From tamasgunda@tigris.klte.hu Tue Aug 29 04:26:03 1995 Received: from tigris.klte.hu for tamasgunda@tigris.klte.hu by www.ccl.net (8.6.10/950822.1) id EAA22898; Tue, 29 Aug 1995 04:19:54 -0400 Message-Id: <199508290819.EAA22898@www.ccl.net> Received: from anti02 (anti02.chem.klte.hu) by tigris.klte.hu (MX V4.1 VAX) with SMTP; Tue, 29 Aug 1995 10:19:48 EDT Sender: X-MX-Warning: Warning -- Invalid "From" header. From: "tamasgunda@tigris.klte.hu" To: chemistry@www.ccl.net Date: Tue, 29 Aug 1995 10:19:42 +1 MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7BIT Subject: Principal components summary Priority: normal X-mailer: Pegasus Mail/Windows (v1.22) Some days ago I sent the following summary to CCL, but somehow only a small part has arrived. Here is it again. --------------------------------------------------------------- Here is the summary of my question concerning the calculation of principal components and plane fitting to points: The original question was: I am looking for algorithm/numerical methods for the determining of the principal components of a data set. In the words of geo- metry, I have a number of points determined by y1, y2 and y3, and I am looking for the best fitting plane to the points, i.e. the plane determined by the two greatest principal components. Do not refer, please, to programs like Mathcad etc. I need it in a numerical method form, as I'd like to use it as a procedure in a program. *************************************************************** ** That looks like a rather straightforward least-squares problem. You want the parameters a,b,c in the equation for a plane y3 = a y1 + b y2 + c. So you make up a matrix X: y1(1) y2(1) 1 y1(2) y2(2) 1 ... y1(n) y2(n) 1 and a vector Z = (y3(1), y3(2), ..., y3(n)), plus a 3-component vector of unknowns A = (a,b,c). Then A = X^+ Z gives you the best values for the parameters, where X^+ is the generalized inverse of X. Konrad Hinsen | E-Mail: hinsenk@ere.umontreal.ca Departement de chimie | Tel.: +1-514-343-6111 ext. 3953 Universite de Montreal | Fax: +1-514-343-7586 C.P. 6128, succ. A | Deutsch/Esperanto/English/Nederlands/ Montreal (QC) H3C 3J7 | Francais (phase experimentale) *************************************************************** ** From: gordonh@chem.QueensU.CA (Heather Gordon) To: tamasgunda@tigris.klte.hu Subject: pca There are listings for FORTRAN programs pertaining to all aspects of factor analyses, including pca in "Factor Analysis of Data Matrices", P. Horst, Holt, Rinehart and Winston, New York, 1965. I also have written my own code, using the routines TRED2 and TQLI >from "Numerical Recipes in Fortran" to diagonalize the covariance matrix and locate the eigenvalues and eigenvectors. Heather Gordon Department of Chemistry Queen's University Kingston, Ontario *************************************************************** What you say you want sounds a lot like the transformation of a molecule to the coordinate system in which the moment-of-inertia tensor is diagonal. If your (y1, y2, y3) components are *orthogonal*, you ought to be able a similar approach. This kind of code is buried in all sorts of programs . It should be a simple matter to adapt a suitable code fragment to you use. ++++++++++++++++++++++++++++++++++++++++++++++++++ FREDERIC A. VAN-CATLEDGE Scientific Computing Division || Office: (302) 695-1187 or 529-2076 Central Research & Development Dept. || The DuPont Company || FAX: (302) 695-9658 P. O. Box 80320 || Wilmington DE 19880-0320 || Internet: fredvc@esvax.dnet.dupont.com ************************************************************* ** Tamas, A friend sent me your e-mail on the subject of an algorithm for calculating the first two PC's from a dataset. I have written PCA (NIPALS) routines in BASIC and C and will be happy to send them to you, if required. I also have a book (Numerical Recipes) which details PCA (Singular Value Decomposition) which does a similar thing. If you still need some help, send me an e-mail Steve Gurden, Bristol Chemometrics Group, School of Chemistry, University of Bristol, Cantock's Close, BRISTOL BS8 1TS Tel: +44 (0)117 9289000 extn. 4421 e-mail: S.P.Gurden@bris.ac.uk *************************************************************** ** Dear Dr. Gunda, One of the most simple and elegant method to perform Principal Component Analysis is that of the NIPALS algorithm. It can be formulated in a few lines of programming and it works really fine. You can find it in the following references: * Analytica Chimica Acta, vol. 185, p. 1-17 (1986) * Chemometrics and Intelligent Laboratory Systems, vol. 2, p. 37-52 (1987) I hope this helps you. Sincerely. Jose I. Garcia -- Dr. Jose Ignacio Garcia-Laureiro Phone : 34-(9)76-350475 Departamento de Quimica Organica Fax : 34-(9)76-567920 Instituto de Ciencia de Materiales de Aragon e-mail: jig@qorg.unizar.es C.S.I.C.-Universidad de Zaragoza jig@msf.unizar.es E-50009 ZARAGOZA (SPAIN) *************************************************************** ** I thought the principle (no pun intended) was quite simple; just calculate the variance/covariance matrix of the variables (or better the correlation matrix) and diagonalize that. You can get the correlation matrix by autoscaling the data and calculating the covariance matrix then. Next, select the number of principal components that you want by calculating the cumulative sum of the eigenvalues divided by the sum of all the eigenvalues (the eigenvalues, eigenvectors need to be sorted by their size by the diagonalization procedure or afterwards) I mean: T = \sum_i \labda_i suppose, as an example: \labda_1 / T = 90 % (\labda_1 + \labda_2) / T = 98 % ... (\labda_1 + \labda_2 + ...) /T = .. if this is sufficiently accurate for you, use the subspace spanned by the first two eigenvectors as your new space, and project all datapoints onto it. otherwise, use more eigenvectors until (\sum_j \labda_j) / T is sufficiently near 100 % (1.00), and project the datapoints on the subspace (unfortunately it is no longer a 2-D plane, which would be most easy for viewing) Hopefully you can use this information, I can't think of any references off-hand.. There is a book by Box (and Hunter?) which describes these multivariate techniques in great detail. Sorry I can't help you better, Frits Frits Daalmans OIO Conformational Analysis Gorlaeus Laboratoria Leiden, The Netherlands E-mail: frits@chemde4.leidenuniv.nl Tel: [+31] (0)71-274505 *************************************************************** ** Check the NIST CD for handwritten characters -- there is a principal component algorithm coded there (may be in C, but you could then recode it in FORTRAN). There are several sites that catalog algorithms, too -- here are a few: http://gams.nist.gov/, http://netlib.att.com/netlib/att/cs/home/1127.ht, http://www.netlib.org/. Good luck. Joe McDaniel joe@psiint.com Dear Dr. Gunda, There are many references to Principal Component Analysis.It is actually a straightforward problem in matrix diagonalization. However, if your real goal is to simply fit a plane to a set of points (minimizing the sum of squares of the distances of the points to the plane) or a line to a set of points (minimizing the sum of squares of the distances of the points to the line), then this is also a simple matrix diagonalization problem whose solution is given and discussed in the following references: "To Fit a Plane or Line to a Set of Points by Least Squares", V. Schomaker, J. Waser, R.E. Marsh, and G. Bergman, Acta Cryst., vol. 12, pp. 600-604 (1959). and "To fit a plane to a set of points by least squares", D.M. Blow, Acta Cryst., vol. 13, p. 168 (1960). Regards, Marvin Waldman, Ph.D. Director, Rational Drug Design Biosym Technologies, Inc. e-mail: marvin@biosym.com *************************************************************** ** Dear Dr. Gunda, There are two ways to do what you want (as I understand it). One is to use a statistical analysis package which supports principal components analysis as statisticians define it. The plane you are looking for is the one defined by the first two principal components. The other is to perform the equivalent of a moment of inertia calculation with all masses set to 1. First find the mean value of each of the three coordinates (call them x, y, z). This is equivlant to your center of mass. Build the symmetric 3x3 tensor with the following elements, summed over all points (distances are relative to the centroid): y*y + z*z -x*y -x*z -x*y x*x + z*z -y*z -x*z -y*z x*x + y*y Diagonalize the tensor (the Jacobi method works well here - see any of the "Numerical Recipes" series by Press et all.) The transformation matrix gives you your three principal coordinates. Hope this helps, Paul Soper ----------------------------------------------------------------- Paul Soper All the usual disclaimers apply DuPont Central Research soperpd@esvax.dnet.dupont.com P.O. Box 80328 Tel (302)-695-1757 Wilmington, DE 19880-0328 FAX (302)-695-8412 *************************************************************** ** According to the math of the normal least square procedure, as far as I know, NOT the distances of the points and the line are minimized (i.e. a perdendicular from a point to the line), but the difference of the measured and the calculated y values,which is represented graphically by a line between the point and the regression line and it is parallel with the y axis: That depends on the error criterion that you use. The one sent yesterday (and probably some others too) does indeed minimize the error along one coordinate axis (and in fact works only if that axis does not lie in the plane that you want to fit). If you want to minimize the distances of the points >from the plane, use the normal form for the plane: n x - d = 0, where n is the normalized normal vector, x is the coordinate vector, and d the distance of the origin from the plane. The distance of any point y >from this plane is simply given by n y - d, so you must minimize __ \ | | 2 / | n y_i - d | -- i with the constraint that n is normalized, so you need a least-squaresminimizer that can handle constraints (e.g. via Lagrange multipliers). Or use a "dirty hack": if you know (or assume) that d will never be zero (i.e. the plane will not contain the coordinate origin), set d = 1 and use an unnormalized normal vector, whose length then is the inverse of the distance from the origin. ------------------------------------------------------------------------------ - Konrad Hinsen | E-Mail: hinsenk@ere.umontreal.ca Departement de chimie | Tel.: +1-514-343-6111 ext. 3953 Universite de Montreal | Fax: +1-514-343-7586 C.P. 6128, succ. A | Deutsch/Esperanto/English/Nederlands/ Montreal (QC) H3C 3J7 | Francais (phase experimentale) ------------------------------------------------------------------------------ *************************************************************** ** What you say you want sounds a lot like the transformation of a molecule to the coordinate system in which the moment-of-inertia tensor is diagonal. If your (y1, y2, y3) components are *orthogonal*, you ought to be able a similar approach. This kind of code is buried in all sorts of programs. It should be a simple matter to adapt a suitable code fragment to you use. FREDERIC A. VAN-CATLEDGE Scientific Computing Division || Office: (302) 695-1187 or 529-2076 Central Research & Development Dept. || The DuPont Company || FAX: (302) 695-9658 P. O. Box 80320 || Wilmington DE 19880-0320 || Internet: fredvc@esvax.dnet.dupont.com Tamas, I think I have solved a problem fairly similar to the one you are interested in. My problem was less general and involved finding thebest-fit plane to a group of four points centred around a fifth point and arose because I am looking at compression strain in tetracoordinate carbon systems. For example the molecule [4.4.4.4]fenestrane C9H12 H H2C__C__CH2 | | | HC-----CH | | | H2C--C--CH2 H contains a central "flattened" tetracoordinate carbon atom. One way to determine the flattening at any C(C)4 moiety is to find the best-fit plane for the four alpha-C atoms passing through the central C atom. This can be done simply by finding the eigenvectors of what we have termed the Geometry Tensor. The Geometry Tensor is constructed by multiplying the matrix D (the vectors to your points -- in this case the 4 alpha-C atoms) by its transpose to give a 3x3 Real Symmetric Matrix (routines to find the eigenvectors/values of any RSM can be found in eispak and similar sets of routines). The eigenvectors of this matrix will correspond to the minimum (Best-Fit) -- this will have the smallest eigenvalue, maximum and one other extremum (i.e. they will form a set of cartesian axes -- x,y and z -- one of which will be the normal to your best-fit plane. The mathematics behind this is actually quite simple and involes setting up the equations to minimise d' = (sum of squared distances of your points to an orbitrary plane). After which the need to form and find the eigenvectors of the "Geometry Tensor" becomes obvious. Naturally this can be readily extended to any number of atoms and any origin. I have used this method to re-orient molecules that were optimised in C1 symmetry (in cartesians) which exhibit higher symmetry but where the axes/planes of symmetry do not lie on the x,y or z axes of my final cartesians. I might be able to send you my fortran code (it is very simple) if you think this would help but I will need to check the origin of the eigenvector solving routines that I use (to make sure we do not break anyone's intellectual property rights as I did not write these routines myself -- I think they came from eispak). Cheers, Danne Danne R Rasmussen, PhD Student phone: +61 6 249-3771 Research School of Chemistry fax: +61 6 249-0750 Australian National University Canberra ACT 0200 e-mail: danne@rsc.anu.edu.au ***************************************************************** Hi, Your assumption about the fitting of the least squares of perpendicular distances is correct .. this is what is done by linear regression. A side note, linear regression is used in teaching software, but programs are always built around the matrix formulation. The way that you can fit perpendicular distances, as well as do non-linear fits (constants other than linear coeficients) is using optimization methods, such as steepest descent, Newton-Rhapson, etc. This is just like doing geometry optimization, only you are minimizing on your perpendicular distances instead of energy. The other possibility is spline methods (cube splines are most commonly used). A spline method is not a least square fit, it is an exact fit. However, if there is noise in the data you fit exactly to the noise. Hope this helps. Dave Young young@slater.cem.msu.edu *************************************************************** ** Dear Tamas, I also am a chemist, a chemometrician in fact, and am interested in chemistry , statistics and computing all mixed togoether. Whenever I have a mathematical problem like yours, I often find an answer in the excellent book "Numerical Recipes in C: The Art of Scientific Computing" by W. H.Press, S.A.Teukolsky, W.T.Vetterling and B.P.Flannery (Cambridge University Press). There are also version for FORTRAN and, I think, BASIC. PCA is also described in various multivariate statistics and chemometrics books, such as "Multivariate Calibration" by H.Martens and T.Naes (Wiley) and "Multivariate pattern recognition in chemometrics" edited by R.G.Brereton (Elsevier). The source code is actually written in Visual Basic, but conversion to Basic, Fortran or C should be quite straightforward PCA: ijX = ikT * kjP where X is an i by j datamatrix, T is an i by k scores matrix and P is a k by j loadings matrix. In your case, the number of components to extract ("kmax" in the source code) will be 2. For this listing, "evals" are the eigenvalues, one for each PC, which I have defined as the sum of squares of the scores. NIPALS calculates the PC's in order of importance (i.e. PC's with the biggest eigenvalues come first). The loadings will be 2 vectors of unit length, which represent the two new PC axis, and the scores matrix gives the coordinates of the points on these new PC axis. Think about whether you wish to mean-centre the data prior to the PCA or not - this will give different results! Maybe you have the answer to your problem anyway, by the time I write this, but best of luck anyway! ****************************************************** Sub NIPALS (X!(), kmax%, mc%, T!(), P!(), evals!()) Dim i%, imax%, j%, jmax%, k%, maxcol% imax = UBound(Xmat, 1) jmax = UBound(Xmat, 2) Dim sm#, ss#, cmaxss#, rmaxss#, smaxss# Dim cmax#(), rmax#(), smax#() ReDim cmax(1 To imax) ReDim rmax(1 To jmax) ReDim smax(1 To imax) ReDim T(1 To imax, 1 To kmax) ReDim P(1 To kmax, 1 To jmax) ReDim evals(1 To kmax) k = 1 ' Mean-centre matrix if required If mc = 1 Then For j = 1 To jmax sm = 0 For i = 1 To imax sm = sm + X(i, j) Next i sm = sm / imax Next j For j = 1 To jmax For i = 1 To imax X(i, j) = X(i, j) - sm Next i Next j End If START1: ' Find column with greatest sum of squares cmaxss = 0# For j = 1 To jmax ss = 0# For i = 1 To imax ss = ss + X(i, j) ^ 2 Next i If ss > cmaxss Then cmaxss = ss maxcol = j End If Next j For i = 1 To imax cmax(i) = X(i, maxcol) Next i START2: ' Calculate row vector and its sum of squares rmaxss = 0# For j = 1 To jmax rmax(j) = 0# For i = 1 To imax rmax(j) = rmax(j) + (cmax(i) * X(i, j)) Next i rmaxss = rmaxss + rmax(j) ^ 2 Next j ' Scale this row vector to unit length, after checking that it is not a ' zero vector If rmaxss = 0# Then MsgBox ("Nipals( ): Exited before all PC's found") Exit Sub End If For j = 1 To jmax rmax(j) = rmax(j) / Sqr(rmaxss) Next j ' Calculate a new estimate of the PC scores smaxss = 0# For i = 1 To imax smax(i) = 0# For j = 1 To jmax smax(i) = smax(i) + (X(i, j) * rmax(j)) Next j smaxss = smaxss + smax(i) ^ 2 Next i ' Test to see if PC has converged. If so, store scores and loadings, ' adjust matrix, and look for next PC. If not, use last scores estimate ' to put back into the algorithm. The convergence criterion can be ' adjusted (i.e. 1e-6 used for single precision). If Abs(smaxss - cmaxss) / cmaxss < .000000000001 Then evals(k) = smaxss For i = 1 To imax T(i, k) = smax(i) Next i For j = 1 To jmax P(k, j) = rmax(j) Next j For i = 1 To imax For j = 1 To jmax X(i, j) = X(i, j) - (smax(i) * rmax(j)) Next j Next i If k = kmax Then Exit Sub End If k = k + 1 GoTo START1 Else For i = 1 To imax cmax(i) = smax(i) Next i cmaxss = smaxss GoTo START2 End If End Sub ================ Steve Gurden, Bristol Chemometrics Group, School of Chemistry, University of Bristol, Cantock's Close, BRISTOL BS8 1TS Tel: +44 (0)117 9289000 extn. 4421 e-mail: S.P.Gurden@bris.ac.uk *************************************************************** ** From: g80@chm.uri.edu Hi Dr. Tamas Gunda, I too have been interested in the question of fitting a plane to a set of points(3D). As you correctly pointed out this is a problem in total least squares, not simple least squares. This problem as been approached many times using a variety of algorithims(its' fundamental to crystallography). I have written a simple FORTRAN program based on the method of D. M. Blow, Acta. Cryst.(1960),13,168. One fundamental paper is V. Schomaker, J. Waser, R. E. Marsh and ?. Bergman, Acta Cryst(1959),12,60. There are many other papers on this subject. I hope this helps and good luck. Brian Schmitz Univ. of Rhode Island *************************************************************** ** end summary ***************************************************************************** Tamas E. Gunda, Ph.D. phone: (+36-52) 316666 ext 2479 Research Group of Antibiotics fax : (+36-52) 310936 L. Kossuth University e-mail: tamasgunda@tigris.klte.hu POBox 36 H-4010 Debrecen Hungary ***************************************************************************** From toukie@zui.unizh.ch Tue Aug 29 05:56:04 1995 Received: from rzusuntk.unizh.ch for toukie@zui.unizh.ch by www.ccl.net (8.6.10/950822.1) id FAA23681; Tue, 29 Aug 1995 05:48:43 -0400 From: Received: from rzurs10.unizh.ch by rzusuntk.unizh.ch (4.1/SMI-4.1.9) id AA24350; Tue, 29 Aug 95 11:48:39 +0200 Received: by rzurs10.unizh.ch (AIX 3.2/UCB 5.64/4.03) id AA38420; Tue, 29 Aug 1995 11:48:40 +0200 Message-Id: <9508290948.AA38420@rzurs10.unizh.ch> Subject: Babel in DOS To: chemistry@www.ccl.net Date: Tue, 29 Aug 1995 11:48:39 +0200 (MEST) X-Mailer: ELM [version 2.4 PL24 PGP2] Content-Type: text Content-Length: 413 Dear Colleagues; Some time ago I asked a question pertaining to a difficulty I had in making file format transformations using the DOS version of Babel. I received lots of useful advice. Probably the _most_ useful suggestion was to type babel -m at the DOS prompt and then proceed from there. (The rest of the steps are extremely self-explanatory.) Regards, S. Shapiro ZH From acp37@rs1.rrz.Uni-Koeln.DE Tue Aug 29 09:41:07 1995 Received: from rs1.rrz.Uni-Koeln.DE for acp37@rs1.rrz.Uni-Koeln.DE by www.ccl.net (8.6.10/950822.1) id JAA26296; Tue, 29 Aug 1995 09:34:36 -0400 From: Received: by rs1.rrz.Uni-Koeln.DE id AA81642 (5.67b/IDA-1.5 for CCL ); Tue, 29 Aug 1995 15:33:48 +0200 Date: Tue, 29 Aug 1995 15:33:48 +0200 (MST) To: CCL Subject: Summary for "Calculation on biradical" Message-Id: Mime-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Dear netters, some days ago I posted a question about calculations on a biradical. The original posting and the answers are attached. Thanks to all who responded! Thorsten Koch The original question: Dear netters, At the moment I am doing ab initio calculations on 1,5-Dehydronaphthalin, so it is supposed to be a biradical. It is an isomer to a ordinary closed-shell molecule which I am interested in. So to compare the energies of the molecules, I am interested in the singlet-groundstate of the biradical. But this doesn't seem to be a trivial problem. It was no problem at all to get the wavefunction for the triplet state. So I optimized the geometry for the triplet state and tried to get some singlet wavefunctions for this geometry. Here is what I got: 1. UHF, 6-31G*, mult=3 : E=-382.059168, =2.2470 (2.0346 after annihilation of first spin contaminant) 2. UHF, 6-31G*, mult=1 : E=-381.906022, =0.0000 This result was the same as the result with RHF. Especially the spin desity was zero everywhere. So I thought that this wouldn't be a proper description of a biradical. 3. UHF, 6-31G*, mult=1, GUESS=mix This calculation didn't converge at all. 4. In fact I was able to get another singlet state wavefunction. The only way to get it was from a stability analysis and optimization from the wavefunction obtained with 2. The result was: E=-382.086893, =2.2807 (8.1431 after annihilation of the first spin contaminant) At least the spin density gave two unpaired electrons with different spin at the right carbons. Because dft is known to give lower spin contaminations, I tried it this way: 5. UHF, 6-31G*/B3LYP, mult=3 : E=-384.510256, =2.0136 (1.9996) 6. UHF, 6-31G*/B3LYP, mult=1 : E=-384.484437, =0.000 Again the same result as with RHF, spin density zero everywhere. 7. UHF, 6-31G*/B3LYP, mult=1, GUESS=mix : E=-384.518499, =0.9392 (0.1510 after annihilation) This calculation had no difficulties to converge. ROHF calculations for the singlet states always gave the same wavefunction as for RHF calculations. So there are a couple of questions: a) Why is it so difficult to get a UHF singlet wavefunktion for a normal biradical? b) Can I rely on the energies obtained when the spin contamination is about 1.0 (Point 7.)? This is especially problematic because I want to compare it to the energiy of a closed-shell isomer. Since I got this results with 6-31G*/B3LYP it seems to be the best I can get... Is a geometry optimization sensible in this case? c) What can I say about the wavefunction of point 4.? The spin contamination is about 2.3 before annihilation (not a very small value for a singlet ;-) and about 8.1 AFTER annihilation of the first spin contaminant. I have no idea what this should mean! d) Am I right when I think that the spin density of a system with to unpaired electrons in the sigma frame separated far enough from another should exhibit these two electrons at the carbons they are attached to? e) Having all these problems in mind: Which way should I go to determine the energiy of the singlet ground state? Sorry for this somewhat lengthy question, but I'll summarize the responses, when there is interest :-) ----------------------------------------------------------------------- The answers: From: Steve Gwaltney Your molecule sounds like an open-shell singlet. If it is, you need two determinants to describe the wavefunction. Thus, any Hartree-Fock calculation will fail. What you need is a MCSCF calculation, or some other way to deal with two determinantal wavefunction. Steve From: Christopher J Cramer > a) Why is it so difficult to get a UHF singlet wavefunktion for a normal > biradical? > Because an open-shell singlet is a two-configuration wavefunction. It requires an MCSCF approach to do it properly. The S^2=1 calculations are for so-called 50:50 wavefunctions (half singlet, half triplet). > b) Can I rely on the energies obtained when the spin contamination is > about 1.0 (Point 7.)? This is especially problematic because I want to > compare it to the energiy of a closed-shell isomer. Since I got this > results with 6-31G*/B3LYP it seems to be the best I can get... > Is a geometry optimization sensible in this case? > No, and no. You can adopt the so-called sum method of Ziegler and double-the energy difference between the triplet and the 50:50 wavefunction to estimate the energy of the open-shell singlet. Although a terrible approach at the HF level, comparing UDFT to RDFT seems to be OK for S-T gaps (we have several papers on this). However, your B3LYP functional includes HF exchange, which will degrade this comparison. > c) What can I say about the wavefunction of point 4.? The spin > contamination is about 2.3 before annihilation (not a very small value > for a singlet ;-) and about 8.1 AFTER annihilation of the first spin > contaminant. I have no idea what this should mean! > It's garbage. > d) Am I right when I think that the spin density of a system with to > unpaired electrons in the sigma frame separated far enough from > another should exhibit these two electrons at the carbons they are > attached to? > For the closed-shell singlet, there is probably hybridization and orbital splitting. For the open-shell and the triplet, your intuition is right. Much of the literature on 1,4-didehydrobenzene discusses these topics. > e) Having all these problems in mind: Which way should I go to determine > the energiy of the singlet ground state? > MCSCF is the best approach. Your DFT results are quite curious because they indicate the open-shell singlet to be lower in energy than the triplet. This is pretty unusual, and I'm not sure I believe it. If you would be interested in seeing some of our work on these issues, I'd be happy to send you some papers. Best regards, Chris -- From: Christian Koelle > > e) Having all these problems in mind: Which way should I go to determine > the energiy of the singlet ground state? > Zwei Wege: einer gut der andere besser: 1) CI Rechnung. Bei einem Diradikal sollte als aktiver Raum HOMO und LUMO ausreichen. Wir benutzen bei unseren Rechnungen mit einer semiempirischen Methode die HOMO-LUMO Einfach- und Doppelanregung, wobei die Doppelanregung fuer den Singulett entscheidend ist. 2) MCSCF-Rechnung. Auch hier HOMO und LUMO als aktiver Raum. Hoffe geholfen zu haben. Gruss Christian Koelle From: Hans Ueli Suter Lieber Herr Koch, ich wuerde Ihr Problem unter einem etwas anderen Gesichtspunkt sehen: Was Sie effektiv rechnen sind die angeregten Zustaende von ... Bei einem Programm wie Gaussian, das keine direkte Symmetriebindung der Orbitale kennt ist es natuerlich schwierig diese in der noetigen Weise zu besetzen. Wie dem auch sei, das Keyword wuerde lauten Guess=Alter und dann am ende des Inputs waeren die Numern vom HOMO und vom LUMO einzusetzen, dann wuerde er eben das HOMO einfach besetzen und das LUMO einfach, womit Sie das hoffentlich richtige Biradikal haetten. Ich war nie in der Lage das auch fuer Dichtefunktionale durchzufuehren, aber UHF sollte gehen. Andererseits koennen Sie versuchen mit CIS die angeregten Zustaende zu rechnen (Das Biradikal wird auch darunter sein). Vom Niveau her ist "CIS" etwa vergleichbar mit HF und nicht mit CI. Die Resultate sind demzufolge mit Vorsicht zu geniessen. Hier wuerde jetzt natuerlich der Hinweis folgen, dass man angeregte Zustaende besser mit ..., aber ich spare mir das, sollte eh klar sein. Ich hoffe, dass Ihnen das hilft. mit freundlichen Gruessen H.U. Suter From: Frank Jensen Thorsten, UHF wave functions that are spin contaminated to the degree you describe are essentially useless, even your triplet UHF is pretty bad. The energetics will be all garbage. You need a small MCSCF (or equivalently GVB) to get a reasonable zero'th order description. For your singlet biradical a [2,2]-CASSCF may be OK (or an OSS GVB type wave function which is the same). But you won't get any dynamic correlation. To include that you need a MRCI, or perhaps a MP2 on top of the MCSCF. Frank From: Matthew.Harbowy@tjlus.sprint.com Singlet biradicals are no trivial problem. You see, your calculations are on the *lowest* singlet, not the biradical. The calculations are converging to (what I believe is called) a Renner-Teller distorted configuration of the orbitals because it wants to do this ------ U D ------- ------ distorts to UD ------ What you want to do is not UHF, where your spins are all crazy mixed up. What you want to do is CI, where you can specify exactly what sort of configuration you want, and send it on it's merry way. I believe QCISD and CASSCF are the methods du jour for these sort of problems. matt >From jkl@ccl.net Sun Aug 27 10:42 EDT 1995 Received: from krakow.ccl.net for jkl@ccl.net by bedrock.ccl.net (8.6.10/950822.1) id KAA29047; Sun, 27 Aug 1995 10:42:57 -0400 Received: for jkl@ccl.net by krakow.ccl.net (8.6.10/920428.1525) id KAA01963; Sun, 27 Aug 1995 10:42:54 -0400 From: Jan Labanowski Date: Sun, 27 Aug 1995 10:42:54 -0400 Message-Id: <199508271442.KAA01963@krakow.ccl.net> To: chemistry@ccl.net Subject: MOPAC Cc: jkl@ccl.net Content-Type: text Content-Length: 296 Status: R This is a test. Mopac . another dots .. and another one . -- Dr. Jan K. Labanowski, Senior Research/Supercomputer Scientist/Specialist, etc. Ohio Supercomputer Center, 1224 Kinnear Rd, Columbus, OH 43212-1163 ph:(614)-292-9279, FAX:(614)-292-7168, E-mail: jkl@ccl.net JKL@OHSTPY.BITNET >From jkl@ccl.net Sun Aug 27 10:42 EDT 1995 Received: from krakow.ccl.net for jkl@ccl.net by bedrock.ccl.net (8.6.10/950822.1) id KAA29047; Sun, 27 Aug 1995 10:42:57 -0400 Received: for jkl@ccl.net by krakow.ccl.net (8.6.10/920428.1525) id KAA01963; Sun, 27 Aug 1995 10:42:54 -0400 From: Jan Labanowski Date: Sun, 27 Aug 1995 10:42:54 -0400 Message-Id: <199508271442.KAA01963@krakow.ccl.net> To: chemistry@ccl.net Subject: MOPAC Cc: jkl@ccl.net Content-Type: text Content-Length: 296 Status: R This is a test. Mopac . another dots .. and another one . -- Dr. Jan K. Labanowski, Senior Research/Supercomputer Scientist/Specialist, etc. Ohio Supercomputer Center, 1224 Kinnear Rd, Columbus, OH 43212-1163 ph:(614)-292-9279, FAX:(614)-292-7168, E-mail: jkl@ccl.net JKL@OHSTPY.BITNET From brianh@scg.scg.fujitsu.com Tue Aug 29 13:56:11 1995 Received: from mail.barrnet.net for brianh@scg.scg.fujitsu.com by www.ccl.net (8.6.10/950822.1) id NAA02916; Tue, 29 Aug 1995 13:52:02 -0400 Received: from fujitsu1.fujitsu.com (fujitsu1.fujitsu.com [133.164.254.1]) by mail.barrnet.net (8.6.10/MAIL-RELAY-LEN) with SMTP id KAA20845 for ; Tue, 29 Aug 1995 10:51:57 -0700 Received: from scg.scg.fai.com (scg.scg.fujitsu.com) by fujitsu1.fujitsu.com (4.1/SMI-4.1) id AA08691; Tue, 29 Aug 95 10:53:14 PDT Received: by scg.scg.fai.com (4.1/SMI-4.1) id AA11489; Tue, 29 Aug 95 10:52:12 PDT Date: Tue, 29 Aug 95 10:52:12 PDT From: brianh@scg.scg.fujitsu.com (Brian Hammond) Message-Id: <9508291752.AA11489@scg.scg.fai.com> To: chemistry@www.ccl.net Subject: Correction to Symposium Announcement CORRECTED SYMPOSIUM ANNOUNCEMENT AND CALL FOR PAPERS Monte Carlo Methods in Chemistry 211'th American Chemical Society National Meeting New Orleans March 24-29, 1996 Sorry I messed up the dates on the previous announcement. The sessions are on the 24th and 25th. Purpose: This symposium will focus on the use of Monte Carlo in all of computational chemistry. There will be sessions on Monte Carlo in classical dynamics, statistical mechanics, and quantum mechanics, both in theory and applications. Sponsor: The Computers in Chemistry Division of the American Chemical Society. Program: Sunday, March 24, 9 a.m. - 12 p.m. Monte Carlo methods for Quantum Systems Sunday, March 24, 2 p.m. - 5 p.m. Monte Carlo methods in Classical Mechanics Monday, March 25, 9 a.m. - 12 p.m. Monte Carlo methods, general (optimization, etc.) Format: There will be six 30 minute talks in each session for a total of 18 talks. All other contributions will be put into the poster sessions. Oral presentations are on a first-come-first-served basis. Deadline: All abstracts must be submitted to me no later than October 1, 1995. Please send me e-mail a.s.a.p if you wish to give an oral presentation. Those people who have already indicated that they will definitely come just need to send me the ACS abstract form. Those who have contacted me, but did not state definitely that they wish to be on the program, please send e-mail to confirm whether or not you are coming. Abstracts: All abstracts must be submitted on an OFFICIAL ACS abstract form. If you don't have one, contact me or you can get sent to you from the ACS web site, www.acs.org. ======================================================================== Brian L. Hammond _/_/_/_/ _/_/_/ _/_/_/_/_/ Computational Research Div. _/ _/ _/ _/ Fujitsu America, Inc. _/_/_/ _/_/_/_/_/ _/ 3055 Orchard Drive _/ _/ _/ _/ San Jose, CA 95134 _/ _/ _/ _/_/_/_/_/ Tel: (408) 456-7322 Fax: (408) 456-7071 Email: brianh@fai.com From smb@smb.chem.niu.edu Tue Aug 29 14:41:11 1995 Received: from mp.cs.niu.edu for smb@smb.chem.niu.edu by www.ccl.net (8.6.10/950822.1) id OAA04071; Tue, 29 Aug 1995 14:31:11 -0400 Received: from cz2.chem.niu.edu by mp.cs.niu.edu with SMTP id AA19296 (5.67b/IDA-1.5 for <@mp.cs.niu.edu.chem.niu.edu:CHEMISTRY@www.ccl.net>); Tue, 29 Aug 1995 13:31:25 -0500 Received: from smb.chem.niu.edu by cz2.chem.niu.edu via SMTP (920330.SGI/890607.SGI) (for @mp.cs.niu.edu.chem.niu.edu:CHEMISTRY@www.ccl.net) id AA28877; Tue, 29 Aug 95 13:14:48 -0500 Received: by smb.chem.niu.edu (920330.SGI/890607.SGI) (for @cz2.chem.niu.edu:CHEMISTRY@www.ccl.net) id AA27111; Tue, 29 Aug 95 13:14:47 -0500 Date: Tue, 29 Aug 95 13:14:47 -0500 From: smb@smb.chem.niu.edu (Steven Bachrach) Message-Id: <9508291814.AA27111@smb.chem.niu.edu> To: CHEMISTRY@www.ccl.net Subject: Proceeding of ECCC-1 Now available The CDROM version of the Proceedings of the First Electronic Computational Chemistry Chemistry will be published in September 1995. Advance ordering of the CDROM is available for a significant discount. Information on the CDROM, poricing and ordering froms are available on the web at URL http://www.ari.net/chemnet/eccc-order.html or by phone to ARInternet (310)459-7171 Steve Steven Bachrach Department of Chemistry Northern Illinois University DeKalb, Il 60115 Phone: (815)753-6863 smb@smb.chem.niu.edu Fax: (815)753-4802 From jlye@tx.ncsu.edu Tue Aug 29 14:42:39 1995 Received: from sparc4.tx.ncsu.edu for jlye@tx.ncsu.edu by www.ccl.net (8.6.10/950822.1) id OAA04121; Tue, 29 Aug 1995 14:32:13 -0400 Received: from hamby.tx.ncsu.edu by sparc4.tx.ncsu.edu (8.6.9/ES17aug94) id OAA19454; Tue, 29 Aug 1995 14:32:18 -0400 Received: by hamby.tx.ncsu.edu (5.65b/SAM 12-13-90 16:56:22) id AA26970; Tue, 29 Aug 95 14:32:06 -0400 Date: Tue, 29 Aug 95 14:32:06 -0400 Posted-Date: Tue, 29 Aug 95 14:32:06 -0400 Message-Id: <9508291832.AA26970@hamby.tx.ncsu.edu> To: chemistry@www.ccl.net Cc: dhinks@tx.ncsu.edu, jlye@tx.ncsu.edu, Harold_Freeman@ncsu.edu From: jlye@tx.ncsu.edu (Jason Lye) Subject: MOPAC: Comments in Output File. X-Mailer: Cem X11/Mailer Version 9.2mgm (Wed Jan 5 12:38:09 EST 1994) Content-Type: text Content-Length: 2626 Dear Net-surfers, I posted a question to the list a few weeks ago concerning certain comments in MOPAC output files. Many thanks to: James Stewart Victor Rosas Garcia Andreas Goeller Richard Bone Dave Giesen for there replies, summarised below: } 1st Comment: } } `GRADIENT TEST NOT PASSED, BUT FURTHER WORK NOT JUSTIFIED' } `SCF FIELD WAS ACHEIVED' The default geometry criteria in MOPAC are set in such a way that the geometry was sufficiently well optimized for most work. If you really want to force the gradient norm down nearer to zero, you can by using other keywords (for example GNORM=1). Some also suggested using PRECISE in addition to GNORM=0.1. I was told that if the gradient is greater than 2, then this is bad, and I need to specify PRECISE to lower this figure in further calculations. } 2nd Comment: } } `HERBERTS TEST WAS SATISFIED IN BFGS' Herbert worked in the Dewar group, and devised a geometric test which Dr. Stewart retained in MOPAC. From a comment in the AMPAC code, "The estimated distance from the current point point to the minimum is less than TOLERA" Where TOLERA is a constant. Some replies have suggested that Herbert's test is not 100% reliable on it's own, and that the gradient norm should always be checked. BFGS stands for the Broyden-Fletcher-Goldfarb-Shanno method for function minimization. Some suggest that the EF (Eigenvector Following) method be used for geometry optimizations instead of the BFGS method, as the EF method is a lot more reliable. EF is a gradient optimizer routine, and can get rid of the gradient problems if used with the keywords EF LET DDMIN=0.0 If problems still persist, add XYZ to this list of keywords. } Finally, can anyone recomend a good book which covers molecular modeling with } MOPAC, or other semi-empirical packages/methods. Sadly, info on good books was thin on the ground: i) Read the MOPAC 93 manual. ii) Practice, Practice, and more Practice with MOPAC (not a book) iii) "A Handbook of comp. Chem.", T. Clark, Wiley Thanks everyone, Hope that this is helpful, Jason. _______________________________________________________________________________ Jason Lye, | Dye Synthesis Research Group, | College Of Textiles, Box 8301, | Type something funny here. North Carolina State University, | Raleigh, N.C. 27695 - 8301 | | Ph: (919) 515-6615 | jlye@tx.ncsu.edu | _______________________________________________________________________________ From jorge.manrique@beilstein.com Tue Aug 29 16:56:16 1995 Received: from uucp-1.csn.net for jorge.manrique@beilstein.com by www.ccl.net (8.6.10/950822.1) id QAA07188; Tue, 29 Aug 1995 16:49:09 -0400 From: Received: from beilstein.com (uucp@localhost) by uucp-1.csn.net (8.6.12/8.6.12) with UUCP id NAA28106 for chemistry@www.ccl.net; Tue, 29 Aug 1995 13:58:30 -0600 Received: by beilstein.com id 0JJB5008 Tue, 29 Aug 95 13:54:23 Message-ID: <9508291354.0JJB500@beilstein.com> Organization: Beilstein Information Systems X-Mailer: TBBS/TIGER v1.0 Date: Tue, 29 Aug 95 13:54:23 Subject: PRESS RELEASE To: chemistry@www.ccl.net PRESS RELEASE August 28, 1995 For Immediate Release Ref.: Jorge Manrique, Beilstein Information Systems Tel: 415 358-9091 Internet: jmanrique@beilstein.com CrossFireplusReactions(TM) and The Beilstein Commander(TM) Enthusiastically Received by Chemists at ACS-Chicago Chicago, IL --CrossFireplusReactions, a fully integrated reactions database, and the Beilstein Commander, a graphical client that provides a unified interface to the entire Beilstein suite of products, were unveiled last week at the 110th meeting of the American Chemical Society in Chicago. The comments of researchers who had an opportunity to visit the Beilstein booth can be summed up as follows: CrossFireplusReactions running under the Commander delivers the performance of CrossFire, with the comprehensive coverage and quality only Beilstein can provide. The Beilstein Commander is a graphical interface designed to integrate the complete Beilstein suite of products in one coherent unit. All interapplication communication is handled transparently to the user. The Beilstein Commander runs on PC computers under Microsoft Windows 3.x, the newly released Windows 95, and on Macintosh computers under System 7.5. CrossFireplusReactions complements the CrossFire system and delivers a complete in-house information solution. The realization of a vision of chemical information, it breaks down artificially created barriers to information about SUBSTANCES (molecules and properties), REACTIONS and CITATIONS. For the first time, scientists are able to search and display information in the way which is most intuitive to them -- letting the format adapt to their requirements. Here, truly, is information at your fingertips. The Beilstein paradigm of chemical information management involves three mutually inclusive domains: Substances, Reactions and Citations. When you review the results of your search in the context of the substances domain, you get the chemical structure and ALL its properties, detailed in up to 350 data fields. If you are interested in how to synthesize a molecule, or wish to see how a structure (or substructure) participates in reactions, the reactions domain presents to you the chemical reactions, reactions' conditions, and literature references. If, from there, you switch to the citations domain, you see ALL the reactions written in the paper that reported the reaction in which you were originally interested. Hyperlinks present in substances, reactions and citations, allow for even more direct and effortless navigation between the various sources of information. The structure search system developed for CrossFire has been extended to include reaction sub-structure searching: you can define the role of a structure or sub-structure in the reaction, and search using reaction attributes such as reaction centers, bond fate, and atom-atom mapping. An intuitive interface allows easy creation of reaction queries. The unique mapping tool, one of many innovations developed for CrossFireplusReactions, provides the simplest and most elegant way to define chemical reactions. Starting with about ten million reactions, with about 300,000 more to come each year, CrossFireplusReactions is the largest and most comprehensive database of chemical reactions available. Reaction conditions such as solvent, temperature, coreactants, catalysts, yield, etc. are displayed with the results, and are also fully searchable as part of the query. As could be expected from Beilstein, all reactions have complete literature citations, and these, also, are fully searchable. Chemists involved in organic synthesis and reaction planning, find in Beilstein an indispensable tool for daily research. Scientists seeking to optimize reaction schemes, have turned to the wealth of data available in CrossFireplusReactions as the optimal solution. CrossFireplusReactions employs RISC technology, a new structure indexing system, and a revolutionary search engine to bring close to 10 million organic reactions and their associated properties and literature references in-house. The client-server system links RISC data servers with clients running on Windows based IBM PC compatible machines and Macintosh computers. Clemens Jochum, Managing Director of Beilstein Information Systems remarked: "Our customers have been looking for a high quality solution to their organic synthesis needs. To date, there is no other information system that can approximate the capabilities of CrossFireplusReactions in richness of the knowledge base, or in the performance of the search engine." Beilstein Informationssysteme GmbH, headquartered in Frankfurt, Germany, and Beilstein Information Systems, Inc., with offices in Englewood, Colorado, also market the Beilstein Handbook of Organic Chemistry: the world's leading authority on organic chemical information. CrossFire, CrossFireplusReactions, and the Beilstein Commander are trademarks of Beilstein Information Systems. All other trademarks are the property of their holders. From cliang@ginger.curagen.com Tue Aug 29 17:26:12 1995 Received: from guarneri.curagen.com for cliang@ginger.curagen.com by www.ccl.net (8.6.10/950822.1) id RAA07749; Tue, 29 Aug 1995 17:24:48 -0400 Received: from ginger.curagen.com by guarneri.curagen.com via SMTP (931110.SGI/930416.SGI.AUTO) for chemistry@www.ccl.net id AA17250; Tue, 29 Aug 95 17:23:38 -0400 Received: by ginger.curagen.com (940816.SGI.8.6.9/930416.SGI.AUTO) id RAA01268; Tue, 29 Aug 1995 17:23:27 -0400 From: "Charlene Liang" Message-Id: <9508291723.ZM1266@ginger.curagen.com> Date: Tue, 29 Aug 1995 17:23:26 -0400 X-Mailer: Z-Mail (3.2.0 26oct94 MediaMail) To: chemistry@www.ccl.net Subject: MM2 lone pair Cc: cliang@ginger.curagen.com Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Dear all, I would appreciate any comments from those of you who have used MM2(87) from QCEP before. The system involve a disulfur (S-S) bond and a peptide bond (-CA-N-C-CA-). | | H O The question is : are lone pairs necessary for N, or O or S ? Personal e-mail response prefered. Thank you in advance. Charlene -- Charlene Liang | (203) 481-1104 ext. 87 cliang@ginger.curagen.com | (FAX) (203) 481-1106 From eknight@wppost.depaul.edu Tue Aug 29 19:41:13 1995 Received: from wppost.depaul.edu for eknight@wppost.depaul.edu by www.ccl.net (8.6.10/950822.1) id TAA08940; Tue, 29 Aug 1995 19:27:39 -0400 From: Received: from ac-lc-Message_Server by wppost.depaul.edu with WordPerfect_Office; Tue, 29 Aug 1995 18:30:18 -0600 Message-Id: X-Mailer: WordPerfect Office 4.0 Date: Tue, 29 Aug 1995 17:51:07 -0600 To: chemistry@www.ccl.net Subject: Parity checking RAM in PCs Parity errors occur when one of the bits in memory randomly flips (usually attributed to radioactivity from the solder). PCs originally supported parity checking on the IBM principal that bad data is worse than no data (Macs have never supported parity checking). A parity error usually caused the machine to halt the current process but that depends on the operating system. The relatively new (less than 1 year old) Triton chip set from Intel that supports the higher speed Pentium processors does not support parity checking of RAM. Most of the new high-speed PC clones thus are not supporting parity checking. The RAM chips themselves have been designed analogously (the so-called EDO RAM does not have Parity checking) so they only have 8 chips on a SIMM as opposed to 9. The new PCs don't support parity checking because (the argument goes) the memory is so reliable that there is no reason to have parity checking. The only quantification I have of this statement is that a parity error should occur every ten years in a 16 megabytes of RAM. (If you have 32 Mb, it would be every 5 years). I got that number from PC Magazine and make no claims to its accuracy. The bottom line is this: if you run a calculation on one of the new PCs, some number of bits _might_ have been randomly flipped. My question is this: Is it legitimate to report calculations done on such a computer in a scientific journal? Here are a few possible solutions to this dilemma: (1) Don't use the cheap computational power of the new PCs. I don't know what the situation is for other computers but they are more expensive, perhaps this is one reason why. (2) Use error correcting RAM. With this, a parity error is corrected as it happens. Unattractive because such RAM should cost at least 50% more than normal due to the larger number of chips and lower demand. I haven't priced it though or even seen it advertised; it could be much more expensive than that. (3) Repeat all of your calculations twice. Unattractive for obvious reasons (you're better off buying a computer with half the speed and parity checking - if you can find one). (4) Report the type of machine that performed your calculation so that the risk is known to those reading your communication. None of these solutions are especially satisfying. Given the chance of programming errors, input errors and transcription errors, perhaps the last solution is acceptable as the hardware manufactures would lead us to believe. What do you think? eknight@wppost.depaul.edu From jkl@ccl.net Tue Aug 29 19:56:14 1995 Received: from bedrock.ccl.net for jkl@ccl.net by www.ccl.net (8.6.10/950822.1) id TAA09196; Tue, 29 Aug 1995 19:55:20 -0400 Received: from pnl.gov for y_zheng@ccmail.pnl.gov by bedrock.ccl.net (8.6.10/950822.1) id TAA07690; Tue, 29 Aug 1995 19:55:17 -0400 From: Received: from ccmail.pnl.gov by pnl.gov (PMDF V4.3-13 #6012) id <01HUNI5ZST0W90MT39@pnl.gov>; Tue, 29 Aug 1995 16:55:15 -0700 (PDT) Date: Tue, 29 Aug 1995 16:55 -0700 (PDT) Subject: Force field parameters for FAD, NAD, and related To: chemistry@ccl.net Message-id: <01HUNI609T8690MT39@pnl.gov> MIME-version: 1.0 Content-transfer-encoding: 7BIT Hi, I am looking for force field parameters for flavin-adenine dinucleotide (FAD), nicotinamide adenine dinucleotide (NAD), and other related compounds. Any help will be appreciated. Please reply to me directly. Thanks. Yajun From jkl@ccl.net Tue Aug 29 22:11:17 1995 Received: from bedrock.ccl.net for jkl@ccl.net by www.ccl.net (8.6.10/950822.1) id WAA10218; Tue, 29 Aug 1995 22:00:51 -0400 Received: from cbdcom.apgea.army.mil for grfamini@cbdcom.apgea.army.mil by bedrock.ccl.net (8.6.10/950822.1) id WAA08102; Tue, 29 Aug 1995 22:00:48 -0400 Date: Tue, 29 Aug 95 21:56:40 EDT From: George R Famini To: chemistry@ccl.net Subject: COMP Program for New Orleans Organization: International Programs Office Message-ID: <9508292156.aa25932@cbdcom.apgea.army.mil> Well Folks, Here is the preliminay schedule for the Computers in Chemistry Division's program for the New Orleans ACS meeting. Please remember that the deadline for getting abstracts in is the end of October (yes, it is earlier than several other divisions, but that's life). Although there is some leeway to the the deadline, I reserve the right to reject any paper submitted after the deadline. In any case, no paper will be accepted after my final program goes to ACS (notice I do not tell you when that date is...). The symposia organizers have made every attempt to accept as many contributed papers as possible. However, many of the symposia are expected to be popular. Therefore, I suggest you contact the organizer now if you are interested in submitting a paper to see if there is space available. There will always be room in the poster session (we had over 300 attendees to the poster session in Chicago, most stayed for the entire evening), as well as the general oral (although I recommend a poster, it seems to attract a larger audience). The preliminary agenda for New Orleans is: Symposium Sunday Monday Tuesday Wdnesday Thursday Applic in Env Chem P D S.E. MO Theory D D Drug Discovery D D P Object Oriented D Monte Carlo D A Experiment Design P D General Oral D General Poster E Sci Mix E Phys Prop Est P D Frugal Chemist D Databasing D Computers in Chemistry Award A D= All Day P= Afternoon A= Morning E= Evening Session Organizers Needed Despite the overall success of offering symposia, COMP still is in need of additional orgranizers for symposia at Las Vegas, Dallas and Boston (Fall 1997 and beyond). Organizing a symposium can be challenging, but at the same time, rewarding and enlightening. If you are interested in organizing a symposium, or have ideas for new symposia within COMP, do not hesitate to contact the Program Chair (that's me, folks). The electronic poster session was a big hit, with over 1000 "visitors" to the infobahn homepage. We are intending to attempt another electronic poster session again in New Orleans, but with some modifications. Hopefully, ACS will not assign us a room for the electronic posters this time (causing a bit of confusion). Plus, it is our intent to a) list the posters in the program for Sunday, and b) have an Internet link outside of the COMP meeting rooms (maybe even at the poster session...). Anyway, Ton, Henry and Steve deserve lots of thanks for making it work in Chicago. George Famini COMP Program Chair The list of symposia (full title) and organizers for New Orleans is: Program Chair: George R. Famini, U.S. Army Edgewood Research, Development and Engineering Center, SCBRD-ASI , Aberdeen Proving Ground, MD 21010; Voice: (410)671-2552; Fax: (410)671-5373; email: grfamini@apgea.army.mil. Four (4) copies of 150-word abstract (Original on ACS Abstract Form) are due by October 20, 1995 to respective session or symposium chairpersons. Molecular Modeling Applications to Environmental Problems - Dr. James Rabinowitz, USEPA, HERL, MD-68, Research Triangle Park, NC 27711; voice: (919)541-5714; fax: (919)541-0694; email: sar@linus.herl.epa.gov. Semi-Empirical Molecular Orbital Methods: Is There a Future? - Dr. Andrew J. Holder, Department of Chemistry, University of Missouri, Kansas City, MO 64110; voice: (816)235- 2293; email: aholder@vax1.umkc.edu. Computational Chemistry Assisted Drug Discovery - Dr. James Damewood, Zeneca Pharmaceuticals, Department of Medicinal/Structural Chemistry, 1800 Concord Pike, Wilmington, DE 19897; voice: (302)886-5792; email: damewoodjr@zen.com. Application of Object Oriented Programming Methodology to Computing in Chemistry - Dr. Dennis J. Gerson, IBM Consulting Group, 1507 LBJ Freeway MS/160601, Dallas, TX 75234; Voice: (214)280-1425; fax: (214)280-1486: email: gerson@vnet.ibm.com. Dr. Kevin Cross, Chemical Abstracts Service, 2540 Olentangy River Road, P.O. Box 3012, Columbus, OH 43210; voice: (614)447-3813 ext. 3192; fax: (614)447-3813; email kcross@acs.org. Monte Carlo Methods in Chemistry - Dr. Brian L. Hammond, Computational Research Div,Fujitsu America, Inc., 3055 Orchard Drive, San Jose, CA 95134; voice: (408) 456-7322; email: brianh@fai.com. Physical/Chemical Property Prediction - Dr. Lionel Carreira, Department of Chemistry, University of Georgia, Athens, GA 30602; voice : (706) 542-2050 or 2051; fax: (706) 542-9454; email: butch@sunlc2.chem.uga.edu. Frugal Chemists Software - Dr. Charles James, Department of Chemistry, University of North Carolina at Asheville, Ashevilee, NC 28804; voice (704)251-6443; fax: (704)251-6041; email: james@unca.edu. Experimental Design for Chemical Models - Dr. Karen Rappaport, Hoechst-Celanese, 86 Morris Ave, Summitt, NJ 07901; voice: (908)522-7868; fax: (908)522-3913; email: kdr1@sumhcc1.hcc.com. New Methods in Databasing - Dr. Scott Kahn,Molecular Simulations, Inc, 555 Oakmead Parkway, Sunnvale, CA 94086; voice: (408)522-0100; fax: (408)522-0199; email: skahn@msi.com. Electronic Poster Session - Dr. Steven Bachrach, Department of Chemistry, Northern Illinois University, DeKalb, Il 60115 Phone: (815)753-6863, smb@smb.chem.niu.edu, Fax: (815)753-4802 General Computational Chemistry - Poster and/or Oral Sessions - Dr. George R. Famini, US Army Edgewood Research, Development and Engineering Center, SCBRD-ASI, APG, MD 21010; voice: (410)671-2552; Fax: (410)671-5373; email: grfamini@apgea.army.mil. From cletner@remcure.bmb.wright.edu Tue Aug 29 22:26:17 1995 Received: from remcure.bmb.wright.edu for cletner@remcure.bmb.wright.edu by www.ccl.net (8.6.10/950822.1) id WAA10470; Tue, 29 Aug 1995 22:20:07 -0400 Received: by remcure.bmb.wright.edu (931110.SGI/921111.SGI.AUTO) for chemistry@www.ccl.net id AA23409; Wed, 30 Aug 95 01:14:15 -0400 Date: Wed, 30 Aug 1995 00:42:21 -0400 (EDT) From: Charles Letner Subject: Re: CCL:Parity checking RAM in PCs To: eknight@wppost.depaul.edu Cc: chemistry@www.ccl.net In-Reply-To: Message-Id: Mime-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII On Tue, 29 Aug 1995 eknight@wppost.depaul.edu wrote: > The bottom line is this: if you run a calculation on one of the new PCs, > some number of bits _might_ have been randomly flipped. My question is > this: Is it legitimate to report calculations done on such a computer in a > scientific journal? When the whole Pentium FPU bug came out we saw basically the same question. That is, how much faith do you put in you hardware? I think the answer is quite a bit. I suppose that the assumption that I work under is that of all the potential errors that could occur, an error due to hardware failure is minimal. In my mind, I as the user am much more likely to make an error than the computer. Who hasn't had to trash a calculation at one time or another because of user error. I know I have. The reasuring point is that an error usually manifest itself in some way that, with experience, the user can see. The other question I have is who decides what machines we can use. IF a Pentium isn't allowed, what about a low end DEC alpha? Are the only valid calculations those that are run on SGI Challenges or Cray's? I don't really think so. The results are really what matters, not the machine used to run the simulation. To carry it one step further, would an author have to report the manufacture of the RAM in the computer used for a calculation? Lets say that to cut cost a owner of an SGI bought third party memory. Because it is less expensive their is always the POSSIBILITY (however small) that over a five year period that memory wouldn't have a higher failure rate? Do we have to start reporting mean time between failures for memory? What about other hardware, disks, cache, FPU, etc...... I think my point is becoming clear. Some things have to be taken for granted. That hardware failure is negligable compared to other problems is one assumption that I think is valid. For those who would say that detracts from the validity of the results, then judge for yourself the validity of the result. But if you want to through out a simulation just because the machine didn't have parity check, better start prepare to through out alot of science. I mean who's to say that the HEPES used to make up the buffer in the last paper you read didn't have some trace contaminate? ;) Having said that, I do believe that reporting, in a general way, the machine used and the cpu time is important. But this is more so that a reader who is considering a similar calculation can estimate the magnatude of the job they are contemplating on the machine they plan on using. There is my $0.02 worth...... Charles Letner Wright State University Department of Biochemistry Dayton, OH 45435 e-mail: cletner@remcure.bmb.wright.edu From jochen+@pitt.edu Tue Aug 29 22:41:19 1995 Received: from post-ofc02.srv.cis.pitt.edu for jochen+@pitt.edu by www.ccl.net (8.6.10/950822.1) id WAA10604; Tue, 29 Aug 1995 22:29:00 -0400 Received: from unixs2.cis.pitt.edu (jochen@unixs2.cis.pitt.edu [136.142.185.29]) by post-ofc02.srv.cis.pitt.edu with SMTP (8.6.10/cispo-2.0) ID ; Tue, 29 Aug 1995 22:25:15 -0400 Date: Tue, 29 Aug 1995 22:25:16 -0400 (EDT) From: Jochen Kuepper Subject: Re: CCL:Parity checking RAM in PCs To: eknight@wppost.depaul.edu cc: chemistry@www.ccl.net In-Reply-To: Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Hi, I still have parity-checking in my PC, and it works - most times the computers crashs, but see IBM... solution 5) buy Pentiums with parity-checking and they will get cheaper. Jochen -------------------------------------------------------------------- Jochen Kuepper University of Pittsburgh (412) 624-8638 Department of Chemistry (412) 624-8665 603 Chevron Science Center jochen+@pitt.edu Pittsburgh, PA 15260 USA Heinrich-Heine-Universitaet Institut fuer Physikalische Chemie und Elektrochemie I Universitaetsstrasse 26.43.02 40225 Duesseldorf kuepperj@uni-duesseldorf.de Germany