Dichroic Mirror

A film of oil on a puddle shows color because light reflecting off the top surface and light reflecting off the bottom surface travel different distances and arrive out of step. At some wavelengths the two waves add; at others they cancel. The film has no pigment in it at all. The color is arithmetic.

A dichroic mirror is that effect committed to, with intent. Alternating layers of high and low refractive index material, each a quarter of a wavelength thick, are deposited onto glass, typically fifty to a hundred of them. Every interface reflects a little. Choose the thicknesses correctly and at your design wavelength all hundred of those small reflections arrive in step and sum to better than 99%; a few tens of nanometers away they arrive out of step and cancel to near zero, so the light passes straight through.

This is why a dichroic can do something no colored glass can: reflect one band almost perfectly while transmitting its neighbor almost perfectly, with a transition between them only a few nanometers wide. Absorption filters throw away the light they reject and turn it into heat. A dichroic does not reject light so much as send it somewhere else, which is exactly what you want when the thing you are rejecting is your excitation beam and the thing you are keeping is a few hundred photons.

The layer thicknesses were chosen for a particular angle — 45 degrees, almost always, because that is the geometry that folds the excitation path into the imaging path. Send light in at a different angle and it travels a longer or shorter distance through each layer, which is the same as having built the layers to a different thickness. The whole spectrum shifts, toward the blue as the angle steepens.

The shift is not subtle. A dichroic designed for 45 degrees and used at 50 can move its edge by ten nanometers or more, which on a part whose entire value is a sharp edge in the right place is the difference between a working filter set and a mystery. It also means a dichroic is not a general-purpose part you can mount however it fits.

There is a second consequence of working off-normal that catches people doing quantitative work. At 45 degrees the two polarizations of light see slightly different effective layer thicknesses, so the edge lands at a different wavelength for each. Every dichroic transmission curve you have ever seen quoted is an average over unpolarized light. If your measurement cares about polarization — fluorescence anisotropy, most obviously — the standard part is quietly introducing a bias, and you need a dichroic specified for it.

Catalogs draw dichroics as step functions. They are not steps. The transition from reflecting to transmitting takes somewhere between five and twenty nanometers, and within that span the part is doing neither job properly: passing some of what it should reflect, reflecting some of what it should pass.

That slope is where the arguments happen. Your fluorophore's emission begins close to its excitation — the Stokes shift for a bright dye is often only twenty or thirty nanometers, so the most useful part of your emission sits directly on the transition. Move the edge closer to the laser to collect more signal and you also pass more excitation light. Move it away and you throw away the brightest part of your emission.

Worth knowing where the responsibility actually sits: the dichroic is not what makes your image dark-background. It contributes perhaps three to four orders of magnitude of rejection at the laser line, and you need nine or ten. The emission filter behind it supplies the rest. The dichroic's real job is geometric: getting the excitation beam and the emission beam onto the same axis and then separating them again, and the blocking is a bonus it is not able to deliver alone.

A single-band dichroic reflects one laser line and transmits everything above it. A multiband — a pentaband, typically — reflects five narrow notches at the common laser lines and transmits everything in between, so five channels can be excited without moving any part.

The simulation on this page makes the consequence visible, and it is not the one most people expect. Put Alexa 488 in the primary channel with DAPI behind it, on a single-band dichroic, and bleedthrough reads zero. Switch to the pentaband, change nothing else, and it jumps past 50%. The filter did not change. What changed is that the single-band part was never delivering the 405 line to the sample at all, so DAPI was dark. A single-band dichroic is sequential acquisition whether you asked for it or not, and its cleanliness is bought with time rather than with optics.

The pentaband hands you all five lines at once and hands the entire separation problem to the emission filter, which now has to reject a neighbor that is emitting at full strength rather than one that is barely lit. That is the trade. It is worth making, but it should be made knowingly, and the number to watch when you make it is channel purity, not speed.