Ice Cream Series: Part 3
As with pizza, music, politics, deities, and whiskey, opinions on ice cream styles range wide. There are aficionados of rich, French custard-based ice creams, with up to eight egg yolks per quart; of the lean, eggless Philadelphia-style that disappears on your tongue and has to be eaten straight from the machine; of the dense, bright, intense, eggless and creamless gelatos of southern Italy, of the chewy, gooey, almost cake-like ice creams of New England, or the even chewier, gooeyer versions from Turkey and Japan that blur the lines between dairy and taffy …
We’re not going to get into the details of every style here, but my hope is to highlight the various textural qualities that define any ice cream, so you’ll be able to see all these versions as existing on a continuum. From that point, with a little ice cream science and technique and experimentation, you should be able to replicate—or invent—just about anything.
Basic Qualities
Richness. Quantified by fat percentage. What’s the ratio of cream to milk? What’s the percentage of egg yolks? Other fats from chocolate, nuts, cheese, etc.?
Density. Quantified by “overrun,” which is simply the percentage the volume has increased by the air whipped into it. 100% overrun means the volume was doubled. 10% overrun means it was increased by 10%. These numbers represent the practical upper and lower limits.
Hardness. When you take the ice cream out of a normal freezer (0°F / -18°C), is it hard as a rock? Is it scoopable? Is it barely cohesive? How about at standard serving temperature (6 to 10° F / -14 to -12°C)? Total solids are important here, especially sugars and the particular sugar blend.
Melt. At a given temperature, does it melt quickly or slowly? So quickly you don’t have time to serve it? So slowly it freaks you out? Dissolved solids and stabilizers and emulsifiers play the biggest roles here.
Body. This refers to the mouthfeel of the ice cream in its frozen state. There can be more or less, but we qualify this rather than quantify it. The body can be firm or yielding, creamy or elastic, long- or short-lasting. These are factors of richness and density, but also of dissolved solids (especially milk solids—more solids=more body) and stabilizers.
Texture. This refers broadly to the mouthfeel of the ice cream’s surface, from its solid state to its melted state. Is it smooth? Icy? Granular? Does it transition from solid to liquid too quickly? Too slowly? In its melted state, does it feel creamy? Milky? Custardy? Sticky?
Finish. After you’ve swallowed it, what’s left? Is it gone without a trace? Is there a lingering creaminess in your mouth, still releasing flavors? Or an oily or pasty film that you want to get rid of?
Flavor. We’re dealing just with the base recipe here, so will be talking about flavor only in a general sense. Is it intense? Muted? Is it multidimensional or flat? Does it hit you immediately or build slowly? Does it linger or vanish? Does it develop over time, revealing new flavors, all the way to the finish? Is it clean and natural? Are there off-flavors, or anything that seems foreign to the flavor ingredients and the dairy ingredients?
These factors are of course influenced by the flavor ingredients themselves. They’re also influenced by the richness and density of the overall formula, and by the stabilizing ingredients. Richer ice cream mutes but extends the release of flavors, especially volatile ones and water soluble ones. Denser ice creams intensify flavors. Some stabilizers, like eggs, mute flavors. Others, like modern gums, are more transparent.
The Master Template
Milk Fat1 5% – 18% (12 – 15% typical)
Total Fat2 (including eggs, cocoa butter, etc.) 10% – 30% (12 – 20% typical)
Nonfat Milk Solids3 7% – 15% (10–12% typical—higher for low-fat ice creams)
Total Nonfat solids4 15 – 32% (22 – 25% typical)
Total solids (total fat plus nonfat solids): 35 – 45% (37 – 42% typical)
Water: 55%–65% (58–63% typical) (generally water = everything that’s not total solids)
Water: 55%–65% (58–63% typical) (generally water = everything that’s not total solids)
Sugars5 (not including milk sugars) 11% – 14% (15% –17% is more typical)
Sucrose 20 – 80% (60 – 65% typical)
Dextrose 0 – 50% (25% typical)
Invert Syrup 10% – 35% (10 – 15% typical)
Egg Yolks: 0 – 12% ( 0 –8% typical)
Gums 0 – 0.3% (0.15% – 0.2% typical)
Salt 0 – 0.2%
1U.S. whole milk is 3.6% fat; heavy cream is 36% fat. Typically
2Egg yolks are about 25% fat, or 4.5g fat per large yolk
3Milk is about 8.8% solids. Cream is about 5.6% solids. Nonfat dry milk is 100% solids.
4Be sure to include solids from flavor ingredients, and from yoks—about 48%, or 8.5g
per yolk
5Most ice cream is TOO SWEET. Every commercial ice cream, every ice cream shop ice cream. They’re formulated for children and sugar addicts, and make it impossible to taste the dairy or any subtleties of the flavors. This blog series will illustrate how to correct excessive sweetness without sacrificing texture.
So yes, I’m sorry—there’s math. If I were a better, kinder person, I’d have built you a spreadsheet.
[Note: by now it should be aparent why we always work with weight measures, and with the metric system. Even when measuring liquids. I do not ever want to hear the world “teaspoon” or “fluid ounce.” When every ingredient is measured in grams, the relationships become clear, and it’s trivial (mostly) to increase or decrease anything by a percentage.
It gets a bit tricky with eggs, which come pre-packaged. Whites keep nicely in the freezer, but not yolks. So I generally design recipes with a discreet numbers of yolks. The yolk of a “large” sized egg weighs about 18 grams. So I’ll make sure the recipes use 18, 36, 54 grams, etc..]
 |
| Dr. Traci Mann, at University of Minnesota’s Health and Eating Lab (Her usual research is not maraschino cherry-centric) |
How to use this Information
You’ll see that a lot of these values are interdependent. If you reduce the fat, you’re also reducing the total solids. So you’ll have to compensate by increasing the nonfat solids.
The least flexible value is the Total Solids. I’m sure there are some good recipes with total solids outside the range I’ve given, but you’ll be safest if you stick with this for now. The total solids value effects the body, the hardness, the melt, and the smoothness.
Think of total solids as everything that is not water. We need water in ice cream, but only the right amount. Too much water = too much ice.
Changing some ingredients will change both the fats and the solids—like eggs, chocolate, cocoa, nut butters.
Ingredients that add water change the solids indirectly, reducing their total percentage. The less water added, the better. But if you add any, you have to compensate. Ingredients like fruits add both water and their own solids—which include a blend of sugars that needs to be compensated for. Fruit flavors are among the trickiest.
Booze flavors introduce alcohol, which has stronger freezing point suppression than any other ingredient. So make the ice cream hard enough, we use a sugar blend that’s nearly all sucrose—and as little of it as possble. We also may reduce the nonfat milk solids, and compensate with some added stabilizers.
Chocolate—the worst most interesting of all—adds sugars, solids, and lots of fat in a form that freezes rock-hard even at room temperature. It can take some rather extreme tweaking of the sugar blend (and everything else) to get intense chocolate flavor and good texture.
These are just a few examples.
Appendix
Whole Milk Composition:
- 87.3% water (range of 85.5% – 88.7%)
- 3.6 % milkfat (range of 2.4% – 5.5%)
- 8.8% solids-not-fat (range of 7.9 – 10.0%):
- protein 3.25% (75% of this is casein)
- lactose 4.6%
- minerals 0.65% – Ca, P, citrate, Mg, K, Na, Zn, Cl, Fe, Cu, sulfate, bicarbonate, many others
- acids 0.18% – citrate, formate, acetate, lactate, oxalate
- enzymes – peroxidase, catalase, phosphatase, lipase
- gases – oxygen, nitrogen
- vitamins – A, C, D, thiamine, riboflavin, others
Heavy Cream Composition:
- 58% water (range of 45.5% – 88.7%)
- 36 % milkfat (range of 25% – 68%)
- 5.6% solids-not-fat (range of 4.5% – 6.8%):
- protein 1.69 – 2.54%
- lactose 4.6%
- ash 0.37% – 0.56%
Egg Yolk Composition:
- 50% water (9g per 18g yolk)
- 23% fat (4g)
- 27% solids-not-fat (5g)
- protein 16%
- 8% Lecithin (1.44g)
- cholesterol 1%
- carbohydrates 1%
- minerals and trace elements 1g
In the next post, we’ll look at this template in action, by taking a typical simple recipe and creating a couple of variations.
Permalink
Permalink
Corn syrup and other glucose syrups are all over the place. I'd generally estimate about 70% solids
Permalink
And if so, what percentage of them is solids?
Permalink
Permalink
Permalink
Permalink
I`m currently working on a web-tool that aims to provide a more comfortable UX than the different spreadsheets that are circulating the web. FDP approximation is done accordingly to the well-known Goff paper (https://www.uoguelph.ca/foodscience/sites/default/files/FreezingCurveCalculation.pdf) that seems to be the most common starting point for this task and is not overly complicated to implement. The additional hardening of cocoa or nut pastes is in fact another issue and it seems there is no common approach on this that is generally aggreed on throughout the literature on this topic. E.g. Angelo Corvitto recommends to account negative PAC values as counter measurement, while other gelatieri just seem to ignore this effects in their calculations. So if anyone has some advice on this it would be quite helpful for me.
The current beta status of the tool can be found at:
https://joernmueller.github.io/Ice-Ed/IceEd.html
I would be very grateful for any comments or feedback.
Permalink
On freezing curve calculation: that equation from the Goff paper will be of limited use. I tried it out in some of my early forays, and found it inaccurate. When I corresponded with Dr. Goff, he affirmed that it's accurate up to 50–55% ice fraction, and then starts to overestimate. The problem is that we mostly care about ice fraction in the 70–75% range.
There are two approaches that work well in my experience. One is to plot the known sucrose calculation / ice fraction data (from Experimental Data from Pickering (1891) and Leighton (1927), adjusted by Smith and Bradley (1983), and further augmented by Goff & Hartel (2013)). You can hunt this down pretty easily. the catch is that it only goes to 180% sucrose concentration, which again isn't high enough. I extrapolated data by continuing the line to 250%—probably not dead accurate, but it gives useful data. Use software like Excel to trace this curve with a 4th order polynomial equation.
You then have to use this equation in regression to find the temperature at each concentration. This result needs to be summed with the freezing point depression of salts, alcohol, and other small molecules. These can be calculated with the standard linear FPD equation that Goff and everyone else uses.
The second approach is a shortcut that I found. There's a standard freezing curve equation for generic foods called the Miles (1974) equation. This requires a starting FPD, for which I use the folowing polynomial equation, fitted to the freezing curve by Dr. Michael Mullan:
[S=g equivalent sucrose / 100g water]
-0.0000000075*S^4+0.0000016739*S^3+0.0000461421*S^2
+0.0571231727*S+0.0235267088
(R² = 0.9999)
The Miles equation requires a "bound protein" number. I use milk and egg proteins times 0.3. Finally, I found it needed to be adjusted with a correction factor of 0.957. This gives a quick formula that lets you plug in a temperature and get an ice fraction. I find it pretty close to the regression method between -10 and -16°C, 10% and 14% MSNF.
TLDR: this gets complicated as fuck, and all you can really hope for is an estimate that accurate enough and flexible enough to be useful. If you create a formula and can consistently get ice cream that's nicely scoopable at the right temperature, you've done well. Don't worry about absolute accuracy, or if your numbers exactly match mine or Dr. Goff's.
Hardening fats: there's even less science here! Negative PAC is an inelegant solution, but I found Corvitto's actual numbers useful. I leave the PAC values alone, but create a hardening fats figure that's based on similar thinking. Pretending it's not a factor (as you've noticed most people doing) is a terrible solution. You'll never be able to calculate a chocolate or hazelnut ice cream without accounting for the saturated fats. My methods are a more complicated but somewhat more flexible and elegant version of what Corvitto does ... the general approach we both use is probably not accurate but has proven useful, at least within normal concentrations and temperature ranges.
The basic idea is to take that negative PAC information and build a separate value. Create a an effective hardness by adding this value to the ice fraction
Questions remain: how different are various saturated fats? What does the temperature / hardness curve for these fats look like? Are they similar to each other (as we assume)? Are they similar to the water freezing curve (as we assume)? Has anyone done real research here that we can poach?Let me know if this is helpful: [email protected]
Permalink
My current implementation pretty much follows the approach you described in the first part.
As an approximation for the Pickering et al. table I extracted the 2nd-order polynomial
FDP(SE) = 0.7592693269656435 * SE² + 6.5366647596977 * SE - 0.1876732904734074
out of the data and found it being sufficient. Be aware that the input value is the normalized sucrose equivalent, so it needs to be in the interval [0, 1.8]. The average absolute error is 0.1 °C which should be acceptable I think. I now did the calculation again also for 4th order and I´m getting slightly different coefficents than Mullan:
[a=-7.413037763056872*10^-9, b=1.605707495080694*10^-6, c=5.894876813534917*10^-5, d=0.05623763887800826, e=0.03774741866354958]
probably because he used some different input data. Maybe I´m going to do a comparision of the three equations and coming back with the results at some time.
The values for salt, alcohol, etc. are just converted to SE and then included straight forward in the calculation, with the one exception of the salt content of the MSNF. Here I´m following the Goff paper and account this as a separate factor it´s own calculation, even if I still wonder why this is required.
The Miles equation is something I have not heard about yet. It sounds very interesting and could be a neat addition to the current state of the tool. Do you have any pointers to the original paper?
Also I would be very curious to learn more on your approach on determining the additional hardening of the saturated fats and it´s exact differences to Corvittos method. So in case you are willing to share your findings I could imaginge this would make up a great topic for a blog post.
I agree with your assumption that the hardening of fats probably is not a simple linear dependency but affected by some parameters, like the type of fat or the amount of emulsifiers in the mixture. And it seems odd in fact that there seems to be so little verified knowledge on this topic. Goff just states:
"Fat, proteins, large molecular weight carbohydrates such as the starch fragments found in CSS, stabilizers, and emulsifiers do not contribute to freezing point depression because fat is immiscible with the aqueous phase, and proteins and polysaccharides are very large molecules. However, as these substances are increased in concentration, there is less water in which solutes can dissolve, so the presence of these materials will result indirectly in depression of the freezing point." So one could think it would be sufficient to maintain the right ratio of total liquid and solid phase and no additional accounting of the saturated fats would be required.
But beside this considerations I also see the need to keep the required input managable. Having to deal with PAC and POD values is from my observation already a bit of a hurdle for some folks. So e.g. observing that relative sweetness is also not linear but dependent on the concentration of the solution and thus replacing POD with some coefficients would make a software tool completely inaccessible for the vast majority of users. So I think the goal has to be to find an approximation model that is good enough to predict most real world scenarios and simple enough to be used without the need to get a diploma in applied physics first.
Permalink
"So one could think it would be sufficient to maintain the right ratio of total liquid and solid phase and no additional accounting of the saturated fats would be required." It's a nice thought, but it just doesn't work that way. Ingredients like cocoa butter get rock-hard in the freezer. It's better to compensate for this, even with an imperfect model, than to try to wish it away.
I agree with the need to keep things simple. But there are different simplicity needs for the end user and for the propellerhead who's building the model. In my case it's the same person! If you're making a product, then you have some UX work to do—you want the customer to be able to use it intuitively.
For the person building the model, the limit is just how much complexity you can handle. I'm actually terrible at math. I studied humanities in college, and most people know me as a writer and an artist. I learned to build models like this just by beating my head against them—and by studying lots of scientific papers, and by asking lots of smart people for help. This has its limits, but I've so far got a better model than any others I've found.
An old adage is that all models are wrong—but some are useful. Mine crossed over into being useful when it was just a couple of months old. But it's still wrong, and always will be. The ongoing goal is to come closer to nailing a formula on the 1st try. Right now I mostly get it on the 2nd or 3rd.