My Math Forum calculating coefficients to fit number series

 Applied Math Applied Math Forum

 April 4th, 2012, 07:19 AM #1 Newbie   Joined: Apr 2012 Posts: 2 Thanks: 0 calculating coefficients to fit number series I have a series of experimental data, let's call it d: d[0], d[1], ..., d[n] I know, according to the theory behind the experiment, that the d series is (approximately, because of noise in the experiment) described by a function of the form: d(i) = p * i * log (i) + q * i + r where p, q and r are constant real numbers and log(i) means the logarithm of i in some base. The base obviously doesn't matter since we have the p factor to accomodate it (i.e. logarithms of n for different bases only differ by a fixed ratio for any n). So we can assume that log (i) is the natural logarithm of i. My question is: what kind of algorithm or numerical analysis tool would allow me to find the factors p and q and r so that the theoretical function best fits the data (e.g. according to the sum of the squares of the differences or some other "distance" metric).
 April 4th, 2012, 09:25 AM #2 Global Moderator     Joined: Nov 2006 From: UTC -5 Posts: 16,046 Thanks: 938 Math Focus: Number theory, computational mathematics, combinatorics, FOM, symbolic logic, TCS, algorithms Re: calculating coefficients to fit number series A least-squares fit is pretty standard here.
 April 4th, 2012, 10:27 AM #3 Newbie   Joined: Apr 2012 Posts: 2 Thanks: 0 Re: calculating coefficients to fit number series Yeah but the least squares is the metric we are trying to minimize. The question is: how do we select successive triplets of p, q, r ? Is it just brute-force or is there a better algorithm in this case? Even under a brute-force approach we have a 3-d space of possible p,q,r values so there must be some theory to allow us to at least bound the space of plausible values or determine what's the right level of granularity at different segments of the solution space or along different dimensions. To give a concrete example intuition would dictate that the granularity one uses when trying different values on the p-axis is a lot more finer than the granularity for the r-axis. I mean do we start with a guess triplet p0,q0,r0 (using intuition) and somehow wander off from that starting point based on the results we get on each point along the way? Do we "spray"the 3-d space uniformly with triplets we wish to check and choose the best one? What is the current state of theory in numerical algorithms on how one approaches that kind of problems? I know about particle swarm algorithms and all that but I was hoping that for closed formulas one can use mathematical analysis (I was thinking derivatives) and not some kind of randomized brute force walk in the solution space.
 April 4th, 2012, 11:42 AM #4 Global Moderator     Joined: Nov 2006 From: UTC -5 Posts: 16,046 Thanks: 938 Math Focus: Number theory, computational mathematics, combinatorics, FOM, symbolic logic, TCS, algorithms Re: calculating coefficients to fit number series There are well-known and widely-implemented algorithms for solving these sorts of problems. You can even do it in a spreadsheet if you like. R is a good tool (or S or SAS or the like), OOo Calc can do it, Mathematica and Matlab can do it. PARI/GP can do it. I imagine Maxima and others can do it as well.

 Tags calculating, coefficients, fit, number, series

 Thread Tools Display Modes Linear Mode

 Similar Threads Thread Thread Starter Forum Replies Last Post hasti Real Analysis 1 February 28th, 2013 04:40 AM xact Algebra 0 April 17th, 2012 08:53 AM icemanfan Linear Algebra 1 February 28th, 2012 03:00 PM bakerconspiracy Advanced Statistics 0 April 14th, 2011 04:49 PM hasti Applied Math 0 December 31st, 1969 04:00 PM

 Contact - Home - Forums - Cryptocurrency Forum - Top