An alternative look at ink wetness

[InesF]

On 6/24/2022 at 8:30 PM, LizEF said:

Hooray!!  Congratulations!

 

On 6/24/2022 at 9:12 PM, txomsy said:

Great work, my most sincere congratulations.

Thank you so much @LizEF and @txomsy for sharing the excitement!

The almost 8 impact caused some heavy review process - the bar was very high and it was difficult to pass the criteria. I was so happy when it was finally accepted. If we ever make it to a penshow, I will open a bottle of this sparkling grape drink.

Sure, it was my curiosity to start the whole thing, but it was your comments and those of many others (mentioned in the paper) who kept me motivated!

 

Thank you all!

[InesF]

On 6/24/2022 at 8:45 PM, mtcn77 said:

did flushing between inks become significant, did leftover water cause any ink dilution, have you checked for any variance in ink properties between the bottle versus in the pen?

As @LizEF already told.

The way how to make sure is not to define a time and look what you get. It is define a goal (the complete drying) and look how long it needs. In the case of the two (originally three) fountain pens, mostly after 36 but always after 48 hours the water residues had dried.

 

It was exactly the long drying times that caused the whole measurement series to last 11 month (instead of 2 weeks).

[dipper]

"Genius is an infinite capacity for taking pains" .... Thomas Carlyle 1858.

 

https://www.encyclopedia.com/humanities/dictionaries-thesauruses-pictures-and-press-releases/genius-infinite-capacity-taking-pains

[dipper]

Here are some notes that may be of interest if you are studying Ines' paper in depth....

 

I have enjoyed reading through pages 1 to 4. Everything on those pages seems well explained, and within the scope of my understanding. Also clear were pages 9-10 text, and Tables 1 & 2, and Fig 1 to 5.

All was well until reaching the text in section 3.4. In section 3.4 (Principal Component Analysis of the full Data Set) my brain began to spin!

After watching half a dozen YouTube videos I recommend this one to help grasp the general meaning of what a "Principal Component" is.

 

Things feel better now, but I still have some problems in understanding.

 

One aspect that is confusing me is that in my (inadequate & simplistic) thinking, I had been considering the four physical properties (Table 1 identifiers: "surf_tens", "viscosity", "pH" and "cond") as potential "causes" of variations in the "effects" (Table 1 identifiers: the twelve columns "W n mm" to "C K m2"). But now I see that PCA takes a broader approach than my previous ideas. All sixteen columns of Table 1 that have an "identifier" are input into the PCA software as err.. just "data".

So that is stretching my brain somewhat.

Some clarity begins to appear when the PCA outputs show that PC1 alone "accounts" for most if the variances in "the data", though I am still not clear what  I just said actually means! (Very different from the example data in the video linked above, where all of PC1 to PC133 were needed to account for 99% of the data variations.)

I want to know more about PC1. Is that a sensible thought?

The answers may be in Fig7 and Fig8? This is where I need more help.

The scattered numbers on Figs 7&8 are individual inks. OK.

The vectors are the sixteen data fields from table 1. OK. (Are these "Projections" onto PC1 and PC2, or onto PC1 and PC3?)

The numbers on all four edges of the figs are .... ?

 

And the paper (pg7 text) draws conclusions by comparing those vectors. I need serious psychiatric help here!

Can anyone find another good video on that particular aspect of interpreting PCA results?

 

 

[LizEF]

1 hour ago, dipper said:

I am still not clear what  I just said actually means!

"I know that you believe you understand what you think I said, but I am not sure that you realize that what you heard is not what I meant."
-- Robert J. McCloskey

 

1 hour ago, dipper said:

I need serious psychiatric help here!

My reading has been delayed and you're making me worry that the entire paper will fly right over my head - like a paper airplane...

[txomsy]

@dipper, you just described PCA without realizing it.

 

ABOUT PCA (if not interested, just think PCA is like posing and solving a riddle and ignore)

 

When I have to explain PCA, I take a book and ask students which are the longer dimensions. Typically they'll point to a canonical dimension (XYZ), length, width, depth. Then I point out it is not: it is the diagonal that goes from one corner (say front-top-left) to the opposite (say, back-bottom-right). Now, if you take that out, which one would be the next? Another diagonal, obviously, once you realize the trend.

 

The problem is we tend to think of Cartesian coordinates and match them to what we see: we see a parallelepiped (a box-like object) and automatically assign each edge an axis. Which leads us to think those edges (variables) are the meaningful dimensions to explain what we see. We look at viscosity, surface tension, etc.. and think they contain the explanation.

 

And it is all OK from a mathematical point of view: we can transform the data to choose any basis we want (even a non-orthogonal one) and data will still be the same, only the coordinates (the values measured that we use for our interpretation) will change. Don't worry about the math terms. Just keep in mind that any set of axis/variables (that do not point in the same direction) works to identify points (to understand Nature).

 

Now, think of experiments: we start from a coordinate system, viscosity, fluidity, surface tension, pH, whatever... those are the properties we can measure and we think of (as the edges of the book). But soon we realize they share common traits: they are not exactly the same, they measure different things, but they are not fully independent from each other either, as you perfectly described. In math words, they are a non-othogonal basis.

 

An orthogonal basis is when changes in one dimension/variable do not affect measures in the others. But if you have a "thicker" ink, it will move slower and affect both, the measure of viscosity, flow, spread and surface tension because all are measured in a way that depends on something common to them.

 

When you realize this, you start pondering: is it possible that I am looking at the problem in the wrong way? Like in the book example. So you perform PCA.

 

You take your data and measure all variables, and see which "direction" (as a mixture of measures, one per variable) is the one that changes the most. In the book, the composite values of some width, height and depth corresponded to a diagonal. Here (in PCA) it will be a combination of the variables used as well. That's easy. Now comes the interpretation.

 

For PCA that is the most difficult part: it is not PCA itself (PCA is a change of coordinate system). You get the dimension of maximal change, the first "component". This component is what is most important to explain and understand your data and is in turn something that is related to each of the variables you measured according to the values you measured (coordinates or "loadings") of each one. Now you need to imagine something that matches those characteristics and hope you identified the right thing. That thing is the one that hopefully does actually explain your data, not the variables you measured.

 

In the book example, you look at the XYZ values and realize it is the diagonal. Maximal physical variation in a book is along a diagonal. Bingo!

 

In other experiments you may see that plant growth associates with water, wind, ice, sun, wearing a coat, wearing a bikini... (some positively, some negatively) and may conclude that "weather" is the actual "thing" that contributes to growth (not just each one separately).

 

That defines an association. Note that wearing a bikini will not make plants grow, it does not identify causation, but associates with heat which associates with sun, which associates with sunlight which does actually help plants carry out photosynthesis. And still, sun is not enough: you may need a second component, like time of year, for winter and summer sun are not the same. Note that bikini wearing may hint you to that second component (e.g. high sun, no bikini, coat, no growth, high sun, high bikini, no coat, dry plants).

 

You may also plot your data in these "mixed" coordinates. That may help interpretation: a typical example is crime: you may see separate groups of points, when plotting versus PC1 and PC2. Then, you look at PC1 and see that, say, all points on one side correspond to big towns and the points in the other side to small towns, you may conclude, say, that the most relevant is rural vs. urban area; in PC2 you may see that on one side you have assault, murder and rape, and on the other, auto theft, burglary, larceny and robbery. What do you make of that? Maybe that PC2 is associated with violence? And if you combine both, how do points cluster in both coordinates? Are big cities more or less violent than rural areas? Anyway, in that example, whatever we were studying (crime) was (in our interpretation) a mixture of "rurality" and "violence", each of which was, in turn, a mixture of the variables we actually measured.

 

The trick here is to note that at the end we are not talking of the original variables measured (viscosity, etc.., assault, etc...) but of something else that we didn't measure (rurality, violence) and that is associated to a combination of the original variables, and which we need to give a name ourselves.

 

It is like a riddle: what is something that's ~1.5-2m high, has two legs, wings, beak, feathers, does not fly, etc...? What would you guess?

 

Depending on our ingenuity, we may identify the correct association or not. It is ingenuity that defines the quality of our interpretation.

 

PCA only points to the fact that, whatever is actually causing what you seek (e.g. "good ink behavior", for whatever that means) depends on something associated with a mixture (as you clearly saw and said) of variables. Actually identifying what that is and naming it is each one's responsibility. You may trust the author of a study or come up with a better explanation.

 

But it is your (each one's) responsibility and ingenuity.

[InesF]

@dipper thank you for taking the pain to read the text carefully!

Before reading through @txomsy's comment, I started to formulate an explanation.

No, it's superfluent.

@txomsyhas provided to us such a well formulated and didactically elegant explanation! Please allow me to let it stand there until new questions arise.

 

For me, PCA is an opportunity to make correlations (with and without causality) visible that are beyond my comprehension because they are beyond 2 (or 3) dimensions. In looking at the data cloud ... and ... I can see nothing, absolutely nothing.

 

Taking data sets in pairs out (2-dimensional diagrams, like Figures 2, 3, 4 and 5) pairing each possible combination, you may get the full picture in K(16;2) diagrams. No, thanks! I prefer PCA which provides me the values I'm looking for in the form of eigenvectors in one or in two diagrams (PC1 vs PC2 and PC1 vs PC3).

As @txomsy described, my part is reduced (ha, ha) to only interpret these vectors and assign to them either measurement data or other recognisable properties which I can prove or I believe in having causality (I like the clever mentioning of sunshine and bikini !).

 

In this case, I know that the measured parameters of ink properties are independent variables while the measured parameter of ink delivery are dependent variables. PCA is neutral, deals with all variables in the same way. So it is on the operator (me) to draw conclusions. When, as in Figure 7, the arrow of surface tension (surf.tens.) shows, more or less, 180° opposite to all the consumption data, I know very well, that surface tension must cause delivery changes and NOT: delivery causes surface tension.

Sunshine causes bikini wear, NOT: bikini wear causes sunshine! 😎

 

So far for today.

You are welcome to discuss further!

[InesF]

7 hours ago, LizEF said:

My reading has been delayed and you're making me worry that the entire paper will fly right over my head - like a paper airplane...

Don't worry. The best read is during having a lemon lemonade at Sunday afternoon in the shadow of big trees ...

[dipper]

 

8 hours ago, LizEF said:

and you're making me worry that the entire paper will fly right over my head -

I think you will enjoy the paper. The section of the paper about "Principal Component Analysis" (PCA) can be understood without brain-stretching if you are happy to take this approach:

1 ) The data is all very clearly shown in Table 1. A little study of the table headings reveals the numers in the table 1 columns are the things that Ines has been telling us about since the start of this forum topic.

2 ) Ines has loaded all the table 1 data into a powerful computer analysis system (called "R"), and then asked "R" to present the most significant patterns and relationships. Some outputs are in tables, and some are in graphical visualisations.

3 ) @InesF and @txomsy know how to interpret the outputs of "R". Now some readers do not know how to interpret those outputs, but that is OK. Ines does continue, in plain language, to write her interpretations of what "R" is saying.

4 ) So, as readers, doing our own interpretations of things like "scree plots" and "vectors" is not enforced! Ines has done that for us.

 

But then there are (a few?) readers like myself who want to understand more. I keep bumping into PCA in other disparate topics. (None that are as interesting as inks and fountain pens of course.) 

 

1 hour ago, InesF said:

So it is on the operator (me) to draw conclusions. When, as in Figure 7, the arrow of surface tension (surf.tens.) shows, more or less, 180° opposite to all the consumption data, I know very well, that surface tension must cause delivery changes and NOT: delivery causes surface tension

Yes, that is well explained.

(And a note for other readers, "all the consumption data" refers to the thick bunch of arrows in the diagram that have small pieces of text at the head of each arrow. That text is the short-format column "identifiers" (such as "C K km") at the tops of the twelve columns in table 1 that list all the values of how much ink is laid down per kilometre of line length etc.)

 

4 hours ago, txomsy said:

You may also plot your data in these "mixed" coordinates. That may help interpretation: a typical example is crime: you may see separate groups of points, when plotting versus PC1 and PC2. Then, you look at PC1 and see that, say, all points on one side correspond to big towns and the points in the other side to small towns, you may conclude, say, that the most relevant is rural vs. urban area;

Thanks for all your explanations. The quote above is but one extract. The use of PCA in revealing "clustering" is where it crops up most often for me. It was also the most common application of PCA presented in the YouTube videos that I watched. Possibly because that use feels easier to understand.

 

It is neat, on screen, turning a 3D mess of dots to view them perpendicular to a different direction, and then see two clusters separated that were previously overlaid. Extend the idea from 3D to unlimited number of dimensions... nice.

 

But in the ink-wetness study we go beyond "clustering" into something else: proportional relationships between data values.....

4 hours ago, txomsy said:

In other experiments you may see that plant growth associates with water, wind, ice, sun, wearing a coat, wearing a bikini... (some positively, some negatively)

(A Quick Check... by "see" you mean that when looking at the PCA outputs, and especially at vector plots such as figs7&8, you "see" the positive and negative associations in those charts?)

 

4 hours ago, txomsy said:

when plotting versus PC1 and PC2. Then, you look at PC1 and see that, say......

.... and that is where I get lost!

Some of the YouTube videos I watched did show "PC1 vs PC2" diagrams, and then the presenter discussed the lengths of some vectors projected onto just PC1.

I think I will grasp this if told simply what is being plotted, and what the (four!) axes are, and (perhaps) that this is actually two things (clustered data points and component vectors) overlaid, two axes for one purpose and two for another?

In comparison - an old-school 2D graph of "x vs y", that seems obscure can often be decoded by looking at the two axis labellings, units, number values etc. Similar decoding is not so easy in the PCA plots because all the data types have been scaled or normalised before the major number crunching begins, and there are no units - just numbers on the axes.

 

@txomsy you have already gone above and beyond with time spent explaining.

[LizEF]

1 hour ago, dipper said:

I think you will enjoy the paper. The section of the paper about "Principal Component Analysis" (PCA) can be understood without brain-stretching if you are happy to take this approach: ...

Phew!   Glad to hear it.

 

1 hour ago, dipper said:

But then there are (a few?) readers like myself who want to understand more.

Never content, huh!   Good for you!  (Really, none of us should ever be content with our current understanding.)

 

Now, I just need to finish playing with the story ideas in my head, and then I can read.  (Fiction makes me dysfunctional - whether my own that I'm making up, or someone else's that I'm reading - until I'm done, the rest of the universe is either ignored or begrudgingly dealt with.)   (FPN seems to be the exception - must keep up with posts.)

[InesF]

5 hours ago, dipper said:

.... and that is where I get lost!

Don't worry, @dipper, its easy to get lost there.

If an eigenvector is (more or less) exactly parallel to an axis, you may interpret the variable responsible as the principal component the axis is representing. That's the simple part.

It gets complicated if none of the vectors is parallel to any of the significant principal components.

In any case, you need to know your data and should know from most of them if they are dependent or not and if there can be causalities or not. If so, you will find a way to interpret the principal components.

The axis are, you are right, relative distances. They cannot be absolute. The data set is centered and normalised (the range of one parameter from 0.01 to 0.10 should have the same importance as another ranging from 1000 to 10000).

From the scree plot (Figure 6) you can interpret an almost 10x higher weight of PC1 than PC2, for example. And so you identify one principal component after the other.

The machine calculates, the human interprets!

[txomsy]

OK, so, let me (try to) put it down in some clearer way.

 

But first: the scree-plot: that is easy: let's take the plant growth as example. We look at the variables and conclude that (among what we measured, there might be other things we didn't think of or we cannot measure), "weather" is the major contributor, followed by "season". The next question is how much more important is it? That is the scree plot: it tells you how "important" is each  "component" by showing how much each contributes to "explain" (how strongly is it associated) whatever you seek.

 

Let us now continue with plants. PCA only helps identify the strongest associations. Which are a mixture of "variables", of things you measured (like water, and bikini).

 

You can get numbers: for each component, each mix of variables: how much does each variable contribute to the mix. If the number (the contribution) is positive, it means that the more you have of that variable, the more you get of the composite, e.g. the more bikini, and water and the less ice, the best the plant grows.

 

Sometimes that is all you need. In the crime example you may get PC2 having positive association with violent crimes and negative with non-violent ones. The numbers are enough

 

Sometimes it isn't or we are no good at numbers. So we plot. We could plot in one dimension, for each PC, but that is boring and wasty. We use two to save space. Say x is PC1 and y is PC2.

 

Now, the trick is to ignore all axis but one (this is tricky for we are not used): take PC1 and only look at where points fall in the X direction (ignore Y), that will tell you which characteristics go towards higher X (contribute positively) or towards lower X (negatively). Same for PC2, only look at the Y direction and ignore the X to interpret the meaning.

 

Again, sometimes that is enough, and sometimes it isn't. That is why it is often so important to have "metadata", things that are additional information. For instance, say you plot crime and when you look at PC1 only see city or state names. That's not much use. But if you know some states or cities are more urban or more rural, then, that additional information allows you to interpret what those state/city names mean. That's extra information, based on your knowledge of what you have measured. And the plot helps you figure it out (usually by repeating the same plot with different colors for different characteristics and seeing the any characteristic -color- associates with the data).

 

So, plots may help interpreting the data.

 

Now comes the funny part: in principle one may plot variables (viscosity, etc.) and see how do they contribute (arrows) to each PC.

 

But one can also plot the data itself (plants or inks, for instance) and see how do they (plants or inks) spread across the dimensions. This may tell us, for instance, that some plants grow better in the sun or the shadow, flower in winter or summer... depending on that, they will go to one side or the other in the corresponding dimension (X, Y, weather, season,...). That is also very helpful, It will tell you some inks have more or less of whatever is important (however you decide to call it) than other inks.

 

It is fun because when we do, we rarely see all points (all plants, all inks) spread uniformly across the space. Instead, very often, we see them grouped. That means that most inks tend to have surfactants and biocides, but not any quantity, at least enough to work, so not all values are seen. This is, you normally see your "subjects" (e.g. inks) grouped.

 

Now, that has nothing to do with PCA itself. That is "clustering".

 

Mind you, they would also be grouped if you used the original variables. But if we use these compound variables (whatever they mean) found with PCA, since we selected them to be the ones that better associate with our data, it means the group would in turn also be more clearly defined. That is why PCA is often used for clustering: as a first step to identify the "previously unknown" or "hidden" variables that better help separate the groups of data.

 

So, some plants will be associated with bikini, and sun, but (hopefully) it is better to see the association with summer Others will associate with coat and ice, but it is (hopefully) better to associate them with winter. Using "season" instead of the original variables is (hopefully again) more meaningful.

 

[dipper]

1 hour ago, txomsy said:

Now, the trick is to ignore all axis but one .....

Ha....!    I would never have guessed without being told, despite seeing tutors on youtube doing exactly that.

 

Thanks to help from @InesF and @txomsy PCA is now becoming much clearer.

 

You have launched me into doing PCA for myself. Found this online PCA tool.....

https://towardsdatascience.com/a-free-online-tool-for-principal-components-analysis-with-full-graphical-output-c9b3725b4f98

 

As an exercise, thinking I will start with a fictitious physics based data set, where some parameter values have been engineered to follow known linear relationships, plus some simulated experimental error. Then try to use PCA to extract or reveal the relationships that I put into the data at the start.

 

[txomsy]

Incidentally...

[txomsy]

Another post to ignore.

 

With no bearing whatsoever in the published paper, and with doubtful consequences... and since the data is available although too limited for what most people would want...

 

In my most humble opinion, it might be interesting in this case to see the results of a Canonical Correspondence Analysis (CCA) of inks and properties. That might help support or question (note I do not say definitively answer) some other common questions posed in the FPN, such as whether some kinds of ink have these or that property.

 

To understand it, CCA is like (you can think of it as) a bidirectional PCA. Instead of using a number of variables and seeing how they contribute to a desired outcome, you take two sets of variables and see how each set is related to the other set through intermediate, composite groups.

 

In effect, it is as if you first did a PCA of the variables measured (surface tension, viscosity, whatever) and tried to see how each contributes, and then took it as a riddle to see what is it that is actually happening (as in the examples above).

 

The difference is that you next do the same to the other set: you try to see how much the variables in this other set associate with the "virtual / composite" variables in the first set. So, you are not looking at many-to-one or a one-to-one direct relationships (say, between viscosity and Noodler's BSB) but to indirect associations. Say we have explanatory variables (surf. tens, viscosity, etc...) and target individuals (inks), you would be looking at the association of a group of variables to a group of subjects.

 

In the end, the real problem, as with PCA, is in solving the (now two) riddles: what does each combination of variables (viscosity, surf.tension...) mean, and what does each group of subjects (Noodler's blue, MB black,...) mean. Hopefully, for inks, we would be able to see if some families of inks (say colors, or made by a specific maker) share a common trend (expressed as a combination of the variables), e.g. whether some maker's inks are better or worse behaved, wetter or drier,... than others, or whether certain colors share a given set of properties different form other colors, or who knows what.

 

I think it may be better understood with plants: say you want to select a plant for your pot/garden. Which is the best one?

 

You may measure, on one side, soil properties, water demand, light hours, latitude, temperature, fertilizers... and on the other side a large number of plants. Now you match them with CCA. Hopefully you would see that some groups of measures can be labelled as "tropical", "temperate", "boreal", "polar", "desert" environments. And you would see which groups of plants can grow in each environment. Most likely most plants will grow in most places, but you would know, for each place which group of plants grows better or worse, or for each family of plants, how well they grow in each different environment.

 

Now, you match your chosen, personal, local conditions to one of those environments as best as you can, look at the group of plants that grow better in that environment and pick the one you like the most. This may tell you which plants grow best on shadow in (say) California... or if you want a specific plant(say a tropical orchid), which kind of conditions you need to keep at home in order to grow it in (say) Helsinki.

 

There are two practical issues for our ink problem.

 

First, the real utility for general users is limited because to date we only have data on ink wetness, and we'd also like to have data on other properties like ease of cleaning or nib-creep or, whatever. Second, although we already have data on some inks, me, personally, I cannot currently undertake it or contribute: I just got out of a serious peritonitis (I'm recovering well, thanks) and after one month and a half out of work, my work stack has piled up too high (students to mentor and defend their theses, projects delayed, proposals to rewrite, many complex  data analyses to do, etc...), plus, it is summertime and holidays are coming (and I still have to undergo some medical tests). Still, if I can find some spare time, I'll likely PM InesF for the data and give it a try, just for the fun of it, as a mock demo.

 

"Beauty" would be it might end up bringing some objective science into ink purchase decisions. "Ugliness" would be the relative loss of "subjective personal touch" in ink selections.

 

My apolgies for the long post. As romantic painter Goya said "The dream of reason produces monsters" (explained by himself as "Imagination abandoned by reason produces impossible monsters; united with her, she is the mother of the arts and source of their wonders."

[InesF]

3 hours ago, txomsy said:

Another post to ignore.

... not even close!

 

4 hours ago, txomsy said:

In my most humble opinion, it might be interesting in this case to see the results of a Canonical Correspondence Analysis (CCA) of inks and properties. That might help support or question (note I do not say definitively answer) some other common questions posed in the FPN, such as whether some kinds of ink have these or that property.

I agree in general. In this case, the structure finding statistic did provide the answer to the asked question: which ink property/properties is/are responsible for "ink wetness" (=ink amount on paper). Other questions were not asked.

 

4 hours ago, txomsy said:

Hopefully, for inks, we would be able to see if some families of inks (say colors, or made by a specific maker) share a common trend (expressed as a combination of the variables), e.g. whether some maker's inks are better or worse behaved, wetter or drier,... than others, or whether certain colors share a given set of properties different form other colors, or who knows what.

To be honest, I started the preparation for this measurement in December 2020 and asked to myself exactly these questions. Only to discard them, as there is no chance to get information about the exact ink composition. "Blue ink" doesn't mean each blue ink is composed the same way. We know, it can be acidic or alkaline, it can be based on one dye or a mixture of dyes and a blue dye can be hydrophilic, hydrophobic, anionic or cationic or non-ionic, can have a low or a high molecular extinction coefficient (so that even colour intensity will not tell anything). And that's only blue.

 

While I thought about conductivity comparably late in the course of the measurements (better late than never), I identified the others as "summary" parameters and hoped so much I would have covered the most influential factors (=dependent variables).

My hope was based on an educated guess: the ink delivery amount will, most probably, be dependent on physical properties, because except for the paper, all other materials in contact with ink should behave chemically inert. Paper is the other "black hole" which I tried to circumnavigate by using 2(3) different types and keeping them constant over the whole series.

 

If we manage to create our own model inks where we can control each ingredient (I'm only one step before doing that) we will be able to widen and diversify the attempt. Working with commercial inks has this "black box" feeling for strict evaluations 🤔 - but will beware this "subjective personal impression" when in use! 🙂

 

[txomsy]

Well, that's the beauty of association statistics: one can use them as a black box (which is why they are now identified "popularly" as "artificial intelligence" or "machine learning").

 

In this case it means that we need not be restricted by anything. We could collect a set of measures on many commercial inks and apply the method. Hopefully that would uncover relevant trends.

 

So, what would be possible is to define a set of properties that pen aficionados consider relevant or associated to good behavior, define an easy measurement procedure that anybody can do, and then just collect the information (say, ask people reviewing inks to please include those measures in their reviews). Then the problem would be collecting all the data from reviews in a database that can be periodically analyzed.

 

The nice thing, too, is that one may use a variant of Canonical Correlation Analysis for categorical variables, (Canonical Correspondence Analysis, also CCA) so that some "properties" could be measured in a qualitative or very rough approximation (say, "nib creep" measured as "no", "easily cleaned", or "permanent stain", or "feathering with M nib in copy paper" as "yes", "no", or "seethrough with M nib in Moleskine" as "terrible", "none", "partial", or similar other scales).

 

This may as well be done for pens or papers and their properties.

 

Thus, one could define some standards to report uniformly those characteristics and then simply wait and let people produce reviews, collect the data and run the algorithms.

 

But, I do not see a way in which most people would agree on a set of measure scales for relevant ink/pen/paper properties. I suppose that if someone would come up with one set of scales, and the moderators approved it, and it were pinned on the review forums, and most readers valued and adopted it and applauded reviews following the standard scale system, it might theoretically work. It's adoption of the system that I see as the main burden.

 

Thinking of it, it is not that different from what any current review does, only that, by having a set of "FPN reference scales" (kinda like Bo Bo's flex scale), reviews would have a somewhat higher objectivity, reproducibility and would be easier to understand and compare. Subject to reviewer trustworthiness. Oh! And that the comparison could be done in batch automatically and automatically updated.

 

Note that one needs to measure things abstractly, for there is no universal definition of "good" or "bad" for any property, so all should be just measured and users would then evaluate which combinations of properties and in which proportions match what they want.

 

The other thing is that nobody will ever do this unless it is us, users, for a given maker will not be interested in maintaining such a system unless all other makers scored too badly. It is us, users, who have an interest in having an (at least roughly approximate) idea of product qualities.

 

It is something that would benefit everybody, but 1) cannot be done by only one person and 2) nobody would be interested in all properties/products relationships. But 1) each one can measure their own inks and 2) everybody is interested in some properties among the few inks at their reach. So, no need to ask if there is someone willing to do it, but to ask if there is enough common interest in doing (participating/using) it.

[InesF]

17 hours ago, txomsy said:

Thus, one could define some standards to report uniformly those characteristics and then simply wait and let people produce reviews, collect the data and run the algorithms.

That would be great!

And as you write later, it may be a small effort for a single review but a big win for the community.

 

To add about parameter that helped me to learn a lot are surface tension and pH-value, with minimal impact also viscosity.

Surface tension helps me for each new fountain pen - ink combination to find a fitting and well behaving pair at the first attempt. No more surprises of chalky dry creaky nib movement, no more flooding of the paper with feathering like no tomorrow.

pH-value helped me to find inks which fit together for mixing.

The counter-intuitive influence of viscosity helped me to increase intensity and shading of inks (which have already some minor shading) and, sometimes, to tame the nicely coloured but nasty low surface tension inks to a certain degree (at least to make them useful with some pens) by shamelessly adding Gum Arabic directly to the bottle.

 

It would be great if ink producers would provide the essential data as, for example, Colorverse does by providing surface tension and pH at least for their series 1 and 2 inks (I do not own others).

 

Effects like nib creep may be harder to catch. Rare random occurrences should not be mismatched with a reproducible ink property. Atlantic Blue from deAtramentis shows nib creep in one of my pens during the winter month (dry air) but not with other pens and not during summertime (hot and damp air). I would not define this ink as nib creep enabler.

 

Have a nice weekend!

[dipper]

13 hours ago, InesF said:

The counter-intuitive influence of viscosity.......

We can construct a plausible explanation for the observed slight increase in wetness with viscosity.

 

Viscosity opposes shearing movements in the flow of a fluid, by definition. (Shearing = layers sliding over other layers.)

So if we are pushing fluids through pipes under applied pressure, from a pump say, then it is intuitive and true that increased viscosity will give a slower total flow rate.

But when writing a line with a fountain pen we are not forcing the ink to flow out of the pen. (Different from tools such as an air-brush or a cake-icing bag or an ink-jet printer head.)

Instead the ink is pulling itself out of the pen and onto the paper, by its own capillary attraction into the narrow gap between the nib and the paper.

 

But then when the nib slides away, moving over the paper surface, the ink says "Oi there... come back over here nib! I want to stay in that nice narrow crevice under the nib tip." The ink tries to pull itself over the paper surface to get back to the nib.

 

Viscosity opposes both movements. The movement out onto the paper, and the movement along the paper surface to catch up with the moving nib.

 

Sliding out through the nib tines tip gap is relatively easy, especially if we adjust the nib to a slightly wider tip gap.

Sliding back along the rough paper surface is more difficult.

 

Viscosity is part of the reason why so much ink gets left behind on the paper.

 

You can play with these ideas by writing on glass with a fairly wet pen. If moving the pen very slowly then a little blob of ink can be seen moving along "attached" to the nib. Move the pen faster and the blob gets left behind, leaving a wetter line on the glass.

Quickly scribble a wet patch on the glass, then move the pen slowly through the wet patch, and the ink in the wet patch will actually go back up the nib into the pen.

[Sholom]

I've been away from this thread for a month or so and am very impressed by how it and the full paper (which I just read) came out. Fortunately, I know something about PCA and statistical data analysis and am a chemist so the paper was an easy and nice read. Thanks very much to Ines for what really was a lot of careful work.

 

I'm still wondering about what the mechanism of conductivity's effect on ink performance is. That conductivity varied widely between inks is not surprising since many dyes in their "commercial" grade (as opposed to extra high purity inks for inkjet inks or biological and optical applications) are loaded with salts that come from the process of isolating the dyes from the reaction mix that produced them. Around 20 years ago I was involved in a project that wanted to use a dye called Basic Blue 3 and was surprised and mildly horrified to discover that the Basic Blue 3 that one could buy in small quantities from chemical supply houses was only about 30% dye; the rest were the salt(s) that were used to precipitate the dye from its synthesis solution. For multiple reasons this particular dye would not be a great choice for a fountain pen ink, even though it is a gorgeous shade of cyan-blue. For many dyes the salt-free form, if commercially available at all, would be nice to use in inks but costs dearly and would not be economically feasible for use in even a high-end fountain pen ink. I wonder whether conductivity really is a proxy for salt content and dye polarity and hence for the overall polarity (in the broad sense of solvent polarity parameters) of the solution? This also gets me thinking, on a hot humid day without air conditioning, that ambient humidity and its effect on paper should also have effects on ink behavior that will also vary with the properties of different papers (sizing agents, fiber types like cotton vs. wood pulp, etc). I think Ines will understand what I'm asking but please don't feel bad if this is all opaque to my fellow readers: this is geeky chemistry and physics talk and I am just thinking out loud. Thanks again Ines!