Comments on Shnoll Method

This page will contain various emails and references discussing the Shnoll work. Thus far we have:

  1. Tatanya Zenchenko's description of the beginning of the work

  2. John Walker's examination of random (radioactive) data using algorithmic methods

  3. Dick Bierman, letter describing Axford confirmation

  4. Edwin Pozharski, Re: human judging -> algorithm

  5. Dean Radin, correlational analysis (not yet available)

  6. Email from Shnoll following Euro-SSE meeting, including abstracts

  7. Tatanya Zenchenko's description of the beginning of the work

    In response to my question about the genesis of Simon Shnoll's quite remarkable procedure, one of his colleagues, Tatanya Zenchenko kindly described the history of his work. This is a slightly edited version of the story.

    Date: Tue, 11 Jul 2000 20:40:39 +0400
    From: Tatyana Zenchenko 
    To: rdnelson 
    Subject: Re: Shnoll method, was Randomness in Amsterdam
    > I would like to repeat this question.  Even though it may be
    > difficult to explain, I and others would appreciate some
    > discussion, which should lead to more efficient effort to
    > develop robust automated analysis, and ultimately to more
    > valuable insight on the possible cosmophysical influences.
    > Best wishes,
    > Roger
    Sorry for delay, I'll try to answer your question (as far as I know the
    history) but it is actually very long story.
    About 45 years ago Simon Shnoll was a biochemist dealing with muscle
    proteins. He was a very good experimentalist, but during experiment he
    obtained two distinct values of subsequent measurements of chemical
    reaction rate. For example, 100 and 120, but almost never - 110. It
    contradicted theory... He repeated his probes up to 300 times every
    day, under absolutely equivalent conditions, but situation remained the
    same. Now we can say that he got one histogram (300 points) per day and
    these histograms didn't look like smooth normal distributions, but very 
    often they were similar to each other: they had two distinct peaks and deep 
    gaps between them. Moreover, sometimes this "two-peak" distribution kept the 
    same form for two or three days, and after that it became "two-and-half peaks" 
    and so on. (Of course the matching of these histograms requared shifting and
    rescaling, but the main idea, "the idea of form" remained the same).
    When all possible artifacts were excluded, he supposed that the point is
    in a special properties of muscle proteins. Then he asked his teacher,
    great russian biochemist Vladimir Engelhardt, "why are these repeated probes
    so different?" and Engelhardt answered: "Don't do so many repetitions 
    and this effect will disappear"...
    Simon Shnoll wrote in his laboratory journal "[I plan] to find out the 
    origin of strange result distributions and return to main question". 
    After that 45 years passed. He did not return yet [to the biochemistry].
    During these years he has examined many systems. If the explanation is special
    properties of myosin, let us take another protein where the distributions
    should not be strange, should not have sharp peaks and "inertia" in their
    form. But other proteins had similar behaviour. If we take a chemical, not
    a biochemical system? The same. It was very labour-intensive work and he could 
    do only one histogram per day. But these histogram sequences had some logic in
    their development. But there were plenty of possible explanations of this
    effect, both scientific and artifactual. All of them needed to be examined,
    so it took many years. During this long time Dr. Shnoll suspected that
    oscillations in homogeneous aqueous system can exist. He was right: under
    his head in his laboratory the Belousov-Zhabotinskii reaction was
    discovered. But in any case these oscillations could not explain the
    effect he was interested in, because around that time he observed the same
    effect in distributions of radioactivity measurements (radioactivity was
    his second speciality).
    Since that discovery (in 1986) Prof. Shnoll and his team use radioactivity
    measurements as a main object of experiments because of absence of
    trivial factors affecting this process.  So, for scientist who usually 
    asks unusual questions it was not so hard to apply this approach 
    (histogram sequences) to abnormalities of biochemical reaction, [and
    then to other systems].  After that there was just the search for an answer.
    The last part of this story you read in our first UFN paper, where an
    exact description of the phenomenon is given. All the words above are only
    the "poetry", but I hope it gives the answer for your question.
    I should note that it was a real (and rare) pleasure to read the
    interpretation of reader who  understood the sense of our articles so
    I am sorry for my English, I hope this story is written clearly to
    understand. But if you want to publish this message in Internet (I guess
    it is so) I would be grateful for some help in English style.
    P.S. Thank for your care, all your emails were received. There was mistake 
    in Prof. Shnoll's address, but he got your message from me.

    John Walker's algorithmic methods

    Date: Wed, 12 Jul 2000 02:25:25 +0200
    From: John Walker 
    To: rdnelson 
    Cc: Dick J Bierman , nick herbert ,
         Jack Sarfatti ,
         Dean Radin ,
    Subject: Statistical pitfalls in interpreting Shnoll et al replications
    Please excuse the unusual format of this message.  Rather
    than including figures as attachments, I've provided URLs
    which download them so as to avoid burdening you with
    images you may not be interested in viewing.
    For the last two weeks, I've been operating a radioactive
    decay based random event generator and tabulating the
    data to test for the effect reported in Shnoll et al [1].
    Having accumulated several days' data, I reduced them by
    computing histograms for 10 minute experiments, each
    consisting of 100 measurements synchronised to the start
    of a minute.  From these I generated smoothed histograms using
    an exponentially smoothed moving average with a smoothing
    constant of 0.2.
    These histograms were read by an analysis program which
    aligned them by their mean values and then computed
    a chi-square goodness of fit between the aligned curves.
    I did not normalise maximum values, adjust variance, or
    test matching of mirrored curves--all of these are easy
    to implement.
    All pairs of histograms were evaluated for closeness of fit
    and the results sorted from the closest match to most distant.  One
    can then plot the time interval for histograms with the closest
    match (in this case, the closest 100,000 of a total of 1,738,442
    pairs of histograms).
    	(NOTE: Use your browser's back button to 
    	 return here after viewing these figures.)
    Closest 100,000 pairs 
    Yaaar!!!  Look at that spike at zero!  How similar it is to
    figures 3 and 4 of [2].  Indeed....
    As a control, let's make the same plot for the the worst 100,000
    matches of our experiment pairs:
    Worst 100,000 pairs
    Hmmm...pretty similar.  If the effect were real, we'd expect the peak
    to be attenuated as we broaden the similarity criterion.  Okay...let's
    make a plot of the *entire* data set--certainly this should suppress
    any "Cosmophysical" influence, as we're comparing every pair of
    experiments regardless of their closeness, however defined:
    All pairs
    Well, hello!  Look at those textbook peaks at 0, 24, and 48 hours.
    What do they mean?  Well, it's all really rather simple.  Note that
    the 0 hour peak in all of the graphs, as well as figures 3 and 4 of [2]
    is almost always about twice the value of the neighbouring bars.
    This is a simple consequence of binning.  If one computes the number
    of hours difference between two experiments conducted at Unix time_t
    second values t1 and t2 as:
            hbin = (t1 - t2) / (60 * 60);
    then the zero bin will encompass all values between -3599 and +3599,
    while the rest of the bins will span only one hour instead of two.
    Consequently, the zero bin will have on the average twice as
    many samples, as shown in these examples and figured 3 and 4 in [2].
    Periodicities at the 24 and 48 hour intervals are clearly
    apparent in the above-mentioned chart.  These are readily
    explained by daily cycles of system administration.  Histogram
    time distance statistics are sensitively dependent on regular
    cycles which interrupt the collection of data.  In this case,
    I tended to shut down the data collection to run
    other programs around 18:00 UTC every day, and the histogram
    analysis fingered it.
    Finally, as a control (and to illustrate the effect of double-counting
    zero hour histograms), I prepared pseudorandom data for 10 days of
    experiments and processed it in the same manner yielding the plot:
    Pseudorandom data
    This make it clear how the delta-T = 0 spike is purely an artefact
    of double-counting assignments to the zero bin.
    With this in mind, take a look at:
    Shnoll assessment of GCP data
    Note how closely it resembles these plots--in particular how
    the central peak is close to twice that of the adjacent values.
    The crucial thing in reducing data from experiments of this
    type is to insist on absolute sign symmetry--binning ±0 into
    one slot will result in spikes around the zero point which
    are not present in the raw data.
    Having developed a modular toolkit for analysing experiments of
    this type, I'll try various other measures of closeness
    over the next few days.  All of this, including source code,
    will eventually be posted once I beat it into something others
    can understand.  If anybody on this list wants a copy of the
    code as it stands, let me know and I'll make a copy available
    to you.  But beware...this is "ad hack" SGI/IRIX and Sun/Solaris code
    written without the slightest thought of portability.  It'll
    probably work on most POSIX-like platforms, but if it doesn't,
    you're on your own; I typically budget three to six months
    for portability testing of Unix code I publish--we're working
    in real-time here, so I'll lighten up if you're willing to
    debug on your own.

    Bierman letter describing Axford confirmation

    Date: Wed, 12 Jul 2000 12:43:59 +0200
    From: Dick J Bierman 
    To: John Walker , rdnelson 
    Cc: nick herbert , Jack Sarfatti ,
         Dean Radin ,
    Subject: Shnoll et al replications
    Hi to all,
    We just recieved a confirmation by W.I. Axford of the Max-Plank 
    Aeronomy  Institute, Lindau, Germany, that he has done an independent 
    replication  of the Schnoll effect.
    If I understand him correctly he just produced sets of  histograms 
    from two (random?) sources, removed all time information, randomized 
    the order and then send them to Tatiana for human judgement.
    She then returned to them the pairs that were simlar and these turned 
    out to be from simultaneous measurements.
    No stats were given but from his words it seems the result is robust. 
    (He is a bit worried about the stretching operation but that can't 
    explain the results; it is worrying from a physics perspective though).
    So this is a good reason (at least for me) to become more optimistic 
    and to invest a bit more time in getting the human judgement replaced 
    by computerized judgement.
    The results produced by John suggest to me that the chi-2 isn't the 
    measure that corresponds very well to their human scored similarity.
    I have been thinking about a bit different approach.
    The idea is that two patterns are more alike if the program that you 
    have to write in order to transform one into the other is shorter.
    One possible instantiation of this idea might be to fit both 
    histograms with a polynomial and then derive the transformation formula.
    Not simple but doable. There might be other and probably better ways to
    fit the histograms. Wavelet analysis comes to mind.
    Anybody ideas?

    Pozharski: Notes on algorithmic efforts

    Date: Fri, 4 Aug 2000 09:11:11 +0200
    From: Dick J Bierman 
    To: Edwin Pozharski 
    Subject: Re: human judging -> algorithm
    Thanks Edwin,
    I was considering neural nets too, so this mail saves me lots of time!
    Anyway, at this point I think we should go for presenting the 
    algorithmic failures to the human judge and ask to do them again while 
    thinking aloud.  Also I would like to see test-retest reliability for 
    the human procedure.  As you noted in an earlier mail it not too 
    difficult to get algorithmic similarity where we see it easily.  We 
    should therefore focus on the cases where the similarity is not obvious.
    My present feeling is that the human guess procedure introduces some
    paranormal effects. These can easily be sorted out because of their
    notorious lack of test-retest reliability.
    The problem of the latter approach is that we need the original human
    judge and much time. So we need to find somebody who is willing to 
    spend his time on this. Or to find considerable amounts of money to 
    hire such a person.
    PS It might be a good idea to summarize somewhere what everybody tried so
    that we don't repeat the same 'mistakes'. I did a.o. Fourier and Wavelet
    analysis on untransformed histograms. Wavelet could be tried again after
    transformation (stretching/shifting). Fourier analysis is rather
    unsensitive to these transformations.
    >once again sorry about the delay with response. Below is the same
    >description I sent to Dean Radin: I could write another one, but it
    >would basically be the same.
    >> Dean,
    >> I will describe briefly what I tried to do and then you will ask
    >> about particular details in case you will find it necessary.
    >> 1. Different measures of the correlation between histogram vectors
    >> itself.
    >> This is, probably, the very first thing which comes to everybody's
    >> mind. Of course, it doesn't work if you simply calculate correlation
    >> coefficient between the unperturbed histograms, since they are
    >> shifted, stretched and probably mirrored. The crucial improvement 
    >> came when I optimized the scalar product (which is the same as the 
    >> correlation coefficient) by shift and stretch.
    >> 2. Neural nets.
    >> I tried backpropagation and Kohonen networks. Backpropagation failed
    >> completely being applied simply to the original histograms. It
    >> definitely needs some conversion of the histogram to the the set of
    >> parameters describing its shape, but I didn't find right one yet. Of
    >> course, this learning uses sets of pairs of similar histograms,
    >> pre-judged by human. The key point of Kohonen learning is how to find
    >> the winning pattern - I used the same optimized scalar product for it.
    >> Actually, it works as good as optimized scalar product itself -
    >> recognizes similar histograms pretty well but produces a little bit
    >> too much wrong pairing. But it gives one unteresting things - the 
    >> set of patterns it finds in the whole dataset.
    >> 3. Converting histogram into a peak sequence.
    >> This approach is based on what we think is how human makes the
    >> judgement. What seems to be the most important is relative positions
    >> and heights of peaks which forms the general shape, or pattern, of the
    >> histogram. Dr Konradov is the one who spent more time trying this as
    >> I did, so in this particular case it makes sense to ask him.
    >> There was one guy in the past, who tried to fit histograms by
    >> polynomials ans then made cluster analysis on polynomial
    >> coefficients.
    >Edwin Pozharski, PhD
    >Postdoctoral Fellow, Department of Biochemistry,
    >Northwestern University, Evanston, IL, 60208

    Radin analysis (not yet available)

    Recent briefs by Shnoll and Kirillov

    From Sat Nov 11 11:07:43 2000
    Date: Sat, 28 Oct 2000 19:44:58 +0300
    From: Simon Shnoll 
    To: rdnelson 
    Cc: Dick J Bierman , Jack Sarfatti ,
         nick herbert , neil slade ,
         Amit Goswami , C Levit ,
         Faustin Bray ,
         Lyle Fuller , Mark Comings ,
         Marcello Truzzi ,
         Paul Zielinski ,
         Russell Targ , Saul-Paul Sirag ,
         Shipi Shtrang , Tony Smith ,
         Vladimir Poponin ,
         zenchenko ,,, John Walker ,
         Hal Puthoff , John Alexander 
    Subject: Re: Shnoll method, was Randomness in Amsterdam
    Dear Dr. Sarfatti, Dr. Nelson, and all colleagues, participating in
    summer 2000 Internet disscussion on Shnoll effect.
    1. Continuation of the discussion initiated by Dr. Sarfatti on effects
    which we have found  was delayed up to the end of SSE Meeting in
    Amsterdam 20-23 October. Now the Meeting is over. An important result of
    contacts in Amsterdam is the understanding that classic statistic
    "criteria of agreement of hypotheses" are inappropriate indeed for
    evaluation of similarity of thin structure of histograms because this is
    determined by certain cosmogonic influences but is not of probability
    nature. We have demonstrated before and during  the Meeting that human
    judging of randomized histogram series gives quite objective and valid
    results. However this work is labour-consuming. Therefore the
    elaboration of first computer programs which can substitute human
    judging are most promising results obtained in our laboratory by
    M.Fedorov and A.Konradov recently. The main manifestations of our
    phenomenon is replicated by new programs.
    2. We  had a fruitful discussion with Dr.J. Walker near computer with
    demonstration  of our expert program, Histogram Manager and handed this
    to him.  Unfortunately Prof. D.Bierman could not discuss our
    disagreements   because pressing  organizing obligations.
    3. I have presented  (See Add.#1)  new results  of our 
    investigations carried out in our laboratory and simultaneously with :
    1) prof. Axford and dr. Wilken  in Max-Plank Institute fur Aeronomy in
    Lindau; 2)  prof. L.Belousov in International Institute of Biophysics
    headed by prof. F. Popp in Neuss (Dusseldorf)  and dr.V. Voeikov in
    Moscow State University. In the last case the precision  of time
    resolution during  histogram comparison was 1 min.
    As previously similar histograms in different places were appeared at
    the same local time.
    This allows to conclude that  during Earth rotation each geographic
    point passes through heterogeneous space  ( wjthin the range of middle
    latitudes) with scale of heterogeneity no more than 20 km. It can be
    also concluded that forces causing histogram patterns are outside solar
    system because cycle of repeated appearance of similar histograms is 23
    h 56 min,  i.e. «star day».
    4. It was important that dr. A.Kirillov presented his  theory of
    Space-Time which explains our phenomenon : Space-Time Fluctuations as a
    possible explanation of the «Shnoll Effect» (See Add. #2).
    5. We did not present our initial results of investigation of temporary
    rows, obtained generators in GCP before discussion with main authors who
    elaborated this system.
    As before we are ready for collaboration,
    				Simon S. Shnoll
    Macroscopic fluctuations in processes of different nature as a  
    result of cosmophysical (cosmogonic) causes. Possible heterogeneity 
    (discretness) of space-time.
    S.E.Shnoll, T.A.Zenchenko, K.I.Zenchenko, M.V.Fedorov, E.V.Pozharskii,
    A.A.Konradov,I.M.Zvereva, V.A.Kolombet
    Moscow State University, Physics Department, Moscow; Institute of
    Theoretical and Experimental Biophysics, Russian Academy of Sciences,
    142290 Puschino, Russia, E-mail:
    50-year  study of  dispersion in  measurements the rates of  different
    processes shows that this is not experimental error but manifestation
    of  fluctuations caused by cosmophysical factors. Our original tool for
    investigation of dispersion in  temporary rows  is comparison the fine
    structure of histograms obtained from experimental time series.
    Histograms were obtained according to small non-overlapping successive
    segments of time series. The  fine structure of histograms distinctly 
    changes in time. The similar histograms are observed with high
    probability simultaneously in different processes and even at a great
    distances between points of measurements. This effect evidences
    cosmogonic phenomenon determining fluctuations in any process
    irrespective of its characteristic scale of energy.
    The phenomenon can be the result  of  fluctuations of four-dimensional
    space-time, related to  non-uniformity (heterogeneity) of gravitational
    structure of the World. During Earth rotation around its axis and along
    near solar orbit particular parts of earth surface are  regularly
    exposed to  different gravitational heterogeneity's and this is
    manifested in respective forms of histograms. The histogram patterns are
    like interferencional pictures and may be the result of interferention
    of coherent cosmogonic waves.
    The statements above are based on many year, long-term investigations
    (the first publication was in 1958) of different processes with careful 
    discriminations of possible artifacts. Reviews of main results have been
    regularly published in Russian and English (see references in
    Physics-Uspekhi 41 (10) 1025-1035 (1998) ; 43 (2) 205-209 (2000)). The
    investigation of the phenomenon was started from biochemical reaction
    rates in the 50's, was continued in chemical reaction rates in the 70's
    and during last 20 years are carried out preferably with radioactive
    decay. The latter allows to exclude trivial "earth's" explanations of
    the observed effects.
    The general conclusions are based on the following experimental results.
    1.  High probability (p<10 -7 - 10 -8 ) of fine structure histogram
    similarity in the nearest, neigbouring  time intervals : "near zone
    2.  High probability of repeated appearance of similar histograms with
    periods near 24 hours, 27 days and a year.
    3.  High probability of histogram similarity at any given time in
    independent measurements of  different processes in the same geographic
    4.  High probability of histogram similarity at the same LOCAL time in
    measurements of different processes in different geographic points.
    5.  Recent data showing that ascertained period of repeated appearance
    of similar histograms is 23 h. 56 min, i.e. star but not solar days.
    Most recent previous data will be  also reported on our study of
    histograms obtained in time series in "egg-generators" of GCP - net. 
    Space-Time Fluctuations as a Possible Explanation of the "Shnoll-Effect".
    Inst. Of Applied Mathematics and Cybernetics Nighnii Novgorod, Russia
    In gravitation theory it is assumed that at Planck scales spacetime
    acquires a foamlike structure as a result of quantum fluctuations. If we
    believe in the fact that  our Universe had a quantum period of evolution
    in the past, then we should expect the existence of traces (relicts) of
    such fluctuations at macroscopic or even cosmological scales. In this
    report we show that a nontrivial quantum structure of our space at
    macroscopic scales (wich may be the result of the fluctuations we just
    pointed out) gives rise to a new fundamental phenomenon: spontaneous
    origin of an interference picture in every physical processes. This
    gives a possible explanation of the fine structure of histograms
    observed in radioactivity measurements (Shnoll Effect) wich, therefore,
    can possible serve as a test of the real structure of space.

GCP Home