Strangely enough I posted a similar question about the Quadcurve EQ recently: http://forum.cakewalk.com...aspx?m=3086433&p=1

Bill mentioned 'peaks' so that got me thinking. I thought perhaps the peak values may have been increasing while the rms (VU) levels were not so much.  Because I was metering with a VU meter I thought perhaps the VU's may have not been showing any upward level shift.
 
Anyway I did a check again with a really great and large peak meter setup and I still got the same result.  ie no change in level upwards no matter what I did.  I am using the Studio One 'Pro EQ' in High Quality mode in case you are wondering.

I researched this a little when I was writing the EQ section of my FX video and was looking into how a regular EQ filter works as opposed to a Linear Phase EQ .
 
A regular EQ will always impart a very slight delay on material and hence a slight phase shift which in turn leads to some comb filtering. This will be more noticeable around the centre frequency. Therefore there will always be a little wave distortion/boost around the cut off frequency of a lo cut filter. The steeper the cut the worse the phase shift, hence the distortion of the wave and boost BTW.
 
Having said that when cutting it's also less noticeable than if applying a boost to say a peak filter. I am no expert by any means but my understanding is while the artefacts introduced are there I would doubt very much if they are capable of introducing clipping unless of course the material is very, very close to clipping already.
 
A linear phase EQ uses a much larger audio buffer hence no phase shift but uses more processing power in the first place.
 
Edit for detail.

Linear phase EQ produces "pre-ringing"... I wouldn't be surprised if an "unanticipated" peak was observed with a linear phase EQ filter.
 
I just got up, will have some coffee, and then describe the circumstances with more detail and try to answer some of the questions asked above.

mike_mccue
Linear phase EQ produces "pre-ringing"... I wouldn't be surprised if an "unanticipated" peak was observed with a linear phase EQ filter.
 
I just got up, will have some coffee, and then describe the circumstances with more detail and try to answer some of the questions asked above.

Me too! Except I am enjoying my coffee as I read this.
 
Waiting...

I'm pretty sure this doesn't have to do with any resonant bump but with phase cancellation. Imagine a 100Hz wave and a 50Hz wave together on a single track, and you push the total level close to 0dBfs. Depending on the existing phase relationship of the two waves, one of them may/will appear to "cut off the peaks" of the other. If you then use an EQ to inhibit/cut either one, the other one will suddenly be able to reach full peak which may clip the signal. We tend to think of sound as stacked bits that increase in loudness depending on how much content you add but that's just not the way it works. We always end up with a single waveform and every thing we define as a single "sound" is affected by all the components.

I found the explanations about the phase interference very interesting and like Jeff, I had not noticed the peaking behavior before.
 
I am also familiar with the *idea* that band pass filters are said to not have "Q" factors. My college instructor from so many years ago use to say things like this to us.
 
Over the years I have learned that many Hi pass and Lo pass filters do have "Q" factors included as an adjustable parameter, and more recently, I have learned that the Q factor, in this context, effectively describes the knee, or corner of the transition to the slope.
 
In the example of the FabFilter Pro-Q2, there is a Q adjustment provided for all of the Lo-cut slope choices except for the 6dB/octave choice.
 
 
 
 
 
Regarding the implication of the interference filtering: I think I have an explanation for why the material I was working with made the issue so apparent.
 
I was working on a listening test that focused on the quietest portion of the trail off of a sustain to hear the effects of a reverb I work with. I was also making another test of listening for the effects of dither choices because I wanted to re-evaluate my opinions about the usefulness of dither.
 
 To that end I was working with a sample of a single hit of a crash cymbal. I made the sample with by combining tracks made with a pair of Coles 4038 overhead mics and a Schoeps MK41 SDC used as a close mic. The 4038s came in through a Chandler TG2, and the MK41 came in through a John Hardy M1. I explain this to point out that the sound of the cymbal was rich and detailed and the sustain was long and clear.
 
 After mixing the three mics into a stereo bus I added an instance of Exponential Audio's Phoenix reverb and set up a "plate" which I adjusted to use an uncharacteristic 10 second reverb time. This served to add detail and texture to the quietest parts of the cymbal's sustain.
 
 Then I added a limiter to the master bus and set it up so that it was easy to hear rather long sustain and used this to make new samples that I could test with my dither tests. (don't get distracted by a discussion of dither... it is inconsequential to this discussion about EQ) The limiter was set to look ahead and had an absolute ceiling of -1.0dBFS. I was not especially concerned with the aesthetic of how low loud the initial transient seemed, and I was not concerned with "rules" that discourage people from working so closely to 0dBFS.
 
 Finally I dropped an instance of FabFilter Pro-Q 2 on the master bus and set up a 48dB slope lo-cut at 30Hz with a Q of 1. That's when I noticed the peaking. Normally I would have dragged the instance of the EQ to precede the limiter and I would not have noticed the peaking so readily, but in this case I made my adjustments to the EQ and saw the very obvious extra peaking on the output meters before I placed the EQ  where I really wanted it in the fx chain.
 
 It was easy to see that with the EQ bypassed the peaking was matching the limiters -1.0dBFS output, but with the EQ engaged it was at +1.0dBFS. Wow.
 
 
 
 So... having explained all the circumstances, and having considered what has been explained by those who have responded, I am going to propose the possibility that the "in-harmonic" nature of a cymbal's "indefinite pitch percussion" voice, and its especially closely spaced and "randomly" mixed harmonic content, increases the possibility that subtracting with an EQ filter can result in interference filtering that produces increased peaking at the output.
 
 I find the idea fascinating and imagine I'll be thinking about it for quite a while.
 
 I prepared an animated gif that compares the playback of the cymbal crash with the EQ engaged and bypassed. The graph shows the "frozen" maximum peaks. You may be able to discern how many of the peaks along the spectrum, specifically in the mid range, seem to be slightly higher in level, and perhaps you may imagine how the small variances can sum to an increase that the broadband PPM meter reports as a difference of 2dBFS. 
 

 
Thanks to everyone for sharing ideas. I look forward to any further comments you may have.
 

Hard to tell from just a picture even an animated one, but I don't see a peak being produced. I see a peak that isn't being reduced as one would expect.

I agree that the picture isn't very helpful, but that's because you're looking at a spectrum rather than a summed wave. You can at least see a pretty powerful low roll off. With a signal that close to 0dBfs it's not hard to see how that might increase peaking. I don't really know how to explain it any better than I already did, nor am I really an expert on the matter.

Some people may be able to see the increased peaking levels in the mid range frequencies, and some people may miss them.
 
The picture isn't meant to serve as a proof that a peak level boost is happening. The broadband PPM meter in SONAR shows the peak level very clearly, and I guess people can either believe me when I claim that it is happening, or not. :-)
 
What I provided the picture for was to casually illustrate, for myself, the nature of the incremental increases and decreases along the spectrum that result in the 2dBFS increase shown on the PPM meter. I thought some other people might want to be able to see how it illustrates the explanation that the components of the series will interact differently when the lo has been cut and when the EQ has been bypassed.
 
 
 
 
 
 
 
"We always end up with a single waveform and every thing we define as a single "sound" is affected by all the components."
 
You can just say "Fourier Series" :-)