Your Mug Crazed. Blame the Glaze Recipe Before the Thick Coat

A crazed mug has a standard diagnosis in many studios: you put the glaze on too thick. I tested that story against published numbers, and I think it blames the wrong thing first. My claim: crazing is mainly a thermal expansion mismatch between glaze and body. A recipe change that lowers glaze expansion fixes it more often than a thinner coat. One part of my own plan failed in the reading. Swapping a flux for more silica is a weak lever. Swapping a high expansion flux for a low expansion one is the strong lever.

The question

A cone 6 glaze crazes. Which change should a potter try first: a thinner coat, or a recipe change that lowers glaze expansion?

I must be fair to the thickness camp. Digitalfire says a glaze that otherwise fits will craze if it goes on very thick. Its example is a tile with twice the glaze thickness of a mug wall. The tile crazed after thermal shock, because a larger gradient formed between the hot clay and the cooling surface [1]. A finite element study also reports a minimum of residual stress at the body surface. The minimum occurs at one particular glaze thickness for a given firing temperature [2]. So thickness is a real variable.

The crux is the order of cause. Does thickness create the stress, or does it only reveal a stress that the recipe already put there? I think the second. The mechanism says so: crazing is a mismatch of expansion between body and glaze, where the glaze has the higher value [3]. The glaze contracts more than the body on cooling, and it is stretched. Thickness changes how the stress spreads. The mismatch sets the stress.

Data and where it came from

I read these in this session. I have no kiln data of my own.

• Digitalfire pages on crazing, calculated expansion, and an expansion adjustment case [1][3][4][5][9].
• Three Digitalfire cone 6 recipes with unity formulas and calculated expansion: G1214W [6], G1214Z [7] and G1214M [8].
• A search summary of Plesingerova (2003) on crack nets. The PDF did not parse for me, so I cite only the summary [10].

Two caveats apply to the data. First, Digitalfire is a technical reference run by one maker's team, not a peer reviewed journal. Second, the 95% figure on its crazing page ("95% of the time the solution is to adjust the thermal expansion of the glaze") is an experience claim. It is not a survey [3]. I know of no census that counts crazing fixes by type. So "more often" in my thesis rests on mechanism and published cases, not on a count.

The recipes, with oxide ratios first (unity formula, flux oxides sum to 1.0), all cone 6 oxidation as the pages imply:

Glaze	Na2O	K2O	CaO	B2O3	Al2O3	SiO2	Calculated expansion
G1214W transparent	0.23	0.03	0.73	0.33	0.48	4.22	6.62 [6][7]
G1214Z matte	0.10	0.01	0.88	0.19	0.47	2.51	7.10 [7]
G1214M 20x5 transparent	0.13	0.07	0.79	0.21	0.34	3.14	7.15 [8]

The recipes by weight: G1214W is silica 25, EPK 25, wollastonite 10, Frit 3134 25, F-4 feldspar 15 [6]. G1214Z is wollastonite 27, Frit 3124 36, EPK 35, silica 5 [7]. G1214M is 20 each of wollastonite, Frit 3134, EPK, silica and Custer feldspar [8]. The author states the cone 6 fit of the matte. He says 7.1 is low enough not to craze on 10 or more mid-range bodies he uses [7]. That is an author's claim on his own bodies, not a general result.

Method

Digitalfire treats expansion as additive. Each oxide has a factor, and the glaze value is the sum of factor times amount [5]. The West and Gerrow weight percent set includes SiO2 at 0.035 and K2O at 0.331 [4]. Na2O and K2O give huge values, CaO a moderate one, and MgO, Al2O3, SiO2 and B2O3 low ones [5]. I did not recompute the figures in the table, because I do not have all the factors and the raw material analyses. I took the calculated values as published. I did the arithmetic below by hand, without the Lab. Each step is simple division or subtraction that a reader can repeat.

The method has stated limits. The number is "primarily a comparative value, not a laboratory measurement." It predicts the direction of change much better than absolute expansion. It gets less reliable with new oxides, a switch from raw materials to frits, a change of firing temperature, or crystallization [4]. Different published factor sets give substantially different results [4]. So I only compare within one family, and I read the differences, not the absolute values.

Result with numbers and uncertainty

1. Normal recipe edits are as large as the shift that matters. Digitalfire advises that a crazing glaze at 7.5 should move toward 7.0. It also says a fall from 7.0 to 6.5 gives a good probability of a lower fired expansion [5]. A move of 0.5 is about 7% of 7.0. That is the target size. Now compare real recipe edits. In the shivering case, a glaze at 6.90 changed in three ways [9]:

• It went to 7.17 when a sodium source (nepheline syenite) became potash feldspar.
• It went to 7.39 when talc (MgO) was removed and whiting (CaO) raised.
• It went to 7.23 when talc was removed with potash feldspar kept.

The changes are +0.27, +0.49 and +0.33. These are 3.9%, 7.1% and 4.8% of 6.90. Ordinary ingredient swaps move calculated expansion by as much as the 0.5 target, in either direction.

2. Within the family, silica and alumina carry the low expansion. G1214W has a SiO2 to Al2O3 ratio of 4.22 / 0.48 = 8.79. G1214M has 3.14 / 0.34 = 9.24. G1214Z has 2.51 / 0.47 = 5.34. The Digitalfire page for G1214Z lists a Si:Al ratio of 5.78 [7]. I cannot reproduce that from the rounded unity values. Rounding cannot explain a gap that large, so I report my own ratio and flag the difference. The matte has the least silica. Its calculated expansion is 7.10 against 6.62 for the transparent, a gap of 0.48 [7]. It has less alkali (Na2O plus K2O is 0.11, against 0.26 for W) but more CaO (0.88 against 0.73) and far less silica. So alkali alone does not rank these glazes. The whole formula does.

3. Silica alone is a weak lever. Digitalfire says that adding 5 or 10% silica to a melting gloss glaze might help. It then gives a case where a 5% silica rise cut calculated expansion only from 6.0 to 5.9. That did not cure the crazing [1]. The change is 0.1, or about 1.7% of 6.0, against a needed shift of about 0.5. This is where my plan changed. I expected silica for flux to be the headline fix. The published number says it is not.

4. The strong levers are oxide and frit swaps. The same page says to reduce sodium and potassium. It says to replace them with oxides of similar function and lower expansion, such as MgO, CaO or ZnO. It adds that a lower expansion frit can be very effective when it is 30% or more of the recipe [1]. The factors agree: K2O at 0.331 is about 9.5 times SiO2 at 0.035 [4] (0.331 / 0.035 = 9.46). Removing one unit of potash removes far more expansion than adding one unit of silica takes away. That is why a feldspar swap beats a silica top-up.

5. Delayed crazing and the mismatch range. Crazing can appear days or months after the firing. Each sudden stretch weakens the bond of glaze and body. Thick ware cools slower inside, which makes it more vulnerable [3]. In my search summary of Plesingerova, a crack net forms with slow cooling. This happens when the body minus glaze expansion difference lies between about -36 and -65 x 10^-7 per kelvin [10]. I read only that summary, so I give it as indicative. It does support the idea that a numeric window of mismatch decides whether a net forms.

Uncertainty. I put my confidence that mismatch is the primary cause of most crazing at about 0.7, which is where it was. For "recipe fixes more often than thinner coats," I put 0.65, a little lower, because no census backs it. The calculated expansion has no error bar in the sources. Digitalfire states only that it is comparative [4]. I will not invent one. A reader should treat the 0.1 and 0.5 shifts above as direction signals, not as measured values with a plus or minus.

Sensitivity: which assumption moves the result most

The assumption that the body stays fixed. All three recipes carry a fit that depends on the clay body. Digitalfire notes that adjustments may be needed on low silica bodies. It also says the body route costs plasticity and raises the risk of thermal cracking. In that route you raise body expansion so the body contracts more and puts the glaze in compression [1][6]. If you change the body, the same glaze moves from craze to fit. That swamps any recipe edit. Mismatch is a difference, and I only changed one side of it.

The factor set. A different published set gives a different table [4]. A rank order might survive, but the size of the 0.5 target would not.

The thickness effect. Suppose glaze thickness shifts the stress by an amount comparable to a 0.3 to 0.5 unit recipe change. Then a thinner coat is as good as a recipe edit for a glaze near the edge of fit. The tile example [1] suggests thickness matters most for glazes already near their limit, but I have no number for how much. This is the assumption that could overturn my view. The finite element result of a stress minimum at one thickness [2] means a coat can be too thin as well as too thick. A one way "thin it out" rule ignores this.

The lab versus the studio. Calculated values come from composition. A real kiln has a cooling rate, and my own blind spot is to assume lab data match a studio kiln. Slow cooling is the condition named in the crack net range [10]. A fast cooled kiln may craze a different set of glazes.

I also should name what I undervalue. Potters learn the feel of a fit from hundreds of firings, and I have none of that in these tables. I take their report that a thin coat worked as real information. I read it as a sign that the glaze sat near its limit.

The test I would trust

Take the crazing glaze. Fire it at cone 6 oxidation on the same body at two thicknesses, one normal and one about half. Fire a third tile with one recipe change that lowers calculated expansion by about 0.5. Swap a sodium or potassium source for CaO or MgO, or use a lower expansion frit at 30% or more. Fire all three in one load, so cooling is equal. If only the thin tile survives, the margin is slim and the recipe is still wrong. If the recipe change tile survives at full thickness, you fixed the cause. If none survive, the body side needs work.

Which test result would change my mind? Suppose thin coats alone cured crazing on most glazes whose calculated expansion sat at 7.5 or above. Then I would drop to 0.4 on the recipe-first claim. I do not expect that.