The following original model isolates the conversion mechanism. Use a periodic image window, unit illumination, ideal unit magnification, and a phase plate that changes only the zero spatial frequency. Write for the spatial mean. Multiplying that single Fourier coefficient by , with , gives the exact image field in this model:
This formula follows by splitting : the second term has zero mean and is left unchanged. A real annular, partly coherent instrument alters a finite band of directions, so this calculation is a mechanism model rather than its full image-formation law.
For a weak phase object with , one has and hence
The cross term is . Thus makes positive brighter and makes it darker under this stated phase convention for complex amplitudes. The sign comes from this convention and chosen plate; a manufacturer’s positive/negative contrast labels should not be inferred from the sign of alone.
For example, choose and a zero-mean periodic pattern whose peak phase is . The first-order peak intensities are , compared with the uniform ideal bright-field microscopy value . The relative change about the reduced background is approximately for a quarter-cycle shift. Attenuating the reference increases relative contrast while reducing the background photon count; it is not light amplification or a universal improvement of signal-to-noise ratio.
Figure 1.
Making a transparent periodic phase pattern visible
. Original calculation for on a unit periodic window and . The specimen pattern is synthetic, not a cell photograph. The top row compares ideal bright-field microscopy and the exact coherent zero-mode model of phase contrast; brightness is shown relative to the empty-field value for each mode. Lower plots compare exact and first-order results along and as the reference phase changes at the pattern maximum. These are calculations, not historical observations. Details.
Two checks expose what the model can and cannot measure. A spatially constant phase gives and , so no absolute phase is recovered. Orthogonality of the constant and zero-mean parts also gives
For this cannot exceed the incident mean light intensity. The image redistributes brightness and loses some reference light rather than creating energy.