Perspectives from our Founder

Article 2: OLI’s History of Modeling of Electrolyte Mixtures



By Marshall Rafal

Founder of OLI Systems, Inc.


The Nature of the Problem

The ultimate model for prediction of complex mixtures of chemicals in polar media, in particular water, is one which would be broadly predictive.  The search for such an ultimate model is at the heart of this essay.  It is impossible for me to tell this story without incorporating broadly the experiences of OLI Systems, Inc. as we have been inexorably drawn into writing this story.  It is a 50-year journey that has taken me a 21 year old “callow youth,” in the words of my Masters Advisor when we reunited for his retirement dinner, to a 78 year old both grateful and in awe of the journey and its ultimate success.


There are five principal thermodynamic properties of principal interest, specifically, partial molal:

  1. Gibbs Free Energy

  2. Enthalpy

  3. Entropy

  4. Heat Capacity

  5. Volume


Each of these properties are comprised of a Standard State term, solely a function of temperature and pressure and an Excess term, a function of temperature, pressure, and composition.  This latter term is normally expressed in terms of an activity coefficient reflecting non-ideality.


The challenge in electrolyte modeling is so formidable because many substances dissociate in polar medium forming ions, both simple and complex.  So, a salt like ferric chloride dissociates into more than 10 ionic substances in water including complexes of Fe and Cl as well as Fe and OH resulting in FeCl3, a salt, being strongly acidic in water.


Ultimately, a broadly predictive model would contain parameters that can only be determined by fitting experimental data on the major subsystems underlying a multicomponent system.  Normally, the experimental data is available for relatively simple systems chemically, binary, ternary and occasionally quaternary systems.  The rest, necessarily must be predictive.

Before the advent of the digital computer one could not hope to predict these principal properties for anything but simple, very dilute systems and yet, a great deal of the underlying work, especially in the area of laboratory data was undertaken starting in the late 1800s.

Before the 1960s

As implied above, laboratories all over the world, including Germany, Russia, France, Japan, the UK and the US, often produced reasonable data applicable to the eventual models deployed on the digital computer.  The data was primarily of physical equilibrium as well as heat effects relative to binary and ternary systems of principal industrial interest.


In parallel, starting in the 1950s, Lev Gurvich and co-workers were compiling data on individual substances, including ions and ion complexes, relative to the Reference State, which was the Standard State at 25 C and 1 Bar.  This compilation along with a later work by the US NBS provided the starting point for predictions of the Standard State properties, which reflected departures from the Reference State.


Finally, in terms of modeling of electrolyte solutions as a function of composition, there was the work of Debye and Huckel, published in 1923, noted that even for very dilute solutions, the reference state properties did not predict the behavior of even simple, very dilute, mixtures of electrolytes.  Rather, an excess term was required, which was described in terms of ionic strength, a property of the very dilute mixture that reflected the concentration of all charged species in the solution.


In the absence of the digital computer, this was about as far as things could go in terms of practical prediction.

The hang-up was simply this.  In terms of the Standard State properties and activity coefficients, there was no way to bring together the wealth of lab data together with models and nonlinear regression capable of fitting coefficients to these models.


The 1970s

In 1973, OLI was two years old, and its software was based upon the user providing a mathematical model in textbook notation and then solving that model in flexible fashion.  The computer timesharing company NCSS, a phenomenon of the 1970s, had taken an interest in the early OLI software.  This eventually, in 1973, led to a call of inquiry from Mac Clarke, a Manager of Technology at Olin Chemical who I consider the Patron Saint of OLI.  Put another way, I have often said that “if Mac Clarke had not existed, I would have had to invent him!”


The way Mac put it was thusly “Marshall, I know that we can write the equations of concern, but I have no idea how to determine the coefficients of the model.”  The process he was trying to model involved chlorine and water … essentially pool chemicals, an important part of Olin’s business.

At this time, I had teamed up with a rather brilliant engineering technologist by the name of Joe Zemaitis who like myself had worked at Esso.  Joe was the thermo guy and I was the computer jock.  We basically had to invent everything from scratch.  The first thing was getting the mathematical model right.  That, mercifully, was straightforward chemical engineering.  It involved writing a proper set of equilibrium and material balance relationships.  The next step was writing a Newton Raphson program to solve the equations.  For the very limited chemistry, H2O, Cl2, and complexes of same, Joe was able to find and regress relevant data to models that had been published by Bromley with extensions by Meissner.  We acknowledged the formidable work by Pitzer but felt the Bromley-Meissner-Zemaitis would result in a more predictive model.  In fact, the Meissner generalized extrapolation curves for activity coefficients led, eventually, to a model that took us from ionic strength of 0.01 (Debye Huckel) to a predictive model up to ionic strength of 6.0 although beyond ionic strength 1.0 things got increasingly ragged.  What is so remarkable for OLI, was that these models were published in 1972 and 1973 just in the nick of time for us to apply them to Mac Clarke’s request.


It is important to take a pause here and recognize Dr. Zemaitis’ brilliance, and most importantly, the imprint that he left on OLI and myself that the best model for electrolytes would be predictive.  This was the lodestar that persistently guided OLI’s future even decades after Joe’s untimely passing in 1983.


At the heart of activity coefficient modeling are interactions primarily between anions and cations but between molecular species and ions as well.  Computer memory was quite limited back then and so when I asked Joe how big an array was needed to store interactions, he said 30 would be more than enough.  Eventually, a decade later when we started to model solutions of incredible breadth of chemistry that array would need to have many thousands of slots for the individual interactions!


By the fall of 1973, the first commercial, generalized electrolyte simulation program ECES (Equilibrium Compositions of Electrolyte Solutions) was delivered to Mac Clarke who liked it so much he raised a contract for us to produce a tower program for simulating electrolytes.  There is an interesting anecdote regarding this history with Mac.  Joe and I were barely surviving when Mac came along.  Then, when Mac came along, he asked me to price development of ECES.  I quoted $100,000.  Mac came back saying he could only afford $30,000.  Knowing that $30K could carry Joe and I for the better part of a year, I told him we could do the job for $30K but we would have to own the resulting software rather than Olin owning the software … a crucial moment in the story of OLI.


Also, in the fall of 1973, Joe arranged to give a paper in New Jersey at an AIChE Meeting in Miniature at, Stevens Institute in Hoboken.  We both attended the talk, which was given by Joe, satisfied in the knowledge that we had accomplished an important first in electrolyte modeling.


During the remainder of the 70s little was added to our fundamental Bromley-Meissner-Zemaitis model.  The aforementioned tower program was successfully developed and delivered to Mac at Olin.  This all proved enough of a success so that DuPont, Sohio, and Phillips Petroleum all came on board and new chemistries were addressed all of which proved out our model.  A solid data regression program, published by Marquardt of DuPont, was deployed and much of what went on at OLI related more to the computer as in deploying the model on different platforms (e.g., IBM, Univac) and in different in-house simulators (e.g., DuPont’s CPES, ICI’s Flowpack, etc.).  The product was a mainframe product and sold as a perpetual license for $95,000 for both ECES ($50,000) and FraChem ($45,000 for the tower program).

The 1980s

The 1980s were a time of transition for OLI as well as modeling of electrolyte systems.  Importantly, in the early 1980s models other than the OLI model were introduced largely through academia (MIT, Technical University of Denmark, etc.).  The MIT model and software became known as Aspen Plus supported by an ultimately very successful commercial enterprise.  One of the seven founders of Aspen Technology was a gifted scientist by the name of Chau-Chyun Chen who produced a mixed solvent electrolyte model (the Lyngby model was also applicable to mixed solvent electrolyte).  Unlike OLI’s search for the ultimate model, which would be predictive, these models were highly interpolative.  That said, for many large chemical companies, that had formidable technical staff, the combination of the Chen model and sound nonlinear regression tools allowed said chemical companies to fit their data of systems of principal interest to the Chen model.   And so, at that moment in time in the 80s, in large chemical companies, some electrolyte systems could be addressed.


In terms of the 1980s at OLI, there was methodical growth in OLI’s client base to more than 10 Fortune 100 companies but, also, there was an event that nearly brought our operation to its knees.  Joe Zemaitis, our brilliant thermodynamicist, was diagnosed, at the age of 41, with pancreatic cancer and passed away in February 1983 at the age of nearly 42.  During the ensuing five years I came to realize that I had to become far more knowledgeable with regard to electrolyte science than I had ever before, when I could lean on Joe.  The first step was to pick up on the half-written Handbook of Aqueous Electrolyte Thermodynamics, a project accorded to OLI by the American Institute of Chemical Engineers.  The book, published in 1984, was highly respected in its field and could not have been written without Joe’s contribution both in the early writing and in leaving a blueprint for its completion.


Other than adding clients, adding chemistry to the growing databank, and improving, incrementally, the user interface, the next five years were a time of consolidation and learning, especially for me.  This inevitably brings me to another hero of OLI.  Noel Scrivner, a superb chemical engineering technologist and visionary, with DuPont, helped not only to guide my education in thermodynamics, but to connect OLI to more and more applications at DuPont, leading to substantial extensions to the databank.  Noel was actually a co-author on the Handbook having come into OLI’s orbit back in 1975 when DuPont purchased the second-ever license to the OLI software.  In 1987, Noel saw an opportunity for DuPont and OLI to back strongly a second Airlie House Conference on electrolytes.  The first in 1979, largely academic, did not get on my radar, but this second one, in 1987, very much did.  Noel headed up the organizing committee and I was on the committee as well.  The master stroke was when Noel realized that electrolyte modeling was of deep interest to certain folks across the academic disciplines of chemistry, chemical engineering, geochemistry, and others as well.  Noel created a conclave of giants … giants if you were deep into this field.  By way of example, Ken Pitzer, from chemistry, and Hal Helgeson, from geochemistry were both in attendance.  As a member of the organizing committee I got to sit between these pillars of electrolytes for the Conference Dinner … learning about the arc of Ken Pitzer’s career, starting back in WWII in DC, then as head of the Atomic Energy Commission in the late 40s and then his seminal work in physical chemistry at UC Berkeley helped me to understand why he was so respected and, I daresay, revered.


What I did not see coming was the first of two seminal steps in creating a generalized predictive model of electrolytes.  The work of Hal Helgeson and co-workers (largely Everett Shock and Dimitri Sverjensky) was known to me just peripherally.  That all changed in a matter of a few days and, I daresay, would not have happened without my involvement in the Airlie House Conference.  Helgeson (the leading geochemist in this field) and co-workers had labored for a decade on a fully predictive equation of state for the prediction of the Standard State properties of electrolytes.  The timing was exquisite as Shock, Sverjensky and other co-workers had just put the finishing touches on a series of four papers, published together in book form, which gave all of the details to turn this work into practical computer code and so, within three years (assimilating this work was formidable), OLI had a universal, predictive model for the standard state properties (most prominently relative to equilibrium constants) requiring only the Reference State properties which were, by now readily available.


OLI prediction of activity coefficients continued, however, continued to rest on the venerable Bromley-Meissner-Zemaitis model with its inherent concentration limitations.

The 90s


The 90s proved to be, perhaps, the most important decade in OLI history and therefore in the predictive model of electrolyte systems.  The decade began with OLI discovering the power of leveraging joint industry support in terms of consortia and closed with OLI discovering a formidable amount of government funding available to a company like ours based on vision and the ability to articulate same.


In terms of the actual “holy grail,” the predictive model for electrolyte simulation, little happened for five years following the introduction of the standard state predictive model of Helgeson and co-workers.  But then, in 1995, the second seminal event occurred, one that by the close of the decade helped to produce the foundational predictive model for not just aqueous electrolytes but mixed solvent electrolytes.


By the approach of the mid-90s a few of OLI’s staunchest customers explained that OLI had within its thermodynamic model, the basis for aiding in the prediction of corrosion.  Ironically, it was easier for me to fund the Corrosion Simulation Program Consortium than to find a person who could actually create such a thing.  My efforts eventually led to my explaining the mission to Professor Richard Riman of Rutgers, already a staunch friend of OLI, who told me, “I have your guy and if you hire him you will thank me every day of your life!”  The “guy” was a young man who had been a post-doc of Ken Pitzer and then an employee with Simulation Sciences, by the name of Dr. Andrzej (Andre) Anderko.  Hiring Andre proved to be the second seminal event in an eventual predictive model for electrolytes.  By the way, I do not speak with Professor Riman daily to thank him, but I have been tempted to do just that.


Reading this, one might be tempted to question just what does corrosion simulation have to do with a predictive model of mixed solvent electrolytes.  The answer is only that the very rare ability to address both problems resided in a single individual, namely, Dr. Anderko.  So, after laboring for several years on the corrosion simulation mission (a highly successful outcome I might add), Andre was able to turn his attention to OLI’s Mixed Solvent Electrolyte Program.  If memory serves it was in the year 1999 that OLI secured funding via a DOE sponsored research project to launch the mission of something as formidable as a predictive software to predict general, mixed solvent electrolyte mixtures.

Post 2000


I should give recognition to three members of Andre’s team, there at the launch of MSE and still major contributors to this very day.  Peiming Wang collaborated with Andre on the original model and led the successful effort in the ensuing decade to implement predictive models for the principal thermophysical properties (e.g., viscosity, thermal conductivity, etc.).  Margaret Lencka was a tireless expert in harvesting, critically judging, and regressing an enormous body of data … in more recent years Margaret and Peiming have mentored young thermodynamics experts at OLI in the OLI methods.  And, Ron Springer who, during the earliest days of implementation of the first MSE model, provided important insights that led to improvement of the initial model.

Other than an important add-on to the MSE model to allow SRK to work with MSE for high temperature, high pressure oil and gas applications, the next two decades have seen a more than 100 work-year effort to populate the databank with coefficients based on regression of underlying binary, ternary and, sometimes, quaternary systems.  The demand for more and more exotic chemistry drove this effort and the databank in 2020 represents what one of OLI clients once referred to as a “dog’s breakfast of chemistry.” 


In a sense, this huge, two-decade long undertaking brought to life the unique predictive model, which was essentially finalized by 2001.  Because OLI was no longer laboring under a concentration limit, a heretofore unimagined array of applications could be realized including:

  • Prediction, via thermodynamics, of corrosion in refinery overheads

  • Wastewater treatment across the CPI

  • Refining of lithium along with battery chemistry

  • CO2 capture and sequestration

  • Refining of rare earth metals

  • Scaling and corrosion in oil wells even in the most hostile, extreme environments

  • Stabilization of nuclear wastes including prediction of chemistry involving radioactive elements and their ionization reactions

  • HF Alkylation


These applications are all in the mainstream of current environmental, safety and production concerns for the United States and, indeed, the world.  As evidence of the worldwide support of US efforts, major funding has come to us, over this period, from consortia of industrial companies, the US government via DOE projects and hubs, governments outside the US (e.g., Norway’s prestigious IFE consortium), and JIPs for which we have been invited to participate.


I would like to close with a highly personal note about the manner in which science evolves.  Dr. Joseph Zemaitis, who passed away in 1983, could not possibly have known just how his voice of advocacy for a universal, predictive model would echo so powerfully in me and the wonderful team that was synthesized in the decades following his passing.  I am so grateful for this journey and for all of us who participated.


Looking to the future

Immediate priorities are to develop scaling kinetics and MSE corrosion models.  OLI is also looking to apply its first principles based predictive tools to operations environments.  Where relevant, OLI will develop hybrid models that leverage insights from operations data and deliver automated solutions for monitoring and analytics.  OLI will also continue to develop its applications expertise in industrial and manufacturing applications with extensions to its database and modeling capabilities.


More information from OLI Systems

Contact OLI at https:/ for more information or to schedule a meeting with an OLI expert to discuss how you can use OLI technologies to address your process modeling challenges.  For more information on OLI’s application specific solutions, please check out the OLI Blogs page