotate by one radian".  All the external and internal correlation
times are in units of s.rad^-1 as they are all rotational correlation
times!  But because the are rotational times, the radians are implied
and do not need to be reported.

2. If the spectral densities are in s rad^-1, the frequencies used to
calculate them should also be in rad s^-1, thus the constant 'c' should
be in rad^2 s^-2 and the constant 'd' also (so we should calculate it
without multiplying by the factor of (mu/4pi)^2). This would be about
using old units instead of the SI units, but then the spectral densities
would be in s rad^-1.

In SI units the spectral density function is in units of s.rad^-1
whereas the physical constants (dipolar, csa, etc.) are in rad^2.s^-2.
Hence the units for the relaxation rates - hold on to your seat - is
in radian Hertz!  The R1 relaxation rate is in rad.s^-1.  Or the T1
relaxation time is in s.rad^-1.

The reason that all of this is hardly ever discussed is because the
units of radians is implied by the fact that this all relates to
angular momentum.  In the rotational world, radians are ubiquitous.
Yet they are silent because they are implied.  Unfortunately this
'hiding' of radians, combined with the CGS vs. SI unit systems,
creates large amounts of confusion.

3. We could calculate everything in SI units (as we do right now) and
normalize to rs ad^-1 in the end.

Either way, I think that Leo's notebook yields spectral densities s and
not the usual s rad^-1 (as in the 2006 paper : JBNMR,36:215-224, for
which I calculated back spectral density values and yielded the same as
published where they say it's in s rad^-1, but the units deriving says
it's s).

relax reports spectral densities in the SI (and CGS) units of s.rad^-1
as dictated by angular momentum.  Hence the input frequency of Hz must
be multiplied by 2pi.  Otherwise the product is not unitless and
hence its square cannot be added to 1 as J(w) = tm/(1+(^2).

Can you please tell me if I'm right with those ideas before I write to
Leo to report this apparent bug or typo or whatever...

I hope that what I've written clarifies a few of the problems.



Related Messages

Powered by MHonArc, Updated Tue Jun 19 17:20:42 2007