Let’s try to sort this out. As me2 stated so well, steel is a mixture of carbon and iron. On a whole it is not a compound, it is just a mixture. Because iron is put together in a crystalline way via metallic bonding it does not have molecules, only atoms stacked in an orderly repeating way, this is why you must be very careful mentioning “molecules” in steel since most often you will be wrong. The exception is carbide which is a chemical compound and the two work differently within the same material we call steel.
Let’s deal with the grains first. In metallurgy “grains” is another word for crystal, but it is also more specifically a unit with a single crystalline orientation. As is illustrated here:
The term “grain direction of the steel” is colloquial nonsense not used in metallurgy, that would be the anisotropic condition and since it is not affected by heat treatment I will not discuss it here. At room temperature the atoms within the crystalline lattice that defines a single grain are stacked in a body centered cubic (bcc) configuration. Bcc does not have many spaces for smaller carbon atoms to fit between the iron atoms so the mixture is very segregated into large iron fields with pockets of carbon in them. But when we heat the iron atoms to 1335F they shift to a new stacking that has lots of spaces for carbon atoms to fit between the iron atoms and we form a more uniform mixture called a solid solution.
This is important because this solid solution is called austenite and it is austenite grains that we refer to whenever we are discussing “grains”. At 1335F when the atoms make the shift to face centered cubic (fcc) they quickly start taking carbon atoms between them to form austenite. New baby grains will form at the corners of the old grain edges and as they dissolve more carbon they will grow to replace the old grains entirely. Once the new grains have replaced the old there will be an equilibrium for a time but with added heat the new grains will then begin to eat each other and continue getting even larger, this is grain growth. Grains reform around the 1335F range (in a simple iron/carbon system) so that defines the temperatures we use for simple grain refinement, which is the easiest and most common occurrence in heated steel. We do it every time we heat the steel to glowing, so it is no big feat and is actually the most basic skill we can master. If you have large grains simply reheat it, with better control this time, and you get all new ones.
Carbides are different, they are the carbon atoms that are not in solution and have gathered in a great enough concentration to chemically bond with the iron (or other metallic elements that may be present); for example simple iron carbide, known as cementite, is Fe3C. These carbides rest outside the iron and iron-carbon solution, and prefer places like the grain boundaries. When we anneal the steel we grow larger carbides by taking as much carbon out of solution as possible to bond in the carbide and leaving soft unfortified fields of iron. When we reheat to make the solution of austenite we dissolve these carbides and put them into the solid solution mixture:
Since carbides are chemically bonded they do not break up and move as easy as the free carbon atoms so we need to throw more heat at them to dissolve them and disperse their carbon evenly, that is why we have soak times and higher temperatures for alloyed steels. When we have things like chrome or vanadium the bonds are very tight and much greater heat is involved in breaking the carbides and that is why you will see temps in excess of 1800F for more richly alloyed steels. If you were to add .5 or .6 percent vanadium to 1084 you could easily reduce its maximum hardness to that of 1040 steel because all of its carbon would be locked in very tough carbides with none left over for proper hardening without extreme overheating.
The other thing to remember about carbide is that it is extremely hard, much harder than the surrounding steel, and this makes it brittle. So having large clusters of carbide is seldom a good thing. It makes machining impossible and can lead to overall embrittlement. Since it is so much harder to abrade than the surrounding material knife edges can only get as fine as the carbides dictate as they will either resist sharpening efforts or pull out and continuously leave large voids in the edge. You can wear away a portion of a grain but no so with a carbide. The worst place to have carbide gather en mass is in the grain boundaries. The preferred path of travel for fracture is often the grain boundaries and if you fill them with the most brittle component of steel you have a weak blade even if it is relatively soft. The ideal carbide condition is as fine as possible and spread and evenly as possible throughout the grains.
How do we control this? It is actually quite easy once we understand the temperature-diffusion relationships. My image of the 1095 with the with the with grain boundaries that me2 linked to is bad but something I did intentionally. I regularly use carbide to highlight grain boundaries when doing metallography. I have had visitors to the shop scratch their heads and ask “you mean you can put carbides exactly where you want inside the steel?” Yes I can, and so can you!
The trick with carbide is tow things, how hot you get it and how fast you cool it. Going hotter dissolves carbides and puts them into solution. Cooling slowly makes carbides, and the slower you cool the larger they get. But be careful since where they will want to grow is seldom what you want, e.g. the grain boundaries. This is why you should avoid slow cooling steels with more than .85% carbon from above “critical”; air cooling is as slow as you want to go. A cooling forge, or a bucket of vermiculite is no place for a hot piece of 1095 or W2!
Proper normalizing, 1600F+, dissolves carbides and even grows grain, but the air cooling leaves you with finer carbides and uniform grain size. Lower temperature thermal cycles will then refine the grain size.
I am sorry if I seem to be repeating me2’s excellent description but I though a reiteration may help people understand better,