Physics Stack Exchange is a question and answer site for active researchers, academics and students of physics. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

This is an extension of a Phys.SE question I asked earlier, Fourier transform of two pulses of light.

I start with 2 Gaussians, represented by the equation $e^{-(t/a)^2}\times e^{2\pi i d t}*(\delta(t-\tau) + \delta(t+\tau))$ where a corresponds to the FWHM of the Gaussian. (FWHM $\approx 15,\ a \approx 9.$ d is the shift in the frequency domain, and $\tau$ is the separation between Gaussians. The $e^{2\pi i d t}$ is simply adding a carrier frequency. I want the Fourier transform to be centered at 900nm light, or a frequency of $3.33\times10^{14} \frac ms$.

When I take the Fourier transform of the equation, I get $2a\sqrt\pi \cos(2\pi p \tau)e^{-(a \pi p)^2} * \delta(p - d)$ I want to know which units I should use for each variable, to get the output in terms wavelength in nm, or frequency in hz. Additionally I would like to know the beat frequency of the fourier transform in terms of $\tau$. The way its written now, $\tau = $ the beat frequency, but if if i have the units wrong, it could change.

share|cite|improve this question
$a$ has dimension of inverse time because $-(t/a)^2$ appears in an exponent, $\sqrt\pi$ is dimensionless much like $2$ or a cosine or exponential of anything. $\delta(y)$ has the same units as $1/y$ because it's the derivative of the dimensionless step function with respect to $y$. Convolution symbol $*$ has the same units as the variable in which the convolution takes place, from the $dt$ etc. integration measure in the convolution. The units of a product should be obvious if you know what the word "unit" means. – Luboš Motl Jul 30 '13 at 14:58
@LubošMotl I think that $a$ has dimensions of time, not inverse of time (frequency), because the exponent has to be dimensionless. – neutrino Jul 30 '13 at 21:41
Moreover I believe you're asking for the units one should use for the transform variables, not the units of the Fourier transform itself. Am I right? – WetSavannaAnimal aka Rod Vance Jul 31 '13 at 4:51
up vote 1 down vote accepted

The Fourier transform definition you are using is:

$\mathfrak{F}(h(t))(\nu) = \tilde{H}(\nu) =\int_{-\infty}^\infty e^{-2\pi\,i\,\nu\,t} \,h(t)\mathrm{d}t$

In this convention, the transform variable is in $\mathrm{Hz}$, as opposed to radians per second, when $t$ is in seconds. So, your independent variable $t$ needs to be in seconds, to get the independent frequency variable to be in hertz.

Wavelength is inversely proportional to frequency $c = \nu \lambda$, so the transformation between frequency and wavelength spectrums is nonlinear. If your pulse is narrowband, so that its frequency spread is much less than its carrier frequency, you can use $\nu = \frac{c}{\lambda} \Rightarrow \Delta \nu \approx -\frac{c}{\lambda^2} \Delta \lambda = -\frac{\nu}{\lambda} \Delta \lambda$. The scaling constant therefore is:

$\nu \approx \nu_0 - \frac{\nu_0}{\lambda_0} (\lambda - \lambda_0) = 3.33\times10^{14} \mathrm{Hz} - 3.7\times10^{11}(\lambda - 900\mathrm{nm}) = 6.67\times10^{14} - 3.7\times10^{11}\lambda $

With $\lambda$ in nanometres. So, to convert your Fourier transform with transform variable of frequency in Hz to one with transform variable of wavelength in nanometres, you plug in:

$\tilde{H}(\nu) = \tilde{H}(6.67\times10^{14} - 3.7\times10^{11}\lambda)$

Again, this formula assumes $\lambda$ is in nanomatres. Beware, though, if your source is broadband (pulse duration less than 0.1 picoseconds) on interpreting power spectrums calculated from the Fourier transform. If your power spectrum is $G(\nu) = |\tilde{H}(\nu)|^2$, and you are trying to interpret $G$ as a power per unit wavelength, you must bring in the Jacobian of the nonlinear transformation. Given $\nu = \frac{c}{\lambda}$, witness that the power in the interval spanned by $[\nu_1,\nu_2]$ is

$\int_{\nu_1}^{\nu_2} G(\nu) d\nu = -\int_{\frac{c}{\nu_1}}^{\frac{c}{\nu_2}} G(\frac{c}{\lambda}) \frac{c}{\lambda^2} d\lambda$

So to get a power spectrum in power per wavelength, you have to use $\left|\tilde{H}\left(\frac{c}{\lambda}\right)\right|^2 \frac{c}{\lambda^2} $, not simply $\left|\tilde{H}\left(\frac{c}{\lambda}\right)\right|^2$

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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