Formulation spaces

In real life (how to be a better cook)

I am a big fan of the the book “Ratio” by Michael Ruhlman. https://www.simonandschuster.com/books/Ratio/Michael-Ruhlman/Ruhlmans-Ratios/

For any home cook, or restaurant chef, this is a way to start thinking about cooking free of recipes.

Or more powerfully being able to conjure “recipes” on the spot that are likely to work and can be customized to your desired use and flavor preferences.

Ruhlman suggests that using a scale and learning ratios is one surefire way to approach this, and I agree.

Beyond formal measurement this becomes intuitively ingrained in certain cooks so that they just know “How much is the right amount of flour” to make a perfect cake with four eggs. Maybe Grandma was an expert at that.

This book teaches you to become a walking formulation space calculator. No computer needed (gram scale useful).

As related to recipes (what do we call this?)

When thinking about “formulation spaces” we are most simply talking about a recipe which in base form is just a list of ingredients and quantities or ratios. Cooking directions and parameters are also important for end users of the recipe. Recipes may also include artistic or historical information that is less relevant to the preparation of the recipe, and more related to the marketing or education about the recipe.

In a vast formulation space such as “all the recipes on the internet” or a large database of popular recipes one will soon realize that there are only “so many” recipes for pot roast (or whatever the dish is).

Most of them share the same core ingredients such as beef, carrot, tomato and onion. Some of them have other ingredients, such as mushrooms, or potatoes, or red wine. Some have all of these ingredients. Some may have a certain cooking method (ex. Dutch Oven), some may have a different cooking method (ex. Instant Pot).

Some may even have novel or shocking ingredients like coffee, but at the end of the day after cooking the variations, most people would say it is still a “pot roast”. *

As we move out to the fringes we may encounter more and more novel ingredients, until at some point it is no longer thought of as a pot roast, it is now something else altogether.

Perhaps there is a classical dish that intersects this point in the formulation space or it may be a novel creation that is described otherwise such as ingredients and cooking methods. Or perhaps you just expand the scope and meaning of the original to encompass the new dish.

What to name novel dishes is a subject separate from this post, however the creativity of cooks and chefs worldwide to increase the menu lexicon through experimentation in the formulation space cannot be underestimated.

The data (what can we compute?)

As we move our thoughts from what sort of delicious foods we can conjure in real life and move into the computational realm, there are a number of reasons one may wish to explore the formulation space.

One reason would be to characterize an ingredient for an application. Suppose a new flour- how does it work for various cookie formulations?

Another reason would be to optimize a formulation for a particular hedonic or functional requirement such as flavor/aroma/mouthfeel, nutritional content, or transportability/durability.

This can be accomplished with a Design of Experiment (DoE) or one factor at a time optimization.

A well thought out DoE can help characterize acceptable variance in formulation and process inputs because the DoE explores a certain part of the formulation space.

While one candidate formulation may not be “the best” and may not be the final formulation, it may also still meet the target product profile.

The formulation space can also be useful for classification and generative recipe applications. For instance prevention of recipes that don’t work (“Not a baked good” in figure below) can be reduced by analysis of the formulation space.

There are a number of opportunities for improvement of classical formulation models using ontological mapping of ingredient functions.

Reduction in dimensionality in the formulation space is useful for a number of reasons, including the fact that an infinite or very large number of experiments is rarely desired for physical optimization, and for computational methods, sparse data is rarely useful without reduction in dimensionality by some method.

The figure below shows a simple cookie formulation space.

* In my experience there is always going to be someone who will argue where this line of this dish is “no longer is a” because (either) it “doesn’t have ….” or “it has … “
This is an impossible argument to win in real life for some of the following reasons:
1. Because everyone’s taste is different, so they may simply just like mushrooms or lots of garlic in pot roast.
2. Because that most difference in taste in food is tied to their unique life experience, childhood, travels, geography and unique personhood. Food involves all senses and is something you ingest for sustenance and gustatory pleasure.
3. The argument is really about the meaning and use of the word. There is always a lot of gray area.
4. Meaning is seeped in and sometimes lost in translations and common names generations old.

Somebody somewhere could be “offended” that I use( in addition to salt) raw sugar, allspice and coriander seed and other spices to cure “duck confit”.

Surely my result is not properly salty as heck and the spices are surely wrong…! I must not know how to “properly make duck confit”.

What I did was definitely not how it was done centuries ago on some European hillside.

Why so salty back then? For good reason! Because they didn’t have refrigerators and the main reason to make Duck Confit was to preserve the duck meat throughout the winter.

As a chef, cooking fundamentally delicious food is a good immunization against this sort of criticism of what you call something.

Best duck: https://libertyducks.com